xref: /openbmc/linux/net/openvswitch/flow.c (revision 54cbac81)
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 *node, *n;
303 		int ver = table->node_ver;
304 
305 		hlist_for_each_entry_safe(flow, node, 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 	struct hlist_node *n;
336 	int ver;
337 	int i;
338 
339 	ver = table->node_ver;
340 	while (*bucket < table->n_buckets) {
341 		i = 0;
342 		head = flex_array_get(table->buckets, *bucket);
343 		hlist_for_each_entry_rcu(flow, n, head, hash_node[ver]) {
344 			if (i < *last) {
345 				i++;
346 				continue;
347 			}
348 			*last = i + 1;
349 			return flow;
350 		}
351 		(*bucket)++;
352 		*last = 0;
353 	}
354 
355 	return NULL;
356 }
357 
358 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
359 {
360 	int old_ver;
361 	int i;
362 
363 	old_ver = old->node_ver;
364 	new->node_ver = !old_ver;
365 
366 	/* Insert in new table. */
367 	for (i = 0; i < old->n_buckets; i++) {
368 		struct sw_flow *flow;
369 		struct hlist_head *head;
370 		struct hlist_node *n;
371 
372 		head = flex_array_get(old->buckets, i);
373 
374 		hlist_for_each_entry(flow, n, head, hash_node[old_ver])
375 			ovs_flow_tbl_insert(new, flow);
376 	}
377 	old->keep_flows = true;
378 }
379 
380 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
381 {
382 	struct flow_table *new_table;
383 
384 	new_table = ovs_flow_tbl_alloc(n_buckets);
385 	if (!new_table)
386 		return ERR_PTR(-ENOMEM);
387 
388 	flow_table_copy_flows(table, new_table);
389 
390 	return new_table;
391 }
392 
393 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
394 {
395 	return __flow_tbl_rehash(table, table->n_buckets);
396 }
397 
398 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
399 {
400 	return __flow_tbl_rehash(table, table->n_buckets * 2);
401 }
402 
403 void ovs_flow_free(struct sw_flow *flow)
404 {
405 	if (unlikely(!flow))
406 		return;
407 
408 	kfree((struct sf_flow_acts __force *)flow->sf_acts);
409 	kmem_cache_free(flow_cache, flow);
410 }
411 
412 /* RCU callback used by ovs_flow_deferred_free. */
413 static void rcu_free_flow_callback(struct rcu_head *rcu)
414 {
415 	struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
416 
417 	ovs_flow_free(flow);
418 }
419 
420 /* Schedules 'flow' to be freed after the next RCU grace period.
421  * The caller must hold rcu_read_lock for this to be sensible. */
422 void ovs_flow_deferred_free(struct sw_flow *flow)
423 {
424 	call_rcu(&flow->rcu, rcu_free_flow_callback);
425 }
426 
427 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
428  * The caller must hold rcu_read_lock for this to be sensible. */
429 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
430 {
431 	kfree_rcu(sf_acts, rcu);
432 }
433 
434 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
435 {
436 	struct qtag_prefix {
437 		__be16 eth_type; /* ETH_P_8021Q */
438 		__be16 tci;
439 	};
440 	struct qtag_prefix *qp;
441 
442 	if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
443 		return 0;
444 
445 	if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
446 					 sizeof(__be16))))
447 		return -ENOMEM;
448 
449 	qp = (struct qtag_prefix *) skb->data;
450 	key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
451 	__skb_pull(skb, sizeof(struct qtag_prefix));
452 
453 	return 0;
454 }
455 
456 static __be16 parse_ethertype(struct sk_buff *skb)
457 {
458 	struct llc_snap_hdr {
459 		u8  dsap;  /* Always 0xAA */
460 		u8  ssap;  /* Always 0xAA */
461 		u8  ctrl;
462 		u8  oui[3];
463 		__be16 ethertype;
464 	};
465 	struct llc_snap_hdr *llc;
466 	__be16 proto;
467 
468 	proto = *(__be16 *) skb->data;
469 	__skb_pull(skb, sizeof(__be16));
470 
471 	if (ntohs(proto) >= 1536)
472 		return proto;
473 
474 	if (skb->len < sizeof(struct llc_snap_hdr))
475 		return htons(ETH_P_802_2);
476 
477 	if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
478 		return htons(0);
479 
480 	llc = (struct llc_snap_hdr *) skb->data;
481 	if (llc->dsap != LLC_SAP_SNAP ||
482 	    llc->ssap != LLC_SAP_SNAP ||
483 	    (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
484 		return htons(ETH_P_802_2);
485 
486 	__skb_pull(skb, sizeof(struct llc_snap_hdr));
487 	return llc->ethertype;
488 }
489 
490 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
491 			int *key_lenp, int nh_len)
492 {
493 	struct icmp6hdr *icmp = icmp6_hdr(skb);
494 	int error = 0;
495 	int key_len;
496 
497 	/* The ICMPv6 type and code fields use the 16-bit transport port
498 	 * fields, so we need to store them in 16-bit network byte order.
