2 * Copyright (c) 2007-2013 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
21 #include <linux/uaccess.h>
22 #include <linux/netdevice.h>
23 #include <linux/etherdevice.h>
24 #include <linux/if_ether.h>
25 #include <linux/if_vlan.h>
26 #include <net/llc_pdu.h>
27 #include <linux/kernel.h>
28 #include <linux/jhash.h>
29 #include <linux/jiffies.h>
30 #include <linux/llc.h>
31 #include <linux/module.h>
33 #include <linux/rcupdate.h>
34 #include <linux/if_arp.h>
36 #include <linux/ipv6.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/icmp.h>
40 #include <linux/icmpv6.h>
41 #include <linux/rculist.h>
44 #include <net/ndisc.h>
48 static struct kmem_cache *flow_cache;
50 static void ovs_sw_flow_mask_set(struct sw_flow_mask *mask,
51 struct sw_flow_key_range *range, u8 val);
53 static void update_range__(struct sw_flow_match *match,
54 size_t offset, size_t size, bool is_mask)
56 struct sw_flow_key_range *range = NULL;
57 size_t start = offset;
58 size_t end = offset + size;
61 range = &match->range;
63 range = &match->mask->range;
68 if (range->start == range->end) {
74 if (range->start > start)
81 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
83 update_range__(match, offsetof(struct sw_flow_key, field), \
84 sizeof((match)->key->field), is_mask); \
87 (match)->mask->key.field = value; \
89 (match)->key->field = value; \
93 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask) \
95 update_range__(match, offsetof(struct sw_flow_key, field), \
99 memcpy(&(match)->mask->key.field, value_p, len);\
101 memcpy(&(match)->key->field, value_p, len); \
105 void ovs_match_init(struct sw_flow_match *match,
106 struct sw_flow_key *key,
107 struct sw_flow_mask *mask)
109 memset(match, 0, sizeof(*match));
113 memset(key, 0, sizeof(*key));
116 memset(&mask->key, 0, sizeof(mask->key));
117 mask->range.start = mask->range.end = 0;
121 static bool ovs_match_validate(const struct sw_flow_match *match,
122 u64 key_attrs, u64 mask_attrs)
124 u64 key_expected = 1ULL << OVS_KEY_ATTR_ETHERNET;
125 u64 mask_allowed = key_attrs; /* At most allow all key attributes */
127 /* The following mask attributes allowed only if they
128 * pass the validation tests. */
129 mask_allowed &= ~((1ULL << OVS_KEY_ATTR_IPV4)
130 | (1ULL << OVS_KEY_ATTR_IPV6)
131 | (1ULL << OVS_KEY_ATTR_TCP)
132 | (1ULL << OVS_KEY_ATTR_UDP)
133 | (1ULL << OVS_KEY_ATTR_ICMP)
134 | (1ULL << OVS_KEY_ATTR_ICMPV6)
135 | (1ULL << OVS_KEY_ATTR_ARP)
136 | (1ULL << OVS_KEY_ATTR_ND));
138 /* Always allowed mask fields. */
139 mask_allowed |= ((1ULL << OVS_KEY_ATTR_TUNNEL)
140 | (1ULL << OVS_KEY_ATTR_IN_PORT)
141 | (1ULL << OVS_KEY_ATTR_ETHERTYPE));
143 /* Check key attributes. */
144 if (match->key->eth.type == htons(ETH_P_ARP)
145 || match->key->eth.type == htons(ETH_P_RARP)) {
146 key_expected |= 1ULL << OVS_KEY_ATTR_ARP;
147 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
148 mask_allowed |= 1ULL << OVS_KEY_ATTR_ARP;
151 if (match->key->eth.type == htons(ETH_P_IP)) {
152 key_expected |= 1ULL << OVS_KEY_ATTR_IPV4;
153 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
154 mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV4;
156 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
157 if (match->key->ip.proto == IPPROTO_UDP) {
158 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
159 if (match->mask && (match->mask->key.ip.proto == 0xff))
160 mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
163 if (match->key->ip.proto == IPPROTO_TCP) {
164 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
165 if (match->mask && (match->mask->key.ip.proto == 0xff))
166 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
169 if (match->key->ip.proto == IPPROTO_ICMP) {
170 key_expected |= 1ULL << OVS_KEY_ATTR_ICMP;
171 if (match->mask && (match->mask->key.ip.proto == 0xff))
172 mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMP;
177 if (match->key->eth.type == htons(ETH_P_IPV6)) {
178 key_expected |= 1ULL << OVS_KEY_ATTR_IPV6;
179 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
180 mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV6;
182 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
183 if (match->key->ip.proto == IPPROTO_UDP) {
184 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
185 if (match->mask && (match->mask->key.ip.proto == 0xff))
186 mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
189 if (match->key->ip.proto == IPPROTO_TCP) {
190 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
191 if (match->mask && (match->mask->key.ip.proto == 0xff))
192 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
195 if (match->key->ip.proto == IPPROTO_ICMPV6) {
196 key_expected |= 1ULL << OVS_KEY_ATTR_ICMPV6;
197 if (match->mask && (match->mask->key.ip.proto == 0xff))
198 mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMPV6;
200 if (match->key->ipv6.tp.src ==
201 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
202 match->key->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
203 key_expected |= 1ULL << OVS_KEY_ATTR_ND;
204 if (match->mask && (match->mask->key.ipv6.tp.src == htons(0xffff)))
205 mask_allowed |= 1ULL << OVS_KEY_ATTR_ND;
211 if ((key_attrs & key_expected) != key_expected) {
212 /* Key attributes check failed. */
213 OVS_NLERR("Missing expected key attributes (key_attrs=%llx, expected=%llx).\n",
214 key_attrs, key_expected);
218 if ((mask_attrs & mask_allowed) != mask_attrs) {
219 /* Mask attributes check failed. */
220 OVS_NLERR("Contain more than allowed mask fields (mask_attrs=%llx, mask_allowed=%llx).\n",
221 mask_attrs, mask_allowed);
228 static int check_header(struct sk_buff *skb, int len)
230 if (unlikely(skb->len < len))
232 if (unlikely(!pskb_may_pull(skb, len)))
237 static bool arphdr_ok(struct sk_buff *skb)
239 return pskb_may_pull(skb, skb_network_offset(skb) +
240 sizeof(struct arp_eth_header));
243 static int check_iphdr(struct sk_buff *skb)
