When testing the previous patch, Zhao Lei reported a similar bug when
attempting to scrub a degraded RAID 5/6 filesystem with a missing
device, leading to NULL pointer dereferences from the RAID 5/6 parity
scrubbing code.

The first cause was the same as in the previous patch: attempting to
call bio_add_page() on a missing block device. To fix this,
scrub_extent_for_parity() can just mark the sectors on the missing
device as errors instead of attempting to read from it.

Additionally, the code uses scrub_remap_extent() to map the extent of
the corresponding data stripe, but the extent wasn't already mapped. If
scrub_remap_extent() finds a missing block device, it doesn't initialize
extent_dev, so we're left with a NULL struct btrfs_device. The solution
is to use btrfs_map_block() directly.
Reported-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>

The original implementation of device replace on RAID 5/6 seems to have
missed support for replacing a missing device. When this is attempted,
we end up calling bio_add_page() on a bio with a NULL ->bi_bdev, which
crashes when we try to dereference it. This happens because
btrfs_map_block() has no choice but to return us the missing device
because RAID 5/6 don't have any alternate mirrors to read from, and a
missing device has a NULL bdev.

The idea implemented here is to handle the missing device case
separately, which better only happen when we're replacing a missing RAID
5/6 device. We use the new BTRFS_RBIO_REBUILD_MISSING operation to
reconstruct the data from parity, check it with
scrub_recheck_block_checksum(), and write it out with
scrub_write_block_to_dev_replace().
Reported-by: Philip <bugzilla@philip-seeger.de>
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=96141Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>

The current RAID 5/6 recovery code isn't quite prepared to handle
missing devices. In particular, it expects a bio that we previously
attempted to use in the read path, meaning that it has valid pages
allocated. However, missing devices have a NULL blkdev, and we can't
call bio_add_page() on a bio with a NULL blkdev. We could do manual
manipulation of bio->bi_io_vec, but that's pretty gross. So instead, add
a separate path that allows us to manually add pages to the rbio.
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>

Commit 5fbc7c59 ("Btrfs: fix unfinished readahead thread for raid5/6
degraded mounting") fixed a problem where we would skip a missing device
when we shouldn't have because there are no other mirrors to read from
in RAID 5/6. After commit 2c8cdd6e ("Btrfs, replace: write dirty
pages into the replace target device"), the fix doesn't work when we're
doing a missing device replace on RAID 5/6 because the replace device is
counted as a mirror so we're tricked into thinking we can safely skip
the missing device. The fix is to count only the real stripes and decide
based on that.
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>

scrub_submit() claims that it can handle a bio with a NULL block device,
but this is misleading, as calling bio_add_page() on a bio with a NULL
->bi_bdev would've already crashed. Delete this, as we're about to
properly handle a missing block device.
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>

Clone and extent same lock their source and target inodes in opposite order.
In addition to this, the range locking in clone doesn't take ordering into
account. Fix this by having clone use the same locking helpers as
btrfs-extent-same.

In addition, I do a small cleanup of the locking helpers, removing a case
(both inodes being the same) which was poorly accounted for and never
actually used by the callers.
Signed-off-by: Mark Fasheh <mfasheh@suse.de>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>

The file layout is

[extent 1]...[extent n][4k extent][HOLE][extent x]

extent 1~n and 4k extent can be merged during defrag, and the whole
defrag bytes is larger than our defrag thresh(256k), 4k extent as a
tail is left unmerged since we check if its next extent can be merged
(the next one is a hole, so the check will fail), the layout thus can
be

[new extent][4k extent][HOLE][extent x]
 (1~n)

To fix it, beside looking at the next one, this also looks at the
previous one by checking @defrag_end, which is set to 0 when we
decide to stop merging contiguous extents, otherwise, we can merge
the previous one with our extent.

Also, this makes btrfs behave consistent with how xfs and ext4 do.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>

When we do backref walking, we search firstly in queued delayed refs
and then the on-disk backrefs, but we parse differently for shared
references, for delayed refs we also add 'ref->root' while for on-disk
backrefs we don't, this can prevent us from merging refs indexed
by the same bytenr and cause find_parent_nodes() to throw a warning at
'WARN_ON(ref->count < 0)', for example, when we have a shared data extent
with 'ref_cnt=1' and a delayed shared data with a BTRFS_DROP_DELAYED_REF,
that happens.

For shared references, no matter if it's delayed or on-disk, ref->root is
not at all used, instead it's ref->parent that really matters, so this has
delayed refs handled as the same way as on-disk refs.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>

When a task trying to double lock a extent buffer, there are no
lockdep warning about it because this lock may be in "blocking_lock"
state, and make us hard to debug.

