Re: [PATCH v7 4/5] sched: Handle set_cpus_allowed_ptr() & sched_setaffinity() race

[Waiman Long <longman@redhat.com>]

On 9/8/22 07:53, Peter Zijlstra wrote:
> On Fri, Sep 02, 2022 at 11:25:55AM -0400, Waiman Long wrote:
>> Racing is possible between set_cpus_allowed_ptr() and sched_setaffinity()
>> or between multiple sched_setaffinity() calls from different
>> CPUs. To resolve these race conditions, we need to update both
>> user_cpus_ptr and cpus_mask in a single lock critical section instead
>> of separated ones. This requires moving the user_cpus_ptr update to
>> set_cpus_allowed_common().
>>
>> The SCA_USER flag will be used to signify that a user_cpus_ptr update
>> will have to be done. The new user_cpus_ptr will be put into the
>> a percpu variable pending_user_mask at the beginning of the lock
>> crtical section. The pending user mask will then be taken up in
>> set_cpus_allowed_common().
>>
>> Ideally, user_cpus_ptr should only be updated if the sched_setaffinity()
>> is successful. However, this patch will update user_cpus_ptr when the
>> first call to __set_cpus_allowed_ptr() is successful. However, if there
>> is racing between sched_setaffinity() and cpuset update, the subsequent
>> calls to __set_cpus_allowed_ptr() may fail but the user_cpus_ptr will
>> still be updated in this corner case. A warning will be printed in this
>> corner case.
>>
>> Signed-off-by: Waiman Long <longman@redhat.com>
> OK, so obviously this is terrible :/
>
> What's wrong with something like so ?

Thanks for the suggestion. I have no problem adding an affinity_context 
structure to pass around the functions. Will incorporate your suggestion 
in the next version of the patch.

Thanks,
Longman