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 /* Tunnel mask is always allowed. */
139 mask_allowed |= (1ULL << OVS_KEY_ATTR_TUNNEL);
141 if (match->key->phy.in_port == DP_MAX_PORTS &&
142 match->mask && (match->mask->key.phy.in_port == 0xffff))
143 mask_allowed |= (1ULL << OVS_KEY_ATTR_IN_PORT);
145 if (match->key->eth.type == htons(ETH_P_802_2) &&
146 match->mask && (match->mask->key.eth.type == htons(0xffff)))
147 mask_allowed |= (1ULL << 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 |= 1ULL << OVS_KEY_ATTR_ARP;
153 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
154 mask_allowed |= 1ULL << OVS_KEY_ATTR_ARP;
157 if (match->key->eth.type == htons(ETH_P_IP)) {
158 key_expected |= 1ULL << OVS_KEY_ATTR_IPV4;
159 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
160 mask_allowed |= 1ULL << 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 |= 1ULL << OVS_KEY_ATTR_UDP;
165 if (match->mask && (match->mask->key.ip.proto == 0xff))
166 mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
169 if (match->key->ip.proto == IPPROTO_TCP) {
170 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
171 if (match->mask && (match->mask->key.ip.proto == 0xff))
172 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
175 if (match->key->ip.proto == IPPROTO_ICMP) {
176 key_expected |= 1ULL << OVS_KEY_ATTR_ICMP;
177 if (match->mask && (match->mask->key.ip.proto == 0xff))
178 mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMP;
183 if (match->key->eth.type == htons(ETH_P_IPV6)) {
184 key_expected |= 1ULL << OVS_KEY_ATTR_IPV6;
185 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
186 mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV6;
188 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
189 if (match->key->ip.proto == IPPROTO_UDP) {
190 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
191 if (match->mask && (match->mask->key.ip.proto == 0xff))
192 mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
195 if (match->key->ip.proto == IPPROTO_TCP) {
196 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
197 if (match->mask && (match->mask->key.ip.proto == 0xff))
198 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
201 if (match->key->ip.proto == IPPROTO_ICMPV6) {
202 key_expected |= 1ULL << OVS_KEY_ATTR_ICMPV6;
203 if (match->mask && (match->mask->key.ip.proto == 0xff))
204 mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMPV6;
206 if (match->key->ipv6.tp.src ==
207 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
208 match->key->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
209 key_expected |= 1ULL << OVS_KEY_ATTR_ND;
210 if (match->mask && (match->mask->key.ipv6.tp.src == htons(0xffff)))
211 mask_allowed |= 1ULL << OVS_KEY_ATTR_ND;
217 if ((key_attrs & key_expected) != key_expected) {
218 /* Key attributes check failed. */
219 OVS_NLERR("Missing expected key attributes (key_attrs=%llx, expected=%llx).\n",
220 key_attrs, key_expected);
224 if ((mask_attrs & mask_allowed) != mask_attrs) {
225 /* Mask attributes check failed. */
226 OVS_NLERR("Contain more than allowed mask fields (mask_attrs=%llx, mask_allowed=%llx).\n",
227 mask_attrs, mask_allowed);
234 static int check_header(struct sk_buff *skb, int len)
236 if (unlikely(skb->len < len))
238 if (unlikely(!pskb_may_pull(skb, len)))
243 static bool arphdr_ok(struct sk_buff *skb)
245 return pskb_may_pull(skb, skb_network_offset(skb) +
246 sizeof(struct arp_eth_header));
249 static int check_iphdr(struct sk_buff *skb)
251 unsigned int nh_ofs = skb_network_offset(skb);
255 err = check_header(skb, nh_ofs + sizeof(struct iphdr));
259 ip_len = ip_hdrlen(skb);
260 if (unlikely(ip_len < sizeof(struct iphdr) ||
261 skb->len < nh_ofs + ip_len))
264 skb_set_transport_header(skb, nh_ofs + ip_len);
268 static bool tcphdr_ok(struct sk_buff *skb)
270 int th_ofs = skb_transport_offset(skb);
273 if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
276 tcp_len = tcp_hdrlen(skb);
277 if (unlikely(tcp_len < sizeof(struct tcphdr) ||
278 skb->len < th_ofs + tcp_len))
284 static bool udphdr_ok(struct sk_buff *skb)
286 return pskb_may_pull(skb, skb_transport_offset(skb) +
287 sizeof(struct udphdr));
290 static bool icmphdr_ok(struct sk_buff *skb)
292 return pskb_may_pull(skb, skb_transport_offset(skb) +
293 sizeof(struct icmphdr));
296 u64 ovs_flow_used_time(unsigned long flow_jiffies)
298 struct timespec cur_ts;
301 ktime_get_ts(&cur_ts);
302 idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
303 cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
304 cur_ts.tv_nsec / NSEC_PER_MSEC;
306 return cur_ms - idle_ms;
309 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
311 unsigned int nh_ofs = skb_network_offset(skb);
319 err = check_header(skb, nh_ofs + sizeof(*nh));
324 nexthdr = nh->nexthdr;
325 payload_ofs = (u8 *)(nh + 1) - skb->data;
327 key->ip.proto = NEXTHDR_NONE;
328 key->ip.tos = ipv6_get_dsfield(nh);
329 key->ip.ttl = nh->hop_limit;
330 key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
331 key->ipv6.addr.src = nh->saddr;
332 key->ipv6.addr.dst = nh->daddr;
334 payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
335 if (unlikely(payload_ofs < 0))
339 if (frag_off & htons(~0x7))
340 key->ip.frag = OVS_FRAG_TYPE_LATER;
