xref: /linux/net/openvswitch/flow.c (revision c4c11dd160a8cc98f402c4e12f94b1572e822ffd)
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/ip_tunnels.h>
44 #include <net/ipv6.h>
45 #include <net/ndisc.h>
46 
47 static struct kmem_cache *flow_cache;
48 
49 static int check_header(struct sk_buff *skb, int len)
50 {
51 	if (unlikely(skb->len < len))
52 		return -EINVAL;
53 	if (unlikely(!pskb_may_pull(skb, len)))
54 		return -ENOMEM;
55 	return 0;
56 }
57 
58 static bool arphdr_ok(struct sk_buff *skb)
59 {
60 	return pskb_may_pull(skb, skb_network_offset(skb) +
61 				  sizeof(struct arp_eth_header));
62 }
63 
64 static int check_iphdr(struct sk_buff *skb)
65 {
66 	unsigned int nh_ofs = skb_network_offset(skb);
67 	unsigned int ip_len;
68 	int err;
69 
70 	err = check_header(skb, nh_ofs + sizeof(struct iphdr));
71 	if (unlikely(err))
72 		return err;
73 
74 	ip_len = ip_hdrlen(skb);
75 	if (unlikely(ip_len < sizeof(struct iphdr) ||
76 		     skb->len < nh_ofs + ip_len))
77 		return -EINVAL;
78 
79 	skb_set_transport_header(skb, nh_ofs + ip_len);
80 	return 0;
81 }
82 
83 static bool tcphdr_ok(struct sk_buff *skb)
84 {
85 	int th_ofs = skb_transport_offset(skb);
86 	int tcp_len;
87 
88 	if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
89 		return false;
90 
91 	tcp_len = tcp_hdrlen(skb);
92 	if (unlikely(tcp_len < sizeof(struct tcphdr) ||
93 		     skb->len < th_ofs + tcp_len))
94 		return false;
95 
96 	return true;
97 }
98 
99 static bool udphdr_ok(struct sk_buff *skb)
100 {
101 	return pskb_may_pull(skb, skb_transport_offset(skb) +
102 				  sizeof(struct udphdr));
103 }
104 
105 static bool icmphdr_ok(struct sk_buff *skb)
106 {
107 	return pskb_may_pull(skb, skb_transport_offset(skb) +
108 				  sizeof(struct icmphdr));
109 }
110 
111 u64 ovs_flow_used_time(unsigned long flow_jiffies)
112 {
113 	struct timespec cur_ts;
114 	u64 cur_ms, idle_ms;
115 
116 	ktime_get_ts(&cur_ts);
117 	idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
118 	cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
119 		 cur_ts.tv_nsec / NSEC_PER_MSEC;
120 
121 	return cur_ms - idle_ms;
122 }
123 
124 #define SW_FLOW_KEY_OFFSET(field)		\
125 	(offsetof(struct sw_flow_key, field) +	\
126 	 FIELD_SIZEOF(struct sw_flow_key, field))
127 
128 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
129 			 int *key_lenp)
130 {
131 	unsigned int nh_ofs = skb_network_offset(skb);
132 	unsigned int nh_len;
133 	int payload_ofs;
134 	struct ipv6hdr *nh;
135 	uint8_t nexthdr;
136 	__be16 frag_off;
137 	int err;
138 
139 	*key_lenp = SW_FLOW_KEY_OFFSET(ipv6.label);
140 
141 	err = check_header(skb, nh_ofs + sizeof(*nh));
142 	if (unlikely(err))
143 		return err;
144 
145 	nh = ipv6_hdr(skb);
146 	nexthdr = nh->nexthdr;
147 	payload_ofs = (u8 *)(nh + 1) - skb->data;
148 
149 	key->ip.proto = NEXTHDR_NONE;
150 	key->ip.tos = ipv6_get_dsfield(nh);
151 	key->ip.ttl = nh->hop_limit;
152 	key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
153 	key->ipv6.addr.src = nh->saddr;
154 	key->ipv6.addr.dst = nh->daddr;
155 
156 	payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
157 	if (unlikely(payload_ofs < 0))
158 		return -EINVAL;
159 
160 	if (frag_off) {
161 		if (frag_off & htons(~0x7))
162 			key->ip.frag = OVS_FRAG_TYPE_LATER;
163 		else
164 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
165 	}
166 
167 	nh_len = payload_ofs - nh_ofs;
168 	skb_set_transport_header(skb, nh_ofs + nh_len);
169 	key->ip.proto = nexthdr;
170 	return nh_len;
171 }
172 
173 static bool icmp6hdr_ok(struct sk_buff *skb)
174 {
175 	return pskb_may_pull(skb, skb_transport_offset(skb) +
176 				  sizeof(struct icmp6hdr));
177 }
178 
179 #define TCP_FLAGS_OFFSET 13
180 #define TCP_FLAG_MASK 0x3f
181 
182 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
183 {
184 	u8 tcp_flags = 0;
185 
186 	if ((flow->key.eth.type == htons(ETH_P_IP) ||
187 	     flow->key.eth.type == htons(ETH_P_IPV6)) &&
188 	    flow->key.ip.proto == IPPROTO_TCP &&
189 	    likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
190 		u8 *tcp = (u8 *)tcp_hdr(skb);
191 		tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
192 	}
193 
194 	spin_lock(&flow->lock);
195 	flow->used = jiffies;
196 	flow->packet_count++;
197 	flow->byte_count += skb->len;
198 	flow->tcp_flags |= tcp_flags;
199 	spin_unlock(&flow->lock);
200 }
201 
202 struct sw_flow_actions *ovs_flow_actions_alloc(int size)
203 {
204 	struct sw_flow_actions *sfa;
205 
206 	if (size > MAX_ACTIONS_BUFSIZE)
207 		return ERR_PTR(-EINVAL);
208 
209 	sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
210 	if (!sfa)
211 		return ERR_PTR(-ENOMEM);
212 
213 	sfa->actions_len = 0;
214 	return sfa;
215 }
216 
217 struct sw_flow *ovs_flow_alloc(void)
218 {
219 	struct sw_flow *flow;
220 
221 	flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
222 	if (!flow)
