openwrt/staging/blogic.git
5 years agobtrfs: Fix bound checking in qgroup_trace_new_subtree_blocks
Nikolay Borisov [Mon, 18 Mar 2019 15:45:19 +0000 (17:45 +0200)]
btrfs: Fix bound checking in qgroup_trace_new_subtree_blocks

If 'cur_level' is 7  then the bound checking at the top of the function
will actually pass. Later on, it's possible to dereference
ds_path->nodes[cur_level+1] which will be an out of bounds.

The correct check will be cur_level >= BTRFS_MAX_LEVEL - 1 .

Fixes-coverty-id: 1440918
Fixes-coverty-id: 1440911
Fixes: ea49f3e73c4b ("btrfs: qgroup: Introduce function to find all new tree blocks of reloc tree")
CC: stable@vger.kernel.org # 4.20+
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>
5 years agobtrfs: raid56: properly unmap parity page in finish_parity_scrub()
Andrea Righi [Thu, 14 Mar 2019 07:56:28 +0000 (08:56 +0100)]
btrfs: raid56: properly unmap parity page in finish_parity_scrub()

Parity page is incorrectly unmapped in finish_parity_scrub(), triggering
a reference counter bug on i386, i.e.:

 [ 157.662401] kernel BUG at mm/highmem.c:349!
 [ 157.666725] invalid opcode: 0000 [#1] SMP PTI

The reason is that kunmap(p_page) was completely left out, so we never
did an unmap for the p_page and the loop unmapping the rbio page was
iterating over the wrong number of stripes: unmapping should be done
with nr_data instead of rbio->real_stripes.

Test case to reproduce the bug:

 - create a raid5 btrfs filesystem:
   # mkfs.btrfs -m raid5 -d raid5 /dev/sdb /dev/sdc /dev/sdd /dev/sde

 - mount it:
   # mount /dev/sdb /mnt

 - run btrfs scrub in a loop:
   # while :; do btrfs scrub start -BR /mnt; done

BugLink: https://bugs.launchpad.net/bugs/1812845
Fixes: 5a6ac9eacb49 ("Btrfs, raid56: support parity scrub on raid56")
CC: stable@vger.kernel.org # 4.4+
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: don't report readahead errors and don't update statistics
David Sterba [Thu, 7 Mar 2019 14:40:50 +0000 (15:40 +0100)]
btrfs: don't report readahead errors and don't update statistics

As readahead is an optimization, all errors are usually filtered out,
but still properly handled when the real read call is done. The commit
5e9d398240b2 ("btrfs: readpages() should submit IO as read-ahead") added
REQ_RAHEAD to readpages() because that's only used for readahead
(despite what one would expect from the callback name).

This causes a flood of messages and inflated read error stats, so skip
reporting in case it's readahead.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=202403
Reported-by: LimeTech <tomm@lime-technology.com>
Fixes: 5e9d398240b2 ("btrfs: readpages() should submit IO as read-ahead")
CC: stable@vger.kernel.org # 4.19+
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agoBtrfs: fix file corruption after snapshotting due to mix of buffered/DIO writes
Filipe Manana [Wed, 27 Feb 2019 13:42:30 +0000 (13:42 +0000)]
Btrfs: fix file corruption after snapshotting due to mix of buffered/DIO writes

When we are mixing buffered writes with direct IO writes against the same
file and snapshotting is happening concurrently, we can end up with a
corrupt file content in the snapshot. Example:

1) Inode/file is empty.

2) Snapshotting starts.

2) Buffered write at offset 0 length 256Kb. This updates the i_size of the
   inode to 256Kb, disk_i_size remains zero. This happens after the task
   doing the snapshot flushes all existing delalloc.

3) DIO write at offset 256Kb length 768Kb. Once the ordered extent
   completes it sets the inode's disk_i_size to 1Mb (256Kb + 768Kb) and
   updates the inode item in the fs tree with a size of 1Mb (which is
   the value of disk_i_size).

4) The dealloc for the range [0, 256Kb[ did not start yet.

5) The transaction used in the DIO ordered extent completion, which updated
   the inode item, is committed by the snapshotting task.

6) Snapshot creation completes.

7) Dealloc for the range [0, 256Kb[ is flushed.

After that when reading the file from the snapshot we always get zeroes for
the range [0, 256Kb[, the file has a size of 1Mb and the data written by
the direct IO write is found. From an application's point of view this is
a corruption, since in the source subvolume it could never read a version
of the file that included the data from the direct IO write without the
data from the buffered write included as well. In the snapshot's tree,
file extent items are missing for the range [0, 256Kb[.

The issue, obviously, does not happen when using the -o flushoncommit
mount option.

Fix this by flushing delalloc for all the roots that are about to be
snapshotted when committing a transaction. This guarantees total ordering
when updating the disk_i_size of an inode since the flush for dealloc is
done when a transaction is in the TRANS_STATE_COMMIT_START state and wait
is done once no more external writers exist. This is similar to what we
do when using the flushoncommit mount option, but we do it only if the
transaction has snapshots to create and only for the roots of the
subvolumes to be snapshotted. The bulk of the dealloc is flushed in the
snapshot creation ioctl, so the flush work we do inside the transaction
is minimized.

This issue, involving buffered and direct IO writes with snapshotting, is
often triggered by fstest btrfs/078, and got reported by fsck when not
using the NO_HOLES features, for example:

  $ cat results/btrfs/078.full
  (...)
  _check_btrfs_filesystem: filesystem on /dev/sdc is inconsistent
  *** fsck.btrfs output ***
  [1/7] checking root items
  [2/7] checking extents
  [3/7] checking free space cache
  [4/7] checking fs roots
  root 258 inode 264 errors 100, file extent discount
  Found file extent holes:
        start: 524288, len: 65536
  ERROR: errors found in fs roots

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: remove WARN_ON in log_dir_items
Josef Bacik [Wed, 6 Mar 2019 22:13:04 +0000 (17:13 -0500)]
btrfs: remove WARN_ON in log_dir_items

When Filipe added the recursive directory logging stuff in
2f2ff0ee5e430 ("Btrfs: fix metadata inconsistencies after directory
fsync") he specifically didn't take the directory i_mutex for the
children directories that we need to log because of lockdep.  This is
generally fine, but can lead to this WARN_ON() tripping if we happen to
run delayed deletion's in between our first search and our second search
of dir_item/dir_indexes for this directory.  We expect this to happen,
so the WARN_ON() isn't necessary.  Drop the WARN_ON() and add a comment
so we know why this case can happen.

CC: stable@vger.kernel.org # 4.4+
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agoBtrfs: fix incorrect file size after shrinking truncate and fsync
Filipe Manana [Mon, 4 Mar 2019 14:06:12 +0000 (14:06 +0000)]
Btrfs: fix incorrect file size after shrinking truncate and fsync

If we do a shrinking truncate against an inode which is already present
in the respective log tree and then rename it, as part of logging the new
name we end up logging an inode item that reflects the old size of the
file (the one which we previously logged) and not the new smaller size.
The decision to preserve the size previously logged was added by commit
1a4bcf470c886b ("Btrfs: fix fsync data loss after adding hard link to
inode") in order to avoid data loss after replaying the log. However that
decision is only needed for the case the logged inode size is smaller then
the current size of the inode, as explained in that commit's change log.
If the current size of the inode is smaller then the previously logged
size, we know a shrinking truncate happened and therefore need to use
that smaller size.

Example to trigger the problem:

  $ mkfs.btrfs -f /dev/sdb
  $ mount /dev/sdb /mnt

  $ xfs_io -f -c "pwrite -S 0xab 0 8000" /mnt/foo
  $ xfs_io -c "fsync" /mnt/foo
  $ xfs_io -c "truncate 3000" /mnt/foo

  $ mv /mnt/foo /mnt/bar
  $ xfs_io -c "fsync" /mnt/bar

  <power failure>

  $ mount /dev/sdb /mnt
  $ od -t x1 -A d /mnt/bar
  0000000 ab ab ab ab ab ab ab ab ab ab ab ab ab ab ab ab
  *
  0008000

Once we rename the file, we log its name (and inode item), and because
the inode was already logged before in the current transaction, we log it
with a size of 8000 bytes because that is the size we previously logged
(with the first fsync). As part of the rename, besides logging the inode,
we do also sync the log, which is done since commit d4682ba03ef618
("Btrfs: sync log after logging new name"), so the next fsync against our
inode is effectively a no-op, since no new changes happened since the
rename operation. Even if did not sync the log during the rename
operation, the same problem (fize size of 8000 bytes instead of 3000
bytes) would be visible after replaying the log if the log ended up
getting synced to disk through some other means, such as for example by
fsyncing some other modified file. In the example above the fsync after
the rename operation is there just because not every filesystem may
guarantee logging/journalling the inode (and syncing the log/journal)
during the rename operation, for example it is needed for f2fs, but not
for ext4 and xfs.

Fix this scenario by, when logging a new name (which is triggered by
rename and link operations), using the current size of the inode instead
of the previously logged inode size.

A test case for fstests follows soon.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=202695
CC: stable@vger.kernel.org # 4.4+
Reported-by: Seulbae Kim <seulbae@gatech.edu>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: zstd: ensure reclaim timer is properly cleaned up
Dennis Zhou [Fri, 22 Feb 2019 19:53:48 +0000 (14:53 -0500)]
btrfs: zstd: ensure reclaim timer is properly cleaned up

The timer function, zstd_reclaim_timer_fn(), reschedules itself under
certain conditions. When cleaning up, take the lock and remove all
workspaces. This prevents the timer from rearming itself. Lastly, switch
to del_timer_sync() to ensure that the timer function can't trigger as
we're unloading.

Signed-off-by: Dennis Zhou <dennis@kernel.org>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: move ulist allocation out of transaction in quota enable
David Sterba [Wed, 19 Dec 2018 18:47:37 +0000 (19:47 +0100)]
btrfs: move ulist allocation out of transaction in quota enable

The allocation happens with GFP_KERNEL after a transaction has been
started, this can potentially cause deadlock if reclaim tries to get the
memory by flushing filesystem data.

The fs_info::qgroup_ulist is not used during transaction start when
quotas are not enabled. The status bit BTRFS_FS_QUOTA_ENABLED is set
later in btrfs_quota_enable so it's safe to move it before the
transaction start.

Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: save drop_progress if we drop refs at all
Josef Bacik [Wed, 6 Feb 2019 20:46:15 +0000 (15:46 -0500)]
btrfs: save drop_progress if we drop refs at all

Previously we only updated the drop_progress key if we were in the
DROP_REFERENCE stage of snapshot deletion.  This is because the
UPDATE_BACKREF stage checks the flags of the blocks it's converting to
FULL_BACKREF, so if we go over a block we processed before it doesn't
matter, we just don't do anything.

The problem is in do_walk_down() we will go ahead and drop the roots
reference to any blocks that we know we won't need to walk into.

Given subvolume A and snapshot B.  The root of B points to all of the
nodes that belong to A, so all of those nodes have a refcnt > 1.  If B
did not modify those blocks it'll hit this condition in do_walk_down

if (!wc->update_ref ||
    generation <= root->root_key.offset)
goto skip;

and in "goto skip" we simply do a btrfs_free_extent() for that bytenr
that we point at.

Now assume we modified some data in B, and then took a snapshot of B and
call it C.  C points to all the nodes in B, making every node the root
of B points to have a refcnt > 1.  This assumes the root level is 2 or
higher.

