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>
Hit this nice little deadlock. What happens is this
__btrfs_end_transaction with throttle set, --use_count so it equals 0
btrfs_commit_transaction
<somebody else actually manages to start the commit>
btrfs_end_transaction --use_count so now its -1 <== BAD
we just return and wait on the transaction
This is bad because we just return after our use_count is -1 and don't let go
of our num_writer count on the transaction, so the guy committing the
transaction just sits there forever. Fix this by inc'ing our use_count if we're
going to call commit_transaction so that if we call btrfs_end_transaction it's
valid. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
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>
When we balanced the chunks across the devices, BUG_ON() in
__finish_chunk_alloc() was triggered.
------------[ cut here ]------------
kernel BUG at fs/btrfs/volumes.c:2568!
[SNIP]
Call Trace:
[<ffffffffa049525e>] btrfs_alloc_chunk+0x8e/0xa0 [btrfs]
[<ffffffffa04546b0>] do_chunk_alloc+0x330/0x3a0 [btrfs]
[<ffffffffa045c654>] btrfs_reserve_extent+0xb4/0x1f0 [btrfs]
[<ffffffffa045c86b>] btrfs_alloc_free_block+0xdb/0x350 [btrfs]
[<ffffffffa048a8d8>] ? read_extent_buffer+0xd8/0x1d0 [btrfs]
[<ffffffffa04476fd>] __btrfs_cow_block+0x14d/0x5e0 [btrfs]
[<ffffffffa044660d>] ? read_block_for_search+0x14d/0x4d0 [btrfs]
[<ffffffffa0447c9b>] btrfs_cow_block+0x10b/0x240 [btrfs]
[<ffffffffa044dd5e>] btrfs_search_slot+0x49e/0x7a0 [btrfs]
[<ffffffffa044f07d>] btrfs_insert_empty_items+0x8d/0xf0 [btrfs]
[<ffffffffa045e973>] insert_with_overflow+0x43/0x110 [btrfs]
[<ffffffffa045eb0d>] btrfs_insert_dir_item+0xcd/0x1f0 [btrfs]
[<ffffffffa0489bd0>] ? map_extent_buffer+0xb0/0xc0 [btrfs]
[<ffffffff812276ad>] ? rb_insert_color+0x9d/0x160
[<ffffffffa046cc40>] ? inode_tree_add+0xf0/0x150 [btrfs]
[<ffffffffa0474801>] btrfs_add_link+0xc1/0x1c0 [btrfs]
[<ffffffff811dacac>] ? security_inode_init_security+0x1c/0x30
[<ffffffffa04a28aa>] ? btrfs_init_acl+0x4a/0x180 [btrfs]
[<ffffffffa047492f>] btrfs_add_nondir+0x2f/0x70 [btrfs]
[<ffffffffa046af16>] ? btrfs_init_inode_security+0x46/0x60 [btrfs]
[<ffffffffa0474ac0>] btrfs_create+0x150/0x1d0 [btrfs]
[<ffffffff81159c63>] ? generic_permission+0x23/0xb0
[<ffffffff8115b415>] vfs_create+0xa5/0xc0
[<ffffffff8115ce6e>] do_last+0x5fe/0x880
[<ffffffff8115dc0d>] path_openat+0xcd/0x3d0
[<ffffffff8115e029>] do_filp_open+0x49/0xa0
[<ffffffff8116a965>] ? alloc_fd+0x95/0x160
[<ffffffff8114f0c7>] do_sys_open+0x107/0x1e0
[<ffffffff810bcc3f>] ? audit_syscall_entry+0x1bf/0x1f0
[<ffffffff8114f1e0>] sys_open+0x20/0x30
[<ffffffff81484ec2>] system_call_fastpath+0x16/0x1b
[SNIP]
RIP [<ffffffffa049444a>] __finish_chunk_alloc+0x20a/0x220 [btrfs]
The reason is:
Task1 Space balance task
do_chunk_alloc()
__finish_chunk_alloc()
update device info
in the chunk tree
alloc system metadata block
relocate system metadata block group
set system metadata block group
readonly, This block group is the
only one that can allocate space. So
there is no free space that can be
allocated now.
find no space and don't try
to alloc new chunk, and then
return ENOSPC
BUG_ON() in __finish_chunk_alloc()
was triggered.
Fix this bug by allocating a new system metadata chunk before relocating the
old one if we find there is no free space which can be allocated after setting
the old block group to be read-only.
Reported-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Everybody else does this, we need to do it too. If we're syncing, we need to
tag the pages we're going to write for writeback so we don't end up writing the
same stuff over and over again if somebody is constantly redirtying our file.