499 	 */
500 	key->ipv6.tp.src = htons(icmp->icmp6_type);
501 	key->ipv6.tp.dst = htons(icmp->icmp6_code);
502 	key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
503 
504 	if (icmp->icmp6_code == 0 &&
505 	    (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
506 	     icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
507 		int icmp_len = skb->len - skb_transport_offset(skb);
508 		struct nd_msg *nd;
509 		int offset;
510 
511 		key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
512 
513 		/* In order to process neighbor discovery options, we need the
514 		 * entire packet.
515 		 */
516 		if (unlikely(icmp_len < sizeof(*nd)))
517 			goto out;
518 		if (unlikely(skb_linearize(skb))) {
519 			error = -ENOMEM;
520 			goto out;
521 		}
522 
523 		nd = (struct nd_msg *)skb_transport_header(skb);
524 		key->ipv6.nd.target = nd->target;
525 		key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
526 
527 		icmp_len -= sizeof(*nd);
528 		offset = 0;
529 		while (icmp_len >= 8) {
530 			struct nd_opt_hdr *nd_opt =
531 				 (struct nd_opt_hdr *)(nd->opt + offset);
532 			int opt_len = nd_opt->nd_opt_len * 8;
533 
534 			if (unlikely(!opt_len || opt_len > icmp_len))
535 				goto invalid;
536 
537 			/* Store the link layer address if the appropriate
538 			 * option is provided.  It is considered an error if
539 			 * the same link layer option is specified twice.
540 			 */
541 			if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
542 			    && opt_len == 8) {
543 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
544 					goto invalid;
545 				memcpy(key->ipv6.nd.sll,
546 				    &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
547 			} else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
548 				   && opt_len == 8) {
549 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
550 					goto invalid;
551 				memcpy(key->ipv6.nd.tll,
552 				    &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
553 			}
554 
555 			icmp_len -= opt_len;
556 			offset += opt_len;
557 		}
558 	}
559 
560 	goto out;
561 
562 invalid:
563 	memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
564 	memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
565 	memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
566 
567 out:
568 	*key_lenp = key_len;
569 	return error;
570 }
571 
572 /**
573  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
574  * @skb: sk_buff that contains the frame, with skb->data pointing to the
575  * Ethernet header
576  * @in_port: port number on which @skb was received.
577  * @key: output flow key
578  * @key_lenp: length of output flow key
579  *
580  * The caller must ensure that skb->len >= ETH_HLEN.
581  *
582  * Returns 0 if successful, otherwise a negative errno value.
583  *
584  * Initializes @skb header pointers as follows:
585  *
586  *    - skb->mac_header: the Ethernet header.
587  *
588  *    - skb->network_header: just past the Ethernet header, or just past the
589  *      VLAN header, to the first byte of the Ethernet payload.
590  *
591  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
592  *      on output, then just past the IP header, if one is present and
593  *      of a correct length, otherwise the same as skb->network_header.
594  *      For other key->dl_type values it is left untouched.
