For unlink transactions and block group removal
btrfs_start_transaction_fallback_global_rsv will first try to start an
ordinary transaction and if it fails it will fall back to reserving the
required amount by stealing from the global reserve. This is problematic
because of all the same reasons we had with previous iterations of the
ENOSPC handling, thundering herd. We get a bunch of failures all at
once, everybody tries to allocate from the global reserve, some win and
some lose, we get an ENSOPC.
Fix this behavior by introducing BTRFS_RESERVE_FLUSH_ALL_STEAL. It's
used to mark unlink reservation. To fix this we need to integrate this
logic into the normal ENOSPC infrastructure. We still go through all of
the normal flushing work, and at the moment we begin to fail all the
tickets we try to satisfy any tickets that are allowed to steal by
stealing from the global reserve. If this works we start the flushing
system over again just like we would with a normal ticket satisfaction.
This serializes our global reserve stealing, so we don't have the
thundering herd problem.
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Tested-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Now that we have proper root ref counting everywhere we can kill the
subvol_srcu.
* removal of fs_info::subvol_srcu reduces size of fs_info by 1176 bytes
* the refcount_t used for the references checks for accidental 0->1
in cases where the root lifetime would not be properly protected
* there's a leak detector for roots to catch unfreed roots at umount
time
* SRCU served us well over the years but is was not a proper
synchronization mechanism for some cases
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ update changelog ]
Signed-off-by: David Sterba <dsterba@suse.com>
If we make sure all the inodes have refs on their root we don't have to
worry about the root disappearing while we have open inodes.
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Getting the end offset for a file extent item requires a bit of code since
the extent can be either inline or regular/prealloc. There are some places
all over the code base that open code this logic and in another patch
later in this series it will be needed again. Therefore encapsulate this
logic in a helper function and use it.
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Currently btrfs doesn't provide a migratepage callback for data pages.
It means that fallback_migrate_page() is used to migrate btrfs pages.
fallback_migrate_page() cannot move dirty pages, instead it tries to
flush them (in sync mode) or just fails (in async mode).
In the sync mode pages which are scheduled to be processed by
btrfs_writepage_fixup_worker() can't be effectively flushed by the
migration code, because there is no established way to wait for the
completion of the delayed work.
It all leads to page migration failures.
To fix it the patch implements a btrs-specific migratepage callback,
which is similar to iomap_migrate_page() used by some other fs, except
it does take care of the PagePrivate2 flag which is used for data
ordering purposes.
Reviewed-by: Chris Mason <clm@fb.com>
Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Currently the non-prefixed version is a simple wrapper used to hide
the 4th argument of the prefixed version. This doesn't bring much value
in practice and only makes the code harder to follow by adding another
level of indirection. Rectify this by removing the __ prefix and
have only one public function to release bytes from a block reservation.
No semantic changes.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
This patch removes all haphazard code implementing nocow writers
exclusion from pending snapshot creation and switches to using the drew
lock to ensure this invariant still holds.
'Readers' are snapshot creators from create_snapshot and 'writers' are
nocow writers from buffered write path or btrfs_setsize. This locking
scheme allows for multiple snapshots to happen while any nocow writers
are blocked, since writes to page cache in the nocow path will make
snapshots inconsistent.
So for performance reasons we'd like to have the ability to run multiple
concurrent snapshots and also favors readers in this case. And in case
there aren't pending snapshots (which will be the majority of the cases)
we rely on the percpu's writers counter to avoid cacheline contention.
The main gain from using the drew lock is it's now a lot easier to
reason about the guarantees of the locking scheme and whether there is
some silent breakage lurking.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The tree pointer can be safely read from the page's inode, use it and
drop the redundant argument.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The tree pointer can be safely read from the inode so we can drop the
redundant argument from btrfs_lock_and_flush_ordered_range.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We are now using these for all roots, rename them to btrfs_put_root()
and btrfs_grab_root();
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Now that all callers of btrfs_get_fs_root are subsequently calling
btrfs_grab_fs_root and handling dropping the ref when they are done
appropriately, go ahead and push btrfs_grab_fs_root up into
btrfs_get_fs_root.
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Looking up the inode from an arbitrary tree means we need to hold a ref
on that root.
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
All this does is call btrfs_get_fs_root() with check_ref == true. Just
use btrfs_get_fs_root() so we don't have a bunch of different helpers
that do the same thing.
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Now that we have a safe way to update the i_size, replace all uses of
btrfs_ordered_update_i_size with btrfs_inode_safe_disk_i_size_write.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We want to use this everywhere we modify the file extent items
permanently. These include:
1) Inserting new file extents for writes and prealloc extents.
2) Truncating inode items.
3) btrfs_cont_expand().
4) Insert inline extents.
5) Insert new extents from log replay.
6) Insert a new extent for clone, as it could be past i_size.
7) Hole punching
For hole punching in particular it might seem it's not necessary because
anybody extending would use btrfs_cont_expand, however there is a corner
that still can give us trouble. Start with an empty file and
fallocate KEEP_SIZE 1M-2M
We now have a 0 length file, and a hole file extent from 0-1M, and a
prealloc extent from 1M-2M. Now
punch 1M-1.5M
Because this is past i_size we have
[HOLE EXTENT][ NOTHING ][PREALLOC]
[0 1M][1M 1.5M][1.5M 2M]
with an i_size of 0. Now if we pwrite 0-1.5M we'll increas our i_size
to 1.5M, but our disk_i_size is still 0 until the ordered extent
completes.
However if we now immediately truncate 2M on the file we'll just call
btrfs_cont_expand(inode, 1.5M, 2M), since our old i_size is 1.5M. If we
commit the transaction here and crash we'll expose the gap.
To fix this we need to clear the file extent mapping for the range that
we punched but didn't insert a corresponding file extent for. This will
mean the truncate will only get an disk_i_size set to 1M if we crash
before the finish ordered io happens.
I've written an xfstest to reproduce the problem and validate this fix.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
In order to keep track of where we have file extents on disk, and thus
where it is safe to adjust the i_size to, we need to have a tree in
place to keep track of the contiguous areas we have file extents for.
Add helpers to use this tree, as it's not required for NO_HOLES file
systems. We will use this by setting DIRTY for areas we know we have
file extent item's set, and clearing it when we remove file extent items
for truncation.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
During a rename whiteout, if btrfs_whiteout_for_rename() returns an error
we can end up returning from btrfs_rename() with the log context object
still in the root's log context list - this happens if 'sync_log' was
set to true before we called btrfs_whiteout_for_rename() and it is
dangerous because we end up with a corrupt linked list (root->log_ctxs)
as the log context object was allocated on the stack.
After btrfs_rename() returns, any task that is running btrfs_sync_log()
concurrently can end up crashing because that linked list is traversed by
btrfs_sync_log() (through btrfs_remove_all_log_ctxs()). That results in
the same issue that commit e6c617102c ("Btrfs: fix log context list
corruption after rename exchange operation") fixed.
