From: Andy Gospodarek <gospo@redhat.com>
Date: Tue, 12 Jan 2010 05:46:54 -0500
Subject: [net] fixup problems with vlans and bonding
Message-id: <20100112054653.GR10827@gospo.rdu.redhat.com>
Patchwork-id: 22425
O-Subject: [RHEL5.5 PATCH] net: fixup problems with vlans and bonding
Bugzilla: 526976
RH-Acked-by: David S. Miller <davem@redhat.com>
RH-Acked-by: John Linville <linville@redhat.com>
RH-Acked-by: Jiri Pirko <jpirko@redhat.com>

The following RHEL5 patch:

	commit 0ee654bf5e4713688d1bdf7f3d39e30883dd39bc
	Author: Andy Gospodarek <gospo@redhat.com>
	Date:   Mon Nov 30 20:22:51 2009 -0500

	    [net] bonding: allow arp_ip_targets on separate vlan from bond device

created some problems with bonding and vlans as well as created some
undesirable behavior when trying to capture frames when using bonding.
Though it was a reasonable backport of the code posted upsream, that code
was not accepted and the following two patches were instead:

	commit ca8d9ea30bc79b2965a1d169dcb2f48f02af4d2d
	Author: Andy Gospodarek <andy@greyhouse.net>

	    fix bonding: allow arp_ip_targets on separate vlans to use arp validation

	commit 1f3c8804acba841b5573b953f5560d2683d2db0d
	Author: Andy Gospodarek <andy@greyhouse.net>

	    bonding: allow arp_ip_targets on separate vlans to use arp validation

I have verified this bug fixes the original request in RHBZ 526976 and
also fixes the problems introduced by the original patch.


diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
index f708fd7..d4d7815 100644
--- a/drivers/net/bonding/bond_main.c
+++ b/drivers/net/bonding/bond_main.c
@@ -2778,8 +2778,16 @@ static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct pack
 	unsigned char *arp_ptr;
 	__be32 sip, tip;
 
-	while (dev->priv_flags & IFF_802_1Q_VLAN)
+	if (dev->priv_flags & IFF_802_1Q_VLAN) {
+		/*
+		 * When using VLANS and bonding, dev and oriv_dev may be
+		 * incorrect if the physical interface supports VLAN
+		 * acceleration.  With this change ARP validation now
+		 * works for hosts only reachable on the VLAN interface.
+		 */
 		dev = VLAN_DEV_INFO(dev)->real_dev;
+		orig_dev = skb->input_dev;
+	}
 
 	if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
 		goto out;
diff --git a/net/core/dev.c b/net/core/dev.c
index e9a472f..7003c23 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -1876,6 +1876,7 @@ int netif_receive_skb(struct sk_buff *skb)
 	struct packet_type *ptype, *pt_prev;
 	struct net_device *orig_dev;
 	struct net_device *null_or_orig;
+	struct net_device *null_or_bond;
 	int ret = NET_RX_DROP;
 	unsigned short type;
 
@@ -1892,8 +1893,8 @@ int netif_receive_skb(struct sk_buff *skb)
 		skb->input_dev = skb->dev;
 
 	null_or_orig = NULL;
-	orig_dev = skb->input_dev;
-	if (orig_dev->master && !(skb->dev->priv_flags & IFF_802_1Q_VLAN)) {
+	orig_dev = skb->dev;
+	if (orig_dev->master) {
 		if (skb_bond_should_drop(skb))
 			null_or_orig = orig_dev; /* deliver only exact match */
 		else
@@ -1964,11 +1965,32 @@ ncls:
 
 	skb_orphan(skb);
 
+	/* Due to kABI constraints, we have to diverge from upstream.
+	 * The code upstream checks to see if this is a vlan device
+	 * and then looks to see if the real_dev has the bonding flags
+	 * set.  If it does, null_or_bond becomes the bonding device.
+	 * Since we cannot check for real_dev, we will look at input_dev
+	 * and determine if input-dev->master is set.  This would also
+	 * indicate bonding was used.
+	 *
+	 * For hw-accelerated vlan devices this will be fine, as
+	 * input_dev is now set to the true input dev.
+	 *
+	 * For legacy devices this will be fine as well since input_dev
+	 * is set above when the tagged frame first runs through
+	 * netif_receive_skb.
+	 */
+	null_or_bond = NULL;
+	if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
+		(skb->input_dev->master)) {
+		null_or_bond = skb->input_dev->master;
+	}
+
 	type = skb->protocol;
 	list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
-		if (ptype->type == type &&
-		    (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
-		     ptype->dev == orig_dev || ptype->dev == orig_dev->master)) {
+		if (ptype->type == type && (ptype->dev == null_or_orig ||
+		    ptype->dev == skb->dev || ptype->dev == orig_dev ||
+		    ptype->dev == null_or_bond)) {
 			if (pt_prev) 
 				ret = deliver_skb(skb, pt_prev, orig_dev);
 			pt_prev = ptype;