Use kernel workqueue to implement a new btrfs_workqueue_struct, which
has the ordering execution feature like the btrfs_worker.

The func is executed in a concurrency way, and the
ordred_func/ordered_free is executed in the sequence them are queued
after the corresponding func is done.

The new btrfs_workqueue works much like the original one, one workqueue
for normal work and a list for ordered work.
When a work is queued, ordered work will be added to the list and helper
function will be queued into the workqueue.
The helper function will execute a normal work and then check and execute as many
ordered work as possible in the sequence they were queued.

At this patch, high priority work queue or thresholding is not added yet.
The high priority feature and thresholding will be added in the following patches.

Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Tested-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Josef Bacik <jbacik@fb.com>

The struct async_sched is not used by any codes and can be removed.

Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Reviewed-by: Josef Bacik <jbacik@fusionio.com>
Tested-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Josef Bacik <jbacik@fb.com>

It is really unnecessary to search tree again for @gen, @mode and @rdev
in the case of REG inodes' creation, as we've got btrfs_inode_item in sctx,
and @gen, @mode and @rdev can easily be fetched.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

We can not release the reserved metadata space for the first write if we
find the write position is pre-allocated. Because the kernel might write
the data on the disk before we do the second write but after the can-nocow
check, if we release the space for the first write, we might fail to update
the metadata because of no space.

Fix this problem by end nocow write if there is dirty data in the range whose
space is pre-allocated.

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

The write range may not be sector-aligned, for example:

       |--------|--------|	<- write range, sector-unaligned, size: 2blocks
  |--------|--------|--------|  <- correct lock range, size: 3blocks

But according to the old code, we used the size of write range to calculate
the lock range directly, not considered the offset, we would get a wrong lock
range:

       |--------|--------|	<- write range, sector-unaligned, size: 2blocks
  |--------|--------|		<- wrong lock range, size: 2blocks

And besides that, the old code also had the same problem when calculating
the real write size. Correct them.

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

Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Josef Bacik <jbacik@fb.com>

In "btrfs: send: lower memory requirements in common case" the code to
save the old_buf_len was incorrectly moved to a wrong place and broke
the original logic.

Reported-by: Filipe David Manana <fdmanana@gmail.com>
Signed-off-by: David Sterba <dsterba@suse.cz>
Reviewed-by: Filipe David Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

While droping extent map structures from the extent cache that cover our
target range, we would remove each extent map structure from the red black
tree and then add either 1 or 2 new extent map structures if the former
extent map covered sections outside our target range.

This change simply attempts to replace the existing extent map structure
with a new one that covers the subsection we're not interested in, instead
of doing a red black remove operation followed by an insertion operation.

The number of elements in an inode's extent map tree can get very high for large
files under random writes. For example, while running the following test:

    sysbench --test=fileio --file-num=1 --file-total-size=10G \
        --file-test-mode=rndrw --num-threads=32 --file-block-size=32768 \
        --max-requests=500000 --file-rw-ratio=2 [prepare|run]

I captured the following histogram capturing the number of extent_map items
in the red black tree while that test was running:

    Count: 122462
    Range:  1.000 - 172231.000; Mean: 96415.831; Median: 101855.000; Stddev: 49700.981
    Percentiles:  90th: 160120.000; 95th: 166335.000; 99th: 171070.000
       1.000 -    5.231:   452 |
       5.231 -  187.392:    87 |
     187.392 -  585.911:   206 |
     585.911 - 1827.438:   623 |
    1827.438 - 5695.245:  1962 #
    5695.245 - 17744.861:  6204 ####
   17744.861 - 55283.764: 21115 ############
   55283.764 - 172231.000: 91813 #####################################################

Benchmark:

    sysbench --test=fileio --file-num=1 --file-total-size=10G --file-test-mode=rndwr \
        --num-threads=64 --file-block-size=32768 --max-requests=0 --max-time=60 \
        --file-io-mode=sync --file-fsync-freq=0 [prepare|run]

Before this change: 122.1Mb/sec
After this change:  125.07Mb/sec
(averages of 5 test runs)

Test machine: quad core intel i5-3570K, 32Gb of ram, SSD

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

When we split an extent state there's no need to start the rbtree search
from the root node - we can start it from the original extent state node,
since we would end up in its subtree if we do the search starting at the
root node anyway.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

We don't need to have an unsigned int field in the extent_map struct
to tell us whether the extent map is in the inode's extent_map tree or
not. We can use the rb_node struct field and the RB_CLEAR_NODE and
RB_EMPTY_NODE macros to achieve the same task.

