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); \
85 if (is_mask && match->mask != NULL) { \
86 (match)->mask->key.field = value; \
88 (match)->key->field = value; \
92 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask) \
94 update_range__(match, offsetof(struct sw_flow_key, field), \
96 if (is_mask && match->mask != NULL) { \
97 memcpy(&(match)->mask->key.field, value_p, len); \
99 memcpy(&(match)->key->field, value_p, len); \
103 void ovs_match_init(struct sw_flow_match *match,
104 struct sw_flow_key *key,
105 struct sw_flow_mask *mask)
107 memset(match, 0, sizeof(*match));
111 memset(key, 0, sizeof(*key));
114 memset(&mask->key, 0, sizeof(mask->key));
115 mask->range.start = mask->range.end = 0;
119 static bool ovs_match_validate(const struct sw_flow_match *match,
120 u64 key_attrs, u64 mask_attrs)
122 u64 key_expected = 1ULL << OVS_KEY_ATTR_ETHERNET;
123 u64 mask_allowed = key_attrs; /* At most allow all key attributes */
125 /* The following mask attributes allowed only if they
126 * pass the validation tests. */
127 mask_allowed &= ~((1ULL << OVS_KEY_ATTR_IPV4)
128 | (1ULL << OVS_KEY_ATTR_IPV6)
129 | (1ULL << OVS_KEY_ATTR_TCP)
130 | (1ULL << OVS_KEY_ATTR_UDP)
131 | (1ULL << OVS_KEY_ATTR_ICMP)
132 | (1ULL << OVS_KEY_ATTR_ICMPV6)
133 | (1ULL << OVS_KEY_ATTR_ARP)
134 | (1ULL << OVS_KEY_ATTR_ND));
136 if (match->key->phy.in_port == DP_MAX_PORTS &&
137 match->mask && (match->mask->key.phy.in_port == 0xffff))
138 mask_allowed |= (1ULL << OVS_KEY_ATTR_IN_PORT);
140 if (match->key->eth.type == htons(ETH_P_802_2) &&
141 match->mask && (match->mask->key.eth.type == htons(0xffff)))
142 mask_allowed |= (1ULL << OVS_KEY_ATTR_ETHERTYPE);
144 /* Check key attributes. */
145 if (match->key->eth.type == htons(ETH_P_ARP)
146 || match->key->eth.type == htons(ETH_P_RARP)) {
147 key_expected |= 1ULL << OVS_KEY_ATTR_ARP;
148 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
149 mask_allowed |= 1ULL << OVS_KEY_ATTR_ARP;
152 if (match->key->eth.type == htons(ETH_P_IP)) {
153 key_expected |= 1ULL << OVS_KEY_ATTR_IPV4;
154 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
155 mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV4;
157 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
158 if (match->key->ip.proto == IPPROTO_UDP) {
159 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
160 if (match->mask && (match->mask->key.ip.proto == 0xff))
161 mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
164 if (match->key->ip.proto == IPPROTO_TCP) {
165 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
166 if (match->mask && (match->mask->key.ip.proto == 0xff))
167 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
170 if (match->key->ip.proto == IPPROTO_ICMP) {
171 key_expected |= 1ULL << OVS_KEY_ATTR_ICMP;
172 if (match->mask && (match->mask->key.ip.proto == 0xff))
173 mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMP;
178 if (match->key->eth.type == htons(ETH_P_IPV6)) {
179 key_expected |= 1ULL << OVS_KEY_ATTR_IPV6;
180 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
181 mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV6;
183 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
184 if (match->key->ip.proto == IPPROTO_UDP) {
185 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
186 if (match->mask && (match->mask->key.ip.proto == 0xff))
187 mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
190 if (match->key->ip.proto == IPPROTO_TCP) {
191 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
192 if (match->mask && (match->mask->key.ip.proto == 0xff))
193 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
196 if (match->key->ip.proto == IPPROTO_ICMPV6) {
197 key_expected |= 1ULL << OVS_KEY_ATTR_ICMPV6;
198 if (match->mask && (match->mask->key.ip.proto == 0xff))
199 mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMPV6;
201 if (match->key->ipv6.tp.src ==
202 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
203 match->key->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
204 key_expected |= 1ULL << OVS_KEY_ATTR_ND;
205 if (match->mask && (match->mask->key.ipv6.tp.src == htons(0xffff)))
206 mask_allowed |= 1ULL << OVS_KEY_ATTR_ND;
212 if ((key_attrs & key_expected) != key_expected) {
213 /* Key attributes check failed. */
214 OVS_NLERR("Missing expected key attributes (key_attrs=%llx, expected=%llx).\n",
215 key_attrs, key_expected);
219 if ((mask_attrs & mask_allowed) != mask_attrs) {
220 /* Mask attributes check failed. */
221 OVS_NLERR("Contain more than allowed mask fields (mask_attrs=%llx, mask_allowed=%llx).\n",
222 mask_attrs, mask_allowed);
229 static int check_header(struct sk_buff *skb, int len)
231 if (unlikely(skb->len < len))
233 if (unlikely(!pskb_may_pull(skb, len)))
238 static bool arphdr_ok(struct sk_buff *skb)
240 return pskb_may_pull(skb, skb_network_offset(skb) +
241 sizeof(struct arp_eth_header));
244 static int check_iphdr(struct sk_buff *skb)
246 unsigned int nh_ofs = skb_network_offset(skb);
250 err = check_header(skb, nh_ofs + sizeof(struct iphdr));
254 ip_len = ip_hdrlen(skb);
255 if (unlikely(ip_len < sizeof(struct iphdr) ||
256 skb->len < nh_ofs + ip_len))
259 skb_set_transport_header(skb, nh_ofs + ip_len);
263 static bool tcphdr_ok(struct sk_buff *skb)
265 int th_ofs = skb_transport_offset(skb);
268 if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
271 tcp_len = tcp_hdrlen(skb);
272 if (unlikely(tcp_len < sizeof(struct tcphdr) ||
273 skb->len < th_ofs + tcp_len))
279 static bool udphdr_ok(struct sk_buff *skb)
281 return pskb_may_pull(skb, skb_transport_offset(skb) +
282 sizeof(struct udphdr));
285 static bool icmphdr_ok(struct sk_buff *skb)
287 return pskb_may_pull(skb, skb_transport_offset(skb) +
288 sizeof(struct icmphdr));
