2 * Distributed under the terms of the GNU GPL version 2.
3 * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
5 * Significant portions of this file may be copied from parts of the Linux
6 * kernel, by Linus Torvalds and others.
11 #include <asm/uaccess.h>
12 #include <linux/netdevice.h>
13 #include <linux/etherdevice.h>
14 #include <linux/if_ether.h>
15 #include <linux/if_vlan.h>
16 #include <net/llc_pdu.h>
17 #include <linux/kernel.h>
18 #include <linux/jhash.h>
19 #include <linux/jiffies.h>
20 #include <linux/llc.h>
21 #include <linux/module.h>
23 #include <linux/rcupdate.h>
24 #include <linux/if_arp.h>
25 #include <linux/if_ether.h>
27 #include <linux/ipv6.h>
28 #include <linux/tcp.h>
29 #include <linux/udp.h>
30 #include <linux/icmp.h>
31 #include <linux/icmpv6.h>
32 #include <net/inet_ecn.h>
35 #include <net/ndisc.h>
37 static struct kmem_cache *flow_cache;
38 static unsigned int hash_seed __read_mostly;
40 static inline bool arphdr_ok(struct sk_buff *skb)
42 return skb->len >= skb_network_offset(skb) + sizeof(struct arp_eth_header);
45 static inline int check_iphdr(struct sk_buff *skb)
47 unsigned int nh_ofs = skb_network_offset(skb);
50 if (skb->len < nh_ofs + sizeof(struct iphdr))
53 ip_len = ip_hdrlen(skb);
54 if (ip_len < sizeof(struct iphdr) || skb->len < nh_ofs + ip_len)
58 * Pull enough header bytes to account for the IP header plus the
59 * longest transport header that we parse, currently 20 bytes for TCP.
61 if (!pskb_may_pull(skb, min(nh_ofs + ip_len + 20, skb->len)))
64 skb_set_transport_header(skb, nh_ofs + ip_len);
68 static inline bool tcphdr_ok(struct sk_buff *skb)
70 int th_ofs = skb_transport_offset(skb);
71 if (skb->len >= th_ofs + sizeof(struct tcphdr)) {
72 int tcp_len = tcp_hdrlen(skb);
73 return (tcp_len >= sizeof(struct tcphdr)
74 && skb->len >= th_ofs + tcp_len);
79 static inline bool udphdr_ok(struct sk_buff *skb)
81 return skb->len >= skb_transport_offset(skb) + sizeof(struct udphdr);
84 static inline bool icmphdr_ok(struct sk_buff *skb)
86 return skb->len >= skb_transport_offset(skb) + sizeof(struct icmphdr);
89 u64 flow_used_time(unsigned long flow_jiffies)
91 struct timespec cur_ts;
94 ktime_get_ts(&cur_ts);
95 idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
96 cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
97 cur_ts.tv_nsec / NSEC_PER_MSEC;
99 return cur_ms - idle_ms;
102 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
104 unsigned int nh_ofs = skb_network_offset(skb);
111 if (unlikely(skb->len < nh_ofs + sizeof(*nh)))
115 nexthdr = nh->nexthdr;
116 payload_ofs = (u8 *)(nh + 1) - skb->data;
117 payload_len = ntohs(nh->payload_len);
119 memcpy(key->ipv6_src, nh->saddr.in6_u.u6_addr8, sizeof(key->ipv6_src));
120 memcpy(key->ipv6_dst, nh->daddr.in6_u.u6_addr8, sizeof(key->ipv6_dst));
121 key->nw_tos = ipv6_get_dsfield(nh) & ~INET_ECN_MASK;
122 key->nw_proto = NEXTHDR_NONE;
124 /* We don't process jumbograms. */
128 if (unlikely(skb->len < nh_ofs + sizeof(*nh) + payload_len))
131 payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr);
132 if (payload_ofs < 0) {
135 nh_len = payload_ofs - nh_ofs;
137 /* Ensure that the payload length claimed is at least large enough
138 * for the headers we've already processed. */
139 if (payload_len < nh_len - sizeof(*nh))
142 /* Pull enough header bytes to account for the IP header plus the
143 * longest transport header that we parse, currently 20 bytes for TCP.