This reduces sizeof(struct extent_map) from 152 bytes to 144 bytes (on a
64 bits system).

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Josef Bacik <jbacik@fb.com>

We won't change commit root, skip locking dance with commit root
when walking backrefs, this can speed up btrfs send operations.

Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

check if @scrubs_running=@scrubs_paused condition inside wait_event()
is not an atomic operation which means we may inc/dec @scrub_running/
paused at any time. Let's wake up @scrub_pause_wait as much as we can
to let commit transaction blocked less.

An example below:

Thread1				Thread2
|->scrub_blocked_if_needed()	|->scrub_pending_trans_workers_inc
  |->increase @scrub_paused
                                       |->increase @scrub_running
  |->wake up scrub_pause_wait list
                                       |->scrub blocked
                                       |->increase @scrub_paused

Thread3 is commiting transaction which is blocked at btrfs_scrub_pause().
So after Thread2 increase @scrub_paused, we meet the condition
@scrub_paused=@scrub_running, but transaction will be still blocked until
another calling to wake up @scrub_pause_wait.

Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

If we fail to commit transaction, we'd better
cancel scrub operations.

Suggested-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

commit cb7ab021

 cause a following deadlock found by
xfstests,btrfs/011:

Thread1 is commiting transaction which is blocked at
btrfs_scrub_pause().

Thread2 is calling btrfs_file_aio_write() which has held
inode's @i_mutex and commit transaction(blocked because
Thread1 is committing transaction).

Thread3 is copy_nocow_page worker which will also try to
hold inode @i_mutex, so thread3 will wait Thread1 finished.

Thread4 is waiting pending workers finished which will wait
Thread3 finished. So the problem is like this:

Thread1--->Thread4--->Thread3--->Thread2---->Thread1

Deadlock happens! we fix it by letting Thread1 go firstly,
which means we won't block transaction commit while we are
waiting pending workers finished.

Reported-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

btrfs_scrub_continue() will be called when cleaning up transaction.However,
this can only be called if btrfs_scrub_pause() is called before.

Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

PTR_RET is deprecated. Use PTR_ERR_OR_ZERO instead. While at it
also include missing err.h header.

Signed-off-by: Sachin Kamat <sachin.kamat@linaro.org>
Signed-off-by: Josef Bacik <jbacik@fb.com>

When doing an incremental send, if we had a directory pending a move/rename
operation and none of its parents, except for the immediate parent, were
pending a move/rename, after processing the directory's references, we would
be issuing utimes, chown and chmod intructions against am outdated path - a
path which matched the one in the parent root.

This change also simplifies a bit the code that deals with building a path
for a directory which has a move/rename operation delayed.

Steps to reproduce:

    $ mkfs.btrfs -f /dev/sdb3
    $ mount /dev/sdb3 /mnt/btrfs
    $ mkdir -p /mnt/btrfs/a/b/c/d/e
    $ mkdir /mnt/btrfs/a/b/c/f
    $ chmod 0777 /mnt/btrfs/a/b/c/d/e
    $ btrfs subvolume snapshot -r /mnt/btrfs /mnt/btrfs/snap1
    $ btrfs send /mnt/btrfs/snap1 -f /tmp/base.send
    $ mv /mnt/btrfs/a/b/c/f /mnt/btrfs/a/b/f2
    $ mv /mnt/btrfs/a/b/c/d/e /mnt/btrfs/a/b/f2/e2
    $ mv /mnt/btrfs/a/b/c /mnt/btrfs/a/b/c2
    $ mv /mnt/btrfs/a/b/c2/d /mnt/btrfs/a/b/c2/d2
    $ chmod 0700 /mnt/btrfs/a/b/f2/e2
    $ btrfs subvolume snapshot -r /mnt/btrfs /mnt/btrfs/snap2
    $ btrfs send -p /mnt/btrfs/snap1 /mnt/btrfs/snap2 -f /tmp/incremental.send