291 u64 ovs_flow_used_time(unsigned long flow_jiffies)
293 struct timespec cur_ts;
296 ktime_get_ts(&cur_ts);
297 idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
298 cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
299 cur_ts.tv_nsec / NSEC_PER_MSEC;
301 return cur_ms - idle_ms;
304 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
306 unsigned int nh_ofs = skb_network_offset(skb);
314 err = check_header(skb, nh_ofs + sizeof(*nh));
319 nexthdr = nh->nexthdr;
320 payload_ofs = (u8 *)(nh + 1) - skb->data;
322 key->ip.proto = NEXTHDR_NONE;
323 key->ip.tos = ipv6_get_dsfield(nh);
324 key->ip.ttl = nh->hop_limit;
325 key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
326 key->ipv6.addr.src = nh->saddr;
327 key->ipv6.addr.dst = nh->daddr;
329 payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
330 if (unlikely(payload_ofs < 0))
334 if (frag_off & htons(~0x7))
335 key->ip.frag = OVS_FRAG_TYPE_LATER;
337 key->ip.frag = OVS_FRAG_TYPE_FIRST;
340 nh_len = payload_ofs - nh_ofs;
341 skb_set_transport_header(skb, nh_ofs + nh_len);
342 key->ip.proto = nexthdr;
346 static bool icmp6hdr_ok(struct sk_buff *skb)
348 return pskb_may_pull(skb, skb_transport_offset(skb) +
349 sizeof(struct icmp6hdr));
352 void ovs_flow_key_mask(struct sw_flow_key *dst, const struct sw_flow_key *src,
353 const struct sw_flow_mask *mask)
355 u8 *m = (u8 *)&mask->key + mask->range.start;
356 u8 *s = (u8 *)src + mask->range.start;
357 u8 *d = (u8 *)dst + mask->range.start;
360 memset(dst, 0, sizeof(*dst));
361 for (i = 0; i < ovs_sw_flow_mask_size_roundup(mask); i++) {
367 #define TCP_FLAGS_OFFSET 13
368 #define TCP_FLAG_MASK 0x3f
370 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
374 if ((flow->key.eth.type == htons(ETH_P_IP) ||
375 flow->key.eth.type == htons(ETH_P_IPV6)) &&
376 flow->key.ip.proto == IPPROTO_TCP &&
377 likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
378 u8 *tcp = (u8 *)tcp_hdr(skb);
379 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
382 spin_lock(&flow->lock);
383 flow->used = jiffies;
384 flow->packet_count++;
385 flow->byte_count += skb->len;
386 flow->tcp_flags |= tcp_flags;
387 spin_unlock(&flow->lock);
390 struct sw_flow_actions *ovs_flow_actions_alloc(int size)
392 struct sw_flow_actions *sfa;
394 if (size > MAX_ACTIONS_BUFSIZE)
395 return ERR_PTR(-EINVAL);
397 sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
399 return ERR_PTR(-ENOMEM);
401 sfa->actions_len = 0;
405 struct sw_flow *ovs_flow_alloc(void)
407 struct sw_flow *flow;
409 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
411 return ERR_PTR(-ENOMEM);
413 spin_lock_init(&flow->lock);
414 flow->sf_acts = NULL;
420 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
422 hash = jhash_1word(hash, table->hash_seed);
423 return flex_array_get(table->buckets,
424 (hash & (table->n_buckets - 1)));
427 static struct flex_array *alloc_buckets(unsigned int n_buckets)
429 struct flex_array *buckets;
432 buckets = flex_array_alloc(sizeof(struct hlist_head *),
433 n_buckets, GFP_KERNEL);
437 err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
439 flex_array_free(buckets);
443 for (i = 0; i < n_buckets; i++)
444 INIT_HLIST_HEAD((struct hlist_head *)
445 flex_array_get(buckets, i));
450 static void free_buckets(struct flex_array *buckets)
452 flex_array_free(buckets);
455 static struct flow_table *__flow_tbl_alloc(int new_size)
457 struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
462 table->buckets = alloc_buckets(new_size);
464 if (!table->buckets) {
468 table->n_buckets = new_size;
471 table->keep_flows = false;
472 get_random_bytes(&table->hash_seed, sizeof(u32));
473 table->mask_list = NULL;
478 static void __flow_tbl_destroy(struct flow_table *table)
482 if (table->keep_flows)
485 for (i = 0; i < table->n_buckets; i++) {
486 struct sw_flow *flow;
487 struct hlist_head *head = flex_array_get(table->buckets, i);
488 struct hlist_node *n;
489 int ver = table->node_ver;
491 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
492 hlist_del_rcu(&flow->hash_node[ver]);
493 ovs_flow_free(flow, false);
497 BUG_ON(!list_empty(table->mask_list));
498 kfree(table->mask_list);
501 free_buckets(table->buckets);
505 struct flow_table *ovs_flow_tbl_alloc(int new_size)
507 struct flow_table *table = __flow_tbl_alloc(new_size);
512 table->mask_list = kmalloc(sizeof(struct list_head), GFP_KERNEL);
513 if (!table->mask_list) {
514 table->keep_flows = true;
515 __flow_tbl_destroy(table);
518 INIT_LIST_HEAD(table->mask_list);
523 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
525 struct flow_table *table = container_of(rcu, struct flow_table, rcu);
527 __flow_tbl_destroy(table);
530 void ovs_flow_tbl_destroy(struct flow_table *table, bool deferred)
536 call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
538 __flow_tbl_destroy(table);
541 struct sw_flow *ovs_flow_dump_next(struct flow_table *table, u32 *bucket, u32 *last)
543 struct sw_flow *flow;
544 struct hlist_head *head;
548 ver = table->node_ver;
549 while (*bucket < table->n_buckets) {
551 head = flex_array_get(table->buckets, *bucket);
552 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
567 static void __tbl_insert(struct flow_table *table, struct sw_flow *flow)
569 struct hlist_head *head;
571 head = find_bucket(table, flow->hash);
572 hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
577 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
582 old_ver = old->node_ver;
583 new->node_ver = !old_ver;
585 /* Insert in new table. */
586 for (i = 0; i < old->n_buckets; i++) {
587 struct sw_flow *flow;
588 struct hlist_head *head;
590 head = flex_array_get(old->buckets, i);
592 hlist_for_each_entry(flow, head, hash_node[old_ver])