245 unsigned int nh_ofs = skb_network_offset(skb);
249 err = check_header(skb, nh_ofs + sizeof(struct iphdr));
253 ip_len = ip_hdrlen(skb);
254 if (unlikely(ip_len < sizeof(struct iphdr) ||
255 skb->len < nh_ofs + ip_len))
258 skb_set_transport_header(skb, nh_ofs + ip_len);
262 static bool tcphdr_ok(struct sk_buff *skb)
264 int th_ofs = skb_transport_offset(skb);
267 if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
270 tcp_len = tcp_hdrlen(skb);
271 if (unlikely(tcp_len < sizeof(struct tcphdr) ||
272 skb->len < th_ofs + tcp_len))
278 static bool udphdr_ok(struct sk_buff *skb)
280 return pskb_may_pull(skb, skb_transport_offset(skb) +
281 sizeof(struct udphdr));
284 static bool icmphdr_ok(struct sk_buff *skb)
286 return pskb_may_pull(skb, skb_transport_offset(skb) +
287 sizeof(struct icmphdr));
290 u64 ovs_flow_used_time(unsigned long flow_jiffies)
292 struct timespec cur_ts;
295 ktime_get_ts(&cur_ts);
296 idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
297 cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
298 cur_ts.tv_nsec / NSEC_PER_MSEC;
300 return cur_ms - idle_ms;
303 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
305 unsigned int nh_ofs = skb_network_offset(skb);
313 err = check_header(skb, nh_ofs + sizeof(*nh));
318 nexthdr = nh->nexthdr;
319 payload_ofs = (u8 *)(nh + 1) - skb->data;
321 key->ip.proto = NEXTHDR_NONE;
322 key->ip.tos = ipv6_get_dsfield(nh);
323 key->ip.ttl = nh->hop_limit;
324 key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
325 key->ipv6.addr.src = nh->saddr;
326 key->ipv6.addr.dst = nh->daddr;
328 payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
329 if (unlikely(payload_ofs < 0))
333 if (frag_off & htons(~0x7))
334 key->ip.frag = OVS_FRAG_TYPE_LATER;
336 key->ip.frag = OVS_FRAG_TYPE_FIRST;
339 nh_len = payload_ofs - nh_ofs;
340 skb_set_transport_header(skb, nh_ofs + nh_len);
341 key->ip.proto = nexthdr;
345 static bool icmp6hdr_ok(struct sk_buff *skb)
347 return pskb_may_pull(skb, skb_transport_offset(skb) +
348 sizeof(struct icmp6hdr));
351 void ovs_flow_key_mask(struct sw_flow_key *dst, const struct sw_flow_key *src,
352 const struct sw_flow_mask *mask)
354 u8 *m = (u8 *)&mask->key + mask->range.start;
355 u8 *s = (u8 *)src + mask->range.start;
356 u8 *d = (u8 *)dst + mask->range.start;
359 memset(dst, 0, sizeof(*dst));
360 for (i = 0; i < ovs_sw_flow_mask_size_roundup(mask); i++) {
366 #define TCP_FLAGS_OFFSET 13
367 #define TCP_FLAG_MASK 0x3f
369 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
373 if ((flow->key.eth.type == htons(ETH_P_IP) ||
374 flow->key.eth.type == htons(ETH_P_IPV6)) &&
375 flow->key.ip.proto == IPPROTO_TCP &&
376 likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
377 u8 *tcp = (u8 *)tcp_hdr(skb);
378 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
381 spin_lock(&flow->lock);
382 flow->used = jiffies;
383 flow->packet_count++;
384 flow->byte_count += skb->len;
385 flow->tcp_flags |= tcp_flags;
386 spin_unlock(&flow->lock);
389 struct sw_flow_actions *ovs_flow_actions_alloc(int size)
391 struct sw_flow_actions *sfa;
393 if (size > MAX_ACTIONS_BUFSIZE)
394 return ERR_PTR(-EINVAL);
396 sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
398 return ERR_PTR(-ENOMEM);
400 sfa->actions_len = 0;
404 struct sw_flow *ovs_flow_alloc(void)
406 struct sw_flow *flow;
408 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
410 return ERR_PTR(-ENOMEM);
412 spin_lock_init(&flow->lock);
413 flow->sf_acts = NULL;
419 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
421 hash = jhash_1word(hash, table->hash_seed);
422 return flex_array_get(table->buckets,
423 (hash & (table->n_buckets - 1)));
426 static struct flex_array *alloc_buckets(unsigned int n_buckets)
428 struct flex_array *buckets;
431 buckets = flex_array_alloc(sizeof(struct hlist_head),
432 n_buckets, GFP_KERNEL);
436 err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
438 flex_array_free(buckets);
442 for (i = 0; i < n_buckets; i++)
443 INIT_HLIST_HEAD((struct hlist_head *)
444 flex_array_get(buckets, i));
449 static void free_buckets(struct flex_array *buckets)
451 flex_array_free(buckets);
454 static struct flow_table *__flow_tbl_alloc(int new_size)
456 struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
461 table->buckets = alloc_buckets(new_size);
463 if (!table->buckets) {
467 table->n_buckets = new_size;
470 table->keep_flows = false;
471 get_random_bytes(&table->hash_seed, sizeof(u32));
472 table->mask_list = NULL;
477 static void __flow_tbl_destroy(struct flow_table *table)
481 if (table->keep_flows)
484 for (i = 0; i < table->n_buckets; i++) {
485 struct sw_flow *flow;
486 struct hlist_head *head = flex_array_get(table->buckets, i);
487 struct hlist_node *n;
488 int ver = table->node_ver;
490 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
491 hlist_del(&flow->hash_node[ver]);
492 ovs_flow_free(flow, false);
496 BUG_ON(!list_empty(table->mask_list));
497 kfree(table->mask_list);
500 free_buckets(table->buckets);
504 struct flow_table *ovs_flow_tbl_alloc(int new_size)
506 struct flow_table *table = __flow_tbl_alloc(new_size);
511 table->mask_list = kmalloc(sizeof(struct list_head), GFP_KERNEL);
512 if (!table->mask_list) {
513 table->keep_flows = true;
514 __flow_tbl_destroy(table);
517 INIT_LIST_HEAD(table->mask_list);
522 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
524 struct flow_table *table = container_of(rcu, struct flow_table, rcu);
526 __flow_tbl_destroy(table);
529 void ovs_flow_tbl_destroy(struct flow_table *table, bool deferred)
535 call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
537 __flow_tbl_destroy(table);
540 struct sw_flow *ovs_flow_dump_next(struct flow_table *table, u32 *bucket, u32 *last)
542 struct sw_flow *flow;
543 struct hlist_head *head;
547 ver = table->node_ver;
548 while (*bucket < table->n_buckets) {