This patch add a WARN_ON() for above condition, it can not report
all deadlock cases(as lock between tasks), but at least helps us
some.
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>

Because it is never used.
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>

These wrong comment was copyed from another function(expired) from
init, this patch fixed them.
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>

When btrfs_reloc_cow_block() failed in __btrfs_cow_block(), current
code just return a err-value to caller, but leave new_created extent
buffer exist and locked.

Then subsequent code (in relocate) try to lock above eb again,
and caused deadlock without any dmesg.
(eb lock use wait_event(), so no lockdep message)

It is hard to do recover work in __btrfs_cow_block() at this error
point, but we can abort transaction to avoid deadlock and operate on
unstable state.a

It also helps developer to find wrong place quickly.
(better than a frozen fs without any dmesg before patch)
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>

These arguments are not used in functions, remove them for cleanup
and make kernel stack happy.
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>

Remove chunk_objectid argument from btrfs_relocate_chunk() because
it is not necessary, it can also cleanup some code in caller for
prepare its value.
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>

objectid's init-value is not used in any case, remove it.
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>

We need error checking code for get_ref_objectid_v0() in
relocate_block_group(), to avoid unpredictable result, especially
for accessing uninitialized value(when function failed) after
this line.
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>

xfstests btrfs/070 sometimes failed.
In my test machine, its fail rate is about 30%.
In another vm(vmware), its fail rate is about 50%.

Reason:
  btrfs/070 do replace and defrag with fsstress simultaneously,
  after above operation, checksum error is found by scrub.

  Actually, it have no relationship with defrag operation, only
  replace with fsstress can trigger this bug.

  New data writen to target device have possibility rewrited by
  old data from source device by replace code in debug, to avoid
  above problem, we can set target block group to readonly in
  replace period, so new data requested by other operation will
  not write to same place with replace code.

  Before patch(4.1-rc3):
    30% failed in 100 xfstests.
  After patch:
    0% failed in 300 xfstests.

It also happened in btrfs/071 as it's another scrub with IO load tests.
Reported-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>

Use new intruduced scrub_pause_on/off() can make this code block
clean and more readable.
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>

It can reduce current duplicated code which is similar to
scrub_blocked_if_needed() but can not call it because little
different.
It also used by my next patch which is in same case.
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>

More than one code call set_block_group_ro() and restore rw in fail.

Old code use bool bit to save blockgroup's ro state, it can not
support parallel case(it is confirmd exist in my debug log).

This patch use ref count to store ro state, and rename
set_block_group_ro/set_block_group_rw
to
inc_block_group_ro/dec_block_group_ro.
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>

When we access extent_root in scrub_stripe() and
scrub_raid56_parity(), we need bypass unrelated tree item firstly
before using its contents to do other condition.

It is not a bug fix, only making code sequence in logic.
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>

We need not load csum of whole strip in scrub because strip is trimed
before use, it is to say, what we really need to calculate csum is
data between [extent_logical, extent_len).

This patch changed to use above segment for btrfs_lookup_csums_range()
in scrub_stripe()
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>

For example, in scrub_raid56_parity(), following lines are used
to judge is all data processed:
 place1: if (key.objectid > logic_end) ...
 place2: if (logic_start >= logic_end) ...
 ...
 (place2 is typo, is should be ">", it is copied from other
  place, where logic_end's meaning is different, long story...)

We can fix above typo directly, but the root reason is ambiguous
meaning of logic_end in scrub raid56 parity.

In other place, XXX_end is pointed to data which is not included,
and we need to process segment of [XXX_start, XXX_end).

But for scrub raid56 parity, logic_end is pointed to lattest data
need to process, and introduced many "+ 1" and "- 1" in code as
below:
 length = sparity->logic_end - sparity->logic_start + 1
 logic_end - logic_start + 1
 stripe_logical + increment - 1

This patch changed logic_end's meaning to make it in normal understanding
in raid56 parity functions and data struct alone with above bugfix.
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>

When scrub_extent() failed, we need to free previois created
checksum list.
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>

Old code checking cancel and pause request inside scrub stripe
operation, like:
  loop() {
    if (parity) {
      scrub_parity_stripe();
      continue;
    }

    check_cancel_and_pause()

    scrub_normal_stripe();
  }

Reason is when introduce raid56 stripe scrub, new code is inserted
simplely to front of loop.

