When testing fsstress with snapshot making background, some snapshot
following problem.

Snapshot 270:
inode 323: size 0

Snapshot 271:
inode 323: size 349145
|-------Hole---|---------Empty gap-------|-------Hole-----|
0	    122880			172032	      349145

Snapshot 272:
inode 323: size 349145
|-------Hole---|------------Data---------|-------Hole-----|
0	    122880			172032	      349145

The fsstress operation on inode 323 is the following:
write: 		offset 	126832 	len 43124
truncate: 	size 	349145

Since the write with offset is consist of 2 operations:
1. punch hole
2. write data
Hole punching is faster than data write, so hole punching in write
and truncate is done first and then buffered write, so the snapshot 271 got
empty gap, which will not pass btrfsck.

To fix the bug, this patch will change the write sequence which will
first punch a hole covering the write end if a hole is needed.
Reported-by: NGui Hecheng <guihc.fnst@cn.fujitsu.com>
Signed-off-by: NQu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: NChris Mason <clm@fb.com>

Print the message only when the device is seen for the first time.
Signed-off-by: NDavid Sterba <dsterba@suse.cz>
Signed-off-by: NChris Mason <clm@fb.com>

When encountering memory pressure, testers have run into the following
lockdep warning. It was caused by __link_block_group calling kobject_add
with the groups_sem held. kobject_add calls kvasprintf with GFP_KERNEL,
which gets us into reclaim context. The kobject doesn't actually need
to be added under the lock -- it just needs to ensure that it's only
added for the first block group to be linked.

=========================================================
[ INFO: possible irq lock inversion dependency detected ]
3.14.0-rc8-default #1 Not tainted
---------------------------------------------------------
kswapd0/169 just changed the state of lock:
 (&delayed_node->mutex){+.+.-.}, at: [<ffffffffa018baea>] __btrfs_release_delayed_node+0x3a/0x200 [btrfs]
but this lock took another, RECLAIM_FS-unsafe lock in the past:
 (&found->groups_sem){+++++.}

and interrupts could create inverse lock ordering between them.

other info that might help us debug this:
 Possible interrupt unsafe locking scenario:
       CPU0                    CPU1
       ----                    ----
  lock(&found->groups_sem);
                               local_irq_disable();
                               lock(&delayed_node->mutex);
                               lock(&found->groups_sem);
  <Interrupt>
    lock(&delayed_node->mutex);

 *** DEADLOCK ***
2 locks held by kswapd0/169:
 #0:  (shrinker_rwsem){++++..}, at: [<ffffffff81159e8a>] shrink_slab+0x3a/0x160
 #1:  (&type->s_umount_key#27){++++..}, at: [<ffffffff811bac6f>] grab_super_passive+0x3f/0x90
Signed-off-by: NJeff Mahoney <jeffm@suse.com>
Signed-off-by: NChris Mason <clm@fb.com>

We currently rely too heavily on roots being read-only to save us from just
accessing root->commit_root.  We can easily balance blocks out from underneath a
read only root, so to save us from getting screwed make sure we only access
root->commit_root under the commit root sem.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Signed-off-by: NChris Mason <clm@fb.com>

Lets try this again.  We can deadlock the box if we send on a box and try to
write onto the same fs with the app that is trying to listen to the send pipe.
This is because the writer could get stuck waiting for a transaction commit
which is being blocked by the send.  So fix this by making sure looking at the
commit roots is always going to be consistent.  We do this by keeping track of
which roots need to have their commit roots swapped during commit, and then
taking the commit_root_sem and swapping them all at once.  Then make sure we
take a read lock on the commit_root_sem in cases where we search the commit root
to make sure we're always looking at a consistent view of the commit roots.
Previously we had problems with this because we would swap a fs tree commit root
and then swap the extent tree commit root independently which would cause the
backref walking code to screw up sometimes.  With this patch we no longer
deadlock and pass all the weird send/receive corner cases.  Thanks,
Reportedy-by: NHugo Mills <hugo@carfax.org.uk>
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Signed-off-by: NChris Mason <clm@fb.com>

So I have an awful exercise script that will run snapshot, balance and
send/receive in parallel.  This sometimes would crash spectacularly and when it
came back up the fs would be completely hosed.  Turns out this is because of a
bad interaction of balance and send/receive.  Send will hold onto its entire
path for the whole send, but its blocks could get relocated out from underneath
it, and because it doesn't old tree locks theres nothing to keep this from
happening.  So it will go to read in a slot with an old transid, and we could
have re-allocated this block for something else and it could have a completely
different transid.  But because we think it is invalid we clear uptodate and
re-read in the block.  If we do this before we actually write out the new block
we could write back stale data to the fs, and boom we're screwed.

