Merge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jesse/openvswitch
[~shefty/rdma-dev.git] / net / openvswitch / flow.c
1 /*
2  * Copyright (c) 2007-2011 Nicira, Inc.
3  *
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.
7  *
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.
12  *
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
16  * 02110-1301, USA
17  */
18
19 #include "flow.h"
20 #include "datapath.h"
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>
32 #include <linux/in.h>
33 #include <linux/rcupdate.h>
34 #include <linux/if_arp.h>
35 #include <linux/ip.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>
42 #include <net/ip.h>
43 #include <net/ipv6.h>
44 #include <net/ndisc.h>
45
46 static struct kmem_cache *flow_cache;
47
48 static int check_header(struct sk_buff *skb, int len)
49 {
50         if (unlikely(skb->len < len))
51                 return -EINVAL;
52         if (unlikely(!pskb_may_pull(skb, len)))
53                 return -ENOMEM;
54         return 0;
55 }
56
57 static bool arphdr_ok(struct sk_buff *skb)
58 {
59         return pskb_may_pull(skb, skb_network_offset(skb) +
60                                   sizeof(struct arp_eth_header));
61 }
62
63 static int check_iphdr(struct sk_buff *skb)
64 {
65         unsigned int nh_ofs = skb_network_offset(skb);
66         unsigned int ip_len;
67         int err;
68
69         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
70         if (unlikely(err))
71                 return err;
72
73         ip_len = ip_hdrlen(skb);
74         if (unlikely(ip_len < sizeof(struct iphdr) ||
75                      skb->len < nh_ofs + ip_len))
76                 return -EINVAL;
77
78         skb_set_transport_header(skb, nh_ofs + ip_len);
79         return 0;
80 }
81
82 static bool tcphdr_ok(struct sk_buff *skb)
83 {
84         int th_ofs = skb_transport_offset(skb);
85         int tcp_len;
86
87         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
88                 return false;
89
90         tcp_len = tcp_hdrlen(skb);
91         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
92                      skb->len < th_ofs + tcp_len))
93                 return false;
94
95         return true;
96 }
97
98 static bool udphdr_ok(struct sk_buff *skb)
99 {
100         return pskb_may_pull(skb, skb_transport_offset(skb) +
101                                   sizeof(struct udphdr));
102 }
103
104 static bool icmphdr_ok(struct sk_buff *skb)
105 {
106         return pskb_may_pull(skb, skb_transport_offset(skb) +
107                                   sizeof(struct icmphdr));
108 }
109
110 u64 ovs_flow_used_time(unsigned long flow_jiffies)
111 {
112         struct timespec cur_ts;
113         u64 cur_ms, idle_ms;
114
115         ktime_get_ts(&cur_ts);
116         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
117         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
118                  cur_ts.tv_nsec / NSEC_PER_MSEC;
119
120         return cur_ms - idle_ms;
121 }
122
123 #define SW_FLOW_KEY_OFFSET(field)               \
124         (offsetof(struct sw_flow_key, field) +  \
125          FIELD_SIZEOF(struct sw_flow_key, field))
126
127 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
128                          int *key_lenp)
129 {
130         unsigned int nh_ofs = skb_network_offset(skb);
131         unsigned int nh_len;
132         int payload_ofs;
133         struct ipv6hdr *nh;
134         uint8_t nexthdr;
135         __be16 frag_off;
136         int err;
137
138         *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.label);
139
140         err = check_header(skb, nh_ofs + sizeof(*nh));
141         if (unlikely(err))
142                 return err;
143
144         nh = ipv6_hdr(skb);
145         nexthdr = nh->nexthdr;
146         payload_ofs = (u8 *)(nh + 1) - skb->data;
147
148         key->ip.proto = NEXTHDR_NONE;
149         key->ip.tos = ipv6_get_dsfield(nh);
150         key->ip.ttl = nh->hop_limit;
151         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
152         key->ipv6.addr.src = nh->saddr;
153         key->ipv6.addr.dst = nh->daddr;
154
155         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
156         if (unlikely(payload_ofs < 0))
157                 return -EINVAL;
158
159         if (frag_off) {
160                 if (frag_off & htons(~0x7))
161                         key->ip.frag = OVS_FRAG_TYPE_LATER;
162                 else
163                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
164         }
165
166         nh_len = payload_ofs - nh_ofs;
167         skb_set_transport_header(skb, nh_ofs + nh_len);
168         key->ip.proto = nexthdr;
169         return nh_len;
170 }
171
172 static bool icmp6hdr_ok(struct sk_buff *skb)
173 {
174         return pskb_may_pull(skb, skb_transport_offset(skb) +
175                                   sizeof(struct icmp6hdr));
176 }
177
178 #define TCP_FLAGS_OFFSET 13
179 #define TCP_FLAG_MASK 0x3f
180
181 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
182 {
183         u8 tcp_flags = 0;
184
185         if ((flow->key.eth.type == htons(ETH_P_IP) ||
186              flow->key.eth.type == htons(ETH_P_IPV6)) &&
187             flow->key.ip.proto == IPPROTO_TCP &&
188             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
189                 u8 *tcp = (u8 *)tcp_hdr(skb);
190                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
191         }
192
193         spin_lock(&flow->lock);
194         flow->used = jiffies;
195         flow->packet_count++;
196         flow->byte_count += skb->len;
197         flow->tcp_flags |= tcp_flags;
198         spin_unlock(&flow->lock);
199 }
200
201 struct sw_flow_actions *ovs_flow_actions_alloc(const struct nlattr *actions)
202 {
203         int actions_len = nla_len(actions);
204         struct sw_flow_actions *sfa;
205
206         if (actions_len > MAX_ACTIONS_BUFSIZE)
207                 return ERR_PTR(-EINVAL);
208
209         sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
