2005-04-16 16:20:36 -06:00
|
|
|
/*
|
|
|
|
* linux/fs/namei.c
|
|
|
|
*
|
|
|
|
* Copyright (C) 1991, 1992 Linus Torvalds
|
|
|
|
*/
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Some corrections by tytso.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
|
|
|
|
* lookup logic.
|
|
|
|
*/
|
|
|
|
/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/init.h>
|
2011-11-16 21:57:37 -07:00
|
|
|
#include <linux/export.h>
|
2012-05-23 21:12:50 -06:00
|
|
|
#include <linux/kernel.h>
|
2005-04-16 16:20:36 -06:00
|
|
|
#include <linux/slab.h>
|
|
|
|
#include <linux/fs.h>
|
|
|
|
#include <linux/namei.h>
|
|
|
|
#include <linux/pagemap.h>
|
[PATCH] inotify
inotify is intended to correct the deficiencies of dnotify, particularly
its inability to scale and its terrible user interface:
* dnotify requires the opening of one fd per each directory
that you intend to watch. This quickly results in too many
open files and pins removable media, preventing unmount.
* dnotify is directory-based. You only learn about changes to
directories. Sure, a change to a file in a directory affects
the directory, but you are then forced to keep a cache of
stat structures.
* dnotify's interface to user-space is awful. Signals?
inotify provides a more usable, simple, powerful solution to file change
notification:
* inotify's interface is a system call that returns a fd, not SIGIO.
You get a single fd, which is select()-able.
* inotify has an event that says "the filesystem that the item
you were watching is on was unmounted."
* inotify can watch directories or files.
Inotify is currently used by Beagle (a desktop search infrastructure),
Gamin (a FAM replacement), and other projects.
See Documentation/filesystems/inotify.txt.
Signed-off-by: Robert Love <rml@novell.com>
Cc: John McCutchan <ttb@tentacle.dhs.org>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-07-12 15:06:03 -06:00
|
|
|
#include <linux/fsnotify.h>
|
2005-04-16 16:20:36 -06:00
|
|
|
#include <linux/personality.h>
|
|
|
|
#include <linux/security.h>
|
2009-02-04 07:06:57 -07:00
|
|
|
#include <linux/ima.h>
|
2005-04-16 16:20:36 -06:00
|
|
|
#include <linux/syscalls.h>
|
|
|
|
#include <linux/mount.h>
|
|
|
|
#include <linux/audit.h>
|
2006-01-11 13:17:46 -07:00
|
|
|
#include <linux/capability.h>
|
2005-10-18 15:20:16 -06:00
|
|
|
#include <linux/file.h>
|
2006-01-18 18:43:53 -07:00
|
|
|
#include <linux/fcntl.h>
|
2008-04-29 02:00:10 -06:00
|
|
|
#include <linux/device_cgroup.h>
|
2009-03-29 17:50:06 -06:00
|
|
|
#include <linux/fs_struct.h>
|
2011-07-22 20:30:19 -06:00
|
|
|
#include <linux/posix_acl.h>
|
vfs: fix bad hashing of dentries
Josef Bacik found a performance regression between 3.2 and 3.10 and
narrowed it down to commit bfcfaa77bdf0 ("vfs: use 'unsigned long'
accesses for dcache name comparison and hashing"). He reports:
"The test case is essentially
for (i = 0; i < 1000000; i++)
mkdir("a$i");
On xfs on a fio card this goes at about 20k dir/sec with 3.2, and 12k
dir/sec with 3.10. This is because we spend waaaaay more time in
__d_lookup on 3.10 than in 3.2.
The new hashing function for strings is suboptimal for <
sizeof(unsigned long) string names (and hell even > sizeof(unsigned
long) string names that I've tested). I broke out the old hashing
function and the new one into a userspace helper to get real numbers
and this is what I'm getting:
Old hash table had 1000000 entries, 0 dupes, 0 max dupes
New hash table had 12628 entries, 987372 dupes, 900 max dupes
We had 11400 buckets with a p50 of 30 dupes, p90 of 240 dupes, p99 of 567 dupes for the new hash
My test does the hash, and then does the d_hash into a integer pointer
array the same size as the dentry hash table on my system, and then
just increments the value at the address we got to see how many
entries we overlap with.
As you can see the old hash function ended up with all 1 million
entries in their own bucket, whereas the new one they are only
distributed among ~12.5k buckets, which is why we're using so much
more CPU in __d_lookup".
The reason for this hash regression is two-fold:
- On 64-bit architectures the down-mixing of the original 64-bit
word-at-a-time hash into the final 32-bit hash value is very
simplistic and suboptimal, and just adds the two 32-bit parts
together.
In particular, because there is no bit shuffling and the mixing
boundary is also a byte boundary, similar character patterns in the
low and high word easily end up just canceling each other out.
- the old byte-at-a-time hash mixed each byte into the final hash as it
hashed the path component name, resulting in the low bits of the hash
generally being a good source of hash data. That is not true for the
word-at-a-time case, and the hash data is distributed among all the
bits.
The fix is the same in both cases: do a better job of mixing the bits up
and using as much of the hash data as possible. We already have the
"hash_32|64()" functions to do that.
Reported-by: Josef Bacik <jbacik@fb.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Chris Mason <clm@fb.com>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-09-13 12:30:10 -06:00
|
|
|
#include <linux/hash.h>
|
2005-04-16 16:20:36 -06:00
|
|
|
#include <asm/uaccess.h>
|
|
|
|
|
2009-12-04 13:47:36 -07:00
|
|
|
#include "internal.h"
|
2011-11-24 16:22:03 -07:00
|
|
|
#include "mount.h"
|
2009-12-04 13:47:36 -07:00
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
/* [Feb-1997 T. Schoebel-Theuer]
|
|
|
|
* Fundamental changes in the pathname lookup mechanisms (namei)
|
|
|
|
* were necessary because of omirr. The reason is that omirr needs
|
|
|
|
* to know the _real_ pathname, not the user-supplied one, in case
|
|
|
|
* of symlinks (and also when transname replacements occur).
|
|
|
|
*
|
|
|
|
* The new code replaces the old recursive symlink resolution with
|
|
|
|
* an iterative one (in case of non-nested symlink chains). It does
|
|
|
|
* this with calls to <fs>_follow_link().
|
|
|
|
* As a side effect, dir_namei(), _namei() and follow_link() are now
|
|
|
|
* replaced with a single function lookup_dentry() that can handle all
|
|
|
|
* the special cases of the former code.
|
|
|
|
*
|
|
|
|
* With the new dcache, the pathname is stored at each inode, at least as
|
|
|
|
* long as the refcount of the inode is positive. As a side effect, the
|
|
|
|
* size of the dcache depends on the inode cache and thus is dynamic.
|
|
|
|
*
|
|
|
|
* [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
|
|
|
|
* resolution to correspond with current state of the code.
|
|
|
|
*
|
|
|
|
* Note that the symlink resolution is not *completely* iterative.
|
|
|
|
* There is still a significant amount of tail- and mid- recursion in
|
|
|
|
* the algorithm. Also, note that <fs>_readlink() is not used in
|
|
|
|
* lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
|
|
|
|
* may return different results than <fs>_follow_link(). Many virtual
|
|
|
|
* filesystems (including /proc) exhibit this behavior.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
|
|
|
|
* New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
|
|
|
|
* and the name already exists in form of a symlink, try to create the new
|
|
|
|
* name indicated by the symlink. The old code always complained that the
|
|
|
|
* name already exists, due to not following the symlink even if its target
|
|
|
|
* is nonexistent. The new semantics affects also mknod() and link() when
|
2011-03-30 19:57:33 -06:00
|
|
|
* the name is a symlink pointing to a non-existent name.
|
2005-04-16 16:20:36 -06:00
|
|
|
*
|
|
|
|
* I don't know which semantics is the right one, since I have no access
|
|
|
|
* to standards. But I found by trial that HP-UX 9.0 has the full "new"
|
|
|
|
* semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
|
|
|
|
* "old" one. Personally, I think the new semantics is much more logical.
|
|
|
|
* Note that "ln old new" where "new" is a symlink pointing to a non-existing
|
|
|
|
* file does succeed in both HP-UX and SunOs, but not in Solaris
|
|
|
|
* and in the old Linux semantics.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
|
|
|
|
* semantics. See the comments in "open_namei" and "do_link" below.
|
|
|
|
*
|
|
|
|
* [10-Sep-98 Alan Modra] Another symlink change.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
|
|
|
|
* inside the path - always follow.
|
|
|
|
* in the last component in creation/removal/renaming - never follow.
|
|
|
|
* if LOOKUP_FOLLOW passed - follow.
|
|
|
|
* if the pathname has trailing slashes - follow.
|
|
|
|
* otherwise - don't follow.
|
|
|
|
* (applied in that order).
|
|
|
|
*
|
|
|
|
* [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
|
|
|
|
* restored for 2.4. This is the last surviving part of old 4.2BSD bug.
|
|
|
|
* During the 2.4 we need to fix the userland stuff depending on it -
|
|
|
|
* hopefully we will be able to get rid of that wart in 2.5. So far only
|
|
|
|
* XEmacs seems to be relying on it...
|
|
|
|
*/
|
|
|
|
/*
|
|
|
|
* [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
|
2006-03-23 04:00:33 -07:00
|
|
|
* implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
|
2005-04-16 16:20:36 -06:00
|
|
|
* any extra contention...
|
|
|
|
*/
|
|
|
|
|
|
|
|
/* In order to reduce some races, while at the same time doing additional
|
|
|
|
* checking and hopefully speeding things up, we copy filenames to the
|
|
|
|
* kernel data space before using them..
|
|
|
|
*
|
|
|
|
* POSIX.1 2.4: an empty pathname is invalid (ENOENT).
|
|
|
|
* PATH_MAX includes the nul terminator --RR.
|
|
|
|
*/
|
2012-10-10 13:25:28 -06:00
|
|
|
|
2015-02-22 18:07:13 -07:00
|
|
|
#define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
|
2012-10-10 14:43:13 -06:00
|
|
|
|
syscalls: implement execveat() system call
This patchset adds execveat(2) for x86, and is derived from Meredydd
Luff's patch from Sept 2012 (https://lkml.org/lkml/2012/9/11/528).
The primary aim of adding an execveat syscall is to allow an
implementation of fexecve(3) that does not rely on the /proc filesystem,
at least for executables (rather than scripts). The current glibc version
of fexecve(3) is implemented via /proc, which causes problems in sandboxed
or otherwise restricted environments.
Given the desire for a /proc-free fexecve() implementation, HPA suggested
(https://lkml.org/lkml/2006/7/11/556) that an execveat(2) syscall would be
an appropriate generalization.
Also, having a new syscall means that it can take a flags argument without
back-compatibility concerns. The current implementation just defines the
AT_EMPTY_PATH and AT_SYMLINK_NOFOLLOW flags, but other flags could be
added in future -- for example, flags for new namespaces (as suggested at
https://lkml.org/lkml/2006/7/11/474).
Related history:
- https://lkml.org/lkml/2006/12/27/123 is an example of someone
realizing that fexecve() is likely to fail in a chroot environment.
- http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=514043 covered
documenting the /proc requirement of fexecve(3) in its manpage, to
"prevent other people from wasting their time".
- https://bugzilla.redhat.com/show_bug.cgi?id=241609 described a
problem where a process that did setuid() could not fexecve()
because it no longer had access to /proc/self/fd; this has since
been fixed.
This patch (of 4):
Add a new execveat(2) system call. execveat() is to execve() as openat()
is to open(): it takes a file descriptor that refers to a directory, and
resolves the filename relative to that.
In addition, if the filename is empty and AT_EMPTY_PATH is specified,
execveat() executes the file to which the file descriptor refers. This
replicates the functionality of fexecve(), which is a system call in other
UNIXen, but in Linux glibc it depends on opening "/proc/self/fd/<fd>" (and
so relies on /proc being mounted).
The filename fed to the executed program as argv[0] (or the name of the
script fed to a script interpreter) will be of the form "/dev/fd/<fd>"
(for an empty filename) or "/dev/fd/<fd>/<filename>", effectively
reflecting how the executable was found. This does however mean that
execution of a script in a /proc-less environment won't work; also, script
execution via an O_CLOEXEC file descriptor fails (as the file will not be
accessible after exec).
Based on patches by Meredydd Luff.
Signed-off-by: David Drysdale <drysdale@google.com>
Cc: Meredydd Luff <meredydd@senatehouse.org>
Cc: Shuah Khan <shuah.kh@samsung.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Rich Felker <dalias@aerifal.cx>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-12-12 17:57:29 -07:00
|
|
|
struct filename *
|
2012-10-10 13:25:28 -06:00
|
|
|
getname_flags(const char __user *filename, int flags, int *empty)
|
|
|
|
{
|
2015-02-22 17:38:03 -07:00
|
|
|
struct filename *result;
|
2012-10-10 14:43:13 -06:00
|
|
|
char *kname;
|
2015-02-22 17:38:03 -07:00
|
|
|
int len;
|
2012-01-03 12:23:08 -07:00
|
|
|
|
2012-10-10 13:25:28 -06:00
|
|
|
result = audit_reusename(filename);
|
|
|
|
if (result)
|
|
|
|
return result;
|
|
|
|
|
2012-10-10 14:43:13 -06:00
|
|
|
result = __getname();
|
2012-04-28 15:38:32 -06:00
|
|
|
if (unlikely(!result))
|
2012-01-03 12:23:08 -07:00
|
|
|
return ERR_PTR(-ENOMEM);
|
|
|
|
|
2012-10-10 14:43:13 -06:00
|
|
|
/*
|
|
|
|
* First, try to embed the struct filename inside the names_cache
|
|
|
|
* allocation
|
|
|
|
*/
|
2015-02-22 18:07:13 -07:00
|
|
|
kname = (char *)result->iname;
|
2012-10-10 13:25:28 -06:00
|
|
|
result->name = kname;
|
2012-10-10 14:43:13 -06:00
|
|
|
|
2015-02-22 17:38:03 -07:00
|
|
|
len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
|
2012-10-10 13:25:28 -06:00
|
|
|
if (unlikely(len < 0)) {
|
2015-02-22 17:38:03 -07:00
|
|
|
__putname(result);
|
|
|
|
return ERR_PTR(len);
|
2012-10-10 13:25:28 -06:00
|
|
|
}
|
2012-04-28 15:38:32 -06:00
|
|
|
|
2012-10-10 14:43:13 -06:00
|
|
|
/*
|
|
|
|
* Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
|
|
|
|
* separate struct filename so we can dedicate the entire
|
|
|
|
* names_cache allocation for the pathname, and re-do the copy from
|
|
|
|
* userland.
|
|
|
|
*/
|
2015-02-22 17:38:03 -07:00
|
|
|
if (unlikely(len == EMBEDDED_NAME_MAX)) {
|
2015-02-22 18:07:13 -07:00
|
|
|
const size_t size = offsetof(struct filename, iname[1]);
|
2012-10-10 14:43:13 -06:00
|
|
|
kname = (char *)result;
|
|
|
|
|
2015-02-22 18:07:13 -07:00
|
|
|
/*
|
|
|
|
* size is chosen that way we to guarantee that
|
|
|
|
* result->iname[0] is within the same object and that
|
|
|
|
* kname can't be equal to result->iname, no matter what.
|
|
|
|
*/
|
|
|
|
result = kzalloc(size, GFP_KERNEL);
|
2015-02-22 17:38:03 -07:00
|
|
|
if (unlikely(!result)) {
|
|
|
|
__putname(kname);
|
|
|
|
return ERR_PTR(-ENOMEM);
|
2012-10-10 14:43:13 -06:00
|
|
|
}
|
|
|
|
result->name = kname;
|
2015-02-22 17:38:03 -07:00
|
|
|
len = strncpy_from_user(kname, filename, PATH_MAX);
|
|
|
|
if (unlikely(len < 0)) {
|
|
|
|
__putname(kname);
|
|
|
|
kfree(result);
|
|
|
|
return ERR_PTR(len);
|
|
|
|
}
|
|
|
|
if (unlikely(len == PATH_MAX)) {
|
|
|
|
__putname(kname);
|
|
|
|
kfree(result);
|
|
|
|
return ERR_PTR(-ENAMETOOLONG);
|
|
|
|
}
|
2012-10-10 14:43:13 -06:00
|
|
|
}
|
|
|
|
|
2015-02-22 17:38:03 -07:00
|
|
|
result->refcnt = 1;
|
2012-04-28 15:38:32 -06:00
|
|
|
/* The empty path is special. */
|
|
|
|
if (unlikely(!len)) {
|
|
|
|
if (empty)
|
2012-01-03 12:23:08 -07:00
|
|
|
*empty = 1;
|
2015-02-22 17:38:03 -07:00
|
|
|
if (!(flags & LOOKUP_EMPTY)) {
|
|
|
|
putname(result);
|
|
|
|
return ERR_PTR(-ENOENT);
|
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
2012-04-28 15:38:32 -06:00
|
|
|
|
2012-10-10 14:43:13 -06:00
|
|
|
result->uptr = filename;
|
2014-02-05 13:54:53 -07:00
|
|
|
result->aname = NULL;
|
2012-10-10 14:43:13 -06:00
|
|
|
audit_getname(result);
|
|
|
|
return result;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2012-10-10 13:25:28 -06:00
|
|
|
struct filename *
|
|
|
|
getname(const char __user * filename)
|
2011-03-14 16:56:51 -06:00
|
|
|
{
|
2012-03-22 17:10:40 -06:00
|
|
|
return getname_flags(filename, 0, NULL);
|
2011-03-14 16:56:51 -06:00
|
|
|
}
|
|
|
|
|
2014-02-05 13:54:53 -07:00
|
|
|
struct filename *
|
|
|
|
getname_kernel(const char * filename)
|
|
|
|
{
|
|
|
|
struct filename *result;
|
2015-01-21 21:59:56 -07:00
|
|
|
int len = strlen(filename) + 1;
|
2014-02-05 13:54:53 -07:00
|
|
|
|
|
|
|
result = __getname();
|
|
|
|
if (unlikely(!result))
|
|
|
|
return ERR_PTR(-ENOMEM);
|
|
|
|
|
2015-01-21 21:59:56 -07:00
|
|
|
if (len <= EMBEDDED_NAME_MAX) {
|
2015-02-22 18:07:13 -07:00
|
|
|
result->name = (char *)result->iname;
|
2015-01-21 21:59:56 -07:00
|
|
|
} else if (len <= PATH_MAX) {
|
|
|
|
struct filename *tmp;
|
|
|
|
|
|
|
|
tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
|
|
|
|
if (unlikely(!tmp)) {
|
|
|
|
__putname(result);
|
|
|
|
return ERR_PTR(-ENOMEM);
|
|
|
|
}
|
|
|
|
tmp->name = (char *)result;
|
|
|
|
result = tmp;
|
|
|
|
} else {
|
|
|
|
__putname(result);
|
|
|
|
return ERR_PTR(-ENAMETOOLONG);
|
|
|
|
}
|
|
|
|
memcpy((char *)result->name, filename, len);
|
2014-02-05 13:54:53 -07:00
|
|
|
result->uptr = NULL;
|
|
|
|
result->aname = NULL;
|
2015-01-21 22:00:23 -07:00
|
|
|
result->refcnt = 1;
|
2015-01-21 22:00:10 -07:00
|
|
|
audit_getname(result);
|
2014-02-05 13:54:53 -07:00
|
|
|
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
2012-10-10 13:25:28 -06:00
|
|
|
void putname(struct filename *name)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2015-01-21 22:00:23 -07:00
|
|
|
BUG_ON(name->refcnt <= 0);
|
|
|
|
|
|
|
|
if (--name->refcnt > 0)
|
|
|
|
return;
|
|
|
|
|
2015-02-22 18:07:13 -07:00
|
|
|
if (name->name != name->iname) {
|
2015-01-21 22:00:23 -07:00
|
|
|
__putname(name->name);
|
|
|
|
kfree(name);
|
|
|
|
} else
|
|
|
|
__putname(name);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2011-07-22 20:30:19 -06:00
|
|
|
static int check_acl(struct inode *inode, int mask)
|
|
|
|
{
|
2011-07-25 23:47:03 -06:00
|
|
|
#ifdef CONFIG_FS_POSIX_ACL
|
2011-07-22 20:30:19 -06:00
|
|
|
struct posix_acl *acl;
|
|
|
|
|
|
|
|
if (mask & MAY_NOT_BLOCK) {
|
2011-08-02 19:32:13 -06:00
|
|
|
acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
|
|
|
|
if (!acl)
|
2011-07-22 20:30:19 -06:00
|
|
|
return -EAGAIN;
|
2011-08-02 19:32:13 -06:00
|
|
|
/* no ->get_acl() calls in RCU mode... */
|
|
|
|
if (acl == ACL_NOT_CACHED)
|
|
|
|
return -ECHILD;
|
2011-08-06 10:43:07 -06:00
|
|
|
return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
|
2011-07-22 20:30:19 -06:00
|
|
|
}
|
|
|
|
|
2013-12-20 06:16:38 -07:00
|
|
|
acl = get_acl(inode, ACL_TYPE_ACCESS);
|
|
|
|
if (IS_ERR(acl))
|
|
|
|
return PTR_ERR(acl);
|
2011-07-22 20:30:19 -06:00
|
|
|
if (acl) {
|
|
|
|
int error = posix_acl_permission(inode, acl, mask);
|
|
|
|
posix_acl_release(acl);
|
|
|
|
return error;
|
|
|
|
}
|
2011-07-25 23:47:03 -06:00
|
|
|
#endif
|
2011-07-22 20:30:19 -06:00
|
|
|
|
|
|
|
return -EAGAIN;
|
|
|
|
}
|
|
|
|
|
2009-08-28 12:51:25 -06:00
|
|
|
/*
|
2011-10-23 11:43:33 -06:00
|
|
|
* This does the basic permission checking
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
2011-06-20 17:12:17 -06:00
|
|
|
static int acl_permission_check(struct inode *inode, int mask)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2011-05-13 12:51:01 -06:00
|
|
|
unsigned int mode = inode->i_mode;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2012-03-03 22:17:15 -07:00
|
|
|
if (likely(uid_eq(current_fsuid(), inode->i_uid)))
|
2005-04-16 16:20:36 -06:00
|
|
|
mode >>= 6;
|
|
|
|
else {
|
2011-07-22 20:30:19 -06:00
|
|
|
if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
|
2011-06-20 17:12:17 -06:00
|
|
|
int error = check_acl(inode, mask);
|
2011-01-06 23:49:58 -07:00
|
|
|
if (error != -EAGAIN)
|
|
|
|
return error;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
if (in_group_p(inode->i_gid))
|
|
|
|
mode >>= 3;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If the DACs are ok we don't need any capability check.
|
|
|
|
*/
|
2011-06-20 17:06:22 -06:00
|
|
|
if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
|
2005-04-16 16:20:36 -06:00
|
|
|
return 0;
|
2009-08-28 12:51:25 -06:00
|
|
|
return -EACCES;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
2011-01-06 23:49:58 -07:00
|
|
|
* generic_permission - check for access rights on a Posix-like filesystem
|
2009-08-28 12:51:25 -06:00
|
|
|
* @inode: inode to check access rights for
|
2011-10-23 11:43:30 -06:00
|
|
|
* @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
|
2009-08-28 12:51:25 -06:00
|
|
|
*
|
|
|
|
* Used to check for read/write/execute permissions on a file.
|
|
|
|
* We use "fsuid" for this, letting us set arbitrary permissions
|
|
|
|
* for filesystem access without changing the "normal" uids which
|
2011-01-06 23:49:58 -07:00
|
|
|
* are used for other things.
|
|
|
|
*
|
|
|
|
* generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
|
|
|
|
* request cannot be satisfied (eg. requires blocking or too much complexity).
|
|
|
|
* It would then be called again in ref-walk mode.
|
2009-08-28 12:51:25 -06:00
|
|
|
*/
|
2011-06-20 17:16:29 -06:00
|
|
|
int generic_permission(struct inode *inode, int mask)
|
2009-08-28 12:51:25 -06:00
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
/*
|
2011-10-23 11:43:33 -06:00
|
|
|
* Do the basic permission checks.
|
2009-08-28 12:51:25 -06:00
|
|
|
*/
|
2011-06-20 17:12:17 -06:00
|
|
|
ret = acl_permission_check(inode, mask);
|
2009-08-28 12:51:25 -06:00
|
|
|
if (ret != -EACCES)
|
|
|
|
return ret;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2011-06-20 17:55:42 -06:00
|
|
|
if (S_ISDIR(inode->i_mode)) {
|
|
|
|
/* DACs are overridable for directories */
|
2014-06-10 13:45:42 -06:00
|
|
|
if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
|
2011-06-20 17:55:42 -06:00
|
|
|
return 0;
|
|
|
|
if (!(mask & MAY_WRITE))
|
2014-06-10 13:45:42 -06:00
|
|
|
if (capable_wrt_inode_uidgid(inode,
|
|
|
|
CAP_DAC_READ_SEARCH))
|
2011-06-20 17:55:42 -06:00
|
|
|
return 0;
|
|
|
|
return -EACCES;
|
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
/*
|
|
|
|
* Read/write DACs are always overridable.
|
2011-06-20 17:55:42 -06:00
|
|
|
* Executable DACs are overridable when there is
|
|
|
|
* at least one exec bit set.
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
2011-06-20 17:55:42 -06:00
|
|
|
if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
|
2014-06-10 13:45:42 -06:00
|
|
|
if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
|
2005-04-16 16:20:36 -06:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Searching includes executable on directories, else just read.
|
|
|
|
*/
|
2009-12-29 13:50:19 -07:00
|
|
|
mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
|
2011-06-20 17:55:42 -06:00
|
|
|
if (mask == MAY_READ)
|
2014-06-10 13:45:42 -06:00
|
|
|
if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
|
2005-04-16 16:20:36 -06:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
return -EACCES;
|
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(generic_permission);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2011-08-06 23:45:50 -06:00
|
|
|
/*
|
|
|
|
* We _really_ want to just do "generic_permission()" without
|
|
|
|
* even looking at the inode->i_op values. So we keep a cache
|
|
|
|
* flag in inode->i_opflags, that says "this has not special
|
|
|
|
* permission function, use the fast case".
|
|
|
|
*/
|
|
|
|
static inline int do_inode_permission(struct inode *inode, int mask)
|
|
|
|
{
|
|
|
|
if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
|
|
|
|
if (likely(inode->i_op->permission))
|
|
|
|
return inode->i_op->permission(inode, mask);
|
|
|
|
|
|
|
|
/* This gets set once for the inode lifetime */
|
|
|
|
spin_lock(&inode->i_lock);
|
|
|
|
inode->i_opflags |= IOP_FASTPERM;
|
|
|
|
spin_unlock(&inode->i_lock);
|
|
|
|
}
|
|
|
|
return generic_permission(inode, mask);
|
|
|
|
}
|
|
|
|
|
2008-10-24 01:59:29 -06:00
|
|
|
/**
|
2012-06-25 05:55:46 -06:00
|
|
|
* __inode_permission - Check for access rights to a given inode
|
|
|
|
* @inode: Inode to check permission on
|
|
|
|
* @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
|
2008-10-24 01:59:29 -06:00
|
|
|
*
|
2012-06-25 05:55:46 -06:00
|
|
|
* Check for read/write/execute permissions on an inode.
|
2011-10-23 11:43:33 -06:00
|
|
|
*
|
|
|
|
* When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
|
2012-06-25 05:55:46 -06:00
|
|
|
*
|
|
|
|
* This does not check for a read-only file system. You probably want
|
|
|
|
* inode_permission().
|
2008-10-24 01:59:29 -06:00
|
|
|
*/
|
2012-06-25 05:55:46 -06:00
|
|
|
int __inode_permission(struct inode *inode, int mask)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2008-07-15 19:03:57 -06:00
|
|
|
int retval;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2011-08-06 23:45:50 -06:00
|
|
|
if (unlikely(mask & MAY_WRITE)) {
|
2005-04-16 16:20:36 -06:00
|
|
|
/*
|
|
|
|
* Nobody gets write access to an immutable file.
|
|
|
|
*/
|
|
|
|
if (IS_IMMUTABLE(inode))
|
|
|
|
return -EACCES;
|
|
|
|
}
|
|
|
|
|
2011-08-06 23:45:50 -06:00
|
|
|
retval = do_inode_permission(inode, mask);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (retval)
|
|
|
|
return retval;
|
|
|
|
|
2008-04-29 02:00:10 -06:00
|
|
|
retval = devcgroup_inode_permission(inode, mask);
|
|
|
|
if (retval)
|
|
|
|
return retval;
|
|
|
|
|
2010-07-23 09:43:57 -06:00
|
|
|
return security_inode_permission(inode, mask);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
2014-10-23 16:14:35 -06:00
|
|
|
EXPORT_SYMBOL(__inode_permission);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2012-06-25 05:55:46 -06:00
|
|
|
/**
|
|
|
|
* sb_permission - Check superblock-level permissions
|
|
|
|
* @sb: Superblock of inode to check permission on
|
2012-08-18 18:39:25 -06:00
|
|
|
* @inode: Inode to check permission on
|
2012-06-25 05:55:46 -06:00
|
|
|
* @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
|
|
|
|
*
|
|
|
|
* Separate out file-system wide checks from inode-specific permission checks.
|
|
|
|
*/
|
|
|
|
static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
|
|
|
|
{
|
|
|
|
if (unlikely(mask & MAY_WRITE)) {
|
|
|
|
umode_t mode = inode->i_mode;
|
|
|
|
|
|
|
|
/* Nobody gets write access to a read-only fs. */
|
|
|
|
if ((sb->s_flags & MS_RDONLY) &&
|
|
|
|
(S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
|
|
|
|
return -EROFS;
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* inode_permission - Check for access rights to a given inode
|
|
|
|
* @inode: Inode to check permission on
|
|
|
|
* @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
|
|
|
|
*
|
|
|
|
* Check for read/write/execute permissions on an inode. We use fs[ug]id for
|
|
|
|
* this, letting us set arbitrary permissions for filesystem access without
|
|
|
|
* changing the "normal" UIDs which are used for other things.
|
|
|
|
*
|
|
|
|
* When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
|
|
|
|
*/
|
|
|
|
int inode_permission(struct inode *inode, int mask)
|
|
|
|
{
|
|
|
|
int retval;
|
|
|
|
|
|
|
|
retval = sb_permission(inode->i_sb, inode, mask);
|
|
|
|
if (retval)
|
|
|
|
return retval;
|
|
|
|
return __inode_permission(inode, mask);
|
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(inode_permission);
|
2012-06-25 05:55:46 -06:00
|
|
|
|
2008-02-14 20:34:38 -07:00
|
|
|
/**
|
|
|
|
* path_get - get a reference to a path
|
|
|
|
* @path: path to get the reference to
|
|
|
|
*
|
|
|
|
* Given a path increment the reference count to the dentry and the vfsmount.
|
|
|
|
*/
|
2013-03-01 21:51:07 -07:00
|
|
|
void path_get(const struct path *path)
|
2008-02-14 20:34:38 -07:00
|
|
|
{
|
|
|
|
mntget(path->mnt);
|
|
|
|
dget(path->dentry);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(path_get);
|
|
|
|
|
2008-02-14 20:34:35 -07:00
|
|
|
/**
|
|
|
|
* path_put - put a reference to a path
|
|
|
|
* @path: path to put the reference to
|
|
|
|
*
|
|
|
|
* Given a path decrement the reference count to the dentry and the vfsmount.
|
|
|
|
*/
|
2013-03-01 21:51:07 -07:00
|
|
|
void path_put(const struct path *path)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2008-02-14 20:34:35 -07:00
|
|
|
dput(path->dentry);
|
|
|
|
mntput(path->mnt);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
2008-02-14 20:34:35 -07:00
|
|
|
EXPORT_SYMBOL(path_put);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2015-05-02 05:16:16 -06:00
|
|
|
#define EMBEDDED_LEVELS 2
|
2014-11-01 17:30:41 -06:00
|
|
|
struct nameidata {
|
|
|
|
struct path path;
|
2015-05-06 14:01:56 -06:00
|
|
|
struct qstr last;
|
2014-11-01 17:30:41 -06:00
|
|
|
struct path root;
|
|
|
|
struct inode *inode; /* path.dentry.d_inode */
|
|
|
|
unsigned int flags;
|
2015-05-13 05:28:08 -06:00
|
|
|
unsigned seq, m_seq;
|
2014-11-01 17:30:41 -06:00
|
|
|
int last_type;
|
|
|
|
unsigned depth;
|
2015-03-22 20:37:38 -06:00
|
|
|
int total_link_count;
|
2015-05-02 17:38:35 -06:00
|
|
|
struct saved {
|
|
|
|
struct path link;
|
2015-12-29 13:58:39 -07:00
|
|
|
struct delayed_call done;
|
2015-05-02 17:38:35 -06:00
|
|
|
const char *name;
|
2015-05-08 11:23:53 -06:00
|
|
|
unsigned seq;
|
2015-05-02 05:16:16 -06:00
|
|
|
} *stack, internal[EMBEDDED_LEVELS];
|
2015-05-13 05:28:08 -06:00
|
|
|
struct filename *name;
|
|
|
|
struct nameidata *saved;
|
2015-12-29 13:58:39 -07:00
|
|
|
struct inode *link_inode;
|
2015-05-13 05:28:08 -06:00
|
|
|
unsigned root_seq;
|
|
|
|
int dfd;
|
2014-11-01 17:30:41 -06:00
|
|
|
};
|
|
|
|
|
2015-05-13 05:28:08 -06:00
|
|
|
static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
|
2015-05-02 05:16:16 -06:00
|
|
|
{
|
2015-03-22 20:37:38 -06:00
|
|
|
struct nameidata *old = current->nameidata;
|
|
|
|
p->stack = p->internal;
|
2015-05-12 16:43:07 -06:00
|
|
|
p->dfd = dfd;
|
|
|
|
p->name = name;
|
2015-03-22 20:37:38 -06:00
|
|
|
p->total_link_count = old ? old->total_link_count : 0;
|
2015-05-13 05:28:08 -06:00
|
|
|
p->saved = old;
|
2015-03-22 20:37:38 -06:00
|
|
|
current->nameidata = p;
|
2015-05-02 05:16:16 -06:00
|
|
|
}
|
|
|
|
|
2015-05-13 05:28:08 -06:00
|
|
|
static void restore_nameidata(void)
|
2015-05-02 05:16:16 -06:00
|
|
|
{
|
2015-05-13 05:28:08 -06:00
|
|
|
struct nameidata *now = current->nameidata, *old = now->saved;
|
2015-03-22 20:37:38 -06:00
|
|
|
|
|
|
|
current->nameidata = old;
|
|
|
|
if (old)
|
|
|
|
old->total_link_count = now->total_link_count;
|
2015-12-05 19:06:33 -07:00
|
|
|
if (now->stack != now->internal)
|
2015-03-22 20:37:38 -06:00
|
|
|
kfree(now->stack);
|
2015-05-02 05:16:16 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
static int __nd_alloc_stack(struct nameidata *nd)
|
|
|
|
{
|
2015-05-09 11:04:24 -06:00
|
|
|
struct saved *p;
|
|
|
|
|
|
|
|
if (nd->flags & LOOKUP_RCU) {
|
|
|
|
p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
|
|
|
|
GFP_ATOMIC);
|
|
|
|
if (unlikely(!p))
|
|
|
|
return -ECHILD;
|
|
|
|
} else {
|
|
|
|
p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
|
2015-05-02 05:16:16 -06:00
|
|
|
GFP_KERNEL);
|
2015-05-09 11:04:24 -06:00
|
|
|
if (unlikely(!p))
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
2015-05-02 05:16:16 -06:00
|
|
|
memcpy(p, nd->internal, sizeof(nd->internal));
|
|
|
|
nd->stack = p;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2015-08-15 19:27:13 -06:00
|
|
|
/**
|
|
|
|
* path_connected - Verify that a path->dentry is below path->mnt.mnt_root
|
|
|
|
* @path: nameidate to verify
|
|
|
|
*
|
|
|
|
* Rename can sometimes move a file or directory outside of a bind
|
|
|
|
* mount, path_connected allows those cases to be detected.
|
|
|
|
*/
|
|
|
|
static bool path_connected(const struct path *path)
|
|
|
|
{
|
|
|
|
struct vfsmount *mnt = path->mnt;
|
|
|
|
|
|
|
|
/* Only bind mounts can have disconnected paths */
|
|
|
|
if (mnt->mnt_root == mnt->mnt_sb->s_root)
|
|
|
|
return true;
|
|
|
|
|
|
|
|
return is_subdir(path->dentry, mnt->mnt_root);
|
|
|
|
}
|
|
|
|
|
2015-05-02 05:16:16 -06:00
|
|
|
static inline int nd_alloc_stack(struct nameidata *nd)
|
|
|
|
{
|
2015-05-03 18:52:15 -06:00
|
|
|
if (likely(nd->depth != EMBEDDED_LEVELS))
|
2015-05-02 05:16:16 -06:00
|
|
|
return 0;
|
|
|
|
if (likely(nd->stack != nd->internal))
|
|
|
|
return 0;
|
|
|
|
return __nd_alloc_stack(nd);
|
|
|
|
}
|
|
|
|
|
2015-05-09 10:55:43 -06:00
|
|
|
static void drop_links(struct nameidata *nd)
|
|
|
|
{
|
|
|
|
int i = nd->depth;
|
|
|
|
while (i--) {
|
|
|
|
struct saved *last = nd->stack + i;
|
2015-12-29 13:58:39 -07:00
|
|
|
do_delayed_call(&last->done);
|
|
|
|
clear_delayed_call(&last->done);
|
2015-05-09 10:55:43 -06:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void terminate_walk(struct nameidata *nd)
|
|
|
|
{
|
|
|
|
drop_links(nd);
|
|
|
|
if (!(nd->flags & LOOKUP_RCU)) {
|
|
|
|
int i;
|
|
|
|
path_put(&nd->path);
|
|
|
|
for (i = 0; i < nd->depth; i++)
|
|
|
|
path_put(&nd->stack[i].link);
|
2015-05-12 15:35:52 -06:00
|
|
|
if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
|
|
|
|
path_put(&nd->root);
|
|
|
|
nd->root.mnt = NULL;
|
|
|
|
}
|
2015-05-09 10:55:43 -06:00
|
|
|
} else {
|
|
|
|
nd->flags &= ~LOOKUP_RCU;
|
|
|
|
if (!(nd->flags & LOOKUP_ROOT))
|
|
|
|
nd->root.mnt = NULL;
|
|
|
|
rcu_read_unlock();
|
|
|
|
}
|
|
|
|
nd->depth = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* path_put is needed afterwards regardless of success or failure */
|
|
|
|
static bool legitimize_path(struct nameidata *nd,
|
|
|
|
struct path *path, unsigned seq)
|
|
|
|
{
|
|
|
|
int res = __legitimize_mnt(path->mnt, nd->m_seq);
|
|
|
|
if (unlikely(res)) {
|
|
|
|
if (res > 0)
|
|
|
|
path->mnt = NULL;
|
|
|
|
path->dentry = NULL;
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
|
|
|
|
path->dentry = NULL;
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
return !read_seqcount_retry(&path->dentry->d_seq, seq);
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool legitimize_links(struct nameidata *nd)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
for (i = 0; i < nd->depth; i++) {
|
|
|
|
struct saved *last = nd->stack + i;
|
|
|
|
if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
|
|
|
|
drop_links(nd);
|
|
|
|
nd->depth = i + 1;
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2011-03-25 08:32:48 -06:00
|
|
|
/*
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
* Path walking has 2 modes, rcu-walk and ref-walk (see
|
2011-03-25 08:32:48 -06:00
|
|
|
* Documentation/filesystems/path-lookup.txt). In situations when we can't
|
|
|
|
* continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
|
2015-11-30 09:11:59 -07:00
|
|
|
* normal reference counts on dentries and vfsmounts to transition to ref-walk
|
2011-03-25 08:32:48 -06:00
|
|
|
* mode. Refcounts are grabbed at the last known good point before rcu-walk
|
|
|
|
* got stuck, so ref-walk may continue from there. If this is not successful
|
|
|
|
* (eg. a seqcount has changed), then failure is returned and it's up to caller
|
|
|
|
* to restart the path walk from the beginning in ref-walk mode.
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
*/
|
|
|
|
|
|
|
|
/**
|
2011-03-25 08:32:48 -06:00
|
|
|
* unlazy_walk - try to switch to ref-walk mode.
|
|
|
|
* @nd: nameidata pathwalk data
|
|
|
|
* @dentry: child of nd->path.dentry or NULL
|
2015-05-05 07:26:05 -06:00
|
|
|
* @seq: seq number to check dentry against
|
2011-01-08 20:36:21 -07:00
|
|
|
* Returns: 0 on success, -ECHILD on failure
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
*
|
2011-03-25 08:32:48 -06:00
|
|
|
* unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
|
|
|
|
* for ref-walk mode. @dentry must be a path found by a do_lookup call on
|
|
|
|
* @nd or NULL. Must be called from rcu-walk context.
|
2015-05-09 10:55:43 -06:00
|
|
|
* Nothing should touch nameidata between unlazy_walk() failure and
|
|
|
|
* terminate_walk().
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
*/
|
2015-05-05 07:26:05 -06:00
|
|
|
static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
{
|
|
|
|
struct dentry *parent = nd->path.dentry;
|
|
|
|
|
|
|
|
BUG_ON(!(nd->flags & LOOKUP_RCU));
|
vfs: fix dentry RCU to refcounting possibly sleeping dput()
This is the fix that the last two commits indirectly led up to - making
sure that we don't call dput() in a bad context on the dentries we've
looked up in RCU mode after the sequence count validation fails.
This basically expands d_rcu_to_refcount() into the callers, and then
fixes the callers to delay the dput() in the failure case until _after_
we've dropped all locks and are no longer in an RCU-locked region.
The case of 'complete_walk()' was trivial, since its failure case did
the unlock_rcu_walk() directly after the call to d_rcu_to_refcount(),
and as such that is just a pure expansion of the function with a trivial
movement of the resulting dput() to after 'unlock_rcu_walk()'.
In contrast, the unlazy_walk() case was much more complicated, because
not only does convert two different dentries from RCU to be reference
counted, but it used to not call unlock_rcu_walk() at all, and instead
just returned an error and let the caller clean everything up in
"terminate_walk()".
Happily, one of the dentries in question (called "parent" inside
unlazy_walk()) is the dentry of "nd->path", which terminate_walk() wants
a refcount to anyway for the non-RCU case.
So what the new and improved unlazy_walk() does is to first turn that
dentry into a refcounted one, and once that is set up, the error cases
can continue to use the terminate_walk() helper for cleanup, but for the
non-RCU case. Which makes it possible to drop out of RCU mode if we
actually hit the sequence number failure case.
Acked-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-08 19:13:49 -06:00
|
|
|
|
|
|
|
nd->flags &= ~LOOKUP_RCU;
|
2015-05-09 10:55:43 -06:00
|
|
|
if (unlikely(!legitimize_links(nd)))
|
|
|
|
goto out2;
|
|
|
|
if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
|
|
|
|
goto out2;
|
|
|
|
if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
|
|
|
|
goto out1;
|
2013-09-29 20:06:07 -06:00
|
|
|
|
2013-09-02 12:38:06 -06:00
|
|
|
/*
|
|
|
|
* For a negative lookup, the lookup sequence point is the parents
|
|
|
|
* sequence point, and it only needs to revalidate the parent dentry.
|
|
|
|
*
|
|
|
|
* For a positive lookup, we need to move both the parent and the
|
|
|
|
* dentry from the RCU domain to be properly refcounted. And the
|
|
|
|
* sequence number in the dentry validates *both* dentry counters,
|
|
|
|
* since we checked the sequence number of the parent after we got
|
|
|
|
* the child sequence number. So we know the parent must still
|
|
|
|
* be valid if the child sequence number is still valid.
|
|
|
|
*/
|
2011-03-25 08:32:48 -06:00
|
|
|
if (!dentry) {
|
vfs: fix dentry RCU to refcounting possibly sleeping dput()
This is the fix that the last two commits indirectly led up to - making
sure that we don't call dput() in a bad context on the dentries we've
looked up in RCU mode after the sequence count validation fails.
This basically expands d_rcu_to_refcount() into the callers, and then
fixes the callers to delay the dput() in the failure case until _after_
we've dropped all locks and are no longer in an RCU-locked region.
The case of 'complete_walk()' was trivial, since its failure case did
the unlock_rcu_walk() directly after the call to d_rcu_to_refcount(),
and as such that is just a pure expansion of the function with a trivial
movement of the resulting dput() to after 'unlock_rcu_walk()'.
In contrast, the unlazy_walk() case was much more complicated, because
not only does convert two different dentries from RCU to be reference
counted, but it used to not call unlock_rcu_walk() at all, and instead
just returned an error and let the caller clean everything up in
"terminate_walk()".
Happily, one of the dentries in question (called "parent" inside
unlazy_walk()) is the dentry of "nd->path", which terminate_walk() wants
a refcount to anyway for the non-RCU case.
So what the new and improved unlazy_walk() does is to first turn that
dentry into a refcounted one, and once that is set up, the error cases
can continue to use the terminate_walk() helper for cleanup, but for the
non-RCU case. Which makes it possible to drop out of RCU mode if we
actually hit the sequence number failure case.
Acked-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-08 19:13:49 -06:00
|
|
|
if (read_seqcount_retry(&parent->d_seq, nd->seq))
|
|
|
|
goto out;
|
2011-03-25 08:32:48 -06:00
|
|
|
BUG_ON(nd->inode != parent->d_inode);
|
|
|
|
} else {
|
vfs: fix dentry RCU to refcounting possibly sleeping dput()
This is the fix that the last two commits indirectly led up to - making
sure that we don't call dput() in a bad context on the dentries we've
looked up in RCU mode after the sequence count validation fails.
This basically expands d_rcu_to_refcount() into the callers, and then
fixes the callers to delay the dput() in the failure case until _after_
we've dropped all locks and are no longer in an RCU-locked region.
The case of 'complete_walk()' was trivial, since its failure case did
the unlock_rcu_walk() directly after the call to d_rcu_to_refcount(),
and as such that is just a pure expansion of the function with a trivial
movement of the resulting dput() to after 'unlock_rcu_walk()'.
In contrast, the unlazy_walk() case was much more complicated, because
not only does convert two different dentries from RCU to be reference
counted, but it used to not call unlock_rcu_walk() at all, and instead
just returned an error and let the caller clean everything up in
"terminate_walk()".
Happily, one of the dentries in question (called "parent" inside
unlazy_walk()) is the dentry of "nd->path", which terminate_walk() wants
a refcount to anyway for the non-RCU case.
So what the new and improved unlazy_walk() does is to first turn that
dentry into a refcounted one, and once that is set up, the error cases
can continue to use the terminate_walk() helper for cleanup, but for the
non-RCU case. Which makes it possible to drop out of RCU mode if we
actually hit the sequence number failure case.
Acked-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-08 19:13:49 -06:00
|
|
|
if (!lockref_get_not_dead(&dentry->d_lockref))
|
|
|
|
goto out;
|
2015-05-05 07:26:05 -06:00
|
|
|
if (read_seqcount_retry(&dentry->d_seq, seq))
|
vfs: fix dentry RCU to refcounting possibly sleeping dput()
This is the fix that the last two commits indirectly led up to - making
sure that we don't call dput() in a bad context on the dentries we've
looked up in RCU mode after the sequence count validation fails.
This basically expands d_rcu_to_refcount() into the callers, and then
fixes the callers to delay the dput() in the failure case until _after_
we've dropped all locks and are no longer in an RCU-locked region.
The case of 'complete_walk()' was trivial, since its failure case did
the unlock_rcu_walk() directly after the call to d_rcu_to_refcount(),
and as such that is just a pure expansion of the function with a trivial
movement of the resulting dput() to after 'unlock_rcu_walk()'.
In contrast, the unlazy_walk() case was much more complicated, because
not only does convert two different dentries from RCU to be reference
counted, but it used to not call unlock_rcu_walk() at all, and instead
just returned an error and let the caller clean everything up in
"terminate_walk()".
Happily, one of the dentries in question (called "parent" inside
unlazy_walk()) is the dentry of "nd->path", which terminate_walk() wants
a refcount to anyway for the non-RCU case.
So what the new and improved unlazy_walk() does is to first turn that
dentry into a refcounted one, and once that is set up, the error cases
can continue to use the terminate_walk() helper for cleanup, but for the
non-RCU case. Which makes it possible to drop out of RCU mode if we
actually hit the sequence number failure case.
Acked-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-08 19:13:49 -06:00
|
|
|
goto drop_dentry;
|
2011-03-25 08:32:48 -06:00
|
|
|
}
|
vfs: fix dentry RCU to refcounting possibly sleeping dput()
This is the fix that the last two commits indirectly led up to - making
sure that we don't call dput() in a bad context on the dentries we've
looked up in RCU mode after the sequence count validation fails.
This basically expands d_rcu_to_refcount() into the callers, and then
fixes the callers to delay the dput() in the failure case until _after_
we've dropped all locks and are no longer in an RCU-locked region.
The case of 'complete_walk()' was trivial, since its failure case did
the unlock_rcu_walk() directly after the call to d_rcu_to_refcount(),
and as such that is just a pure expansion of the function with a trivial
movement of the resulting dput() to after 'unlock_rcu_walk()'.
In contrast, the unlazy_walk() case was much more complicated, because
not only does convert two different dentries from RCU to be reference
counted, but it used to not call unlock_rcu_walk() at all, and instead
just returned an error and let the caller clean everything up in
"terminate_walk()".
Happily, one of the dentries in question (called "parent" inside
unlazy_walk()) is the dentry of "nd->path", which terminate_walk() wants
a refcount to anyway for the non-RCU case.
So what the new and improved unlazy_walk() does is to first turn that
dentry into a refcounted one, and once that is set up, the error cases
can continue to use the terminate_walk() helper for cleanup, but for the
non-RCU case. Which makes it possible to drop out of RCU mode if we
actually hit the sequence number failure case.
Acked-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-08 19:13:49 -06:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Sequence counts matched. Now make sure that the root is
|
|
|
|
* still valid and get it if required.
|
|
|
|
*/
|
|
|
|
if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
|
2015-05-11 22:10:18 -06:00
|
|
|
if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
|
|
|
|
rcu_read_unlock();
|
|
|
|
dput(dentry);
|
|
|
|
return -ECHILD;
|
2015-05-09 10:55:43 -06:00
|
|
|
}
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
}
|
|
|
|
|
2013-11-08 10:45:01 -07:00
|
|
|
rcu_read_unlock();
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
return 0;
|
2011-03-25 08:32:48 -06:00
|
|
|
|
vfs: fix dentry RCU to refcounting possibly sleeping dput()
This is the fix that the last two commits indirectly led up to - making
sure that we don't call dput() in a bad context on the dentries we've
looked up in RCU mode after the sequence count validation fails.
This basically expands d_rcu_to_refcount() into the callers, and then
fixes the callers to delay the dput() in the failure case until _after_
we've dropped all locks and are no longer in an RCU-locked region.
The case of 'complete_walk()' was trivial, since its failure case did
the unlock_rcu_walk() directly after the call to d_rcu_to_refcount(),
and as such that is just a pure expansion of the function with a trivial
movement of the resulting dput() to after 'unlock_rcu_walk()'.
In contrast, the unlazy_walk() case was much more complicated, because
not only does convert two different dentries from RCU to be reference
counted, but it used to not call unlock_rcu_walk() at all, and instead
just returned an error and let the caller clean everything up in
"terminate_walk()".
Happily, one of the dentries in question (called "parent" inside
unlazy_walk()) is the dentry of "nd->path", which terminate_walk() wants
a refcount to anyway for the non-RCU case.
So what the new and improved unlazy_walk() does is to first turn that
dentry into a refcounted one, and once that is set up, the error cases
can continue to use the terminate_walk() helper for cleanup, but for the
non-RCU case. Which makes it possible to drop out of RCU mode if we
actually hit the sequence number failure case.
Acked-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-08 19:13:49 -06:00
|
|
|
drop_dentry:
|
2013-11-08 10:45:01 -07:00
|
|
|
rcu_read_unlock();
|
2013-09-02 12:38:06 -06:00
|
|
|
dput(dentry);
|
2013-09-10 13:17:49 -06:00
|
|
|
goto drop_root_mnt;
|
2015-05-09 10:55:43 -06:00
|
|
|
out2:
|
|
|
|
nd->path.mnt = NULL;
|
|
|
|
out1:
|
|
|
|
nd->path.dentry = NULL;
|
vfs: fix dentry RCU to refcounting possibly sleeping dput()
This is the fix that the last two commits indirectly led up to - making
sure that we don't call dput() in a bad context on the dentries we've
looked up in RCU mode after the sequence count validation fails.
This basically expands d_rcu_to_refcount() into the callers, and then
fixes the callers to delay the dput() in the failure case until _after_
we've dropped all locks and are no longer in an RCU-locked region.
The case of 'complete_walk()' was trivial, since its failure case did
the unlock_rcu_walk() directly after the call to d_rcu_to_refcount(),
and as such that is just a pure expansion of the function with a trivial
movement of the resulting dput() to after 'unlock_rcu_walk()'.
In contrast, the unlazy_walk() case was much more complicated, because
not only does convert two different dentries from RCU to be reference
counted, but it used to not call unlock_rcu_walk() at all, and instead
just returned an error and let the caller clean everything up in
"terminate_walk()".
Happily, one of the dentries in question (called "parent" inside
unlazy_walk()) is the dentry of "nd->path", which terminate_walk() wants
a refcount to anyway for the non-RCU case.
So what the new and improved unlazy_walk() does is to first turn that
dentry into a refcounted one, and once that is set up, the error cases
can continue to use the terminate_walk() helper for cleanup, but for the
non-RCU case. Which makes it possible to drop out of RCU mode if we
actually hit the sequence number failure case.
