Hi,
On 2017/9/9 1:42, Brian Foster wrote:
> On Thu, Sep 07, 2017 at 04:04:57PM +0800, Hou Tao wrote:
>> A umount hang is possible when a race occurs between the umount
>> process and the xfsaild kthread. The following sequences outline
>> the race:
>>
>> xfsaild: kthread_should_stop()
>> => return false, so xfsaild continue
>>
>> umount: set_bit(KTHREAD_SHOULD_STOP, &kthread->flags)
>> => by kthread_stop()
>> umount: wake_up_process()
>> => because xfsaild is still running, so 0 is returned
>>
>> xfsaild: __set_current_state(TASK_INTERRUPTIBLE)
>> xfsaild: schedule()
>> => now, xfsaild will wait indefinitely
>>
>> umount: wait_for_completion()
>> => and umount will hang
>>
>> To fix that, we need to check kthread_should_stop() after we set
>> the task state, so the xfsaild will either see the stop bit and
>> exit or the task state is reset to runnable by wake_up_process()
>> such that it isn't blocked indefinitely and detects the stop bit
>> at the next iteration.
>>
>> Signed-off-by: Hou Tao <houtao1@xxxxxxxxxx>
>> ---
>
> I assume you've verified this against your local reproducer? Otherwise
> just a nit on the comment...
Yes, the patch fixes the test case with the delay hacks. After applying the
patch, I also have tried to move the artificial delay in xfsaild down to the
next of kthread_should_stop() and there is no hang of umount.
>> fs/xfs/xfs_trans_ail.c | 17 ++++++++++++++---
>> 1 file changed, 14 insertions(+), 3 deletions(-)
>>
>> diff --git a/fs/xfs/xfs_trans_ail.c b/fs/xfs/xfs_trans_ail.c
>> index 9056c0f..cd6e185 100644
>> --- a/fs/xfs/xfs_trans_ail.c
>> +++ b/fs/xfs/xfs_trans_ail.c
>> @@ -499,11 +499,22 @@ xfsaild(
>> current->flags |= PF_MEMALLOC;
>> set_freezable();
>>
>> - while (!kthread_should_stop()) {
>> + while (1) {
>> if (tout && tout <= 20)
>> - __set_current_state(TASK_KILLABLE);
>> + set_current_state(TASK_KILLABLE);
>> else
>> - __set_current_state(TASK_INTERRUPTIBLE);
>> + set_current_state(TASK_INTERRUPTIBLE);
>> +
>> + /*
>> + * Check kthread_should_stop() after we set the task state
>> + * to guarantee that we either see the stop bit and exit or the
>> + * task state is reset to runnable such that it's not blocked
>> + * indefinitely and detects the stop bit at the next iteration.
>> + */
>
> I'd change the "it's not blocked indefinitely" wording to something like
> "it's not scheduled out indefinitely." Also, a mention that the task
> state sets above include a memory barrier to serialize against
> kthread_stop() couldn't hurt. Otherwise this looks fine to me:
>
> Reviewed-by: Brian Foster <bfoster@xxxxxxxxxx>
Thanks for your comments. I will send v2 soon.
Regards,
Tao
--
To unsubscribe from this list: send the line "unsubscribe linux-xfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
