Re: [PATCH] cgroup/freezer: walk exiting tasks in the v1 freezer
From: Tao Cui
Date: Wed Nov 04 2026 - 08:26:05 EST
Hi Mani,
在 2026/11/4 02:09, Mani Tofigh via B4 Relay 写道:
> From: Mani Tofigh <mt4022@xxxxxxxxxxxx>
>
> An exiting task can stay frozen after its v1 freezer cgroup is thawed.
> Since commit a72f73c4dd9b ("cgroup: Don't expose dead tasks in cgroup")
> the css task walk hides an exiting task once its process has no live
> threads left. That count drops when another thread of the process
> exits, which can happen between the freeze and the thaw walk. A pid
> namespace init hits this easily since it waits freezably in
> zap_pid_ns_processes() after commit 81933038f5e2 ("pid_namespace: make
> zap_pid_ns_processes() freezable"):
>
> Task A (pid ns init) Task B (A's thread) Task C
> do_exit()
> exit_signals()
> do_exit()
> zap_pid_ns_processes()
> write FROZEN
> /* A frozen */
> /* live count 0 */
> write THAWED
> /* A skipped */
>
> A then stays frozen even after SIGKILL:
>
> [ 10.775155] task:fz-init state:D stack:28152 pid:65 tgid:65 ppid:1 task_flags:0xc0004c flags:0x00080000
> [ 10.775176] Call Trace:
> [ 10.775180] <TASK>
> [ 10.775188] __schedule+0x10f1/0x4760
> [ 10.775252] schedule+0xd7/0x2c0
> [ 10.775265] zap_pid_ns_processes+0x2c7/0x5f0
> [ 10.775346] do_exit+0x1bf4/0x2560
> [ 10.775403] __x64_sys_exit+0x3d/0x50
> [ 10.775414] x64_sys_call+0x156c/0x1570
> [ 10.775425] do_syscall_64+0xfb/0x590
> [ 10.775450] entry_SYSCALL_64_after_hwframe+0x77/0x7f
> [ 10.775529] </TASK>
>
> Use CSS_TASK_ITER_WITH_DEAD in the three v1 freezer walks so they all
> see the same tasks, as commit 60f21a264930 ("cgroup, sched_ext: Include
> exiting tasks in cgroup iter") did for sched_ext. With the fix, A is
> thawed and exits.
>
Thanks for the clear writeup and reproducer. Ran freezev1.c against
next-20261103 on a 2-vCPU VM. Base reproduced the hang on round 1
(5.1s), with the patch 600s / 6124 rounds ran clean.
Reviewed-by: Tao Cui <cuitao@xxxxxxxxxx>
Tested-by: Tao Cui <cuitao@xxxxxxxxxx>
Thanks,
Tao
> This bug was identified and confirmed with the help of a deterministic
> hypervisor currently under development. The C reproducer below however
> hits it within about 10s.
>
> Fixes: a72f73c4dd9b ("cgroup: Don't expose dead tasks in cgroup")
> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: Mani Tofigh <mt4022@xxxxxxxxxxxx>
> ---
> Reproducer: https://github.com/manitofigh/files/blob/main/linux-repros/freezev1.c
> ---
> kernel/cgroup/legacy_freezer.c | 6 +++---
> 1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/kernel/cgroup/legacy_freezer.c b/kernel/cgroup/legacy_freezer.c
> index 8545e0d1ba3d..bd11047f2311 100644
> --- a/kernel/cgroup/legacy_freezer.c
> +++ b/kernel/cgroup/legacy_freezer.c
> @@ -265,7 +265,7 @@ static void update_if_frozen(struct cgroup_subsys_state *css)
> rcu_read_unlock();
>
> /* are all tasks frozen? */
> - css_task_iter_start(css, 0, &it);
> + css_task_iter_start(css, CSS_TASK_ITER_WITH_DEAD, &it);
>
> while ((task = css_task_iter_next(&it))) {
> if (freezing(task) && !frozen(task))
> @@ -309,7 +309,7 @@ static void freeze_cgroup(struct freezer *freezer)
> struct css_task_iter it;
> struct task_struct *task;
>
> - css_task_iter_start(&freezer->css, 0, &it);
> + css_task_iter_start(&freezer->css, CSS_TASK_ITER_WITH_DEAD, &it);
> while ((task = css_task_iter_next(&it)))
> freeze_task(task);
> css_task_iter_end(&it);
> @@ -320,7 +320,7 @@ static void unfreeze_cgroup(struct freezer *freezer)
> struct css_task_iter it;
> struct task_struct *task;
>
> - css_task_iter_start(&freezer->css, 0, &it);
> + css_task_iter_start(&freezer->css, CSS_TASK_ITER_WITH_DEAD, &it);
> while ((task = css_task_iter_next(&it)))
> __thaw_task(task);
> css_task_iter_end(&it);
>
> ---
> base-commit: fffbdad76a57eed26cff1be19f4cbb407f0b6e7c
> change-id: 20261102-cgroup-freezer-exiting-c7f60e4c3200
>
> Best regards,
> --
> Mani Tofigh <mt4022@xxxxxxxxxxxx>