Acked-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-08 19:13:49 -06:00
|
|
|
out:
|
2013-11-08 10:45:01 -07:00
|
|
|
rcu_read_unlock();
|
2013-09-10 13:17:49 -06:00
|
|
|
drop_root_mnt:
|
|
|
|
if (!(nd->flags & LOOKUP_ROOT))
|
|
|
|
nd->root.mnt = NULL;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
return -ECHILD;
|
|
|
|
}
|
|
|
|
|
2015-05-09 10:55:43 -06:00
|
|
|
static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
|
|
|
|
{
|
|
|
|
if (unlikely(!legitimize_path(nd, link, seq))) {
|
|
|
|
drop_links(nd);
|
|
|
|
nd->depth = 0;
|
|
|
|
nd->flags &= ~LOOKUP_RCU;
|
|
|
|
nd->path.mnt = NULL;
|
|
|
|
nd->path.dentry = NULL;
|
|
|
|
if (!(nd->flags & LOOKUP_ROOT))
|
|
|
|
nd->root.mnt = NULL;
|
|
|
|
rcu_read_unlock();
|
|
|
|
} else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
path_put(link);
|
|
|
|
return -ECHILD;
|
|
|
|
}
|
|
|
|
|
2012-06-10 14:10:59 -06:00
|
|
|
static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
|
2011-01-06 23:49:57 -07:00
|
|
|
{
|
2012-06-10 14:10:59 -06:00
|
|
|
return dentry->d_op->d_revalidate(dentry, flags);
|
2011-01-06 23:49:57 -07:00
|
|
|
}
|
|
|
|
|
2011-03-25 09:00:12 -06:00
|
|
|
/**
|
|
|
|
* complete_walk - successful completion of path walk
|
|
|
|
* @nd: pointer nameidata
|
2009-12-07 10:01:50 -07:00
|
|
|
*
|
2011-03-25 09:00:12 -06:00
|
|
|
* If we had been in RCU mode, drop out of it and legitimize nd->path.
|
|
|
|
* Revalidate the final result, unless we'd already done that during
|
|
|
|
* the path walk or the filesystem doesn't ask for it. Return 0 on
|
|
|
|
* success, -error on failure. In case of failure caller does not
|
|
|
|
* need to drop nd->path.
|
2009-12-07 10:01:50 -07:00
|
|
|
*/
|
2011-03-25 09:00:12 -06:00
|
|
|
static int complete_walk(struct nameidata *nd)
|
2009-12-07 10:01:50 -07:00
|
|
|
{
|
2011-02-22 13:50:10 -07:00
|
|
|
struct dentry *dentry = nd->path.dentry;
|
2009-12-07 10:01:50 -07:00
|
|
|
int status;
|
|
|
|
|
2011-03-25 09:00:12 -06:00
|
|
|
if (nd->flags & LOOKUP_RCU) {
|
|
|
|
if (!(nd->flags & LOOKUP_ROOT))
|
|
|
|
nd->root.mnt = NULL;
|
2015-05-05 07:26:05 -06:00
|
|
|
if (unlikely(unlazy_walk(nd, NULL, 0)))
|
2011-03-25 09:00:12 -06:00
|
|
|
return -ECHILD;
|
|
|
|
}
|
|
|
|
|
2011-02-22 13:50:10 -07:00
|
|
|
if (likely(!(nd->flags & LOOKUP_JUMPED)))
|
|
|
|
return 0;
|
|
|
|
|
2013-02-20 09:19:05 -07:00
|
|
|
if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
|
2009-12-07 10:01:50 -07:00
|
|
|
return 0;
|
|
|
|
|
2013-02-20 09:19:05 -07:00
|
|
|
status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
|
2009-12-07 10:01:50 -07:00
|
|
|
if (status > 0)
|
|
|
|
return 0;
|
|
|
|
|
2011-02-22 13:50:10 -07:00
|
|
|
if (!status)
|
2009-12-07 10:01:50 -07:00
|
|
|
status = -ESTALE;
|
2011-02-22 13:50:10 -07:00
|
|
|
|
2009-12-07 10:01:50 -07:00
|
|
|
return status;
|
|
|
|
}
|
|
|
|
|
2015-05-12 14:32:34 -06:00
|
|
|
static void set_root(struct nameidata *nd)
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
{
|
2014-09-13 19:55:46 -06:00
|
|
|
struct fs_struct *fs = current->fs;
|
2011-01-06 23:49:53 -07:00
|
|
|
|
2015-12-05 18:07:21 -07:00
|
|
|
if (nd->flags & LOOKUP_RCU) {
|
|
|
|
unsigned seq;
|
|
|
|
|
|
|
|
do {
|
|
|
|
seq = read_seqcount_begin(&fs->seq);
|
|
|
|
nd->root = fs->root;
|
|
|
|
nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
|
|
|
|
} while (read_seqcount_retry(&fs->seq, seq));
|
|
|
|
} else {
|
|
|
|
get_fs_root(fs, &nd->root);
|
|
|
|
}
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
}
|
|
|
|
|
2008-02-14 20:34:35 -07:00
|
|
|
static void path_put_conditional(struct path *path, struct nameidata *nd)
|
2006-03-27 02:14:53 -07:00
|
|
|
{
|
|
|
|
dput(path->dentry);
|
2008-02-14 20:34:32 -07:00
|
|
|
if (path->mnt != nd->path.mnt)
|
2006-03-27 02:14:53 -07:00
|
|
|
mntput(path->mnt);
|
|
|
|
}
|
|
|
|
|
2011-01-14 01:42:43 -07:00
|
|
|
static inline void path_to_nameidata(const struct path *path,
|
|
|
|
struct nameidata *nd)
|
2006-03-27 02:14:53 -07:00
|
|
|
{
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
if (!(nd->flags & LOOKUP_RCU)) {
|
|
|
|
dput(nd->path.dentry);
|
|
|
|
if (nd->path.mnt != path->mnt)
|
|
|
|
mntput(nd->path.mnt);
|
2010-04-02 03:37:13 -06:00
|
|
|
}
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
nd->path.mnt = path->mnt;
|
2008-02-14 20:34:32 -07:00
|
|
|
nd->path.dentry = path->dentry;
|
2006-03-27 02:14:53 -07:00
|
|
|
}
|
|
|
|
|
2015-12-05 18:51:58 -07:00
|
|
|
static int nd_jump_root(struct nameidata *nd)
|
|
|
|
{
|
|
|
|
if (nd->flags & LOOKUP_RCU) {
|
|
|
|
struct dentry *d;
|
|
|
|
nd->path = nd->root;
|
|
|
|
d = nd->path.dentry;
|
|
|
|
nd->inode = d->d_inode;
|
|
|
|
nd->seq = nd->root_seq;
|
|
|
|
if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
|
|
|
|
return -ECHILD;
|
|
|
|
} else {
|
|
|
|
path_put(&nd->path);
|
|
|
|
nd->path = nd->root;
|
|
|
|
path_get(&nd->path);
|
|
|
|
nd->inode = nd->path.dentry->d_inode;
|
|
|
|
}
|
|
|
|
nd->flags |= LOOKUP_JUMPED;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2012-06-18 08:47:04 -06:00
|
|
|
/*
|
2015-11-17 08:20:54 -07:00
|
|
|
* Helper to directly jump to a known parsed path from ->get_link,
|
2012-06-18 08:47:04 -06:00
|
|
|
* caller must have taken a reference to path beforehand.
|
|
|
|
*/
|
2015-05-02 11:37:52 -06:00
|
|
|
void nd_jump_link(struct path *path)
|
2012-06-18 08:47:04 -06:00
|
|
|
{
|
2015-05-02 11:37:52 -06:00
|
|
|
struct nameidata *nd = current->nameidata;
|
2012-06-18 08:47:04 -06:00
|
|
|
path_put(&nd->path);
|
|
|
|
|
|
|
|
nd->path = *path;
|
|
|
|
nd->inode = nd->path.dentry->d_inode;
|
|
|
|
nd->flags |= LOOKUP_JUMPED;
|
|
|
|
}
|
|
|
|
|
2015-05-02 18:19:23 -06:00
|
|
|
static inline void put_link(struct nameidata *nd)
|
2011-03-14 20:20:34 -06:00
|
|
|
{
|
2015-05-03 19:06:24 -06:00
|
|
|
struct saved *last = nd->stack + --nd->depth;
|
2015-12-29 13:58:39 -07:00
|
|
|
do_delayed_call(&last->done);
|
2015-05-07 18:32:22 -06:00
|
|
|
if (!(nd->flags & LOOKUP_RCU))
|
|
|
|
path_put(&last->link);
|
2011-03-14 20:20:34 -06:00
|
|
|
}
|
|
|
|
|
VFS: don't do protected {sym,hard}links by default
In commit 800179c9b8a1 ("This adds symlink and hardlink restrictions to
the Linux VFS"), the new link protections were enabled by default, in
the hope that no actual application would care, despite it being
technically against legacy UNIX (and documented POSIX) behavior.
However, it does turn out to break some applications. It's rare, and
it's unfortunate, but it's unacceptable to break existing systems, so
we'll have to default to legacy behavior.
In particular, it has broken the way AFD distributes files, see
http://www.dwd.de/AFD/
along with some legacy scripts.
Distributions can end up setting this at initrd time or in system
scripts: if you have security problems due to link attacks during your
early boot sequence, you have bigger problems than some kernel sysctl
setting. Do:
echo 1 > /proc/sys/fs/protected_symlinks
echo 1 > /proc/sys/fs/protected_hardlinks
to re-enable the link protections.
Alternatively, we may at some point introduce a kernel config option
that sets these kinds of "more secure but not traditional" behavioural
options automatically.
Reported-by: Nick Bowler <nbowler@elliptictech.com>
Reported-by: Holger Kiehl <Holger.Kiehl@dwd.de>
Cc: Kees Cook <keescook@chromium.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org # v3.6
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-26 11:05:07 -06:00
|
|
|
int sysctl_protected_symlinks __read_mostly = 0;
|
|
|
|
int sysctl_protected_hardlinks __read_mostly = 0;
|
fs: add link restrictions
This adds symlink and hardlink restrictions to the Linux VFS.
Symlinks:
A long-standing class of security issues is the symlink-based
time-of-check-time-of-use race, most commonly seen in world-writable
directories like /tmp. The common method of exploitation of this flaw
is to cross privilege boundaries when following a given symlink (i.e. a
root process follows a symlink belonging to another user). For a likely
incomplete list of hundreds of examples across the years, please see:
http://cve.mitre.org/cgi-bin/cvekey.cgi?keyword=/tmp
The solution is to permit symlinks to only be followed when outside
a sticky world-writable directory, or when the uid of the symlink and
follower match, or when the directory owner matches the symlink's owner.
Some pointers to the history of earlier discussion that I could find:
1996 Aug, Zygo Blaxell
http://marc.info/?l=bugtraq&m=87602167419830&w=2
1996 Oct, Andrew Tridgell
http://lkml.indiana.edu/hypermail/linux/kernel/9610.2/0086.html
1997 Dec, Albert D Cahalan
http://lkml.org/lkml/1997/12/16/4
2005 Feb, Lorenzo Hernández GarcÃa-Hierro
http://lkml.indiana.edu/hypermail/linux/kernel/0502.0/1896.html
2010 May, Kees Cook
https://lkml.org/lkml/2010/5/30/144
Past objections and rebuttals could be summarized as:
- Violates POSIX.
- POSIX didn't consider this situation and it's not useful to follow
a broken specification at the cost of security.
- Might break unknown applications that use this feature.
- Applications that break because of the change are easy to spot and
fix. Applications that are vulnerable to symlink ToCToU by not having
the change aren't. Additionally, no applications have yet been found
that rely on this behavior.
- Applications should just use mkstemp() or O_CREATE|O_EXCL.
- True, but applications are not perfect, and new software is written
all the time that makes these mistakes; blocking this flaw at the
kernel is a single solution to the entire class of vulnerability.
- This should live in the core VFS.
- This should live in an LSM. (https://lkml.org/lkml/2010/5/31/135)
- This should live in an LSM.
- This should live in the core VFS. (https://lkml.org/lkml/2010/8/2/188)
Hardlinks:
On systems that have user-writable directories on the same partition
as system files, a long-standing class of security issues is the
hardlink-based time-of-check-time-of-use race, most commonly seen in
world-writable directories like /tmp. The common method of exploitation
of this flaw is to cross privilege boundaries when following a given
hardlink (i.e. a root process follows a hardlink created by another
user). Additionally, an issue exists where users can "pin" a potentially
vulnerable setuid/setgid file so that an administrator will not actually
upgrade a system fully.
The solution is to permit hardlinks to only be created when the user is
already the existing file's owner, or if they already have read/write
access to the existing file.
Many Linux users are surprised when they learn they can link to files
they have no access to, so this change appears to follow the doctrine
of "least surprise". Additionally, this change does not violate POSIX,
which states "the implementation may require that the calling process
has permission to access the existing file"[1].
This change is known to break some implementations of the "at" daemon,
though the version used by Fedora and Ubuntu has been fixed[2] for
a while. Otherwise, the change has been undisruptive while in use in
Ubuntu for the last 1.5 years.
[1] http://pubs.opengroup.org/onlinepubs/9699919799/functions/linkat.html
[2] http://anonscm.debian.org/gitweb/?p=collab-maint/at.git;a=commitdiff;h=f4114656c3a6c6f6070e315ffdf940a49eda3279
This patch is based on the patches in Openwall and grsecurity, along with
suggestions from Al Viro. I have added a sysctl to enable the protected
behavior, and documentation.
Signed-off-by: Kees Cook <keescook@chromium.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2012-07-25 18:29:07 -06:00
|
|
|
|
|
|
|
/**
|
|
|
|
* may_follow_link - Check symlink following for unsafe situations
|
2012-08-18 18:39:25 -06:00
|
|
|
* @nd: nameidata pathwalk data
|
fs: add link restrictions
This adds symlink and hardlink restrictions to the Linux VFS.
Symlinks:
A long-standing class of security issues is the symlink-based
time-of-check-time-of-use race, most commonly seen in world-writable
directories like /tmp. The common method of exploitation of this flaw
is to cross privilege boundaries when following a given symlink (i.e. a
root process follows a symlink belonging to another user). For a likely
incomplete list of hundreds of examples across the years, please see:
http://cve.mitre.org/cgi-bin/cvekey.cgi?keyword=/tmp
The solution is to permit symlinks to only be followed when outside
a sticky world-writable directory, or when the uid of the symlink and
follower match, or when the directory owner matches the symlink's owner.
Some pointers to the history of earlier discussion that I could find:
1996 Aug, Zygo Blaxell
http://marc.info/?l=bugtraq&m=87602167419830&w=2
1996 Oct, Andrew Tridgell
http://lkml.indiana.edu/hypermail/linux/kernel/9610.2/0086.html
1997 Dec, Albert D Cahalan
http://lkml.org/lkml/1997/12/16/4
2005 Feb, Lorenzo Hernández GarcÃa-Hierro
http://lkml.indiana.edu/hypermail/linux/kernel/0502.0/1896.html
2010 May, Kees Cook
https://lkml.org/lkml/2010/5/30/144
Past objections and rebuttals could be summarized as:
- Violates POSIX.
- POSIX didn't consider this situation and it's not useful to follow
a broken specification at the cost of security.
- Might break unknown applications that use this feature.
- Applications that break because of the change are easy to spot and
fix. Applications that are vulnerable to symlink ToCToU by not having
the change aren't. Additionally, no applications have yet been found
that rely on this behavior.
- Applications should just use mkstemp() or O_CREATE|O_EXCL.
- True, but applications are not perfect, and new software is written
all the time that makes these mistakes; blocking this flaw at the
kernel is a single solution to the entire class of vulnerability.
- This should live in the core VFS.
- This should live in an LSM. (https://lkml.org/lkml/2010/5/31/135)
- This should live in an LSM.
- This should live in the core VFS. (https://lkml.org/lkml/2010/8/2/188)
Hardlinks:
On systems that have user-writable directories on the same partition
as system files, a long-standing class of security issues is the
hardlink-based time-of-check-time-of-use race, most commonly seen in
world-writable directories like /tmp. The common method of exploitation
of this flaw is to cross privilege boundaries when following a given
hardlink (i.e. a root process follows a hardlink created by another
user). Additionally, an issue exists where users can "pin" a potentially
vulnerable setuid/setgid file so that an administrator will not actually
upgrade a system fully.
The solution is to permit hardlinks to only be created when the user is
already the existing file's owner, or if they already have read/write
access to the existing file.
Many Linux users are surprised when they learn they can link to files
they have no access to, so this change appears to follow the doctrine
of "least surprise". Additionally, this change does not violate POSIX,
which states "the implementation may require that the calling process
has permission to access the existing file"[1].
This change is known to break some implementations of the "at" daemon,
though the version used by Fedora and Ubuntu has been fixed[2] for
a while. Otherwise, the change has been undisruptive while in use in
Ubuntu for the last 1.5 years.
[1] http://pubs.opengroup.org/onlinepubs/9699919799/functions/linkat.html
[2] http://anonscm.debian.org/gitweb/?p=collab-maint/at.git;a=commitdiff;h=f4114656c3a6c6f6070e315ffdf940a49eda3279
This patch is based on the patches in Openwall and grsecurity, along with
suggestions from Al Viro. I have added a sysctl to enable the protected
behavior, and documentation.
Signed-off-by: Kees Cook <keescook@chromium.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2012-07-25 18:29:07 -06:00
|
|
|
*
|
|
|
|
* In the case of the sysctl_protected_symlinks sysctl being enabled,
|
|
|
|
* CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
|
|
|
|
* in a sticky world-writable directory. This is to protect privileged
|
|
|
|
* processes from failing races against path names that may change out
|
|
|
|
* from under them by way of other users creating malicious symlinks.
|
|
|
|
* It will permit symlinks to be followed only when outside a sticky
|
|
|
|
* world-writable directory, or when the uid of the symlink and follower
|
|
|
|
* match, or when the directory owner matches the symlink's owner.
|
|
|
|
*
|
|
|
|
* Returns 0 if following the symlink is allowed, -ve on error.
|
|
|
|
*/
|
2015-05-06 13:58:18 -06:00
|
|
|
static inline int may_follow_link(struct nameidata *nd)
|
fs: add link restrictions
This adds symlink and hardlink restrictions to the Linux VFS.
Symlinks:
A long-standing class of security issues is the symlink-based
time-of-check-time-of-use race, most commonly seen in world-writable
directories like /tmp. The common method of exploitation of this flaw
is to cross privilege boundaries when following a given symlink (i.e. a
root process follows a symlink belonging to another user). For a likely
incomplete list of hundreds of examples across the years, please see:
http://cve.mitre.org/cgi-bin/cvekey.cgi?keyword=/tmp
The solution is to permit symlinks to only be followed when outside
a sticky world-writable directory, or when the uid of the symlink and
follower match, or when the directory owner matches the symlink's owner.
Some pointers to the history of earlier discussion that I could find:
1996 Aug, Zygo Blaxell
http://marc.info/?l=bugtraq&m=87602167419830&w=2
1996 Oct, Andrew Tridgell
http://lkml.indiana.edu/hypermail/linux/kernel/9610.2/0086.html
1997 Dec, Albert D Cahalan
http://lkml.org/lkml/1997/12/16/4
2005 Feb, Lorenzo Hernández GarcÃa-Hierro
http://lkml.indiana.edu/hypermail/linux/kernel/0502.0/1896.html
2010 May, Kees Cook
https://lkml.org/lkml/2010/5/30/144
Past objections and rebuttals could be summarized as:
- Violates POSIX.
- POSIX didn't consider this situation and it's not useful to follow
a broken specification at the cost of security.
- Might break unknown applications that use this feature.
- Applications that break because of the change are easy to spot and
fix. Applications that are vulnerable to symlink ToCToU by not having
the change aren't. Additionally, no applications have yet been found
that rely on this behavior.
- Applications should just use mkstemp() or O_CREATE|O_EXCL.
- True, but applications are not perfect, and new software is written
all the time that makes these mistakes; blocking this flaw at the
kernel is a single solution to the entire class of vulnerability.
- This should live in the core VFS.
- This should live in an LSM. (https://lkml.org/lkml/2010/5/31/135)
- This should live in an LSM.
- This should live in the core VFS. (https://lkml.org/lkml/2010/8/2/188)
Hardlinks:
On systems that have user-writable directories on the same partition
as system files, a long-standing class of security issues is the
hardlink-based time-of-check-time-of-use race, most commonly seen in
world-writable directories like /tmp. The common method of exploitation
of this flaw is to cross privilege boundaries when following a given
hardlink (i.e. a root process follows a hardlink created by another
user). Additionally, an issue exists where users can "pin" a potentially
vulnerable setuid/setgid file so that an administrator will not actually
upgrade a system fully.
The solution is to permit hardlinks to only be created when the user is
already the existing file's owner, or if they already have read/write
access to the existing file.
Many Linux users are surprised when they learn they can link to files
they have no access to, so this change appears to follow the doctrine
of "least surprise". Additionally, this change does not violate POSIX,
which states "the implementation may require that the calling process
has permission to access the existing file"[1].
This change is known to break some implementations of the "at" daemon,
though the version used by Fedora and Ubuntu has been fixed[2] for
a while. Otherwise, the change has been undisruptive while in use in
Ubuntu for the last 1.5 years.
[1] http://pubs.opengroup.org/onlinepubs/9699919799/functions/linkat.html
[2] http://anonscm.debian.org/gitweb/?p=collab-maint/at.git;a=commitdiff;h=f4114656c3a6c6f6070e315ffdf940a49eda3279
This patch is based on the patches in Openwall and grsecurity, along with
suggestions from Al Viro. I have added a sysctl to enable the protected
behavior, and documentation.
Signed-off-by: Kees Cook <keescook@chromium.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2012-07-25 18:29:07 -06:00
|
|
|
{
|
|
|
|
const struct inode *inode;
|
|
|
|
const struct inode *parent;
|
|
|
|
|
|
|
|
if (!sysctl_protected_symlinks)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* Allowed if owner and follower match. */
|
2015-12-29 13:58:39 -07:00
|
|
|
inode = nd->link_inode;
|
2012-08-03 10:38:08 -06:00
|
|
|
if (uid_eq(current_cred()->fsuid, inode->i_uid))
|
fs: add link restrictions
This adds symlink and hardlink restrictions to the Linux VFS.
Symlinks:
A long-standing class of security issues is the symlink-based
time-of-check-time-of-use race, most commonly seen in world-writable
directories like /tmp. The common method of exploitation of this flaw
is to cross privilege boundaries when following a given symlink (i.e. a
root process follows a symlink belonging to another user). For a likely
incomplete list of hundreds of examples across the years, please see:
http://cve.mitre.org/cgi-bin/cvekey.cgi?keyword=/tmp
The solution is to permit symlinks to only be followed when outside
a sticky world-writable directory, or when the uid of the symlink and
follower match, or when the directory owner matches the symlink's owner.
Some pointers to the history of earlier discussion that I could find:
1996 Aug, Zygo Blaxell
http://marc.info/?l=bugtraq&m=87602167419830&w=2
1996 Oct, Andrew Tridgell
http://lkml.indiana.edu/hypermail/linux/kernel/9610.2/0086.html
1997 Dec, Albert D Cahalan
http://lkml.org/lkml/1997/12/16/4
2005 Feb, Lorenzo Hernández GarcÃa-Hierro
http://lkml.indiana.edu/hypermail/linux/kernel/0502.0/1896.html
2010 May, Kees Cook
https://lkml.org/lkml/2010/5/30/144
Past objections and rebuttals could be summarized as:
- Violates POSIX.
- POSIX didn't consider this situation and it's not useful to follow
a broken specification at the cost of security.
- Might break unknown applications that use this feature.
- Applications that break because of the change are easy to spot and
fix. Applications that are vulnerable to symlink ToCToU by not having
the change aren't. Additionally, no applications have yet been found
that rely on this behavior.
- Applications should just use mkstemp() or O_CREATE|O_EXCL.
- True, but applications are not perfect, and new software is written
all the time that makes these mistakes; blocking this flaw at the
kernel is a single solution to the entire class of vulnerability.
- This should live in the core VFS.
- This should live in an LSM. (https://lkml.org/lkml/2010/5/31/135)
- This should live in an LSM.
- This should live in the core VFS. (https://lkml.org/lkml/2010/8/2/188)
Hardlinks:
On systems that have user-writable directories on the same partition
as system files, a long-standing class of security issues is the
hardlink-based time-of-check-time-of-use race, most commonly seen in
world-writable directories like /tmp. The common method of exploitation
of this flaw is to cross privilege boundaries when following a given
hardlink (i.e. a root process follows a hardlink created by another
user). Additionally, an issue exists where users can "pin" a potentially
vulnerable setuid/setgid file so that an administrator will not actually
upgrade a system fully.
The solution is to permit hardlinks to only be created when the user is
already the existing file's owner, or if they already have read/write
access to the existing file.
Many Linux users are surprised when they learn they can link to files
they have no access to, so this change appears to follow the doctrine
of "least surprise". Additionally, this change does not violate POSIX,
which states "the implementation may require that the calling process
has permission to access the existing file"[1].
This change is known to break some implementations of the "at" daemon,
though the version used by Fedora and Ubuntu has been fixed[2] for
a while. Otherwise, the change has been undisruptive while in use in
Ubuntu for the last 1.5 years.
[1] http://pubs.opengroup.org/onlinepubs/9699919799/functions/linkat.html
[2] http://anonscm.debian.org/gitweb/?p=collab-maint/at.git;a=commitdiff;h=f4114656c3a6c6f6070e315ffdf940a49eda3279
This patch is based on the patches in Openwall and grsecurity, along with
suggestions from Al Viro. I have added a sysctl to enable the protected
behavior, and documentation.
Signed-off-by: Kees Cook <keescook@chromium.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2012-07-25 18:29:07 -06:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* Allowed if parent directory not sticky and world-writable. */
|
2015-08-04 21:23:50 -06:00
|
|
|
parent = nd->inode;
|
fs: add link restrictions
This adds symlink and hardlink restrictions to the Linux VFS.
Symlinks:
A long-standing class of security issues is the symlink-based
time-of-check-time-of-use race, most commonly seen in world-writable
directories like /tmp. The common method of exploitation of this flaw
is to cross privilege boundaries when following a given symlink (i.e. a
root process follows a symlink belonging to another user). For a likely
incomplete list of hundreds of examples across the years, please see:
http://cve.mitre.org/cgi-bin/cvekey.cgi?keyword=/tmp
The solution is to permit symlinks to only be followed when outside
a sticky world-writable directory, or when the uid of the symlink and
follower match, or when the directory owner matches the symlink's owner.
Some pointers to the history of earlier discussion that I could find:
1996 Aug, Zygo Blaxell
http://marc.info/?l=bugtraq&m=87602167419830&w=2
1996 Oct, Andrew Tridgell
http://lkml.indiana.edu/hypermail/linux/kernel/9610.2/0086.html
1997 Dec, Albert D Cahalan
http://lkml.org/lkml/1997/12/16/4
2005 Feb, Lorenzo Hernández GarcÃa-Hierro
http://lkml.indiana.edu/hypermail/linux/kernel/0502.0/1896.html
2010 May, Kees Cook
https://lkml.org/lkml/2010/5/30/144
Past objections and rebuttals could be summarized as:
- Violates POSIX.
- POSIX didn't consider this situation and it's not useful to follow
a broken specification at the cost of security.
- Might break unknown applications that use this feature.
- Applications that break because of the change are easy to spot and
fix. Applications that are vulnerable to symlink ToCToU by not having
the change aren't. Additionally, no applications have yet been found
that rely on this behavior.
- Applications should just use mkstemp() or O_CREATE|O_EXCL.
- True, but applications are not perfect, and new software is written
all the time that makes these mistakes; blocking this flaw at the
kernel is a single solution to the entire class of vulnerability.
- This should live in the core VFS.
- This should live in an LSM. (https://lkml.org/lkml/2010/5/31/135)
- This should live in an LSM.
- This should live in the core VFS. (https://lkml.org/lkml/2010/8/2/188)
Hardlinks:
On systems that have user-writable directories on the same partition
as system files, a long-standing class of security issues is the
hardlink-based time-of-check-time-of-use race, most commonly seen in
world-writable directories like /tmp. The common method of exploitation
of this flaw is to cross privilege boundaries when following a given
hardlink (i.e. a root process follows a hardlink created by another
user). Additionally, an issue exists where users can "pin" a potentially
vulnerable setuid/setgid file so that an administrator will not actually
upgrade a system fully.
The solution is to permit hardlinks to only be created when the user is
already the existing file's owner, or if they already have read/write
access to the existing file.
Many Linux users are surprised when they learn they can link to files
they have no access to, so this change appears to follow the doctrine
of "least surprise". Additionally, this change does not violate POSIX,
which states "the implementation may require that the calling process
has permission to access the existing file"[1].
This change is known to break some implementations of the "at" daemon,
though the version used by Fedora and Ubuntu has been fixed[2] for
a while. Otherwise, the change has been undisruptive while in use in
Ubuntu for the last 1.5 years.
[1] http://pubs.opengroup.org/onlinepubs/9699919799/functions/linkat.html
[2] http://anonscm.debian.org/gitweb/?p=collab-maint/at.git;a=commitdiff;h=f4114656c3a6c6f6070e315ffdf940a49eda3279
This patch is based on the patches in Openwall and grsecurity, along with
suggestions from Al Viro. I have added a sysctl to enable the protected
behavior, and documentation.
Signed-off-by: Kees Cook <keescook@chromium.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2012-07-25 18:29:07 -06:00
|
|
|
if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* Allowed if parent directory and link owner match. */
|
2012-08-03 10:38:08 -06:00
|
|
|
if (uid_eq(parent->i_uid, inode->i_uid))
|
fs: add link restrictions
This adds symlink and hardlink restrictions to the Linux VFS.
Symlinks:
A long-standing class of security issues is the symlink-based
time-of-check-time-of-use race, most commonly seen in world-writable
directories like /tmp. The common method of exploitation of this flaw
is to cross privilege boundaries when following a given symlink (i.e. a
root process follows a symlink belonging to another user). For a likely
incomplete list of hundreds of examples across the years, please see:
http://cve.mitre.org/cgi-bin/cvekey.cgi?keyword=/tmp
The solution is to permit symlinks to only be followed when outside
a sticky world-writable directory, or when the uid of the symlink and
follower match, or when the directory owner matches the symlink's owner.
Some pointers to the history of earlier discussion that I could find:
1996 Aug, Zygo Blaxell
http://marc.info/?l=bugtraq&m=87602167419830&w=2
1996 Oct, Andrew Tridgell
http://lkml.indiana.edu/hypermail/linux/kernel/9610.2/0086.html
1997 Dec, Albert D Cahalan
http://lkml.org/lkml/1997/12/16/4
2005 Feb, Lorenzo Hernández GarcÃa-Hierro
http://lkml.indiana.edu/hypermail/linux/kernel/0502.0/1896.html
2010 May, Kees Cook
https://lkml.org/lkml/2010/5/30/144
Past objections and rebuttals could be summarized as:
- Violates POSIX.
- POSIX didn't consider this situation and it's not useful to follow
a broken specification at the cost of security.
- Might break unknown applications that use this feature.
- Applications that break because of the change are easy to spot and
fix. Applications that are vulnerable to symlink ToCToU by not having
the change aren't. Additionally, no applications have yet been found
that rely on this behavior.
- Applications should just use mkstemp() or O_CREATE|O_EXCL.
- True, but applications are not perfect, and new software is written
all the time that makes these mistakes; blocking this flaw at the
kernel is a single solution to the entire class of vulnerability.
- This should live in the core VFS.
- This should live in an LSM. (https://lkml.org/lkml/2010/5/31/135)
- This should live in an LSM.
- This should live in the core VFS. (https://lkml.org/lkml/2010/8/2/188)
Hardlinks:
On systems that have user-writable directories on the same partition
as system files, a long-standing class of security issues is the
hardlink-based time-of-check-time-of-use race, most commonly seen in
world-writable directories like /tmp. The common method of exploitation
of this flaw is to cross privilege boundaries when following a given
hardlink (i.e. a root process follows a hardlink created by another
user). Additionally, an issue exists where users can "pin" a potentially
vulnerable setuid/setgid file so that an administrator will not actually
upgrade a system fully.
The solution is to permit hardlinks to only be created when the user is
already the existing file's owner, or if they already have read/write
access to the existing file.
Many Linux users are surprised when they learn they can link to files
they have no access to, so this change appears to follow the doctrine
of "least surprise". Additionally, this change does not violate POSIX,
which states "the implementation may require that the calling process
has permission to access the existing file"[1].
This change is known to break some implementations of the "at" daemon,
though the version used by Fedora and Ubuntu has been fixed[2] for
a while. Otherwise, the change has been undisruptive while in use in
Ubuntu for the last 1.5 years.
[1] http://pubs.opengroup.org/onlinepubs/9699919799/functions/linkat.html
[2] http://anonscm.debian.org/gitweb/?p=collab-maint/at.git;a=commitdiff;h=f4114656c3a6c6f6070e315ffdf940a49eda3279
This patch is based on the patches in Openwall and grsecurity, along with
suggestions from Al Viro. I have added a sysctl to enable the protected
behavior, and documentation.
Signed-off-by: Kees Cook <keescook@chromium.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2012-07-25 18:29:07 -06:00
|
|
|
return 0;
|
|
|
|
|
2015-05-07 18:37:40 -06:00
|
|
|
if (nd->flags & LOOKUP_RCU)
|
|
|
|
return -ECHILD;
|
|
|
|
|
2015-05-06 14:01:56 -06:00
|
|
|
audit_log_link_denied("follow_link", &nd->stack[0].link);
|
fs: add link restrictions
This adds symlink and hardlink restrictions to the Linux VFS.
Symlinks:
A long-standing class of security issues is the symlink-based
time-of-check-time-of-use race, most commonly seen in world-writable
directories like /tmp. The common method of exploitation of this flaw
is to cross privilege boundaries when following a given symlink (i.e. a
root process follows a symlink belonging to another user). For a likely
incomplete list of hundreds of examples across the years, please see:
http://cve.mitre.org/cgi-bin/cvekey.cgi?keyword=/tmp
The solution is to permit symlinks to only be followed when outside
a sticky world-writable directory, or when the uid of the symlink and
follower match, or when the directory owner matches the symlink's owner.
Some pointers to the history of earlier discussion that I could find:
1996 Aug, Zygo Blaxell
http://marc.info/?l=bugtraq&m=87602167419830&w=2
1996 Oct, Andrew Tridgell
http://lkml.indiana.edu/hypermail/linux/kernel/9610.2/0086.html
1997 Dec, Albert D Cahalan
http://lkml.org/lkml/1997/12/16/4
2005 Feb, Lorenzo Hernández GarcÃa-Hierro
http://lkml.indiana.edu/hypermail/linux/kernel/0502.0/1896.html
2010 May, Kees Cook
https://lkml.org/lkml/2010/5/30/144
Past objections and rebuttals could be summarized as:
- Violates POSIX.
- POSIX didn't consider this situation and it's not useful to follow
a broken specification at the cost of security.
- Might break unknown applications that use this feature.
- Applications that break because of the change are easy to spot and
fix. Applications that are vulnerable to symlink ToCToU by not having
the change aren't. Additionally, no applications have yet been found
that rely on this behavior.
- Applications should just use mkstemp() or O_CREATE|O_EXCL.
- True, but applications are not perfect, and new software is written
all the time that makes these mistakes; blocking this flaw at the
kernel is a single solution to the entire class of vulnerability.
- This should live in the core VFS.
- This should live in an LSM. (https://lkml.org/lkml/2010/5/31/135)
- This should live in an LSM.
- This should live in the core VFS. (https://lkml.org/lkml/2010/8/2/188)
Hardlinks:
On systems that have user-writable directories on the same partition
as system files, a long-standing class of security issues is the
hardlink-based time-of-check-time-of-use race, most commonly seen in
world-writable directories like /tmp. The common method of exploitation
of this flaw is to cross privilege boundaries when following a given
hardlink (i.e. a root process follows a hardlink created by another
user). Additionally, an issue exists where users can "pin" a potentially
vulnerable setuid/setgid file so that an administrator will not actually
upgrade a system fully.
The solution is to permit hardlinks to only be created when the user is
already the existing file's owner, or if they already have read/write
access to the existing file.
Many Linux users are surprised when they learn they can link to files
they have no access to, so this change appears to follow the doctrine
of "least surprise". Additionally, this change does not violate POSIX,
which states "the implementation may require that the calling process
has permission to access the existing file"[1].
This change is known to break some implementations of the "at" daemon,
though the version used by Fedora and Ubuntu has been fixed[2] for
a while. Otherwise, the change has been undisruptive while in use in
Ubuntu for the last 1.5 years.
[1] http://pubs.opengroup.org/onlinepubs/9699919799/functions/linkat.html
[2] http://anonscm.debian.org/gitweb/?p=collab-maint/at.git;a=commitdiff;h=f4114656c3a6c6f6070e315ffdf940a49eda3279
This patch is based on the patches in Openwall and grsecurity, along with
suggestions from Al Viro. I have added a sysctl to enable the protected
behavior, and documentation.
Signed-off-by: Kees Cook <keescook@chromium.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2012-07-25 18:29:07 -06:00
|
|
|
return -EACCES;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* safe_hardlink_source - Check for safe hardlink conditions
|
|
|
|
* @inode: the source inode to hardlink from
|
|
|
|
*
|
|
|
|
* Return false if at least one of the following conditions:
|
|
|
|
* - inode is not a regular file
|
|
|
|
* - inode is setuid
|
|
|
|
* - inode is setgid and group-exec
|
|
|
|
* - access failure for read and write
|
|
|
|
*
|
|
|
|
* Otherwise returns true.
|
|
|
|
*/
|
|
|
|
static bool safe_hardlink_source(struct inode *inode)
|
|
|
|
{
|
|
|
|
umode_t mode = inode->i_mode;
|
|
|
|
|
|
|
|
/* Special files should not get pinned to the filesystem. */
|
|
|
|
if (!S_ISREG(mode))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
/* Setuid files should not get pinned to the filesystem. */
|
|
|
|
if (mode & S_ISUID)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
/* Executable setgid files should not get pinned to the filesystem. */
|
|
|
|
if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
/* Hardlinking to unreadable or unwritable sources is dangerous. */
|
|
|
|
if (inode_permission(inode, MAY_READ | MAY_WRITE))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* may_linkat - Check permissions for creating a hardlink
|
|
|
|
* @link: the source to hardlink from
|
|
|
|
*
|
|
|
|
* Block hardlink when all of:
|
|
|
|
* - sysctl_protected_hardlinks enabled
|
|
|
|
* - fsuid does not match inode
|
|
|
|
* - hardlink source is unsafe (see safe_hardlink_source() above)
|
2015-10-20 08:09:19 -06:00
|
|
|
* - not CAP_FOWNER in a namespace with the inode owner uid mapped
|
fs: add link restrictions
This adds symlink and hardlink restrictions to the Linux VFS.
Symlinks:
A long-standing class of security issues is the symlink-based
time-of-check-time-of-use race, most commonly seen in world-writable
directories like /tmp. The common method of exploitation of this flaw
is to cross privilege boundaries when following a given symlink (i.e. a
root process follows a symlink belonging to another user). For a likely
incomplete list of hundreds of examples across the years, please see:
http://cve.mitre.org/cgi-bin/cvekey.cgi?keyword=/tmp
The solution is to permit symlinks to only be followed when outside
a sticky world-writable directory, or when the uid of the symlink and
follower match, or when the directory owner matches the symlink's owner.
Some pointers to the history of earlier discussion that I could find:
1996 Aug, Zygo Blaxell
http://marc.info/?l=bugtraq&m=87602167419830&w=2
1996 Oct, Andrew Tridgell
http://lkml.indiana.edu/hypermail/linux/kernel/9610.2/0086.html
1997 Dec, Albert D Cahalan
http://lkml.org/lkml/1997/12/16/4
2005 Feb, Lorenzo Hernández GarcÃa-Hierro
http://lkml.indiana.edu/hypermail/linux/kernel/0502.0/1896.html
2010 May, Kees Cook
https://lkml.org/lkml/2010/5/30/144
Past objections and rebuttals could be summarized as:
- Violates POSIX.
- POSIX didn't consider this situation and it's not useful to follow
a broken specification at the cost of security.
- Might break unknown applications that use this feature.
- Applications that break because of the change are easy to spot and
fix. Applications that are vulnerable to symlink ToCToU by not having
the change aren't. Additionally, no applications have yet been found
that rely on this behavior.
- Applications should just use mkstemp() or O_CREATE|O_EXCL.
- True, but applications are not perfect, and new software is written
all the time that makes these mistakes; blocking this flaw at the
kernel is a single solution to the entire class of vulnerability.
- This should live in the core VFS.
- This should live in an LSM. (https://lkml.org/lkml/2010/5/31/135)
- This should live in an LSM.
- This should live in the core VFS. (https://lkml.org/lkml/2010/8/2/188)
Hardlinks:
On systems that have user-writable directories on the same partition
as system files, a long-standing class of security issues is the
hardlink-based time-of-check-time-of-use race, most commonly seen in
world-writable directories like /tmp. The common method of exploitation
of this flaw is to cross privilege boundaries when following a given
hardlink (i.e. a root process follows a hardlink created by another
user). Additionally, an issue exists where users can "pin" a potentially
vulnerable setuid/setgid file so that an administrator will not actually
upgrade a system fully.
The solution is to permit hardlinks to only be created when the user is
already the existing file's owner, or if they already have read/write
access to the existing file.
Many Linux users are surprised when they learn they can link to files
they have no access to, so this change appears to follow the doctrine
of "least surprise". Additionally, this change does not violate POSIX,
which states "the implementation may require that the calling process
has permission to access the existing file"[1].
This change is known to break some implementations of the "at" daemon,
though the version used by Fedora and Ubuntu has been fixed[2] for
a while. Otherwise, the change has been undisruptive while in use in
Ubuntu for the last 1.5 years.
[1] http://pubs.opengroup.org/onlinepubs/9699919799/functions/linkat.html
[2] http://anonscm.debian.org/gitweb/?p=collab-maint/at.git;a=commitdiff;h=f4114656c3a6c6f6070e315ffdf940a49eda3279
This patch is based on the patches in Openwall and grsecurity, along with
suggestions from Al Viro. I have added a sysctl to enable the protected
behavior, and documentation.
Signed-off-by: Kees Cook <keescook@chromium.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2012-07-25 18:29:07 -06:00
|
|
|
*
|
|
|
|
* Returns 0 if successful, -ve on error.
|
|
|
|
*/
|
|
|
|
static int may_linkat(struct path *link)
|
|
|
|
{
|
|
|
|
struct inode *inode;
|
|
|
|
|
|
|
|
if (!sysctl_protected_hardlinks)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
inode = link->dentry->d_inode;
|
|
|
|
|
|
|
|
/* Source inode owner (or CAP_FOWNER) can hardlink all they like,
|
|
|
|
* otherwise, it must be a safe source.
|
|
|
|
*/
|
2015-10-20 08:09:19 -06:00
|
|
|
if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
|
fs: add link restrictions
This adds symlink and hardlink restrictions to the Linux VFS.
Symlinks:
A long-standing class of security issues is the symlink-based
time-of-check-time-of-use race, most commonly seen in world-writable
directories like /tmp. The common method of exploitation of this flaw
is to cross privilege boundaries when following a given symlink (i.e. a
root process follows a symlink belonging to another user). For a likely
incomplete list of hundreds of examples across the years, please see:
http://cve.mitre.org/cgi-bin/cvekey.cgi?keyword=/tmp
The solution is to permit symlinks to only be followed when outside
a sticky world-writable directory, or when the uid of the symlink and
follower match, or when the directory owner matches the symlink's owner.
Some pointers to the history of earlier discussion that I could find:
1996 Aug, Zygo Blaxell
http://marc.info/?l=bugtraq&m=87602167419830&w=2
1996 Oct, Andrew Tridgell
http://lkml.indiana.edu/hypermail/linux/kernel/9610.2/0086.html
1997 Dec, Albert D Cahalan
http://lkml.org/lkml/1997/12/16/4
2005 Feb, Lorenzo Hernández GarcÃa-Hierro
http://lkml.indiana.edu/hypermail/linux/kernel/0502.0/1896.html
2010 May, Kees Cook
https://lkml.org/lkml/2010/5/30/144
Past objections and rebuttals could be summarized as:
- Violates POSIX.
- POSIX didn't consider this situation and it's not useful to follow
a broken specification at the cost of security.
- Might break unknown applications that use this feature.
- Applications that break because of the change are easy to spot and
fix. Applications that are vulnerable to symlink ToCToU by not having
the change aren't. Additionally, no applications have yet been found
that rely on this behavior.
- Applications should just use mkstemp() or O_CREATE|O_EXCL.
- True, but applications are not perfect, and new software is written
all the time that makes these mistakes; blocking this flaw at the
kernel is a single solution to the entire class of vulnerability.
- This should live in the core VFS.
- This should live in an LSM. (https://lkml.org/lkml/2010/5/31/135)
- This should live in an LSM.
- This should live in the core VFS. (https://lkml.org/lkml/2010/8/2/188)
Hardlinks:
On systems that have user-writable directories on the same partition
as system files, a long-standing class of security issues is the
hardlink-based time-of-check-time-of-use race, most commonly seen in
world-writable directories like /tmp. The common method of exploitation
of this flaw is to cross privilege boundaries when following a given
hardlink (i.e. a root process follows a hardlink created by another
user). Additionally, an issue exists where users can "pin" a potentially
vulnerable setuid/setgid file so that an administrator will not actually
upgrade a system fully.
The solution is to permit hardlinks to only be created when the user is
already the existing file's owner, or if they already have read/write
access to the existing file.
Many Linux users are surprised when they learn they can link to files
they have no access to, so this change appears to follow the doctrine
of "least surprise". Additionally, this change does not violate POSIX,
which states "the implementation may require that the calling process
has permission to access the existing file"[1].
This change is known to break some implementations of the "at" daemon,
though the version used by Fedora and Ubuntu has been fixed[2] for
a while. Otherwise, the change has been undisruptive while in use in
Ubuntu for the last 1.5 years.
[1] http://pubs.opengroup.org/onlinepubs/9699919799/functions/linkat.html
[2] http://anonscm.debian.org/gitweb/?p=collab-maint/at.git;a=commitdiff;h=f4114656c3a6c6f6070e315ffdf940a49eda3279
This patch is based on the patches in Openwall and grsecurity, along with
suggestions from Al Viro. I have added a sysctl to enable the protected
behavior, and documentation.
Signed-off-by: Kees Cook <keescook@chromium.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2012-07-25 18:29:07 -06:00
|
|
|
return 0;
|
|
|
|
|
2012-07-25 18:29:08 -06:00
|
|
|
audit_log_link_denied("linkat", link);
|
fs: add link restrictions
This adds symlink and hardlink restrictions to the Linux VFS.
Symlinks:
A long-standing class of security issues is the symlink-based
time-of-check-time-of-use race, most commonly seen in world-writable
directories like /tmp. The common method of exploitation of this flaw
is to cross privilege boundaries when following a given symlink (i.e. a
root process follows a symlink belonging to another user). For a likely
incomplete list of hundreds of examples across the years, please see:
http://cve.mitre.org/cgi-bin/cvekey.cgi?keyword=/tmp
The solution is to permit symlinks to only be followed when outside
a sticky world-writable directory, or when the uid of the symlink and
follower match, or when the directory owner matches the symlink's owner.
Some pointers to the history of earlier discussion that I could find:
1996 Aug, Zygo Blaxell
http://marc.info/?l=bugtraq&m=87602167419830&w=2
1996 Oct, Andrew Tridgell
http://lkml.indiana.edu/hypermail/linux/kernel/9610.2/0086.html
1997 Dec, Albert D Cahalan
http://lkml.org/lkml/1997/12/16/4
2005 Feb, Lorenzo Hernández GarcÃa-Hierro
http://lkml.indiana.edu/hypermail/linux/kernel/0502.0/1896.html
2010 May, Kees Cook
https://lkml.org/lkml/2010/5/30/144
Past objections and rebuttals could be summarized as:
- Violates POSIX.
- POSIX didn't consider this situation and it's not useful to follow
a broken specification at the cost of security.
- Might break unknown applications that use this feature.
- Applications that break because of the change are easy to spot and
fix. Applications that are vulnerable to symlink ToCToU by not having
the change aren't. Additionally, no applications have yet been found
that rely on this behavior.
- Applications should just use mkstemp() or O_CREATE|O_EXCL.
- True, but applications are not perfect, and new software is written
all the time that makes these mistakes; blocking this flaw at the
kernel is a single solution to the entire class of vulnerability.
- This should live in the core VFS.
- This should live in an LSM. (https://lkml.org/lkml/2010/5/31/135)
- This should live in an LSM.
- This should live in the core VFS. (https://lkml.org/lkml/2010/8/2/188)
Hardlinks:
On systems that have user-writable directories on the same partition
as system files, a long-standing class of security issues is the
hardlink-based time-of-check-time-of-use race, most commonly seen in
world-writable directories like /tmp. The common method of exploitation
of this flaw is to cross privilege boundaries when following a given
hardlink (i.e. a root process follows a hardlink created by another
user). Additionally, an issue exists where users can "pin" a potentially
vulnerable setuid/setgid file so that an administrator will not actually
upgrade a system fully.
The solution is to permit hardlinks to only be created when the user is
already the existing file's owner, or if they already have read/write
access to the existing file.
Many Linux users are surprised when they learn they can link to files
they have no access to, so this change appears to follow the doctrine
of "least surprise". Additionally, this change does not violate POSIX,
which states "the implementation may require that the calling process
has permission to access the existing file"[1].
This change is known to break some implementations of the "at" daemon,
though the version used by Fedora and Ubuntu has been fixed[2] for
a while. Otherwise, the change has been undisruptive while in use in
Ubuntu for the last 1.5 years.
[1] http://pubs.opengroup.org/onlinepubs/9699919799/functions/linkat.html
[2] http://anonscm.debian.org/gitweb/?p=collab-maint/at.git;a=commitdiff;h=f4114656c3a6c6f6070e315ffdf940a49eda3279
This patch is based on the patches in Openwall and grsecurity, along with
suggestions from Al Viro. I have added a sysctl to enable the protected
behavior, and documentation.
Signed-off-by: Kees Cook <keescook@chromium.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2012-07-25 18:29:07 -06:00
|
|
|
return -EPERM;
|
|
|
|
}
|
|
|
|
|
2015-04-18 22:53:50 -06:00
|
|
|
static __always_inline
|
|
|
|
const char *get_link(struct nameidata *nd)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2015-05-10 09:50:01 -06:00
|
|
|
struct saved *last = nd->stack + nd->depth - 1;
|
2015-05-06 14:01:56 -06:00
|
|
|
struct dentry *dentry = last->link.dentry;
|
2015-12-29 13:58:39 -07:00
|
|
|
struct inode *inode = nd->link_inode;
|
2012-06-10 02:15:17 -06:00
|
|
|
int error;
|
2015-04-18 16:23:41 -06:00
|
|
|
const char *res;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2015-03-22 20:37:40 -06:00
|
|
|
if (!(nd->flags & LOOKUP_RCU)) {
|
|
|
|
touch_atime(&last->link);
|
|
|
|
cond_resched();
|
|
|
|
} else if (atime_needs_update(&last->link, inode)) {
|
2015-05-09 11:04:24 -06:00
|
|
|
if (unlikely(unlazy_walk(nd, NULL, 0)))
|
|
|
|
return ERR_PTR(-ECHILD);
|
2015-03-22 20:37:40 -06:00
|
|
|
touch_atime(&last->link);
|
2015-05-09 11:04:24 -06:00
|
|
|
}
|
2011-03-14 20:20:34 -06:00
|
|
|
|
2015-03-22 20:37:39 -06:00
|
|
|
error = security_inode_follow_link(dentry, inode,
|
|
|
|
nd->flags & LOOKUP_RCU);
|
|
|
|
if (unlikely(error))
|
2015-05-10 08:43:46 -06:00
|
|
|
return ERR_PTR(error);
|
2011-02-22 19:24:38 -07:00
|
|
|
|
2009-12-22 21:45:11 -07:00
|
|
|
nd->last_type = LAST_BIND;
|
2015-04-30 18:08:02 -06:00
|
|
|
res = inode->i_link;
|
|
|
|
if (!res) {
|
2015-12-29 13:58:39 -07:00
|
|
|
const char * (*get)(struct dentry *, struct inode *,
|
|
|
|
struct delayed_call *);
|
|
|
|
get = inode->i_op->get_link;
|
2015-05-09 16:15:21 -06:00
|
|
|
if (nd->flags & LOOKUP_RCU) {
|
2015-12-29 13:58:39 -07:00
|
|
|
res = get(NULL, inode, &last->done);
|
2015-11-17 08:20:54 -07:00
|
|
|
if (res == ERR_PTR(-ECHILD)) {
|
|
|
|
if (unlikely(unlazy_walk(nd, NULL, 0)))
|
|
|
|
return ERR_PTR(-ECHILD);
|
2015-12-29 13:58:39 -07:00
|
|
|
res = get(dentry, inode, &last->done);
|
2015-11-17 08:20:54 -07:00
|
|
|
}
|
|
|
|
} else {
|
2015-12-29 13:58:39 -07:00
|
|
|
res = get(dentry, inode, &last->done);
|
2015-05-09 16:15:21 -06:00
|
|
|
}
|
2015-12-29 13:58:39 -07:00
|
|
|
if (IS_ERR_OR_NULL(res))
|
2015-05-10 09:01:00 -06:00
|
|
|
return res;
|
|
|
|
}
|
|
|
|
if (*res == '/') {
|
2015-12-05 18:07:21 -07:00
|
|
|
if (!nd->root.mnt)
|
|
|
|
set_root(nd);
|
2015-12-05 18:51:58 -07:00
|
|
|
if (unlikely(nd_jump_root(nd)))
|
|
|
|
return ERR_PTR(-ECHILD);
|
2015-05-10 09:01:00 -06:00
|
|
|
while (unlikely(*++res == '/'))
|
|
|
|
;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
2015-05-10 09:01:00 -06:00
|
|
|
if (!*res)
|
|
|
|
res = NULL;
|
2015-04-18 16:23:41 -06:00
|
|
|
return res;
|
|
|
|
}
|
2012-06-10 02:15:17 -06:00
|
|
|
|
2012-06-25 05:55:28 -06:00
|
|
|
/*
|
|
|
|
* follow_up - Find the mountpoint of path's vfsmount
|
|
|
|
*
|
|
|
|
* Given a path, find the mountpoint of its source file system.
|
|
|
|
* Replace @path with the path of the mountpoint in the parent mount.
|
|
|
|
* Up is towards /.
|
|
|
|
*
|
|
|
|
* Return 1 if we went up a level and 0 if we were already at the
|
|
|
|
* root.
|
|
|
|
*/
|
2009-04-18 01:26:48 -06:00
|
|
|
int follow_up(struct path *path)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2011-11-24 20:19:58 -07:00
|
|
|
struct mount *mnt = real_mount(path->mnt);
|
|
|
|
struct mount *parent;
|
2005-04-16 16:20:36 -06:00
|
|
|
struct dentry *mountpoint;
|
fs: brlock vfsmount_lock
fs: brlock vfsmount_lock
Use a brlock for the vfsmount lock. It must be taken for write whenever
modifying the mount hash or associated fields, and may be taken for read when
performing mount hash lookups.
A new lock is added for the mnt-id allocator, so it doesn't need to take
the heavy vfsmount write-lock.
The number of atomics should remain the same for fastpath rlock cases, though
code would be slightly slower due to per-cpu access. Scalability is not not be
much improved in common cases yet, due to other locks (ie. dcache_lock) getting
in the way. However path lookups crossing mountpoints should be one case where
scalability is improved (currently requiring the global lock).
The slowpath is slower due to use of brlock. On a 64 core, 64 socket, 32 node
Altix system (high latency to remote nodes), a simple umount microbenchmark
(mount --bind mnt mnt2 ; umount mnt2 loop 1000 times), before this patch it
took 6.8s, afterwards took 7.1s, about 5% slower.
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2010-08-17 12:37:39 -06:00
|
|
|
|
2013-09-29 20:06:07 -06:00
|
|
|
read_seqlock_excl(&mount_lock);
|
2011-11-24 20:19:58 -07:00
|
|
|
parent = mnt->mnt_parent;
|
2012-07-18 07:32:50 -06:00
|
|
|
if (parent == mnt) {
|
2013-09-29 20:06:07 -06:00
|
|
|
read_sequnlock_excl(&mount_lock);
|
2005-04-16 16:20:36 -06:00
|
|
|
return 0;
|
|
|
|
}
|
2011-11-24 20:19:58 -07:00
|
|
|
mntget(&parent->mnt);
|
2011-11-24 20:25:07 -07:00
|
|
|
mountpoint = dget(mnt->mnt_mountpoint);
|
2013-09-29 20:06:07 -06:00
|
|
|
read_sequnlock_excl(&mount_lock);
|
2009-04-18 01:26:48 -06:00
|
|
|
dput(path->dentry);
|
|
|
|
path->dentry = mountpoint;
|
|
|
|
mntput(path->mnt);
|
2011-11-24 20:19:58 -07:00
|
|
|
path->mnt = &parent->mnt;
|
2005-04-16 16:20:36 -06:00
|
|
|
return 1;
|
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(follow_up);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2011-01-06 23:49:38 -07:00
|
|
|
/*
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
* Perform an automount
|
|
|
|
* - return -EISDIR to tell follow_managed() to stop and return the path we
|
|
|
|
* were called with.
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
2015-03-22 20:37:38 -06:00
|
|
|
static int follow_automount(struct path *path, struct nameidata *nd,
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
bool *need_mntput)
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
{
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
struct vfsmount *mnt;
|
2011-01-14 12:10:03 -07:00
|
|
|
int err;
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
|
|
|
|
if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
|
|
|
|
return -EREMOTE;
|
|
|
|
|
2011-09-05 10:06:26 -06:00
|
|
|
/* We don't want to mount if someone's just doing a stat -
|
|
|
|
* unless they're stat'ing a directory and appended a '/' to
|
|
|
|
* the name.
|
|
|
|
*
|
|
|
|
* We do, however, want to mount if someone wants to open or
|
|
|
|
* create a file of any type under the mountpoint, wants to
|
|
|
|
* traverse through the mountpoint or wants to open the
|
|
|
|
* mounted directory. Also, autofs may mark negative dentries
|
|
|
|
* as being automount points. These will need the attentions
|
|
|
|
* of the daemon to instantiate them before they can be used.
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
*/
|
2015-03-22 20:37:38 -06:00
|
|
|
if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
|
|
|
|
LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
|
2011-09-05 10:06:26 -06:00
|
|
|
path->dentry->d_inode)
|
|
|
|
return -EISDIR;
|
|
|
|
|
2015-03-22 20:37:38 -06:00
|
|
|
nd->total_link_count++;
|
|
|
|
if (nd->total_link_count >= 40)
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
return -ELOOP;
|
|
|
|
|
|
|
|
mnt = path->dentry->d_op->d_automount(path);
|
|
|
|
if (IS_ERR(mnt)) {
|
|
|
|
/*
|
|
|
|
* The filesystem is allowed to return -EISDIR here to indicate
|
|
|
|
* it doesn't want to automount. For instance, autofs would do
|
|
|
|
* this so that its userspace daemon can mount on this dentry.
|
|
|
|
*
|
|
|
|
* However, we can only permit this if it's a terminal point in
|
|
|
|
* the path being looked up; if it wasn't then the remainder of
|
|
|
|
* the path is inaccessible and we should say so.
|
|
|
|
*/
|
2015-03-22 20:37:38 -06:00
|
|
|
if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
return -EREMOTE;
|
|
|
|
return PTR_ERR(mnt);
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
}
|
2011-01-14 12:10:03 -07:00
|
|
|
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
if (!mnt) /* mount collision */
|
|
|
|
return 0;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
|
VFS: Fix vfsmount overput on simultaneous automount
[Kudos to dhowells for tracking that crap down]
If two processes attempt to cause automounting on the same mountpoint at the
same time, the vfsmount holding the mountpoint will be left with one too few
references on it, causing a BUG when the kernel tries to clean up.
The problem is that lock_mount() drops the caller's reference to the
mountpoint's vfsmount in the case where it finds something already mounted on
the mountpoint as it transits to the mounted filesystem and replaces path->mnt
with the new mountpoint vfsmount.
During a pathwalk, however, we don't take a reference on the vfsmount if it is
the same as the one in the nameidata struct, but do_add_mount() doesn't know
this.
The fix is to make sure we have a ref on the vfsmount of the mountpoint before
calling do_add_mount(). However, if lock_mount() doesn't transit, we're then
left with an extra ref on the mountpoint vfsmount which needs releasing.
We can handle that in follow_managed() by not making assumptions about what
we can and what we cannot get from lookup_mnt() as the current code does.