We delete snapshot B, which does the above work in do_walk_down,
free'ing our ref for nodes we share with A that we didn't modify.  Now
we hit a node we _did_ modify, thus we own.  We need to walk down into
this node and we set wc->stage == UPDATE_BACKREF.  We walk down to level
0 which we also own because we modified data.  We can't walk any further
down and thus now need to walk up and start the next part of the
deletion.  Now walk_up_proc is supposed to put us back into
DROP_REFERENCE, but there's an exception to this

if (level < wc->shared_level)
goto out;

we are at level == 0, and our shared_level == 1.  We skip out of this
one and go up to level 1.  Since path->slots[1] < nritems we
path->slots[1]++ and break out of walk_up_tree to stop our transaction
and loop back around.  Now in btrfs_drop_snapshot we have this snippet

if (wc->stage == DROP_REFERENCE) {
level = wc->level;
btrfs_node_key(path->nodes[level],
       &root_item->drop_progress,
       path->slots[level]);
root_item->drop_level = level;
}

our stage == UPDATE_BACKREF still, so we don't update the drop_progress
key.  This is a problem because we would have done btrfs_free_extent()
for the nodes leading up to our current position.  If we crash or
unmount here and go to remount we'll start over where we were before and
try to free our ref for blocks we've already freed, and thus abort()
out.

Fix this by keeping track of the last place we dropped a reference for
our block in do_walk_down.  Then if wc->stage == UPDATE_BACKREF we know
we'll start over from a place we meant to, and otherwise things continue
to work as they did before.

I have a complicated reproducer for this problem, without this patch
we'll fail to fsck the fs when replaying the log writes log.  With this
patch we can replay the whole log without any fsck or mount failures.

The steps to reproduce this easily are sort of tricky, I had to add a
couple of debug patches to the kernel in order to make it easy,
basically I just needed to make sure we did actually commit the
transaction every time we finished a walk_down_tree/walk_up_tree combo.

The reproducer:

1) Creates a base subvolume.
2) Creates 100k files in the subvolume.
3) Snapshots the base subvolume (snap1).
4) Touches files 5000-6000 in snap1.
5) Snapshots snap1 (snap2).
6) Deletes snap1.

I do this with dm-log-writes, and then replay to every FUA in the log
and fsck the fs.

Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
[ copy reproducer steps ]
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: check for refs on snapshot delete resume
Josef Bacik [Wed, 6 Feb 2019 20:46:14 +0000 (15:46 -0500)]
btrfs: check for refs on snapshot delete resume

There's a bug in snapshot deletion where we won't update the
drop_progress key if we're in the UPDATE_BACKREF stage.  This is a
problem because we could drop refs for blocks we know don't belong to
ours.  If we crash or umount at the right time we could experience
messages such as the following when snapshot deletion resumes

 BTRFS error (device dm-3): unable to find ref byte nr 66797568 parent 0 root 258  owner 1 offset 0
 ------------[ cut here ]------------
 WARNING: CPU: 3 PID: 16052 at fs/btrfs/extent-tree.c:7108 __btrfs_free_extent.isra.78+0x62c/0xb30 [btrfs]
 CPU: 3 PID: 16052 Comm: umount Tainted: G        W  OE     5.0.0-rc4+ #147
 Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./890FX Deluxe5, BIOS P1.40 05/03/2011
 RIP: 0010:__btrfs_free_extent.isra.78+0x62c/0xb30 [btrfs]
 RSP: 0018:ffffc90005cd7b18 EFLAGS: 00010286
 RAX: 0000000000000000 RBX: 0000000000000001 RCX: 0000000000000000
 RDX: ffff88842fade680 RSI: ffff88842fad6b18 RDI: ffff88842fad6b18
 RBP: ffffc90005cd7bc8 R08: 0000000000000000 R09: 0000000000000001
 R10: 0000000000000001 R11: ffffffff822696b8 R12: 0000000003fb4000
 R13: 0000000000000001 R14: 0000000000000102 R15: ffff88819c9d67e0
 FS:  00007f08bb138fc0(0000) GS:ffff88842fac0000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 00007f8f5d861ea0 CR3: 00000003e99fe000 CR4: 00000000000006e0
 Call Trace:
 ? _raw_spin_unlock+0x27/0x40
 ? btrfs_merge_delayed_refs+0x356/0x3e0 [btrfs]
 __btrfs_run_delayed_refs+0x75a/0x13c0 [btrfs]
 ? join_transaction+0x2b/0x460 [btrfs]
 btrfs_run_delayed_refs+0xf3/0x1c0 [btrfs]
 btrfs_commit_transaction+0x52/0xa50 [btrfs]
 ? start_transaction+0xa6/0x510 [btrfs]
 btrfs_sync_fs+0x79/0x1c0 [btrfs]
 sync_filesystem+0x70/0x90
 generic_shutdown_super+0x27/0x120
 kill_anon_super+0x12/0x30
 btrfs_kill_super+0x16/0xa0 [btrfs]
 deactivate_locked_super+0x43/0x70
 deactivate_super+0x40/0x60
 cleanup_mnt+0x3f/0x80
 __cleanup_mnt+0x12/0x20
 task_work_run+0x8b/0xc0
 exit_to_usermode_loop+0xce/0xd0
 do_syscall_64+0x20b/0x210
 entry_SYSCALL_64_after_hwframe+0x49/0xbe

To fix this simply mark dead roots we read from disk as DEAD and then
set the walk_control->restarted flag so we know we have a restarted
deletion.  From here whenever we try to drop refs for blocks we check to
verify our ref is set on them, and if it is not we skip it.  Once we
find a ref that is set we unset walk_control->restarted since the tree
should be in a normal state from then on, and any problems we run into
from there are different issues.  I tested this with an existing broken
fs and my reproducer that creates a broken fs and it fixed both file
systems.

Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agoBtrfs: fix deadlock between clone/dedupe and rename
Filipe Manana [Tue, 26 Feb 2019 12:06:09 +0000 (12:06 +0000)]
Btrfs: fix deadlock between clone/dedupe and rename

Reflinking (clone/dedupe) and rename are operations that operate on two
inodes and therefore need to lock them in the same order to avoid ABBA
deadlocks. It happens that Btrfs' reflink implementation always locked
them in a different order from VFS's lock_two_nondirectories() helper,
which is used by the rename code in VFS, resulting in ABBA type deadlocks.

Btrfs' locking order:

  static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
  {
         if (inode1 < inode2)
                swap(inode1, inode2);

         inode_lock_nested(inode1, I_MUTEX_PARENT);
         inode_lock_nested(inode2, I_MUTEX_CHILD);
  }

VFS's locking order:

  void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
  {
        if (inode1 > inode2)
                swap(inode1, inode2);

        if (inode1 && !S_ISDIR(inode1->i_mode))
                inode_lock(inode1);
        if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
                inode_lock_nested(inode2, I_MUTEX_NONDIR2);
}

Fix this by killing the btrfs helper function that does the double inode
locking and replace it with VFS's helper lock_two_nondirectories().

Reported-by: Zygo Blaxell <ce3g8jdj@umail.furryterror.org>
Fixes: 416161db9b63e3 ("btrfs: offline dedupe")
CC: stable@vger.kernel.org # 4.4+
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agoBtrfs: fix corruption reading shared and compressed extents after hole punching
Filipe Manana [Thu, 14 Feb 2019 15:17:20 +0000 (15:17 +0000)]
Btrfs: fix corruption reading shared and compressed extents after hole punching

In the past we had data corruption when reading compressed extents that
are shared within the same file and they are consecutive, this got fixed
by commit 005efedf2c7d0 ("Btrfs: fix read corruption of compressed and
shared extents") and by commit 808f80b46790f ("Btrfs: update fix for read
corruption of compressed and shared extents"). However there was a case
that was missing in those fixes, which is when the shared and compressed
extents are referenced with a non-zero offset. The following shell script
creates a reproducer for this issue:

  #!/bin/bash

  mkfs.btrfs -f /dev/sdc &> /dev/null
  mount -o compress /dev/sdc /mnt/sdc

  # Create a file with 3 consecutive compressed extents, each has an
  # uncompressed size of 128Kb and a compressed size of 4Kb.
  for ((i = 1; i <= 3; i++)); do
      head -c 4096 /dev/zero
      for ((j = 1; j <= 31; j++)); do
          head -c 4096 /dev/zero | tr '\0' "\377"
      done
  done > /mnt/sdc/foobar
  sync

  echo "Digest after file creation:   $(md5sum /mnt/sdc/foobar)"

  # Clone the first extent into offsets 128K and 256K.
  xfs_io -c "reflink /mnt/sdc/foobar 0 128K 128K" /mnt/sdc/foobar
  xfs_io -c "reflink /mnt/sdc/foobar 0 256K 128K" /mnt/sdc/foobar
  sync

  echo "Digest after cloning:         $(md5sum /mnt/sdc/foobar)"

  # Punch holes into the regions that are already full of zeroes.
  xfs_io -c "fpunch 0 4K" /mnt/sdc/foobar
  xfs_io -c "fpunch 128K 4K" /mnt/sdc/foobar
  xfs_io -c "fpunch 256K 4K" /mnt/sdc/foobar
  sync

  echo "Digest after hole punching:   $(md5sum /mnt/sdc/foobar)"

  echo "Dropping page cache..."
  sysctl -q vm.drop_caches=1
  echo "Digest after hole punching:   $(md5sum /mnt/sdc/foobar)"

  umount /dev/sdc

When running the script we get the following output:

  Digest after file creation:   5a0888d80d7ab1fd31c229f83a3bbcc8  /mnt/sdc/foobar
  linked 131072/131072 bytes at offset 131072
  128 KiB, 1 ops; 0.0033 sec (36.960 MiB/sec and 295.6830 ops/sec)
  linked 131072/131072 bytes at offset 262144
  128 KiB, 1 ops; 0.0015 sec (78.567 MiB/sec and 628.5355 ops/sec)
  Digest after cloning:         5a0888d80d7ab1fd31c229f83a3bbcc8  /mnt/sdc/foobar
  Digest after hole punching:   5a0888d80d7ab1fd31c229f83a3bbcc8  /mnt/sdc/foobar
  Dropping page cache...
  Digest after hole punching:   fba694ae8664ed0c2e9ff8937e7f1484  /mnt/sdc/foobar

This happens because after reading all the pages of the extent in the
range from 128K to 256K for example, we read the hole at offset 256K
and then when reading the page at offset 260K we don't submit the
existing bio, which is responsible for filling all the page in the
range 128K to 256K only, therefore adding the pages from range 260K
to 384K to the existing bio and submitting it after iterating over the
entire range. Once the bio completes, the uncompressed data fills only
the pages in the range 128K to 256K because there's no more data read
from disk, leaving the pages in the range 260K to 384K unfilled. It is
just a slightly different variant of what was solved by commit
005efedf2c7d0 ("Btrfs: fix read corruption of compressed and shared
extents").

Fix this by forcing a bio submit, during readpages(), whenever we find a
compressed extent map for a page that is different from the extent map
for the previous page or has a different starting offset (in case it's
the same compressed extent), instead of the extent map's original start
offset.

A test case for fstests follows soon.

Reported-by: Zygo Blaxell <ce3g8jdj@umail.furryterror.org>
Fixes: 808f80b46790f ("Btrfs: update fix for read corruption of compressed and shared extents")
Fixes: 005efedf2c7d0 ("Btrfs: fix read corruption of compressed and shared extents")
Cc: stable@vger.kernel.org # 4.3+
Tested-by: Zygo Blaxell <ce3g8jdj@umail.furryterror.org>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: Remove unnecessary casts in btrfs_read_root_item
YueHaibing [Wed, 20 Feb 2019 12:32:02 +0000 (12:32 +0000)]
btrfs: Remove unnecessary casts in btrfs_read_root_item

There is a messy cast here:
min_t(int, len, (int)sizeof(*item)));

min_t() should normally cast to unsigned.  It's not possible for "len"
to be negative, but if it were then we definitely wouldn't want to pass
negatives to read_extent_buffer().  Also there is an extra cast.