210         if (!sfa)
211                 return ERR_PTR(-ENOMEM);
212
213         sfa->actions_len = actions_len;
214         memcpy(sfa->actions, nla_data(actions), actions_len);
215         return sfa;
216 }
217
218 struct sw_flow *ovs_flow_alloc(void)
219 {
220         struct sw_flow *flow;
221
222         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
223         if (!flow)
224                 return ERR_PTR(-ENOMEM);
225
226         spin_lock_init(&flow->lock);
227         flow->sf_acts = NULL;
228
229         return flow;
230 }
231
232 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
233 {
234         hash = jhash_1word(hash, table->hash_seed);
235         return flex_array_get(table->buckets,
236                                 (hash & (table->n_buckets - 1)));
237 }
238
239 static struct flex_array *alloc_buckets(unsigned int n_buckets)
240 {
241         struct flex_array *buckets;
242         int i, err;
243
244         buckets = flex_array_alloc(sizeof(struct hlist_head *),
245                                    n_buckets, GFP_KERNEL);
246         if (!buckets)
247                 return NULL;
248
249         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
250         if (err) {
251                 flex_array_free(buckets);
252                 return NULL;
253         }
254
255         for (i = 0; i < n_buckets; i++)
256                 INIT_HLIST_HEAD((struct hlist_head *)
257                                         flex_array_get(buckets, i));
258
259         return buckets;
260 }
261
262 static void free_buckets(struct flex_array *buckets)
263 {
264         flex_array_free(buckets);
265 }
266
267 struct flow_table *ovs_flow_tbl_alloc(int new_size)
268 {
269         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
270
271         if (!table)
272                 return NULL;
273
274         table->buckets = alloc_buckets(new_size);
275
276         if (!table->buckets) {
277                 kfree(table);
278                 return NULL;
279         }
280         table->n_buckets = new_size;
281         table->count = 0;
282         table->node_ver = 0;
283         table->keep_flows = false;
284         get_random_bytes(&table->hash_seed, sizeof(u32));
285
286         return table;
287 }
288
289 void ovs_flow_tbl_destroy(struct flow_table *table)
290 {
291         int i;
292
293         if (!table)
294                 return;
295
296         if (table->keep_flows)
297                 goto skip_flows;
298
299         for (i = 0; i < table->n_buckets; i++) {
300                 struct sw_flow *flow;
301                 struct hlist_head *head = flex_array_get(table->buckets, i);
302                 struct hlist_node *n;
303                 int ver = table->node_ver;
304
305                 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
306                         hlist_del_rcu(&flow->hash_node[ver]);
307                         ovs_flow_free(flow);
308                 }
309         }
310
311 skip_flows:
312         free_buckets(table->buckets);
313         kfree(table);
314 }
315
316 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
317 {
318         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
319
320         ovs_flow_tbl_destroy(table);
321 }
322
323 void ovs_flow_tbl_deferred_destroy(struct flow_table *table)
324 {
325         if (!table)
326                 return;
327
328         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
329 }
330
331 struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
332 {
333         struct sw_flow *flow;
334         struct hlist_head *head;
335         int ver;
336         int i;
337
338         ver = table->node_ver;
339         while (*bucket < table->n_buckets) {
340                 i = 0;
341                 head = flex_array_get(table->buckets, *bucket);
342                 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
343                         if (i < *last) {
344                                 i++;
345                                 continue;
346                         }
347                         *last = i + 1;
348                         return flow;
349                 }
350                 (*bucket)++;
351                 *last = 0;
352         }
353
354         return NULL;
355 }
356
357 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
358 {
359         int old_ver;
360         int i;
361
362         old_ver = old->node_ver;
363         new->node_ver = !old_ver;
364
365         /* Insert in new table. */
366         for (i = 0; i < old->n_buckets; i++) {
367                 struct sw_flow *flow;
368                 struct hlist_head *head;
369
370                 head = flex_array_get(old->buckets, i);
371
372                 hlist_for_each_entry(flow, head, hash_node[old_ver])
373                         ovs_flow_tbl_insert(new, flow);
374         }
375         old->keep_flows = true;
376 }
377
378 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
379 {
380         struct flow_table *new_table;
381
382         new_table = ovs_flow_tbl_alloc(n_buckets);
383         if (!new_table)
384                 return ERR_PTR(-ENOMEM);
385
386         flow_table_copy_flows(table, new_table);
387
388         return new_table;
389 }
390
391 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
392 {
393         return __flow_tbl_rehash(table, table->n_buckets);
394 }
395
396 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
397 {
398         return __flow_tbl_rehash(table, table->n_buckets * 2);
399 }
400
401 void ovs_flow_free(struct sw_flow *flow)
402 {
403         if (unlikely(!flow))
404                 return;
405
406         kfree((struct sf_flow_acts __force *)flow->sf_acts);
407         kmem_cache_free(flow_cache, flow);
408 }
409
410 /* RCU callback used by ovs_flow_deferred_free. */
411 static void rcu_free_flow_callback(struct rcu_head *rcu)
412 {
413         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
414
415         ovs_flow_free(flow);
416 }
417
418 /* Schedules 'flow' to be freed after the next RCU grace period.
