2 * Copyright (c) 2007-2013 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
21 #include <linux/uaccess.h>
22 #include <linux/netdevice.h>
23 #include <linux/etherdevice.h>
24 #include <linux/if_ether.h>
25 #include <linux/if_vlan.h>
26 #include <net/llc_pdu.h>
27 #include <linux/kernel.h>
28 #include <linux/jhash.h>
29 #include <linux/jiffies.h>
30 #include <linux/llc.h>
31 #include <linux/module.h>
33 #include <linux/rcupdate.h>
34 #include <linux/if_arp.h>
36 #include <linux/ipv6.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/icmp.h>
40 #include <linux/icmpv6.h>
41 #include <linux/rculist.h>
44 #include <net/ndisc.h>
48 static struct kmem_cache *flow_cache;
50 static void ovs_sw_flow_mask_set(struct sw_flow_mask *mask,
51 struct sw_flow_key_range *range, u8 val);
53 static void update_range__(struct sw_flow_match *match,
54 size_t offset, size_t size, bool is_mask)
56 struct sw_flow_key_range *range = NULL;
57 size_t start = offset;
58 size_t end = offset + size;
61 range = &match->range;
63 range = &match->mask->range;
68 if (range->start == range->end) {
74 if (range->start > start)
81 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
83 update_range__(match, offsetof(struct sw_flow_key, field), \
84 sizeof((match)->key->field), is_mask); \
87 (match)->mask->key.field = value; \
89 (match)->key->field = value; \
93 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask) \
95 update_range__(match, offsetof(struct sw_flow_key, field), \
99 memcpy(&(match)->mask->key.field, value_p, len);\
101 memcpy(&(match)->key->field, value_p, len); \
105 void ovs_match_init(struct sw_flow_match *match,
106 struct sw_flow_key *key,
107 struct sw_flow_mask *mask)
109 memset(match, 0, sizeof(*match));
113 memset(key, 0, sizeof(*key));
116 memset(&mask->key, 0, sizeof(mask->key));
117 mask->range.start = mask->range.end = 0;
121 static bool ovs_match_validate(const struct sw_flow_match *match,
122 u64 key_attrs, u64 mask_attrs)
124 u64 key_expected = 1ULL << OVS_KEY_ATTR_ETHERNET;
125 u64 mask_allowed = key_attrs; /* At most allow all key attributes */
127 /* The following mask attributes allowed only if they
128 * pass the validation tests. */
129 mask_allowed &= ~((1ULL << OVS_KEY_ATTR_IPV4)
130 | (1ULL << OVS_KEY_ATTR_IPV6)
131 | (1ULL << OVS_KEY_ATTR_TCP)
132 | (1ULL << OVS_KEY_ATTR_UDP)
133 | (1ULL << OVS_KEY_ATTR_ICMP)
134 | (1ULL << OVS_KEY_ATTR_ICMPV6)
135 | (1ULL << OVS_KEY_ATTR_ARP)
136 | (1ULL << OVS_KEY_ATTR_ND));
138 if (match->key->phy.in_port == DP_MAX_PORTS &&
139 match->mask && (match->mask->key.phy.in_port == 0xffff))
140 mask_allowed |= (1ULL << OVS_KEY_ATTR_IN_PORT);
142 if (match->key->eth.type == htons(ETH_P_802_2) &&
143 match->mask && (match->mask->key.eth.type == htons(0xffff)))
144 mask_allowed |= (1ULL << OVS_KEY_ATTR_ETHERTYPE);
146 /* Check key attributes. */
147 if (match->key->eth.type == htons(ETH_P_ARP)
148 || match->key->eth.type == htons(ETH_P_RARP)) {
149 key_expected |= 1ULL << OVS_KEY_ATTR_ARP;
150 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
151 mask_allowed |= 1ULL << OVS_KEY_ATTR_ARP;
154 if (match->key->eth.type == htons(ETH_P_IP)) {
155 key_expected |= 1ULL << OVS_KEY_ATTR_IPV4;
156 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
157 mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV4;
159 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
160 if (match->key->ip.proto == IPPROTO_UDP) {
161 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
162 if (match->mask && (match->mask->key.ip.proto == 0xff))
163 mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
166 if (match->key->ip.proto == IPPROTO_TCP) {
167 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
168 if (match->mask && (match->mask->key.ip.proto == 0xff))
169 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
172 if (match->key->ip.proto == IPPROTO_ICMP) {
173 key_expected |= 1ULL << OVS_KEY_ATTR_ICMP;
174 if (match->mask && (match->mask->key.ip.proto == 0xff))
175 mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMP;
180 if (match->key->eth.type == htons(ETH_P_IPV6)) {
181 key_expected |= 1ULL << OVS_KEY_ATTR_IPV6;
182 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
183 mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV6;
185 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
186 if (match->key->ip.proto == IPPROTO_UDP) {
187 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
188 if (match->mask && (match->mask->key.ip.proto == 0xff))
189 mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
192 if (match->key->ip.proto == IPPROTO_TCP) {
193 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
194 if (match->mask && (match->mask->key.ip.proto == 0xff))
195 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
198 if (match->key->ip.proto == IPPROTO_ICMPV6) {
199 key_expected |= 1ULL << OVS_KEY_ATTR_ICMPV6;
200 if (match->mask && (match->mask->key.ip.proto == 0xff))
201 mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMPV6;
203 if (match->key->ipv6.tp.src ==
204 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
205 match->key->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
206 key_expected |= 1ULL << OVS_KEY_ATTR_ND;
207 if (match->mask && (match->mask->key.ipv6.tp.src == htons(0xffff)))
208 mask_allowed |= 1ULL << OVS_KEY_ATTR_ND;
214 if ((key_attrs & key_expected) != key_expected) {
215 /* Key attributes check failed. */
216 OVS_NLERR("Missing expected key attributes (key_attrs=%llx, expected=%llx).\n",
217 key_attrs, key_expected);
221 if ((mask_attrs & mask_allowed) != mask_attrs) {
222 /* Mask attributes check failed. */
223 OVS_NLERR("Contain more than allowed mask fields (mask_attrs=%llx, mask_allowed=%llx).\n",
224 mask_attrs, mask_allowed);
231 static int check_header(struct sk_buff *skb, int len)
233 if (unlikely(skb->len < len))
235 if (unlikely(!pskb_may_pull(skb, len)))
