datapath: Avoid freeing wild pointer in corner case.
[sliver-openvswitch.git] / datapath / flow.c
index 4365e22..d670925 100644 (file)
@@ -32,6 +32,9 @@
 #include <net/inet_ecn.h>
 #include <net/ip.h>
 #include <net/ipv6.h>
+#include <net/ndisc.h>
+
+#include "vlan.h"
 
 static struct kmem_cache *flow_cache;
 static unsigned int hash_seed __read_mostly;
@@ -103,7 +106,6 @@ static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
        unsigned int nh_ofs = skb_network_offset(skb);
        unsigned int nh_len;
        int payload_ofs;
-       int payload_len;
        struct ipv6hdr *nh;
        uint8_t nexthdr;
 
@@ -113,30 +115,17 @@ static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
        nh = ipv6_hdr(skb);
        nexthdr = nh->nexthdr;
        payload_ofs = (u8 *)(nh + 1) - skb->data;
-       payload_len = ntohs(nh->payload_len);
 
-       memcpy(key->ipv6_src, nh->saddr.in6_u.u6_addr8, sizeof(key->ipv6_src));
-       memcpy(key->ipv6_dst, nh->daddr.in6_u.u6_addr8, sizeof(key->ipv6_dst));
+       ipv6_addr_copy(&key->ipv6_src, &nh->saddr);
+       ipv6_addr_copy(&key->ipv6_dst, &nh->daddr);
        key->nw_tos = ipv6_get_dsfield(nh) & ~INET_ECN_MASK;
        key->nw_proto = NEXTHDR_NONE;
 
-       /* We don't process jumbograms. */
-       if (!payload_len)
-               return -EINVAL;
-
-       if (unlikely(skb->len < nh_ofs + sizeof(*nh) + payload_len))
-               return -EINVAL;
-
        payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr);
-       if (payload_ofs < 0) {
+       if (unlikely(payload_ofs < 0))
                return -EINVAL;
-       }
-       nh_len = payload_ofs - nh_ofs;
 
-       /* Ensure that the payload length claimed is at least large enough
-        * for the headers we've already processed. */
-       if (payload_len < nh_len - sizeof(*nh))
-               return -EINVAL;
+       nh_len = payload_ofs - nh_ofs;
 
        /* Pull enough header bytes to account for the IP header plus the
         * longest transport header that we parse, currently 20 bytes for TCP.
@@ -207,6 +196,7 @@ struct sw_flow *flow_alloc(void)
 
        spin_lock_init(&flow->lock);
        atomic_set(&flow->refcnt, 1);
+       flow->sf_acts = NULL;
        flow->dead = false;
 
        return flow;
@@ -314,6 +304,77 @@ static __be16 parse_ethertype(struct sk_buff *skb)
        return llc->ethertype;
 }
 
