datapath: Drop useless WARN_ON_ONCE during flow conversion.
[openvswitch] / datapath / flow.c
1 /*
2  * Distributed under the terms of the GNU GPL version 2.
3  * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
4  *
5  * Significant portions of this file may be copied from parts of the Linux
6  * kernel, by Linus Torvalds and others.
7  */
8
9 #include "flow.h"
10 #include "datapath.h"
11 #include <asm/uaccess.h>
12 #include <linux/netdevice.h>
13 #include <linux/etherdevice.h>
14 #include <linux/if_ether.h>
15 #include <linux/if_vlan.h>
16 #include <net/llc_pdu.h>
17 #include <linux/kernel.h>
18 #include <linux/jhash.h>
19 #include <linux/jiffies.h>
20 #include <linux/llc.h>
21 #include <linux/module.h>
22 #include <linux/in.h>
23 #include <linux/rcupdate.h>
24 #include <linux/if_arp.h>
25 #include <linux/if_ether.h>
26 #include <linux/ip.h>
27 #include <linux/ipv6.h>
28 #include <linux/tcp.h>
29 #include <linux/udp.h>
30 #include <linux/icmp.h>
31 #include <linux/icmpv6.h>
32 #include <linux/rculist.h>
33 #include <net/inet_ecn.h>
34 #include <net/ip.h>
35 #include <net/ipv6.h>
36 #include <net/ndisc.h>
37
38 #include "vlan.h"
39
40 static struct kmem_cache *flow_cache;
41 static unsigned int hash_seed __read_mostly;
42
43 static int check_header(struct sk_buff *skb, int len)
44 {
45         if (unlikely(skb->len < len))
46                 return -EINVAL;
47         if (unlikely(!pskb_may_pull(skb, len)))
48                 return -ENOMEM;
49         return 0;
50 }
51
52 static inline bool arphdr_ok(struct sk_buff *skb)
53 {
54         return pskb_may_pull(skb, skb_network_offset(skb) +
55                                   sizeof(struct arp_eth_header));
56 }
57
58 static inline int check_iphdr(struct sk_buff *skb)
59 {
60         unsigned int nh_ofs = skb_network_offset(skb);
61         unsigned int ip_len;
62         int err;
63
64         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
65         if (unlikely(err))
66                 return err;
67
68         ip_len = ip_hdrlen(skb);
69         if (unlikely(ip_len < sizeof(struct iphdr) ||
70                      skb->len < nh_ofs + ip_len))
71                 return -EINVAL;
72
73         skb_set_transport_header(skb, nh_ofs + ip_len);
74         return 0;
75 }
76
77 static inline bool tcphdr_ok(struct sk_buff *skb)
78 {
79         int th_ofs = skb_transport_offset(skb);
80         int tcp_len;
81
82         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
83                 return false;
84
85         tcp_len = tcp_hdrlen(skb);
86         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
87                      skb->len < th_ofs + tcp_len))
88                 return false;
89
90         return true;
91 }
92
93 static inline bool udphdr_ok(struct sk_buff *skb)
94 {
95         return pskb_may_pull(skb, skb_transport_offset(skb) +
96                                   sizeof(struct udphdr));
97 }
98
99 static inline bool icmphdr_ok(struct sk_buff *skb)
100 {
101         return pskb_may_pull(skb, skb_transport_offset(skb) +
102                                   sizeof(struct icmphdr));
103 }
104
105 u64 flow_used_time(unsigned long flow_jiffies)
106 {
107         struct timespec cur_ts;
108         u64 cur_ms, idle_ms;
109
110         ktime_get_ts(&cur_ts);
111         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
112         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
113                  cur_ts.tv_nsec / NSEC_PER_MSEC;
114
115         return cur_ms - idle_ms;
116 }
117
118 #define SW_FLOW_KEY_OFFSET(field)               \
119         offsetof(struct sw_flow_key, field) +   \
120         FIELD_SIZEOF(struct sw_flow_key, field)
121
122 /**
123  * skip_exthdr - skip any IPv6 extension headers
124  * @skb: skbuff to parse
125  * @start: offset of first extension header
126  * @nexthdrp: Initially, points to the type of the extension header at @start.
127  * This function updates it to point to the extension header at the final
128  * offset.
129  * @tos_frag: Points to the @tos_frag member in a &struct sw_flow_key.  This
130  * function sets an appropriate %OVS_FRAG_TYPE_* value.
131  *
132  * This is based on ipv6_skip_exthdr() but adds the updates to *@tos_frag.
133  *
134  * When there is more than one fragment header, this version reports whether
135  * the final fragment header that it examines is a first fragment.
136  *
137  * Returns the final payload offset, or -1 on error.
