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 <linux/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 <linux/rculist.h>
35 #include <net/ndisc.h>
39 static struct kmem_cache *flow_cache;
40 static unsigned int hash_seed __read_mostly;
42 static int check_header(struct sk_buff *skb, int len)
44 if (unlikely(skb->len < len))
46 if (unlikely(!pskb_may_pull(skb, len)))
51 static bool arphdr_ok(struct sk_buff *skb)
53 return pskb_may_pull(skb, skb_network_offset(skb) +
54 sizeof(struct arp_eth_header));
57 static int check_iphdr(struct sk_buff *skb)
59 unsigned int nh_ofs = skb_network_offset(skb);
63 err = check_header(skb, nh_ofs + sizeof(struct iphdr));
67 ip_len = ip_hdrlen(skb);
68 if (unlikely(ip_len < sizeof(struct iphdr) ||
69 skb->len < nh_ofs + ip_len))
72 skb_set_transport_header(skb, nh_ofs + ip_len);
76 static bool tcphdr_ok(struct sk_buff *skb)
78 int th_ofs = skb_transport_offset(skb);
81 if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
84 tcp_len = tcp_hdrlen(skb);
85 if (unlikely(tcp_len < sizeof(struct tcphdr) ||
86 skb->len < th_ofs + tcp_len))
92 static bool udphdr_ok(struct sk_buff *skb)
94 return pskb_may_pull(skb, skb_transport_offset(skb) +
95 sizeof(struct udphdr));
98 static bool icmphdr_ok(struct sk_buff *skb)
100 return pskb_may_pull(skb, skb_transport_offset(skb) +
101 sizeof(struct icmphdr));
104 u64 flow_used_time(unsigned long flow_jiffies)
106 struct timespec cur_ts;
109 ktime_get_ts(&cur_ts);
110 idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
111 cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
112 cur_ts.tv_nsec / NSEC_PER_MSEC;
114 return cur_ms - idle_ms;
117 #define SW_FLOW_KEY_OFFSET(field) \
118 (offsetof(struct sw_flow_key, field) + \
119 FIELD_SIZEOF(struct sw_flow_key, field))
122 * skip_exthdr - skip any IPv6 extension headers
123 * @skb: skbuff to parse
124 * @start: offset of first extension header
125 * @nexthdrp: Initially, points to the type of the extension header at @start.
126 * This function updates it to point to the extension header at the final
128 * @frag: Points to the @frag member in a &struct sw_flow_key. This
129 * function sets an appropriate %OVS_FRAG_TYPE_* value.
131 * This is based on ipv6_skip_exthdr() but adds the updates to *@frag.
133 * When there is more than one fragment header, this version reports whether
134 * the final fragment header that it examines is a first fragment.
136 * Returns the final payload offset, or -1 on error.
138 static int skip_exthdr(const struct sk_buff *skb, int start, u8 *nexthdrp,
141 u8 nexthdr = *nexthdrp;
143 while (ipv6_ext_hdr(nexthdr)) {
144 struct ipv6_opt_hdr _hdr, *hp;
147 if (nexthdr == NEXTHDR_NONE)
149 hp = skb_header_pointer(skb, start, sizeof(_hdr), &_hdr);
152 if (nexthdr == NEXTHDR_FRAGMENT) {
153 __be16 _frag_off, *fp;
154 fp = skb_header_pointer(skb,
155 start+offsetof(struct frag_hdr,
162 if (ntohs(*fp) & ~0x7) {
163 *frag = OVS_FRAG_TYPE_LATER;
166 *frag = OVS_FRAG_TYPE_FIRST;
168 } else if (nexthdr == NEXTHDR_AUTH)
169 hdrlen = (hp->hdrlen+2)<<2;
171 hdrlen = ipv6_optlen(hp);
173 nexthdr = hp->nexthdr;
181 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
184 unsigned int nh_ofs = skb_network_offset(skb);
191 *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.label);
193 err = check_header(skb, nh_ofs + sizeof(*nh));
198 nexthdr = nh->nexthdr;
199 payload_ofs = (u8 *)(nh + 1) - skb->data;