595  */
596 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
597 		 int *key_lenp)
598 {
599 	int error = 0;
600 	int key_len = SW_FLOW_KEY_OFFSET(eth);
601 	struct ethhdr *eth;
602 
603 	memset(key, 0, sizeof(*key));
604 
605 	key->phy.priority = skb->priority;
606 	key->phy.in_port = in_port;
607 	key->phy.skb_mark = skb->mark;
608 
609 	skb_reset_mac_header(skb);
610 
611 	/* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
612 	 * header in the linear data area.
613 	 */
614 	eth = eth_hdr(skb);
615 	memcpy(key->eth.src, eth->h_source, ETH_ALEN);
616 	memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
617 
618 	__skb_pull(skb, 2 * ETH_ALEN);
619 
620 	if (vlan_tx_tag_present(skb))
621 		key->eth.tci = htons(skb->vlan_tci);
622 	else if (eth->h_proto == htons(ETH_P_8021Q))
623 		if (unlikely(parse_vlan(skb, key)))
624 			return -ENOMEM;
625 
626 	key->eth.type = parse_ethertype(skb);
627 	if (unlikely(key->eth.type == htons(0)))
628 		return -ENOMEM;
629 
630 	skb_reset_network_header(skb);
631 	__skb_push(skb, skb->data - skb_mac_header(skb));
632 
633 	/* Network layer. */
634 	if (key->eth.type == htons(ETH_P_IP)) {
635 		struct iphdr *nh;
636 		__be16 offset;
637 
638 		key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
639 
640 		error = check_iphdr(skb);
641 		if (unlikely(error)) {
642 			if (error == -EINVAL) {
643 				skb->transport_header = skb->network_header;
644 				error = 0;
645 			}
646 			goto out;
647 		}
648 
649 		nh = ip_hdr(skb);
650 		key->ipv4.addr.src = nh->saddr;
651 		key->ipv4.addr.dst = nh->daddr;
652 
653 		key->ip.proto = nh->protocol;
654 		key->ip.tos = nh->tos;
655 		key->ip.ttl = nh->ttl;
656 
657 		offset = nh->frag_off & htons(IP_OFFSET);
658 		if (offset) {
659 			key->ip.frag = OVS_FRAG_TYPE_LATER;
660 			goto out;
661 		}
662 		if (nh->frag_off & htons(IP_MF) ||
663 			 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
664 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
665 
666 		/* Transport layer. */
667 		if (key->ip.proto == IPPROTO_TCP) {
668 			key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
669 			if (tcphdr_ok(skb)) {
670 				struct tcphdr *tcp = tcp_hdr(skb);
671 				key->ipv4.tp.src = tcp->source;
672 				key->ipv4.tp.dst = tcp->dest;
673 			}
674 		} else if (key->ip.proto == IPPROTO_UDP) {
675 			key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
676 			if (udphdr_ok(skb)) {
677 				struct udphdr *udp = udp_hdr(skb);
678 				key->ipv4.tp.src = udp->source;
679 				key->ipv4.tp.dst = udp->dest;
680 			}
681 		} else if (key->ip.proto == IPPROTO_ICMP) {
682 			key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
683 			if (icmphdr_ok(skb)) {
684 				struct icmphdr *icmp = icmp_hdr(skb);
685 				/* The ICMP type and code fields use the 16-bit
686 				 * transport port fields, so we need to store
687 				 * them in 16-bit network byte order. */
688 				key->ipv4.tp.src = htons(icmp->type);
689 				key->ipv4.tp.dst = htons(icmp->code);
690 			}
691 		}
692 
693 	} else if ((key->eth.type == htons(ETH_P_ARP) ||
694 		   key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
695 		struct arp_eth_header *arp;
696 
697 		arp = (struct arp_eth_header *)skb_network_header(skb);
698 
699 		if (arp->ar_hrd == htons(ARPHRD_ETHER)
700 				&& arp->ar_pro == htons(ETH_P_IP)
701 				&& arp->ar_hln == ETH_ALEN
702 				&& arp->ar_pln == 4) {
703 
704 			/* We only match on the lower 8 bits of the opcode. */
705 			if (ntohs(arp->ar_op) <= 0xff)
706 				key->ip.proto = ntohs(arp->ar_op);
707 			memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
708 			memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
709 			memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
710 			memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
711 			key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
712 		}
713 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
714 		int nh_len;             /* IPv6 Header + Extensions */
715 
716 		nh_len = parse_ipv6hdr(skb, key, &key_len);