    $ umount /mnt/btrfs
    $ mkfs.btrfs -f /dev/sdb3
    $ mount /dev/sdb3 /mnt/btrfs
    $ btrfs receive /mnt/btrfs -f /tmp/base.send
    $ btrfs receive /mnt/btrfs -f /tmp/incremental.send

The second btrfs receive command failed with:

    ERROR: chmod a/b/c/d/e failed. No such file or directory

A test case for xfstests follows.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

Just comparing the pointers (logical disk addresses) of the btree nodes is
not completely bullet proof, we have to check if their generation numbers
match too.

It is guaranteed that a COW operation will result in a block with a different
logical disk address than the original block's address, but over time we can
reuse that former logical disk address.

For example, creating a 2Gb filesystem on a loop device, and having a script
running in a loop always updating the access timestamp of a file, resulted in
the same logical disk address being reused for the same fs btree block in about
only 4 minutes.

This could make us skip entire subtrees when doing an incremental send (which
is currently the only user of btrfs_compare_trees). However the odds of getting
2 blocks at the same tree level, with the same logical disk address, equal first
slot keys and different generations, should hopefully be very low.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

The incremental send algorithm assumed that it was possible to issue
a directory remove (rmdir) if the the inode number it was currently
processing was greater than (or equal) to any inode that referenced
the directory's inode. This wasn't a valid assumption because any such
inode might be a child directory that is pending a move/rename operation,
because it was moved into a directory that has a higher inode number and
was moved/renamed too - in other words, the case the following commit
addressed:

    9f03740a


    (Btrfs: fix infinite path build loops in incremental send)

This made an incremental send issue an rmdir operation before the
target directory was actually empty, which made btrfs receive fail.
Therefore it needs to wait for all pending child directory inodes to
be moved/renamed before sending an rmdir operation.

Simple steps to reproduce this issue:

    $ mkfs.btrfs -f /dev/sdb3
    $ mount /dev/sdb3 /mnt/btrfs
    $ mkdir -p /mnt/btrfs/a/b/c/x
    $ mkdir /mnt/btrfs/a/b/y
    $ btrfs subvolume snapshot -r /mnt/btrfs /mnt/btrfs/snap1
    $ btrfs send /mnt/btrfs/snap1 -f /tmp/base.send
    $ mv /mnt/btrfs/a/b/y /mnt/btrfs/a/b/YY
    $ mv /mnt/btrfs/a/b/c/x /mnt/btrfs/a/b/YY
    $ rmdir /mnt/btrfs/a/b/c
    $ btrfs subvolume snapshot -r /mnt/btrfs /mnt/btrfs/snap2
    $ btrfs send -p /mnt/btrfs/snap1 /mnt/btrfs/snap2 -f /tmp/incremental.send

    $ umount /mnt/btrfs
    $ mkfs.btrfs -f /dev/sdb3
    $ mount /dev/sdb3 /mnt/btrfs
    $ btrfs receive /mnt/btrfs -f /tmp/base.send
    $ btrfs receive /mnt/btrfs -f /tmp/incremental.send

The second btrfs receive command failed with:

    ERROR: rmdir o259-6-0 failed. Directory not empty

A test case for xfstests follows.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

When doing an incremental send, if we delete a directory that has N > 1
hardlinks for the same file and that file has the highest inode number
inside the directory contents, an incremental send would send N times an
rmdir operation against the directory. This made the btrfs receive command
fail on the second rmdir instruction, as the target directory didn't exist
anymore.

