xref: /openbmc/linux/net/openvswitch/flow.c (revision f7777dcc)
1 /*
2  * Copyright (c) 2007-2013 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/sctp.h>
38 #include <linux/tcp.h>
39 #include <linux/udp.h>
40 #include <linux/icmp.h>
41 #include <linux/icmpv6.h>
42 #include <linux/rculist.h>
43 #include <net/ip.h>
44 #include <net/ip_tunnels.h>
45 #include <net/ipv6.h>
46 #include <net/ndisc.h>
47 
48 static struct kmem_cache *flow_cache;
49 
50 static void ovs_sw_flow_mask_set(struct sw_flow_mask *mask,
51 		struct sw_flow_key_range *range, u8 val);
52 
53 static void update_range__(struct sw_flow_match *match,
54 			  size_t offset, size_t size, bool is_mask)
55 {
56 	struct sw_flow_key_range *range = NULL;
57 	size_t start = rounddown(offset, sizeof(long));
58 	size_t end = roundup(offset + size, sizeof(long));
59 
60 	if (!is_mask)
61 		range = &match->range;
62 	else if (match->mask)
63 		range = &match->mask->range;
64 
65 	if (!range)
66 		return;
67 
68 	if (range->start == range->end) {
69 		range->start = start;
70 		range->end = end;
71 		return;
72 	}
73 
74 	if (range->start > start)
75 		range->start = start;
76 
77 	if (range->end < end)
78 		range->end = end;
79 }
80 
81 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
82 	do { \
83 		update_range__(match, offsetof(struct sw_flow_key, field),  \
84 				     sizeof((match)->key->field), is_mask); \
85 		if (is_mask) {						    \
86 			if ((match)->mask)				    \
87 				(match)->mask->key.field = value;	    \
88 		} else {                                                    \
89 			(match)->key->field = value;		            \
90 		}                                                           \
91 	} while (0)
92 
93 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask) \
94 	do { \
95 		update_range__(match, offsetof(struct sw_flow_key, field),  \
96 				len, is_mask);                              \
97 		if (is_mask) {						    \
98 			if ((match)->mask)				    \
99 				memcpy(&(match)->mask->key.field, value_p, len);\
100 		} else {                                                    \
101 			memcpy(&(match)->key->field, value_p, len);         \
102 		}                                                           \
103 	} while (0)
104 
105 static u16 range_n_bytes(const struct sw_flow_key_range *range)
106 {
107 	return range->end - range->start;
108 }
109 
110 void ovs_match_init(struct sw_flow_match *match,
111 		    struct sw_flow_key *key,
112 		    struct sw_flow_mask *mask)
113 {
114 	memset(match, 0, sizeof(*match));
115 	match->key = key;
116 	match->mask = mask;
117 
118 	memset(key, 0, sizeof(*key));
119 
120 	if (mask) {
121 		memset(&mask->key, 0, sizeof(mask->key));
122 		mask->range.start = mask->range.end = 0;
123 	}
124 }
125 
126 static bool ovs_match_validate(const struct sw_flow_match *match,
127 		u64 key_attrs, u64 mask_attrs)
128 {
129 	u64 key_expected = 1 << OVS_KEY_ATTR_ETHERNET;
130 	u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
131 
132 	/* The following mask attributes allowed only if they
133 	 * pass the validation tests. */
134 	mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
135 			| (1 << OVS_KEY_ATTR_IPV6)
136 			| (1 << OVS_KEY_ATTR_TCP)
137 			| (1 << OVS_KEY_ATTR_UDP)
138 			| (1 << OVS_KEY_ATTR_SCTP)
139 			| (1 << OVS_KEY_ATTR_ICMP)
140 			| (1 << OVS_KEY_ATTR_ICMPV6)
141 			| (1 << OVS_KEY_ATTR_ARP)
142 			| (1 << OVS_KEY_ATTR_ND));
143 
144 	/* Always allowed mask fields. */
145 	mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
146 		       | (1 << OVS_KEY_ATTR_IN_PORT)
147 		       | (1 << OVS_KEY_ATTR_ETHERTYPE));
148 
149 	/* Check key attributes. */
150 	if (match->key->eth.type == htons(ETH_P_ARP)
151 			|| match->key->eth.type == htons(ETH_P_RARP)) {
152 		key_expected |= 1 << OVS_KEY_ATTR_ARP;
153 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
154 			mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
155 	}
156 
157 	if (match->key->eth.type == htons(ETH_P_IP)) {
158 		key_expected |= 1 << OVS_KEY_ATTR_IPV4;
159 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
160 			mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
161 
162 		if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
163 			if (match->key->ip.proto == IPPROTO_UDP) {
164 				key_expected |= 1 << OVS_KEY_ATTR_UDP;
165 				if (match->mask && (match->mask->key.ip.proto == 0xff))
166 					mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
167 			}
168 
169 			if (match->key->ip.proto == IPPROTO_SCTP) {
170 				key_expected |= 1 << OVS_KEY_ATTR_SCTP;
171 				if (match->mask && (match->mask->key.ip.proto == 0xff))
172 					mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
173 			}
174 
175 			if (match->key->ip.proto == IPPROTO_TCP) {
176 				key_expected |= 1 << OVS_KEY_ATTR_TCP;
177 				if (match->mask && (match->mask->key.ip.proto == 0xff))
178 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
179 			}
180 
181 			if (match->key->ip.proto == IPPROTO_ICMP) {
182 				key_expected |= 1 << OVS_KEY_ATTR_ICMP;
183 				if (match->mask && (match->mask->key.ip.proto == 0xff))
184 					mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
185 			}
186 		}
187 	}
188 
189 	if (match->key->eth.type == htons(ETH_P_IPV6)) {
190 		key_expected |= 1 << OVS_KEY_ATTR_IPV6;
191 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
192 			mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
193 
194 		if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
195 			if (match->key->ip.proto == IPPROTO_UDP) {
196 				key_expected |= 1 << OVS_KEY_ATTR_UDP;
197 				if (match->mask && (match->mask->key.ip.proto == 0xff))