240 static bool arphdr_ok(struct sk_buff *skb)
242 return pskb_may_pull(skb, skb_network_offset(skb) +
243 sizeof(struct arp_eth_header));
246 static int check_iphdr(struct sk_buff *skb)
248 unsigned int nh_ofs = skb_network_offset(skb);
252 err = check_header(skb, nh_ofs + sizeof(struct iphdr));
256 ip_len = ip_hdrlen(skb);
257 if (unlikely(ip_len < sizeof(struct iphdr) ||
258 skb->len < nh_ofs + ip_len))
261 skb_set_transport_header(skb, nh_ofs + ip_len);
265 static bool tcphdr_ok(struct sk_buff *skb)
267 int th_ofs = skb_transport_offset(skb);
270 if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
273 tcp_len = tcp_hdrlen(skb);
274 if (unlikely(tcp_len < sizeof(struct tcphdr) ||
275 skb->len < th_ofs + tcp_len))
281 static bool udphdr_ok(struct sk_buff *skb)
283 return pskb_may_pull(skb, skb_transport_offset(skb) +
284 sizeof(struct udphdr));
287 static bool icmphdr_ok(struct sk_buff *skb)
289 return pskb_may_pull(skb, skb_transport_offset(skb) +
290 sizeof(struct icmphdr));
293 u64 ovs_flow_used_time(unsigned long flow_jiffies)
295 struct timespec cur_ts;
298 ktime_get_ts(&cur_ts);
299 idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
300 cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
301 cur_ts.tv_nsec / NSEC_PER_MSEC;
303 return cur_ms - idle_ms;
306 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
308 unsigned int nh_ofs = skb_network_offset(skb);
316 err = check_header(skb, nh_ofs + sizeof(*nh));
321 nexthdr = nh->nexthdr;
322 payload_ofs = (u8 *)(nh + 1) - skb->data;
324 key->ip.proto = NEXTHDR_NONE;
325 key->ip.tos = ipv6_get_dsfield(nh);
326 key->ip.ttl = nh->hop_limit;
327 key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
328 key->ipv6.addr.src = nh->saddr;
329 key->ipv6.addr.dst = nh->daddr;
331 payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
332 if (unlikely(payload_ofs < 0))
336 if (frag_off & htons(~0x7))
337 key->ip.frag = OVS_FRAG_TYPE_LATER;
339 key->ip.frag = OVS_FRAG_TYPE_FIRST;
342 nh_len = payload_ofs - nh_ofs;
343 skb_set_transport_header(skb, nh_ofs + nh_len);
344 key->ip.proto = nexthdr;
348 static bool icmp6hdr_ok(struct sk_buff *skb)
350 return pskb_may_pull(skb, skb_transport_offset(skb) +
351 sizeof(struct icmp6hdr));
354 void ovs_flow_key_mask(struct sw_flow_key *dst, const struct sw_flow_key *src,
355 const struct sw_flow_mask *mask)
357 u8 *m = (u8 *)&mask->key + mask->range.start;
358 u8 *s = (u8 *)src + mask->range.start;
359 u8 *d = (u8 *)dst + mask->range.start;
362 memset(dst, 0, sizeof(*dst));
363 for (i = 0; i < ovs_sw_flow_mask_size_roundup(mask); i++) {
369 #define TCP_FLAGS_OFFSET 13
370 #define TCP_FLAG_MASK 0x3f
372 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
376 if ((flow->key.eth.type == htons(ETH_P_IP) ||
377 flow->key.eth.type == htons(ETH_P_IPV6)) &&
378 flow->key.ip.proto == IPPROTO_TCP &&
379 likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
380 u8 *tcp = (u8 *)tcp_hdr(skb);
381 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
384 spin_lock(&flow->lock);
385 flow->used = jiffies;
386 flow->packet_count++;
387 flow->byte_count += skb->len;
388 flow->tcp_flags |= tcp_flags;
389 spin_unlock(&flow->lock);
392 struct sw_flow_actions *ovs_flow_actions_alloc(int size)
394 struct sw_flow_actions *sfa;
396 if (size > MAX_ACTIONS_BUFSIZE)
397 return ERR_PTR(-EINVAL);
399 sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
401 return ERR_PTR(-ENOMEM);
403 sfa->actions_len = 0;
407 struct sw_flow *ovs_flow_alloc(void)
409 struct sw_flow *flow;
411 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
413 return ERR_PTR(-ENOMEM);
415 spin_lock_init(&flow->lock);
416 flow->sf_acts = NULL;
422 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
424 hash = jhash_1word(hash, table->hash_seed);
425 return flex_array_get(table->buckets,
426 (hash & (table->n_buckets - 1)));
429 static struct flex_array *alloc_buckets(unsigned int n_buckets)
431 struct flex_array *buckets;
434 buckets = flex_array_alloc(sizeof(struct hlist_head),
435 n_buckets, GFP_KERNEL);
439 err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
441 flex_array_free(buckets);
445 for (i = 0; i < n_buckets; i++)
446 INIT_HLIST_HEAD((struct hlist_head *)
447 flex_array_get(buckets, i));
452 static void free_buckets(struct flex_array *buckets)
454 flex_array_free(buckets);
457 static struct flow_table *__flow_tbl_alloc(int new_size)
459 struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
464 table->buckets = alloc_buckets(new_size);
466 if (!table->buckets) {
470 table->n_buckets = new_size;
473 table->keep_flows = false;
474 get_random_bytes(&table->hash_seed, sizeof(u32));
475 table->mask_list = NULL;
480 static void __flow_tbl_destroy(struct flow_table *table)
484 if (table->keep_flows)
487 for (i = 0; i < table->n_buckets; i++) {
488 struct sw_flow *flow;
489 struct hlist_head *head = flex_array_get(table->buckets, i);
490 struct hlist_node *n;
491 int ver = table->node_ver;
493 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
494 hlist_del(&flow->hash_node[ver]);
495 ovs_flow_free(flow, false);
499 BUG_ON(!list_empty(table->mask_list));
500 kfree(table->mask_list);
503 free_buckets(table->buckets);
507 struct flow_table *ovs_flow_tbl_alloc(int new_size)
509 struct flow_table *table = __flow_tbl_alloc(new_size);
514 table->mask_list = kmalloc(sizeof(struct list_head), GFP_KERNEL);
515 if (!table->mask_list) {
516 table->keep_flows = true;
517 __flow_tbl_destroy(table);
520 INIT_LIST_HEAD(table->mask_list);
525 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
527 struct flow_table *table = container_of(rcu, struct flow_table, rcu);
529 __flow_tbl_destroy(table);
532 void ovs_flow_tbl_destroy(struct flow_table *table, bool deferred)
538 call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
540 __flow_tbl_destroy(table);