550 head = flex_array_get(table->buckets, *bucket);
551 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
566 static void __tbl_insert(struct flow_table *table, struct sw_flow *flow)
568 struct hlist_head *head;
570 head = find_bucket(table, flow->hash);
571 hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
576 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
581 old_ver = old->node_ver;
582 new->node_ver = !old_ver;
584 /* Insert in new table. */
585 for (i = 0; i < old->n_buckets; i++) {
586 struct sw_flow *flow;
587 struct hlist_head *head;
589 head = flex_array_get(old->buckets, i);
591 hlist_for_each_entry(flow, head, hash_node[old_ver])
592 __tbl_insert(new, flow);
595 new->mask_list = old->mask_list;
596 old->keep_flows = true;
599 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
601 struct flow_table *new_table;
603 new_table = __flow_tbl_alloc(n_buckets);
605 return ERR_PTR(-ENOMEM);
607 flow_table_copy_flows(table, new_table);
612 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
614 return __flow_tbl_rehash(table, table->n_buckets);
617 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
619 return __flow_tbl_rehash(table, table->n_buckets * 2);
622 static void __flow_free(struct sw_flow *flow)
624 kfree((struct sf_flow_acts __force *)flow->sf_acts);
625 kmem_cache_free(flow_cache, flow);
628 static void rcu_free_flow_callback(struct rcu_head *rcu)
630 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
635 void ovs_flow_free(struct sw_flow *flow, bool deferred)
640 ovs_sw_flow_mask_del_ref(flow->mask, deferred);
643 call_rcu(&flow->rcu, rcu_free_flow_callback);
648 /* RCU callback used by ovs_flow_deferred_free_acts. */
649 static void rcu_free_acts_callback(struct rcu_head *rcu)
651 struct sw_flow_actions *sf_acts = container_of(rcu,
652 struct sw_flow_actions, rcu);
656 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
657 * The caller must hold rcu_read_lock for this to be sensible. */
658 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
660 call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
663 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
666 __be16 eth_type; /* ETH_P_8021Q */
669 struct qtag_prefix *qp;
671 if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
674 if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
678 qp = (struct qtag_prefix *) skb->data;
679 key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
680 __skb_pull(skb, sizeof(struct qtag_prefix));
685 static __be16 parse_ethertype(struct sk_buff *skb)
687 struct llc_snap_hdr {
688 u8 dsap; /* Always 0xAA */
689 u8 ssap; /* Always 0xAA */
694 struct llc_snap_hdr *llc;
697 proto = *(__be16 *) skb->data;
698 __skb_pull(skb, sizeof(__be16));
700 if (ntohs(proto) >= ETH_P_802_3_MIN)
703 if (skb->len < sizeof(struct llc_snap_hdr))
704 return htons(ETH_P_802_2);
706 if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
709 llc = (struct llc_snap_hdr *) skb->data;
710 if (llc->dsap != LLC_SAP_SNAP ||
711 llc->ssap != LLC_SAP_SNAP ||
712 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
713 return htons(ETH_P_802_2);
715 __skb_pull(skb, sizeof(struct llc_snap_hdr));
717 if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
718 return llc->ethertype;
720 return htons(ETH_P_802_2);
723 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
726 struct icmp6hdr *icmp = icmp6_hdr(skb);
728 /* The ICMPv6 type and code fields use the 16-bit transport port
729 * fields, so we need to store them in 16-bit network byte order.
731 key->ipv6.tp.src = htons(icmp->icmp6_type);
732 key->ipv6.tp.dst = htons(icmp->icmp6_code);
734 if (icmp->icmp6_code == 0 &&
735 (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
736 icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
737 int icmp_len = skb->len - skb_transport_offset(skb);
741 /* In order to process neighbor discovery options, we need the
744 if (unlikely(icmp_len < sizeof(*nd)))
747 if (unlikely(skb_linearize(skb)))
750 nd = (struct nd_msg *)skb_transport_header(skb);
751 key->ipv6.nd.target = nd->target;
753 icmp_len -= sizeof(*nd);
755 while (icmp_len >= 8) {
756 struct nd_opt_hdr *nd_opt =
757 (struct nd_opt_hdr *)(nd->opt + offset);
758 int opt_len = nd_opt->nd_opt_len * 8;
760 if (unlikely(!opt_len || opt_len > icmp_len))
763 /* Store the link layer address if the appropriate
764 * option is provided. It is considered an error if
765 * the same link layer option is specified twice.
767 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
769 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
771 memcpy(key->ipv6.nd.sll,
772 &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
773 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
775 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
777 memcpy(key->ipv6.nd.tll,
778 &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
789 memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
790 memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
791 memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
797 * ovs_flow_extract - extracts a flow key from an Ethernet frame.
798 * @skb: sk_buff that contains the frame, with skb->data pointing to the
800 * @in_port: port number on which @skb was received.
801 * @key: output flow key
803 * The caller must ensure that skb->len >= ETH_HLEN.
805 * Returns 0 if successful, otherwise a negative errno value.
807 * Initializes @skb header pointers as follows:
809 * - skb->mac_header: the Ethernet header.
811 * - skb->network_header: just past the Ethernet header, or just past the
812 * VLAN header, to the first byte of the Ethernet payload.
814 * - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
815 * on output, then just past the IP header, if one is present and
816 * of a correct length, otherwise the same as skb->network_header.
817 * For other key->eth.type values it is left untouched.