593 __tbl_insert(new, flow);
596 new->mask_list = old->mask_list;
597 old->keep_flows = true;
600 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
602 struct flow_table *new_table;
604 new_table = __flow_tbl_alloc(n_buckets);
606 return ERR_PTR(-ENOMEM);
608 flow_table_copy_flows(table, new_table);
613 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
615 return __flow_tbl_rehash(table, table->n_buckets);
618 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
620 return __flow_tbl_rehash(table, table->n_buckets * 2);
623 static void __flow_free(struct sw_flow *flow)
625 kfree((struct sf_flow_acts __force *)flow->sf_acts);
626 kmem_cache_free(flow_cache, flow);
629 static void rcu_free_flow_callback(struct rcu_head *rcu)
631 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
636 void ovs_flow_free(struct sw_flow *flow, bool deferred)
641 ovs_sw_flow_mask_del_ref((struct sw_flow_mask __force *)flow->mask,
645 call_rcu(&flow->rcu, rcu_free_flow_callback);
650 /* RCU callback used by ovs_flow_deferred_free_acts. */
651 static void rcu_free_acts_callback(struct rcu_head *rcu)
653 struct sw_flow_actions *sf_acts = container_of(rcu,
654 struct sw_flow_actions, rcu);
658 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
659 * The caller must hold rcu_read_lock for this to be sensible. */
660 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
662 call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
665 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
668 __be16 eth_type; /* ETH_P_8021Q */
671 struct qtag_prefix *qp;
673 if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
676 if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
680 qp = (struct qtag_prefix *) skb->data;
681 key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
682 __skb_pull(skb, sizeof(struct qtag_prefix));
687 static __be16 parse_ethertype(struct sk_buff *skb)
689 struct llc_snap_hdr {
690 u8 dsap; /* Always 0xAA */
691 u8 ssap; /* Always 0xAA */
696 struct llc_snap_hdr *llc;
699 proto = *(__be16 *) skb->data;
700 __skb_pull(skb, sizeof(__be16));
702 if (ntohs(proto) >= ETH_P_802_3_MIN)
705 if (skb->len < sizeof(struct llc_snap_hdr))
706 return htons(ETH_P_802_2);
708 if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
711 llc = (struct llc_snap_hdr *) skb->data;
712 if (llc->dsap != LLC_SAP_SNAP ||
713 llc->ssap != LLC_SAP_SNAP ||
714 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
715 return htons(ETH_P_802_2);
717 __skb_pull(skb, sizeof(struct llc_snap_hdr));
719 if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
720 return llc->ethertype;
722 return htons(ETH_P_802_2);
725 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
728 struct icmp6hdr *icmp = icmp6_hdr(skb);
730 /* The ICMPv6 type and code fields use the 16-bit transport port
731 * fields, so we need to store them in 16-bit network byte order.
733 key->ipv6.tp.src = htons(icmp->icmp6_type);
734 key->ipv6.tp.dst = htons(icmp->icmp6_code);
736 if (icmp->icmp6_code == 0 &&
737 (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
738 icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
739 int icmp_len = skb->len - skb_transport_offset(skb);
743 /* In order to process neighbor discovery options, we need the
746 if (unlikely(icmp_len < sizeof(*nd)))
749 if (unlikely(skb_linearize(skb)))
752 nd = (struct nd_msg *)skb_transport_header(skb);
753 key->ipv6.nd.target = nd->target;
755 icmp_len -= sizeof(*nd);
757 while (icmp_len >= 8) {
758 struct nd_opt_hdr *nd_opt =
759 (struct nd_opt_hdr *)(nd->opt + offset);
760 int opt_len = nd_opt->nd_opt_len * 8;
762 if (unlikely(!opt_len || opt_len > icmp_len))
765 /* Store the link layer address if the appropriate
766 * option is provided. It is considered an error if
767 * the same link layer option is specified twice.
769 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
771 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
773 memcpy(key->ipv6.nd.sll,
774 &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
775 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
777 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
779 memcpy(key->ipv6.nd.tll,
780 &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
791 memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
792 memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
793 memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
799 * ovs_flow_extract - extracts a flow key from an Ethernet frame.
800 * @skb: sk_buff that contains the frame, with skb->data pointing to the
802 * @in_port: port number on which @skb was received.
803 * @key: output flow key
804 * @key_lenp: length of output flow key
806 * The caller must ensure that skb->len >= ETH_HLEN.
808 * Returns 0 if successful, otherwise a negative errno value.
810 * Initializes @skb header pointers as follows:
812 * - skb->mac_header: the Ethernet header.
814 * - skb->network_header: just past the Ethernet header, or just past the
815 * VLAN header, to the first byte of the Ethernet payload.
817 * - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
818 * on output, then just past the IP header, if one is present and
819 * of a correct length, otherwise the same as skb->network_header.
820 * For other key->eth.type values it is left untouched.
822 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
827 memset(key, 0, sizeof(*key));
829 key->phy.priority = skb->priority;
830 if (OVS_CB(skb)->tun_key)
831 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
832 key->phy.in_port = in_port;
833 key->phy.skb_mark = skb_get_mark(skb);
835 skb_reset_mac_header(skb);
837 /* Link layer. We are guaranteed to have at least the 14 byte Ethernet
838 * header in the linear data area.