342 key->ip.frag = OVS_FRAG_TYPE_FIRST;
345 nh_len = payload_ofs - nh_ofs;
346 skb_set_transport_header(skb, nh_ofs + nh_len);
347 key->ip.proto = nexthdr;
351 static bool icmp6hdr_ok(struct sk_buff *skb)
353 return pskb_may_pull(skb, skb_transport_offset(skb) +
354 sizeof(struct icmp6hdr));
357 void ovs_flow_key_mask(struct sw_flow_key *dst, const struct sw_flow_key *src,
358 const struct sw_flow_mask *mask)
360 u8 *m = (u8 *)&mask->key + mask->range.start;
361 u8 *s = (u8 *)src + mask->range.start;
362 u8 *d = (u8 *)dst + mask->range.start;
365 memset(dst, 0, sizeof(*dst));
366 for (i = 0; i < ovs_sw_flow_mask_size_roundup(mask); i++) {
372 #define TCP_FLAGS_OFFSET 13
373 #define TCP_FLAG_MASK 0x3f
375 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
379 if ((flow->key.eth.type == htons(ETH_P_IP) ||
380 flow->key.eth.type == htons(ETH_P_IPV6)) &&
381 flow->key.ip.proto == IPPROTO_TCP &&
382 likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
383 u8 *tcp = (u8 *)tcp_hdr(skb);
384 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
387 spin_lock(&flow->lock);
388 flow->used = jiffies;
389 flow->packet_count++;
390 flow->byte_count += skb->len;
391 flow->tcp_flags |= tcp_flags;
392 spin_unlock(&flow->lock);
395 struct sw_flow_actions *ovs_flow_actions_alloc(int size)
397 struct sw_flow_actions *sfa;
399 if (size > MAX_ACTIONS_BUFSIZE)
400 return ERR_PTR(-EINVAL);
402 sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
404 return ERR_PTR(-ENOMEM);
406 sfa->actions_len = 0;
410 struct sw_flow *ovs_flow_alloc(void)
412 struct sw_flow *flow;
414 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
416 return ERR_PTR(-ENOMEM);
418 spin_lock_init(&flow->lock);
419 flow->sf_acts = NULL;
425 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
427 hash = jhash_1word(hash, table->hash_seed);
428 return flex_array_get(table->buckets,
429 (hash & (table->n_buckets - 1)));
432 static struct flex_array *alloc_buckets(unsigned int n_buckets)
434 struct flex_array *buckets;
437 buckets = flex_array_alloc(sizeof(struct hlist_head),
438 n_buckets, GFP_KERNEL);
442 err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
444 flex_array_free(buckets);
448 for (i = 0; i < n_buckets; i++)
449 INIT_HLIST_HEAD((struct hlist_head *)
450 flex_array_get(buckets, i));
455 static void free_buckets(struct flex_array *buckets)
457 flex_array_free(buckets);
460 static struct flow_table *__flow_tbl_alloc(int new_size)
462 struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
467 table->buckets = alloc_buckets(new_size);
469 if (!table->buckets) {
473 table->n_buckets = new_size;
476 table->keep_flows = false;
477 get_random_bytes(&table->hash_seed, sizeof(u32));
478 table->mask_list = NULL;
483 static void __flow_tbl_destroy(struct flow_table *table)
487 if (table->keep_flows)
490 for (i = 0; i < table->n_buckets; i++) {
491 struct sw_flow *flow;
492 struct hlist_head *head = flex_array_get(table->buckets, i);
493 struct hlist_node *n;
494 int ver = table->node_ver;
496 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
497 hlist_del(&flow->hash_node[ver]);
498 ovs_flow_free(flow, false);
502 BUG_ON(!list_empty(table->mask_list));
503 kfree(table->mask_list);
506 free_buckets(table->buckets);
510 struct flow_table *ovs_flow_tbl_alloc(int new_size)
512 struct flow_table *table = __flow_tbl_alloc(new_size);
517 table->mask_list = kmalloc(sizeof(struct list_head), GFP_KERNEL);
518 if (!table->mask_list) {
519 table->keep_flows = true;
520 __flow_tbl_destroy(table);
523 INIT_LIST_HEAD(table->mask_list);
528 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
530 struct flow_table *table = container_of(rcu, struct flow_table, rcu);
532 __flow_tbl_destroy(table);
535 void ovs_flow_tbl_destroy(struct flow_table *table, bool deferred)
541 call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
543 __flow_tbl_destroy(table);
546 struct sw_flow *ovs_flow_dump_next(struct flow_table *table, u32 *bucket, u32 *last)
548 struct sw_flow *flow;
549 struct hlist_head *head;
553 ver = table->node_ver;
554 while (*bucket < table->n_buckets) {
556 head = flex_array_get(table->buckets, *bucket);
557 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
572 static void __tbl_insert(struct flow_table *table, struct sw_flow *flow)
574 struct hlist_head *head;
576 head = find_bucket(table, flow->hash);
577 hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
582 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
587 old_ver = old->node_ver;
588 new->node_ver = !old_ver;
590 /* Insert in new table. */
591 for (i = 0; i < old->n_buckets; i++) {
592 struct sw_flow *flow;
593 struct hlist_head *head;
595 head = flex_array_get(old->buckets, i);
597 hlist_for_each_entry(flow, head, hash_node[old_ver])
598 __tbl_insert(new, flow);
601 new->mask_list = old->mask_list;
602 old->keep_flows = true;
605 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
607 struct flow_table *new_table;
609 new_table = __flow_tbl_alloc(n_buckets);
611 return ERR_PTR(-ENOMEM);
613 flow_table_copy_flows(table, new_table);
618 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
620 return __flow_tbl_rehash(table, table->n_buckets);
623 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