543 struct sw_flow *ovs_flow_dump_next(struct flow_table *table, u32 *bucket, u32 *last)
545 struct sw_flow *flow;
546 struct hlist_head *head;
550 ver = table->node_ver;
551 while (*bucket < table->n_buckets) {
553 head = flex_array_get(table->buckets, *bucket);
554 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
569 static void __tbl_insert(struct flow_table *table, struct sw_flow *flow)
571 struct hlist_head *head;
573 head = find_bucket(table, flow->hash);
574 hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
579 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
584 old_ver = old->node_ver;
585 new->node_ver = !old_ver;
587 /* Insert in new table. */
588 for (i = 0; i < old->n_buckets; i++) {
589 struct sw_flow *flow;
590 struct hlist_head *head;
592 head = flex_array_get(old->buckets, i);
594 hlist_for_each_entry(flow, head, hash_node[old_ver])
595 __tbl_insert(new, flow);
598 new->mask_list = old->mask_list;
599 old->keep_flows = true;
602 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
604 struct flow_table *new_table;
606 new_table = __flow_tbl_alloc(n_buckets);
608 return ERR_PTR(-ENOMEM);
610 flow_table_copy_flows(table, new_table);
615 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
617 return __flow_tbl_rehash(table, table->n_buckets);
620 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
622 return __flow_tbl_rehash(table, table->n_buckets * 2);
625 static void __flow_free(struct sw_flow *flow)
627 kfree((struct sf_flow_acts __force *)flow->sf_acts);
628 kmem_cache_free(flow_cache, flow);
631 static void rcu_free_flow_callback(struct rcu_head *rcu)
633 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
638 void ovs_flow_free(struct sw_flow *flow, bool deferred)
643 ovs_sw_flow_mask_del_ref(flow->mask, deferred);
646 call_rcu(&flow->rcu, rcu_free_flow_callback);
651 /* RCU callback used by ovs_flow_deferred_free_acts. */
652 static void rcu_free_acts_callback(struct rcu_head *rcu)
654 struct sw_flow_actions *sf_acts = container_of(rcu,
655 struct sw_flow_actions, rcu);
659 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
660 * The caller must hold rcu_read_lock for this to be sensible. */
661 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
663 call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
666 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
669 __be16 eth_type; /* ETH_P_8021Q */
672 struct qtag_prefix *qp;
674 if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
677 if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
681 qp = (struct qtag_prefix *) skb->data;
682 key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
683 __skb_pull(skb, sizeof(struct qtag_prefix));
688 static __be16 parse_ethertype(struct sk_buff *skb)
690 struct llc_snap_hdr {
691 u8 dsap; /* Always 0xAA */
692 u8 ssap; /* Always 0xAA */
697 struct llc_snap_hdr *llc;
700 proto = *(__be16 *) skb->data;
701 __skb_pull(skb, sizeof(__be16));
703 if (ntohs(proto) >= ETH_P_802_3_MIN)
706 if (skb->len < sizeof(struct llc_snap_hdr))
707 return htons(ETH_P_802_2);
709 if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
712 llc = (struct llc_snap_hdr *) skb->data;
713 if (llc->dsap != LLC_SAP_SNAP ||
714 llc->ssap != LLC_SAP_SNAP ||
715 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
716 return htons(ETH_P_802_2);
718 __skb_pull(skb, sizeof(struct llc_snap_hdr));
720 if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
721 return llc->ethertype;
723 return htons(ETH_P_802_2);
726 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
729 struct icmp6hdr *icmp = icmp6_hdr(skb);
731 /* The ICMPv6 type and code fields use the 16-bit transport port
732 * fields, so we need to store them in 16-bit network byte order.
734 key->ipv6.tp.src = htons(icmp->icmp6_type);
735 key->ipv6.tp.dst = htons(icmp->icmp6_code);
737 if (icmp->icmp6_code == 0 &&
738 (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
739 icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
740 int icmp_len = skb->len - skb_transport_offset(skb);
744 /* In order to process neighbor discovery options, we need the
747 if (unlikely(icmp_len < sizeof(*nd)))
750 if (unlikely(skb_linearize(skb)))
753 nd = (struct nd_msg *)skb_transport_header(skb);
754 key->ipv6.nd.target = nd->target;
756 icmp_len -= sizeof(*nd);
758 while (icmp_len >= 8) {
759 struct nd_opt_hdr *nd_opt =
760 (struct nd_opt_hdr *)(nd->opt + offset);
761 int opt_len = nd_opt->nd_opt_len * 8;
763 if (unlikely(!opt_len || opt_len > icmp_len))
766 /* Store the link layer address if the appropriate
767 * option is provided. It is considered an error if
768 * the same link layer option is specified twice.
770 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
772 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
774 memcpy(key->ipv6.nd.sll,
775 &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
776 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
778 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
780 memcpy(key->ipv6.nd.tll,
781 &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
792 memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
793 memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
794 memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
800 * ovs_flow_extract - extracts a flow key from an Ethernet frame.
801 * @skb: sk_buff that contains the frame, with skb->data pointing to the
803 * @in_port: port number on which @skb was received.
804 * @key: output flow key
805 * @key_lenp: length of output flow key
807 * The caller must ensure that skb->len >= ETH_HLEN.
809 * Returns 0 if successful, otherwise a negative errno value.
811 * Initializes @skb header pointers as follows:
813 * - skb->mac_header: the Ethernet header.
815 * - skb->network_header: just past the Ethernet header, or just past the
816 * VLAN header, to the first byte of the Ethernet payload.