841 memcpy(key->eth.src, eth->h_source, ETH_ALEN);
842 memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
844 __skb_pull(skb, 2 * ETH_ALEN);
845 /* We are going to push all headers that we pull, so no need to
846 * update skb->csum here. */
848 if (vlan_tx_tag_present(skb))
849 key->eth.tci = htons(vlan_get_tci(skb));
850 else if (eth->h_proto == htons(ETH_P_8021Q))
851 if (unlikely(parse_vlan(skb, key)))
854 key->eth.type = parse_ethertype(skb);
855 if (unlikely(key->eth.type == htons(0)))
858 skb_reset_network_header(skb);
859 __skb_push(skb, skb->data - skb_mac_header(skb));
862 if (key->eth.type == htons(ETH_P_IP)) {
866 error = check_iphdr(skb);
867 if (unlikely(error)) {
868 if (error == -EINVAL) {
869 skb->transport_header = skb->network_header;
876 key->ipv4.addr.src = nh->saddr;
877 key->ipv4.addr.dst = nh->daddr;
879 key->ip.proto = nh->protocol;
880 key->ip.tos = nh->tos;
881 key->ip.ttl = nh->ttl;
883 offset = nh->frag_off & htons(IP_OFFSET);
885 key->ip.frag = OVS_FRAG_TYPE_LATER;
888 if (nh->frag_off & htons(IP_MF) ||
889 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
890 key->ip.frag = OVS_FRAG_TYPE_FIRST;
892 /* Transport layer. */
893 if (key->ip.proto == IPPROTO_TCP) {
894 if (tcphdr_ok(skb)) {
895 struct tcphdr *tcp = tcp_hdr(skb);
896 key->ipv4.tp.src = tcp->source;
897 key->ipv4.tp.dst = tcp->dest;
899 } else if (key->ip.proto == IPPROTO_UDP) {
900 if (udphdr_ok(skb)) {
901 struct udphdr *udp = udp_hdr(skb);
902 key->ipv4.tp.src = udp->source;
903 key->ipv4.tp.dst = udp->dest;
905 } else if (key->ip.proto == IPPROTO_ICMP) {
906 if (icmphdr_ok(skb)) {
907 struct icmphdr *icmp = icmp_hdr(skb);
908 /* The ICMP type and code fields use the 16-bit
909 * transport port fields, so we need to store
910 * them in 16-bit network byte order. */
911 key->ipv4.tp.src = htons(icmp->type);
912 key->ipv4.tp.dst = htons(icmp->code);
916 } else if ((key->eth.type == htons(ETH_P_ARP) ||
917 key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
918 struct arp_eth_header *arp;
920 arp = (struct arp_eth_header *)skb_network_header(skb);
922 if (arp->ar_hrd == htons(ARPHRD_ETHER)
923 && arp->ar_pro == htons(ETH_P_IP)
924 && arp->ar_hln == ETH_ALEN
925 && arp->ar_pln == 4) {
927 /* We only match on the lower 8 bits of the opcode. */
928 if (ntohs(arp->ar_op) <= 0xff)
929 key->ip.proto = ntohs(arp->ar_op);
930 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
931 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
932 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
933 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
935 } else if (key->eth.type == htons(ETH_P_IPV6)) {
936 int nh_len; /* IPv6 Header + Extensions */
938 nh_len = parse_ipv6hdr(skb, key);
939 if (unlikely(nh_len < 0)) {
940 if (nh_len == -EINVAL) {
941 skb->transport_header = skb->network_header;
949 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
951 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
952 key->ip.frag = OVS_FRAG_TYPE_FIRST;
954 /* Transport layer. */
955 if (key->ip.proto == NEXTHDR_TCP) {
956 if (tcphdr_ok(skb)) {
957 struct tcphdr *tcp = tcp_hdr(skb);
958 key->ipv6.tp.src = tcp->source;
959 key->ipv6.tp.dst = tcp->dest;
961 } else if (key->ip.proto == NEXTHDR_UDP) {
962 if (udphdr_ok(skb)) {
963 struct udphdr *udp = udp_hdr(skb);
964 key->ipv6.tp.src = udp->source;
965 key->ipv6.tp.dst = udp->dest;
967 } else if (key->ip.proto == NEXTHDR_ICMP) {
968 if (icmp6hdr_ok(skb)) {
969 error = parse_icmpv6(skb, key, nh_len);
979 static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start, int key_len)
981 return jhash2((u32 *)((u8 *)key + key_start),
982 DIV_ROUND_UP(key_len - key_start, sizeof(u32)), 0);
985 static int flow_key_start(const struct sw_flow_key *key)
987 if (key->tun_key.ipv4_dst)
990 return offsetof(struct sw_flow_key, phy);
993 static bool __cmp_key(const struct sw_flow_key *key1,
994 const struct sw_flow_key *key2, int key_start, int key_len)
996 return !memcmp((u8 *)key1 + key_start,
997 (u8 *)key2 + key_start, (key_len - key_start));
1000 static bool __flow_cmp_key(const struct sw_flow *flow,
1001 const struct sw_flow_key *key, int key_start, int key_len)
1003 return __cmp_key(&flow->key, key, key_start, key_len);
1006 static bool __flow_cmp_unmasked_key(const struct sw_flow *flow,
1007 const struct sw_flow_key *key, int key_start, int key_len)
1009 return __cmp_key(&flow->unmasked_key, key, key_start, key_len);
1012 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
1013 const struct sw_flow_key *key, int key_len)
1016 key_start = flow_key_start(key);
1018 return __flow_cmp_unmasked_key(flow, key, key_start, key_len);
1022 struct sw_flow *ovs_flow_lookup_unmasked_key(struct flow_table *table,
1023 struct sw_flow_match *match)
1025 struct sw_flow_key *unmasked = match->key;
1026 int key_len = match->range.end;
1027 struct sw_flow *flow;
1029 flow = ovs_flow_lookup(table, unmasked);
1030 if (flow && (!ovs_flow_cmp_unmasked_key(flow, unmasked, key_len)))
1036 static struct sw_flow *ovs_masked_flow_lookup(struct flow_table *table,
1037 const struct sw_flow_key *flow_key,
1038 struct sw_flow_mask *mask)
1040 struct sw_flow *flow;
1041 struct hlist_head *head;
1042 int key_start = mask->range.start;
1043 int key_len = mask->range.end;
1045 struct sw_flow_key masked_key;
1047 ovs_flow_key_mask(&masked_key, flow_key, mask);
1048 hash = ovs_flow_hash(&masked_key, key_start, key_len);
1049 head = find_bucket(table, hash);
1050 hlist_for_each_entry_rcu(flow, head, hash_node[table->node_ver]) {
1051 if (__flow_cmp_key(flow, &masked_key, key_start, key_len))
1057 struct sw_flow *ovs_flow_lookup(struct flow_table *tbl,
1058 const struct sw_flow_key *key)
1060 struct sw_flow *flow = NULL;
1061 struct sw_flow_mask *mask;
1063 list_for_each_entry_rcu(mask, tbl->mask_list, list) {
1064 flow = ovs_masked_flow_lookup(tbl, key, mask);
1065 if (flow) /* Found */
1073 void ovs_flow_insert(struct flow_table *table, struct sw_flow *flow)
1075 flow->hash = ovs_flow_hash(&flow->key,
1076 ovsl_dereference(flow->mask)->range.start,
1077 ovsl_dereference(flow->mask)->range.end);
1078 __tbl_insert(table, flow);
1081 void ovs_flow_remove(struct flow_table *table, struct sw_flow *flow)
1083 BUG_ON(table->count == 0);
1084 hlist_del_rcu(&flow->hash_node[table->node_ver]);
1088 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute. */
1089 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
1090 [OVS_KEY_ATTR_ENCAP] = -1,
1091 [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
1092 [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
1093 [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