201 key->ip.proto = NEXTHDR_NONE;
202 key->ip.tos = ipv6_get_dsfield(nh);
203 key->ip.ttl = nh->hop_limit;
204 key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
205 ipv6_addr_copy(&key->ipv6.addr.src, &nh->saddr);
206 ipv6_addr_copy(&key->ipv6.addr.dst, &nh->daddr);
208 payload_ofs = skip_exthdr(skb, payload_ofs, &nexthdr, &key->ip.frag);
209 if (unlikely(payload_ofs < 0))
212 nh_len = payload_ofs - nh_ofs;
213 skb_set_transport_header(skb, nh_ofs + nh_len);
214 key->ip.proto = nexthdr;
218 static bool icmp6hdr_ok(struct sk_buff *skb)
220 return pskb_may_pull(skb, skb_transport_offset(skb) +
221 sizeof(struct icmp6hdr));
224 #define TCP_FLAGS_OFFSET 13
225 #define TCP_FLAG_MASK 0x3f
227 void flow_used(struct sw_flow *flow, struct sk_buff *skb)
231 if (flow->key.eth.type == htons(ETH_P_IP) &&
232 flow->key.ip.proto == IPPROTO_TCP) {
233 u8 *tcp = (u8 *)tcp_hdr(skb);
234 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
237 spin_lock(&flow->lock);
238 flow->used = jiffies;
239 flow->packet_count++;
240 flow->byte_count += skb->len;
241 flow->tcp_flags |= tcp_flags;
242 spin_unlock(&flow->lock);
245 struct sw_flow_actions *flow_actions_alloc(const struct nlattr *actions)
247 int actions_len = nla_len(actions);
248 struct sw_flow_actions *sfa;
250 /* At least DP_MAX_PORTS actions are required to be able to flood a
251 * packet to every port. Factor of 2 allows for setting VLAN tags,
253 if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
254 return ERR_PTR(-EINVAL);
256 sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
258 return ERR_PTR(-ENOMEM);
260 sfa->actions_len = actions_len;
261 memcpy(sfa->actions, nla_data(actions), actions_len);
265 struct sw_flow *flow_alloc(void)
267 struct sw_flow *flow;
269 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
271 return ERR_PTR(-ENOMEM);
273 spin_lock_init(&flow->lock);
274 atomic_set(&flow->refcnt, 1);
275 flow->sf_acts = NULL;
281 static struct hlist_head __rcu *find_bucket(struct flow_table * table, u32 hash)
283 return flex_array_get(table->buckets,
284 (hash & (table->n_buckets - 1)));
287 static struct flex_array __rcu *alloc_buckets(unsigned int n_buckets)
289 struct flex_array __rcu *buckets;
292 buckets = flex_array_alloc(sizeof(struct hlist_head *),
293 n_buckets, GFP_KERNEL);
297 err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
299 flex_array_free(buckets);
303 for (i = 0; i < n_buckets; i++)
304 INIT_HLIST_HEAD((struct hlist_head *)
305 flex_array_get(buckets, i));
310 static void free_buckets(struct flex_array *buckets)
312 flex_array_free(buckets);
315 struct flow_table *flow_tbl_alloc(int new_size)
317 struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
322 table->buckets = alloc_buckets(new_size);
324 if (!table->buckets) {
328 table->n_buckets = new_size;
334 static void flow_free(struct sw_flow *flow)
340 void flow_tbl_destroy(struct flow_table *table)
347 for (i = 0; i < table->n_buckets; i++) {
348 struct sw_flow *flow;
349 struct hlist_head *head = flex_array_get(table->buckets, i);
350 struct hlist_node *node, *n;
352 hlist_for_each_entry_safe(flow, node, n, head, hash_node) {
353 hlist_del_init_rcu(&flow->hash_node);
358 free_buckets(table->buckets);
362 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
364 struct flow_table *table = container_of(rcu, struct flow_table, rcu);
366 flow_tbl_destroy(table);
369 void flow_tbl_deferred_destroy(struct flow_table *table)