818 * - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
819 * on output, then just past the IP header, if one is present and
820 * of a correct length, otherwise the same as skb->network_header.
821 * For other key->eth.type values it is left untouched.
823 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
828 memset(key, 0, sizeof(*key));
830 key->phy.priority = skb->priority;
831 if (OVS_CB(skb)->tun_key)
832 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
833 key->phy.in_port = in_port;
834 key->phy.skb_mark = skb_get_mark(skb);
836 skb_reset_mac_header(skb);
838 /* Link layer. We are guaranteed to have at least the 14 byte Ethernet
839 * header in the linear data area.
842 memcpy(key->eth.src, eth->h_source, ETH_ALEN);
843 memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
845 __skb_pull(skb, 2 * ETH_ALEN);
846 /* We are going to push all headers that we pull, so no need to
847 * update skb->csum here. */
849 if (vlan_tx_tag_present(skb))
850 key->eth.tci = htons(vlan_get_tci(skb));
851 else if (eth->h_proto == htons(ETH_P_8021Q))
852 if (unlikely(parse_vlan(skb, key)))
855 key->eth.type = parse_ethertype(skb);
856 if (unlikely(key->eth.type == htons(0)))
859 skb_reset_network_header(skb);
860 __skb_push(skb, skb->data - skb_mac_header(skb));
863 if (key->eth.type == htons(ETH_P_IP)) {
867 error = check_iphdr(skb);
868 if (unlikely(error)) {
869 if (error == -EINVAL) {
870 skb->transport_header = skb->network_header;
877 key->ipv4.addr.src = nh->saddr;
878 key->ipv4.addr.dst = nh->daddr;
880 key->ip.proto = nh->protocol;
881 key->ip.tos = nh->tos;
882 key->ip.ttl = nh->ttl;
884 offset = nh->frag_off & htons(IP_OFFSET);
886 key->ip.frag = OVS_FRAG_TYPE_LATER;
889 if (nh->frag_off & htons(IP_MF) ||
890 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
891 key->ip.frag = OVS_FRAG_TYPE_FIRST;
893 /* Transport layer. */
894 if (key->ip.proto == IPPROTO_TCP) {
895 if (tcphdr_ok(skb)) {
896 struct tcphdr *tcp = tcp_hdr(skb);
897 key->ipv4.tp.src = tcp->source;
898 key->ipv4.tp.dst = tcp->dest;
900 } else if (key->ip.proto == IPPROTO_UDP) {
901 if (udphdr_ok(skb)) {
902 struct udphdr *udp = udp_hdr(skb);
903 key->ipv4.tp.src = udp->source;
904 key->ipv4.tp.dst = udp->dest;
906 } else if (key->ip.proto == IPPROTO_ICMP) {
907 if (icmphdr_ok(skb)) {
908 struct icmphdr *icmp = icmp_hdr(skb);
909 /* The ICMP type and code fields use the 16-bit
910 * transport port fields, so we need to store
911 * them in 16-bit network byte order. */
912 key->ipv4.tp.src = htons(icmp->type);
913 key->ipv4.tp.dst = htons(icmp->code);
917 } else if ((key->eth.type == htons(ETH_P_ARP) ||
918 key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
919 struct arp_eth_header *arp;
921 arp = (struct arp_eth_header *)skb_network_header(skb);
923 if (arp->ar_hrd == htons(ARPHRD_ETHER)
924 && arp->ar_pro == htons(ETH_P_IP)
925 && arp->ar_hln == ETH_ALEN
926 && arp->ar_pln == 4) {
928 /* We only match on the lower 8 bits of the opcode. */
929 if (ntohs(arp->ar_op) <= 0xff)
930 key->ip.proto = ntohs(arp->ar_op);
931 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
932 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
933 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
934 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
936 } else if (key->eth.type == htons(ETH_P_IPV6)) {
937 int nh_len; /* IPv6 Header + Extensions */
939 nh_len = parse_ipv6hdr(skb, key);
940 if (unlikely(nh_len < 0)) {
941 if (nh_len == -EINVAL) {
942 skb->transport_header = skb->network_header;
950 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
952 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
953 key->ip.frag = OVS_FRAG_TYPE_FIRST;
955 /* Transport layer. */
956 if (key->ip.proto == NEXTHDR_TCP) {
957 if (tcphdr_ok(skb)) {
958 struct tcphdr *tcp = tcp_hdr(skb);
959 key->ipv6.tp.src = tcp->source;
960 key->ipv6.tp.dst = tcp->dest;
962 } else if (key->ip.proto == NEXTHDR_UDP) {
963 if (udphdr_ok(skb)) {
964 struct udphdr *udp = udp_hdr(skb);
965 key->ipv6.tp.src = udp->source;
966 key->ipv6.tp.dst = udp->dest;
968 } else if (key->ip.proto == NEXTHDR_ICMP) {
969 if (icmp6hdr_ok(skb)) {
970 error = parse_icmpv6(skb, key, nh_len);
980 static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start, int key_len)
982 return jhash2((u32 *)((u8 *)key + key_start),
983 DIV_ROUND_UP(key_len - key_start, sizeof(u32)), 0);
986 static int flow_key_start(const struct sw_flow_key *key)
988 if (key->tun_key.ipv4_dst)
991 return offsetof(struct sw_flow_key, phy);
994 static bool __cmp_key(const struct sw_flow_key *key1,
995 const struct sw_flow_key *key2, int key_start, int key_len)
997 return !memcmp((u8 *)key1 + key_start,
998 (u8 *)key2 + key_start, (key_len - key_start));
1001 static bool __flow_cmp_key(const struct sw_flow *flow,
1002 const struct sw_flow_key *key, int key_start, int key_len)
1004 return __cmp_key(&flow->key, key, key_start, key_len);
1007 static bool __flow_cmp_unmasked_key(const struct sw_flow *flow,
1008 const struct sw_flow_key *key, int key_start, int key_len)
1010 return __cmp_key(&flow->unmasked_key, key, key_start, key_len);
1013 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
1014 const struct sw_flow_key *key, int key_len)
1017 key_start = flow_key_start(key);
1019 return __flow_cmp_unmasked_key(flow, key, key_start, key_len);
1023 struct sw_flow *ovs_flow_lookup_unmasked_key(struct flow_table *table,
1024 struct sw_flow_match *match)
1026 struct sw_flow_key *unmasked = match->key;
1027 int key_len = match->range.end;
1028 struct sw_flow *flow;
1030 flow = ovs_flow_lookup(table, unmasked);
1031 if (flow && (!ovs_flow_cmp_unmasked_key(flow, unmasked, key_len)))
1037 static struct sw_flow *ovs_masked_flow_lookup(struct flow_table *table,
1038 const struct sw_flow_key *flow_key,
1039 struct sw_flow_mask *mask)
1041 struct sw_flow *flow;
1042 struct hlist_head *head;
1043 int key_start = mask->range.start;
1044 int key_len = mask->range.end;
1046 struct sw_flow_key masked_key;
1048 ovs_flow_key_mask(&masked_key, flow_key, mask);
1049 hash = ovs_flow_hash(&masked_key, key_start, key_len);
1050 head = find_bucket(table, hash);
1051 hlist_for_each_entry_rcu(flow, head, hash_node[table->node_ver]) {
1052 if (flow->mask == mask &&
1053 __flow_cmp_key(flow, &masked_key, key_start, key_len))
1059 struct sw_flow *ovs_flow_lookup(struct flow_table *tbl,
1060 const struct sw_flow_key *key)
1062 struct sw_flow *flow = NULL;
1063 struct sw_flow_mask *mask;
1065 list_for_each_entry_rcu(mask, tbl->mask_list, list) {
1066 flow = ovs_masked_flow_lookup(tbl, key, mask);
1067 if (flow) /* Found */
1075 void ovs_flow_insert(struct flow_table *table, struct sw_flow *flow)
1077 flow->hash = ovs_flow_hash(&flow->key, flow->mask->range.start,
1078 flow->mask->range.end);
1079 __tbl_insert(table, flow);
1082 void ovs_flow_remove(struct flow_table *table, struct sw_flow *flow)
1084 BUG_ON(table->count == 0);