The callers of follow_managed() expect that reference to path->mnt will be
grabbed iff path->mnt has been changed. follow_managed() and follow_automount()
keep track of whether such reference has been grabbed and assume that it'll
happen in those and only those cases that'll have us return with changed
path->mnt. That assumption is almost correct - it breaks in case of
racing automounts and in even harder to hit race between following a mountpoint
and a couple of mount --move. The thing is, we don't need to make that
assumption at all - after the end of loop in follow_manage() we can check
if path->mnt has ended up unchanged and do mntput() if needed.
The BUG can be reproduced with the following test program:
#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <sys/wait.h>
int main(int argc, char **argv)
{
int pid, ws;
struct stat buf;
pid = fork();
stat(argv[1], &buf);
if (pid > 0) wait(&ws);
return 0;
}
and the following procedure:
(1) Mount an NFS volume that on the server has something else mounted on a
subdirectory. For instance, I can mount / from my server:
mount warthog:/ /mnt -t nfs4 -r
On the server /data has another filesystem mounted on it, so NFS will see
a change in FSID as it walks down the path, and will mark /mnt/data as
being a mountpoint. This will cause the automount code to be triggered.
!!! Do not look inside the mounted fs at this point !!!
(2) Run the above program on a file within the submount to generate two
simultaneous automount requests:
/tmp/forkstat /mnt/data/testfile
(3) Unmount the automounted submount:
umount /mnt/data
(4) Unmount the original mount:
umount /mnt
At this point the kernel should throw a BUG with something like the
following:
BUG: Dentry ffff880032e3c5c0{i=2,n=} still in use (1) [unmount of nfs4 0:12]
Note that the bug appears on the root dentry of the original mount, not the
mountpoint and not the submount because sys_umount() hasn't got to its final
mntput_no_expire() yet, but this isn't so obvious from the call trace:
[<ffffffff8117cd82>] shrink_dcache_for_umount+0x69/0x82
[<ffffffff8116160e>] generic_shutdown_super+0x37/0x15b
[<ffffffffa00fae56>] ? nfs_super_return_all_delegations+0x2e/0x1b1 [nfs]
[<ffffffff811617f3>] kill_anon_super+0x1d/0x7e
[<ffffffffa00d0be1>] nfs4_kill_super+0x60/0xb6 [nfs]
[<ffffffff81161c17>] deactivate_locked_super+0x34/0x83
[<ffffffff811629ff>] deactivate_super+0x6f/0x7b
[<ffffffff81186261>] mntput_no_expire+0x18d/0x199
[<ffffffff811862a8>] mntput+0x3b/0x44
[<ffffffff81186d87>] release_mounts+0xa2/0xbf
[<ffffffff811876af>] sys_umount+0x47a/0x4ba
[<ffffffff8109e1ca>] ? trace_hardirqs_on_caller+0x1fd/0x22f
[<ffffffff816ea86b>] system_call_fastpath+0x16/0x1b
as do_umount() is inlined. However, you can see release_mounts() in there.
Note also that it may be necessary to have multiple CPU cores to be able to
trigger this bug.
Tested-by: Jeff Layton <jlayton@redhat.com>
Tested-by: Ian Kent <raven@themaw.net>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-06-16 08:10:06 -06:00
|
|
|
if (!*need_mntput) {
|
|
|
|
/* lock_mount() may release path->mnt on error */
|
|
|
|
mntget(path->mnt);
|
|
|
|
*need_mntput = true;
|
|
|
|
}
|
2011-01-16 23:35:23 -07:00
|
|
|
err = finish_automount(mnt, path);
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
|
2011-01-14 12:10:03 -07:00
|
|
|
switch (err) {
|
|
|
|
case -EBUSY:
|
|
|
|
/* Someone else made a mount here whilst we were busy */
|
2011-01-16 23:35:23 -07:00
|
|
|
return 0;
|
2011-01-14 12:10:03 -07:00
|
|
|
case 0:
|
VFS: Fix vfsmount overput on simultaneous automount
[Kudos to dhowells for tracking that crap down]
If two processes attempt to cause automounting on the same mountpoint at the
same time, the vfsmount holding the mountpoint will be left with one too few
references on it, causing a BUG when the kernel tries to clean up.
The problem is that lock_mount() drops the caller's reference to the
mountpoint's vfsmount in the case where it finds something already mounted on
the mountpoint as it transits to the mounted filesystem and replaces path->mnt
with the new mountpoint vfsmount.
During a pathwalk, however, we don't take a reference on the vfsmount if it is
the same as the one in the nameidata struct, but do_add_mount() doesn't know
this.
The fix is to make sure we have a ref on the vfsmount of the mountpoint before
calling do_add_mount(). However, if lock_mount() doesn't transit, we're then
left with an extra ref on the mountpoint vfsmount which needs releasing.
We can handle that in follow_managed() by not making assumptions about what
we can and what we cannot get from lookup_mnt() as the current code does.
The callers of follow_managed() expect that reference to path->mnt will be
grabbed iff path->mnt has been changed. follow_managed() and follow_automount()
keep track of whether such reference has been grabbed and assume that it'll
happen in those and only those cases that'll have us return with changed
path->mnt. That assumption is almost correct - it breaks in case of
racing automounts and in even harder to hit race between following a mountpoint
and a couple of mount --move. The thing is, we don't need to make that
assumption at all - after the end of loop in follow_manage() we can check
if path->mnt has ended up unchanged and do mntput() if needed.
The BUG can be reproduced with the following test program:
#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <sys/wait.h>
int main(int argc, char **argv)
{
int pid, ws;
struct stat buf;
pid = fork();
stat(argv[1], &buf);
if (pid > 0) wait(&ws);
return 0;
}
and the following procedure:
(1) Mount an NFS volume that on the server has something else mounted on a
subdirectory. For instance, I can mount / from my server:
mount warthog:/ /mnt -t nfs4 -r
On the server /data has another filesystem mounted on it, so NFS will see
a change in FSID as it walks down the path, and will mark /mnt/data as
being a mountpoint. This will cause the automount code to be triggered.
!!! Do not look inside the mounted fs at this point !!!
(2) Run the above program on a file within the submount to generate two
simultaneous automount requests:
/tmp/forkstat /mnt/data/testfile
(3) Unmount the automounted submount:
umount /mnt/data
(4) Unmount the original mount:
umount /mnt
At this point the kernel should throw a BUG with something like the
following:
BUG: Dentry ffff880032e3c5c0{i=2,n=} still in use (1) [unmount of nfs4 0:12]
Note that the bug appears on the root dentry of the original mount, not the
mountpoint and not the submount because sys_umount() hasn't got to its final
mntput_no_expire() yet, but this isn't so obvious from the call trace:
[<ffffffff8117cd82>] shrink_dcache_for_umount+0x69/0x82
[<ffffffff8116160e>] generic_shutdown_super+0x37/0x15b
[<ffffffffa00fae56>] ? nfs_super_return_all_delegations+0x2e/0x1b1 [nfs]
[<ffffffff811617f3>] kill_anon_super+0x1d/0x7e
[<ffffffffa00d0be1>] nfs4_kill_super+0x60/0xb6 [nfs]
[<ffffffff81161c17>] deactivate_locked_super+0x34/0x83
[<ffffffff811629ff>] deactivate_super+0x6f/0x7b
[<ffffffff81186261>] mntput_no_expire+0x18d/0x199
[<ffffffff811862a8>] mntput+0x3b/0x44
[<ffffffff81186d87>] release_mounts+0xa2/0xbf
[<ffffffff811876af>] sys_umount+0x47a/0x4ba
[<ffffffff8109e1ca>] ? trace_hardirqs_on_caller+0x1fd/0x22f
[<ffffffff816ea86b>] system_call_fastpath+0x16/0x1b
as do_umount() is inlined. However, you can see release_mounts() in there.
Note also that it may be necessary to have multiple CPU cores to be able to
trigger this bug.
Tested-by: Jeff Layton <jlayton@redhat.com>
Tested-by: Ian Kent <raven@themaw.net>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-06-16 08:10:06 -06:00
|
|
|
path_put(path);
|
2011-01-14 12:10:03 -07:00
|
|
|
path->mnt = mnt;
|
|
|
|
path->dentry = dget(mnt->mnt_root);
|
|
|
|
return 0;
|
2011-01-16 23:35:23 -07:00
|
|
|
default:
|
|
|
|
return err;
|
2011-01-14 12:10:03 -07:00
|
|
|
}
|
2011-01-16 23:35:23 -07:00
|
|
|
|
2005-06-06 14:36:05 -06:00
|
|
|
}
|
|
|
|
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
/*
|
|
|
|
* Handle a dentry that is managed in some way.
|
Add a dentry op to allow processes to be held during pathwalk transit
Add a dentry op (d_manage) to permit a filesystem to hold a process and make it
sleep when it tries to transit away from one of that filesystem's directories
during a pathwalk. The operation is keyed off a new dentry flag
(DCACHE_MANAGE_TRANSIT).
The filesystem is allowed to be selective about which processes it holds and
which it permits to continue on or prohibits from transiting from each flagged
directory. This will allow autofs to hold up client processes whilst letting
its userspace daemon through to maintain the directory or the stuff behind it
or mounted upon it.
The ->d_manage() dentry operation:
int (*d_manage)(struct path *path, bool mounting_here);
takes a pointer to the directory about to be transited away from and a flag
indicating whether the transit is undertaken by do_add_mount() or
do_move_mount() skipping through a pile of filesystems mounted on a mountpoint.
It should return 0 if successful and to let the process continue on its way;
-EISDIR to prohibit the caller from skipping to overmounted filesystems or
automounting, and to use this directory; or some other error code to return to
the user.
->d_manage() is called with namespace_sem writelocked if mounting_here is true
and no other locks held, so it may sleep. However, if mounting_here is true,
it may not initiate or wait for a mount or unmount upon the parameter
directory, even if the act is actually performed by userspace.
Within fs/namei.c, follow_managed() is extended to check with d_manage() first
on each managed directory, before transiting away from it or attempting to
automount upon it.
follow_down() is renamed follow_down_one() and should only be used where the
filesystem deliberately intends to avoid management steps (e.g. autofs).
A new follow_down() is added that incorporates the loop done by all other
callers of follow_down() (do_add/move_mount(), autofs and NFSD; whilst AFS, NFS
and CIFS do use it, their use is removed by converting them to use
d_automount()). The new follow_down() calls d_manage() as appropriate. It
also takes an extra parameter to indicate if it is being called from mount code
(with namespace_sem writelocked) which it passes to d_manage(). follow_down()
ignores automount points so that it can be used to mount on them.
__follow_mount_rcu() is made to abort rcu-walk mode if it hits a directory with
DCACHE_MANAGE_TRANSIT set on the basis that we're probably going to have to
sleep. It would be possible to enter d_manage() in rcu-walk mode too, and have
that determine whether to abort or not itself. That would allow the autofs
daemon to continue on in rcu-walk mode.
Note that DCACHE_MANAGE_TRANSIT on a directory should be cleared when it isn't
required as every tranist from that directory will cause d_manage() to be
invoked. It can always be set again when necessary.
==========================
WHAT THIS MEANS FOR AUTOFS
==========================
Autofs currently uses the lookup() inode op and the d_revalidate() dentry op to
trigger the automounting of indirect mounts, and both of these can be called
with i_mutex held.
autofs knows that the i_mutex will be held by the caller in lookup(), and so
can drop it before invoking the daemon - but this isn't so for d_revalidate(),
since the lock is only held on _some_ of the code paths that call it. This
means that autofs can't risk dropping i_mutex from its d_revalidate() function
before it calls the daemon.
The bug could manifest itself as, for example, a process that's trying to
validate an automount dentry that gets made to wait because that dentry is
expired and needs cleaning up:
mkdir S ffffffff8014e05a 0 32580 24956
Call Trace:
[<ffffffff885371fd>] :autofs4:autofs4_wait+0x674/0x897
[<ffffffff80127f7d>] avc_has_perm+0x46/0x58
[<ffffffff8009fdcf>] autoremove_wake_function+0x0/0x2e
[<ffffffff88537be6>] :autofs4:autofs4_expire_wait+0x41/0x6b
[<ffffffff88535cfc>] :autofs4:autofs4_revalidate+0x91/0x149
[<ffffffff80036d96>] __lookup_hash+0xa0/0x12f
[<ffffffff80057a2f>] lookup_create+0x46/0x80
[<ffffffff800e6e31>] sys_mkdirat+0x56/0xe4
versus the automount daemon which wants to remove that dentry, but can't
because the normal process is holding the i_mutex lock:
automount D ffffffff8014e05a 0 32581 1 32561
Call Trace:
[<ffffffff80063c3f>] __mutex_lock_slowpath+0x60/0x9b
[<ffffffff8000ccf1>] do_path_lookup+0x2ca/0x2f1
[<ffffffff80063c89>] .text.lock.mutex+0xf/0x14
[<ffffffff800e6d55>] do_rmdir+0x77/0xde
[<ffffffff8005d229>] tracesys+0x71/0xe0
[<ffffffff8005d28d>] tracesys+0xd5/0xe0
which means that the system is deadlocked.
This patch allows autofs to hold up normal processes whilst the daemon goes
ahead and does things to the dentry tree behind the automouter point without
risking a deadlock as almost no locks are held in d_manage() and none in
d_automount().
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:26 -07:00
|
|
|
* - Flagged for transit management (autofs)
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
* - Flagged as mountpoint
|
|
|
|
* - Flagged as automount point
|
|
|
|
*
|
|
|
|
* This may only be called in refwalk mode.
|
|
|
|
*
|
|
|
|
* Serialization is taken care of in namespace.c
|
|
|
|
*/
|
2015-03-22 20:37:38 -06:00
|
|
|
static int follow_managed(struct path *path, struct nameidata *nd)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
VFS: Fix vfsmount overput on simultaneous automount
[Kudos to dhowells for tracking that crap down]
If two processes attempt to cause automounting on the same mountpoint at the
same time, the vfsmount holding the mountpoint will be left with one too few
references on it, causing a BUG when the kernel tries to clean up.
The problem is that lock_mount() drops the caller's reference to the
mountpoint's vfsmount in the case where it finds something already mounted on
the mountpoint as it transits to the mounted filesystem and replaces path->mnt
with the new mountpoint vfsmount.
During a pathwalk, however, we don't take a reference on the vfsmount if it is
the same as the one in the nameidata struct, but do_add_mount() doesn't know
this.
The fix is to make sure we have a ref on the vfsmount of the mountpoint before
calling do_add_mount(). However, if lock_mount() doesn't transit, we're then
left with an extra ref on the mountpoint vfsmount which needs releasing.
We can handle that in follow_managed() by not making assumptions about what
we can and what we cannot get from lookup_mnt() as the current code does.
The callers of follow_managed() expect that reference to path->mnt will be
grabbed iff path->mnt has been changed. follow_managed() and follow_automount()
keep track of whether such reference has been grabbed and assume that it'll
happen in those and only those cases that'll have us return with changed
path->mnt. That assumption is almost correct - it breaks in case of
racing automounts and in even harder to hit race between following a mountpoint
and a couple of mount --move. The thing is, we don't need to make that
assumption at all - after the end of loop in follow_manage() we can check
if path->mnt has ended up unchanged and do mntput() if needed.
The BUG can be reproduced with the following test program:
#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <sys/wait.h>
int main(int argc, char **argv)
{
int pid, ws;
struct stat buf;
pid = fork();
stat(argv[1], &buf);
if (pid > 0) wait(&ws);
return 0;
}
and the following procedure:
(1) Mount an NFS volume that on the server has something else mounted on a
subdirectory. For instance, I can mount / from my server:
mount warthog:/ /mnt -t nfs4 -r
On the server /data has another filesystem mounted on it, so NFS will see
a change in FSID as it walks down the path, and will mark /mnt/data as
being a mountpoint. This will cause the automount code to be triggered.
!!! Do not look inside the mounted fs at this point !!!
(2) Run the above program on a file within the submount to generate two
simultaneous automount requests:
/tmp/forkstat /mnt/data/testfile
(3) Unmount the automounted submount:
umount /mnt/data
(4) Unmount the original mount:
umount /mnt
At this point the kernel should throw a BUG with something like the
following:
BUG: Dentry ffff880032e3c5c0{i=2,n=} still in use (1) [unmount of nfs4 0:12]
Note that the bug appears on the root dentry of the original mount, not the
mountpoint and not the submount because sys_umount() hasn't got to its final
mntput_no_expire() yet, but this isn't so obvious from the call trace:
[<ffffffff8117cd82>] shrink_dcache_for_umount+0x69/0x82
[<ffffffff8116160e>] generic_shutdown_super+0x37/0x15b
[<ffffffffa00fae56>] ? nfs_super_return_all_delegations+0x2e/0x1b1 [nfs]
[<ffffffff811617f3>] kill_anon_super+0x1d/0x7e
[<ffffffffa00d0be1>] nfs4_kill_super+0x60/0xb6 [nfs]
[<ffffffff81161c17>] deactivate_locked_super+0x34/0x83
[<ffffffff811629ff>] deactivate_super+0x6f/0x7b
[<ffffffff81186261>] mntput_no_expire+0x18d/0x199
[<ffffffff811862a8>] mntput+0x3b/0x44
[<ffffffff81186d87>] release_mounts+0xa2/0xbf
[<ffffffff811876af>] sys_umount+0x47a/0x4ba
[<ffffffff8109e1ca>] ? trace_hardirqs_on_caller+0x1fd/0x22f
[<ffffffff816ea86b>] system_call_fastpath+0x16/0x1b
as do_umount() is inlined. However, you can see release_mounts() in there.
Note also that it may be necessary to have multiple CPU cores to be able to
trigger this bug.
Tested-by: Jeff Layton <jlayton@redhat.com>
Tested-by: Ian Kent <raven@themaw.net>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-06-16 08:10:06 -06:00
|
|
|
struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
unsigned managed;
|
|
|
|
bool need_mntput = false;
|
VFS: Fix vfsmount overput on simultaneous automount
[Kudos to dhowells for tracking that crap down]
If two processes attempt to cause automounting on the same mountpoint at the
same time, the vfsmount holding the mountpoint will be left with one too few
references on it, causing a BUG when the kernel tries to clean up.
The problem is that lock_mount() drops the caller's reference to the
mountpoint's vfsmount in the case where it finds something already mounted on
the mountpoint as it transits to the mounted filesystem and replaces path->mnt
with the new mountpoint vfsmount.
During a pathwalk, however, we don't take a reference on the vfsmount if it is
the same as the one in the nameidata struct, but do_add_mount() doesn't know
this.
The fix is to make sure we have a ref on the vfsmount of the mountpoint before
calling do_add_mount(). However, if lock_mount() doesn't transit, we're then
left with an extra ref on the mountpoint vfsmount which needs releasing.
We can handle that in follow_managed() by not making assumptions about what
we can and what we cannot get from lookup_mnt() as the current code does.
The callers of follow_managed() expect that reference to path->mnt will be
grabbed iff path->mnt has been changed. follow_managed() and follow_automount()
keep track of whether such reference has been grabbed and assume that it'll
happen in those and only those cases that'll have us return with changed
path->mnt. That assumption is almost correct - it breaks in case of
racing automounts and in even harder to hit race between following a mountpoint
and a couple of mount --move. The thing is, we don't need to make that
assumption at all - after the end of loop in follow_manage() we can check
if path->mnt has ended up unchanged and do mntput() if needed.
The BUG can be reproduced with the following test program:
#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <sys/wait.h>
int main(int argc, char **argv)
{
int pid, ws;
struct stat buf;
pid = fork();
stat(argv[1], &buf);
if (pid > 0) wait(&ws);
return 0;
}
and the following procedure:
(1) Mount an NFS volume that on the server has something else mounted on a
subdirectory. For instance, I can mount / from my server:
mount warthog:/ /mnt -t nfs4 -r
On the server /data has another filesystem mounted on it, so NFS will see
a change in FSID as it walks down the path, and will mark /mnt/data as
being a mountpoint. This will cause the automount code to be triggered.
!!! Do not look inside the mounted fs at this point !!!
(2) Run the above program on a file within the submount to generate two
simultaneous automount requests:
/tmp/forkstat /mnt/data/testfile
(3) Unmount the automounted submount:
umount /mnt/data
(4) Unmount the original mount:
umount /mnt
At this point the kernel should throw a BUG with something like the
following:
BUG: Dentry ffff880032e3c5c0{i=2,n=} still in use (1) [unmount of nfs4 0:12]
Note that the bug appears on the root dentry of the original mount, not the
mountpoint and not the submount because sys_umount() hasn't got to its final
mntput_no_expire() yet, but this isn't so obvious from the call trace:
[<ffffffff8117cd82>] shrink_dcache_for_umount+0x69/0x82
[<ffffffff8116160e>] generic_shutdown_super+0x37/0x15b
[<ffffffffa00fae56>] ? nfs_super_return_all_delegations+0x2e/0x1b1 [nfs]
[<ffffffff811617f3>] kill_anon_super+0x1d/0x7e
[<ffffffffa00d0be1>] nfs4_kill_super+0x60/0xb6 [nfs]
[<ffffffff81161c17>] deactivate_locked_super+0x34/0x83
[<ffffffff811629ff>] deactivate_super+0x6f/0x7b
[<ffffffff81186261>] mntput_no_expire+0x18d/0x199
[<ffffffff811862a8>] mntput+0x3b/0x44
[<ffffffff81186d87>] release_mounts+0xa2/0xbf
[<ffffffff811876af>] sys_umount+0x47a/0x4ba
[<ffffffff8109e1ca>] ? trace_hardirqs_on_caller+0x1fd/0x22f
[<ffffffff816ea86b>] system_call_fastpath+0x16/0x1b
as do_umount() is inlined. However, you can see release_mounts() in there.
Note also that it may be necessary to have multiple CPU cores to be able to
trigger this bug.
Tested-by: Jeff Layton <jlayton@redhat.com>
Tested-by: Ian Kent <raven@themaw.net>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-06-16 08:10:06 -06:00
|
|
|
int ret = 0;
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
|
|
|
|
/* Given that we're not holding a lock here, we retain the value in a
|
|
|
|
* local variable for each dentry as we look at it so that we don't see
|
|
|
|
* the components of that value change under us */
|
|
|
|
while (managed = ACCESS_ONCE(path->dentry->d_flags),
|
|
|
|
managed &= DCACHE_MANAGED_DENTRY,
|
|
|
|
unlikely(managed != 0)) {
|
Add a dentry op to allow processes to be held during pathwalk transit
Add a dentry op (d_manage) to permit a filesystem to hold a process and make it
sleep when it tries to transit away from one of that filesystem's directories
during a pathwalk. The operation is keyed off a new dentry flag
(DCACHE_MANAGE_TRANSIT).
The filesystem is allowed to be selective about which processes it holds and
which it permits to continue on or prohibits from transiting from each flagged
directory. This will allow autofs to hold up client processes whilst letting
its userspace daemon through to maintain the directory or the stuff behind it
or mounted upon it.
The ->d_manage() dentry operation:
int (*d_manage)(struct path *path, bool mounting_here);
takes a pointer to the directory about to be transited away from and a flag
indicating whether the transit is undertaken by do_add_mount() or
do_move_mount() skipping through a pile of filesystems mounted on a mountpoint.
It should return 0 if successful and to let the process continue on its way;
-EISDIR to prohibit the caller from skipping to overmounted filesystems or
automounting, and to use this directory; or some other error code to return to
the user.
->d_manage() is called with namespace_sem writelocked if mounting_here is true
and no other locks held, so it may sleep. However, if mounting_here is true,
it may not initiate or wait for a mount or unmount upon the parameter
directory, even if the act is actually performed by userspace.
Within fs/namei.c, follow_managed() is extended to check with d_manage() first
on each managed directory, before transiting away from it or attempting to
automount upon it.
follow_down() is renamed follow_down_one() and should only be used where the
filesystem deliberately intends to avoid management steps (e.g. autofs).
A new follow_down() is added that incorporates the loop done by all other
callers of follow_down() (do_add/move_mount(), autofs and NFSD; whilst AFS, NFS
and CIFS do use it, their use is removed by converting them to use
d_automount()). The new follow_down() calls d_manage() as appropriate. It
also takes an extra parameter to indicate if it is being called from mount code
(with namespace_sem writelocked) which it passes to d_manage(). follow_down()
ignores automount points so that it can be used to mount on them.
__follow_mount_rcu() is made to abort rcu-walk mode if it hits a directory with
DCACHE_MANAGE_TRANSIT set on the basis that we're probably going to have to
sleep. It would be possible to enter d_manage() in rcu-walk mode too, and have
that determine whether to abort or not itself. That would allow the autofs
daemon to continue on in rcu-walk mode.
Note that DCACHE_MANAGE_TRANSIT on a directory should be cleared when it isn't
required as every tranist from that directory will cause d_manage() to be
invoked. It can always be set again when necessary.
==========================
WHAT THIS MEANS FOR AUTOFS
==========================
Autofs currently uses the lookup() inode op and the d_revalidate() dentry op to
trigger the automounting of indirect mounts, and both of these can be called
with i_mutex held.
autofs knows that the i_mutex will be held by the caller in lookup(), and so
can drop it before invoking the daemon - but this isn't so for d_revalidate(),
since the lock is only held on _some_ of the code paths that call it. This
means that autofs can't risk dropping i_mutex from its d_revalidate() function
before it calls the daemon.
The bug could manifest itself as, for example, a process that's trying to
validate an automount dentry that gets made to wait because that dentry is
expired and needs cleaning up:
mkdir S ffffffff8014e05a 0 32580 24956
Call Trace:
[<ffffffff885371fd>] :autofs4:autofs4_wait+0x674/0x897
[<ffffffff80127f7d>] avc_has_perm+0x46/0x58
[<ffffffff8009fdcf>] autoremove_wake_function+0x0/0x2e
[<ffffffff88537be6>] :autofs4:autofs4_expire_wait+0x41/0x6b
[<ffffffff88535cfc>] :autofs4:autofs4_revalidate+0x91/0x149
[<ffffffff80036d96>] __lookup_hash+0xa0/0x12f
[<ffffffff80057a2f>] lookup_create+0x46/0x80
[<ffffffff800e6e31>] sys_mkdirat+0x56/0xe4
versus the automount daemon which wants to remove that dentry, but can't
because the normal process is holding the i_mutex lock:
automount D ffffffff8014e05a 0 32581 1 32561
Call Trace:
[<ffffffff80063c3f>] __mutex_lock_slowpath+0x60/0x9b
[<ffffffff8000ccf1>] do_path_lookup+0x2ca/0x2f1
[<ffffffff80063c89>] .text.lock.mutex+0xf/0x14
[<ffffffff800e6d55>] do_rmdir+0x77/0xde
[<ffffffff8005d229>] tracesys+0x71/0xe0
[<ffffffff8005d28d>] tracesys+0xd5/0xe0
which means that the system is deadlocked.
This patch allows autofs to hold up normal processes whilst the daemon goes
ahead and does things to the dentry tree behind the automouter point without
risking a deadlock as almost no locks are held in d_manage() and none in
d_automount().
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:26 -07:00
|
|
|
/* Allow the filesystem to manage the transit without i_mutex
|
|
|
|
* being held. */
|
|
|
|
if (managed & DCACHE_MANAGE_TRANSIT) {
|
|
|
|
BUG_ON(!path->dentry->d_op);
|
|
|
|
BUG_ON(!path->dentry->d_op->d_manage);
|
2011-03-18 07:09:02 -06:00
|
|
|
ret = path->dentry->d_op->d_manage(path->dentry, false);
|
Add a dentry op to allow processes to be held during pathwalk transit
Add a dentry op (d_manage) to permit a filesystem to hold a process and make it
sleep when it tries to transit away from one of that filesystem's directories
during a pathwalk. The operation is keyed off a new dentry flag
(DCACHE_MANAGE_TRANSIT).
The filesystem is allowed to be selective about which processes it holds and
which it permits to continue on or prohibits from transiting from each flagged
directory. This will allow autofs to hold up client processes whilst letting
its userspace daemon through to maintain the directory or the stuff behind it
or mounted upon it.
The ->d_manage() dentry operation:
int (*d_manage)(struct path *path, bool mounting_here);
takes a pointer to the directory about to be transited away from and a flag
indicating whether the transit is undertaken by do_add_mount() or
do_move_mount() skipping through a pile of filesystems mounted on a mountpoint.
It should return 0 if successful and to let the process continue on its way;
-EISDIR to prohibit the caller from skipping to overmounted filesystems or
automounting, and to use this directory; or some other error code to return to
the user.
->d_manage() is called with namespace_sem writelocked if mounting_here is true
and no other locks held, so it may sleep. However, if mounting_here is true,
it may not initiate or wait for a mount or unmount upon the parameter
directory, even if the act is actually performed by userspace.
Within fs/namei.c, follow_managed() is extended to check with d_manage() first
on each managed directory, before transiting away from it or attempting to
automount upon it.
follow_down() is renamed follow_down_one() and should only be used where the
filesystem deliberately intends to avoid management steps (e.g. autofs).
A new follow_down() is added that incorporates the loop done by all other
callers of follow_down() (do_add/move_mount(), autofs and NFSD; whilst AFS, NFS
and CIFS do use it, their use is removed by converting them to use
d_automount()). The new follow_down() calls d_manage() as appropriate. It
also takes an extra parameter to indicate if it is being called from mount code
(with namespace_sem writelocked) which it passes to d_manage(). follow_down()
ignores automount points so that it can be used to mount on them.
__follow_mount_rcu() is made to abort rcu-walk mode if it hits a directory with
DCACHE_MANAGE_TRANSIT set on the basis that we're probably going to have to
sleep. It would be possible to enter d_manage() in rcu-walk mode too, and have
that determine whether to abort or not itself. That would allow the autofs
daemon to continue on in rcu-walk mode.
Note that DCACHE_MANAGE_TRANSIT on a directory should be cleared when it isn't
required as every tranist from that directory will cause d_manage() to be
invoked. It can always be set again when necessary.
==========================
WHAT THIS MEANS FOR AUTOFS
==========================
Autofs currently uses the lookup() inode op and the d_revalidate() dentry op to
trigger the automounting of indirect mounts, and both of these can be called
with i_mutex held.
autofs knows that the i_mutex will be held by the caller in lookup(), and so
can drop it before invoking the daemon - but this isn't so for d_revalidate(),
since the lock is only held on _some_ of the code paths that call it. This
means that autofs can't risk dropping i_mutex from its d_revalidate() function
before it calls the daemon.
The bug could manifest itself as, for example, a process that's trying to
validate an automount dentry that gets made to wait because that dentry is
expired and needs cleaning up:
mkdir S ffffffff8014e05a 0 32580 24956
Call Trace:
[<ffffffff885371fd>] :autofs4:autofs4_wait+0x674/0x897
[<ffffffff80127f7d>] avc_has_perm+0x46/0x58
[<ffffffff8009fdcf>] autoremove_wake_function+0x0/0x2e
[<ffffffff88537be6>] :autofs4:autofs4_expire_wait+0x41/0x6b
[<ffffffff88535cfc>] :autofs4:autofs4_revalidate+0x91/0x149
[<ffffffff80036d96>] __lookup_hash+0xa0/0x12f
[<ffffffff80057a2f>] lookup_create+0x46/0x80
[<ffffffff800e6e31>] sys_mkdirat+0x56/0xe4
versus the automount daemon which wants to remove that dentry, but can't
because the normal process is holding the i_mutex lock:
automount D ffffffff8014e05a 0 32581 1 32561
Call Trace:
[<ffffffff80063c3f>] __mutex_lock_slowpath+0x60/0x9b
[<ffffffff8000ccf1>] do_path_lookup+0x2ca/0x2f1
[<ffffffff80063c89>] .text.lock.mutex+0xf/0x14
[<ffffffff800e6d55>] do_rmdir+0x77/0xde
[<ffffffff8005d229>] tracesys+0x71/0xe0
[<ffffffff8005d28d>] tracesys+0xd5/0xe0
which means that the system is deadlocked.
This patch allows autofs to hold up normal processes whilst the daemon goes
ahead and does things to the dentry tree behind the automouter point without
risking a deadlock as almost no locks are held in d_manage() and none in
d_automount().
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:26 -07:00
|
|
|
if (ret < 0)
|
VFS: Fix vfsmount overput on simultaneous automount
[Kudos to dhowells for tracking that crap down]
If two processes attempt to cause automounting on the same mountpoint at the
same time, the vfsmount holding the mountpoint will be left with one too few
references on it, causing a BUG when the kernel tries to clean up.
The problem is that lock_mount() drops the caller's reference to the
mountpoint's vfsmount in the case where it finds something already mounted on
the mountpoint as it transits to the mounted filesystem and replaces path->mnt
with the new mountpoint vfsmount.
During a pathwalk, however, we don't take a reference on the vfsmount if it is
the same as the one in the nameidata struct, but do_add_mount() doesn't know
this.
The fix is to make sure we have a ref on the vfsmount of the mountpoint before
calling do_add_mount(). However, if lock_mount() doesn't transit, we're then
left with an extra ref on the mountpoint vfsmount which needs releasing.
We can handle that in follow_managed() by not making assumptions about what
we can and what we cannot get from lookup_mnt() as the current code does.
The callers of follow_managed() expect that reference to path->mnt will be
grabbed iff path->mnt has been changed. follow_managed() and follow_automount()
keep track of whether such reference has been grabbed and assume that it'll
happen in those and only those cases that'll have us return with changed
path->mnt. That assumption is almost correct - it breaks in case of
racing automounts and in even harder to hit race between following a mountpoint
and a couple of mount --move. The thing is, we don't need to make that
assumption at all - after the end of loop in follow_manage() we can check
if path->mnt has ended up unchanged and do mntput() if needed.
The BUG can be reproduced with the following test program:
#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <sys/wait.h>
int main(int argc, char **argv)
{
int pid, ws;
struct stat buf;
pid = fork();
stat(argv[1], &buf);
if (pid > 0) wait(&ws);
return 0;
}
and the following procedure:
(1) Mount an NFS volume that on the server has something else mounted on a
subdirectory. For instance, I can mount / from my server:
mount warthog:/ /mnt -t nfs4 -r
On the server /data has another filesystem mounted on it, so NFS will see
a change in FSID as it walks down the path, and will mark /mnt/data as
being a mountpoint. This will cause the automount code to be triggered.
!!! Do not look inside the mounted fs at this point !!!
(2) Run the above program on a file within the submount to generate two
simultaneous automount requests:
/tmp/forkstat /mnt/data/testfile
(3) Unmount the automounted submount:
umount /mnt/data
(4) Unmount the original mount:
umount /mnt
At this point the kernel should throw a BUG with something like the
following:
BUG: Dentry ffff880032e3c5c0{i=2,n=} still in use (1) [unmount of nfs4 0:12]
Note that the bug appears on the root dentry of the original mount, not the
mountpoint and not the submount because sys_umount() hasn't got to its final
mntput_no_expire() yet, but this isn't so obvious from the call trace:
[<ffffffff8117cd82>] shrink_dcache_for_umount+0x69/0x82
[<ffffffff8116160e>] generic_shutdown_super+0x37/0x15b
[<ffffffffa00fae56>] ? nfs_super_return_all_delegations+0x2e/0x1b1 [nfs]
[<ffffffff811617f3>] kill_anon_super+0x1d/0x7e
[<ffffffffa00d0be1>] nfs4_kill_super+0x60/0xb6 [nfs]
[<ffffffff81161c17>] deactivate_locked_super+0x34/0x83
[<ffffffff811629ff>] deactivate_super+0x6f/0x7b
[<ffffffff81186261>] mntput_no_expire+0x18d/0x199
[<ffffffff811862a8>] mntput+0x3b/0x44
[<ffffffff81186d87>] release_mounts+0xa2/0xbf
[<ffffffff811876af>] sys_umount+0x47a/0x4ba
[<ffffffff8109e1ca>] ? trace_hardirqs_on_caller+0x1fd/0x22f
[<ffffffff816ea86b>] system_call_fastpath+0x16/0x1b
as do_umount() is inlined. However, you can see release_mounts() in there.
Note also that it may be necessary to have multiple CPU cores to be able to
trigger this bug.
Tested-by: Jeff Layton <jlayton@redhat.com>
Tested-by: Ian Kent <raven@themaw.net>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-06-16 08:10:06 -06:00
|
|
|
break;
|
Add a dentry op to allow processes to be held during pathwalk transit
Add a dentry op (d_manage) to permit a filesystem to hold a process and make it
sleep when it tries to transit away from one of that filesystem's directories
during a pathwalk. The operation is keyed off a new dentry flag
(DCACHE_MANAGE_TRANSIT).
The filesystem is allowed to be selective about which processes it holds and
which it permits to continue on or prohibits from transiting from each flagged
directory. This will allow autofs to hold up client processes whilst letting
its userspace daemon through to maintain the directory or the stuff behind it
or mounted upon it.
The ->d_manage() dentry operation:
int (*d_manage)(struct path *path, bool mounting_here);
takes a pointer to the directory about to be transited away from and a flag
indicating whether the transit is undertaken by do_add_mount() or
do_move_mount() skipping through a pile of filesystems mounted on a mountpoint.
It should return 0 if successful and to let the process continue on its way;
-EISDIR to prohibit the caller from skipping to overmounted filesystems or
automounting, and to use this directory; or some other error code to return to
the user.
->d_manage() is called with namespace_sem writelocked if mounting_here is true
and no other locks held, so it may sleep. However, if mounting_here is true,
it may not initiate or wait for a mount or unmount upon the parameter
directory, even if the act is actually performed by userspace.
Within fs/namei.c, follow_managed() is extended to check with d_manage() first
on each managed directory, before transiting away from it or attempting to
automount upon it.
follow_down() is renamed follow_down_one() and should only be used where the
filesystem deliberately intends to avoid management steps (e.g. autofs).
A new follow_down() is added that incorporates the loop done by all other
callers of follow_down() (do_add/move_mount(), autofs and NFSD; whilst AFS, NFS
and CIFS do use it, their use is removed by converting them to use
d_automount()). The new follow_down() calls d_manage() as appropriate. It
also takes an extra parameter to indicate if it is being called from mount code
(with namespace_sem writelocked) which it passes to d_manage(). follow_down()
ignores automount points so that it can be used to mount on them.
__follow_mount_rcu() is made to abort rcu-walk mode if it hits a directory with
DCACHE_MANAGE_TRANSIT set on the basis that we're probably going to have to
sleep. It would be possible to enter d_manage() in rcu-walk mode too, and have
that determine whether to abort or not itself. That would allow the autofs
daemon to continue on in rcu-walk mode.
Note that DCACHE_MANAGE_TRANSIT on a directory should be cleared when it isn't
required as every tranist from that directory will cause d_manage() to be
invoked. It can always be set again when necessary.
==========================
WHAT THIS MEANS FOR AUTOFS
==========================
Autofs currently uses the lookup() inode op and the d_revalidate() dentry op to
trigger the automounting of indirect mounts, and both of these can be called
with i_mutex held.
autofs knows that the i_mutex will be held by the caller in lookup(), and so
can drop it before invoking the daemon - but this isn't so for d_revalidate(),
since the lock is only held on _some_ of the code paths that call it. This
means that autofs can't risk dropping i_mutex from its d_revalidate() function
before it calls the daemon.
The bug could manifest itself as, for example, a process that's trying to
validate an automount dentry that gets made to wait because that dentry is
expired and needs cleaning up:
mkdir S ffffffff8014e05a 0 32580 24956
Call Trace:
[<ffffffff885371fd>] :autofs4:autofs4_wait+0x674/0x897
[<ffffffff80127f7d>] avc_has_perm+0x46/0x58
[<ffffffff8009fdcf>] autoremove_wake_function+0x0/0x2e
[<ffffffff88537be6>] :autofs4:autofs4_expire_wait+0x41/0x6b
[<ffffffff88535cfc>] :autofs4:autofs4_revalidate+0x91/0x149
[<ffffffff80036d96>] __lookup_hash+0xa0/0x12f
[<ffffffff80057a2f>] lookup_create+0x46/0x80
[<ffffffff800e6e31>] sys_mkdirat+0x56/0xe4
versus the automount daemon which wants to remove that dentry, but can't
because the normal process is holding the i_mutex lock:
automount D ffffffff8014e05a 0 32581 1 32561
Call Trace:
[<ffffffff80063c3f>] __mutex_lock_slowpath+0x60/0x9b
[<ffffffff8000ccf1>] do_path_lookup+0x2ca/0x2f1
[<ffffffff80063c89>] .text.lock.mutex+0xf/0x14
[<ffffffff800e6d55>] do_rmdir+0x77/0xde
[<ffffffff8005d229>] tracesys+0x71/0xe0
[<ffffffff8005d28d>] tracesys+0xd5/0xe0
which means that the system is deadlocked.
This patch allows autofs to hold up normal processes whilst the daemon goes
ahead and does things to the dentry tree behind the automouter point without
risking a deadlock as almost no locks are held in d_manage() and none in
d_automount().
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:26 -07:00
|
|
|
}
|
|
|
|
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
/* Transit to a mounted filesystem. */
|
|
|
|
if (managed & DCACHE_MOUNTED) {
|
|
|
|
struct vfsmount *mounted = lookup_mnt(path);
|
|
|
|
if (mounted) {
|
|
|
|
dput(path->dentry);
|
|
|
|
if (need_mntput)
|
|
|
|
mntput(path->mnt);
|
|
|
|
path->mnt = mounted;
|
|
|
|
path->dentry = dget(mounted->mnt_root);
|
|
|
|
need_mntput = true;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Something is mounted on this dentry in another
|
|
|
|
* namespace and/or whatever was mounted there in this
|
2013-09-29 20:06:07 -06:00
|
|
|
* namespace got unmounted before lookup_mnt() could
|
|
|
|
* get it */
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Handle an automount point */
|
|
|
|
if (managed & DCACHE_NEED_AUTOMOUNT) {
|
2015-03-22 20:37:38 -06:00
|
|
|
ret = follow_automount(path, nd, &need_mntput);
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
if (ret < 0)
|
VFS: Fix vfsmount overput on simultaneous automount
[Kudos to dhowells for tracking that crap down]
If two processes attempt to cause automounting on the same mountpoint at the
same time, the vfsmount holding the mountpoint will be left with one too few
references on it, causing a BUG when the kernel tries to clean up.
The problem is that lock_mount() drops the caller's reference to the
mountpoint's vfsmount in the case where it finds something already mounted on
the mountpoint as it transits to the mounted filesystem and replaces path->mnt
with the new mountpoint vfsmount.
During a pathwalk, however, we don't take a reference on the vfsmount if it is
the same as the one in the nameidata struct, but do_add_mount() doesn't know
this.
The fix is to make sure we have a ref on the vfsmount of the mountpoint before
calling do_add_mount(). However, if lock_mount() doesn't transit, we're then
left with an extra ref on the mountpoint vfsmount which needs releasing.
We can handle that in follow_managed() by not making assumptions about what
we can and what we cannot get from lookup_mnt() as the current code does.
The callers of follow_managed() expect that reference to path->mnt will be
grabbed iff path->mnt has been changed. follow_managed() and follow_automount()
keep track of whether such reference has been grabbed and assume that it'll
happen in those and only those cases that'll have us return with changed
path->mnt. That assumption is almost correct - it breaks in case of
racing automounts and in even harder to hit race between following a mountpoint
and a couple of mount --move. The thing is, we don't need to make that
assumption at all - after the end of loop in follow_manage() we can check
if path->mnt has ended up unchanged and do mntput() if needed.
The BUG can be reproduced with the following test program:
#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <sys/wait.h>
int main(int argc, char **argv)
{
int pid, ws;
struct stat buf;
pid = fork();
stat(argv[1], &buf);
if (pid > 0) wait(&ws);
return 0;
}
and the following procedure:
(1) Mount an NFS volume that on the server has something else mounted on a
subdirectory. For instance, I can mount / from my server:
mount warthog:/ /mnt -t nfs4 -r
On the server /data has another filesystem mounted on it, so NFS will see
a change in FSID as it walks down the path, and will mark /mnt/data as
being a mountpoint. This will cause the automount code to be triggered.
!!! Do not look inside the mounted fs at this point !!!
(2) Run the above program on a file within the submount to generate two
simultaneous automount requests:
/tmp/forkstat /mnt/data/testfile
(3) Unmount the automounted submount:
umount /mnt/data
(4) Unmount the original mount:
umount /mnt
At this point the kernel should throw a BUG with something like the
following:
BUG: Dentry ffff880032e3c5c0{i=2,n=} still in use (1) [unmount of nfs4 0:12]
Note that the bug appears on the root dentry of the original mount, not the
mountpoint and not the submount because sys_umount() hasn't got to its final
mntput_no_expire() yet, but this isn't so obvious from the call trace:
[<ffffffff8117cd82>] shrink_dcache_for_umount+0x69/0x82
[<ffffffff8116160e>] generic_shutdown_super+0x37/0x15b
[<ffffffffa00fae56>] ? nfs_super_return_all_delegations+0x2e/0x1b1 [nfs]
[<ffffffff811617f3>] kill_anon_super+0x1d/0x7e
[<ffffffffa00d0be1>] nfs4_kill_super+0x60/0xb6 [nfs]
[<ffffffff81161c17>] deactivate_locked_super+0x34/0x83
[<ffffffff811629ff>] deactivate_super+0x6f/0x7b
[<ffffffff81186261>] mntput_no_expire+0x18d/0x199
[<ffffffff811862a8>] mntput+0x3b/0x44
[<ffffffff81186d87>] release_mounts+0xa2/0xbf
[<ffffffff811876af>] sys_umount+0x47a/0x4ba
[<ffffffff8109e1ca>] ? trace_hardirqs_on_caller+0x1fd/0x22f
[<ffffffff816ea86b>] system_call_fastpath+0x16/0x1b
as do_umount() is inlined. However, you can see release_mounts() in there.
Note also that it may be necessary to have multiple CPU cores to be able to
trigger this bug.
Tested-by: Jeff Layton <jlayton@redhat.com>
Tested-by: Ian Kent <raven@themaw.net>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-06-16 08:10:06 -06:00
|
|
|
break;
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* We didn't change the current path point */
|
|
|
|
break;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
VFS: Fix vfsmount overput on simultaneous automount
[Kudos to dhowells for tracking that crap down]
If two processes attempt to cause automounting on the same mountpoint at the
same time, the vfsmount holding the mountpoint will be left with one too few
references on it, causing a BUG when the kernel tries to clean up.
The problem is that lock_mount() drops the caller's reference to the
mountpoint's vfsmount in the case where it finds something already mounted on
the mountpoint as it transits to the mounted filesystem and replaces path->mnt
with the new mountpoint vfsmount.
During a pathwalk, however, we don't take a reference on the vfsmount if it is
the same as the one in the nameidata struct, but do_add_mount() doesn't know
this.
The fix is to make sure we have a ref on the vfsmount of the mountpoint before
calling do_add_mount(). However, if lock_mount() doesn't transit, we're then
left with an extra ref on the mountpoint vfsmount which needs releasing.
We can handle that in follow_managed() by not making assumptions about what
we can and what we cannot get from lookup_mnt() as the current code does.
The callers of follow_managed() expect that reference to path->mnt will be
grabbed iff path->mnt has been changed. follow_managed() and follow_automount()
keep track of whether such reference has been grabbed and assume that it'll
happen in those and only those cases that'll have us return with changed
path->mnt. That assumption is almost correct - it breaks in case of
racing automounts and in even harder to hit race between following a mountpoint
and a couple of mount --move. The thing is, we don't need to make that
assumption at all - after the end of loop in follow_manage() we can check
if path->mnt has ended up unchanged and do mntput() if needed.
The BUG can be reproduced with the following test program:
#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <sys/wait.h>
int main(int argc, char **argv)
{
int pid, ws;
struct stat buf;
pid = fork();
stat(argv[1], &buf);
if (pid > 0) wait(&ws);
return 0;
}
and the following procedure:
(1) Mount an NFS volume that on the server has something else mounted on a
subdirectory. For instance, I can mount / from my server:
mount warthog:/ /mnt -t nfs4 -r
On the server /data has another filesystem mounted on it, so NFS will see
a change in FSID as it walks down the path, and will mark /mnt/data as
being a mountpoint. This will cause the automount code to be triggered.
!!! Do not look inside the mounted fs at this point !!!
(2) Run the above program on a file within the submount to generate two
simultaneous automount requests:
/tmp/forkstat /mnt/data/testfile
(3) Unmount the automounted submount:
umount /mnt/data
(4) Unmount the original mount:
umount /mnt
At this point the kernel should throw a BUG with something like the
following:
BUG: Dentry ffff880032e3c5c0{i=2,n=} still in use (1) [unmount of nfs4 0:12]
Note that the bug appears on the root dentry of the original mount, not the
mountpoint and not the submount because sys_umount() hasn't got to its final
mntput_no_expire() yet, but this isn't so obvious from the call trace:
[<ffffffff8117cd82>] shrink_dcache_for_umount+0x69/0x82
[<ffffffff8116160e>] generic_shutdown_super+0x37/0x15b
[<ffffffffa00fae56>] ? nfs_super_return_all_delegations+0x2e/0x1b1 [nfs]
[<ffffffff811617f3>] kill_anon_super+0x1d/0x7e
[<ffffffffa00d0be1>] nfs4_kill_super+0x60/0xb6 [nfs]
[<ffffffff81161c17>] deactivate_locked_super+0x34/0x83
[<ffffffff811629ff>] deactivate_super+0x6f/0x7b
[<ffffffff81186261>] mntput_no_expire+0x18d/0x199
[<ffffffff811862a8>] mntput+0x3b/0x44
[<ffffffff81186d87>] release_mounts+0xa2/0xbf
[<ffffffff811876af>] sys_umount+0x47a/0x4ba
[<ffffffff8109e1ca>] ? trace_hardirqs_on_caller+0x1fd/0x22f
[<ffffffff816ea86b>] system_call_fastpath+0x16/0x1b
as do_umount() is inlined. However, you can see release_mounts() in there.
Note also that it may be necessary to have multiple CPU cores to be able to
trigger this bug.
Tested-by: Jeff Layton <jlayton@redhat.com>
Tested-by: Ian Kent <raven@themaw.net>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-06-16 08:10:06 -06:00
|
|
|
|
|
|
|
if (need_mntput && path->mnt == mnt)
|
|
|
|
mntput(path->mnt);
|
2016-03-05 20:04:59 -07:00
|
|
|
if (ret == -EISDIR || !ret)
|
|
|
|
ret = 1;
|
2015-04-22 08:30:08 -06:00
|
|
|
if (need_mntput)
|
|
|
|
nd->flags |= LOOKUP_JUMPED;
|
|
|
|
if (unlikely(ret < 0))
|
|
|
|
path_put_conditional(path, nd);
|
|
|
|
return ret;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
Add a dentry op to allow processes to be held during pathwalk transit
Add a dentry op (d_manage) to permit a filesystem to hold a process and make it
sleep when it tries to transit away from one of that filesystem's directories
during a pathwalk. The operation is keyed off a new dentry flag
(DCACHE_MANAGE_TRANSIT).
The filesystem is allowed to be selective about which processes it holds and
which it permits to continue on or prohibits from transiting from each flagged
directory. This will allow autofs to hold up client processes whilst letting
its userspace daemon through to maintain the directory or the stuff behind it
or mounted upon it.
The ->d_manage() dentry operation:
int (*d_manage)(struct path *path, bool mounting_here);
takes a pointer to the directory about to be transited away from and a flag
indicating whether the transit is undertaken by do_add_mount() or
do_move_mount() skipping through a pile of filesystems mounted on a mountpoint.
It should return 0 if successful and to let the process continue on its way;
-EISDIR to prohibit the caller from skipping to overmounted filesystems or
automounting, and to use this directory; or some other error code to return to
the user.
->d_manage() is called with namespace_sem writelocked if mounting_here is true
and no other locks held, so it may sleep. However, if mounting_here is true,
it may not initiate or wait for a mount or unmount upon the parameter
directory, even if the act is actually performed by userspace.
Within fs/namei.c, follow_managed() is extended to check with d_manage() first
on each managed directory, before transiting away from it or attempting to
automount upon it.
follow_down() is renamed follow_down_one() and should only be used where the
filesystem deliberately intends to avoid management steps (e.g. autofs).
A new follow_down() is added that incorporates the loop done by all other
callers of follow_down() (do_add/move_mount(), autofs and NFSD; whilst AFS, NFS
and CIFS do use it, their use is removed by converting them to use
d_automount()). The new follow_down() calls d_manage() as appropriate. It
also takes an extra parameter to indicate if it is being called from mount code
(with namespace_sem writelocked) which it passes to d_manage(). follow_down()
ignores automount points so that it can be used to mount on them.
__follow_mount_rcu() is made to abort rcu-walk mode if it hits a directory with
DCACHE_MANAGE_TRANSIT set on the basis that we're probably going to have to
sleep. It would be possible to enter d_manage() in rcu-walk mode too, and have
that determine whether to abort or not itself. That would allow the autofs
daemon to continue on in rcu-walk mode.
Note that DCACHE_MANAGE_TRANSIT on a directory should be cleared when it isn't
required as every tranist from that directory will cause d_manage() to be
invoked. It can always be set again when necessary.
==========================
WHAT THIS MEANS FOR AUTOFS
==========================
Autofs currently uses the lookup() inode op and the d_revalidate() dentry op to
trigger the automounting of indirect mounts, and both of these can be called
with i_mutex held.
autofs knows that the i_mutex will be held by the caller in lookup(), and so
can drop it before invoking the daemon - but this isn't so for d_revalidate(),
since the lock is only held on _some_ of the code paths that call it. This
means that autofs can't risk dropping i_mutex from its d_revalidate() function
before it calls the daemon.
The bug could manifest itself as, for example, a process that's trying to
validate an automount dentry that gets made to wait because that dentry is
expired and needs cleaning up:
mkdir S ffffffff8014e05a 0 32580 24956
Call Trace:
[<ffffffff885371fd>] :autofs4:autofs4_wait+0x674/0x897
[<ffffffff80127f7d>] avc_has_perm+0x46/0x58
[<ffffffff8009fdcf>] autoremove_wake_function+0x0/0x2e
[<ffffffff88537be6>] :autofs4:autofs4_expire_wait+0x41/0x6b
[<ffffffff88535cfc>] :autofs4:autofs4_revalidate+0x91/0x149
[<ffffffff80036d96>] __lookup_hash+0xa0/0x12f
[<ffffffff80057a2f>] lookup_create+0x46/0x80
[<ffffffff800e6e31>] sys_mkdirat+0x56/0xe4
versus the automount daemon which wants to remove that dentry, but can't
because the normal process is holding the i_mutex lock:
automount D ffffffff8014e05a 0 32581 1 32561
Call Trace:
[<ffffffff80063c3f>] __mutex_lock_slowpath+0x60/0x9b
[<ffffffff8000ccf1>] do_path_lookup+0x2ca/0x2f1
[<ffffffff80063c89>] .text.lock.mutex+0xf/0x14
[<ffffffff800e6d55>] do_rmdir+0x77/0xde
[<ffffffff8005d229>] tracesys+0x71/0xe0
[<ffffffff8005d28d>] tracesys+0xd5/0xe0
which means that the system is deadlocked.
This patch allows autofs to hold up normal processes whilst the daemon goes
ahead and does things to the dentry tree behind the automouter point without
risking a deadlock as almost no locks are held in d_manage() and none in
d_automount().
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:26 -07:00
|
|
|
int follow_down_one(struct path *path)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
|
|
|
struct vfsmount *mounted;
|
|
|
|
|
2009-04-18 12:06:57 -06:00
|
|
|
mounted = lookup_mnt(path);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (mounted) {
|
2009-04-18 11:58:15 -06:00
|
|
|
dput(path->dentry);
|
|
|
|
mntput(path->mnt);
|
|
|
|
path->mnt = mounted;
|
|
|
|
path->dentry = dget(mounted->mnt_root);
|
2005-04-16 16:20:36 -06:00
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(follow_down_one);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2014-08-04 01:06:29 -06:00
|
|
|
static inline int managed_dentry_rcu(struct dentry *dentry)
|
2011-03-24 11:51:02 -06:00
|
|
|
{
|
2014-08-04 01:06:29 -06:00
|
|
|
return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
|
|
|
|
dentry->d_op->d_manage(dentry, true) : 0;
|
2011-03-24 11:51:02 -06:00
|
|
|
}
|
|
|
|
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
/*
|
2011-05-27 04:50:06 -06:00
|
|
|
* Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
|
|
|
|
* we meet a managed dentry that would need blocking.
