block: an exiting task should be allowed to create io_context
authorTejun Heo <tj@kernel.org>
Sun, 25 Dec 2011 13:29:14 +0000 (14:29 +0100)
committerJens Axboe <axboe@kernel.dk>
Sun, 25 Dec 2011 13:29:14 +0000 (14:29 +0100)
commitfd63836811d6e5b5f5f608abf865bc9e91762c8c
tree9262ccdc0319b331c497701a851c8c5752dedb00
parent64c42998f14d5894ea3138625897d620b30c8e4e
block: an exiting task should be allowed to create io_context

While fixing io_context creation / task exit race condition,
6e736be7f2 "block: make ioc get/put interface more conventional and
fix race on alloction" also prevented an exiting (%PF_EXITING) task
from creating its own io_context.  This is incorrect as exit path may
issue IOs, e.g. from exit_files(), and if those IOs are the first ones
issued by the task, io_context needs to be created to process the IOs.

Combined with the existing problem of io_context / io_cq creation
failure having the possibility of stalling IO, this problem results in
deterministic full IO lockup with certain workloads.

Fix it by allowing io_context creation regardless of %PF_EXITING for
%current.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
block/blk-ioc.c