Fixes: d4682ba03e ("Btrfs: sync log after logging new name")
CC: stable@vger.kernel.org # 4.19+
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
btrfs_lookup_and_bind_dio_csum() does pointer arithmetic which assumes
32-bit checksums. If using a larger checksum, this leads to spurious
failures when a direct I/O read crosses a stripe. This is easy
to reproduce:
# mkfs.btrfs -f --checksum blake2 -d raid0 /dev/vdc /dev/vdd
...
# mount /dev/vdc /mnt
# cd /mnt
# dd if=/dev/urandom of=foo bs=1M count=1 status=none
# dd if=foo of=/dev/null bs=1M iflag=direct status=none
dd: error reading 'foo': Input/output error
# dmesg | tail -1
[ 135.821568] BTRFS warning (device vdc): csum failed root 5 ino 257 off 421888 ...
Fix it by using the actual checksum size.
Fixes: 1e25a2e3ca ("btrfs: don't assume ordered sums to be 4 bytes")
CC: stable@vger.kernel.org # 5.4+
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
While logging the prealloc extents of an inode during a fast fsync we call
btrfs_truncate_inode_items(), through btrfs_log_prealloc_extents(), while
holding a read lock on a leaf of the inode's root (not the log root, the
fs/subvol root), and then that function locks the file range in the inode's
iotree. This can lead to a deadlock when:
* the fsync is ranged
* the file has prealloc extents beyond eof
* writeback for a range different from the fsync range starts
during the fsync
* the size of the file is not sector size aligned
Because when finishing an ordered extent we lock first a file range and
then try to COW the fs/subvol tree to insert an extent item.
The following diagram shows how the deadlock can happen.
CPU 1 CPU 2
btrfs_sync_file()
--> for range [0, 1MiB)
--> inode has a size of
1MiB and has 1 prealloc
extent beyond the
i_size, starting at offset
4MiB
flushes all delalloc for the
range [0MiB, 1MiB) and waits
for the respective ordered
extents to complete
--> before task at CPU 1 locks the
inode, a write into file range
[1MiB, 2MiB + 1KiB) is made
--> i_size is updated to 2MiB + 1KiB
--> writeback is started for that
range, [1MiB, 2MiB + 4KiB)
--> end offset rounded up to
be sector size aligned
btrfs_log_dentry_safe()
btrfs_log_inode_parent()
btrfs_log_inode()
btrfs_log_changed_extents()
btrfs_log_prealloc_extents()
--> does a search on the
inode's root
--> holds a read lock on
leaf X
btrfs_finish_ordered_io()
--> locks range [1MiB, 2MiB + 4KiB)
--> end offset rounded up
to be sector size aligned
--> tries to cow leaf X, through
insert_reserved_file_extent()
--> already locked by the
task at CPU 1
btrfs_truncate_inode_items()
--> gets an i_size of
2MiB + 1KiB, which is
not sector size
aligned
--> tries to lock file
range [2MiB, (u64)-1)
--> the start range
is rounded down
from 2MiB + 1K
to 2MiB to be sector
size aligned
--> but the subrange
[2MiB, 2MiB + 4KiB) is
already locked by
task at CPU 2 which
is waiting to get a
write lock on leaf X
for which we are
holding a read lock
*** deadlock ***
This results in a stack trace like the following, triggered by test case
generic/561 from fstests:
[ 2779.973608] INFO: task kworker/u8:6:247 blocked for more than 120 seconds.
[ 2779.979536] Not tainted 5.6.0-rc2-btrfs-next-53 #1
[ 2779.984503] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[ 2779.990136] kworker/u8:6 D 0 247 2 0x80004000
[ 2779.990457] Workqueue: btrfs-endio-write btrfs_work_helper [btrfs]
[ 2779.990466] Call Trace:
[ 2779.990491] ? __schedule+0x384/0xa30
[ 2779.990521] schedule+0x33/0xe0
[ 2779.990616] btrfs_tree_read_lock+0x19e/0x2e0 [btrfs]
[ 2779.990632] ? remove_wait_queue+0x60/0x60
[ 2779.990730] btrfs_read_lock_root_node+0x2f/0x40 [btrfs]
[ 2779.990782] btrfs_search_slot+0x510/0x1000 [btrfs]
[ 2779.990869] btrfs_lookup_file_extent+0x4a/0x70 [btrfs]
[ 2779.990944] __btrfs_drop_extents+0x161/0x1060 [btrfs]
[ 2779.990987] ? mark_held_locks+0x6d/0xc0
[ 2779.990994] ? __slab_alloc.isra.49+0x99/0x100
[ 2779.991060] ? insert_reserved_file_extent.constprop.19+0x64/0x300 [btrfs]
[ 2779.991145] insert_reserved_file_extent.constprop.19+0x97/0x300 [btrfs]
[ 2779.991222] ? start_transaction+0xdd/0x5c0 [btrfs]
[ 2779.991291] btrfs_finish_ordered_io+0x4f4/0x840 [btrfs]
[ 2779.991405] btrfs_work_helper+0xaa/0x720 [btrfs]
[ 2779.991432] process_one_work+0x26d/0x6a0
[ 2779.991460] worker_thread+0x4f/0x3e0
[ 2779.991481] ? process_one_work+0x6a0/0x6a0
[ 2779.991489] kthread+0x103/0x140
[ 2779.991499] ? kthread_create_worker_on_cpu+0x70/0x70
[ 2779.991515] ret_from_fork+0x3a/0x50
(...)
[ 2780.026211] INFO: task fsstress:17375 blocked for more than 120 seconds.
[ 2780.027480] Not tainted 5.6.0-rc2-btrfs-next-53 #1
[ 2780.028482] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[ 2780.030035] fsstress D 0 17375 17373 0x00004000
[ 2780.030038] Call Trace:
[ 2780.030044] ? __schedule+0x384/0xa30
[ 2780.030052] schedule+0x33/0xe0
[ 2780.030075] lock_extent_bits+0x20c/0x320 [btrfs]
[ 2780.030094] ? btrfs_truncate_inode_items+0xf4/0x1150 [btrfs]
[ 2780.030098] ? rcu_read_lock_sched_held+0x59/0xa0
[ 2780.030102] ? remove_wait_queue+0x60/0x60
[ 2780.030122] btrfs_truncate_inode_items+0x133/0x1150 [btrfs]
[ 2780.030151] ? btrfs_set_path_blocking+0xb2/0x160 [btrfs]
[ 2780.030165] ? btrfs_search_slot+0x379/0x1000 [btrfs]
[ 2780.030195] btrfs_log_changed_extents.isra.8+0x841/0x93e [btrfs]
[ 2780.030202] ? do_raw_spin_unlock+0x49/0xc0
[ 2780.030215] ? btrfs_get_num_csums+0x10/0x10 [btrfs]
[ 2780.030239] btrfs_log_inode+0xf83/0x1124 [btrfs]
[ 2780.030251] ? __mutex_unlock_slowpath+0x45/0x2a0
[ 2780.030275] btrfs_log_inode_parent+0x2a0/0xe40 [btrfs]
[ 2780.030282] ? dget_parent+0xa1/0x370
[ 2780.030309] btrfs_log_dentry_safe+0x4a/0x70 [btrfs]
[ 2780.030329] btrfs_sync_file+0x3f3/0x490 [btrfs]
[ 2780.030339] do_fsync+0x38/0x60
[ 2780.030343] __x64_sys_fdatasync+0x13/0x20
[ 2780.030345] do_syscall_64+0x5c/0x280
[ 2780.030348] entry_SYSCALL_64_after_hwframe+0x49/0xbe
[ 2780.030356] RIP: 0033:0x7f2d80f6d5f0
[ 2780.030361] Code: Bad RIP value.
