[PATCH linux-next] cgroup: fix suspicious rcu_dereference_check() usage warning

[PATCH linux-next] cgroup: fix suspicious rcu_dereference_check() usage warning

[Chengming Zhou]

task_css_set_check() will use rcu_dereference_check() to check for
rcu_read_lock_held() on the read-side, which is not true after commit
dc6e0818bc9a ("sched/cpuacct: Optimize away RCU read lock"). This
commit drop explicit rcu_read_lock(), change to RCU-sched read-side
critical section. So fix the RCU warning by adding check for
rcu_read_lock_sched_held().

Fixes: dc6e0818bc9a ("sched/cpuacct: Optimize away RCU read lock")
Reported-by: Linux Kernel Functional Testing <lkft-QSEj5FYQhm4dnm+yROfE0A@public.gmane.org>
Reported-by: syzbot+16e3f2c77e7c5a0113f9-Pl5Pbv+GP7P466ipTTIvnc23WoclnBCfAL8bYrjMMd8@public.gmane.org
Tested-by: Zhouyi Zhou <zhouzhouyi-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
Signed-off-by: Chengming Zhou <zhouchengming-EC8Uxl6Npydl57MIdRCFDg@public.gmane.org>
---
 include/linux/cgroup.h | 1 +
 1 file changed, 1 insertion(+)

diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h
index 1e356c222756..0d1ada8968d7 100644
--- a/include/linux/cgroup.h
+++ b/include/linux/cgroup.h
@@ -450,6 +450,7 @@ extern struct mutex cgroup_mutex;
 extern spinlock_t css_set_lock;
 #define task_css_set_check(task, __c)					\
 	rcu_dereference_check((task)->cgroups,				\
+		rcu_read_lock_sched_held() ||				\
 		lockdep_is_held(&cgroup_mutex) ||			\
 		lockdep_is_held(&css_set_lock) ||			\
 		((task)->flags & PF_EXITING) || (__c))
-- 
2.20.1

Re: [PATCH linux-next] cgroup: fix suspicious rcu_dereference_check() usage warning

[Marek Szyprowski]

On 05.03.2022 04:41, Chengming Zhou wrote:
> task_css_set_check() will use rcu_dereference_check() to check for
> rcu_read_lock_held() on the read-side, which is not true after commit
> dc6e0818bc9a ("sched/cpuacct: Optimize away RCU read lock"). This
> commit drop explicit rcu_read_lock(), change to RCU-sched read-side
> critical section. So fix the RCU warning by adding check for
> rcu_read_lock_sched_held().
>
> Fixes: dc6e0818bc9a ("sched/cpuacct: Optimize away RCU read lock")
> Reported-by: Linux Kernel Functional Testing <lkft-QSEj5FYQhm4dnm+yROfE0A@public.gmane.org>
> Reported-by: syzbot+16e3f2c77e7c5a0113f9-Pl5Pbv+GP7P466ipTTIvnc23WoclnBCfAL8bYrjMMd8@public.gmane.org
> Tested-by: Zhouyi Zhou <zhouzhouyi-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
> Signed-off-by: Chengming Zhou <zhouchengming-EC8Uxl6Npydl57MIdRCFDg@public.gmane.org>

Right, this fixes the issue I've reported here:

https://lore.kernel.org/all/f4bc652b-115f-35b5-91db-bad3b30fed9b-Sze3O3UU22JBDgjK7y7TUQ@public.gmane.org/

Tested-by: Marek Szyprowski <m.szyprowski-Sze3O3UU22JBDgjK7y7TUQ@public.gmane.org>

> ---
>   include/linux/cgroup.h | 1 +
>   1 file changed, 1 insertion(+)
>
> diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h
> index 1e356c222756..0d1ada8968d7 100644
> --- a/include/linux/cgroup.h
> +++ b/include/linux/cgroup.h
> @@ -450,6 +450,7 @@ extern struct mutex cgroup_mutex;
>   extern spinlock_t css_set_lock;
>   #define task_css_set_check(task, __c)					\
>   	rcu_dereference_check((task)->cgroups,				\
> +		rcu_read_lock_sched_held() ||				\
>   		lockdep_is_held(&cgroup_mutex) ||			\
>   		lockdep_is_held(&css_set_lock) ||			\
>   		((task)->flags & PF_EXITING) || (__c))

Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland

Re: [PATCH linux-next] cgroup: fix suspicious rcu_dereference_check() usage warning

[Tejun Heo]

On Sat, Mar 05, 2022 at 11:41:03AM +0800, Chengming Zhou wrote:
> task_css_set_check() will use rcu_dereference_check() to check for
> rcu_read_lock_held() on the read-side, which is not true after commit
> dc6e0818bc9a ("sched/cpuacct: Optimize away RCU read lock"). This
> commit drop explicit rcu_read_lock(), change to RCU-sched read-side
> critical section. So fix the RCU warning by adding check for
> rcu_read_lock_sched_held().
> 
> Fixes: dc6e0818bc9a ("sched/cpuacct: Optimize away RCU read lock")
> Reported-by: Linux Kernel Functional Testing <lkft@linaro.org>
> Reported-by: syzbot+16e3f2c77e7c5a0113f9@syzkaller.appspotmail.com
> Tested-by: Zhouyi Zhou <zhouzhouyi@gmail.com>
> Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
> ---
>  include/linux/cgroup.h | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h
> index 1e356c222756..0d1ada8968d7 100644
> --- a/include/linux/cgroup.h
> +++ b/include/linux/cgroup.h
> @@ -450,6 +450,7 @@ extern struct mutex cgroup_mutex;
>  extern spinlock_t css_set_lock;
>  #define task_css_set_check(task, __c)					\
>  	rcu_dereference_check((task)->cgroups,				\
> +		rcu_read_lock_sched_held() ||				\
>  		lockdep_is_held(&cgroup_mutex) ||			\
>  		lockdep_is_held(&css_set_lock) ||			\
>  		((task)->flags & PF_EXITING) || (__c))

Acked-by: Tejun Heo <tj@kernel.org>

Thanks.