This patch shouldn't affect runtime, it's just a clean up.

Reviewed-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agoBtrfs: remove assertion when searching for a key in a node/leaf
Filipe Manana [Wed, 20 Feb 2019 11:11:43 +0000 (11:11 +0000)]
Btrfs: remove assertion when searching for a key in a node/leaf

At ctree.c:key_search(), the assertion that verifies the first key on a
child extent buffer corresponds to the key at a specific slot in the
parent has a disadvantage: we effectively hit a BUG_ON() which requires
rebooting the machine later. It also does not tell any information about
which extent buffer is affected, from which root, the expected and found
keys, etc.

However as of commit 581c1760415c48 ("btrfs: Validate child tree block's
level and first key"), that assertion is not needed since at the time we
read an extent buffer from disk we validate that its first key matches the
key, at the respective slot, in the parent extent buffer. Therefore just
remove the assertion at key_search().

Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agoBtrfs: add missing error handling after doing leaf/node binary search
Filipe Manana [Mon, 18 Feb 2019 16:57:26 +0000 (16:57 +0000)]
Btrfs: add missing error handling after doing leaf/node binary search

The function map_private_extent_buffer() can return an -EINVAL error, and
it is called by generic_bin_search() which will return back the error. The
btrfs_bin_search() function in turn calls generic_bin_search() and the
key_search() function calls btrfs_bin_search(), so both can return the
-EINVAL error coming from the map_private_extent_buffer() function. Some
callers of these functions were ignoring that these functions can return
an error, so fix them to deal with error return values.

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>
5 years agobtrfs: drop the lock on error in btrfs_dev_replace_cancel
Dan Carpenter [Mon, 11 Feb 2019 18:32:10 +0000 (21:32 +0300)]
btrfs: drop the lock on error in btrfs_dev_replace_cancel

We should drop the lock on this error path.  This has been found by a
static tool.

The lock needs to be released, it's there to protect access to the
dev_replace members and is not supposed to be left locked. The value of
state that's being switched would need to be artifically changed to an
invalid value so the default: branch is taken.

Fixes: d189dd70e255 ("btrfs: fix use-after-free due to race between replace start and cancel")
CC: stable@vger.kernel.org # 5.0+
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: ensure that a DUP or RAID1 block group has exactly two stripes
Johannes Thumshirn [Mon, 18 Feb 2019 10:28:37 +0000 (11:28 +0100)]
btrfs: ensure that a DUP or RAID1 block group has exactly two stripes

We recently had a customer issue with a corrupted filesystem. When
trying to mount this image btrfs panicked with a division by zero in
calc_stripe_length().

The corrupt chunk had a 'num_stripes' value of 1. calc_stripe_length()
takes this value and divides it by the number of copies the RAID profile
is expected to have to calculate the amount of data stripes. As a DUP
profile is expected to have 2 copies this division resulted in 1/2 = 0.
Later then the 'data_stripes' variable is used as a divisor in the
stripe length calculation which results in a division by 0 and thus a
kernel panic.

When encountering a filesystem with a DUP block group and a
'num_stripes' value unequal to 2, refuse mounting as the image is
corrupted and will lead to unexpected behaviour.

Code inspection showed a RAID1 block group has the same issues.

Fixes: e06cd3dd7cea ("Btrfs: add validadtion checks for chunk loading")
CC: stable@vger.kernel.org # 4.4+
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: init csum_list before possible free
Dan Robertson [Tue, 19 Feb 2019 02:56:43 +0000 (02:56 +0000)]
btrfs: init csum_list before possible free

The scrub_ctx csum_list member must be initialized before scrub_free_ctx
is called. If the csum_list is not initialized beforehand, the
list_empty call in scrub_free_csums will result in a null deref if the
allocation fails in the for loop.

Fixes: a2de733c78fa ("btrfs: scrub")
CC: stable@vger.kernel.org # 3.0+
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Dan Robertson <dan@dlrobertson.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agoBtrfs: remove no longer needed range length checks for deduplication
Filipe Manana [Wed, 12 Dec 2018 18:05:59 +0000 (18:05 +0000)]
Btrfs: remove no longer needed range length checks for deduplication

Comparing the content of the pages in the range to deduplicate is now
done in generic_remap_checks called by the generic helper
generic_remap_file_range_prep(), which takes care of ensuring we do not
compare/deduplicate undefined data beyond a file's EOF (range from EOF
to the next block boundary). So remove these checks which are now
redundant.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agoBtrfs: fix fsync after succession of renames and unlink/rmdir
Filipe Manana [Wed, 13 Feb 2019 12:14:09 +0000 (12:14 +0000)]
Btrfs: fix fsync after succession of renames and unlink/rmdir

After a succession of renames operations of different files and unlinking
one of them, if we fsync one of the renamed files we can end up with a
log that will either fail to replay at mount time or result in a filesystem
that is in an inconsistent state. One example scenario:

  $ mkfs.btrfs -f /dev/sdb
  $ mount /dev/sdb /mnt

  $ mkdir /mnt/testdir
  $ touch /mnt/testdir/fname1
  $ touch /mnt/testdir/fname2

  $ sync

  $ mv /mnt/testdir/fname1 /mnt/testdir/fname3
  $ rm -f /mnt/testdir/fname2
  $ ln /mnt/testdir/fname3 /mnt/testdir/fname2

  $ touch /mnt/testdir/fname1
  $ xfs_io -c "fsync" /mnt/testdir/fname1

  <power failure>

  $ mount /dev/sdb /mnt
  $ umount /mnt
  $ btrfs check /dev/sdb
  [1/7] checking root items
  [2/7] checking extents
  [3/7] checking free space cache
  [4/7] checking fs roots
  root 5 inode 259 errors 2, no orphan item
  ERROR: errors found in fs roots
  Opening filesystem to check...
  Checking filesystem on /dev/sdc
  UUID: 20e4abb8-5a19-4492-8bb4-6084125c2d0d
  found 393216 bytes used, error(s) found
  total csum bytes: 0
  total tree bytes: 131072
  total fs tree bytes: 32768
  total extent tree bytes: 16384
  btree space waste bytes: 122986
  file data blocks allocated: 262144
   referenced 262144

On a kernel without the first patch in this series, titled
"[PATCH] Btrfs: fix fsync after succession of renames of different files",
we get instead an error when mounting the filesystem due to failure of
replaying the log:

  $ mount /dev/sdb /mnt
  mount: mount /dev/sdb on /mnt failed: File exists

Fix this by logging the parent directory of an inode whenever we find an
inode that no longer exists (was unlinked in the current transaction),
during the procedure which finds inodes that have old names that collide
with new names of other inodes.

A test case for fstests follows soon.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agoBtrfs: fix fsync after succession of renames of different files
Filipe Manana [Wed, 13 Feb 2019 12:14:03 +0000 (12:14 +0000)]
Btrfs: fix fsync after succession of renames of different files

After a succession of rename operations of different files and fsyncing
one of them, such that each file gets a new name that corresponds to an
old name of another file, we can end up with a log that will cause a
failure when attempted to replay at mount time (an EEXIST error).
We currently have correct behaviour when such succession of renames
involves only two files, but if there are more files involved, we end up
not logging all the inodes that are needed, therefore resulting in a
failure when attempting to replay the log.

Example:

  $ mkfs.btrfs -f /dev/sdb
  $ mount /dev/sdb /mnt

  $ mkdir /mnt/testdir
  $ touch /mnt/testdir/fname1
  $ touch /mnt/testdir/fname2

  $ sync

  $ mv /mnt/testdir/fname1 /mnt/testdir/fname3
  $ mv /mnt/testdir/fname2 /mnt/testdir/fname4
  $ ln /mnt/testdir/fname3 /mnt/testdir/fname2

  $ touch /mnt/testdir/fname1
  $ xfs_io -c "fsync" /mnt/testdir/fname1

  <power failure>

  $ mount /dev/sdb /mnt
  mount: mount /dev/sdb on /mnt failed: File exists

So fix this by checking all inode dependencies when logging an inode. That
is, if one logged inode A has a new name that matches the old name of some
other inode B, check if inode B has a new name that matches the old name
of some other inode C, and so on. This fix is implemented not by doing any
recursive function calls but by using an iterative method using a linked
list that is used in a first-in-first-out fashion.

A test case for fstests follows soon.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: honor path->skip_locking in backref code
Josef Bacik [Wed, 16 Jan 2019 16:00:57 +0000 (11:00 -0500)]
btrfs: honor path->skip_locking in backref code

Qgroups will do the old roots lookup at delayed ref time, which could be
while walking down the extent root while running a delayed ref.  This
should be fine, except we specifically lock eb's in the backref walking
code irrespective of path->skip_locking, which deadlocks the system.
Fix up the backref code to honor path->skip_locking, nobody will be
modifying the commit_root when we're searching so it's completely safe
to do.

This happens since fb235dc06fac ("btrfs: qgroup: Move half of the qgroup
accounting time out of commit trans"), kernel may lockup with quota
enabled.

There is one backref trace triggered by snapshot dropping along with
write operation in the source subvolume.  The example can be reliably
reproduced:

  btrfs-cleaner   D    0  4062      2 0x80000000
  Call Trace:
   schedule+0x32/0x90
   btrfs_tree_read_lock+0x93/0x130 [btrfs]
   find_parent_nodes+0x29b/0x1170 [btrfs]
   btrfs_find_all_roots_safe+0xa8/0x120 [btrfs]
   btrfs_find_all_roots+0x57/0x70 [btrfs]
   btrfs_qgroup_trace_extent_post+0x37/0x70 [btrfs]
   btrfs_qgroup_trace_leaf_items+0x10b/0x140 [btrfs]
   btrfs_qgroup_trace_subtree+0xc8/0xe0 [btrfs]
   do_walk_down+0x541/0x5e3 [btrfs]
   walk_down_tree+0xab/0xe7 [btrfs]
   btrfs_drop_snapshot+0x356/0x71a [btrfs]
   btrfs_clean_one_deleted_snapshot+0xb8/0xf0 [btrfs]
   cleaner_kthread+0x12b/0x160 [btrfs]
   kthread+0x112/0x130
   ret_from_fork+0x27/0x50

When dropping snapshots with qgroup enabled, we will trigger backref
walk.

However such backref walk at that timing is pretty dangerous, as if one
of the parent nodes get WRITE locked by other thread, we could cause a
dead lock.

For example:

           FS 260     FS 261 (Dropped)
            node A        node B
           /      \      /      \
       node C      node D      node E
      /   \         /  \        /     \
  leaf F|leaf G|leaf H|leaf I|leaf J|leaf K

The lock sequence would be:

      Thread A (cleaner)             |       Thread B (other writer)
-----------------------------------------------------------------------
write_lock(B)                        |
write_lock(D)                        |
^^^ called by walk_down_tree()       |
                                     |       write_lock(A)
                                     |       write_lock(D) << Stall
read_lock(H) << for backref walk     |
read_lock(D) << lock owner is        |
                the same thread A    |
                so read lock is OK   |
read_lock(A) << Stall                |

So thread A hold write lock D, and needs read lock A to unlock.
While thread B holds write lock A, while needs lock D to unlock.

This will cause a deadlock.

This is not only limited to snapshot dropping case.  As the backref
walk, even only happens on commit trees, is breaking the normal top-down
locking order, makes it deadlock prone.