[ 2780.030362] RSP: 002b:00007ffdba3c8548 EFLAGS: 00000246 ORIG_RAX: 000000000000004b
[ 2780.030364] RAX: ffffffffffffffda RBX: 0000000000000003 RCX: 00007f2d80f6d5f0
[ 2780.030365] RDX: 00007ffdba3c84b0 RSI: 00007ffdba3c84b0 RDI: 0000000000000003
[ 2780.030367] RBP: 000000000000004a R08: 0000000000000001 R09: 00007ffdba3c855c
[ 2780.030368] R10: 0000000000000078 R11: 0000000000000246 R12: 00000000000001f4
[ 2780.030369] R13: 0000000051eb851f R14: 00007ffdba3c85f0 R15: 0000557a49220d90
So fix this by making btrfs_truncate_inode_items() not lock the range in
the inode's iotree when the target root is a log root, since it's not
needed to lock the range for log roots as the protection from the inode's
lock and log_mutex are all that's needed.
Fixes: 28553fa992 ("Btrfs: fix race between shrinking truncate and fiemap")
CC: stable@vger.kernel.org # 4.4+
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
I hit the following warning while running my error injection stress
testing:
WARNING: CPU: 3 PID: 1453 at fs/btrfs/space-info.h:108 btrfs_free_reserved_data_space_noquota+0xfd/0x160 [btrfs]
RIP: 0010:btrfs_free_reserved_data_space_noquota+0xfd/0x160 [btrfs]
Call Trace:
btrfs_free_reserved_data_space+0x4f/0x70 [btrfs]
__btrfs_prealloc_file_range+0x378/0x470 [btrfs]
elfcorehdr_read+0x40/0x40
? elfcorehdr_read+0x40/0x40
? btrfs_commit_transaction+0xca/0xa50 [btrfs]
? dput+0xb4/0x2a0
? btrfs_log_dentry_safe+0x55/0x70 [btrfs]
? btrfs_sync_file+0x30e/0x420 [btrfs]
? do_fsync+0x38/0x70
? __x64_sys_fdatasync+0x13/0x20
? do_syscall_64+0x5b/0x1b0
? entry_SYSCALL_64_after_hwframe+0x44/0xa9
This happens if we fail to insert our reserved file extent. At this
point we've already converted our reservation from ->bytes_may_use to
->bytes_reserved. However once we break we will attempt to free
everything from [cur_offset, end] from ->bytes_may_use, but our extent
reservation will overlap part of this.
Fix this problem by adding ins.offset (our extent allocation size) to
cur_offset so we remove the actual remaining part from ->bytes_may_use.
I validated this fix using my inject-error.py script
python inject-error.py -o should_fail_bio -t cache_save_setup -t \
__btrfs_prealloc_file_range \
-t insert_reserved_file_extent.constprop.0 \
-r "-5" ./run-fsstress.sh
where run-fsstress.sh simply mounts and runs fsstress on a disk.
CC: stable@vger.kernel.org # 4.4+
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The only time we actually leave the path spinning is if we're truncating
a small amount and don't actually free an extent, which is not a common
occurrence. We have to set the path blocking in order to add the
delayed ref anyway, so the first extent we find we set the path to
blocking and stay blocking for the duration of the operation. With the
upcoming file extent map stuff there will be another case that we have
to have the path blocking, so just swap to blocking always.
Note: this patch also fixes a warning after 28553fa992 ("Btrfs: fix
race between shrinking truncate and fiemap") got merged that inserts
extent locks around truncation so the path must not leave spinning locks
after btrfs_search_slot.
[70.794783] BUG: sleeping function called from invalid context at mm/slab.h:565
[70.794834] in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 1141, name: rsync
[70.794863] 5 locks held by rsync/1141:
[70.794876] #0: ffff888417b9c408 (sb_writers#17){.+.+}, at: mnt_want_write+0x20/0x50
[70.795030] #1: ffff888428de28e8 (&type->i_mutex_dir_key#13/1){+.+.}, at: lock_rename+0xf1/0x100
[70.795051] #2: ffff888417b9c608 (sb_internal#2){.+.+}, at: start_transaction+0x394/0x560
[70.795124] #3: ffff888403081768 (btrfs-fs-01){++++}, at: btrfs_try_tree_write_lock+0x2f/0x160
[70.795203] #4: ffff888403086568 (btrfs-fs-00){++++}, at: btrfs_try_tree_write_lock+0x2f/0x160
[70.795222] CPU: 5 PID: 1141 Comm: rsync Not tainted 5.6.0-rc2-backup+ #2
[70.795362] Call Trace:
[70.795374] dump_stack+0x71/0xa0
[70.795445] ___might_sleep.part.96.cold.106+0xa6/0xb6
[70.795459] kmem_cache_alloc+0x1d3/0x290
[70.795471] alloc_extent_state+0x22/0x1c0
[70.795544] __clear_extent_bit+0x3ba/0x580
[70.795557] ? _raw_spin_unlock_irq+0x24/0x30
[70.795569] btrfs_truncate_inode_items+0x339/0xe50
[70.795647] btrfs_evict_inode+0x269/0x540
[70.795659] ? dput.part.38+0x29/0x460
[70.795671] evict+0xcd/0x190
[70.795682] __dentry_kill+0xd6/0x180
[70.795754] dput.part.38+0x2ad/0x460
[70.795765] do_renameat2+0x3cb/0x540
[70.795777] __x64_sys_rename+0x1c/0x20
Reported-by: Dave Jones <davej@codemonkey.org.uk>
Fixes: 28553fa992 ("Btrfs: fix race between shrinking truncate and fiemap")
CC: stable@vger.kernel.org # 4.4+
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ add note ]
Signed-off-by: David Sterba <dsterba@suse.com>
When there is a fiemap executing in parallel with a shrinking truncate
we can end up in a situation where we have extent maps for which we no
longer have corresponding file extent items. This is generally harmless
and at the moment the only consequences are missing file extent items
representing holes after we expand the file size again after the
truncate operation removed the prealloc extent items, and stale
information for future fiemap calls (reporting extents that no longer
exist or may have been reallocated to other files for example).
