With the tree mod log, we may end up with two roots (the current root and a
rewinded version of it) both pointing to two leaves, l1 and l2, of which l2
had already been cow-ed in the current transaction. If we don't rewind any
tree blocks, we cannot have two roots both pointing to an already cowed tree
block.
Now there is btrfs_next_leaf, which has a leaf locked and wants a lock on
the next (right) leaf. And there is push_leaf_left, which has a (cowed!)
leaf locked and wants a lock on the previous (left) leaf.
In order to solve this dead lock situation, we use try_lock in
btrfs_next_leaf (only in case it's called with a tree mod log time_seq
paramter) and if we fail to get a lock on the next leaf, we give up our lock
on the current leaf and retry from the very beginning.
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
When a MOD_LOG_KEY_ADD operation is rewinded, we remove the key from the
tree block. If its not the last key, removal involves a move operation.
This move operation was explicitly done before this commit.
However, at insertion time, there's a move operation before the actual
addition to make room for the new key, which is recorded in the tree mod
log as well. This means, we must drop the move operation when rewinding the
add operation, because the next operation we'll be rewinding will be the
corresponding MOD_LOG_MOVE_KEYS operation.
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
Several callers of insert_ptr set the tree_mod_log parameter to 0 to avoid
addition to the tree mod log. In fact, we need all of those operations. This
commit simply removes the additional parameter and makes addition to the
tree mod log unconditional.
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
For the tree mod log, we don't log any operations at leaf level. If the root
is at the leaf level (i.e. the tree consists only of the root), then
__tree_mod_log_oldest_root will find a ROOT_REPLACE operation in the log
(because we always log that one no matter which level), but no other
operations.
With this patch __tree_mod_log_oldest_root exits cleanly instead of
BUGging in this situation. get_old_root checks if its really a root at leaf
level in case we don't have any operations and WARNs if this assumption
breaks.
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
gcc was giving an uninit variable warning here. Strictly
speaking we don't need to init it, but this will make things
much less error prone.
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
When adding to the tree modification log, we grab two locks at different
stages. We must not drop the outer lock until we're done with section
protected by the inner lock. This moves the unlock call for the outer lock
to the appropriate position.
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
To make sense of the tree mod log, the backref walker not only needs
btrfs_search_old_slot, but it also called btrfs_next_leaf, which in turn was
calling btrfs_search_slot. This obviously didn't give the correct result.
This commit adds btrfs_next_old_leaf, a drop-in replacement for
btrfs_next_leaf with a time_seq parameter. If it is zero, it behaves exactly
like btrfs_next_leaf. If it is non-zero, it will use btrfs_search_old_slot
with this time_seq parameter.
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
In __tree_mod_log_oldest_root() we must return the found operation even if
it's not a ROOT_REPLACE operation. Otherwise, the caller assumes that there
are no operations to be rewinded and returns immediately.
The code in the caller is modified to improve readability.
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
get_old_root could race with root node updates because we weren't locking
the node early enough. Use btrfs_read_lock_root_node to grab the root locked
in the very beginning and release the lock as soon as possible (just like
btrfs_search_slot does).
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
This is a leftover from cleanup patch 559af821. Before the cleanup,
btrfs_header_nritems was called inside an if condition. As it has no side
effects we need to preserve here, it should simply be dropped.
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
When we rewind REMOVE_WHILE_FREEING operations, there's code that allocates
a fresh buffer instead of cloning the old one. Setting that buffer's level
correctly was missing in this case.
When rewinding a MOVE_KEYS operation, btrfs_node_key_ptr_offset(slot) was
missing for memmove_extent_buffer()'s arguments.
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
Logging for del_ptr when we're not deleting the last pointer was wrong. This
fixes both, duplicate log entries and log sequence.
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
btrfs_read_buffer() has the possibility of returning the error.
Therefore, I add the code in which the return value of btrfs_read_buffer()
is checked.