Now we definitely need to fix this disconnect between send and balance, but we
really really need to not allow ourselves to accidently read in stale data over
new data.  So make sure we check if the extent buffer is not under io before
clearing uptodate, this will kick back EIO to the caller instead of reading in
stale data and keep us from corrupting the fs.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Signed-off-by: NChris Mason <clm@fb.com>

We could have possibly added an extent_op to the locked_ref while we dropped
locked_ref->lock, so check for this case as well and loop around.  Otherwise we
could lose flag updates which would lead to extent tree corruption.  Thanks,

cc: stable@vger.kernel.org
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Signed-off-by: NChris Mason <clm@fb.com>

This was done to allow NO_COW to continue to be NO_COW after relocation but it
is not right.  When relocating we will convert blocks to FULL_BACKREF that we
relocate.  We can leave some of these full backref blocks behind if they are not
cow'ed out during the relocation, like if we fail the relocation with ENOSPC and
then just drop the reloc tree.  Then when we go to cow the block again we won't
lookup the extent flags because we won't think there has been a snapshot
recently which means we will do our normal ref drop thing instead of adding back
a tree ref and dropping the shared ref.  This will cause btrfs_free_extent to
blow up because it can't find the ref we are trying to free.  This was found
with my ref verifying tool.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Signed-off-by: NChris Mason <clm@fb.com>

xfstests's btrfs/035 triggers a BUG_ON, which we use to detect the split
of inline extents in __btrfs_drop_extents().

For inline extents, we cannot duplicate another EXTENT_DATA item, because
it breaks the rule of inline extents, that is, 'start offset' needs to be 0.

We have set limitations for the source inode's compressed inline extents,
because it needs to decompress and recompress.  Now the destination inode's
inline extents also need similar limitations.

With this, xfstests btrfs/035 doesn't run into panic.
Signed-off-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NChris Mason <clm@fb.com>

fs/btrfs/send.c:2926: warning: ‘entry’ may be used uninitialized in this
function
Signed-off-by: NChris Mason <clm@fb.com>

I added an optimization for large files where we would stop searching for
backrefs once we had looked at the number of references we currently had for
this extent.  This works great most of the time, but for snapshots that point to
this extent and has changes in the original root this assumption falls on it
face.  So keep track of any delayed ref mods made and add in the actual ref
count as reported by the extent item and use that to limit how far down an inode
we'll search for extents.  Thanks,
Reportedy-by: NHugo Mills <hugo@carfax.org.uk>
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Reported-by: NHugo Mills <hugo@carfax.org.uk>
Tested-by: NHugo Mills <hugo@carfax.org.uk>
Signed-off-by: NChris Mason <clm@fb.com>

For an incremental send, fix the process of determining whether the directory
inode we're currently processing needs to have its move/rename operation delayed.

We were ignoring the fact that if the inode's new immediate ancestor has a higher
inode number than ours but wasn't renamed/moved, we might still need to delay our
move/rename, because some other ancestor directory higher in the hierarchy might
have an inode number higher than ours *and* was renamed/moved too - in this case
we have to wait for rename/move of that ancestor to happen before our current
directory's rename/move operation.

Simple steps to reproduce this issue:

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

      $ mkdir -p /mnt/a/x1/x2
      $ mkdir /mnt/a/Z
      $ mkdir -p /mnt/a/x1/x2/x3/x4/x5

      $ btrfs subvolume snapshot -r /mnt /mnt/snap1
      $ btrfs send /mnt/snap1 -f /tmp/base.send

      $ mv /mnt/a/x1/x2/x3 /mnt/a/Z/X33
      $ mv /mnt/a/x1/x2 /mnt/a/Z/X33/x4/x5/X22

      $ btrfs subvolume snapshot -r /mnt /mnt/snap2
      $ btrfs send -p /mnt/snap1 /mnt/snap2 -f /tmp/incremental.send

The incremental send caused the kernel code to enter an infinite loop when
building the path string for directory Z after its references are processed.