419  * The caller must hold rcu_read_lock for this to be sensible. */
420 void ovs_flow_deferred_free(struct sw_flow *flow)
421 {
422         call_rcu(&flow->rcu, rcu_free_flow_callback);
423 }
424
425 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
426  * The caller must hold rcu_read_lock for this to be sensible. */
427 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
428 {
429         kfree_rcu(sf_acts, rcu);
430 }
431
432 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
433 {
434         struct qtag_prefix {
435                 __be16 eth_type; /* ETH_P_8021Q */
436                 __be16 tci;
437         };
438         struct qtag_prefix *qp;
439
440         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
441                 return 0;
442
443         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
444                                          sizeof(__be16))))
445                 return -ENOMEM;
446
447         qp = (struct qtag_prefix *) skb->data;
448         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
449         __skb_pull(skb, sizeof(struct qtag_prefix));
450
451         return 0;
452 }
453
454 static __be16 parse_ethertype(struct sk_buff *skb)
455 {
456         struct llc_snap_hdr {
457                 u8  dsap;  /* Always 0xAA */
458                 u8  ssap;  /* Always 0xAA */
459                 u8  ctrl;
460                 u8  oui[3];
461                 __be16 ethertype;
462         };
463         struct llc_snap_hdr *llc;
464         __be16 proto;
465
466         proto = *(__be16 *) skb->data;
467         __skb_pull(skb, sizeof(__be16));
468
469         if (ntohs(proto) >= 1536)
470                 return proto;
471
472         if (skb->len < sizeof(struct llc_snap_hdr))
473                 return htons(ETH_P_802_2);
474
475         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
476                 return htons(0);
477
478         llc = (struct llc_snap_hdr *) skb->data;
479         if (llc->dsap != LLC_SAP_SNAP ||
480             llc->ssap != LLC_SAP_SNAP ||
481             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
482                 return htons(ETH_P_802_2);
483
484         __skb_pull(skb, sizeof(struct llc_snap_hdr));
485
486         if (ntohs(llc->ethertype) >= 1536)
487                 return llc->ethertype;
488
489         return htons(ETH_P_802_2);
490 }
491
492 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
493                         int *key_lenp, int nh_len)
494 {
495         struct icmp6hdr *icmp = icmp6_hdr(skb);
496         int error = 0;
497         int key_len;
498
499         /* The ICMPv6 type and code fields use the 16-bit transport port
500          * fields, so we need to store them in 16-bit network byte order.
501          */
502         key->ipv6.tp.src = htons(icmp->icmp6_type);
503         key->ipv6.tp.dst = htons(icmp->icmp6_code);
504         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
505
506         if (icmp->icmp6_code == 0 &&
507             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
508              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
509                 int icmp_len = skb->len - skb_transport_offset(skb);
510                 struct nd_msg *nd;
511                 int offset;
512
513                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
514
515                 /* In order to process neighbor discovery options, we need the
516                  * entire packet.
517                  */
518                 if (unlikely(icmp_len < sizeof(*nd)))
519                         goto out;
520                 if (unlikely(skb_linearize(skb))) {
521                         error = -ENOMEM;
522                         goto out;
523                 }
524
525                 nd = (struct nd_msg *)skb_transport_header(skb);
526                 key->ipv6.nd.target = nd->target;
527                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
528
529                 icmp_len -= sizeof(*nd);
530                 offset = 0;
531                 while (icmp_len >= 8) {
532                         struct nd_opt_hdr *nd_opt =
533                                  (struct nd_opt_hdr *)(nd->opt + offset);
534                         int opt_len = nd_opt->nd_opt_len * 8;
535
536                         if (unlikely(!opt_len || opt_len > icmp_len))
537                                 goto invalid;
538
539                         /* Store the link layer address if the appropriate
540                          * option is provided.  It is considered an error if
541                          * the same link layer option is specified twice.
