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/sctp.h>
38 #include <linux/tcp.h>
39 #include <linux/udp.h>
40 #include <linux/icmp.h>
41 #include <linux/icmpv6.h>
42 #include <linux/rculist.h>
44 #include <net/ip_tunnels.h>
46 #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 = rounddown(offset, sizeof(long));
58 size_t end = roundup(offset + size, sizeof(long));
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 static u16 range_n_bytes(const struct sw_flow_key_range *range)
107 return range->end - range->start;
110 void ovs_match_init(struct sw_flow_match *match,
111 struct sw_flow_key *key,
112 struct sw_flow_mask *mask)
114 memset(match, 0, sizeof(*match));
118 memset(key, 0, sizeof(*key));
121 memset(&mask->key, 0, sizeof(mask->key));
122 mask->range.start = mask->range.end = 0;
126 static bool ovs_match_validate(const struct sw_flow_match *match,
127 u64 key_attrs, u64 mask_attrs)
129 u64 key_expected = 1 << OVS_KEY_ATTR_ETHERNET;
130 u64 mask_allowed = key_attrs; /* At most allow all key attributes */
132 /* The following mask attributes allowed only if they
133 * pass the validation tests. */
134 mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
135 | (1 << OVS_KEY_ATTR_IPV6)
136 | (1 << OVS_KEY_ATTR_TCP)
137 | (1 << OVS_KEY_ATTR_UDP)
138 | (1 << OVS_KEY_ATTR_SCTP)
139 | (1 << OVS_KEY_ATTR_ICMP)
140 | (1 << OVS_KEY_ATTR_ICMPV6)
141 | (1 << OVS_KEY_ATTR_ARP)
142 | (1 << OVS_KEY_ATTR_ND));
144 /* Always allowed mask fields. */
145 mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
146 | (1 << OVS_KEY_ATTR_IN_PORT)
147 | (1 << OVS_KEY_ATTR_ETHERTYPE));
149 /* Check key attributes. */
150 if (match->key->eth.type == htons(ETH_P_ARP)
151 || match->key->eth.type == htons(ETH_P_RARP)) {
152 key_expected |= 1 << OVS_KEY_ATTR_ARP;
153 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
154 mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
157 if (match->key->eth.type == htons(ETH_P_IP)) {
158 key_expected |= 1 << OVS_KEY_ATTR_IPV4;
159 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
160 mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
162 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
163 if (match->key->ip.proto == IPPROTO_UDP) {
164 key_expected |= 1 << OVS_KEY_ATTR_UDP;
165 if (match->mask && (match->mask->key.ip.proto == 0xff))
166 mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
169 if (match->key->ip.proto == IPPROTO_SCTP) {
170 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
171 if (match->mask && (match->mask->key.ip.proto == 0xff))
172 mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
175 if (match->key->ip.proto == IPPROTO_TCP) {
176 key_expected |= 1 << OVS_KEY_ATTR_TCP;
177 if (match->mask && (match->mask->key.ip.proto == 0xff))
178 mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
181 if (match->key->ip.proto == IPPROTO_ICMP) {
182 key_expected |= 1 << OVS_KEY_ATTR_ICMP;
183 if (match->mask && (match->mask->key.ip.proto == 0xff))
184 mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
189 if (match->key->eth.type == htons(ETH_P_IPV6)) {
190 key_expected |= 1 << OVS_KEY_ATTR_IPV6;
191 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
192 mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
194 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
195 if (match->key->ip.proto == IPPROTO_UDP) {
196 key_expected |= 1 << OVS_KEY_ATTR_UDP;
197 if (match->mask && (match->mask->key.ip.proto == 0xff))
198 mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
201 if (match->key->ip.proto == IPPROTO_SCTP) {
202 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
203 if (match->mask && (match->mask->key.ip.proto == 0xff))
204 mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
207 if (match->key->ip.proto == IPPROTO_TCP) {
208 key_expected |= 1 << OVS_KEY_ATTR_TCP;
209 if (match->mask && (match->mask->key.ip.proto == 0xff))
210 mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
213 if (match->key->ip.proto == IPPROTO_ICMPV6) {
214 key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
215 if (match->mask && (match->mask->key.ip.proto == 0xff))
216 mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
218 if (match->key->ipv6.tp.src ==
219 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
220 match->key->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
221 key_expected |= 1 << OVS_KEY_ATTR_ND;
222 if (match->mask && (match->mask->key.ipv6.tp.src == htons(0xffff)))
223 mask_allowed |= 1 << OVS_KEY_ATTR_ND;
229 if ((key_attrs & key_expected) != key_expected) {
230 /* Key attributes check failed. */
231 OVS_NLERR("Missing expected key attributes (key_attrs=%llx, expected=%llx).\n",
232 key_attrs, key_expected);
236 if ((mask_attrs & mask_allowed) != mask_attrs) {
237 /* Mask attributes check failed. */
238 OVS_NLERR("Contain more than allowed mask fields (mask_attrs=%llx, mask_allowed=%llx).\n",
239 mask_attrs, mask_allowed);
246 static int check_header(struct sk_buff *skb, int len)
248 if (unlikely(skb->len < len))
250 if (unlikely(!pskb_may_pull(skb, len)))
255 static bool arphdr_ok(struct sk_buff *skb)
257 return pskb_may_pull(skb, skb_network_offset(skb) +
258 sizeof(struct arp_eth_header));
261 static int check_iphdr(struct sk_buff *skb)
263 unsigned int nh_ofs = skb_network_offset(skb);
267 err = check_header(skb, nh_ofs + sizeof(struct iphdr));
271 ip_len = ip_hdrlen(skb);
272 if (unlikely(ip_len < sizeof(struct iphdr) ||
273 skb->len < nh_ofs + ip_len))
276 skb_set_transport_header(skb, nh_ofs + ip_len);
280 static bool tcphdr_ok(struct sk_buff *skb)
282 int th_ofs = skb_transport_offset(skb);
285 if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
288 tcp_len = tcp_hdrlen(skb);
289 if (unlikely(tcp_len < sizeof(struct tcphdr) ||
290 skb->len < th_ofs + tcp_len))
296 static bool udphdr_ok(struct sk_buff *skb)
298 return pskb_may_pull(skb, skb_transport_offset(skb) +
