signal: don't allow STOP on PF_IO_WORKER threads
authorEric W. Biederman <ebiederm@xmission.com>
Sun, 21 Mar 2021 15:37:48 +0000 (09:37 -0600)
committerJens Axboe <axboe@kernel.dk>
Sun, 21 Mar 2021 15:41:07 +0000 (09:41 -0600)
Just like we don't allow normal signals to IO threads, don't deliver a
STOP to a task that has PF_IO_WORKER set. The IO threads don't take
signals in general, and have no means of flushing out a stop either.

Longer term, we may want to look into allowing stop of these threads,
as it relates to eg process freezing. For now, this prevents a spin
issue if a SIGSTOP is delivered to the parent task.

Reported-by: Stefan Metzmacher <metze@samba.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
kernel/signal.c

index 11cabcf20e7a6b18fb8917c59b74b305862a9761..f2a1b898da29fcb5adc6cc5732f53150b25d51b3 100644 (file)
@@ -288,7 +288,8 @@ bool task_set_jobctl_pending(struct task_struct *task, unsigned long mask)
                        JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
        BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
 
-       if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
+       if (unlikely(fatal_signal_pending(task) ||
+                    (task->flags & (PF_EXITING | PF_IO_WORKER))))
                return false;
 
        if (mask & JOBCTL_STOP_SIGMASK)