138  */
139 static int skip_exthdr(const struct sk_buff *skb, int start, u8 *nexthdrp,
140                        u8 *tos_frag)
141 {
142         u8 nexthdr = *nexthdrp;
143
144         while (ipv6_ext_hdr(nexthdr)) {
145                 struct ipv6_opt_hdr _hdr, *hp;
146                 int hdrlen;
147
148                 if (nexthdr == NEXTHDR_NONE)
149                         return -1;
150                 hp = skb_header_pointer(skb, start, sizeof(_hdr), &_hdr);
151                 if (hp == NULL)
152                         return -1;
153                 if (nexthdr == NEXTHDR_FRAGMENT) {
154                         __be16 _frag_off, *fp;
155                         fp = skb_header_pointer(skb,
156                                                 start+offsetof(struct frag_hdr,
157                                                                frag_off),
158                                                 sizeof(_frag_off),
159                                                 &_frag_off);
160                         if (fp == NULL)
161                                 return -1;
162
163                         *tos_frag &= ~OVS_FRAG_TYPE_MASK;
164                         if (ntohs(*fp) & ~0x7) {
165                                 *tos_frag |= OVS_FRAG_TYPE_LATER;
166                                 break;
167                         }
168                         *tos_frag |= OVS_FRAG_TYPE_FIRST;
169                         hdrlen = 8;
170                 } else if (nexthdr == NEXTHDR_AUTH)
171                         hdrlen = (hp->hdrlen+2)<<2;
172                 else
173                         hdrlen = ipv6_optlen(hp);
174
175                 nexthdr = hp->nexthdr;
176                 start += hdrlen;
177         }
178
179         *nexthdrp = nexthdr;
180         return start;
181 }
182
183 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
184                          int *key_lenp)
185 {
186         unsigned int nh_ofs = skb_network_offset(skb);
187         unsigned int nh_len;
188         int payload_ofs;
189         struct ipv6hdr *nh;
190         uint8_t nexthdr;
191         int err;
192
193         *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.addr);
194
195         err = check_header(skb, nh_ofs + sizeof(*nh));
196         if (unlikely(err))
197                 return err;
198
199         nh = ipv6_hdr(skb);
200         nexthdr = nh->nexthdr;
201         payload_ofs = (u8 *)(nh + 1) - skb->data;
202
203         key->ip.proto = NEXTHDR_NONE;
204         key->ip.tos_frag = ipv6_get_dsfield(nh) & ~INET_ECN_MASK;
205         ipv6_addr_copy(&key->ipv6.addr.src, &nh->saddr);
206         ipv6_addr_copy(&key->ipv6.addr.dst, &nh->daddr);
207
208         payload_ofs = skip_exthdr(skb, payload_ofs, &nexthdr, &key->ip.tos_frag);
209         if (unlikely(payload_ofs < 0))
210                 return -EINVAL;
211
212         nh_len = payload_ofs - nh_ofs;
213         skb_set_transport_header(skb, nh_ofs + nh_len);
214         key->ip.proto = nexthdr;
215         return nh_len;
216 }
217
218 static bool icmp6hdr_ok(struct sk_buff *skb)
219 {
220         return pskb_may_pull(skb, skb_transport_offset(skb) +
221                                   sizeof(struct icmp6hdr));
222 }
223
224 #define TCP_FLAGS_OFFSET 13
225 #define TCP_FLAG_MASK 0x3f
226
227 void flow_used(struct sw_flow *flow, struct sk_buff *skb)
228 {
229         u8 tcp_flags = 0;
230
231         if (flow->key.eth.type == htons(ETH_P_IP) &&
232             flow->key.ip.proto == IPPROTO_TCP) {
233                 u8 *tcp = (u8 *)tcp_hdr(skb);
234                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
235         }
236
237         spin_lock(&flow->lock);
238         flow->used = jiffies;
239         flow->packet_count++;
240         flow->byte_count += skb->len;
241         flow->tcp_flags |= tcp_flags;
242         spin_unlock(&flow->lock);
243 }
244
245 struct sw_flow_actions *flow_actions_alloc(const struct nlattr *actions)
246 {
247         int actions_len = nla_len(actions);
248         struct sw_flow_actions *sfa;
249
250         /* At least DP_MAX_PORTS actions are required to be able to flood a
251          * packet to every port.  Factor of 2 allows for setting VLAN tags,
252          * etc. */
253         if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
254                 return ERR_PTR(-EINVAL);
255
256         sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
257         if (!sfa)
258                 return ERR_PTR(-ENOMEM);
259
260         sfa->actions_len = actions_len;
261         memcpy(sfa->actions, nla_data(actions), actions_len);
262         return sfa;
263 }
264
265 struct sw_flow *flow_alloc(void)
266 {
267         struct sw_flow *flow;
268
269         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
270         if (!flow)
271                 return ERR_PTR(-ENOMEM);
272
273         spin_lock_init(&flow->lock);
274         atomic_set(&flow->refcnt, 1);
275         flow->sf_acts = NULL;
276         flow->dead = false;
277
278         return flow;
279 }
280
281 static struct hlist_head __rcu *find_bucket(struct flow_table * table, u32 hash)
282 {
283         return flex_array_get(table->buckets,
284                                 (hash & (table->n_buckets - 1)));
285 }
286
287 static struct flex_array  __rcu *alloc_buckets(unsigned int n_buckets)
288 {
289         struct flex_array  __rcu * buckets;
290         int i, err;
291
292         buckets = flex_array_alloc(sizeof(struct hlist_head *),
293                                    n_buckets, GFP_KERNEL);
294         if (!buckets)
295                 return NULL;
296
297         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
298         if (err) {
299                 flex_array_free(buckets);
300                 return NULL;
301         }
302
303         for (i = 0; i < n_buckets; i++)
304                 INIT_HLIST_HEAD((struct hlist_head *)
305                                         flex_array_get(buckets, i));
306
307         return buckets;
308 }
309
310 static void free_buckets(struct flex_array * buckets)
311 {
312         flex_array_free(buckets);
313 }
314
315 struct flow_table *flow_tbl_alloc(int new_size)
316 {
317         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
318
319         if (!table)
320                 return NULL;
321
322         table->buckets = alloc_buckets(new_size);
323
324         if (!table->buckets) {
325                 kfree(table);
326                 return NULL;
327         }
328         table->n_buckets = new_size;
329         table->count = 0;
330
331         return table;
332 }
333
334 static void flow_free(struct sw_flow *flow)
335 {
336         flow->dead = true;
337         flow_put(flow);
338 }
339
340 void flow_tbl_destroy(struct flow_table *table)
341 {
342         int i;
343
344         if (!table)
345                 return;
346
347         for (i = 0; i < table->n_buckets; i++) {
348                 struct sw_flow *flow;
349                 struct hlist_head *head = flex_array_get(table->buckets, i);
350                 struct hlist_node *node, *n;
351
352                 hlist_for_each_entry_safe(flow, node, n, head, hash_node) {
353                         hlist_del_init_rcu(&flow->hash_node);
354                         flow_free(flow);
355                 }
356         }
357
358         free_buckets(table->buckets);
359         kfree(table);
360 }
361
362 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
363 {
364         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
365
366         flow_tbl_destroy(table);
367 }
368
369 void flow_tbl_deferred_destroy(struct flow_table *table)
370 {
371         if (!table)
372                 return;
373
374         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
375 }
376
377 struct sw_flow *flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
378 {
379         struct sw_flow *flow;
380         struct hlist_head *head;
381         struct hlist_node *n;
382         int i;
383
384         while (*bucket < table->n_buckets) {
385                 i = 0;
386                 head = flex_array_get(table->buckets, *bucket);
387                 hlist_for_each_entry_rcu(flow, n, head, hash_node) {
388                         if (i < *last) {
389                                 i++;
390                                 continue;
391                         }
392                         *last = i + 1;
393                         return flow;
394                 }
395                 (*bucket)++;
396                 *last = 0;
397         }
398
399         return NULL;
400 }
401
402 struct flow_table *flow_tbl_expand(struct flow_table *table)
403 {
404         struct flow_table *new_table;
405         int n_buckets = table->n_buckets * 2;
406         int i;
407
408         new_table = flow_tbl_alloc(n_buckets);
409         if (!new_table)
410                 return ERR_PTR(-ENOMEM);
411
412         for (i = 0; i < table->n_buckets; i++) {
413                 struct sw_flow *flow;
414                 struct hlist_head *head;
415                 struct hlist_node *n, *pos;
416
417                 head = flex_array_get(table->buckets, i);
418
419                 hlist_for_each_entry_safe(flow, n, pos, head, hash_node) {
420                         hlist_del_init_rcu(&flow->hash_node);
421                         flow_tbl_insert(new_table, flow);
422                 }
423         }
424
425         return new_table;
426 }
427
428 /* RCU callback used by flow_deferred_free. */
429 static void rcu_free_flow_callback(struct rcu_head *rcu)
430 {
431         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
432
433         flow->dead = true;
434         flow_put(flow);
435 }
436
437 /* Schedules 'flow' to be freed after the next RCU grace period.