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
*/
|
|
|
|
static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
|
2015-05-05 07:40:46 -06:00
|
|
|
struct inode **inode, unsigned *seqp)
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
{
|
2011-03-24 11:51:02 -06:00
|
|
|
for (;;) {
|
2011-11-24 16:22:03 -07:00
|
|
|
struct mount *mounted;
|
2011-03-24 11:51:02 -06:00
|
|
|
/*
|
|
|
|
* Don't forget we might have a non-mountpoint managed dentry
|
|
|
|
* that wants to block transit.
|
|
|
|
*/
|
2014-08-04 01:06:29 -06:00
|
|
|
switch (managed_dentry_rcu(path->dentry)) {
|
|
|
|
case -ECHILD:
|
|
|
|
default:
|
2011-01-14 11:46:51 -07:00
|
|
|
return false;
|
2014-08-04 01:06:29 -06:00
|
|
|
case -EISDIR:
|
|
|
|
return true;
|
|
|
|
case 0:
|
|
|
|
break;
|
|
|
|
}
|
2011-03-24 11:51:02 -06:00
|
|
|
|
|
|
|
if (!d_mountpoint(path->dentry))
|
2014-08-04 01:06:29 -06:00
|
|
|
return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
|
2011-03-24 11:51:02 -06:00
|
|
|
|
2013-10-01 14:11:26 -06:00
|
|
|
mounted = __lookup_mnt(path->mnt, path->dentry);
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
if (!mounted)
|
|
|
|
break;
|
2011-11-24 16:22:03 -07:00
|
|
|
path->mnt = &mounted->mnt;
|
|
|
|
path->dentry = mounted->mnt.mnt_root;
|
2011-11-07 14:21:26 -07:00
|
|
|
nd->flags |= LOOKUP_JUMPED;
|
2015-05-05 07:40:46 -06:00
|
|
|
*seqp = read_seqcount_begin(&path->dentry->d_seq);
|
2011-07-18 16:43:29 -06:00
|
|
|
/*
|
|
|
|
* Update the inode too. We don't need to re-check the
|
|
|
|
* dentry sequence number here after this d_inode read,
|
|
|
|
* because a mount-point is always pinned.
|
|
|
|
*/
|
|
|
|
*inode = path->dentry->d_inode;
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
}
|
2014-09-13 19:50:45 -06:00
|
|
|
return !read_seqretry(&mount_lock, nd->m_seq) &&
|
2014-08-04 01:06:29 -06:00
|
|
|
!(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
|
2011-05-27 04:50:06 -06:00
|
|
|
}
|
|
|
|
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
static int follow_dotdot_rcu(struct nameidata *nd)
|
|
|
|
{
|
2014-09-13 19:59:43 -06:00
|
|
|
struct inode *inode = nd->inode;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
while (1) {
|
2015-05-12 10:22:47 -06:00
|
|
|
if (path_equal(&nd->path, &nd->root))
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
break;
|
|
|
|
if (nd->path.dentry != nd->path.mnt->mnt_root) {
|
|
|
|
struct dentry *old = nd->path.dentry;
|
|
|
|
struct dentry *parent = old->d_parent;
|
|
|
|
unsigned seq;
|
|
|
|
|
2014-09-13 19:59:43 -06:00
|
|
|
inode = parent->d_inode;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
seq = read_seqcount_begin(&parent->d_seq);
|
2015-05-12 10:22:47 -06:00
|
|
|
if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
|
|
|
|
return -ECHILD;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
nd->path.dentry = parent;
|
|
|
|
nd->seq = seq;
|
2015-08-15 19:27:13 -06:00
|
|
|
if (unlikely(!path_connected(&nd->path)))
|
|
|
|
return -ENOENT;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
break;
|
2015-05-12 10:22:47 -06:00
|
|
|
} else {
|
|
|
|
struct mount *mnt = real_mount(nd->path.mnt);
|
|
|
|
struct mount *mparent = mnt->mnt_parent;
|
|
|
|
struct dentry *mountpoint = mnt->mnt_mountpoint;
|
|
|
|
struct inode *inode2 = mountpoint->d_inode;
|
|
|
|
unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
|
|
|
|
if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
|
|
|
|
return -ECHILD;
|
|
|
|
if (&mparent->mnt == nd->path.mnt)
|
|
|
|
break;
|
|
|
|
/* we know that mountpoint was pinned */
|
|
|
|
nd->path.dentry = mountpoint;
|
|
|
|
nd->path.mnt = &mparent->mnt;
|
|
|
|
inode = inode2;
|
|
|
|
nd->seq = seq;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
}
|
|
|
|
}
|
2015-05-12 10:22:47 -06:00
|
|
|
while (unlikely(d_mountpoint(nd->path.dentry))) {
|
2014-03-20 13:18:22 -06:00
|
|
|
struct mount *mounted;
|
|
|
|
mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
|
2015-05-12 10:22:47 -06:00
|
|
|
if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
|
|
|
|
return -ECHILD;
|
2014-03-20 13:18:22 -06:00
|
|
|
if (!mounted)
|
|
|
|
break;
|
|
|
|
nd->path.mnt = &mounted->mnt;
|
|
|
|
nd->path.dentry = mounted->mnt.mnt_root;
|
2014-09-13 19:59:43 -06:00
|
|
|
inode = nd->path.dentry->d_inode;
|
2014-03-20 13:18:22 -06:00
|
|
|
nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
|
|
|
|
}
|
2014-09-13 19:59:43 -06:00
|
|
|
nd->inode = inode;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
Add a dentry op to allow processes to be held during pathwalk transit
Add a dentry op (d_manage) to permit a filesystem to hold a process and make it
sleep when it tries to transit away from one of that filesystem's directories
during a pathwalk. The operation is keyed off a new dentry flag
(DCACHE_MANAGE_TRANSIT).
The filesystem is allowed to be selective about which processes it holds and
which it permits to continue on or prohibits from transiting from each flagged
directory. This will allow autofs to hold up client processes whilst letting
its userspace daemon through to maintain the directory or the stuff behind it
or mounted upon it.
The ->d_manage() dentry operation:
int (*d_manage)(struct path *path, bool mounting_here);
takes a pointer to the directory about to be transited away from and a flag
indicating whether the transit is undertaken by do_add_mount() or
do_move_mount() skipping through a pile of filesystems mounted on a mountpoint.
It should return 0 if successful and to let the process continue on its way;
-EISDIR to prohibit the caller from skipping to overmounted filesystems or
automounting, and to use this directory; or some other error code to return to
the user.
->d_manage() is called with namespace_sem writelocked if mounting_here is true
and no other locks held, so it may sleep. However, if mounting_here is true,
it may not initiate or wait for a mount or unmount upon the parameter
directory, even if the act is actually performed by userspace.
Within fs/namei.c, follow_managed() is extended to check with d_manage() first
on each managed directory, before transiting away from it or attempting to
automount upon it.
follow_down() is renamed follow_down_one() and should only be used where the
filesystem deliberately intends to avoid management steps (e.g. autofs).
A new follow_down() is added that incorporates the loop done by all other
callers of follow_down() (do_add/move_mount(), autofs and NFSD; whilst AFS, NFS
and CIFS do use it, their use is removed by converting them to use
d_automount()). The new follow_down() calls d_manage() as appropriate. It
also takes an extra parameter to indicate if it is being called from mount code
(with namespace_sem writelocked) which it passes to d_manage(). follow_down()
ignores automount points so that it can be used to mount on them.
__follow_mount_rcu() is made to abort rcu-walk mode if it hits a directory with
DCACHE_MANAGE_TRANSIT set on the basis that we're probably going to have to
sleep. It would be possible to enter d_manage() in rcu-walk mode too, and have
that determine whether to abort or not itself. That would allow the autofs
daemon to continue on in rcu-walk mode.
Note that DCACHE_MANAGE_TRANSIT on a directory should be cleared when it isn't
required as every tranist from that directory will cause d_manage() to be
invoked. It can always be set again when necessary.
==========================
WHAT THIS MEANS FOR AUTOFS
==========================
Autofs currently uses the lookup() inode op and the d_revalidate() dentry op to
trigger the automounting of indirect mounts, and both of these can be called
with i_mutex held.
autofs knows that the i_mutex will be held by the caller in lookup(), and so
can drop it before invoking the daemon - but this isn't so for d_revalidate(),
since the lock is only held on _some_ of the code paths that call it. This
means that autofs can't risk dropping i_mutex from its d_revalidate() function
before it calls the daemon.
The bug could manifest itself as, for example, a process that's trying to
validate an automount dentry that gets made to wait because that dentry is
expired and needs cleaning up:
mkdir S ffffffff8014e05a 0 32580 24956
Call Trace:
[<ffffffff885371fd>] :autofs4:autofs4_wait+0x674/0x897
[<ffffffff80127f7d>] avc_has_perm+0x46/0x58
[<ffffffff8009fdcf>] autoremove_wake_function+0x0/0x2e
[<ffffffff88537be6>] :autofs4:autofs4_expire_wait+0x41/0x6b
[<ffffffff88535cfc>] :autofs4:autofs4_revalidate+0x91/0x149
[<ffffffff80036d96>] __lookup_hash+0xa0/0x12f
[<ffffffff80057a2f>] lookup_create+0x46/0x80
[<ffffffff800e6e31>] sys_mkdirat+0x56/0xe4
versus the automount daemon which wants to remove that dentry, but can't
because the normal process is holding the i_mutex lock:
automount D ffffffff8014e05a 0 32581 1 32561
Call Trace:
[<ffffffff80063c3f>] __mutex_lock_slowpath+0x60/0x9b
[<ffffffff8000ccf1>] do_path_lookup+0x2ca/0x2f1
[<ffffffff80063c89>] .text.lock.mutex+0xf/0x14
[<ffffffff800e6d55>] do_rmdir+0x77/0xde
[<ffffffff8005d229>] tracesys+0x71/0xe0
[<ffffffff8005d28d>] tracesys+0xd5/0xe0
which means that the system is deadlocked.
This patch allows autofs to hold up normal processes whilst the daemon goes
ahead and does things to the dentry tree behind the automouter point without
risking a deadlock as almost no locks are held in d_manage() and none in
d_automount().
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:26 -07:00
|
|
|
/*
|
|
|
|
* Follow down to the covering mount currently visible to userspace. At each
|
|
|
|
* point, the filesystem owning that dentry may be queried as to whether the
|
|
|
|
* caller is permitted to proceed or not.
|
|
|
|
*/
|
2011-03-18 07:04:20 -06:00
|
|
|
int follow_down(struct path *path)
|
Add a dentry op to allow processes to be held during pathwalk transit
Add a dentry op (d_manage) to permit a filesystem to hold a process and make it
sleep when it tries to transit away from one of that filesystem's directories
during a pathwalk. The operation is keyed off a new dentry flag
(DCACHE_MANAGE_TRANSIT).
The filesystem is allowed to be selective about which processes it holds and
which it permits to continue on or prohibits from transiting from each flagged
directory. This will allow autofs to hold up client processes whilst letting
its userspace daemon through to maintain the directory or the stuff behind it
or mounted upon it.
The ->d_manage() dentry operation:
int (*d_manage)(struct path *path, bool mounting_here);
takes a pointer to the directory about to be transited away from and a flag
indicating whether the transit is undertaken by do_add_mount() or
do_move_mount() skipping through a pile of filesystems mounted on a mountpoint.
It should return 0 if successful and to let the process continue on its way;
-EISDIR to prohibit the caller from skipping to overmounted filesystems or
automounting, and to use this directory; or some other error code to return to
the user.
->d_manage() is called with namespace_sem writelocked if mounting_here is true
and no other locks held, so it may sleep. However, if mounting_here is true,
it may not initiate or wait for a mount or unmount upon the parameter
directory, even if the act is actually performed by userspace.
Within fs/namei.c, follow_managed() is extended to check with d_manage() first
on each managed directory, before transiting away from it or attempting to
automount upon it.
follow_down() is renamed follow_down_one() and should only be used where the
filesystem deliberately intends to avoid management steps (e.g. autofs).
A new follow_down() is added that incorporates the loop done by all other
callers of follow_down() (do_add/move_mount(), autofs and NFSD; whilst AFS, NFS
and CIFS do use it, their use is removed by converting them to use
d_automount()). The new follow_down() calls d_manage() as appropriate. It
also takes an extra parameter to indicate if it is being called from mount code
(with namespace_sem writelocked) which it passes to d_manage(). follow_down()
ignores automount points so that it can be used to mount on them.
__follow_mount_rcu() is made to abort rcu-walk mode if it hits a directory with
DCACHE_MANAGE_TRANSIT set on the basis that we're probably going to have to
sleep. It would be possible to enter d_manage() in rcu-walk mode too, and have
that determine whether to abort or not itself. That would allow the autofs
daemon to continue on in rcu-walk mode.
Note that DCACHE_MANAGE_TRANSIT on a directory should be cleared when it isn't
required as every tranist from that directory will cause d_manage() to be
invoked. It can always be set again when necessary.
==========================
WHAT THIS MEANS FOR AUTOFS
==========================
Autofs currently uses the lookup() inode op and the d_revalidate() dentry op to
trigger the automounting of indirect mounts, and both of these can be called
with i_mutex held.
autofs knows that the i_mutex will be held by the caller in lookup(), and so
can drop it before invoking the daemon - but this isn't so for d_revalidate(),
since the lock is only held on _some_ of the code paths that call it. This
means that autofs can't risk dropping i_mutex from its d_revalidate() function
before it calls the daemon.
The bug could manifest itself as, for example, a process that's trying to
validate an automount dentry that gets made to wait because that dentry is
expired and needs cleaning up:
mkdir S ffffffff8014e05a 0 32580 24956
Call Trace:
[<ffffffff885371fd>] :autofs4:autofs4_wait+0x674/0x897
[<ffffffff80127f7d>] avc_has_perm+0x46/0x58
[<ffffffff8009fdcf>] autoremove_wake_function+0x0/0x2e
[<ffffffff88537be6>] :autofs4:autofs4_expire_wait+0x41/0x6b
[<ffffffff88535cfc>] :autofs4:autofs4_revalidate+0x91/0x149
[<ffffffff80036d96>] __lookup_hash+0xa0/0x12f
[<ffffffff80057a2f>] lookup_create+0x46/0x80
[<ffffffff800e6e31>] sys_mkdirat+0x56/0xe4
versus the automount daemon which wants to remove that dentry, but can't
because the normal process is holding the i_mutex lock:
automount D ffffffff8014e05a 0 32581 1 32561
Call Trace:
[<ffffffff80063c3f>] __mutex_lock_slowpath+0x60/0x9b
[<ffffffff8000ccf1>] do_path_lookup+0x2ca/0x2f1
[<ffffffff80063c89>] .text.lock.mutex+0xf/0x14
[<ffffffff800e6d55>] do_rmdir+0x77/0xde
[<ffffffff8005d229>] tracesys+0x71/0xe0
[<ffffffff8005d28d>] tracesys+0xd5/0xe0
which means that the system is deadlocked.
This patch allows autofs to hold up normal processes whilst the daemon goes
ahead and does things to the dentry tree behind the automouter point without
risking a deadlock as almost no locks are held in d_manage() and none in
d_automount().
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:26 -07:00
|
|
|
{
|
|
|
|
unsigned managed;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
while (managed = ACCESS_ONCE(path->dentry->d_flags),
|
|
|
|
unlikely(managed & DCACHE_MANAGED_DENTRY)) {
|
|
|
|
/* Allow the filesystem to manage the transit without i_mutex
|
|
|
|
* being held.
|
|
|
|
*
|
|
|
|
* We indicate to the filesystem if someone is trying to mount
|
|
|
|
* something here. This gives autofs the chance to deny anyone
|
|
|
|
* other than its daemon the right to mount on its
|
|
|
|
* superstructure.
|
|
|
|
*
|
|
|
|
* The filesystem may sleep at this point.
|
|
|
|
*/
|
|
|
|
if (managed & DCACHE_MANAGE_TRANSIT) {
|
|
|
|
BUG_ON(!path->dentry->d_op);
|
|
|
|
BUG_ON(!path->dentry->d_op->d_manage);
|
2011-01-14 11:46:51 -07:00
|
|
|
ret = path->dentry->d_op->d_manage(
|
2011-03-18 07:09:02 -06:00
|
|
|
path->dentry, false);
|
Add a dentry op to allow processes to be held during pathwalk transit
Add a dentry op (d_manage) to permit a filesystem to hold a process and make it
sleep when it tries to transit away from one of that filesystem's directories
during a pathwalk. The operation is keyed off a new dentry flag
(DCACHE_MANAGE_TRANSIT).
The filesystem is allowed to be selective about which processes it holds and
which it permits to continue on or prohibits from transiting from each flagged
directory. This will allow autofs to hold up client processes whilst letting
its userspace daemon through to maintain the directory or the stuff behind it
or mounted upon it.
The ->d_manage() dentry operation:
int (*d_manage)(struct path *path, bool mounting_here);
takes a pointer to the directory about to be transited away from and a flag
indicating whether the transit is undertaken by do_add_mount() or
do_move_mount() skipping through a pile of filesystems mounted on a mountpoint.
It should return 0 if successful and to let the process continue on its way;
-EISDIR to prohibit the caller from skipping to overmounted filesystems or
automounting, and to use this directory; or some other error code to return to
the user.
->d_manage() is called with namespace_sem writelocked if mounting_here is true
and no other locks held, so it may sleep. However, if mounting_here is true,
it may not initiate or wait for a mount or unmount upon the parameter
directory, even if the act is actually performed by userspace.
Within fs/namei.c, follow_managed() is extended to check with d_manage() first
on each managed directory, before transiting away from it or attempting to
automount upon it.
follow_down() is renamed follow_down_one() and should only be used where the
filesystem deliberately intends to avoid management steps (e.g. autofs).
A new follow_down() is added that incorporates the loop done by all other
callers of follow_down() (do_add/move_mount(), autofs and NFSD; whilst AFS, NFS
and CIFS do use it, their use is removed by converting them to use
d_automount()). The new follow_down() calls d_manage() as appropriate. It
also takes an extra parameter to indicate if it is being called from mount code
(with namespace_sem writelocked) which it passes to d_manage(). follow_down()
ignores automount points so that it can be used to mount on them.
__follow_mount_rcu() is made to abort rcu-walk mode if it hits a directory with
DCACHE_MANAGE_TRANSIT set on the basis that we're probably going to have to
sleep. It would be possible to enter d_manage() in rcu-walk mode too, and have
that determine whether to abort or not itself. That would allow the autofs
daemon to continue on in rcu-walk mode.
Note that DCACHE_MANAGE_TRANSIT on a directory should be cleared when it isn't
required as every tranist from that directory will cause d_manage() to be
invoked. It can always be set again when necessary.
==========================
WHAT THIS MEANS FOR AUTOFS
==========================
Autofs currently uses the lookup() inode op and the d_revalidate() dentry op to
trigger the automounting of indirect mounts, and both of these can be called
with i_mutex held.
autofs knows that the i_mutex will be held by the caller in lookup(), and so
can drop it before invoking the daemon - but this isn't so for d_revalidate(),
since the lock is only held on _some_ of the code paths that call it. This
means that autofs can't risk dropping i_mutex from its d_revalidate() function
before it calls the daemon.
The bug could manifest itself as, for example, a process that's trying to
validate an automount dentry that gets made to wait because that dentry is
expired and needs cleaning up:
mkdir S ffffffff8014e05a 0 32580 24956
Call Trace:
[<ffffffff885371fd>] :autofs4:autofs4_wait+0x674/0x897
[<ffffffff80127f7d>] avc_has_perm+0x46/0x58
[<ffffffff8009fdcf>] autoremove_wake_function+0x0/0x2e
[<ffffffff88537be6>] :autofs4:autofs4_expire_wait+0x41/0x6b
[<ffffffff88535cfc>] :autofs4:autofs4_revalidate+0x91/0x149
[<ffffffff80036d96>] __lookup_hash+0xa0/0x12f
[<ffffffff80057a2f>] lookup_create+0x46/0x80
[<ffffffff800e6e31>] sys_mkdirat+0x56/0xe4
versus the automount daemon which wants to remove that dentry, but can't
because the normal process is holding the i_mutex lock:
automount D ffffffff8014e05a 0 32581 1 32561
Call Trace:
[<ffffffff80063c3f>] __mutex_lock_slowpath+0x60/0x9b
[<ffffffff8000ccf1>] do_path_lookup+0x2ca/0x2f1
[<ffffffff80063c89>] .text.lock.mutex+0xf/0x14
[<ffffffff800e6d55>] do_rmdir+0x77/0xde
[<ffffffff8005d229>] tracesys+0x71/0xe0
[<ffffffff8005d28d>] tracesys+0xd5/0xe0
which means that the system is deadlocked.
This patch allows autofs to hold up normal processes whilst the daemon goes
ahead and does things to the dentry tree behind the automouter point without
risking a deadlock as almost no locks are held in d_manage() and none in
d_automount().
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:26 -07:00
|
|
|
if (ret < 0)
|
|
|
|
return ret == -EISDIR ? 0 : ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Transit to a mounted filesystem. */
|
|
|
|
if (managed & DCACHE_MOUNTED) {
|
|
|
|
struct vfsmount *mounted = lookup_mnt(path);
|
|
|
|
if (!mounted)
|
|
|
|
break;
|
|
|
|
dput(path->dentry);
|
|
|
|
mntput(path->mnt);
|
|
|
|
path->mnt = mounted;
|
|
|
|
path->dentry = dget(mounted->mnt_root);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Don't handle automount points here */
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(follow_down);
|
Add a dentry op to allow processes to be held during pathwalk transit
Add a dentry op (d_manage) to permit a filesystem to hold a process and make it
sleep when it tries to transit away from one of that filesystem's directories
during a pathwalk. The operation is keyed off a new dentry flag
(DCACHE_MANAGE_TRANSIT).
The filesystem is allowed to be selective about which processes it holds and
which it permits to continue on or prohibits from transiting from each flagged
directory. This will allow autofs to hold up client processes whilst letting
its userspace daemon through to maintain the directory or the stuff behind it
or mounted upon it.
The ->d_manage() dentry operation:
int (*d_manage)(struct path *path, bool mounting_here);
takes a pointer to the directory about to be transited away from and a flag
indicating whether the transit is undertaken by do_add_mount() or
do_move_mount() skipping through a pile of filesystems mounted on a mountpoint.
It should return 0 if successful and to let the process continue on its way;
-EISDIR to prohibit the caller from skipping to overmounted filesystems or
automounting, and to use this directory; or some other error code to return to
the user.
->d_manage() is called with namespace_sem writelocked if mounting_here is true
and no other locks held, so it may sleep. However, if mounting_here is true,
it may not initiate or wait for a mount or unmount upon the parameter
directory, even if the act is actually performed by userspace.
Within fs/namei.c, follow_managed() is extended to check with d_manage() first
on each managed directory, before transiting away from it or attempting to
automount upon it.
follow_down() is renamed follow_down_one() and should only be used where the
filesystem deliberately intends to avoid management steps (e.g. autofs).
A new follow_down() is added that incorporates the loop done by all other
callers of follow_down() (do_add/move_mount(), autofs and NFSD; whilst AFS, NFS
and CIFS do use it, their use is removed by converting them to use
d_automount()). The new follow_down() calls d_manage() as appropriate. It
also takes an extra parameter to indicate if it is being called from mount code
(with namespace_sem writelocked) which it passes to d_manage(). follow_down()
ignores automount points so that it can be used to mount on them.
__follow_mount_rcu() is made to abort rcu-walk mode if it hits a directory with
DCACHE_MANAGE_TRANSIT set on the basis that we're probably going to have to
sleep. It would be possible to enter d_manage() in rcu-walk mode too, and have
that determine whether to abort or not itself. That would allow the autofs
daemon to continue on in rcu-walk mode.
Note that DCACHE_MANAGE_TRANSIT on a directory should be cleared when it isn't
required as every tranist from that directory will cause d_manage() to be
invoked. It can always be set again when necessary.
==========================
WHAT THIS MEANS FOR AUTOFS
==========================
Autofs currently uses the lookup() inode op and the d_revalidate() dentry op to
trigger the automounting of indirect mounts, and both of these can be called
with i_mutex held.
autofs knows that the i_mutex will be held by the caller in lookup(), and so
can drop it before invoking the daemon - but this isn't so for d_revalidate(),
since the lock is only held on _some_ of the code paths that call it. This
means that autofs can't risk dropping i_mutex from its d_revalidate() function
before it calls the daemon.
The bug could manifest itself as, for example, a process that's trying to
validate an automount dentry that gets made to wait because that dentry is
expired and needs cleaning up:
mkdir S ffffffff8014e05a 0 32580 24956
Call Trace:
[<ffffffff885371fd>] :autofs4:autofs4_wait+0x674/0x897
[<ffffffff80127f7d>] avc_has_perm+0x46/0x58
[<ffffffff8009fdcf>] autoremove_wake_function+0x0/0x2e
[<ffffffff88537be6>] :autofs4:autofs4_expire_wait+0x41/0x6b
[<ffffffff88535cfc>] :autofs4:autofs4_revalidate+0x91/0x149
[<ffffffff80036d96>] __lookup_hash+0xa0/0x12f
[<ffffffff80057a2f>] lookup_create+0x46/0x80
[<ffffffff800e6e31>] sys_mkdirat+0x56/0xe4
versus the automount daemon which wants to remove that dentry, but can't
because the normal process is holding the i_mutex lock:
automount D ffffffff8014e05a 0 32581 1 32561
Call Trace:
[<ffffffff80063c3f>] __mutex_lock_slowpath+0x60/0x9b
[<ffffffff8000ccf1>] do_path_lookup+0x2ca/0x2f1
[<ffffffff80063c89>] .text.lock.mutex+0xf/0x14
[<ffffffff800e6d55>] do_rmdir+0x77/0xde
[<ffffffff8005d229>] tracesys+0x71/0xe0
[<ffffffff8005d28d>] tracesys+0xd5/0xe0
which means that the system is deadlocked.
This patch allows autofs to hold up normal processes whilst the daemon goes
ahead and does things to the dentry tree behind the automouter point without
risking a deadlock as almost no locks are held in d_manage() and none in
d_automount().
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:26 -07:00
|
|
|
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
/*
|
|
|
|
* Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
|
|
|
|
*/
|
|
|
|
static void follow_mount(struct path *path)
|
|
|
|
{
|
|
|
|
while (d_mountpoint(path->dentry)) {
|
|
|
|
struct vfsmount *mounted = lookup_mnt(path);
|
|
|
|
if (!mounted)
|
|
|
|
break;
|
|
|
|
dput(path->dentry);
|
|
|
|
mntput(path->mnt);
|
|
|
|
path->mnt = mounted;
|
|
|
|
path->dentry = dget(mounted->mnt_root);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-08-15 19:27:13 -06:00
|
|
|
static int follow_dotdot(struct nameidata *nd)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
|
|
|
while(1) {
|
2008-02-14 20:34:32 -07:00
|
|
|
struct dentry *old = nd->path.dentry;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2009-04-07 09:49:53 -06:00
|
|
|
if (nd->path.dentry == nd->root.dentry &&
|
|
|
|
nd->path.mnt == nd->root.mnt) {
|
2005-04-16 16:20:36 -06:00
|
|
|
break;
|
|
|
|
}
|
2008-02-14 20:34:32 -07:00
|
|
|
if (nd->path.dentry != nd->path.mnt->mnt_root) {
|
2010-01-30 13:47:29 -07:00
|
|
|
/* rare case of legitimate dget_parent()... */
|
|
|
|
nd->path.dentry = dget_parent(nd->path.dentry);
|
2005-04-16 16:20:36 -06:00
|
|
|
dput(old);
|
2015-08-15 19:27:13 -06:00
|
|
|
if (unlikely(!path_connected(&nd->path)))
|
|
|
|
return -ENOENT;
|
2005-04-16 16:20:36 -06:00
|
|
|
break;
|
|
|
|
}
|
2010-01-30 13:47:29 -07:00
|
|
|
if (!follow_up(&nd->path))
|
2005-04-16 16:20:36 -06:00
|
|
|
break;
|
|
|
|
}
|
2009-04-18 11:59:41 -06:00
|
|
|
follow_mount(&nd->path);
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
nd->inode = nd->path.dentry->d_inode;
|
2015-08-15 19:27:13 -06:00
|
|
|
return 0;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2010-08-17 12:37:31 -06:00
|
|
|
/*
|
2012-03-26 04:54:24 -06:00
|
|
|
* This looks up the name in dcache, possibly revalidates the old dentry and
|
|
|
|
* allocates a new one if not found or not valid. In the need_lookup argument
|
|
|
|
* returns whether i_op->lookup is necessary.
|
|
|
|
*
|
|
|
|
* dir->d_inode->i_mutex must be held
|
2010-08-17 12:37:31 -06:00
|
|
|
*/
|
2012-03-26 04:54:24 -06:00
|
|
|
static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
|
2016-03-05 18:09:32 -07:00
|
|
|
unsigned int flags)
|
2010-08-17 12:37:31 -06:00
|
|
|
{
|
|
|
|
struct dentry *dentry;
|
2012-03-26 04:54:24 -06:00
|
|
|
int error;
|
2010-08-17 12:37:31 -06:00
|
|
|
|
2012-03-26 04:54:24 -06:00
|
|
|
dentry = d_lookup(dir, name);
|
|
|
|
if (dentry) {
|
2012-11-28 09:30:53 -07:00
|
|
|
if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
|
2012-06-22 02:42:10 -06:00
|
|
|
error = d_revalidate(dentry, flags);
|
2012-03-26 04:54:24 -06:00
|
|
|
if (unlikely(error <= 0)) {
|
|
|
|
if (error < 0) {
|
|
|
|
dput(dentry);
|
|
|
|
return ERR_PTR(error);
|
2014-02-13 10:46:25 -07:00
|
|
|
} else {
|
|
|
|
d_invalidate(dentry);
|
2012-03-26 04:54:24 -06:00
|
|
|
dput(dentry);
|
|
|
|
dentry = NULL;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2010-08-17 12:37:31 -06:00
|
|
|
return dentry;
|
|
|
|
}
|
|
|
|
|
2011-05-31 09:58:49 -06:00
|
|
|
/*
|
2013-10-23 14:09:16 -06:00
|
|
|
* Call i_op->lookup on the dentry. The dentry must be negative and
|
|
|
|
* unhashed.
|
2012-03-26 04:54:24 -06:00
|
|
|
*
|
|
|
|
* dir->d_inode->i_mutex must be held
|
2011-05-31 09:58:49 -06:00
|
|
|
*/
|
2012-03-26 04:54:24 -06:00
|
|
|
static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
|
2012-06-10 15:17:17 -06:00
|
|
|
unsigned int flags)
|
2011-05-31 09:58:49 -06:00
|
|
|
{
|
|
|
|
struct dentry *old;
|
|
|
|
|
|
|
|
/* Don't create child dentry for a dead directory. */
|
2012-03-26 04:54:24 -06:00
|
|
|
if (unlikely(IS_DEADDIR(dir))) {
|
2012-02-03 06:25:18 -07:00
|
|
|
dput(dentry);
|
2011-05-31 09:58:49 -06:00
|
|
|
return ERR_PTR(-ENOENT);
|
2012-02-03 06:25:18 -07:00
|
|
|
}
|
2011-05-31 09:58:49 -06:00
|
|
|
|
2012-06-10 15:17:17 -06:00
|
|
|
old = dir->i_op->lookup(dir, dentry, flags);
|
2011-05-31 09:58:49 -06:00
|
|
|
if (unlikely(old)) {
|
|
|
|
dput(dentry);
|
|
|
|
dentry = old;
|
|
|
|
}
|
|
|
|
return dentry;
|
|
|
|
}
|
|
|
|
|
2012-03-30 12:41:51 -06:00
|
|
|
static struct dentry *__lookup_hash(struct qstr *name,
|
2012-06-10 15:17:17 -06:00
|
|
|
struct dentry *base, unsigned int flags)
|
2012-03-30 12:41:51 -06:00
|
|
|
{
|
2016-03-05 18:09:32 -07:00
|
|
|
struct dentry *dentry = lookup_dcache(name, base, flags);
|
2012-03-30 12:41:51 -06:00
|
|
|
|
2016-03-05 18:09:32 -07:00
|
|
|
if (dentry)
|
2012-03-26 04:54:24 -06:00
|
|
|
return dentry;
|
2012-03-30 12:41:51 -06:00
|
|
|
|
2016-03-05 18:09:32 -07:00
|
|
|
dentry = d_alloc(base, name);
|
|
|
|
if (unlikely(!dentry))
|
|
|
|
return ERR_PTR(-ENOMEM);
|
|
|
|
|
2012-06-10 15:17:17 -06:00
|
|
|
return lookup_real(base->d_inode, dentry, flags);
|
2012-03-30 12:41:51 -06:00
|
|
|
}
|
|
|
|
|
2013-01-24 16:16:00 -07:00
|
|
|
static int lookup_fast(struct nameidata *nd,
|
2015-05-05 07:40:46 -06:00
|
|
|
struct path *path, struct inode **inode,
|
|
|
|
unsigned *seqp)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2008-02-14 20:34:32 -07:00
|
|
|
struct vfsmount *mnt = nd->path.mnt;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
struct dentry *dentry, *parent = nd->path.dentry;
|
untangle do_lookup()
That thing has devolved into rats nest of gotos; sane use of unlikely()
gets rid of that horror and gives much more readable structure:
* make a fast attempt to find a dentry; false negatives are OK.
In RCU mode if everything went fine, we are done, otherwise just drop
out of RCU. If we'd done (RCU) ->d_revalidate() and it had not refused
outright (i.e. didn't give us -ECHILD), remember its result.
* now we are not in RCU mode and hopefully have a dentry. If we
do not, lock parent, do full d_lookup() and if that has not found anything,
allocate and call ->lookup(). If we'd done that ->lookup(), remember that
dentry is good and we don't need to revalidate it.
* now we have a dentry. If it has ->d_revalidate() and we can't
skip it, call it.
* hopefully dentry is good; if not, either fail (in case of error)
or try to invalidate it. If d_invalidate() has succeeded, drop it and
retry everything as if original attempt had not found a dentry.
* now we can finish it up - deal with mountpoint crossing and
automount.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-03-11 02:44:53 -07:00
|
|
|
int status = 1;
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
int err;
|
|
|
|
|
fs: remove extra lookup in __lookup_hash
fs: remove extra lookup in __lookup_hash
Optimize lookup for create operations, where no dentry should often be
common-case. In cases where it is not, such as unlink, the added overhead
is much smaller than the removed.
Also, move comments about __d_lookup racyness to the __d_lookup call site.
d_lookup is intuitive; __d_lookup is what needs commenting. So in that same
vein, add kerneldoc comments to __d_lookup and clean up some of the comments:
- We are interested in how the RCU lookup works here, particularly with
renames. Make that explicit, and point to the document where it is explained
in more detail.
- RCU is pretty standard now, and macros make implementations pretty mindless.
If we want to know about RCU barrier details, we look in RCU code.
- Delete some boring legacy comments because we don't care much about how the
code used to work, more about the interesting parts of how it works now. So
comments about lazy LRU may be interesting, but would better be done in the
LRU or refcount management code.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2010-08-17 12:37:34 -06:00
|
|
|
/*
|
|
|
|
* Rename seqlock is not required here because in the off chance
|
2016-03-05 19:32:53 -07:00
|
|
|
* of a false negative due to a concurrent rename, the caller is
|
|
|
|
* going to fall back to non-racy lookup.
|
fs: remove extra lookup in __lookup_hash
fs: remove extra lookup in __lookup_hash
Optimize lookup for create operations, where no dentry should often be
common-case. In cases where it is not, such as unlink, the added overhead
is much smaller than the removed.
Also, move comments about __d_lookup racyness to the __d_lookup call site.
d_lookup is intuitive; __d_lookup is what needs commenting. So in that same
vein, add kerneldoc comments to __d_lookup and clean up some of the comments:
- We are interested in how the RCU lookup works here, particularly with
renames. Make that explicit, and point to the document where it is explained
in more detail.
- RCU is pretty standard now, and macros make implementations pretty mindless.
If we want to know about RCU barrier details, we look in RCU code.
- Delete some boring legacy comments because we don't care much about how the
code used to work, more about the interesting parts of how it works now. So
comments about lazy LRU may be interesting, but would better be done in the
LRU or refcount management code.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2010-08-17 12:37:34 -06:00
|
|
|
*/
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
if (nd->flags & LOOKUP_RCU) {
|
|
|
|
unsigned seq;
|
2015-05-07 17:24:57 -06:00
|
|
|
bool negative;
|
2013-05-21 16:22:44 -06:00
|
|
|
dentry = __d_lookup_rcu(parent, &nd->last, &seq);
|
2016-03-05 19:32:53 -07:00
|
|
|
if (unlikely(!dentry)) {
|
|
|
|
if (unlazy_walk(nd, NULL, 0))
|
|
|
|
return -ECHILD;
|
2016-03-05 20:04:59 -07:00
|
|
|
return 0;
|
2016-03-05 19:32:53 -07:00
|
|
|
}
|
untangle do_lookup()
That thing has devolved into rats nest of gotos; sane use of unlikely()
gets rid of that horror and gives much more readable structure:
* make a fast attempt to find a dentry; false negatives are OK.
In RCU mode if everything went fine, we are done, otherwise just drop
out of RCU. If we'd done (RCU) ->d_revalidate() and it had not refused
outright (i.e. didn't give us -ECHILD), remember its result.
* now we are not in RCU mode and hopefully have a dentry. If we
do not, lock parent, do full d_lookup() and if that has not found anything,
allocate and call ->lookup(). If we'd done that ->lookup(), remember that
dentry is good and we don't need to revalidate it.
* now we have a dentry. If it has ->d_revalidate() and we can't
skip it, call it.
* hopefully dentry is good; if not, either fail (in case of error)
or try to invalidate it. If d_invalidate() has succeeded, drop it and
retry everything as if original attempt had not found a dentry.
* now we can finish it up - deal with mountpoint crossing and
automount.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-03-11 02:44:53 -07:00
|
|
|
|
vfs: clean up __d_lookup_rcu() and dentry_cmp() interfaces
The calling conventions for __d_lookup_rcu() and dentry_cmp() are
annoying in different ways, and there is actually one single underlying
reason for both of the annoyances.
The fundamental reason is that we do the returned dentry sequence number
check inside __d_lookup_rcu() instead of doing it in the caller. This
results in two annoyances:
- __d_lookup_rcu() now not only needs to return the dentry and the
sequence number that goes along with the lookup, it also needs to
return the inode pointer that was validated by that sequence number
check.
- and because we did the sequence number check early (to validate the
name pointer and length) we also couldn't just pass the dentry itself
to dentry_cmp(), we had to pass the counted string that contained the
name.
So that sequence number decision caused two separate ugly calling
conventions.
Both of these problems would be solved if we just did the sequence
number check in the caller instead. There's only one caller, and that
caller already has to do the sequence number check for the parent
anyway, so just do that.
That allows us to stop returning the dentry->d_inode in that in-out
argument (pointer-to-pointer-to-inode), so we can make the inode
argument just a regular input inode pointer. The caller can just load
the inode from dentry->d_inode, and then do the sequence number check
after that to make sure that it's synchronized with the name we looked
up.
And it allows us to just pass in the dentry to dentry_cmp(), which is
what all the callers really wanted. Sure, dentry_cmp() has to be a bit
careful about the dentry (which is not stable during RCU lookup), but
that's actually very simple.
And now that dentry_cmp() can clearly see that the first string argument
is a dentry, we can use the direct word access for that, instead of the
careful unaligned zero-padding. The dentry name is always properly
aligned, since it is a single path component that is either embedded
into the dentry itself, or was allocated with kmalloc() (see __d_alloc).
Finally, this also uninlines the nasty slow-case for dentry comparisons:
that one *does* need to do a sequence number check, since it will call
in to the low-level filesystems, and we want to give those a stable
inode pointer and path component length/start arguments. Doing an extra
sequence check for that slow case is not a problem, though.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-04 15:59:14 -06:00
|
|
|
/*
|
|
|
|
* This sequence count validates that the inode matches
|
|
|
|
* the dentry name information from lookup.
|
|
|
|
*/
|
2015-05-06 08:59:00 -06:00
|
|
|
*inode = d_backing_inode(dentry);
|
2015-05-07 17:24:57 -06:00
|
|
|
negative = d_is_negative(dentry);
|
2016-03-05 19:32:53 -07:00
|
|
|
if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
|
vfs: clean up __d_lookup_rcu() and dentry_cmp() interfaces
The calling conventions for __d_lookup_rcu() and dentry_cmp() are
annoying in different ways, and there is actually one single underlying
reason for both of the annoyances.
The fundamental reason is that we do the returned dentry sequence number
check inside __d_lookup_rcu() instead of doing it in the caller. This
results in two annoyances:
- __d_lookup_rcu() now not only needs to return the dentry and the
sequence number that goes along with the lookup, it also needs to
return the inode pointer that was validated by that sequence number
check.
- and because we did the sequence number check early (to validate the
name pointer and length) we also couldn't just pass the dentry itself
to dentry_cmp(), we had to pass the counted string that contained the
name.
So that sequence number decision caused two separate ugly calling
conventions.
Both of these problems would be solved if we just did the sequence
number check in the caller instead. There's only one caller, and that
caller already has to do the sequence number check for the parent
anyway, so just do that.
That allows us to stop returning the dentry->d_inode in that in-out
argument (pointer-to-pointer-to-inode), so we can make the inode
argument just a regular input inode pointer. The caller can just load
the inode from dentry->d_inode, and then do the sequence number check
after that to make sure that it's synchronized with the name we looked
up.
And it allows us to just pass in the dentry to dentry_cmp(), which is
what all the callers really wanted. Sure, dentry_cmp() has to be a bit
careful about the dentry (which is not stable during RCU lookup), but
that's actually very simple.
And now that dentry_cmp() can clearly see that the first string argument
is a dentry, we can use the direct word access for that, instead of the
careful unaligned zero-padding. The dentry name is always properly
aligned, since it is a single path component that is either embedded
into the dentry itself, or was allocated with kmalloc() (see __d_alloc).
Finally, this also uninlines the nasty slow-case for dentry comparisons:
that one *does* need to do a sequence number check, since it will call
in to the low-level filesystems, and we want to give those a stable
inode pointer and path component length/start arguments. Doing an extra
sequence check for that slow case is not a problem, though.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-04 15:59:14 -06:00
|
|
|
return -ECHILD;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This sequence count validates that the parent had no
|
|
|
|
* changes while we did the lookup of the dentry above.
|
|
|
|
*
|
|
|
|
* The memory barrier in read_seqcount_begin of child is
|
|
|
|
* enough, we can use __read_seqcount_retry here.
|
|
|
|
*/
|
2016-03-05 19:32:53 -07:00
|
|
|
if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
return -ECHILD;
|
untangle do_lookup()
That thing has devolved into rats nest of gotos; sane use of unlikely()
gets rid of that horror and gives much more readable structure:
* make a fast attempt to find a dentry; false negatives are OK.
In RCU mode if everything went fine, we are done, otherwise just drop
out of RCU. If we'd done (RCU) ->d_revalidate() and it had not refused
outright (i.e. didn't give us -ECHILD), remember its result.
* now we are not in RCU mode and hopefully have a dentry. If we
do not, lock parent, do full d_lookup() and if that has not found anything,
allocate and call ->lookup(). If we'd done that ->lookup(), remember that
dentry is good and we don't need to revalidate it.
* now we have a dentry. If it has ->d_revalidate() and we can't
skip it, call it.
* hopefully dentry is good; if not, either fail (in case of error)
or try to invalidate it. If d_invalidate() has succeeded, drop it and
retry everything as if original attempt had not found a dentry.
* now we can finish it up - deal with mountpoint crossing and
automount.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-03-11 02:44:53 -07:00
|
|
|
|
2015-05-05 07:40:46 -06:00
|
|
|
*seqp = seq;
|
2016-03-05 19:32:53 -07:00
|
|
|
if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
|
2012-06-10 14:10:59 -06:00
|
|
|
status = d_revalidate(dentry, nd->flags);
|
2016-03-05 19:32:53 -07:00
|
|
|
if (unlikely(status <= 0)) {
|
|
|
|
if (unlazy_walk(nd, dentry, seq))
|
|
|
|
return -ECHILD;
|
|
|
|
if (status == -ECHILD)
|
|
|
|
status = d_revalidate(dentry, nd->flags);
|
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* Note: do negative dentry check after revalidation in
|
|
|
|
* case that drops it.
|
|
|
|
*/
|
|
|
|
if (unlikely(negative))
|
|
|
|
return -ENOENT;
|
|
|
|
path->mnt = mnt;
|
|
|
|
path->dentry = dentry;
|
|
|
|
if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
|
2016-03-05 20:04:59 -07:00
|
|
|
return 1;
|
2016-03-05 19:32:53 -07:00
|
|
|
if (unlazy_walk(nd, dentry, seq))
|
|
|
|
return -ECHILD;
|
2011-02-14 23:26:22 -07:00
|
|
|
}
|
untangle do_lookup()
That thing has devolved into rats nest of gotos; sane use of unlikely()
gets rid of that horror and gives much more readable structure:
* make a fast attempt to find a dentry; false negatives are OK.
In RCU mode if everything went fine, we are done, otherwise just drop
out of RCU. If we'd done (RCU) ->d_revalidate() and it had not refused
outright (i.e. didn't give us -ECHILD), remember its result.
* now we are not in RCU mode and hopefully have a dentry. If we
do not, lock parent, do full d_lookup() and if that has not found anything,
allocate and call ->lookup(). If we'd done that ->lookup(), remember that
dentry is good and we don't need to revalidate it.
* now we have a dentry. If it has ->d_revalidate() and we can't
skip it, call it.
* hopefully dentry is good; if not, either fail (in case of error)
or try to invalidate it. If d_invalidate() has succeeded, drop it and
retry everything as if original attempt had not found a dentry.
* now we can finish it up - deal with mountpoint crossing and
automount.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-03-11 02:44:53 -07:00
|
|
|
} else {
|
2013-01-24 16:16:00 -07:00
|
|
|
dentry = __d_lookup(parent, &nd->last);
|
2016-03-05 19:32:53 -07:00
|
|
|
if (unlikely(!dentry))
|
2016-03-05 20:04:59 -07:00
|
|
|
return 0;
|
2016-03-05 19:32:53 -07:00
|
|
|
if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
|
|
|
|
status = d_revalidate(dentry, nd->flags);
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
}
|
untangle do_lookup()
That thing has devolved into rats nest of gotos; sane use of unlikely()
gets rid of that horror and gives much more readable structure:
* make a fast attempt to find a dentry; false negatives are OK.
In RCU mode if everything went fine, we are done, otherwise just drop
out of RCU. If we'd done (RCU) ->d_revalidate() and it had not refused
outright (i.e. didn't give us -ECHILD), remember its result.
* now we are not in RCU mode and hopefully have a dentry. If we
do not, lock parent, do full d_lookup() and if that has not found anything,
allocate and call ->lookup(). If we'd done that ->lookup(), remember that
dentry is good and we don't need to revalidate it.
* now we have a dentry. If it has ->d_revalidate() and we can't
skip it, call it.
* hopefully dentry is good; if not, either fail (in case of error)
or try to invalidate it. If d_invalidate() has succeeded, drop it and
retry everything as if original attempt had not found a dentry.
* now we can finish it up - deal with mountpoint crossing and
automount.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-03-11 02:44:53 -07:00
|
|
|
if (unlikely(status <= 0)) {
|
2016-03-05 20:04:59 -07:00
|
|
|
if (!status)
|
2016-03-05 19:32:53 -07:00
|
|
|
d_invalidate(dentry);
|
2014-02-13 10:46:25 -07:00
|
|
|
dput(dentry);
|
2016-03-05 19:32:53 -07:00
|
|
|
return status;
|
2011-02-14 23:26:22 -07:00
|
|
|
}
|
2015-05-07 17:24:57 -06:00
|
|
|
if (unlikely(d_is_negative(dentry))) {
|
|
|
|
dput(dentry);
|
|
|
|
return -ENOENT;
|
|
|
|
}
|
2016-03-05 19:32:53 -07:00
|
|
|
|
Add a dentry op to handle automounting rather than abusing follow_link()
Add a dentry op (d_automount) to handle automounting directories rather than
abusing the follow_link() inode operation. The operation is keyed off a new
dentry flag (DCACHE_NEED_AUTOMOUNT).
This also makes it easier to add an AT_ flag to suppress terminal segment
automount during pathwalk and removes the need for the kludge code in the
pathwalk algorithm to handle directories with follow_link() semantics.
The ->d_automount() dentry operation:
struct vfsmount *(*d_automount)(struct path *mountpoint);
takes a pointer to the directory to be mounted upon, which is expected to
provide sufficient data to determine what should be mounted. If successful, it
should return the vfsmount struct it creates (which it should also have added
to the namespace using do_add_mount() or similar). If there's a collision with
another automount attempt, NULL should be returned. If the directory specified
by the parameter should be used directly rather than being mounted upon,
-EISDIR should be returned. In any other case, an error code should be
returned.
The ->d_automount() operation is called with no locks held and may sleep. At
this point the pathwalk algorithm will be in ref-walk mode.
Within fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is
added to handle mountpoints. It will return -EREMOTE if the automount flag was
set, but no d_automount() op was supplied, -ELOOP if we've encountered too many
symlinks or mountpoints, -EISDIR if the walk point should be used without
mounting and 0 if successful. The path will be updated to point to the mounted
filesystem if a successful automount took place.
__follow_mount() is replaced by follow_managed() which is more generic
(especially with the patch that adds ->d_manage()). This handles transits from
directories during pathwalk, including automounting and skipping over
mountpoints (and holding processes with the next patch).
__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an
automount point with nothing mounted on it.
follow_dotdot*() does not handle automounts as you don't want to trigger them
whilst following "..".
I've also extracted the mount/don't-mount logic from autofs4 and included it
here. It makes the mount go ahead anyway if someone calls open() or creat(),
tries to traverse the directory, tries to chdir/chroot/etc. into the directory,
or sticks a '/' on the end of the pathname. If they do a stat(), however,
they'll only trigger the automount if they didn't also say O_NOFOLLOW.
I've also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their
inodes as automount points. This flag is automatically propagated to the
dentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could
save AFS a private flag bit apiece, but is not strictly necessary. It would be
preferable to do the propagation in d_set_d_op(), but that doesn't normally
have access to the inode.
[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()
succeeds in RCU case of do_lookup(); we need to fall through to non-RCU case after
that, rather than just returning with ungrabbed *path]
Signed-off-by: David Howells <dhowells@redhat.com>
Was-Acked-by: Ian Kent <raven@themaw.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-14 11:45:21 -07:00
|
|
|
path->mnt = mnt;
|
|
|
|
path->dentry = dentry;
|
2015-03-22 20:37:38 -06:00
|
|
|
err = follow_managed(path, nd);
|
2016-03-05 20:04:59 -07:00
|
|
|
if (likely(err > 0))
|
2015-05-06 08:59:00 -06:00
|
|
|
*inode = d_backing_inode(path->dentry);
|
2015-04-22 08:30:08 -06:00
|
|
|
return err;
|
2012-05-21 09:30:05 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Fast lookup failed, do it the slow way */
|
2013-01-24 16:19:49 -07:00
|
|
|
static int lookup_slow(struct nameidata *nd, struct path *path)
|
2012-05-21 09:30:05 -06:00
|
|
|
{
|
|
|
|
struct dentry *dentry, *parent;
|
|
|
|
|
|
|
|
parent = nd->path.dentry;
|
2012-03-30 12:48:04 -06:00
|
|
|
BUG_ON(nd->inode != parent->d_inode);
|
|
|
|
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_lock(parent->d_inode);
|
2013-01-24 16:19:49 -07:00
|
|
|
dentry = __lookup_hash(&nd->last, parent, nd->flags);
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_unlock(parent->d_inode);
|
2012-03-30 12:48:04 -06:00
|
|
|
if (IS_ERR(dentry))
|
|
|
|
return PTR_ERR(dentry);
|
2012-05-21 09:30:05 -06:00
|
|
|
path->mnt = nd->path.mnt;
|
|
|
|
path->dentry = dentry;
|
2015-04-22 08:30:08 -06:00
|
|
|
return follow_managed(path, nd);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2011-02-21 19:34:47 -07:00
|
|
|
static inline int may_lookup(struct nameidata *nd)
|
|
|
|
{
|
|
|
|
if (nd->flags & LOOKUP_RCU) {
|
2011-06-20 17:57:03 -06:00
|
|
|
int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
|
2011-02-21 19:34:47 -07:00
|
|
|
if (err != -ECHILD)
|
|
|
|
return err;
|
2015-05-05 07:26:05 -06:00
|
|
|
if (unlazy_walk(nd, NULL, 0))
|
2011-02-21 19:34:47 -07:00
|
|
|
return -ECHILD;
|
|
|
|
}
|
2011-06-20 17:57:03 -06:00
|
|
|
return inode_permission(nd->inode, MAY_EXEC);
|
2011-02-21 19:34:47 -07:00
|
|
|
}
|
|
|
|
|
2011-03-04 12:22:06 -07:00
|
|
|
static inline int handle_dots(struct nameidata *nd, int type)
|
|
|
|
{
|
|
|
|
if (type == LAST_DOTDOT) {
|
2015-12-05 18:07:21 -07:00
|
|
|
if (!nd->root.mnt)
|
|
|
|
set_root(nd);
|
2011-03-04 12:22:06 -07:00
|
|
|
if (nd->flags & LOOKUP_RCU) {
|
2015-05-04 05:53:00 -06:00
|
|
|
return follow_dotdot_rcu(nd);
|
2011-03-04 12:22:06 -07:00
|
|
|
} else
|
2015-08-15 19:27:13 -06:00
|
|
|
return follow_dotdot(nd);
|
2011-03-04 12:22:06 -07:00
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2015-05-07 17:54:34 -06:00
|
|
|
static int pick_link(struct nameidata *nd, struct path *link,
|
|
|
|
struct inode *inode, unsigned seq)
|
2015-05-04 16:13:23 -06:00
|
|
|
{
|
2015-05-04 16:26:59 -06:00
|
|
|
int error;
|
2015-05-06 14:01:56 -06:00
|
|
|
struct saved *last;
|
2015-03-22 20:37:38 -06:00
|
|
|
if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
|
2015-05-04 16:26:59 -06:00
|
|
|
path_to_nameidata(link, nd);
|
|
|
|
return -ELOOP;
|
|
|
|
}
|
2015-05-09 11:04:24 -06:00
|
|
|
if (!(nd->flags & LOOKUP_RCU)) {
|
2015-05-09 10:55:43 -06:00
|
|
|
if (link->mnt == nd->path.mnt)
|
|
|
|
mntget(link->mnt);
|
2015-05-04 16:13:23 -06:00
|
|
|
}
|
2015-05-04 16:26:59 -06:00
|
|
|
error = nd_alloc_stack(nd);
|
|
|
|
if (unlikely(error)) {
|
2015-05-09 11:04:24 -06:00
|
|
|
if (error == -ECHILD) {
|
|
|
|
if (unlikely(unlazy_link(nd, link, seq)))
|
|
|
|
return -ECHILD;
|
|
|
|
error = nd_alloc_stack(nd);
|
|
|
|
}
|
|
|
|
if (error) {
|
|
|
|
path_put(link);
|
|
|
|
return error;
|
|
|
|
}
|
2015-05-04 16:26:59 -06:00
|
|
|
}
|
|
|
|
|
2015-05-10 09:50:01 -06:00
|
|
|
last = nd->stack + nd->depth++;
|
2015-05-06 14:01:56 -06:00
|
|
|
last->link = *link;
|
2015-12-29 13:58:39 -07:00
|
|
|
clear_delayed_call(&last->done);
|
|
|
|
nd->link_inode = inode;
|
2015-05-08 11:23:53 -06:00
|
|
|
last->seq = seq;
|
2015-05-04 16:13:23 -06:00
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
2011-08-06 23:45:50 -06:00
|
|
|
/*
|
|
|
|
* Do we need to follow links? We _really_ want to be able
|
|
|
|
* to do this check without having to look at inode->i_op,
|
|
|
|
* so we keep a cache of "no, this doesn't need follow_link"
|
|
|
|
* for the common case.
|
|
|
|
*/
|
2015-05-05 07:40:46 -06:00
|
|
|
static inline int should_follow_link(struct nameidata *nd, struct path *link,
|
2015-05-07 17:54:34 -06:00
|
|
|
int follow,
|
|
|
|
struct inode *inode, unsigned seq)
|
2011-08-06 23:45:50 -06:00
|
|
|
{
|
2015-05-04 16:13:23 -06:00
|
|
|
if (likely(!d_is_symlink(link->dentry)))
|
|
|
|
return 0;
|
|
|
|
if (!follow)
|
|
|
|
return 0;
|
2016-02-27 17:31:01 -07:00
|
|
|
/* make sure that d_is_symlink above matches inode */
|
|
|
|
if (nd->flags & LOOKUP_RCU) {
|
|
|
|
if (read_seqcount_retry(&link->dentry->d_seq, seq))
|
|
|
|
return -ECHILD;
|
|
|
|
}
|
2015-05-07 17:54:34 -06:00
|
|
|
return pick_link(nd, link, inode, seq);
|
2011-08-06 23:45:50 -06:00
|
|
|
}
|
|
|
|
|
2015-05-04 15:47:11 -06:00
|
|
|
enum {WALK_GET = 1, WALK_PUT = 2};
|
|
|
|
|
|
|
|
static int walk_component(struct nameidata *nd, int flags)
|
2011-03-13 17:58:58 -06:00
|
|
|
{
|
2015-04-22 15:52:47 -06:00
|
|
|
struct path path;
|
2011-03-13 17:58:58 -06:00
|
|
|
struct inode *inode;
|
2015-05-05 07:40:46 -06:00
|
|
|
unsigned seq;
|
2011-03-13 17:58:58 -06:00
|
|
|
int err;
|
|
|
|
/*
|
|
|
|
* "." and ".." are special - ".." especially so because it has
|
|
|
|
* to be able to know about the current root directory and
|
|
|
|
* parent relationships.
|
|
|
|
*/
|
2015-05-04 15:47:11 -06:00
|
|
|
if (unlikely(nd->last_type != LAST_NORM)) {
|
|
|
|
err = handle_dots(nd, nd->last_type);
|
|
|
|
if (flags & WALK_PUT)
|
|
|
|
put_link(nd);
|
|
|
|
return err;
|
|
|
|
}
|
2015-05-05 07:40:46 -06:00
|
|
|
err = lookup_fast(nd, &path, &inode, &seq);
|
2016-03-05 20:04:59 -07:00
|
|
|
if (unlikely(err <= 0)) {
|
2012-05-21 09:30:05 -06:00
|
|
|
if (err < 0)
|
2015-05-04 05:59:30 -06:00
|
|
|
return err;
|
2012-05-21 09:30:05 -06:00
|
|
|
|
2015-04-22 15:52:47 -06:00
|
|
|
err = lookup_slow(nd, &path);
|
2012-05-21 09:30:05 -06:00
|
|
|
if (err < 0)
|
2015-05-04 05:59:30 -06:00
|
|
|
return err;
|
2012-05-21 09:30:05 -06:00
|
|
|
|
2015-05-05 07:40:46 -06:00
|
|
|
seq = 0; /* we are already out of RCU mode */
|
2015-05-07 17:24:57 -06:00
|
|
|
err = -ENOENT;
|
2015-04-22 15:52:47 -06:00
|
|
|
if (d_is_negative(path.dentry))
|
2015-05-07 17:24:57 -06:00
|
|
|
goto out_path_put;
|
2016-02-27 17:23:16 -07:00
|
|
|
inode = d_backing_inode(path.dentry);
|
2011-03-13 17:58:58 -06:00
|
|
|
}
|
2012-05-21 09:30:05 -06:00
|
|
|
|
2015-05-04 15:47:11 -06:00
|
|
|
if (flags & WALK_PUT)
|
|
|
|
put_link(nd);
|
2015-05-07 17:54:34 -06:00
|
|
|
err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
|
2015-05-04 16:13:23 -06:00
|
|
|
if (unlikely(err))
|
|
|
|
return err;
|
2015-04-22 15:52:47 -06:00
|
|
|
path_to_nameidata(&path, nd);
|
2011-03-13 17:58:58 -06:00
|
|
|
nd->inode = inode;
|
2015-05-05 07:40:46 -06:00
|
|
|
nd->seq = seq;
|
2011-03-13 17:58:58 -06:00
|
|
|
return 0;
|
2012-05-21 09:30:05 -06:00
|
|
|
|
|
|
|
out_path_put:
|
2015-04-22 15:52:47 -06:00
|
|
|
path_to_nameidata(&path, nd);
|
2012-05-21 09:30:05 -06:00
|
|
|
return err;
|
2011-03-13 17:58:58 -06:00
|
|
|
}
|
|
|
|
|
2012-03-06 12:16:17 -07:00
|
|
|
/*
|
|
|
|
* We can do the critical dentry name comparison and hashing
|
|
|
|
* operations one word at a time, but we are limited to:
|
|
|
|
*
|
|
|
|
* - Architectures with fast unaligned word accesses. We could
|
|
|
|
* do a "get_unaligned()" if this helps and is sufficiently
|
|
|
|
* fast.
|
|
|
|
*
|
|
|
|
* - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
|
|
|
|
* do not trap on the (extremely unlikely) case of a page
|
|
|
|
* crossing operation.
|
|
|
|
*
|
|
|
|
* - Furthermore, we need an efficient 64-bit compile for the
|
|
|
|
* 64-bit case in order to generate the "number of bytes in
|
|
|
|
* the final mask". Again, that could be replaced with a
|
|
|
|
* efficient population count instruction or similar.
|
|
|
|
*/
|
|
|
|
#ifdef CONFIG_DCACHE_WORD_ACCESS
|
|
|
|
|
2012-04-06 14:54:56 -06:00
|
|
|
#include <asm/word-at-a-time.h>
|
2012-03-06 12:16:17 -07:00
|
|
|
|
2012-04-06 14:54:56 -06:00
|
|
|
#ifdef CONFIG_64BIT
|
2012-03-06 12:16:17 -07:00
|
|
|
|
|
|
|
static inline unsigned int fold_hash(unsigned long hash)
|
|
|
|
{
|
vfs: fix bad hashing of dentries
Josef Bacik found a performance regression between 3.2 and 3.10 and
narrowed it down to commit bfcfaa77bdf0 ("vfs: use 'unsigned long'
accesses for dcache name comparison and hashing"). He reports:
"The test case is essentially
for (i = 0; i < 1000000; i++)
mkdir("a$i");
On xfs on a fio card this goes at about 20k dir/sec with 3.2, and 12k
dir/sec with 3.10. This is because we spend waaaaay more time in
__d_lookup on 3.10 than in 3.2.
The new hashing function for strings is suboptimal for <
sizeof(unsigned long) string names (and hell even > sizeof(unsigned
long) string names that I've tested). I broke out the old hashing
function and the new one into a userspace helper to get real numbers
and this is what I'm getting:
Old hash table had 1000000 entries, 0 dupes, 0 max dupes
New hash table had 12628 entries, 987372 dupes, 900 max dupes
We had 11400 buckets with a p50 of 30 dupes, p90 of 240 dupes, p99 of 567 dupes for the new hash
My test does the hash, and then does the d_hash into a integer pointer
array the same size as the dentry hash table on my system, and then
just increments the value at the address we got to see how many
entries we overlap with.
