[NEIGH] Fix add_timer race in neigh_add_timer
authorHerbert Xu <herbert@gondor.apana.org.au>
Sun, 23 Oct 2005 06:37:48 +0000 (16:37 +1000)
committerHerbert Xu <herbert@gondor.apana.org.au>
Sun, 23 Oct 2005 06:37:48 +0000 (16:37 +1000)
neigh_add_timer cannot use add_timer unconditionally.  The reason is that
by the time it has obtained the write lock someone else (e.g., neigh_update)
could have already added a new timer.

So it should only use mod_timer and deal with its return value accordingly.

This bug would have led to rare neighbour cache entry leaks.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
net/core/neighbour.c

index 766caa0dd9305be10468da8af20f72acb1f6922f..37d8d8c295226b214c95de9134317391886e387e 100644 (file)
@@ -816,10 +816,10 @@ static void neigh_timer_handler(unsigned long arg)
        }
 
        if (neigh->nud_state & NUD_IN_TIMER) {
-               neigh_hold(neigh);
                if (time_before(next, jiffies + HZ/2))
                        next = jiffies + HZ/2;
-               neigh_add_timer(neigh, next);
+               if (!mod_timer(&neigh->timer, next))
+                       neigh_hold(neigh);
        }
        if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
                struct sk_buff *skb = skb_peek(&neigh->arp_queue);