438  * The caller must hold rcu_read_lock for this to be sensible. */
439 void flow_deferred_free(struct sw_flow *flow)
440 {
441         call_rcu(&flow->rcu, rcu_free_flow_callback);
442 }
443
444 void flow_hold(struct sw_flow *flow)
445 {
446         atomic_inc(&flow->refcnt);
447 }
448
449 void flow_put(struct sw_flow *flow)
450 {
451         if (unlikely(!flow))
452                 return;
453
454         if (atomic_dec_and_test(&flow->refcnt)) {
455                 kfree((struct sf_flow_acts __force *)flow->sf_acts);
456                 kmem_cache_free(flow_cache, flow);
457         }
458 }
459
460 /* RCU callback used by flow_deferred_free_acts. */
461 static void rcu_free_acts_callback(struct rcu_head *rcu)
462 {
463         struct sw_flow_actions *sf_acts = container_of(rcu,
464                         struct sw_flow_actions, rcu);
465         kfree(sf_acts);
466 }
467
468 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
469  * The caller must hold rcu_read_lock for this to be sensible. */
470 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
471 {
472         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
473 }
474
475 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
476 {
477         struct qtag_prefix {
478                 __be16 eth_type; /* ETH_P_8021Q */
479                 __be16 tci;
480         };
481         struct qtag_prefix *qp;
482
483         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
484                                          sizeof(__be16))))
485                 return -ENOMEM;
486
487         qp = (struct qtag_prefix *) skb->data;
488         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
489         __skb_pull(skb, sizeof(struct qtag_prefix));
490
491         return 0;
492 }
493
494 static __be16 parse_ethertype(struct sk_buff *skb)
495 {
496         struct llc_snap_hdr {
497                 u8  dsap;  /* Always 0xAA */
498                 u8  ssap;  /* Always 0xAA */
499                 u8  ctrl;
500                 u8  oui[3];
501                 __be16 ethertype;
502         };
503         struct llc_snap_hdr *llc;
504         __be16 proto;
505
506         proto = *(__be16 *) skb->data;
507         __skb_pull(skb, sizeof(__be16));
508
509         if (ntohs(proto) >= 1536)
510                 return proto;
511
512         if (skb->len < sizeof(struct llc_snap_hdr))
513                 return htons(ETH_P_802_2);
514
515         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
516                 return htons(0);
517
518         llc = (struct llc_snap_hdr *) skb->data;
519         if (llc->dsap != LLC_SAP_SNAP ||
520             llc->ssap != LLC_SAP_SNAP ||
521             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
522                 return htons(ETH_P_802_2);
523
524         __skb_pull(skb, sizeof(struct llc_snap_hdr));
525         return llc->ethertype;
526 }
527
528 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
529                         int *key_lenp, int nh_len)
530 {
531         struct icmp6hdr *icmp = icmp6_hdr(skb);
532         int error = 0;
533         int key_len;
534
535         /* The ICMPv6 type and code fields use the 16-bit transport port
536          * fields, so we need to store them in 16-bit network byte order.
537          */
538         key->ipv6.tp.src = htons(icmp->icmp6_type);
539         key->ipv6.tp.dst = htons(icmp->icmp6_code);
540         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
541
542         if (icmp->icmp6_code == 0 &&
543             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
544              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
545                 int icmp_len = skb->len - skb_transport_offset(skb);
546                 struct nd_msg *nd;
547                 int offset;
548
549                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
550
551                 /* In order to process neighbor discovery options, we need the
552                  * entire packet.
553                  */
554                 if (unlikely(icmp_len < sizeof(*nd)))
555                         goto out;
556                 if (unlikely(skb_linearize(skb))) {
557                         error = -ENOMEM;
558                         goto out;
559                 }
560
561                 nd = (struct nd_msg *)skb_transport_header(skb);
562                 ipv6_addr_copy(&key->ipv6.nd.target, &nd->target);
563                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
564
565                 icmp_len -= sizeof(*nd);
566                 offset = 0;
567                 while (icmp_len >= 8) {
568                         struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd->opt + offset);
569                         int opt_len = nd_opt->nd_opt_len * 8;
570
571                         if (unlikely(!opt_len || opt_len > icmp_len))
572                                 goto invalid;
573
574                         /* Store the link layer address if the appropriate
575                          * option is provided.  It is considered an error if
576                          * the same link layer option is specified twice.