819 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
824 memset(key, 0, sizeof(*key));
826 key->phy.priority = skb->priority;
827 if (OVS_CB(skb)->tun_key)
828 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
829 key->phy.in_port = in_port;
830 key->phy.skb_mark = skb_get_mark(skb);
832 skb_reset_mac_header(skb);
834 /* Link layer. We are guaranteed to have at least the 14 byte Ethernet
835 * header in the linear data area.
838 memcpy(key->eth.src, eth->h_source, ETH_ALEN);
839 memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
841 __skb_pull(skb, 2 * ETH_ALEN);
842 /* We are going to push all headers that we pull, so no need to
843 * update skb->csum here. */
845 if (vlan_tx_tag_present(skb))
846 key->eth.tci = htons(vlan_get_tci(skb));
847 else if (eth->h_proto == htons(ETH_P_8021Q))
848 if (unlikely(parse_vlan(skb, key)))
851 key->eth.type = parse_ethertype(skb);
852 if (unlikely(key->eth.type == htons(0)))
855 skb_reset_network_header(skb);
856 __skb_push(skb, skb->data - skb_mac_header(skb));
859 if (key->eth.type == htons(ETH_P_IP)) {
863 error = check_iphdr(skb);
864 if (unlikely(error)) {
865 if (error == -EINVAL) {
866 skb->transport_header = skb->network_header;
873 key->ipv4.addr.src = nh->saddr;
874 key->ipv4.addr.dst = nh->daddr;
876 key->ip.proto = nh->protocol;
877 key->ip.tos = nh->tos;
878 key->ip.ttl = nh->ttl;
880 offset = nh->frag_off & htons(IP_OFFSET);
882 key->ip.frag = OVS_FRAG_TYPE_LATER;
885 if (nh->frag_off & htons(IP_MF) ||
886 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
887 key->ip.frag = OVS_FRAG_TYPE_FIRST;
889 /* Transport layer. */
890 if (key->ip.proto == IPPROTO_TCP) {
891 if (tcphdr_ok(skb)) {
892 struct tcphdr *tcp = tcp_hdr(skb);
893 key->ipv4.tp.src = tcp->source;
894 key->ipv4.tp.dst = tcp->dest;
896 } else if (key->ip.proto == IPPROTO_UDP) {
897 if (udphdr_ok(skb)) {
898 struct udphdr *udp = udp_hdr(skb);
899 key->ipv4.tp.src = udp->source;
900 key->ipv4.tp.dst = udp->dest;
902 } else if (key->ip.proto == IPPROTO_ICMP) {
903 if (icmphdr_ok(skb)) {
904 struct icmphdr *icmp = icmp_hdr(skb);
905 /* The ICMP type and code fields use the 16-bit
906 * transport port fields, so we need to store
907 * them in 16-bit network byte order. */
908 key->ipv4.tp.src = htons(icmp->type);
909 key->ipv4.tp.dst = htons(icmp->code);
913 } else if ((key->eth.type == htons(ETH_P_ARP) ||
914 key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
915 struct arp_eth_header *arp;
917 arp = (struct arp_eth_header *)skb_network_header(skb);
919 if (arp->ar_hrd == htons(ARPHRD_ETHER)
920 && arp->ar_pro == htons(ETH_P_IP)
921 && arp->ar_hln == ETH_ALEN
922 && arp->ar_pln == 4) {
924 /* We only match on the lower 8 bits of the opcode. */
925 if (ntohs(arp->ar_op) <= 0xff)
926 key->ip.proto = ntohs(arp->ar_op);
927 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
928 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
929 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
930 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
932 } else if (key->eth.type == htons(ETH_P_IPV6)) {
933 int nh_len; /* IPv6 Header + Extensions */
935 nh_len = parse_ipv6hdr(skb, key);
936 if (unlikely(nh_len < 0)) {
937 if (nh_len == -EINVAL) {
938 skb->transport_header = skb->network_header;
946 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
948 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
949 key->ip.frag = OVS_FRAG_TYPE_FIRST;
951 /* Transport layer. */
952 if (key->ip.proto == NEXTHDR_TCP) {
953 if (tcphdr_ok(skb)) {
954 struct tcphdr *tcp = tcp_hdr(skb);
955 key->ipv6.tp.src = tcp->source;
956 key->ipv6.tp.dst = tcp->dest;
958 } else if (key->ip.proto == NEXTHDR_UDP) {
959 if (udphdr_ok(skb)) {
960 struct udphdr *udp = udp_hdr(skb);
961 key->ipv6.tp.src = udp->source;
962 key->ipv6.tp.dst = udp->dest;
964 } else if (key->ip.proto == NEXTHDR_ICMP) {
965 if (icmp6hdr_ok(skb)) {
966 error = parse_icmpv6(skb, key, nh_len);
976 static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start, int key_len)
978 return jhash2((u32 *)((u8 *)key + key_start),
979 DIV_ROUND_UP(key_len - key_start, sizeof(u32)), 0);
982 static int flow_key_start(const struct sw_flow_key *key)
984 if (key->tun_key.ipv4_dst)
987 return offsetof(struct sw_flow_key, phy);
990 static bool __cmp_key(const struct sw_flow_key *key1,
991 const struct sw_flow_key *key2, int key_start, int key_len)
993 return !memcmp((u8 *)key1 + key_start,
994 (u8 *)key2 + key_start, (key_len - key_start));
997 static bool __flow_cmp_key(const struct sw_flow *flow,
998 const struct sw_flow_key *key, int key_start, int key_len)
1000 return __cmp_key(&flow->key, key, key_start, key_len);
1003 static bool __flow_cmp_unmasked_key(const struct sw_flow *flow,
1004 const struct sw_flow_key *key, int key_start, int key_len)
1006 return __cmp_key(&flow->unmasked_key, key, key_start, key_len);
1009 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
1010 const struct sw_flow_key *key, int key_len)
1013 key_start = flow_key_start(key);
1015 return __flow_cmp_unmasked_key(flow, key, key_start, key_len);
1019 struct sw_flow *ovs_flow_lookup_unmasked_key(struct flow_table *table,
1020 struct sw_flow_match *match)
1022 struct sw_flow_key *unmasked = match->key;
1023 int key_len = match->range.end;
1024 struct sw_flow *flow;
1026 flow = ovs_flow_lookup(table, unmasked);
1027 if (flow && (!ovs_flow_cmp_unmasked_key(flow, unmasked, key_len)))
1033 static struct sw_flow *ovs_masked_flow_lookup(struct flow_table *table,
1034 const struct sw_flow_key *flow_key,
1035 struct sw_flow_mask *mask)
1037 struct sw_flow *flow;
1038 struct hlist_head *head;
1039 int key_start = mask->range.start;
1040 int key_len = mask->range.end;
1042 struct sw_flow_key masked_key;
1044 ovs_flow_key_mask(&masked_key, flow_key, mask);
1045 hash = ovs_flow_hash(&masked_key, key_start, key_len);
1046 head = find_bucket(table, hash);
1047 hlist_for_each_entry_rcu(flow, head, hash_node[table->node_ver]) {
1048 if (flow->mask == mask &&
1049 __flow_cmp_key(flow, &masked_key, key_start, key_len))
1055 struct sw_flow *ovs_flow_lookup(struct flow_table *tbl,
1056 const struct sw_flow_key *key)
1058 struct sw_flow *flow = NULL;
1059 struct sw_flow_mask *mask;
1061 list_for_each_entry_rcu(mask, tbl->mask_list, list) {
1062 flow = ovs_masked_flow_lookup(tbl, key, mask);
1063 if (flow) /* Found */
1071 void ovs_flow_insert(struct flow_table *table, struct sw_flow *flow)
1073 flow->hash = ovs_flow_hash(&flow->key, flow->mask->range.start,
1074 flow->mask->range.end);
1075 __tbl_insert(table, flow);
1078 void ovs_flow_remove(struct flow_table *table, struct sw_flow *flow)
1080 BUG_ON(table->count == 0);
1081 hlist_del_rcu(&flow->hash_node[table->node_ver]);
1085 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute. */
1086 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
1087 [OVS_KEY_ATTR_ENCAP] = -1,