As you can see the old hash function ended up with all 1 million
entries in their own bucket, whereas the new one they are only
distributed among ~12.5k buckets, which is why we're using so much
more CPU in __d_lookup".
The reason for this hash regression is two-fold:
- On 64-bit architectures the down-mixing of the original 64-bit
word-at-a-time hash into the final 32-bit hash value is very
simplistic and suboptimal, and just adds the two 32-bit parts
together.
In particular, because there is no bit shuffling and the mixing
boundary is also a byte boundary, similar character patterns in the
low and high word easily end up just canceling each other out.
- the old byte-at-a-time hash mixed each byte into the final hash as it
hashed the path component name, resulting in the low bits of the hash
generally being a good source of hash data. That is not true for the
word-at-a-time case, and the hash data is distributed among all the
bits.
The fix is the same in both cases: do a better job of mixing the bits up
and using as much of the hash data as possible. We already have the
"hash_32|64()" functions to do that.
Reported-by: Josef Bacik <jbacik@fb.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Chris Mason <clm@fb.com>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-09-13 12:30:10 -06:00
|
|
|
return hash_64(hash, 32);
|
2012-03-06 12:16:17 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
#else /* 32-bit case */
|
|
|
|
|
|
|
|
#define fold_hash(x) (x)
|
|
|
|
|
|
|
|
#endif
|
|
|
|
|
|
|
|
unsigned int full_name_hash(const unsigned char *name, unsigned int len)
|
|
|
|
{
|
|
|
|
unsigned long a, mask;
|
|
|
|
unsigned long hash = 0;
|
|
|
|
|
|
|
|
for (;;) {
|
2012-05-03 11:16:43 -06:00
|
|
|
a = load_unaligned_zeropad(name);
|
2012-03-06 12:16:17 -07:00
|
|
|
if (len < sizeof(unsigned long))
|
|
|
|
break;
|
|
|
|
hash += a;
|
2012-03-22 15:59:52 -06:00
|
|
|
hash *= 9;
|
2012-03-06 12:16:17 -07:00
|
|
|
name += sizeof(unsigned long);
|
|
|
|
len -= sizeof(unsigned long);
|
|
|
|
if (!len)
|
|
|
|
goto done;
|
|
|
|
}
|
2013-12-12 10:40:21 -07:00
|
|
|
mask = bytemask_from_count(len);
|
2012-03-06 12:16:17 -07:00
|
|
|
hash += mask & a;
|
|
|
|
done:
|
|
|
|
return fold_hash(hash);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(full_name_hash);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Calculate the length and hash of the path component, and
|
vfs: simplify and shrink stack frame of link_path_walk()
Commit 9226b5b440f2 ("vfs: avoid non-forwarding large load after small
store in path lookup") made link_path_walk() always access the
"hash_len" field as a single 64-bit entity, in order to avoid mixed size
accesses to the members.
However, what I didn't notice was that that effectively means that the
whole "struct qstr this" is now basically redundant. We already
explicitly track the "const char *name", and if we just use "u64
hash_len" instead of "long len", there is nothing else left of the
"struct qstr".
We do end up wanting the "struct qstr" if we have a filesystem with a
"d_hash()" function, but that's a rare case, and we might as well then
just squirrell away the name and hash_len at that point.
End result: fewer live variables in the loop, a smaller stack frame, and
better code generation. And we don't need to pass in pointers variables
to helper functions any more, because the return value contains all the
relevant information. So this removes more lines than it adds, and the
source code is clearer too.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-09-15 11:51:07 -06:00
|
|
|
* return the "hash_len" as the result.
|
2012-03-06 12:16:17 -07:00
|
|
|
*/
|
vfs: simplify and shrink stack frame of link_path_walk()
Commit 9226b5b440f2 ("vfs: avoid non-forwarding large load after small
store in path lookup") made link_path_walk() always access the
"hash_len" field as a single 64-bit entity, in order to avoid mixed size
accesses to the members.
However, what I didn't notice was that that effectively means that the
whole "struct qstr this" is now basically redundant. We already
explicitly track the "const char *name", and if we just use "u64
hash_len" instead of "long len", there is nothing else left of the
"struct qstr".
We do end up wanting the "struct qstr" if we have a filesystem with a
"d_hash()" function, but that's a rare case, and we might as well then
just squirrell away the name and hash_len at that point.
End result: fewer live variables in the loop, a smaller stack frame, and
better code generation. And we don't need to pass in pointers variables
to helper functions any more, because the return value contains all the
relevant information. So this removes more lines than it adds, and the
source code is clearer too.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-09-15 11:51:07 -06:00
|
|
|
static inline u64 hash_name(const char *name)
|
2012-03-06 12:16:17 -07:00
|
|
|
{
|
word-at-a-time: make the interfaces truly generic
This changes the interfaces in <asm/word-at-a-time.h> to be a bit more
complicated, but a lot more generic.
In particular, it allows us to really do the operations efficiently on
both little-endian and big-endian machines, pretty much regardless of
machine details. For example, if you can rely on a fast population
count instruction on your architecture, this will allow you to make your
optimized <asm/word-at-a-time.h> file with that.
NOTE! The "generic" version in include/asm-generic/word-at-a-time.h is
not truly generic, it actually only works on big-endian. Why? Because
on little-endian the generic algorithms are wasteful, since you can
inevitably do better. The x86 implementation is an example of that.
(The only truly non-generic part of the asm-generic implementation is
the "find_zero()" function, and you could make a little-endian version
of it. And if the Kbuild infrastructure allowed us to pick a particular
header file, that would be lovely)
The <asm/word-at-a-time.h> functions are as follows:
- WORD_AT_A_TIME_CONSTANTS: specific constants that the algorithm
uses.
- has_zero(): take a word, and determine if it has a zero byte in it.
It gets the word, the pointer to the constant pool, and a pointer to
an intermediate "data" field it can set.
This is the "quick-and-dirty" zero tester: it's what is run inside
the hot loops.
- "prep_zero_mask()": take the word, the data that has_zero() produced,
and the constant pool, and generate an *exact* mask of which byte had
the first zero. This is run directly *outside* the loop, and allows
the "has_zero()" function to answer the "is there a zero byte"
question without necessarily getting exactly *which* byte is the
first one to contain a zero.
If you do multiple byte lookups concurrently (eg "hash_name()", which
looks for both NUL and '/' bytes), after you've done the prep_zero_mask()
phase, the result of those can be or'ed together to get the "either
or" case.
- The result from "prep_zero_mask()" can then be fed into "find_zero()"
(to find the byte offset of the first byte that was zero) or into
"zero_bytemask()" (to find the bytemask of the bytes preceding the
zero byte).
The existence of zero_bytemask() is optional, and is not necessary
for the normal string routines. But dentry name hashing needs it, so
if you enable DENTRY_WORD_AT_A_TIME you need to expose it.
This changes the generic strncpy_from_user() function and the dentry
hashing functions to use these modified word-at-a-time interfaces. This
gets us back to the optimized state of the x86 strncpy that we lost in
the previous commit when moving over to the generic version.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-26 11:43:17 -06:00
|
|
|
unsigned long a, b, adata, bdata, mask, hash, len;
|
|
|
|
const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
|
2012-03-06 12:16:17 -07:00
|
|
|
|
|
|
|
hash = a = 0;
|
|
|
|
len = -sizeof(unsigned long);
|
|
|
|
do {
|
|
|
|
hash = (hash + a) * 9;
|
|
|
|
len += sizeof(unsigned long);
|
2012-05-03 11:16:43 -06:00
|
|
|
a = load_unaligned_zeropad(name+len);
|
word-at-a-time: make the interfaces truly generic
This changes the interfaces in <asm/word-at-a-time.h> to be a bit more
complicated, but a lot more generic.
In particular, it allows us to really do the operations efficiently on
both little-endian and big-endian machines, pretty much regardless of
machine details. For example, if you can rely on a fast population
count instruction on your architecture, this will allow you to make your
optimized <asm/word-at-a-time.h> file with that.
NOTE! The "generic" version in include/asm-generic/word-at-a-time.h is
not truly generic, it actually only works on big-endian. Why? Because
on little-endian the generic algorithms are wasteful, since you can
inevitably do better. The x86 implementation is an example of that.
(The only truly non-generic part of the asm-generic implementation is
the "find_zero()" function, and you could make a little-endian version
of it. And if the Kbuild infrastructure allowed us to pick a particular
header file, that would be lovely)
The <asm/word-at-a-time.h> functions are as follows:
- WORD_AT_A_TIME_CONSTANTS: specific constants that the algorithm
uses.
- has_zero(): take a word, and determine if it has a zero byte in it.
It gets the word, the pointer to the constant pool, and a pointer to
an intermediate "data" field it can set.
This is the "quick-and-dirty" zero tester: it's what is run inside
the hot loops.
- "prep_zero_mask()": take the word, the data that has_zero() produced,
and the constant pool, and generate an *exact* mask of which byte had
the first zero. This is run directly *outside* the loop, and allows
the "has_zero()" function to answer the "is there a zero byte"
question without necessarily getting exactly *which* byte is the
first one to contain a zero.
If you do multiple byte lookups concurrently (eg "hash_name()", which
looks for both NUL and '/' bytes), after you've done the prep_zero_mask()
phase, the result of those can be or'ed together to get the "either
or" case.
- The result from "prep_zero_mask()" can then be fed into "find_zero()"
(to find the byte offset of the first byte that was zero) or into
"zero_bytemask()" (to find the bytemask of the bytes preceding the
zero byte).
The existence of zero_bytemask() is optional, and is not necessary
for the normal string routines. But dentry name hashing needs it, so
if you enable DENTRY_WORD_AT_A_TIME you need to expose it.
This changes the generic strncpy_from_user() function and the dentry
hashing functions to use these modified word-at-a-time interfaces. This
gets us back to the optimized state of the x86 strncpy that we lost in
the previous commit when moving over to the generic version.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-26 11:43:17 -06:00
|
|
|
b = a ^ REPEAT_BYTE('/');
|
|
|
|
} while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
|
|
|
|
|
|
|
|
adata = prep_zero_mask(a, adata, &constants);
|
|
|
|
bdata = prep_zero_mask(b, bdata, &constants);
|
|
|
|
|
|
|
|
mask = create_zero_mask(adata | bdata);
|
|
|
|
|
|
|
|
hash += a & zero_bytemask(mask);
|
2014-09-14 18:28:32 -06:00
|
|
|
len += find_zero(mask);
|
vfs: simplify and shrink stack frame of link_path_walk()
Commit 9226b5b440f2 ("vfs: avoid non-forwarding large load after small
store in path lookup") made link_path_walk() always access the
"hash_len" field as a single 64-bit entity, in order to avoid mixed size
accesses to the members.
However, what I didn't notice was that that effectively means that the
whole "struct qstr this" is now basically redundant. We already
explicitly track the "const char *name", and if we just use "u64
hash_len" instead of "long len", there is nothing else left of the
"struct qstr".
We do end up wanting the "struct qstr" if we have a filesystem with a
"d_hash()" function, but that's a rare case, and we might as well then
just squirrell away the name and hash_len at that point.
End result: fewer live variables in the loop, a smaller stack frame, and
better code generation. And we don't need to pass in pointers variables
to helper functions any more, because the return value contains all the
relevant information. So this removes more lines than it adds, and the
source code is clearer too.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-09-15 11:51:07 -06:00
|
|
|
return hashlen_create(fold_hash(hash), len);
|
2012-03-06 12:16:17 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
#else
|
|
|
|
|
2012-03-02 15:32:59 -07:00
|
|
|
unsigned int full_name_hash(const unsigned char *name, unsigned int len)
|
|
|
|
{
|
|
|
|
unsigned long hash = init_name_hash();
|
|
|
|
while (len--)
|
|
|
|
hash = partial_name_hash(*name++, hash);
|
|
|
|
return end_name_hash(hash);
|
|
|
|
}
|
2012-03-02 20:40:57 -07:00
|
|
|
EXPORT_SYMBOL(full_name_hash);
|
2012-03-02 15:32:59 -07:00
|
|
|
|
2012-03-02 15:49:24 -07:00
|
|
|
/*
|
|
|
|
* We know there's a real path component here of at least
|
|
|
|
* one character.
|
|
|
|
*/
|
vfs: simplify and shrink stack frame of link_path_walk()
Commit 9226b5b440f2 ("vfs: avoid non-forwarding large load after small
store in path lookup") made link_path_walk() always access the
"hash_len" field as a single 64-bit entity, in order to avoid mixed size
accesses to the members.
However, what I didn't notice was that that effectively means that the
whole "struct qstr this" is now basically redundant. We already
explicitly track the "const char *name", and if we just use "u64
hash_len" instead of "long len", there is nothing else left of the
"struct qstr".
We do end up wanting the "struct qstr" if we have a filesystem with a
"d_hash()" function, but that's a rare case, and we might as well then
just squirrell away the name and hash_len at that point.
End result: fewer live variables in the loop, a smaller stack frame, and
better code generation. And we don't need to pass in pointers variables
to helper functions any more, because the return value contains all the
relevant information. So this removes more lines than it adds, and the
source code is clearer too.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-09-15 11:51:07 -06:00
|
|
|
static inline u64 hash_name(const char *name)
|
2012-03-02 15:49:24 -07:00
|
|
|
{
|
|
|
|
unsigned long hash = init_name_hash();
|
|
|
|
unsigned long len = 0, c;
|
|
|
|
|
|
|
|
c = (unsigned char)*name;
|
|
|
|
do {
|
|
|
|
len++;
|
|
|
|
hash = partial_name_hash(c, hash);
|
|
|
|
c = (unsigned char)name[len];
|
|
|
|
} while (c && c != '/');
|
vfs: simplify and shrink stack frame of link_path_walk()
Commit 9226b5b440f2 ("vfs: avoid non-forwarding large load after small
store in path lookup") made link_path_walk() always access the
"hash_len" field as a single 64-bit entity, in order to avoid mixed size
accesses to the members.
However, what I didn't notice was that that effectively means that the
whole "struct qstr this" is now basically redundant. We already
explicitly track the "const char *name", and if we just use "u64
hash_len" instead of "long len", there is nothing else left of the
"struct qstr".
We do end up wanting the "struct qstr" if we have a filesystem with a
"d_hash()" function, but that's a rare case, and we might as well then
just squirrell away the name and hash_len at that point.
End result: fewer live variables in the loop, a smaller stack frame, and
better code generation. And we don't need to pass in pointers variables
to helper functions any more, because the return value contains all the
relevant information. So this removes more lines than it adds, and the
source code is clearer too.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-09-15 11:51:07 -06:00
|
|
|
return hashlen_create(end_name_hash(hash), len);
|
2012-03-02 15:49:24 -07:00
|
|
|
}
|
|
|
|
|
2012-03-06 12:16:17 -07:00
|
|
|
#endif
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
/*
|
|
|
|
* Name resolution.
|
2005-04-29 09:00:17 -06:00
|
|
|
* This is the basic name resolution function, turning a pathname into
|
|
|
|
* the final dentry. We expect 'base' to be positive and a directory.
|
2005-04-16 16:20:36 -06:00
|
|
|
*
|
2005-04-29 09:00:17 -06:00
|
|
|
* Returns 0 and nd will have valid dentry and mnt on success.
|
|
|
|
* Returns error and drops reference to input namei data on failure.
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
2009-08-08 15:41:57 -06:00
|
|
|
static int link_path_walk(const char *name, struct nameidata *nd)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
|
|
|
int err;
|
2015-04-18 18:30:49 -06:00
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
while (*name=='/')
|
|
|
|
name++;
|
|
|
|
if (!*name)
|
2015-04-18 18:44:34 -06:00
|
|
|
return 0;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
/* At this point we know we have a real path component. */
|
|
|
|
for(;;) {
|
vfs: simplify and shrink stack frame of link_path_walk()
Commit 9226b5b440f2 ("vfs: avoid non-forwarding large load after small
store in path lookup") made link_path_walk() always access the
"hash_len" field as a single 64-bit entity, in order to avoid mixed size
accesses to the members.
However, what I didn't notice was that that effectively means that the
whole "struct qstr this" is now basically redundant. We already
explicitly track the "const char *name", and if we just use "u64
hash_len" instead of "long len", there is nothing else left of the
"struct qstr".
We do end up wanting the "struct qstr" if we have a filesystem with a
"d_hash()" function, but that's a rare case, and we might as well then
just squirrell away the name and hash_len at that point.
End result: fewer live variables in the loop, a smaller stack frame, and
better code generation. And we don't need to pass in pointers variables
to helper functions any more, because the return value contains all the
relevant information. So this removes more lines than it adds, and the
source code is clearer too.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-09-15 11:51:07 -06:00
|
|
|
u64 hash_len;
|
2011-02-22 13:10:03 -07:00
|
|
|
int type;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2011-02-21 19:34:47 -07:00
|
|
|
err = may_lookup(nd);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (err)
|
2015-05-09 14:54:45 -06:00
|
|
|
return err;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
vfs: simplify and shrink stack frame of link_path_walk()
Commit 9226b5b440f2 ("vfs: avoid non-forwarding large load after small
store in path lookup") made link_path_walk() always access the
"hash_len" field as a single 64-bit entity, in order to avoid mixed size
accesses to the members.
However, what I didn't notice was that that effectively means that the
whole "struct qstr this" is now basically redundant. We already
explicitly track the "const char *name", and if we just use "u64
hash_len" instead of "long len", there is nothing else left of the
"struct qstr".
We do end up wanting the "struct qstr" if we have a filesystem with a
"d_hash()" function, but that's a rare case, and we might as well then
just squirrell away the name and hash_len at that point.
End result: fewer live variables in the loop, a smaller stack frame, and
better code generation. And we don't need to pass in pointers variables
to helper functions any more, because the return value contains all the
relevant information. So this removes more lines than it adds, and the
source code is clearer too.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-09-15 11:51:07 -06:00
|
|
|
hash_len = hash_name(name);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2011-02-22 13:10:03 -07:00
|
|
|
type = LAST_NORM;
|
vfs: simplify and shrink stack frame of link_path_walk()
Commit 9226b5b440f2 ("vfs: avoid non-forwarding large load after small
store in path lookup") made link_path_walk() always access the
"hash_len" field as a single 64-bit entity, in order to avoid mixed size
accesses to the members.
However, what I didn't notice was that that effectively means that the
whole "struct qstr this" is now basically redundant. We already
explicitly track the "const char *name", and if we just use "u64
hash_len" instead of "long len", there is nothing else left of the
"struct qstr".
We do end up wanting the "struct qstr" if we have a filesystem with a
"d_hash()" function, but that's a rare case, and we might as well then
just squirrell away the name and hash_len at that point.
End result: fewer live variables in the loop, a smaller stack frame, and
better code generation. And we don't need to pass in pointers variables
to helper functions any more, because the return value contains all the
relevant information. So this removes more lines than it adds, and the
source code is clearer too.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-09-15 11:51:07 -06:00
|
|
|
if (name[0] == '.') switch (hashlen_len(hash_len)) {
|
2011-02-22 13:10:03 -07:00
|
|
|
case 2:
|
2012-03-02 15:49:24 -07:00
|
|
|
if (name[1] == '.') {
|
2011-02-22 13:10:03 -07:00
|
|
|
type = LAST_DOTDOT;
|
2011-02-22 13:50:10 -07:00
|
|
|
nd->flags |= LOOKUP_JUMPED;
|
|
|
|
}
|
2011-02-22 13:10:03 -07:00
|
|
|
break;
|
|
|
|
case 1:
|
|
|
|
type = LAST_DOT;
|
|
|
|
}
|
2011-03-08 12:17:44 -07:00
|
|
|
if (likely(type == LAST_NORM)) {
|
|
|
|
struct dentry *parent = nd->path.dentry;
|
2011-02-22 13:50:10 -07:00
|
|
|
nd->flags &= ~LOOKUP_JUMPED;
|
2011-03-08 12:17:44 -07:00
|
|
|
if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
|
2014-09-16 06:07:35 -06:00
|
|
|
struct qstr this = { { .hash_len = hash_len }, .name = name };
|
2013-05-21 16:22:44 -06:00
|
|
|
err = parent->d_op->d_hash(parent, &this);
|
2011-03-08 12:17:44 -07:00
|
|
|
if (err < 0)
|
2015-05-09 14:54:45 -06:00
|
|
|
return err;
|
vfs: simplify and shrink stack frame of link_path_walk()
Commit 9226b5b440f2 ("vfs: avoid non-forwarding large load after small
store in path lookup") made link_path_walk() always access the
"hash_len" field as a single 64-bit entity, in order to avoid mixed size
accesses to the members.
However, what I didn't notice was that that effectively means that the
whole "struct qstr this" is now basically redundant. We already
explicitly track the "const char *name", and if we just use "u64
hash_len" instead of "long len", there is nothing else left of the
"struct qstr".
We do end up wanting the "struct qstr" if we have a filesystem with a
"d_hash()" function, but that's a rare case, and we might as well then
just squirrell away the name and hash_len at that point.
End result: fewer live variables in the loop, a smaller stack frame, and
better code generation. And we don't need to pass in pointers variables
to helper functions any more, because the return value contains all the
relevant information. So this removes more lines than it adds, and the
source code is clearer too.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-09-15 11:51:07 -06:00
|
|
|
hash_len = this.hash_len;
|
|
|
|
name = this.name;
|
2011-03-08 12:17:44 -07:00
|
|
|
}
|
|
|
|
}
|
2011-02-22 13:10:03 -07:00
|
|
|
|
vfs: simplify and shrink stack frame of link_path_walk()
Commit 9226b5b440f2 ("vfs: avoid non-forwarding large load after small
store in path lookup") made link_path_walk() always access the
"hash_len" field as a single 64-bit entity, in order to avoid mixed size
accesses to the members.
However, what I didn't notice was that that effectively means that the
whole "struct qstr this" is now basically redundant. We already
explicitly track the "const char *name", and if we just use "u64
hash_len" instead of "long len", there is nothing else left of the
"struct qstr".
We do end up wanting the "struct qstr" if we have a filesystem with a
"d_hash()" function, but that's a rare case, and we might as well then
just squirrell away the name and hash_len at that point.
End result: fewer live variables in the loop, a smaller stack frame, and
better code generation. And we don't need to pass in pointers variables
to helper functions any more, because the return value contains all the
relevant information. So this removes more lines than it adds, and the
source code is clearer too.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-09-15 11:51:07 -06:00
|
|
|
nd->last.hash_len = hash_len;
|
|
|
|
nd->last.name = name;
|
2013-01-24 16:04:22 -07:00
|
|
|
nd->last_type = type;
|
|
|
|
|
vfs: simplify and shrink stack frame of link_path_walk()
Commit 9226b5b440f2 ("vfs: avoid non-forwarding large load after small
store in path lookup") made link_path_walk() always access the
"hash_len" field as a single 64-bit entity, in order to avoid mixed size
accesses to the members.
However, what I didn't notice was that that effectively means that the
whole "struct qstr this" is now basically redundant. We already
explicitly track the "const char *name", and if we just use "u64
hash_len" instead of "long len", there is nothing else left of the
"struct qstr".
We do end up wanting the "struct qstr" if we have a filesystem with a
"d_hash()" function, but that's a rare case, and we might as well then
just squirrell away the name and hash_len at that point.
End result: fewer live variables in the loop, a smaller stack frame, and
better code generation. And we don't need to pass in pointers variables
to helper functions any more, because the return value contains all the
relevant information. So this removes more lines than it adds, and the
source code is clearer too.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-09-15 11:51:07 -06:00
|
|
|
name += hashlen_len(hash_len);
|
|
|
|
if (!*name)
|
2015-04-18 18:21:40 -06:00
|
|
|
goto OK;
|
2012-03-02 15:49:24 -07:00
|
|
|
/*
|
|
|
|
* If it wasn't NUL, we know it was '/'. Skip that
|
|
|
|
* slash, and continue until no more slashes.
|
|
|
|
*/
|
|
|
|
do {
|
vfs: simplify and shrink stack frame of link_path_walk()
Commit 9226b5b440f2 ("vfs: avoid non-forwarding large load after small
store in path lookup") made link_path_walk() always access the
"hash_len" field as a single 64-bit entity, in order to avoid mixed size
accesses to the members.
However, what I didn't notice was that that effectively means that the
whole "struct qstr this" is now basically redundant. We already
explicitly track the "const char *name", and if we just use "u64
hash_len" instead of "long len", there is nothing else left of the
"struct qstr".
We do end up wanting the "struct qstr" if we have a filesystem with a
"d_hash()" function, but that's a rare case, and we might as well then
just squirrell away the name and hash_len at that point.
End result: fewer live variables in the loop, a smaller stack frame, and
better code generation. And we don't need to pass in pointers variables
to helper functions any more, because the return value contains all the
relevant information. So this removes more lines than it adds, and the
source code is clearer too.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-09-15 11:51:07 -06:00
|
|
|
name++;
|
|
|
|
} while (unlikely(*name == '/'));
|
2015-05-04 06:58:35 -06:00
|
|
|
if (unlikely(!*name)) {
|
|
|
|
OK:
|
2015-05-08 15:19:59 -06:00
|
|
|
/* pathname body, done */
|
2015-05-04 06:58:35 -06:00
|
|
|
if (!nd->depth)
|
|
|
|
return 0;
|
|
|
|
name = nd->stack[nd->depth - 1].name;
|
2015-05-08 15:19:59 -06:00
|
|
|
/* trailing symlink, done */
|
2015-05-04 06:58:35 -06:00
|
|
|
if (!name)
|
|
|
|
return 0;
|
|
|
|
/* last component of nested symlink */
|
2015-05-04 15:47:11 -06:00
|
|
|
err = walk_component(nd, WALK_GET | WALK_PUT);
|
2015-05-04 06:58:35 -06:00
|
|
|
} else {
|
2015-05-04 15:47:11 -06:00
|
|
|
err = walk_component(nd, WALK_GET);
|
2015-05-04 06:58:35 -06:00
|
|
|
}
|
2011-03-13 17:58:58 -06:00
|
|
|
if (err < 0)
|
2015-05-09 14:54:45 -06:00
|
|
|
return err;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2011-03-13 17:58:58 -06:00
|
|
|
if (err) {
|
2015-05-04 16:26:59 -06:00
|
|
|
const char *s = get_link(nd);
|
2015-04-17 21:44:45 -06:00
|
|
|
|
2015-08-12 04:29:44 -06:00
|
|
|
if (IS_ERR(s))
|
2015-05-09 14:54:45 -06:00
|
|
|
return PTR_ERR(s);
|
2015-04-18 18:03:03 -06:00
|
|
|
err = 0;
|
|
|
|
if (unlikely(!s)) {
|
|
|
|
/* jumped */
|
2015-05-02 18:19:23 -06:00
|
|
|
put_link(nd);
|
2015-04-18 18:03:03 -06:00
|
|
|
} else {
|
2015-05-10 09:01:00 -06:00
|
|
|
nd->stack[nd->depth - 1].name = name;
|
|
|
|
name = s;
|
|
|
|
continue;
|
2015-04-18 18:03:03 -06:00
|
|
|
}
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
}
|
2015-08-01 17:59:28 -06:00
|
|
|
if (unlikely(!d_can_lookup(nd->path.dentry))) {
|
|
|
|
if (nd->flags & LOOKUP_RCU) {
|
|
|
|
if (unlazy_walk(nd, NULL, 0))
|
|
|
|
return -ECHILD;
|
|
|
|
}
|
2015-05-09 14:54:45 -06:00
|
|
|
return -ENOTDIR;
|
2015-08-01 17:59:28 -06:00
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-05-12 16:43:07 -06:00
|
|
|
static const char *path_init(struct nameidata *nd, unsigned flags)
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
{
|
|
|
|
int retval = 0;
|
2015-05-12 16:43:07 -06:00
|
|
|
const char *s = nd->name->name;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
|
|
|
|
nd->last_type = LAST_ROOT; /* if there are only slashes... */
|
2014-11-20 12:20:24 -07:00
|
|
|
nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
nd->depth = 0;
|
2011-03-09 21:04:47 -07:00
|
|
|
if (flags & LOOKUP_ROOT) {
|
2013-09-12 12:22:53 -06:00
|
|
|
struct dentry *root = nd->root.dentry;
|
|
|
|
struct inode *inode = root->d_inode;
|
2015-02-22 18:07:13 -07:00
|
|
|
if (*s) {
|
2014-04-01 09:08:41 -06:00
|
|
|
if (!d_can_lookup(root))
|
2015-05-08 15:19:59 -06:00
|
|
|
return ERR_PTR(-ENOTDIR);
|
2011-03-11 10:08:24 -07:00
|
|
|
retval = inode_permission(inode, MAY_EXEC);
|
|
|
|
if (retval)
|
2015-05-08 15:19:59 -06:00
|
|
|
return ERR_PTR(retval);
|
2011-03-11 10:08:24 -07:00
|
|
|
}
|
2011-03-09 21:04:47 -07:00
|
|
|
nd->path = nd->root;
|
|
|
|
nd->inode = inode;
|
|
|
|
if (flags & LOOKUP_RCU) {
|
2013-11-08 10:45:01 -07:00
|
|
|
rcu_read_lock();
|
2011-03-09 21:04:47 -07:00
|
|
|
nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
|
2015-05-09 17:02:01 -06:00
|
|
|
nd->root_seq = nd->seq;
|
2013-09-29 20:06:07 -06:00
|
|
|
nd->m_seq = read_seqbegin(&mount_lock);
|
2011-03-09 21:04:47 -07:00
|
|
|
} else {
|
|
|
|
path_get(&nd->path);
|
|
|
|
}
|
2015-05-08 15:19:59 -06:00
|
|
|
return s;
|
2011-03-09 21:04:47 -07:00
|
|
|
}
|
|
|
|
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
nd->root.mnt = NULL;
|
2015-12-05 18:51:58 -07:00
|
|
|
nd->path.mnt = NULL;
|
|
|
|
nd->path.dentry = NULL;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
|
2013-09-29 20:06:07 -06:00
|
|
|
nd->m_seq = read_seqbegin(&mount_lock);
|
2015-02-22 18:07:13 -07:00
|
|
|
if (*s == '/') {
|
2015-12-05 18:07:21 -07:00
|
|
|
if (flags & LOOKUP_RCU)
|
2013-11-08 10:45:01 -07:00
|
|
|
rcu_read_lock();
|
2015-12-05 18:07:21 -07:00
|
|
|
set_root(nd);
|
2015-12-05 18:51:58 -07:00
|
|
|
if (likely(!nd_jump_root(nd)))
|
2015-12-05 18:25:06 -07:00
|
|
|
return s;
|
2015-12-05 18:51:58 -07:00
|
|
|
nd->root.mnt = NULL;
|
2015-12-05 18:25:06 -07:00
|
|
|
rcu_read_unlock();
|
|
|
|
return ERR_PTR(-ECHILD);
|
2015-05-12 16:43:07 -06:00
|
|
|
} else if (nd->dfd == AT_FDCWD) {
|
2011-02-22 12:02:58 -07:00
|
|
|
if (flags & LOOKUP_RCU) {
|
|
|
|
struct fs_struct *fs = current->fs;
|
|
|
|
unsigned seq;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
|
2013-11-08 10:45:01 -07:00
|
|
|
rcu_read_lock();
|
2011-01-06 23:49:53 -07:00
|
|
|
|
2011-02-22 12:02:58 -07:00
|
|
|
do {
|
|
|
|
seq = read_seqcount_begin(&fs->seq);
|
|
|
|
nd->path = fs->pwd;
|
2015-12-05 18:25:06 -07:00
|
|
|
nd->inode = nd->path.dentry->d_inode;
|
2011-02-22 12:02:58 -07:00
|
|
|
nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
|
|
|
|
} while (read_seqcount_retry(&fs->seq, seq));
|
|
|
|
} else {
|
|
|
|
get_fs_pwd(current->fs, &nd->path);
|
2015-12-05 18:25:06 -07:00
|
|
|
nd->inode = nd->path.dentry->d_inode;
|
2011-02-22 12:02:58 -07:00
|
|
|
}
|
2015-12-05 18:25:06 -07:00
|
|
|
return s;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
} else {
|
2012-12-11 06:56:16 -07:00
|
|
|
/* Caller must check execute permissions on the starting path component */
|
2015-05-12 16:43:07 -06:00
|
|
|
struct fd f = fdget_raw(nd->dfd);
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
struct dentry *dentry;
|
|
|
|
|
2012-08-28 10:52:22 -06:00
|
|
|
if (!f.file)
|
2015-05-08 15:19:59 -06:00
|
|
|
return ERR_PTR(-EBADF);
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
|
2012-08-28 10:52:22 -06:00
|
|
|
dentry = f.file->f_path.dentry;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
|
2015-02-22 18:07:13 -07:00
|
|
|
if (*s) {
|
2014-04-01 09:08:41 -06:00
|
|
|
if (!d_can_lookup(dentry)) {
|
2012-08-28 10:52:22 -06:00
|
|
|
fdput(f);
|
2015-05-08 15:19:59 -06:00
|
|
|
return ERR_PTR(-ENOTDIR);
|
2012-08-28 10:52:22 -06:00
|
|
|
}
|
2011-03-14 16:56:51 -06:00
|
|
|
}
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
|
2012-08-28 10:52:22 -06:00
|
|
|
nd->path = f.file->f_path;
|
2011-02-22 12:02:58 -07:00
|
|
|
if (flags & LOOKUP_RCU) {
|
2013-11-08 10:45:01 -07:00
|
|
|
rcu_read_lock();
|
2015-05-11 06:05:05 -06:00
|
|
|
nd->inode = nd->path.dentry->d_inode;
|
|
|
|
nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
|
2011-02-22 12:02:58 -07:00
|
|
|
} else {
|
2012-08-28 10:52:22 -06:00
|
|
|
path_get(&nd->path);
|
2015-05-11 06:05:05 -06:00
|
|
|
nd->inode = nd->path.dentry->d_inode;
|
2011-02-22 12:02:58 -07:00
|
|
|
}
|
2015-05-11 06:05:05 -06:00
|
|
|
fdput(f);
|
2015-05-08 15:19:59 -06:00
|
|
|
return s;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
}
|
2009-04-07 09:44:16 -06:00
|
|
|
}
|
|
|
|
|
2015-05-08 15:37:07 -06:00
|
|
|
static const char *trailing_symlink(struct nameidata *nd)
|
2015-04-22 11:46:57 -06:00
|
|
|
{
|
|
|
|
const char *s;
|
2015-05-06 13:58:18 -06:00
|
|
|
int error = may_follow_link(nd);
|
2015-05-08 16:05:21 -06:00
|
|
|
if (unlikely(error))
|
2015-05-08 15:37:07 -06:00
|
|
|
return ERR_PTR(error);
|
2015-04-22 11:46:57 -06:00
|
|
|
nd->flags |= LOOKUP_PARENT;
|
2015-05-10 09:01:00 -06:00
|
|
|
nd->stack[0].name = NULL;
|
2015-04-18 22:53:50 -06:00
|
|
|
s = get_link(nd);
|
2015-05-08 16:05:21 -06:00
|
|
|
return s ? s : "";
|
2015-04-22 11:46:57 -06:00
|
|
|
}
|
|
|
|
|
2015-04-22 15:52:47 -06:00
|
|
|
static inline int lookup_last(struct nameidata *nd)
|
2011-03-14 17:54:59 -06:00
|
|
|
{
|
|
|
|
if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
|
|
|
|
nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
|
|
|
|
|
|
|
|
nd->flags &= ~LOOKUP_PARENT;
|
2015-05-08 16:05:21 -06:00
|
|
|
return walk_component(nd,
|
2015-05-04 15:47:11 -06:00
|
|
|
nd->flags & LOOKUP_FOLLOW
|
|
|
|
? nd->depth
|
|
|
|
? WALK_PUT | WALK_GET
|
|
|
|
: WALK_GET
|
|
|
|
: 0);
|
2011-03-14 17:54:59 -06:00
|
|
|
}
|
|
|
|
|
2009-04-07 09:44:16 -06:00
|
|
|
/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
|
2015-05-12 16:43:07 -06:00
|
|
|
static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
|
2009-04-07 09:44:16 -06:00
|
|
|
{
|
2015-05-12 16:43:07 -06:00
|
|
|
const char *s = path_init(nd, flags);
|
2011-03-14 17:54:59 -06:00
|
|
|
int err;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
|
2015-05-08 15:19:59 -06:00
|
|
|
if (IS_ERR(s))
|
|
|
|
return PTR_ERR(s);
|
2015-05-08 15:37:07 -06:00
|
|
|
while (!(err = link_path_walk(s, nd))
|
|
|
|
&& ((err = lookup_last(nd)) > 0)) {
|
|
|
|
s = trailing_symlink(nd);
|
|
|
|
if (IS_ERR(s)) {
|
|
|
|
err = PTR_ERR(s);
|
|
|
|
break;
|
2011-03-14 17:54:59 -06:00
|
|
|
}
|
|
|
|
}
|
2011-03-25 09:00:12 -06:00
|
|
|
if (!err)
|
|
|
|
err = complete_walk(nd);
|
2011-03-14 17:54:59 -06:00
|
|
|
|
2015-05-08 16:05:21 -06:00
|
|
|
if (!err && nd->flags & LOOKUP_DIRECTORY)
|
|
|
|
if (!d_can_lookup(nd->path.dentry))
|
2011-03-23 07:56:30 -06:00
|
|
|
err = -ENOTDIR;
|
2015-05-12 14:36:12 -06:00
|
|
|
if (!err) {
|
|
|
|
*path = nd->path;
|
|
|
|
nd->path.mnt = NULL;
|
|
|
|
nd->path.dentry = NULL;
|
|
|
|
}
|
|
|
|
terminate_walk(nd);
|
2011-03-14 17:54:59 -06:00
|
|
|
return err;
|
2011-02-21 21:38:09 -07:00
|
|
|
}
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
|
2015-05-12 14:36:12 -06:00
|
|
|
static int filename_lookup(int dfd, struct filename *name, unsigned flags,
|
2015-05-12 14:44:39 -06:00
|
|
|
struct path *path, struct path *root)
|
2011-02-21 21:38:09 -07:00
|
|
|
{
|
2015-05-02 05:16:16 -06:00
|
|
|
int retval;
|
2015-05-13 05:28:08 -06:00
|
|
|
struct nameidata nd;
|
2015-05-12 14:53:42 -06:00
|
|
|
if (IS_ERR(name))
|
|
|
|
return PTR_ERR(name);
|
2015-05-12 14:44:39 -06:00
|
|
|
if (unlikely(root)) {
|
|
|
|
nd.root = *root;
|
|
|
|
flags |= LOOKUP_ROOT;
|
|
|
|
}
|
2015-05-13 05:28:08 -06:00
|
|
|
set_nameidata(&nd, dfd, name);
|
2015-05-12 16:43:07 -06:00
|
|
|
retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
|
2011-02-21 21:38:09 -07:00
|
|
|
if (unlikely(retval == -ECHILD))
|
2015-05-12 16:43:07 -06:00
|
|
|
retval = path_lookupat(&nd, flags, path);
|
2011-02-21 21:38:09 -07:00
|
|
|
if (unlikely(retval == -ESTALE))
|
2015-05-12 16:43:07 -06:00
|
|
|
retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
|
2012-10-10 13:25:20 -06:00
|
|
|
if (likely(!retval))
|
2015-05-12 14:36:12 -06:00
|
|
|
audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
|
2015-05-13 05:28:08 -06:00
|
|
|
restore_nameidata();
|
2015-05-12 14:40:39 -06:00
|
|
|
putname(name);
|
2006-02-05 00:28:02 -07:00
|
|
|
return retval;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2015-05-08 14:59:20 -06:00
|
|
|
/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
|
2015-05-12 16:43:07 -06:00
|
|
|
static int path_parentat(struct nameidata *nd, unsigned flags,
|
2015-05-09 09:19:16 -06:00
|
|
|
struct path *parent)
|
2015-05-08 14:59:20 -06:00
|
|
|
{
|
2015-05-12 16:43:07 -06:00
|
|
|
const char *s = path_init(nd, flags);
|
2015-05-08 15:19:59 -06:00
|
|
|
int err;
|
|
|
|
if (IS_ERR(s))
|
|
|
|
return PTR_ERR(s);
|
|
|
|
err = link_path_walk(s, nd);
|
2015-05-08 14:59:20 -06:00
|
|
|
if (!err)
|
|
|
|
err = complete_walk(nd);
|
2015-05-09 09:19:16 -06:00
|
|
|
if (!err) {
|
|
|
|
*parent = nd->path;
|
|
|
|
nd->path.mnt = NULL;
|
|
|
|
nd->path.dentry = NULL;
|
|
|
|
}
|
|
|
|
terminate_walk(nd);
|
2015-05-08 14:59:20 -06:00
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2015-05-12 15:32:54 -06:00
|
|
|
static struct filename *filename_parentat(int dfd, struct filename *name,
|
2015-05-09 09:19:16 -06:00
|
|
|
unsigned int flags, struct path *parent,
|
|
|
|
struct qstr *last, int *type)
|
2015-05-08 14:59:20 -06:00
|
|
|
{
|
|
|
|
int retval;
|
2015-05-13 05:28:08 -06:00
|
|
|
struct nameidata nd;
|
2015-05-08 14:59:20 -06:00
|
|
|
|
2015-05-12 15:32:54 -06:00
|
|
|
if (IS_ERR(name))
|
|
|
|
return name;
|
2015-05-13 05:28:08 -06:00
|
|
|
set_nameidata(&nd, dfd, name);
|
2015-05-12 16:43:07 -06:00
|
|
|
retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
|
2015-05-08 14:59:20 -06:00
|
|
|
if (unlikely(retval == -ECHILD))
|
2015-05-12 16:43:07 -06:00
|
|
|
retval = path_parentat(&nd, flags, parent);
|
2015-05-08 14:59:20 -06:00
|
|
|
if (unlikely(retval == -ESTALE))
|
2015-05-12 16:43:07 -06:00
|
|
|
retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
|
2015-05-09 09:19:16 -06:00
|
|
|
if (likely(!retval)) {
|
|
|
|
*last = nd.last;
|
|
|
|
*type = nd.last_type;
|
|
|
|
audit_inode(name, parent->dentry, LOOKUP_PARENT);
|
2015-05-12 15:32:54 -06:00
|
|
|
} else {
|
|
|
|
putname(name);
|
|
|
|
name = ERR_PTR(retval);
|
2015-05-09 09:19:16 -06:00
|
|
|
}
|
2015-05-13 05:28:08 -06:00
|
|
|
restore_nameidata();
|
2015-05-12 15:32:54 -06:00
|
|
|
return name;
|
2015-05-08 14:59:20 -06:00
|
|
|
}
|
|
|
|
|
2012-06-14 17:01:42 -06:00
|
|
|
/* does lookup, returns the object with parent locked */
|
|
|
|
struct dentry *kern_path_locked(const char *name, struct path *path)
|
2006-01-18 18:43:53 -07:00
|
|
|
{
|
2015-05-12 15:32:54 -06:00
|
|
|
struct filename *filename;
|
|
|
|
struct dentry *d;
|
2015-05-09 09:19:16 -06:00
|
|
|
struct qstr last;
|
|
|
|
int type;
|
2015-01-21 22:00:03 -07:00
|
|
|
|
2015-05-12 15:32:54 -06:00
|
|
|
filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
|
|
|
|
&last, &type);
|
2015-01-21 22:00:03 -07:00
|
|
|
if (IS_ERR(filename))
|
|
|
|
return ERR_CAST(filename);
|
2015-05-12 15:32:54 -06:00
|
|
|
if (unlikely(type != LAST_NORM)) {
|
2015-05-09 09:19:16 -06:00
|
|
|
path_put(path);
|
2015-05-12 15:32:54 -06:00
|
|
|
putname(filename);
|
|
|
|
return ERR_PTR(-EINVAL);
|
2012-06-14 17:01:42 -06:00
|
|
|
}
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
|
2015-05-09 09:19:16 -06:00
|
|
|
d = __lookup_hash(&last, path->dentry, 0);
|
2012-06-14 17:01:42 -06:00
|
|
|
if (IS_ERR(d)) {
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_unlock(path->dentry->d_inode);
|
2015-05-09 09:19:16 -06:00
|
|
|
path_put(path);
|
2012-06-14 17:01:42 -06:00
|
|
|
}
|
2015-01-21 22:00:03 -07:00
|
|
|
putname(filename);
|
2012-06-14 17:01:42 -06:00
|
|
|
return d;
|
2006-01-18 18:43:53 -07:00
|
|
|
}
|
|
|
|
|
2008-08-01 22:49:18 -06:00
|
|
|
int kern_path(const char *name, unsigned int flags, struct path *path)
|
|
|
|
{
|
2015-05-12 14:53:42 -06:00
|
|
|
return filename_lookup(AT_FDCWD, getname_kernel(name),
|
|
|
|
flags, path, NULL);
|
2008-08-01 22:49:18 -06:00
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(kern_path);
|
2008-08-01 22:49:18 -06:00
|
|
|
|
fs: introduce vfs_path_lookup
Stackable file systems, among others, frequently need to lookup paths or
path components starting from an arbitrary point in the namespace
(identified by a dentry and a vfsmount). Currently, such file systems use
lookup_one_len, which is frowned upon [1] as it does not pass the lookup
intent along; not passing a lookup intent, for example, can trigger BUG_ON's
when stacking on top of NFSv4.
The first patch introduces a new lookup function to allow lookup starting
from an arbitrary point in the namespace. This approach has been suggested
by Christoph Hellwig [2].
The second patch changes sunrpc to use vfs_path_lookup.
The third patch changes nfsctl.c to use vfs_path_lookup.
The fourth patch marks link_path_walk static.
The fifth, and last patch, unexports path_walk because it is no longer
unnecessary to call it directly, and using the new vfs_path_lookup is
cleaner.
For example, the following snippet of code, looks up "some/path/component"
in a directory pointed to by parent_{dentry,vfsmnt}:
err = vfs_path_lookup(parent_dentry, parent_vfsmnt,
"some/path/component", 0, &nd);
if (!err) {
/* exits */
...
/* once done, release the references */
path_release(&nd);
} else if (err == -ENOENT) {
/* doesn't exist */
} else {
/* other error */
}
VFS functions such as lookup_create can be used on the nameidata structure
to pass the create intent to the file system.
Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Acked-by: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@suse.de>
Cc: Michael Halcrow <mhalcrow@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 02:48:18 -06:00
|
|
|
/**
|
|
|
|
* vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
|
|
|
|
* @dentry: pointer to dentry of the base directory
|
|
|
|
* @mnt: pointer to vfs mount of the base directory
|
|
|
|
* @name: pointer to file name
|
|
|
|
* @flags: lookup flags
|
2011-06-27 15:00:37 -06:00
|
|
|
* @path: pointer to struct path to fill
|
fs: introduce vfs_path_lookup
Stackable file systems, among others, frequently need to lookup paths or
path components starting from an arbitrary point in the namespace
(identified by a dentry and a vfsmount). Currently, such file systems use
lookup_one_len, which is frowned upon [1] as it does not pass the lookup
intent along; not passing a lookup intent, for example, can trigger BUG_ON's
when stacking on top of NFSv4.
The first patch introduces a new lookup function to allow lookup starting
from an arbitrary point in the namespace. This approach has been suggested
by Christoph Hellwig [2].
The second patch changes sunrpc to use vfs_path_lookup.
The third patch changes nfsctl.c to use vfs_path_lookup.
The fourth patch marks link_path_walk static.
The fifth, and last patch, unexports path_walk because it is no longer
unnecessary to call it directly, and using the new vfs_path_lookup is
cleaner.
For example, the following snippet of code, looks up "some/path/component"
in a directory pointed to by parent_{dentry,vfsmnt}:
err = vfs_path_lookup(parent_dentry, parent_vfsmnt,
"some/path/component", 0, &nd);
if (!err) {
/* exits */
...
/* once done, release the references */
path_release(&nd);
} else if (err == -ENOENT) {
/* doesn't exist */
} else {
/* other error */
}
VFS functions such as lookup_create can be used on the nameidata structure
to pass the create intent to the file system.
Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Acked-by: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@suse.de>
Cc: Michael Halcrow <mhalcrow@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 02:48:18 -06:00
|
|
|
*/
|
|
|
|
int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
|
|
|
|
const char *name, unsigned int flags,
|
2011-06-27 15:00:37 -06:00
|
|
|
struct path *path)
|
fs: introduce vfs_path_lookup
Stackable file systems, among others, frequently need to lookup paths or
path components starting from an arbitrary point in the namespace
(identified by a dentry and a vfsmount). Currently, such file systems use
lookup_one_len, which is frowned upon [1] as it does not pass the lookup
intent along; not passing a lookup intent, for example, can trigger BUG_ON's
when stacking on top of NFSv4.
The first patch introduces a new lookup function to allow lookup starting
from an arbitrary point in the namespace. This approach has been suggested
by Christoph Hellwig [2].
The second patch changes sunrpc to use vfs_path_lookup.
The third patch changes nfsctl.c to use vfs_path_lookup.
The fourth patch marks link_path_walk static.
The fifth, and last patch, unexports path_walk because it is no longer
unnecessary to call it directly, and using the new vfs_path_lookup is
cleaner.
For example, the following snippet of code, looks up "some/path/component"
in a directory pointed to by parent_{dentry,vfsmnt}:
err = vfs_path_lookup(parent_dentry, parent_vfsmnt,
"some/path/component", 0, &nd);
if (!err) {
/* exits */
...
/* once done, release the references */
path_release(&nd);
} else if (err == -ENOENT) {
/* doesn't exist */
} else {
/* other error */
}
VFS functions such as lookup_create can be used on the nameidata structure
to pass the create intent to the file system.
Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Acked-by: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@suse.de>
Cc: Michael Halcrow <mhalcrow@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 02:48:18 -06:00
|
|
|
{
|
2015-05-12 14:44:39 -06:00
|
|
|
struct path root = {.mnt = mnt, .dentry = dentry};
|
|
|
|
/* the first argument of filename_lookup() is ignored with root */
|
2015-05-12 14:53:42 -06:00
|
|
|
return filename_lookup(AT_FDCWD, getname_kernel(name),
|
|
|
|
flags , path, &root);
|
fs: introduce vfs_path_lookup
Stackable file systems, among others, frequently need to lookup paths or
path components starting from an arbitrary point in the namespace
(identified by a dentry and a vfsmount). Currently, such file systems use
lookup_one_len, which is frowned upon [1] as it does not pass the lookup
intent along; not passing a lookup intent, for example, can trigger BUG_ON's
when stacking on top of NFSv4.
The first patch introduces a new lookup function to allow lookup starting
from an arbitrary point in the namespace. This approach has been suggested
by Christoph Hellwig [2].
The second patch changes sunrpc to use vfs_path_lookup.
The third patch changes nfsctl.c to use vfs_path_lookup.
The fourth patch marks link_path_walk static.
The fifth, and last patch, unexports path_walk because it is no longer
unnecessary to call it directly, and using the new vfs_path_lookup is
cleaner.
For example, the following snippet of code, looks up "some/path/component"
in a directory pointed to by parent_{dentry,vfsmnt}:
err = vfs_path_lookup(parent_dentry, parent_vfsmnt,
"some/path/component", 0, &nd);
if (!err) {
/* exits */
...
/* once done, release the references */
path_release(&nd);
} else if (err == -ENOENT) {
/* doesn't exist */
} else {
/* other error */
}
VFS functions such as lookup_create can be used on the nameidata structure
to pass the create intent to the file system.
Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Acked-by: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@suse.de>
Cc: Michael Halcrow <mhalcrow@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 02:48:18 -06:00
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(vfs_path_lookup);
|
fs: introduce vfs_path_lookup
Stackable file systems, among others, frequently need to lookup paths or
path components starting from an arbitrary point in the namespace
(identified by a dentry and a vfsmount). Currently, such file systems use
lookup_one_len, which is frowned upon [1] as it does not pass the lookup
intent along; not passing a lookup intent, for example, can trigger BUG_ON's
when stacking on top of NFSv4.
The first patch introduces a new lookup function to allow lookup starting
from an arbitrary point in the namespace. This approach has been suggested
by Christoph Hellwig [2].
The second patch changes sunrpc to use vfs_path_lookup.
The third patch changes nfsctl.c to use vfs_path_lookup.
The fourth patch marks link_path_walk static.
The fifth, and last patch, unexports path_walk because it is no longer
unnecessary to call it directly, and using the new vfs_path_lookup is
cleaner.
For example, the following snippet of code, looks up "some/path/component"
in a directory pointed to by parent_{dentry,vfsmnt}:
err = vfs_path_lookup(parent_dentry, parent_vfsmnt,
"some/path/component", 0, &nd);
if (!err) {
/* exits */
...
/* once done, release the references */
path_release(&nd);
} else if (err == -ENOENT) {
/* doesn't exist */
} else {
/* other error */
}
VFS functions such as lookup_create can be used on the nameidata structure
to pass the create intent to the file system.
Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Acked-by: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@suse.de>
Cc: Michael Halcrow <mhalcrow@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 02:48:18 -06:00
|
|
|
|
2007-10-17 00:25:38 -06:00
|
|
|
/**
|
2008-03-19 18:01:00 -06:00
|
|
|
* lookup_one_len - filesystem helper to lookup single pathname component
|
2007-10-17 00:25:38 -06:00
|
|
|
* @name: pathname component to lookup
|
|
|
|
* @base: base directory to lookup from
|
|
|
|
* @len: maximum length @len should be interpreted to
|
|
|
|
*
|
2008-03-19 18:01:00 -06:00
|
|
|
* Note that this routine is purely a helper for filesystem usage and should
|
2015-02-23 00:49:48 -07:00
|
|
|
* not be called by generic code.
|
2016-01-07 14:08:20 -07:00
|
|
|
*
|
|
|
|
* The caller must hold base->i_mutex.
|
2007-10-17 00:25:38 -06:00
|
|
|
*/
|
2007-04-26 01:12:05 -06:00
|
|
|
struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
|
|
|
|
{
|
|
|
|
struct qstr this;
|
2011-03-07 21:49:20 -07:00
|
|
|
unsigned int c;
|
2012-03-26 04:54:21 -06:00
|
|
|
int err;
|
2007-04-26 01:12:05 -06:00
|
|
|
|
2016-01-22 13:40:57 -07:00
|
|
|
WARN_ON_ONCE(!inode_is_locked(base->d_inode));
|
2009-04-20 16:18:37 -06:00
|
|
|
|
2011-03-07 21:49:20 -07:00
|
|
|
this.name = name;
|
|
|
|
this.len = len;
|
2012-03-02 15:32:59 -07:00
|
|
|
this.hash = full_name_hash(name, len);
|
2011-03-07 21:49:20 -07:00
|
|
|
if (!len)
|
|
|
|
return ERR_PTR(-EACCES);
|
|
|
|
|
2012-11-29 20:17:21 -07:00
|
|
|
if (unlikely(name[0] == '.')) {
|
|
|
|
if (len < 2 || (len == 2 && name[1] == '.'))
|
|
|
|
return ERR_PTR(-EACCES);
|
|
|
|
}
|
|
|
|
|
2011-03-07 21:49:20 -07:00
|
|
|
while (len--) {
|
|
|
|
c = *(const unsigned char *)name++;
|
|
|
|
if (c == '/' || c == '\0')
|
|
|
|
return ERR_PTR(-EACCES);
|
|
|
|
}
|
2011-03-08 12:17:44 -07:00
|
|
|
/*
|
|
|
|
* See if the low-level filesystem might want
|
|
|
|
* to use its own hash..
|
|
|
|
*/
|
|
|
|
if (base->d_flags & DCACHE_OP_HASH) {
|
2013-05-21 16:22:44 -06:00
|
|
|
int err = base->d_op->d_hash(base, &this);
|
2011-03-08 12:17:44 -07:00
|
|
|
if (err < 0)
|
|
|
|
return ERR_PTR(err);
|
|
|
|
}
|
2007-10-17 00:25:38 -06:00
|
|
|
|
2012-03-26 04:54:21 -06:00
|
|
|
err = inode_permission(base->d_inode, MAY_EXEC);
|
|
|
|
if (err)
|
|
|
|
return ERR_PTR(err);
|
|
|
|
|
2012-06-10 15:17:17 -06:00
|
|
|
return __lookup_hash(&this, base, 0);
|
2007-04-26 01:12:05 -06:00
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(lookup_one_len);
|
2007-04-26 01:12:05 -06:00
|
|
|
|
2016-01-07 14:08:20 -07:00
|
|
|
/**
|
|
|
|
* lookup_one_len_unlocked - filesystem helper to lookup single pathname component
|
|
|
|
* @name: pathname component to lookup
|
|
|
|
* @base: base directory to lookup from
|
|
|
|
* @len: maximum length @len should be interpreted to
|
|
|
|
*
|
|
|
|
* Note that this routine is purely a helper for filesystem usage and should
|
|
|
|
* not be called by generic code.
|
|
|
|
*
|
|
|
|
* Unlike lookup_one_len, it should be called without the parent
|
|
|
|
* i_mutex held, and will take the i_mutex itself if necessary.
|
|
|
|
*/
|
|
|
|
struct dentry *lookup_one_len_unlocked(const char *name,
|
|
|
|
struct dentry *base, int len)
|
|
|
|
{
|
|
|
|
struct qstr this;
|
|
|
|
unsigned int c;
|
|
|
|
int err;
|
|
|
|
struct dentry *ret;
|
|
|
|
|
|
|
|
this.name = name;
|
|
|
|
this.len = len;
|
|
|
|
this.hash = full_name_hash(name, len);
|
|
|
|
if (!len)
|
|
|
|
return ERR_PTR(-EACCES);
|
|
|
|
|
|
|
|
if (unlikely(name[0] == '.')) {
|
|
|
|
if (len < 2 || (len == 2 && name[1] == '.'))
|
|
|
|
return ERR_PTR(-EACCES);
|
|
|
|
}
|
|
|
|
|
|
|
|
while (len--) {
|
|
|
|
c = *(const unsigned char *)name++;
|
|
|
|
if (c == '/' || c == '\0')
|
|
|
|
return ERR_PTR(-EACCES);
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* See if the low-level filesystem might want
|
|
|
|
* to use its own hash..
|
|
|
|
*/
|
|
|
|
if (base->d_flags & DCACHE_OP_HASH) {
|
|
|
|
int err = base->d_op->d_hash(base, &this);
|
|
|
|
if (err < 0)
|
|
|
|
return ERR_PTR(err);
|
|
|
|
}
|
|
|
|
|
|
|
|
err = inode_permission(base->d_inode, MAY_EXEC);
|
|
|
|
if (err)
|
|
|
|
return ERR_PTR(err);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* __d_lookup() is used to try to get a quick answer and avoid the
|
|
|
|
* mutex. A false-negative does no harm.
|
|
|
|
*/
|
|
|
|
ret = __d_lookup(base, &this);
|
|
|
|
if (ret && unlikely(ret->d_flags & DCACHE_OP_REVALIDATE)) {
|
|
|
|
dput(ret);
|
|
|
|
ret = NULL;
|
|
|
|
}
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_lock(base->d_inode);
|
2016-01-07 14:08:20 -07:00
|
|
|
ret = __lookup_hash(&this, base, 0);
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_unlock(base->d_inode);
|
2016-01-07 14:08:20 -07:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(lookup_one_len_unlocked);
|
|
|
|
|
2011-11-02 02:44:39 -06:00
|
|
|
int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
|
|
|
|
struct path *path, int *empty)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2015-05-12 14:53:42 -06:00
|
|
|
return filename_lookup(dfd, getname_flags(name, flags, empty),
|
|
|
|
flags, path, NULL);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
2015-05-13 07:12:02 -06:00
|
|
|
EXPORT_SYMBOL(user_path_at_empty);
|
2011-11-02 02:44:39 -06:00
|
|
|
|
2012-10-10 13:25:29 -06:00
|
|
|
/*
|
|
|
|
* NB: most callers don't do anything directly with the reference to the
|
|
|
|
* to struct filename, but the nd->last pointer points into the name string
|
|
|
|
* allocated by getname. So we must hold the reference to it until all
|
|
|
|
* path-walking is complete.
|
|
|
|
*/
|
2015-05-13 04:57:49 -06:00
|
|
|
static inline struct filename *
|
2015-04-30 14:09:11 -06:00
|
|
|
user_path_parent(int dfd, const char __user *path,
|
|
|
|
struct path *parent,
|
|
|
|
struct qstr *last,
|
|
|
|
int *type,
|
2012-12-11 10:10:09 -07:00
|
|
|
unsigned int flags)
|
2008-07-21 07:32:51 -06:00
|
|
|
{
|
2012-12-11 10:10:09 -07:00
|
|
|
/* only LOOKUP_REVAL is allowed in extra flags */
|
2015-05-12 15:32:54 -06:00
|
|
|
return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
|
|
|
|
parent, last, type);
|
2008-07-21 07:32:51 -06:00
|
|
|
}
|
|
|
|
|
2013-07-26 04:23:25 -06:00
|
|
|
/**
|
2013-09-08 12:03:27 -06:00
|
|
|
* mountpoint_last - look up last component for umount
|
2013-07-26 04:23:25 -06:00
|
|
|
* @nd: pathwalk nameidata - currently pointing at parent directory of "last"
|
|
|
|
* @path: pointer to container for result
|
|
|
|
*
|
|
|
|
* This is a special lookup_last function just for umount. In this case, we
|
|
|
|
* need to resolve the path without doing any revalidation.
|
|
|
|
*
|
|
|
|
* The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
|
|
|
|
* mountpoints are always pinned in the dcache, their ancestors are too. Thus,
|
|
|
|
* in almost all cases, this lookup will be served out of the dcache. The only
|
|
|
|
* cases where it won't are if nd->last refers to a symlink or the path is
|
|
|
|
* bogus and it doesn't exist.
|
|
|
|
*
|
|
|
|
* Returns:
|
|
|
|
* -error: if there was an error during lookup. This includes -ENOENT if the
|
|
|
|
* lookup found a negative dentry. The nd->path reference will also be
|
|
|
|
* put in this case.
|
|
|
|
*
|
|
|
|
* 0: if we successfully resolved nd->path and found it to not to be a
|
|
|
|
* symlink that needs to be followed. "path" will also be populated.
|
|
|
|
* The nd->path reference will also be put.
|
|
|
|
*
|
|
|
|
* 1: if we successfully resolved nd->last and found it to be a symlink
|
|
|
|
* that needs to be followed. "path" will be populated with the path
|
|
|
|
* to the link, and nd->path will *not* be put.
|
|
|
|
*/
|
|
|
|
static int
|
2013-09-08 12:03:27 -06:00
|
|
|
mountpoint_last(struct nameidata *nd, struct path *path)
|
2013-07-26 04:23:25 -06:00
|
|
|
{
|
|
|
|
int error = 0;
|
|
|
|
struct dentry *dentry;
|
|
|
|
struct dentry *dir = nd->path.dentry;
|
|
|
|
|
2013-09-08 11:41:33 -06:00
|
|
|
/* If we're in rcuwalk, drop out of it to handle last component */
|
|
|
|
if (nd->flags & LOOKUP_RCU) {
|
2015-05-05 07:26:05 -06:00
|
|
|
if (unlazy_walk(nd, NULL, 0))
|
2015-05-08 16:05:21 -06:00
|
|
|
return -ECHILD;
|
2013-07-26 04:23:25 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
nd->flags &= ~LOOKUP_PARENT;
|
|
|
|
|
|
|
|
if (unlikely(nd->last_type != LAST_NORM)) {
|
|
|
|
error = handle_dots(nd, nd->last_type);
|
2013-09-08 11:41:33 -06:00
|
|
|
if (error)
|
2015-05-08 16:05:21 -06:00
|
|
|
return error;
|
2013-09-08 11:41:33 -06:00
|
|
|
dentry = dget(nd->path.dentry);
|
|
|
|
goto done;
|
2013-07-26 04:23:25 -06:00
|
|
|
}
|
|
|
|
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_lock(dir->d_inode);
|
2013-07-26 04:23:25 -06:00
|
|
|
dentry = d_lookup(dir, &nd->last);
|
|
|
|
if (!dentry) {
|
|
|
|
/*
|
|
|
|
* No cached dentry. Mounted dentries are pinned in the cache,
|
|
|
|
* so that means that this dentry is probably a symlink or the
|
|
|
|
* path doesn't actually point to a mounted dentry.
|
|
|
|
*/
|
|
|
|
dentry = d_alloc(dir, &nd->last);
|
|
|
|
if (!dentry) {
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_unlock(dir->d_inode);
|
2015-05-08 16:05:21 -06:00
|
|
|
return -ENOMEM;
|
2013-07-26 04:23:25 -06:00
|
|
|
}
|
2013-09-08 11:41:33 -06:00
|
|
|
dentry = lookup_real(dir->d_inode, dentry, nd->flags);
|
2013-09-10 15:04:25 -06:00
|
|
|
if (IS_ERR(dentry)) {
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_unlock(dir->d_inode);
|
2015-05-08 16:05:21 -06:00
|
|
|
return PTR_ERR(dentry);
|
2013-09-10 15:04:25 -06:00
|
|
|
}
|
2013-07-26 04:23:25 -06:00
|
|
|
}
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_unlock(dir->d_inode);
|
2013-07-26 04:23:25 -06:00
|
|
|
|
2013-09-08 11:41:33 -06:00
|
|
|
done:
|
2015-03-17 11:33:52 -06:00
|
|
|
if (d_is_negative(dentry)) {
|
2013-09-08 11:41:33 -06:00
|
|
|
dput(dentry);
|
2015-05-08 16:05:21 -06:00
|
|
|
return -ENOENT;
|
2013-07-26 04:23:25 -06:00
|
|
|
}
|
2015-05-04 06:26:45 -06:00
|
|
|
if (nd->depth)
|
|
|
|
put_link(nd);
|
2013-09-08 11:41:33 -06:00
|
|
|
path->dentry = dentry;
|
2014-07-21 02:30:23 -06:00
|
|
|
path->mnt = nd->path.mnt;
|
2015-05-07 17:54:34 -06:00
|
|
|
error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
|
|
|
|
d_backing_inode(dentry), 0);
|
2015-05-08 16:05:21 -06:00
|
|
|
if (unlikely(error))
|
2015-05-04 16:13:23 -06:00
|
|
|
return error;
|
2014-07-21 02:30:23 -06:00
|
|
|
mntget(path->mnt);
|
2013-09-08 11:41:33 -06:00
|
|
|
follow_mount(path);
|
2015-05-08 16:05:21 -06:00
|
|
|
return 0;
|
2013-07-26 04:23:25 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
2013-09-08 12:03:27 -06:00
|
|
|
* path_mountpoint - look up a path to be umounted
|
2015-09-09 16:39:23 -06:00
|
|
|
* @nd: lookup context
|
2013-07-26 04:23:25 -06:00
|
|
|
* @flags: lookup flags
|
2015-05-12 16:43:07 -06:00
|
|
|
* @path: pointer to container for result
|
2013-07-26 04:23:25 -06:00
|
|
|
*
|
|
|
|
* Look up the given name, but don't attempt to revalidate the last component.
|
2013-10-19 15:56:55 -06:00
|
|
|
* Returns 0 and "path" will be valid on success; Returns error otherwise.
|
2013-07-26 04:23:25 -06:00
|
|
|
*/
|
|
|
|
static int
|
2015-05-12 16:43:07 -06:00
|
|
|
path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
|
2013-07-26 04:23:25 -06:00
|
|
|
{
|
2015-05-12 16:43:07 -06:00
|
|
|
const char *s = path_init(nd, flags);
|
2015-05-08 15:19:59 -06:00
|
|
|
int err;
|
|
|
|
if (IS_ERR(s))
|
|
|
|
return PTR_ERR(s);
|
2015-05-08 15:37:07 -06:00
|
|
|
while (!(err = link_path_walk(s, nd)) &&
|
|
|
|
(err = mountpoint_last(nd, path)) > 0) {
|
|
|
|
s = trailing_symlink(nd);
|
|
|
|
if (IS_ERR(s)) {
|
|
|
|
err = PTR_ERR(s);
|
2013-07-26 04:23:25 -06:00
|
|
|
break;
|
2015-05-08 15:37:07 -06:00
|
|
|
}
|
2013-07-26 04:23:25 -06:00
|
|
|
}
|
2015-05-08 16:05:21 -06:00
|
|
|
terminate_walk(nd);
|
2013-07-26 04:23:25 -06:00
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2013-09-08 18:18:44 -06:00
|
|
|
static int
|
2015-02-22 17:44:00 -07:00
|
|
|
filename_mountpoint(int dfd, struct filename *name, struct path *path,
|
2013-09-08 18:18:44 -06:00
|
|
|
unsigned int flags)
|
|
|
|
{
|
2015-05-13 05:28:08 -06:00
|
|
|
struct nameidata nd;
|
2015-01-22 00:49:00 -07:00
|
|
|
int error;
|
2015-02-22 17:44:00 -07:00
|
|
|
if (IS_ERR(name))
|
|
|
|
return PTR_ERR(name);
|
2015-05-13 05:28:08 -06:00
|
|
|
set_nameidata(&nd, dfd, name);
|
2015-05-12 16:43:07 -06:00
|
|
|
error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
|
2013-09-08 18:18:44 -06:00
|
|
|
if (unlikely(error == -ECHILD))
|
2015-05-12 16:43:07 -06:00
|
|
|
error = path_mountpoint(&nd, flags, path);
|
2013-09-08 18:18:44 -06:00
|
|
|
if (unlikely(error == -ESTALE))
|
2015-05-12 16:43:07 -06:00
|
|
|
error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
|
2013-09-08 18:18:44 -06:00
|
|
|
if (likely(!error))
|
2015-02-22 17:44:00 -07:00
|
|
|
audit_inode(name, path->dentry, 0);
|
2015-05-13 05:28:08 -06:00
|
|
|
restore_nameidata();
|
2015-02-22 17:44:00 -07:00
|
|
|
putname(name);
|
2013-09-08 18:18:44 -06:00
|
|
|
return error;
|
|
|
|
}
|
|
|
|
|
2013-07-26 04:23:25 -06:00
|
|
|
/**
|
2013-09-08 12:03:27 -06:00
|
|
|
* user_path_mountpoint_at - lookup a path from userland in order to umount it
|
2013-07-26 04:23:25 -06:00
|
|
|
* @dfd: directory file descriptor
|
|
|
|
* @name: pathname from userland
|
|
|
|
* @flags: lookup flags
|
|
|
|
* @path: pointer to container to hold result
|
|
|
|
*
|
|
|
|
* A umount is a special case for path walking. We're not actually interested
|
|
|
|
* in the inode in this situation, and ESTALE errors can be a problem. We
|
|
|
|
* simply want track down the dentry and vfsmount attached at the mountpoint
|
|
|
|
* and avoid revalidating the last component.
|
|
|
|
*
|
|
|
|
* Returns 0 and populates "path" on success.
|
|
|
|
*/
|
|
|
|
int
|
2013-09-08 12:03:27 -06:00
|
|
|
user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
|
2013-07-26 04:23:25 -06:00
|
|
|
struct path *path)
|
|
|
|
{
|
2015-01-22 00:49:00 -07:00
|
|
|
return filename_mountpoint(dfd, getname(name), path, flags);
|
2013-07-26 04:23:25 -06:00
|
|
|
}
|
|
|
|
|
2013-09-08 18:18:44 -06:00
|
|
|
int
|
|
|
|
kern_path_mountpoint(int dfd, const char *name, struct path *path,
|
|
|
|
unsigned int flags)
|
|
|
|
{
|
2015-01-22 00:49:00 -07:00
|
|
|
return filename_mountpoint(dfd, getname_kernel(name), path, flags);
|
2013-09-08 18:18:44 -06:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(kern_path_mountpoint);
|
|
|
|
|
2014-10-23 16:14:36 -06:00
|
|
|
int __check_sticky(struct inode *dir, struct inode *inode)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2012-03-03 22:17:15 -07:00
|
|
|
kuid_t fsuid = current_fsuid();
|
2008-11-13 16:39:05 -07:00
|
|
|
|
2012-03-03 22:17:15 -07:00
|
|
|
if (uid_eq(inode->i_uid, fsuid))
|
2005-04-16 16:20:36 -06:00
|
|
|
return 0;
|
2012-03-03 22:17:15 -07:00
|
|
|
if (uid_eq(dir->i_uid, fsuid))
|
2005-04-16 16:20:36 -06:00
|
|
|
return 0;
|
2014-06-10 13:45:42 -06:00
|
|
|
return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
2014-10-23 16:14:36 -06:00
|
|
|
EXPORT_SYMBOL(__check_sticky);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Check whether we can remove a link victim from directory dir, check
|
|
|
|
* whether the type of victim is right.
|
|
|
|
* 1. We can't do it if dir is read-only (done in permission())
|
|
|
|
* 2. We should have write and exec permissions on dir
|
|
|
|
* 3. We can't remove anything from append-only dir
|
|
|
|
* 4. We can't do anything with immutable dir (done in permission())
|
|
|
|
* 5. If the sticky bit on dir is set we should either
|
|
|
|
* a. be owner of dir, or
|
|
|
|
* b. be owner of victim, or
|
|
|
|
* c. have CAP_FOWNER capability
|
|
|
|
* 6. If the victim is append-only or immutable we can't do antyhing with
|
|
|
|
* links pointing to it.
|
|
|
|
* 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
|
|
|
|
* 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
|
|
|
|
* 9. We can't remove a root or mountpoint.
|
|
|
|
* 10. We don't allow removal of NFS sillyrenamed files; it's handled by
|
|
|
|
* nfs_async_unlink().
|
|
|
|
*/
|
2013-09-12 12:22:53 -06:00
|
|
|
static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2015-05-06 08:59:00 -06:00
|
|
|
struct inode *inode = d_backing_inode(victim);
|
2005-04-16 16:20:36 -06:00
|
|
|
int error;
|
|
|
|
|
2013-09-12 12:22:53 -06:00
|
|
|
if (d_is_negative(victim))
|
2005-04-16 16:20:36 -06:00
|
|
|
return -ENOENT;
|
2013-09-12 12:22:53 -06:00
|
|
|
BUG_ON(!inode);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
BUG_ON(victim->d_parent->d_inode != dir);
|
2012-10-10 13:25:25 -06:00
|
|
|
audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2008-07-21 22:07:17 -06:00
|
|
|
error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
if (IS_APPEND(dir))
|
|
|
|
return -EPERM;
|
2013-09-12 12:22:53 -06:00
|
|
|
|
|
|
|
if (check_sticky(dir, inode) || IS_APPEND(inode) ||
|
|
|
|
IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
|
2005-04-16 16:20:36 -06:00
|
|
|
return -EPERM;
|
|
|
|
if (isdir) {
|
2014-04-01 09:08:41 -06:00
|
|
|
if (!d_is_dir(victim))
|
2005-04-16 16:20:36 -06:00
|
|
|
return -ENOTDIR;
|
|
|
|
if (IS_ROOT(victim))
|
|
|
|
return -EBUSY;
|
2014-04-01 09:08:41 -06:00
|
|
|
} else if (d_is_dir(victim))
|
2005-04-16 16:20:36 -06:00
|
|
|
return -EISDIR;
|
|
|
|
if (IS_DEADDIR(dir))
|
|
|
|
return -ENOENT;
|
|
|
|
if (victim->d_flags & DCACHE_NFSFS_RENAMED)
|
|
|
|
return -EBUSY;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Check whether we can create an object with dentry child in directory
|
|
|
|
* dir.
|
|
|
|
* 1. We can't do it if child already exists (open has special treatment for
|
|
|
|
* this case, but since we are inlined it's OK)
|
|
|
|
* 2. We can't do it if dir is read-only (done in permission())
|
|
|
|
* 3. We should have write and exec permissions on dir
|
|
|
|
* 4. We can't do it if dir is immutable (done in permission())
|
|
|
|
*/
|
2008-07-30 07:08:48 -06:00
|
|
|
static inline int may_create(struct inode *dir, struct dentry *child)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2013-05-08 08:25:58 -06:00
|
|
|
audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (child->d_inode)
|
|
|
|
return -EEXIST;
|
|
|
|
if (IS_DEADDIR(dir))
|
|
|
|
return -ENOENT;
|
2008-07-21 22:07:17 -06:00
|
|
|
return inode_permission(dir, MAY_WRITE | MAY_EXEC);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* p1 and p2 should be directories on the same fs.
|
|
|
|
*/
|
|
|
|
struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
|
|
|
|
{
|
|
|
|
struct dentry *p;
|
|
|
|
|
|
|
|
if (p1 == p2) {
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
|
2005-04-16 16:20:36 -06:00
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2006-03-23 04:00:33 -07:00
|
|
|
mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2008-10-15 16:50:28 -06:00
|
|
|
p = d_ancestor(p2, p1);
|
|
|
|
if (p) {
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
|
|
|
|
inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
|
2008-10-15 16:50:28 -06:00
|
|
|
return p;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2008-10-15 16:50:28 -06:00
|
|
|
p = d_ancestor(p1, p2);
|
|
|
|
if (p) {
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
|
|
|
|
inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
|
2008-10-15 16:50:28 -06:00
|
|
|
return p;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
|
|
|
|
inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
|
2005-04-16 16:20:36 -06:00
|
|
|
return NULL;
|
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(lock_rename);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
void unlock_rename(struct dentry *p1, struct dentry *p2)
|
|
|
|
{
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_unlock(p1->d_inode);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (p1 != p2) {
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_unlock(p2->d_inode);
|
2006-03-23 04:00:33 -07:00
|
|
|
mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(unlock_rename);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2011-07-25 23:42:34 -06:00
|
|
|
int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
|
2012-06-10 16:09:36 -06:00
|
|
|
bool want_excl)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2008-07-30 07:08:48 -06:00
|
|
|
int error = may_create(dir, dentry);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
|
2008-12-04 08:06:33 -07:00
|
|
|
if (!dir->i_op->create)
|
2005-04-16 16:20:36 -06:00
|
|
|
return -EACCES; /* shouldn't it be ENOSYS? */
|
|
|
|
mode &= S_IALLUGO;
|
|
|
|
mode |= S_IFREG;
|
|
|
|
error = security_inode_create(dir, dentry, mode);
|
|
|
|
if (error)
|
|
|
|
return error;
|
2012-06-10 16:09:36 -06:00
|
|
|
error = dir->i_op->create(dir, dentry, mode, want_excl);
|
2005-09-09 14:01:44 -06:00
|
|
|
if (!error)
|
2005-11-03 08:57:06 -07:00
|
|
|
fsnotify_create(dir, dentry);
|
2005-04-16 16:20:36 -06:00
|
|
|
return error;
|
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(vfs_create);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2011-03-11 10:08:24 -07:00
|
|
|
static int may_open(struct path *path, int acc_mode, int flag)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2008-10-24 01:58:10 -06:00
|
|
|
struct dentry *dentry = path->dentry;
|
2005-04-16 16:20:36 -06:00
|
|
|
struct inode *inode = dentry->d_inode;
|
|
|
|
int error;
|
|
|
|
|
|
|
|
if (!inode)
|
|
|
|
return -ENOENT;
|
|
|
|
|
2009-01-05 11:27:23 -07:00
|
|
|
switch (inode->i_mode & S_IFMT) {
|
|
|
|
case S_IFLNK:
|
2005-04-16 16:20:36 -06:00
|
|
|
return -ELOOP;
|
2009-01-05 11:27:23 -07:00
|
|
|
case S_IFDIR:
|
|
|
|
if (acc_mode & MAY_WRITE)
|
|
|
|
return -EISDIR;
|
|
|
|
break;
|
|
|
|
case S_IFBLK:
|
|
|
|
case S_IFCHR:
|
2008-10-24 01:58:10 -06:00
|
|
|
if (path->mnt->mnt_flags & MNT_NODEV)
|
2005-04-16 16:20:36 -06:00
|
|
|
return -EACCES;
|
2009-01-05 11:27:23 -07:00
|
|
|
/*FALLTHRU*/
|
|
|
|
case S_IFIFO:
|
|
|
|
case S_IFSOCK:
|
2005-04-16 16:20:36 -06:00
|
|
|
flag &= ~O_TRUNC;
|
2009-01-05 11:27:23 -07:00
|
|
|
break;
|
2008-02-15 15:37:48 -07:00
|
|
|
}
|
2007-10-17 00:31:14 -06:00
|
|
|
|
2015-12-26 20:33:24 -07:00
|
|
|
error = inode_permission(inode, MAY_OPEN | acc_mode);
|
2007-10-17 00:31:14 -06:00
|
|
|
if (error)
|
|
|
|
return error;
|
2009-02-04 07:06:57 -07:00
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
/*
|
|
|
|
* An append-only file must be opened in append mode for writing.
|
|
|
|
*/
|
|
|
|
if (IS_APPEND(inode)) {
|
2009-12-24 04:47:55 -07:00
|
|
|
if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
|
2009-12-16 01:54:00 -07:00
|
|
|
return -EPERM;
|
2005-04-16 16:20:36 -06:00
|
|
|
if (flag & O_TRUNC)
|
2009-12-16 01:54:00 -07:00
|
|
|
return -EPERM;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
/* O_NOATIME can only be set by the owner or superuser */
|
2011-03-23 17:43:26 -06:00
|
|
|
if (flag & O_NOATIME && !inode_owner_or_capable(inode))
|
2009-12-16 01:54:00 -07:00
|
|
|
return -EPERM;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2011-09-21 08:58:13 -06:00
|
|
|
return 0;
|
2009-12-16 01:54:00 -07:00
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2010-12-07 14:19:50 -07:00
|
|
|
static int handle_truncate(struct file *filp)
|
2009-12-16 01:54:00 -07:00
|
|
|
{
|
2010-12-07 14:19:50 -07:00
|
|
|
struct path *path = &filp->f_path;
|
2009-12-16 01:54:00 -07:00
|
|
|
struct inode *inode = path->dentry->d_inode;
|
|
|
|
int error = get_write_access(inode);
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
/*
|
|
|
|
* Refuse to truncate files with mandatory locks held on them.
|
|
|
|
*/
|
2014-03-10 07:54:15 -06:00
|
|
|
error = locks_verify_locked(filp);
|
2009-12-16 01:54:00 -07:00
|
|
|
if (!error)
|
2010-06-01 22:24:43 -06:00
|
|
|
error = security_path_truncate(path);
|
2009-12-16 01:54:00 -07:00
|
|
|
if (!error) {
|
|
|
|
error = do_truncate(path->dentry, 0,
|
|
|
|
ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
|
2010-12-07 14:19:50 -07:00
|
|
|
filp);
|
2009-12-16 01:54:00 -07:00
|
|
|
}
|
|
|
|
put_write_access(inode);
|
2009-09-04 11:08:46 -06:00
|
|
|
return error;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2008-02-15 15:37:27 -07:00
|
|
|
static inline int open_to_namei_flags(int flag)
|
|
|
|
{
|
2011-06-25 17:15:54 -06:00
|
|
|
if ((flag & O_ACCMODE) == 3)
|
|
|
|
flag--;
|
2008-02-15 15:37:27 -07:00
|
|
|
return flag;
|
|
|
|
}
|
|
|
|
|
2012-06-05 07:10:17 -06:00
|
|
|
static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
|
|
|
|
{
|
|
|
|
int error = security_path_mknod(dir, dentry, mode, 0);
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
|
|
|
|
error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
|
|
|
|
return security_inode_create(dir->dentry->d_inode, dentry, mode);
|
|
|
|
}
|
|
|
|
|
2012-06-14 09:13:46 -06:00
|
|
|
/*
|
|
|
|
* Attempt to atomically look up, create and open a file from a negative
|
|
|
|
* dentry.
|
|
|
|
*
|
|
|
|
* Returns 0 if successful. The file will have been created and attached to
|
|
|
|
* @file by the filesystem calling finish_open().
|
|
|
|
*
|
|
|
|
* Returns 1 if the file was looked up only or didn't need creating. The
|
|
|
|
* caller will need to perform the open themselves. @path will have been
|
|
|
|
* updated to point to the new dentry. This may be negative.
|
|
|
|
*
|
|
|
|
* Returns an error code otherwise.
|
|
|
|
*/
|
2012-06-22 02:41:10 -06:00
|
|
|
static int atomic_open(struct nameidata *nd, struct dentry *dentry,
|
|
|
|
struct path *path, struct file *file,
|
|
|
|
const struct open_flags *op,
|
2012-07-30 14:53:35 -06:00
|
|
|
bool got_write, bool need_lookup,
|
2012-06-22 02:41:10 -06:00
|
|
|
int *opened)
|
2012-06-05 07:10:17 -06:00
|
|
|
{
|
|
|
|
struct inode *dir = nd->path.dentry->d_inode;
|
|
|
|
unsigned open_flag = open_to_namei_flags(op->open_flag);
|
|
|
|
umode_t mode;
|
|
|
|
int error;
|
|
|
|
int acc_mode;
|
|
|
|
int create_error = 0;
|
|
|
|
struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
|
2013-09-16 06:52:05 -06:00
|
|
|
bool excl;
|
2012-06-05 07:10:17 -06:00
|
|
|
|
|
|
|
BUG_ON(dentry->d_inode);
|
|
|
|
|
|
|
|
/* Don't create child dentry for a dead directory. */
|
|
|
|
if (unlikely(IS_DEADDIR(dir))) {
|
2012-06-22 02:41:10 -06:00
|
|
|
error = -ENOENT;
|
2012-06-05 07:10:17 -06:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2012-08-15 05:01:24 -06:00
|
|
|
mode = op->mode;
|
2012-06-05 07:10:17 -06:00
|
|
|
if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
|
|
|
|
mode &= ~current_umask();
|
|
|
|
|
2013-09-16 06:52:05 -06:00
|
|
|
excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
|
|
|
|
if (excl)
|
2012-06-05 07:10:17 -06:00
|
|
|
open_flag &= ~O_TRUNC;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Checking write permission is tricky, bacuse we don't know if we are
|
|
|
|
* going to actually need it: O_CREAT opens should work as long as the
|
|
|
|
* file exists. But checking existence breaks atomicity. The trick is
|
|
|
|
* to check access and if not granted clear O_CREAT from the flags.
|
|
|
|
*
|
|
|
|
* Another problem is returing the "right" error value (e.g. for an
|
|
|
|
* O_EXCL open we want to return EEXIST not EROFS).
|
|
|
|
*/
|
2012-07-30 14:53:35 -06:00
|
|
|
if (((open_flag & (O_CREAT | O_TRUNC)) ||
|
|
|
|
(open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
|
|
|
|
if (!(open_flag & O_CREAT)) {
|
2012-06-05 07:10:17 -06:00
|
|
|
/*
|
|
|
|
* No O_CREATE -> atomicity not a requirement -> fall
|
|
|
|
* back to lookup + open
|
|
|
|
*/
|
|
|
|
goto no_open;
|
|
|
|
} else if (open_flag & (O_EXCL | O_TRUNC)) {
|
|
|
|
/* Fall back and fail with the right error */
|
2012-07-30 14:53:35 -06:00
|
|
|
create_error = -EROFS;
|
2012-06-05 07:10:17 -06:00
|
|
|
goto no_open;
|
|
|
|
} else {
|
|
|
|
/* No side effects, safe to clear O_CREAT */
|
2012-07-30 14:53:35 -06:00
|
|
|
create_error = -EROFS;
|
2012-06-05 07:10:17 -06:00
|
|
|
open_flag &= ~O_CREAT;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (open_flag & O_CREAT) {
|
2012-08-15 05:01:24 -06:00
|
|
|
error = may_o_create(&nd->path, dentry, mode);
|
2012-06-05 07:10:17 -06:00
|
|
|
if (error) {
|
|
|
|
create_error = error;
|
|
|
|
if (open_flag & O_EXCL)
|
|
|
|
goto no_open;
|
|
|
|
open_flag &= ~O_CREAT;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (nd->flags & LOOKUP_DIRECTORY)
|
|
|
|
open_flag |= O_DIRECTORY;
|
|
|
|
|
2012-06-22 02:40:19 -06:00
|
|
|
file->f_path.dentry = DENTRY_NOT_SET;
|
|
|
|
file->f_path.mnt = nd->path.mnt;
|
|
|
|
error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
|
2012-06-10 03:01:45 -06:00
|
|
|
opened);
|
2012-06-22 02:39:14 -06:00
|
|
|
if (error < 0) {
|
|
|
|
if (create_error && error == -ENOENT)
|
|
|
|
error = create_error;
|
2012-06-05 07:10:17 -06:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2012-06-22 02:39:14 -06:00
|
|
|
if (error) { /* returned 1, that is */
|
2012-06-22 02:40:19 -06:00
|
|
|
if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
|
2012-06-22 02:41:10 -06:00
|
|
|
error = -EIO;
|
2012-06-05 07:10:17 -06:00
|
|
|
goto out;
|
|
|
|
}
|
2012-06-22 02:40:19 -06:00
|
|
|
if (file->f_path.dentry) {
|
2012-06-05 07:10:17 -06:00
|
|
|
dput(dentry);
|
2012-06-22 02:40:19 -06:00
|
|
|
dentry = file->f_path.dentry;
|
2012-06-05 07:10:17 -06:00
|
|
|
}
|
2013-09-16 17:22:33 -06:00
|
|
|
if (*opened & FILE_CREATED)
|
|
|
|
fsnotify_create(dir, dentry);
|
|
|
|
if (!dentry->d_inode) {
|
|
|
|
WARN_ON(*opened & FILE_CREATED);
|
|
|
|
if (create_error) {
|
|
|
|
error = create_error;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if (excl && !(*opened & FILE_CREATED)) {
|
|
|
|
error = -EEXIST;
|
|
|
|
goto out;
|
|
|
|
}
|
2012-08-15 14:30:12 -06:00
|
|
|
}
|
2012-06-05 07:10:17 -06:00
|
|
|
goto looked_up;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We didn't have the inode before the open, so check open permission
|
|
|
|
* here.
|
|
|
|
*/
|
2013-09-16 17:22:33 -06:00
|
|
|
acc_mode = op->acc_mode;
|
|
|
|
if (*opened & FILE_CREATED) {
|
|
|
|
WARN_ON(!(open_flag & O_CREAT));
|
|
|
|
fsnotify_create(dir, dentry);
|
2015-12-26 20:33:24 -07:00
|
|
|
acc_mode = 0;
|
2013-09-16 17:22:33 -06:00
|
|
|
}
|
2012-06-22 02:41:10 -06:00
|
|
|
error = may_open(&file->f_path, acc_mode, open_flag);
|
|
|
|
if (error)
|
|
|
|
fput(file);
|
2012-06-05 07:10:17 -06:00
|
|
|
|
|
|
|
out:
|
|
|
|
dput(dentry);
|
2012-06-22 02:41:10 -06:00
|
|
|
return error;
|
2012-06-05 07:10:17 -06:00
|
|
|
|
|
|
|
no_open:
|
|
|
|
if (need_lookup) {
|
2012-06-10 15:17:17 -06:00
|
|
|
dentry = lookup_real(dir, dentry, nd->flags);
|
2012-06-05 07:10:17 -06:00
|
|
|
if (IS_ERR(dentry))
|
2012-06-22 02:41:10 -06:00
|
|
|
return PTR_ERR(dentry);
|
2012-06-05 07:10:17 -06:00
|
|
|
|
|
|
|
if (create_error) {
|
|
|
|
int open_flag = op->open_flag;
|
|
|
|
|
2012-06-22 02:41:10 -06:00
|
|
|
error = create_error;
|
2012-06-05 07:10:17 -06:00
|
|
|
if ((open_flag & O_EXCL)) {
|
|
|
|
if (!dentry->d_inode)
|
|
|
|
goto out;
|
|
|
|
} else if (!dentry->d_inode) {
|
|
|
|
goto out;
|
|
|
|
} else if ((open_flag & O_TRUNC) &&
|
VFS: (Scripted) Convert S_ISLNK/DIR/REG(dentry->d_inode) to d_is_*(dentry)
Convert the following where appropriate:
(1) S_ISLNK(dentry->d_inode) to d_is_symlink(dentry).
(2) S_ISREG(dentry->d_inode) to d_is_reg(dentry).
(3) S_ISDIR(dentry->d_inode) to d_is_dir(dentry). This is actually more
complicated than it appears as some calls should be converted to
d_can_lookup() instead. The difference is whether the directory in
question is a real dir with a ->lookup op or whether it's a fake dir with
a ->d_automount op.
In some circumstances, we can subsume checks for dentry->d_inode not being
NULL into this, provided we the code isn't in a filesystem that expects
d_inode to be NULL if the dirent really *is* negative (ie. if we're going to
use d_inode() rather than d_backing_inode() to get the inode pointer).
Note that the dentry type field may be set to something other than
DCACHE_MISS_TYPE when d_inode is NULL in the case of unionmount, where the VFS
manages the fall-through from a negative dentry to a lower layer. In such a
case, the dentry type of the negative union dentry is set to the same as the
type of the lower dentry.
However, if you know d_inode is not NULL at the call site, then you can use
the d_is_xxx() functions even in a filesystem.
There is one further complication: a 0,0 chardev dentry may be labelled
DCACHE_WHITEOUT_TYPE rather than DCACHE_SPECIAL_TYPE. Strictly, this was
intended for special directory entry types that don't have attached inodes.
The following perl+coccinelle script was used:
use strict;
my @callers;
open($fd, 'git grep -l \'S_IS[A-Z].*->d_inode\' |') ||
die "Can't grep for S_ISDIR and co. callers";
@callers = <$fd>;
close($fd);
unless (@callers) {
print "No matches\n";
exit(0);
}
my @cocci = (
'@@',
'expression E;',
'@@',
'',
'- S_ISLNK(E->d_inode->i_mode)',
'+ d_is_symlink(E)',
'',
'@@',
'expression E;',
'@@',
'',
'- S_ISDIR(E->d_inode->i_mode)',
'+ d_is_dir(E)',
'',
'@@',
'expression E;',
'@@',
'',
'- S_ISREG(E->d_inode->i_mode)',
'+ d_is_reg(E)' );
my $coccifile = "tmp.sp.cocci";
open($fd, ">$coccifile") || die $coccifile;
print($fd "$_\n") || die $coccifile foreach (@cocci);
close($fd);
foreach my $file (@callers) {
chomp $file;
print "Processing ", $file, "\n";
system("spatch", "--sp-file", $coccifile, $file, "--in-place", "--no-show-diff") == 0 ||
die "spatch failed";
}
[AV: overlayfs parts skipped]
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2015-01-29 05:02:35 -07:00
|
|
|
d_is_reg(dentry)) {
|
2012-06-05 07:10:17 -06:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
/* will fail later, go on to get the right error */
|
|
|
|
}
|
|
|
|
}
|
|
|
|
looked_up:
|
|
|
|
path->dentry = dentry;
|
|
|
|
path->mnt = nd->path.mnt;
|
2012-06-22 02:41:10 -06:00
|
|
|
return 1;
|
2012-06-05 07:10:17 -06:00
|
|
|
}
|
|
|
|
|
2012-06-05 07:10:15 -06:00
|
|
|
/*
|
2012-06-14 09:13:46 -06:00
|
|
|
* Look up and maybe create and open the last component.
|
2012-06-05 07:10:15 -06:00
|
|
|
*
|
|
|
|
* Must be called with i_mutex held on parent.
|
|
|
|
*
|
2012-06-14 09:13:46 -06:00
|
|
|
* Returns 0 if the file was successfully atomically created (if necessary) and
|
|
|
|
* opened. In this case the file will be returned attached to @file.
|
|
|
|
*
|
|
|
|
* Returns 1 if the file was not completely opened at this time, though lookups
|
|
|
|
* and creations will have been performed and the dentry returned in @path will
|
|
|
|
* be positive upon return if O_CREAT was specified. If O_CREAT wasn't
|
|
|
|
* specified then a negative dentry may be returned.
|
|
|
|
*
|
|
|
|
* An error code is returned otherwise.
|
|
|
|
*
|
|
|
|
* FILE_CREATE will be set in @*opened if the dentry was created and will be
|
|
|
|
* cleared otherwise prior to returning.
|
2012-06-05 07:10:15 -06:00
|
|
|
*/
|
2012-06-22 02:41:10 -06:00
|
|
|
static int lookup_open(struct nameidata *nd, struct path *path,
|
|
|
|
struct file *file,
|
|
|
|
const struct open_flags *op,
|
2012-07-30 14:53:35 -06:00
|
|
|
bool got_write, int *opened)
|
2012-06-05 07:10:15 -06:00
|
|
|
{
|
|
|
|
struct dentry *dir = nd->path.dentry;
|
2012-06-05 07:10:16 -06:00
|
|
|
struct inode *dir_inode = dir->d_inode;
|
2012-06-05 07:10:15 -06:00
|
|
|
struct dentry *dentry;
|
|
|
|
int error;
|
2016-03-05 18:09:32 -07:00
|
|
|
bool need_lookup = false;
|
2012-06-05 07:10:15 -06:00
|
|
|
|
2012-06-10 03:01:45 -06:00
|
|
|
*opened &= ~FILE_CREATED;
|
2016-03-05 18:09:32 -07:00
|
|
|
dentry = lookup_dcache(&nd->last, dir, nd->flags);
|
2012-06-05 07:10:15 -06:00
|
|
|
if (IS_ERR(dentry))
|
2012-06-22 02:41:10 -06:00
|
|
|
return PTR_ERR(dentry);
|
2012-06-05 07:10:15 -06:00
|
|
|
|
2016-03-05 18:09:32 -07:00
|
|
|
if (!dentry) {
|
|
|
|
dentry = d_alloc(dir, &nd->last);
|
|
|
|
if (unlikely(!dentry))
|
|
|
|
return -ENOMEM;
|
|
|
|
need_lookup = true;
|
|
|
|
} else if (dentry->d_inode) {
|
|
|
|
/* Cached positive dentry: will open in f_op->open */
|
2012-06-05 07:10:17 -06:00
|
|
|
goto out_no_open;
|
2016-03-05 18:09:32 -07:00
|
|
|
}
|
2012-06-05 07:10:17 -06:00
|
|
|
|
|
|
|
if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
|
2012-07-30 14:53:35 -06:00
|
|
|
return atomic_open(nd, dentry, path, file, op, got_write,
|
2012-06-10 03:01:45 -06:00
|
|
|
need_lookup, opened);
|
2012-06-05 07:10:17 -06:00
|
|
|
}
|
|
|
|
|
2012-06-05 07:10:16 -06:00
|
|
|
if (need_lookup) {
|
|
|
|
BUG_ON(dentry->d_inode);
|
|
|
|
|
2012-06-10 15:17:17 -06:00
|
|
|
dentry = lookup_real(dir_inode, dentry, nd->flags);
|
2012-06-05 07:10:16 -06:00
|
|
|
if (IS_ERR(dentry))
|
2012-06-22 02:41:10 -06:00
|
|
|
return PTR_ERR(dentry);
|
2012-06-05 07:10:16 -06:00
|
|
|
}
|
|
|
|
|
2012-06-05 07:10:15 -06:00
|
|
|
/* Negative dentry, just create the file */
|
|
|
|
if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
|
|
|
|
umode_t mode = op->mode;
|
|
|
|
if (!IS_POSIXACL(dir->d_inode))
|
|
|
|
mode &= ~current_umask();
|
|
|
|
/*
|
|
|
|
* This write is needed to ensure that a
|
|
|
|
* rw->ro transition does not occur between
|
|
|
|
* the time when the file is created and when
|
|
|
|
* a permanent write count is taken through
|
2012-06-05 07:10:27 -06:00
|
|
|
* the 'struct file' in finish_open().
|
2012-06-05 07:10:15 -06:00
|
|
|
*/
|
2012-07-30 14:53:35 -06:00
|
|
|
if (!got_write) {
|
|
|
|
error = -EROFS;
|
2012-06-05 07:10:15 -06:00
|
|
|
goto out_dput;
|
2012-07-30 14:53:35 -06:00
|
|
|
}
|
2012-06-10 03:01:45 -06:00
|
|
|
*opened |= FILE_CREATED;
|
2012-06-05 07:10:15 -06:00
|
|
|
error = security_path_mknod(&nd->path, dentry, mode, 0);
|
|
|
|
if (error)
|
|
|
|
goto out_dput;
|
2012-06-10 16:09:36 -06:00
|
|
|
error = vfs_create(dir->d_inode, dentry, mode,
|
|
|
|
nd->flags & LOOKUP_EXCL);
|
2012-06-05 07:10:15 -06:00
|
|
|
if (error)
|
|
|
|
goto out_dput;
|
|
|
|
}
|
2012-06-05 07:10:17 -06:00
|
|
|
out_no_open:
|
2012-06-05 07:10:15 -06:00
|
|
|
path->dentry = dentry;
|
|
|
|
path->mnt = nd->path.mnt;
|
2012-06-22 02:41:10 -06:00
|
|
|
return 1;
|
2012-06-05 07:10:15 -06:00
|
|
|
|
|
|
|
out_dput:
|
|
|
|
dput(dentry);
|
2012-06-22 02:41:10 -06:00
|
|
|
return error;
|
2012-06-05 07:10:15 -06:00
|
|
|
}
|
|
|
|
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
/*
|
2011-03-05 20:58:25 -07:00
|
|
|
* Handle the last step of open()
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
*/
|
2015-04-22 16:02:17 -06:00
|
|
|
static int do_last(struct nameidata *nd,
|
2012-06-22 02:41:10 -06:00
|
|
|
struct file *file, const struct open_flags *op,
|
2015-05-12 16:44:32 -06:00
|
|
|
int *opened)
|
2009-12-23 23:58:28 -07:00
|
|
|
{
|
2009-12-24 00:12:06 -07:00
|
|
|
struct dentry *dir = nd->path.dentry;
|
2011-03-08 22:36:45 -07:00
|
|
|
int open_flag = op->open_flag;
|
2012-06-05 07:10:30 -06:00
|
|
|
bool will_truncate = (open_flag & O_TRUNC) != 0;
|
2012-07-30 14:53:35 -06:00
|
|
|
bool got_write = false;
|
2011-03-13 14:42:14 -06:00
|
|
|
int acc_mode = op->acc_mode;
|
2015-05-05 07:40:46 -06:00
|
|
|
unsigned seq;
|
2012-05-21 09:30:07 -06:00
|
|
|
struct inode *inode;
|
2012-05-21 09:30:19 -06:00
|
|
|
struct path save_parent = { .dentry = NULL, .mnt = NULL };
|
2015-04-22 16:02:17 -06:00
|
|
|
struct path path;
|
2012-05-21 09:30:19 -06:00
|
|
|
bool retried = false;
|
2011-02-22 13:50:10 -07:00
|
|
|
int error;
|
2009-12-26 08:56:19 -07:00
|
|
|
|
2011-02-23 11:39:45 -07:00
|
|
|
nd->flags &= ~LOOKUP_PARENT;
|
|
|
|
nd->flags |= op->intent;
|
|
|
|
|
2013-06-06 07:12:33 -06:00
|
|
|
if (nd->last_type != LAST_NORM) {
|
2011-03-05 20:58:25 -07:00
|
|
|
error = handle_dots(nd, nd->last_type);
|
2015-05-08 16:05:21 -06:00
|
|
|
if (unlikely(error))
|
2012-06-22 02:41:10 -06:00
|
|
|
return error;
|
2012-06-05 07:10:29 -06:00
|
|
|
goto finish_open;
|
2009-12-26 08:56:19 -07:00
|
|
|
}
|
2009-12-26 05:01:01 -07:00
|
|
|
|
2011-03-08 22:36:45 -07:00
|
|
|
if (!(open_flag & O_CREAT)) {
|
2011-03-05 20:58:25 -07:00
|
|
|
if (nd->last.name[nd->last.len])
|
|
|
|
nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
|
|
|
|
/* we _can_ be in RCU mode here */
|
2015-05-05 07:40:46 -06:00
|
|
|
error = lookup_fast(nd, &path, &inode, &seq);
|
2016-03-05 20:04:59 -07:00
|
|
|
if (likely(error > 0))
|
2012-06-05 07:10:14 -06:00
|
|
|
goto finish_lookup;
|
|
|
|
|
|
|
|
if (error < 0)
|
2015-05-08 16:05:21 -06:00
|
|
|
return error;
|
2012-06-05 07:10:14 -06:00
|
|
|
|
|
|
|
BUG_ON(nd->inode != dir->d_inode);
|
2016-03-05 16:14:03 -07:00
|
|
|
BUG_ON(nd->flags & LOOKUP_RCU);
|
2012-06-05 07:10:13 -06:00
|
|
|
} else {
|
|
|
|
/* create side of things */
|
|
|
|
/*
|
|
|
|
* This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
|
|
|
|
* has been cleared when we got to the last component we are
|
|
|
|
* about to look up
|
|
|
|
*/
|
|
|
|
error = complete_walk(nd);
|
2015-05-08 14:28:42 -06:00
|
|
|
if (error)
|
2012-06-22 02:41:10 -06:00
|
|
|
return error;
|
2011-03-05 20:58:25 -07:00
|
|
|
|
2015-05-12 16:44:32 -06:00
|
|
|
audit_inode(nd->name, dir, LOOKUP_PARENT);
|
2012-06-05 07:10:13 -06:00
|
|
|
/* trailing slashes? */
|
2015-05-08 16:05:21 -06:00
|
|
|
if (unlikely(nd->last.name[nd->last.len]))
|
|
|
|
return -EISDIR;
|
2012-06-05 07:10:13 -06:00
|
|
|
}
|
2009-12-24 01:39:50 -07:00
|
|
|
|
2012-05-21 09:30:19 -06:00
|
|
|
retry_lookup:
|
2012-07-30 14:53:35 -06:00
|
|
|
if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
|
|
|
|
error = mnt_want_write(nd->path.mnt);
|
|
|
|
if (!error)
|
|
|
|
got_write = true;
|
|
|
|
/*
|
|
|
|
* do _not_ fail yet - we might not need that or fail with
|
|
|
|
* a different error; let lookup_open() decide; we'll be
|
|
|
|
* dropping this one anyway.
|
|
|
|
*/
|
|
|
|
}
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_lock(dir->d_inode);
|
2015-04-22 16:02:17 -06:00
|
|
|
error = lookup_open(nd, &path, file, op, got_write, opened);
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_unlock(dir->d_inode);
|
2009-12-24 00:12:06 -07:00
|
|
|
|
2012-06-22 02:41:10 -06:00
|
|
|
if (error <= 0) {
|
|
|
|
if (error)
|
2012-06-05 07:10:17 -06:00
|
|
|
goto out;
|
|
|
|
|
2012-06-10 03:01:45 -06:00
|
|
|
if ((*opened & FILE_CREATED) ||
|
2013-01-23 15:07:38 -07:00
|
|
|
!S_ISREG(file_inode(file)->i_mode))
|
2012-06-05 07:10:30 -06:00
|
|
|
will_truncate = false;
|
2012-06-05 07:10:17 -06:00
|
|
|
|
2015-05-12 16:44:32 -06:00
|
|
|
audit_inode(nd->name, file->f_path.dentry, 0);
|
2012-06-05 07:10:17 -06:00
|
|
|
goto opened;
|
|
|
|
}
|
2009-12-23 23:58:28 -07:00
|
|
|
|
2012-06-10 03:01:45 -06:00
|
|
|
if (*opened & FILE_CREATED) {
|
2011-03-08 22:17:27 -07:00
|
|
|
/* Don't check for write permission, don't truncate */
|
2011-03-08 22:36:45 -07:00
|
|
|
open_flag &= ~O_TRUNC;
|
2012-06-05 07:10:30 -06:00
|
|
|
will_truncate = false;
|
2015-12-26 20:33:24 -07:00
|
|
|
acc_mode = 0;
|
2015-04-22 16:02:17 -06:00
|
|
|
path_to_nameidata(&path, nd);
|
2012-06-05 07:10:29 -06:00
|
|
|
goto finish_open_created;
|
2009-12-23 23:58:28 -07:00
|
|
|
}
|
|
|
|
|
2012-06-05 07:10:17 -06:00
|
|
|
/*
|
|
|
|
* If atomic_open() acquired write access it is dropped now due to
|
|
|
|
* possible mount and symlink following (this might be optimized away if
|
|
|
|
* necessary...)
|
|
|
|
*/
|
2012-07-30 14:53:35 -06:00
|
|
|
if (got_write) {
|
2012-06-05 07:10:17 -06:00
|
|
|
mnt_drop_write(nd->path.mnt);
|
2012-07-30 14:53:35 -06:00
|
|
|
got_write = false;
|
2012-06-05 07:10:17 -06:00
|
|
|
}
|
|
|
|
|
2016-03-05 16:14:03 -07:00
|
|
|
if (unlikely(d_is_negative(path.dentry))) {
|
|
|
|
path_to_nameidata(&path, nd);
|
|
|
|
return -ENOENT;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* create/update audit record if it already exists.