577                          */
578                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
579                             && opt_len == 8) {
580                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
581                                         goto invalid;
582                                 memcpy(key->ipv6.nd.sll,
583                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
584                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
585                                    && opt_len == 8) {
586                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
587                                         goto invalid;
588                                 memcpy(key->ipv6.nd.tll,
589                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
590                         }
591
592                         icmp_len -= opt_len;
593                         offset += opt_len;
594                 }
595         }
596
597         goto out;
598
599 invalid:
600         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
601         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
602         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
603
604 out:
605         *key_lenp = key_len;
606         return error;
607 }
608
609 /**
610  * flow_extract - extracts a flow key from an Ethernet frame.
611  * @skb: sk_buff that contains the frame, with skb->data pointing to the
612  * Ethernet header
613  * @in_port: port number on which @skb was received.
614  * @key: output flow key
615  * @key_lenp: length of output flow key
616  *
617  * The caller must ensure that skb->len >= ETH_HLEN.
618  *
619  * Returns 0 if successful, otherwise a negative errno value.
620  *
621  * Initializes @skb header pointers as follows:
622  *
623  *    - skb->mac_header: the Ethernet header.
624  *
625  *    - skb->network_header: just past the Ethernet header, or just past the
626  *      VLAN header, to the first byte of the Ethernet payload.
627  *
628  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
629  *      on output, then just past the IP header, if one is present and
630  *      of a correct length, otherwise the same as skb->network_header.
631  *      For other key->dl_type values it is left untouched.
632  */
633 int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
634                  int *key_lenp)
635 {
636         int error = 0;
637         int key_len = SW_FLOW_KEY_OFFSET(eth);
638         struct ethhdr *eth;
639
640         memset(key, 0, sizeof(*key));
641
642         key->phy.priority = skb->priority;
643         key->phy.tun_id = OVS_CB(skb)->tun_id;
644         key->phy.in_port = in_port;
645
646         skb_reset_mac_header(skb);
647
648         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
649          * header in the linear data area.
650          */
651         eth = eth_hdr(skb);
652         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
653         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
654
655         __skb_pull(skb, 2 * ETH_ALEN);
656
657         if (vlan_tx_tag_present(skb))
658                 key->eth.tci = htons(vlan_get_tci(skb));
659         else if (eth->h_proto == htons(ETH_P_8021Q))
660                 if (unlikely(parse_vlan(skb, key)))
661                         return -ENOMEM;
662
663         key->eth.type = parse_ethertype(skb);
664         if (unlikely(key->eth.type == htons(0)))
665                 return -ENOMEM;
666
667         skb_reset_network_header(skb);
668         __skb_push(skb, skb->data - skb_mac_header(skb));
669
670         /* Network layer. */
671         if (key->eth.type == htons(ETH_P_IP)) {
672                 struct iphdr *nh;
673                 __be16 offset;
674
675                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
676
677                 error = check_iphdr(skb);
678                 if (unlikely(error)) {
679                         if (error == -EINVAL) {
680                                 skb->transport_header = skb->network_header;
681                                 error = 0;
682                         }
683                         goto out;
684                 }
685
686                 nh = ip_hdr(skb);
687                 key->ipv4.addr.src = nh->saddr;
688                 key->ipv4.addr.dst = nh->daddr;
689
690                 key->ip.proto = nh->protocol;
691                 key->ip.tos_frag = nh->tos & ~INET_ECN_MASK;
692
693                 offset = nh->frag_off & htons(IP_OFFSET);
694                 if (offset) {
695                         key->ip.tos_frag |= OVS_FRAG_TYPE_LATER;
696                         goto out;
697                 }
698                 if (nh->frag_off & htons(IP_MF) ||
699                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
700                         key->ip.tos_frag |= OVS_FRAG_TYPE_FIRST;
701
702                 /* Transport layer. */
703                 if (key->ip.proto == IPPROTO_TCP) {
704                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
705                         if (tcphdr_ok(skb)) {
706                                 struct tcphdr *tcp = tcp_hdr(skb);
707                                 key->ipv4.tp.src = tcp->source;
708                                 key->ipv4.tp.dst = tcp->dest;
709                         }
710                 } else if (key->ip.proto == IPPROTO_UDP) {
711                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
712                         if (udphdr_ok(skb)) {
713                                 struct udphdr *udp = udp_hdr(skb);
714                                 key->ipv4.tp.src = udp->source;
715                                 key->ipv4.tp.dst = udp->dest;
716                         }
717                 } else if (key->ip.proto == IPPROTO_ICMP) {
718                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
719                         if (icmphdr_ok(skb)) {
720                                 struct icmphdr *icmp = icmp_hdr(skb);
721                                 /* The ICMP type and code fields use the 16-bit
722                                  * transport port fields, so we need to store them
723                                  * in 16-bit network byte order. */
724                                 key->ipv4.tp.src = htons(icmp->type);
725                                 key->ipv4.tp.dst = htons(icmp->code);
726                         }
727                 }
728
729         } else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
730                 struct arp_eth_header *arp;
731
732                 arp = (struct arp_eth_header *)skb_network_header(skb);
733
734                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
735                                 && arp->ar_pro == htons(ETH_P_IP)
736                                 && arp->ar_hln == ETH_ALEN
737                                 && arp->ar_pln == 4) {
738
739                         /* We only match on the lower 8 bits of the opcode. */
740                         if (ntohs(arp->ar_op) <= 0xff)
741                                 key->ip.proto = ntohs(arp->ar_op);
742
743                         if (key->ip.proto == ARPOP_REQUEST
744                                         || key->ip.proto == ARPOP_REPLY) {
745                                 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
746                                 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