299 sizeof(struct udphdr));
302 static bool sctphdr_ok(struct sk_buff *skb)
304 return pskb_may_pull(skb, skb_transport_offset(skb) +
305 sizeof(struct sctphdr));
308 static bool icmphdr_ok(struct sk_buff *skb)
310 return pskb_may_pull(skb, skb_transport_offset(skb) +
311 sizeof(struct icmphdr));
314 u64 ovs_flow_used_time(unsigned long flow_jiffies)
316 struct timespec cur_ts;
319 ktime_get_ts(&cur_ts);
320 idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
321 cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
322 cur_ts.tv_nsec / NSEC_PER_MSEC;
324 return cur_ms - idle_ms;
327 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
329 unsigned int nh_ofs = skb_network_offset(skb);
337 err = check_header(skb, nh_ofs + sizeof(*nh));
342 nexthdr = nh->nexthdr;
343 payload_ofs = (u8 *)(nh + 1) - skb->data;
345 key->ip.proto = NEXTHDR_NONE;
346 key->ip.tos = ipv6_get_dsfield(nh);
347 key->ip.ttl = nh->hop_limit;
348 key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
349 key->ipv6.addr.src = nh->saddr;
350 key->ipv6.addr.dst = nh->daddr;
352 payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
353 if (unlikely(payload_ofs < 0))
357 if (frag_off & htons(~0x7))
358 key->ip.frag = OVS_FRAG_TYPE_LATER;
360 key->ip.frag = OVS_FRAG_TYPE_FIRST;
363 nh_len = payload_ofs - nh_ofs;
364 skb_set_transport_header(skb, nh_ofs + nh_len);
365 key->ip.proto = nexthdr;
369 static bool icmp6hdr_ok(struct sk_buff *skb)
371 return pskb_may_pull(skb, skb_transport_offset(skb) +
372 sizeof(struct icmp6hdr));
375 void ovs_flow_key_mask(struct sw_flow_key *dst, const struct sw_flow_key *src,
376 const struct sw_flow_mask *mask)
378 const long *m = (long *)((u8 *)&mask->key + mask->range.start);
379 const long *s = (long *)((u8 *)src + mask->range.start);
380 long *d = (long *)((u8 *)dst + mask->range.start);
383 /* The memory outside of the 'mask->range' are not set since
384 * further operations on 'dst' only uses contents within
387 for (i = 0; i < range_n_bytes(&mask->range); i += sizeof(long))
391 #define TCP_FLAGS_OFFSET 13
392 #define TCP_FLAG_MASK 0x3f
394 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
398 if ((flow->key.eth.type == htons(ETH_P_IP) ||
399 flow->key.eth.type == htons(ETH_P_IPV6)) &&
400 flow->key.ip.proto == IPPROTO_TCP &&
401 likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
402 u8 *tcp = (u8 *)tcp_hdr(skb);
403 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
406 spin_lock(&flow->lock);
407 flow->used = jiffies;
408 flow->packet_count++;
409 flow->byte_count += skb->len;
410 flow->tcp_flags |= tcp_flags;
411 spin_unlock(&flow->lock);
414 struct sw_flow_actions *ovs_flow_actions_alloc(int size)
416 struct sw_flow_actions *sfa;
418 if (size > MAX_ACTIONS_BUFSIZE)
419 return ERR_PTR(-EINVAL);
421 sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
423 return ERR_PTR(-ENOMEM);
425 sfa->actions_len = 0;
429 struct sw_flow *ovs_flow_alloc(void)
431 struct sw_flow *flow;
433 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
435 return ERR_PTR(-ENOMEM);
437 spin_lock_init(&flow->lock);
438 flow->sf_acts = NULL;
444 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
446 hash = jhash_1word(hash, table->hash_seed);
447 return flex_array_get(table->buckets,
448 (hash & (table->n_buckets - 1)));
451 static struct flex_array *alloc_buckets(unsigned int n_buckets)
453 struct flex_array *buckets;
456 buckets = flex_array_alloc(sizeof(struct hlist_head),
457 n_buckets, GFP_KERNEL);
461 err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
463 flex_array_free(buckets);
467 for (i = 0; i < n_buckets; i++)
468 INIT_HLIST_HEAD((struct hlist_head *)
469 flex_array_get(buckets, i));
474 static void free_buckets(struct flex_array *buckets)
476 flex_array_free(buckets);
479 static struct flow_table *__flow_tbl_alloc(int new_size)
481 struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
486 table->buckets = alloc_buckets(new_size);
488 if (!table->buckets) {
492 table->n_buckets = new_size;
495 table->keep_flows = false;
496 get_random_bytes(&table->hash_seed, sizeof(u32));
497 table->mask_list = NULL;
502 static void __flow_tbl_destroy(struct flow_table *table)
506 if (table->keep_flows)
509 for (i = 0; i < table->n_buckets; i++) {
510 struct sw_flow *flow;
511 struct hlist_head *head = flex_array_get(table->buckets, i);
512 struct hlist_node *n;
513 int ver = table->node_ver;
515 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
516 hlist_del(&flow->hash_node[ver]);
517 ovs_flow_free(flow, false);
521 BUG_ON(!list_empty(table->mask_list));
522 kfree(table->mask_list);
525 free_buckets(table->buckets);
529 struct flow_table *ovs_flow_tbl_alloc(int new_size)
531 struct flow_table *table = __flow_tbl_alloc(new_size);
536 table->mask_list = kmalloc(sizeof(struct list_head), GFP_KERNEL);
537 if (!table->mask_list) {
538 table->keep_flows = true;
539 __flow_tbl_destroy(table);
542 INIT_LIST_HEAD(table->mask_list);
547 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
549 struct flow_table *table = container_of(rcu, struct flow_table, rcu);
551 __flow_tbl_destroy(table);
554 void ovs_flow_tbl_destroy(struct flow_table *table, bool deferred)
560 call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
562 __flow_tbl_destroy(table);
565 struct sw_flow *ovs_flow_dump_next(struct flow_table *table, u32 *bucket, u32 *last)
567 struct sw_flow *flow;
568 struct hlist_head *head;
572 ver = table->node_ver;
573 while (*bucket < table->n_buckets) {
575 head = flex_array_get(table->buckets, *bucket);
576 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
591 static void __tbl_insert(struct flow_table *table, struct sw_flow *flow)
593 struct hlist_head *head;
595 head = find_bucket(table, flow->hash);
596 hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
601 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
606 old_ver = old->node_ver;
607 new->node_ver = !old_ver;