A more complex scenario:

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

      $ mkdir -p /mnt/a/b/c/d
      $ mkdir /mnt/a/b/c/d/e
      $ mkdir /mnt/a/b/c/d/f
      $ mv /mnt/a/b/c/d/e /mnt/a/b/c/d/f/E2
      $ mkdir /mmt/a/b/c/g
      $ mv /mnt/a/b/c/d /mnt/a/b/D2

      $ btrfs subvolume snapshot -r /mnt /mnt/snap1
      $ btrfs send /mnt/snap1 -f /tmp/base.send

      $ mkdir /mnt/a/o
      $ mv /mnt/a/b/c/g /mnt/a/b/D2/f/G2
      $ mv /mnt/a/b/D2 /mnt/a/b/dd
      $ mv /mnt/a/b/c /mnt/a/C2
      $ mv /mnt/a/b/dd/f /mnt/a/o/FF
      $ mv /mnt/a/b /mnt/a/o/FF/E2/BB

      $ btrfs subvolume snapshot -r /mnt /mnt/snap2
      $ btrfs send -p /mnt/snap1 /mnt/snap2 -f /tmp/incremental.send

A test case for xfstests follows.
Signed-off-by: NFilipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: NChris Mason <clm@fb.com>

It's possible to change the parent/child relationship between directories
in such a way that if a child directory has a higher inode number than
its parent, it doesn't necessarily means the child rename/move operation
can be performed immediately. The parent migth have its own rename/move
operation delayed, therefore in this case the child needs to have its
rename/move operation delayed too, and be performed after its new parent's
rename/move.

Steps to reproduce the issue:

      $ umount /mnt
      $ mkfs.btrfs -f /dev/sdd
      $ mount /dev/sdd /mnt

      $ mkdir /mnt/A
      $ mkdir /mnt/B
      $ mkdir /mnt/C
      $ mv /mnt/C /mnt/A
      $ mv /mnt/B /mnt/A/C
      $ mkdir /mnt/A/C/D

      $ btrfs subvolume snapshot -r /mnt /mnt/snap1
      $ btrfs send /mnt/snap1 -f /tmp/base.send

      $ mv /mnt/A/C/D /mnt/A/D2
      $ mv /mnt/A/C/B /mnt/A/D2/B2
      $ mv /mnt/A/C /mnt/A/D2/B2/C2

      $ btrfs subvolume snapshot -r /mnt /mnt/snap2
      $ btrfs send -p /mnt/snap1 /mnt/snap2 -f /tmp/incremental.send

The incremental send caused the kernel code to enter an infinite loop when
building the path string for directory C after its references are processed.

The necessary conditions here are that C has an inode number higher than both
A and B, and B as an higher inode number higher than A, and D has the highest
inode number, that is:
    inode_number(A) < inode_number(B) < inode_number(C) < inode_number(D)

The same issue could happen if after the first snapshot there's any number
of intermediary parent directories between A2 and B2, and between B2 and C2.

A test case for xfstests follows, covering this simple case and more advanced
ones, with files and hard links created inside the directories.
Signed-off-by: NFilipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: NChris Mason <clm@fb.com>

No need to search in the send tree for the generation number of the inode,
we already have it in the recorded_ref structure passed to us.
Signed-off-by: NFilipe David Borba Manana <fdmanana@gmail.com>
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NChris Mason <clm@fb.com>

While we update an existing ref head's extent_op, we're not holding
its spinlock, so while we're updating its extent_op contents (key,
flags) we can have a task running __btrfs_run_delayed_refs() that
holds the ref head's lock and sets its extent_op to NULL right after
the task updating the ref head just checked its extent_op was not NULL.
Signed-off-by: NFilipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: NChris Mason <clm@fb.com>

Since most of the btrfs_workqueue is printed as pointer address,
for easier analysis, add trace for btrfs_workqueue alloc/destroy.
So it is possible to determine the workqueue that a given work belongs
to(by comparing the wq pointer address with alloc trace event).
Signed-off-by: NQu Wenruo <quenruo@cn.fujitsu.com>
Signed-off-by: NChris Mason <clm@fb.com>

When finding new extents during an autodefrag, don't do so many fs tree
lookups to find an extent with a size smaller then the target treshold.
Instead, after each fs tree forward search immediately unlock upper
levels and process the entire leaf while holding a read lock on the leaf,
since our leaf processing is very fast.
This reduces lock contention, allowing for higher concurrency when other
tasks want to write/update items related to other inodes in the fs tree,
as we're not holding read locks on upper tree levels while processing the
leaf and we do less tree searches.