1085 hlist_del_rcu(&flow->hash_node[table->node_ver]);
1089 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute. */
1090 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
1091 [OVS_KEY_ATTR_ENCAP] = -1,
1092 [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
1093 [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
1094 [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
1095 [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
1096 [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
1097 [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
1098 [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
1099 [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
1100 [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
1101 [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
1102 [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
1103 [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
1104 [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
1105 [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
1106 [OVS_KEY_ATTR_TUNNEL] = -1,
1109 static bool is_all_zero(const u8 *fp, size_t size)
1116 for (i = 0; i < size; i++)
1123 static int __parse_flow_nlattrs(const struct nlattr *attr,
1124 const struct nlattr *a[],
1125 u64 *attrsp, bool nz)
1127 const struct nlattr *nla;
1132 nla_for_each_nested(nla, attr, rem) {
1133 u16 type = nla_type(nla);
1136 if (type > OVS_KEY_ATTR_MAX) {
1137 OVS_NLERR("Unknown key attribute (type=%d, max=%d).\n",
1138 type, OVS_KEY_ATTR_MAX);
1141 if (attrs & (1ULL << type)) {
1142 OVS_NLERR("Duplicate key attribute (type %d).\n", type);
1146 expected_len = ovs_key_lens[type];
1147 if (nla_len(nla) != expected_len && expected_len != -1) {
1148 OVS_NLERR("Key attribute has unexpected length (type=%d"
1149 ", length=%d, expected=%d).\n", type,
1150 nla_len(nla), expected_len);
1154 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
1155 attrs |= 1ULL << type;
1160 OVS_NLERR("Message has %d unknown bytes.\n", rem);
1168 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
1169 const struct nlattr *a[], u64 *attrsp)
1171 return __parse_flow_nlattrs(attr, a, attrsp, true);
1174 static int parse_flow_nlattrs(const struct nlattr *attr,
1175 const struct nlattr *a[], u64 *attrsp)
1177 return __parse_flow_nlattrs(attr, a, attrsp, false);
1180 int ipv4_tun_from_nlattr(const struct nlattr *attr,
1181 struct sw_flow_match *match, bool is_mask)
1186 __be16 tun_flags = 0;
1188 nla_for_each_nested(a, attr, rem) {
1189 int type = nla_type(a);
1190 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
1191 [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
1192 [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
1193 [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
1194 [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
1195 [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
1196 [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
1197 [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
1200 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
1201 OVS_NLERR("Unknown IPv4 tunnel attribute (type=%d, max=%d).\n",
1202 type, OVS_TUNNEL_KEY_ATTR_MAX);
1206 if (ovs_tunnel_key_lens[type] != nla_len(a)) {
1207 OVS_NLERR("IPv4 tunnel attribute type has unexpected "
1208 " legnth (type=%d, length=%d, expected=%d).\n",
1209 type, nla_len(a), ovs_tunnel_key_lens[type]);
1214 case OVS_TUNNEL_KEY_ATTR_ID:
1215 SW_FLOW_KEY_PUT(match, tun_key.tun_id,
1216 nla_get_be64(a), is_mask);
1217 tun_flags |= TUNNEL_KEY;
1219 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1220 SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
1221 nla_get_be32(a), is_mask);
1223 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1224 SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
1225 nla_get_be32(a), is_mask);
1227 case OVS_TUNNEL_KEY_ATTR_TOS:
1228 SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
1229 nla_get_u8(a), is_mask);
1231 case OVS_TUNNEL_KEY_ATTR_TTL:
1232 SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
1233 nla_get_u8(a), is_mask);
1236 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1237 tun_flags |= TUNNEL_DONT_FRAGMENT;
1239 case OVS_TUNNEL_KEY_ATTR_CSUM:
1240 tun_flags |= TUNNEL_CSUM;
1247 SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
1250 OVS_NLERR("IPv4 tunnel attribute has %d unknown bytes.\n", rem);
1254 if (!match->key->tun_key.ipv4_dst) {
1255 OVS_NLERR("IPv4 tunnel destination address is zero.\n");
1260 OVS_NLERR("IPv4 tunnel TTL not specified.\n");
1267 int ipv4_tun_to_nlattr(struct sk_buff *skb,
1268 const struct ovs_key_ipv4_tunnel *tun_key,
1269 const struct ovs_key_ipv4_tunnel *output)
1273 nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
1277 if (output->tun_flags & TUNNEL_KEY &&
1278 nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
1280 if (output->ipv4_src &&
1281 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
1283 if (output->ipv4_dst &&
1284 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
1286 if (output->ipv4_tos &&
1287 nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
1289 if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
1291 if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
1292 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
1294 if ((output->tun_flags & TUNNEL_CSUM) &&
1295 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
1298 nla_nest_end(skb, nla);
1303 static int metadata_from_nlattrs(struct sw_flow_match *match, u64 *attrs,
1304 const struct nlattr **a, bool is_mask)
1306 if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
1307 SW_FLOW_KEY_PUT(match, phy.priority,
1308 nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1309 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
1312 if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
1313 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1315 if (!is_mask && in_port >= DP_MAX_PORTS)
1317 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1318 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
1319 } else if (!is_mask) {
1320 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1323 if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
1324 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1325 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1326 if (!is_mask && mark != 0) {
1327 OVS_NLERR("skb->mark must be zero on this kernel (mark=%d).\n", mark);
1331 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1332 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
1334 if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
1335 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1338 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
1343 static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs,
1344 const struct nlattr **a, bool is_mask)
1347 u64 orig_attrs = attrs;
1349 err = metadata_from_nlattrs(match, &attrs, a, is_mask);
1353 if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