542                          */
543                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
544                             && opt_len == 8) {
545                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
546                                         goto invalid;
547                                 memcpy(key->ipv6.nd.sll,
548                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
549                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
550                                    && opt_len == 8) {
551                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
552                                         goto invalid;
553                                 memcpy(key->ipv6.nd.tll,
554                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
555                         }
556
557                         icmp_len -= opt_len;
558                         offset += opt_len;
559                 }
560         }
561
562         goto out;
563
564 invalid:
565         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
566         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
567         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
568
569 out:
570         *key_lenp = key_len;
571         return error;
572 }
573
574 /**
575  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
576  * @skb: sk_buff that contains the frame, with skb->data pointing to the
577  * Ethernet header
578  * @in_port: port number on which @skb was received.
579  * @key: output flow key
580  * @key_lenp: length of output flow key
581  *
582  * The caller must ensure that skb->len >= ETH_HLEN.
583  *
584  * Returns 0 if successful, otherwise a negative errno value.
585  *
586  * Initializes @skb header pointers as follows:
587  *
588  *    - skb->mac_header: the Ethernet header.
589  *
590  *    - skb->network_header: just past the Ethernet header, or just past the
591  *      VLAN header, to the first byte of the Ethernet payload.
592  *
593  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
594  *      on output, then just past the IP header, if one is present and
595  *      of a correct length, otherwise the same as skb->network_header.
596  *      For other key->dl_type values it is left untouched.
597  */
598 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
599                  int *key_lenp)
600 {
601         int error = 0;
602         int key_len = SW_FLOW_KEY_OFFSET(eth);
603         struct ethhdr *eth;
604
605         memset(key, 0, sizeof(*key));
606
607         key->phy.priority = skb->priority;
608         key->phy.in_port = in_port;
609         key->phy.skb_mark = skb->mark;
610
611         skb_reset_mac_header(skb);
612
613         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
614          * header in the linear data area.
615          */
616         eth = eth_hdr(skb);
617         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
618         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
619
620         __skb_pull(skb, 2 * ETH_ALEN);
621
622         if (vlan_tx_tag_present(skb))
623                 key->eth.tci = htons(skb->vlan_tci);
624         else if (eth->h_proto == htons(ETH_P_8021Q))
625                 if (unlikely(parse_vlan(skb, key)))
626                         return -ENOMEM;
627
628         key->eth.type = parse_ethertype(skb);
629         if (unlikely(key->eth.type == htons(0)))
630                 return -ENOMEM;
631
632         skb_reset_network_header(skb);
633         __skb_push(skb, skb->data - skb_mac_header(skb));
634
635         /* Network layer. */
636         if (key->eth.type == htons(ETH_P_IP)) {
637                 struct iphdr *nh;
638                 __be16 offset;
639
640                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
641
642                 error = check_iphdr(skb);
643                 if (unlikely(error)) {
644                         if (error == -EINVAL) {
645                                 skb->transport_header = skb->network_header;
646                                 error = 0;
647                         }
648                         goto out;
649                 }
650
651                 nh = ip_hdr(skb);
652                 key->ipv4.addr.src = nh->saddr;
653                 key->ipv4.addr.dst = nh->daddr;
654
655                 key->ip.proto = nh->protocol;
656                 key->ip.tos = nh->tos;
657                 key->ip.ttl = nh->ttl;
658
659                 offset = nh->frag_off & htons(IP_OFFSET);
660                 if (offset) {
661                         key->ip.frag = OVS_FRAG_TYPE_LATER;
662                         goto out;
663                 }
664                 if (nh->frag_off & htons(IP_MF) ||
665                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
666                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
667
668                 /* Transport layer. */
669                 if (key->ip.proto == IPPROTO_TCP) {
670                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
671                         if (tcphdr_ok(skb)) {
672                                 struct tcphdr *tcp = tcp_hdr(skb);
673                                 key->ipv4.tp.src = tcp->source;
674                                 key->ipv4.tp.dst = tcp->dest;
675                         }
676                 } else if (key->ip.proto == IPPROTO_UDP) {
677                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
678                         if (udphdr_ok(skb)) {
679                                 struct udphdr *udp = udp_hdr(skb);
680                                 key->ipv4.tp.src = udp->source;
681                                 key->ipv4.tp.dst = udp->dest;
682                         }
683                 } else if (key->ip.proto == IPPROTO_ICMP) {
684                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
685                         if (icmphdr_ok(skb)) {
686                                 struct icmphdr *icmp = icmp_hdr(skb);
687                                 /* The ICMP type and code fields use the 16-bit
688                                  * transport port fields, so we need to store
689                                  * them in 16-bit network byte order. */
690                                 key->ipv4.tp.src = htons(icmp->type);
691                                 key->ipv4.tp.dst = htons(icmp->code);
692                         }
693                 }
694
695         } else if ((key->eth.type == htons(ETH_P_ARP) ||
696                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
697                 struct arp_eth_header *arp;
698
699                 arp = (struct arp_eth_header *)skb_network_header(skb);
700
701                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
702                                 && arp->ar_pro == htons(ETH_P_IP)
703                                 && arp->ar_hln == ETH_ALEN
704                                 && arp->ar_pln == 4) {
705
706                         /* We only match on the lower 8 bits of the opcode. */
707                         if (ntohs(arp->ar_op) <= 0xff)
708                                 key->ip.proto = ntohs(arp->ar_op);
709                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