223 		return ERR_PTR(-ENOMEM);
224 
225 	spin_lock_init(&flow->lock);
226 	flow->sf_acts = NULL;
227 
228 	return flow;
229 }
230 
231 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
232 {
233 	hash = jhash_1word(hash, table->hash_seed);
234 	return flex_array_get(table->buckets,
235 				(hash & (table->n_buckets - 1)));
236 }
237 
238 static struct flex_array *alloc_buckets(unsigned int n_buckets)
239 {
240 	struct flex_array *buckets;
241 	int i, err;
242 
243 	buckets = flex_array_alloc(sizeof(struct hlist_head *),
244 				   n_buckets, GFP_KERNEL);
245 	if (!buckets)
246 		return NULL;
247 
248 	err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
249 	if (err) {
250 		flex_array_free(buckets);
251 		return NULL;
252 	}
253 
254 	for (i = 0; i < n_buckets; i++)
255 		INIT_HLIST_HEAD((struct hlist_head *)
256 					flex_array_get(buckets, i));
257 
258 	return buckets;
259 }
260 
261 static void free_buckets(struct flex_array *buckets)
262 {
263 	flex_array_free(buckets);
264 }
265 
266 struct flow_table *ovs_flow_tbl_alloc(int new_size)
267 {
268 	struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
269 
270 	if (!table)
271 		return NULL;
272 
273 	table->buckets = alloc_buckets(new_size);
274 
275 	if (!table->buckets) {
276 		kfree(table);
277 		return NULL;
278 	}
279 	table->n_buckets = new_size;
280 	table->count = 0;
281 	table->node_ver = 0;
282 	table->keep_flows = false;
283 	get_random_bytes(&table->hash_seed, sizeof(u32));
284 
285 	return table;
286 }
287 
288 void ovs_flow_tbl_destroy(struct flow_table *table)
289 {
290 	int i;
291 
292 	if (!table)
293 		return;
294 
295 	if (table->keep_flows)
296 		goto skip_flows;
297 
298 	for (i = 0; i < table->n_buckets; i++) {
299 		struct sw_flow *flow;
300 		struct hlist_head *head = flex_array_get(table->buckets, i);
301 		struct hlist_node *n;
302 		int ver = table->node_ver;
303 
304 		hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
305 			hlist_del_rcu(&flow->hash_node[ver]);
306 			ovs_flow_free(flow);
307 		}
308 	}
309 
310 skip_flows:
311 	free_buckets(table->buckets);
312 	kfree(table);
313 }
314 
315 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
316 {
317 	struct flow_table *table = container_of(rcu, struct flow_table, rcu);
318 
319 	ovs_flow_tbl_destroy(table);
320 }
321 
322 void ovs_flow_tbl_deferred_destroy(struct flow_table *table)
323 {
324 	if (!table)
325 		return;
326 
327 	call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
328 }
329 
330 struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
331 {
332 	struct sw_flow *flow;
333 	struct hlist_head *head;
334 	int ver;
335 	int i;
336 
337 	ver = table->node_ver;
338 	while (*bucket < table->n_buckets) {
339 		i = 0;
340 		head = flex_array_get(table->buckets, *bucket);
341 		hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
342 			if (i < *last) {
343 				i++;
344 				continue;
345 			}
346 			*last = i + 1;
347 			return flow;
348 		}
349 		(*bucket)++;
350 		*last = 0;
351 	}
352 
353 	return NULL;
354 }
355 
356 static void __flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
357 {
358 	struct hlist_head *head;
359 	head = find_bucket(table, flow->hash);
360 	hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
361 	table->count++;
362 }
363 
364 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
365 {
366 	int old_ver;
367 	int i;
368 
369 	old_ver = old->node_ver;
370 	new->node_ver = !old_ver;
371 
372 	/* Insert in new table. */
373 	for (i = 0; i < old->n_buckets; i++) {
374 		struct sw_flow *flow;
375 		struct hlist_head *head;
376 
377 		head = flex_array_get(old->buckets, i);
378 
379 		hlist_for_each_entry(flow, head, hash_node[old_ver])
380 			__flow_tbl_insert(new, flow);
381 	}
382 	old->keep_flows = true;
383 }
384 
385 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
386 {
387 	struct flow_table *new_table;
388 
389 	new_table = ovs_flow_tbl_alloc(n_buckets);
390 	if (!new_table)
391 		return ERR_PTR(-ENOMEM);
392 
393 	flow_table_copy_flows(table, new_table);
394 
395 	return new_table;
396 }
397 
398 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
399 {
400 	return __flow_tbl_rehash(table, table->n_buckets);
401 }
402 
403 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
404 {
405 	return __flow_tbl_rehash(table, table->n_buckets * 2);
406 }
407 
408 void ovs_flow_free(struct sw_flow *flow)
409 {
410 	if (unlikely(!flow))
411 		return;
412 
413 	kfree((struct sf_flow_acts __force *)flow->sf_acts);
414 	kmem_cache_free(flow_cache, flow);
415 }
416 
417 /* RCU callback used by ovs_flow_deferred_free. */
418 static void rcu_free_flow_callback(struct rcu_head *rcu)
419 {
420 	struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
421 
422 	ovs_flow_free(flow);
423 }
424 
425 /* Schedules 'flow' to be freed after the next RCU grace period.