1354 const struct ovs_key_ethernet *eth_key;
1356 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1357 SW_FLOW_KEY_MEMCPY(match, eth.src,
1358 eth_key->eth_src, ETH_ALEN, is_mask);
1359 SW_FLOW_KEY_MEMCPY(match, eth.dst,
1360 eth_key->eth_dst, ETH_ALEN, is_mask);
1361 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
1364 if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
1367 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1368 if (!(tci & htons(VLAN_TAG_PRESENT))) {
1370 OVS_NLERR("VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.\n");
1372 OVS_NLERR("VLAN TCI does not have VLAN_TAG_PRESENT bit set.\n");
1377 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
1378 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
1379 } else if (!is_mask)
1380 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1382 if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
1385 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1386 if (!is_mask && ntohs(eth_type) < ETH_P_802_3_MIN) {
1387 OVS_NLERR("EtherType is less than mimimum (type=%x, min=%x).\n",
1388 ntohs(eth_type), ETH_P_802_3_MIN);
1392 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1393 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1394 } else if (!is_mask) {
1395 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1398 if (attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1399 const struct ovs_key_ipv4 *ipv4_key;
1401 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1402 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1403 OVS_NLERR("Unknown IPv4 fragment type (value=%d, max=%d).\n",
1404 ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1407 SW_FLOW_KEY_PUT(match, ip.proto,
1408 ipv4_key->ipv4_proto, is_mask);
1409 SW_FLOW_KEY_PUT(match, ip.tos,
1410 ipv4_key->ipv4_tos, is_mask);
1411 SW_FLOW_KEY_PUT(match, ip.ttl,
1412 ipv4_key->ipv4_ttl, is_mask);
1413 SW_FLOW_KEY_PUT(match, ip.frag,
1414 ipv4_key->ipv4_frag, is_mask);
1415 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1416 ipv4_key->ipv4_src, is_mask);
1417 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1418 ipv4_key->ipv4_dst, is_mask);
1419 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4);
1422 if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
1423 const struct ovs_key_ipv6 *ipv6_key;
1425 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1426 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1427 OVS_NLERR("Unknown IPv6 fragment type (value=%d, max=%d).\n",
1428 ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1431 SW_FLOW_KEY_PUT(match, ipv6.label,
1432 ipv6_key->ipv6_label, is_mask);
1433 SW_FLOW_KEY_PUT(match, ip.proto,
1434 ipv6_key->ipv6_proto, is_mask);
1435 SW_FLOW_KEY_PUT(match, ip.tos,
1436 ipv6_key->ipv6_tclass, is_mask);
1437 SW_FLOW_KEY_PUT(match, ip.ttl,
1438 ipv6_key->ipv6_hlimit, is_mask);
1439 SW_FLOW_KEY_PUT(match, ip.frag,
1440 ipv6_key->ipv6_frag, is_mask);
1441 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1443 sizeof(match->key->ipv6.addr.src),
1445 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1447 sizeof(match->key->ipv6.addr.dst),
1450 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
1453 if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
1454 const struct ovs_key_arp *arp_key;
1456 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1457 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1458 OVS_NLERR("Unknown ARP opcode (opcode=%d).\n",
1463 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1464 arp_key->arp_sip, is_mask);
1465 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1466 arp_key->arp_tip, is_mask);
1467 SW_FLOW_KEY_PUT(match, ip.proto,
1468 ntohs(arp_key->arp_op), is_mask);
1469 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1470 arp_key->arp_sha, ETH_ALEN, is_mask);
1471 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1472 arp_key->arp_tha, ETH_ALEN, is_mask);
1474 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
1477 if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
1478 const struct ovs_key_tcp *tcp_key;
1480 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1481 if (orig_attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1482 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1483 tcp_key->tcp_src, is_mask);
1484 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1485 tcp_key->tcp_dst, is_mask);
1487 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1488 tcp_key->tcp_src, is_mask);
1489 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1490 tcp_key->tcp_dst, is_mask);
1492 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
1495 if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
1496 const struct ovs_key_udp *udp_key;
1498 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1499 if (orig_attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1500 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1501 udp_key->udp_src, is_mask);
1502 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1503 udp_key->udp_dst, is_mask);
1505 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1506 udp_key->udp_src, is_mask);
1507 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1508 udp_key->udp_dst, is_mask);
1510 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
1513 if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
1514 const struct ovs_key_icmp *icmp_key;
1516 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1517 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1518 htons(icmp_key->icmp_type), is_mask);
1519 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1520 htons(icmp_key->icmp_code), is_mask);
1521 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
1524 if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
1525 const struct ovs_key_icmpv6 *icmpv6_key;
1527 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1528 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1529 htons(icmpv6_key->icmpv6_type), is_mask);
1530 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1531 htons(icmpv6_key->icmpv6_code), is_mask);
1532 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
1535 if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
1536 const struct ovs_key_nd *nd_key;
1538 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1539 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1541 sizeof(match->key->ipv6.nd.target),
1543 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1544 nd_key->nd_sll, ETH_ALEN, is_mask);
1545 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1546 nd_key->nd_tll, ETH_ALEN, is_mask);
1547 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
1557 * ovs_match_from_nlattrs - parses Netlink attributes into a flow key and
1558 * mask. In case the 'mask' is NULL, the flow is treated as exact match
1559 * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1560 * does not include any don't care bit.