710                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
711                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
712                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
713                         key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
714                 }
715         } else if (key->eth.type == htons(ETH_P_IPV6)) {
716                 int nh_len;             /* IPv6 Header + Extensions */
717
718                 nh_len = parse_ipv6hdr(skb, key, &key_len);
719                 if (unlikely(nh_len < 0)) {
720                         if (nh_len == -EINVAL)
721                                 skb->transport_header = skb->network_header;
722                         else
723                                 error = nh_len;
724                         goto out;
725                 }
726
727                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
728                         goto out;
729                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
730                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
731
732                 /* Transport layer. */
733                 if (key->ip.proto == NEXTHDR_TCP) {
734                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
735                         if (tcphdr_ok(skb)) {
736                                 struct tcphdr *tcp = tcp_hdr(skb);
737                                 key->ipv6.tp.src = tcp->source;
738                                 key->ipv6.tp.dst = tcp->dest;
739                         }
740                 } else if (key->ip.proto == NEXTHDR_UDP) {
741                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
742                         if (udphdr_ok(skb)) {
743                                 struct udphdr *udp = udp_hdr(skb);
744                                 key->ipv6.tp.src = udp->source;
745                                 key->ipv6.tp.dst = udp->dest;
746                         }
747                 } else if (key->ip.proto == NEXTHDR_ICMP) {
748                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
749                         if (icmp6hdr_ok(skb)) {
750                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
751                                 if (error < 0)
752                                         goto out;
753                         }
754                 }
755         }
756
757 out:
758         *key_lenp = key_len;
759         return error;
760 }
761
762 u32 ovs_flow_hash(const struct sw_flow_key *key, int key_len)
763 {
764         return jhash2((u32 *)key, DIV_ROUND_UP(key_len, sizeof(u32)), 0);
765 }
766
767 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
768                                 struct sw_flow_key *key, int key_len)
769 {
770         struct sw_flow *flow;
771         struct hlist_head *head;
772         u32 hash;
773
774         hash = ovs_flow_hash(key, key_len);
775
776         head = find_bucket(table, hash);
777         hlist_for_each_entry_rcu(flow, head, hash_node[table->node_ver]) {
778
779                 if (flow->hash == hash &&
780                     !memcmp(&flow->key, key, key_len)) {
781                         return flow;
782                 }
783         }
784         return NULL;
785 }
786
787 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
788 {
789         struct hlist_head *head;
790
791         head = find_bucket(table, flow->hash);
792         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
793         table->count++;
794 }
795
796 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
797 {
798         hlist_del_rcu(&flow->hash_node[table->node_ver]);
799         table->count--;
800         BUG_ON(table->count < 0);
801 }
802
803 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
804 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
805         [OVS_KEY_ATTR_ENCAP] = -1,
806         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
807         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
808         [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
809         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
810         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
811         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
812         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
813         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
814         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
815         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
816         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
817         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
818         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
819         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
820 };
821
822 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
823                                   const struct nlattr *a[], u32 *attrs)
824 {
825         const struct ovs_key_icmp *icmp_key;
826         const struct ovs_key_tcp *tcp_key;
827         const struct ovs_key_udp *udp_key;
828
829         switch (swkey->ip.proto) {
830         case IPPROTO_TCP:
831                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
832                         return -EINVAL;
833                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
834
835                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
836                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
837                 swkey->ipv4.tp.src = tcp_key->tcp_src;
838                 swkey->ipv4.tp.dst = tcp_key->tcp_dst;
839                 break;
840
841         case IPPROTO_UDP:
842                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
843                         return -EINVAL;
844                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
845
846                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
847                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
848                 swkey->ipv4.tp.src = udp_key->udp_src;
849                 swkey->ipv4.tp.dst = udp_key->udp_dst;
850                 break;
851
852         case IPPROTO_ICMP:
853                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
854                         return -EINVAL;
855                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
856
857                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
858                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
859                 swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
860                 swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
861                 break;
862         }
863
864         return 0;
865 }
866
867 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
868                                   const struct nlattr *a[], u32 *attrs)
869 {
870         const struct ovs_key_icmpv6 *icmpv6_key;
871         const struct ovs_key_tcp *tcp_key;
872         const struct ovs_key_udp *udp_key;
873
874         switch (swkey->ip.proto) {
875         case IPPROTO_TCP:
876                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
877                         return -EINVAL;