144 * To dig deeper than the transport header, transport parsers may need
145 * to pull more header bytes.
147 if (unlikely(!pskb_may_pull(skb, min(nh_ofs + nh_len + 20, skb->len))))
150 skb_set_transport_header(skb, nh_ofs + nh_len);
151 key->nw_proto = nexthdr;
155 static bool icmp6hdr_ok(struct sk_buff *skb)
157 return skb->len >= skb_transport_offset(skb) + sizeof(struct icmp6hdr);
160 #define TCP_FLAGS_OFFSET 13
161 #define TCP_FLAG_MASK 0x3f
163 void flow_used(struct sw_flow *flow, struct sk_buff *skb)
167 if (flow->key.dl_type == htons(ETH_P_IP) &&
168 flow->key.nw_proto == IPPROTO_TCP) {
169 u8 *tcp = (u8 *)tcp_hdr(skb);
170 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
173 spin_lock_bh(&flow->lock);
174 flow->used = jiffies;
175 flow->packet_count++;
176 flow->byte_count += skb->len;
177 flow->tcp_flags |= tcp_flags;
178 spin_unlock_bh(&flow->lock);
181 struct sw_flow_actions *flow_actions_alloc(const struct nlattr *actions)
183 int actions_len = nla_len(actions);
184 struct sw_flow_actions *sfa;
186 /* At least DP_MAX_PORTS actions are required to be able to flood a
187 * packet to every port. Factor of 2 allows for setting VLAN tags,
189 if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
190 return ERR_PTR(-EINVAL);
192 sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
194 return ERR_PTR(-ENOMEM);
196 sfa->actions_len = actions_len;
197 memcpy(sfa->actions, nla_data(actions), actions_len);
201 struct sw_flow *flow_alloc(void)
203 struct sw_flow *flow;
205 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
207 return ERR_PTR(-ENOMEM);
209 spin_lock_init(&flow->lock);
210 atomic_set(&flow->refcnt, 1);
216 void flow_free_tbl(struct tbl_node *node)
218 struct sw_flow *flow = flow_cast(node);
224 /* RCU callback used by flow_deferred_free. */
225 static void rcu_free_flow_callback(struct rcu_head *rcu)
227 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
233 /* Schedules 'flow' to be freed after the next RCU grace period.
234 * The caller must hold rcu_read_lock for this to be sensible. */
235 void flow_deferred_free(struct sw_flow *flow)
237 call_rcu(&flow->rcu, rcu_free_flow_callback);
240 void flow_hold(struct sw_flow *flow)
242 atomic_inc(&flow->refcnt);
245 void flow_put(struct sw_flow *flow)
250 if (atomic_dec_and_test(&flow->refcnt)) {
251 kfree((struct sf_flow_acts __force *)flow->sf_acts);
252 kmem_cache_free(flow_cache, flow);
256 /* RCU callback used by flow_deferred_free_acts. */
257 static void rcu_free_acts_callback(struct rcu_head *rcu)
259 struct sw_flow_actions *sf_acts = container_of(rcu,
260 struct sw_flow_actions, rcu);
264 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
265 * The caller must hold rcu_read_lock for this to be sensible. */
266 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
268 call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
271 static void parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
274 __be16 eth_type; /* ETH_P_8021Q */
277 struct qtag_prefix *qp;
279 if (skb->len < sizeof(struct qtag_prefix) + sizeof(__be16))
282 qp = (struct qtag_prefix *) skb->data;
283 key->dl_tci = qp->tci | htons(VLAN_TAG_PRESENT);
284 __skb_pull(skb, sizeof(struct qtag_prefix));
287 static __be16 parse_ethertype(struct sk_buff *skb)
289 struct llc_snap_hdr {
290 u8 dsap; /* Always 0xAA */
291 u8 ssap; /* Always 0xAA */
296 struct llc_snap_hdr *llc;
299 proto = *(__be16 *) skb->data;
300 __skb_pull(skb, sizeof(__be16));
302 if (ntohs(proto) >= 1536)
305 if (unlikely(skb->len < sizeof(struct llc_snap_hdr)))
306 return htons(ETH_P_802_2);
308 llc = (struct llc_snap_hdr *) skb->data;
309 if (llc->dsap != LLC_SAP_SNAP ||