198 					mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
199 			}
200 
201 			if (match->key->ip.proto == IPPROTO_SCTP) {
202 				key_expected |= 1 << OVS_KEY_ATTR_SCTP;
203 				if (match->mask && (match->mask->key.ip.proto == 0xff))
204 					mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
205 			}
206 
207 			if (match->key->ip.proto == IPPROTO_TCP) {
208 				key_expected |= 1 << OVS_KEY_ATTR_TCP;
209 				if (match->mask && (match->mask->key.ip.proto == 0xff))
210 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
211 			}
212 
213 			if (match->key->ip.proto == IPPROTO_ICMPV6) {
214 				key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
215 				if (match->mask && (match->mask->key.ip.proto == 0xff))
216 					mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
217 
218 				if (match->key->ipv6.tp.src ==
219 						htons(NDISC_NEIGHBOUR_SOLICITATION) ||
220 				    match->key->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
221 					key_expected |= 1 << OVS_KEY_ATTR_ND;
222 					if (match->mask && (match->mask->key.ipv6.tp.src == htons(0xffff)))
223 						mask_allowed |= 1 << OVS_KEY_ATTR_ND;
224 				}
225 			}
226 		}
227 	}
228 
229 	if ((key_attrs & key_expected) != key_expected) {
230 		/* Key attributes check failed. */
231 		OVS_NLERR("Missing expected key attributes (key_attrs=%llx, expected=%llx).\n",
232 				key_attrs, key_expected);
233 		return false;
234 	}
235 
236 	if ((mask_attrs & mask_allowed) != mask_attrs) {
237 		/* Mask attributes check failed. */
238 		OVS_NLERR("Contain more than allowed mask fields (mask_attrs=%llx, mask_allowed=%llx).\n",
239 				mask_attrs, mask_allowed);
240 		return false;
241 	}
242 
243 	return true;
244 }
245 
246 static int check_header(struct sk_buff *skb, int len)
247 {
248 	if (unlikely(skb->len < len))
249 		return -EINVAL;
250 	if (unlikely(!pskb_may_pull(skb, len)))
251 		return -ENOMEM;
252 	return 0;
253 }
254 
255 static bool arphdr_ok(struct sk_buff *skb)
256 {
257 	return pskb_may_pull(skb, skb_network_offset(skb) +
258 				  sizeof(struct arp_eth_header));
259 }
260 
261 static int check_iphdr(struct sk_buff *skb)
262 {
263 	unsigned int nh_ofs = skb_network_offset(skb);
264 	unsigned int ip_len;
265 	int err;
266 
267 	err = check_header(skb, nh_ofs + sizeof(struct iphdr));
268 	if (unlikely(err))
269 		return err;
270 
271 	ip_len = ip_hdrlen(skb);
272 	if (unlikely(ip_len < sizeof(struct iphdr) ||
273 		     skb->len < nh_ofs + ip_len))
274 		return -EINVAL;
275 
276 	skb_set_transport_header(skb, nh_ofs + ip_len);
277 	return 0;
278 }
279 
280 static bool tcphdr_ok(struct sk_buff *skb)
281 {
282 	int th_ofs = skb_transport_offset(skb);
283 	int tcp_len;
284 
285 	if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
286 		return false;
287 
288 	tcp_len = tcp_hdrlen(skb);
289 	if (unlikely(tcp_len < sizeof(struct tcphdr) ||
290 		     skb->len < th_ofs + tcp_len))
291 		return false;
292 
293 	return true;
294 }
295 
296 static bool udphdr_ok(struct sk_buff *skb)
297 {
298 	return pskb_may_pull(skb, skb_transport_offset(skb) +
299 				  sizeof(struct udphdr));
300 }
301 
302 static bool sctphdr_ok(struct sk_buff *skb)
303 {
304 	return pskb_may_pull(skb, skb_transport_offset(skb) +
305 				  sizeof(struct sctphdr));
306 }
307 
308 static bool icmphdr_ok(struct sk_buff *skb)
309 {
310 	return pskb_may_pull(skb, skb_transport_offset(skb) +
311 				  sizeof(struct icmphdr));
312 }
313 
314 u64 ovs_flow_used_time(unsigned long flow_jiffies)
315 {
316 	struct timespec cur_ts;
317 	u64 cur_ms, idle_ms;
318 
319 	ktime_get_ts(&cur_ts);
320 	idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
321 	cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
322 		 cur_ts.tv_nsec / NSEC_PER_MSEC;
323 
324 	return cur_ms - idle_ms;
325 }
326 
327 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
328 {
329 	unsigned int nh_ofs = skb_network_offset(skb);
330 	unsigned int nh_len;
331 	int payload_ofs;
332 	struct ipv6hdr *nh;
333 	uint8_t nexthdr;
334 	__be16 frag_off;
335 	int err;
336 
337 	err = check_header(skb, nh_ofs + sizeof(*nh));
338 	if (unlikely(err))
339 		return err;
340 
341 	nh = ipv6_hdr(skb);
342 	nexthdr = nh->nexthdr;
343 	payload_ofs = (u8 *)(nh + 1) - skb->data;
344 
345 	key->ip.proto = NEXTHDR_NONE;
346 	key->ip.tos = ipv6_get_dsfield(nh);
347 	key->ip.ttl = nh->hop_limit;
348 	key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
349 	key->ipv6.addr.src = nh->saddr;
350 	key->ipv6.addr.dst = nh->daddr;
351 
352 	payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
353 	if (unlikely(payload_ofs < 0))
354 		return -EINVAL;
355 
356 	if (frag_off) {
357 		if (frag_off & htons(~0x7))
358 			key->ip.frag = OVS_FRAG_TYPE_LATER;
359 		else
360 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
361 	}
362 
363 	nh_len = payload_ofs - nh_ofs;
364 	skb_set_transport_header(skb, nh_ofs + nh_len);
365 	key->ip.proto = nexthdr;
366 	return nh_len;
367 }
368 
369 static bool icmp6hdr_ok(struct sk_buff *skb)
370 {
371 	return pskb_may_pull(skb, skb_transport_offset(skb) +
372 				  sizeof(struct icmp6hdr));
373 }
374 
375 void ovs_flow_key_mask(struct sw_flow_key *dst, const struct sw_flow_key *src,
376 		       const struct sw_flow_mask *mask)
377 {
378 	const long *m = (long *)((u8 *)&mask->key + mask->range.start);
379 	const long *s = (long *)((u8 *)src + mask->range.start);
380 	long *d = (long *)((u8 *)dst + mask->range.start);
381 	int i;
382 
383 	/* The memory outside of the 'mask->range' are not set since
384 	 * further operations on 'dst' only uses contents within
385 	 * 'mask->range'.