+static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
+                       int nh_len)
+{
+       struct icmp6hdr *icmp = icmp6_hdr(skb);
+
+       /* The ICMPv6 type and code fields use the 16-bit transport port
+        * fields, so we need to store them in 16-bit network byte order.
+        */
+       key->tp_src = htons(icmp->icmp6_type);
+       key->tp_dst = htons(icmp->icmp6_code);
+
+       if (icmp->icmp6_code == 0 &&
+           (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
+            icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
+               int icmp_len = skb->len - skb_transport_offset(skb);
+               struct nd_msg *nd;
+               int offset;
+
+               /* In order to process neighbor discovery options, we need the
+                * entire packet.
+                */
+               if (unlikely(icmp_len < sizeof(*nd)))
+                       return 0;
+               if (unlikely(skb_linearize(skb)))
+                       return -ENOMEM;
+
+               nd = (struct nd_msg *)skb_transport_header(skb);
+               ipv6_addr_copy(&key->nd_target, &nd->target);
+
+               icmp_len -= sizeof(*nd);
+               offset = 0;
+               while (icmp_len >= 8) {
+                       struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd->opt + offset);
+                       int opt_len = nd_opt->nd_opt_len * 8;
+
+                       if (unlikely(!opt_len || opt_len > icmp_len))
+                               goto invalid;
+
+                       /* Store the link layer address if the appropriate
+                        * option is provided.  It is considered an error if
+                        * the same link layer option is specified twice.
+                        */
+                       if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
+                           && opt_len == 8) {
+                               if (unlikely(!is_zero_ether_addr(key->arp_sha)))
+                                       goto invalid;
+                               memcpy(key->arp_sha,
+                                   &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
+                       } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
+                                  && opt_len == 8) {
+                               if (unlikely(!is_zero_ether_addr(key->arp_tha)))
+                                       goto invalid;
+                               memcpy(key->arp_tha,
+                                   &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
+                       }
+
+                       icmp_len -= opt_len;
+                       offset += opt_len;
+               }
+       }
+
+       return 0;
+
+invalid:
+       memset(&key->nd_target, 0, sizeof(key->nd_target));
+       memset(key->arp_sha, 0, sizeof(key->arp_sha));
+       memset(key->arp_tha, 0, sizeof(key->arp_tha));
+
+       return 0;
+}
+
 /**
  * flow_extract - extracts a flow key from an Ethernet frame.
  * @skb: sk_buff that contains the frame, with skb->data pointing to the
@@ -379,8 +440,12 @@ int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 
        /* dl_type, dl_vlan, dl_vlan_pcp. */
        __skb_pull(skb, 2 * ETH_ALEN);
-       if (eth->h_proto == htons(ETH_P_8021Q))
+
+       if (vlan_tx_tag_present(skb))
+               key->dl_tci = htons(vlan_get_tci(skb));
+       else if (eth->h_proto == htons(ETH_P_8021Q))
                parse_vlan(skb, key);
+
        key->dl_type = parse_ethertype(skb);
        skb_reset_network_header(skb);
        __skb_push(skb, skb->data - (unsigned char *)eth);
@@ -482,12 +547,9 @@ int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
                        }
                } else if (key->nw_proto == NEXTHDR_ICMP) {
                        if (icmp6hdr_ok(skb)) {
-                               struct icmp6hdr *icmp = icmp6_hdr(skb);
-                               /* The ICMPv6 type and code fields use the 16-bit
-                                * transport port fields, so we need to store them
-                                * in 16-bit network byte order. */
-                               key->tp_src = htons(icmp->icmp6_type);
-                               key->tp_dst = htons(icmp->icmp6_code);
+                               int error = parse_icmpv6(skb, key, nh_len);
+                               if (error < 0)
+                                       return error;
                        }
                }
        }
@@ -517,7 +579,7 @@ int flow_cmp(const struct tbl_node *node, void *key2_)
  * elements and | for alternatives:
  *
  * [tun_id] in_port ethernet [8021q] [ethertype \
- *              [IPv4 [TCP|UDP|ICMP] | IPv6 [TCP|UDP|ICMPv6] | ARP]]
+ *              [IPv4 [TCP|UDP|ICMP] | IPv6 [TCP|UDP|ICMPv6 [ND]] | ARP]]
  */
 int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *attr)
 {
@@ -543,6 +605,7 @@ int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *attr)
                        [ODP_KEY_ATTR_ICMP] = sizeof(struct odp_key_icmp),
                        [ODP_KEY_ATTR_ICMPV6] = sizeof(struct odp_key_icmpv6),
                        [ODP_KEY_ATTR_ARP] = sizeof(struct odp_key_arp),
+                       [ODP_KEY_ATTR_ND] = sizeof(struct odp_key_nd),
                };
 
                const struct odp_key_ethernet *eth_key;
@@ -554,6 +617,7 @@ int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *attr)
                const struct odp_key_icmp *icmp_key;
                const struct odp_key_icmpv6 *icmpv6_key;
                const struct odp_key_arp *arp_key;
+               const struct odp_key_nd *nd_key;
 
                 int type = nla_type(nla);
 
