bonding: fix l23 and l34 load balancing in forwarding path
Eric Dumazet [Mon, 15 Apr 2013 17:03:24 +0000 (17:03 +0000)]
Since commit 6b923cb7188d46 (bonding: support for IPv6 transmit hashing)
bonding doesn't properly hash traffic in forwarding setups.

Vitaly V. Bursov diagnosed that skb_network_header_len() returned 0 in
this case.

More generally, the transport header might not be in the skb head.

Use pskb_may_pull() & skb_header_pointer() to get it right, and use
proto_ports_offset() in bond_xmit_hash_policy_l34() to get support for
more protocols than TCP and UDP.

Reported-by: Vitaly V. Bursov <vitalyb@telenet.dn.ua>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Jay Vosburgh <fubar@us.ibm.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Cc: John Eaglesham <linux@8192.net>
Tested-by: Vitaly V. Bursov <vitalyb@telenet.dn.ua>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/bonding/bond_main.c

index a61a760..dea8ce2 100644 (file)
@@ -3296,20 +3296,22 @@ static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
  */
 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
 {
-       struct ethhdr *data = (struct ethhdr *)skb->data;
-       struct iphdr *iph;
-       struct ipv6hdr *ipv6h;
+       const struct ethhdr *data;
+       const struct iphdr *iph;
+       const struct ipv6hdr *ipv6h;
        u32 v6hash;
-       __be32 *s, *d;
+       const __be32 *s, *d;
 
        if (skb->protocol == htons(ETH_P_IP) &&
-           skb_network_header_len(skb) >= sizeof(*iph)) {
+           pskb_network_may_pull(skb, sizeof(*iph))) {
                iph = ip_hdr(skb);
+               data = (struct ethhdr *)skb->data;
                return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
                        (data->h_dest[5] ^ data->h_source[5])) % count;
        } else if (skb->protocol == htons(ETH_P_IPV6) &&
-                  skb_network_header_len(skb) >= sizeof(*ipv6h)) {
+                  pskb_network_may_pull(skb, sizeof(*ipv6h))) {
                ipv6h = ipv6_hdr(skb);
+               data = (struct ethhdr *)skb->data;
                s = &ipv6h->saddr.s6_addr32[0];
                d = &ipv6h->daddr.s6_addr32[0];
                v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
@@ -3328,33 +3330,36 @@ static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
 {
        u32 layer4_xor = 0;
-       struct iphdr *iph;
-       struct ipv6hdr *ipv6h;
-       __be32 *s, *d;
-       __be16 *layer4hdr;
+       const struct iphdr *iph;
+       const struct ipv6hdr *ipv6h;
+       const __be32 *s, *d;
+       const __be16 *l4 = NULL;
+       __be16 _l4[2];
+       int noff = skb_network_offset(skb);
+       int poff;
 
        if (skb->protocol == htons(ETH_P_IP) &&
-           skb_network_header_len(skb) >= sizeof(*iph)) {
+           pskb_may_pull(skb, noff + sizeof(*iph))) {
                iph = ip_hdr(skb);
-               if (!ip_is_fragment(iph) &&
-                   (iph->protocol == IPPROTO_TCP ||
-                    iph->protocol == IPPROTO_UDP) &&
-                   (skb_headlen(skb) - skb_network_offset(skb) >=
-                    iph->ihl * sizeof(u32) + sizeof(*layer4hdr) * 2)) {
-                       layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
-                       layer4_xor = ntohs(*layer4hdr ^ *(layer4hdr + 1));
+               poff = proto_ports_offset(iph->protocol);
+
+               if (!ip_is_fragment(iph) && poff >= 0) {
+                       l4 = skb_header_pointer(skb, noff + (iph->ihl << 2) + poff,
+                                               sizeof(_l4), &_l4);
+                       if (l4)
+                               layer4_xor = ntohs(l4[0] ^ l4[1]);
                }
                return (layer4_xor ^
                        ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
        } else if (skb->protocol == htons(ETH_P_IPV6) &&
-                  skb_network_header_len(skb) >= sizeof(*ipv6h)) {
+                  pskb_may_pull(skb, noff + sizeof(*ipv6h))) {
                ipv6h = ipv6_hdr(skb);
-               if ((ipv6h->nexthdr == IPPROTO_TCP ||
-                    ipv6h->nexthdr == IPPROTO_UDP) &&
-                   (skb_headlen(skb) - skb_network_offset(skb) >=
-                    sizeof(*ipv6h) + sizeof(*layer4hdr) * 2)) {
-                       layer4hdr = (__be16 *)(ipv6h + 1);
-                       layer4_xor = ntohs(*layer4hdr ^ *(layer4hdr + 1));
+               poff = proto_ports_offset(ipv6h->nexthdr);
+               if (poff >= 0) {
+                       l4 = skb_header_pointer(skb, noff + sizeof(*ipv6h) + poff,
+                                               sizeof(_l4), &_l4);
+                       if (l4)
+                               layer4_xor = ntohs(l4[0] ^ l4[1]);
                }
                s = &ipv6h->saddr.s6_addr32[0];
                d = &ipv6h->daddr.s6_addr32[0];