|
|
|
|
*/
|
|
|
|
audit_inode(nd->name, path.dentry, 0);
|
|
|
|
|
2015-05-08 16:05:21 -06:00
|
|
|
if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
|
|
|
|
path_to_nameidata(&path, nd);
|
|
|
|
return -EEXIST;
|
|
|
|
}
|
2009-12-23 23:58:28 -07:00
|
|
|
|
2015-03-22 20:37:38 -06:00
|
|
|
error = follow_managed(&path, nd);
|
2015-05-08 16:05:21 -06:00
|
|
|
if (unlikely(error < 0))
|
|
|
|
return error;
|
2011-11-07 14:21:26 -07:00
|
|
|
|
2015-05-05 07:40:46 -06:00
|
|
|
seq = 0; /* out of RCU mode, so the value doesn't matter */
|
2016-02-27 17:23:16 -07:00
|
|
|
inode = d_backing_inode(path.dentry);
|
2015-05-07 17:24:57 -06:00
|
|
|
finish_lookup:
|
2015-05-04 16:13:23 -06:00
|
|
|
if (nd->depth)
|
|
|
|
put_link(nd);
|
2015-05-07 17:54:34 -06:00
|
|
|
error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
|
|
|
|
inode, seq);
|
2015-05-08 16:05:21 -06:00
|
|
|
if (unlikely(error))
|
2015-05-04 16:13:23 -06:00
|
|
|
return error;
|
2009-12-23 23:58:28 -07:00
|
|
|
|
2015-04-22 16:02:17 -06:00
|
|
|
if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
|
|
|
|
path_to_nameidata(&path, nd);
|
2012-05-21 09:30:19 -06:00
|
|
|
} else {
|
|
|
|
save_parent.dentry = nd->path.dentry;
|
2015-04-22 16:02:17 -06:00
|
|
|
save_parent.mnt = mntget(path.mnt);
|
|
|
|
nd->path.dentry = path.dentry;
|
2012-05-21 09:30:19 -06:00
|
|
|
|
|
|
|
}
|
2012-05-21 09:30:08 -06:00
|
|
|
nd->inode = inode;
|
2015-05-05 07:40:46 -06:00
|
|
|
nd->seq = seq;
|
2011-11-07 14:21:26 -07:00
|
|
|
/* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
|
2013-06-06 07:12:33 -06:00
|
|
|
finish_open:
|
2011-11-07 14:21:26 -07:00
|
|
|
error = complete_walk(nd);
|
2012-05-21 09:30:19 -06:00
|
|
|
if (error) {
|
|
|
|
path_put(&save_parent);
|
2012-06-22 02:41:10 -06:00
|
|
|
return error;
|
2012-05-21 09:30:19 -06:00
|
|
|
}
|
2015-05-12 16:44:32 -06:00
|
|
|
audit_inode(nd->name, nd->path.dentry, 0);
|
2016-02-27 17:37:37 -07:00
|
|
|
if (unlikely(d_is_symlink(nd->path.dentry)) && !(open_flag & O_PATH)) {
|
|
|
|
error = -ELOOP;
|
|
|
|
goto out;
|
|
|
|
}
|
2009-12-23 23:58:28 -07:00
|
|
|
error = -EISDIR;
|
2014-04-01 09:08:41 -06:00
|
|
|
if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
|
2012-06-22 02:41:10 -06:00
|
|
|
goto out;
|
2012-05-21 09:30:11 -06:00
|
|
|
error = -ENOTDIR;
|
2014-04-01 09:08:41 -06:00
|
|
|
if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
|
2012-06-22 02:41:10 -06:00
|
|
|
goto out;
|
2015-03-17 16:16:40 -06:00
|
|
|
if (!d_is_reg(nd->path.dentry))
|
2012-06-05 07:10:30 -06:00
|
|
|
will_truncate = false;
|
2011-03-08 22:59:59 -07:00
|
|
|
|
2011-03-08 22:13:14 -07:00
|
|
|
if (will_truncate) {
|
|
|
|
error = mnt_want_write(nd->path.mnt);
|
|
|
|
if (error)
|
2012-06-22 02:41:10 -06:00
|
|
|
goto out;
|
2012-07-30 14:53:35 -06:00
|
|
|
got_write = true;
|
2011-03-08 22:13:14 -07:00
|
|
|
}
|
2012-06-05 07:10:29 -06:00
|
|
|
finish_open_created:
|
2015-12-26 20:33:24 -07:00
|
|
|
if (likely(!(open_flag & O_PATH))) {
|
|
|
|
error = may_open(&nd->path, acc_mode, open_flag);
|
|
|
|
if (error)
|
|
|
|
goto out;
|
|
|
|
}
|
2014-10-23 16:14:35 -06:00
|
|
|
BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
|
|
|
|
error = vfs_open(&nd->path, file, current_cred());
|
|
|
|
if (!error) {
|
|
|
|
*opened |= FILE_OPENED;
|
|
|
|
} else {
|
2012-06-22 02:40:19 -06:00
|
|
|
if (error == -EOPENSTALE)
|
2012-06-05 07:10:31 -06:00
|
|
|
goto stale_open;
|
2012-06-05 07:10:27 -06:00
|
|
|
goto out;
|
2012-06-05 07:10:31 -06:00
|
|
|
}
|
2012-06-05 07:10:32 -06:00
|
|
|
opened:
|
2012-06-22 02:41:10 -06:00
|
|
|
error = open_check_o_direct(file);
|
2012-06-05 07:10:27 -06:00
|
|
|
if (error)
|
|
|
|
goto exit_fput;
|
2014-06-27 09:15:44 -06:00
|
|
|
error = ima_file_check(file, op->acc_mode, *opened);
|
2012-06-05 07:10:28 -06:00
|
|
|
if (error)
|
|
|
|
goto exit_fput;
|
|
|
|
|
|
|
|
if (will_truncate) {
|
2012-06-22 02:41:10 -06:00
|
|
|
error = handle_truncate(file);
|
2012-06-05 07:10:28 -06:00
|
|
|
if (error)
|
|
|
|
goto exit_fput;
|
2011-03-08 22:13:14 -07:00
|
|
|
}
|
2011-03-08 22:36:45 -07:00
|
|
|
out:
|
2016-02-27 17:17:33 -07:00
|
|
|
if (unlikely(error > 0)) {
|
|
|
|
WARN_ON(1);
|
|
|
|
error = -EINVAL;
|
|
|
|
}
|
2012-07-30 14:53:35 -06:00
|
|
|
if (got_write)
|
2011-03-08 22:13:14 -07:00
|
|
|
mnt_drop_write(nd->path.mnt);
|
2012-05-21 09:30:19 -06:00
|
|
|
path_put(&save_parent);
|
2012-06-22 02:41:10 -06:00
|
|
|
return error;
|
2009-12-23 23:58:28 -07:00
|
|
|
|
2012-06-05 07:10:27 -06:00
|
|
|
exit_fput:
|
2012-06-22 02:41:10 -06:00
|
|
|
fput(file);
|
|
|
|
goto out;
|
2012-06-05 07:10:27 -06:00
|
|
|
|
2012-06-05 07:10:31 -06:00
|
|
|
stale_open:
|
|
|
|
/* If no saved parent or already retried then can't retry */
|
|
|
|
if (!save_parent.dentry || retried)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
BUG_ON(save_parent.dentry != dir);
|
|
|
|
path_put(&nd->path);
|
|
|
|
nd->path = save_parent;
|
|
|
|
nd->inode = dir->d_inode;
|
|
|
|
save_parent.mnt = NULL;
|
|
|
|
save_parent.dentry = NULL;
|
2012-07-30 14:53:35 -06:00
|
|
|
if (got_write) {
|
2012-06-05 07:10:31 -06:00
|
|
|
mnt_drop_write(nd->path.mnt);
|
2012-07-30 14:53:35 -06:00
|
|
|
got_write = false;
|
2012-06-05 07:10:31 -06:00
|
|
|
}
|
|
|
|
retried = true;
|
|
|
|
goto retry_lookup;
|
2009-12-23 23:58:28 -07:00
|
|
|
}
|
|
|
|
|
2015-05-12 16:43:07 -06:00
|
|
|
static int do_tmpfile(struct nameidata *nd, unsigned flags,
|
2013-06-06 23:20:27 -06:00
|
|
|
const struct open_flags *op,
|
|
|
|
struct file *file, int *opened)
|
|
|
|
{
|
|
|
|
static const struct qstr name = QSTR_INIT("/", 1);
|
2015-05-12 14:36:12 -06:00
|
|
|
struct dentry *child;
|
2013-06-06 23:20:27 -06:00
|
|
|
struct inode *dir;
|
2015-05-12 14:36:12 -06:00
|
|
|
struct path path;
|
2015-05-12 16:43:07 -06:00
|
|
|
int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
|
2013-06-06 23:20:27 -06:00
|
|
|
if (unlikely(error))
|
|
|
|
return error;
|
2015-05-12 14:36:12 -06:00
|
|
|
error = mnt_want_write(path.mnt);
|
2013-06-06 23:20:27 -06:00
|
|
|
if (unlikely(error))
|
|
|
|
goto out;
|
2015-05-12 14:36:12 -06:00
|
|
|
dir = path.dentry->d_inode;
|
2013-06-06 23:20:27 -06:00
|
|
|
/* we want directory to be writable */
|
2015-05-12 14:36:12 -06:00
|
|
|
error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
|
2013-06-06 23:20:27 -06:00
|
|
|
if (error)
|
|
|
|
goto out2;
|
|
|
|
if (!dir->i_op->tmpfile) {
|
|
|
|
error = -EOPNOTSUPP;
|
|
|
|
goto out2;
|
|
|
|
}
|
2015-05-12 14:36:12 -06:00
|
|
|
child = d_alloc(path.dentry, &name);
|
2013-06-06 23:20:27 -06:00
|
|
|
if (unlikely(!child)) {
|
|
|
|
error = -ENOMEM;
|
|
|
|
goto out2;
|
|
|
|
}
|
2015-05-12 14:36:12 -06:00
|
|
|
dput(path.dentry);
|
|
|
|
path.dentry = child;
|
|
|
|
error = dir->i_op->tmpfile(dir, child, op->mode);
|
2013-06-06 23:20:27 -06:00
|
|
|
if (error)
|
|
|
|
goto out2;
|
2015-05-12 16:43:07 -06:00
|
|
|
audit_inode(nd->name, child, 0);
|
fs: allow open(dir, O_TMPFILE|..., 0) with mode 0
The man page for open(2) indicates that when O_CREAT is specified, the
'mode' argument applies only to future accesses to the file:
Note that this mode applies only to future accesses of the newly
created file; the open() call that creates a read-only file
may well return a read/write file descriptor.
The man page for open(2) implies that 'mode' is treated identically by
O_CREAT and O_TMPFILE.
O_TMPFILE, however, behaves differently:
int fd = open("/tmp", O_TMPFILE | O_RDWR, 0);
assert(fd == -1);
assert(errno == EACCES);
int fd = open("/tmp", O_TMPFILE | O_RDWR, 0600);
assert(fd > 0);
For O_CREAT, do_last() sets acc_mode to MAY_OPEN only:
if (*opened & FILE_CREATED) {
/* Don't check for write permission, don't truncate */
open_flag &= ~O_TRUNC;
will_truncate = false;
acc_mode = MAY_OPEN;
path_to_nameidata(path, nd);
goto finish_open_created;
}
But for O_TMPFILE, do_tmpfile() passes the full op->acc_mode to
may_open().
This patch lines up the behavior of O_TMPFILE with O_CREAT. After the
inode is created, may_open() is called with acc_mode = MAY_OPEN, in
do_tmpfile().
A different, but related glibc bug revealed the discrepancy:
https://sourceware.org/bugzilla/show_bug.cgi?id=17523
The glibc lazily loads the 'mode' argument of open() and openat() using
va_arg() only if O_CREAT is present in 'flags' (to support both the 2
argument and the 3 argument forms of open; same idea for openat()).
However, the glibc ignores the 'mode' argument if O_TMPFILE is in
'flags'.
On x86_64, for open(), it magically works anyway, as 'mode' is in
RDX when entering open(), and is still in RDX on SYSCALL, which is where
the kernel looks for the 3rd argument of a syscall.
But openat() is not quite so lucky: 'mode' is in RCX when entering the
glibc wrapper for openat(), while the kernel looks for the 4th argument
of a syscall in R10. Indeed, the syscall calling convention differs from
the regular calling convention in this respect on x86_64. So the kernel
sees mode = 0 when trying to use glibc openat() with O_TMPFILE, and
fails with EACCES.
Signed-off-by: Eric Rannaud <e@nanocritical.com>
Acked-by: Andy Lutomirski <luto@amacapital.net>
Cc: stable@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-10-30 02:51:01 -06:00
|
|
|
/* Don't check for other permissions, the inode was just created */
|
2015-12-26 20:33:24 -07:00
|
|
|
error = may_open(&path, 0, op->open_flag);
|
2013-06-06 23:20:27 -06:00
|
|
|
if (error)
|
|
|
|
goto out2;
|
2015-05-12 14:36:12 -06:00
|
|
|
file->f_path.mnt = path.mnt;
|
|
|
|
error = finish_open(file, child, NULL, opened);
|
2013-06-06 23:20:27 -06:00
|
|
|
if (error)
|
|
|
|
goto out2;
|
|
|
|
error = open_check_o_direct(file);
|
2013-06-10 22:34:36 -06:00
|
|
|
if (error) {
|
2013-06-06 23:20:27 -06:00
|
|
|
fput(file);
|
2013-06-10 22:34:36 -06:00
|
|
|
} else if (!(op->open_flag & O_EXCL)) {
|
|
|
|
struct inode *inode = file_inode(file);
|
|
|
|
spin_lock(&inode->i_lock);
|
|
|
|
inode->i_state |= I_LINKABLE;
|
|
|
|
spin_unlock(&inode->i_lock);
|
|
|
|
}
|
2013-06-06 23:20:27 -06:00
|
|
|
out2:
|
2015-05-12 14:36:12 -06:00
|
|
|
mnt_drop_write(path.mnt);
|
2013-06-06 23:20:27 -06:00
|
|
|
out:
|
2015-05-12 14:36:12 -06:00
|
|
|
path_put(&path);
|
2013-06-06 23:20:27 -06:00
|
|
|
return error;
|
|
|
|
}
|
|
|
|
|
2015-05-12 16:43:07 -06:00
|
|
|
static struct file *path_openat(struct nameidata *nd,
|
|
|
|
const struct open_flags *op, unsigned flags)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2015-05-08 15:19:59 -06:00
|
|
|
const char *s;
|
2012-06-22 02:40:19 -06:00
|
|
|
struct file *file;
|
2012-06-10 03:01:45 -06:00
|
|
|
int opened = 0;
|
2011-02-23 15:54:08 -07:00
|
|
|
int error;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
|
2012-06-22 02:40:19 -06:00
|
|
|
file = get_empty_filp();
|
2013-02-14 18:41:04 -07:00
|
|
|
if (IS_ERR(file))
|
|
|
|
return file;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
|
2012-06-22 02:40:19 -06:00
|
|
|
file->f_flags = op->open_flag;
|
fs: rcu-walk for path lookup
Perform common cases of path lookups without any stores or locking in the
ancestor dentry elements. This is called rcu-walk, as opposed to the current
algorithm which is a refcount based walk, or ref-walk.
This results in far fewer atomic operations on every path element,
significantly improving path lookup performance. It also avoids cacheline
bouncing on common dentries, significantly improving scalability.
The overall design is like this:
* LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk.
* Take the RCU lock for the entire path walk, starting with the acquiring
of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are
not required for dentry persistence.
* synchronize_rcu is called when unregistering a filesystem, so we can
access d_ops and i_ops during rcu-walk.
* Similarly take the vfsmount lock for the entire path walk. So now mnt
refcounts are not required for persistence. Also we are free to perform mount
lookups, and to assume dentry mount points and mount roots are stable up and
down the path.
* Have a per-dentry seqlock to protect the dentry name, parent, and inode,
so we can load this tuple atomically, and also check whether any of its
members have changed.
* Dentry lookups (based on parent, candidate string tuple) recheck the parent
sequence after the child is found in case anything changed in the parent
during the path walk.
* inode is also RCU protected so we can load d_inode and use the inode for
limited things.
* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.
* i_op can be loaded.
When we reach the destination dentry, we lock it, recheck lookup sequence,
and increment its refcount and mountpoint refcount. RCU and vfsmount locks
are dropped. This is termed "dropping rcu-walk". If the dentry refcount does
not match, we can not drop rcu-walk gracefully at the current point in the
lokup, so instead return -ECHILD (for want of a better errno). This signals the
path walking code to re-do the entire lookup with a ref-walk.
Aside from the final dentry, there are other situations that may be encounted
where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take
a reference on the last good dentry) and continue with a ref-walk. Again, if
we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup
using ref-walk. But it is very important that we can continue with ref-walk
for most cases, particularly to avoid the overhead of double lookups, and to
gain the scalability advantages on common path elements (like cwd and root).
The cases where rcu-walk cannot continue are:
* NULL dentry (ie. any uncached path element)
* parent with d_inode->i_op->permission or ACLs
* dentries with d_revalidate
* Following links
In future patches, permission checks and d_revalidate become rcu-walk aware. It
may be possible eventually to make following links rcu-walk aware.
Uncached path elements will always require dropping to ref-walk mode, at the
very least because i_mutex needs to be grabbed, and objects allocated.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-06 23:49:52 -07:00
|
|
|
|
2013-07-13 03:26:37 -06:00
|
|
|
if (unlikely(file->f_flags & __O_TMPFILE)) {
|
2015-05-12 16:43:07 -06:00
|
|
|
error = do_tmpfile(nd, flags, op, file, &opened);
|
2015-05-08 20:53:15 -06:00
|
|
|
goto out2;
|
2013-06-06 23:20:27 -06:00
|
|
|
}
|
|
|
|
|
2015-05-12 16:43:07 -06:00
|
|
|
s = path_init(nd, flags);
|
2015-05-08 15:19:59 -06:00
|
|
|
if (IS_ERR(s)) {
|
|
|
|
put_filp(file);
|
|
|
|
return ERR_CAST(s);
|
|
|
|
}
|
2015-05-08 15:37:07 -06:00
|
|
|
while (!(error = link_path_walk(s, nd)) &&
|
2015-05-12 16:44:32 -06:00
|
|
|
(error = do_last(nd, file, op, &opened)) > 0) {
|
2011-03-11 10:08:24 -07:00
|
|
|
nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
|
2015-05-08 15:37:07 -06:00
|
|
|
s = trailing_symlink(nd);
|
|
|
|
if (IS_ERR(s)) {
|
|
|
|
error = PTR_ERR(s);
|
2012-06-22 02:41:10 -06:00
|
|
|
break;
|
2015-05-08 15:37:07 -06:00
|
|
|
}
|
2009-12-26 05:16:40 -07:00
|
|
|
}
|
2015-05-08 16:05:21 -06:00
|
|
|
terminate_walk(nd);
|
2015-05-08 20:53:15 -06:00
|
|
|
out2:
|
2012-06-22 02:41:10 -06:00
|
|
|
if (!(opened & FILE_OPENED)) {
|
|
|
|
BUG_ON(!error);
|
2012-06-22 02:40:19 -06:00
|
|
|
put_filp(file);
|
2012-05-21 09:30:19 -06:00
|
|
|
}
|
2012-06-22 02:41:10 -06:00
|
|
|
if (unlikely(error)) {
|
|
|
|
if (error == -EOPENSTALE) {
|
|
|
|
if (flags & LOOKUP_RCU)
|
|
|
|
error = -ECHILD;
|
|
|
|
else
|
|
|
|
error = -ESTALE;
|
|
|
|
}
|
|
|
|
file = ERR_PTR(error);
|
|
|
|
}
|
|
|
|
return file;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2012-10-10 14:43:10 -06:00
|
|
|
struct file *do_filp_open(int dfd, struct filename *pathname,
|
2013-06-10 22:23:01 -06:00
|
|
|
const struct open_flags *op)
|
2011-02-23 15:54:08 -07:00
|
|
|
{
|
2015-05-13 05:28:08 -06:00
|
|
|
struct nameidata nd;
|
2013-06-10 22:23:01 -06:00
|
|
|
int flags = op->lookup_flags;
|
2011-02-23 15:54:08 -07:00
|
|
|
struct file *filp;
|
|
|
|
|
2015-05-13 05:28:08 -06:00
|
|
|
set_nameidata(&nd, dfd, pathname);
|
2015-05-12 16:43:07 -06:00
|
|
|
filp = path_openat(&nd, op, flags | LOOKUP_RCU);
|
2011-02-23 15:54:08 -07:00
|
|
|
if (unlikely(filp == ERR_PTR(-ECHILD)))
|
2015-05-12 16:43:07 -06:00
|
|
|
filp = path_openat(&nd, op, flags);
|
2011-02-23 15:54:08 -07:00
|
|
|
if (unlikely(filp == ERR_PTR(-ESTALE)))
|
2015-05-12 16:43:07 -06:00
|
|
|
filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
|
2015-05-13 05:28:08 -06:00
|
|
|
restore_nameidata();
|
2011-02-23 15:54:08 -07:00
|
|
|
return filp;
|
|
|
|
}
|
|
|
|
|
2011-03-11 10:08:24 -07:00
|
|
|
struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
|
2013-06-10 22:23:01 -06:00
|
|
|
const char *name, const struct open_flags *op)
|
2011-03-11 10:08:24 -07:00
|
|
|
{
|
2015-05-13 05:28:08 -06:00
|
|
|
struct nameidata nd;
|
2011-03-11 10:08:24 -07:00
|
|
|
struct file *file;
|
2015-01-21 22:00:03 -07:00
|
|
|
struct filename *filename;
|
2013-06-10 22:23:01 -06:00
|
|
|
int flags = op->lookup_flags | LOOKUP_ROOT;
|
2011-03-11 10:08:24 -07:00
|
|
|
|
|
|
|
nd.root.mnt = mnt;
|
|
|
|
nd.root.dentry = dentry;
|
|
|
|
|
2013-09-12 12:22:53 -06:00
|
|
|
if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
|
2011-03-11 10:08:24 -07:00
|
|
|
return ERR_PTR(-ELOOP);
|
|
|
|
|
2015-01-21 22:00:03 -07:00
|
|
|
filename = getname_kernel(name);
|
2015-08-12 04:29:44 -06:00
|
|
|
if (IS_ERR(filename))
|
2015-01-21 22:00:03 -07:00
|
|
|
return ERR_CAST(filename);
|
|
|
|
|
2015-05-13 05:28:08 -06:00
|
|
|
set_nameidata(&nd, -1, filename);
|
2015-05-12 16:43:07 -06:00
|
|
|
file = path_openat(&nd, op, flags | LOOKUP_RCU);
|
2011-03-11 10:08:24 -07:00
|
|
|
if (unlikely(file == ERR_PTR(-ECHILD)))
|
2015-05-12 16:43:07 -06:00
|
|
|
file = path_openat(&nd, op, flags);
|
2011-03-11 10:08:24 -07:00
|
|
|
if (unlikely(file == ERR_PTR(-ESTALE)))
|
2015-05-12 16:43:07 -06:00
|
|
|
file = path_openat(&nd, op, flags | LOOKUP_REVAL);
|
2015-05-13 05:28:08 -06:00
|
|
|
restore_nameidata();
|
2015-01-21 22:00:03 -07:00
|
|
|
putname(filename);
|
2011-03-11 10:08:24 -07:00
|
|
|
return file;
|
|
|
|
}
|
|
|
|
|
2015-01-22 00:16:49 -07:00
|
|
|
static struct dentry *filename_create(int dfd, struct filename *name,
|
2012-12-11 10:10:06 -07:00
|
|
|
struct path *path, unsigned int lookup_flags)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2005-06-23 01:09:49 -06:00
|
|
|
struct dentry *dentry = ERR_PTR(-EEXIST);
|
2015-05-09 09:19:16 -06:00
|
|
|
struct qstr last;
|
|
|
|
int type;
|
2012-06-12 08:20:30 -06:00
|
|
|
int err2;
|
2012-12-11 10:10:06 -07:00
|
|
|
int error;
|
|
|
|
bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
|
|
|
|
* other flags passed in are ignored!
|
|
|
|
*/
|
|
|
|
lookup_flags &= LOOKUP_REVAL;
|
|
|
|
|
2015-05-12 15:32:54 -06:00
|
|
|
name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
|
|
|
|
if (IS_ERR(name))
|
|
|
|
return ERR_CAST(name);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2005-06-23 01:09:49 -06:00
|
|
|
/*
|
|
|
|
* Yucky last component or no last component at all?
|
|
|
|
* (foo/., foo/.., /////)
|
|
|
|
*/
|
2015-05-12 15:32:54 -06:00
|
|
|
if (unlikely(type != LAST_NORM))
|
2011-06-27 14:53:43 -06:00
|
|
|
goto out;
|
2005-06-23 01:09:49 -06:00
|
|
|
|
2012-06-12 08:20:30 -06:00
|
|
|
/* don't fail immediately if it's r/o, at least try to report other errors */
|
2015-05-09 09:19:16 -06:00
|
|
|
err2 = mnt_want_write(path->mnt);
|
2005-06-23 01:09:49 -06:00
|
|
|
/*
|
|
|
|
* Do the final lookup.
|
|
|
|
*/
|
2015-05-09 09:19:16 -06:00
|
|
|
lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
|
2015-05-09 09:19:16 -06:00
|
|
|
dentry = __lookup_hash(&last, path->dentry, lookup_flags);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (IS_ERR(dentry))
|
2012-07-19 16:25:00 -06:00
|
|
|
goto unlock;
|
2005-06-23 01:09:49 -06:00
|
|
|
|
2012-07-19 16:25:00 -06:00
|
|
|
error = -EEXIST;
|
2013-09-12 12:22:53 -06:00
|
|
|
if (d_is_positive(dentry))
|
2012-07-19 16:25:00 -06:00
|
|
|
goto fail;
|
2013-09-12 12:22:53 -06:00
|
|
|
|
2005-06-23 01:09:49 -06:00
|
|
|
/*
|
|
|
|
* Special case - lookup gave negative, but... we had foo/bar/
|
|
|
|
* From the vfs_mknod() POV we just have a negative dentry -
|
|
|
|
* all is fine. Let's be bastards - you had / on the end, you've
|
|
|
|
* been asking for (non-existent) directory. -ENOENT for you.
|
|
|
|
*/
|
2015-05-09 09:19:16 -06:00
|
|
|
if (unlikely(!is_dir && last.name[last.len])) {
|
2012-07-19 16:25:00 -06:00
|
|
|
error = -ENOENT;
|
2011-06-27 14:53:43 -06:00
|
|
|
goto fail;
|
2008-05-15 02:49:12 -06:00
|
|
|
}
|
2012-06-12 08:20:30 -06:00
|
|
|
if (unlikely(err2)) {
|
|
|
|
error = err2;
|
2012-07-19 16:25:00 -06:00
|
|
|
goto fail;
|
2012-06-12 08:20:30 -06:00
|
|
|
}
|
2015-05-12 15:21:25 -06:00
|
|
|
putname(name);
|
2005-04-16 16:20:36 -06:00
|
|
|
return dentry;
|
|
|
|
fail:
|
2012-07-19 16:25:00 -06:00
|
|
|
dput(dentry);
|
|
|
|
dentry = ERR_PTR(error);
|
|
|
|
unlock:
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_unlock(path->dentry->d_inode);
|
2012-06-12 08:20:30 -06:00
|
|
|
if (!err2)
|
2015-05-09 09:19:16 -06:00
|
|
|
mnt_drop_write(path->mnt);
|
2011-06-27 14:53:43 -06:00
|
|
|
out:
|
2015-05-09 09:19:16 -06:00
|
|
|
path_put(path);
|
2015-05-12 15:21:25 -06:00
|
|
|
putname(name);
|
2005-04-16 16:20:36 -06:00
|
|
|
return dentry;
|
|
|
|
}
|
2015-01-22 00:16:49 -07:00
|
|
|
|
|
|
|
struct dentry *kern_path_create(int dfd, const char *pathname,
|
|
|
|
struct path *path, unsigned int lookup_flags)
|
|
|
|
{
|
2015-05-12 15:21:25 -06:00
|
|
|
return filename_create(dfd, getname_kernel(pathname),
|
|
|
|
path, lookup_flags);
|
2015-01-22 00:16:49 -07:00
|
|
|
}
|
2011-06-26 09:50:15 -06:00
|
|
|
EXPORT_SYMBOL(kern_path_create);
|
|
|
|
|
2012-07-19 15:15:31 -06:00
|
|
|
void done_path_create(struct path *path, struct dentry *dentry)
|
|
|
|
{
|
|
|
|
dput(dentry);
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_unlock(path->dentry->d_inode);
|
2012-07-19 16:25:00 -06:00
|
|
|
mnt_drop_write(path->mnt);
|
2012-07-19 15:15:31 -06:00
|
|
|
path_put(path);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(done_path_create);
|
|
|
|
|
2015-05-13 05:00:28 -06:00
|
|
|
inline struct dentry *user_path_create(int dfd, const char __user *pathname,
|
2012-12-11 10:10:06 -07:00
|
|
|
struct path *path, unsigned int lookup_flags)
|
2011-06-26 09:50:15 -06:00
|
|
|
{
|
2015-05-12 15:21:25 -06:00
|
|
|
return filename_create(dfd, getname(pathname), path, lookup_flags);
|
2011-06-26 09:50:15 -06:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(user_path_create);
|
|
|
|
|
2011-07-25 23:52:52 -06:00
|
|
|
int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2008-07-30 07:08:48 -06:00
|
|
|
int error = may_create(dir, dentry);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
|
2011-11-13 13:16:43 -07:00
|
|
|
if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
|
2005-04-16 16:20:36 -06:00
|
|
|
return -EPERM;
|
|
|
|
|
2008-12-04 08:06:33 -07:00
|
|
|
if (!dir->i_op->mknod)
|
2005-04-16 16:20:36 -06:00
|
|
|
return -EPERM;
|
|
|
|
|
2008-04-29 02:00:10 -06:00
|
|
|
error = devcgroup_inode_mknod(mode, dev);
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
error = security_inode_mknod(dir, dentry, mode, dev);
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
|
|
|
|
error = dir->i_op->mknod(dir, dentry, mode, dev);
|
2005-09-09 14:01:44 -06:00
|
|
|
if (!error)
|
2005-11-03 08:57:06 -07:00
|
|
|
fsnotify_create(dir, dentry);
|
2005-04-16 16:20:36 -06:00
|
|
|
return error;
|
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(vfs_mknod);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2011-07-26 02:31:40 -06:00
|
|
|
static int may_mknod(umode_t mode)
|
2008-02-15 15:37:57 -07:00
|
|
|
{
|
|
|
|
switch (mode & S_IFMT) {
|
|
|
|
case S_IFREG:
|
|
|
|
case S_IFCHR:
|
|
|
|
case S_IFBLK:
|
|
|
|
case S_IFIFO:
|
|
|
|
case S_IFSOCK:
|
|
|
|
case 0: /* zero mode translates to S_IFREG */
|
|
|
|
return 0;
|
|
|
|
case S_IFDIR:
|
|
|
|
return -EPERM;
|
|
|
|
default:
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2011-07-25 15:32:17 -06:00
|
|
|
SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
|
2009-01-14 06:14:31 -07:00
|
|
|
unsigned, dev)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2008-07-21 07:32:51 -06:00
|
|
|
struct dentry *dentry;
|
2011-06-26 09:50:15 -06:00
|
|
|
struct path path;
|
|
|
|
int error;
|
2012-12-20 14:00:10 -07:00
|
|
|
unsigned int lookup_flags = 0;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2012-07-19 15:17:26 -06:00
|
|
|
error = may_mknod(mode);
|
|
|
|
if (error)
|
|
|
|
return error;
|
2012-12-20 14:00:10 -07:00
|
|
|
retry:
|
|
|
|
dentry = user_path_create(dfd, filename, &path, lookup_flags);
|
2011-06-26 09:50:15 -06:00
|
|
|
if (IS_ERR(dentry))
|
|
|
|
return PTR_ERR(dentry);
|
2008-07-21 07:32:51 -06:00
|
|
|
|
2011-06-26 09:50:15 -06:00
|
|
|
if (!IS_POSIXACL(path.dentry->d_inode))
|
2009-03-29 17:08:22 -06:00
|
|
|
mode &= ~current_umask();
|
2011-06-26 09:50:15 -06:00
|
|
|
error = security_path_mknod(&path, dentry, mode, dev);
|
2008-12-16 21:24:15 -07:00
|
|
|
if (error)
|
2012-07-19 16:25:00 -06:00
|
|
|
goto out;
|
2008-02-15 15:37:57 -07:00
|
|
|
switch (mode & S_IFMT) {
|
2005-04-16 16:20:36 -06:00
|
|
|
case 0: case S_IFREG:
|
2012-06-10 16:09:36 -06:00
|
|
|
error = vfs_create(path.dentry->d_inode,dentry,mode,true);
|
2005-04-16 16:20:36 -06:00
|
|
|
break;
|
|
|
|
case S_IFCHR: case S_IFBLK:
|
2011-06-26 09:50:15 -06:00
|
|
|
error = vfs_mknod(path.dentry->d_inode,dentry,mode,
|
2005-04-16 16:20:36 -06:00
|
|
|
new_decode_dev(dev));
|
|
|
|
break;
|
|
|
|
case S_IFIFO: case S_IFSOCK:
|
2011-06-26 09:50:15 -06:00
|
|
|
error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
|
2005-04-16 16:20:36 -06:00
|
|
|
break;
|
|
|
|
}
|
2012-07-19 16:25:00 -06:00
|
|
|
out:
|
2012-07-19 15:15:31 -06:00
|
|
|
done_path_create(&path, dentry);
|
2012-12-20 14:00:10 -07:00
|
|
|
if (retry_estale(error, lookup_flags)) {
|
|
|
|
lookup_flags |= LOOKUP_REVAL;
|
|
|
|
goto retry;
|
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
return error;
|
|
|
|
}
|
|
|
|
|
2011-07-25 15:32:17 -06:00
|
|
|
SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
|
2006-01-18 18:43:53 -07:00
|
|
|
{
|
|
|
|
return sys_mknodat(AT_FDCWD, filename, mode, dev);
|
|
|
|
}
|
|
|
|
|
2011-07-25 23:41:39 -06:00
|
|
|
int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2008-07-30 07:08:48 -06:00
|
|
|
int error = may_create(dir, dentry);
|
2012-02-06 10:45:27 -07:00
|
|
|
unsigned max_links = dir->i_sb->s_max_links;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
|
2008-12-04 08:06:33 -07:00
|
|
|
if (!dir->i_op->mkdir)
|
2005-04-16 16:20:36 -06:00
|
|
|
return -EPERM;
|
|
|
|
|
|
|
|
mode &= (S_IRWXUGO|S_ISVTX);
|
|
|
|
error = security_inode_mkdir(dir, dentry, mode);
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
|
2012-02-06 10:45:27 -07:00
|
|
|
if (max_links && dir->i_nlink >= max_links)
|
|
|
|
return -EMLINK;
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
error = dir->i_op->mkdir(dir, dentry, mode);
|
2005-09-09 14:01:44 -06:00
|
|
|
if (!error)
|
2005-11-03 08:57:06 -07:00
|
|
|
fsnotify_mkdir(dir, dentry);
|
2005-04-16 16:20:36 -06:00
|
|
|
return error;
|
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(vfs_mkdir);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2011-11-21 12:59:34 -07:00
|
|
|
SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2006-10-01 00:29:01 -06:00
|
|
|
struct dentry *dentry;
|
2011-06-26 09:50:15 -06:00
|
|
|
struct path path;
|
|
|
|
int error;
|
2012-12-20 14:04:09 -07:00
|
|
|
unsigned int lookup_flags = LOOKUP_DIRECTORY;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2012-12-20 14:04:09 -07:00
|
|
|
retry:
|
|
|
|
dentry = user_path_create(dfd, pathname, &path, lookup_flags);
|
2006-10-01 00:29:01 -06:00
|
|
|
if (IS_ERR(dentry))
|
2011-06-26 09:50:15 -06:00
|
|
|
return PTR_ERR(dentry);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2011-06-26 09:50:15 -06:00
|
|
|
if (!IS_POSIXACL(path.dentry->d_inode))
|
2009-03-29 17:08:22 -06:00
|
|
|
mode &= ~current_umask();
|
2011-06-26 09:50:15 -06:00
|
|
|
error = security_path_mkdir(&path, dentry, mode);
|
2012-07-19 16:25:00 -06:00
|
|
|
if (!error)
|
|
|
|
error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
|
2012-07-19 15:15:31 -06:00
|
|
|
done_path_create(&path, dentry);
|
2012-12-20 14:04:09 -07:00
|
|
|
if (retry_estale(error, lookup_flags)) {
|
|
|
|
lookup_flags |= LOOKUP_REVAL;
|
|
|
|
goto retry;
|
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
return error;
|
|
|
|
}
|
|
|
|
|
2011-11-21 12:59:34 -07:00
|
|
|
SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
|
2006-01-18 18:43:53 -07:00
|
|
|
{
|
|
|
|
return sys_mkdirat(AT_FDCWD, pathname, mode);
|
|
|
|
}
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
/*
|
2011-05-24 14:06:08 -06:00
|
|
|
* The dentry_unhash() helper will try to drop the dentry early: we
|
2012-02-15 09:48:40 -07:00
|
|
|
* should have a usage count of 1 if we're the only user of this
|
2011-05-24 14:06:08 -06:00
|
|
|
* dentry, and if that is true (possibly after pruning the dcache),
|
|
|
|
* then we drop the dentry now.
|
2005-04-16 16:20:36 -06:00
|
|
|
*
|
|
|
|
* A low-level filesystem can, if it choses, legally
|
|
|
|
* do a
|
|
|
|
*
|
|
|
|
* if (!d_unhashed(dentry))
|
|
|
|
* return -EBUSY;
|
|
|
|
*
|
|
|
|
* if it cannot handle the case of removing a directory
|
|
|
|
* that is still in use by something else..
|
|
|
|
*/
|
|
|
|
void dentry_unhash(struct dentry *dentry)
|
|
|
|
{
|
2006-12-06 21:37:07 -07:00
|
|
|
shrink_dcache_parent(dentry);
|
2005-04-16 16:20:36 -06:00
|
|
|
spin_lock(&dentry->d_lock);
|
2013-08-28 19:24:59 -06:00
|
|
|
if (dentry->d_lockref.count == 1)
|
2005-04-16 16:20:36 -06:00
|
|
|
__d_drop(dentry);
|
|
|
|
spin_unlock(&dentry->d_lock);
|
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(dentry_unhash);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
int vfs_rmdir(struct inode *dir, struct dentry *dentry)
|
|
|
|
{
|
|
|
|
int error = may_delete(dir, dentry, 1);
|
|
|
|
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
|
2008-12-04 08:06:33 -07:00
|
|
|
if (!dir->i_op->rmdir)
|
2005-04-16 16:20:36 -06:00
|
|
|
return -EPERM;
|
|
|
|
|
2011-09-14 11:55:41 -06:00
|
|
|
dget(dentry);
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_lock(dentry->d_inode);
|
2011-05-24 14:06:11 -06:00
|
|
|
|
|
|
|
error = -EBUSY;
|
2013-10-04 20:15:13 -06:00
|
|
|
if (is_local_mountpoint(dentry))
|
2011-05-24 14:06:11 -06:00
|
|
|
goto out;
|
|
|
|
|
|
|
|
error = security_inode_rmdir(dir, dentry);
|
|
|
|
if (error)
|
|
|
|
goto out;
|
|
|
|
|
2011-05-29 22:20:59 -06:00
|
|
|
shrink_dcache_parent(dentry);
|
2011-05-24 14:06:11 -06:00
|
|
|
error = dir->i_op->rmdir(dir, dentry);
|
|
|
|
if (error)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
dentry->d_inode->i_flags |= S_DEAD;
|
|
|
|
dont_mount(dentry);
|
vfs: Lazily remove mounts on unlinked files and directories.
With the introduction of mount namespaces and bind mounts it became
possible to access files and directories that on some paths are mount
points but are not mount points on other paths. It is very confusing
when rm -rf somedir returns -EBUSY simply because somedir is mounted
somewhere else. With the addition of user namespaces allowing
unprivileged mounts this condition has gone from annoying to allowing
a DOS attack on other users in the system.
The possibility for mischief is removed by updating the vfs to support
rename, unlink and rmdir on a dentry that is a mountpoint and by
lazily unmounting mountpoints on deleted dentries.
In particular this change allows rename, unlink and rmdir system calls
on a dentry without a mountpoint in the current mount namespace to
succeed, and it allows rename, unlink, and rmdir performed on a
distributed filesystem to update the vfs cache even if when there is a
mount in some namespace on the original dentry.
There are two common patterns of maintaining mounts: Mounts on trusted
paths with the parent directory of the mount point and all ancestory
directories up to / owned by root and modifiable only by root
(i.e. /media/xxx, /dev, /dev/pts, /proc, /sys, /sys/fs/cgroup/{cpu,
cpuacct, ...}, /usr, /usr/local). Mounts on unprivileged directories
maintained by fusermount.
In the case of mounts in trusted directories owned by root and
modifiable only by root the current parent directory permissions are
sufficient to ensure a mount point on a trusted path is not removed
or renamed by anyone other than root, even if there is a context
where the there are no mount points to prevent this.
In the case of mounts in directories owned by less privileged users
races with users modifying the path of a mount point are already a
danger. fusermount already uses a combination of chdir,
/proc/<pid>/fd/NNN, and UMOUNT_NOFOLLOW to prevent these races. The
removable of global rename, unlink, and rmdir protection really adds
nothing new to consider only a widening of the attack window, and
fusermount is already safe against unprivileged users modifying the
directory simultaneously.
In principle for perfect userspace programs returning -EBUSY for
unlink, rmdir, and rename of dentires that have mounts in the local
namespace is actually unnecessary. Unfortunately not all userspace
programs are perfect so retaining -EBUSY for unlink, rmdir and rename
of dentries that have mounts in the current mount namespace plays an
important role of maintaining consistency with historical behavior and
making imperfect userspace applications hard to exploit.
v2: Remove spurious old_dentry.
v3: Optimized shrink_submounts_and_drop
Removed unsued afs label
v4: Simplified the changes to check_submounts_and_drop
Do not rename check_submounts_and_drop shrink_submounts_and_drop
Document what why we need atomicity in check_submounts_and_drop
Rely on the parent inode mutex to make d_revalidate and d_invalidate
an atomic unit.
v5: Refcount the mountpoint to detach in case of simultaneous
renames.
Reviewed-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-10-01 19:33:48 -06:00
|
|
|
detach_mounts(dentry);
|
2011-05-24 14:06:11 -06:00
|
|
|
|
|
|
|
out:
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_unlock(dentry->d_inode);
|
2011-09-14 11:55:41 -06:00
|
|
|
dput(dentry);
|
2011-05-24 14:06:11 -06:00
|
|
|
if (!error)
|
2005-04-16 16:20:36 -06:00
|
|
|
d_delete(dentry);
|
|
|
|
return error;
|
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(vfs_rmdir);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2006-01-18 18:43:53 -07:00
|
|
|
static long do_rmdir(int dfd, const char __user *pathname)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
|
|
|
int error = 0;
|
2012-10-10 13:25:28 -06:00
|
|
|
struct filename *name;
|
2005-04-16 16:20:36 -06:00
|
|
|
struct dentry *dentry;
|
2015-04-30 14:09:11 -06:00
|
|
|
struct path path;
|
|
|
|
struct qstr last;
|
|
|
|
int type;
|
2012-12-20 14:28:33 -07:00
|
|
|
unsigned int lookup_flags = 0;
|
|
|
|
retry:
|
2015-04-30 14:09:11 -06:00
|
|
|
name = user_path_parent(dfd, pathname,
|
|
|
|
&path, &last, &type, lookup_flags);
|
2012-10-10 13:25:28 -06:00
|
|
|
if (IS_ERR(name))
|
|
|
|
return PTR_ERR(name);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2015-04-30 14:09:11 -06:00
|
|
|
switch (type) {
|
2008-10-15 16:50:29 -06:00
|
|
|
case LAST_DOTDOT:
|
|
|
|
error = -ENOTEMPTY;
|
|
|
|
goto exit1;
|
|
|
|
case LAST_DOT:
|
|
|
|
error = -EINVAL;
|
|
|
|
goto exit1;
|
|
|
|
case LAST_ROOT:
|
|
|
|
error = -EBUSY;
|
|
|
|
goto exit1;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
2008-10-15 16:50:29 -06:00
|
|
|
|
2015-04-30 14:09:11 -06:00
|
|
|
error = mnt_want_write(path.mnt);
|
2012-06-12 08:20:30 -06:00
|
|
|
if (error)
|
|
|
|
goto exit1;
|
2008-10-15 16:50:29 -06:00
|
|
|
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
|
2015-04-30 14:09:11 -06:00
|
|
|
dentry = __lookup_hash(&last, path.dentry, lookup_flags);
|
2005-04-16 16:20:36 -06:00
|
|
|
error = PTR_ERR(dentry);
|
2006-10-01 00:29:01 -06:00
|
|
|
if (IS_ERR(dentry))
|
|
|
|
goto exit2;
|
2011-06-06 17:19:40 -06:00
|
|
|
if (!dentry->d_inode) {
|
|
|
|
error = -ENOENT;
|
|
|
|
goto exit3;
|
|
|
|
}
|
2015-04-30 14:09:11 -06:00
|
|
|
error = security_path_rmdir(&path, dentry);
|
2008-12-16 21:24:15 -07:00
|
|
|
if (error)
|
2012-06-12 08:20:30 -06:00
|
|
|
goto exit3;
|
2015-04-30 14:09:11 -06:00
|
|
|
error = vfs_rmdir(path.dentry->d_inode, dentry);
|
2008-02-15 15:37:34 -07:00
|
|
|
exit3:
|
2006-10-01 00:29:01 -06:00
|
|
|
dput(dentry);
|
|
|
|
exit2:
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_unlock(path.dentry->d_inode);
|
2015-04-30 14:09:11 -06:00
|
|
|
mnt_drop_write(path.mnt);
|
2005-04-16 16:20:36 -06:00
|
|
|
exit1:
|
2015-04-30 14:09:11 -06:00
|
|
|
path_put(&path);
|
2005-04-16 16:20:36 -06:00
|
|
|
putname(name);
|
2012-12-20 14:28:33 -07:00
|
|
|
if (retry_estale(error, lookup_flags)) {
|
|
|
|
lookup_flags |= LOOKUP_REVAL;
|
|
|
|
goto retry;
|
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
return error;
|
|
|
|
}
|
|
|
|
|
2009-01-14 06:14:22 -07:00
|
|
|
SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
|
2006-01-18 18:43:53 -07:00
|
|
|
{
|
|
|
|
return do_rmdir(AT_FDCWD, pathname);
|
|
|
|
}
|
|
|
|
|
2011-09-20 07:14:34 -06:00
|
|
|
/**
|
|
|
|
* vfs_unlink - unlink a filesystem object
|
|
|
|
* @dir: parent directory
|
|
|
|
* @dentry: victim
|
|
|
|
* @delegated_inode: returns victim inode, if the inode is delegated.
|
|
|
|
*
|
|
|
|
* The caller must hold dir->i_mutex.
|
|
|
|
*
|
|
|
|
* If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
|
|
|
|
* return a reference to the inode in delegated_inode. The caller
|
|
|
|
* should then break the delegation on that inode and retry. Because
|
|
|
|
* breaking a delegation may take a long time, the caller should drop
|
|
|
|
* dir->i_mutex before doing so.
|
|
|
|
*
|
|
|
|
* Alternatively, a caller may pass NULL for delegated_inode. This may
|
|
|
|
* be appropriate for callers that expect the underlying filesystem not
|
|
|
|
* to be NFS exported.
|
|
|
|
*/
|
|
|
|
int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2012-08-28 05:03:24 -06:00
|
|
|
struct inode *target = dentry->d_inode;
|
2005-04-16 16:20:36 -06:00
|
|
|
int error = may_delete(dir, dentry, 0);
|
|
|
|
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
|
2008-12-04 08:06:33 -07:00
|
|
|
if (!dir->i_op->unlink)
|
2005-04-16 16:20:36 -06:00
|
|
|
return -EPERM;
|
|
|
|
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_lock(target);
|
vfs: Lazily remove mounts on unlinked files and directories.
With the introduction of mount namespaces and bind mounts it became
possible to access files and directories that on some paths are mount
points but are not mount points on other paths. It is very confusing
when rm -rf somedir returns -EBUSY simply because somedir is mounted
somewhere else. With the addition of user namespaces allowing
unprivileged mounts this condition has gone from annoying to allowing
a DOS attack on other users in the system.
The possibility for mischief is removed by updating the vfs to support
rename, unlink and rmdir on a dentry that is a mountpoint and by
lazily unmounting mountpoints on deleted dentries.
In particular this change allows rename, unlink and rmdir system calls
on a dentry without a mountpoint in the current mount namespace to
succeed, and it allows rename, unlink, and rmdir performed on a
distributed filesystem to update the vfs cache even if when there is a
mount in some namespace on the original dentry.
There are two common patterns of maintaining mounts: Mounts on trusted
paths with the parent directory of the mount point and all ancestory
directories up to / owned by root and modifiable only by root
(i.e. /media/xxx, /dev, /dev/pts, /proc, /sys, /sys/fs/cgroup/{cpu,
cpuacct, ...}, /usr, /usr/local). Mounts on unprivileged directories
maintained by fusermount.
In the case of mounts in trusted directories owned by root and
modifiable only by root the current parent directory permissions are
sufficient to ensure a mount point on a trusted path is not removed
or renamed by anyone other than root, even if there is a context
where the there are no mount points to prevent this.
In the case of mounts in directories owned by less privileged users
races with users modifying the path of a mount point are already a
danger. fusermount already uses a combination of chdir,
/proc/<pid>/fd/NNN, and UMOUNT_NOFOLLOW to prevent these races. The
removable of global rename, unlink, and rmdir protection really adds
nothing new to consider only a widening of the attack window, and
fusermount is already safe against unprivileged users modifying the
directory simultaneously.
In principle for perfect userspace programs returning -EBUSY for
unlink, rmdir, and rename of dentires that have mounts in the local
namespace is actually unnecessary. Unfortunately not all userspace
programs are perfect so retaining -EBUSY for unlink, rmdir and rename
of dentries that have mounts in the current mount namespace plays an
important role of maintaining consistency with historical behavior and
making imperfect userspace applications hard to exploit.
v2: Remove spurious old_dentry.
v3: Optimized shrink_submounts_and_drop
Removed unsued afs label
v4: Simplified the changes to check_submounts_and_drop
Do not rename check_submounts_and_drop shrink_submounts_and_drop
Document what why we need atomicity in check_submounts_and_drop
Rely on the parent inode mutex to make d_revalidate and d_invalidate
an atomic unit.
v5: Refcount the mountpoint to detach in case of simultaneous
renames.
Reviewed-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-10-01 19:33:48 -06:00
|
|
|
if (is_local_mountpoint(dentry))
|
2005-04-16 16:20:36 -06:00
|
|
|
error = -EBUSY;
|
|
|
|
else {
|
|
|
|
error = security_inode_unlink(dir, dentry);
|
2010-03-03 12:12:08 -07:00
|
|
|
if (!error) {
|
2012-08-28 08:50:40 -06:00
|
|
|
error = try_break_deleg(target, delegated_inode);
|
|
|
|
if (error)
|
2011-09-20 07:14:34 -06:00
|
|
|
goto out;
|
2005-04-16 16:20:36 -06:00
|
|
|
error = dir->i_op->unlink(dir, dentry);
|
vfs: Lazily remove mounts on unlinked files and directories.
With the introduction of mount namespaces and bind mounts it became
possible to access files and directories that on some paths are mount
points but are not mount points on other paths. It is very confusing
when rm -rf somedir returns -EBUSY simply because somedir is mounted
somewhere else. With the addition of user namespaces allowing
unprivileged mounts this condition has gone from annoying to allowing
a DOS attack on other users in the system.
The possibility for mischief is removed by updating the vfs to support
rename, unlink and rmdir on a dentry that is a mountpoint and by
lazily unmounting mountpoints on deleted dentries.
In particular this change allows rename, unlink and rmdir system calls
on a dentry without a mountpoint in the current mount namespace to
succeed, and it allows rename, unlink, and rmdir performed on a
distributed filesystem to update the vfs cache even if when there is a
mount in some namespace on the original dentry.
There are two common patterns of maintaining mounts: Mounts on trusted
paths with the parent directory of the mount point and all ancestory
directories up to / owned by root and modifiable only by root
(i.e. /media/xxx, /dev, /dev/pts, /proc, /sys, /sys/fs/cgroup/{cpu,
cpuacct, ...}, /usr, /usr/local). Mounts on unprivileged directories
maintained by fusermount.
In the case of mounts in trusted directories owned by root and
modifiable only by root the current parent directory permissions are
sufficient to ensure a mount point on a trusted path is not removed
or renamed by anyone other than root, even if there is a context
where the there are no mount points to prevent this.
In the case of mounts in directories owned by less privileged users
races with users modifying the path of a mount point are already a
danger. fusermount already uses a combination of chdir,
/proc/<pid>/fd/NNN, and UMOUNT_NOFOLLOW to prevent these races. The
removable of global rename, unlink, and rmdir protection really adds
nothing new to consider only a widening of the attack window, and
fusermount is already safe against unprivileged users modifying the
directory simultaneously.
In principle for perfect userspace programs returning -EBUSY for
unlink, rmdir, and rename of dentires that have mounts in the local
namespace is actually unnecessary. Unfortunately not all userspace
programs are perfect so retaining -EBUSY for unlink, rmdir and rename
of dentries that have mounts in the current mount namespace plays an
important role of maintaining consistency with historical behavior and
making imperfect userspace applications hard to exploit.
v2: Remove spurious old_dentry.
v3: Optimized shrink_submounts_and_drop
Removed unsued afs label
v4: Simplified the changes to check_submounts_and_drop
Do not rename check_submounts_and_drop shrink_submounts_and_drop
Document what why we need atomicity in check_submounts_and_drop
Rely on the parent inode mutex to make d_revalidate and d_invalidate
an atomic unit.
v5: Refcount the mountpoint to detach in case of simultaneous
renames.
Reviewed-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-10-01 19:33:48 -06:00
|
|
|
if (!error) {
|
2010-04-30 15:17:09 -06:00
|
|
|
dont_mount(dentry);
|
vfs: Lazily remove mounts on unlinked files and directories.
With the introduction of mount namespaces and bind mounts it became
possible to access files and directories that on some paths are mount
points but are not mount points on other paths. It is very confusing
when rm -rf somedir returns -EBUSY simply because somedir is mounted
somewhere else. With the addition of user namespaces allowing
unprivileged mounts this condition has gone from annoying to allowing
a DOS attack on other users in the system.
The possibility for mischief is removed by updating the vfs to support
rename, unlink and rmdir on a dentry that is a mountpoint and by
lazily unmounting mountpoints on deleted dentries.
In particular this change allows rename, unlink and rmdir system calls
on a dentry without a mountpoint in the current mount namespace to
succeed, and it allows rename, unlink, and rmdir performed on a
distributed filesystem to update the vfs cache even if when there is a
mount in some namespace on the original dentry.
There are two common patterns of maintaining mounts: Mounts on trusted
paths with the parent directory of the mount point and all ancestory
directories up to / owned by root and modifiable only by root
(i.e. /media/xxx, /dev, /dev/pts, /proc, /sys, /sys/fs/cgroup/{cpu,
cpuacct, ...}, /usr, /usr/local). Mounts on unprivileged directories
maintained by fusermount.
In the case of mounts in trusted directories owned by root and
modifiable only by root the current parent directory permissions are
sufficient to ensure a mount point on a trusted path is not removed
or renamed by anyone other than root, even if there is a context
where the there are no mount points to prevent this.
In the case of mounts in directories owned by less privileged users
races with users modifying the path of a mount point are already a
danger. fusermount already uses a combination of chdir,
/proc/<pid>/fd/NNN, and UMOUNT_NOFOLLOW to prevent these races. The
removable of global rename, unlink, and rmdir protection really adds
nothing new to consider only a widening of the attack window, and
fusermount is already safe against unprivileged users modifying the
directory simultaneously.
In principle for perfect userspace programs returning -EBUSY for
unlink, rmdir, and rename of dentires that have mounts in the local
namespace is actually unnecessary. Unfortunately not all userspace
programs are perfect so retaining -EBUSY for unlink, rmdir and rename
of dentries that have mounts in the current mount namespace plays an
important role of maintaining consistency with historical behavior and
making imperfect userspace applications hard to exploit.
v2: Remove spurious old_dentry.
v3: Optimized shrink_submounts_and_drop
Removed unsued afs label
v4: Simplified the changes to check_submounts_and_drop
Do not rename check_submounts_and_drop shrink_submounts_and_drop
Document what why we need atomicity in check_submounts_and_drop
Rely on the parent inode mutex to make d_revalidate and d_invalidate
an atomic unit.
v5: Refcount the mountpoint to detach in case of simultaneous
renames.
Reviewed-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-10-01 19:33:48 -06:00
|
|
|
detach_mounts(dentry);
|
|
|
|
}
|
2010-03-03 12:12:08 -07:00
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
2011-09-20 07:14:34 -06:00
|
|
|
out:
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_unlock(target);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
/* We don't d_delete() NFS sillyrenamed files--they still exist. */
|
|
|
|
if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
|
2012-08-28 05:03:24 -06:00
|
|
|
fsnotify_link_count(target);
|
2005-08-04 14:07:08 -06:00
|
|
|
d_delete(dentry);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
[PATCH] inotify
inotify is intended to correct the deficiencies of dnotify, particularly
its inability to scale and its terrible user interface:
* dnotify requires the opening of one fd per each directory
that you intend to watch. This quickly results in too many
open files and pins removable media, preventing unmount.
* dnotify is directory-based. You only learn about changes to
directories. Sure, a change to a file in a directory affects
the directory, but you are then forced to keep a cache of
stat structures.
* dnotify's interface to user-space is awful. Signals?
inotify provides a more usable, simple, powerful solution to file change
notification:
* inotify's interface is a system call that returns a fd, not SIGIO.
You get a single fd, which is select()-able.
* inotify has an event that says "the filesystem that the item
you were watching is on was unmounted."
* inotify can watch directories or files.
Inotify is currently used by Beagle (a desktop search infrastructure),
Gamin (a FAM replacement), and other projects.
See Documentation/filesystems/inotify.txt.
