Re: [patch net-next-2.6 V3] net: convert bonding to use rx_handler

Sun, Feb 27, 2011 at 03:17:01PM CET, nicolas.2p.debian@gmail.com wrote:

Le 23/02/2011 20:05, Jiri Pirko a écrit :

quoted

This patch converts bonding to use rx_handler. Results in cleaner
__netif_receive_skb() with much less exceptions needed. Also
bond-specific work is moved into bond code.

Did performance test using pktgen and counting incoming packets by
iptables. No regression noted.

Signed-off-by: Jiri Pirko<redacted>

v1->v2:
        using skb_iif instead of new input_dev to remember original
device

v2->v3:
do another loop in case skb->dev is changed. That way orig_dev
core can be left untouched.

Signed-off-by: Jiri Pirko<redacted>
---

[snip]

quoted

+static struct sk_buff *bond_handle_frame(struct sk_buff *skb)
+{
+	struct net_device *slave_dev;
+	struct net_device *bond_dev;
+
+	skb = skb_share_check(skb, GFP_ATOMIC);
+	if (unlikely(!skb))
+		return NULL;
+	slave_dev = skb->dev;
+	bond_dev = ACCESS_ONCE(slave_dev->master);
+	if (unlikely(!bond_dev))
+		return skb;
+
+	if (bond_dev->priv_flags&  IFF_MASTER_ARPMON)
+		slave_dev->last_rx = jiffies;
+
+	if (bond_should_deliver_exact_match(skb, slave_dev, bond_dev)) {
+		skb->deliver_no_wcard = 1;
+		return skb;

Shouldn't we return NULL here ?

No we shouldn't. We need sbk to be delivered to exact match.

quoted

+	}
+
+	skb->dev = bond_dev;
+
+	if (bond_dev->priv_flags&  IFF_MASTER_ALB&&
+	    bond_dev->priv_flags&  IFF_BRIDGE_PORT&&
+	    skb->pkt_type == PACKET_HOST) {
+		u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
+
+		memcpy(dest, bond_dev->dev_addr, ETH_ALEN);
+	}
+
+	return skb;
+}
+

[snip]

quoted

+static void vlan_on_bond_hook(struct sk_buff *skb)
 {
-	if (skb->pkt_type == PACKET_HOST) {
-		u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
+	/*
+	 * Make sure ARP frames received on VLAN interfaces stacked on
+	 * bonding interfaces still make their way to any base bonding
+	 * device that may have registered for a specific ptype.
+	 */
+	if (skb->dev->priv_flags&  IFF_802_1Q_VLAN&&
+	    vlan_dev_real_dev(skb->dev)->priv_flags&  IFF_BONDING&&
+	    skb->protocol == htons(ETH_P_ARP)) {

The vlan_on_bond case used to be cost effective. Now, we clone the skb and call netif_rx...

This should not cost too much overhead considering only few packets are
going thru this. This hook shouldn't have exited in the fisrt place. I
think introducing this functionality was a big mistake.

quoted

+		struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);

-		memcpy(dest, master->dev_addr, ETH_ALEN);
+		if (!skb2)
+			return;
+		skb2->dev = vlan_dev_real_dev(skb->dev);
+		netif_rx(skb2);
 	}
 }

[snip]

quoted

 	if (rx_handler) {
+		struct net_device *prev_dev;
+
 		if (pt_prev) {
 			ret = deliver_skb(skb, pt_prev, orig_dev);
 			pt_prev = NULL;
 		}
+		prev_dev = skb->dev;
 		skb = rx_handler(skb);
 		if (!skb)
 			goto out;

I would instead consider NULL as meaning exact-match-delivery-only.
(The same effect as dev_bond_should_drop() returning true).

we can change the behaviour later on.

quoted

+		if (skb->dev != prev_dev)
+			goto another_round;
 	}

Anyway, all my comments can't be postponed to follow-up patchs. Thanks Jiri.

Reviewed-by: Nicolas de Pesloüan <redacted>