1088 [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
1089 [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
1090 [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
1091 [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
1092 [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
1093 [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
1094 [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
1095 [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
1096 [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
1097 [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
1098 [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
1099 [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
1100 [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
1101 [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
1102 [OVS_KEY_ATTR_TUNNEL] = -1,
1105 static bool is_all_zero(const u8 *fp, size_t size)
1112 for (i = 0; i < size; i++)
1119 static int __parse_flow_nlattrs(const struct nlattr *attr,
1120 const struct nlattr *a[],
1121 u64 *attrsp, bool nz)
1123 const struct nlattr *nla;
1128 nla_for_each_nested(nla, attr, rem) {
1129 u16 type = nla_type(nla);
1132 if (type > OVS_KEY_ATTR_MAX) {
1133 OVS_NLERR("Unknown key attribute (type=%d, max=%d).\n",
1134 type, OVS_KEY_ATTR_MAX);
1137 if (attrs & (1ULL << type)) {
1138 OVS_NLERR("Duplicate key attribute (type %d).\n", type);
1142 expected_len = ovs_key_lens[type];
1143 if (nla_len(nla) != expected_len && expected_len != -1) {
1144 OVS_NLERR("Key attribute has unexpected length (type=%d"
1145 ", length=%d, expected=%d).\n", type,
1146 nla_len(nla), expected_len);
1150 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
1151 attrs |= 1ULL << type;
1156 OVS_NLERR("Message has %d unknown bytes.\n", rem);
1164 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
1165 const struct nlattr *a[], u64 *attrsp)
1167 return __parse_flow_nlattrs(attr, a, attrsp, true);
1170 static int parse_flow_nlattrs(const struct nlattr *attr,
1171 const struct nlattr *a[], u64 *attrsp)
1173 return __parse_flow_nlattrs(attr, a, attrsp, false);
1176 int ovs_ipv4_tun_from_nlattr(const struct nlattr *attr,
1177 struct sw_flow_match *match, bool is_mask)
1182 __be16 tun_flags = 0;
1184 nla_for_each_nested(a, attr, rem) {
1185 int type = nla_type(a);
1186 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
1187 [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
1188 [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
1189 [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
1190 [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
1191 [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
1192 [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
1193 [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
1196 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
1197 OVS_NLERR("Unknown IPv4 tunnel attribute (type=%d, max=%d).\n",
1198 type, OVS_TUNNEL_KEY_ATTR_MAX);
1202 if (ovs_tunnel_key_lens[type] != nla_len(a)) {
1203 OVS_NLERR("IPv4 tunnel attribute type has unexpected "
1204 " length (type=%d, length=%d, expected=%d).\n",
1205 type, nla_len(a), ovs_tunnel_key_lens[type]);
1210 case OVS_TUNNEL_KEY_ATTR_ID:
1211 SW_FLOW_KEY_PUT(match, tun_key.tun_id,
1212 nla_get_be64(a), is_mask);
1213 tun_flags |= TUNNEL_KEY;
1215 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1216 SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
1217 nla_get_be32(a), is_mask);
1219 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1220 SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
1221 nla_get_be32(a), is_mask);
1223 case OVS_TUNNEL_KEY_ATTR_TOS:
1224 SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
1225 nla_get_u8(a), is_mask);
1227 case OVS_TUNNEL_KEY_ATTR_TTL:
1228 SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
1229 nla_get_u8(a), is_mask);
1232 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1233 tun_flags |= TUNNEL_DONT_FRAGMENT;
1235 case OVS_TUNNEL_KEY_ATTR_CSUM:
1236 tun_flags |= TUNNEL_CSUM;
1243 SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
1246 OVS_NLERR("IPv4 tunnel attribute has %d unknown bytes.\n", rem);
1251 if (!match->key->tun_key.ipv4_dst) {
1252 OVS_NLERR("IPv4 tunnel destination address is zero.\n");
1257 OVS_NLERR("IPv4 tunnel TTL not specified.\n");
1265 int ovs_ipv4_tun_to_nlattr(struct sk_buff *skb,
1266 const struct ovs_key_ipv4_tunnel *tun_key,
1267 const struct ovs_key_ipv4_tunnel *output)
1271 nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
1275 if (output->tun_flags & TUNNEL_KEY &&
1276 nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
1278 if (output->ipv4_src &&
1279 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
1281 if (output->ipv4_dst &&
1282 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
1284 if (output->ipv4_tos &&
1285 nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
1287 if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
1289 if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
1290 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
1292 if ((output->tun_flags & TUNNEL_CSUM) &&
1293 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
1296 nla_nest_end(skb, nla);
1301 static int metadata_from_nlattrs(struct sw_flow_match *match, u64 *attrs,
1302 const struct nlattr **a, bool is_mask)
1304 if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
1305 SW_FLOW_KEY_PUT(match, phy.priority,
1306 nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1307 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
1310 if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
1311 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1314 in_port = 0xffffffff; /* Always exact match in_port. */
1315 else if (in_port >= DP_MAX_PORTS)
1318 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1319 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
1320 } else if (!is_mask) {
1321 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1324 if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
1325 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1326 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1327 if (!is_mask && mark != 0) {
1328 OVS_NLERR("skb->mark must be zero on this kernel (mark=%d).\n", mark);
1332 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1333 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
1335 if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
1336 if (ovs_ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1339 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
1344 static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs,
1345 const struct nlattr **a, bool is_mask)
1348 u64 orig_attrs = attrs;
1350 err = metadata_from_nlattrs(match, &attrs, a, is_mask);
1354 if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
1355 const struct ovs_key_ethernet *eth_key;
