writeback: don't drain bdi_writeback_congested on bdi destruction
authorTejun Heo <tj@kernel.org>
Thu, 2 Jul 2015 00:53:37 +0000 (20:53 -0400)
committerJens Axboe <axboe@fb.com>
Thu, 2 Jul 2015 14:46:00 +0000 (08:46 -0600)
commita20135ffbc44545596f9b99c970de097fb497bdd
tree18d7267ad1c0b0d603b4ca9cb1e43755c344f632
parenta13f35e8714009145e32ebe2bf25b84e1376e314
writeback: don't drain bdi_writeback_congested on bdi destruction

52ebea749aae ("writeback: make backing_dev_info host cgroup-specific
bdi_writebacks") made bdi (backing_dev_info) host per-cgroup wb's
(bdi_writeback's).  As the congested state needs to be per-wb and
referenced from blkcg side and multiple wbs, the patch made all
non-root cong's (bdi_writeback_congested's) reference counted and
indexed on bdi.

When a bdi is destroyed, cgwb_bdi_destroy() tries to drain all
non-root cong's; however, this can hang indefinitely because wb's can
also be referenced from blkcg_gq's which are destroyed after bdi
destruction is complete.

This patch fixes the bug by updating bdi destruction to not wait for
cong's to drain.  A cong is unlinked from bdi->cgwb_congested_tree on
bdi destuction regardless of its reference count as the bdi may go
away any point after destruction.  wb_congested_put() checks whether
the cong is already unlinked on release.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Jon Christopherson <jon@jons.org>
Link: https://bugzilla.kernel.org/show_bug.cgi?id=100681
Fixes: 52ebea749aae ("writeback: make backing_dev_info host cgroup-specific bdi_writebacks")
Tested-by: Jon Christopherson <jon@jons.org>
Signed-off-by: Jens Axboe <axboe@fb.com>
mm/backing-dev.c