Test:

    sysbench --test=fileio --file-num=512 --file-total-size=16G \
       --file-test-mode=rndrw --num-threads=32 --file-block-size=32768 \
       --file-rw-ratio=3 --file-io-mode=sync --max-time=1800 \
       --max-requests=10000000000 [prepare|run]

(fileystem mounted with -o autodefrag, averages of 5 runs)

Before this change: 58.852Mb/sec throughtput, read 77.589Gb, written 25.863Gb
After this change:  63.034Mb/sec throughtput, read 83.102Gb, written 27.701Gb

Test machine: quad core intel i5-3570K, 32Gb of RAM, SSD.
Signed-off-by: NFilipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: NChris Mason <clm@fb.com>

The error message is confusing:

 # btrfs sub delete /mnt/mysub/
 Delete subvolume '/mnt/mysub'
 ERROR: cannot delete '/mnt/mysub' - Directory not empty

The error message does not make sense to me: It's not about deleting a
directory but it's a subvolume, and it doesn't matter if the subvolume is
empty or not.

Maybe EPERM or is more appropriate in this case, combined with an explanatory
kernel log message. (e.g. "subvolume with ID 123 cannot be deleted because
it is configured as default subvolume.")
Reported-by: NKoen De Wit <koen.de.wit@oracle.com>
Signed-off-by: NGuangyu Sun <guangyu.sun@oracle.com>
Reviewed-by: NDavid Sterba <dsterba@suse.cz>
Signed-off-by: NChris Mason <clm@fb.com>

Signed-off-by: NFilipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: NChris Mason <clm@fb.com>

When locking file ranges in the inode's io_tree, cache the first
extent state that belongs to the target range, so that when unlocking
the range we don't need to search in the io_tree again, reducing cpu
time and making and therefore holding the io_tree's lock for a shorter
period.
Signed-off-by: NFilipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: NChris Mason <clm@fb.com>

Zach found this deadlock that would happen like this

btrfs_end_transaction <- reduce trans->use_count to 0
  btrfs_run_delayed_refs
    btrfs_cow_block
      find_free_extent
	btrfs_start_transaction <- increase trans->use_count to 1
          allocate chunk
	btrfs_end_transaction <- decrease trans->use_count to 0
	  btrfs_run_delayed_refs
	    lock tree block we are cowing above ^^

We need to only decrease trans->use_count if it is above 1, otherwise leave it
alone.  This will make nested trans be the only ones who decrease their added
ref, and will let us get rid of the trans->use_count++ hack if we have to commit
the transaction.  Thanks,

cc: stable@vger.kernel.org
Reported-by: NZach Brown <zab@redhat.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Tested-by: NZach Brown <zab@redhat.com>
Signed-off-by: NChris Mason <clm@fb.com>

We didn't have a lock to protect the access to the delalloc inodes list, that is
we might access a empty delalloc inodes list if someone start flushing delalloc
inodes because the delalloc inodes were moved into a other list temporarily.
Fix it by wrapping the access with a lock.
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>

When we create a snapshot, we just need wait the ordered extents in
the source fs/file root, but because we use the global mutex to protect
this ordered extents list of the source fs/file root to avoid accessing
a empty list, if someone got the mutex to access the ordered extents list
of the other fs/file root, we had to wait.

This patch splits the above global mutex, now every fs/file root has
its own mutex to protect its own list.
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>

We needn't flush all delalloc inodes when we doesn't get s_umount lock,
or we would make the tasks wait for a long time.
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>

generic/074 in xfstests failed sometimes because of the enospc error,
the reason of this problem is that we just reclaimed the space we need
from the reserved space for delalloc, and then tried to reserve the space,
but if some task did no-flush reservation between the above reclamation
and reservation,
	Task1			Task2
	shrink_delalloc()
	reclaim 1 block
	(The space that can
	 be reserved now is 1
	 block)
				do no-flush reservation
				reserve 1 block
				(The space that can
				 be reserved now is 0
				 block)
	reserving 1 block failed
the reservation of Task1 failed, but in fact, there was enough space to
reserve if we could reclaim more space before.

Fix this problem by the aggressive reclamation of the reserved delalloc
metadata space.
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>

The reason is:
- The per-cpu counter has its own lock to protect itself.
- Here we needn't get a exact value.
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>

Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>

As the comment in the btrfs_direct_IO says, only the compressed pages need be
flush again to make sure they are on the disk, but the common pages needn't,
so we add a if statement to check if the inode has compressed pages or not,
if no, skip the flush.