1357 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1358 SW_FLOW_KEY_MEMCPY(match, eth.src,
1359 eth_key->eth_src, ETH_ALEN, is_mask);
1360 SW_FLOW_KEY_MEMCPY(match, eth.dst,
1361 eth_key->eth_dst, ETH_ALEN, is_mask);
1362 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
1365 if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
1368 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1369 if (!(tci & htons(VLAN_TAG_PRESENT))) {
1371 OVS_NLERR("VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.\n");
1373 OVS_NLERR("VLAN TCI does not have VLAN_TAG_PRESENT bit set.\n");
1378 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
1379 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
1380 } else if (!is_mask)
1381 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1383 if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
1386 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1388 /* Always exact match EtherType. */
1389 eth_type = htons(0xffff);
1390 } else if (ntohs(eth_type) < ETH_P_802_3_MIN) {
1391 OVS_NLERR("EtherType is less than minimum (type=%x, min=%x).\n",
1392 ntohs(eth_type), ETH_P_802_3_MIN);
1396 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1397 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1398 } else if (!is_mask) {
1399 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1402 if (attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1403 const struct ovs_key_ipv4 *ipv4_key;
1405 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1406 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1407 OVS_NLERR("Unknown IPv4 fragment type (value=%d, max=%d).\n",
1408 ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1411 SW_FLOW_KEY_PUT(match, ip.proto,
1412 ipv4_key->ipv4_proto, is_mask);
1413 SW_FLOW_KEY_PUT(match, ip.tos,
1414 ipv4_key->ipv4_tos, is_mask);
1415 SW_FLOW_KEY_PUT(match, ip.ttl,
1416 ipv4_key->ipv4_ttl, is_mask);
1417 SW_FLOW_KEY_PUT(match, ip.frag,
1418 ipv4_key->ipv4_frag, is_mask);
1419 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1420 ipv4_key->ipv4_src, is_mask);
1421 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1422 ipv4_key->ipv4_dst, is_mask);
1423 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4);
1426 if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
1427 const struct ovs_key_ipv6 *ipv6_key;
1429 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1430 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1431 OVS_NLERR("Unknown IPv6 fragment type (value=%d, max=%d).\n",
1432 ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1435 SW_FLOW_KEY_PUT(match, ipv6.label,
1436 ipv6_key->ipv6_label, is_mask);
1437 SW_FLOW_KEY_PUT(match, ip.proto,
1438 ipv6_key->ipv6_proto, is_mask);
1439 SW_FLOW_KEY_PUT(match, ip.tos,
1440 ipv6_key->ipv6_tclass, is_mask);
1441 SW_FLOW_KEY_PUT(match, ip.ttl,
1442 ipv6_key->ipv6_hlimit, is_mask);
1443 SW_FLOW_KEY_PUT(match, ip.frag,
1444 ipv6_key->ipv6_frag, is_mask);
1445 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1447 sizeof(match->key->ipv6.addr.src),
1449 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1451 sizeof(match->key->ipv6.addr.dst),
1454 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
1457 if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
1458 const struct ovs_key_arp *arp_key;
1460 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1461 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1462 OVS_NLERR("Unknown ARP opcode (opcode=%d).\n",
1467 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1468 arp_key->arp_sip, is_mask);
1469 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1470 arp_key->arp_tip, is_mask);
1471 SW_FLOW_KEY_PUT(match, ip.proto,
1472 ntohs(arp_key->arp_op), is_mask);
1473 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1474 arp_key->arp_sha, ETH_ALEN, is_mask);
1475 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1476 arp_key->arp_tha, ETH_ALEN, is_mask);
1478 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
1481 if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
1482 const struct ovs_key_tcp *tcp_key;
1484 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1485 if (orig_attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1486 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1487 tcp_key->tcp_src, is_mask);
1488 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1489 tcp_key->tcp_dst, is_mask);
1491 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1492 tcp_key->tcp_src, is_mask);
1493 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1494 tcp_key->tcp_dst, is_mask);
1496 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
1499 if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
1500 const struct ovs_key_udp *udp_key;
1502 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1503 if (orig_attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1504 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1505 udp_key->udp_src, is_mask);
1506 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1507 udp_key->udp_dst, is_mask);
1509 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1510 udp_key->udp_src, is_mask);
1511 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1512 udp_key->udp_dst, is_mask);
1514 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
1517 if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
1518 const struct ovs_key_icmp *icmp_key;
1520 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1521 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1522 htons(icmp_key->icmp_type), is_mask);
1523 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1524 htons(icmp_key->icmp_code), is_mask);
1525 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
1528 if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
1529 const struct ovs_key_icmpv6 *icmpv6_key;
1531 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1532 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1533 htons(icmpv6_key->icmpv6_type), is_mask);
1534 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1535 htons(icmpv6_key->icmpv6_code), is_mask);
1536 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
1539 if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
1540 const struct ovs_key_nd *nd_key;
1542 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1543 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1545 sizeof(match->key->ipv6.nd.target),
1547 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1548 nd_key->nd_sll, ETH_ALEN, is_mask);
1549 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1550 nd_key->nd_tll, ETH_ALEN, is_mask);
1551 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
1561 * ovs_match_from_nlattrs - parses Netlink attributes into a flow key and
1562 * mask. In case the 'mask' is NULL, the flow is treated as exact match
1563 * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1564 * does not include any don't care bit.