386 	 */
387 	for (i = 0; i < range_n_bytes(&mask->range); i += sizeof(long))
388 		*d++ = *s++ & *m++;
389 }
390 
391 #define TCP_FLAGS_OFFSET 13
392 #define TCP_FLAG_MASK 0x3f
393 
394 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
395 {
396 	u8 tcp_flags = 0;
397 
398 	if ((flow->key.eth.type == htons(ETH_P_IP) ||
399 	     flow->key.eth.type == htons(ETH_P_IPV6)) &&
400 	    flow->key.ip.proto == IPPROTO_TCP &&
401 	    likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
402 		u8 *tcp = (u8 *)tcp_hdr(skb);
403 		tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
404 	}
405 
406 	spin_lock(&flow->lock);
407 	flow->used = jiffies;
408 	flow->packet_count++;
409 	flow->byte_count += skb->len;
410 	flow->tcp_flags |= tcp_flags;
411 	spin_unlock(&flow->lock);
412 }
413 
414 struct sw_flow_actions *ovs_flow_actions_alloc(int size)
415 {
416 	struct sw_flow_actions *sfa;
417 
418 	if (size > MAX_ACTIONS_BUFSIZE)
419 		return ERR_PTR(-EINVAL);
420 
421 	sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
422 	if (!sfa)
423 		return ERR_PTR(-ENOMEM);
424 
425 	sfa->actions_len = 0;
426 	return sfa;
427 }
428 
429 struct sw_flow *ovs_flow_alloc(void)
430 {
431 	struct sw_flow *flow;
432 
433 	flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
434 	if (!flow)
435 		return ERR_PTR(-ENOMEM);
436 
437 	spin_lock_init(&flow->lock);
438 	flow->sf_acts = NULL;
439 	flow->mask = NULL;
440 
441 	return flow;
442 }
443 
444 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
445 {
446 	hash = jhash_1word(hash, table->hash_seed);
447 	return flex_array_get(table->buckets,
448 				(hash & (table->n_buckets - 1)));
449 }
450 
451 static struct flex_array *alloc_buckets(unsigned int n_buckets)
452 {
453 	struct flex_array *buckets;
454 	int i, err;
455 
456 	buckets = flex_array_alloc(sizeof(struct hlist_head),
457 				   n_buckets, GFP_KERNEL);
458 	if (!buckets)
459 		return NULL;
460 
461 	err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
462 	if (err) {
463 		flex_array_free(buckets);
464 		return NULL;
465 	}
466 
467 	for (i = 0; i < n_buckets; i++)
468 		INIT_HLIST_HEAD((struct hlist_head *)
469 					flex_array_get(buckets, i));
470 
471 	return buckets;
472 }
473 
474 static void free_buckets(struct flex_array *buckets)
475 {
476 	flex_array_free(buckets);
477 }
478 
479 static struct flow_table *__flow_tbl_alloc(int new_size)
480 {
481 	struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
482 
483 	if (!table)
484 		return NULL;
485 
486 	table->buckets = alloc_buckets(new_size);
487 
488 	if (!table->buckets) {
489 		kfree(table);
490 		return NULL;
491 	}
492 	table->n_buckets = new_size;
493 	table->count = 0;
494 	table->node_ver = 0;
495 	table->keep_flows = false;
496 	get_random_bytes(&table->hash_seed, sizeof(u32));
497 	table->mask_list = NULL;
498 
499 	return table;
500 }
501 
502 static void __flow_tbl_destroy(struct flow_table *table)
503 {
504 	int i;
505 
506 	if (table->keep_flows)
507 		goto skip_flows;
508 
509 	for (i = 0; i < table->n_buckets; i++) {
510 		struct sw_flow *flow;
511 		struct hlist_head *head = flex_array_get(table->buckets, i);
512 		struct hlist_node *n;
513 		int ver = table->node_ver;
514 
515 		hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
516 			hlist_del(&flow->hash_node[ver]);
517 			ovs_flow_free(flow, false);
518 		}
519 	}
520 
521 	BUG_ON(!list_empty(table->mask_list));
522 	kfree(table->mask_list);
523 
524 skip_flows:
525 	free_buckets(table->buckets);
526 	kfree(table);
527 }
528 
529 struct flow_table *ovs_flow_tbl_alloc(int new_size)
530 {
531 	struct flow_table *table = __flow_tbl_alloc(new_size);
532 
533 	if (!table)
534 		return NULL;
535 
536 	table->mask_list = kmalloc(sizeof(struct list_head), GFP_KERNEL);
537 	if (!table->mask_list) {
538 		table->keep_flows = true;
539 		__flow_tbl_destroy(table);
540 		return NULL;
541 	}
542 	INIT_LIST_HEAD(table->mask_list);
543 
544 	return table;
545 }
546 
547 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
548 {
549 	struct flow_table *table = container_of(rcu, struct flow_table, rcu);
550 
551 	__flow_tbl_destroy(table);
552 }
553 
554 void ovs_flow_tbl_destroy(struct flow_table *table, bool deferred)
555 {
556 	if (!table)
557 		return;
558 
559 	if (deferred)
560 		call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
561 	else
562 		__flow_tbl_destroy(table);
563 }
564 
565 struct sw_flow *ovs_flow_dump_next(struct flow_table *table, u32 *bucket, u32 *last)
566 {
567 	struct sw_flow *flow;