310 llc->ssap != LLC_SAP_SNAP ||
311 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
312 return htons(ETH_P_802_2);
314 __skb_pull(skb, sizeof(struct llc_snap_hdr));
315 return llc->ethertype;
318 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
321 struct ipv6hdr *nh = ipv6_hdr(skb);
322 int icmp_len = ntohs(nh->payload_len) + sizeof(*nh) - nh_len;
323 struct icmp6hdr *icmp = icmp6_hdr(skb);
325 /* The ICMPv6 type and code fields use the 16-bit transport port
326 * fields, so we need to store them in 16-bit network byte order. */
327 key->tp_src = htons(icmp->icmp6_type);
328 key->tp_dst = htons(icmp->icmp6_code);
330 if (!icmp->icmp6_code
331 && ((icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
332 || (icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT))) {
336 /* In order to process neighbor discovery options, we need the
338 if (icmp_len < sizeof(*nd))
340 if (!pskb_may_pull(skb, skb_transport_offset(skb) + icmp_len))
343 nd = (struct nd_msg *)skb_transport_header(skb);
344 memcpy(key->nd_target, &nd->target, sizeof(key->nd_target));
346 icmp_len -= sizeof(*nd);
348 while (icmp_len >= 8) {
349 struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd->opt + offset);
350 int opt_len = nd_opt->nd_opt_len * 8;
352 if (!opt_len || (opt_len > icmp_len))
355 /* Store the link layer address if the appropriate option is
356 * provided. It is considered an error if the same link
357 * layer option is specified twice. */
358 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
360 if (!is_zero_ether_addr(key->arp_sha))
363 &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
364 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
366 if (!is_zero_ether_addr(key->arp_tha))
369 &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
380 memset(key->nd_target, 0, sizeof(key->nd_target));
381 memset(key->arp_sha, 0, sizeof(key->arp_sha));
382 memset(key->arp_tha, 0, sizeof(key->arp_tha));
388 * flow_extract - extracts a flow key from an Ethernet frame.
389 * @skb: sk_buff that contains the frame, with skb->data pointing to the
391 * @in_port: port number on which @skb was received.
392 * @key: output flow key
393 * @is_frag: set to 1 if @skb contains an IPv4 fragment, or to 0 if @skb does
394 * not contain an IPv4 packet or if it is not a fragment.
396 * The caller must ensure that skb->len >= ETH_HLEN.
398 * Returns 0 if successful, otherwise a negative errno value.
400 * Initializes @skb header pointers as follows:
402 * - skb->mac_header: the Ethernet header.
404 * - skb->network_header: just past the Ethernet header, or just past the
405 * VLAN header, to the first byte of the Ethernet payload.
407 * - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
408 * on output, then just past the IP header, if one is present and
409 * of a correct length, otherwise the same as skb->network_header.
410 * For other key->dl_type values it is left untouched.
412 int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
417 memset(key, 0, sizeof(*key));
418 key->tun_id = OVS_CB(skb)->tun_id;
419 key->in_port = in_port;
423 * We would really like to pull as many bytes as we could possibly
424 * want to parse into the linear data area. Currently, for IPv4,
429 * 60 max IP header with options
430 * 20 max TCP/UDP/ICMP header (don't care about options)
434 * But Xen only allocates 64 or 72 bytes for the linear data area in
435 * netback, which means that we would reallocate and copy the skb's
436 * linear data on every packet if we did that. So instead just pull 64
437 * bytes, which is always sufficient without IP options, and then check
438 * whether we need to pull more later when we look at the IP header.