1565 * @match: receives the extracted flow match information.
1566 * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1567 * sequence. The fields should of the packet that triggered the creation
1569 * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1570 * attribute specifies the mask field of the wildcarded flow.
1572 int ovs_match_from_nlattrs(struct sw_flow_match *match,
1573 const struct nlattr *key,
1574 const struct nlattr *mask)
1576 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1577 const struct nlattr *encap;
1580 bool encap_valid = false;
1583 err = parse_flow_nlattrs(key, a, &key_attrs);
1587 if (key_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
1588 encap = a[OVS_KEY_ATTR_ENCAP];
1589 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1590 if (nla_len(encap)) {
1591 __be16 eth_type = 0; /* ETH_P_8021Q */
1593 if (a[OVS_KEY_ATTR_ETHERTYPE])
1594 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1596 if ((eth_type == htons(ETH_P_8021Q)) && (a[OVS_KEY_ATTR_VLAN])) {
1598 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1599 err = parse_flow_nlattrs(encap, a, &key_attrs);
1601 OVS_NLERR("Encap attribute is set for a non-VLAN frame.\n");
1610 err = ovs_key_from_nlattrs(match, key_attrs, a, false);
1615 err = parse_flow_mask_nlattrs(mask, a, &mask_attrs);
1619 if ((mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) && encap_valid) {
1620 __be16 eth_type = 0;
1622 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1623 if (a[OVS_KEY_ATTR_ETHERTYPE])
1624 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1625 if (eth_type == htons(0xffff)) {
1626 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1627 encap = a[OVS_KEY_ATTR_ENCAP];
1628 err = parse_flow_mask_nlattrs(encap, a, &mask_attrs);
1630 OVS_NLERR("VLAN frames must have an exact match"
1631 " on the TPID (mask=%x).\n",
1640 err = ovs_key_from_nlattrs(match, mask_attrs, a, true);
1644 /* Populate exact match flow's key mask. */
1646 ovs_sw_flow_mask_set(match->mask, &match->range, 0xff);
1649 if (!ovs_match_validate(match, key_attrs, mask_attrs))
1656 * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1657 * @flow: Receives extracted in_port, priority, tun_key and skb_mark.
1658 * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1661 * This parses a series of Netlink attributes that form a flow key, which must
1662 * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1663 * get the metadata, that is, the parts of the flow key that cannot be
1664 * extracted from the packet itself.
1667 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow,
1668 const struct nlattr *attr)
1670 struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1671 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1674 struct sw_flow_match match;
1676 flow->key.phy.in_port = DP_MAX_PORTS;
1677 flow->key.phy.priority = 0;
1678 flow->key.phy.skb_mark = 0;
1679 memset(tun_key, 0, sizeof(flow->key.tun_key));
1681 err = parse_flow_nlattrs(attr, a, &attrs);
1685 memset(&match, 0, sizeof(match));
1686 match.key = &flow->key;
1688 err = metadata_from_nlattrs(&match, &attrs, a, false);
1695 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey,
1696 const struct sw_flow_key *output, struct sk_buff *skb)
1698 struct ovs_key_ethernet *eth_key;
1699 struct nlattr *nla, *encap;
1700 bool is_mask = (swkey != output);
1702 if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1703 goto nla_put_failure;
1705 if ((swkey->tun_key.ipv4_dst || is_mask) &&
1706 ovs_ipv4_tun_to_nlattr(skb, &swkey->tun_key, &output->tun_key))
1707 goto nla_put_failure;
1709 if (swkey->phy.in_port == DP_MAX_PORTS) {
1710 if (is_mask && (output->phy.in_port == 0xffff))
1711 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1712 goto nla_put_failure;
1715 upper_u16 = !is_mask ? 0 : 0xffff;
1717 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1718 (upper_u16 << 16) | output->phy.in_port))
1719 goto nla_put_failure;
1722 if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1723 goto nla_put_failure;
1725 nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1727 goto nla_put_failure;
1729 eth_key = nla_data(nla);
1730 memcpy(eth_key->eth_src, output->eth.src, ETH_ALEN);
1731 memcpy(eth_key->eth_dst, output->eth.dst, ETH_ALEN);
1733 if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1735 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1736 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1737 nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1738 goto nla_put_failure;
1739 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1740 if (!swkey->eth.tci)
1745 if (swkey->eth.type == htons(ETH_P_802_2)) {
1747 * Ethertype 802.2 is represented in the netlink with omitted
1748 * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1749 * 0xffff in the mask attribute. Ethertype can also
1752 if (is_mask && output->eth.type)
1753 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1755 goto nla_put_failure;
1759 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1760 goto nla_put_failure;
1762 if (swkey->eth.type == htons(ETH_P_IP)) {
1763 struct ovs_key_ipv4 *ipv4_key;
1765 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1767 goto nla_put_failure;
1768 ipv4_key = nla_data(nla);
1769 ipv4_key->ipv4_src = output->ipv4.addr.src;
1770 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1771 ipv4_key->ipv4_proto = output->ip.proto;
1772 ipv4_key->ipv4_tos = output->ip.tos;
1773 ipv4_key->ipv4_ttl = output->ip.ttl;
1774 ipv4_key->ipv4_frag = output->ip.frag;
1775 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1776 struct ovs_key_ipv6 *ipv6_key;