568 	struct hlist_head *head;
569 	int ver;
570 	int i;
571 
572 	ver = table->node_ver;
573 	while (*bucket < table->n_buckets) {
574 		i = 0;
575 		head = flex_array_get(table->buckets, *bucket);
576 		hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
577 			if (i < *last) {
578 				i++;
579 				continue;
580 			}
581 			*last = i + 1;
582 			return flow;
583 		}
584 		(*bucket)++;
585 		*last = 0;
586 	}
587 
588 	return NULL;
589 }
590 
591 static void __tbl_insert(struct flow_table *table, struct sw_flow *flow)
592 {
593 	struct hlist_head *head;
594 
595 	head = find_bucket(table, flow->hash);
596 	hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
597 
598 	table->count++;
599 }
600 
601 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
602 {
603 	int old_ver;
604 	int i;
605 
606 	old_ver = old->node_ver;
607 	new->node_ver = !old_ver;
608 
609 	/* Insert in new table. */
610 	for (i = 0; i < old->n_buckets; i++) {
611 		struct sw_flow *flow;
612 		struct hlist_head *head;
613 
614 		head = flex_array_get(old->buckets, i);
615 
616 		hlist_for_each_entry(flow, head, hash_node[old_ver])
617 			__tbl_insert(new, flow);
618 	}
619 
620 	new->mask_list = old->mask_list;
621 	old->keep_flows = true;
622 }
623 
624 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
625 {
626 	struct flow_table *new_table;
627 
628 	new_table = __flow_tbl_alloc(n_buckets);
629 	if (!new_table)
630 		return ERR_PTR(-ENOMEM);
631 
632 	flow_table_copy_flows(table, new_table);
633 
634 	return new_table;
635 }
636 
637 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
638 {
639 	return __flow_tbl_rehash(table, table->n_buckets);
640 }
641 
642 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
643 {
644 	return __flow_tbl_rehash(table, table->n_buckets * 2);
645 }
646 
647 static void __flow_free(struct sw_flow *flow)
648 {
649 	kfree((struct sf_flow_acts __force *)flow->sf_acts);
650 	kmem_cache_free(flow_cache, flow);
651 }
652 
653 static void rcu_free_flow_callback(struct rcu_head *rcu)
654 {
655 	struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
656 
657 	__flow_free(flow);
658 }
659 
660 void ovs_flow_free(struct sw_flow *flow, bool deferred)
661 {
662 	if (!flow)
663 		return;
664 
665 	ovs_sw_flow_mask_del_ref(flow->mask, deferred);
666 
667 	if (deferred)
668 		call_rcu(&flow->rcu, rcu_free_flow_callback);
669 	else
670 		__flow_free(flow);
671 }
672 
673 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
674  * The caller must hold rcu_read_lock for this to be sensible. */
675 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
676 {
677 	kfree_rcu(sf_acts, rcu);
678 }
679 
680 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
681 {
682 	struct qtag_prefix {
683 		__be16 eth_type; /* ETH_P_8021Q */
684 		__be16 tci;
685 	};
686 	struct qtag_prefix *qp;
687 
688 	if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
689 		return 0;
690 
691 	if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
692 					 sizeof(__be16))))
693 		return -ENOMEM;
694 
695 	qp = (struct qtag_prefix *) skb->data;
696 	key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
697 	__skb_pull(skb, sizeof(struct qtag_prefix));
698 
699 	return 0;
700 }
701 
702 static __be16 parse_ethertype(struct sk_buff *skb)
703 {
704 	struct llc_snap_hdr {
705 		u8  dsap;  /* Always 0xAA */
706 		u8  ssap;  /* Always 0xAA */
707 		u8  ctrl;
708 		u8  oui[3];
709 		__be16 ethertype;
710 	};
711 	struct llc_snap_hdr *llc;
712 	__be16 proto;
713 
714 	proto = *(__be16 *) skb->data;
715 	__skb_pull(skb, sizeof(__be16));
716 
717 	if (ntohs(proto) >= ETH_P_802_3_MIN)
718 		return proto;
719 
720 	if (skb->len < sizeof(struct llc_snap_hdr))
721 		return htons(ETH_P_802_2);
722 
723 	if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
724 		return htons(0);
725 
726 	llc = (struct llc_snap_hdr *) skb->data;
727 	if (llc->dsap != LLC_SAP_SNAP ||
728 	    llc->ssap != LLC_SAP_SNAP ||
729 	    (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
730 		return htons(ETH_P_802_2);
731 
732 	__skb_pull(skb, sizeof(struct llc_snap_hdr));
733 
734 	if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
735 		return llc->ethertype;
736 
737 	return htons(ETH_P_802_2);
738 }
739 
740 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
741 			int nh_len)
742 {
743 	struct icmp6hdr *icmp = icmp6_hdr(skb);
744 
745 	/* The ICMPv6 type and code fields use the 16-bit transport port
746 	 * fields, so we need to store them in 16-bit network byte order.