374 call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
377 struct sw_flow *flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
379 struct sw_flow *flow;
380 struct hlist_head *head;
381 struct hlist_node *n;
384 while (*bucket < table->n_buckets) {
386 head = flex_array_get(table->buckets, *bucket);
387 hlist_for_each_entry_rcu(flow, n, head, hash_node) {
402 struct flow_table *flow_tbl_expand(struct flow_table *table)
404 struct flow_table *new_table;
405 int n_buckets = table->n_buckets * 2;
408 new_table = flow_tbl_alloc(n_buckets);
410 return ERR_PTR(-ENOMEM);
412 for (i = 0; i < table->n_buckets; i++) {
413 struct sw_flow *flow;
414 struct hlist_head *head;
415 struct hlist_node *n, *pos;
417 head = flex_array_get(table->buckets, i);
419 hlist_for_each_entry_safe(flow, n, pos, head, hash_node) {
420 hlist_del_init_rcu(&flow->hash_node);
421 flow_tbl_insert(new_table, flow);
428 /* RCU callback used by flow_deferred_free. */
429 static void rcu_free_flow_callback(struct rcu_head *rcu)
431 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
437 /* Schedules 'flow' to be freed after the next RCU grace period.
438 * The caller must hold rcu_read_lock for this to be sensible. */
439 void flow_deferred_free(struct sw_flow *flow)
441 call_rcu(&flow->rcu, rcu_free_flow_callback);
444 void flow_hold(struct sw_flow *flow)
446 atomic_inc(&flow->refcnt);
449 void flow_put(struct sw_flow *flow)
454 if (atomic_dec_and_test(&flow->refcnt)) {
455 kfree((struct sf_flow_acts __force *)flow->sf_acts);
456 kmem_cache_free(flow_cache, flow);
460 /* RCU callback used by flow_deferred_free_acts. */
461 static void rcu_free_acts_callback(struct rcu_head *rcu)
463 struct sw_flow_actions *sf_acts = container_of(rcu,
464 struct sw_flow_actions, rcu);
468 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
469 * The caller must hold rcu_read_lock for this to be sensible. */
470 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
472 call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
475 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
478 __be16 eth_type; /* ETH_P_8021Q */
481 struct qtag_prefix *qp;
483 if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
486 if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
490 qp = (struct qtag_prefix *) skb->data;
491 key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
492 __skb_pull(skb, sizeof(struct qtag_prefix));
497 static __be16 parse_ethertype(struct sk_buff *skb)
499 struct llc_snap_hdr {
500 u8 dsap; /* Always 0xAA */
501 u8 ssap; /* Always 0xAA */
506 struct llc_snap_hdr *llc;
509 proto = *(__be16 *) skb->data;
510 __skb_pull(skb, sizeof(__be16));
512 if (ntohs(proto) >= 1536)
515 if (skb->len < sizeof(struct llc_snap_hdr))
516 return htons(ETH_P_802_2);
518 if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
521 llc = (struct llc_snap_hdr *) skb->data;
522 if (llc->dsap != LLC_SAP_SNAP ||
523 llc->ssap != LLC_SAP_SNAP ||
524 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
525 return htons(ETH_P_802_2);
527 __skb_pull(skb, sizeof(struct llc_snap_hdr));
528 return llc->ethertype;
531 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
532 int *key_lenp, int nh_len)
534 struct icmp6hdr *icmp = icmp6_hdr(skb);
538 /* The ICMPv6 type and code fields use the 16-bit transport port
539 * fields, so we need to store them in 16-bit network byte order.