1094 [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
1095 [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
1096 [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
1097 [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
1098 [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
1099 [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
1100 [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
1101 [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
1102 [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
1103 [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
1104 [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
1105 [OVS_KEY_ATTR_TUNNEL] = -1,
1108 static bool is_all_zero(const u8 *fp, size_t size)
1115 for (i = 0; i < size; i++)
1122 static int __parse_flow_nlattrs(const struct nlattr *attr,
1123 const struct nlattr *a[],
1124 u64 *attrsp, bool nz)
1126 const struct nlattr *nla;
1131 nla_for_each_nested(nla, attr, rem) {
1132 u16 type = nla_type(nla);
1135 if (type > OVS_KEY_ATTR_MAX) {
1136 OVS_NLERR("Unknown key attribute (type=%d, max=%d).\n",
1137 type, OVS_KEY_ATTR_MAX);
1140 if (attrs & (1ULL << type)) {
1141 OVS_NLERR("Duplicate key attribute (type %d).\n", type);
1145 expected_len = ovs_key_lens[type];
1146 if (nla_len(nla) != expected_len && expected_len != -1) {
1147 OVS_NLERR("Key attribute has unexpected length (type=%d"
1148 ", length=%d, expected=%d).\n", type,
1149 nla_len(nla), expected_len);
1153 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
1154 attrs |= 1ULL << type;
1159 OVS_NLERR("Message has %d unknown bytes.\n", rem);
1167 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
1168 const struct nlattr *a[], u64 *attrsp)
1170 return __parse_flow_nlattrs(attr, a, attrsp, true);
1173 static int parse_flow_nlattrs(const struct nlattr *attr,
1174 const struct nlattr *a[], u64 *attrsp)
1176 return __parse_flow_nlattrs(attr, a, attrsp, false);
1179 int ipv4_tun_from_nlattr(const struct nlattr *attr,
1180 struct sw_flow_match *match, bool is_mask)
1185 __be16 tun_flags = 0;
1187 nla_for_each_nested(a, attr, rem) {
1188 int type = nla_type(a);
1189 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
1190 [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
1191 [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
1192 [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
1193 [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
1194 [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
1195 [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
1196 [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
1199 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
1200 OVS_NLERR("Unknown IPv4 tunnel attribute (type=%d, max=%d).\n",
1201 type, OVS_TUNNEL_KEY_ATTR_MAX);
1205 if (ovs_tunnel_key_lens[type] != nla_len(a)) {
1206 OVS_NLERR("IPv4 tunnel attribute type has unexpected "
1207 " legnth (type=%d, length=%d, expected=%d).\n",
1208 type, nla_len(a), ovs_tunnel_key_lens[type]);
1213 case OVS_TUNNEL_KEY_ATTR_ID:
1214 SW_FLOW_KEY_PUT(match, tun_key.tun_id,
1215 nla_get_be64(a), is_mask);
1216 tun_flags |= TUNNEL_KEY;
1218 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1219 SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
1220 nla_get_be32(a), is_mask);
1222 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1223 SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
1224 nla_get_be32(a), is_mask);
1226 case OVS_TUNNEL_KEY_ATTR_TOS:
1227 SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
1228 nla_get_u8(a), is_mask);
1230 case OVS_TUNNEL_KEY_ATTR_TTL:
1231 SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
1232 nla_get_u8(a), is_mask);
1235 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1236 tun_flags |= TUNNEL_DONT_FRAGMENT;
1238 case OVS_TUNNEL_KEY_ATTR_CSUM:
1239 tun_flags |= TUNNEL_CSUM;
1246 SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
1249 OVS_NLERR("IPv4 tunnel attribute has %d unknown bytes.\n", rem);
1253 if (!match->key->tun_key.ipv4_dst) {
1254 OVS_NLERR("IPv4 tunnel destination address is zero.\n");
1259 OVS_NLERR("IPv4 tunnel TTL not specified.\n");
1266 int ipv4_tun_to_nlattr(struct sk_buff *skb,
1267 const struct ovs_key_ipv4_tunnel *tun_key,
1268 const struct ovs_key_ipv4_tunnel *output)
1272 nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
1276 if (output->tun_flags & TUNNEL_KEY &&
1277 nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
1279 if (output->ipv4_src &&
1280 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
1282 if (output->ipv4_dst &&
1283 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
1285 if (output->ipv4_tos &&
1286 nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
1288 if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
1290 if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
1291 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
1293 if ((output->tun_flags & TUNNEL_CSUM) &&
1294 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
1297 nla_nest_end(skb, nla);
1302 static int metadata_from_nlattrs(struct sw_flow_match *match, u64 *attrs,
1303 const struct nlattr **a, bool is_mask)
1305 if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
1306 SW_FLOW_KEY_PUT(match, phy.priority,
1307 nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1308 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
1311 if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
1312 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1314 if (!is_mask && in_port >= DP_MAX_PORTS)
1316 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1317 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
1318 } else if (!is_mask) {
1319 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1322 if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
1323 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1324 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1325 if (!is_mask && mark != 0) {
1326 OVS_NLERR("skb->mark must be zero on this kernel (mark=%d).\n", mark);
1330 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1331 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
1333 if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
1334 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1337 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
1342 static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs,
1343 const struct nlattr **a, bool is_mask)
1346 u64 orig_attrs = attrs;
1348 err = metadata_from_nlattrs(match, &attrs, a, is_mask);
1352 if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
1353 const struct ovs_key_ethernet *eth_key;