717 		if (unlikely(nh_len < 0)) {
718 			if (nh_len == -EINVAL)
719 				skb->transport_header = skb->network_header;
720 			else
721 				error = nh_len;
722 			goto out;
723 		}
724 
725 		if (key->ip.frag == OVS_FRAG_TYPE_LATER)
726 			goto out;
727 		if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
728 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
729 
730 		/* Transport layer. */
731 		if (key->ip.proto == NEXTHDR_TCP) {
732 			key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
733 			if (tcphdr_ok(skb)) {
734 				struct tcphdr *tcp = tcp_hdr(skb);
735 				key->ipv6.tp.src = tcp->source;
736 				key->ipv6.tp.dst = tcp->dest;
737 			}
738 		} else if (key->ip.proto == NEXTHDR_UDP) {
739 			key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
740 			if (udphdr_ok(skb)) {
741 				struct udphdr *udp = udp_hdr(skb);
742 				key->ipv6.tp.src = udp->source;
743 				key->ipv6.tp.dst = udp->dest;
744 			}
745 		} else if (key->ip.proto == NEXTHDR_ICMP) {
746 			key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
747 			if (icmp6hdr_ok(skb)) {
748 				error = parse_icmpv6(skb, key, &key_len, nh_len);
749 				if (error < 0)
750 					goto out;
751 			}
752 		}
753 	}
754 
755 out:
756 	*key_lenp = key_len;
757 	return error;
758 }
759 
760 u32 ovs_flow_hash(const struct sw_flow_key *key, int key_len)
761 {
762 	return jhash2((u32 *)key, DIV_ROUND_UP(key_len, sizeof(u32)), 0);
763 }
764 
765 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
766 				struct sw_flow_key *key, int key_len)
767 {
768 	struct sw_flow *flow;
769 	struct hlist_node *n;
770 	struct hlist_head *head;
771 	u32 hash;
772 
773 	hash = ovs_flow_hash(key, key_len);
774 
775 	head = find_bucket(table, hash);
776 	hlist_for_each_entry_rcu(flow, n, head, hash_node[table->node_ver]) {
777 
778 		if (flow->hash == hash &&
779 		    !memcmp(&flow->key, key, key_len)) {
780 			return flow;
781 		}
782 	}
783 	return NULL;
784 }
785 
786 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
787 {
788 	struct hlist_head *head;
789 
790 	head = find_bucket(table, flow->hash);
791 	hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
792 	table->count++;
793 }
794 
795 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
796 {
797 	hlist_del_rcu(&flow->hash_node[table->node_ver]);
798 	table->count--;
799 	BUG_ON(table->count < 0);
800 }
801 
802 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
803 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
804 	[OVS_KEY_ATTR_ENCAP] = -1,
805 	[OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
806 	[OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
807 	[OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
808 	[OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
809 	[OVS_KEY_ATTR_VLAN] = sizeof(__be16),
810 	[OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
811 	[OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
812 	[OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
813 	[OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
814 	[OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
815 	[OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
816 	[OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
817 	[OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
818 	[OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
819 };
820 
821 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
822 				  const struct nlattr *a[], u32 *attrs)
823 {
824 	const struct ovs_key_icmp *icmp_key;
825 	const struct ovs_key_tcp *tcp_key;
826 	const struct ovs_key_udp *udp_key;
827 
828 	switch (swkey->ip.proto) {
829 	case IPPROTO_TCP:
830 		if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
831 			return -EINVAL;
832 		*attrs &= ~(1 << OVS_KEY_ATTR_TCP);
833 
834 		*key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
835 		tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
836 		swkey->ipv4.tp.src = tcp_key->tcp_src;
837 		swkey->ipv4.tp.dst = tcp_key->tcp_dst;