747 	 */
748 	key->ipv6.tp.src = htons(icmp->icmp6_type);
749 	key->ipv6.tp.dst = htons(icmp->icmp6_code);
750 
751 	if (icmp->icmp6_code == 0 &&
752 	    (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
753 	     icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
754 		int icmp_len = skb->len - skb_transport_offset(skb);
755 		struct nd_msg *nd;
756 		int offset;
757 
758 		/* In order to process neighbor discovery options, we need the
759 		 * entire packet.
760 		 */
761 		if (unlikely(icmp_len < sizeof(*nd)))
762 			return 0;
763 
764 		if (unlikely(skb_linearize(skb)))
765 			return -ENOMEM;
766 
767 		nd = (struct nd_msg *)skb_transport_header(skb);
768 		key->ipv6.nd.target = nd->target;
769 
770 		icmp_len -= sizeof(*nd);
771 		offset = 0;
772 		while (icmp_len >= 8) {
773 			struct nd_opt_hdr *nd_opt =
774 				 (struct nd_opt_hdr *)(nd->opt + offset);
775 			int opt_len = nd_opt->nd_opt_len * 8;
776 
777 			if (unlikely(!opt_len || opt_len > icmp_len))
778 				return 0;
779 
780 			/* Store the link layer address if the appropriate
781 			 * option is provided.  It is considered an error if
782 			 * the same link layer option is specified twice.
783 			 */
784 			if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
785 			    && opt_len == 8) {
786 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
787 					goto invalid;
788 				memcpy(key->ipv6.nd.sll,
789 				    &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
790 			} else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
791 				   && opt_len == 8) {
792 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
793 					goto invalid;
794 				memcpy(key->ipv6.nd.tll,
795 				    &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
796 			}
797 
798 			icmp_len -= opt_len;
799 			offset += opt_len;
800 		}
801 	}
802 
803 	return 0;
804 
805 invalid:
806 	memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
807 	memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
808 	memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
809 
810 	return 0;
811 }
812 
813 /**
814  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
815  * @skb: sk_buff that contains the frame, with skb->data pointing to the
816  * Ethernet header
817  * @in_port: port number on which @skb was received.
818  * @key: output flow key
819  *
820  * The caller must ensure that skb->len >= ETH_HLEN.
821  *
822  * Returns 0 if successful, otherwise a negative errno value.
823  *
824  * Initializes @skb header pointers as follows:
825  *
826  *    - skb->mac_header: the Ethernet header.
827  *
828  *    - skb->network_header: just past the Ethernet header, or just past the
829  *      VLAN header, to the first byte of the Ethernet payload.
830  *
831  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
832  *      on output, then just past the IP header, if one is present and
833  *      of a correct length, otherwise the same as skb->network_header.
834  *      For other key->eth.type values it is left untouched.
835  */
836 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
837 {
838 	int error;
839 	struct ethhdr *eth;
840 
841 	memset(key, 0, sizeof(*key));
842 
843 	key->phy.priority = skb->priority;
844 	if (OVS_CB(skb)->tun_key)
845 		memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
846 	key->phy.in_port = in_port;
847 	key->phy.skb_mark = skb->mark;
848 
849 	skb_reset_mac_header(skb);
850 
851 	/* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
852 	 * header in the linear data area.
853 	 */
854 	eth = eth_hdr(skb);
855 	memcpy(key->eth.src, eth->h_source, ETH_ALEN);
856 	memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
857 
858 	__skb_pull(skb, 2 * ETH_ALEN);
859 	/* We are going to push all headers that we pull, so no need to
860 	 * update skb->csum here.