440 if (!pskb_may_pull(skb, min(skb->len, 64u)))
443 skb_reset_mac_header(skb);
447 memcpy(key->dl_src, eth->h_source, ETH_ALEN);
448 memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
450 /* dl_type, dl_vlan, dl_vlan_pcp. */
451 __skb_pull(skb, 2 * ETH_ALEN);
452 if (eth->h_proto == htons(ETH_P_8021Q))
453 parse_vlan(skb, key);
454 key->dl_type = parse_ethertype(skb);
455 skb_reset_network_header(skb);
456 __skb_push(skb, skb->data - (unsigned char *)eth);
459 if (key->dl_type == htons(ETH_P_IP)) {
463 error = check_iphdr(skb);
464 if (unlikely(error)) {
465 if (error == -EINVAL) {
466 skb->transport_header = skb->network_header;
473 key->ipv4_src = nh->saddr;
474 key->ipv4_dst = nh->daddr;
475 key->nw_tos = nh->tos & ~INET_ECN_MASK;
476 key->nw_proto = nh->protocol;
478 /* Transport layer. */
479 if (!(nh->frag_off & htons(IP_MF | IP_OFFSET)) &&
480 !(skb_shinfo(skb)->gso_type & SKB_GSO_UDP)) {
481 if (key->nw_proto == IPPROTO_TCP) {
482 if (tcphdr_ok(skb)) {
483 struct tcphdr *tcp = tcp_hdr(skb);
484 key->tp_src = tcp->source;
485 key->tp_dst = tcp->dest;
487 } else if (key->nw_proto == IPPROTO_UDP) {
488 if (udphdr_ok(skb)) {
489 struct udphdr *udp = udp_hdr(skb);
490 key->tp_src = udp->source;
491 key->tp_dst = udp->dest;
493 } else if (key->nw_proto == IPPROTO_ICMP) {
494 if (icmphdr_ok(skb)) {
495 struct icmphdr *icmp = icmp_hdr(skb);
496 /* The ICMP type and code fields use the 16-bit
497 * transport port fields, so we need to store them
498 * in 16-bit network byte order. */
499 key->tp_src = htons(icmp->type);
500 key->tp_dst = htons(icmp->code);
506 } else if (key->dl_type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
507 struct arp_eth_header *arp;
509 arp = (struct arp_eth_header *)skb_network_header(skb);
511 if (arp->ar_hrd == htons(ARPHRD_ETHER)
512 && arp->ar_pro == htons(ETH_P_IP)
513 && arp->ar_hln == ETH_ALEN
514 && arp->ar_pln == 4) {
516 /* We only match on the lower 8 bits of the opcode. */
517 if (ntohs(arp->ar_op) <= 0xff)
518 key->nw_proto = ntohs(arp->ar_op);
520 if (key->nw_proto == ARPOP_REQUEST
521 || key->nw_proto == ARPOP_REPLY) {
522 memcpy(&key->ipv4_src, arp->ar_sip, sizeof(key->ipv4_src));
523 memcpy(&key->ipv4_dst, arp->ar_tip, sizeof(key->ipv4_dst));
524 memcpy(key->arp_sha, arp->ar_sha, ETH_ALEN);
525 memcpy(key->arp_tha, arp->ar_tha, ETH_ALEN);
528 } else if (key->dl_type == htons(ETH_P_IPV6)) {
529 int nh_len; /* IPv6 Header + Extensions */
531 nh_len = parse_ipv6hdr(skb, key);
532 if (unlikely(nh_len < 0)) {
533 if (nh_len == -EINVAL) {
534 skb->transport_header = skb->network_header;
540 /* Transport layer. */
541 if (key->nw_proto == NEXTHDR_TCP) {
542 if (tcphdr_ok(skb)) {
543 struct tcphdr *tcp = tcp_hdr(skb);
544 key->tp_src = tcp->source;
545 key->tp_dst = tcp->dest;
547 } else if (key->nw_proto == NEXTHDR_UDP) {
548 if (udphdr_ok(skb)) {
549 struct udphdr *udp = udp_hdr(skb);
550 key->tp_src = udp->source;
551 key->tp_dst = udp->dest;
553 } else if (key->nw_proto == NEXTHDR_ICMP) {
554 if (icmp6hdr_ok(skb)) {
555 int error = parse_icmpv6(skb, key, nh_len);
564 u32 flow_hash(const struct sw_flow_key *key)
566 return jhash2((u32*)key, sizeof(*key) / sizeof(u32), hash_seed);
569 int flow_cmp(const struct tbl_node *node, void *key2_)
571 const struct sw_flow_key *key1 = &flow_cast(node)->key;
572 const struct sw_flow_key *key2 = key2_;
574 return !memcmp(key1, key2, sizeof(struct sw_flow_key));
578 * flow_from_nlattrs - parses Netlink attributes into a flow key.