878                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
879
880                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
881                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
882                 swkey->ipv6.tp.src = tcp_key->tcp_src;
883                 swkey->ipv6.tp.dst = tcp_key->tcp_dst;
884                 break;
885
886         case IPPROTO_UDP:
887                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
888                         return -EINVAL;
889                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
890
891                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
892                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
893                 swkey->ipv6.tp.src = udp_key->udp_src;
894                 swkey->ipv6.tp.dst = udp_key->udp_dst;
895                 break;
896
897         case IPPROTO_ICMPV6:
898                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
899                         return -EINVAL;
900                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
901
902                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
903                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
904                 swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
905                 swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
906
907                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
908                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
909                         const struct ovs_key_nd *nd_key;
910
911                         if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
912                                 return -EINVAL;
913                         *attrs &= ~(1 << OVS_KEY_ATTR_ND);
914
915                         *key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
916                         nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
917                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
918                                sizeof(swkey->ipv6.nd.target));
919                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
920                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
921                 }
922                 break;
923         }
924
925         return 0;
926 }
927
928 static int parse_flow_nlattrs(const struct nlattr *attr,
929                               const struct nlattr *a[], u32 *attrsp)
930 {
931         const struct nlattr *nla;
932         u32 attrs;
933         int rem;
934
935         attrs = 0;
936         nla_for_each_nested(nla, attr, rem) {
937                 u16 type = nla_type(nla);
938                 int expected_len;
939
940                 if (type > OVS_KEY_ATTR_MAX || attrs & (1 << type))
941                         return -EINVAL;
942
943                 expected_len = ovs_key_lens[type];
944                 if (nla_len(nla) != expected_len && expected_len != -1)
945                         return -EINVAL;
946
947                 attrs |= 1 << type;
948                 a[type] = nla;
949         }
950         if (rem)
951                 return -EINVAL;
952
953         *attrsp = attrs;
954         return 0;
955 }
956
957 /**
958  * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
959  * @swkey: receives the extracted flow key.
960  * @key_lenp: number of bytes used in @swkey.
961  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
962  * sequence.
963  */
964 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
965                       const struct nlattr *attr)
966 {
967         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
968         const struct ovs_key_ethernet *eth_key;
969         int key_len;
970         u32 attrs;
971         int err;
972
973         memset(swkey, 0, sizeof(struct sw_flow_key));
974         key_len = SW_FLOW_KEY_OFFSET(eth);
975
976         err = parse_flow_nlattrs(attr, a, &attrs);
977         if (err)
978                 return err;
979
980         /* Metadata attributes. */
981         if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
982                 swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
983                 attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
984         }
985         if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
986                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
987                 if (in_port >= DP_MAX_PORTS)
988                         return -EINVAL;
989                 swkey->phy.in_port = in_port;
990                 attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
991         } else {
992                 swkey->phy.in_port = DP_MAX_PORTS;
993         }
994         if (attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
995                 swkey->phy.skb_mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
996                 attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
997         }
998
999         /* Data attributes. */
1000         if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1001                 return -EINVAL;
1002         attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1003
1004         eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1005         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1006         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1007
1008         if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1009             nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1010                 const struct nlattr *encap;
1011                 __be16 tci;
1012
1013                 if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1014                               (1 << OVS_KEY_ATTR_ETHERTYPE) |
1015                               (1 << OVS_KEY_ATTR_ENCAP)))
1016                         return -EINVAL;
1017
1018                 encap = a[OVS_KEY_ATTR_ENCAP];
1019                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1020                 if (tci & htons(VLAN_TAG_PRESENT)) {
1021                         swkey->eth.tci = tci;
1022
1023                         err = parse_flow_nlattrs(encap, a, &attrs);
1024                         if (err)
1025                                 return err;
1026                 } else if (!tci) {
1027                         /* Corner case for truncated 802.1Q header. */
1028                         if (nla_len(encap))
1029                                 return -EINVAL;
1030
1031                         swkey->eth.type = htons(ETH_P_8021Q);
1032                         *key_lenp = key_len;
1033                         return 0;
1034                 } else {
1035                         return -EINVAL;
1036                 }
1037         }
1038
1039         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1040                 swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1041                 if (ntohs(swkey->eth.type) < 1536)
1042                         return -EINVAL;
1043                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1044         } else {
1045                 swkey->eth.type = htons(ETH_P_802_2);
1046         }
1047
1048         if (swkey->eth.type == htons(ETH_P_IP)) {