Signed-off-by: Robert Love <rml@novell.com>
Cc: John McCutchan <ttb@tentacle.dhs.org>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-07-12 15:06:03 -06:00
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
return error;
|
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(vfs_unlink);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Make sure that the actual truncation of the file will occur outside its
|
2006-01-09 16:59:24 -07:00
|
|
|
* directory's i_mutex. Truncate can take a long time if there is a lot of
|
2005-04-16 16:20:36 -06:00
|
|
|
* writeout happening, and we don't want to prevent access to the directory
|
|
|
|
* while waiting on the I/O.
|
|
|
|
*/
|
2006-01-18 18:43:53 -07:00
|
|
|
static long do_unlinkat(int dfd, const char __user *pathname)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2008-07-21 07:32:51 -06:00
|
|
|
int error;
|
2012-10-10 13:25:28 -06:00
|
|
|
struct filename *name;
|
2005-04-16 16:20:36 -06:00
|
|
|
struct dentry *dentry;
|
2015-04-30 14:09:11 -06:00
|
|
|
struct path path;
|
|
|
|
struct qstr last;
|
|
|
|
int type;
|
2005-04-16 16:20:36 -06:00
|
|
|
struct inode *inode = NULL;
|
2011-09-20 07:14:34 -06:00
|
|
|
struct inode *delegated_inode = NULL;
|
2012-12-20 14:38:04 -07:00
|
|
|
unsigned int lookup_flags = 0;
|
|
|
|
retry:
|
2015-04-30 14:09:11 -06:00
|
|
|
name = user_path_parent(dfd, pathname,
|
|
|
|
&path, &last, &type, lookup_flags);
|
2012-10-10 13:25:28 -06:00
|
|
|
if (IS_ERR(name))
|
|
|
|
return PTR_ERR(name);
|
2008-07-21 07:32:51 -06:00
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
error = -EISDIR;
|
2015-04-30 14:09:11 -06:00
|
|
|
if (type != LAST_NORM)
|
2005-04-16 16:20:36 -06:00
|
|
|
goto exit1;
|
2008-10-15 16:50:29 -06:00
|
|
|
|
2015-04-30 14:09:11 -06:00
|
|
|
error = mnt_want_write(path.mnt);
|
2012-06-12 08:20:30 -06:00
|
|
|
if (error)
|
|
|
|
goto exit1;
|
2011-09-20 07:14:34 -06:00
|
|
|
retry_deleg:
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
|
2015-04-30 14:09:11 -06:00
|
|
|
dentry = __lookup_hash(&last, path.dentry, lookup_flags);
|
2005-04-16 16:20:36 -06:00
|
|
|
error = PTR_ERR(dentry);
|
|
|
|
if (!IS_ERR(dentry)) {
|
|
|
|
/* Why not before? Because we want correct error value */
|
2015-04-30 14:09:11 -06:00
|
|
|
if (last.name[last.len])
|
2011-06-15 15:06:14 -06:00
|
|
|
goto slashes;
|
2005-04-16 16:20:36 -06:00
|
|
|
inode = dentry->d_inode;
|
2013-09-12 12:22:53 -06:00
|
|
|
if (d_is_negative(dentry))
|
2011-06-06 17:19:40 -06:00
|
|
|
goto slashes;
|
|
|
|
ihold(inode);
|
2015-04-30 14:09:11 -06:00
|
|
|
error = security_path_unlink(&path, dentry);
|
2008-12-16 21:24:15 -07:00
|
|
|
if (error)
|
2012-06-12 08:20:30 -06:00
|
|
|
goto exit2;
|
2015-04-30 14:09:11 -06:00
|
|
|
error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
|
2012-06-12 08:20:30 -06:00
|
|
|
exit2:
|
2005-04-16 16:20:36 -06:00
|
|
|
dput(dentry);
|
|
|
|
}
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_unlock(path.dentry->d_inode);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (inode)
|
|
|
|
iput(inode); /* truncate the inode here */
|
2011-09-20 07:14:34 -06:00
|
|
|
inode = NULL;
|
|
|
|
if (delegated_inode) {
|
2012-08-28 08:50:40 -06:00
|
|
|
error = break_deleg_wait(&delegated_inode);
|
2011-09-20 07:14:34 -06:00
|
|
|
if (!error)
|
|
|
|
goto retry_deleg;
|
|
|
|
}
|
2015-04-30 14:09:11 -06:00
|
|
|
mnt_drop_write(path.mnt);
|
2005-04-16 16:20:36 -06:00
|
|
|
exit1:
|
2015-04-30 14:09:11 -06:00
|
|
|
path_put(&path);
|
2005-04-16 16:20:36 -06:00
|
|
|
putname(name);
|
2012-12-20 14:38:04 -07:00
|
|
|
if (retry_estale(error, lookup_flags)) {
|
|
|
|
lookup_flags |= LOOKUP_REVAL;
|
|
|
|
inode = NULL;
|
|
|
|
goto retry;
|
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
return error;
|
|
|
|
|
|
|
|
slashes:
|
2013-09-12 12:22:53 -06:00
|
|
|
if (d_is_negative(dentry))
|
|
|
|
error = -ENOENT;
|
2014-04-01 09:08:41 -06:00
|
|
|
else if (d_is_dir(dentry))
|
2013-09-12 12:22:53 -06:00
|
|
|
error = -EISDIR;
|
|
|
|
else
|
|
|
|
error = -ENOTDIR;
|
2005-04-16 16:20:36 -06:00
|
|
|
goto exit2;
|
|
|
|
}
|
|
|
|
|
2009-01-14 06:14:31 -07:00
|
|
|
SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
|
2006-01-18 18:43:53 -07:00
|
|
|
{
|
|
|
|
if ((flag & ~AT_REMOVEDIR) != 0)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
if (flag & AT_REMOVEDIR)
|
|
|
|
return do_rmdir(dfd, pathname);
|
|
|
|
|
|
|
|
return do_unlinkat(dfd, pathname);
|
|
|
|
}
|
|
|
|
|
2009-01-14 06:14:16 -07:00
|
|
|
SYSCALL_DEFINE1(unlink, const char __user *, pathname)
|
2006-01-18 18:43:53 -07:00
|
|
|
{
|
|
|
|
return do_unlinkat(AT_FDCWD, pathname);
|
|
|
|
}
|
|
|
|
|
2008-06-24 08:50:16 -06:00
|
|
|
int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2008-07-30 07:08:48 -06:00
|
|
|
int error = may_create(dir, dentry);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
|
2008-12-04 08:06:33 -07:00
|
|
|
if (!dir->i_op->symlink)
|
2005-04-16 16:20:36 -06:00
|
|
|
return -EPERM;
|
|
|
|
|
|
|
|
error = security_inode_symlink(dir, dentry, oldname);
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
|
|
|
|
error = dir->i_op->symlink(dir, dentry, oldname);
|
2005-09-09 14:01:44 -06:00
|
|
|
if (!error)
|
2005-11-03 08:57:06 -07:00
|
|
|
fsnotify_create(dir, dentry);
|
2005-04-16 16:20:36 -06:00
|
|
|
return error;
|
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(vfs_symlink);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2009-01-14 06:14:31 -07:00
|
|
|
SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
|
|
|
|
int, newdfd, const char __user *, newname)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2008-07-21 07:32:51 -06:00
|
|
|
int error;
|
2012-10-10 13:25:28 -06:00
|
|
|
struct filename *from;
|
2006-10-01 00:29:01 -06:00
|
|
|
struct dentry *dentry;
|
2011-06-26 09:50:15 -06:00
|
|
|
struct path path;
|
2012-12-11 10:10:08 -07:00
|
|
|
unsigned int lookup_flags = 0;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
from = getname(oldname);
|
2008-07-21 07:32:51 -06:00
|
|
|
if (IS_ERR(from))
|
2005-04-16 16:20:36 -06:00
|
|
|
return PTR_ERR(from);
|
2012-12-11 10:10:08 -07:00
|
|
|
retry:
|
|
|
|
dentry = user_path_create(newdfd, newname, &path, lookup_flags);
|
2006-10-01 00:29:01 -06:00
|
|
|
error = PTR_ERR(dentry);
|
|
|
|
if (IS_ERR(dentry))
|
2011-06-26 09:50:15 -06:00
|
|
|
goto out_putname;
|
2006-10-01 00:29:01 -06:00
|
|
|
|
2012-10-10 13:25:28 -06:00
|
|
|
error = security_path_symlink(&path, dentry, from->name);
|
2012-07-19 16:25:00 -06:00
|
|
|
if (!error)
|
2012-10-10 13:25:28 -06:00
|
|
|
error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
|
2012-07-19 15:15:31 -06:00
|
|
|
done_path_create(&path, dentry);
|
2012-12-11 10:10:08 -07:00
|
|
|
if (retry_estale(error, lookup_flags)) {
|
|
|
|
lookup_flags |= LOOKUP_REVAL;
|
|
|
|
goto retry;
|
|
|
|
}
|
2006-10-01 00:29:01 -06:00
|
|
|
out_putname:
|
2005-04-16 16:20:36 -06:00
|
|
|
putname(from);
|
|
|
|
return error;
|
|
|
|
}
|
|
|
|
|
2009-01-14 06:14:16 -07:00
|
|
|
SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
|
2006-01-18 18:43:53 -07:00
|
|
|
{
|
|
|
|
return sys_symlinkat(oldname, AT_FDCWD, newname);
|
|
|
|
}
|
|
|
|
|
2011-09-20 15:14:31 -06:00
|
|
|
/**
|
|
|
|
* vfs_link - create a new link
|
|
|
|
* @old_dentry: object to be linked
|
|
|
|
* @dir: new parent
|
|
|
|
* @new_dentry: where to create the new link
|
|
|
|
* @delegated_inode: returns inode needing a delegation break
|
|
|
|
*
|
|
|
|
* The caller must hold dir->i_mutex
|
|
|
|
*
|
|
|
|
* If vfs_link discovers a delegation on the to-be-linked file in need
|
|
|
|
* of breaking, it will return -EWOULDBLOCK and return a reference to the
|
|
|
|
* inode in delegated_inode. The caller should then break the delegation
|
|
|
|
* and retry. Because breaking a delegation may take a long time, the
|
|
|
|
* caller should drop the i_mutex before doing so.
|
|
|
|
*
|
|
|
|
* Alternatively, a caller may pass NULL for delegated_inode. This may
|
|
|
|
* be appropriate for callers that expect the underlying filesystem not
|
|
|
|
* to be NFS exported.
|
|
|
|
*/
|
|
|
|
int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
|
|
|
struct inode *inode = old_dentry->d_inode;
|
2012-02-06 10:45:27 -07:00
|
|
|
unsigned max_links = dir->i_sb->s_max_links;
|
2005-04-16 16:20:36 -06:00
|
|
|
int error;
|
|
|
|
|
|
|
|
if (!inode)
|
|
|
|
return -ENOENT;
|
|
|
|
|
2008-07-30 07:08:48 -06:00
|
|
|
error = may_create(dir, new_dentry);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
|
|
|
|
if (dir->i_sb != inode->i_sb)
|
|
|
|
return -EXDEV;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* A link to an append-only or immutable file cannot be created.
|
|
|
|
*/
|
|
|
|
if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
|
|
|
|
return -EPERM;
|
2008-12-04 08:06:33 -07:00
|
|
|
if (!dir->i_op->link)
|
2005-04-16 16:20:36 -06:00
|
|
|
return -EPERM;
|
2008-06-24 08:50:15 -06:00
|
|
|
if (S_ISDIR(inode->i_mode))
|
2005-04-16 16:20:36 -06:00
|
|
|
return -EPERM;
|
|
|
|
|
|
|
|
error = security_inode_link(old_dentry, dir, new_dentry);
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_lock(inode);
|
2011-01-29 06:13:27 -07:00
|
|
|
/* Make sure we don't allow creating hardlink to an unlinked file */
|
2013-06-10 22:34:36 -06:00
|
|
|
if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
|
2011-01-29 06:13:27 -07:00
|
|
|
error = -ENOENT;
|
2012-02-06 10:45:27 -07:00
|
|
|
else if (max_links && inode->i_nlink >= max_links)
|
|
|
|
error = -EMLINK;
|
2011-09-20 15:14:31 -06:00
|
|
|
else {
|
|
|
|
error = try_break_deleg(inode, delegated_inode);
|
|
|
|
if (!error)
|
|
|
|
error = dir->i_op->link(old_dentry, dir, new_dentry);
|
|
|
|
}
|
2013-06-10 22:34:36 -06:00
|
|
|
|
|
|
|
if (!error && (inode->i_state & I_LINKABLE)) {
|
|
|
|
spin_lock(&inode->i_lock);
|
|
|
|
inode->i_state &= ~I_LINKABLE;
|
|
|
|
spin_unlock(&inode->i_lock);
|
|
|
|
}
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_unlock(inode);
|
2005-09-09 14:01:45 -06:00
|
|
|
if (!error)
|
2008-06-24 08:50:15 -06:00
|
|
|
fsnotify_link(dir, inode, new_dentry);
|
2005-04-16 16:20:36 -06:00
|
|
|
return error;
|
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(vfs_link);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Hardlinks are often used in delicate situations. We avoid
|
|
|
|
* security-related surprises by not following symlinks on the
|
|
|
|
* newname. --KAB
|
|
|
|
*
|
|
|
|
* We don't follow them on the oldname either to be compatible
|
|
|
|
* with linux 2.0, and to avoid hard-linking to directories
|
|
|
|
* and other special files. --ADM
|
|
|
|
*/
|
2009-01-14 06:14:31 -07:00
|
|
|
SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
|
|
|
|
int, newdfd, const char __user *, newname, int, flags)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
|
|
|
struct dentry *new_dentry;
|
2011-06-26 09:50:15 -06:00
|
|
|
struct path old_path, new_path;
|
2011-09-20 15:14:31 -06:00
|
|
|
struct inode *delegated_inode = NULL;
|
2011-01-29 06:13:42 -07:00
|
|
|
int how = 0;
|
2005-04-16 16:20:36 -06:00
|
|
|
int error;
|
|
|
|
|
2011-01-29 06:13:42 -07:00
|
|
|
if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
|
2006-02-24 14:04:21 -07:00
|
|
|
return -EINVAL;
|
2011-01-29 06:13:42 -07:00
|
|
|
/*
|
2013-08-28 10:18:05 -06:00
|
|
|
* To use null names we require CAP_DAC_READ_SEARCH
|
|
|
|
* This ensures that not everyone will be able to create
|
|
|
|
* handlink using the passed filedescriptor.
|
2011-01-29 06:13:42 -07:00
|
|
|
*/
|
2013-08-28 10:18:05 -06:00
|
|
|
if (flags & AT_EMPTY_PATH) {
|
|
|
|
if (!capable(CAP_DAC_READ_SEARCH))
|
|
|
|
return -ENOENT;
|
2011-01-29 06:13:42 -07:00
|
|
|
how = LOOKUP_EMPTY;
|
2013-08-28 10:18:05 -06:00
|
|
|
}
|
2011-01-29 06:13:42 -07:00
|
|
|
|
|
|
|
if (flags & AT_SYMLINK_FOLLOW)
|
|
|
|
how |= LOOKUP_FOLLOW;
|
2012-12-20 14:15:38 -07:00
|
|
|
retry:
|
2011-01-29 06:13:42 -07:00
|
|
|
error = user_path_at(olddfd, oldname, how, &old_path);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (error)
|
2008-07-21 07:32:51 -06:00
|
|
|
return error;
|
|
|
|
|
2012-12-20 14:15:38 -07:00
|
|
|
new_dentry = user_path_create(newdfd, newname, &new_path,
|
|
|
|
(how & LOOKUP_REVAL));
|
2005-04-16 16:20:36 -06:00
|
|
|
error = PTR_ERR(new_dentry);
|
2006-10-01 00:29:01 -06:00
|
|
|
if (IS_ERR(new_dentry))
|
2011-06-26 09:50:15 -06:00
|
|
|
goto out;
|
|
|
|
|
|
|
|
error = -EXDEV;
|
|
|
|
if (old_path.mnt != new_path.mnt)
|
|
|
|
goto out_dput;
|
fs: add link restrictions
This adds symlink and hardlink restrictions to the Linux VFS.
Symlinks:
A long-standing class of security issues is the symlink-based
time-of-check-time-of-use race, most commonly seen in world-writable
directories like /tmp. The common method of exploitation of this flaw
is to cross privilege boundaries when following a given symlink (i.e. a
root process follows a symlink belonging to another user). For a likely
incomplete list of hundreds of examples across the years, please see:
http://cve.mitre.org/cgi-bin/cvekey.cgi?keyword=/tmp
The solution is to permit symlinks to only be followed when outside
a sticky world-writable directory, or when the uid of the symlink and
follower match, or when the directory owner matches the symlink's owner.
Some pointers to the history of earlier discussion that I could find:
1996 Aug, Zygo Blaxell
http://marc.info/?l=bugtraq&m=87602167419830&w=2
1996 Oct, Andrew Tridgell
http://lkml.indiana.edu/hypermail/linux/kernel/9610.2/0086.html
1997 Dec, Albert D Cahalan
http://lkml.org/lkml/1997/12/16/4
2005 Feb, Lorenzo Hernández GarcÃa-Hierro
http://lkml.indiana.edu/hypermail/linux/kernel/0502.0/1896.html
2010 May, Kees Cook
https://lkml.org/lkml/2010/5/30/144
Past objections and rebuttals could be summarized as:
- Violates POSIX.
- POSIX didn't consider this situation and it's not useful to follow
a broken specification at the cost of security.
- Might break unknown applications that use this feature.
- Applications that break because of the change are easy to spot and
fix. Applications that are vulnerable to symlink ToCToU by not having
the change aren't. Additionally, no applications have yet been found
that rely on this behavior.
- Applications should just use mkstemp() or O_CREATE|O_EXCL.
- True, but applications are not perfect, and new software is written
all the time that makes these mistakes; blocking this flaw at the
kernel is a single solution to the entire class of vulnerability.
- This should live in the core VFS.
- This should live in an LSM. (https://lkml.org/lkml/2010/5/31/135)
- This should live in an LSM.
- This should live in the core VFS. (https://lkml.org/lkml/2010/8/2/188)
Hardlinks:
On systems that have user-writable directories on the same partition
as system files, a long-standing class of security issues is the
hardlink-based time-of-check-time-of-use race, most commonly seen in
world-writable directories like /tmp. The common method of exploitation
of this flaw is to cross privilege boundaries when following a given
hardlink (i.e. a root process follows a hardlink created by another
user). Additionally, an issue exists where users can "pin" a potentially
vulnerable setuid/setgid file so that an administrator will not actually
upgrade a system fully.
The solution is to permit hardlinks to only be created when the user is
already the existing file's owner, or if they already have read/write
access to the existing file.
Many Linux users are surprised when they learn they can link to files
they have no access to, so this change appears to follow the doctrine
of "least surprise". Additionally, this change does not violate POSIX,
which states "the implementation may require that the calling process
has permission to access the existing file"[1].
This change is known to break some implementations of the "at" daemon,
though the version used by Fedora and Ubuntu has been fixed[2] for
a while. Otherwise, the change has been undisruptive while in use in
Ubuntu for the last 1.5 years.
[1] http://pubs.opengroup.org/onlinepubs/9699919799/functions/linkat.html
[2] http://anonscm.debian.org/gitweb/?p=collab-maint/at.git;a=commitdiff;h=f4114656c3a6c6f6070e315ffdf940a49eda3279
This patch is based on the patches in Openwall and grsecurity, along with
suggestions from Al Viro. I have added a sysctl to enable the protected
behavior, and documentation.
Signed-off-by: Kees Cook <keescook@chromium.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2012-07-25 18:29:07 -06:00
|
|
|
error = may_linkat(&old_path);
|
|
|
|
if (unlikely(error))
|
|
|
|
goto out_dput;
|
2011-06-26 09:50:15 -06:00
|
|
|
error = security_path_link(old_path.dentry, &new_path, new_dentry);
|
2008-12-16 21:24:15 -07:00
|
|
|
if (error)
|
2012-07-19 16:25:00 -06:00
|
|
|
goto out_dput;
|
2011-09-20 15:14:31 -06:00
|
|
|
error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
|
2008-02-15 15:37:45 -07:00
|
|
|
out_dput:
|
2012-07-19 15:15:31 -06:00
|
|
|
done_path_create(&new_path, new_dentry);
|
2011-09-20 15:14:31 -06:00
|
|
|
if (delegated_inode) {
|
|
|
|
error = break_deleg_wait(&delegated_inode);
|
2014-01-31 13:41:58 -07:00
|
|
|
if (!error) {
|
|
|
|
path_put(&old_path);
|
2011-09-20 15:14:31 -06:00
|
|
|
goto retry;
|
2014-01-31 13:41:58 -07:00
|
|
|
}
|
2011-09-20 15:14:31 -06:00
|
|
|
}
|
2012-12-20 14:15:38 -07:00
|
|
|
if (retry_estale(error, how)) {
|
2014-01-31 13:41:58 -07:00
|
|
|
path_put(&old_path);
|
2012-12-20 14:15:38 -07:00
|
|
|
how |= LOOKUP_REVAL;
|
|
|
|
goto retry;
|
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
out:
|
2008-07-22 07:59:21 -06:00
|
|
|
path_put(&old_path);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
return error;
|
|
|
|
}
|
|
|
|
|
2009-01-14 06:14:16 -07:00
|
|
|
SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
|
2006-01-18 18:43:53 -07:00
|
|
|
{
|
2006-02-24 14:04:21 -07:00
|
|
|
return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
|
2006-01-18 18:43:53 -07:00
|
|
|
}
|
|
|
|
|
2014-04-01 09:08:42 -06:00
|
|
|
/**
|
|
|
|
* vfs_rename - rename a filesystem object
|
|
|
|
* @old_dir: parent of source
|
|
|
|
* @old_dentry: source
|
|
|
|
* @new_dir: parent of destination
|
|
|
|
* @new_dentry: destination
|
|
|
|
* @delegated_inode: returns an inode needing a delegation break
|
2014-04-01 09:08:42 -06:00
|
|
|
* @flags: rename flags
|
2014-04-01 09:08:42 -06:00
|
|
|
*
|
|
|
|
* The caller must hold multiple mutexes--see lock_rename()).
|
|
|
|
*
|
|
|
|
* If vfs_rename discovers a delegation in need of breaking at either
|
|
|
|
* the source or destination, it will return -EWOULDBLOCK and return a
|
|
|
|
* reference to the inode in delegated_inode. The caller should then
|
|
|
|
* break the delegation and retry. Because breaking a delegation may
|
|
|
|
* take a long time, the caller should drop all locks before doing
|
|
|
|
* so.
|
|
|
|
*
|
|
|
|
* Alternatively, a caller may pass NULL for delegated_inode. This may
|
|
|
|
* be appropriate for callers that expect the underlying filesystem not
|
|
|
|
* to be NFS exported.
|
|
|
|
*
|
2005-04-16 16:20:36 -06:00
|
|
|
* The worst of all namespace operations - renaming directory. "Perverted"
|
|
|
|
* doesn't even start to describe it. Somebody in UCB had a heck of a trip...
|
|
|
|
* Problems:
|
2014-02-17 14:52:33 -07:00
|
|
|
* a) we can get into loop creation.
|
2005-04-16 16:20:36 -06:00
|
|
|
* b) race potential - two innocent renames can create a loop together.
|
|
|
|
* That's where 4.4 screws up. Current fix: serialization on
|
2006-03-23 04:00:33 -07:00
|
|
|
* sb->s_vfs_rename_mutex. We might be more accurate, but that's another
|
2005-04-16 16:20:36 -06:00
|
|
|
* story.
|
2012-03-05 09:40:41 -07:00
|
|
|
* c) we have to lock _four_ objects - parents and victim (if it exists),
|
|
|
|
* and source (if it is not a directory).
|
2006-01-09 16:59:24 -07:00
|
|
|
* And that - after we got ->i_mutex on parents (until then we don't know
|
2005-04-16 16:20:36 -06:00
|
|
|
* whether the target exists). Solution: try to be smart with locking
|
|
|
|
* order for inodes. We rely on the fact that tree topology may change
|
2006-03-23 04:00:33 -07:00
|
|
|
* only under ->s_vfs_rename_mutex _and_ that parent of the object we
|
2005-04-16 16:20:36 -06:00
|
|
|
* move will be locked. Thus we can rank directories by the tree
|
|
|
|
* (ancestors first) and rank all non-directories after them.
|
|
|
|
* That works since everybody except rename does "lock parent, lookup,
|
2006-03-23 04:00:33 -07:00
|
|
|
* lock child" and rename is under ->s_vfs_rename_mutex.
|
2005-04-16 16:20:36 -06:00
|
|
|
* HOWEVER, it relies on the assumption that any object with ->lookup()
|
|
|
|
* has no more than 1 dentry. If "hybrid" objects will ever appear,
|
|
|
|
* we'd better make sure that there's no link(2) for them.
|
2011-05-24 14:06:07 -06:00
|
|
|
* d) conversion from fhandle to dentry may come in the wrong moment - when
|
2006-01-09 16:59:24 -07:00
|
|
|
* we are removing the target. Solution: we will have to grab ->i_mutex
|
2005-04-16 16:20:36 -06:00
|
|
|
* in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
|
2009-12-11 14:35:39 -07:00
|
|
|
* ->i_mutex on parents, which works but leads to some truly excessive
|
2005-04-16 16:20:36 -06:00
|
|
|
* locking].
|
|
|
|
*/
|
2014-04-01 09:08:42 -06:00
|
|
|
int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
|
|
|
|
struct inode *new_dir, struct dentry *new_dentry,
|
2014-04-01 09:08:42 -06:00
|
|
|
struct inode **delegated_inode, unsigned int flags)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2014-04-01 09:08:42 -06:00
|
|
|
int error;
|
|
|
|
bool is_dir = d_is_dir(old_dentry);
|
|
|
|
const unsigned char *old_name;
|
|
|
|
struct inode *source = old_dentry->d_inode;
|
2011-05-24 14:06:12 -06:00
|
|
|
struct inode *target = new_dentry->d_inode;
|
2014-04-01 09:08:43 -06:00
|
|
|
bool new_is_dir = false;
|
|
|
|
unsigned max_links = new_dir->i_sb->s_max_links;
|
2014-04-01 09:08:42 -06:00
|
|
|
|
|
|
|
if (source == target)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
error = may_delete(old_dir, old_dentry, is_dir);
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
|
2014-04-01 09:08:43 -06:00
|
|
|
if (!target) {
|
2014-04-01 09:08:42 -06:00
|
|
|
error = may_create(new_dir, new_dentry);
|
2014-04-01 09:08:43 -06:00
|
|
|
} else {
|
|
|
|
new_is_dir = d_is_dir(new_dentry);
|
|
|
|
|
|
|
|
if (!(flags & RENAME_EXCHANGE))
|
|
|
|
error = may_delete(new_dir, new_dentry, is_dir);
|
|
|
|
else
|
|
|
|
error = may_delete(new_dir, new_dentry, new_is_dir);
|
|
|
|
}
|
2014-04-01 09:08:42 -06:00
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
|
2014-07-23 07:15:30 -06:00
|
|
|
if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
|
2014-04-01 09:08:42 -06:00
|
|
|
return -EPERM;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2014-04-01 09:08:42 -06:00
|
|
|
if (flags && !old_dir->i_op->rename2)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
/*
|
|
|
|
* If we are going to change the parent - check write permissions,
|
|
|
|
* we'll need to flip '..'.
|
|
|
|
*/
|
2014-04-01 09:08:43 -06:00
|
|
|
if (new_dir != old_dir) {
|
|
|
|
if (is_dir) {
|
|
|
|
error = inode_permission(source, MAY_WRITE);
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
}
|
|
|
|
if ((flags & RENAME_EXCHANGE) && new_is_dir) {
|
|
|
|
error = inode_permission(target, MAY_WRITE);
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2014-04-01 09:08:43 -06:00
|
|
|
error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
|
|
|
|
flags);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
|
2014-04-01 09:08:42 -06:00
|
|
|
old_name = fsnotify_oldname_init(old_dentry->d_name.name);
|
2011-09-14 11:55:41 -06:00
|
|
|
dget(new_dentry);
|
2014-04-01 09:08:43 -06:00
|
|
|
if (!is_dir || (flags & RENAME_EXCHANGE))
|
2014-04-01 09:08:42 -06:00
|
|
|
lock_two_nondirectories(source, target);
|
|
|
|
else if (target)
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_lock(target);
|
2011-05-24 14:06:12 -06:00
|
|
|
|
|
|
|
error = -EBUSY;
|
2013-10-04 20:15:13 -06:00
|
|
|
if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
|
2011-05-24 14:06:12 -06:00
|
|
|
goto out;
|
|
|
|
|
2014-04-01 09:08:43 -06:00
|
|
|
if (max_links && new_dir != old_dir) {
|
2014-04-01 09:08:42 -06:00
|
|
|
error = -EMLINK;
|
2014-04-01 09:08:43 -06:00
|
|
|
if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
|
2014-04-01 09:08:42 -06:00
|
|
|
goto out;
|
2014-04-01 09:08:43 -06:00
|
|
|
if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
|
|
|
|
old_dir->i_nlink >= max_links)
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (is_dir && !(flags & RENAME_EXCHANGE) && target)
|
|
|
|
shrink_dcache_parent(new_dentry);
|
|
|
|
if (!is_dir) {
|
2014-04-01 09:08:42 -06:00
|
|
|
error = try_break_deleg(source, delegated_inode);
|
2011-09-20 14:59:58 -06:00
|
|
|
if (error)
|
|
|
|
goto out;
|
2014-04-01 09:08:43 -06:00
|
|
|
}
|
|
|
|
if (target && !new_is_dir) {
|
|
|
|
error = try_break_deleg(target, delegated_inode);
|
|
|
|
if (error)
|
|
|
|
goto out;
|
2011-09-20 14:59:58 -06:00
|
|
|
}
|
2014-07-23 07:15:30 -06:00
|
|
|
if (!old_dir->i_op->rename2) {
|
2014-04-01 09:08:42 -06:00
|
|
|
error = old_dir->i_op->rename(old_dir, old_dentry,
|
|
|
|
new_dir, new_dentry);
|
|
|
|
} else {
|
2014-07-23 07:15:30 -06:00
|
|
|
WARN_ON(old_dir->i_op->rename != NULL);
|
2014-04-01 09:08:42 -06:00
|
|
|
error = old_dir->i_op->rename2(old_dir, old_dentry,
|
|
|
|
new_dir, new_dentry, flags);
|
|
|
|
}
|
2011-05-24 14:06:13 -06:00
|
|
|
if (error)
|
|
|
|
goto out;
|
|
|
|
|
2014-04-01 09:08:43 -06:00
|
|
|
if (!(flags & RENAME_EXCHANGE) && target) {
|
2014-04-01 09:08:42 -06:00
|
|
|
if (is_dir)
|
|
|
|
target->i_flags |= S_DEAD;
|
2011-05-24 14:06:13 -06:00
|
|
|
dont_mount(new_dentry);
|
vfs: Lazily remove mounts on unlinked files and directories.
With the introduction of mount namespaces and bind mounts it became
possible to access files and directories that on some paths are mount
points but are not mount points on other paths. It is very confusing
when rm -rf somedir returns -EBUSY simply because somedir is mounted
somewhere else. With the addition of user namespaces allowing
unprivileged mounts this condition has gone from annoying to allowing
a DOS attack on other users in the system.
The possibility for mischief is removed by updating the vfs to support
rename, unlink and rmdir on a dentry that is a mountpoint and by
lazily unmounting mountpoints on deleted dentries.
In particular this change allows rename, unlink and rmdir system calls
on a dentry without a mountpoint in the current mount namespace to
succeed, and it allows rename, unlink, and rmdir performed on a
distributed filesystem to update the vfs cache even if when there is a
mount in some namespace on the original dentry.
There are two common patterns of maintaining mounts: Mounts on trusted
paths with the parent directory of the mount point and all ancestory
directories up to / owned by root and modifiable only by root
(i.e. /media/xxx, /dev, /dev/pts, /proc, /sys, /sys/fs/cgroup/{cpu,
cpuacct, ...}, /usr, /usr/local). Mounts on unprivileged directories
maintained by fusermount.
In the case of mounts in trusted directories owned by root and
modifiable only by root the current parent directory permissions are
sufficient to ensure a mount point on a trusted path is not removed
or renamed by anyone other than root, even if there is a context
where the there are no mount points to prevent this.
In the case of mounts in directories owned by less privileged users
races with users modifying the path of a mount point are already a
danger. fusermount already uses a combination of chdir,
/proc/<pid>/fd/NNN, and UMOUNT_NOFOLLOW to prevent these races. The
removable of global rename, unlink, and rmdir protection really adds
nothing new to consider only a widening of the attack window, and
fusermount is already safe against unprivileged users modifying the
directory simultaneously.
In principle for perfect userspace programs returning -EBUSY for
unlink, rmdir, and rename of dentires that have mounts in the local
namespace is actually unnecessary. Unfortunately not all userspace
programs are perfect so retaining -EBUSY for unlink, rmdir and rename
of dentries that have mounts in the current mount namespace plays an
important role of maintaining consistency with historical behavior and
making imperfect userspace applications hard to exploit.
v2: Remove spurious old_dentry.
v3: Optimized shrink_submounts_and_drop
Removed unsued afs label
v4: Simplified the changes to check_submounts_and_drop
Do not rename check_submounts_and_drop shrink_submounts_and_drop
Document what why we need atomicity in check_submounts_and_drop
Rely on the parent inode mutex to make d_revalidate and d_invalidate
an atomic unit.
v5: Refcount the mountpoint to detach in case of simultaneous
renames.
Reviewed-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-10-01 19:33:48 -06:00
|
|
|
detach_mounts(new_dentry);
|
2014-04-01 09:08:42 -06:00
|
|
|
}
|
2014-04-01 09:08:43 -06:00
|
|
|
if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
|
|
|
|
if (!(flags & RENAME_EXCHANGE))
|
|
|
|
d_move(old_dentry, new_dentry);
|
|
|
|
else
|
|
|
|
d_exchange(old_dentry, new_dentry);
|
|
|
|
}
|
2011-05-24 14:06:13 -06:00
|
|
|
out:
|
2014-04-01 09:08:43 -06:00
|
|
|
if (!is_dir || (flags & RENAME_EXCHANGE))
|
2014-04-01 09:08:42 -06:00
|
|
|
unlock_two_nondirectories(source, target);
|
|
|
|
else if (target)
|
2016-01-22 13:40:57 -07:00
|
|
|
inode_unlock(target);
|
2005-04-16 16:20:36 -06:00
|
|
|
dput(new_dentry);
|
2014-04-01 09:08:43 -06:00
|
|
|
if (!error) {
|
2009-12-25 02:57:57 -07:00
|
|
|
fsnotify_move(old_dir, new_dir, old_name, is_dir,
|
2014-04-01 09:08:43 -06:00
|
|
|
!(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
|
|
|
|
if (flags & RENAME_EXCHANGE) {
|
|
|
|
fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
|
|
|
|
new_is_dir, NULL, new_dentry);
|
|
|
|
}
|
|
|
|
}
|
[PATCH] inotify
inotify is intended to correct the deficiencies of dnotify, particularly
its inability to scale and its terrible user interface:
* dnotify requires the opening of one fd per each directory
that you intend to watch. This quickly results in too many
open files and pins removable media, preventing unmount.
* dnotify is directory-based. You only learn about changes to
directories. Sure, a change to a file in a directory affects
the directory, but you are then forced to keep a cache of
stat structures.
* dnotify's interface to user-space is awful. Signals?
inotify provides a more usable, simple, powerful solution to file change
notification:
* inotify's interface is a system call that returns a fd, not SIGIO.
You get a single fd, which is select()-able.
* inotify has an event that says "the filesystem that the item
you were watching is on was unmounted."
* inotify can watch directories or files.
Inotify is currently used by Beagle (a desktop search infrastructure),
Gamin (a FAM replacement), and other projects.
See Documentation/filesystems/inotify.txt.
Signed-off-by: Robert Love <rml@novell.com>
Cc: John McCutchan <ttb@tentacle.dhs.org>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-07-12 15:06:03 -06:00
|
|
|
fsnotify_oldname_free(old_name);
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
return error;
|
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(vfs_rename);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2014-04-01 09:08:42 -06:00
|
|
|
SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
|
|
|
|
int, newdfd, const char __user *, newname, unsigned int, flags)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2008-07-21 07:32:51 -06:00
|
|
|
struct dentry *old_dentry, *new_dentry;
|
|
|
|
struct dentry *trap;
|
2015-04-30 14:09:11 -06:00
|
|
|
struct path old_path, new_path;
|
|
|
|
struct qstr old_last, new_last;
|
|
|
|
int old_type, new_type;
|
2011-09-20 14:59:58 -06:00
|
|
|
struct inode *delegated_inode = NULL;
|
2012-10-10 13:25:28 -06:00
|
|
|
struct filename *from;
|
|
|
|
struct filename *to;
|
2015-04-30 14:09:11 -06:00
|
|
|
unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
|
2012-12-11 10:10:10 -07:00
|
|
|
bool should_retry = false;
|
2008-07-21 07:32:51 -06:00
|
|
|
int error;
|
2014-04-01 09:08:42 -06:00
|
|
|
|
2014-10-23 16:14:37 -06:00
|
|
|
if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
|
2014-04-01 09:08:43 -06:00
|
|
|
return -EINVAL;
|
|
|
|
|
2014-10-23 16:14:37 -06:00
|
|
|
if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
|
|
|
|
(flags & RENAME_EXCHANGE))
|
2014-04-01 09:08:42 -06:00
|
|
|
return -EINVAL;
|
|
|
|
|
2014-10-23 16:14:37 -06:00
|
|
|
if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
|
|
|
|
return -EPERM;
|
|
|
|
|
2015-04-30 14:09:11 -06:00
|
|
|
if (flags & RENAME_EXCHANGE)
|
|
|
|
target_flags = 0;
|
|
|
|
|
2012-12-11 10:10:10 -07:00
|
|
|
retry:
|
2015-04-30 14:09:11 -06:00
|
|
|
from = user_path_parent(olddfd, oldname,
|
|
|
|
&old_path, &old_last, &old_type, lookup_flags);
|
2012-10-10 13:25:28 -06:00
|
|
|
if (IS_ERR(from)) {
|
|
|
|
error = PTR_ERR(from);
|
2005-04-16 16:20:36 -06:00
|
|
|
goto exit;
|
2012-10-10 13:25:28 -06:00
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2015-04-30 14:09:11 -06:00
|
|
|
to = user_path_parent(newdfd, newname,
|
|
|
|
&new_path, &new_last, &new_type, lookup_flags);
|
2012-10-10 13:25:28 -06:00
|
|
|
if (IS_ERR(to)) {
|
|
|
|
error = PTR_ERR(to);
|
2005-04-16 16:20:36 -06:00
|
|
|
goto exit1;
|
2012-10-10 13:25:28 -06:00
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
error = -EXDEV;
|
2015-04-30 14:09:11 -06:00
|
|
|
if (old_path.mnt != new_path.mnt)
|
2005-04-16 16:20:36 -06:00
|
|
|
goto exit2;
|
|
|
|
|
|
|
|
error = -EBUSY;
|
2015-04-30 14:09:11 -06:00
|
|
|
if (old_type != LAST_NORM)
|
2005-04-16 16:20:36 -06:00
|
|
|
goto exit2;
|
|
|
|
|
2014-04-01 09:08:43 -06:00
|
|
|
if (flags & RENAME_NOREPLACE)
|
|
|
|
error = -EEXIST;
|
2015-04-30 14:09:11 -06:00
|
|
|
if (new_type != LAST_NORM)
|
2005-04-16 16:20:36 -06:00
|
|
|
goto exit2;
|
|
|
|
|
2015-04-30 14:09:11 -06:00
|
|
|
error = mnt_want_write(old_path.mnt);
|
2012-06-12 08:20:30 -06:00
|
|
|
if (error)
|
|
|
|
goto exit2;
|
|
|
|
|
2011-09-20 14:59:58 -06:00
|
|
|
retry_deleg:
|
2015-04-30 14:09:11 -06:00
|
|
|
trap = lock_rename(new_path.dentry, old_path.dentry);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2015-04-30 14:09:11 -06:00
|
|
|
old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
|
2005-04-16 16:20:36 -06:00
|
|
|
error = PTR_ERR(old_dentry);
|
|
|
|
if (IS_ERR(old_dentry))
|
|
|
|
goto exit3;
|
|
|
|
/* source must exist */
|
|
|
|
error = -ENOENT;
|
2013-09-12 12:22:53 -06:00
|
|
|
if (d_is_negative(old_dentry))
|
2005-04-16 16:20:36 -06:00
|
|
|
goto exit4;
|
2015-04-30 14:09:11 -06:00
|
|
|
new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
|
2014-04-01 09:08:43 -06:00
|
|
|
error = PTR_ERR(new_dentry);
|
|
|
|
if (IS_ERR(new_dentry))
|
|
|
|
goto exit4;
|
|
|
|
error = -EEXIST;
|
|
|
|
if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
|
|
|
|
goto exit5;
|
2014-04-01 09:08:43 -06:00
|
|
|
if (flags & RENAME_EXCHANGE) {
|
|
|
|
error = -ENOENT;
|
|
|
|
if (d_is_negative(new_dentry))
|
|
|
|
goto exit5;
|
|
|
|
|
|
|
|
if (!d_is_dir(new_dentry)) {
|
|
|
|
error = -ENOTDIR;
|
2015-04-30 14:09:11 -06:00
|
|
|
if (new_last.name[new_last.len])
|
2014-04-01 09:08:43 -06:00
|
|
|
goto exit5;
|
|
|
|
}
|
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
/* unless the source is a directory trailing slashes give -ENOTDIR */
|
2014-04-01 09:08:41 -06:00
|
|
|
if (!d_is_dir(old_dentry)) {
|
2005-04-16 16:20:36 -06:00
|
|
|
error = -ENOTDIR;
|
2015-04-30 14:09:11 -06:00
|
|
|
if (old_last.name[old_last.len])
|
2014-04-01 09:08:43 -06:00
|
|
|
goto exit5;
|
2015-04-30 14:09:11 -06:00
|
|
|
if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
|
2014-04-01 09:08:43 -06:00
|
|
|
goto exit5;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
/* source should not be ancestor of target */
|
|
|
|
error = -EINVAL;
|
|
|
|
if (old_dentry == trap)
|
2014-04-01 09:08:43 -06:00
|
|
|
goto exit5;
|
2005-04-16 16:20:36 -06:00
|
|
|
/* target should not be an ancestor of source */
|
2014-04-01 09:08:43 -06:00
|
|
|
if (!(flags & RENAME_EXCHANGE))
|
|
|
|
error = -ENOTEMPTY;
|
2005-04-16 16:20:36 -06:00
|
|
|
if (new_dentry == trap)
|
|
|
|
goto exit5;
|
|
|
|
|
2015-04-30 14:09:11 -06:00
|
|
|
error = security_path_rename(&old_path, old_dentry,
|
|
|
|
&new_path, new_dentry, flags);
|
2008-12-16 21:24:15 -07:00
|
|
|
if (error)
|
2012-06-12 08:20:30 -06:00
|
|
|
goto exit5;
|
2015-04-30 14:09:11 -06:00
|
|
|
error = vfs_rename(old_path.dentry->d_inode, old_dentry,
|
|
|
|
new_path.dentry->d_inode, new_dentry,
|
2014-04-01 09:08:42 -06:00
|
|
|
&delegated_inode, flags);
|
2005-04-16 16:20:36 -06:00
|
|
|
exit5:
|
|
|
|
dput(new_dentry);
|
|
|
|
exit4:
|
|
|
|
dput(old_dentry);
|
|
|
|
exit3:
|
2015-04-30 14:09:11 -06:00
|
|
|
unlock_rename(new_path.dentry, old_path.dentry);
|
2011-09-20 14:59:58 -06:00
|
|
|
if (delegated_inode) {
|
|
|
|
error = break_deleg_wait(&delegated_inode);
|
|
|
|
if (!error)
|
|
|
|
goto retry_deleg;
|
|
|
|
}
|
2015-04-30 14:09:11 -06:00
|
|
|
mnt_drop_write(old_path.mnt);
|
2005-04-16 16:20:36 -06:00
|
|
|
exit2:
|
2012-12-11 10:10:10 -07:00
|
|
|
if (retry_estale(error, lookup_flags))
|
|
|
|
should_retry = true;
|
2015-04-30 14:09:11 -06:00
|
|
|
path_put(&new_path);
|
2008-07-21 07:32:51 -06:00
|
|
|
putname(to);
|
2005-04-16 16:20:36 -06:00
|
|
|
exit1:
|
2015-04-30 14:09:11 -06:00
|
|
|
path_put(&old_path);
|
2005-04-16 16:20:36 -06:00
|
|
|
putname(from);
|
2012-12-11 10:10:10 -07:00
|
|
|
if (should_retry) {
|
|
|
|
should_retry = false;
|
|
|
|
lookup_flags |= LOOKUP_REVAL;
|
|
|
|
goto retry;
|
|
|
|
}
|
2008-07-21 07:32:51 -06:00
|
|
|
exit:
|
2005-04-16 16:20:36 -06:00
|
|
|
return error;
|
|
|
|
}
|
|
|
|
|
2014-04-01 09:08:42 -06:00
|
|
|
SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
|
|
|
|
int, newdfd, const char __user *, newname)
|
|
|
|
{
|
|
|
|
return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
|
|
|
|
}
|
|
|
|
|
2009-01-14 06:14:17 -07:00
|
|
|
SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
|
2006-01-18 18:43:53 -07:00
|
|
|
{
|
2014-04-01 09:08:42 -06:00
|
|
|
return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
|
2006-01-18 18:43:53 -07:00
|
|
|
}
|
|
|
|
|
2014-10-23 16:14:36 -06:00
|
|
|
int vfs_whiteout(struct inode *dir, struct dentry *dentry)
|
|
|
|
{
|
|
|
|
int error = may_create(dir, dentry);
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
|
|
|
|
if (!dir->i_op->mknod)
|
|
|
|
return -EPERM;
|
|
|
|
|
|
|
|
return dir->i_op->mknod(dir, dentry,
|
|
|
|
S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(vfs_whiteout);
|
|
|
|
|
2014-03-14 11:42:45 -06:00
|
|
|
int readlink_copy(char __user *buffer, int buflen, const char *link)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2014-03-14 11:42:45 -06:00
|
|
|
int len = PTR_ERR(link);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (IS_ERR(link))
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
len = strlen(link);
|
|
|
|
if (len > (unsigned) buflen)
|
|
|
|
len = buflen;
|
|
|
|
if (copy_to_user(buffer, link, len))
|
|
|
|
len = -EFAULT;
|
|
|
|
out:
|
|
|
|
return len;
|
|
|
|
}
|
2014-03-14 11:42:45 -06:00
|
|
|
EXPORT_SYMBOL(readlink_copy);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
/*
|
|
|
|
* A helper for ->readlink(). This should be used *ONLY* for symlinks that
|
2015-11-17 08:20:54 -07:00
|
|
|
* have ->get_link() not calling nd_jump_link(). Using (or not using) it
|
|
|
|
* for any given inode is up to filesystem.
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
|
|
|
int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
|
|
|
|
{
|
2015-12-29 13:58:39 -07:00
|
|
|
DEFINE_DELAYED_CALL(done);
|
2015-05-07 09:14:26 -06:00
|
|
|
struct inode *inode = d_inode(dentry);
|
|
|
|
const char *link = inode->i_link;
|
2008-06-09 17:40:37 -06:00
|
|
|
int res;
|
2005-08-19 19:02:56 -06:00
|
|
|
|
2015-04-30 18:08:02 -06:00
|
|
|
if (!link) {
|
2015-12-29 13:58:39 -07:00
|
|
|
link = inode->i_op->get_link(dentry, inode, &done);
|
2015-04-30 18:08:02 -06:00
|
|
|
if (IS_ERR(link))
|
|
|
|
return PTR_ERR(link);
|
|
|
|
}
|
2015-05-02 11:32:22 -06:00
|
|
|
res = readlink_copy(buffer, buflen, link);
|
2015-12-29 13:58:39 -07:00
|
|
|
do_delayed_call(&done);
|
2008-06-09 17:40:37 -06:00
|
|
|
return res;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(generic_readlink);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
/* get the link contents into pagecache */
|
2015-11-17 08:20:54 -07:00
|
|
|
const char *page_get_link(struct dentry *dentry, struct inode *inode,
|
2015-12-29 13:58:39 -07:00
|
|
|
struct delayed_call *callback)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2008-12-19 13:47:12 -07:00
|
|
|
char *kaddr;
|
|
|
|
struct page *page;
|
2015-11-17 08:20:54 -07:00
|
|
|
struct address_space *mapping = inode->i_mapping;
|
|
|
|
|
2015-11-17 08:41:04 -07:00
|
|
|
if (!dentry) {
|
|
|
|
page = find_get_page(mapping, 0);
|
|
|
|
if (!page)
|
|
|
|
return ERR_PTR(-ECHILD);
|
|
|
|
if (!PageUptodate(page)) {
|
|
|
|
put_page(page);
|
|
|
|
return ERR_PTR(-ECHILD);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
page = read_mapping_page(mapping, 0, NULL);
|
|
|
|
if (IS_ERR(page))
|
|
|
|
return (char*)page;
|
|
|
|
}
|
2015-12-29 13:58:39 -07:00
|
|
|
set_delayed_call(callback, page_put_link, page);
|
2015-11-16 23:07:57 -07:00
|
|
|
BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
|
|
|
|
kaddr = page_address(page);
|
2015-11-17 08:20:54 -07:00
|
|
|
nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
|
2008-12-19 13:47:12 -07:00
|
|
|
return kaddr;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2015-11-17 08:20:54 -07:00
|
|
|
EXPORT_SYMBOL(page_get_link);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2015-12-29 13:58:39 -07:00
|
|
|
void page_put_link(void *arg)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2015-12-29 13:58:39 -07:00
|
|
|
put_page(arg);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(page_put_link);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2015-11-16 16:26:34 -07:00
|
|
|
int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
|
|
|
|
{
|
2015-12-29 13:58:39 -07:00
|
|
|
DEFINE_DELAYED_CALL(done);
|
2015-11-17 08:20:54 -07:00
|
|
|
int res = readlink_copy(buffer, buflen,
|
|
|
|
page_get_link(dentry, d_inode(dentry),
|
2015-12-29 13:58:39 -07:00
|
|
|
&done));
|
|
|
|
do_delayed_call(&done);
|
2015-11-16 16:26:34 -07:00
|
|
|
return res;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(page_readlink);
|
|
|
|
|
fs: symlink write_begin allocation context fix
With the write_begin/write_end aops, page_symlink was broken because it
could no longer pass a GFP_NOFS type mask into the point where the
allocations happened. They are done in write_begin, which would always
assume that the filesystem can be entered from reclaim. This bug could
cause filesystem deadlocks.
The funny thing with having a gfp_t mask there is that it doesn't really
allow the caller to arbitrarily tinker with the context in which it can be
called. It couldn't ever be GFP_ATOMIC, for example, because it needs to
take the page lock. The only thing any callers care about is __GFP_FS
anyway, so turn that into a single flag.
Add a new flag for write_begin, AOP_FLAG_NOFS. Filesystems can now act on
this flag in their write_begin function. Change __grab_cache_page to
accept a nofs argument as well, to honour that flag (while we're there,
change the name to grab_cache_page_write_begin which is more instructive
and does away with random leading underscores).
This is really a more flexible way to go in the end anyway -- if a
filesystem happens to want any extra allocations aside from the pagecache
ones in ints write_begin function, it may now use GFP_KERNEL (rather than
GFP_NOFS) for common case allocations (eg. ocfs2_alloc_write_ctxt, for a
random example).
[kosaki.motohiro@jp.fujitsu.com: fix ubifs]
[kosaki.motohiro@jp.fujitsu.com: fix fuse]
Signed-off-by: Nick Piggin <npiggin@suse.de>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: <stable@kernel.org> [2.6.28.x]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
[ Cleaned up the calling convention: just pass in the AOP flags
untouched to the grab_cache_page_write_begin() function. That
just simplifies everybody, and may even allow future expansion of the
logic. - Linus ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-04 13:00:53 -07:00
|
|
|
/*
|
|
|
|
* The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
|
|
|
|
*/
|
|
|
|
int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
|
|
|
struct address_space *mapping = inode->i_mapping;
|
2006-03-11 04:27:13 -07:00
|
|
|
struct page *page;
|
2007-10-16 02:25:01 -06:00
|
|
|
void *fsdata;
|
2007-02-16 02:27:18 -07:00
|
|
|
int err;
|
fs: symlink write_begin allocation context fix
With the write_begin/write_end aops, page_symlink was broken because it
could no longer pass a GFP_NOFS type mask into the point where the
allocations happened. They are done in write_begin, which would always
assume that the filesystem can be entered from reclaim. This bug could
cause filesystem deadlocks.
The funny thing with having a gfp_t mask there is that it doesn't really
allow the caller to arbitrarily tinker with the context in which it can be
called. It couldn't ever be GFP_ATOMIC, for example, because it needs to
take the page lock. The only thing any callers care about is __GFP_FS
anyway, so turn that into a single flag.
Add a new flag for write_begin, AOP_FLAG_NOFS. Filesystems can now act on
this flag in their write_begin function. Change __grab_cache_page to
accept a nofs argument as well, to honour that flag (while we're there,
change the name to grab_cache_page_write_begin which is more instructive
and does away with random leading underscores).
This is really a more flexible way to go in the end anyway -- if a
filesystem happens to want any extra allocations aside from the pagecache
ones in ints write_begin function, it may now use GFP_KERNEL (rather than
GFP_NOFS) for common case allocations (eg. ocfs2_alloc_write_ctxt, for a
random example).
[kosaki.motohiro@jp.fujitsu.com: fix ubifs]
[kosaki.motohiro@jp.fujitsu.com: fix fuse]
Signed-off-by: Nick Piggin <npiggin@suse.de>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: <stable@kernel.org> [2.6.28.x]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
[ Cleaned up the calling convention: just pass in the AOP flags
untouched to the grab_cache_page_write_begin() function. That
just simplifies everybody, and may even allow future expansion of the
logic. - Linus ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-04 13:00:53 -07:00
|
|
|
unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
|
|
|
|
if (nofs)
|
|
|
|
flags |= AOP_FLAG_NOFS;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2006-03-25 04:07:57 -07:00
|
|
|
retry:
|
2007-10-16 02:25:01 -06:00
|
|
|
err = pagecache_write_begin(NULL, mapping, 0, len-1,
|
fs: symlink write_begin allocation context fix
With the write_begin/write_end aops, page_symlink was broken because it
could no longer pass a GFP_NOFS type mask into the point where the
allocations happened. They are done in write_begin, which would always
assume that the filesystem can be entered from reclaim. This bug could
cause filesystem deadlocks.
The funny thing with having a gfp_t mask there is that it doesn't really
allow the caller to arbitrarily tinker with the context in which it can be
called. It couldn't ever be GFP_ATOMIC, for example, because it needs to
take the page lock. The only thing any callers care about is __GFP_FS
anyway, so turn that into a single flag.
Add a new flag for write_begin, AOP_FLAG_NOFS. Filesystems can now act on
this flag in their write_begin function. Change __grab_cache_page to
accept a nofs argument as well, to honour that flag (while we're there,
change the name to grab_cache_page_write_begin which is more instructive
and does away with random leading underscores).
This is really a more flexible way to go in the end anyway -- if a
filesystem happens to want any extra allocations aside from the pagecache
ones in ints write_begin function, it may now use GFP_KERNEL (rather than
GFP_NOFS) for common case allocations (eg. ocfs2_alloc_write_ctxt, for a
random example).
[kosaki.motohiro@jp.fujitsu.com: fix ubifs]
[kosaki.motohiro@jp.fujitsu.com: fix fuse]
Signed-off-by: Nick Piggin <npiggin@suse.de>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: <stable@kernel.org> [2.6.28.x]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
[ Cleaned up the calling convention: just pass in the AOP flags
untouched to the grab_cache_page_write_begin() function. That
just simplifies everybody, and may even allow future expansion of the
logic. - Linus ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-04 13:00:53 -07:00
|
|
|
flags, &page, &fsdata);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (err)
|
2007-10-16 02:25:01 -06:00
|
|
|
goto fail;
|
|
|
|
|
2015-11-16 23:07:57 -07:00
|
|
|
memcpy(page_address(page), symname, len-1);
|
2007-10-16 02:25:01 -06:00
|
|
|
|
|
|
|
err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
|
|
|
|
page, fsdata);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (err < 0)
|
|
|
|
goto fail;
|
2007-10-16 02:25:01 -06:00
|
|
|
if (err < len-1)
|
|
|
|
goto retry;
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
mark_inode_dirty(inode);
|
|
|
|
return 0;
|
|
|
|
fail:
|
|
|
|
return err;
|
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(__page_symlink);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2006-03-11 04:27:13 -07:00
|
|
|
int page_symlink(struct inode *inode, const char *symname, int len)
|
|
|
|
{
|
|
|
|
return __page_symlink(inode, symname, len,
|
2015-11-06 17:28:49 -07:00
|
|
|
!mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
|
2006-03-11 04:27:13 -07:00
|
|
|
}
|
2014-03-14 10:20:17 -06:00
|
|
|
EXPORT_SYMBOL(page_symlink);
|
2006-03-11 04:27:13 -07:00
|
|
|
|
2007-02-12 01:55:39 -07:00
|
|
|
const struct inode_operations page_symlink_inode_operations = {
|
2005-04-16 16:20:36 -06:00
|
|
|
.readlink = generic_readlink,
|
2015-11-17 08:20:54 -07:00
|
|
|
.get_link = page_get_link,
|
2005-04-16 16:20:36 -06:00
|
|
|
};
|
|
|
|
EXPORT_SYMBOL(page_symlink_inode_operations);
|