Status goes in ret, before fix status remained -ENOMEM
and was returned from dev_queue_xmit, that is not desired
behaviour - we want to drop packet silently and say
NETDEV_TX_OK(that it was sent).

Return NULL, before the fix we were very lucky that in
dev_queue_xmit, we do not get to kfree_skb_list(-ENOMEM),
as dev_xmit_complete(-ENOMEM) is true =)

In both sch_direct_xmit and dev_queue_xmit we will
never use returned value of dev_hard_start_xmit in
case of fencing so it is safe to return NULL here.

Signed-off-by: Pavel Tikhomirov <ptikhomi...@virtuozzo.com>
---
 net/core/dev.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index be92f4f..0137947 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2559,7 +2559,8 @@ struct sk_buff *dev_hard_start_xmit(struct sk_buff 
*first, struct net_device *de
 #ifdef CONFIG_FENCE_WATCHDOG
        if (unlikely(fence_wdog_check_timer())) {
                kfree_skb(skb);
-               return NETDEV_TX_OK;
+               *ret = rc;
+               return NULL;
        }
 #endif
 
-- 
2.5.5

_______________________________________________
Devel mailing list
Devel@openvz.org
https://lists.openvz.org/mailman/listinfo/devel

Reply via email to