1355 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1356 SW_FLOW_KEY_MEMCPY(match, eth.src,
1357 eth_key->eth_src, ETH_ALEN, is_mask);
1358 SW_FLOW_KEY_MEMCPY(match, eth.dst,
1359 eth_key->eth_dst, ETH_ALEN, is_mask);
1360 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
1363 if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
1366 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1368 if (!(tci & htons(VLAN_TAG_PRESENT))) {
1369 OVS_NLERR("VLAN TCI does not have VLAN_TAG_PRESENT bit set.\n");
1373 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
1374 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
1377 if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
1380 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1381 if (!is_mask && ntohs(eth_type) < ETH_P_802_3_MIN) {
1382 OVS_NLERR("EtherType is less than mimimum (type=%x, min=%x).\n",
1383 ntohs(eth_type), ETH_P_802_3_MIN);
1387 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1388 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1389 } else if (!is_mask) {
1390 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1393 if (attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1394 const struct ovs_key_ipv4 *ipv4_key;
1396 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1397 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1398 OVS_NLERR("Unknown IPv4 fragment type (value=%d, max=%d).\n",
1399 ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1402 SW_FLOW_KEY_PUT(match, ip.proto,
1403 ipv4_key->ipv4_proto, is_mask);
1404 SW_FLOW_KEY_PUT(match, ip.tos,
1405 ipv4_key->ipv4_tos, is_mask);
1406 SW_FLOW_KEY_PUT(match, ip.ttl,
1407 ipv4_key->ipv4_ttl, is_mask);
1408 SW_FLOW_KEY_PUT(match, ip.frag,
1409 ipv4_key->ipv4_frag, is_mask);
1410 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1411 ipv4_key->ipv4_src, is_mask);
1412 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1413 ipv4_key->ipv4_dst, is_mask);
1414 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4);
1417 if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
1418 const struct ovs_key_ipv6 *ipv6_key;
1420 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1421 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1422 OVS_NLERR("Unknown IPv6 fragment type (value=%d, max=%d).\n",
1423 ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1426 SW_FLOW_KEY_PUT(match, ipv6.label,
1427 ipv6_key->ipv6_label, is_mask);
1428 SW_FLOW_KEY_PUT(match, ip.proto,
1429 ipv6_key->ipv6_proto, is_mask);
1430 SW_FLOW_KEY_PUT(match, ip.tos,
1431 ipv6_key->ipv6_tclass, is_mask);
1432 SW_FLOW_KEY_PUT(match, ip.ttl,
1433 ipv6_key->ipv6_hlimit, is_mask);
1434 SW_FLOW_KEY_PUT(match, ip.frag,
1435 ipv6_key->ipv6_frag, is_mask);
1436 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1438 sizeof(match->key->ipv6.addr.src),
1440 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1442 sizeof(match->key->ipv6.addr.dst),
1445 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
1448 if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
1449 const struct ovs_key_arp *arp_key;
1451 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1452 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1453 OVS_NLERR("Unknown ARP opcode (opcode=%d).\n",
1458 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1459 arp_key->arp_sip, is_mask);
1460 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1461 arp_key->arp_tip, is_mask);
1462 SW_FLOW_KEY_PUT(match, ip.proto,
1463 ntohs(arp_key->arp_op), is_mask);
1464 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1465 arp_key->arp_sha, ETH_ALEN, is_mask);
1466 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1467 arp_key->arp_tha, ETH_ALEN, is_mask);
1469 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
1472 if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
1473 const struct ovs_key_tcp *tcp_key;
1475 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1476 if (orig_attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1477 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1478 tcp_key->tcp_src, is_mask);
1479 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1480 tcp_key->tcp_dst, is_mask);
1482 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1483 tcp_key->tcp_src, is_mask);
1484 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1485 tcp_key->tcp_dst, is_mask);
1487 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
1490 if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
1491 const struct ovs_key_udp *udp_key;
1493 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1494 if (orig_attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1495 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1496 udp_key->udp_src, is_mask);
1497 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1498 udp_key->udp_dst, is_mask);
1500 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1501 udp_key->udp_src, is_mask);
1502 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1503 udp_key->udp_dst, is_mask);
1505 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
1508 if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
1509 const struct ovs_key_icmp *icmp_key;
1511 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1512 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1513 htons(icmp_key->icmp_type), is_mask);
1514 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1515 htons(icmp_key->icmp_code), is_mask);
1516 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
1519 if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
1520 const struct ovs_key_icmpv6 *icmpv6_key;
1522 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1523 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1524 htons(icmpv6_key->icmpv6_type), is_mask);
1525 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1526 htons(icmpv6_key->icmpv6_code), is_mask);
1527 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
1530 if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
1531 const struct ovs_key_nd *nd_key;
1533 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1534 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1536 sizeof(match->key->ipv6.nd.target),
1538 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1539 nd_key->nd_sll, ETH_ALEN, is_mask);
1540 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1541 nd_key->nd_tll, ETH_ALEN, is_mask);
1542 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
1552 * ovs_match_from_nlattrs - parses Netlink attributes into a flow key and
1553 * mask. In case the 'mask' is NULL, the flow is treated as exact match
1554 * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1555 * does not include any don't care bit.