609 /* Insert in new table. */
610 for (i = 0; i < old->n_buckets; i++) {
611 struct sw_flow *flow;
612 struct hlist_head *head;
614 head = flex_array_get(old->buckets, i);
616 hlist_for_each_entry(flow, head, hash_node[old_ver])
617 __tbl_insert(new, flow);
620 new->mask_list = old->mask_list;
621 old->keep_flows = true;
624 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
626 struct flow_table *new_table;
628 new_table = __flow_tbl_alloc(n_buckets);
630 return ERR_PTR(-ENOMEM);
632 flow_table_copy_flows(table, new_table);
637 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
639 return __flow_tbl_rehash(table, table->n_buckets);
642 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
644 return __flow_tbl_rehash(table, table->n_buckets * 2);
647 static void __flow_free(struct sw_flow *flow)
649 kfree((struct sf_flow_acts __force *)flow->sf_acts);
650 kmem_cache_free(flow_cache, flow);
653 static void rcu_free_flow_callback(struct rcu_head *rcu)
655 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
660 void ovs_flow_free(struct sw_flow *flow, bool deferred)
665 ovs_sw_flow_mask_del_ref(flow->mask, deferred);
668 call_rcu(&flow->rcu, rcu_free_flow_callback);
673 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
674 * The caller must hold rcu_read_lock for this to be sensible. */
675 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
677 kfree_rcu(sf_acts, rcu);
680 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
683 __be16 eth_type; /* ETH_P_8021Q */
686 struct qtag_prefix *qp;
688 if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
691 if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
695 qp = (struct qtag_prefix *) skb->data;
696 key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
697 __skb_pull(skb, sizeof(struct qtag_prefix));
702 static __be16 parse_ethertype(struct sk_buff *skb)
704 struct llc_snap_hdr {
705 u8 dsap; /* Always 0xAA */
706 u8 ssap; /* Always 0xAA */
711 struct llc_snap_hdr *llc;
714 proto = *(__be16 *) skb->data;
715 __skb_pull(skb, sizeof(__be16));
717 if (ntohs(proto) >= ETH_P_802_3_MIN)
720 if (skb->len < sizeof(struct llc_snap_hdr))
721 return htons(ETH_P_802_2);
723 if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
726 llc = (struct llc_snap_hdr *) skb->data;
727 if (llc->dsap != LLC_SAP_SNAP ||
728 llc->ssap != LLC_SAP_SNAP ||
729 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
730 return htons(ETH_P_802_2);
732 __skb_pull(skb, sizeof(struct llc_snap_hdr));
734 if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
735 return llc->ethertype;
737 return htons(ETH_P_802_2);
740 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
743 struct icmp6hdr *icmp = icmp6_hdr(skb);
745 /* The ICMPv6 type and code fields use the 16-bit transport port
746 * fields, so we need to store them in 16-bit network byte order.
748 key->ipv6.tp.src = htons(icmp->icmp6_type);
749 key->ipv6.tp.dst = htons(icmp->icmp6_code);
751 if (icmp->icmp6_code == 0 &&
752 (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
753 icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
754 int icmp_len = skb->len - skb_transport_offset(skb);
758 /* In order to process neighbor discovery options, we need the
761 if (unlikely(icmp_len < sizeof(*nd)))
764 if (unlikely(skb_linearize(skb)))
767 nd = (struct nd_msg *)skb_transport_header(skb);
768 key->ipv6.nd.target = nd->target;
770 icmp_len -= sizeof(*nd);
772 while (icmp_len >= 8) {
773 struct nd_opt_hdr *nd_opt =
774 (struct nd_opt_hdr *)(nd->opt + offset);
775 int opt_len = nd_opt->nd_opt_len * 8;
777 if (unlikely(!opt_len || opt_len > icmp_len))
780 /* Store the link layer address if the appropriate
781 * option is provided. It is considered an error if
782 * the same link layer option is specified twice.
784 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
786 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
788 memcpy(key->ipv6.nd.sll,
789 &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
790 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
792 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
794 memcpy(key->ipv6.nd.tll,
795 &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
806 memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
807 memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
808 memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
814 * ovs_flow_extract - extracts a flow key from an Ethernet frame.
815 * @skb: sk_buff that contains the frame, with skb->data pointing to the
817 * @in_port: port number on which @skb was received.
818 * @key: output flow key
820 * The caller must ensure that skb->len >= ETH_HLEN.
822 * Returns 0 if successful, otherwise a negative errno value.
824 * Initializes @skb header pointers as follows:
826 * - skb->mac_header: the Ethernet header.
828 * - skb->network_header: just past the Ethernet header, or just past the
829 * VLAN header, to the first byte of the Ethernet payload.
831 * - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
832 * on output, then just past the IP header, if one is present and
833 * of a correct length, otherwise the same as skb->network_header.
834 * For other key->eth.type values it is left untouched.
836 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
841 memset(key, 0, sizeof(*key));
843 key->phy.priority = skb->priority;
844 if (OVS_CB(skb)->tun_key)
845 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
846 key->phy.in_port = in_port;
847 key->phy.skb_mark = skb->mark;
849 skb_reset_mac_header(skb);
851 /* Link layer. We are guaranteed to have at least the 14 byte Ethernet
852 * header in the linear data area.
855 memcpy(key->eth.src, eth->h_source, ETH_ALEN);
856 memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
858 __skb_pull(skb, 2 * ETH_ALEN);
859 /* We are going to push all headers that we pull, so no need to
860 * update skb->csum here.