541 key->ipv6.tp.src = htons(icmp->icmp6_type);
542 key->ipv6.tp.dst = htons(icmp->icmp6_code);
543 key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
545 if (icmp->icmp6_code == 0 &&
546 (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
547 icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
548 int icmp_len = skb->len - skb_transport_offset(skb);
552 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
554 /* In order to process neighbor discovery options, we need the
557 if (unlikely(icmp_len < sizeof(*nd)))
559 if (unlikely(skb_linearize(skb))) {
564 nd = (struct nd_msg *)skb_transport_header(skb);
565 ipv6_addr_copy(&key->ipv6.nd.target, &nd->target);
566 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
568 icmp_len -= sizeof(*nd);
570 while (icmp_len >= 8) {
571 struct nd_opt_hdr *nd_opt =
572 (struct nd_opt_hdr *)(nd->opt + offset);
573 int opt_len = nd_opt->nd_opt_len * 8;
575 if (unlikely(!opt_len || opt_len > icmp_len))
578 /* Store the link layer address if the appropriate
579 * option is provided. It is considered an error if
580 * the same link layer option is specified twice.
582 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
584 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
586 memcpy(key->ipv6.nd.sll,
587 &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
588 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
590 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
592 memcpy(key->ipv6.nd.tll,
593 &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
604 memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
605 memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
606 memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
614 * flow_extract - extracts a flow key from an Ethernet frame.
615 * @skb: sk_buff that contains the frame, with skb->data pointing to the
617 * @in_port: port number on which @skb was received.
618 * @key: output flow key
619 * @key_lenp: length of output flow key
621 * The caller must ensure that skb->len >= ETH_HLEN.
623 * Returns 0 if successful, otherwise a negative errno value.
625 * Initializes @skb header pointers as follows:
627 * - skb->mac_header: the Ethernet header.
629 * - skb->network_header: just past the Ethernet header, or just past the
630 * VLAN header, to the first byte of the Ethernet payload.
632 * - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
633 * on output, then just past the IP header, if one is present and
634 * of a correct length, otherwise the same as skb->network_header.
635 * For other key->dl_type values it is left untouched.
637 int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
641 int key_len = SW_FLOW_KEY_OFFSET(eth);
644 memset(key, 0, sizeof(*key));
646 key->phy.priority = skb->priority;
647 key->phy.tun_id = OVS_CB(skb)->tun_id;
648 key->phy.in_port = in_port;
650 skb_reset_mac_header(skb);
652 /* Link layer. We are guaranteed to have at least the 14 byte Ethernet
653 * header in the linear data area.
656 memcpy(key->eth.src, eth->h_source, ETH_ALEN);
657 memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
659 __skb_pull(skb, 2 * ETH_ALEN);
661 if (vlan_tx_tag_present(skb))
662 key->eth.tci = htons(vlan_get_tci(skb));
663 else if (eth->h_proto == htons(ETH_P_8021Q))
664 if (unlikely(parse_vlan(skb, key)))
667 key->eth.type = parse_ethertype(skb);
668 if (unlikely(key->eth.type == htons(0)))
671 skb_reset_network_header(skb);
672 __skb_push(skb, skb->data - skb_mac_header(skb));
675 if (key->eth.type == htons(ETH_P_IP)) {
679 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
681 error = check_iphdr(skb);
682 if (unlikely(error)) {
683 if (error == -EINVAL) {
684 skb->transport_header = skb->network_header;
691 key->ipv4.addr.src = nh->saddr;
692 key->ipv4.addr.dst = nh->daddr;
694 key->ip.proto = nh->protocol;
695 key->ip.tos = nh->tos;
696 key->ip.ttl = nh->ttl;
698 offset = nh->frag_off & htons(IP_OFFSET);
700 key->ip.frag = OVS_FRAG_TYPE_LATER;
703 if (nh->frag_off & htons(IP_MF) ||
704 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
705 key->ip.frag = OVS_FRAG_TYPE_FIRST;
707 /* Transport layer. */
708 if (key->ip.proto == IPPROTO_TCP) {
709 key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
710 if (tcphdr_ok(skb)) {
711 struct tcphdr *tcp = tcp_hdr(skb);
712 key->ipv4.tp.src = tcp->source;
713 key->ipv4.tp.dst = tcp->dest;
715 } else if (key->ip.proto == IPPROTO_UDP) {
716 key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
717 if (udphdr_ok(skb)) {
718 struct udphdr *udp = udp_hdr(skb);
719 key->ipv4.tp.src = udp->source;
720 key->ipv4.tp.dst = udp->dest;
722 } else if (key->ip.proto == IPPROTO_ICMP) {
723 key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
