xfs: make deferred processing safe for embedded dfops
authorBrian Foster <bfoster@redhat.com>
Tue, 24 Jul 2018 20:43:09 +0000 (13:43 -0700)
committerDarrick J. Wong <darrick.wong@oracle.com>
Thu, 26 Jul 2018 17:15:12 +0000 (10:15 -0700)
commit03f4e4b26cd5f6eed728f82d90039a19d1b51ce3
treed0f08d48590568df7cea64eb92fdf478ab010d98
parentdcbd44f79986e55691600b969c14db004d741883
xfs: make deferred processing safe for embedded dfops

xfs_defer_finish() has a couple quirks that are not safe with
respect to the upcoming internal dfops functionality. First,
xfs_defer_finish() attaches the passed in dfops structure to
->t_dfops and caches and restores the original value. Second, it
continues to use the initial dfops reference before and after the
transaction roll.

These behaviors assume that dop is an independent memory allocation
from the transaction itself, which may not always be true once
transactions begin to use an embedded dfops structure. In the latter
model, dfops processing creates a new xfs_defer_ops structure with
each transaction and the associated state is migrated across to the
new transaction.

Fix up xfs_defer_finish() to handle the possibility of the current
dfops changing after a transaction roll. Since ->t_dfops is used
unconditionally in this path, it is no longer necessary to
attach/restore ->t_dfops and pass it explicitly down to
xfs_defer_trans_roll(). Update dop in the latter function and the
caller to ensure that it always refers to the current dfops
structure.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bill O'Donnell <billodo@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
fs/xfs/libxfs/xfs_defer.c