Hi,
This patchset contains two patches in the series. The fist patch fixes
the incorrect lock ordering between queue ->sysfs_lock and freeze-lock.
And the second patch avoids using q->sysfs_lock while accessing sysfs
attribute files.
After we modeled the freeze & enter queue as lock for supporting lockdep
under commit f1be1788a32e ("block: model freeze & enter queue as lock
for supporting lockdep"), we received numerous lockdep splats. And one
of those splats[1] reported the potential deadlock due to incorrect lock
ordering issue between q->sysfs-lock and q->q_usage_counter. So the
first patch in the series addresses this lockdep splat.
The second patch in the series removes unnecessary holding of q->sysfs_lock
while we access the sysfs block layer attribute files. Ideally, we don't
need to hold any lock while accessing sysfs attribute files as attribute
file is already protected with internal sysfs/kernfs locking. For instance,
accessing a sysfs attribute file for reading, follows the following
code path:
vfs_read
kernfs_fop_read_iter
seq_read_iter
kernfs_seq_start ==> acquires kernfs internal mutex
kernfs_seq_show
sysfs_kf_seq_show
queue_attr_show
Similarly, accessing a sysfs attribute file for writing, follows the
the following code path:
vfs_write
kernfs_fop_write_iter ==> acquires kernfs internal mutex
sysfs_kf_write
queue_attr_store
As shown in the above code path, kernfs internal mutex already protects
sysfs store/show operations, and so we could safely get rid off holding
q->sysfs_lock while accessing sysfs attribute files.
Please note that above changes were unit tested against blktests and
quick xfstests with lockdep enabled.
Nilay Shroff (2):
block: fix lock ordering between the queue ->sysfs_lock and
freeze-lock
block: avoid acquiring q->sysfs_lock while accessing sysfs attributes
block/blk-mq-sysfs.c | 6 +-----
block/blk-mq.c | 49 +++++++++++++++++++++++++++++---------------
block/blk-sysfs.c | 10 +++------
block/elevator.c | 9 ++++++++
4 files changed, 46 insertions(+), 28 deletions(-)
--
2.47.1