426  * The caller must hold rcu_read_lock for this to be sensible. */
427 void ovs_flow_deferred_free(struct sw_flow *flow)
428 {
429 	call_rcu(&flow->rcu, rcu_free_flow_callback);
430 }
431 
432 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
433  * The caller must hold rcu_read_lock for this to be sensible. */
434 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
435 {
436 	kfree_rcu(sf_acts, rcu);
437 }
438 
439 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
440 {
441 	struct qtag_prefix {
442 		__be16 eth_type; /* ETH_P_8021Q */
443 		__be16 tci;
444 	};
445 	struct qtag_prefix *qp;
446 
447 	if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
448 		return 0;
449 
450 	if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
451 					 sizeof(__be16))))
452 		return -ENOMEM;
453 
454 	qp = (struct qtag_prefix *) skb->data;
455 	key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
456 	__skb_pull(skb, sizeof(struct qtag_prefix));
457 
458 	return 0;
459 }
460 
461 static __be16 parse_ethertype(struct sk_buff *skb)
462 {
463 	struct llc_snap_hdr {
464 		u8  dsap;  /* Always 0xAA */
465 		u8  ssap;  /* Always 0xAA */
466 		u8  ctrl;
467 		u8  oui[3];
468 		__be16 ethertype;
469 	};
470 	struct llc_snap_hdr *llc;
471 	__be16 proto;
472 
473 	proto = *(__be16 *) skb->data;
474 	__skb_pull(skb, sizeof(__be16));
475 
476 	if (ntohs(proto) >= ETH_P_802_3_MIN)
477 		return proto;
478 
479 	if (skb->len < sizeof(struct llc_snap_hdr))
480 		return htons(ETH_P_802_2);
481 
482 	if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
483 		return htons(0);
484 
485 	llc = (struct llc_snap_hdr *) skb->data;
486 	if (llc->dsap != LLC_SAP_SNAP ||
487 	    llc->ssap != LLC_SAP_SNAP ||
488 	    (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
489 		return htons(ETH_P_802_2);
490 
491 	__skb_pull(skb, sizeof(struct llc_snap_hdr));
492 
493 	if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
494 		return llc->ethertype;
495 
496 	return htons(ETH_P_802_2);
497 }
498 
499 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
500 			int *key_lenp, int nh_len)
501 {
502 	struct icmp6hdr *icmp = icmp6_hdr(skb);
503 	int error = 0;
504 	int key_len;
505 
506 	/* The ICMPv6 type and code fields use the 16-bit transport port
507 	 * fields, so we need to store them in 16-bit network byte order.
508 	 */
509 	key->ipv6.tp.src = htons(icmp->icmp6_type);
510 	key->ipv6.tp.dst = htons(icmp->icmp6_code);
511 	key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
512 
513 	if (icmp->icmp6_code == 0 &&
514 	    (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
515 	     icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
516 		int icmp_len = skb->len - skb_transport_offset(skb);
517 		struct nd_msg *nd;
518 		int offset;
519 
520 		key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
521 
522 		/* In order to process neighbor discovery options, we need the
523 		 * entire packet.
524 		 */
525 		if (unlikely(icmp_len < sizeof(*nd)))
526 			goto out;
527 		if (unlikely(skb_linearize(skb))) {
528 			error = -ENOMEM;
529 			goto out;
530 		}
531 
532 		nd = (struct nd_msg *)skb_transport_header(skb);
533 		key->ipv6.nd.target = nd->target;
534 		key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
535 
536 		icmp_len -= sizeof(*nd);
537 		offset = 0;
538 		while (icmp_len >= 8) {
539 			struct nd_opt_hdr *nd_opt =
540 				 (struct nd_opt_hdr *)(nd->opt + offset);
541 			int opt_len = nd_opt->nd_opt_len * 8;
542 
543 			if (unlikely(!opt_len || opt_len > icmp_len))
544 				goto invalid;
545 
546 			/* Store the link layer address if the appropriate
547 			 * option is provided.  It is considered an error if
548 			 * the same link layer option is specified twice.
549 			 */
550 			if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
551 			    && opt_len == 8) {
552 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
553 					goto invalid;
554 				memcpy(key->ipv6.nd.sll,
555 				    &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
556 			} else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
557 				   && opt_len == 8) {
558 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
559 					goto invalid;
560 				memcpy(key->ipv6.nd.tll,
561 				    &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
562 			}
563 
564 			icmp_len -= opt_len;
565 			offset += opt_len;
566 		}
567 	}
568 
569 	goto out;
570 
571 invalid:
572 	memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
573 	memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
574 	memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
575 
576 out:
577 	*key_lenp = key_len;
578 	return error;
579 }
580 
581 /**
582  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
583  * @skb: sk_buff that contains the frame, with skb->data pointing to the
584  * Ethernet header
585  * @in_port: port number on which @skb was received.
586  * @key: output flow key
587  * @key_lenp: length of output flow key
588  *
589  * The caller must ensure that skb->len >= ETH_HLEN.
590  *
591  * Returns 0 if successful, otherwise a negative errno value.
592  *
593  * Initializes @skb header pointers as follows:
594  *
595  *    - skb->mac_header: the Ethernet header.
596  *
597  *    - skb->network_header: just past the Ethernet header, or just past the
598  *      VLAN header, to the first byte of the Ethernet payload.
599  *
600  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
601  *      on output, then just past the IP header, if one is present and
602  *      of a correct length, otherwise the same as skb->network_header.
603  *      For other key->eth.type values it is left untouched.
604  */
605 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
606 		 int *key_lenp)
607 {
608 	int error = 0;
609 	int key_len = SW_FLOW_KEY_OFFSET(eth);
610 	struct ethhdr *eth;
611 
612 	memset(key, 0, sizeof(*key));
613 
614 	key->phy.priority = skb->priority;
615 	if (OVS_CB(skb)->tun_key)
616 		memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
617 	key->phy.in_port = in_port;
618 	key->phy.skb_mark = skb->mark;
619 
620 	skb_reset_mac_header(skb);
621 
622 	/* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
623 	 * header in the linear data area.