@@ -612,9 +676,9 @@ int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *attr)
                        if (swkey->dl_type != htons(ETH_P_IPV6))
                                return -EINVAL;
                        ipv6_key = nla_data(nla);
-                       memcpy(swkey->ipv6_src, ipv6_key->ipv6_src,
+                       memcpy(&swkey->ipv6_src, ipv6_key->ipv6_src,
                                        sizeof(swkey->ipv6_src));
-                       memcpy(swkey->ipv6_dst, ipv6_key->ipv6_dst,
+                       memcpy(&swkey->ipv6_dst, ipv6_key->ipv6_dst,
                                        sizeof(swkey->ipv6_dst));
                        swkey->nw_proto = ipv6_key->ipv6_proto;
                        swkey->nw_tos = ipv6_key->ipv6_tos;
@@ -669,6 +733,17 @@ int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *attr)
                        memcpy(swkey->arp_tha, arp_key->arp_tha, ETH_ALEN);
                        break;
 
+               case TRANSITION(ODP_KEY_ATTR_ICMPV6, ODP_KEY_ATTR_ND):
+                       if (swkey->tp_src != htons(NDISC_NEIGHBOUR_SOLICITATION)
+                           && swkey->tp_src != htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
+                               return -EINVAL;
+                       nd_key = nla_data(nla);
+                       memcpy(&swkey->nd_target, nd_key->nd_target,
+                                       sizeof(swkey->nd_target));
+                       memcpy(swkey->arp_sha, nd_key->nd_sll, ETH_ALEN);
+                       memcpy(swkey->arp_tha, nd_key->nd_tll, ETH_ALEN);
+                       break;
+
                default:
                        return -EINVAL;
                }
@@ -710,11 +785,17 @@ int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *attr)
                        return -EINVAL;
                return 0;
 
+       case ODP_KEY_ATTR_ICMPV6:
+               if (swkey->tp_src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
+                   swkey->tp_src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
+                       return -EINVAL;
+               return 0;
+
        case ODP_KEY_ATTR_TCP:
        case ODP_KEY_ATTR_UDP:
        case ODP_KEY_ATTR_ICMP:
-       case ODP_KEY_ATTR_ICMPV6:
        case ODP_KEY_ATTR_ARP:
+       case ODP_KEY_ATTR_ND:
                return 0;
        }
 
@@ -727,6 +808,11 @@ int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
        struct odp_key_ethernet *eth_key;
        struct nlattr *nla;
 
+       /* This is an imperfect sanity-check that FLOW_BUFSIZE doesn't need
+        * to be updated, but will at least raise awareness when new ODP key
+        * types are added. */
+       BUILD_BUG_ON(__ODP_KEY_ATTR_MAX != 14);
+
        if (swkey->tun_id != cpu_to_be64(0))
                NLA_PUT_BE64(skb, ODP_KEY_ATTR_TUN_ID, swkey->tun_id);
 