Better to:
  loop() {
    check_cancel_and_pause()

    if (parity)
      scrub_parity_stripe();
    else
      scrub_normal_stripe();
  }

This patch adjusted code place to realize above sequence.
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>

When mount failed because missing device, we can see following
dmesg:
 [ 1060.267743] BTRFS: too many missing devices, writeable mount is not allowed
 [ 1060.273158] BTRFS: open_ctree failed

This patch add missing_device_number and tolerated_missing_device_number
to above output, to let user know what really happened, and helps
bug-report and debug.

dmesg after patch:
 [  127.050367] BTRFS: missing devices(1) exceeds the limit(0), writeable mount is not allowed
 [  127.056099] BTRFS: open_ctree failed

Changelog v1->v2:
1: Changed to more clear description, suggested-by:
   Anand Jain <anand.jain@oracle.com>
Suggested-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>

Scrub panic in following operation:
  mkfs.ext4 /dev/vdh
  btrfs-convert /dev/vdh
  mount /dev/vdh /mnt/tmp1
  btrfs scrub start -B /dev/vdh
  (panic)

Reason:
  1: In some case, leaf created by btrfs-convert was splited into 2
     strips.
  2: Scrub bypassed part of above wrong leaf data, but remain data
     caused panic in scrub_checksum_tree_block().

For reason 1:
  we can get following information after some simple operation.
  a. mkfs.ext4 /dev/vdh
     btrfs-convert /dev/vdh
  b. btrfs-debug-tree /dev/vdh
     we can see following item in extent tree:
     item 25 key (27054080 METADATA_ITEM 0) itemoff 15083 itemsize 33
     Its logical address is [27054080, 27070464)
     and acrossed 2 strips:
     [27000832, 27066368)
     [27066368, 27131904)
  Will be fixed in btrfs-progs(btrfs-convert, btrfsck, ...)

For reason 2:
  Scrub is trying to do a "bypass" in this case, but the result is
  "panic", because current code lacks of some condition in bypass,
  and let some wrong leaf data escaped.

This patch fixed above scrub code.

Before patch:
  # btrfs scrub start -B /dev/vdh
  (panic)

After patch:
  # btrfs scrub start -B /dev/vdh
  scrub done for 353cec8f-da31-4a94-aa35-be72d997b06e
  ...
  # dmesg
  ...
  [   59.088697] BTRFS error (device vdh): scrub: tree block 27054080 spanning stripes, ignored. logical=27000832
  [   59.089929] BTRFS error (device vdh): scrub: tree block 27054080 spanning stripes, ignored. logical=27066368
  #
Reported-by: Chris Murphy <lists@colorremedies.com>
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>

We have one more case where after a log tree is replayed we get
inconsistent metadata leading to stale directory entries, due to
some directories having entries pointing to some inode while the
inode does not have a matching BTRFS_INODE_[REF|EXTREF]_KEY item.

To trigger the problem we need to have a file with multiple hard links
belonging to different parent directories. Then if one of those hard
links is removed and we fsync the file using one of its other links
that belongs to a different parent directory, we end up not logging
the fact that the removed hard link doesn't exists anymore in the
parent directory.

Simple reproducer:

  seq=`basename $0`
  seqres=$RESULT_DIR/$seq
  echo "QA output created by $seq"
  tmp=/tmp/$$
  status=1	# failure is the default!
  trap "_cleanup; exit \$status" 0 1 2 3 15

  _cleanup()
  {
      _cleanup_flakey
      rm -f $tmp.*
  }

  # get standard environment, filters and checks
  . ./common/rc
  . ./common/filter
  . ./common/dmflakey

  # real QA test starts here
  _need_to_be_root
  _supported_fs generic
  _supported_os Linux
  _require_scratch
  _require_dm_flakey
  _require_metadata_journaling $SCRATCH_DEV

  rm -f $seqres.full

  _scratch_mkfs >>$seqres.full 2>&1
  _init_flakey
  _mount_flakey

  # Create our test directory and file.
  mkdir $SCRATCH_MNT/testdir
  touch $SCRATCH_MNT/foo
  ln $SCRATCH_MNT/foo $SCRATCH_MNT/testdir/foo2
  ln $SCRATCH_MNT/foo $SCRATCH_MNT/testdir/foo3

  # Make sure everything done so far is durably persisted.
  sync

  # Now we remove one of our file's hardlinks in the directory testdir.
  unlink $SCRATCH_MNT/testdir/foo3

  # We now fsync our file using the "foo" link, which has a parent that
  # is not the directory "testdir".
  $XFS_IO_PROG -c "fsync" $SCRATCH_MNT/foo

  # Silently drop all writes and unmount to simulate a crash/power
  # failure.
  _load_flakey_table $FLAKEY_DROP_WRITES
  _unmount_flakey

  # Allow writes again, mount to trigger journal/log replay.
  _load_flakey_table $FLAKEY_ALLOW_WRITES
  _mount_flakey

  # After the journal/log is replayed we expect to not see the "foo3"
  # link anymore and we should be able to remove all names in the
  # directory "testdir" and then remove it (no stale directory entries
  # left after the journal/log replay).
  echo "Entries in testdir:"
  ls -1 $SCRATCH_MNT/testdir

  rm -f $SCRATCH_MNT/testdir/*
  rmdir $SCRATCH_MNT/testdir

  _unmount_flakey

  status=0
  exit

The test fails with:

  $ ./check generic/107
  FSTYP         -- btrfs
  PLATFORM      -- Linux/x86_64 debian3 4.1.0-rc6-btrfs-next-11+
  MKFS_OPTIONS  -- /dev/sdc
  MOUNT_OPTIONS -- /dev/sdc /home/fdmanana/btrfs-tests/scratch_1

  generic/107 3s ... - output mismatch (see .../results/generic/107.out.bad)
    --- tests/generic/107.out	2015-08-01 01:39:45.807462161 +0100
    +++ /home/fdmanana/git/hub/xfstests/results//generic/107.out.bad
    @@ -1,3 +1,5 @@
     QA output created by 107
     Entries in testdir:
     foo2
    +foo3
    +rmdir: failed to remove '/home/fdmanana/btrfs-tests/scratch_1/testdir': Directory not empty
    ...
    _check_btrfs_filesystem: filesystem on /dev/sdc is inconsistent \
      (see /home/fdmanana/git/hub/xfstests/results//generic/107.full)
    _check_dmesg: something found in dmesg (see .../results/generic/107.dmesg)
  Ran: generic/107
  Failures: generic/107
  Failed 1 of 1 tests

  $ cat /home/fdmanana/git/hub/xfstests/results//generic/107.full
  (...)
  checking fs roots
  root 5 inode 257 errors 200, dir isize wrong
	unresolved ref dir 257 index 3 namelen 4 name foo3 filetype 1 errors 5, no dir item, no inode ref
  (...)

And produces the following warning in dmesg:

  [127298.759064] BTRFS info (device dm-0): failed to delete reference to foo3, inode 258 parent 257
  [127298.762081] ------------[ cut here ]------------
  [127298.763311] WARNING: CPU: 10 PID: 7891 at fs/btrfs/inode.c:3956 __btrfs_unlink_inode+0x182/0x35a [btrfs]()
  [127298.767327] BTRFS: Transaction aborted (error -2)
  (...)
  [127298.788611] Call Trace:
  [127298.789137]  [<ffffffff8145f077>] dump_stack+0x4f/0x7b
  [127298.790090]  [<ffffffff81095de5>] ? console_unlock+0x356/0x3a2
  [127298.791157]  [<ffffffff8104b3b0>] warn_slowpath_common+0xa1/0xbb
  [127298.792323]  [<ffffffffa065ad09>] ? __btrfs_unlink_inode+0x182/0x35a [btrfs]
  [127298.793633]  [<ffffffff8104b410>] warn_slowpath_fmt+0x46/0x48
  [127298.794699]  [<ffffffffa065ad09>] __btrfs_unlink_inode+0x182/0x35a [btrfs]
  [127298.797640]  [<ffffffffa065be8f>] btrfs_unlink_inode+0x1e/0x40 [btrfs]
  [127298.798876]  [<ffffffffa065bf11>] btrfs_unlink+0x60/0x9b [btrfs]
  [127298.800154]  [<ffffffff8116fb48>] vfs_unlink+0x9c/0xed
  [127298.801303]  [<ffffffff81173481>] do_unlinkat+0x12b/0x1fb
  [127298.802450]  [<ffffffff81253855>] ? lockdep_sys_exit_thunk+0x12/0x14
  [127298.803797]  [<ffffffff81174056>] SyS_unlinkat+0x29/0x2b
  [127298.805017]  [<ffffffff81465197>] system_call_fastpath+0x12/0x6f
  [127298.806310] ---[ end trace bbfddacb7aaada7b ]---
  [127298.807325] BTRFS warning (device dm-0): __btrfs_unlink_inode:3956: Aborting unused transaction(No such entry).

So fix this by logging all parent inodes, current and old ones, to make
sure we do not get stale entries after log replay. This is not a simple
solution such as triggering a full transaction commit because it would
imply full transaction commit when an inode is fsynced in the same
transaction that modified it and reloaded it after eviction (because its
last_unlink_trans is set to the same value as its last_trans as of the
commit with the title "Btrfs: fix stale dir entries after unlink, inode
eviction and fsync"), and it would also make fstest generic/066 fail
since one of the fsyncs triggers a full commit and the next fsync will
not find the inode in the log anymore (therefore not removing the xattr).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>

The search key advancing condition used in copy_to_sk() is loose. It can
advance the key even if it reaches sk->max_*: e.g. when the max key = (512,
1024, -1) and the current key = (512, 1025, 10), it increments the
offset by 1, continues hopeless search from (512, 1025, 11). This issue
make ioctl() to take unexpectedly long time scanning all the leaf a blocks
one by one.

This commit fix the problem using standard way of key comparison:
btrfs_comp_cpu_keys()
Signed-off-by: Naohiro Aota <naota@elisp.net>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>

When cloning/deduplicating file extents (through the clone and extent_same
ioctls) we can get data back references with offset values that are a
result of an unsigned integer arithmetic underflow, that is, values that
are much larger then they could be otherwise.