624 	 */
625 	eth = eth_hdr(skb);
626 	memcpy(key->eth.src, eth->h_source, ETH_ALEN);
627 	memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
628 
629 	__skb_pull(skb, 2 * ETH_ALEN);
630 	/* We are going to push all headers that we pull, so no need to
631 	 * update skb->csum here.
632 	 */
633 
634 	if (vlan_tx_tag_present(skb))
635 		key->eth.tci = htons(skb->vlan_tci);
636 	else if (eth->h_proto == htons(ETH_P_8021Q))
637 		if (unlikely(parse_vlan(skb, key)))
638 			return -ENOMEM;
639 
640 	key->eth.type = parse_ethertype(skb);
641 	if (unlikely(key->eth.type == htons(0)))
642 		return -ENOMEM;
643 
644 	skb_reset_network_header(skb);
645 	__skb_push(skb, skb->data - skb_mac_header(skb));
646 
647 	/* Network layer. */
648 	if (key->eth.type == htons(ETH_P_IP)) {
649 		struct iphdr *nh;
650 		__be16 offset;
651 
652 		key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
653 
654 		error = check_iphdr(skb);
655 		if (unlikely(error)) {
656 			if (error == -EINVAL) {
657 				skb->transport_header = skb->network_header;
658 				error = 0;
659 			}
660 			goto out;
661 		}
662 
663 		nh = ip_hdr(skb);
664 		key->ipv4.addr.src = nh->saddr;
665 		key->ipv4.addr.dst = nh->daddr;
666 
667 		key->ip.proto = nh->protocol;
668 		key->ip.tos = nh->tos;
669 		key->ip.ttl = nh->ttl;
670 
671 		offset = nh->frag_off & htons(IP_OFFSET);
672 		if (offset) {
673 			key->ip.frag = OVS_FRAG_TYPE_LATER;
674 			goto out;
675 		}
676 		if (nh->frag_off & htons(IP_MF) ||
677 			 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
678 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
679 
680 		/* Transport layer. */
681 		if (key->ip.proto == IPPROTO_TCP) {
682 			key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
683 			if (tcphdr_ok(skb)) {
684 				struct tcphdr *tcp = tcp_hdr(skb);
685 				key->ipv4.tp.src = tcp->source;
686 				key->ipv4.tp.dst = tcp->dest;
687 			}
688 		} else if (key->ip.proto == IPPROTO_UDP) {
689 			key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
690 			if (udphdr_ok(skb)) {
691 				struct udphdr *udp = udp_hdr(skb);
692 				key->ipv4.tp.src = udp->source;
693 				key->ipv4.tp.dst = udp->dest;
694 			}
695 		} else if (key->ip.proto == IPPROTO_ICMP) {
696 			key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
697 			if (icmphdr_ok(skb)) {
698 				struct icmphdr *icmp = icmp_hdr(skb);
699 				/* The ICMP type and code fields use the 16-bit
700 				 * transport port fields, so we need to store
701 				 * them in 16-bit network byte order. */
702 				key->ipv4.tp.src = htons(icmp->type);
703 				key->ipv4.tp.dst = htons(icmp->code);
704 			}
705 		}
706 
707 	} else if ((key->eth.type == htons(ETH_P_ARP) ||
708 		   key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
709 		struct arp_eth_header *arp;
710 
711 		arp = (struct arp_eth_header *)skb_network_header(skb);
712 
713 		if (arp->ar_hrd == htons(ARPHRD_ETHER)
714 				&& arp->ar_pro == htons(ETH_P_IP)
715 				&& arp->ar_hln == ETH_ALEN
716 				&& arp->ar_pln == 4) {
717 
718 			/* We only match on the lower 8 bits of the opcode. */
719 			if (ntohs(arp->ar_op) <= 0xff)
720 				key->ip.proto = ntohs(arp->ar_op);
721 			memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
722 			memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
723 			memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
724 			memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
725 			key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
726 		}
727 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
728 		int nh_len;             /* IPv6 Header + Extensions */
729 
730 		nh_len = parse_ipv6hdr(skb, key, &key_len);
731 		if (unlikely(nh_len < 0)) {
732 			if (nh_len == -EINVAL)
733 				skb->transport_header = skb->network_header;
734 			else
735 				error = nh_len;
736 			goto out;
737 		}
738 
739 		if (key->ip.frag == OVS_FRAG_TYPE_LATER)
740 			goto out;
741 		if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
742 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
743 
744 		/* Transport layer. */
745 		if (key->ip.proto == NEXTHDR_TCP) {
746 			key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
747 			if (tcphdr_ok(skb)) {
748 				struct tcphdr *tcp = tcp_hdr(skb);
749 				key->ipv6.tp.src = tcp->source;
750 				key->ipv6.tp.dst = tcp->dest;
751 			}
752 		} else if (key->ip.proto == NEXTHDR_UDP) {
753 			key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
754 			if (udphdr_ok(skb)) {
755 				struct udphdr *udp = udp_hdr(skb);
756 				key->ipv6.tp.src = udp->source;
757 				key->ipv6.tp.dst = udp->dest;
758 			}
759 		} else if (key->ip.proto == NEXTHDR_ICMP) {
760 			key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
761 			if (icmp6hdr_ok(skb)) {
762 				error = parse_icmpv6(skb, key, &key_len, nh_len);
763 				if (error < 0)
764 					goto out;
765 			}
766 		}
767 	}
768 
769 out:
770 	*key_lenp = key_len;
771 	return error;
772 }
773 
774 static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start, int key_len)
775 {
776 	return jhash2((u32 *)((u8 *)key + key_start),
777 		      DIV_ROUND_UP(key_len - key_start, sizeof(u32)), 0);
778 }
779 
780 static int flow_key_start(struct sw_flow_key *key)
781 {
782 	if (key->tun_key.ipv4_dst)
783 		return 0;
784 	else
785 		return offsetof(struct sw_flow_key, phy);
786 }
787 
788 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
789 				struct sw_flow_key *key, int key_len)
790 {
791 	struct sw_flow *flow;
792 	struct hlist_head *head;
793 	u8 *_key;
794 	int key_start;
795 	u32 hash;
796 
797 	key_start = flow_key_start(key);
798 	hash = ovs_flow_hash(key, key_start, key_len);
799 
800 	_key = (u8 *) key + key_start;
801 	head = find_bucket(table, hash);
802 	hlist_for_each_entry_rcu(flow, head, hash_node[table->node_ver]) {
803 
804 		if (flow->hash == hash &&
805 		    !memcmp((u8 *)&flow->key + key_start, _key, key_len - key_start)) {
806 			return flow;
807 		}
808 	}
809 	return NULL;
810 }
811 
812 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