863 if (vlan_tx_tag_present(skb))
864 key->eth.tci = htons(skb->vlan_tci);
865 else if (eth->h_proto == htons(ETH_P_8021Q))
866 if (unlikely(parse_vlan(skb, key)))
869 key->eth.type = parse_ethertype(skb);
870 if (unlikely(key->eth.type == htons(0)))
873 skb_reset_network_header(skb);
874 __skb_push(skb, skb->data - skb_mac_header(skb));
877 if (key->eth.type == htons(ETH_P_IP)) {
881 error = check_iphdr(skb);
882 if (unlikely(error)) {
883 if (error == -EINVAL) {
884 skb->transport_header = skb->network_header;
891 key->ipv4.addr.src = nh->saddr;
892 key->ipv4.addr.dst = nh->daddr;
894 key->ip.proto = nh->protocol;
895 key->ip.tos = nh->tos;
896 key->ip.ttl = nh->ttl;
898 offset = nh->frag_off & htons(IP_OFFSET);
900 key->ip.frag = OVS_FRAG_TYPE_LATER;
903 if (nh->frag_off & htons(IP_MF) ||
904 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
905 key->ip.frag = OVS_FRAG_TYPE_FIRST;
907 /* Transport layer. */
908 if (key->ip.proto == IPPROTO_TCP) {
909 if (tcphdr_ok(skb)) {
910 struct tcphdr *tcp = tcp_hdr(skb);
911 key->ipv4.tp.src = tcp->source;
912 key->ipv4.tp.dst = tcp->dest;
914 } else if (key->ip.proto == IPPROTO_UDP) {
915 if (udphdr_ok(skb)) {
916 struct udphdr *udp = udp_hdr(skb);
917 key->ipv4.tp.src = udp->source;
918 key->ipv4.tp.dst = udp->dest;
920 } else if (key->ip.proto == IPPROTO_SCTP) {
921 if (sctphdr_ok(skb)) {
922 struct sctphdr *sctp = sctp_hdr(skb);
923 key->ipv4.tp.src = sctp->source;
924 key->ipv4.tp.dst = sctp->dest;
926 } else if (key->ip.proto == IPPROTO_ICMP) {
927 if (icmphdr_ok(skb)) {
928 struct icmphdr *icmp = icmp_hdr(skb);
929 /* The ICMP type and code fields use the 16-bit
930 * transport port fields, so we need to store
931 * them in 16-bit network byte order. */
932 key->ipv4.tp.src = htons(icmp->type);
933 key->ipv4.tp.dst = htons(icmp->code);
937 } else if ((key->eth.type == htons(ETH_P_ARP) ||
938 key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
939 struct arp_eth_header *arp;
941 arp = (struct arp_eth_header *)skb_network_header(skb);
943 if (arp->ar_hrd == htons(ARPHRD_ETHER)
944 && arp->ar_pro == htons(ETH_P_IP)
945 && arp->ar_hln == ETH_ALEN
946 && arp->ar_pln == 4) {
948 /* We only match on the lower 8 bits of the opcode. */
949 if (ntohs(arp->ar_op) <= 0xff)
950 key->ip.proto = ntohs(arp->ar_op);
951 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
952 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
953 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
954 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
956 } else if (key->eth.type == htons(ETH_P_IPV6)) {
957 int nh_len; /* IPv6 Header + Extensions */
959 nh_len = parse_ipv6hdr(skb, key);
960 if (unlikely(nh_len < 0)) {
961 if (nh_len == -EINVAL) {
962 skb->transport_header = skb->network_header;
970 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
972 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
973 key->ip.frag = OVS_FRAG_TYPE_FIRST;
975 /* Transport layer. */
976 if (key->ip.proto == NEXTHDR_TCP) {
977 if (tcphdr_ok(skb)) {
978 struct tcphdr *tcp = tcp_hdr(skb);
979 key->ipv6.tp.src = tcp->source;
980 key->ipv6.tp.dst = tcp->dest;
982 } else if (key->ip.proto == NEXTHDR_UDP) {
983 if (udphdr_ok(skb)) {
984 struct udphdr *udp = udp_hdr(skb);
985 key->ipv6.tp.src = udp->source;
986 key->ipv6.tp.dst = udp->dest;
988 } else if (key->ip.proto == NEXTHDR_SCTP) {
989 if (sctphdr_ok(skb)) {
990 struct sctphdr *sctp = sctp_hdr(skb);
991 key->ipv6.tp.src = sctp->source;
992 key->ipv6.tp.dst = sctp->dest;
994 } else if (key->ip.proto == NEXTHDR_ICMP) {
995 if (icmp6hdr_ok(skb)) {
996 error = parse_icmpv6(skb, key, nh_len);
1006 static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start,
1009 u32 *hash_key = (u32 *)((u8 *)key + key_start);
1010 int hash_u32s = (key_end - key_start) >> 2;
1012 /* Make sure number of hash bytes are multiple of u32. */
1013 BUILD_BUG_ON(sizeof(long) % sizeof(u32));
1015 return jhash2(hash_key, hash_u32s, 0);
1018 static int flow_key_start(const struct sw_flow_key *key)
1020 if (key->tun_key.ipv4_dst)
1023 return rounddown(offsetof(struct sw_flow_key, phy),
1027 static bool __cmp_key(const struct sw_flow_key *key1,
1028 const struct sw_flow_key *key2, int key_start, int key_end)
1030 const long *cp1 = (long *)((u8 *)key1 + key_start);
1031 const long *cp2 = (long *)((u8 *)key2 + key_start);
1035 for (i = key_start; i < key_end; i += sizeof(long))
1036 diffs |= *cp1++ ^ *cp2++;
1041 static bool __flow_cmp_masked_key(const struct sw_flow *flow,
1042 const struct sw_flow_key *key, int key_start, int key_end)
1044 return __cmp_key(&flow->key, key, key_start, key_end);
1047 static bool __flow_cmp_unmasked_key(const struct sw_flow *flow,
1048 const struct sw_flow_key *key, int key_start, int key_end)
1050 return __cmp_key(&flow->unmasked_key, key, key_start, key_end);
1053 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
1054 const struct sw_flow_key *key, int key_end)
1057 key_start = flow_key_start(key);
1059 return __flow_cmp_unmasked_key(flow, key, key_start, key_end);
1063 struct sw_flow *ovs_flow_lookup_unmasked_key(struct flow_table *table,
1064 struct sw_flow_match *match)
1066 struct sw_flow_key *unmasked = match->key;
1067 int key_end = match->range.end;
1068 struct sw_flow *flow;
1070 flow = ovs_flow_lookup(table, unmasked);
1071 if (flow && (!ovs_flow_cmp_unmasked_key(flow, unmasked, key_end)))
1077 static struct sw_flow *ovs_masked_flow_lookup(struct flow_table *table,
1078 const struct sw_flow_key *unmasked,
1079 struct sw_flow_mask *mask)
1081 struct sw_flow *flow;
1082 struct hlist_head *head;
1083 int key_start = mask->range.start;
1084 int key_end = mask->range.end;
1086 struct sw_flow_key masked_key;
1088 ovs_flow_key_mask(&masked_key, unmasked, mask);
1089 hash = ovs_flow_hash(&masked_key, key_start, key_end);
1090 head = find_bucket(table, hash);
1091 hlist_for_each_entry_rcu(flow, head, hash_node[table->node_ver]) {
1092 if (flow->mask == mask &&
1093 __flow_cmp_masked_key(flow, &masked_key,
1094 key_start, key_end))
1100 struct sw_flow *ovs_flow_lookup(struct flow_table *tbl,
1101 const struct sw_flow_key *key)
1103 struct sw_flow *flow = NULL;
1104 struct sw_flow_mask *mask;
1106 list_for_each_entry_rcu(mask, tbl->mask_list, list) {
1107 flow = ovs_masked_flow_lookup(tbl, key, mask);
1108 if (flow) /* Found */
1116 void ovs_flow_insert(struct flow_table *table, struct sw_flow *flow)
1118 flow->hash = ovs_flow_hash(&flow->key, flow->mask->range.start,
1119 flow->mask->range.end);
1120 __tbl_insert(table, flow);
1123 void ovs_flow_remove(struct flow_table *table, struct sw_flow *flow)
1125 BUG_ON(table->count == 0);
1126 hlist_del_rcu(&flow->hash_node[table->node_ver]);
1130 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute. */
1131 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
1132 [OVS_KEY_ATTR_ENCAP] = -1,
1133 [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
1134 [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
1135 [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