838 		break;
839 
840 	case IPPROTO_UDP:
841 		if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
842 			return -EINVAL;
843 		*attrs &= ~(1 << OVS_KEY_ATTR_UDP);
844 
845 		*key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
846 		udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
847 		swkey->ipv4.tp.src = udp_key->udp_src;
848 		swkey->ipv4.tp.dst = udp_key->udp_dst;
849 		break;
850 
851 	case IPPROTO_ICMP:
852 		if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
853 			return -EINVAL;
854 		*attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
855 
856 		*key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
857 		icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
858 		swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
859 		swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
860 		break;
861 	}
862 
863 	return 0;
864 }
865 
866 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
867 				  const struct nlattr *a[], u32 *attrs)
868 {
869 	const struct ovs_key_icmpv6 *icmpv6_key;
870 	const struct ovs_key_tcp *tcp_key;
871 	const struct ovs_key_udp *udp_key;
872 
873 	switch (swkey->ip.proto) {
874 	case IPPROTO_TCP:
875 		if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
876 			return -EINVAL;
877 		*attrs &= ~(1 << OVS_KEY_ATTR_TCP);
878 
879 		*key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
880 		tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
881 		swkey->ipv6.tp.src = tcp_key->tcp_src;
882 		swkey->ipv6.tp.dst = tcp_key->tcp_dst;
883 		break;
884 
885 	case IPPROTO_UDP:
886 		if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
887 			return -EINVAL;
888 		*attrs &= ~(1 << OVS_KEY_ATTR_UDP);
889 
890 		*key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
891 		udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
892 		swkey->ipv6.tp.src = udp_key->udp_src;
893 		swkey->ipv6.tp.dst = udp_key->udp_dst;
894 		break;
895 
896 	case IPPROTO_ICMPV6:
897 		if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
898 			return -EINVAL;
899 		*attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
900 
901 		*key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
902 		icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
903 		swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
904 		swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
905 
906 		if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
907 		    swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
908 			const struct ovs_key_nd *nd_key;
909 
910 			if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
911 				return -EINVAL;
912 			*attrs &= ~(1 << OVS_KEY_ATTR_ND);
913 
914 			*key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
915 			nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
916 			memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
917 			       sizeof(swkey->ipv6.nd.target));
918 			memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
919 			memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
920 		}
921 		break;
922 	}
923 
924 	return 0;
925 }
926 
927 static int parse_flow_nlattrs(const struct nlattr *attr,
928 			      const struct nlattr *a[], u32 *attrsp)
929 {
930 	const struct nlattr *nla;
931 	u32 attrs;
932 	int rem;
933 
934 	attrs = 0;
935 	nla_for_each_nested(nla, attr, rem) {
936 		u16 type = nla_type(nla);
937 		int expected_len;
938 
939 		if (type > OVS_KEY_ATTR_MAX || attrs & (1 << type))
940 			return -EINVAL;
941 
942 		expected_len = ovs_key_lens[type];
943 		if (nla_len(nla) != expected_len && expected_len != -1)
944 			return -EINVAL;
945 
946 		attrs |= 1 << type;
947 		a[type] = nla;
948 	}
949 	if (rem)
950 		return -EINVAL;
951 
952 	*attrsp = attrs;
953 	return 0;
954 }
955 
956 /**
957  * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
958  * @swkey: receives the extracted flow key.
959  * @key_lenp: number of bytes used in @swkey.
960  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
961  * sequence.