This will keep us from having latencies with heavy sync workloads. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
So I had this brilliant idea to use atomic counters for outstanding and reserved
extents, but this turned out to be a bad idea. Consider this where we have 1
outstanding extent and 1 reserved extent
Reserver Releaser
atomic_dec(outstanding) now 0
atomic_read(outstanding)+1 get 1
atomic_read(reserved) get 1
don't actually reserve anything because
they are the same
atomic_cmpxchg(reserved, 1, 0)
atomic_inc(outstanding)
atomic_add(0, reserved)
free reserved space for 1 extent
Then the reserver now has no actual space reserved for it, and when it goes to
finish the ordered IO it won't have enough space to do it's allocation and you
get those lovely warnings.
Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Kill the check to see if we have 512mb of reserved space in delalloc and
shrink_delalloc if we do. This causes unexpected latencies and we have other
logic to see if we need to throttle. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
grab_cache_page will use mapping_gfp_mask(), which for all inodes is set to
GFP_HIGHUSER_MOVABLE. So instead use find_or_create_page in all cases where we
need GFP_NOFS so we don't deadlock. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
A user reported a deadlock when copying a bunch of files. This is because they
were low on memory and kthreadd got hung up trying to migrate pages for an
allocation when starting the caching kthread. The page was locked by the person
starting the caching kthread. To fix this we just need to use the async thread
stuff so that the threads are already created and we don't have to worry about
deadlocks. Thanks,
Reported-by: Roman Mamedov <rm@romanrm.ru>
Signed-off-by: Josef Bacik <josef@redhat.com>
First, we can sometimes free the state we're merging, which means anybody who
calls merge_state() may have the state it passed in free'ed. This is
problematic because we could end up caching the state, which makes caching
useless as the state will no longer be part of the tree. So instead of free'ing
the state we passed into merge_state(), set it's end to the other->end and free
the other state. This way we are sure to cache the correct state. Also because
we can merge states together, instead of only using the cache'd state if it's
start == the start we are looking for, go ahead and use it if the start we are
looking for is within the range of the cached state. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
We used to store the checksums of the space cache directly in the space cache,
however that doesn't work out too well if we have more space than we can fit the
checksums into the first page. So instead use the normal checksumming
infrastructure. There were problems with doing this originally but those
problems don't exist now so this works out fine. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
We keep having problems with early enospc, and that's because our method of
making space is inherently racy. The problem is we can have one guy trying to
make space for himself, and in the meantime people come in and steal his
reservation. In order to stop this we make a waitqueue and put anybody who
comes into reserve_metadata_bytes on that waitqueue if somebody is trying to
make more space. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
We have to do weird things when handling enospc in the transaction joining code.
Because we've already joined the transaction we cannot commit the transaction
within the reservation code since it will deadlock, so we have to return EAGAIN
and then make sure we don't retry too many times. Instead of doing this, just
do the reservation the normal way before we join the transaction, that way we
can do whatever we want to try and reclaim space, and then if it fails we know
for sure we are out of space and we can return ENOSPC. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
I've been watching how many btrfs_search_slot()'s we do and I noticed that when
we create a file with selinux enabled we were doing 2 each time we initialize
the security context. That's because we lookup the xattr first so we can delete
it if we're setting a new value to an existing xattr. But in the create case we
don't have any xattrs, so it is completely useless to have the extra lookup. So
re-arrange things so that we only lookup first if we specifically have
XATTR_REPLACE. That way in the basic case we only do 1 search, and in the more
complicated case we do the normal 2 lookups. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
A user reported an error where if we try to balance an fs after a device has
been removed it will blow up. This is because we get an EIO back and this is
where BUG_ON(ret) bites us in the ass. To fix we just exit. Thanks,
Reported-by: Anand Jain <Anand.Jain@oracle.com>
Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
There are three missed mount options settable by user which are not
currently displayed in mount output.
Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
When iputting the inode, We may leave the delayed nodes if they have some
delayed items that have not been dealt with. So when the inode is read again,
we must look up the relative delayed node, and use the information in it to
initialize the inode. Or we will get inconsonant inode information, it may
cause that the same directory index number is allocated again, and hit the
following oops:
[ 5447.554187] err add delayed dir index item(name: pglog_0.965_0) into the
insertion tree of the delayed node(root id: 262, inode id: 258, errno: -17)
[ 5447.569766] ------------[ cut here ]------------
[ 5447.575361] kernel BUG at fs/btrfs/delayed-inode.c:1301!