579 * @swkey: receives the extracted flow key.
580 * @key: Netlink attribute holding nested %ODP_KEY_ATTR_* Netlink attribute
583 * This state machine accepts the following forms, with [] for optional
584 * elements and | for alternatives:
586 * [tun_id] in_port ethernet [8021q] [ethertype \
587 * [IPv4 [TCP|UDP|ICMP] | IPv6 [TCP|UDP|ICMPv6 [ND]] | ARP]]
589 int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *attr)
591 const struct nlattr *nla;
595 memset(swkey, 0, sizeof(*swkey));
596 swkey->dl_type = htons(ETH_P_802_2);
598 prev_type = ODP_KEY_ATTR_UNSPEC;
599 nla_for_each_nested(nla, attr, rem) {
600 static const u32 key_lens[ODP_KEY_ATTR_MAX + 1] = {
601 [ODP_KEY_ATTR_TUN_ID] = 8,
602 [ODP_KEY_ATTR_IN_PORT] = 4,
603 [ODP_KEY_ATTR_ETHERNET] = sizeof(struct odp_key_ethernet),
604 [ODP_KEY_ATTR_8021Q] = sizeof(struct odp_key_8021q),
605 [ODP_KEY_ATTR_ETHERTYPE] = 2,
606 [ODP_KEY_ATTR_IPV4] = sizeof(struct odp_key_ipv4),
607 [ODP_KEY_ATTR_IPV6] = sizeof(struct odp_key_ipv6),
608 [ODP_KEY_ATTR_TCP] = sizeof(struct odp_key_tcp),
609 [ODP_KEY_ATTR_UDP] = sizeof(struct odp_key_udp),
610 [ODP_KEY_ATTR_ICMP] = sizeof(struct odp_key_icmp),
611 [ODP_KEY_ATTR_ICMPV6] = sizeof(struct odp_key_icmpv6),
612 [ODP_KEY_ATTR_ARP] = sizeof(struct odp_key_arp),
613 [ODP_KEY_ATTR_ND] = sizeof(struct odp_key_nd),
616 const struct odp_key_ethernet *eth_key;
617 const struct odp_key_8021q *q_key;
618 const struct odp_key_ipv4 *ipv4_key;
619 const struct odp_key_ipv6 *ipv6_key;
620 const struct odp_key_tcp *tcp_key;
621 const struct odp_key_udp *udp_key;
622 const struct odp_key_icmp *icmp_key;
623 const struct odp_key_icmpv6 *icmpv6_key;
624 const struct odp_key_arp *arp_key;
625 const struct odp_key_nd *nd_key;
627 int type = nla_type(nla);
629 if (type > ODP_KEY_ATTR_MAX || nla_len(nla) != key_lens[type])
632 #define TRANSITION(PREV_TYPE, TYPE) (((PREV_TYPE) << 16) | (TYPE))
633 switch (TRANSITION(prev_type, type)) {
634 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_TUN_ID):
635 swkey->tun_id = nla_get_be64(nla);
638 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_IN_PORT):
639 case TRANSITION(ODP_KEY_ATTR_TUN_ID, ODP_KEY_ATTR_IN_PORT):
640 if (nla_get_u32(nla) >= DP_MAX_PORTS)
642 swkey->in_port = nla_get_u32(nla);
645 case TRANSITION(ODP_KEY_ATTR_IN_PORT, ODP_KEY_ATTR_ETHERNET):
646 eth_key = nla_data(nla);
647 memcpy(swkey->dl_src, eth_key->eth_src, ETH_ALEN);
648 memcpy(swkey->dl_dst, eth_key->eth_dst, ETH_ALEN);
651 case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_8021Q):
652 q_key = nla_data(nla);
653 /* Only standard 0x8100 VLANs currently supported. */
654 if (q_key->q_tpid != htons(ETH_P_8021Q))
656 if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