Signed-off-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
The tree modification log together with the current state of the tree gives
a consistent, old version of the tree. btrfs_search_old_slot is used to
search through this old version and return old (dummy!) extent buffers.
Naturally, this function cannot do any tree modifications.
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
Record all relevant modifications to block pointers in the tree mod log so
that we can rewind them later on for backref walking.
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
When running functions that can make changes to the internal trees
(e.g. btrfs_search_slot), we check if somebody may be interested in the
block we're currently modifying. If so, we record our modification to be
able to rewind it later on.
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
The tree mod log will log modifications made fs-tree nodes. Most
modifications are done by autobalance of the tree. Such changes are recorded
as long as a block entry exists. When released, the log is cleaned.
With the tree modification log, it's possible to reconstruct a consistent
old state of the tree. This is required to do backref walking on a busy
file system.
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
Three callers of btrfs_free_tree_block or btrfs_alloc_tree_block passed
parameter for_cow = 1. In fact, these two functions should never mark
their tree modification operations as for_cow, because they can change
the number of blocks referenced by a tree.
Hence, we remove the extra for_cow parameter from these functions and
make them pass a zero down.
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
verify_parent_transid needs to lock the extent range to make
sure no IO is underway, and so it can safely clear the
uptodate bits if our checks fail.
But, a few callers are using it with spinlocks held. Most
of the time, the generation numbers are going to match, and
we don't want to switch to a blocking lock just for the error
case. This adds an atomic flag to verify_parent_transid,
and changes it to return EAGAIN if it needs to block to
properly verifiy things.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
add_root_to_dirty_list happens once at the very beginning of the
transaction, but it is still racey.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
btrfs_search_slot sometimes needs write locks on high levels of
the tree. It remembers the highest level that needs a write lock
and will use that for all future searches through the tree in a given
call.
But, very often we'll just cow the top level or the level below and we
won't really need write locks on the root again after that. This patch
changes things to adjust the write lock requirement as it unlocks
levels.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Because btrfs cow's we can end up with extent buffers that are no longer
necessary just sitting around in memory. So instead of evicting these pages, we
could end up evicting things we actually care about. Thus we have
free_extent_buffer_stale for use when we are freeing tree blocks. This will
make it so that the ref for the eb being in the radix tree is dropped as soon as
possible and then is freed when the refcount hits 0 instead of waiting to be
released by releasepage. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
btrfs currently handles most errors with BUG_ON. This patch is a work-in-
progress but aims to handle most errors other than internal logic
errors and ENOMEM more gracefully.
This iteration prevents most crashes but can run into lockups with
the page lock on occasion when the timing "works out."
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
balace_level() seems to deal with missing tree nodes by BUG_ON(). Instead,
we can easily just set the file system readonly and bubble -EROFS back up
the stack.
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
__btrfs_cow_block(), the only caller of update_ref_for_cow() will BUG_ON()
any error return. Instead, we can go read-only fs as update_ref_for_cow()
manipulates disk data in a way which doesn't look like it's easily rolled
back.
Signed-off-by: Mark Fasheh <mfasheh@suse.de>
update_ref_for_cow() will BUG_ON() after it's call to
btrfs_lookup_extent_info() if no existing references are found. Since refs
are computed directly from disk, this should be treated as a corruption
instead of a logic error.
Signed-off-by: Mark Fasheh <mfasheh@suse.de>
The only caller of update_ref_for_cow() is __btrfs_cow_block() which was
originally ignoring any return values. update_ref_for_cow() however doesn't
look like a candidate to become a void function - there are a few places
where errors can occur.
So instead I changed update_ref_for_cow() to bubble all errors up (instead
of BUG_ON). __btrfs_cow_block() was then updated to catch and BUG_ON() any
errors from update_ref_for_cow(). The end effect is that we have no change
in behavior, but about 8 different places where a BUG_ON(ret) was removed.