1049                 const struct ovs_key_ipv4 *ipv4_key;
1050
1051                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1052                         return -EINVAL;
1053                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1054
1055                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1056                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1057                 if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1058                         return -EINVAL;
1059                 swkey->ip.proto = ipv4_key->ipv4_proto;
1060                 swkey->ip.tos = ipv4_key->ipv4_tos;
1061                 swkey->ip.ttl = ipv4_key->ipv4_ttl;
1062                 swkey->ip.frag = ipv4_key->ipv4_frag;
1063                 swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1064                 swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1065
1066                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1067                         err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1068                         if (err)
1069                                 return err;
1070                 }
1071         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1072                 const struct ovs_key_ipv6 *ipv6_key;
1073
1074                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1075                         return -EINVAL;
1076                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1077
1078                 key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1079                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1080                 if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1081                         return -EINVAL;
1082                 swkey->ipv6.label = ipv6_key->ipv6_label;
1083                 swkey->ip.proto = ipv6_key->ipv6_proto;
1084                 swkey->ip.tos = ipv6_key->ipv6_tclass;
1085                 swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1086                 swkey->ip.frag = ipv6_key->ipv6_frag;
1087                 memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1088                        sizeof(swkey->ipv6.addr.src));
1089                 memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1090                        sizeof(swkey->ipv6.addr.dst));
1091
1092                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1093                         err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1094                         if (err)
1095                                 return err;
1096                 }
1097         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1098                    swkey->eth.type == htons(ETH_P_RARP)) {
1099                 const struct ovs_key_arp *arp_key;
1100
1101                 if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1102                         return -EINVAL;
1103                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1104
1105                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1106                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1107                 swkey->ipv4.addr.src = arp_key->arp_sip;
1108                 swkey->ipv4.addr.dst = arp_key->arp_tip;
1109                 if (arp_key->arp_op & htons(0xff00))
1110                         return -EINVAL;
1111                 swkey->ip.proto = ntohs(arp_key->arp_op);
1112                 memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1113                 memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1114         }
1115
1116         if (attrs)
1117                 return -EINVAL;
1118         *key_lenp = key_len;
1119
1120         return 0;
1121 }
1122
1123 /**
1124  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1125  * @priority: receives the skb priority
1126  * @mark: receives the skb mark
1127  * @in_port: receives the extracted input port.
1128  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1129  * sequence.
1130  *
1131  * This parses a series of Netlink attributes that form a flow key, which must
1132  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1133  * get the metadata, that is, the parts of the flow key that cannot be
1134  * extracted from the packet itself.
1135  */
1136 int ovs_flow_metadata_from_nlattrs(u32 *priority, u32 *mark, u16 *in_port,
1137                                const struct nlattr *attr)
1138 {
1139         const struct nlattr *nla;
1140         int rem;
1141
1142         *in_port = DP_MAX_PORTS;
1143         *priority = 0;
1144         *mark = 0;
1145
1146         nla_for_each_nested(nla, attr, rem) {
1147                 int type = nla_type(nla);
1148
1149                 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1150                         if (nla_len(nla) != ovs_key_lens[type])
1151                                 return -EINVAL;
1152
1153                         switch (type) {
1154                         case OVS_KEY_ATTR_PRIORITY:
1155                                 *priority = nla_get_u32(nla);
1156                                 break;
1157
1158                         case OVS_KEY_ATTR_IN_PORT:
1159                                 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1160                                         return -EINVAL;
1161                                 *in_port = nla_get_u32(nla);
1162                                 break;
1163
1164                         case OVS_KEY_ATTR_SKB_MARK:
1165                                 *mark = nla_get_u32(nla);
1166                                 break;
1167                         }
1168                 }
1169         }
1170         if (rem)
1171                 return -EINVAL;
1172         return 0;
1173 }
1174
1175 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1176 {
1177         struct ovs_key_ethernet *eth_key;
1178         struct nlattr *nla, *encap;
1179
1180         if (swkey->phy.priority &&
1181             nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority))
1182                 goto nla_put_failure;
1183
1184         if (swkey->phy.in_port != DP_MAX_PORTS &&
1185             nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port))
1186                 goto nla_put_failure;
1187
1188         if (swkey->phy.skb_mark &&
1189             nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, swkey->phy.skb_mark))
1190                 goto nla_put_failure;
1191
1192         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1193         if (!nla)
1194                 goto nla_put_failure;
1195         eth_key = nla_data(nla);
1196         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1197         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1198
1199         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1200                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q)) ||
1201                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci))
1202                         goto nla_put_failure;
1203                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1204                 if (!swkey->eth.tci)
1205                         goto unencap;
1206         } else {
1207                 encap = NULL;