658 swkey->dl_tci = q_key->q_tci | htons(VLAN_TAG_PRESENT);
661 case TRANSITION(ODP_KEY_ATTR_8021Q, ODP_KEY_ATTR_ETHERTYPE):
662 case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_ETHERTYPE):
663 swkey->dl_type = nla_get_be16(nla);
664 if (ntohs(swkey->dl_type) < 1536)
668 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_IPV4):
669 if (swkey->dl_type != htons(ETH_P_IP))
671 ipv4_key = nla_data(nla);
672 swkey->ipv4_src = ipv4_key->ipv4_src;
673 swkey->ipv4_dst = ipv4_key->ipv4_dst;
674 swkey->nw_proto = ipv4_key->ipv4_proto;
675 swkey->nw_tos = ipv4_key->ipv4_tos;
676 if (swkey->nw_tos & INET_ECN_MASK)
680 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_IPV6):
681 if (swkey->dl_type != htons(ETH_P_IPV6))
683 ipv6_key = nla_data(nla);
684 memcpy(swkey->ipv6_src, ipv6_key->ipv6_src,
685 sizeof(swkey->ipv6_src));
686 memcpy(swkey->ipv6_dst, ipv6_key->ipv6_dst,
687 sizeof(swkey->ipv6_dst));
688 swkey->nw_proto = ipv6_key->ipv6_proto;
689 swkey->nw_tos = ipv6_key->ipv6_tos;
690 if (swkey->nw_tos & INET_ECN_MASK)
694 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_TCP):
695 case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_TCP):
696 if (swkey->nw_proto != IPPROTO_TCP)
698 tcp_key = nla_data(nla);
699 swkey->tp_src = tcp_key->tcp_src;
700 swkey->tp_dst = tcp_key->tcp_dst;
703 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_UDP):
704 case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_UDP):
705 if (swkey->nw_proto != IPPROTO_UDP)
707 udp_key = nla_data(nla);
708 swkey->tp_src = udp_key->udp_src;
709 swkey->tp_dst = udp_key->udp_dst;
712 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_ICMP):
713 if (swkey->nw_proto != IPPROTO_ICMP)
715 icmp_key = nla_data(nla);
716 swkey->tp_src = htons(icmp_key->icmp_type);
717 swkey->tp_dst = htons(icmp_key->icmp_code);
720 case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_ICMPV6):
721 if (swkey->nw_proto != IPPROTO_ICMPV6)
723 icmpv6_key = nla_data(nla);
724 swkey->tp_src = htons(icmpv6_key->icmpv6_type);
725 swkey->tp_dst = htons(icmpv6_key->icmpv6_code);
728 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_ARP):
729 if (swkey->dl_type != htons(ETH_P_ARP))
731 arp_key = nla_data(nla);
732 swkey->ipv4_src = arp_key->arp_sip;
733 swkey->ipv4_dst = arp_key->arp_tip;
734 if (arp_key->arp_op & htons(0xff00))
736 swkey->nw_proto = ntohs(arp_key->arp_op);
737 memcpy(swkey->arp_sha, arp_key->arp_sha, ETH_ALEN);
738 memcpy(swkey->arp_tha, arp_key->arp_tha, ETH_ALEN);
741 case TRANSITION(ODP_KEY_ATTR_ICMPV6, ODP_KEY_ATTR_ND):
742 if (swkey->tp_src != htons(NDISC_NEIGHBOUR_SOLICITATION)
743 && swkey->tp_src != htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
745 nd_key = nla_data(nla);