962  */
963 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
964 		      const struct nlattr *attr)
965 {
966 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
967 	const struct ovs_key_ethernet *eth_key;
968 	int key_len;
969 	u32 attrs;
970 	int err;
971 
972 	memset(swkey, 0, sizeof(struct sw_flow_key));
973 	key_len = SW_FLOW_KEY_OFFSET(eth);
974 
975 	err = parse_flow_nlattrs(attr, a, &attrs);
976 	if (err)
977 		return err;
978 
979 	/* Metadata attributes. */
980 	if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
981 		swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
982 		attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
983 	}
984 	if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
985 		u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
986 		if (in_port >= DP_MAX_PORTS)
987 			return -EINVAL;
988 		swkey->phy.in_port = in_port;
989 		attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
990 	} else {
991 		swkey->phy.in_port = DP_MAX_PORTS;
992 	}
993 	if (attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
994 		swkey->phy.skb_mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
995 		attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
996 	}
997 
998 	/* Data attributes. */
999 	if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1000 		return -EINVAL;
1001 	attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1002 
1003 	eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1004 	memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1005 	memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1006 
1007 	if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1008 	    nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1009 		const struct nlattr *encap;
1010 		__be16 tci;
1011 
1012 		if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1013 			      (1 << OVS_KEY_ATTR_ETHERTYPE) |
1014 			      (1 << OVS_KEY_ATTR_ENCAP)))
1015 			return -EINVAL;
1016 
1017 		encap = a[OVS_KEY_ATTR_ENCAP];
1018 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1019 		if (tci & htons(VLAN_TAG_PRESENT)) {
1020 			swkey->eth.tci = tci;
1021 
1022 			err = parse_flow_nlattrs(encap, a, &attrs);
1023 			if (err)
1024 				return err;
1025 		} else if (!tci) {
1026 			/* Corner case for truncated 802.1Q header. */
1027 			if (nla_len(encap))
1028 				return -EINVAL;
1029 
1030 			swkey->eth.type = htons(ETH_P_8021Q);
1031 			*key_lenp = key_len;
1032 			return 0;
1033 		} else {
1034 			return -EINVAL;
1035 		}
1036 	}
1037 
1038 	if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1039 		swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1040 		if (ntohs(swkey->eth.type) < 1536)
1041 			return -EINVAL;
1042 		attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1043 	} else {
1044 		swkey->eth.type = htons(ETH_P_802_2);
1045 	}
1046 
1047 	if (swkey->eth.type == htons(ETH_P_IP)) {
1048 		const struct ovs_key_ipv4 *ipv4_key;
1049 
1050 		if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1051 			return -EINVAL;
1052 		attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1053 
1054 		key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1055 		ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1056 		if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1057 			return -EINVAL;
1058 		swkey->ip.proto = ipv4_key->ipv4_proto;
1059 		swkey->ip.tos = ipv4_key->ipv4_tos;
1060 		swkey->ip.ttl = ipv4_key->ipv4_ttl;
1061 		swkey->ip.frag = ipv4_key->ipv4_frag;
1062 		swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1063 		swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1064 
1065 		if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1066 			err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1067 			if (err)
1068 				return err;
1069 		}
1070 	} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1071 		const struct ovs_key_ipv6 *ipv6_key;
1072 
1073 		if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1074 			return -EINVAL;
1075 		attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1076 
1077 		key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1078 		ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1079 		if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1080 			return -EINVAL;
1081 		swkey->ipv6.label = ipv6_key->ipv6_label;
1082 		swkey->ip.proto = ipv6_key->ipv6_proto;
1083 		swkey->ip.tos = ipv6_key->ipv6_tclass;
1084 		swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1085 		swkey->ip.frag = ipv6_key->ipv6_frag;
1086 		memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1087 		       sizeof(swkey->ipv6.addr.src));
1088 		memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1089 		       sizeof(swkey->ipv6.addr.dst));
1090 
1091 		if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1092 			err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1093 			if (err)
1094 				return err;
1095 		}
1096 	} else if (swkey->eth.type == htons(ETH_P_ARP) ||
1097 		   swkey->eth.type == htons(ETH_P_RARP)) {
1098 		const struct ovs_key_arp *arp_key;
1099 
1100 		if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1101 			return -EINVAL;
1102 		attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1103 
1104 		key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1105 		arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1106 		swkey->ipv4.addr.src = arp_key->arp_sip;
1107 		swkey->ipv4.addr.dst = arp_key->arp_tip;
1108 		if (arp_key->arp_op & htons(0xff00))
1109 			return -EINVAL;
1110 		swkey->ip.proto = ntohs(arp_key->arp_op);
1111 		memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1112 		memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1113 	}
1114 
1115 	if (attrs)
1116 		return -EINVAL;
1117 	*key_lenp = key_len;
1118 
1119 	return 0;
1120 }
1121 
1122 /**
1123  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1124  * @priority: receives the skb priority
1125  * @mark: receives the skb mark
1126  * @in_port: receives the extracted input port.