Fixes: fb235dc06fac ("btrfs: qgroup: Move half of the qgroup accounting time out of commit trans")
CC: stable@vger.kernel.org # 4.14+
Reported-and-tested-by: David Sterba <dsterba@suse.com>
Reported-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
[ rebase to latest branch and fix lock assert bug in btrfs/007 ]
Signed-off-by: Qu Wenruo <wqu@suse.com>
[ copy logs and deadlock analysis from Qu's patch ]
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: qgroup: Make qgroup async transaction commit more aggressive
Qu Wenruo [Thu, 24 Jan 2019 23:55:27 +0000 (07:55 +0800)]
btrfs: qgroup: Make qgroup async transaction commit more aggressive

[BUG]
Btrfs qgroup will still hit EDQUOT under the following case:

  $ dev=/dev/test/test
  $ mnt=/mnt/btrfs
  $ umount $mnt &> /dev/null
  $ umount $dev &> /dev/null

  $ mkfs.btrfs -f $dev
  $ mount $dev $mnt -o nospace_cache

  $ btrfs subv create $mnt/subv
  $ btrfs quota enable $mnt
  $ btrfs quota rescan -w $mnt
  $ btrfs qgroup limit -e 1G $mnt/subv

  $ fallocate -l 900M $mnt/subv/padding
  $ sync

  $ rm $mnt/subv/padding

  # Hit EDQUOT
  $ xfs_io -f -c "pwrite 0 512M" $mnt/subv/real_file

[CAUSE]
Since commit a514d63882c3 ("btrfs: qgroup: Commit transaction in advance
to reduce early EDQUOT"), btrfs is not forced to commit transaction to
reclaim more quota space.

Instead, we just check pertrans metadata reservation against some
threshold and try to do asynchronously transaction commit.

However in above case, the pertrans metadata reservation is pretty small
thus it will never trigger asynchronous transaction commit.

[FIX]
Instead of only accounting pertrans metadata reservation, we calculate
how much free space we have, and if there isn't much free space left,
commit transaction asynchronously to try to free some space.

This may slow down the fs when we have less than 32M free qgroup space,
but should reduce a lot of false EDQUOT, so the cost should be
acceptable.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: qgroup: Move reserved data accounting from btrfs_delayed_ref_head to btrfs_qgr...
Qu Wenruo [Wed, 23 Jan 2019 07:15:12 +0000 (15:15 +0800)]
btrfs: qgroup: Move reserved data accounting from btrfs_delayed_ref_head to btrfs_qgroup_extent_record

[BUG]
Btrfs/139 will fail with a high probability if the testing machine (VM)
has only 2G RAM.

Resulting the final write success while it should fail due to EDQUOT,
and the fs will have quota exceeding the limit by 16K.

The simplified reproducer will be: (needs a 2G ram VM)

  $ mkfs.btrfs -f $dev
  $ mount $dev $mnt

  $ btrfs subv create $mnt/subv
  $ btrfs quota enable $mnt
  $ btrfs quota rescan -w $mnt
  $ btrfs qgroup limit -e 1G $mnt/subv

  $ for i in $(seq -w  1 8); do
   xfs_io -f -c "pwrite 0 128M" $mnt/subv/file_$i > /dev/null
   echo "file $i written" > /dev/kmsg
    done
  $ sync
  $ btrfs qgroup show -pcre --raw $mnt

The last pwrite will not trigger EDQUOT and final 'qgroup show' will
show something like:

  qgroupid         rfer         excl     max_rfer     max_excl parent  child
  --------         ----         ----     --------     -------- ------  -----
  0/5             16384        16384         none         none ---     ---
  0/256      1073758208   1073758208         none   1073741824 ---     ---

And 1073758208 is larger than
  > 1073741824.

[CAUSE]
It's a bug in btrfs qgroup data reserved space management.

For quota limit, we must ensure that:
  reserved (data + metadata) + rfer/excl <= limit

Since rfer/excl is only updated at transaction commmit time, reserved
space needs to be taken special care.

One important part of reserved space is data, and for a new data extent
written to disk, we still need to take the reserved space until
rfer/excl numbers get updated.

Originally when an ordered extent finishes, we migrate the reserved
qgroup data space from extent_io tree to delayed ref head of the data
extent, expecting delayed ref will only be cleaned up at commit
transaction time.

However for small RAM machine, due to memory pressure dirty pages can be
flushed back to disk without committing a transaction.

The related events will be something like:

  file 1 written
  btrfs_finish_ordered_io: ino=258 ordered offset=0 len=54947840
  btrfs_finish_ordered_io: ino=258 ordered offset=54947840 len=5636096
  btrfs_finish_ordered_io: ino=258 ordered offset=61153280 len=57344
  btrfs_finish_ordered_io: ino=258 ordered offset=61210624 len=8192
  btrfs_finish_ordered_io: ino=258 ordered offset=60583936 len=569344
  cleanup_ref_head: num_bytes=54947840
  cleanup_ref_head: num_bytes=5636096
  cleanup_ref_head: num_bytes=569344
  cleanup_ref_head: num_bytes=57344
  cleanup_ref_head: num_bytes=8192
  ^^^^^^^^^^^^^^^^ This will free qgroup data reserved space
  file 2 written
  ...
  file 8 written
  cleanup_ref_head: num_bytes=8192
  ...
  btrfs_commit_transaction  <<< the only transaction committed during
the test

When file 2 is written, we have already freed 128M reserved qgroup data
space for ino 258. Thus later write won't trigger EDQUOT.

This allows us to write more data beyond qgroup limit.

In my 2G ram VM, it could reach about 1.2G before hitting EDQUOT.

[FIX]
By moving reserved qgroup data space from btrfs_delayed_ref_head to
btrfs_qgroup_extent_record, we can ensure that reserved qgroup data
space won't be freed half way before commit transaction, thus fix the
problem.

Fixes: f64d5ca86821 ("btrfs: delayed_ref: Add new function to record reserved space into delayed ref")
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: scrub: remove unused nocow worker pointer
David Sterba [Tue, 12 Feb 2019 17:20:04 +0000 (18:20 +0100)]
btrfs: scrub: remove unused nocow worker pointer

The member btrfs_fs_info::scrub_nocow_workers is unused since the nocow
optimization was removed from scrub in 9bebe665c3e4 ("btrfs: scrub:
Remove unused copy_nocow_pages and its callchain").

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: scrub: add assertions for worker pointers
David Sterba [Tue, 12 Feb 2019 15:51:18 +0000 (16:51 +0100)]
btrfs: scrub: add assertions for worker pointers

The scrub worker pointers are not NULL iff the scrub is running, so
reset them back once the last reference is dropped. Add assertions to
the initial phase of scrub to verify that.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: scrub: convert scrub_workers_refcnt to refcount_t
Anand Jain [Wed, 30 Jan 2019 06:45:02 +0000 (14:45 +0800)]
btrfs: scrub: convert scrub_workers_refcnt to refcount_t

Use the refcount_t for fs_info::scrub_workers_refcnt instead of int so
we get the extra checks. All reference changes are still done under
scrub_lock.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: scrub: add scrub_lock lockdep check in scrub_workers_get
Anand Jain [Wed, 30 Jan 2019 06:45:01 +0000 (14:45 +0800)]
btrfs: scrub: add scrub_lock lockdep check in scrub_workers_get

scrub_workers_refcnt is protected by scrub_lock, add lockdep_assert_held()
in scrub_workers_get().

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Suggested-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: scrub: fix circular locking dependency warning
Anand Jain [Wed, 30 Jan 2019 06:45:00 +0000 (14:45 +0800)]
btrfs: scrub: fix circular locking dependency warning

This fixes a longstanding lockdep warning triggered by
fstests/btrfs/011.

Circular locking dependency check reports warning[1], that's because the
btrfs_scrub_dev() calls the stack #0 below with, the fs_info::scrub_lock
held. The test case leading to this warning:

  $ mkfs.btrfs -f /dev/sdb
  $ mount /dev/sdb /btrfs
  $ btrfs scrub start -B /btrfs

In fact we have fs_info::scrub_workers_refcnt to track if the init and destroy
of the scrub workers are needed. So once we have incremented and decremented
the fs_info::scrub_workers_refcnt value in the thread, its ok to drop the
scrub_lock, and then actually do the btrfs_destroy_workqueue() part. So this
patch drops the scrub_lock before calling btrfs_destroy_workqueue().