861 	 */
862 
863 	if (vlan_tx_tag_present(skb))
864 		key->eth.tci = htons(skb->vlan_tci);
865 	else if (eth->h_proto == htons(ETH_P_8021Q))
866 		if (unlikely(parse_vlan(skb, key)))
867 			return -ENOMEM;
868 
869 	key->eth.type = parse_ethertype(skb);
870 	if (unlikely(key->eth.type == htons(0)))
871 		return -ENOMEM;
872 
873 	skb_reset_network_header(skb);
874 	__skb_push(skb, skb->data - skb_mac_header(skb));
875 
876 	/* Network layer. */
877 	if (key->eth.type == htons(ETH_P_IP)) {
878 		struct iphdr *nh;
879 		__be16 offset;
880 
881 		error = check_iphdr(skb);
882 		if (unlikely(error)) {
883 			if (error == -EINVAL) {
884 				skb->transport_header = skb->network_header;
885 				error = 0;
886 			}
887 			return error;
888 		}
889 
890 		nh = ip_hdr(skb);
891 		key->ipv4.addr.src = nh->saddr;
892 		key->ipv4.addr.dst = nh->daddr;
893 
894 		key->ip.proto = nh->protocol;
895 		key->ip.tos = nh->tos;
896 		key->ip.ttl = nh->ttl;
897 
898 		offset = nh->frag_off & htons(IP_OFFSET);
899 		if (offset) {
900 			key->ip.frag = OVS_FRAG_TYPE_LATER;
901 			return 0;
902 		}
903 		if (nh->frag_off & htons(IP_MF) ||
904 			 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
905 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
906 
907 		/* Transport layer. */
908 		if (key->ip.proto == IPPROTO_TCP) {
909 			if (tcphdr_ok(skb)) {
910 				struct tcphdr *tcp = tcp_hdr(skb);
911 				key->ipv4.tp.src = tcp->source;
912 				key->ipv4.tp.dst = tcp->dest;
913 			}
914 		} else if (key->ip.proto == IPPROTO_UDP) {
915 			if (udphdr_ok(skb)) {
916 				struct udphdr *udp = udp_hdr(skb);
917 				key->ipv4.tp.src = udp->source;
918 				key->ipv4.tp.dst = udp->dest;
919 			}
920 		} else if (key->ip.proto == IPPROTO_SCTP) {
921 			if (sctphdr_ok(skb)) {
922 				struct sctphdr *sctp = sctp_hdr(skb);
923 				key->ipv4.tp.src = sctp->source;
924 				key->ipv4.tp.dst = sctp->dest;
925 			}
926 		} else if (key->ip.proto == IPPROTO_ICMP) {
927 			if (icmphdr_ok(skb)) {
928 				struct icmphdr *icmp = icmp_hdr(skb);
929 				/* The ICMP type and code fields use the 16-bit
930 				 * transport port fields, so we need to store
931 				 * them in 16-bit network byte order. */
932 				key->ipv4.tp.src = htons(icmp->type);
933 				key->ipv4.tp.dst = htons(icmp->code);
934 			}
935 		}
936 
937 	} else if ((key->eth.type == htons(ETH_P_ARP) ||
938 		   key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
939 		struct arp_eth_header *arp;
940 
941 		arp = (struct arp_eth_header *)skb_network_header(skb);
942 
943 		if (arp->ar_hrd == htons(ARPHRD_ETHER)
944 				&& arp->ar_pro == htons(ETH_P_IP)
945 				&& arp->ar_hln == ETH_ALEN
946 				&& arp->ar_pln == 4) {
947 
948 			/* We only match on the lower 8 bits of the opcode. */
949 			if (ntohs(arp->ar_op) <= 0xff)
950 				key->ip.proto = ntohs(arp->ar_op);
951 			memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
952 			memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
953 			memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
954 			memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
955 		}
956 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
957 		int nh_len;             /* IPv6 Header + Extensions */
958 
959 		nh_len = parse_ipv6hdr(skb, key);
960 		if (unlikely(nh_len < 0)) {
961 			if (nh_len == -EINVAL) {
962 				skb->transport_header = skb->network_header;
963 				error = 0;
964 			} else {
965 				error = nh_len;
966 			}
967 			return error;
968 		}
969 
970 		if (key->ip.frag == OVS_FRAG_TYPE_LATER)
971 			return 0;
972 		if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
973 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
974 
975 		/* Transport layer. */
976 		if (key->ip.proto == NEXTHDR_TCP) {
977 			if (tcphdr_ok(skb)) {
978 				struct tcphdr *tcp = tcp_hdr(skb);
979 				key->ipv6.tp.src = tcp->source;
980 				key->ipv6.tp.dst = tcp->dest;
981 			}
982 		} else if (key->ip.proto == NEXTHDR_UDP) {
983 			if (udphdr_ok(skb)) {
984 				struct udphdr *udp = udp_hdr(skb);
985 				key->ipv6.tp.src = udp->source;
986 				key->ipv6.tp.dst = udp->dest;
987 			}
988 		} else if (key->ip.proto == NEXTHDR_SCTP) {
989 			if (sctphdr_ok(skb)) {
990 				struct sctphdr *sctp = sctp_hdr(skb);
991 				key->ipv6.tp.src = sctp->source;
992 				key->ipv6.tp.dst = sctp->dest;
993 			}
994 		} else if (key->ip.proto == NEXTHDR_ICMP) {
995 			if (icmp6hdr_ok(skb)) {
996 				error = parse_icmpv6(skb, key, nh_len);
997 				if (error)
998 					return error;
999 			}
1000 		}
1001 	}
1002 
1003 	return 0;
1004 }
1005 
1006 static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start,
1007 			 int key_end)
1008 {
1009 	u32 *hash_key = (u32 *)((u8 *)key + key_start);
1010 	int hash_u32s = (key_end - key_start) >> 2;
1011 
1012 	/* Make sure number of hash bytes are multiple of u32. */
1013 	BUILD_BUG_ON(sizeof(long) % sizeof(u32));
1014 
1015 	return jhash2(hash_key, hash_u32s, 0);
1016 }
1017 
1018 static int flow_key_start(const struct sw_flow_key *key)
1019 {
1020 	if (key->tun_key.ipv4_dst)
1021 		return 0;
1022 	else
1023 		return rounddown(offsetof(struct sw_flow_key, phy),
1024 					  sizeof(long));
1025 }
1026 