And in order to prevent the write ranges from intersecting, we need wait for
the running ordered extents. But the current code waits for them twice, one
is done before the direct IO starts (in btrfs_wait_ordered_range()), the other
is before we get the blocks, it is unnecessary. because we can do the direct
IO without holding i_mutex, it means that the intersected ordered extents may
happen during the direct IO, the first wait can not avoid this problem. So we
use filemap_fdatawrite_range() instead of btrfs_wait_ordered_range() to remove
the first wait.
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>

The tasks that wait for the IO_DONE flag just care about the io of the dirty
pages, so it is better to wake up them immediately after all the pages are
written, not the whole process of the io completes.
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>

btrfs_wait_ordered_roots() moves all the list entries to a new list,
and then deals with them one by one. But if the other task invokes this
function at that time, it would get a empty list. It makes the enospc
error happens more early. Fix it.
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>

If the snapshot creation happened after the nocow write but before the dirty
data flush, we would fail to flush the dirty data because of no space.

So we must keep track of when those nocow write operations start and when they
end, if there are nocow writers, the snapshot creators must wait. In order
to implement this function, I introduce btrfs_{start, end}_nocow_write(),
which is similar to mnt_{want,drop}_write().

These two functions are only used for nocow file write operations.
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>

Add ftrace for btrfs_workqueue for further workqueue tunning.
This patch needs to applied after the workqueue replace patchset.
Signed-off-by: NQu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>

The new btrfs_workqueue still use open-coded function defition,
this patch will change them into btrfs_func_t type which is much the
same as kernel workqueue.
Signed-off-by: NQu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>

Btrfs send reads data from disk and then writes to a stream via pipe or
a file via flush.

Currently we're going to read each page a time, so every page results
in a disk read, which is not friendly to disks, esp. HDD.  Given that,
the performance can be gained by adding readahead for those pages.

Here is a quick test:
$ btrfs subvolume create send
$ xfs_io -f -c "pwrite 0 1G" send/foobar
$ btrfs subvolume snap -r send ro
$ time "btrfs send ro -f /dev/null"

           w/o             w
real    1m37.527s       0m9.097s
user    0m0.122s        0m0.086s
sys     0m53.191s       0m12.857s
Signed-off-by: NLiu Bo <bo.li.liu@oracle.com>
Reviewed-by: NDavid Sterba <dsterba@suse.cz>
Signed-off-by: NJosef Bacik <jbacik@fb.com>

This has no functional change, only picks out the same part of two functions,
and makes it shared.
Signed-off-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>

When we're finishing processing of an inode, if we're dealing with a
directory inode that has a pending move/rename operation, we don't
need to send a utimes update instruction to the send stream, as we'll
do it later after doing the move/rename operation. Therefore we save
some time here building paths and doing btree lookups.
Signed-off-by: NFilipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>

When using prealloc extents, a file defragment operation may actually
fragment the file and increase the amount of data space used by the file.
This change fixes that behaviour.

Example:

$ mkfs.btrfs -f /dev/sdb3
$ mount /dev/sdb3 /mnt
$ cd /mnt
$ xfs_io -f -c 'falloc 0 1048576' foobar && sync
$ xfs_io -c 'pwrite -S 0xff -b 100000 5000 100000' foobar
$ xfs_io -c 'pwrite -S 0xac -b 100000 200000 100000' foobar
$ xfs_io -c 'pwrite -S 0xe1 -b 100000 900000 100000' foobar && sync

Before defragmenting the file:

$ btrfs filesystem df /mnt
Data, single: total=8.00MiB, used=1.25MiB
System, DUP: total=8.00MiB, used=16.00KiB
System, single: total=4.00MiB, used=0.00
Metadata, DUP: total=1.00GiB, used=112.00KiB
Metadata, single: total=8.00MiB, used=0.00

$ btrfs-debug-tree /dev/sdb3
(...)
	item 6 key (257 EXTENT_DATA 0) itemoff 15810 itemsize 53
		prealloc data disk byte 12845056 nr 1048576
		prealloc data offset 0 nr 4096
	item 7 key (257 EXTENT_DATA 4096) itemoff 15757 itemsize 53
		extent data disk byte 12845056 nr 1048576
		extent data offset 4096 nr 102400 ram 1048576
		extent compression 0
	item 8 key (257 EXTENT_DATA 106496) itemoff 15704 itemsize 53
		prealloc data disk byte 12845056 nr 1048576
		prealloc data offset 106496 nr 90112
	item 9 key (257 EXTENT_DATA 196608) itemoff 15651 itemsize 53
		extent data disk byte 12845056 nr 1048576
		extent data offset 196608 nr 106496 ram 1048576
		extent compression 0
	item 10 key (257 EXTENT_DATA 303104) itemoff 15598 itemsize 53
		prealloc data disk byte 12845056 nr 1048576
		prealloc data offset 303104 nr 593920
	item 11 key (257 EXTENT_DATA 897024) itemoff 15545 itemsize 53
		extent data disk byte 12845056 nr 1048576
		extent data offset 897024 nr 106496 ram 1048576
		extent compression 0
	item 12 key (257 EXTENT_DATA 1003520) itemoff 15492 itemsize 53
		prealloc data disk byte 12845056 nr 1048576
		prealloc data offset 1003520 nr 45056
(...)