Obviously a future patch will have to address the BUG_ON() in
__btrfs_cow_block().
Signed-off-by: Mark Fasheh <mfasheh@suse.de>
Add a for_cow parameter to add_delayed_*_ref and pass the appropriate value
from every call site. The for_cow parameter will later on be used to
determine if a ref will change anything with respect to qgroups.
Delayed refs coming from relocation are always counted as for_cow, as they
don't change subvol quota.
Also pass in the fs_info for later use.
btrfs_find_all_roots() will use this as an optimization, as changes that are
for_cow will not change anything with respect to which root points to a
certain leaf. Thus, we don't need to add the current sequence number to
those delayed refs.
Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
The btrfs snapshotting code requires that once a root has been
snapshotted, we don't change it during a commit.
But there are two cases to lead to tree corruptions:
1) multi-thread snapshots can commit serveral snapshots in a transaction,
and this may change the src root when processing the following pending
snapshots, which lead to the former snapshots corruptions;
2) the free inode cache was changing the roots when it root the cache,
which lead to corruptions.
This fixes things by making sure we force COW the block after we create a
snapshot during commiting a transaction, then any changes to the roots
will result in COW, and we get all the fs roots and snapshot roots to be
consistent.
Signed-off-by: Liu Bo <liubo2009@cn.fujitsu.com>
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Before the reader/writer locks, btrfs_next_leaf needed to keep
the path blocking to avoid making lockdep upset.
Now that btrfs_next_leaf only takes read locks, this isn't required.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
The btrfs metadata btree is the source of significant
lock contention, especially in the root node. This
commit changes our locking to use a reader/writer
lock.
The lock is built on top of rw spinlocks, and it
extends the lock tracking to remember if we have a
read lock or a write lock when we go to blocking. Atomics
count the number of blocking readers or writers at any
given time.
It removes all of the adaptive spinning from the old code
and uses only the spinning/blocking hints inside of btrfs
to decide when it should continue spinning.
In read heavy workloads this is dramatically faster. In write
heavy workloads we're still faster because of less contention
on the root node lock.
We suffer slightly in dbench because we schedule more often
during write locks, but all other benchmarks so far are improved.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
The extent_buffers have a very complex interface where
we use HIGHMEM for metadata and try to cache a kmap mapping
to access the memory.
The next commit adds reader/writer locks, and concurrent use
of this kmap cache would make it even more complex.
This commit drops the ability to use HIGHMEM with extent buffers,
and rips out all of the related code.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Arne's scrub stuff exposed a problem with mapping the extent buffer in
reada_for_search. He searches the commit root with multiple threads and with
skip_locking set, so we can race and overwrite node->map_token since node isn't
locked. So fix this so that we only map the extent buffer if we don't already
have a map_token and skip_locking isn't set. Without this patch scrub would
panic almost immediately, with the patch it doesn't panic anymore. Thanks,
Reported-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Josef Bacik <josef@redhat.com>
Currently, btrfs_truncate_item and btrfs_extend_item returns only 0.
So, the check by BUG_ON in the caller is unnecessary.
Signed-off-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Our readahead is sort of sloppy, and really isn't always needed. For example if
ls is doing a stating ls (which is the default) it's going to stat in non-disk
order, so if say you have a directory with a stupid amount of files, readahead
is going to do nothing but waste time in the case of doing the stat. Taking the
unconditional readahead out made my test go from 57 minutes to 36 minutes. This
means that everywhere we do loop through the tree we want to make sure we do set
path->reada properly, so I went through and found all of the places where we
loop through the path and set reada to 1. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
We have a bit of debugging in btrfs_search_slot to make sure the level of the
cow block is the same as the original block we were cow'ing. I don't think I've
ever seen this tripped, so kill it. This saves us 2 kmap's per level in our
search. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
If we have particularly full nodes, we could call btrfs_node_blockptr up to 32
times, which is 32 pairs of kmap/kunmap, which _sucks_. So go ahead and map the
extent buffer while we look for readahead targets. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
Changelog V5 -> V6:
- Fix oom when the memory load is high, by storing the delayed nodes into the
root's radix tree, and letting btrfs inodes go.