625 return __flow_tbl_rehash(table, table->n_buckets * 2);
628 static void __flow_free(struct sw_flow *flow)
630 kfree((struct sf_flow_acts __force *)flow->sf_acts);
631 kmem_cache_free(flow_cache, flow);
634 static void rcu_free_flow_callback(struct rcu_head *rcu)
636 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
641 void ovs_flow_free(struct sw_flow *flow, bool deferred)
646 ovs_sw_flow_mask_del_ref(flow->mask, deferred);
649 call_rcu(&flow->rcu, rcu_free_flow_callback);
654 /* RCU callback used by ovs_flow_deferred_free_acts. */
655 static void rcu_free_acts_callback(struct rcu_head *rcu)
657 struct sw_flow_actions *sf_acts = container_of(rcu,
658 struct sw_flow_actions, rcu);
662 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
663 * The caller must hold rcu_read_lock for this to be sensible. */
664 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
666 call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
669 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
672 __be16 eth_type; /* ETH_P_8021Q */
675 struct qtag_prefix *qp;
677 if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
680 if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
684 qp = (struct qtag_prefix *) skb->data;
685 key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
686 __skb_pull(skb, sizeof(struct qtag_prefix));
691 static __be16 parse_ethertype(struct sk_buff *skb)
693 struct llc_snap_hdr {
694 u8 dsap; /* Always 0xAA */
695 u8 ssap; /* Always 0xAA */
700 struct llc_snap_hdr *llc;
703 proto = *(__be16 *) skb->data;
704 __skb_pull(skb, sizeof(__be16));
706 if (ntohs(proto) >= ETH_P_802_3_MIN)
709 if (skb->len < sizeof(struct llc_snap_hdr))
710 return htons(ETH_P_802_2);
712 if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
715 llc = (struct llc_snap_hdr *) skb->data;
716 if (llc->dsap != LLC_SAP_SNAP ||
717 llc->ssap != LLC_SAP_SNAP ||
718 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
719 return htons(ETH_P_802_2);
721 __skb_pull(skb, sizeof(struct llc_snap_hdr));
723 if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
724 return llc->ethertype;
726 return htons(ETH_P_802_2);
729 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
732 struct icmp6hdr *icmp = icmp6_hdr(skb);
734 /* The ICMPv6 type and code fields use the 16-bit transport port
735 * fields, so we need to store them in 16-bit network byte order.
737 key->ipv6.tp.src = htons(icmp->icmp6_type);
738 key->ipv6.tp.dst = htons(icmp->icmp6_code);
740 if (icmp->icmp6_code == 0 &&
741 (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
742 icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
743 int icmp_len = skb->len - skb_transport_offset(skb);
747 /* In order to process neighbor discovery options, we need the
750 if (unlikely(icmp_len < sizeof(*nd)))
753 if (unlikely(skb_linearize(skb)))
756 nd = (struct nd_msg *)skb_transport_header(skb);
757 key->ipv6.nd.target = nd->target;
759 icmp_len -= sizeof(*nd);
761 while (icmp_len >= 8) {
762 struct nd_opt_hdr *nd_opt =
763 (struct nd_opt_hdr *)(nd->opt + offset);
764 int opt_len = nd_opt->nd_opt_len * 8;
766 if (unlikely(!opt_len || opt_len > icmp_len))
769 /* Store the link layer address if the appropriate
770 * option is provided. It is considered an error if
771 * the same link layer option is specified twice.
773 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
775 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
777 memcpy(key->ipv6.nd.sll,
778 &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
779 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
781 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
783 memcpy(key->ipv6.nd.tll,
784 &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
795 memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
796 memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
797 memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
803 * ovs_flow_extract - extracts a flow key from an Ethernet frame.
804 * @skb: sk_buff that contains the frame, with skb->data pointing to the
806 * @in_port: port number on which @skb was received.
807 * @key: output flow key
808 * @key_lenp: length of output flow key
810 * The caller must ensure that skb->len >= ETH_HLEN.
812 * Returns 0 if successful, otherwise a negative errno value.
814 * Initializes @skb header pointers as follows:
816 * - skb->mac_header: the Ethernet header.
818 * - skb->network_header: just past the Ethernet header, or just past the
819 * VLAN header, to the first byte of the Ethernet payload.
821 * - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
822 * on output, then just past the IP header, if one is present and
823 * of a correct length, otherwise the same as skb->network_header.
824 * For other key->eth.type values it is left untouched.
826 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
831 memset(key, 0, sizeof(*key));
833 key->phy.priority = skb->priority;
834 if (OVS_CB(skb)->tun_key)
835 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
836 key->phy.in_port = in_port;
837 key->phy.skb_mark = skb_get_mark(skb);
839 skb_reset_mac_header(skb);
841 /* Link layer. We are guaranteed to have at least the 14 byte Ethernet
842 * header in the linear data area.