  [359.258534] ======================================================
  [359.260305] WARNING: possible circular locking dependency detected
  [359.261938] 5.0.0-rc6-default #461 Not tainted
  [359.263135] ------------------------------------------------------
  [359.264672] btrfs/20975 is trying to acquire lock:
  [359.265927] 00000000d4d32bea ((wq_completion)"%s-%s""btrfs", name){+.+.}, at: flush_workqueue+0x87/0x540
  [359.268416]
  [359.268416] but task is already holding lock:
  [359.270061] 0000000053ea26a6 (&fs_info->scrub_lock){+.+.}, at: btrfs_scrub_dev+0x322/0x590 [btrfs]
  [359.272418]
  [359.272418] which lock already depends on the new lock.
  [359.272418]
  [359.274692]
  [359.274692] the existing dependency chain (in reverse order) is:
  [359.276671]
  [359.276671] -> #3 (&fs_info->scrub_lock){+.+.}:
  [359.278187]        __mutex_lock+0x86/0x9c0
  [359.279086]        btrfs_scrub_pause+0x31/0x100 [btrfs]
  [359.280421]        btrfs_commit_transaction+0x1e4/0x9e0 [btrfs]
  [359.281931]        close_ctree+0x30b/0x350 [btrfs]
  [359.283208]        generic_shutdown_super+0x64/0x100
  [359.284516]        kill_anon_super+0x14/0x30
  [359.285658]        btrfs_kill_super+0x12/0xa0 [btrfs]
  [359.286964]        deactivate_locked_super+0x29/0x60
  [359.288242]        cleanup_mnt+0x3b/0x70
  [359.289310]        task_work_run+0x98/0xc0
  [359.290428]        exit_to_usermode_loop+0x83/0x90
  [359.291445]        do_syscall_64+0x15b/0x180
  [359.292598]        entry_SYSCALL_64_after_hwframe+0x49/0xbe
  [359.294011]
  [359.294011] -> #2 (sb_internal#2){.+.+}:
  [359.295432]        __sb_start_write+0x113/0x1d0
  [359.296394]        start_transaction+0x369/0x500 [btrfs]
  [359.297471]        btrfs_finish_ordered_io+0x2aa/0x7c0 [btrfs]
  [359.298629]        normal_work_helper+0xcd/0x530 [btrfs]
  [359.299698]        process_one_work+0x246/0x610
  [359.300898]        worker_thread+0x3c/0x390
  [359.302020]        kthread+0x116/0x130
  [359.303053]        ret_from_fork+0x24/0x30
  [359.304152]
  [359.304152] -> #1 ((work_completion)(&work->normal_work)){+.+.}:
  [359.306100]        process_one_work+0x21f/0x610
  [359.307302]        worker_thread+0x3c/0x390
  [359.308465]        kthread+0x116/0x130
  [359.309357]        ret_from_fork+0x24/0x30
  [359.310229]
  [359.310229] -> #0 ((wq_completion)"%s-%s""btrfs", name){+.+.}:
  [359.311812]        lock_acquire+0x90/0x180
  [359.312929]        flush_workqueue+0xaa/0x540
  [359.313845]        drain_workqueue+0xa1/0x180
  [359.314761]        destroy_workqueue+0x17/0x240
  [359.315754]        btrfs_destroy_workqueue+0x57/0x200 [btrfs]
  [359.317245]        scrub_workers_put+0x2c/0x60 [btrfs]
  [359.318585]        btrfs_scrub_dev+0x336/0x590 [btrfs]
  [359.319944]        btrfs_dev_replace_by_ioctl.cold.19+0x179/0x1bb [btrfs]
  [359.321622]        btrfs_ioctl+0x28a4/0x2e40 [btrfs]
  [359.322908]        do_vfs_ioctl+0xa2/0x6d0
  [359.324021]        ksys_ioctl+0x3a/0x70
  [359.325066]        __x64_sys_ioctl+0x16/0x20
  [359.326236]        do_syscall_64+0x54/0x180
  [359.327379]        entry_SYSCALL_64_after_hwframe+0x49/0xbe
  [359.328772]
  [359.328772] other info that might help us debug this:
  [359.328772]
  [359.330990] Chain exists of:
  [359.330990]   (wq_completion)"%s-%s""btrfs", name --> sb_internal#2 --> &fs_info->scrub_lock
  [359.330990]
  [359.334376]  Possible unsafe locking scenario:
  [359.334376]
  [359.336020]        CPU0                    CPU1
  [359.337070]        ----                    ----
  [359.337821]   lock(&fs_info->scrub_lock);
  [359.338506]                                lock(sb_internal#2);
  [359.339506]                                lock(&fs_info->scrub_lock);
  [359.341461]   lock((wq_completion)"%s-%s""btrfs", name);
  [359.342437]
  [359.342437]  *** DEADLOCK ***
  [359.342437]
  [359.343745] 1 lock held by btrfs/20975:
  [359.344788]  #0: 0000000053ea26a6 (&fs_info->scrub_lock){+.+.}, at: btrfs_scrub_dev+0x322/0x590 [btrfs]
  [359.346778]
  [359.346778] stack backtrace:
  [359.347897] CPU: 0 PID: 20975 Comm: btrfs Not tainted 5.0.0-rc6-default #461
  [359.348983] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.11.2-0-gf9626cc-prebuilt.qemu-project.org 04/01/2014
  [359.350501] Call Trace:
  [359.350931]  dump_stack+0x67/0x90
  [359.351676]  print_circular_bug.isra.37.cold.56+0x15c/0x195
  [359.353569]  check_prev_add.constprop.44+0x4f9/0x750
  [359.354849]  ? check_prev_add.constprop.44+0x286/0x750
  [359.356505]  __lock_acquire+0xb84/0xf10
  [359.357505]  lock_acquire+0x90/0x180
  [359.358271]  ? flush_workqueue+0x87/0x540
  [359.359098]  flush_workqueue+0xaa/0x540
  [359.359912]  ? flush_workqueue+0x87/0x540
  [359.360740]  ? drain_workqueue+0x1e/0x180
  [359.361565]  ? drain_workqueue+0xa1/0x180
  [359.362391]  drain_workqueue+0xa1/0x180
  [359.363193]  destroy_workqueue+0x17/0x240
  [359.364539]  btrfs_destroy_workqueue+0x57/0x200 [btrfs]
  [359.365673]  scrub_workers_put+0x2c/0x60 [btrfs]
  [359.366618]  btrfs_scrub_dev+0x336/0x590 [btrfs]
  [359.367594]  ? start_transaction+0xa1/0x500 [btrfs]
  [359.368679]  btrfs_dev_replace_by_ioctl.cold.19+0x179/0x1bb [btrfs]
  [359.369545]  btrfs_ioctl+0x28a4/0x2e40 [btrfs]
  [359.370186]  ? __lock_acquire+0x263/0xf10
  [359.370777]  ? kvm_clock_read+0x14/0x30
  [359.371392]  ? kvm_sched_clock_read+0x5/0x10
  [359.372248]  ? sched_clock+0x5/0x10
  [359.372786]  ? sched_clock_cpu+0xc/0xc0
  [359.373662]  ? do_vfs_ioctl+0xa2/0x6d0
  [359.374552]  do_vfs_ioctl+0xa2/0x6d0
  [359.375378]  ? do_sigaction+0xff/0x250
  [359.376233]  ksys_ioctl+0x3a/0x70
  [359.376954]  __x64_sys_ioctl+0x16/0x20
  [359.377772]  do_syscall_64+0x54/0x180
  [359.378841]  entry_SYSCALL_64_after_hwframe+0x49/0xbe
  [359.380422] RIP: 0033:0x7f5429296a97

Backporting to older kernels: scrub_nocow_workers must be freed the same
way as the others.

CC: stable@vger.kernel.org # 4.4+
Signed-off-by: Anand Jain <anand.jain@oracle.com>
[ update changelog ]
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: fix comment its device list mutex not volume lock
Anand Jain [Fri, 8 Feb 2019 07:39:37 +0000 (15:39 +0800)]
btrfs: fix comment its device list mutex not volume lock

We have killed volume mutex (commit: dccdb07bc996
btrfs: kill btrfs_fs_info::volume_mutex). This a trival one seems to have
escaped.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: extent_io: Kill the forward declaration of flush_write_bio
Qu Wenruo [Fri, 25 Jan 2019 05:09:15 +0000 (13:09 +0800)]
btrfs: extent_io: Kill the forward declaration of flush_write_bio

There is no need to forward declare flush_write_bio(), as it only
depends on submit_one_bio().  Both of them are pretty small, just move
them to kill the forward declaration.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: Fix grossly misleading argument names in extent io search
Nikolay Borisov [Wed, 30 Jan 2019 14:51:00 +0000 (16:51 +0200)]
btrfs: Fix grossly misleading argument names in extent io search

The variables and function parameters of __etree_search which pertain to
prev/next are grossly misnamed. Namely, prev_ret holds the next state
and not the previous. Similarly, next_ret actually holds the previous
extent state relating to the offset we are interested in. Fix this by
renaming the variables as well as switching the arguments order. No
functional changes.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: Remove EXTENT_FIRST_DELALLOC bit
Nikolay Borisov [Wed, 30 Jan 2019 14:50:50 +0000 (16:50 +0200)]
btrfs: Remove EXTENT_FIRST_DELALLOC bit

With the refactoring introduced in 8b62f87bad9c ("Btrfs: reworki
outstanding_extents") this flag became unused. Remove it and renumber
the following flags accordingly. No functional changes.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: use WARN_ON in a canonical form btrfs_remove_block_group
Nikolay Borisov [Wed, 30 Jan 2019 14:50:49 +0000 (16:50 +0200)]
btrfs: use WARN_ON in a canonical form btrfs_remove_block_group

There is no point in using a construct like 'if (!condition)
WARN_ON(1)'. Use WARN_ON(!condition) directly. No functional changes.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: reserve extra space during evict
Josef Bacik [Wed, 21 Nov 2018 19:03:13 +0000 (14:03 -0500)]
btrfs: reserve extra space during evict

We could generate a lot of delayed refs in evict but never have any left
over space from our block rsv to make up for that fact.  So reserve some
extra space and give it to the transaction so it can be used to refill
the delayed refs rsv every loop through the truncate path.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: be more explicit about allowed flush states
Josef Bacik [Wed, 21 Nov 2018 19:03:12 +0000 (14:03 -0500)]
btrfs: be more explicit about allowed flush states

For FLUSH_LIMIT flushers we really can only allocate chunks and flush
delayed inode items, everything else is problematic.  I added a bunch of
new states and it lead to weirdness in the FLUSH_LIMIT case because I
forgot about how it worked.  So instead explicitly declare the states
that are ok for flushing with FLUSH_LIMIT and use that for our state
machine.  Then as we add new things that are safe we can just add them
to this list.

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>
5 years agobtrfs: loop in inode_rsv_refill
Josef Bacik [Wed, 21 Nov 2018 19:03:11 +0000 (14:03 -0500)]
btrfs: loop in inode_rsv_refill

With severe fragmentation we can end up with our inode rsv size being
huge during writeout, which would cause us to need to make very large
metadata reservations.

However we may not actually need that much once writeout is complete,
because of the over-reservation for the worst case.

So instead try to make our reservation, and if we couldn't make it
re-calculate our new reservation size and try again.  If our reservation
size doesn't change between tries then we know we are actually out of
space and can error. Flushing that could have been running in parallel
did not make any space.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
[ rename to calc_refill_bytes, update comment and changelog ]
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: don't enospc all tickets on flush failure
Josef Bacik [Wed, 21 Nov 2018 19:03:10 +0000 (14:03 -0500)]
btrfs: don't enospc all tickets on flush failure

With the introduction of the per-inode block_rsv it became possible to
have really really large reservation requests made because of data
fragmentation.  Since the ticket stuff assumed that we'd always have
relatively small reservation requests it just killed all tickets if we
were unable to satisfy the current request.

However, this is generally not the case anymore.  So fix this logic to
instead see if we had a ticket that we were able to give some
reservation to, and if we were continue the flushing loop again.

Likewise we make the tickets use the space_info_add_old_bytes() method
of returning what reservation they did receive in hopes that it could
satisfy reservations down the line.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: don't use global reserve for chunk allocation
Josef Bacik [Wed, 21 Nov 2018 19:03:08 +0000 (14:03 -0500)]
btrfs: don't use global reserve for chunk allocation

We've done this forever because of the voodoo around knowing how much
space we have.  However, we have better ways of doing this now, and on
normal file systems we'll easily have a global reserve of 512MiB, and
since metadata chunks are usually 1GiB that means we'll allocate
metadata chunks more readily.  Instead use the actual used amount when
determining if we need to allocate a chunk or not.

This has a side effect for mixed block group fs'es where we are no
longer allocating enough chunks for the data/metadata requirements.  To
deal with this add a ALLOC_CHUNK_FORCE step to the flushing state
machine.  This will only get used if we've already made a full loop
through the flushing machinery and tried committing the transaction.

If we have then we can try and force a chunk allocation since we likely
need it to make progress.  This resolves issues I was seeing with
the mixed bg tests in xfstests without the new flushing state.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
[ merged with patch "add ALLOC_CHUNK_FORCE to the flushing code" ]
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: dump block_rsv details when dumping space info
Josef Bacik [Wed, 21 Nov 2018 19:03:07 +0000 (14:03 -0500)]
btrfs: dump block_rsv details when dumping space info

For enospc_debug having the block rsvs is super helpful to see if we've
done something wrong.

Reviewed-by: Omar Sandoval <osandov@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>
5 years agobtrfs: check if there are free block groups for commit
Josef Bacik [Wed, 21 Nov 2018 19:03:06 +0000 (14:03 -0500)]
btrfs: check if there are free block groups for commit

may_commit_transaction will skip committing the transaction if we don't
have enough pinned space or if we're trying to find space for a SYSTEM
chunk.  However, if we have pending free block groups in this transaction
we still want to commit as we may be able to allocate a chunk to make
our reservation.  So instead of just returning ENOSPC, check if we have
free block groups pending, and if so commit the transaction to allow us
to use that free space.

Reviewed-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: add zstd compression level support
Dennis Zhou [Mon, 4 Feb 2019 20:20:08 +0000 (15:20 -0500)]
btrfs: add zstd compression level support

Zstd compression requires different amounts of memory for each level of
compression. The prior patches implemented indirection to allow for each
compression type to manage their workspaces independently. This patch
uses this indirection to implement compression level support for zstd.

To manage the additional memory require, each compression level has its
own queue of workspaces. A global LRU is used to help with reclaim.
Reclaim is done via a timer which provides a mechanism to decrease
memory utilization by keeping only workspaces around that are sized
appropriately. Forward progress is guaranteed by a preallocated max
workspace hidden from the LRU.