747                                 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
748                                 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
749                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
750                         }
751                 }
752         } else if (key->eth.type == htons(ETH_P_IPV6)) {
753                 int nh_len;             /* IPv6 Header + Extensions */
754
755                 nh_len = parse_ipv6hdr(skb, key, &key_len);
756                 if (unlikely(nh_len < 0)) {
757                         if (nh_len == -EINVAL)
758                                 skb->transport_header = skb->network_header;
759                         else
760                                 error = nh_len;
761                         goto out;
762                 }
763
764                 if ((key->ip.tos_frag & OVS_FRAG_TYPE_MASK) == OVS_FRAG_TYPE_LATER)
765                         goto out;
766                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
767                         key->ip.tos_frag |= OVS_FRAG_TYPE_FIRST;
768
769                 /* Transport layer. */
770                 if (key->ip.proto == NEXTHDR_TCP) {
771                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
772                         if (tcphdr_ok(skb)) {
773                                 struct tcphdr *tcp = tcp_hdr(skb);
774                                 key->ipv6.tp.src = tcp->source;
775                                 key->ipv6.tp.dst = tcp->dest;
776                         }
777                 } else if (key->ip.proto == NEXTHDR_UDP) {
778                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
779                         if (udphdr_ok(skb)) {
780                                 struct udphdr *udp = udp_hdr(skb);
781                                 key->ipv6.tp.src = udp->source;
782                                 key->ipv6.tp.dst = udp->dest;
783                         }
784                 } else if (key->ip.proto == NEXTHDR_ICMP) {
785                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
786                         if (icmp6hdr_ok(skb)) {
787                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
788                                 if (error < 0)
789                                         goto out;
790                         }
791                 }
792         }
793
794 out:
795         *key_lenp = key_len;
796         return error;
797 }
798
799 u32 flow_hash(const struct sw_flow_key *key, int key_len)
800 {
801         return jhash2((u32*)key, DIV_ROUND_UP(key_len, sizeof(u32)), hash_seed);
802 }
803
804 struct sw_flow * flow_tbl_lookup(struct flow_table *table,
805                                 struct sw_flow_key *key, int key_len)
806 {
807         struct sw_flow *flow;
808         struct hlist_node *n;
809         struct hlist_head *head;
810         u32 hash;
811
812         hash = flow_hash(key, key_len);
813
814         head = find_bucket(table, hash);
815         hlist_for_each_entry_rcu(flow, n, head, hash_node) {
816
817                 if (flow->hash == hash &&
818                     !memcmp(&flow->key, key, key_len)) {
819                         return flow;
820                 }
821         }
822         return NULL;
823 }
824
825 void flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
826 {
827         struct hlist_head *head;
828
829         head = find_bucket(table, flow->hash);
830         hlist_add_head_rcu(&flow->hash_node, head);
831         table->count++;
832 }
833
834 void flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
835 {
836         if (!hlist_unhashed(&flow->hash_node)) {
837                 hlist_del_init_rcu(&flow->hash_node);
838                 table->count--;
839                 BUG_ON(table->count < 0);
840         }
841 }
842
843 static int parse_tos_frag(struct sw_flow_key *swkey, u8 tos, u8 frag)
844 {
845         if (tos & INET_ECN_MASK || frag > OVS_FRAG_TYPE_MAX)
846                 return -EINVAL;
847
848         swkey->ip.tos_frag = tos | frag;
849         return 0;
850 }
851
852 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
853 const u32 ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
854         [OVS_KEY_ATTR_PRIORITY] = 4,
855         [OVS_KEY_ATTR_TUN_ID] = 8,
856         [OVS_KEY_ATTR_IN_PORT] = 4,
857         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
858         [OVS_KEY_ATTR_8021Q] = sizeof(struct ovs_key_8021q),
859         [OVS_KEY_ATTR_ETHERTYPE] = 2,
860         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
861         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
862         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
863         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
864         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
865         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
866         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
867         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
868 };
869
870 /**
871  * flow_from_nlattrs - parses Netlink attributes into a flow key.
872  * @swkey: receives the extracted flow key.
873  * @key_lenp: number of bytes used in @swkey.
874  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
875  * sequence.
876  *
877  * This state machine accepts the following forms, with [] for optional
878  * elements and | for alternatives:
879  *
880  * [priority] [tun_id] [in_port] ethernet [8021q] [ethertype \
881  *              [IPv4 [TCP|UDP|ICMP] | IPv6 [TCP|UDP|ICMPv6 [ND]] | ARP]]
882  *
883  * except that IPv4 or IPv6 terminates the sequence if its @ipv4_frag or
884  * @ipv6_frag member, respectively, equals %OVS_FRAG_TYPE_LATER.
885  */
886 int flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
887                       const struct nlattr *attr)
888 {
889         int error = 0;
890         enum ovs_frag_type frag_type;
891         const struct nlattr *nla;
892         u16 prev_type;
893         int rem;
894         int key_len;
895
896         memset(swkey, 0, sizeof(*swkey));
897         swkey->phy.in_port = USHRT_MAX;
898         swkey->eth.type = htons(ETH_P_802_2);
899         key_len = SW_FLOW_KEY_OFFSET(eth);
900
901         prev_type = OVS_KEY_ATTR_UNSPEC;
902         nla_for_each_nested(nla, attr, rem) {
903                 const struct ovs_key_ethernet *eth_key;
904                 const struct ovs_key_8021q *q_key;
905                 const struct ovs_key_ipv4 *ipv4_key;
906                 const struct ovs_key_ipv6 *ipv6_key;
907                 const struct ovs_key_tcp *tcp_key;
908                 const struct ovs_key_udp *udp_key;
909                 const struct ovs_key_icmp *icmp_key;
910                 const struct ovs_key_icmpv6 *icmpv6_key;
911                 const struct ovs_key_arp *arp_key;
912                 const struct ovs_key_nd *nd_key;
913
914                 int type = nla_type(nla);
915
916                 if (type > OVS_KEY_ATTR_MAX || nla_len(nla) != ovs_key_lens[type])
917                         goto invalid;
918
919 #define TRANSITION(PREV_TYPE, TYPE) (((PREV_TYPE) << 16) | (TYPE))
920                 switch (TRANSITION(prev_type, type)) {
921                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_PRIORITY):
922                         swkey->phy.priority = nla_get_u32(nla);
923                         break;
924