1136 [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
1137 [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
1138 [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
1139 [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
1140 [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
1141 [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
1142 [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
1143 [OVS_KEY_ATTR_SCTP] = sizeof(struct ovs_key_sctp),
1144 [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
1145 [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
1146 [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
1147 [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
1148 [OVS_KEY_ATTR_TUNNEL] = -1,
1151 static bool is_all_zero(const u8 *fp, size_t size)
1158 for (i = 0; i < size; i++)
1165 static int __parse_flow_nlattrs(const struct nlattr *attr,
1166 const struct nlattr *a[],
1167 u64 *attrsp, bool nz)
1169 const struct nlattr *nla;
1174 nla_for_each_nested(nla, attr, rem) {
1175 u16 type = nla_type(nla);
1178 if (type > OVS_KEY_ATTR_MAX) {
1179 OVS_NLERR("Unknown key attribute (type=%d, max=%d).\n",
1180 type, OVS_KEY_ATTR_MAX);
1183 if (attrs & (1 << type)) {
1184 OVS_NLERR("Duplicate key attribute (type %d).\n", type);
1188 expected_len = ovs_key_lens[type];
1189 if (nla_len(nla) != expected_len && expected_len != -1) {
1190 OVS_NLERR("Key attribute has unexpected length (type=%d"
1191 ", length=%d, expected=%d).\n", type,
1192 nla_len(nla), expected_len);
1196 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
1202 OVS_NLERR("Message has %d unknown bytes.\n", rem);
1210 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
1211 const struct nlattr *a[], u64 *attrsp)
1213 return __parse_flow_nlattrs(attr, a, attrsp, true);
1216 static int parse_flow_nlattrs(const struct nlattr *attr,
1217 const struct nlattr *a[], u64 *attrsp)
1219 return __parse_flow_nlattrs(attr, a, attrsp, false);
1222 int ovs_ipv4_tun_from_nlattr(const struct nlattr *attr,
1223 struct sw_flow_match *match, bool is_mask)
1228 __be16 tun_flags = 0;
1230 nla_for_each_nested(a, attr, rem) {
1231 int type = nla_type(a);
1232 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
1233 [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
1234 [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
1235 [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
1236 [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
1237 [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
1238 [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
1239 [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
1242 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
1243 OVS_NLERR("Unknown IPv4 tunnel attribute (type=%d, max=%d).\n",
1244 type, OVS_TUNNEL_KEY_ATTR_MAX);
1248 if (ovs_tunnel_key_lens[type] != nla_len(a)) {
1249 OVS_NLERR("IPv4 tunnel attribute type has unexpected "
1250 " length (type=%d, length=%d, expected=%d).\n",
1251 type, nla_len(a), ovs_tunnel_key_lens[type]);
1256 case OVS_TUNNEL_KEY_ATTR_ID:
1257 SW_FLOW_KEY_PUT(match, tun_key.tun_id,
1258 nla_get_be64(a), is_mask);
1259 tun_flags |= TUNNEL_KEY;
1261 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1262 SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
1263 nla_get_be32(a), is_mask);
1265 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1266 SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
1267 nla_get_be32(a), is_mask);
1269 case OVS_TUNNEL_KEY_ATTR_TOS:
1270 SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
1271 nla_get_u8(a), is_mask);
1273 case OVS_TUNNEL_KEY_ATTR_TTL:
1274 SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
1275 nla_get_u8(a), is_mask);
1278 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1279 tun_flags |= TUNNEL_DONT_FRAGMENT;
1281 case OVS_TUNNEL_KEY_ATTR_CSUM:
1282 tun_flags |= TUNNEL_CSUM;
1289 SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
1292 OVS_NLERR("IPv4 tunnel attribute has %d unknown bytes.\n", rem);
1297 if (!match->key->tun_key.ipv4_dst) {
1298 OVS_NLERR("IPv4 tunnel destination address is zero.\n");
1303 OVS_NLERR("IPv4 tunnel TTL not specified.\n");
1311 int ovs_ipv4_tun_to_nlattr(struct sk_buff *skb,
1312 const struct ovs_key_ipv4_tunnel *tun_key,
1313 const struct ovs_key_ipv4_tunnel *output)
1317 nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
1321 if (output->tun_flags & TUNNEL_KEY &&
1322 nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
1324 if (output->ipv4_src &&
1325 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
1327 if (output->ipv4_dst &&
1328 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
1330 if (output->ipv4_tos &&
1331 nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
1333 if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
1335 if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
1336 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
1338 if ((output->tun_flags & TUNNEL_CSUM) &&
1339 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
1342 nla_nest_end(skb, nla);
1346 static int metadata_from_nlattrs(struct sw_flow_match *match, u64 *attrs,
1347 const struct nlattr **a, bool is_mask)
1349 if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1350 SW_FLOW_KEY_PUT(match, phy.priority,
1351 nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1352 *attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1355 if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1356 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1359 in_port = 0xffffffff; /* Always exact match in_port. */
1360 else if (in_port >= DP_MAX_PORTS)
1363 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1364 *attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1365 } else if (!is_mask) {
1366 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1369 if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1370 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1372 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1373 *attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1375 if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
1376 if (ovs_ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1379 *attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
1384 static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs,
1385 const struct nlattr **a, bool is_mask)
1388 u64 orig_attrs = attrs;
1390 err = metadata_from_nlattrs(match, &attrs, a, is_mask);
1394 if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
1395 const struct ovs_key_ethernet *eth_key;
1397 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1398 SW_FLOW_KEY_MEMCPY(match, eth.src,
1399 eth_key->eth_src, ETH_ALEN, is_mask);
1400 SW_FLOW_KEY_MEMCPY(match, eth.dst,
1401 eth_key->eth_dst, ETH_ALEN, is_mask);
1402 attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1405 if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
1408 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1409 if (!(tci & htons(VLAN_TAG_PRESENT))) {