724 if (icmphdr_ok(skb)) {
725 struct icmphdr *icmp = icmp_hdr(skb);
726 /* The ICMP type and code fields use the 16-bit
727 * transport port fields, so we need to store
728 * them in 16-bit network byte order. */
729 key->ipv4.tp.src = htons(icmp->type);
730 key->ipv4.tp.dst = htons(icmp->code);
734 } else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
735 struct arp_eth_header *arp;
737 arp = (struct arp_eth_header *)skb_network_header(skb);
739 if (arp->ar_hrd == htons(ARPHRD_ETHER)
740 && arp->ar_pro == htons(ETH_P_IP)
741 && arp->ar_hln == ETH_ALEN
742 && arp->ar_pln == 4) {
744 /* We only match on the lower 8 bits of the opcode. */
745 if (ntohs(arp->ar_op) <= 0xff)
746 key->ip.proto = ntohs(arp->ar_op);
748 if (key->ip.proto == ARPOP_REQUEST
749 || key->ip.proto == ARPOP_REPLY) {
750 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
751 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
752 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
753 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
754 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
757 } else if (key->eth.type == htons(ETH_P_IPV6)) {
758 int nh_len; /* IPv6 Header + Extensions */
760 nh_len = parse_ipv6hdr(skb, key, &key_len);
761 if (unlikely(nh_len < 0)) {
762 if (nh_len == -EINVAL)
763 skb->transport_header = skb->network_header;
769 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
771 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
772 key->ip.frag = OVS_FRAG_TYPE_FIRST;
774 /* Transport layer. */
775 if (key->ip.proto == NEXTHDR_TCP) {
776 key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
777 if (tcphdr_ok(skb)) {
778 struct tcphdr *tcp = tcp_hdr(skb);
779 key->ipv6.tp.src = tcp->source;
780 key->ipv6.tp.dst = tcp->dest;
782 } else if (key->ip.proto == NEXTHDR_UDP) {
783 key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
784 if (udphdr_ok(skb)) {
785 struct udphdr *udp = udp_hdr(skb);
786 key->ipv6.tp.src = udp->source;
787 key->ipv6.tp.dst = udp->dest;
789 } else if (key->ip.proto == NEXTHDR_ICMP) {
790 key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
791 if (icmp6hdr_ok(skb)) {
792 error = parse_icmpv6(skb, key, &key_len, nh_len);
804 u32 flow_hash(const struct sw_flow_key *key, int key_len)
806 return jhash2((u32 *)key, DIV_ROUND_UP(key_len, sizeof(u32)), hash_seed);
809 struct sw_flow *flow_tbl_lookup(struct flow_table *table,
810 struct sw_flow_key *key, int key_len)
812 struct sw_flow *flow;
813 struct hlist_node *n;
814 struct hlist_head *head;
817 hash = flow_hash(key, key_len);
819 head = find_bucket(table, hash);
820 hlist_for_each_entry_rcu(flow, n, head, hash_node) {
822 if (flow->hash == hash &&
823 !memcmp(&flow->key, key, key_len)) {
830 void flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
832 struct hlist_head *head;
834 head = find_bucket(table, flow->hash);
835 hlist_add_head_rcu(&flow->hash_node, head);
839 void flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
841 if (!hlist_unhashed(&flow->hash_node)) {
842 hlist_del_init_rcu(&flow->hash_node);
844 BUG_ON(table->count < 0);
848 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute. */
849 const u32 ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
850 [OVS_KEY_ATTR_ENCAP] = 0,
851 [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
852 [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
853 [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
854 [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
855 [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
856 [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
857 [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
858 [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
859 [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
860 [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
861 [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
862 [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
863 [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
866 [OVS_KEY_ATTR_TUN_ID] = sizeof(__be64),
869 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
870 const struct nlattr *a[], u64 *attrs)
872 const struct ovs_key_icmp *icmp_key;
873 const struct ovs_key_tcp *tcp_key;
874 const struct ovs_key_udp *udp_key;
876 switch (swkey->ip.proto) {
878 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
880 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
882 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
883 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
884 swkey->ipv4.tp.src = tcp_key->tcp_src;
885 swkey->ipv4.tp.dst = tcp_key->tcp_dst;
889 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
891 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