Changelog V4 -> V5:
- Fix the race on adding the delayed node to the inode, which is spotted by
Chris Mason.
- Merge Chris Mason's incremental patch into this patch.
- Fix deadlock between readdir() and memory fault, which is reported by
Itaru Kitayama.
Changelog V3 -> V4:
- Fix nested lock, which is reported by Itaru Kitayama, by updating space cache
inode in time.
Changelog V2 -> V3:
- Fix the race between the delayed worker and the task which does delayed items
balance, which is reported by Tsutomu Itoh.
- Modify the patch address David Sterba's comment.
- Fix the bug of the cpu recursion spinlock, reported by Chris Mason
Changelog V1 -> V2:
- break up the global rb-tree, use a list to manage the delayed nodes,
which is created for every directory and file, and used to manage the
delayed directory name index items and the delayed inode item.
- introduce a worker to deal with the delayed nodes.
Compare with Ext3/4, the performance of file creation and deletion on btrfs
is very poor. the reason is that btrfs must do a lot of b+ tree insertions,
such as inode item, directory name item, directory name index and so on.
If we can do some delayed b+ tree insertion or deletion, we can improve the
performance, so we made this patch which implemented delayed directory name
index insertion/deletion and delayed inode update.
Implementation:
- introduce a delayed root object into the filesystem, that use two lists to
manage the delayed nodes which are created for every file/directory.
One is used to manage all the delayed nodes that have delayed items. And the
other is used to manage the delayed nodes which is waiting to be dealt with
by the work thread.
- Every delayed node has two rb-tree, one is used to manage the directory name
index which is going to be inserted into b+ tree, and the other is used to
manage the directory name index which is going to be deleted from b+ tree.
- introduce a worker to deal with the delayed operation. This worker is used
to deal with the works of the delayed directory name index items insertion
and deletion and the delayed inode update.
When the delayed items is beyond the lower limit, we create works for some
delayed nodes and insert them into the work queue of the worker, and then
go back.
When the delayed items is beyond the upper bound, we create works for all
the delayed nodes that haven't been dealt with, and insert them into the work
queue of the worker, and then wait for that the untreated items is below some
threshold value.
- When we want to insert a directory name index into b+ tree, we just add the
information into the delayed inserting rb-tree.
And then we check the number of the delayed items and do delayed items
balance. (The balance policy is above.)
- When we want to delete a directory name index from the b+ tree, we search it
in the inserting rb-tree at first. If we look it up, just drop it. If not,
add the key of it into the delayed deleting rb-tree.
Similar to the delayed inserting rb-tree, we also check the number of the
delayed items and do delayed items balance.
(The same to inserting manipulation)
- When we want to update the metadata of some inode, we cached the data of the
inode into the delayed node. the worker will flush it into the b+ tree after
dealing with the delayed insertion and deletion.
- We will move the delayed node to the tail of the list after we access the
delayed node, By this way, we can cache more delayed items and merge more
inode updates.
- If we want to commit transaction, we will deal with all the delayed node.
- the delayed node will be freed when we free the btrfs inode.
- Before we log the inode items, we commit all the directory name index items
and the delayed inode update.
I did a quick test by the benchmark tool[1] and found we can improve the
performance of file creation by ~15%, and file deletion by ~20%.
Before applying this patch:
Create files:
Total files: 50000
Total time: 1.096108
Average time: 0.000022
Delete files:
Total files: 50000
Total time: 1.510403
Average time: 0.000030
After applying this patch:
Create files:
Total files: 50000
Total time: 0.932899
Average time: 0.000019
Delete files:
Total files: 50000
Total time: 1.215732
Average time: 0.000024
[1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3
Many thanks for Kitayama-san's help!
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Reviewed-by: David Sterba <dave@jikos.cz>
Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
parameter tree root it's not used since commit
5f39d397df ("Btrfs: Create extent_buffer
interface for large blocksizes")
Signed-off-by: David Sterba <dsterba@suse.cz>