1411 OVS_NLERR("VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.\n");
1413 OVS_NLERR("VLAN TCI does not have VLAN_TAG_PRESENT bit set.\n");
1418 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
1419 attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
1420 } else if (!is_mask)
1421 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1423 if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1426 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1428 /* Always exact match EtherType. */
1429 eth_type = htons(0xffff);
1430 } else if (ntohs(eth_type) < ETH_P_802_3_MIN) {
1431 OVS_NLERR("EtherType is less than minimum (type=%x, min=%x).\n",
1432 ntohs(eth_type), ETH_P_802_3_MIN);
1436 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1437 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1438 } else if (!is_mask) {
1439 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1442 if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
1443 const struct ovs_key_ipv4 *ipv4_key;
1445 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1446 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1447 OVS_NLERR("Unknown IPv4 fragment type (value=%d, max=%d).\n",
1448 ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1451 SW_FLOW_KEY_PUT(match, ip.proto,
1452 ipv4_key->ipv4_proto, is_mask);
1453 SW_FLOW_KEY_PUT(match, ip.tos,
1454 ipv4_key->ipv4_tos, is_mask);
1455 SW_FLOW_KEY_PUT(match, ip.ttl,
1456 ipv4_key->ipv4_ttl, is_mask);
1457 SW_FLOW_KEY_PUT(match, ip.frag,
1458 ipv4_key->ipv4_frag, is_mask);
1459 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1460 ipv4_key->ipv4_src, is_mask);
1461 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1462 ipv4_key->ipv4_dst, is_mask);
1463 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1466 if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
1467 const struct ovs_key_ipv6 *ipv6_key;
1469 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1470 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1471 OVS_NLERR("Unknown IPv6 fragment type (value=%d, max=%d).\n",
1472 ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1475 SW_FLOW_KEY_PUT(match, ipv6.label,
1476 ipv6_key->ipv6_label, is_mask);
1477 SW_FLOW_KEY_PUT(match, ip.proto,
1478 ipv6_key->ipv6_proto, is_mask);
1479 SW_FLOW_KEY_PUT(match, ip.tos,
1480 ipv6_key->ipv6_tclass, is_mask);
1481 SW_FLOW_KEY_PUT(match, ip.ttl,
1482 ipv6_key->ipv6_hlimit, is_mask);
1483 SW_FLOW_KEY_PUT(match, ip.frag,
1484 ipv6_key->ipv6_frag, is_mask);
1485 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1487 sizeof(match->key->ipv6.addr.src),
1489 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1491 sizeof(match->key->ipv6.addr.dst),
1494 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1497 if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
1498 const struct ovs_key_arp *arp_key;
1500 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1501 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1502 OVS_NLERR("Unknown ARP opcode (opcode=%d).\n",
1507 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1508 arp_key->arp_sip, is_mask);
1509 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1510 arp_key->arp_tip, is_mask);
1511 SW_FLOW_KEY_PUT(match, ip.proto,
1512 ntohs(arp_key->arp_op), is_mask);
1513 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1514 arp_key->arp_sha, ETH_ALEN, is_mask);
1515 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1516 arp_key->arp_tha, ETH_ALEN, is_mask);
1518 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1521 if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
1522 const struct ovs_key_tcp *tcp_key;
1524 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1525 if (orig_attrs & (1 << OVS_KEY_ATTR_IPV4)) {
1526 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1527 tcp_key->tcp_src, is_mask);
1528 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1529 tcp_key->tcp_dst, is_mask);
1531 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1532 tcp_key->tcp_src, is_mask);
1533 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1534 tcp_key->tcp_dst, is_mask);
1536 attrs &= ~(1 << OVS_KEY_ATTR_TCP);
1539 if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
1540 const struct ovs_key_udp *udp_key;
1542 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1543 if (orig_attrs & (1 << OVS_KEY_ATTR_IPV4)) {
1544 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1545 udp_key->udp_src, is_mask);
1546 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1547 udp_key->udp_dst, is_mask);
1549 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1550 udp_key->udp_src, is_mask);
1551 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1552 udp_key->udp_dst, is_mask);
1554 attrs &= ~(1 << OVS_KEY_ATTR_UDP);
1557 if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
1558 const struct ovs_key_sctp *sctp_key;
1560 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1561 if (orig_attrs & (1 << OVS_KEY_ATTR_IPV4)) {
1562 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1563 sctp_key->sctp_src, is_mask);
1564 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1565 sctp_key->sctp_dst, is_mask);
1567 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1568 sctp_key->sctp_src, is_mask);
1569 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1570 sctp_key->sctp_dst, is_mask);
1572 attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
1575 if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1576 const struct ovs_key_icmp *icmp_key;
1578 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1579 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1580 htons(icmp_key->icmp_type), is_mask);
1581 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1582 htons(icmp_key->icmp_code), is_mask);
1583 attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1586 if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1587 const struct ovs_key_icmpv6 *icmpv6_key;
1589 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1590 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1591 htons(icmpv6_key->icmpv6_type), is_mask);
1592 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1593 htons(icmpv6_key->icmpv6_code), is_mask);
1594 attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1597 if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1598 const struct ovs_key_nd *nd_key;
1600 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1601 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1603 sizeof(match->key->ipv6.nd.target),
1605 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1606 nd_key->nd_sll, ETH_ALEN, is_mask);
1607 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1608 nd_key->nd_tll, ETH_ALEN, is_mask);
1609 attrs &= ~(1 << OVS_KEY_ATTR_ND);
1619 * ovs_match_from_nlattrs - parses Netlink attributes into a flow key and
1620 * mask. In case the 'mask' is NULL, the flow is treated as exact match
1621 * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1622 * does not include any don't care bit.