845 memcpy(key->eth.src, eth->h_source, ETH_ALEN);
846 memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
848 __skb_pull(skb, 2 * ETH_ALEN);
849 /* We are going to push all headers that we pull, so no need to
850 * update skb->csum here. */
852 if (vlan_tx_tag_present(skb))
853 key->eth.tci = htons(vlan_get_tci(skb));
854 else if (eth->h_proto == htons(ETH_P_8021Q))
855 if (unlikely(parse_vlan(skb, key)))
858 key->eth.type = parse_ethertype(skb);
859 if (unlikely(key->eth.type == htons(0)))
862 skb_reset_network_header(skb);
863 __skb_push(skb, skb->data - skb_mac_header(skb));
866 if (key->eth.type == htons(ETH_P_IP)) {
870 error = check_iphdr(skb);
871 if (unlikely(error)) {
872 if (error == -EINVAL) {
873 skb->transport_header = skb->network_header;
880 key->ipv4.addr.src = nh->saddr;
881 key->ipv4.addr.dst = nh->daddr;
883 key->ip.proto = nh->protocol;
884 key->ip.tos = nh->tos;
885 key->ip.ttl = nh->ttl;
887 offset = nh->frag_off & htons(IP_OFFSET);
889 key->ip.frag = OVS_FRAG_TYPE_LATER;
892 if (nh->frag_off & htons(IP_MF) ||
893 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
894 key->ip.frag = OVS_FRAG_TYPE_FIRST;
896 /* Transport layer. */
897 if (key->ip.proto == IPPROTO_TCP) {
898 if (tcphdr_ok(skb)) {
899 struct tcphdr *tcp = tcp_hdr(skb);
900 key->ipv4.tp.src = tcp->source;
901 key->ipv4.tp.dst = tcp->dest;
903 } else if (key->ip.proto == IPPROTO_UDP) {
904 if (udphdr_ok(skb)) {
905 struct udphdr *udp = udp_hdr(skb);
906 key->ipv4.tp.src = udp->source;
907 key->ipv4.tp.dst = udp->dest;
909 } else if (key->ip.proto == IPPROTO_ICMP) {
910 if (icmphdr_ok(skb)) {
911 struct icmphdr *icmp = icmp_hdr(skb);
912 /* The ICMP type and code fields use the 16-bit
913 * transport port fields, so we need to store
914 * them in 16-bit network byte order. */
915 key->ipv4.tp.src = htons(icmp->type);
916 key->ipv4.tp.dst = htons(icmp->code);
920 } else if ((key->eth.type == htons(ETH_P_ARP) ||
921 key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
922 struct arp_eth_header *arp;
924 arp = (struct arp_eth_header *)skb_network_header(skb);
926 if (arp->ar_hrd == htons(ARPHRD_ETHER)
927 && arp->ar_pro == htons(ETH_P_IP)
928 && arp->ar_hln == ETH_ALEN
929 && arp->ar_pln == 4) {
931 /* We only match on the lower 8 bits of the opcode. */
932 if (ntohs(arp->ar_op) <= 0xff)
933 key->ip.proto = ntohs(arp->ar_op);
934 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
935 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
936 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
937 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
939 } else if (key->eth.type == htons(ETH_P_IPV6)) {
940 int nh_len; /* IPv6 Header + Extensions */
942 nh_len = parse_ipv6hdr(skb, key);
943 if (unlikely(nh_len < 0)) {
944 if (nh_len == -EINVAL) {
945 skb->transport_header = skb->network_header;
953 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
955 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
956 key->ip.frag = OVS_FRAG_TYPE_FIRST;
958 /* Transport layer. */
959 if (key->ip.proto == NEXTHDR_TCP) {
960 if (tcphdr_ok(skb)) {
961 struct tcphdr *tcp = tcp_hdr(skb);
962 key->ipv6.tp.src = tcp->source;
963 key->ipv6.tp.dst = tcp->dest;
965 } else if (key->ip.proto == NEXTHDR_UDP) {
966 if (udphdr_ok(skb)) {
967 struct udphdr *udp = udp_hdr(skb);
968 key->ipv6.tp.src = udp->source;
969 key->ipv6.tp.dst = udp->dest;
971 } else if (key->ip.proto == NEXTHDR_ICMP) {
972 if (icmp6hdr_ok(skb)) {
973 error = parse_icmpv6(skb, key, nh_len);
983 static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start, int key_len)
985 return jhash2((u32 *)((u8 *)key + key_start),
986 DIV_ROUND_UP(key_len - key_start, sizeof(u32)), 0);
989 static int flow_key_start(const struct sw_flow_key *key)
991 if (key->tun_key.ipv4_dst)
994 return offsetof(struct sw_flow_key, phy);
997 static bool __cmp_key(const struct sw_flow_key *key1,
998 const struct sw_flow_key *key2, int key_start, int key_len)
1000 return !memcmp((u8 *)key1 + key_start,
1001 (u8 *)key2 + key_start, (key_len - key_start));
1004 static bool __flow_cmp_key(const struct sw_flow *flow,
1005 const struct sw_flow_key *key, int key_start, int key_len)
1007 return __cmp_key(&flow->key, key, key_start, key_len);
1010 static bool __flow_cmp_unmasked_key(const struct sw_flow *flow,
1011 const struct sw_flow_key *key, int key_start, int key_len)
1013 return __cmp_key(&flow->unmasked_key, key, key_start, key_len);
1016 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
1017 const struct sw_flow_key *key, int key_len)
1020 key_start = flow_key_start(key);
1022 return __flow_cmp_unmasked_key(flow, key, key_start, key_len);
1026 struct sw_flow *ovs_flow_lookup_unmasked_key(struct flow_table *table,
1027 struct sw_flow_match *match)
1029 struct sw_flow_key *unmasked = match->key;
1030 int key_len = match->range.end;
1031 struct sw_flow *flow;
1033 flow = ovs_flow_lookup(table, unmasked);
1034 if (flow && (!ovs_flow_cmp_unmasked_key(flow, unmasked, key_len)))
1040 static struct sw_flow *ovs_masked_flow_lookup(struct flow_table *table,
1041 const struct sw_flow_key *flow_key,
1042 struct sw_flow_mask *mask)
1044 struct sw_flow *flow;
1045 struct hlist_head *head;
1046 int key_start = mask->range.start;
1047 int key_len = mask->range.end;
1049 struct sw_flow_key masked_key;
1051 ovs_flow_key_mask(&masked_key, flow_key, mask);
1052 hash = ovs_flow_hash(&masked_key, key_start, key_len);
1053 head = find_bucket(table, hash);
1054 hlist_for_each_entry_rcu(flow, head, hash_node[table->node_ver]) {
1055 if (flow->mask == mask &&
1056 __flow_cmp_key(flow, &masked_key, key_start, key_len))
1062 struct sw_flow *ovs_flow_lookup(struct flow_table *tbl,
1063 const struct sw_flow_key *key)
1065 struct sw_flow *flow = NULL;
1066 struct sw_flow_mask *mask;
1068 list_for_each_entry_rcu(mask, tbl->mask_list, list) {
1069 flow = ovs_masked_flow_lookup(tbl, key, mask);
1070 if (flow) /* Found */
1078 void ovs_flow_insert(struct flow_table *table, struct sw_flow *flow)
1080 flow->hash = ovs_flow_hash(&flow->key, flow->mask->range.start,
1081 flow->mask->range.end);
1082 __tbl_insert(table, flow);
1085 void ovs_flow_remove(struct flow_table *table, struct sw_flow *flow)
1087 BUG_ON(table->count == 0);
1088 hlist_del_rcu(&flow->hash_node[table->node_ver]);
1092 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute. */
1093 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
1094 [OVS_KEY_ATTR_ENCAP] = -1,
1095 [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
1096 [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
1097 [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
1098 [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
1099 [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
1100 [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