When getting a workspace, it uses a bitmap to identify the levels that
are populated and scans up. If it finds a workspace that is greater than
it, it uses it, but does not update the last_used time and the
corresponding place in the LRU. If we hit memory pressure, we sleep on
the max level workspace. We continue to rescan in case we can use a
smaller workspace, but eventually should be able to obtain the max level
workspace or allocate one again should memory pressure subside.

The memory requirement for decompression is the same as level 1, and
therefore can use any of available workspace.

The number of workspaces is bound by an upper limit of the workqueue's
limit which currently is 2 (percpu limit). The reclaim timer is used to
free inactive/improperly sized workspaces and is set to 307s to avoid
colliding with transaction commit (every 30s).

Repeating the experiment from v2 [1], the Silesia corpus was copied to a
btrfs filesystem 10 times and then read back after dropping the caches.
The btrfs filesystem was on an SSD.

Level   Ratio   Compression (MB/s)  Decompression (MB/s)  Memory (KB)
1       2.658        438.47                910.51            780
2       2.744        364.86                886.55           1004
3       2.801        336.33                828.41           1260
4       2.858        286.71                886.55           1260
5       2.916        212.77                556.84           1388
6       2.363        119.82                990.85           1516
7       3.000        154.06                849.30           1516
8       3.011        159.54                875.03           1772
9       3.025        100.51                940.15           1772
10      3.033        118.97                616.26           1772
11      3.036         94.19                802.11           1772
12      3.037         73.45                931.49           1772
13      3.041         55.17                835.26           2284
14      3.087         44.70                716.78           2547
15      3.126         37.30                878.84           2547

[1] https://lore.kernel.org/linux-btrfs/20181031181108.289340-1-terrelln@fb.com/

Cc: Nick Terrell <terrelln@fb.com>
Cc: Omar Sandoval <osandov@osandov.com>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: make zstd memory requirements monotonic
Dennis Zhou [Mon, 4 Feb 2019 20:20:07 +0000 (15:20 -0500)]
btrfs: make zstd memory requirements monotonic

It is possible based on the level configurations that a higher level
workspace uses less memory than a lower level workspace. In order to
reuse workspaces, this must be made a monotonic relationship. This
precomputes the required memory for each level and enforces the
monotonicity between level and memory required. This is also done
in upstream zstd in [1].

[1] https://github.com/facebook/zstd/commit/a68b76afefec6876f8e8a538155109a5aeac0143

Cc: Nick Terrell <terrelln@fb.com>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: zstd use the passed through level instead of default
Dennis Zhou [Mon, 4 Feb 2019 20:20:06 +0000 (15:20 -0500)]
btrfs: zstd use the passed through level instead of default

Zstd currently only supports the default level of compression. This
patch switches to using the level passed in for btrfs zstd
configuration.

Zstd workspaces now keep track of the requested level as this can differ
from the size of the workspace.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: change set_level() to bound the level passed in
Dennis Zhou [Mon, 4 Feb 2019 20:20:05 +0000 (15:20 -0500)]
btrfs: change set_level() to bound the level passed in

Currently, the only user of set_level() is zlib which sets an internal
workspace parameter. As level is now plumbed into get_workspace(), this
can be handled there rather than separately.

This repurposes set_level() to bound the level passed in so it can be
used when setting the mounts compression level and as well as verifying
the level before getting a workspace. The other benefit is this divides
the meaning of compress(0) and get_workspace(0). The former means we
want to use the default compression level of the compression type. The
latter means we can use any workspace available.

Signed-off-by: Dennis Zhou <dennis@kernel.org>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: plumb level through the compression interface
Dennis Zhou [Mon, 4 Feb 2019 20:20:04 +0000 (15:20 -0500)]
btrfs: plumb level through the compression interface

Zlib compression supports multiple levels, but doesn't require changing
in how a workspace itself is created and managed. Zstd introduces a
different memory requirement such that higher levels of compression
require more memory.

This requires changes in how the alloc()/get() methods work for zstd.
This pach plumbs compression level through the interface as a parameter
in preparation for zstd compression levels.  This gives the compression
types opportunity to create/manage based on the compression level.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: move to function pointers for get/put workspaces
Dennis Zhou [Mon, 4 Feb 2019 20:20:03 +0000 (15:20 -0500)]
btrfs: move to function pointers for get/put workspaces

The previous patch added generic helpers for get_workspace() and
put_workspace(). Now, we can migrate ownership of the workspace_manager
to be in the compression type code as the compression code itself
doesn't care beyond being able to get a workspace. The init/cleanup and
get/put methods are abstracted so each compression algorithm can decide
how they want to manage their workspaces.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: add compression interface in (get/put)_workspace
Dennis Zhou [Mon, 4 Feb 2019 20:20:02 +0000 (15:20 -0500)]
btrfs: add compression interface in (get/put)_workspace

There are two levels of workspace management. First, alloc()/free()
which are responsible for actually creating and destroy workspaces.
Second, at a higher level, get()/put() which is the compression code
asking for a workspace from a workspace_manager.

The compression code shouldn't really care how it gets a workspace, but
that it got a workspace. This adds get_workspace() and put_workspace()
to be the higher level interface which is responsible for indexing into
the appropriate compression type. It also introduces
btrfs_put_workspace() and btrfs_get_workspace() to be the generic
implementations of the higher interface.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: add helper methods for workspace manager init and cleanup
Dennis Zhou [Mon, 4 Feb 2019 20:20:01 +0000 (15:20 -0500)]
btrfs: add helper methods for workspace manager init and cleanup

Workspace manager init and cleanup code is open coded inside a for loop
over the compression types. This forces each compression type to rely on
the same workspace manager implementation. This patch creates helper
methods that will be the generic implementation for btrfs workspace
management.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: unify compression ops with workspace_manager
Dennis Zhou [Mon, 4 Feb 2019 20:20:00 +0000 (15:20 -0500)]
btrfs: unify compression ops with workspace_manager

Make the workspace_manager own the interface operations rather than
managing index-paired arrays for the workspace_manager and compression
operations.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: manage heuristic workspace as index 0
Dennis Zhou [Mon, 4 Feb 2019 20:19:59 +0000 (15:19 -0500)]
btrfs: manage heuristic workspace as index 0

While the heuristic workspaces aren't really compression workspaces,
they use the same interface for managing them. So rather than branching,
let's just handle them once again as the index 0 compression type.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: rename workspaces_list to workspace_manager
Dennis Zhou [Mon, 4 Feb 2019 20:19:58 +0000 (15:19 -0500)]
btrfs: rename workspaces_list to workspace_manager

This is in preparation for zstd compression levels. As each level will
require different size of workspace, workspaces_list is no longer a
really fitting name.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: add helpers for compression type and level
Dennis Zhou [Mon, 4 Feb 2019 20:19:57 +0000 (15:19 -0500)]
btrfs: add helpers for compression type and level

It is very easy to miss places that rely on a certain bitshifting for
decoding the type_level overloading. Add helpers to do this instead.

Cc: Omar Sandoval <osandov@osandov.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: introduce new ioctl to unregister a btrfs device
Anand Jain [Fri, 4 Jan 2019 05:31:54 +0000 (13:31 +0800)]
btrfs: introduce new ioctl to unregister a btrfs device

Support for a new command that can be used eg. as a command

  $ btrfs device scan --forget [dev]'
(the final name may change though)

to undo the effects of 'btrfs device scan [dev]'. For this purpose
this patch proposes to use ioctl #5 as it was empty and is next to the
SCAN ioctl.

The new ioctl BTRFS_IOC_FORGET_DEV works only on the control device
(/dev/btrfs-control) to unregister one or all devices, devices that are
not mounted.

The argument is struct btrfs_ioctl_vol_args, ::name specifies the device
path. To unregister all device, the path is an empty string.

Again, the devices are removed only if they aren't part of a mounte
filesystem.

This new ioctl provides:

- release of unwanted btrfs_fs_devices and btrfs_devices structures
  from memory if the device is not going to be mounted

- ability to mount filesystem in degraded mode, when one devices is
  corrupted like in split brain raid1

- running test cases which would require reloading the kernel module
  but this is not possible eg. due to mounted filesystem or built-in

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ update changelog ]
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: replace cleaner_delayed_iput_mutex with a waitqueue
Josef Bacik [Mon, 3 Dec 2018 16:06:52 +0000 (11:06 -0500)]
btrfs: replace cleaner_delayed_iput_mutex with a waitqueue

The throttle path doesn't take cleaner_delayed_iput_mutex, which means
we could think we're done flushing iputs in the data space reservation
path when we could have a throttler doing an iput.  There's no real
reason to serialize the delayed iput flushing, so instead of taking the
cleaner_delayed_iput_mutex whenever we flush the delayed iputs just
replace it with an atomic counter and a waitqueue.  This removes the
short (or long depending on how big the inode is) window where we think
there are no more pending iputs when there really are some.

The waiting is killable as it could be indirectly called from user
operations like fallocate or zero-range. Such call sites should handle
the error but otherwise it's not necessary. Eg. flush_space just needs
to attempt to make space by waiting on iputs.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
[ add killable comment and changelog parts ]
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: Output ENOSPC debug info in inc_block_group_ro
Qu Wenruo [Wed, 30 Jan 2019 05:07:51 +0000 (13:07 +0800)]
btrfs: Output ENOSPC debug info in inc_block_group_ro

Since inc_block_group_ro() would return -ENOSPC, outputting debug info
for enospc_debug mount option would be helpful to debug some balance
false ENOSPC report.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: qgroup: Remove duplicated trace points for qgroup_rsv_add/release
Qu Wenruo [Tue, 13 Nov 2018 07:05:08 +0000 (15:05 +0800)]
btrfs: qgroup: Remove duplicated trace points for qgroup_rsv_add/release

Inside qgroup_rsv_add/release(), we have trace events
trace_qgroup_update_reserve() to catch reserved space update.

However we still have two manual trace_qgroup_update_reserve() calls
just outside these functions.  Remove these duplicated calls.

Fixes: 64ee4e751a1c ("btrfs: qgroup: Update trace events to use new separate rsv types")
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: let the assertion expression compile in all configs
Anders Roxell [Tue, 29 Jan 2019 13:01:46 +0000 (14:01 +0100)]
btrfs: let the assertion expression compile in all configs

A compiler warning (in a patch in development) pointed to a variable
that was used only inside and ASSERT:

  u64 root_objectid = root->root_key.objectid;
  ASSERT(root_objectid == ...);

  fs/btrfs/relocation.c: In function â€˜insert_dirty_subv’:
  fs/btrfs/relocation.c:2138:6: warning: unused variable â€˜root_objectid’ [-Wunused-variable]
    u64 root_objectid = root->root_key.objectid;
^~~~~~~~~~~~~

When CONFIG_BRTFS_ASSERT isn't enabled, variable root_objectid isn't used.

Rework the assertion helper by adding a runtime check instead of the
'#ifdef CONFIG_BTRFS_ASSERT #else ...", so the compiler sees the
condition being passed into an inline function after preprocessing.

Signed-off-by: Anders Roxell <anders.roxell@linaro.org>
Reviewed-by: David Sterba <dsterba@suse.com>
[ update changelog ]
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: merge btrfs_set_lock_blocking_rw with it's caller
David Sterba [Wed, 23 Jan 2019 17:07:14 +0000 (18:07 +0100)]
btrfs: merge btrfs_set_lock_blocking_rw with it's caller

The last caller that does not have a fixed value of lock is
btrfs_set_path_blocking, that actually does the same conditional swtich
by the lock type so we can merge the branches together and remove the
helper.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: simplify waiting loop in btrfs_tree_lock
David Sterba [Wed, 4 Apr 2018 00:11:50 +0000 (02:11 +0200)]
btrfs: simplify waiting loop in btrfs_tree_lock

Currently, the number of readers and writers is checked and in case
there are any, wait and redo the locks. There's some duplication
before the branches go back to again label, eg. calling wait_event on
blocking_readers twice.

