Bug 201793

Description of problem:
I hit a XFS corruption several times on ppc64le, by running generic/299 on 2k blocksize XFS (reflink=1,rmapbt=1):
#
# FSQA Test No. 299
#
# AIO/DIO stress test
# Run random AIO/DIO activity and fallocate/truncate simultaneously
# Test will operate on huge sparsed files so ENOSPC is expected.
#
FSTYP         -- xfs (non-debug)
PLATFORM      -- Linux/ppc64le xxxxxxx 4.20.0-rc1.kasan
MKFS_OPTIONS  -- -f -b size=2048 -m crc=1,finobt=1,reflink=1,rmapbt=1 -i sparse=1 /dev/vda5
MOUNT_OPTIONS -- -o context=system_u:object_r:nfs_t:s0 /dev/vda5 /mnt/xfstests/mnt2

generic/299 176s ... _check_xfs_filesystem: filesystem on /dev/vda5 is inconsistent (r)
(see /var/lib/xfstests/results//generic/299.full for details)

Ran: generic/299
Failures: generic/299
Failed 1 of 1 tests

Version-Release number of selected component (if applicable):
https://git.kernel.org/pub/scm/fs/xfs/xfs-linux.git
xfs-4.20-fixes-2

How reproducible:
1~2% on my ppc64le kvm machine. But it's hard to reproduce on my x86_64 real machine.

Steps to Reproduce:
Loop running generic/299, but maybe not easy to hit it

Actual results:
XFS corruption

Expected results:
test always pass

Additional info:
Please check the full output and xfs metadump from attachments.