1101 [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
1102 [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
1103 [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
1104 [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
1105 [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
1106 [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
1107 [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
1108 [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
1109 [OVS_KEY_ATTR_TUNNEL] = -1,
1112 static bool is_all_zero(const u8 *fp, size_t size)
1119 for (i = 0; i < size; i++)
1126 static int __parse_flow_nlattrs(const struct nlattr *attr,
1127 const struct nlattr *a[],
1128 u64 *attrsp, bool nz)
1130 const struct nlattr *nla;
1135 nla_for_each_nested(nla, attr, rem) {
1136 u16 type = nla_type(nla);
1139 if (type > OVS_KEY_ATTR_MAX) {
1140 OVS_NLERR("Unknown key attribute (type=%d, max=%d).\n",
1141 type, OVS_KEY_ATTR_MAX);
1144 if (attrs & (1ULL << type)) {
1145 OVS_NLERR("Duplicate key attribute (type %d).\n", type);
1149 expected_len = ovs_key_lens[type];
1150 if (nla_len(nla) != expected_len && expected_len != -1) {
1151 OVS_NLERR("Key attribute has unexpected length (type=%d"
1152 ", length=%d, expected=%d).\n", type,
1153 nla_len(nla), expected_len);
1157 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
1158 attrs |= 1ULL << type;
1163 OVS_NLERR("Message has %d unknown bytes.\n", rem);
1171 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
1172 const struct nlattr *a[], u64 *attrsp)
1174 return __parse_flow_nlattrs(attr, a, attrsp, true);
1177 static int parse_flow_nlattrs(const struct nlattr *attr,
1178 const struct nlattr *a[], u64 *attrsp)
1180 return __parse_flow_nlattrs(attr, a, attrsp, false);
1183 int ipv4_tun_from_nlattr(const struct nlattr *attr,
1184 struct sw_flow_match *match, bool is_mask)
1189 __be16 tun_flags = 0;
1191 nla_for_each_nested(a, attr, rem) {
1192 int type = nla_type(a);
1193 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
1194 [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
1195 [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
1196 [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
1197 [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
1198 [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
1199 [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
1200 [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
1203 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
1204 OVS_NLERR("Unknown IPv4 tunnel attribute (type=%d, max=%d).\n",
1205 type, OVS_TUNNEL_KEY_ATTR_MAX);
1209 if (ovs_tunnel_key_lens[type] != nla_len(a)) {
1210 OVS_NLERR("IPv4 tunnel attribute type has unexpected "
1211 " legnth (type=%d, length=%d, expected=%d).\n",
1212 type, nla_len(a), ovs_tunnel_key_lens[type]);
1217 case OVS_TUNNEL_KEY_ATTR_ID:
1218 SW_FLOW_KEY_PUT(match, tun_key.tun_id,
1219 nla_get_be64(a), is_mask);
1220 tun_flags |= TUNNEL_KEY;
1222 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1223 SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
1224 nla_get_be32(a), is_mask);
1226 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1227 SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
1228 nla_get_be32(a), is_mask);
1230 case OVS_TUNNEL_KEY_ATTR_TOS:
1231 SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
1232 nla_get_u8(a), is_mask);
1234 case OVS_TUNNEL_KEY_ATTR_TTL:
1235 SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
1236 nla_get_u8(a), is_mask);
1239 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1240 tun_flags |= TUNNEL_DONT_FRAGMENT;
1242 case OVS_TUNNEL_KEY_ATTR_CSUM:
1243 tun_flags |= TUNNEL_CSUM;
1250 SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
1253 OVS_NLERR("IPv4 tunnel attribute has %d unknown bytes.\n", rem);
1257 if (!match->key->tun_key.ipv4_dst) {
1258 OVS_NLERR("IPv4 tunnel destination address is zero.\n");
1263 OVS_NLERR("IPv4 tunnel TTL not specified.\n");
1270 int ipv4_tun_to_nlattr(struct sk_buff *skb,
1271 const struct ovs_key_ipv4_tunnel *tun_key,
1272 const struct ovs_key_ipv4_tunnel *output)
1276 nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
1280 if (output->tun_flags & TUNNEL_KEY &&
1281 nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
1283 if (output->ipv4_src &&
1284 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
1286 if (output->ipv4_dst &&
1287 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
1289 if (output->ipv4_tos &&
1290 nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
1292 if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
1294 if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
1295 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
1297 if ((output->tun_flags & TUNNEL_CSUM) &&
1298 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
1301 nla_nest_end(skb, nla);
1306 static int metadata_from_nlattrs(struct sw_flow_match *match, u64 *attrs,
1307 const struct nlattr **a, bool is_mask)
1309 if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
1310 SW_FLOW_KEY_PUT(match, phy.priority,
1311 nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1312 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
1315 if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
1316 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1318 if (!is_mask && in_port >= DP_MAX_PORTS)
1320 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1321 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
1322 } else if (!is_mask) {
1323 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1326 if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
1327 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1328 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1329 if (!is_mask && mark != 0) {
1330 OVS_NLERR("skb->mark must be zero on this kernel (mark=%d).\n", mark);
1334 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1335 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
1337 if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
1338 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1341 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
1346 static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs,
1347 const struct nlattr **a, bool is_mask)
1350 u64 orig_attrs = attrs;
1352 err = metadata_from_nlattrs(match, &attrs, a, is_mask);
1356 if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
1357 const struct ovs_key_ethernet *eth_key;
1359 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1360 SW_FLOW_KEY_MEMCPY(match, eth.src,
1361 eth_key->eth_src, ETH_ALEN, is_mask);
1362 SW_FLOW_KEY_MEMCPY(match, eth.dst,
1363 eth_key->eth_dst, ETH_ALEN, is_mask);
1364 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
1367 if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