Consider the following example:
1) Our inode has a size of 128KiB, one 128KiB extent at file offset 0
and a 1MiB prealloc extent at file offset 128KiB;
2) Task A starts doing a shrinking truncate of our inode to reduce it to
a size of 64KiB. Before it searches the subvolume tree for file
extent items to delete, it drops all the extent maps in the range
from 64KiB to (u64)-1 by calling btrfs_drop_extent_cache();
3) Task B starts doing a fiemap against our inode. When looking up for
the inode's extent maps in the range from 128KiB to (u64)-1, it
doesn't find any in the inode's extent map tree, since they were
removed by task A. Because it didn't find any in the extent map
tree, it scans the inode's subvolume tree for file extent items, and
it finds the 1MiB prealloc extent at file offset 128KiB, then it
creates an extent map based on that file extent item and adds it to
inode's extent map tree (this ends up being done by
btrfs_get_extent() <- btrfs_get_extent_fiemap() <-
get_extent_skip_holes());
4) Task A then drops the prealloc extent at file offset 128KiB and
shrinks the 128KiB extent file offset 0 to a length of 64KiB. The
truncation operation finishes and we end up with an extent map
representing a 1MiB prealloc extent at file offset 128KiB, despite we
don't have any more that extent;
After this the two types of problems we have are:
1) Future calls to fiemap always report that a 1MiB prealloc extent
exists at file offset 128KiB. This is stale information, no longer
correct;
2) If the size of the file is increased, by a truncate operation that
increases the file size or by a write into a file offset > 64KiB for
example, we end up not inserting file extent items to represent holes
for any range between 128KiB and 128KiB + 1MiB, since the hole
expansion function, btrfs_cont_expand() will skip hole insertion for
any range for which an extent map exists that represents a prealloc
extent. This causes fsck to complain about missing file extent items
when not using the NO_HOLES feature.
The second issue could be often triggered by test case generic/561 from
fstests, which runs fsstress and duperemove in parallel, and duperemove
does frequent fiemap calls.
Essentially the problems happens because fiemap does not acquire the
inode's lock while truncate does, and fiemap locks the file range in the
inode's iotree while truncate does not. So fix the issue by making
btrfs_truncate_inode_items() lock the file range from the new file size
to (u64)-1, so that it serializes with fiemap.
CC: stable@vger.kernel.org # 4.4+
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We ran into a deadlock in production with the fixup worker. The stack
traces were as follows:
Thread responsible for the writeout, waiting on the page lock
[<0>] io_schedule+0x12/0x40
[<0>] __lock_page+0x109/0x1e0
[<0>] extent_write_cache_pages+0x206/0x360
[<0>] extent_writepages+0x40/0x60
[<0>] do_writepages+0x31/0xb0
[<0>] __writeback_single_inode+0x3d/0x350
[<0>] writeback_sb_inodes+0x19d/0x3c0
[<0>] __writeback_inodes_wb+0x5d/0xb0
[<0>] wb_writeback+0x231/0x2c0
[<0>] wb_workfn+0x308/0x3c0
[<0>] process_one_work+0x1e0/0x390
[<0>] worker_thread+0x2b/0x3c0
[<0>] kthread+0x113/0x130
[<0>] ret_from_fork+0x35/0x40
[<0>] 0xffffffffffffffff
Thread of the fixup worker who is holding the page lock
[<0>] start_delalloc_inodes+0x241/0x2d0
[<0>] btrfs_start_delalloc_roots+0x179/0x230
[<0>] btrfs_alloc_data_chunk_ondemand+0x11b/0x2e0
[<0>] btrfs_check_data_free_space+0x53/0xa0
[<0>] btrfs_delalloc_reserve_space+0x20/0x70
[<0>] btrfs_writepage_fixup_worker+0x1fc/0x2a0
[<0>] normal_work_helper+0x11c/0x360
[<0>] process_one_work+0x1e0/0x390
[<0>] worker_thread+0x2b/0x3c0
[<0>] kthread+0x113/0x130
[<0>] ret_from_fork+0x35/0x40
[<0>] 0xffffffffffffffff
Thankfully the stars have to align just right to hit this. First you
have to end up in the fixup worker, which is tricky by itself (my
reproducer does DIO reads into a MMAP'ed region, so not a common
operation). Then you have to have less than a page size of free data
space and 0 unallocated space so you go down the "commit the transaction
to free up pinned space" path. This was accomplished by a random
balance that was running on the host. Then you get this deadlock.
I'm still in the process of trying to force the deadlock to happen on
demand, but I've hit other issues. I can still trigger the fixup worker
path itself so this patch has been tested in that regard, so the normal
case is fine.
Fixes: 87826df0ec ("btrfs: delalloc for page dirtied out-of-band in fixup worker")
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
For COW, btrfs expects pages dirty pages to have been through a few setup
steps. This includes reserving space for the new block allocations and marking
the range in the state tree for delayed allocation.
A few places outside btrfs will dirty pages directly, especially when unmapping
mmap'd pages. In order for these to properly go through COW, we run them
through a fixup worker to wait for stable pages, and do the delalloc prep.
87826df0ec added a window where the dirty pages were cleaned, but pending
more action from the fixup worker. We clear_page_dirty_for_io() before
we call into writepage, so the page is no longer dirty. The commit
changed it so now we leave the page clean between unlocking it here and
the fixup worker starting at some point in the future.
During this window, page migration can jump in and relocate the page. Once our
fixup work actually starts, it finds page->mapping is NULL and we end up
freeing the page without ever writing it.
This leads to crc errors and other exciting problems, since it screws up the
whole statemachine for waiting for ordered extents. The fix here is to keep
the page dirty while we're waiting for the fixup worker to get to work.
This is accomplished by returning -EAGAIN from btrfs_writepage_cow_fixup
if we queued the page up for fixup, which will cause the writepage
function to redirty the page.
Because we now expect the page to be dirty once it gets to the fixup
worker we must adjust the error cases to call clear_page_dirty_for_io()
on the page. That is the bulk of the patch, but it is not the fix, the
fix is the -EAGAIN from btrfs_writepage_cow_fixup. We cannot separate
these two changes out because the error conditions change with the new
expectations.
Signed-off-by: Chris Mason <clm@fb.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
This series introduces async discard which will use the flag
DISCARD_ASYNC, so rename the original flag to DISCARD_SYNC as it is
synchronously done in transaction commit.
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We only pass this as 1 from __extent_writepage_io(). The parameter
basically means "pretend I didn't pass in a page". This is silly since
we can simply not pass in the page. Get rid of the parameter from
btrfs_get_extent(), and since it's used as a get_extent_t callback,
remove it from get_extent_t and btree_get_extent(), neither of which
need it.