813 			 struct sw_flow_key *key, int key_len)
814 {
815 	flow->hash = ovs_flow_hash(key, flow_key_start(key), key_len);
816 	memcpy(&flow->key, key, sizeof(flow->key));
817 	__flow_tbl_insert(table, flow);
818 }
819 
820 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
821 {
822 	BUG_ON(table->count == 0);
823 	hlist_del_rcu(&flow->hash_node[table->node_ver]);
824 	table->count--;
825 }
826 
827 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
828 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
829 	[OVS_KEY_ATTR_ENCAP] = -1,
830 	[OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
831 	[OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
832 	[OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
833 	[OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
834 	[OVS_KEY_ATTR_VLAN] = sizeof(__be16),
835 	[OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
836 	[OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
837 	[OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
838 	[OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
839 	[OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
840 	[OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
841 	[OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
842 	[OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
843 	[OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
844 	[OVS_KEY_ATTR_TUNNEL] = -1,
845 };
846 
847 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
848 				  const struct nlattr *a[], u32 *attrs)
849 {
850 	const struct ovs_key_icmp *icmp_key;
851 	const struct ovs_key_tcp *tcp_key;
852 	const struct ovs_key_udp *udp_key;
853 
854 	switch (swkey->ip.proto) {
855 	case IPPROTO_TCP:
856 		if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
857 			return -EINVAL;
858 		*attrs &= ~(1 << OVS_KEY_ATTR_TCP);
859 
860 		*key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
861 		tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
862 		swkey->ipv4.tp.src = tcp_key->tcp_src;
863 		swkey->ipv4.tp.dst = tcp_key->tcp_dst;
864 		break;
865 
866 	case IPPROTO_UDP:
867 		if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
868 			return -EINVAL;
869 		*attrs &= ~(1 << OVS_KEY_ATTR_UDP);
870 
871 		*key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
872 		udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
873 		swkey->ipv4.tp.src = udp_key->udp_src;
874 		swkey->ipv4.tp.dst = udp_key->udp_dst;
875 		break;
876 
877 	case IPPROTO_ICMP:
878 		if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
879 			return -EINVAL;
880 		*attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
881 
882 		*key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
883 		icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
884 		swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
885 		swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
886 		break;
887 	}
888 
889 	return 0;
890 }
891 
892 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
893 				  const struct nlattr *a[], u32 *attrs)
894 {
895 	const struct ovs_key_icmpv6 *icmpv6_key;
896 	const struct ovs_key_tcp *tcp_key;
897 	const struct ovs_key_udp *udp_key;
898 
899 	switch (swkey->ip.proto) {
900 	case IPPROTO_TCP:
901 		if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
902 			return -EINVAL;
903 		*attrs &= ~(1 << OVS_KEY_ATTR_TCP);
904 
905 		*key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
906 		tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
907 		swkey->ipv6.tp.src = tcp_key->tcp_src;
908 		swkey->ipv6.tp.dst = tcp_key->tcp_dst;
909 		break;
910 
911 	case IPPROTO_UDP:
912 		if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
913 			return -EINVAL;
914 		*attrs &= ~(1 << OVS_KEY_ATTR_UDP);
915 
916 		*key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
917 		udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
918 		swkey->ipv6.tp.src = udp_key->udp_src;
919 		swkey->ipv6.tp.dst = udp_key->udp_dst;
920 		break;
921 
922 	case IPPROTO_ICMPV6:
923 		if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
924 			return -EINVAL;
925 		*attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
926 
927 		*key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
928 		icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
929 		swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
930 		swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
931 
932 		if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
933 		    swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
934 			const struct ovs_key_nd *nd_key;
935 
936 			if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
937 				return -EINVAL;
938 			*attrs &= ~(1 << OVS_KEY_ATTR_ND);
939 
940 			*key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
941 			nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
942 			memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
943 			       sizeof(swkey->ipv6.nd.target));
944 			memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
945 			memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
946 		}
947 		break;
948 	}
949 
950 	return 0;
951 }
952 
953 static int parse_flow_nlattrs(const struct nlattr *attr,
954 			      const struct nlattr *a[], u32 *attrsp)
955 {
956 	const struct nlattr *nla;
957 	u32 attrs;
958 	int rem;
959 
960 	attrs = 0;
961 	nla_for_each_nested(nla, attr, rem) {
962 		u16 type = nla_type(nla);
963 		int expected_len;
964 
965 		if (type > OVS_KEY_ATTR_MAX || attrs & (1 << type))
966 			return -EINVAL;
967 
968 		expected_len = ovs_key_lens[type];
969 		if (nla_len(nla) != expected_len && expected_len != -1)
970 			return -EINVAL;
971 
972 		attrs |= 1 << type;
973 		a[type] = nla;
974 	}
975 	if (rem)
976 		return -EINVAL;
977 
978 	*attrsp = attrs;
979 	return 0;
980 }
981 
982 int ovs_ipv4_tun_from_nlattr(const struct nlattr *attr,
983 			     struct ovs_key_ipv4_tunnel *tun_key)
984 {
985 	struct nlattr *a;
986 	int rem;
987 	bool ttl = false;
988 
989 	memset(tun_key, 0, sizeof(*tun_key));
990 