1027 static bool __cmp_key(const struct sw_flow_key *key1,
1028 		const struct sw_flow_key *key2,  int key_start, int key_end)
1029 {
1030 	const long *cp1 = (long *)((u8 *)key1 + key_start);
1031 	const long *cp2 = (long *)((u8 *)key2 + key_start);
1032 	long diffs = 0;
1033 	int i;
1034 
1035 	for (i = key_start; i < key_end;  i += sizeof(long))
1036 		diffs |= *cp1++ ^ *cp2++;
1037 
1038 	return diffs == 0;
1039 }
1040 
1041 static bool __flow_cmp_masked_key(const struct sw_flow *flow,
1042 		const struct sw_flow_key *key, int key_start, int key_end)
1043 {
1044 	return __cmp_key(&flow->key, key, key_start, key_end);
1045 }
1046 
1047 static bool __flow_cmp_unmasked_key(const struct sw_flow *flow,
1048 		  const struct sw_flow_key *key, int key_start, int key_end)
1049 {
1050 	return __cmp_key(&flow->unmasked_key, key, key_start, key_end);
1051 }
1052 
1053 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
1054 		const struct sw_flow_key *key, int key_end)
1055 {
1056 	int key_start;
1057 	key_start = flow_key_start(key);
1058 
1059 	return __flow_cmp_unmasked_key(flow, key, key_start, key_end);
1060 
1061 }
1062 
1063 struct sw_flow *ovs_flow_lookup_unmasked_key(struct flow_table *table,
1064 				       struct sw_flow_match *match)
1065 {
1066 	struct sw_flow_key *unmasked = match->key;
1067 	int key_end = match->range.end;
1068 	struct sw_flow *flow;
1069 
1070 	flow = ovs_flow_lookup(table, unmasked);
1071 	if (flow && (!ovs_flow_cmp_unmasked_key(flow, unmasked, key_end)))
1072 		flow = NULL;
1073 
1074 	return flow;
1075 }
1076 
1077 static struct sw_flow *ovs_masked_flow_lookup(struct flow_table *table,
1078 				    const struct sw_flow_key *unmasked,
1079 				    struct sw_flow_mask *mask)
1080 {
1081 	struct sw_flow *flow;
1082 	struct hlist_head *head;
1083 	int key_start = mask->range.start;
1084 	int key_end = mask->range.end;
1085 	u32 hash;
1086 	struct sw_flow_key masked_key;
1087 
1088 	ovs_flow_key_mask(&masked_key, unmasked, mask);
1089 	hash = ovs_flow_hash(&masked_key, key_start, key_end);
1090 	head = find_bucket(table, hash);
1091 	hlist_for_each_entry_rcu(flow, head, hash_node[table->node_ver]) {
1092 		if (flow->mask == mask &&
1093 		    __flow_cmp_masked_key(flow, &masked_key,
1094 					  key_start, key_end))
1095 			return flow;
1096 	}
1097 	return NULL;
1098 }
1099 
1100 struct sw_flow *ovs_flow_lookup(struct flow_table *tbl,
1101 				const struct sw_flow_key *key)
1102 {
1103 	struct sw_flow *flow = NULL;
1104 	struct sw_flow_mask *mask;
1105 
1106 	list_for_each_entry_rcu(mask, tbl->mask_list, list) {
1107 		flow = ovs_masked_flow_lookup(tbl, key, mask);
1108 		if (flow)  /* Found */
1109 			break;
1110 	}
1111 
1112 	return flow;
1113 }
1114 
1115 
1116 void ovs_flow_insert(struct flow_table *table, struct sw_flow *flow)
1117 {
1118 	flow->hash = ovs_flow_hash(&flow->key, flow->mask->range.start,
1119 			flow->mask->range.end);
1120 	__tbl_insert(table, flow);
1121 }
1122 
1123 void ovs_flow_remove(struct flow_table *table, struct sw_flow *flow)
1124 {
1125 	BUG_ON(table->count == 0);
1126 	hlist_del_rcu(&flow->hash_node[table->node_ver]);
1127 	table->count--;
1128 }
1129 
1130 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
1131 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
1132 	[OVS_KEY_ATTR_ENCAP] = -1,
1133 	[OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
1134 	[OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
1135 	[OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
1136 	[OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
1137 	[OVS_KEY_ATTR_VLAN] = sizeof(__be16),
1138 	[OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
1139 	[OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
1140 	[OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
1141 	[OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
1142 	[OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
1143 	[OVS_KEY_ATTR_SCTP] = sizeof(struct ovs_key_sctp),
1144 	[OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
1145 	[OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
1146 	[OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
1147 	[OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
1148 	[OVS_KEY_ATTR_TUNNEL] = -1,
1149 };
1150 
1151 static bool is_all_zero(const u8 *fp, size_t size)
1152 {
1153 	int i;
1154 
1155 	if (!fp)
1156 		return false;
1157 
1158 	for (i = 0; i < size; i++)
1159 		if (fp[i])
1160 			return false;
1161 
1162 	return true;
1163 }
1164 
1165 static int __parse_flow_nlattrs(const struct nlattr *attr,
1166 			      const struct nlattr *a[],
1167 			      u64 *attrsp, bool nz)
1168 {
1169 	const struct nlattr *nla;
1170 	u32 attrs;
1171 	int rem;
1172 
1173 	attrs = *attrsp;
1174 	nla_for_each_nested(nla, attr, rem) {
1175 		u16 type = nla_type(nla);
1176 		int expected_len;
1177 
1178 		if (type > OVS_KEY_ATTR_MAX) {
1179 			OVS_NLERR("Unknown key attribute (type=%d, max=%d).\n",
1180 				  type, OVS_KEY_ATTR_MAX);
1181 			return -EINVAL;
1182 		}
1183 
1184 		if (attrs & (1 << type)) {
1185 			OVS_NLERR("Duplicate key attribute (type %d).\n", type);
1186 			return -EINVAL;
1187 		}
1188 