1778 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1780 goto nla_put_failure;
1781 ipv6_key = nla_data(nla);
1782 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1783 sizeof(ipv6_key->ipv6_src));
1784 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1785 sizeof(ipv6_key->ipv6_dst));
1786 ipv6_key->ipv6_label = output->ipv6.label;
1787 ipv6_key->ipv6_proto = output->ip.proto;
1788 ipv6_key->ipv6_tclass = output->ip.tos;
1789 ipv6_key->ipv6_hlimit = output->ip.ttl;
1790 ipv6_key->ipv6_frag = output->ip.frag;
1791 } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1792 swkey->eth.type == htons(ETH_P_RARP)) {
1793 struct ovs_key_arp *arp_key;
1795 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1797 goto nla_put_failure;
1798 arp_key = nla_data(nla);
1799 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1800 arp_key->arp_sip = output->ipv4.addr.src;
1801 arp_key->arp_tip = output->ipv4.addr.dst;
1802 arp_key->arp_op = htons(output->ip.proto);
1803 memcpy(arp_key->arp_sha, output->ipv4.arp.sha, ETH_ALEN);
1804 memcpy(arp_key->arp_tha, output->ipv4.arp.tha, ETH_ALEN);
1807 if ((swkey->eth.type == htons(ETH_P_IP) ||
1808 swkey->eth.type == htons(ETH_P_IPV6)) &&
1809 swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1811 if (swkey->ip.proto == IPPROTO_TCP) {
1812 struct ovs_key_tcp *tcp_key;
1814 nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1816 goto nla_put_failure;
1817 tcp_key = nla_data(nla);
1818 if (swkey->eth.type == htons(ETH_P_IP)) {
1819 tcp_key->tcp_src = output->ipv4.tp.src;
1820 tcp_key->tcp_dst = output->ipv4.tp.dst;
1821 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1822 tcp_key->tcp_src = output->ipv6.tp.src;
1823 tcp_key->tcp_dst = output->ipv6.tp.dst;
1825 } else if (swkey->ip.proto == IPPROTO_UDP) {
1826 struct ovs_key_udp *udp_key;
1828 nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1830 goto nla_put_failure;
1831 udp_key = nla_data(nla);
1832 if (swkey->eth.type == htons(ETH_P_IP)) {
1833 udp_key->udp_src = output->ipv4.tp.src;
1834 udp_key->udp_dst = output->ipv4.tp.dst;
1835 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1836 udp_key->udp_src = output->ipv6.tp.src;
1837 udp_key->udp_dst = output->ipv6.tp.dst;
1839 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1840 swkey->ip.proto == IPPROTO_ICMP) {
1841 struct ovs_key_icmp *icmp_key;
1843 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1845 goto nla_put_failure;
1846 icmp_key = nla_data(nla);
1847 icmp_key->icmp_type = ntohs(output->ipv4.tp.src);
1848 icmp_key->icmp_code = ntohs(output->ipv4.tp.dst);
1849 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1850 swkey->ip.proto == IPPROTO_ICMPV6) {
1851 struct ovs_key_icmpv6 *icmpv6_key;
1853 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1854 sizeof(*icmpv6_key));
1856 goto nla_put_failure;
1857 icmpv6_key = nla_data(nla);
1858 icmpv6_key->icmpv6_type = ntohs(output->ipv6.tp.src);
1859 icmpv6_key->icmpv6_code = ntohs(output->ipv6.tp.dst);
1861 if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1862 icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1863 struct ovs_key_nd *nd_key;
1865 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1867 goto nla_put_failure;
1868 nd_key = nla_data(nla);
1869 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1870 sizeof(nd_key->nd_target));
1871 memcpy(nd_key->nd_sll, output->ipv6.nd.sll, ETH_ALEN);
1872 memcpy(nd_key->nd_tll, output->ipv6.nd.tll, ETH_ALEN);
1879 nla_nest_end(skb, encap);
1887 /* Initializes the flow module.
1888 * Returns zero if successful or a negative error code. */
1889 int ovs_flow_init(void)
1891 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1893 if (flow_cache == NULL)
1899 /* Uninitializes the flow module. */
1900 void ovs_flow_exit(void)
1902 kmem_cache_destroy(flow_cache);
1905 struct sw_flow_mask *ovs_sw_flow_mask_alloc(void)
1907 struct sw_flow_mask *mask;
1909 mask = kmalloc(sizeof(*mask), GFP_KERNEL);
1911 mask->ref_count = 0;
1916 void ovs_sw_flow_mask_add_ref(struct sw_flow_mask *mask)
1921 static void rcu_free_sw_flow_mask_cb(struct rcu_head *rcu)
1923 struct sw_flow_mask *mask = container_of(rcu, struct sw_flow_mask, rcu);
1928 void ovs_sw_flow_mask_del_ref(struct sw_flow_mask *mask, bool deferred)
1933 BUG_ON(!mask->ref_count);
1936 if (!mask->ref_count) {
1937 list_del_rcu(&mask->list);
1939 call_rcu(&mask->rcu, rcu_free_sw_flow_mask_cb);
1945 static bool ovs_sw_flow_mask_equal(const struct sw_flow_mask *a,
1946 const struct sw_flow_mask *b)
1948 u8 *a_ = (u8 *)&a->key + a->range.start;
1949 u8 *b_ = (u8 *)&b->key + b->range.start;
1951 return (a->range.end == b->range.end)
1952 && (a->range.start == b->range.start)
1953 && (memcmp(a_, b_, ovs_sw_flow_mask_actual_size(a)) == 0);
1956 struct sw_flow_mask *ovs_sw_flow_mask_find(const struct flow_table *tbl,
1957 const struct sw_flow_mask *mask)
1959 struct list_head *ml;
1961 list_for_each(ml, tbl->mask_list) {
1962 struct sw_flow_mask *m;
1963 m = container_of(ml, struct sw_flow_mask, list);
1964 if (ovs_sw_flow_mask_equal(mask, m))
1972 * add a new mask into the mask list.
1973 * The caller needs to make sure that 'mask' is not the same
1974 * as any masks that are already on the list.
1976 void ovs_sw_flow_mask_insert(struct flow_table *tbl, struct sw_flow_mask *mask)
1978 list_add_rcu(&mask->list, tbl->mask_list);
1982 * Set 'range' fields in the mask to the value of 'val'.
1984 static void ovs_sw_flow_mask_set(struct sw_flow_mask *mask,
1985 struct sw_flow_key_range *range, u8 val)
1987 u8 *m = (u8 *)&mask->key + range->start;
1989 mask->range = *range;
1990 memset(m, val, ovs_sw_flow_mask_size_roundup(mask));