@@ -759,6 +845,7 @@ int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
                if (!nla)
                        goto nla_put_failure;
                ipv4_key = nla_data(nla);
+               memset(ipv4_key, 0, sizeof(struct odp_key_ipv4));
                ipv4_key->ipv4_src = swkey->ipv4_src;
                ipv4_key->ipv4_dst = swkey->ipv4_dst;
                ipv4_key->ipv4_proto = swkey->nw_proto;
@@ -770,9 +857,10 @@ int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
                if (!nla)
                        goto nla_put_failure;
                ipv6_key = nla_data(nla);
-               memcpy(ipv6_key->ipv6_src, swkey->ipv6_src,
+               memset(ipv6_key, 0, sizeof(struct odp_key_ipv6));
+               memcpy(ipv6_key->ipv6_src, &swkey->ipv6_src,
                                sizeof(ipv6_key->ipv6_src));
-               memcpy(ipv6_key->ipv6_dst, swkey->ipv6_dst,
+               memcpy(ipv6_key->ipv6_dst, &swkey->ipv6_dst,
                                sizeof(ipv6_key->ipv6_dst));
                ipv6_key->ipv6_proto = swkey->nw_proto;
                ipv6_key->ipv6_tos = swkey->nw_tos;
@@ -783,6 +871,7 @@ int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
                if (!nla)
                        goto nla_put_failure;
                arp_key = nla_data(nla);
+               memset(arp_key, 0, sizeof(struct odp_key_arp));
                arp_key->arp_sip = swkey->ipv4_src;
                arp_key->arp_tip = swkey->ipv4_dst;
                arp_key->arp_op = htons(swkey->nw_proto);
@@ -790,8 +879,8 @@ int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
                memcpy(arp_key->arp_tha, swkey->arp_tha, ETH_ALEN);
        }
 
-       if (swkey->dl_type == htons(ETH_P_IP)
-                       || swkey->dl_type == htons(ETH_P_IPV6)) {
+       if (swkey->dl_type == htons(ETH_P_IP) ||
+           swkey->dl_type == htons(ETH_P_IPV6)) {
 
                if (swkey->nw_proto == IPPROTO_TCP) {
                        struct odp_key_tcp *tcp_key;
@@ -811,8 +900,8 @@ int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
                        udp_key = nla_data(nla);
                        udp_key->udp_src = swkey->tp_src;
                        udp_key->udp_dst = swkey->tp_dst;
-               } else if (swkey->dl_type == htons(ETH_P_IP)
-                               && swkey->nw_proto == IPPROTO_ICMP) {
+               } else if (swkey->dl_type == htons(ETH_P_IP) &&
+                          swkey->nw_proto == IPPROTO_ICMP) {
                        struct odp_key_icmp *icmp_key;
 
                        nla = nla_reserve(skb, ODP_KEY_ATTR_ICMP, sizeof(*icmp_key));
@@ -821,16 +910,31 @@ int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
                        icmp_key = nla_data(nla);
                        icmp_key->icmp_type = ntohs(swkey->tp_src);
                        icmp_key->icmp_code = ntohs(swkey->tp_dst);
-               } else if (swkey->dl_type == htons(ETH_P_IPV6)
-                               && swkey->nw_proto == IPPROTO_ICMPV6) {
+               } else if (swkey->dl_type == htons(ETH_P_IPV6) &&
+                          swkey->nw_proto == IPPROTO_ICMPV6) {
                        struct odp_key_icmpv6 *icmpv6_key;
 
-                       nla = nla_reserve(skb, ODP_KEY_ATTR_ICMPV6, sizeof(*icmpv6_key));
+                       nla = nla_reserve(skb, ODP_KEY_ATTR_ICMPV6,
+                                               sizeof(*icmpv6_key));
                        if (!nla)
                                goto nla_put_failure;
                        icmpv6_key = nla_data(nla);
                        icmpv6_key->icmpv6_type = ntohs(swkey->tp_src);
                        icmpv6_key->icmpv6_code = ntohs(swkey->tp_dst);
+
+                       if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
+                           icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
+                               struct odp_key_nd *nd_key;
+
+                               nla = nla_reserve(skb, ODP_KEY_ATTR_ND, sizeof(*nd_key));
+                               if (!nla)
+                                       goto nla_put_failure;
+                               nd_key = nla_data(nla);
+                               memcpy(nd_key->nd_target, &swkey->nd_target,
+                                                       sizeof(nd_key->nd_target));
+                               memcpy(nd_key->nd_sll, swkey->arp_sha, ETH_ALEN);
+                               memcpy(nd_key->nd_tll, swkey->arp_tha, ETH_ALEN);
+                       }
                }
        }