1370 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1371 if (!(tci & htons(VLAN_TAG_PRESENT))) {
1373 OVS_NLERR("VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.\n");
1375 OVS_NLERR("VLAN TCI does not have VLAN_TAG_PRESENT bit set.\n");
1380 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
1381 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
1382 } else if (!is_mask)
1383 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1385 if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
1388 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1389 if (!is_mask && ntohs(eth_type) < ETH_P_802_3_MIN) {
1390 OVS_NLERR("EtherType is less than mimimum (type=%x, min=%x).\n",
1391 ntohs(eth_type), ETH_P_802_3_MIN);
1395 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1396 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1397 } else if (!is_mask) {
1398 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1401 if (attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1402 const struct ovs_key_ipv4 *ipv4_key;
1404 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1405 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1406 OVS_NLERR("Unknown IPv4 fragment type (value=%d, max=%d).\n",
1407 ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1410 SW_FLOW_KEY_PUT(match, ip.proto,
1411 ipv4_key->ipv4_proto, is_mask);
1412 SW_FLOW_KEY_PUT(match, ip.tos,
1413 ipv4_key->ipv4_tos, is_mask);
1414 SW_FLOW_KEY_PUT(match, ip.ttl,
1415 ipv4_key->ipv4_ttl, is_mask);
1416 SW_FLOW_KEY_PUT(match, ip.frag,
1417 ipv4_key->ipv4_frag, is_mask);
1418 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1419 ipv4_key->ipv4_src, is_mask);
1420 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1421 ipv4_key->ipv4_dst, is_mask);
1422 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4);
1425 if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
1426 const struct ovs_key_ipv6 *ipv6_key;
1428 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1429 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1430 OVS_NLERR("Unknown IPv6 fragment type (value=%d, max=%d).\n",
1431 ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1434 SW_FLOW_KEY_PUT(match, ipv6.label,
1435 ipv6_key->ipv6_label, is_mask);
1436 SW_FLOW_KEY_PUT(match, ip.proto,
1437 ipv6_key->ipv6_proto, is_mask);
1438 SW_FLOW_KEY_PUT(match, ip.tos,
1439 ipv6_key->ipv6_tclass, is_mask);
1440 SW_FLOW_KEY_PUT(match, ip.ttl,
1441 ipv6_key->ipv6_hlimit, is_mask);
1442 SW_FLOW_KEY_PUT(match, ip.frag,
1443 ipv6_key->ipv6_frag, is_mask);
1444 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1446 sizeof(match->key->ipv6.addr.src),
1448 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1450 sizeof(match->key->ipv6.addr.dst),
1453 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
1456 if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
1457 const struct ovs_key_arp *arp_key;
1459 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1460 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1461 OVS_NLERR("Unknown ARP opcode (opcode=%d).\n",
1466 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1467 arp_key->arp_sip, is_mask);
1468 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1469 arp_key->arp_tip, is_mask);
1470 SW_FLOW_KEY_PUT(match, ip.proto,
1471 ntohs(arp_key->arp_op), is_mask);
1472 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1473 arp_key->arp_sha, ETH_ALEN, is_mask);
1474 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1475 arp_key->arp_tha, ETH_ALEN, is_mask);
1477 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
1480 if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
1481 const struct ovs_key_tcp *tcp_key;
1483 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1484 if (orig_attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1485 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1486 tcp_key->tcp_src, is_mask);
1487 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1488 tcp_key->tcp_dst, is_mask);
1490 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1491 tcp_key->tcp_src, is_mask);
1492 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1493 tcp_key->tcp_dst, is_mask);
1495 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
1498 if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
1499 const struct ovs_key_udp *udp_key;
1501 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1502 if (orig_attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1503 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1504 udp_key->udp_src, is_mask);
1505 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1506 udp_key->udp_dst, is_mask);
1508 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1509 udp_key->udp_src, is_mask);
1510 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1511 udp_key->udp_dst, is_mask);
1513 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
1516 if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
1517 const struct ovs_key_icmp *icmp_key;
1519 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1520 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1521 htons(icmp_key->icmp_type), is_mask);
1522 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1523 htons(icmp_key->icmp_code), is_mask);
1524 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
1527 if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
1528 const struct ovs_key_icmpv6 *icmpv6_key;
1530 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1531 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1532 htons(icmpv6_key->icmpv6_type), is_mask);
1533 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1534 htons(icmpv6_key->icmpv6_code), is_mask);
1535 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
1538 if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
1539 const struct ovs_key_nd *nd_key;
1541 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1542 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1544 sizeof(match->key->ipv6.nd.target),
1546 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1547 nd_key->nd_sll, ETH_ALEN, is_mask);
1548 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1549 nd_key->nd_tll, ETH_ALEN, is_mask);
1550 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
1560 * ovs_match_from_nlattrs - parses Netlink attributes into a flow key and
1561 * mask. In case the 'mask' is NULL, the flow is treated as exact match
1562 * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1563 * does not include any don't care bit.
1564 * @match: receives the extracted flow match information.
1565 * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1566 * sequence. The fields should of the packet that triggered the creation
1568 * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1569 * attribute specifies the mask field of the wildcarded flow.