This is not a problem when decrementing or dropping the back references
(happens when we overwrite the extents or punch a hole for example, through
__btrfs_drop_extents()), since we compute the same too large offset value,
but it is a problem for the backref walking code, used by an incremental
send and the ioctls that are used by the btrfs tool "inspect-internal"
commands, as it makes it miss the corresponding file extent items because
the search key is set for an extent item that starts at an offset matching
the exceptionally large offset value of the data back reference. For an
incremental send this causes the send ioctl to fail with -EIO.

So teach the backref walking code to deal with these cases by setting the
search key's offset to 0 if the backref's offset value is larger than
LLONG_MAX (the largest possible file offset). This makes sure the backref
walking code finds the corresponding file extent items at the expense of
scanning more items and leafs in the btree.

Fixing the clone/dedup ioctls to not produce such underflowed results would
require major changes breaking backward compatibility, updating user space
tools, etc.

Simple reproducer case for fstests:

  seq=`basename $0`
  seqres=$RESULT_DIR/$seq
  echo "QA output created by $seq"

  tmp=/tmp/$$
  status=1	# failure is the default!
  trap "_cleanup; exit \$status" 0 1 2 3 15

  _cleanup()
  {
      rm -fr $send_files_dir
      rm -f $tmp.*
  }

  # get standard environment, filters and checks
  . ./common/rc
  . ./common/filter

  # real QA test starts here
  _supported_fs btrfs
  _supported_os Linux
  _require_scratch
  _require_cloner
  _need_to_be_root

  send_files_dir=$TEST_DIR/btrfs-test-$seq

  rm -f $seqres.full
  rm -fr $send_files_dir
  mkdir $send_files_dir

  _scratch_mkfs >>$seqres.full 2>&1
  _scratch_mount

  # Create our test file with a single extent of 64K starting at file
  # offset 128K.
  $XFS_IO_PROG -f -c "pwrite -S 0xaa 128K 64K" $SCRATCH_MNT/foo \
      | _filter_xfs_io

  _run_btrfs_util_prog subvolume snapshot -r $SCRATCH_MNT \
      $SCRATCH_MNT/mysnap1

  # Now clone parts of the original extent into lower offsets of the file.
  #
  # The first clone operation adds a file extent item to file offset 0
  # that points to our initial extent with a data offset of 16K. The
  # corresponding data back reference in the extent tree has an offset of
  # 18446744073709535232, which is the result of file_offset - data_offset
  # = 0 - 16K.
  #
  # The second clone operation adds a file extent item to file offset 16K
  # that points to our initial extent with a data offset of 48K. The
  # corresponding data back reference in the extent tree has an offset of
  # 18446744073709518848, which is the result of file_offset - data_offset
  # = 16K - 48K.
  #
  # Those large back reference offsets (result of unsigned arithmetic
  # underflow) confused the back reference walking code (used by an
  # incremental send and the multiple inspect-internal ioctls) and made it
  # miss the back references, which for the case of an incremental send it
  # made it fail with -EIO and print a message like the following to
  # dmesg:
  #
  # "BTRFS error (device sdc): did not find backref in send_root. \
  #  inode=257, offset=0, disk_byte=12845056 found extent=12845056"
  #
  $CLONER_PROG -s $(((128 + 16) * 1024)) -d 0 -l $((16 * 1024)) \
      $SCRATCH_MNT/foo $SCRATCH_MNT/foo
  $CLONER_PROG -s $(((128 + 48) * 1024)) -d $((16 * 1024)) \
      -l $((16 * 1024)) $SCRATCH_MNT/foo $SCRATCH_MNT/foo

  _run_btrfs_util_prog subvolume snapshot -r $SCRATCH_MNT \
      $SCRATCH_MNT/mysnap2

  _run_btrfs_util_prog send $SCRATCH_MNT/mysnap1 -f $send_files_dir/1.snap
  _run_btrfs_util_prog send -p $SCRATCH_MNT/mysnap1 $SCRATCH_MNT/mysnap2 \
      -f $send_files_dir/2.snap

  echo "File digest in the original filesystem:"
  md5sum $SCRATCH_MNT/mysnap2/foo | _filter_scratch

  # Now recreate the filesystem by receiving both send streams and verify
  # we get the same file contents that the original filesystem had.
  _scratch_unmount
  _scratch_mkfs >>$seqres.full 2>&1
  _scratch_mount