Now defragmenting the file results in more data space used than before:

$ btrfs filesystem defragment -f foobar && sync
$ btrfs filesystem df /mnt
Data, single: total=8.00MiB, used=1.55MiB
System, DUP: total=8.00MiB, used=16.00KiB
System, single: total=4.00MiB, used=0.00
Metadata, DUP: total=1.00GiB, used=112.00KiB
Metadata, single: total=8.00MiB, used=0.00

And the corresponding file extent items are now no longer perfectly sequential
as before, and we're now needlessly using more space from data block groups:

$ btrfs-debug-tree /dev/sdb3
(...)
	item 6 key (257 EXTENT_DATA 0) itemoff 15810 itemsize 53
		extent data disk byte 12845056 nr 1048576
		extent data offset 0 nr 4096 ram 1048576
		extent compression 0
	item 7 key (257 EXTENT_DATA 4096) itemoff 15757 itemsize 53
		extent data disk byte 13893632 nr 102400
		extent data offset 0 nr 102400 ram 102400
		extent compression 0
	item 8 key (257 EXTENT_DATA 106496) itemoff 15704 itemsize 53
		extent data disk byte 12845056 nr 1048576
		extent data offset 106496 nr 90112 ram 1048576
		extent compression 0
	item 9 key (257 EXTENT_DATA 196608) itemoff 15651 itemsize 53
		extent data disk byte 13996032 nr 106496
		extent data offset 0 nr 106496 ram 106496
		extent compression 0
	item 10 key (257 EXTENT_DATA 303104) itemoff 15598 itemsize 53
		prealloc data disk byte 12845056 nr 1048576
		prealloc data offset 303104 nr 593920
	item 11 key (257 EXTENT_DATA 897024) itemoff 15545 itemsize 53
		extent data disk byte 14102528 nr 106496
		extent data offset 0 nr 106496 ram 106496
		extent compression 0
	item 12 key (257 EXTENT_DATA 1003520) itemoff 15492 itemsize 53
		extent data disk byte 12845056 nr 1048576
		extent data offset 1003520 nr 45056 ram 1048576
		extent compression 0
(...)

With this change, the above example will no longer cause allocation of new data
space nor change the sequentiality of the file extents, that is, defragment will
be effectless, leaving all extent items pointing to the extent starting at disk
byte 12845056.

In a 20Gb filesystem I had, mounted with the autodefrag option and 20 files of
400Mb each, initially consisting of a single prealloc extent of 400Mb, having
random writes happening at a low rate, lead to a total of over ~17Gb of data
space used, not far from eventually reaching an ENOSPC state.
Signed-off-by: NFilipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>

When the defrag flag BTRFS_DEFRAG_RANGE_START_IO is set and compression
enabled, we weren't flushing completely, as writing compressed extents
is a 2 steps process, one to compress the data and another one to write
the compressed data to disk.
Signed-off-by: NFilipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>

Since the "_struct" suffix is mainly used for distinguish the differnt
btrfs_work between the original and the newly created one,
there is no need using the suffix since all btrfs_workers are changed
into btrfs_workqueue.

Also this patch fixed some codes whose code style is changed due to the
too long "_struct" suffix.
Signed-off-by: NQu Wenruo <quwenruo@cn.fujitsu.com>
Tested-by: NDavid Sterba <dsterba@suse.cz>
Signed-off-by: NJosef Bacik <jbacik@fb.com>

Since all the btrfs_worker is replaced with the newly created
btrfs_workqueue, the old codes can be easily remove.
Signed-off-by: NQuwenruo <quwenruo@cn.fujitsu.com>
Tested-by: NDavid Sterba <dsterba@suse.cz>
Signed-off-by: NJosef Bacik <jbacik@fb.com>