1127  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1128  * sequence.
1129  *
1130  * This parses a series of Netlink attributes that form a flow key, which must
1131  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1132  * get the metadata, that is, the parts of the flow key that cannot be
1133  * extracted from the packet itself.
1134  */
1135 int ovs_flow_metadata_from_nlattrs(u32 *priority, u32 *mark, u16 *in_port,
1136 			       const struct nlattr *attr)
1137 {
1138 	const struct nlattr *nla;
1139 	int rem;
1140 
1141 	*in_port = DP_MAX_PORTS;
1142 	*priority = 0;
1143 	*mark = 0;
1144 
1145 	nla_for_each_nested(nla, attr, rem) {
1146 		int type = nla_type(nla);
1147 
1148 		if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1149 			if (nla_len(nla) != ovs_key_lens[type])
1150 				return -EINVAL;
1151 
1152 			switch (type) {
1153 			case OVS_KEY_ATTR_PRIORITY:
1154 				*priority = nla_get_u32(nla);
1155 				break;
1156 
1157 			case OVS_KEY_ATTR_IN_PORT:
1158 				if (nla_get_u32(nla) >= DP_MAX_PORTS)
1159 					return -EINVAL;
1160 				*in_port = nla_get_u32(nla);
1161 				break;
1162 
1163 			case OVS_KEY_ATTR_SKB_MARK:
1164 				*mark = nla_get_u32(nla);
1165 				break;
1166 			}
1167 		}
1168 	}
1169 	if (rem)
1170 		return -EINVAL;
1171 	return 0;
1172 }
1173 
1174 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1175 {
1176 	struct ovs_key_ethernet *eth_key;
1177 	struct nlattr *nla, *encap;
1178 
1179 	if (swkey->phy.priority &&
1180 	    nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority))
1181 		goto nla_put_failure;
1182 
1183 	if (swkey->phy.in_port != DP_MAX_PORTS &&
1184 	    nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port))
1185 		goto nla_put_failure;
1186 
1187 	if (swkey->phy.skb_mark &&
1188 	    nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, swkey->phy.skb_mark))
1189 		goto nla_put_failure;
1190 
1191 	nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1192 	if (!nla)
1193 		goto nla_put_failure;
1194 	eth_key = nla_data(nla);
1195 	memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1196 	memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1197 
1198 	if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1199 		if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q)) ||
1200 		    nla_put_be16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci))
1201 			goto nla_put_failure;
1202 		encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1203 		if (!swkey->eth.tci)
1204 			goto unencap;
1205 	} else {
1206 		encap = NULL;
1207 	}
1208 
1209 	if (swkey->eth.type == htons(ETH_P_802_2))
1210 		goto unencap;
1211 
1212 	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type))
1213 		goto nla_put_failure;
1214 
1215 	if (swkey->eth.type == htons(ETH_P_IP)) {
1216 		struct ovs_key_ipv4 *ipv4_key;
1217 
1218 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1219 		if (!nla)
1220 			goto nla_put_failure;
1221 		ipv4_key = nla_data(nla);
1222 		ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1223 		ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1224 		ipv4_key->ipv4_proto = swkey->ip.proto;
1225 		ipv4_key->ipv4_tos = swkey->ip.tos;
1226 		ipv4_key->ipv4_ttl = swkey->ip.ttl;
1227 		ipv4_key->ipv4_frag = swkey->ip.frag;
1228 	} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1229 		struct ovs_key_ipv6 *ipv6_key;
1230 
1231 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1232 		if (!nla)
1233 			goto nla_put_failure;
1234 		ipv6_key = nla_data(nla);
1235 		memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1236 				sizeof(ipv6_key->ipv6_src));
1237 		memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1238 				sizeof(ipv6_key->ipv6_dst));
1239 		ipv6_key->ipv6_label = swkey->ipv6.label;
1240 		ipv6_key->ipv6_proto = swkey->ip.proto;
1241 		ipv6_key->ipv6_tclass = swkey->ip.tos;
1242 		ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1243 		ipv6_key->ipv6_frag = swkey->ip.frag;