991 	nla_for_each_nested(a, attr, rem) {
992 		int type = nla_type(a);
993 		static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
994 			[OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
995 			[OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
996 			[OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
997 			[OVS_TUNNEL_KEY_ATTR_TOS] = 1,
998 			[OVS_TUNNEL_KEY_ATTR_TTL] = 1,
999 			[OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
1000 			[OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
1001 		};
1002 
1003 		if (type > OVS_TUNNEL_KEY_ATTR_MAX ||
1004 			ovs_tunnel_key_lens[type] != nla_len(a))
1005 			return -EINVAL;
1006 
1007 		switch (type) {
1008 		case OVS_TUNNEL_KEY_ATTR_ID:
1009 			tun_key->tun_id = nla_get_be64(a);
1010 			tun_key->tun_flags |= TUNNEL_KEY;
1011 			break;
1012 		case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1013 			tun_key->ipv4_src = nla_get_be32(a);
1014 			break;
1015 		case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1016 			tun_key->ipv4_dst = nla_get_be32(a);
1017 			break;
1018 		case OVS_TUNNEL_KEY_ATTR_TOS:
1019 			tun_key->ipv4_tos = nla_get_u8(a);
1020 			break;
1021 		case OVS_TUNNEL_KEY_ATTR_TTL:
1022 			tun_key->ipv4_ttl = nla_get_u8(a);
1023 			ttl = true;
1024 			break;
1025 		case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1026 			tun_key->tun_flags |= TUNNEL_DONT_FRAGMENT;
1027 			break;
1028 		case OVS_TUNNEL_KEY_ATTR_CSUM:
1029 			tun_key->tun_flags |= TUNNEL_CSUM;
1030 			break;
1031 		default:
1032 			return -EINVAL;
1033 
1034 		}
1035 	}
1036 	if (rem > 0)
1037 		return -EINVAL;
1038 
1039 	if (!tun_key->ipv4_dst)
1040 		return -EINVAL;
1041 
1042 	if (!ttl)
1043 		return -EINVAL;
1044 
1045 	return 0;
1046 }
1047 
1048 int ovs_ipv4_tun_to_nlattr(struct sk_buff *skb,
1049 			   const struct ovs_key_ipv4_tunnel *tun_key)
1050 {
1051 	struct nlattr *nla;
1052 
1053 	nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
1054 	if (!nla)
1055 		return -EMSGSIZE;
1056 
1057 	if (tun_key->tun_flags & TUNNEL_KEY &&
1058 	    nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, tun_key->tun_id))
1059 		return -EMSGSIZE;
1060 	if (tun_key->ipv4_src &&
1061 	    nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, tun_key->ipv4_src))
1062 		return -EMSGSIZE;
1063 	if (nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, tun_key->ipv4_dst))
1064 		return -EMSGSIZE;
1065 	if (tun_key->ipv4_tos &&
1066 	    nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, tun_key->ipv4_tos))
1067 		return -EMSGSIZE;
1068 	if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, tun_key->ipv4_ttl))
1069 		return -EMSGSIZE;
1070 	if ((tun_key->tun_flags & TUNNEL_DONT_FRAGMENT) &&
1071 		nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
1072 		return -EMSGSIZE;
1073 	if ((tun_key->tun_flags & TUNNEL_CSUM) &&
1074 		nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
1075 		return -EMSGSIZE;
1076 
1077 	nla_nest_end(skb, nla);
1078 	return 0;
1079 }
1080 
1081 /**
1082  * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
1083  * @swkey: receives the extracted flow key.
1084  * @key_lenp: number of bytes used in @swkey.
1085  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1086  * sequence.
1087  */
1088 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
1089 		      const struct nlattr *attr)
1090 {
1091 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1092 	const struct ovs_key_ethernet *eth_key;
1093 	int key_len;
1094 	u32 attrs;
1095 	int err;
1096 
1097 	memset(swkey, 0, sizeof(struct sw_flow_key));
1098 	key_len = SW_FLOW_KEY_OFFSET(eth);
1099 
1100 	err = parse_flow_nlattrs(attr, a, &attrs);
1101 	if (err)
1102 		return err;
1103 
1104 	/* Metadata attributes. */
1105 	if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1106 		swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
1107 		attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1108 	}
1109 	if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1110 		u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1111 		if (in_port >= DP_MAX_PORTS)
1112 			return -EINVAL;
1113 		swkey->phy.in_port = in_port;
1114 		attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1115 	} else {
1116 		swkey->phy.in_port = DP_MAX_PORTS;
1117 	}
1118 	if (attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1119 		swkey->phy.skb_mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1120 		attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1121 	}
1122 
1123 	if (attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
1124 		err = ovs_ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], &swkey->tun_key);
1125 		if (err)
1126 			return err;
1127 
1128 		attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
1129 	}
1130 
1131 	/* Data attributes. */
1132 	if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1133 		return -EINVAL;
1134 	attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1135 
1136 	eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1137 	memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1138 	memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1139 
1140 	if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1141 	    nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1142 		const struct nlattr *encap;
1143 		__be16 tci;
1144 
1145 		if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1146 			      (1 << OVS_KEY_ATTR_ETHERTYPE) |
1147 			      (1 << OVS_KEY_ATTR_ENCAP)))
1148 			return -EINVAL;
1149 
1150 		encap = a[OVS_KEY_ATTR_ENCAP];
1151 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1152 		if (tci & htons(VLAN_TAG_PRESENT)) {
1153 			swkey->eth.tci = tci;
1154 
1155 			err = parse_flow_nlattrs(encap, a, &attrs);
1156 			if (err)
1157 				return err;
1158 		} else if (!tci) {
1159 			/* Corner case for truncated 802.1Q header. */
1160 			if (nla_len(encap))
1161 				return -EINVAL;
1162 
1163 			swkey->eth.type = htons(ETH_P_8021Q);
1164 			*key_lenp = key_len;
1165 			return 0;
1166 		} else {