1571 int ovs_match_from_nlattrs(struct sw_flow_match *match,
1572 const struct nlattr *key,
1573 const struct nlattr *mask)
1575 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1576 const struct nlattr *encap;
1579 bool encap_valid = false;
1582 err = parse_flow_nlattrs(key, a, &key_attrs);
1586 if (key_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
1587 encap = a[OVS_KEY_ATTR_ENCAP];
1588 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1589 if (nla_len(encap)) {
1590 __be16 eth_type = 0; /* ETH_P_8021Q */
1592 if (a[OVS_KEY_ATTR_ETHERTYPE])
1593 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1595 if ((eth_type == htons(ETH_P_8021Q)) && (a[OVS_KEY_ATTR_VLAN])) {
1597 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1598 err = parse_flow_nlattrs(encap, a, &key_attrs);
1600 OVS_NLERR("Encap attribute is set for a non-VLAN frame.\n");
1609 err = ovs_key_from_nlattrs(match, key_attrs, a, false);
1614 err = parse_flow_mask_nlattrs(mask, a, &mask_attrs);
1618 if ((mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) && encap_valid) {
1619 __be16 eth_type = 0;
1621 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1622 if (a[OVS_KEY_ATTR_ETHERTYPE])
1623 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1624 if (eth_type == htons(0xffff)) {
1625 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1626 encap = a[OVS_KEY_ATTR_ENCAP];
1627 err = parse_flow_mask_nlattrs(encap, a, &mask_attrs);
1629 OVS_NLERR("VLAN frames must have an exact match"
1630 " on the TPID (mask=%x).\n",
1639 err = ovs_key_from_nlattrs(match, mask_attrs, a, true);
1643 /* Populate exact match flow's key mask. */
1645 ovs_sw_flow_mask_set(match->mask, &match->range, 0xff);
1648 if (!ovs_match_validate(match, key_attrs, mask_attrs))
1655 * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1656 * @flow: Receives extracted in_port, priority, tun_key and skb_mark.
1657 * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1660 * This parses a series of Netlink attributes that form a flow key, which must
1661 * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1662 * get the metadata, that is, the parts of the flow key that cannot be
1663 * extracted from the packet itself.
1666 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow,
1667 const struct nlattr *attr)
1669 struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1670 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1673 struct sw_flow_match match;
1675 flow->key.phy.in_port = DP_MAX_PORTS;
1676 flow->key.phy.priority = 0;
1677 flow->key.phy.skb_mark = 0;
1678 memset(tun_key, 0, sizeof(flow->key.tun_key));
1680 err = parse_flow_nlattrs(attr, a, &attrs);
1684 memset(&match, 0, sizeof(match));
1685 match.key = &flow->key;
1687 err = metadata_from_nlattrs(&match, &attrs, a, false);
1694 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey,
1695 const struct sw_flow_key *output, struct sk_buff *skb)
1697 struct ovs_key_ethernet *eth_key;
1698 struct nlattr *nla, *encap;
1700 if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1701 goto nla_put_failure;
1703 if (swkey->tun_key.ipv4_dst &&
1704 ipv4_tun_to_nlattr(skb, &swkey->tun_key, &output->tun_key))
1705 goto nla_put_failure;
1707 if (swkey->phy.in_port == DP_MAX_PORTS) {
1708 if ((swkey != output) && (output->phy.in_port == 0xffff))
1709 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1710 goto nla_put_failure;
1713 upper_u16 = (swkey == output) ? 0 : 0xffff;
1715 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1716 (upper_u16 << 16) | output->phy.in_port))
1717 goto nla_put_failure;
1720 if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1721 goto nla_put_failure;
1723 nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1725 goto nla_put_failure;
1727 eth_key = nla_data(nla);
1728 memcpy(eth_key->eth_src, output->eth.src, ETH_ALEN);
1729 memcpy(eth_key->eth_dst, output->eth.dst, ETH_ALEN);
1731 if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1733 eth_type = (swkey == output) ? htons(ETH_P_8021Q) : htons(0xffff) ;
1734 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1735 nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1736 goto nla_put_failure;
1737 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1738 if (!swkey->eth.tci)
1743 if (swkey->eth.type == htons(ETH_P_802_2)) {
1745 * Ethertype 802.2 is represented in the netlink with omitted
1746 * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1747 * 0xffff in the mask attribute. Ethertype can also
1750 if (swkey != output && output->eth.type)
1751 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1753 goto nla_put_failure;
1757 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1758 goto nla_put_failure;
1760 if (swkey->eth.type == htons(ETH_P_IP)) {
1761 struct ovs_key_ipv4 *ipv4_key;
1763 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1765 goto nla_put_failure;
1766 ipv4_key = nla_data(nla);
1767 ipv4_key->ipv4_src = output->ipv4.addr.src;
1768 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1769 ipv4_key->ipv4_proto = output->ip.proto;
1770 ipv4_key->ipv4_tos = output->ip.tos;
1771 ipv4_key->ipv4_ttl = output->ip.ttl;
1772 ipv4_key->ipv4_frag = output->ip.frag;
1773 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1774 struct ovs_key_ipv6 *ipv6_key;
1776 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1778 goto nla_put_failure;
1779 ipv6_key = nla_data(nla);
1780 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1781 sizeof(ipv6_key->ipv6_src));