While we're here, let's document btrfs_get_extent().
Signed-off-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
ordered->start, ordered->len, and ordered->disk_len correspond to
fi->disk_bytenr, fi->num_bytes, and fi->disk_num_bytes, respectively.
It's confusing to translate between the two naming schemes. Since a
btrfs_ordered_extent is basically a pending btrfs_file_extent_item,
let's make the former use the naming from the latter.
Note that I didn't touch the names in tracepoints just in case there are
scripts depending on the current naming.
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Snapshot-aware defrag has been disabled since commit 8101c8dbf6
("Btrfs: disable snapshot aware defrag for now") almost 6 years ago.
Let's remove the dead code. If someone is up to the task of bringing it
back, they can dig it up from git.
This is logically a revert of commit 38c227d87c ("Btrfs:
snapshot-aware defrag") except that now we have to clear the
EXTENT_DEFRAG bit to avoid need_force_cow() returning true forever.
The reasons to disable were caused by runtime problems (like long stalls
or memory consumption) on heavily referenced extents (eg. thousands of
snapshots). There were attempts to fix that but never finished.
Current defrag breaks the extent references and some users prefer that
behaviour over the one implemented by snapshot aware (ie. keeping links
for defragmentation). To enable both usecases we'd need to extend
defrag ioctl but let's do that properly from scratch.
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ enhance ]
Signed-off-by: David Sterba <dsterba@suse.com>
We can encode this in the offset parameter: -1 means use the page
offsets, anything else is a valid offset.
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Currently, we have two wrappers for __btrfs_lookup_bio_sums():
btrfs_lookup_bio_sums_dio(), which is used for direct I/O, and
btrfs_lookup_bio_sums(), which is used everywhere else. The only
difference is that the _dio variant looks up csums starting at the given
offset instead of using the page index, which isn't actually direct
I/O-specific. Let's clean up the signature and return value of
__btrfs_lookup_bio_sums(), rename it to btrfs_lookup_bio_sums(), and get
rid of the trivial helpers.
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When you snapshot a subvolume containing a subvolume, you get a
placeholder directory where the subvolume would be. These directories
have their own btrfs_dir_ro_inode_operations.
Al pointed out [1] that these directories can use simple_lookup()
instead of btrfs_lookup(), as they are always empty. Furthermore, they
can use the default generic_permission() instead of btrfs_permission();
the additional checks in the latter don't matter because we can't write
to the directory anyways. Finally, they can use the default
generic_update_time() instead of btrfs_update_time(), as the inode
doesn't exist on disk and doesn't need any special handling.
All together, this means that we can get rid of
btrfs_dir_ro_inode_operations and use simple_dir_inode_operations
instead.
1: https://lore.kernel.org/linux-btrfs/20190929052934.GY26530@ZenIV.linux.org.uk/
Cc: Al Viro <viro@zeniv.linux.org.uk>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ add comment ]
Signed-off-by: David Sterba <dsterba@suse.com>
The condition '!ret2' is always true. commit 717beb96d9 ("Btrfs: fix
regression in btrfs_page_mkwrite() from vm_fault_t conversion") left
behind the check after moving this code out of the goto, so remove the
unused condition check.
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Yunfeng Ye <yeyunfeng@huawei.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
All callers of btrfs_free_reserved_extent (respectively
__btrfs_free_reserved_extent with in set to 0) pass in extents which
have only been reserved but not yet written to. Namely,
* in cow_file_range that function is called only if create_io_em fails
or btrfs_add_ordered_extent fail, both of which happen _before_ any IO
is submitted to the newly reserved range
* in submit_compressed_extents the code flow is similar -
out_free_reserve can be called only before
btrfs_submit_compressed_write which is where any writes to the range
could occur
* btrfs_new_extent_direct also calls btrfs_free_reserved_extent only
if extent_map fails, before any IO is issued
* __btrfs_prealloc_file_range also calls btrfs_free_reserved_extent
in case insertion of the metadata fails
* btrfs_alloc_tree_block again can only be called in case in-memory
operations fail, before any IO is submitted
* btrfs_finish_ordered_io - this is the only caller where discarding
the extent could have a material effect, since it can be called for
an extent which was partially written.
With this change the submission of discards is optimised since discards
are now not being created for extents which are known to not have been
touched on disk.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
If we have the following sequence of events
btrfs sub create A
btrfs sub create A/B
btrfs sub snap A C
mkdir C/foo
mv A/B C/foo
rm -rf *
We will end up with a transaction abort.
The reason for this is because we create a root ref for B pointing to A.
When we create a snapshot of C we still have B in our tree, but because
the root ref points to A and not C we will make it appear to be empty.
The problem happens when we move B into C. This removes the root ref
for B pointing to A and adds a ref of B pointing to C. When we rmdir C
we'll see that we have a ref to our root and remove the root ref,
despite not actually matching our reference name.
Now btrfs_del_root_ref() allowing this to work is a bug as well, however
we know that this inode does not actually point to a root ref in the
first place, so we shouldn't be calling btrfs_del_root_ref() in the
first place and instead simply look up our dir index for this item and
do the rest of the removal.
CC: stable@vger.kernel.org # 4.4+
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
btrfs_unlink_subvol takes the name of the dentry and the root objectid
based on what kind of inode this is, either a real subvolume link or a
empty one that we inherited as a snapshot. We need to fix how we unlink
in the case for BTRFS_EMPTY_SUBVOL_DIR_OBJECTID in the future, so rework
btrfs_unlink_subvol to just take the dentry and handle getting the right
objectid given the type of inode this is. There is no functional change
here, simply pushing the work into btrfs_unlink_subvol() proper.
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When starting writeback for a range that covers part of a preallocated
extent, due to a race with writeback for another range that also covers
another part of the same preallocated extent, we can end up in an infinite
loop.