893 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
894 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
895 swkey->ipv4.tp.src = udp_key->udp_src;
896 swkey->ipv4.tp.dst = udp_key->udp_dst;
900 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
902 *attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
904 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
905 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
906 swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
907 swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
914 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
915 const struct nlattr *a[], u64 *attrs)
917 const struct ovs_key_icmpv6 *icmpv6_key;
918 const struct ovs_key_tcp *tcp_key;
919 const struct ovs_key_udp *udp_key;
921 switch (swkey->ip.proto) {
923 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
925 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
927 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
928 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
929 swkey->ipv6.tp.src = tcp_key->tcp_src;
930 swkey->ipv6.tp.dst = tcp_key->tcp_dst;
934 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
936 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
938 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
939 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
940 swkey->ipv6.tp.src = udp_key->udp_src;
941 swkey->ipv6.tp.dst = udp_key->udp_dst;
945 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
947 *attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
949 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
950 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
951 swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
952 swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
954 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
955 swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
956 const struct ovs_key_nd *nd_key;
958 if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
960 *attrs &= ~(1 << OVS_KEY_ATTR_ND);
962 *key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
963 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
964 memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
965 sizeof(swkey->ipv6.nd.target));
966 memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
967 memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
975 static int parse_flow_nlattrs(const struct nlattr *attr,
976 const struct nlattr *a[], u64 *attrsp)
978 const struct nlattr *nla;
983 nla_for_each_nested(nla, attr, rem) {
984 u16 type = nla_type(nla);
986 if (type > OVS_KEY_ATTR_MAX || attrs & (1ULL << type) ||
987 nla_len(nla) != ovs_key_lens[type])
989 attrs |= 1ULL << type;
1000 * flow_from_nlattrs - parses Netlink attributes into a flow key.
1001 * @swkey: receives the extracted flow key.
1002 * @key_lenp: number of bytes used in @swkey.
1003 * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1006 int flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
1007 const struct nlattr *attr)
1009 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1010 const struct ovs_key_ethernet *eth_key;
1015 memset(swkey, 0, sizeof(struct sw_flow_key));
1016 key_len = SW_FLOW_KEY_OFFSET(eth);
1018 err = parse_flow_nlattrs(attr, a, &attrs);
1022 /* Metadata attributes. */
1023 if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1024 swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
1025 attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1027 if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1028 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1029 if (in_port >= DP_MAX_PORTS)
1031 swkey->phy.in_port = in_port;
1032 attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1034 swkey->phy.in_port = USHRT_MAX;
1037 if (attrs & (1ULL << OVS_KEY_ATTR_TUN_ID)) {
1038 swkey->phy.tun_id = nla_get_be64(a[OVS_KEY_ATTR_TUN_ID]);
1039 attrs &= ~(1ULL << OVS_KEY_ATTR_TUN_ID);
1042 /* Data attributes. */
1043 if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1045 attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1047 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1048 memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1049 memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1051 if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1052 nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1053 const struct nlattr *encap;
1056 if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1057 (1 << OVS_KEY_ATTR_ETHERTYPE) |
1058 (1 << OVS_KEY_ATTR_ENCAP)))
1061 encap = a[OVS_KEY_ATTR_ENCAP];
1062 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1063 if (tci & htons(VLAN_TAG_PRESENT)) {
