gianfar: fix BUG under load after introduction of skb recycling
authorLennert Buytenhek <buytenh@wantstofly.org>
Mon, 25 May 2009 07:42:34 +0000 (00:42 -0700)
committerDavid S. Miller <davem@davemloft.net>
Mon, 25 May 2009 07:42:34 +0000 (00:42 -0700)
Since commit 0fd56bb5be6455d0d42241e65aed057244665e5e ("gianfar:
Add support for skb recycling"), gianfar puts skbuffs that are in
the rx ring back onto the recycle list as-is in case there was a
receive error, but this breaks the following invariant: that all
skbuffs on the recycle list have skb->data = skb->head + NET_SKB_PAD.

The RXBUF_ALIGNMENT realignment done in gfar_new_skb() will be done
twice on skbuffs recycled in this way, causing there not to be enough
room in the skb anymore to receive a full packet, eventually leading
to an skb_over_panic from gfar_clean_rx_ring() -> skb_put().

Resetting the skb->data pointer to skb->head + NET_SKB_PAD before
putting the skb back onto the recycle list restores the mentioned
invariant, and should fix this issue.

Reported-by: Michael Guntsche <mike@it-loops.com>
Tested-by: Michael Guntsche <mike@it-loops.com>
Signed-off-by: Lennert Buytenhek <buytenh@wantstofly.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/gianfar.c

index b2c4967..a051918 100644 (file)
@@ -1885,8 +1885,17 @@ int gfar_clean_rx_ring(struct net_device *dev, int rx_work_limit)
 
                        if (unlikely(!newskb))
                                newskb = skb;
-                       else if (skb)
+                       else if (skb) {
+                               /*
+                                * We need to reset ->data to what it
+                                * was before gfar_new_skb() re-aligned
+                                * it to an RXBUF_ALIGNMENT boundary
+                                * before we put the skb back on the
+                                * recycle list.
+                                */
+                               skb->data = skb->head + NET_SKB_PAD;
                                __skb_queue_head(&priv->rx_recycle, skb);
+                       }
                } else {
                        /* Increment the number of packets */
                        dev->stats.rx_packets++;