925                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_TUN_ID):
926                 case TRANSITION(OVS_KEY_ATTR_PRIORITY, OVS_KEY_ATTR_TUN_ID):
927                         swkey->phy.tun_id = nla_get_be64(nla);
928                         break;
929
930                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_IN_PORT):
931                 case TRANSITION(OVS_KEY_ATTR_PRIORITY, OVS_KEY_ATTR_IN_PORT):
932                 case TRANSITION(OVS_KEY_ATTR_TUN_ID, OVS_KEY_ATTR_IN_PORT):
933                         if (nla_get_u32(nla) >= DP_MAX_PORTS)
934                                 goto invalid;
935                         swkey->phy.in_port = nla_get_u32(nla);
936                         break;
937
938                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_ETHERNET):
939                 case TRANSITION(OVS_KEY_ATTR_PRIORITY, OVS_KEY_ATTR_ETHERNET):
940                 case TRANSITION(OVS_KEY_ATTR_TUN_ID, OVS_KEY_ATTR_ETHERNET):
941                 case TRANSITION(OVS_KEY_ATTR_IN_PORT, OVS_KEY_ATTR_ETHERNET):
942                         eth_key = nla_data(nla);
943                         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
944                         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
945                         break;
946
947                 case TRANSITION(OVS_KEY_ATTR_ETHERNET, OVS_KEY_ATTR_8021Q):
948                         q_key = nla_data(nla);
949                         /* Only standard 0x8100 VLANs currently supported. */
950                         if (q_key->q_tpid != htons(ETH_P_8021Q))
951                                 goto invalid;
952                         if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
953                                 goto invalid;
954                         swkey->eth.tci = q_key->q_tci | htons(VLAN_TAG_PRESENT);
955                         break;
956
957                 case TRANSITION(OVS_KEY_ATTR_8021Q, OVS_KEY_ATTR_ETHERTYPE):
958                 case TRANSITION(OVS_KEY_ATTR_ETHERNET, OVS_KEY_ATTR_ETHERTYPE):
959                         swkey->eth.type = nla_get_be16(nla);
960                         if (ntohs(swkey->eth.type) < 1536)
961                                 goto invalid;
962                         break;
963
964                 case TRANSITION(OVS_KEY_ATTR_ETHERTYPE, OVS_KEY_ATTR_IPV4):
965                         key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
966                         if (swkey->eth.type != htons(ETH_P_IP))
967                                 goto invalid;
968                         ipv4_key = nla_data(nla);
969                         swkey->ip.proto = ipv4_key->ipv4_proto;
970                         if (parse_tos_frag(swkey, ipv4_key->ipv4_tos,
971                                            ipv4_key->ipv4_frag))
972                                 goto invalid;
973                         swkey->ipv4.addr.src = ipv4_key->ipv4_src;
974                         swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
975                         break;
976
977                 case TRANSITION(OVS_KEY_ATTR_ETHERTYPE, OVS_KEY_ATTR_IPV6):
978                         key_len = SW_FLOW_KEY_OFFSET(ipv6.addr);
979                         if (swkey->eth.type != htons(ETH_P_IPV6))
980                                 goto invalid;
981                         ipv6_key = nla_data(nla);
982                         swkey->ip.proto = ipv6_key->ipv6_proto;
983                         if (parse_tos_frag(swkey, ipv6_key->ipv6_tos,
984                                            ipv6_key->ipv6_frag))
985                                 goto invalid;
986                         memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
987                                         sizeof(swkey->ipv6.addr.src));
988                         memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
989                                         sizeof(swkey->ipv6.addr.dst));
990                         break;
991
992                 case TRANSITION(OVS_KEY_ATTR_IPV4, OVS_KEY_ATTR_TCP):
993                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
994                         if (swkey->ip.proto != IPPROTO_TCP)
995                                 goto invalid;
996                         tcp_key = nla_data(nla);
997                         swkey->ipv4.tp.src = tcp_key->tcp_src;
998                         swkey->ipv4.tp.dst = tcp_key->tcp_dst;
999                         break;
1000
1001                 case TRANSITION(OVS_KEY_ATTR_IPV6, OVS_KEY_ATTR_TCP):
1002                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
1003                         if (swkey->ip.proto != IPPROTO_TCP)
1004                                 goto invalid;
1005                         tcp_key = nla_data(nla);
1006                         swkey->ipv6.tp.src = tcp_key->tcp_src;
1007                         swkey->ipv6.tp.dst = tcp_key->tcp_dst;
1008                         break;
1009
1010                 case TRANSITION(OVS_KEY_ATTR_IPV4, OVS_KEY_ATTR_UDP):
1011                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
1012                         if (swkey->ip.proto != IPPROTO_UDP)
1013                                 goto invalid;
1014                         udp_key = nla_data(nla);
1015                         swkey->ipv4.tp.src = udp_key->udp_src;
1016                         swkey->ipv4.tp.dst = udp_key->udp_dst;
1017                         break;
1018
1019                 case TRANSITION(OVS_KEY_ATTR_IPV6, OVS_KEY_ATTR_UDP):
1020                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
1021                         if (swkey->ip.proto != IPPROTO_UDP)
1022                                 goto invalid;
1023                         udp_key = nla_data(nla);
1024                         swkey->ipv6.tp.src = udp_key->udp_src;
1025                         swkey->ipv6.tp.dst = udp_key->udp_dst;
1026                         break;
1027
1028                 case TRANSITION(OVS_KEY_ATTR_IPV4, OVS_KEY_ATTR_ICMP):
1029                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
1030                         if (swkey->ip.proto != IPPROTO_ICMP)
1031                                 goto invalid;
1032                         icmp_key = nla_data(nla);
1033                         swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
1034                         swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
1035                         break;
1036
1037                 case TRANSITION(OVS_KEY_ATTR_IPV6, OVS_KEY_ATTR_ICMPV6):
1038                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
1039                         if (swkey->ip.proto != IPPROTO_ICMPV6)
1040                                 goto invalid;
1041                         icmpv6_key = nla_data(nla);
1042                         swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
1043                         swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
1044                         break;
1045
1046                 case TRANSITION(OVS_KEY_ATTR_ETHERTYPE, OVS_KEY_ATTR_ARP):
1047                         key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1048                         if (swkey->eth.type != htons(ETH_P_ARP))
1049                                 goto invalid;
1050                         arp_key = nla_data(nla);
1051                         swkey->ipv4.addr.src = arp_key->arp_sip;