  _run_btrfs_util_prog receive $SCRATCH_MNT -f $send_files_dir/1.snap
  _run_btrfs_util_prog receive $SCRATCH_MNT -f $send_files_dir/2.snap

  echo "File digest in the new filesystem:"
  md5sum $SCRATCH_MNT/mysnap2/foo | _filter_scratch

  status=0
  exit

The test's expected golden output is:

  wrote 65536/65536 bytes at offset 131072
  XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
  File digest in the original filesystem:
  6c6079335cff141b8a31233ead04cbff  SCRATCH_MNT/mysnap2/foo
  File digest in the new filesystem:
  6c6079335cff141b8a31233ead04cbff  SCRATCH_MNT/mysnap2/foo

But it failed with:

    (...)
    @@ -1,7 +1,5 @@
     QA output created by 097
     wrote 65536/65536 bytes at offset 131072
     XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
    -File digest in the original filesystem:
    -6c6079335cff141b8a31233ead04cbff  SCRATCH_MNT/mysnap2/foo
    -File digest in the new filesystem:
    -6c6079335cff141b8a31233ead04cbff  SCRATCH_MNT/mysnap2/foo
    ...

  $ cat /home/fdmanana/git/hub/xfstests/results//btrfs/097.full
  (...)
  ERROR: send ioctl failed with -5: Input/output error
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>

If we remove a hard link from an inode, the inode gets evicted, then
we fsync the inode and then power fail/crash, when the log tree is
replayed, the parent directory inode still has entries pointing to
the name that no longer exists, while our inode no longer has the
BTRFS_INODE_REF_KEY item matching the deleted hard link (as expected),
leaving the filesystem in an inconsistent state. The stale directory
entries can not be deleted (an attempt to delete them causes -ESTALE
errors), which makes it impossible to delete the parent directory.

This happens because we track the id of the transaction where the last
unlink operation for the inode happened (last_unlink_trans) in an
in-memory only field of the inode, that is, a value that is never
persisted in the inode item stored on the fs/subvol btree. So if an
inode is evicted and loaded again, the value for last_unlink_trans is
set to 0, which prevents the fsync from logging the parent directory
at btrfs_log_inode_parent(). So fix this by setting last_unlink_trans
to the id of the transaction that last modified the inode when we
load the inode. This is a pessimistic approach but it always ensures
correctness with the trade off of ocassional full transaction commits
when an fsync is done against the inode in the same transaction where
it was evicted and reloaded when our inode is a directory and often
logging its parent unnecessarily when our inode is not a directory.

The following test case for fstests triggers the problem:

  seq=`basename $0`
  seqres=$RESULT_DIR/$seq
  echo "QA output created by $seq"
  tmp=/tmp/$$
  status=1	# failure is the default!
  trap "_cleanup; exit \$status" 0 1 2 3 15

  _cleanup()
  {
      _cleanup_flakey
      rm -f $tmp.*
  }

  # get standard environment, filters and checks
  . ./common/rc
  . ./common/filter
  . ./common/dmflakey

  # real QA test starts here
  _need_to_be_root
  _supported_fs generic
  _supported_os Linux
  _require_scratch
  _require_dm_flakey
  _require_metadata_journaling $SCRATCH_DEV

  rm -f $seqres.full

  _scratch_mkfs >>$seqres.full 2>&1
  _init_flakey
  _mount_flakey

  # Create our test file with 2 hard links.
  mkdir $SCRATCH_MNT/testdir
  touch $SCRATCH_MNT/testdir/foo
  ln $SCRATCH_MNT/testdir/foo $SCRATCH_MNT/testdir/bar

  # Make sure everything done so far is durably persisted.
  sync

  # Now remove one of the links, trigger inode eviction and then fsync
  # our inode.
  unlink $SCRATCH_MNT/testdir/bar
  echo 2 > /proc/sys/vm/drop_caches
  $XFS_IO_PROG -c "fsync" $SCRATCH_MNT/testdir/foo

  # Silently drop all writes on our scratch device to simulate a power failure.
  _load_flakey_table $FLAKEY_DROP_WRITES
  _unmount_flakey

  # Allow writes again and mount the fs to trigger log/journal replay.
  _load_flakey_table $FLAKEY_ALLOW_WRITES
  _mount_flakey

  # Now verify our directory entries.
  echo "Entries in testdir:"
  ls -1 $SCRATCH_MNT/testdir

  # If we remove our inode, its parent should become empty and therefore we should
  # be able to remove the parent.
  rm -f $SCRATCH_MNT/testdir/*
  rmdir $SCRATCH_MNT/testdir

  _unmount_flakey

  # The fstests framework will call fsck against our filesystem which will verify
  # that all metadata is in a consistent state.