Steps to reproduce the issue:

    $ mkfs.btrfs -f /dev/sdb3
    $ mount /dev/sdb3 /mnt/btrfs
    $ mkdir -p /mnt/btrfs/a/b/c
    $ echo 'ola mundo' > /mnt/btrfs/a/b/c/foo.txt
    $ ln /mnt/btrfs/a/b/c/foo.txt /mnt/btrfs/a/b/c/bar.txt
    $ btrfs subvolume snapshot -r /mnt/btrfs /mnt/btrfs/snap1
    $ btrfs send /mnt/btrfs/snap1 -f /tmp/base.send
    $ rm -f /mnt/btrfs/a/b/c/foo.txt
    $ rm -f /mnt/btrfs/a/b/c/bar.txt
    $ rmdir /mnt/btrfs/a/b/c
    $ btrfs subvolume snapshot -r /mnt/btrfs /mnt/btrfs/snap2
    $ btrfs send -p /mnt/btrfs/snap1 /mnt/btrfs/snap2 -f /tmp/incremental.send

    $ umount /mnt/btrfs
    $ mkfs.btrfs -f /dev/sdb3
    $ mount /dev/sdb3 /mnt/btrfs
    $ btrfs receive /mnt/btrfs -f /tmp/base.send
    $ btrfs receive /mnt/btrfs -f /tmp/incremental.send

The second btrfs receive command failed with:

    ERROR: rmdir o259-6-0 failed. No such file or directory

A test case for xfstests follows.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

This fixes yet one more case not caught by the commit titled:

   Btrfs: fix infinite path build loops in incremental send

In this case, even before the initial full send, we have a directory
which is a child of a directory with a higher inode number. Then we
perform the initial send, and after we rename both the child and the
parent, without moving them around. After doing these 2 renames, an
incremental send sent a rename instruction for the child directory
which contained an invalid "from" path (referenced the parent's old
name, not the new one), which made the btrfs receive command fail.

Steps to reproduce:

    $ mkfs.btrfs -f /dev/sdb3
    $ mount /dev/sdb3 /mnt/btrfs
    $ mkdir -p /mnt/btrfs/a/b
    $ mkdir /mnt/btrfs/d
    $ mkdir /mnt/btrfs/a/b/c
    $ mv /mnt/btrfs/d /mnt/btrfs/a/b/c
    $ btrfs subvolume snapshot -r /mnt/btrfs /mnt/btrfs/snap1
    $ btrfs send /mnt/btrfs/snap1 -f /tmp/base.send
    $ mv /mnt/btrfs/a/b/c /mnt/btrfs/a/b/x
    $ mv /mnt/btrfs/a/b/x/d /mnt/btrfs/a/b/x/y
    $ btrfs subvolume snapshot -r /mnt/btrfs /mnt/btrfs/snap2
    $ btrfs send -p /mnt/btrfs/snap1 /mnt/btrfs/snap2 -f /tmp/incremental.send

    $ umout /mnt/btrfs
    $ mkfs.btrfs -f /dev/sdb3
    $ mount /dev/sdb3 /mnt/btrfs
    $ btrfs receive /mnt/btrfs -f /tmp/base.send
    $ btrfs receive /mnt/btrfs -f /tmp/incremental.send

The second btrfs receive command failed with:
  "ERROR: rename a/b/c/d -> a/b/x/y failed. No such file or directory"

A test case for xfstests follows.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

If we punch beyond the size of an inode, we'll correctly remove any prealloc extents,
but we'll also insert file extent items representing holes (disk bytenr == 0) that start
with a key offset that lies beyond the inode's size and are not contiguous with the last
file extent item.

Example:

  $XFS_IO_PROG -f -c "truncate 118811" $SCRATCH_MNT/foo
  $XFS_IO_PROG -c "fpunch 582007 864596" $SCRATCH_MNT/foo
  $XFS_IO_PROG -c "pwrite -S 0x0d -b 39987 92267 39987" $SCRATCH_MNT/foo

btrfs-debug-tree output:

  item 4 key (257 INODE_ITEM 0) itemoff 15885 itemsize 160
	inode generation 6 transid 6 size 132254 block group 0 mode 100600 links 1
  item 5 key (257 INODE_REF 256) itemoff 15872 itemsize 13
	inode ref index 2 namelen 3 name: foo
  item 6 key (257 EXTENT_DATA 0) itemoff 15819 itemsize 53
	extent data disk byte 0 nr 0 gen 6
	extent data offset 0 nr 90112 ram 122880
	extent compression 0
  item 7 key (257 EXTENT_DATA 90112) itemoff 15766 itemsize 53
	extent data disk byte 12845056 nr 4096 gen 6
	extent data offset 0 nr 45056 ram 45056
	extent compression 2
  item 8 key (257 EXTENT_DATA 585728) itemoff 15713 itemsize 53
	extent data disk byte 0 nr 0 gen 6
	extent data offset 0 nr 860160 ram 860160
	extent compression 0