1561 * @match: receives the extracted flow match information.
1562 * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1563 * sequence. The fields should of the packet that triggered the creation
1565 * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1566 * attribute specifies the mask field of the wildcarded flow.
1568 int ovs_match_from_nlattrs(struct sw_flow_match *match,
1569 const struct nlattr *key,
1570 const struct nlattr *mask)
1572 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1573 const struct nlattr *encap;
1576 bool encap_valid = false;
1579 err = parse_flow_nlattrs(key, a, &key_attrs);
1583 if (key_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
1584 encap = a[OVS_KEY_ATTR_ENCAP];
1585 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1586 if (nla_len(encap)) {
1587 __be16 eth_type = 0; /* ETH_P_8021Q */
1589 if (a[OVS_KEY_ATTR_ETHERTYPE])
1590 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1592 if ((eth_type == htons(ETH_P_8021Q)) && (a[OVS_KEY_ATTR_VLAN])) {
1594 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1595 err = parse_flow_nlattrs(encap, a, &key_attrs);
1597 OVS_NLERR("Encap attribute is set for a non-VLAN frame.\n");
1606 err = ovs_key_from_nlattrs(match, key_attrs, a, false);
1611 err = parse_flow_mask_nlattrs(mask, a, &mask_attrs);
1615 if ((mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) && encap_valid) {
1616 __be16 eth_type = 0;
1618 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1619 if (a[OVS_KEY_ATTR_ETHERTYPE])
1620 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1621 if (eth_type == htons(0xffff)) {
1622 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1623 encap = a[OVS_KEY_ATTR_ENCAP];
1624 err = parse_flow_mask_nlattrs(encap, a, &mask_attrs);
1626 OVS_NLERR("VLAN frames must have an exact match"
1627 " on the TPID (mask=%x).\n",
1636 err = ovs_key_from_nlattrs(match, mask_attrs, a, true);
1640 /* Populate exact match flow's key mask. */
1642 ovs_sw_flow_mask_set(match->mask, &match->range, 0xff);
1645 if (!ovs_match_validate(match, key_attrs, mask_attrs))
1652 * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1653 * @flow: Receives extracted in_port, priority, tun_key and skb_mark.
1654 * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1657 * This parses a series of Netlink attributes that form a flow key, which must
1658 * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1659 * get the metadata, that is, the parts of the flow key that cannot be
1660 * extracted from the packet itself.
1663 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow,
1664 const struct nlattr *attr)
1666 struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1667 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1670 struct sw_flow_match match;
1672 flow->key.phy.in_port = DP_MAX_PORTS;
1673 flow->key.phy.priority = 0;
1674 flow->key.phy.skb_mark = 0;
1675 memset(tun_key, 0, sizeof(flow->key.tun_key));
1677 err = parse_flow_nlattrs(attr, a, &attrs);
1681 memset(&match, 0, sizeof(match));
1682 match.key = &flow->key;
1684 err = metadata_from_nlattrs(&match, &attrs, a, false);
1691 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey,
1692 const struct sw_flow_key *output, struct sk_buff *skb)
1694 struct ovs_key_ethernet *eth_key;
1695 struct nlattr *nla, *encap;
1697 if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1698 goto nla_put_failure;
1700 if (swkey->tun_key.ipv4_dst &&
1701 ipv4_tun_to_nlattr(skb, &swkey->tun_key, &output->tun_key))
1702 goto nla_put_failure;
1704 if (swkey->phy.in_port == DP_MAX_PORTS) {
1705 if ((swkey != output) && (output->phy.in_port == 0xffff))
1706 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1707 goto nla_put_failure;
1710 upper_u16 = (swkey == output) ? 0 : 0xffff;
1712 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1713 (upper_u16 << 16) | output->phy.in_port))
1714 goto nla_put_failure;
1717 if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1718 goto nla_put_failure;
1720 nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1722 goto nla_put_failure;
1724 eth_key = nla_data(nla);
1725 memcpy(eth_key->eth_src, output->eth.src, ETH_ALEN);
1726 memcpy(eth_key->eth_dst, output->eth.dst, ETH_ALEN);
1728 if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1730 eth_type = (swkey == output) ? htons(ETH_P_8021Q) : htons(0xffff) ;
1731 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1732 nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1733 goto nla_put_failure;
1734 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1735 if (!swkey->eth.tci)
1740 if (swkey->eth.type == htons(ETH_P_802_2)) {
1742 * Ethertype 802.2 is represented in the netlink with omitted
1743 * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1744 * 0xffff in the mask attribute. Ethertype can also
1747 if (swkey != output && output->eth.type)
1748 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1750 goto nla_put_failure;
1754 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1755 goto nla_put_failure;
1757 if (swkey->eth.type == htons(ETH_P_IP)) {
1758 struct ovs_key_ipv4 *ipv4_key;
1760 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1762 goto nla_put_failure;
1763 ipv4_key = nla_data(nla);
1764 ipv4_key->ipv4_src = output->ipv4.addr.src;
1765 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1766 ipv4_key->ipv4_proto = output->ip.proto;
1767 ipv4_key->ipv4_tos = output->ip.tos;
1768 ipv4_key->ipv4_ttl = output->ip.ttl;
1769 ipv4_key->ipv4_frag = output->ip.frag;
1770 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1771 struct ovs_key_ipv6 *ipv6_key;
