[PATCH] send_sigqueue: simplify and fix the race
Oleg Nesterov [Wed, 29 Mar 2006 00:11:30 +0000 (16:11 -0800)]
send_sigqueue() checks PF_EXITING, then locks p->sighand->siglock.  This is
unsafe: 'p' can exit in between and set ->sighand = NULL.  The race is
theoretical, the window is tiny and irqs are disabled by the caller, so I
don't think we need the fix for -stable tree.

Convert send_sigqueue() to use lock_task_sighand() helper.

Also, delete 'p->flags & PF_EXITING' re-check, it is unneeded and the
comment is wrong.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

kernel/signal.c

index 0528a95..4922928 100644 (file)
@@ -1309,12 +1309,10 @@ void sigqueue_free(struct sigqueue *q)
        __sigqueue_free(q);
 }
 
-int
-send_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
+int send_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
 {
        unsigned long flags;
        int ret = 0;
-       struct sighand_struct *sh;
 
        BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
 
@@ -1328,48 +1326,17 @@ send_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
         */
        rcu_read_lock();
 
-       if (unlikely(p->flags & PF_EXITING)) {
+       if (!likely(lock_task_sighand(p, &flags))) {
                ret = -1;
                goto out_err;
        }
 
-retry:
-       sh = rcu_dereference(p->sighand);
-
-       spin_lock_irqsave(&sh->siglock, flags);
-       if (p->sighand != sh) {
-               /* We raced with exec() in a multithreaded process... */
-               spin_unlock_irqrestore(&sh->siglock, flags);
-               goto retry;
-       }
-
-       /*
-        * We do the check here again to handle the following scenario:
-        *
-        * CPU 0                CPU 1
-        * send_sigqueue
-        * check PF_EXITING
-        * interrupt            exit code running
-        *                      __exit_signal
-        *                      lock sighand->siglock
-        *                      unlock sighand->siglock
-        * lock sh->siglock
-        * add(tsk->pending)    flush_sigqueue(tsk->pending)
-        *
-        */
-
-       if (unlikely(p->flags & PF_EXITING)) {
-               ret = -1;
-               goto out;
-       }
-
        if (unlikely(!list_empty(&q->list))) {
                /*
                 * If an SI_TIMER entry is already queue just increment
                 * the overrun count.
                 */
-               if (q->info.si_code != SI_TIMER)
-                       BUG();
+               BUG_ON(q->info.si_code != SI_TIMER);
                q->info.si_overrun++;
                goto out;
        }
@@ -1385,7 +1352,7 @@ retry:
                signal_wake_up(p, sig == SIGKILL);
 
 out:
-       spin_unlock_irqrestore(&sh->siglock, flags);
+       unlock_task_sighand(p, &flags);
 out_err:
        rcu_read_unlock();