The last extent item, which represents a hole, is useless as it lies beyond the inode's
size.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

The argument last wasn't used, all callers supplied a NULL value
for it. Also removed unnecessary intermediate storage of the result
of key comparisons.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

When we didn't find the exact ref head we were looking for, if
return_bigger != 0 we set a new search key to match either the
next node after the last one we found or the first one in the
ref heads rb tree, and then did another full tree search. For both
cases this ended up being pointless as we would end up returning
an entry we already had before repeating the search.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

Now that we can adjust the commit interval with a remount, we need
to wake up the transaction thread or else he will continue to sleep
until the previous transaction interval has elapsed before waking
up.  So, if we go from a large commit interval to something smaller,
the transaction thread will not wake up until the large interval has
expired.  This also causes the cleaner thread to stay sleeping, since
it gets woken up by the transaction thread.

Fix it by simply waking up the transaction thread during a remount.

Signed-off-by: Justin Maggard <jmaggard10@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

If the log sync fails, there is something wrong in the log tree, we
should not continue to join the log transaction and log the metadata.
What we should do is to do a full commit.

This patch fixes this problem by setting ->last_trans_log_full_commit
to the current transaction id, it will tell the tasks not to join
the log transaction, and do a full commit.

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

We might commit the log sub-transaction which didn't contain the metadata we
logged. It was because we didn't record the log transid and just select
the current log sub-transaction to commit, but the right one might be
committed by the other task already. Actually, we needn't do anything
and it is safe that we go back directly in this case.

This patch improves the log sync by the above idea. We record the transid
of the log sub-transaction in which we log the metadata, and the transid
of the log sub-transaction we have committed. If the committed transid
is >= the transid we record when logging the metadata, we just go back.

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

It is possible that many tasks sync the log tree at the same time, but
only one task can do the sync work, the others will wait for it. But those
wait tasks didn't get the result of the log sync, and returned 0 when they
ended the wait. It caused those tasks skipped the error handle, and the
serious problem was they told the users the file sync succeeded but in
fact they failed.

This patch fixes this problem by introducing a log context structure,
we insert it into the a global list. When the sync fails, we will set
the error number of every log context in the list, then the waiting tasks
get the error number of the log context and handle the error if need.

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

The log trans id is initialized to be 0 every time we create a log tree,
and the log tree need be re-created after a new transaction is started,
it means the log trans id is unlikely to be a huge number, so we can use
signed integer instead of unsigned long integer to save a bit space.

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

Mutex unlock implies certain memory barriers to make sure all the memory
operation completes before the unlock, and the next mutex lock implies memory
barriers to make sure the all the memory happens after the lock. So it is
a full memory barrier(smp_mb), we needn't add memory barriers. Remove them.

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

The old code would start the log transaction even the log tree init
failed, it was unnecessary. Fix it.

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

We may abort the wait earlier if ->last_trans_log_full_commit was set to
the current transaction id, at this case, we need commit the current
transaction instead of the log sub-transaction. But the current code
didn't tell the caller to do it (return 0, not -EAGAIN). Fix it.

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

->last_trans_log_full_commit may be changed by the other tasks without lock,
so we need prevent the compiler from the optimize access just like
	tmp = fs_info->last_trans_log_full_commit
	if (tmp == ...)
		...

	<do something>

	if (tmp == ...)
		...

In fact, we need get the new value of ->last_trans_log_full_commit during
the second access. Fix it by ACCESS_ONCE().