[SNIP]
[ 5447.790721] Call Trace:
[ 5447.793191] [<ffffffffa0641c4e>] btrfs_insert_dir_item+0x189/0x1bb [btrfs]
[ 5447.800156] [<ffffffffa0651a45>] btrfs_add_link+0x12b/0x191 [btrfs]
[ 5447.806517] [<ffffffffa0651adc>] btrfs_add_nondir+0x31/0x58 [btrfs]
[ 5447.812876] [<ffffffffa0651d6a>] btrfs_create+0xf9/0x197 [btrfs]
[ 5447.818961] [<ffffffff8111f840>] vfs_create+0x72/0x92
[ 5447.824090] [<ffffffff8111fa8c>] do_last+0x22c/0x40b
[ 5447.829133] [<ffffffff8112076a>] path_openat+0xc0/0x2ef
[ 5447.834438] [<ffffffff810c58e2>] ? __perf_event_task_sched_out+0x24/0x44
[ 5447.841216] [<ffffffff8103ecdd>] ? perf_event_task_sched_out+0x59/0x67
[ 5447.847846] [<ffffffff81121a79>] do_filp_open+0x3d/0x87
[ 5447.853156] [<ffffffff811e126c>] ? strncpy_from_user+0x43/0x4d
[ 5447.859072] [<ffffffff8111f1f5>] ? getname_flags+0x2e/0x80
[ 5447.864636] [<ffffffff8111f179>] ? do_getname+0x14b/0x173
[ 5447.870112] [<ffffffff8111f1b7>] ? audit_getname+0x16/0x26
[ 5447.875682] [<ffffffff8112b1ab>] ? spin_lock+0xe/0x10
[ 5447.880882] [<ffffffff81112d39>] do_sys_open+0x69/0xae
[ 5447.886153] [<ffffffff81112db1>] sys_open+0x20/0x22
[ 5447.891114] [<ffffffff813b9aab>] system_call_fastpath+0x16/0x1b
Fix it by reusing the old delayed node.
Reported-by: Jim Schutt <jaschut@sandia.gov>
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Tested-by: Jim Schutt <jaschut@sandia.gov>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
A user reported this bug again where we have more bitmaps than we are supposed
to. This is because we failed to load the free space cache, but don't update
the ctl->total_bitmaps counter when we remove entries from the tree. This patch
fixes this problem and we should be good to go again. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
data parameter should be u64 because a full-sized chunk flags field is
passed instead of 0/1 for distinguishing data from metadata. All
underlying functions expect u64.
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
When we create a new inode, we aren't filling in the
field that records the transaction that last changed this
inode.
If we then go to fsync that inode, it will be skipped because the field
isn't filled in.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Snapshot creation has two phases. One is the initial snapshot setup,
and the second is done during commit, while nobody is allowed to modify
the root we are snapshotting.
The delayed metadata insertion code can break that rule, it does a
delayed inode update on the inode of the parent of the snapshot,
and delayed directory item insertion.
This makes sure to run the pending delayed operations before we
record the snapshot root, which avoids corruptions.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
When allocation fails in btrfs_read_fs_root_no_name, ret is not set
although it is returned, holding a garbage value.
Signed-off-by: David Sterba <dsterba@suse.cz>
Reviewed-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Removes code no longer used. The sysfs file itself is kept, because the
btrfs developers expressed interest in putting new entries to sysfs.
Signed-off-by: Maarten Lankhorst <m.b.lankhorst@gmail.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
The recent commit to get rid of our trans_mutex introduced
some races with block group relocation. The problem is that relocation
needs to do some record keeping about each root, and it was relying
on the transaction mutex to coordinate things in subtle ways.
This fix adds a mutex just for the relocation code and makes sure
it doesn't have a big impact on normal operations. The race is
really fixed in btrfs_record_root_in_trans, which is where we
step back and wait for the relocation code to finish accounting
setup.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
We can lockup if we try to allow new writers join the transaction and we have
flushoncommit set or have a pending snapshot. This is because we set
no_trans_join and then loop around and try to wait for ordered extents again.
The problem is the ordered endio stuff needs to join the transaction, which it
can't do because no_trans_join is set. So instead wait until after this loop to
set no_trans_join and then make sure to wait for num_writers == 1 in case
anybody got started in between us exiting the loop and setting no_trans_join.