Consider the following example where for inode 280 we have two dirty
ranges:
range A, from 294912 to 303103, 8192 bytes
range B, from 348160 to 438271, 90112 bytes
and we have the following file extent item layout for our inode:
leaf 38895616 gen 24544 total ptrs 29 free space 13820 owner 5
(...)
item 27 key (280 108 200704) itemoff 14598 itemsize 53
extent data disk bytenr 0 nr 0 type 1 (regular)
extent data offset 0 nr 94208 ram 94208
item 28 key (280 108 294912) itemoff 14545 itemsize 53
extent data disk bytenr 10433052672 nr 81920 type 2 (prealloc)
extent data offset 0 nr 81920 ram 81920
Then the following happens:
1) Writeback starts for range B (from 348160 to 438271), execution of
run_delalloc_nocow() starts;
2) The first iteration of run_delalloc_nocow()'s whil loop leaves us at
the extent item at slot 28, pointing to the prealloc extent item
covering the range from 294912 to 376831. This extent covers part of
our range;
3) An ordered extent is created against that extent, covering the file
range from 348160 to 376831 (28672 bytes);
4) We adjust 'cur_offset' to 376832 and move on to the next iteration of
the while loop;
5) The call to btrfs_lookup_file_extent() leaves us at the same leaf,
pointing to slot 29, 1 slot after the last item (the extent item
we processed in the previous iteration);
6) Because we are a slot beyond the last item, we call btrfs_next_leaf(),
which releases the search path before doing a another search for the
last key of the leaf (280 108 294912);
7) Right after btrfs_next_leaf() released the path, and before it did
another search for the last key of the leaf, writeback for the range
A (from 294912 to 303103) completes (it was previously started at
some point);
8) Upon completion of the ordered extent for range A, the prealloc extent
we previously found got split into two extent items, one covering the
range from 294912 to 303103 (8192 bytes), with a type of regular extent
(and no longer prealloc) and another covering the range from 303104 to
376831 (73728 bytes), with a type of prealloc and an offset of 8192
bytes. So our leaf now has the following layout:
leaf 38895616 gen 24544 total ptrs 31 free space 13664 owner 5
(...)
item 27 key (280 108 200704) itemoff 14598 itemsize 53
extent data disk bytenr 0 nr 0 type 1
extent data offset 0 nr 8192 ram 94208
item 28 key (280 108 208896) itemoff 14545 itemsize 53
extent data disk bytenr 10433142784 nr 86016 type 1
extent data offset 0 nr 86016 ram 86016
item 29 key (280 108 294912) itemoff 14492 itemsize 53
extent data disk bytenr 10433052672 nr 81920 type 1
extent data offset 0 nr 8192 ram 81920
item 30 key (280 108 303104) itemoff 14439 itemsize 53
extent data disk bytenr 10433052672 nr 81920 type 2
extent data offset 8192 nr 73728 ram 81920
9) After btrfs_next_leaf() returns, we have our path pointing to that same
leaf and at slot 30, since it has a key we didn't have before and it's
the first key greater then the key that was previously the last key of
the leaf (key (280 108 294912));
10) The extent item at slot 30 covers the range from 303104 to 376831
which is in our target range, so we process it, despite having already
created an ordered extent against this extent for the file range from
348160 to 376831. This is because we skip to the next extent item only
if its end is less than or equals to the start of our delalloc range,
and not less than or equals to the current offset ('cur_offset');
11) As a result we compute 'num_bytes' as:
num_bytes = min(end + 1, extent_end) - cur_offset;
= min(438271 + 1, 376832) - 376832 = 0
12) We then call create_io_em() for a 0 bytes range starting at offset
376832;
13) Then create_io_em() enters an infinite loop because its calls to
btrfs_drop_extent_cache() do nothing due to the 0 length range
passed to it. So no existing extent maps that cover the offset
376832 get removed, and therefore calls to add_extent_mapping()
return -EEXIST, resulting in an infinite loop. This loop from
create_io_em() is the following:
do {
btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
em->start + em->len - 1, 0);
write_lock(&em_tree->lock);
ret = add_extent_mapping(em_tree, em, 1);
write_unlock(&em_tree->lock);
/*
* The caller has taken lock_extent(), who could race with us
* to add em?
*/
} while (ret == -EEXIST);
Also, each call to btrfs_drop_extent_cache() triggers a warning because
the start offset passed to it (376832) is smaller then the end offset
(376832 - 1) passed to it by -1, due to the 0 length:
[258532.052621] ------------[ cut here ]------------
[258532.052643] WARNING: CPU: 0 PID: 9987 at fs/btrfs/file.c:602 btrfs_drop_extent_cache+0x3f4/0x590 [btrfs]
(...)
[258532.052672] CPU: 0 PID: 9987 Comm: fsx Tainted: G W 5.4.0-rc7-btrfs-next-64 #1
[258532.052673] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.0-0-ga698c8995f-prebuilt.qemu.org 04/01/2014
[258532.052691] RIP: 0010:btrfs_drop_extent_cache+0x3f4/0x590 [btrfs]
(...)
[258532.052695] RSP: 0018:ffffb4be0153f860 EFLAGS: 00010287
[258532.052700] RAX: ffff975b445ee360 RBX: ffff975b44eb3e08 RCX: 0000000000000000
[258532.052700] RDX: 0000000000038fff RSI: 0000000000039000 RDI: ffff975b445ee308
[258532.052700] RBP: 0000000000038fff R08: 0000000000000000 R09: 0000000000000001
[258532.052701] R10: ffff975b513c5c10 R11: 00000000e3c0cfa9 R12: 0000000000039000
[258532.052703] R13: ffff975b445ee360 R14: 00000000ffffffef R15: ffff975b445ee308
[258532.052705] FS: 00007f86a821de80(0000) GS:ffff975b76a00000(0000) knlGS:0000000000000000
[258532.052707] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[258532.052708] CR2: 00007fdacf0f3ab4 CR3: 00000001f9d26002 CR4: 00000000003606f0
[258532.052712] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[258532.052717] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[258532.052717] Call Trace:
[258532.052718] ? preempt_schedule_common+0x32/0x70
[258532.052722] ? ___preempt_schedule+0x16/0x20
[258532.052741] create_io_em+0xff/0x180 [btrfs]
[258532.052767] run_delalloc_nocow+0x942/0xb10 [btrfs]
[258532.052791] btrfs_run_delalloc_range+0x30b/0x520 [btrfs]
[258532.052812] ? find_lock_delalloc_range+0x221/0x250 [btrfs]
[258532.052834] writepage_delalloc+0xe4/0x140 [btrfs]
[258532.052855] __extent_writepage+0x110/0x4e0 [btrfs]
[258532.052876] extent_write_cache_pages+0x21c/0x480 [btrfs]
[258532.052906] extent_writepages+0x52/0xb0 [btrfs]
[258532.052911] do_writepages+0x23/0x80
[258532.052915] __filemap_fdatawrite_range+0xd2/0x110
[258532.052938] btrfs_fdatawrite_range+0x1b/0x50 [btrfs]
[258532.052954] start_ordered_ops+0x57/0xa0 [btrfs]
[258532.052973] ? btrfs_sync_file+0x225/0x490 [btrfs]
[258532.052988] btrfs_sync_file+0x225/0x490 [btrfs]
[258532.052997] __x64_sys_msync+0x199/0x200
[258532.053004] do_syscall_64+0x5c/0x250
[258532.053007] entry_SYSCALL_64_after_hwframe+0x49/0xbe
[258532.053010] RIP: 0033:0x7f86a7dfd760
(...)