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

So after transaction is aborted, we need to cleanup inode resources by
calling btrfs_invalidate_inodes(), and btrfs_invalidate_inodes() hopes
roots' refs to be zero in old times and sets a WARN_ON(), however, this
is not always true within cleaning up transaction, so we get to detect
transaction abortion and not warn at all.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

[13654.480669] ======================================================
[13654.480905] [ INFO: possible circular locking dependency detected ]
[13654.481003] 3.12.0+ #4 Tainted: G        W  O
[13654.481060] -------------------------------------------------------
[13654.481060] btrfs-transacti/9347 is trying to acquire lock:
[13654.481060]  (&(&root->ordered_extent_lock)->rlock){+.+...}, at: [<ffffffffa02d30a1>] btrfs_cleanup_transaction+0x271/0x570 [btrfs]
[13654.481060] but task is already holding lock:
[13654.481060]  (&(&fs_info->ordered_root_lock)->rlock){+.+...}, at: [<ffffffffa02d3015>] btrfs_cleanup_transaction+0x1e5/0x570 [btrfs]
[13654.481060] which lock already depends on the new lock.

[13654.481060] the existing dependency chain (in reverse order) is:
[13654.481060] -> #1 (&(&fs_info->ordered_root_lock)->rlock){+.+...}:
[13654.481060]        [<ffffffff810c4103>] lock_acquire+0x93/0x130
[13654.481060]        [<ffffffff81689991>] _raw_spin_lock+0x41/0x50
[13654.481060]        [<ffffffffa02f011b>] __btrfs_add_ordered_extent+0x39b/0x450 [btrfs]
[13654.481060]        [<ffffffffa02f0202>] btrfs_add_ordered_extent+0x32/0x40 [btrfs]
[13654.481060]        [<ffffffffa02df6aa>] run_delalloc_nocow+0x78a/0x9d0 [btrfs]
[13654.481060]        [<ffffffffa02dfc0d>] run_delalloc_range+0x31d/0x390 [btrfs]
[13654.481060]        [<ffffffffa02f7c00>] __extent_writepage+0x310/0x780 [btrfs]
[13654.481060]        [<ffffffffa02f830a>] extent_write_cache_pages.isra.29.constprop.48+0x29a/0x410 [btrfs]
[13654.481060]        [<ffffffffa02f879d>] extent_writepages+0x4d/0x70 [btrfs]
[13654.481060]        [<ffffffffa02d9f68>] btrfs_writepages+0x28/0x30 [btrfs]
[13654.481060]        [<ffffffff8114be91>] do_writepages+0x21/0x50
[13654.481060]        [<ffffffff81140d49>] __filemap_fdatawrite_range+0x59/0x60
[13654.481060]        [<ffffffff81140e13>] filemap_fdatawrite_range+0x13/0x20
[13654.481060]        [<ffffffffa02f1db9>] btrfs_wait_ordered_range+0x49/0x140 [btrfs]
[13654.481060]        [<ffffffffa0318fe2>] __btrfs_write_out_cache+0x682/0x8b0 [btrfs]
[13654.481060]        [<ffffffffa031952d>] btrfs_write_out_cache+0x8d/0xe0 [btrfs]
[13654.481060]        [<ffffffffa02c7083>] btrfs_write_dirty_block_groups+0x593/0x680 [btrfs]
[13654.481060]        [<ffffffffa0345307>] commit_cowonly_roots+0x14b/0x20d [btrfs]
[13654.481060]        [<ffffffffa02d7c1a>] btrfs_commit_transaction+0x43a/0x9d0 [btrfs]
[13654.481060]        [<ffffffffa030061a>] btrfs_create_uuid_tree+0x5a/0x100 [btrfs]
[13654.481060]        [<ffffffffa02d5a8a>] open_ctree+0x21da/0x2210 [btrfs]
[13654.481060]        [<ffffffffa02ab6fe>] btrfs_mount+0x68e/0x870 [btrfs]
[13654.481060]        [<ffffffff811b2409>] mount_fs+0x39/0x1b0
[13654.481060]        [<ffffffff811cd653>] vfs_kern_mount+0x63/0xf0
[13654.481060]        [<ffffffff811cfcce>] do_mount+0x23e/0xa90
[13654.481060]        [<ffffffff811d05a3>] SyS_mount+0x83/0xc0
[13654.481060]        [<ffffffff81692b52>] system_call_fastpath+0x16/0x1b
[13654.481060] -> #0 (&(&root->ordered_extent_lock)->rlock){+.+...}:
[13654.481060]        [<ffffffff810c340a>] __lock_acquire+0x150a/0x1a70
[13654.481060]        [<ffffffff810c4103>] lock_acquire+0x93/0x130
[13654.481060]        [<ffffffff81689991>] _raw_spin_lock+0x41/0x50
[13654.481060]        [<ffffffffa02d30a1>] btrfs_cleanup_transaction+0x271/0x570 [btrfs]
[13654.481060]        [<ffffffffa02d35ce>] transaction_kthread+0x22e/0x270 [btrfs]
[13654.481060]        [<ffffffff81079efa>] kthread+0xea/0xf0
[13654.481060]        [<ffffffff81692aac>] ret_from_fork+0x7c/0xb0
[13654.481060] other info that might help us debug this:

[13654.481060]  Possible unsafe locking scenario:

[13654.481060]        CPU0                    CPU1
[13654.481060]        ----                    ----
[13654.481060]   lock(&(&fs_info->ordered_root_lock)->rlock);
[13654.481060]				 lock(&(&root->ordered_extent_lock)->rlock);
[13654.481060]				 lock(&(&fs_info->ordered_root_lock)->rlock);
[13654.481060]   lock(&(&root->ordered_extent_lock)->rlock);
[13654.481060]
 *** DEADLOCK ***
[...]

======================================================

btrfs_destroy_all_ordered_extents()
gets &fs_info->ordered_root_lock __BEFORE__ acquiring &root->ordered_extent_lock,
while btrfs_[add,remove]_ordered_extent()
acquires &fs_info->ordered_root_lock __AFTER__ getting &root->ordered_extent_lock.

This patch fixes the above problem.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

This is an extension to my previous commit titled:

  "Btrfs: faster file extent item replace operations"
  (hash 1acae57b

)

Instead of inserting the new file extent item if we deleted existing
file extent items covering our target file range, also allow to insert
the new file extent item if we didn't find any existing items to delete
and replace_extent != 0, since in this case our caller would do another
tree search to insert the new file extent item anyway, therefore just
combine the two tree searches into a single one, saving cpu time, reducing
lock contention and reducing btree node/leaf COW operations.

This covers the case where applications keep doing tail append writes to
files, which for example is the case of Apache CouchDB (its database and
view index files are always open with O_APPEND).

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

In case we do not refill, we can overwrite cur pointer from prio_head
by one from not prioritized head, what looks as something that was
not intended.

This change make we always take works from prio_head first until it's
not empty.

Signed-off-by: Stanislaw Gruszka <stf_xl@wp.pl>
Signed-off-by: Josef Bacik <jbacik@fb.com>

This reverts commit 41ce9970

.
Previously i was thinking we can use readonly root's commit root
safely while it is not true, readonly root may be cowed with the
following cases.

1.snapshot send root will cow source root.
2.balance,device operations will also cow readonly send root
to relocate.

So i have two ideas to make us safe to use commit root.

-->approach 1:
make it protected by transaction and end transaction properly and we research
next item from root node(see btrfs_search_slot_for_read()).

-->approach 2:
add another counter to local root structure to sync snapshot with send.
and add a global counter to sync send with exclusive device operations.

So with approach 2, send can use commit root safely, because we make sure
send root can not be cowed during send. Unfortunately, it make codes *ugly*
and more complex to maintain.

To make snapshot and send exclusively, device operations and send operation
exclusively with each other is a little confusing for common users.

So why not drop into previous way.

Cc: Josef Bacik <jbacik@fb.com>
Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

Btrfs send is assuming readonly root won't change, let's skip readonly root.

Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>

Since we have introduced btrfs_previous_extent_item() to search previous
extent item, just switch into it.

Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Reviewed-by: Filipe Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>