1556 * @match: receives the extracted flow match information.
1557 * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1558 * sequence. The fields should of the packet that triggered the creation
1560 * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1561 * attribute specifies the mask field of the wildcarded flow.
1563 int ovs_match_from_nlattrs(struct sw_flow_match *match,
1564 const struct nlattr *key,
1565 const struct nlattr *mask)
1567 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1568 const struct nlattr *encap;
1571 bool encap_valid = false;
1574 err = parse_flow_nlattrs(key, a, &key_attrs);
1578 if (key_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
1579 encap = a[OVS_KEY_ATTR_ENCAP];
1580 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1581 if (nla_len(encap)) {
1582 __be16 eth_type = 0; /* ETH_P_8021Q */
1584 if (a[OVS_KEY_ATTR_ETHERTYPE])
1585 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1587 if ((eth_type == htons(ETH_P_8021Q)) && (a[OVS_KEY_ATTR_VLAN])) {
1589 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1590 err = parse_flow_nlattrs(encap, a, &key_attrs);
1592 OVS_NLERR("Encap attribute is set for a non-VLAN frame.\n");
1601 err = ovs_key_from_nlattrs(match, key_attrs, a, false);
1606 err = parse_flow_mask_nlattrs(mask, a, &mask_attrs);
1610 if ((mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) && encap_valid) {
1611 __be16 eth_type = 0;
1613 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1614 if (a[OVS_KEY_ATTR_ETHERTYPE])
1615 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1616 if (eth_type == htons(0xffff)) {
1617 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1618 encap = a[OVS_KEY_ATTR_ENCAP];
1619 err = parse_flow_mask_nlattrs(encap, a, &mask_attrs);
1621 OVS_NLERR("VLAN frames must have an exact match"
1622 " on the TPID (mask=%x).\n",
1631 err = ovs_key_from_nlattrs(match, mask_attrs, a, true);
1635 /* Populate exact match flow's key mask. */
1637 ovs_sw_flow_mask_set(match->mask, &match->range, 0xff);
1640 if (!ovs_match_validate(match, key_attrs, mask_attrs))
1647 * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1648 * @flow: Receives extracted in_port, priority, tun_key and skb_mark.
1649 * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1652 * This parses a series of Netlink attributes that form a flow key, which must
1653 * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1654 * get the metadata, that is, the parts of the flow key that cannot be
1655 * extracted from the packet itself.
1658 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow,
1659 const struct nlattr *attr)
1661 struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1662 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1665 struct sw_flow_match match;
1667 flow->key.phy.in_port = DP_MAX_PORTS;
1668 flow->key.phy.priority = 0;
1669 flow->key.phy.skb_mark = 0;
1670 memset(tun_key, 0, sizeof(flow->key.tun_key));
1672 err = parse_flow_nlattrs(attr, a, &attrs);
1676 memset(&match, 0, sizeof(match));
1677 match.key = &flow->key;
1679 err = metadata_from_nlattrs(&match, &attrs, a, false);
1686 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey,
1687 const struct sw_flow_key *output, struct sk_buff *skb)
1689 struct ovs_key_ethernet *eth_key;
1690 struct nlattr *nla, *encap;
1692 if (output->phy.priority &&
1693 nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1694 goto nla_put_failure;
1696 if (swkey->tun_key.ipv4_dst &&
1697 ipv4_tun_to_nlattr(skb, &swkey->tun_key, &output->tun_key))
1698 goto nla_put_failure;
1700 if (swkey->phy.in_port == DP_MAX_PORTS) {
1701 if ((swkey != output) && (output->phy.in_port == 0xffff))
1702 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1703 goto nla_put_failure;
1706 upper_u16 = (swkey == output) ? 0 : 0xffff;
1708 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1709 (upper_u16 << 16) | output->phy.in_port))
1710 goto nla_put_failure;
1713 if (output->phy.skb_mark &&
1714 nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1715 goto nla_put_failure;
1717 nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1719 goto nla_put_failure;
1721 eth_key = nla_data(nla);
1722 memcpy(eth_key->eth_src, output->eth.src, ETH_ALEN);
1723 memcpy(eth_key->eth_dst, output->eth.dst, ETH_ALEN);
1725 if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1727 eth_type = (swkey == output) ? htons(ETH_P_8021Q) : htons(0xffff) ;
1728 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1729 nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1730 goto nla_put_failure;
1731 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1732 if (!swkey->eth.tci)
1737 if (swkey->eth.type == htons(ETH_P_802_2)) {
1739 * Ethertype 802.2 is represented in the netlink with omitted
1740 * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1741 * 0xffff in the mask attribute. Ethertype can also
1744 if (swkey != output && output->eth.type)
1745 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1747 goto nla_put_failure;
1751 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1752 goto nla_put_failure;
1754 if (swkey->eth.type == htons(ETH_P_IP)) {
1755 struct ovs_key_ipv4 *ipv4_key;
1757 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1759 goto nla_put_failure;
1760 ipv4_key = nla_data(nla);
1761 ipv4_key->ipv4_src = output->ipv4.addr.src;
1762 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1763 ipv4_key->ipv4_proto = output->ip.proto;
1764 ipv4_key->ipv4_tos = output->ip.tos;
1765 ipv4_key->ipv4_ttl = output->ip.ttl;
1766 ipv4_key->ipv4_frag = output->ip.frag;
1767 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1768 struct ovs_key_ipv6 *ipv6_key;