  status=0
  exit

The test failed on btrfs with:

  generic/098 4s ... - output mismatch (see /home/fdmanana/git/hub/xfstests/results//generic/098.out.bad)
    --- tests/generic/098.out	2015-07-23 18:01:12.616175932 +0100
    +++ /home/fdmanana/git/hub/xfstests/results//generic/098.out.bad	2015-07-23 18:04:58.924138308 +0100
    @@ -1,3 +1,6 @@
     QA output created by 098
     Entries in testdir:
    +bar
     foo
    +rm: cannot remove '/home/fdmanana/btrfs-tests/scratch_1/testdir/foo': Stale file handle
    +rmdir: failed to remove '/home/fdmanana/btrfs-tests/scratch_1/testdir': Directory not empty
    ...
    (Run 'diff -u tests/generic/098.out /home/fdmanana/git/hub/xfstests/results//generic/098.out.bad'  to see the entire diff)
  _check_btrfs_filesystem: filesystem on /dev/sdc is inconsistent (see /home/fdmanana/git/hub/xfstests/results//generic/098.full)

  $ cat /home/fdmanana/git/hub/xfstests/results//generic/098.full
  (...)
  checking fs roots
  root 5 inode 258 errors 2001, no inode item, link count wrong
     unresolved ref dir 257 index 0 namelen 3 name foo filetype 1 errors 6, no dir index, no inode ref
     unresolved ref dir 257 index 3 namelen 3 name bar filetype 1 errors 5, no dir item, no inode ref
  Checking filesystem on /dev/sdc
  (...)
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>

We have another case where after an fsync log replay we get an inode with
a wrong link count (smaller than it should be) and a number of directory
entries greater than its link count. This happens when we add a new link
hard link to our inode A and then we fsync some other inode B that has
the side effect of logging the parent directory inode too. In this case
at log replay time we add the new hard link to our inode (the item with
key BTRFS_INODE_REF_KEY) when processing the parent directory but we
never adjust the link count of our inode A. As a result we get stale dir
entries for our inode A that can never be deleted and therefore it makes
it impossible to remove the parent directory (as its i_size can never
decrease back to 0).

A simple reproducer for fstests that triggers this issue:

  seq=`basename $0`
  seqres=$RESULT_DIR/$seq
  echo "QA output created by $seq"
  tmp=/tmp/$$
  status=1	# failure is the default!
  trap "_cleanup; exit \$status" 0 1 2 3 15

  _cleanup()
  {
      _cleanup_flakey
      rm -f $tmp.*
  }

  # get standard environment, filters and checks
  . ./common/rc
  . ./common/filter
  . ./common/dmflakey

  # real QA test starts here
  _need_to_be_root
  _supported_fs generic
  _supported_os Linux
  _require_scratch
  _require_dm_flakey
  _require_metadata_journaling $SCRATCH_DEV

  rm -f $seqres.full

  _scratch_mkfs >>$seqres.full 2>&1
  _init_flakey
  _mount_flakey

  # Create our test directory and files.
  mkdir $SCRATCH_MNT/testdir
  touch $SCRATCH_MNT/testdir/foo
  touch $SCRATCH_MNT/testdir/bar

  # Make sure everything done so far is durably persisted.
  sync

  # Create one hard link for file foo and another one for file bar. After
  # that fsync only the file bar.
  ln $SCRATCH_MNT/testdir/bar $SCRATCH_MNT/testdir/bar_link
  ln $SCRATCH_MNT/testdir/foo $SCRATCH_MNT/testdir/foo_link
  $XFS_IO_PROG -c "fsync" $SCRATCH_MNT/testdir/bar

  # Silently drop all writes on scratch device to simulate power failure.
  _load_flakey_table $FLAKEY_DROP_WRITES
  _unmount_flakey

  # Allow writes again and mount the fs to trigger log/journal replay.
  _load_flakey_table $FLAKEY_ALLOW_WRITES
  _mount_flakey

  # Now verify both our files have a link count of 2.
  echo "Link count for file foo: $(stat --format=%h $SCRATCH_MNT/testdir/foo)"
  echo "Link count for file bar: $(stat --format=%h $SCRATCH_MNT/testdir/bar)"

  # We should be able to remove all the links of our files in testdir, and
  # after that the parent directory should become empty and therefore
  # possible to remove it.
  rm -f $SCRATCH_MNT/testdir/*
  rmdir $SCRATCH_MNT/testdir

  _unmount_flakey

  # The fstests framework will call fsck against our filesystem which will verify
  # that all metadata is in a consistent state.

  status=0
  exit

The test fails with:

 -Link count for file foo: 2
 +Link count for file foo: 1
  Link count for file bar: 2
 +rm: cannot remove '/home/fdmanana/btrfs-tests/scratch_1/testdir/foo_link': Stale file handle
 +rmdir: failed to remove '/home/fdmanana/btrfs-tests/scratch_1/testdir': Directory not empty
 (...)
 _check_btrfs_filesystem: filesystem on /dev/sdc is inconsistent

And fsck's output:

  (...)
  checking fs roots
  root 5 inode 258 errors 2001, no inode item, link count wrong
      unresolved ref dir 257 index 5 namelen 8 name foo_link filetype 1 errors 4, no inode ref
  Checking filesystem on /dev/sdc
  (...)