The sequence is transformed

loop:
* wait for readers
* wait for writers
* write_lock
* check readers, unlock and wait for readers, loop
* check writers, unlock and wait for writers, loop

The new sequence is not exactly the same due to the simplification, for
readers it's slightly faster. For the writers, original code does

* wait for writers
* (loop) wait for readers
*        wait for writers -- again

while the new goes directly to the reader check. This should behave the
same on a contended lock with multiple writers and readers, but can
reduce number of times we're waiting on something.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: open code now trivial btrfs_set_lock_blocking
David Sterba [Wed, 4 Apr 2018 00:03:48 +0000 (02:03 +0200)]
btrfs: open code now trivial btrfs_set_lock_blocking

btrfs_set_lock_blocking is now only a simple wrapper around
btrfs_set_lock_blocking_write. The name does not bring any semantic
value that could not be inferred from the new function so there's no
point keeping it.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: replace btrfs_set_lock_blocking_rw with appropriate helpers
David Sterba [Wed, 4 Apr 2018 00:00:17 +0000 (02:00 +0200)]
btrfs: replace btrfs_set_lock_blocking_rw with appropriate helpers

We can use the right helper where the lock type is a fixed parameter.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: split btrfs_clear_lock_blocking_rw to read and write helpers
David Sterba [Tue, 3 Apr 2018 23:52:31 +0000 (01:52 +0200)]
btrfs: split btrfs_clear_lock_blocking_rw to read and write helpers

There are many callers that hardcode the desired lock type so we can
avoid the switch and call them directly. Split the current function to
two. There are no remaining users of btrfs_clear_lock_blocking_rw so
it's removed.  The call sites will be converted in followup patches.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: split btrfs_set_lock_blocking_rw to read and write helpers
David Sterba [Tue, 3 Apr 2018 23:43:05 +0000 (01:43 +0200)]
btrfs: split btrfs_set_lock_blocking_rw to read and write helpers

There are many callers that hardcode the desired lock type so we can
avoid the switch and call them directly. Split the current function to
two but leave a helper that still takes the variable lock type to make
current code compile.  The call sites will be converted in followup
patches.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: qgroup: Cleanup old subtree swap code
Qu Wenruo [Wed, 23 Jan 2019 07:15:18 +0000 (15:15 +0800)]
btrfs: qgroup: Cleanup old subtree swap code

Since it's replaced by new delayed subtree swap code, remove the
original code.

The cleanup is small since most of its core function is still used by
delayed subtree swap trace.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: qgroup: Use delayed subtree rescan for balance
Qu Wenruo [Wed, 23 Jan 2019 07:15:17 +0000 (15:15 +0800)]
btrfs: qgroup: Use delayed subtree rescan for balance

Before this patch, qgroup code traces the whole subtree of subvolume and
reloc trees unconditionally.

This makes qgroup numbers consistent, but it could cause tons of
unnecessary extent tracing, which causes a lot of overhead.

However for subtree swap of balance, just swap both subtrees because
they contain the same contents and tree structure, so qgroup numbers
won't change.

It's the race window between subtree swap and transaction commit could
cause qgroup number change.

This patch will delay the qgroup subtree scan until COW happens for the
subtree root.

So if there is no other operations for the fs, balance won't cause extra
qgroup overhead. (best case scenario)
Depending on the workload, most of the subtree scan can still be
avoided.

Only for worst case scenario, it will fall back to old subtree swap
overhead. (scan all swapped subtrees)

[[Benchmark]]
Hardware:
VM 4G vRAM, 8 vCPUs,
disk is using 'unsafe' cache mode,
backing device is SAMSUNG 850 evo SSD.
Host has 16G ram.

Mkfs parameter:
--nodesize 4K (To bump up tree size)

Initial subvolume contents:
4G data copied from /usr and /lib.
(With enough regular small files)

Snapshots:
16 snapshots of the original subvolume.
each snapshot has 3 random files modified.

balance parameter:
-m

So the content should be pretty similar to a real world root fs layout.

And after file system population, there is no other activity, so it
should be the best case scenario.

                     | v4.20-rc1            | w/ patchset    | diff
-----------------------------------------------------------------------
relocated extents    | 22615                | 22457          | -0.1%
qgroup dirty extents | 163457               | 121606         | -25.6%
time (sys)           | 22.884s              | 18.842s        | -17.6%
time (real)          | 27.724s              | 22.884s        | -17.5%

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: qgroup: Introduce per-root swapped blocks infrastructure
Qu Wenruo [Wed, 23 Jan 2019 07:15:16 +0000 (15:15 +0800)]
btrfs: qgroup: Introduce per-root swapped blocks infrastructure

To allow delayed subtree swap rescan, btrfs needs to record per-root
information about which tree blocks get swapped.  This patch introduces
the required infrastructure.

The designed workflow will be:

1) Record the subtree root block that gets swapped.

   During subtree swap:
   O = Old tree blocks
   N = New tree blocks
         reloc tree                         subvolume tree X
            Root                               Root
           /    \                             /    \
         NA     OB                          OA      OB
       /  |     |  \                      /  |      |  \
     NC  ND     OE  OF                   OC  OD     OE  OF

  In this case, NA and OA are going to be swapped, record (NA, OA) into
  subvolume tree X.

2) After subtree swap.
         reloc tree                         subvolume tree X
            Root                               Root
           /    \                             /    \
         OA     OB                          NA      OB
       /  |     |  \                      /  |      |  \
     OC  OD     OE  OF                   NC  ND     OE  OF

3a) COW happens for OB
    If we are going to COW tree block OB, we check OB's bytenr against
    tree X's swapped_blocks structure.
    If it doesn't fit any, nothing will happen.

3b) COW happens for NA
    Check NA's bytenr against tree X's swapped_blocks, and get a hit.
    Then we do subtree scan on both subtrees OA and NA.
    Resulting 6 tree blocks to be scanned (OA, OC, OD, NA, NC, ND).

    Then no matter what we do to subvolume tree X, qgroup numbers will
    still be correct.
    Then NA's record gets removed from X's swapped_blocks.

4)  Transaction commit
    Any record in X's swapped_blocks gets removed, since there is no
    modification to swapped subtrees, no need to trigger heavy qgroup
    subtree rescan for them.

This will introduce 128 bytes overhead for each btrfs_root even qgroup
is not enabled. This is to reduce memory allocations and potential
failures.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: qgroup: Refactor btrfs_qgroup_trace_subtree_swap
Qu Wenruo [Wed, 23 Jan 2019 07:15:15 +0000 (15:15 +0800)]
btrfs: qgroup: Refactor btrfs_qgroup_trace_subtree_swap

Refactor btrfs_qgroup_trace_subtree_swap() into
qgroup_trace_subtree_swap(), which only needs two extent buffer and some
other bool to control the behavior.

This provides the basis for later delayed subtree scan work.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: relocation: Delay reloc tree deletion after merge_reloc_roots
Qu Wenruo [Wed, 23 Jan 2019 07:15:14 +0000 (15:15 +0800)]
btrfs: relocation: Delay reloc tree deletion after merge_reloc_roots

Relocation code will drop btrfs_root::reloc_root as soon as
merge_reloc_root() finishes.

However later qgroup code will need to access btrfs_root::reloc_root
after merge_reloc_root() for delayed subtree rescan.

So alter the timming of resetting btrfs_root:::reloc_root, make it
happens after transaction commit.

With this patch, we will introduce a new btrfs_root::state,
BTRFS_ROOT_DEAD_RELOC_TREE, to info part of btrfs_root::reloc_tree user
that although btrfs_root::reloc_tree is still non-NULL, but still it's
not used any more.

The lifespan of btrfs_root::reloc tree will become:
          Old behavior            |              New
------------------------------------------------------------------------
btrfs_init_reloc_root()      ---  | btrfs_init_reloc_root()      ---
  set reloc_root              |   |   set reloc_root              |
                              |   |                               |
                              |   |                               |
merge_reloc_root()            |   | merge_reloc_root()            |
|- btrfs_update_reloc_root() ---  | |- btrfs_update_reloc_root() -+-
     clear btrfs_root::reloc_root |      set ROOT_DEAD_RELOC_TREE |
                                  |      record root into dirty   |
                                  |      roots rbtree             |
                                  |                               |
                                  | reloc_block_group() Or        |
                                  | btrfs_recover_relocation()    |
                                  | | After transaction commit    |
                                  | |- clean_dirty_subvols()     ---
                                  |     clear btrfs_root::reloc_root

During ROOT_DEAD_RELOC_TREE set lifespan, the only user of
btrfs_root::reloc_tree should be qgroup.

Since reloc root needs a longer life-span, this patch will also delay
btrfs_drop_snapshot() call.
Now btrfs_drop_snapshot() is called in clean_dirty_subvols().

This patch will increase the size of btrfs_root by 16 bytes.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: call btrfs_create_pending_block_groups unconditionally
Josef Bacik [Wed, 21 Nov 2018 19:05:42 +0000 (14:05 -0500)]
btrfs: call btrfs_create_pending_block_groups unconditionally

The first thing we do is loop through the list, this

if (!list_empty())
btrfs_create_pending_block_groups();

thing is just wasted space.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: make btrfs_destroy_delayed_refs use btrfs_delete_ref_head
Josef Bacik [Wed, 21 Nov 2018 19:05:40 +0000 (14:05 -0500)]
btrfs: make btrfs_destroy_delayed_refs use btrfs_delete_ref_head

Instead of open coding this stuff use the helper instead.

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>
5 years agobtrfs: make btrfs_destroy_delayed_refs use btrfs_delayed_ref_lock
Josef Bacik [Wed, 21 Nov 2018 19:05:39 +0000 (14:05 -0500)]
btrfs: make btrfs_destroy_delayed_refs use btrfs_delayed_ref_lock

We have this open coded in btrfs_destroy_delayed_refs, use the helper
instead.

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>
5 years agobtrfs: scrub: print messages when started or finished
Anand Jain [Thu, 3 Jan 2019 08:17:40 +0000 (16:17 +0800)]
btrfs: scrub: print messages when started or finished

The kernel log messages help debugging and audit, add them for scrub

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: simplify workqueue name when allocating
David Sterba [Thu, 17 Jan 2019 16:15:18 +0000 (17:15 +0100)]
btrfs: simplify workqueue name when allocating

The workqueue name is constructed from a format string but the prefix
does not need to be set by %s.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: merge btrfs_find_device and find_device
Anand Jain [Sat, 19 Jan 2019 06:48:55 +0000 (14:48 +0800)]
btrfs: merge btrfs_find_device and find_device

Both btrfs_find_device() and find_device() does the same thing except
that the latter does not take the seed device onto account in the device
scanning context. We can merge them.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: refactor btrfs_free_stale_devices() to get return value
Anand Jain [Fri, 4 Jan 2019 05:31:53 +0000 (13:31 +0800)]
btrfs: refactor btrfs_free_stale_devices() to get return value

Preparatory patch to add ioctl that allows to forget a device (ie.
reverse of scan).

Refactors btrfs_free_stale_devices() to obtain return status. As this
function can fail if it can't find the given path (returns -ENOENT) or
trying to delete a mounted device (returns -EBUSY).