1782 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1783 sizeof(ipv6_key->ipv6_dst));
1784 ipv6_key->ipv6_label = output->ipv6.label;
1785 ipv6_key->ipv6_proto = output->ip.proto;
1786 ipv6_key->ipv6_tclass = output->ip.tos;
1787 ipv6_key->ipv6_hlimit = output->ip.ttl;
1788 ipv6_key->ipv6_frag = output->ip.frag;
1789 } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1790 swkey->eth.type == htons(ETH_P_RARP)) {
1791 struct ovs_key_arp *arp_key;
1793 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1795 goto nla_put_failure;
1796 arp_key = nla_data(nla);
1797 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1798 arp_key->arp_sip = output->ipv4.addr.src;
1799 arp_key->arp_tip = output->ipv4.addr.dst;
1800 arp_key->arp_op = htons(output->ip.proto);
1801 memcpy(arp_key->arp_sha, output->ipv4.arp.sha, ETH_ALEN);
1802 memcpy(arp_key->arp_tha, output->ipv4.arp.tha, ETH_ALEN);
1805 if ((swkey->eth.type == htons(ETH_P_IP) ||
1806 swkey->eth.type == htons(ETH_P_IPV6)) &&
1807 swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1809 if (swkey->ip.proto == IPPROTO_TCP) {
1810 struct ovs_key_tcp *tcp_key;
1812 nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1814 goto nla_put_failure;
1815 tcp_key = nla_data(nla);
1816 if (swkey->eth.type == htons(ETH_P_IP)) {
1817 tcp_key->tcp_src = output->ipv4.tp.src;
1818 tcp_key->tcp_dst = output->ipv4.tp.dst;
1819 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1820 tcp_key->tcp_src = output->ipv6.tp.src;
1821 tcp_key->tcp_dst = output->ipv6.tp.dst;
1823 } else if (swkey->ip.proto == IPPROTO_UDP) {
1824 struct ovs_key_udp *udp_key;
1826 nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1828 goto nla_put_failure;
1829 udp_key = nla_data(nla);
1830 if (swkey->eth.type == htons(ETH_P_IP)) {
1831 udp_key->udp_src = output->ipv4.tp.src;
1832 udp_key->udp_dst = output->ipv4.tp.dst;
1833 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1834 udp_key->udp_src = output->ipv6.tp.src;
1835 udp_key->udp_dst = output->ipv6.tp.dst;
1837 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1838 swkey->ip.proto == IPPROTO_ICMP) {
1839 struct ovs_key_icmp *icmp_key;
1841 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1843 goto nla_put_failure;
1844 icmp_key = nla_data(nla);
1845 icmp_key->icmp_type = ntohs(output->ipv4.tp.src);
1846 icmp_key->icmp_code = ntohs(output->ipv4.tp.dst);
1847 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1848 swkey->ip.proto == IPPROTO_ICMPV6) {
1849 struct ovs_key_icmpv6 *icmpv6_key;
1851 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1852 sizeof(*icmpv6_key));
1854 goto nla_put_failure;
1855 icmpv6_key = nla_data(nla);
1856 icmpv6_key->icmpv6_type = ntohs(output->ipv6.tp.src);
1857 icmpv6_key->icmpv6_code = ntohs(output->ipv6.tp.dst);
1859 if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1860 icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1861 struct ovs_key_nd *nd_key;
1863 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1865 goto nla_put_failure;
1866 nd_key = nla_data(nla);
1867 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1868 sizeof(nd_key->nd_target));
1869 memcpy(nd_key->nd_sll, output->ipv6.nd.sll, ETH_ALEN);
1870 memcpy(nd_key->nd_tll, output->ipv6.nd.tll, ETH_ALEN);
1877 nla_nest_end(skb, encap);
1885 /* Initializes the flow module.
1886 * Returns zero if successful or a negative error code. */
1887 int ovs_flow_init(void)
1889 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1891 if (flow_cache == NULL)
1897 /* Uninitializes the flow module. */
1898 void ovs_flow_exit(void)
1900 kmem_cache_destroy(flow_cache);
1903 struct sw_flow_mask *ovs_sw_flow_mask_alloc(void)
1905 struct sw_flow_mask *mask;
1907 mask = kmalloc(sizeof(*mask), GFP_KERNEL);
1909 mask->ref_count = 0;
1914 void ovs_sw_flow_mask_add_ref(struct sw_flow_mask *mask)
1919 static void rcu_free_sw_flow_mask_cb(struct rcu_head *rcu)
1921 struct sw_flow_mask *mask = container_of(rcu, struct sw_flow_mask, rcu);
1926 void ovs_sw_flow_mask_del_ref(struct sw_flow_mask *mask, bool deferred)
1931 BUG_ON(!mask->ref_count);
1934 if (!mask->ref_count) {
1935 list_del_rcu(&mask->list);
1937 call_rcu(&mask->rcu, rcu_free_sw_flow_mask_cb);
1943 static bool ovs_sw_flow_mask_equal(const struct sw_flow_mask *a,
1944 const struct sw_flow_mask *b)
1946 u8 *a_ = (u8 *)&a->key + a->range.start;
1947 u8 *b_ = (u8 *)&b->key + b->range.start;
1949 return (a->range.end == b->range.end)
1950 && (a->range.start == b->range.start)
1951 && (memcmp(a_, b_, ovs_sw_flow_mask_actual_size(a)) == 0);
1954 struct sw_flow_mask *ovs_sw_flow_mask_find(const struct flow_table *tbl,
1955 const struct sw_flow_mask *mask)
1957 struct list_head *ml;
1959 list_for_each(ml, tbl->mask_list) {
1960 struct sw_flow_mask *m;
1961 m = container_of(ml, struct sw_flow_mask, list);
1962 if (ovs_sw_flow_mask_equal(mask, m))
1970 * add a new mask into the mask list.
1971 * The caller needs to make sure that 'mask' is not the same
1972 * as any masks that are already on the list.
1974 void ovs_sw_flow_mask_insert(struct flow_table *tbl, struct sw_flow_mask *mask)
1976 list_add_rcu(&mask->list, tbl->mask_list);
1980 * Set 'range' fields in the mask to the value of 'val'.
1982 static void ovs_sw_flow_mask_set(struct sw_flow_mask *mask,
1983 struct sw_flow_key_range *range, u8 val)
1985 u8 *m = (u8 *)&mask->key + range->start;
1987 mask->range = *range;
1988 memset(m, val, ovs_sw_flow_mask_size_roundup(mask));