1167 			return -EINVAL;
1168 		}
1169 	}
1170 
1171 	if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1172 		swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1173 		if (ntohs(swkey->eth.type) < ETH_P_802_3_MIN)
1174 			return -EINVAL;
1175 		attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1176 	} else {
1177 		swkey->eth.type = htons(ETH_P_802_2);
1178 	}
1179 
1180 	if (swkey->eth.type == htons(ETH_P_IP)) {
1181 		const struct ovs_key_ipv4 *ipv4_key;
1182 
1183 		if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1184 			return -EINVAL;
1185 		attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1186 
1187 		key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1188 		ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1189 		if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1190 			return -EINVAL;
1191 		swkey->ip.proto = ipv4_key->ipv4_proto;
1192 		swkey->ip.tos = ipv4_key->ipv4_tos;
1193 		swkey->ip.ttl = ipv4_key->ipv4_ttl;
1194 		swkey->ip.frag = ipv4_key->ipv4_frag;
1195 		swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1196 		swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1197 
1198 		if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1199 			err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1200 			if (err)
1201 				return err;
1202 		}
1203 	} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1204 		const struct ovs_key_ipv6 *ipv6_key;
1205 
1206 		if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1207 			return -EINVAL;
1208 		attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1209 
1210 		key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1211 		ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1212 		if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1213 			return -EINVAL;
1214 		swkey->ipv6.label = ipv6_key->ipv6_label;
1215 		swkey->ip.proto = ipv6_key->ipv6_proto;
1216 		swkey->ip.tos = ipv6_key->ipv6_tclass;
1217 		swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1218 		swkey->ip.frag = ipv6_key->ipv6_frag;
1219 		memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1220 		       sizeof(swkey->ipv6.addr.src));
1221 		memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1222 		       sizeof(swkey->ipv6.addr.dst));
1223 
1224 		if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1225 			err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1226 			if (err)
1227 				return err;
1228 		}
1229 	} else if (swkey->eth.type == htons(ETH_P_ARP) ||
1230 		   swkey->eth.type == htons(ETH_P_RARP)) {
1231 		const struct ovs_key_arp *arp_key;
1232 
1233 		if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1234 			return -EINVAL;
1235 		attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1236 
1237 		key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1238 		arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1239 		swkey->ipv4.addr.src = arp_key->arp_sip;
1240 		swkey->ipv4.addr.dst = arp_key->arp_tip;
1241 		if (arp_key->arp_op & htons(0xff00))
1242 			return -EINVAL;
1243 		swkey->ip.proto = ntohs(arp_key->arp_op);
1244 		memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1245 		memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1246 	}
1247 
1248 	if (attrs)
1249 		return -EINVAL;
1250 	*key_lenp = key_len;
1251 
1252 	return 0;
1253 }
1254 
1255 /**
1256  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1257  * @flow: Receives extracted in_port, priority, tun_key and skb_mark.
1258  * @key_len: Length of key in @flow.  Used for calculating flow hash.
1259  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1260  * sequence.
1261  *
1262  * This parses a series of Netlink attributes that form a flow key, which must
1263  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1264  * get the metadata, that is, the parts of the flow key that cannot be
1265  * extracted from the packet itself.
1266  */
1267 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow, int key_len,
1268 				   const struct nlattr *attr)
1269 {
1270 	struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1271 	const struct nlattr *nla;
1272 	int rem;
1273 
1274 	flow->key.phy.in_port = DP_MAX_PORTS;
1275 	flow->key.phy.priority = 0;
1276 	flow->key.phy.skb_mark = 0;
1277 	memset(tun_key, 0, sizeof(flow->key.tun_key));
1278 
1279 	nla_for_each_nested(nla, attr, rem) {
1280 		int type = nla_type(nla);
1281 
1282 		if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1283 			int err;
1284 
1285 			if (nla_len(nla) != ovs_key_lens[type])
1286 				return -EINVAL;
1287 
1288 			switch (type) {
1289 			case OVS_KEY_ATTR_PRIORITY:
1290 				flow->key.phy.priority = nla_get_u32(nla);
1291 				break;
1292 
1293 			case OVS_KEY_ATTR_TUNNEL:
1294 				err = ovs_ipv4_tun_from_nlattr(nla, tun_key);
1295 				if (err)
1296 					return err;
1297 				break;
1298 
1299 			case OVS_KEY_ATTR_IN_PORT:
1300 				if (nla_get_u32(nla) >= DP_MAX_PORTS)
1301 					return -EINVAL;
1302 				flow->key.phy.in_port = nla_get_u32(nla);
1303 				break;
1304 
1305 			case OVS_KEY_ATTR_SKB_MARK:
1306 				flow->key.phy.skb_mark = nla_get_u32(nla);
1307 				break;
1308 			}
1309 		}
1310 	}
1311 	if (rem)
1312 		return -EINVAL;
1313 
1314 	flow->hash = ovs_flow_hash(&flow->key,
1315 				   flow_key_start(&flow->key), key_len);
1316 
1317 	return 0;
1318 }
1319 
1320 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1321 {
1322 	struct ovs_key_ethernet *eth_key;
1323 	struct nlattr *nla, *encap;
1324 
1325 	if (swkey->phy.priority &&
1326 	    nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority))
1327 		goto nla_put_failure;
1328 
1329 	if (swkey->tun_key.ipv4_dst &&
1330 	    ovs_ipv4_tun_to_nlattr(skb, &swkey->tun_key))
1331 		goto nla_put_failure;
1332 
1333 	if (swkey->phy.in_port != DP_MAX_PORTS &&
1334 	    nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port))
1335 		goto nla_put_failure;
1336 
1337 	if (swkey->phy.skb_mark &&
1338 	    nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, swkey->phy.skb_mark))
1339 		goto nla_put_failure;
1340 