1052                         swkey->ipv4.addr.dst = arp_key->arp_tip;
1053                         if (arp_key->arp_op & htons(0xff00))
1054                                 goto invalid;
1055                         swkey->ip.proto = ntohs(arp_key->arp_op);
1056                         memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1057                         memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1058                         break;
1059
1060                 case TRANSITION(OVS_KEY_ATTR_ICMPV6, OVS_KEY_ATTR_ND):
1061                         key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
1062                         if (swkey->ipv6.tp.src != htons(NDISC_NEIGHBOUR_SOLICITATION)
1063                             && swkey->ipv6.tp.src != htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
1064                                 goto invalid;
1065                         nd_key = nla_data(nla);
1066                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
1067                                         sizeof(swkey->ipv6.nd.target));
1068                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
1069                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
1070                         break;
1071
1072                 default:
1073                         goto invalid;
1074                 }
1075
1076                 prev_type = type;
1077         }
1078         if (rem)
1079                 goto invalid;
1080
1081         frag_type = swkey->ip.tos_frag & OVS_FRAG_TYPE_MASK;
1082         switch (prev_type) {
1083         case OVS_KEY_ATTR_UNSPEC:
1084                 goto invalid;
1085
1086         case OVS_KEY_ATTR_PRIORITY:
1087         case OVS_KEY_ATTR_TUN_ID:
1088         case OVS_KEY_ATTR_IN_PORT:
1089                 goto invalid;
1090
1091         case OVS_KEY_ATTR_ETHERNET:
1092         case OVS_KEY_ATTR_8021Q:
1093                 goto ok;
1094
1095         case OVS_KEY_ATTR_ETHERTYPE:
1096                 if (swkey->eth.type == htons(ETH_P_IP) ||
1097                     swkey->eth.type == htons(ETH_P_IPV6) ||
1098                     swkey->eth.type == htons(ETH_P_ARP))
1099                         goto invalid;
1100                 goto ok;
1101
1102         case OVS_KEY_ATTR_IPV4:
1103                 if (frag_type == OVS_FRAG_TYPE_LATER)
1104                         goto ok;
1105                 if (swkey->ip.proto == IPPROTO_TCP ||
1106                     swkey->ip.proto == IPPROTO_UDP ||
1107                     swkey->ip.proto == IPPROTO_ICMP)
1108                         goto invalid;
1109                 goto ok;
1110
1111         case OVS_KEY_ATTR_IPV6:
1112                 if (frag_type == OVS_FRAG_TYPE_LATER)
1113                         goto ok;
1114                 if (swkey->ip.proto == IPPROTO_TCP ||
1115                     swkey->ip.proto == IPPROTO_UDP ||
1116                     swkey->ip.proto == IPPROTO_ICMPV6)
1117                         goto invalid;
1118                 goto ok;
1119
1120         case OVS_KEY_ATTR_ICMPV6:
1121                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
1122                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT) ||
1123                     frag_type == OVS_FRAG_TYPE_LATER)
1124                         goto invalid;
1125                 goto ok;
1126
1127         case OVS_KEY_ATTR_TCP:
1128         case OVS_KEY_ATTR_UDP:
1129         case OVS_KEY_ATTR_ICMP:
1130         case OVS_KEY_ATTR_ND:
1131                 if (frag_type == OVS_FRAG_TYPE_LATER)
1132                         goto invalid;
1133                 goto ok;
1134
1135         case OVS_KEY_ATTR_ARP:
1136                 goto ok;
1137
1138         default:
1139                 WARN_ON_ONCE(1);
1140         }
1141
1142 invalid:
1143         error = -EINVAL;
1144
1145 ok:
1146         *key_lenp = key_len;
1147         return error;
1148 }
1149
1150 /**
1151  * flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1152  * @in_port: receives the extracted input port.
1153  * @tun_id: receives the extracted tunnel ID.
1154  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1155  * sequence.
1156  *
1157  * This parses a series of Netlink attributes that form a flow key, which must
1158  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1159  * get the metadata, that is, the parts of the flow key that cannot be
1160  * extracted from the packet itself.
1161  */
1162 int flow_metadata_from_nlattrs(u32 *priority, u16 *in_port, __be64 *tun_id,
1163                                const struct nlattr *attr)
1164 {
1165         const struct nlattr *nla;
1166         u16 prev_type;
1167         int rem;
1168
1169         *in_port = USHRT_MAX;
1170         *tun_id = 0;
1171         *priority = 0;
1172
1173         prev_type = OVS_KEY_ATTR_UNSPEC;
1174         nla_for_each_nested(nla, attr, rem) {
1175                 int type = nla_type(nla);
1176
1177                 if (type > OVS_KEY_ATTR_MAX || nla_len(nla) != ovs_key_lens[type])
1178                         return -EINVAL;
1179
1180                 switch (TRANSITION(prev_type, type)) {
1181                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_PRIORITY):
1182                         *priority = nla_get_u32(nla);
1183                         break;
1184
1185                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_TUN_ID):
1186                 case TRANSITION(OVS_KEY_ATTR_PRIORITY, OVS_KEY_ATTR_TUN_ID):
1187                         *tun_id = nla_get_be64(nla);
1188                         break;
1189
1190                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_IN_PORT):
1191                 case TRANSITION(OVS_KEY_ATTR_PRIORITY, OVS_KEY_ATTR_IN_PORT):
1192                 case TRANSITION(OVS_KEY_ATTR_TUN_ID, OVS_KEY_ATTR_IN_PORT):
1193                         if (nla_get_u32(nla) >= DP_MAX_PORTS)
1194                                 return -EINVAL;
1195                         *in_port = nla_get_u32(nla);
1196                         break;
1197
1198                 default:
1199                         return 0;
1200                 }
1201
1202                 prev_type = type;
1203         }
1204         if (rem)
1205                 return -EINVAL;
1206         return 0;
1207 }
1208
1209 int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1210 {
1211         struct ovs_key_ethernet *eth_key;
1212         struct nlattr *nla;
1213
1214         /* This is an imperfect sanity-check that FLOW_BUFSIZE doesn't need
1215          * to be updated, but will at least raise awareness when new
1216          * datapath key types are added. */
1217         BUILD_BUG_ON(__OVS_KEY_ATTR_MAX != 15);
1218
1219         if (swkey->phy.priority)
1220                 NLA_PUT_U32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority);
1221
1222         if (swkey->phy.tun_id != cpu_to_be64(0))
1223                 NLA_PUT_BE64(skb, OVS_KEY_ATTR_TUN_ID, swkey->phy.tun_id);
1224
1225         if (swkey->phy.in_port != USHRT_MAX)
1226                 NLA_PUT_U32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port);
1227
1228         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1229         if (!nla)
1230                 goto nla_put_failure;
1231         eth_key = nla_data(nla);
1232         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1233         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1234
1235         if (swkey->eth.tci != htons(0)) {
