在 2022/3/14 16:06, Greg Kroah-Hartman 写道:
> On Tue, Mar 08, 2022 at 04:12:32PM +0100, Michal Koutný wrote:
>> Hello.
>>
>> On Mon, Feb 28, 2022 at 06:24:07PM +0100, Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
>>> [...]
>>> cpuset_attach() cpu hotplug
>>> --------------------------- ----------------------
>>> down_write(cpuset_rwsem)
>>> guarantee_online_cpus() // (load cpus_attach)
>>> sched_cpu_deactivate
>>> set_cpu_active()
>>> // will change cpu_active_mask
>>> set_cpus_allowed_ptr(cpus_attach)
>>> __set_cpus_allowed_ptr_locked()
>>> // (if the intersection of cpus_attach and
>>> cpu_active_mask is empty, will return -EINVAL)
>>> up_write(cpuset_rwsem)
>>> [...]
>>> --- a/kernel/cgroup/cpuset.c
>>> +++ b/kernel/cgroup/cpuset.c
>>> @@ -1528,6 +1528,7 @@ static void cpuset_attach(struct cgroup_
>>> cgroup_taskset_first(tset, &css);
>>> cs = css_cs(css);
>>>
>>> + cpus_read_lock();
>>> mutex_lock(&cpuset_mutex);
>>
>> This backport (and possible older kernels) looks suspicious since it comes
>> before commit d74b27d63a8b ("cgroup/cpuset: Change cpuset_rwsem and
>> hotplug lock order") v5.4-rc1~176^2~30 when the locking order was:
>> cpuset lock, cpus lock.
>>
>> At the same time it also comes before commit 710da3c8ea7d ("sched/core:
>> Prevent race condition between cpuset and __sched_setscheduler()")
>> v5.4-rc1~176^2~27 when neither __sched_setscheduler() cared and this
>> race is similar. (The swapped locking may still conflict with
>> rebuild_sched_domains() before d74b27d63a8b.)
>
> Thanks for noticing this. What do you recommend to do to resolve this?
>
> thanks,
>
hi, Please review the following patch to fix it.
thanks.
patch:
[PATCH] cpuset: Fix unsafe lock order between cpuset lock and cpus lock
The backport commit 4eec5fe1c680a ("cgroup/cpuset: Fix a race
between cpuset_attach() and cpu hotplug") looks suspicious since
it comes before commit d74b27d63a8b ("cgroup/cpuset: Change
cpuset_rwsem and hotplug lock order") v5.4-rc1~176^2~30 when
the locking order was: cpuset lock, cpus lock.
Fix it with the correct locking order and reduce the cpus locking
range because only set_cpus_allowed_ptr() needs the protection of
cpus lock.
Fixes: 4eec5fe1c680a ("cgroup/cpuset: Fix a race between cpuset_attach() and cpu hotplug")
Reported-by: Michal Koutný <mkoutny@xxxxxxxx>
Signed-off-by: Zhang Qiao <zhangqiao22@xxxxxxxxxx>
---
kernel/cgroup/cpuset.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index d43d25acc..4e1c4232e 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -1528,9 +1528,13 @@ static void cpuset_attach(struct cgroup_taskset *tset)
cgroup_taskset_first(tset, &css);
cs = css_cs(css);
- cpus_read_lock();
mutex_lock(&cpuset_mutex);
+ /*
+ * It should hold cpus lock because a cpu offline event can
+ * cause set_cpus_allowed_ptr() failed.
+ */
+ cpus_read_lock();
/* prepare for attach */
if (cs == &top_cpuset)
cpumask_copy(cpus_attach, cpu_possible_mask);
@@ -1549,6 +1553,7 @@ static void cpuset_attach(struct cgroup_taskset *tset)
cpuset_change_task_nodemask(task, &cpuset_attach_nodemask_to);
cpuset_update_task_spread_flag(cs, task);
}
+ cpus_read_unlock();
/*
* Change mm for all threadgroup leaders. This is expensive and may
@@ -1584,7 +1589,6 @@ static void cpuset_attach(struct cgroup_taskset *tset)
wake_up(&cpuset_attach_wq);
mutex_unlock(&cpuset_mutex);
- cpus_read_unlock();
}
/* The various types of files and directories in a cpuset file system */
--
2.18.0
> greg k-h
> .
>