[258532.053014] RSP: 002b:00007ffd99af0368 EFLAGS: 00000246 ORIG_RAX: 000000000000001a
[258532.053016] RAX: ffffffffffffffda RBX: 0000000000000ec9 RCX: 00007f86a7dfd760
[258532.053017] RDX: 0000000000000004 RSI: 000000000000836c RDI: 00007f86a8221000
[258532.053019] RBP: 0000000000021ec9 R08: 0000000000000003 R09: 00007f86a812037c
[258532.053020] R10: 0000000000000001 R11: 0000000000000246 R12: 00000000000074a3
[258532.053021] R13: 00007f86a8221000 R14: 000000000000836c R15: 0000000000000001
[258532.053032] irq event stamp: 1653450494
[258532.053035] hardirqs last enabled at (1653450493): [<ffffffff9dec69f9>] _raw_spin_unlock_irq+0x29/0x50
[258532.053037] hardirqs last disabled at (1653450494): [<ffffffff9d4048ea>] trace_hardirqs_off_thunk+0x1a/0x20
[258532.053039] softirqs last enabled at (1653449852): [<ffffffff9e200466>] __do_softirq+0x466/0x6bd
[258532.053042] softirqs last disabled at (1653449845): [<ffffffff9d4c8a0c>] irq_exit+0xec/0x120
[258532.053043] ---[ end trace 8476fce13d9ce20a ]---
Which results in flooding dmesg/syslog since btrfs_drop_extent_cache()
uses WARN_ON() and not WARN_ON_ONCE().
So fix this issue by changing run_delalloc_nocow()'s loop to move to the
next extent item when the current extent item ends at at offset less than
or equals to the current offset instead of the start offset.
Fixes: 80ff385665 ("Btrfs: update nodatacow code v2")
CC: stable@vger.kernel.org # 4.4+
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
If we're rename exchanging two subvols we'll try to lock this lock
twice, which is bad. Just lock once if either of the ino's are subvols.
Fixes: cdd1fedf82 ("btrfs: add support for RENAME_EXCHANGE and RENAME_WHITEOUT")
CC: stable@vger.kernel.org # 4.4+
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Testing with the new fsstress uncovered a pretty nasty deadlock with
lookup and snapshot deletion.
Process A
unlink
-> final iput
-> inode_tree_del
-> synchronize_srcu(subvol_srcu)
Process B
btrfs_lookup <- srcu_read_lock() acquired here
-> btrfs_iget
-> find inode that has I_FREEING set
-> __wait_on_freeing_inode()
We're holding the srcu_read_lock() while doing the iget in order to make
sure our fs root doesn't go away, and then we are waiting for the inode
to finish freeing. However because the free'ing process is doing a
synchronize_srcu() we deadlock.
Fix this by dropping the synchronize_srcu() in inode_tree_del(). We
don't need people to stop accessing the fs root at this point, we're
only adding our empty root to the dead roots list.
A larger much more invasive fix is forthcoming to address how we deal
with fs roots, but this fixes the immediate problem.
Fixes: 76dda93c6a ("Btrfs: add snapshot/subvolume destroy ioctl")
CC: stable@vger.kernel.org # 4.4+
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We can now remove the bdev from extent_map. Previous patches made sure
that bio_set_dev is correctly in all places and that we don't need to
grab it from latest_bdev or pass it around inside the extent map.
Signed-off-by: David Sterba <dsterba@suse.com>
Testing with the new fsstress support for subvolumes uncovered a pretty
bad problem with rename exchange on subvolumes. We're modifying two
different subvolumes, but we only start the transaction on one of them,
so the other one is not added to the dirty root list. This is caught by
btrfs_cow_block() with a warning because the root has not been updated,
however if we do not modify this root again we'll end up pointing at an
invalid root because the root item is never updated.
Fix this by making sure we add the destination root to the trans list,
the same as we do with normal renames. This fixes the corruption.
Fixes: cdd1fedf82 ("btrfs: add support for RENAME_EXCHANGE and RENAME_WHITEOUT")
CC: stable@vger.kernel.org # 4.9+
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The type name is misleading, a single entry is named 'cache' while this
normally means a collection of objects. Rename that everywhere. Also the
identifier was quite long, making function prototypes harder to format.
Suggested-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
In commit "Btrfs: use REQ_CGROUP_PUNT for worker thread submitted bios",
cow_file_range_async gained wbc as a parameter and this makes passing
write flags redundant. Set it inside the function and remove the
parameter.
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The inode delalloc mutex was added a long time ago by commit f248679e86
("Btrfs: add a delalloc mutex to inodes for delalloc reservations"), and
the reason for its introduction is not very clear from the change log. It
claims it solves bogus warnings from lockdep, however it lacks an example
report/warning from lockdep, or any explanation.
Since we have enough concurrentcy protection from the locks of the space
info and block reserve objects, and such lockdep warnings don't seem to
exist anymore (at least on a 5.3 kernel I couldn't get them with fstests,
ltp, fs_mark, etc), remove it, simplifying things a bit and decreasing
the size of the btrfs_inode structure. With some quick fio tests doing
direct IO and mmap writes I couldn't observe any significant performance
increase either (direct IO writes that don't increase the file's size
don't hold the inode's lock for their entire duration and mmap writes
don't hold the inode's lock at all), which are the only type of writes
that could see any performance gain due to less serialization.
Review feedback from Josef:
The problem was taking the i_mutex in mmap, which is how I was
protecting delalloc reservations originally. The delalloc mutex didn't
come with all of the other dependencies. That's what the lockdep
messages were about, removing the lock isn't going to make them appear
again.
We _had_ to lock around this because we used to do tricks to keep from
over-reserving, and if we didn't serialize delalloc reservations we'd
end up with ugly accounting problems when we tried to clean things up.
However with my recentish changes this isn't the case anymore. Every
operation is responsible for reserving its space, and then adding it to
the inode. Then cleaning up is straightforward and can't be mucked up
by other users. So we no longer need the delalloc mutex to safe us from
ourselves.
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
To remove use of extent_map::bdev we need to find a replacement, and the
latest_bdev is the only one we can use here, because inode::i_bdev and
superblock::s_bdev are NULL.
The DIO code uses bdev in two places:
* to read blocksize to perform alignment checks in
do_blockdev_direct_IO, but we do them in btrfs code before any call to
DIO
* in the following call chain:
do_direct_IO
get_more_blocks
sdio->get_block() <-- this is btrfs_get_blocks_direct
subsequently the map_bh->b_dev member is used in clean_bdev_aliases
and dio_new_bio to set the bio's bdev to that of the buffer_head.
However, because we have provided a submit function dio_bio_submit
calls our submission function and ignores the bdev.
So it's safe to pass any valid bdev that's used within the filesystem.
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
In the fixup worker, if we fail to mark the range as delalloc in the io
tree, we must release the previously reserved metadata, as well as update
the outstanding extents counter for the inode, otherwise we leak metadata
space.