1236                 struct ovs_key_8021q q_key;
1237
1238                 q_key.q_tpid = htons(ETH_P_8021Q);
1239                 q_key.q_tci = swkey->eth.tci & ~htons(VLAN_TAG_PRESENT);
1240                 NLA_PUT(skb, OVS_KEY_ATTR_8021Q, sizeof(q_key), &q_key);
1241         }
1242
1243         if (swkey->eth.type == htons(ETH_P_802_2))
1244                 return 0;
1245
1246         NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type);
1247
1248         if (swkey->eth.type == htons(ETH_P_IP)) {
1249                 struct ovs_key_ipv4 *ipv4_key;
1250
1251                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1252                 if (!nla)
1253                         goto nla_put_failure;
1254                 ipv4_key = nla_data(nla);
1255                 memset(ipv4_key, 0, sizeof(struct ovs_key_ipv4));
1256                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1257                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1258                 ipv4_key->ipv4_proto = swkey->ip.proto;
1259                 ipv4_key->ipv4_tos = swkey->ip.tos_frag & ~INET_ECN_MASK;
1260                 ipv4_key->ipv4_frag = swkey->ip.tos_frag & OVS_FRAG_TYPE_MASK;
1261         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1262                 struct ovs_key_ipv6 *ipv6_key;
1263
1264                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1265                 if (!nla)
1266                         goto nla_put_failure;
1267                 ipv6_key = nla_data(nla);
1268                 memset(ipv6_key, 0, sizeof(struct ovs_key_ipv6));
1269                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1270                                 sizeof(ipv6_key->ipv6_src));
1271                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1272                                 sizeof(ipv6_key->ipv6_dst));
1273                 ipv6_key->ipv6_proto = swkey->ip.proto;
1274                 ipv6_key->ipv6_tos = swkey->ip.tos_frag & ~INET_ECN_MASK;
1275                 ipv6_key->ipv6_frag = swkey->ip.tos_frag & OVS_FRAG_TYPE_MASK;
1276         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1277                 struct ovs_key_arp *arp_key;
1278
1279                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1280                 if (!nla)
1281                         goto nla_put_failure;
1282                 arp_key = nla_data(nla);
1283                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1284                 arp_key->arp_sip = swkey->ipv4.addr.src;
1285                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1286                 arp_key->arp_op = htons(swkey->ip.proto);
1287                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1288                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1289         }
1290
1291         if ((swkey->eth.type == htons(ETH_P_IP) ||
1292              swkey->eth.type == htons(ETH_P_IPV6)) &&
1293             (swkey->ip.tos_frag & OVS_FRAG_TYPE_MASK) != OVS_FRAG_TYPE_LATER) {
1294
1295                 if (swkey->ip.proto == IPPROTO_TCP) {
1296                         struct ovs_key_tcp *tcp_key;
1297
1298                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1299                         if (!nla)
1300                                 goto nla_put_failure;
1301                         tcp_key = nla_data(nla);
1302                         if (swkey->eth.type == htons(ETH_P_IP)) {
1303                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1304                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1305                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1306                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1307                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1308                         }
1309                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1310                         struct ovs_key_udp *udp_key;
1311
1312                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1313                         if (!nla)
1314                                 goto nla_put_failure;
1315                         udp_key = nla_data(nla);
1316                         if (swkey->eth.type == htons(ETH_P_IP)) {
1317                                 udp_key->udp_src = swkey->ipv4.tp.src;
1318                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1319                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1320                                 udp_key->udp_src = swkey->ipv6.tp.src;
1321                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1322                         }
1323                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1324                            swkey->ip.proto == IPPROTO_ICMP) {
1325                         struct ovs_key_icmp *icmp_key;
1326
1327                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1328                         if (!nla)
1329                                 goto nla_put_failure;
1330                         icmp_key = nla_data(nla);
1331                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1332                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1333                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1334                            swkey->ip.proto == IPPROTO_ICMPV6) {
1335                         struct ovs_key_icmpv6 *icmpv6_key;
1336
1337                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1338                                                 sizeof(*icmpv6_key));
1339                         if (!nla)
1340                                 goto nla_put_failure;
1341                         icmpv6_key = nla_data(nla);
1342                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1343                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1344
1345                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1346                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1347                                 struct ovs_key_nd *nd_key;
1348
1349                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1350                                 if (!nla)
1351                                         goto nla_put_failure;
1352                                 nd_key = nla_data(nla);
1353                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1354                                                         sizeof(nd_key->nd_target));
1355                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1356                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1357                         }
1358                 }
1359         }
1360
1361         return 0;
1362
1363 nla_put_failure:
1364         return -EMSGSIZE;
1365 }
1366
1367 /* Initializes the flow module.
1368  * Returns zero if successful or a negative error code. */
1369 int flow_init(void)
1370 {
1371         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1372                                         0, NULL);
1373         if (flow_cache == NULL)
1374                 return -ENOMEM;
1375
1376         get_random_bytes(&hash_seed, sizeof(hash_seed));
1377
1378         return 0;
1379 }
1380
1381 /* Uninitializes the flow module. */
1382 void flow_exit(void)
1383 {
1384         kmem_cache_destroy(flow_cache);
1385 }