[openib-general] Re: IPoIB question/problem

[Michael S. Tsirkin]

How does the following strike you?
I didnt notice any performance impact - could more people test this please?

---

Assuming that a remote node is replaced and its address changes
(e.g. gid change), it seems that the ha field will gets out of sync with
the address handle stored in ipoib_neigh->ah, with the result that the
ah field would point to an incorrect path, resulting in all packets
being lost.

Signed-off-by: Michael S. Tsirkin <mst at mellanox.co.il>

Index: linux-2.6.14/drivers/infiniband/ulp/ipoib/ipoib_main.c
===================================================================
--- linux-2.6.14.orig/drivers/infiniband/ulp/ipoib/ipoib_main.c	2005-11-14 18:29:40.000000000 +0200
+++ linux-2.6.14/drivers/infiniband/ulp/ipoib/ipoib_main.c	2005-11-14 20:26:43.000000000 +0200
@@ -394,6 +394,7 @@ static void path_rec_completion(int stat
 		list_for_each_entry(neigh, &path->neigh_list, list) {
 			kref_get(&path->ah->ref);
 			neigh->ah = path->ah;
+			memcpy(neigh->dgid.raw, path->pathrec.dgid.raw, sizeof (union ib_gid));
 
 			while ((skb = __skb_dequeue(&neigh->queue)))
 				__skb_queue_tail(&skqueue, skb);
@@ -503,6 +504,7 @@ static void neigh_add_path(struct sk_buf
 	if (path->pathrec.dlid) {
 		kref_get(&path->ah->ref);
 		neigh->ah = path->ah;
+		memcpy(neigh->dgid.raw, path->pathrec.dgid.raw, sizeof (union ib_gid));
 
 		ipoib_send(dev, skb, path->ah,
 			   be32_to_cpup((__be32 *) skb->dst->neighbour->ha));
@@ -633,6 +635,17 @@ static int ipoib_start_xmit(struct sk_bu
 		neigh = *to_ipoib_neigh(skb->dst->neighbour);
 
 		if (likely(neigh->ah)) {
+			if (unlikely(memcmp(neigh->dgid.raw,
+					    skb->dst->neighbour->ha + 4,
+					    sizeof (union ib_gid)))) {
+				ipoib_put_ah(neigh->ah);
+				*to_ipoib_neigh(skb->dst->neighbour) = NULL;
+				skb->dst->neighbour->ops->destructor = NULL;
+				list_del(&neigh->list);
+				kfree(neigh);
+				ipoib_path_lookup(skb, dev);
+				goto out;
+			}
 			ipoib_send(dev, skb, neigh->ah,
 				   be32_to_cpup((__be32 *) skb->dst->neighbour->ha));
 			goto out;
Index: linux-2.6.14/drivers/infiniband/ulp/ipoib/ipoib.h
===================================================================
--- linux-2.6.14.orig/drivers/infiniband/ulp/ipoib/ipoib.h	2005-11-14 18:29:40.000000000 +0200
+++ linux-2.6.14/drivers/infiniband/ulp/ipoib/ipoib.h	2005-11-14 18:29:42.000000000 +0200
@@ -209,6 +209,7 @@ struct ipoib_path {
 
 struct ipoib_neigh {
 	struct ipoib_ah    *ah;
+	union ib_gid        dgid;
 	struct sk_buff_head queue;
 
 	struct neighbour   *neighbour;
-- 
MST