-- 
tejun

Re: [PATCH linux-next] cgroup: fix suspicious rcu_dereference_check() usage warning

[Peter Zijlstra]

On Sat, Mar 05, 2022 at 11:41:03AM +0800, Chengming Zhou wrote:
> task_css_set_check() will use rcu_dereference_check() to check for
> rcu_read_lock_held() on the read-side, which is not true after commit
> dc6e0818bc9a ("sched/cpuacct: Optimize away RCU read lock"). This
> commit drop explicit rcu_read_lock(), change to RCU-sched read-side
> critical section. So fix the RCU warning by adding check for
> rcu_read_lock_sched_held().
> 
> Fixes: dc6e0818bc9a ("sched/cpuacct: Optimize away RCU read lock")
> Reported-by: Linux Kernel Functional Testing <lkft@linaro.org>
> Reported-by: syzbot+16e3f2c77e7c5a0113f9@syzkaller.appspotmail.com
> Tested-by: Zhouyi Zhou <zhouzhouyi@gmail.com>
> Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>

Thanks, I'll go stick this in sched/core so it's in the same branch that
caused the problem.

Re: [PATCH linux-next] cgroup: fix suspicious rcu_dereference_check() usage warning

[Peter Zijlstra]

On Sat, Mar 12, 2022 at 01:19:13PM +0100, Peter Zijlstra wrote:
> On Sat, Mar 05, 2022 at 11:41:03AM +0800, Chengming Zhou wrote:
> > task_css_set_check() will use rcu_dereference_check() to check for
> > rcu_read_lock_held() on the read-side, which is not true after commit
> > dc6e0818bc9a ("sched/cpuacct: Optimize away RCU read lock"). This
> > commit drop explicit rcu_read_lock(), change to RCU-sched read-side
> > critical section. So fix the RCU warning by adding check for
> > rcu_read_lock_sched_held().
> > 
> > Fixes: dc6e0818bc9a ("sched/cpuacct: Optimize away RCU read lock")
> > Reported-by: Linux Kernel Functional Testing <lkft-QSEj5FYQhm4dnm+yROfE0A@public.gmane.org>
> > Reported-by: syzbot+16e3f2c77e7c5a0113f9-Pl5Pbv+GP7P466ipTTIvnc23WoclnBCfAL8bYrjMMd8@public.gmane.org
> > Tested-by: Zhouyi Zhou <zhouzhouyi-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
> > Signed-off-by: Chengming Zhou <zhouchengming-EC8Uxl6Npydl57MIdRCFDg@public.gmane.org>
> 
> Thanks, I'll go stick this in sched/core so it's in the same branch that
> caused the problem.

FWIW I never saw this patch because it doesn't instantly look like a
patch I should be interested in. It's classified as 'for-next' and I
don't run -next, sfr does that. Then it's tagged as cgroup, which I also
don't do.

Nowhere does that look like a patch that wants to go in sched/core and
fixes a cpuacct issue.

On top of that, I still don't agree with this, I really think
rcu_dereference_check() itself should be changed.

Re: [External] Re: [PATCH linux-next] cgroup: fix suspicious rcu_dereference_check() usage warning

[Chengming Zhou]

On 2022/3/12 9:34 下午, Peter Zijlstra wrote:
> On Sat, Mar 12, 2022 at 01:19:13PM +0100, Peter Zijlstra wrote:
>> On Sat, Mar 05, 2022 at 11:41:03AM +0800, Chengming Zhou wrote:
>>> task_css_set_check() will use rcu_dereference_check() to check for
>>> rcu_read_lock_held() on the read-side, which is not true after commit
>>> dc6e0818bc9a ("sched/cpuacct: Optimize away RCU read lock"). This
>>> commit drop explicit rcu_read_lock(), change to RCU-sched read-side
>>> critical section. So fix the RCU warning by adding check for
>>> rcu_read_lock_sched_held().
>>>
>>> Fixes: dc6e0818bc9a ("sched/cpuacct: Optimize away RCU read lock")
>>> Reported-by: Linux Kernel Functional Testing <lkft@linaro.org>
>>> Reported-by: syzbot+16e3f2c77e7c5a0113f9@syzkaller.appspotmail.com
>>> Tested-by: Zhouyi Zhou <zhouzhouyi@gmail.com>
>>> Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
>>
>> Thanks, I'll go stick this in sched/core so it's in the same branch that
>> caused the problem.
> 
> FWIW I never saw this patch because it doesn't instantly look like a
> patch I should be interested in. It's classified as 'for-next' and I
> don't run -next, sfr does that. Then it's tagged as cgroup, which I also
> don't do.

Oh, sorry for this.. I should've add "cpuacct" in the subject. The
"linux-next" prefix was added because I thought any patch based on
the linux-next branch should add this prefix.

> 
> Nowhere does that look like a patch that wants to go in sched/core and
> fixes a cpuacct issue.
> 
> On top of that, I still don't agree with this, I really think
> rcu_dereference_check() itself should be changed.

Yes, I think so too. This patch is workaround to fix the warning to
follow the usage in RCU Documentation.

Maybe changes should be made in RCU code to make rcu_dereference_check()
more flexible as you expressed in the conversation with Paul.

Thanks.