1770 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1772 goto nla_put_failure;
1773 ipv6_key = nla_data(nla);
1774 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1775 sizeof(ipv6_key->ipv6_src));
1776 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1777 sizeof(ipv6_key->ipv6_dst));
1778 ipv6_key->ipv6_label = output->ipv6.label;
1779 ipv6_key->ipv6_proto = output->ip.proto;
1780 ipv6_key->ipv6_tclass = output->ip.tos;
1781 ipv6_key->ipv6_hlimit = output->ip.ttl;
1782 ipv6_key->ipv6_frag = output->ip.frag;
1783 } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1784 swkey->eth.type == htons(ETH_P_RARP)) {
1785 struct ovs_key_arp *arp_key;
1787 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1789 goto nla_put_failure;
1790 arp_key = nla_data(nla);
1791 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1792 arp_key->arp_sip = output->ipv4.addr.src;
1793 arp_key->arp_tip = output->ipv4.addr.dst;
1794 arp_key->arp_op = htons(output->ip.proto);
1795 memcpy(arp_key->arp_sha, output->ipv4.arp.sha, ETH_ALEN);
1796 memcpy(arp_key->arp_tha, output->ipv4.arp.tha, ETH_ALEN);
1799 if ((swkey->eth.type == htons(ETH_P_IP) ||
1800 swkey->eth.type == htons(ETH_P_IPV6)) &&
1801 swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1803 if (swkey->ip.proto == IPPROTO_TCP) {
1804 struct ovs_key_tcp *tcp_key;
1806 nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1808 goto nla_put_failure;
1809 tcp_key = nla_data(nla);
1810 if (swkey->eth.type == htons(ETH_P_IP)) {
1811 tcp_key->tcp_src = output->ipv4.tp.src;
1812 tcp_key->tcp_dst = output->ipv4.tp.dst;
1813 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1814 tcp_key->tcp_src = output->ipv6.tp.src;
1815 tcp_key->tcp_dst = output->ipv6.tp.dst;
1817 } else if (swkey->ip.proto == IPPROTO_UDP) {
1818 struct ovs_key_udp *udp_key;
1820 nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1822 goto nla_put_failure;
1823 udp_key = nla_data(nla);
1824 if (swkey->eth.type == htons(ETH_P_IP)) {
1825 udp_key->udp_src = output->ipv4.tp.src;
1826 udp_key->udp_dst = output->ipv4.tp.dst;
1827 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1828 udp_key->udp_src = output->ipv6.tp.src;
1829 udp_key->udp_dst = output->ipv6.tp.dst;
1831 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1832 swkey->ip.proto == IPPROTO_ICMP) {
1833 struct ovs_key_icmp *icmp_key;
1835 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1837 goto nla_put_failure;
1838 icmp_key = nla_data(nla);
1839 icmp_key->icmp_type = ntohs(output->ipv4.tp.src);
1840 icmp_key->icmp_code = ntohs(output->ipv4.tp.dst);
1841 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1842 swkey->ip.proto == IPPROTO_ICMPV6) {
1843 struct ovs_key_icmpv6 *icmpv6_key;
1845 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1846 sizeof(*icmpv6_key));
1848 goto nla_put_failure;
1849 icmpv6_key = nla_data(nla);
1850 icmpv6_key->icmpv6_type = ntohs(output->ipv6.tp.src);
1851 icmpv6_key->icmpv6_code = ntohs(output->ipv6.tp.dst);
1853 if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1854 icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1855 struct ovs_key_nd *nd_key;
1857 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1859 goto nla_put_failure;
1860 nd_key = nla_data(nla);
1861 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1862 sizeof(nd_key->nd_target));
1863 memcpy(nd_key->nd_sll, output->ipv6.nd.sll, ETH_ALEN);
1864 memcpy(nd_key->nd_tll, output->ipv6.nd.tll, ETH_ALEN);
1871 nla_nest_end(skb, encap);
1879 /* Initializes the flow module.
1880 * Returns zero if successful or a negative error code. */
1881 int ovs_flow_init(void)
1883 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1885 if (flow_cache == NULL)
1891 /* Uninitializes the flow module. */
1892 void ovs_flow_exit(void)
1894 kmem_cache_destroy(flow_cache);
1897 struct sw_flow_mask *ovs_sw_flow_mask_alloc(void)
1899 struct sw_flow_mask *mask;
1901 mask = kmalloc(sizeof(*mask), GFP_KERNEL);
1903 mask->ref_count = 0;
1908 void ovs_sw_flow_mask_add_ref(struct sw_flow_mask *mask)
1913 static void rcu_free_sw_flow_mask_cb(struct rcu_head *rcu)
1915 struct sw_flow_mask *mask = container_of(rcu, struct sw_flow_mask, rcu);
1920 void ovs_sw_flow_mask_del_ref(struct sw_flow_mask *mask, bool deferred)
1925 BUG_ON(!mask->ref_count);
1928 if (!mask->ref_count) {
1929 list_del_rcu(&mask->list);
1931 call_rcu(&mask->rcu, rcu_free_sw_flow_mask_cb);
1937 static bool ovs_sw_flow_mask_equal(const struct sw_flow_mask *a,
1938 const struct sw_flow_mask *b)
1940 u8 *a_ = (u8 *)&a->key + a->range.start;
1941 u8 *b_ = (u8 *)&b->key + b->range.start;
1943 return (a->range.end == b->range.end)
1944 && (a->range.start == b->range.start)
1945 && (memcmp(a_, b_, ovs_sw_flow_mask_actual_size(a)) == 0);
1948 struct sw_flow_mask *ovs_sw_flow_mask_find(const struct flow_table *tbl,
1949 const struct sw_flow_mask *mask)
1951 struct list_head *ml;
1953 list_for_each(ml, tbl->mask_list) {
1954 struct sw_flow_mask *m;
1955 m = container_of(ml, struct sw_flow_mask, list);
1956 if (ovs_sw_flow_mask_equal(mask, m))
1964 * add a new mask into the mask list.
1965 * The caller needs to make sure that 'mask' is not the same
1966 * as any masks that are already on the list.
1968 void ovs_sw_flow_mask_insert(struct flow_table *tbl, struct sw_flow_mask *mask)
1970 list_add_rcu(&mask->list, tbl->mask_list);
1974 * Set 'range' fields in the mask to the value of 'val'.
1976 static void ovs_sw_flow_mask_set(struct sw_flow_mask *mask,
1977 struct sw_flow_key_range *range, u8 val)
1979 u8 *m = (u8 *)&mask->key + range->start;
1981 mask->range = *range;
1982 memset(m, val, ovs_sw_flow_mask_size_roundup(mask));