In pratice we can't return an error from btrfs_set_extent_delalloc(),
which is just a wrapper around __set_extent_bit(), as for most errors
__set_extent_bit() does a BUG_ON() (or panics which hits a BUG_ON() as
well) and returning an -EEXIST error doesn't happen in this case since
the exclusive bits parameter always has a value of 0 through this code
path. Nevertheless, just fix the error handling in the fixup worker,
in case one day __set_extent_bit() can return an error to this code
path.
Fixes: f3038ee3a3 ("btrfs: Handle btrfs_set_extent_delalloc failure in fixup worker")
CC: stable@vger.kernel.org # 4.19+
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
This function is used only during the final phase of freespace cache
writeout. This is necessary since using the plain btrfs_join_transaction
api is deadlock prone. The deadlock looks like:
T1:
btrfs_commit_transaction
commit_cowonly_roots
btrfs_write_dirty_block_groups
btrfs_wait_cache_io
__btrfs_wait_cache_io
btrfs_wait_ordered_range <-- Triggers ordered IO for freespace
inode and blocks transaction commit
until freespace cache writeout
T2: <-- after T1 has triggered the writeout
finish_ordered_fn
btrfs_finish_ordered_io
btrfs_join_transaction <--- this would block waiting for current
transaction to commit, but since trans
commit is waiting for this writeout to
finish
The special purpose functions prevents it by simply skipping the "wait
for writeout" since it's guaranteed the transaction won't proceed until
we are done.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Async CRCs and compression submit IO through helper threads, which means
they have IO priority inversions when cgroup IO controllers are in use.
This flags all of the writes submitted by btrfs helper threads as
REQ_CGROUP_PUNT. submit_bio() will punt these to dedicated per-blkcg
work items to avoid the priority inversion.
For the compression code, we take a reference on the wbc's blkg css and
pass it down to the async workers.
For the async CRCs, the bio already has the correct css, we just need to
tell the block layer to use REQ_CGROUP_PUNT.
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Chris Mason <clm@fb.com>
Modified-and-reviewed-by: Tejun Heo <tj@kernel.org>
Signed-off-by: David Sterba <dsterba@suse.com>
The btrfs writepages function collects a large range of pages flagged
for delayed allocation, and then sends them down through the COW code
for processing. When compression is on, we allocate one async_chunk
structure for every 512K, and then run those pages through the
compression code for IO submission.
writepages starts all of this off with a single page, locked by the
original call to extent_write_cache_pages(), and it's important to keep
track of this page because it has already been through
clear_page_dirty_for_io().
The btrfs async_chunk struct has a pointer to the locked_page, and when
we're redirtying the page because compression had to fallback to
uncompressed IO, we use page->index to decide if a given async_chunk
struct really owns that page.
But, this is racey. If a given delalloc range is broken up into two
async_chunks (chunkA and chunkB), we can end up with something like
this:
compress_file_range(chunkA)
submit_compress_extents(chunkA)
submit compressed bios(chunkA)
put_page(locked_page)
compress_file_range(chunkB)
...
Or:
async_cow_submit
submit_compressed_extents <--- falls back to buffered writeout
cow_file_range
extent_clear_unlock_delalloc
__process_pages_contig
put_page(locked_pages)
async_cow_submit
The end result is that chunkA is completed and cleaned up before chunkB
even starts processing. This means we can free locked_page() and reuse
it elsewhere. If we get really lucky, it'll have the same page->index
in its new home as it did before.
While we're processing chunkB, we might decide we need to fall back to
uncompressed IO, and so compress_file_range() will call
__set_page_dirty_nobufers() on chunkB->locked_page.
Without cgroups in use, this creates as a phantom dirty page, which
isn't great but isn't the end of the world. What can happen, it can go
through the fixup worker and the whole COW machinery again:
in submit_compressed_extents():
while (async extents) {
...
cow_file_range
if (!page_started ...)
extent_write_locked_range
else if (...)
unlock_page
continue;
This hasn't been observed in practice but is still possible.
With cgroups in use, we might crash in the accounting code because
page->mapping->i_wb isn't set.
BUG: unable to handle kernel NULL pointer dereference at 00000000000000d0
IP: percpu_counter_add_batch+0x11/0x70
PGD 66534e067 P4D 66534e067 PUD 66534f067 PMD 0
Oops: 0000 [#1] SMP DEBUG_PAGEALLOC
CPU: 16 PID: 2172 Comm: rm Not tainted
RIP: 0010:percpu_counter_add_batch+0x11/0x70
RSP: 0018:ffffc9000a97bbe0 EFLAGS: 00010286
RAX: 0000000000000005 RBX: 0000000000000090 RCX: 0000000000026115
RDX: 0000000000000030 RSI: ffffffffffffffff RDI: 0000000000000090
RBP: 0000000000000000 R08: fffffffffffffff5 R09: 0000000000000000
R10: 00000000000260c0 R11: ffff881037fc26c0 R12: ffffffffffffffff
R13: ffff880fe4111548 R14: ffffc9000a97bc90 R15: 0000000000000001
FS: 00007f5503ced480(0000) GS:ffff880ff7200000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00000000000000d0 CR3: 00000001e0459005 CR4: 0000000000360ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
account_page_cleaned+0x15b/0x1f0
__cancel_dirty_page+0x146/0x200
truncate_cleanup_page+0x92/0xb0
truncate_inode_pages_range+0x202/0x7d0
btrfs_evict_inode+0x92/0x5a0
evict+0xc1/0x190
do_unlinkat+0x176/0x280
do_syscall_64+0x63/0x1a0
entry_SYSCALL_64_after_hwframe+0x42/0xb7
The fix here is to make asyc_chunk->locked_page NULL everywhere but the
one async_chunk struct that's allowed to do things to the locked page.
Link: https://lore.kernel.org/linux-btrfs/c2419d01-5c84-3fb4-189e-4db519d08796@suse.com/
Fixes: 771ed689d2 ("Btrfs: Optimize compressed writeback and reads")
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Chris Mason <clm@fb.com>
[ update changelog from mail thread discussion ]
Signed-off-by: David Sterba <dsterba@suse.com>
btrfs_schedule_bio() hands IO off to a helper thread to do the actual
submit_bio() call. This has been used to make sure async crc and
compression helpers don't get stuck on IO submission. To maintain good
performance, over time the IO submission threads duplicated some IO
scheduler characteristics such as high and low priority IOs and they
also made some ugly assumptions about request allocation batch sizes.
All of this cost at least one extra context switch during IO submission,
and doesn't fit well with the modern blkmq IO stack. So, this commit stops
using btrfs_schedule_bio(). We may need to adjust the number of async
helper threads for crcs and compression, but long term it's a better
path.
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The parameter is now always set to NULL and could be dropped. The last
user was get_default_root but that got reworked in 05dbe6837b ("Btrfs:
unify subvol= and subvolid= mounting") and the parameter became unused.
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