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: refactor btrfs_find_device() take fs_devices as argument
Anand Jain [Thu, 17 Jan 2019 15:32:31 +0000 (23:32 +0800)]
btrfs: refactor btrfs_find_device() take fs_devices as argument

btrfs_find_device() accepts fs_info as an argument and retrieves
fs_devices from fs_info.

Instead use fs_devices, so that this function can be used in non-mount
(during device scanning) context as well.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: cleanup btrfs_find_device_by_devspec()
Anand Jain [Thu, 17 Jan 2019 15:32:29 +0000 (23:32 +0800)]
btrfs: cleanup btrfs_find_device_by_devspec()

btrfs_find_device_by_devspec() finds the device by @devid or by
@device_path. This patch makes code flow easy to read by open coding the
else part and renames devpath to device_path.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: merge btrfs_find_device_missing_or_by_path() into parent
Anand Jain [Thu, 17 Jan 2019 15:32:28 +0000 (23:32 +0800)]
btrfs: merge btrfs_find_device_missing_or_by_path() into parent

btrfs_find_device_missing_or_by_path() is relatively small function, and
its only parent btrfs_find_device_by_devspec() is small as well. Besides
there are a number of find_device functions. Merge
btrfs_find_device_missing_or_by_path() into its parent.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: Remove not_found_em label from btrfs_get_extent
Nikolay Borisov [Mon, 17 Dec 2018 08:36:02 +0000 (10:36 +0200)]
btrfs: Remove not_found_em label from btrfs_get_extent

In order to avoid duplicating init code for em there is an additional
label, not_found_em, which is used to only set ->block_start. The only
case when it will be used is if the extent we are adding overlaps with
an existing extent. Make that case more obvious by:

 1. Adding a comment hinting at what's going on
 2. Assigning EXTENT_MAP_HOLE and directly going to insert.

 No functional changes.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: Consolidate retval checking of core btree functions
Nikolay Borisov [Mon, 17 Dec 2018 09:49:00 +0000 (11:49 +0200)]
btrfs: Consolidate retval checking of core btree functions

Core btree functions in btrfs generally return 0 when an item is found,
1 in case the sought item cannot be found and <0 when an error happens.
Consolidate the checks for those conditions in one 'if () {} else if ()
{}' construct rather than 2 separate 'if () {}' statements. This
emphasizes that the handling code pertains to a single function. No
functional changes.

Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: Rename found_type to extent_type in btrfs_get_extent
Nikolay Borisov [Mon, 17 Dec 2018 08:35:59 +0000 (10:35 +0200)]
btrfs: Rename found_type to extent_type in btrfs_get_extent

found_type really holds the type of extent and is guaranteed to to have
a value between [0, 2]. The only time it can contain anything different
is if btrfs_lookup_file_extent returned a positive value and the
previous item is different than an extent. Avoid this situation by
simply checking found_key.type rather than assigning the item type to
found_type intermittently. Also make the variable an u8 to reduce stack
usage. No functional changes.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agoBtrfs: move duplicated nodatasum check into common reflink/dedupe helper
Filipe Manana [Wed, 12 Dec 2018 18:05:56 +0000 (18:05 +0000)]
Btrfs: move duplicated nodatasum check into common reflink/dedupe helper

Move the check that verifies if both inodes have checksums disabled or
both have them enabled, from the clone and deduplication functions into
the new common helper btrfs_remap_file_range_prep().

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: Remove impossible condition from mergable_maps
Nikolay Borisov [Tue, 8 Jan 2019 14:53:46 +0000 (16:53 +0200)]
btrfs: Remove impossible condition from mergable_maps

We can never have extents marked as EXTENT_MAP_DELALLOC since this
value is only ever used by btrfs_get_extent_fiemap. In this case the
extent map is created by btrfs_get_extent_fiemap and is never really
published, this flag is used to return the corresponding userspace one.
Considering this, it's pointless having a check for EXTENT_MAP_DELALLOC
in mergable_maps. Just remove it.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agoBtrfs: do not overwrite error return value in the balance ioctl
Filipe Manana [Tue, 8 Jan 2019 11:42:01 +0000 (11:42 +0000)]
Btrfs: do not overwrite error return value in the balance ioctl

If the call to btrfs_balance() failed we would overwrite the error
returned to user space with -EFAULT if the call to copy_to_user() failed
as well. Fix that by calling copy_to_user() only if btrfs_balance()
returned success or was canceled.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agoBtrfs: do not overwrite error return value in the device replace ioctl
Filipe Manana [Tue, 8 Jan 2019 11:42:09 +0000 (11:42 +0000)]
Btrfs: do not overwrite error return value in the device replace ioctl

If the call to btrfs_dev_replace_by_ioctl() failed we would overwrite the
error returned to user space with -EFAULT if the call to copy_to_user()
failed as well. Fix that by calling copy_to_user() only if no error
happened before or a device replace operation was canceled.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agoBtrfs: remove redundant check for swapfiles when reflinking
Filipe Manana [Tue, 8 Jan 2019 11:43:18 +0000 (11:43 +0000)]
Btrfs: remove redundant check for swapfiles when reflinking

Checking if either of the inodes corresponds to a swapfile is already
performed by generic_remap_file_range_prep(), so we do not need to do
it in the btrfs clone and deduplication functions.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: Refactor shrink_delalloc
Nikolay Borisov [Thu, 3 Jan 2019 08:50:05 +0000 (10:50 +0200)]
btrfs: Refactor shrink_delalloc

Add a couple of comments regarding the logic flow in shrink_delalloc.
Then, cease using max_reclaim as a temporary variable when calculating
nr_pages. Finally give max_reclaim a more becoming name, which
uneqivocally shows at what this variable really holds. No functional
changes.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: Document logic regarding inode in async_cow_submit
Nikolay Borisov [Thu, 3 Jan 2019 08:50:03 +0000 (10:50 +0200)]
btrfs: Document logic regarding inode in async_cow_submit

Add a comment explaining when ->inode could be NULL and why we always
perform the ->async_delalloc_pages modification.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: Remove WARN_ON in btrfs_alloc_delalloc_work
Nikolay Borisov [Thu, 3 Jan 2019 08:50:02 +0000 (10:50 +0200)]
btrfs: Remove WARN_ON in btrfs_alloc_delalloc_work

It can never trigger since before calling alloc_delalloc_work we have
called igrab in start_delalloc_inodes.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: Use ihold instead of igrab in cow_file_range_async
Nikolay Borisov [Thu, 3 Jan 2019 08:50:01 +0000 (10:50 +0200)]
btrfs: Use ihold instead of igrab in cow_file_range_async

ihold is supposed to be used when the caller already has a reference to
the inode. In the case of cow_file_range_async this invariants holds,
since the 3 call chains leading to this function all take a reference:

btrfs_writepage  <--- does igrab
 extent_write_full_page
  __extent_writepage
   writepage_delalloc
     btrfs_run_delalloc_range
      cow_file_range_async

extent_write_cache_pages <--- does igrab
 __extent_writepage (same callchain as above)

and

submit_compressed_extents <-- already called from async CoW submit path,
      which would have done ihold.
 extent_write_locked_range
  __extent_writepage

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ add comment ]
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: Remove isize local variable in compress_file_range
Nikolay Borisov [Thu, 3 Jan 2019 08:50:00 +0000 (10:50 +0200)]
btrfs: Remove isize local variable in compress_file_range

It's used only once so just inline the call to i_size_read. The
semantics regarding the inode size are not changed, the pages in the
range are locked and i_size cannot change between the time it was set
and used.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: Remove inode argument from async_cow_submit
Nikolay Borisov [Thu, 3 Jan 2019 08:49:59 +0000 (10:49 +0200)]
btrfs: Remove inode argument from async_cow_submit

We already pass the async_cow struct that holds a reference to the
inode. Exploit this fact and remove the extra inode argument. No
functional changes.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: remove set but not used variable 'num_pages'
YueHaibing [Sat, 15 Dec 2018 06:31:07 +0000 (06:31 +0000)]
btrfs: remove set but not used variable 'num_pages'

Fixes gcc '-Wunused-but-set-variable' warning:

fs/btrfs/ioctl.c: In function 'btrfs_extent_same':
fs/btrfs/ioctl.c:3260:6: warning:
 variable 'num_pages' set but not used [-Wunused-but-set-variable]

It not used any more since commit 9ee8234e6220 ("Btrfs: use
generic_remap_file_range_prep() for cloning and deduplication")

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: Remove redundant assignment in btrfs_get_extent_fiemap
Nikolay Borisov [Wed, 12 Dec 2018 07:42:34 +0000 (09:42 +0200)]
btrfs: Remove redundant assignment in btrfs_get_extent_fiemap

hole_len is only used if the hole falls within the requested range. Make
that explicitly clear by only assigning in the corresponding branch.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: Refactor btrfs_get_extent_fiemap
Nikolay Borisov [Wed, 12 Dec 2018 07:42:33 +0000 (09:42 +0200)]
btrfs: Refactor btrfs_get_extent_fiemap

Make btrfs_get_extent_fiemap a bit more friendly. First step is to
rename the closely related, yet arbitrary named
range_start/found_end/found variables. They define the delalloc range
that is found in case a real extent wasn't found. Subsequently remove
an unnecessary check for hole_em since it's guaranteed to be set i.e the
check is always true. Top it off by giving all comments a refresh.

No functional changes.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ reformatted a few more comments ]
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agobtrfs: Remove unused arguments from btrfs_get_extent_fiemap
Nikolay Borisov [Wed, 12 Dec 2018 07:42:32 +0000 (09:42 +0200)]
btrfs: Remove unused arguments from btrfs_get_extent_fiemap

This function is a simple wrapper over btrfs_get_extent that returns
either:

a) A real extent in the passed range or
b) Adjusted extent based on whether delalloc bytes are found backing up
   a hole.

To support these semantics it doesn't need the page/pg_offset/create
arguments which are passed to btrfs_get_extent in case an extent is to
be created. So simplify the function by removing the unused arguments.
No functional changes.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
5 years agoBtrfs: setup a nofs context for memory allocation at __btrfs_set_acl
Filipe Manana [Thu, 13 Dec 2018 21:16:56 +0000 (21:16 +0000)]
Btrfs: setup a nofs context for memory allocation at __btrfs_set_acl

We are holding a transaction handle when setting an acl, therefore we can
not allocate the xattr value buffer using GFP_KERNEL, as we could deadlock
if reclaim is triggered by the allocation, therefore setup a nofs context.

Fixes: 39a27ec1004e8 ("btrfs: use GFP_KERNEL for xattr and acl allocations")
CC: stable@vger.kernel.org # 4.9+
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>
5 years agoBtrfs: setup a nofs context for memory allocation at btrfs_create_tree()
Filipe Manana [Thu, 13 Dec 2018 21:16:45 +0000 (21:16 +0000)]
Btrfs: setup a nofs context for memory allocation at btrfs_create_tree()

We are holding a transaction handle when creating a tree, therefore we can
not allocate the root using GFP_KERNEL, as we could deadlock if reclaim is
triggered by the allocation, therefore setup a nofs context.

Fixes: 74e4d82757f74 ("btrfs: let callers of btrfs_alloc_root pass gfp flags")
CC: stable@vger.kernel.org # 4.9+
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>
5 years agoBtrfs: do not overwrite error return value in the get device stats ioctl
Filipe Manana [Fri, 14 Dec 2018 19:45:22 +0000 (19:45 +0000)]
Btrfs: do not overwrite error return value in the get device stats ioctl

If the call to btrfs_get_dev_stats() failed we would overwrite the error
returned to user space with -EFAULT if the call to copy_to_user() failed
as well. Fix that by calling copy_to_user() only if btrfs_get_dev_stats()
returned success.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>