This could easily be reproduced by mounting -o flushoncommit and running xfstest
13. It cannot be reproduced with this patch. Thanks,
Reported-by: Jim Schutt <jaschut@sandia.gov>
Signed-off-by: Josef Bacik <josef@redhat.com>
Currently there is nothing protecting the pending_snapshots list on the
transaction. We only hold the directory mutex that we are snapshotting and a
read lock on the subvol_sem, so we could race with somebody else creating a
snapshot in a different directory and end up with list corruption. So protect
this list with the trans_lock. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
The delayed ref patch accidently removed the btrfs_free_path in
btrfs_unlink_subvol, this puts it back and means we don't leak a path. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
The WARN_ON() in start_transaction() was triggered while balancing.
The cause is btrfs_relocate_chunk() started a transaction and
then called iput() on the inode that stores free space cache,
and iput() called btrfs_start_transaction() again.
Reported-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Reviewed-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Get rid of FIXME comment. Uuids from dmesg are now the same as uuids
given by btrfs-progs.
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
When encountering an EIO while reading from a nodatasum extent, we
insert an error record into the inode's failure tree.
btrfs_readpage_end_io_hook returns early for nodatasum inodes. We'd
better clear the failure tree in that case, otherwise the kernel
complains about
BUG extent_state: Objects remaining on kmem_cache_close()
on rmmod.
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
list_splice_init will make delalloc_inodes empty, but without a spinlock
around, this may produce corrupted list head, accessed in many placess,
The race window is very tight and nobody seems to have hit it so far.
Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
The size of struct btrfs_ioctl_fs_info_args is as big as 1KB, so
don't declare the variable on stack.
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Reviewed-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Reorder extent_buffer to remove 8 bytes of alignment padding on 64 bit
builds. This shrinks its size to 128 bytes allowing it to fit into one
fewer cache lines and allows more objects per slab in its kmem_cache.
slabinfo extent_buffer reports :-
before:-
Sizes (bytes) Slabs
----------------------------------
Object : 136 Total : 123
SlabObj: 136 Full : 121
SlabSiz: 4096 Partial: 0
Loss : 0 CpuSlab: 2
Align : 8 Objects: 30
after :-
Object : 128 Total : 4
SlabObj: 128 Full : 2
SlabSiz: 4096 Partial: 0
Loss : 0 CpuSlab: 2
Align : 8 Objects: 32
Signed-off-by: Richard Kennedy <richard@rsk.demon.co.uk>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Normally current->jouranl_info is cleared by commit_transaction. For an
async snap or subvol creation, though, it runs in a work queue. Clear
it in btrfs_commit_transaction_async() to avoid leaking a non-NULL
journal_info when we return to userspace. When the actual commit runs in
the other thread it won't care that it's current->journal_info is already
NULL.
Signed-off-by: Sage Weil <sage@newdream.net>
Tested-by: Jim Schutt <jaschut@sandia.gov>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Josef recently changed the free extent cache to look in
the block group cluster for any bitmaps before trying to
add a new bitmap for the same offset. This avoids BUG_ON()s due
covering duplicate ranges.
But it didn't go quite far enough. A given free range might span
between one or more bitmaps or free space entries. The code has
looping to cover this, but it doesn't check for clustered bitmaps
every time.
This shuffles our gotos to check for a bitmap in the cluster
for every new bitmap entry we try to add.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Scrub starts the workers each time a scrub starts and stops them after it
finished. This patch adds an initialization for the workers before each
start, otherwise the workers behave strangely.
Signed-off-by: Arne Jansen <sensille@gmx.net>
due to the semantics of btrfs_search_slot the path can point to an
invalid slot when ret > 0. This condition went unnoticed, which in
turn could have led to an incomplete scrubbing.
Signed-off-by: Arne Jansen <sensille@gmx.net>
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>
In btrfs_wait_for_commit if we came upon a transaction that had committed we
just exited, but that's bad since we are holding the trans_lock. So break
instead so that the lock is dropped. Thanks,
Reported-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Josef Bacik <josef@redhat.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>
When merging my code into the integration test the second check for duplicate
entries got screwed up. This patch fixes it by dropping ret2 and just using ret
for the return value, and checking if we got an error before adding the bitmap
to the local list. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
I was testing with empty_cluster = 0 to try and reproduce a problem and kept
hitting early enospc panics. This was because our loop logic was a little
confused. So this is what I did
1) Make the loop variable the ultimate decider on wether we should loop again
isntead of checking to see if we had an uncached bg, empty size or empty
cluster.
2) Increment loop before checking to see what we are on to make the loop
definitions make more sense.
3) If we are on the chunk alloc loop don't set empty_size/empty_cluster to 0
unless we didn't actually allocate a chunk. If we did allocate a chunk we
should be able to easily setup a new cluster so clearing
empty_size/empty_cluster makes us less efficient.
This kept me from hitting panics while trying to reproduce the other problem.
Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
Chris Mason