So fix this by marking inodes for link count fixup at log replay time
whenever a directory entry is replayed if the entry was created in the
transaction where the fsync was made and if it points to a non-directory
inode.

This isn't a new problem/regression, the issue exists for a long time,
possibly since the log tree feature was added (2008).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>

Pull powerpc fixes from Michael Ellerman:
 - TCE table memory calculation fix from Alexey
 - Build fix for ans-lcd from Luis
 - Unbalanced IRQ warning fix from Alistair

* tag 'powerpc-4.2-3' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux:
  powerpc/eeh-powernv: Fix unbalanced IRQ warning
  macintosh/ans-lcd: fix build failure after module_init/exit relocation
  powerpc/powernv/ioda2: Fix calculation for memory allocated for TCE table

Ted Ts'o reports that his Lenovo T540p ThinkPad crashes at boot if
attached to the docking station.  This is a regression that he was able
to bisect to commit 8c7b5ccb: "drm/i915: Use atomic helpers for
computing changed flags:"

The reason seems to be the new call to drm_atomic_helper_check_modeset()
added to intel_modeset_compute_config(), which in turn calls
update_connector_routing(), and somehow ends up picking a NULL crtc for
the connector state, causing the subsequent drm_crtc_index() to OOPS.

Daniel Vetter says that the fundamental issue seems to be confusion in
the encoder selection, and this isn't the right fix, but while he chases
down the proper fix, this at least avoids the NULL pointer dereference
and makes Ted's docking station work again.
Reported-bisected-and-tested-by: Theodore Ts'o <tytso@mit.edu>
Cc: Daniel Vetter <daniel.vetter@intel.com>
Cc: Mani Nikula <jani.nikula@linux.intel.com>
Cc: Dave Airlie <airlied@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Pull SCSI fixes from James Bottomley:
 "A set of three fixes for the ipr driver and one fairly major one for
  memory leaks in the mq path of SCSI"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: fix memory leak with scsi-mq
  ipr: Fix invalid array indexing for HRRQ
  ipr: Fix incorrect trace indexing
  ipr: Fix locking for unit attention handling

Pull ARM SoC fixes from Olof Johansson:
 "Things are calming down nicely here w.r.t. fixes.  This batch
  includes two week's worth since I missed to send before -rc4.

  Nothing particularly scary to point out, smaller fixes here and there.
  Shortlog describes it pretty well"

* tag 'armsoc-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc:
  ARM: dts: keystone: fix dt bindings to use post div register for mainpll
  ARM: nomadik: disable UART0 on Nomadik boards
  ARM: dts: i.MX35: Fix can support.
  ARM: OMAP2+: hwmod: Fix _wait_target_ready() for hwmods without sysc
  ARM: dts: add CPU OPP and regulator supply property for exynos4210
  ARM: dts: Update video-phy node with syscon phandle for exynos3250
  ARM: DRA7: hwmod: fix gpmc hwmod

Pull VFS fix from Al Viro:
 "Spurious ENOTDIR fix"

This should fix the problems reported by Dominique Martinet and Hugh
Dickins.

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  link_path_walk(): be careful when failing with ENOTDIR

In RCU mode we might end up with dentry evicted just we check
that it's a directory.  In such case we should return ECHILD
rather than ENOTDIR, so that pathwalk would be retries in non-RCU
mode.

Breakage had been introduced in commit b18825a7 - prior to that
we were looking at nd->inode, which had been fetched before
verifying that ->d_seq was still valid.  That form of check
would only be satisfied if at some point the pathname prefix
would indeed have resolved to a non-directory.  The fix consists
of checking ->d_seq after we'd run into a non-directory dentry,
and failing with ECHILD in case of mismatch.

Note that all branches since 3.12 have that problem...
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

Pull dmaengine fixes from Vinod Koul:
 "We had a regression due to reuse of descriptor so we have reverted
  that.

  The rest are driver fixes:

   - at_hdmac and at_xdmac for residue, trannfer width, and channel config
   - pl330 final fix for dma fails and overflow issue
   - xgene resouce map fix
   - mv_xor big endian op fix"

* tag 'dmaengine-fix-4.2-rc5' of git://git.infradead.org/users/vkoul/slave-dma:
  Revert "dmaengine: virt-dma: don't always free descriptor upon completion"
  dmaengine: mv_xor: fix big endian operation in register mode
  dmaengine: xgene-dma: Fix the resource map to handle overlapping
  dmaengine: at_xdmac: fix transfer data width in at_xdmac_prep_slave_sg()
  dmaengine: at_hdmac: fix residue computation
  dmaengine: at_xdmac: fix bug about channel configuration
  dmaengine: pl330: Really fix choppy sound because of wrong residue calculation
  dmaengine: pl330: Fix overflow when reporting residue in memcpy