1341 	nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1342 	if (!nla)
1343 		goto nla_put_failure;
1344 	eth_key = nla_data(nla);
1345 	memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1346 	memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1347 
1348 	if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1349 		if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q)) ||
1350 		    nla_put_be16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci))
1351 			goto nla_put_failure;
1352 		encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1353 		if (!swkey->eth.tci)
1354 			goto unencap;
1355 	} else {
1356 		encap = NULL;
1357 	}
1358 
1359 	if (swkey->eth.type == htons(ETH_P_802_2))
1360 		goto unencap;
1361 
1362 	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type))
1363 		goto nla_put_failure;
1364 
1365 	if (swkey->eth.type == htons(ETH_P_IP)) {
1366 		struct ovs_key_ipv4 *ipv4_key;
1367 
1368 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1369 		if (!nla)
1370 			goto nla_put_failure;
1371 		ipv4_key = nla_data(nla);
1372 		ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1373 		ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1374 		ipv4_key->ipv4_proto = swkey->ip.proto;
1375 		ipv4_key->ipv4_tos = swkey->ip.tos;
1376 		ipv4_key->ipv4_ttl = swkey->ip.ttl;
1377 		ipv4_key->ipv4_frag = swkey->ip.frag;
1378 	} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1379 		struct ovs_key_ipv6 *ipv6_key;
1380 
1381 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1382 		if (!nla)
1383 			goto nla_put_failure;
1384 		ipv6_key = nla_data(nla);
1385 		memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1386 				sizeof(ipv6_key->ipv6_src));
1387 		memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1388 				sizeof(ipv6_key->ipv6_dst));
1389 		ipv6_key->ipv6_label = swkey->ipv6.label;
1390 		ipv6_key->ipv6_proto = swkey->ip.proto;
1391 		ipv6_key->ipv6_tclass = swkey->ip.tos;
1392 		ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1393 		ipv6_key->ipv6_frag = swkey->ip.frag;
1394 	} else if (swkey->eth.type == htons(ETH_P_ARP) ||
1395 		   swkey->eth.type == htons(ETH_P_RARP)) {
1396 		struct ovs_key_arp *arp_key;
1397 
1398 		nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1399 		if (!nla)
1400 			goto nla_put_failure;
1401 		arp_key = nla_data(nla);
1402 		memset(arp_key, 0, sizeof(struct ovs_key_arp));
1403 		arp_key->arp_sip = swkey->ipv4.addr.src;
1404 		arp_key->arp_tip = swkey->ipv4.addr.dst;
1405 		arp_key->arp_op = htons(swkey->ip.proto);
1406 		memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1407 		memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1408 	}
1409 
1410 	if ((swkey->eth.type == htons(ETH_P_IP) ||
1411 	     swkey->eth.type == htons(ETH_P_IPV6)) &&
1412 	     swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1413 
1414 		if (swkey->ip.proto == IPPROTO_TCP) {
1415 			struct ovs_key_tcp *tcp_key;
1416 
1417 			nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1418 			if (!nla)
1419 				goto nla_put_failure;
1420 			tcp_key = nla_data(nla);
1421 			if (swkey->eth.type == htons(ETH_P_IP)) {
1422 				tcp_key->tcp_src = swkey->ipv4.tp.src;
1423 				tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1424 			} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1425 				tcp_key->tcp_src = swkey->ipv6.tp.src;
1426 				tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1427 			}
1428 		} else if (swkey->ip.proto == IPPROTO_UDP) {
1429 			struct ovs_key_udp *udp_key;
1430 
1431 			nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1432 			if (!nla)
1433 				goto nla_put_failure;
1434 			udp_key = nla_data(nla);
1435 			if (swkey->eth.type == htons(ETH_P_IP)) {
1436 				udp_key->udp_src = swkey->ipv4.tp.src;
1437 				udp_key->udp_dst = swkey->ipv4.tp.dst;
1438 			} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1439 				udp_key->udp_src = swkey->ipv6.tp.src;
1440 				udp_key->udp_dst = swkey->ipv6.tp.dst;
1441 			}
1442 		} else if (swkey->eth.type == htons(ETH_P_IP) &&
1443 			   swkey->ip.proto == IPPROTO_ICMP) {
1444 			struct ovs_key_icmp *icmp_key;
1445 
1446 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1447 			if (!nla)
1448 				goto nla_put_failure;
1449 			icmp_key = nla_data(nla);
1450 			icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1451 			icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1452 		} else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1453 			   swkey->ip.proto == IPPROTO_ICMPV6) {
1454 			struct ovs_key_icmpv6 *icmpv6_key;
1455 
1456 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1457 						sizeof(*icmpv6_key));
1458 			if (!nla)
1459 				goto nla_put_failure;
1460 			icmpv6_key = nla_data(nla);
1461 			icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1462 			icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1463 
1464 			if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1465 			    icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1466 				struct ovs_key_nd *nd_key;
1467 
1468 				nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1469 				if (!nla)
1470 					goto nla_put_failure;
1471 				nd_key = nla_data(nla);
1472 				memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1473 							sizeof(nd_key->nd_target));
1474 				memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1475 				memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1476 			}
1477 		}
1478 	}
1479 
1480 unencap:
1481 	if (encap)
1482 		nla_nest_end(skb, encap);
1483 
1484 	return 0;
1485 
1486 nla_put_failure:
1487 	return -EMSGSIZE;
1488 }
1489 
1490 /* Initializes the flow module.
1491  * Returns zero if successful or a negative error code. */
1492 int ovs_flow_init(void)
1493 {
1494 	flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1495 					0, NULL);
1496 	if (flow_cache == NULL)
1497 		return -ENOMEM;
1498 
1499 	return 0;
1500 }
1501 
1502 /* Uninitializes the flow module. */
1503 void ovs_flow_exit(void)
1504 {
1505 	kmem_cache_destroy(flow_cache);
1506 }
1507