1064 swkey->eth.tci = tci;
1066 err = parse_flow_nlattrs(encap, a, &attrs);
1070 /* Corner case for truncated 802.1Q header. */
1074 swkey->eth.type = htons(ETH_P_8021Q);
1075 *key_lenp = key_len;
1082 if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1083 swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1084 if (ntohs(swkey->eth.type) < 1536)
1086 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1088 swkey->eth.type = htons(ETH_P_802_2);
1091 if (swkey->eth.type == htons(ETH_P_IP)) {
1092 const struct ovs_key_ipv4 *ipv4_key;
1094 if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1096 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1098 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1099 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1100 if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1102 swkey->ip.proto = ipv4_key->ipv4_proto;
1103 swkey->ip.tos = ipv4_key->ipv4_tos;
1104 swkey->ip.ttl = ipv4_key->ipv4_ttl;
1105 swkey->ip.frag = ipv4_key->ipv4_frag;
1106 swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1107 swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1109 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1110 err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1114 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1115 const struct ovs_key_ipv6 *ipv6_key;
1117 if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1119 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1121 key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1122 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1123 if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1125 swkey->ipv6.label = ipv6_key->ipv6_label;
1126 swkey->ip.proto = ipv6_key->ipv6_proto;
1127 swkey->ip.tos = ipv6_key->ipv6_tclass;
1128 swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1129 swkey->ip.frag = ipv6_key->ipv6_frag;
1130 memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1131 sizeof(swkey->ipv6.addr.src));
1132 memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1133 sizeof(swkey->ipv6.addr.dst));
1135 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1136 err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1140 } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1141 const struct ovs_key_arp *arp_key;
1143 if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1145 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1147 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1148 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1149 swkey->ipv4.addr.src = arp_key->arp_sip;
1150 swkey->ipv4.addr.dst = arp_key->arp_tip;
1151 if (arp_key->arp_op & htons(0xff00))
1153 swkey->ip.proto = ntohs(arp_key->arp_op);
1154 memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1155 memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1160 *key_lenp = key_len;
1166 * flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1167 * @in_port: receives the extracted input port.
1168 * @tun_id: receives the extracted tunnel ID.
1169 * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1172 * This parses a series of Netlink attributes that form a flow key, which must
1173 * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1174 * get the metadata, that is, the parts of the flow key that cannot be
1175 * extracted from the packet itself.
1177 int flow_metadata_from_nlattrs(u32 *priority, u16 *in_port, __be64 *tun_id,
1178 const struct nlattr *attr)
1180 const struct nlattr *nla;
1183 *in_port = USHRT_MAX;
1187 nla_for_each_nested(nla, attr, rem) {
1188 int type = nla_type(nla);
1190 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] != 0) {
1191 if (nla_len(nla) != ovs_key_lens[type])
1195 case OVS_KEY_ATTR_PRIORITY:
1196 *priority = nla_get_u32(nla);
1199 case OVS_KEY_ATTR_TUN_ID:
1200 *tun_id = nla_get_be64(nla);
1203 case OVS_KEY_ATTR_IN_PORT:
1204 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1206 *in_port = nla_get_u32(nla);
1216 int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1218 struct ovs_key_ethernet *eth_key;
1219 struct nlattr *nla, *encap;
1221 if (swkey->phy.priority)
1222 NLA_PUT_U32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority);
1224 if (swkey->phy.tun_id != cpu_to_be64(0))
1225 NLA_PUT_BE64(skb, OVS_KEY_ATTR_TUN_ID, swkey->phy.tun_id);
1227 if (swkey->phy.in_port != USHRT_MAX)
1228 NLA_PUT_U32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port);
1230 nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1232 goto nla_put_failure;
1233 eth_key = nla_data(nla);
1234 memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1235 memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1237 if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1238 NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q));