1208         }
1209
1210         if (swkey->eth.type == htons(ETH_P_802_2))
1211                 goto unencap;
1212
1213         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type))
1214                 goto nla_put_failure;
1215
1216         if (swkey->eth.type == htons(ETH_P_IP)) {
1217                 struct ovs_key_ipv4 *ipv4_key;
1218
1219                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1220                 if (!nla)
1221                         goto nla_put_failure;
1222                 ipv4_key = nla_data(nla);
1223                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1224                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1225                 ipv4_key->ipv4_proto = swkey->ip.proto;
1226                 ipv4_key->ipv4_tos = swkey->ip.tos;
1227                 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1228                 ipv4_key->ipv4_frag = swkey->ip.frag;
1229         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1230                 struct ovs_key_ipv6 *ipv6_key;
1231
1232                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1233                 if (!nla)
1234                         goto nla_put_failure;
1235                 ipv6_key = nla_data(nla);
1236                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1237                                 sizeof(ipv6_key->ipv6_src));
1238                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1239                                 sizeof(ipv6_key->ipv6_dst));
1240                 ipv6_key->ipv6_label = swkey->ipv6.label;
1241                 ipv6_key->ipv6_proto = swkey->ip.proto;
1242                 ipv6_key->ipv6_tclass = swkey->ip.tos;
1243                 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1244                 ipv6_key->ipv6_frag = swkey->ip.frag;
1245         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1246                    swkey->eth.type == htons(ETH_P_RARP)) {
1247                 struct ovs_key_arp *arp_key;
1248
1249                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1250                 if (!nla)
1251                         goto nla_put_failure;
1252                 arp_key = nla_data(nla);
1253                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1254                 arp_key->arp_sip = swkey->ipv4.addr.src;
1255                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1256                 arp_key->arp_op = htons(swkey->ip.proto);
1257                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1258                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1259         }
1260
1261         if ((swkey->eth.type == htons(ETH_P_IP) ||
1262              swkey->eth.type == htons(ETH_P_IPV6)) &&
1263              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1264
1265                 if (swkey->ip.proto == IPPROTO_TCP) {
1266                         struct ovs_key_tcp *tcp_key;
1267
1268                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1269                         if (!nla)
1270                                 goto nla_put_failure;
1271                         tcp_key = nla_data(nla);
1272                         if (swkey->eth.type == htons(ETH_P_IP)) {
1273                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1274                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1275                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1276                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1277                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1278                         }
1279                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1280                         struct ovs_key_udp *udp_key;
1281
1282                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1283                         if (!nla)
1284                                 goto nla_put_failure;
1285                         udp_key = nla_data(nla);
1286                         if (swkey->eth.type == htons(ETH_P_IP)) {
1287                                 udp_key->udp_src = swkey->ipv4.tp.src;
1288                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1289                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1290                                 udp_key->udp_src = swkey->ipv6.tp.src;
1291                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1292                         }
1293                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1294                            swkey->ip.proto == IPPROTO_ICMP) {
1295                         struct ovs_key_icmp *icmp_key;
1296
1297                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1298                         if (!nla)
1299                                 goto nla_put_failure;
1300                         icmp_key = nla_data(nla);
1301                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1302                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1303                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1304                            swkey->ip.proto == IPPROTO_ICMPV6) {
1305                         struct ovs_key_icmpv6 *icmpv6_key;
1306
1307                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1308                                                 sizeof(*icmpv6_key));
1309                         if (!nla)
1310                                 goto nla_put_failure;
1311                         icmpv6_key = nla_data(nla);
1312                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1313                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1314
1315                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1316                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1317                                 struct ovs_key_nd *nd_key;
1318
1319                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1320                                 if (!nla)
1321                                         goto nla_put_failure;
1322                                 nd_key = nla_data(nla);
1323                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1324                                                         sizeof(nd_key->nd_target));
1325                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1326                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1327                         }
1328                 }
1329         }
1330
1331 unencap:
1332         if (encap)
1333                 nla_nest_end(skb, encap);
1334
1335         return 0;
1336
1337 nla_put_failure:
1338         return -EMSGSIZE;
1339 }
1340
1341 /* Initializes the flow module.
1342  * Returns zero if successful or a negative error code. */
1343 int ovs_flow_init(void)
1344 {
1345         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1346                                         0, NULL);
1347         if (flow_cache == NULL)
1348                 return -ENOMEM;
1349
1350         return 0;
1351 }
1352
1353 /* Uninitializes the flow module. */
1354 void ovs_flow_exit(void)
1355 {
1356         kmem_cache_destroy(flow_cache);
1357 }