neigh: do not modify unlinked entries
authorJulian Anastasov <ja@ssi.bg>
Tue, 16 Jun 2015 19:56:39 +0000 (22:56 +0300)
committerBen Hutchings <ben@decadent.org.uk>
Thu, 6 Aug 2015 23:32:17 +0000 (00:32 +0100)
[ Upstream commit 2c51a97f76d20ebf1f50fef908b986cb051fdff9 ]

The lockless lookups can return entry that is unlinked.
Sometimes they get reference before last neigh_cleanup_and_release,
sometimes they do not need reference. Later, any
modification attempts may result in the following problems:

1. entry is not destroyed immediately because neigh_update
can start the timer for dead entry, eg. on change to NUD_REACHABLE
state. As result, entry lives for some time but is invisible
and out of control.

2. __neigh_event_send can run in parallel with neigh_destroy
while refcnt=0 but if timer is started and expired refcnt can
reach 0 for second time leading to second neigh_destroy and
possible crash.

Thanks to Eric Dumazet and Ying Xue for their work and analyze
on the __neigh_event_send change.

Fixes: 767e97e1e0db ("neigh: RCU conversion of struct neighbour")
Fixes: a263b3093641 ("ipv4: Make neigh lookups directly in output packet path.")
Fixes: 6fd6ce2056de ("ipv6: Do not depend on rt->n in ip6_finish_output2().")
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Ying Xue <ying.xue@windriver.com>
Signed-off-by: Julian Anastasov <ja@ssi.bg>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
[bwh: Backported to 3.2: drop change to __neigh_set_probe_once() which
 we don't have]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
net/core/neighbour.c

index 0ea3fd3..c8c2645 100644 (file)
@@ -955,6 +955,8 @@ int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
        rc = 0;
        if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
                goto out_unlock_bh;
+       if (neigh->dead)
+               goto out_dead;
 
        if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
                if (neigh->parms->mcast_probes + neigh->parms->app_probes) {
@@ -1003,6 +1005,13 @@ out_unlock_bh:
                write_unlock(&neigh->lock);
        local_bh_enable();
        return rc;
+
+out_dead:
+       if (neigh->nud_state & NUD_STALE)
+               goto out_unlock_bh;
+       write_unlock_bh(&neigh->lock);
+       kfree_skb(skb);
+       return 1;
 }
 EXPORT_SYMBOL(__neigh_event_send);
 
@@ -1066,6 +1075,8 @@ int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
        if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
            (old & (NUD_NOARP | NUD_PERMANENT)))
                goto out;
+       if (neigh->dead)
+               goto out;
 
        if (!(new & NUD_VALID)) {
                neigh_del_timer(neigh);