1623 * @match: receives the extracted flow match information.
1624 * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1625 * sequence. The fields should of the packet that triggered the creation
1627 * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1628 * attribute specifies the mask field of the wildcarded flow.
1630 int ovs_match_from_nlattrs(struct sw_flow_match *match,
1631 const struct nlattr *key,
1632 const struct nlattr *mask)
1634 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1635 const struct nlattr *encap;
1638 bool encap_valid = false;
1641 err = parse_flow_nlattrs(key, a, &key_attrs);
1645 if ((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
1646 (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
1647 (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1650 if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
1651 (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
1652 OVS_NLERR("Invalid Vlan frame.\n");
1656 key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1657 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1658 encap = a[OVS_KEY_ATTR_ENCAP];
1659 key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1662 if (tci & htons(VLAN_TAG_PRESENT)) {
1663 err = parse_flow_nlattrs(encap, a, &key_attrs);
1667 /* Corner case for truncated 802.1Q header. */
1668 if (nla_len(encap)) {
1669 OVS_NLERR("Truncated 802.1Q header has non-zero encap attribute.\n");
1673 OVS_NLERR("Encap attribute is set for a non-VLAN frame.\n");
1678 err = ovs_key_from_nlattrs(match, key_attrs, a, false);
1683 err = parse_flow_mask_nlattrs(mask, a, &mask_attrs);
1687 if (mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
1688 __be16 eth_type = 0;
1692 OVS_NLERR("Encap mask attribute is set for non-VLAN frame.\n");
1696 mask_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1697 if (a[OVS_KEY_ATTR_ETHERTYPE])
1698 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1700 if (eth_type == htons(0xffff)) {
1701 mask_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1702 encap = a[OVS_KEY_ATTR_ENCAP];
1703 err = parse_flow_mask_nlattrs(encap, a, &mask_attrs);
1705 OVS_NLERR("VLAN frames must have an exact match on the TPID (mask=%x).\n",
1710 if (a[OVS_KEY_ATTR_VLAN])
1711 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1713 if (!(tci & htons(VLAN_TAG_PRESENT))) {
1714 OVS_NLERR("VLAN tag present bit must have an exact match (tci_mask=%x).\n", ntohs(tci));
1719 err = ovs_key_from_nlattrs(match, mask_attrs, a, true);
1723 /* Populate exact match flow's key mask. */
1725 ovs_sw_flow_mask_set(match->mask, &match->range, 0xff);
1728 if (!ovs_match_validate(match, key_attrs, mask_attrs))
1735 * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1736 * @flow: Receives extracted in_port, priority, tun_key and skb_mark.
1737 * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1740 * This parses a series of Netlink attributes that form a flow key, which must
1741 * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1742 * get the metadata, that is, the parts of the flow key that cannot be
1743 * extracted from the packet itself.
1746 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow,
1747 const struct nlattr *attr)
1749 struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1750 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1753 struct sw_flow_match match;
1755 flow->key.phy.in_port = DP_MAX_PORTS;
1756 flow->key.phy.priority = 0;
1757 flow->key.phy.skb_mark = 0;
1758 memset(tun_key, 0, sizeof(flow->key.tun_key));
1760 err = parse_flow_nlattrs(attr, a, &attrs);
1764 memset(&match, 0, sizeof(match));
1765 match.key = &flow->key;
1767 err = metadata_from_nlattrs(&match, &attrs, a, false);
1774 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey,
1775 const struct sw_flow_key *output, struct sk_buff *skb)
1777 struct ovs_key_ethernet *eth_key;
1778 struct nlattr *nla, *encap;
1779 bool is_mask = (swkey != output);
1781 if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1782 goto nla_put_failure;
1784 if ((swkey->tun_key.ipv4_dst || is_mask) &&
1785 ovs_ipv4_tun_to_nlattr(skb, &swkey->tun_key, &output->tun_key))
1786 goto nla_put_failure;
1788 if (swkey->phy.in_port == DP_MAX_PORTS) {
1789 if (is_mask && (output->phy.in_port == 0xffff))
1790 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1791 goto nla_put_failure;
1794 upper_u16 = !is_mask ? 0 : 0xffff;
1796 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1797 (upper_u16 << 16) | output->phy.in_port))
1798 goto nla_put_failure;
1801 if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1802 goto nla_put_failure;
1804 nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1806 goto nla_put_failure;
1808 eth_key = nla_data(nla);
1809 memcpy(eth_key->eth_src, output->eth.src, ETH_ALEN);
1810 memcpy(eth_key->eth_dst, output->eth.dst, ETH_ALEN);
1812 if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1814 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1815 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1816 nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1817 goto nla_put_failure;
1818 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1819 if (!swkey->eth.tci)
1824 if (swkey->eth.type == htons(ETH_P_802_2)) {
1826 * Ethertype 802.2 is represented in the netlink with omitted
1827 * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1828 * 0xffff in the mask attribute. Ethertype can also
1831 if (is_mask && output->eth.type)
1832 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1834 goto nla_put_failure;
1838 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1839 goto nla_put_failure;
1841 if (swkey->eth.type == htons(ETH_P_IP)) {
1842 struct ovs_key_ipv4 *ipv4_key;
1844 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1846 goto nla_put_failure;
1847 ipv4_key = nla_data(nla);
1848 ipv4_key->ipv4_src = output->ipv4.addr.src;
1849 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1850 ipv4_key->ipv4_proto = output->ip.proto;
1851 ipv4_key->ipv4_tos = output->ip.tos;
1852 ipv4_key->ipv4_ttl = output->ip.ttl;
1853 ipv4_key->ipv4_frag = output->ip.frag;