1244 	} else if (swkey->eth.type == htons(ETH_P_ARP) ||
1245 		   swkey->eth.type == htons(ETH_P_RARP)) {
1246 		struct ovs_key_arp *arp_key;
1247 
1248 		nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1249 		if (!nla)
1250 			goto nla_put_failure;
1251 		arp_key = nla_data(nla);
1252 		memset(arp_key, 0, sizeof(struct ovs_key_arp));
1253 		arp_key->arp_sip = swkey->ipv4.addr.src;
1254 		arp_key->arp_tip = swkey->ipv4.addr.dst;
1255 		arp_key->arp_op = htons(swkey->ip.proto);
1256 		memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1257 		memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1258 	}
1259 
1260 	if ((swkey->eth.type == htons(ETH_P_IP) ||
1261 	     swkey->eth.type == htons(ETH_P_IPV6)) &&
1262 	     swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1263 
1264 		if (swkey->ip.proto == IPPROTO_TCP) {
1265 			struct ovs_key_tcp *tcp_key;
1266 
1267 			nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1268 			if (!nla)
1269 				goto nla_put_failure;
1270 			tcp_key = nla_data(nla);
1271 			if (swkey->eth.type == htons(ETH_P_IP)) {
1272 				tcp_key->tcp_src = swkey->ipv4.tp.src;
1273 				tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1274 			} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1275 				tcp_key->tcp_src = swkey->ipv6.tp.src;
1276 				tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1277 			}
1278 		} else if (swkey->ip.proto == IPPROTO_UDP) {
1279 			struct ovs_key_udp *udp_key;
1280 
1281 			nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1282 			if (!nla)
1283 				goto nla_put_failure;
1284 			udp_key = nla_data(nla);
1285 			if (swkey->eth.type == htons(ETH_P_IP)) {
1286 				udp_key->udp_src = swkey->ipv4.tp.src;
1287 				udp_key->udp_dst = swkey->ipv4.tp.dst;
1288 			} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1289 				udp_key->udp_src = swkey->ipv6.tp.src;
1290 				udp_key->udp_dst = swkey->ipv6.tp.dst;
1291 			}
1292 		} else if (swkey->eth.type == htons(ETH_P_IP) &&
1293 			   swkey->ip.proto == IPPROTO_ICMP) {
1294 			struct ovs_key_icmp *icmp_key;
1295 
1296 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1297 			if (!nla)
1298 				goto nla_put_failure;
1299 			icmp_key = nla_data(nla);
1300 			icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1301 			icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1302 		} else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1303 			   swkey->ip.proto == IPPROTO_ICMPV6) {
1304 			struct ovs_key_icmpv6 *icmpv6_key;
1305 
1306 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1307 						sizeof(*icmpv6_key));
1308 			if (!nla)
1309 				goto nla_put_failure;
1310 			icmpv6_key = nla_data(nla);
1311 			icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1312 			icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1313 
1314 			if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1315 			    icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1316 				struct ovs_key_nd *nd_key;
1317 
1318 				nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1319 				if (!nla)
1320 					goto nla_put_failure;
1321 				nd_key = nla_data(nla);
1322 				memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1323 							sizeof(nd_key->nd_target));
1324 				memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1325 				memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1326 			}
1327 		}
1328 	}
1329 
1330 unencap:
1331 	if (encap)
1332 		nla_nest_end(skb, encap);
1333 
1334 	return 0;
1335 
1336 nla_put_failure:
1337 	return -EMSGSIZE;
1338 }
1339 
1340 /* Initializes the flow module.
1341  * Returns zero if successful or a negative error code. */
1342 int ovs_flow_init(void)
1343 {
1344 	flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1345 					0, NULL);
1346 	if (flow_cache == NULL)
1347 		return -ENOMEM;
1348 
1349 	return 0;
1350 }
1351 
1352 /* Uninitializes the flow module. */
1353 void ovs_flow_exit(void)
1354 {
1355 	kmem_cache_destroy(flow_cache);
1356 }
1357