cgroup_freezer: cgroup_freezing: Check if not frozen
authorTim Van Patten <timvp@google.com>
Wed, 15 Nov 2023 16:20:43 +0000 (09:20 -0700)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 13 Dec 2023 17:39:19 +0000 (18:39 +0100)
commit4c00f01832f27eb90e3e5a803a91ac28049d281f
tree41dfc2638d0addb71a94828285447a9bf7847b1b
parent3907e9cc85ffc49b232dd0b0e8ebe12d3c835a2d
cgroup_freezer: cgroup_freezing: Check if not frozen

commit cff5f49d433fcd0063c8be7dd08fa5bf190c6c37 upstream.

__thaw_task() was recently updated to warn if the task being thawed was
part of a freezer cgroup that is still currently freezing:

void __thaw_task(struct task_struct *p)
{
...
if (WARN_ON_ONCE(freezing(p)))
goto unlock;

This has exposed a bug in cgroup1 freezing where when CGROUP_FROZEN is
asserted, the CGROUP_FREEZING bits are not also cleared at the same
time. Meaning, when a cgroup is marked FROZEN it continues to be marked
FREEZING as well. This causes the WARNING to trigger, because
cgroup_freezing() thinks the cgroup is still freezing.

There are two ways to fix this:

1. Whenever FROZEN is set, clear FREEZING for the cgroup and all
children cgroups.
2. Update cgroup_freezing() to also verify that FROZEN is not set.

This patch implements option (2), since it's smaller and more
straightforward.

Signed-off-by: Tim Van Patten <timvp@google.com>
Tested-by: Mark Hasemeyer <markhas@chromium.org>
Fixes: f5d39b020809 ("freezer,sched: Rewrite core freezer logic")
Cc: stable@vger.kernel.org # v6.1+
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
kernel/cgroup/legacy_freezer.c