746 memcpy(swkey->nd_target, nd_key->nd_target,
747 sizeof(swkey->nd_target));
748 memcpy(swkey->arp_sha, nd_key->nd_sll, ETH_ALEN);
749 memcpy(swkey->arp_tha, nd_key->nd_tll, ETH_ALEN);
762 case ODP_KEY_ATTR_UNSPEC:
765 case ODP_KEY_ATTR_TUN_ID:
766 case ODP_KEY_ATTR_IN_PORT:
769 case ODP_KEY_ATTR_ETHERNET:
770 case ODP_KEY_ATTR_8021Q:
773 case ODP_KEY_ATTR_ETHERTYPE:
774 if (swkey->dl_type == htons(ETH_P_IP) ||
775 swkey->dl_type == htons(ETH_P_ARP))
779 case ODP_KEY_ATTR_IPV4:
780 if (swkey->nw_proto == IPPROTO_TCP ||
781 swkey->nw_proto == IPPROTO_UDP ||
782 swkey->nw_proto == IPPROTO_ICMP)
786 case ODP_KEY_ATTR_IPV6:
787 if (swkey->nw_proto == IPPROTO_TCP ||
788 swkey->nw_proto == IPPROTO_UDP ||
789 swkey->nw_proto == IPPROTO_ICMPV6)
793 case ODP_KEY_ATTR_ICMPV6:
794 if (swkey->tp_src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
795 swkey->tp_src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
799 case ODP_KEY_ATTR_TCP:
800 case ODP_KEY_ATTR_UDP:
801 case ODP_KEY_ATTR_ICMP:
802 case ODP_KEY_ATTR_ARP:
803 case ODP_KEY_ATTR_ND:
811 int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
813 struct odp_key_ethernet *eth_key;
816 if (swkey->tun_id != cpu_to_be64(0))
817 NLA_PUT_BE64(skb, ODP_KEY_ATTR_TUN_ID, swkey->tun_id);
819 NLA_PUT_U32(skb, ODP_KEY_ATTR_IN_PORT, swkey->in_port);
821 nla = nla_reserve(skb, ODP_KEY_ATTR_ETHERNET, sizeof(*eth_key));
823 goto nla_put_failure;
824 eth_key = nla_data(nla);
825 memcpy(eth_key->eth_src, swkey->dl_src, ETH_ALEN);
826 memcpy(eth_key->eth_dst, swkey->dl_dst, ETH_ALEN);
828 if (swkey->dl_tci != htons(0)) {
829 struct odp_key_8021q q_key;
831 q_key.q_tpid = htons(ETH_P_8021Q);
832 q_key.q_tci = swkey->dl_tci & ~htons(VLAN_TAG_PRESENT);
833 NLA_PUT(skb, ODP_KEY_ATTR_8021Q, sizeof(q_key), &q_key);
836 if (swkey->dl_type == htons(ETH_P_802_2))
839 NLA_PUT_BE16(skb, ODP_KEY_ATTR_ETHERTYPE, swkey->dl_type);
841 if (swkey->dl_type == htons(ETH_P_IP)) {
842 struct odp_key_ipv4 *ipv4_key;
844 nla = nla_reserve(skb, ODP_KEY_ATTR_IPV4, sizeof(*ipv4_key));
846 goto nla_put_failure;
847 ipv4_key = nla_data(nla);
848 memset(ipv4_key, 0, sizeof(struct odp_key_ipv4));
849 ipv4_key->ipv4_src = swkey->ipv4_src;
850 ipv4_key->ipv4_dst = swkey->ipv4_dst;
851 ipv4_key->ipv4_proto = swkey->nw_proto;
852 ipv4_key->ipv4_tos = swkey->nw_tos;
853 } else if (swkey->dl_type == htons(ETH_P_IPV6)) {
854 struct odp_key_ipv6 *ipv6_key;
856 nla = nla_reserve(skb, ODP_KEY_ATTR_IPV6, sizeof(*ipv6_key));
858 goto nla_put_failure;
859 ipv6_key = nla_data(nla);
860 memset(ipv6_key, 0, sizeof(struct odp_key_ipv6));
861 memcpy(ipv6_key->ipv6_src, swkey->ipv6_src,
862 sizeof(ipv6_key->ipv6_src));
863 memcpy(ipv6_key->ipv6_dst, swkey->ipv6_dst,