1854 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1855 struct ovs_key_ipv6 *ipv6_key;
1857 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1859 goto nla_put_failure;
1860 ipv6_key = nla_data(nla);
1861 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1862 sizeof(ipv6_key->ipv6_src));
1863 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1864 sizeof(ipv6_key->ipv6_dst));
1865 ipv6_key->ipv6_label = output->ipv6.label;
1866 ipv6_key->ipv6_proto = output->ip.proto;
1867 ipv6_key->ipv6_tclass = output->ip.tos;
1868 ipv6_key->ipv6_hlimit = output->ip.ttl;
1869 ipv6_key->ipv6_frag = output->ip.frag;
1870 } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1871 swkey->eth.type == htons(ETH_P_RARP)) {
1872 struct ovs_key_arp *arp_key;
1874 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1876 goto nla_put_failure;
1877 arp_key = nla_data(nla);
1878 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1879 arp_key->arp_sip = output->ipv4.addr.src;
1880 arp_key->arp_tip = output->ipv4.addr.dst;
1881 arp_key->arp_op = htons(output->ip.proto);
1882 memcpy(arp_key->arp_sha, output->ipv4.arp.sha, ETH_ALEN);
1883 memcpy(arp_key->arp_tha, output->ipv4.arp.tha, ETH_ALEN);
1886 if ((swkey->eth.type == htons(ETH_P_IP) ||
1887 swkey->eth.type == htons(ETH_P_IPV6)) &&
1888 swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1890 if (swkey->ip.proto == IPPROTO_TCP) {
1891 struct ovs_key_tcp *tcp_key;
1893 nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1895 goto nla_put_failure;
1896 tcp_key = nla_data(nla);
1897 if (swkey->eth.type == htons(ETH_P_IP)) {
1898 tcp_key->tcp_src = output->ipv4.tp.src;
1899 tcp_key->tcp_dst = output->ipv4.tp.dst;
1900 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1901 tcp_key->tcp_src = output->ipv6.tp.src;
1902 tcp_key->tcp_dst = output->ipv6.tp.dst;
1904 } else if (swkey->ip.proto == IPPROTO_UDP) {
1905 struct ovs_key_udp *udp_key;
1907 nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1909 goto nla_put_failure;
1910 udp_key = nla_data(nla);
1911 if (swkey->eth.type == htons(ETH_P_IP)) {
1912 udp_key->udp_src = output->ipv4.tp.src;
1913 udp_key->udp_dst = output->ipv4.tp.dst;
1914 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1915 udp_key->udp_src = output->ipv6.tp.src;
1916 udp_key->udp_dst = output->ipv6.tp.dst;
1918 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1919 struct ovs_key_sctp *sctp_key;
1921 nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1923 goto nla_put_failure;
1924 sctp_key = nla_data(nla);
1925 if (swkey->eth.type == htons(ETH_P_IP)) {
1926 sctp_key->sctp_src = swkey->ipv4.tp.src;
1927 sctp_key->sctp_dst = swkey->ipv4.tp.dst;
1928 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1929 sctp_key->sctp_src = swkey->ipv6.tp.src;
1930 sctp_key->sctp_dst = swkey->ipv6.tp.dst;
1932 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1933 swkey->ip.proto == IPPROTO_ICMP) {
1934 struct ovs_key_icmp *icmp_key;
1936 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1938 goto nla_put_failure;
1939 icmp_key = nla_data(nla);
1940 icmp_key->icmp_type = ntohs(output->ipv4.tp.src);
1941 icmp_key->icmp_code = ntohs(output->ipv4.tp.dst);
1942 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1943 swkey->ip.proto == IPPROTO_ICMPV6) {
1944 struct ovs_key_icmpv6 *icmpv6_key;
1946 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1947 sizeof(*icmpv6_key));
1949 goto nla_put_failure;
1950 icmpv6_key = nla_data(nla);
1951 icmpv6_key->icmpv6_type = ntohs(output->ipv6.tp.src);
1952 icmpv6_key->icmpv6_code = ntohs(output->ipv6.tp.dst);
1954 if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1955 icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1956 struct ovs_key_nd *nd_key;
1958 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1960 goto nla_put_failure;
1961 nd_key = nla_data(nla);
1962 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1963 sizeof(nd_key->nd_target));
1964 memcpy(nd_key->nd_sll, output->ipv6.nd.sll, ETH_ALEN);
1965 memcpy(nd_key->nd_tll, output->ipv6.nd.tll, ETH_ALEN);
1972 nla_nest_end(skb, encap);
1980 /* Initializes the flow module.
1981 * Returns zero if successful or a negative error code. */
1982 int ovs_flow_init(void)
1984 BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long));
1985 BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
1987 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1989 if (flow_cache == NULL)
1995 /* Uninitializes the flow module. */
1996 void ovs_flow_exit(void)
1998 kmem_cache_destroy(flow_cache);
2001 struct sw_flow_mask *ovs_sw_flow_mask_alloc(void)
2003 struct sw_flow_mask *mask;
2005 mask = kmalloc(sizeof(*mask), GFP_KERNEL);
2007 mask->ref_count = 0;
2012 void ovs_sw_flow_mask_add_ref(struct sw_flow_mask *mask)
2017 void ovs_sw_flow_mask_del_ref(struct sw_flow_mask *mask, bool deferred)
2022 BUG_ON(!mask->ref_count);
2025 if (!mask->ref_count) {
2026 list_del_rcu(&mask->list);
2028 kfree_rcu(mask, rcu);
2034 static bool ovs_sw_flow_mask_equal(const struct sw_flow_mask *a,
2035 const struct sw_flow_mask *b)
2037 u8 *a_ = (u8 *)&a->key + a->range.start;
2038 u8 *b_ = (u8 *)&b->key + b->range.start;
2040 return (a->range.end == b->range.end)
2041 && (a->range.start == b->range.start)
2042 && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0);
2045 struct sw_flow_mask *ovs_sw_flow_mask_find(const struct flow_table *tbl,
2046 const struct sw_flow_mask *mask)
2048 struct list_head *ml;
2050 list_for_each(ml, tbl->mask_list) {
2051 struct sw_flow_mask *m;
2052 m = container_of(ml, struct sw_flow_mask, list);
2053 if (ovs_sw_flow_mask_equal(mask, m))
2061 * add a new mask into the mask list.
2062 * The caller needs to make sure that 'mask' is not the same
2063 * as any masks that are already on the list.
2065 void ovs_sw_flow_mask_insert(struct flow_table *tbl, struct sw_flow_mask *mask)
2067 list_add_rcu(&mask->list, tbl->mask_list);
2071 * Set 'range' fields in the mask to the value of 'val'.
2073 static void ovs_sw_flow_mask_set(struct sw_flow_mask *mask,
2074 struct sw_flow_key_range *range, u8 val)
2076 u8 *m = (u8 *)&mask->key + range->start;
2078 mask->range = *range;
2079 memset(m, val, range_n_bytes(range));