1239 NLA_PUT_BE16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci);
1240 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1241 if (!swkey->eth.tci)
1247 if (swkey->eth.type == htons(ETH_P_802_2))
1250 NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type);
1252 if (swkey->eth.type == htons(ETH_P_IP)) {
1253 struct ovs_key_ipv4 *ipv4_key;
1255 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1257 goto nla_put_failure;
1258 ipv4_key = nla_data(nla);
1259 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1260 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1261 ipv4_key->ipv4_proto = swkey->ip.proto;
1262 ipv4_key->ipv4_tos = swkey->ip.tos;
1263 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1264 ipv4_key->ipv4_frag = swkey->ip.frag;
1265 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1266 struct ovs_key_ipv6 *ipv6_key;
1268 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1270 goto nla_put_failure;
1271 ipv6_key = nla_data(nla);
1272 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1273 sizeof(ipv6_key->ipv6_src));
1274 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1275 sizeof(ipv6_key->ipv6_dst));
1276 ipv6_key->ipv6_label = swkey->ipv6.label;
1277 ipv6_key->ipv6_proto = swkey->ip.proto;
1278 ipv6_key->ipv6_tclass = swkey->ip.tos;
1279 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1280 ipv6_key->ipv6_frag = swkey->ip.frag;
1281 } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1282 struct ovs_key_arp *arp_key;
1284 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1286 goto nla_put_failure;
1287 arp_key = nla_data(nla);
1288 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1289 arp_key->arp_sip = swkey->ipv4.addr.src;
1290 arp_key->arp_tip = swkey->ipv4.addr.dst;
1291 arp_key->arp_op = htons(swkey->ip.proto);
1292 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1293 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1296 if ((swkey->eth.type == htons(ETH_P_IP) ||
1297 swkey->eth.type == htons(ETH_P_IPV6)) &&
1298 swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1300 if (swkey->ip.proto == IPPROTO_TCP) {
1301 struct ovs_key_tcp *tcp_key;
1303 nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1305 goto nla_put_failure;
1306 tcp_key = nla_data(nla);
1307 if (swkey->eth.type == htons(ETH_P_IP)) {
1308 tcp_key->tcp_src = swkey->ipv4.tp.src;
1309 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1310 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1311 tcp_key->tcp_src = swkey->ipv6.tp.src;
1312 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1314 } else if (swkey->ip.proto == IPPROTO_UDP) {
1315 struct ovs_key_udp *udp_key;
1317 nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1319 goto nla_put_failure;
1320 udp_key = nla_data(nla);
1321 if (swkey->eth.type == htons(ETH_P_IP)) {
1322 udp_key->udp_src = swkey->ipv4.tp.src;
1323 udp_key->udp_dst = swkey->ipv4.tp.dst;
1324 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1325 udp_key->udp_src = swkey->ipv6.tp.src;
1326 udp_key->udp_dst = swkey->ipv6.tp.dst;
1328 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1329 swkey->ip.proto == IPPROTO_ICMP) {
1330 struct ovs_key_icmp *icmp_key;
1332 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1334 goto nla_put_failure;
1335 icmp_key = nla_data(nla);
1336 icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1337 icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1338 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1339 swkey->ip.proto == IPPROTO_ICMPV6) {
1340 struct ovs_key_icmpv6 *icmpv6_key;
1342 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1343 sizeof(*icmpv6_key));
1345 goto nla_put_failure;
1346 icmpv6_key = nla_data(nla);
1347 icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1348 icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1350 if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1351 icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1352 struct ovs_key_nd *nd_key;
1354 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1356 goto nla_put_failure;
1357 nd_key = nla_data(nla);
1358 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1359 sizeof(nd_key->nd_target));
1360 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1361 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1368 nla_nest_end(skb, encap);
1376 /* Initializes the flow module.
1377 * Returns zero if successful or a negative error code. */
1380 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1382 if (flow_cache == NULL)
1385 get_random_bytes(&hash_seed, sizeof(hash_seed));
1390 /* Uninitializes the flow module. */
1391 void flow_exit(void)
1393 kmem_cache_destroy(flow_cache);