864 sizeof(ipv6_key->ipv6_dst));
865 ipv6_key->ipv6_proto = swkey->nw_proto;
866 ipv6_key->ipv6_tos = swkey->nw_tos;
867 } else if (swkey->dl_type == htons(ETH_P_ARP)) {
868 struct odp_key_arp *arp_key;
870 nla = nla_reserve(skb, ODP_KEY_ATTR_ARP, sizeof(*arp_key));
872 goto nla_put_failure;
873 arp_key = nla_data(nla);
874 memset(arp_key, 0, sizeof(struct odp_key_arp));
875 arp_key->arp_sip = swkey->ipv4_src;
876 arp_key->arp_tip = swkey->ipv4_dst;
877 arp_key->arp_op = htons(swkey->nw_proto);
878 memcpy(arp_key->arp_sha, swkey->arp_sha, ETH_ALEN);
879 memcpy(arp_key->arp_tha, swkey->arp_tha, ETH_ALEN);
882 if (swkey->dl_type == htons(ETH_P_IP)
883 || swkey->dl_type == htons(ETH_P_IPV6)) {
885 if (swkey->nw_proto == IPPROTO_TCP) {
886 struct odp_key_tcp *tcp_key;
888 nla = nla_reserve(skb, ODP_KEY_ATTR_TCP, sizeof(*tcp_key));
890 goto nla_put_failure;
891 tcp_key = nla_data(nla);
892 tcp_key->tcp_src = swkey->tp_src;
893 tcp_key->tcp_dst = swkey->tp_dst;
894 } else if (swkey->nw_proto == IPPROTO_UDP) {
895 struct odp_key_udp *udp_key;
897 nla = nla_reserve(skb, ODP_KEY_ATTR_UDP, sizeof(*udp_key));
899 goto nla_put_failure;
900 udp_key = nla_data(nla);
901 udp_key->udp_src = swkey->tp_src;
902 udp_key->udp_dst = swkey->tp_dst;
903 } else if (swkey->dl_type == htons(ETH_P_IP)
904 && swkey->nw_proto == IPPROTO_ICMP) {
905 struct odp_key_icmp *icmp_key;
907 nla = nla_reserve(skb, ODP_KEY_ATTR_ICMP, sizeof(*icmp_key));
909 goto nla_put_failure;
910 icmp_key = nla_data(nla);
911 icmp_key->icmp_type = ntohs(swkey->tp_src);
912 icmp_key->icmp_code = ntohs(swkey->tp_dst);
913 } else if (swkey->dl_type == htons(ETH_P_IPV6)
914 && swkey->nw_proto == IPPROTO_ICMPV6) {
915 struct odp_key_icmpv6 *icmpv6_key;
917 nla = nla_reserve(skb, ODP_KEY_ATTR_ICMPV6, sizeof(*icmpv6_key));
919 goto nla_put_failure;
920 icmpv6_key = nla_data(nla);
921 icmpv6_key->icmpv6_type = ntohs(swkey->tp_src);
922 icmpv6_key->icmpv6_code = ntohs(swkey->tp_dst);
924 if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION
925 || icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
926 struct odp_key_nd *nd_key;
928 nla = nla_reserve(skb, ODP_KEY_ATTR_ND, sizeof(*nd_key));
930 goto nla_put_failure;
931 nd_key = nla_data(nla);
932 memcpy(nd_key->nd_target, swkey->nd_target,
933 sizeof(nd_key->nd_target));
934 memcpy(nd_key->nd_sll, swkey->arp_sha, ETH_ALEN);
935 memcpy(nd_key->nd_tll, swkey->arp_tha, ETH_ALEN);
946 /* Initializes the flow module.
947 * Returns zero if successful or a negative error code. */
950 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
952 if (flow_cache == NULL)
955 get_random_bytes(&hash_seed, sizeof(hash_seed));
960 /* Uninitializes the flow module. */
963 kmem_cache_destroy(flow_cache);