1773 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1775 goto nla_put_failure;
1776 ipv6_key = nla_data(nla);
1777 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1778 sizeof(ipv6_key->ipv6_src));
1779 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1780 sizeof(ipv6_key->ipv6_dst));
1781 ipv6_key->ipv6_label = output->ipv6.label;
1782 ipv6_key->ipv6_proto = output->ip.proto;
1783 ipv6_key->ipv6_tclass = output->ip.tos;
1784 ipv6_key->ipv6_hlimit = output->ip.ttl;
1785 ipv6_key->ipv6_frag = output->ip.frag;
1786 } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1787 swkey->eth.type == htons(ETH_P_RARP)) {
1788 struct ovs_key_arp *arp_key;
1790 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1792 goto nla_put_failure;
1793 arp_key = nla_data(nla);
1794 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1795 arp_key->arp_sip = output->ipv4.addr.src;
1796 arp_key->arp_tip = output->ipv4.addr.dst;
1797 arp_key->arp_op = htons(output->ip.proto);
1798 memcpy(arp_key->arp_sha, output->ipv4.arp.sha, ETH_ALEN);
1799 memcpy(arp_key->arp_tha, output->ipv4.arp.tha, ETH_ALEN);
1802 if ((swkey->eth.type == htons(ETH_P_IP) ||
1803 swkey->eth.type == htons(ETH_P_IPV6)) &&
1804 swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1806 if (swkey->ip.proto == IPPROTO_TCP) {
1807 struct ovs_key_tcp *tcp_key;
1809 nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1811 goto nla_put_failure;
1812 tcp_key = nla_data(nla);
1813 if (swkey->eth.type == htons(ETH_P_IP)) {
1814 tcp_key->tcp_src = output->ipv4.tp.src;
1815 tcp_key->tcp_dst = output->ipv4.tp.dst;
1816 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1817 tcp_key->tcp_src = output->ipv6.tp.src;
1818 tcp_key->tcp_dst = output->ipv6.tp.dst;
1820 } else if (swkey->ip.proto == IPPROTO_UDP) {
1821 struct ovs_key_udp *udp_key;
1823 nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1825 goto nla_put_failure;
1826 udp_key = nla_data(nla);
1827 if (swkey->eth.type == htons(ETH_P_IP)) {
1828 udp_key->udp_src = output->ipv4.tp.src;
1829 udp_key->udp_dst = output->ipv4.tp.dst;
1830 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1831 udp_key->udp_src = output->ipv6.tp.src;
1832 udp_key->udp_dst = output->ipv6.tp.dst;
1834 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1835 swkey->ip.proto == IPPROTO_ICMP) {
1836 struct ovs_key_icmp *icmp_key;
1838 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1840 goto nla_put_failure;
1841 icmp_key = nla_data(nla);
1842 icmp_key->icmp_type = ntohs(output->ipv4.tp.src);
1843 icmp_key->icmp_code = ntohs(output->ipv4.tp.dst);
1844 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1845 swkey->ip.proto == IPPROTO_ICMPV6) {
1846 struct ovs_key_icmpv6 *icmpv6_key;
1848 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1849 sizeof(*icmpv6_key));
1851 goto nla_put_failure;
1852 icmpv6_key = nla_data(nla);
1853 icmpv6_key->icmpv6_type = ntohs(output->ipv6.tp.src);
1854 icmpv6_key->icmpv6_code = ntohs(output->ipv6.tp.dst);
1856 if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1857 icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1858 struct ovs_key_nd *nd_key;
1860 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1862 goto nla_put_failure;
1863 nd_key = nla_data(nla);
1864 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1865 sizeof(nd_key->nd_target));
1866 memcpy(nd_key->nd_sll, output->ipv6.nd.sll, ETH_ALEN);
1867 memcpy(nd_key->nd_tll, output->ipv6.nd.tll, ETH_ALEN);
1874 nla_nest_end(skb, encap);
1882 /* Initializes the flow module.
1883 * Returns zero if successful or a negative error code. */
1884 int ovs_flow_init(void)
1886 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1888 if (flow_cache == NULL)
1894 /* Uninitializes the flow module. */
1895 void ovs_flow_exit(void)
1897 kmem_cache_destroy(flow_cache);
1900 struct sw_flow_mask *ovs_sw_flow_mask_alloc(void)
1902 struct sw_flow_mask *mask;
1904 mask = kmalloc(sizeof(*mask), GFP_KERNEL);
1906 mask->ref_count = 0;
1911 void ovs_sw_flow_mask_add_ref(struct sw_flow_mask *mask)
1916 static void rcu_free_sw_flow_mask_cb(struct rcu_head *rcu)
1918 struct sw_flow_mask *mask = container_of(rcu, struct sw_flow_mask, rcu);
1923 void ovs_sw_flow_mask_del_ref(struct sw_flow_mask *mask, bool deferred)
1928 BUG_ON(!mask->ref_count);
1931 if (!mask->ref_count) {
1932 list_del_rcu(&mask->list);
1934 call_rcu(&mask->rcu, rcu_free_sw_flow_mask_cb);
1940 static bool ovs_sw_flow_mask_equal(const struct sw_flow_mask *a,
1941 const struct sw_flow_mask *b)
1943 u8 *a_ = (u8 *)&a->key + a->range.start;
1944 u8 *b_ = (u8 *)&b->key + b->range.start;
1946 return (a->range.end == b->range.end)
1947 && (a->range.start == b->range.start)
1948 && (memcmp(a_, b_, ovs_sw_flow_mask_actual_size(a)) == 0);
1951 struct sw_flow_mask *ovs_sw_flow_mask_find(const struct flow_table *tbl,
1952 const struct sw_flow_mask *mask)
1954 struct list_head *ml;
1956 list_for_each(ml, tbl->mask_list) {
1957 struct sw_flow_mask *m;
1958 m = container_of(ml, struct sw_flow_mask, list);
1959 if (ovs_sw_flow_mask_equal(mask, m))
1967 * add a new mask into the mask list.
1968 * The caller needs to make sure that 'mask' is not the same
1969 * as any masks that are already on the list.
1971 void ovs_sw_flow_mask_insert(struct flow_table *tbl, struct sw_flow_mask *mask)
1973 list_add_rcu(&mask->list, tbl->mask_list);
1977 * Set 'range' fields in the mask to the value of 'val'.
1979 static void ovs_sw_flow_mask_set(struct sw_flow_mask *mask,
1980 struct sw_flow_key_range *range, u8 val)
1982 u8 *m = (u8 *)&mask->key + range->start;
1984 mask->range = *range;
1985 memset(m, val, ovs_sw_flow_mask_size_roundup(mask));