1189 		expected_len = ovs_key_lens[type];
1190 		if (nla_len(nla) != expected_len && expected_len != -1) {
1191 			OVS_NLERR("Key attribute has unexpected length (type=%d"
1192 				  ", length=%d, expected=%d).\n", type,
1193 				  nla_len(nla), expected_len);
1194 			return -EINVAL;
1195 		}
1196 
1197 		if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
1198 			attrs |= 1 << type;
1199 			a[type] = nla;
1200 		}
1201 	}
1202 	if (rem) {
1203 		OVS_NLERR("Message has %d unknown bytes.\n", rem);
1204 		return -EINVAL;
1205 	}
1206 
1207 	*attrsp = attrs;
1208 	return 0;
1209 }
1210 
1211 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
1212 			      const struct nlattr *a[], u64 *attrsp)
1213 {
1214 	return __parse_flow_nlattrs(attr, a, attrsp, true);
1215 }
1216 
1217 static int parse_flow_nlattrs(const struct nlattr *attr,
1218 			      const struct nlattr *a[], u64 *attrsp)
1219 {
1220 	return __parse_flow_nlattrs(attr, a, attrsp, false);
1221 }
1222 
1223 int ovs_ipv4_tun_from_nlattr(const struct nlattr *attr,
1224 			     struct sw_flow_match *match, bool is_mask)
1225 {
1226 	struct nlattr *a;
1227 	int rem;
1228 	bool ttl = false;
1229 	__be16 tun_flags = 0;
1230 
1231 	nla_for_each_nested(a, attr, rem) {
1232 		int type = nla_type(a);
1233 		static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
1234 			[OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
1235 			[OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
1236 			[OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
1237 			[OVS_TUNNEL_KEY_ATTR_TOS] = 1,
1238 			[OVS_TUNNEL_KEY_ATTR_TTL] = 1,
1239 			[OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
1240 			[OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
1241 		};
1242 
1243 		if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
1244 			OVS_NLERR("Unknown IPv4 tunnel attribute (type=%d, max=%d).\n",
1245 			type, OVS_TUNNEL_KEY_ATTR_MAX);
1246 			return -EINVAL;
1247 		}
1248 
1249 		if (ovs_tunnel_key_lens[type] != nla_len(a)) {
1250 			OVS_NLERR("IPv4 tunnel attribute type has unexpected "
1251 				  " length (type=%d, length=%d, expected=%d).\n",
1252 				  type, nla_len(a), ovs_tunnel_key_lens[type]);
1253 			return -EINVAL;
1254 		}
1255 
1256 		switch (type) {
1257 		case OVS_TUNNEL_KEY_ATTR_ID:
1258 			SW_FLOW_KEY_PUT(match, tun_key.tun_id,
1259 					nla_get_be64(a), is_mask);
1260 			tun_flags |= TUNNEL_KEY;
1261 			break;
1262 		case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1263 			SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
1264 					nla_get_be32(a), is_mask);
1265 			break;
1266 		case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1267 			SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
1268 					nla_get_be32(a), is_mask);
1269 			break;
1270 		case OVS_TUNNEL_KEY_ATTR_TOS:
1271 			SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
1272 					nla_get_u8(a), is_mask);
1273 			break;
1274 		case OVS_TUNNEL_KEY_ATTR_TTL:
1275 			SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
1276 					nla_get_u8(a), is_mask);
1277 			ttl = true;
1278 			break;
1279 		case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1280 			tun_flags |= TUNNEL_DONT_FRAGMENT;
1281 			break;
1282 		case OVS_TUNNEL_KEY_ATTR_CSUM:
1283 			tun_flags |= TUNNEL_CSUM;
1284 			break;
1285 		default:
1286 			return -EINVAL;
1287 		}
1288 	}
1289 
1290 	SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
1291 
1292 	if (rem > 0) {
1293 		OVS_NLERR("IPv4 tunnel attribute has %d unknown bytes.\n", rem);
1294 		return -EINVAL;
1295 	}
1296 
1297 	if (!is_mask) {
1298 		if (!match->key->tun_key.ipv4_dst) {
1299 			OVS_NLERR("IPv4 tunnel destination address is zero.\n");
1300 			return -EINVAL;
1301 		}
1302 
1303 		if (!ttl) {
1304 			OVS_NLERR("IPv4 tunnel TTL not specified.\n");
1305 			return -EINVAL;
1306 		}
1307 	}
1308 
1309 	return 0;
1310 }
1311 
1312 int ovs_ipv4_tun_to_nlattr(struct sk_buff *skb,
1313 			   const struct ovs_key_ipv4_tunnel *tun_key,
1314 			   const struct ovs_key_ipv4_tunnel *output)
1315 {
1316 	struct nlattr *nla;
1317 
1318 	nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
1319 	if (!nla)
1320 		return -EMSGSIZE;
1321 
1322 	if (output->tun_flags & TUNNEL_KEY &&
1323 	    nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
1324 		return -EMSGSIZE;
1325 	if (output->ipv4_src &&
1326 		nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
1327 		return -EMSGSIZE;
1328 	if (output->ipv4_dst &&
1329 		nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
1330 		return -EMSGSIZE;
1331 	if (output->ipv4_tos &&
1332 		nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
1333 		return -EMSGSIZE;
1334 	if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
1335 		return -EMSGSIZE;
1336 	if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
1337 		nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
1338 		return -EMSGSIZE;
1339 	if ((output->tun_flags & TUNNEL_CSUM) &&
1340 		nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
1341 		return -EMSGSIZE;
1342 
1343 	nla_nest_end(skb, nla);
1344 	return 0;
1345 }
1346 
1347 static int metadata_from_nlattrs(struct sw_flow_match *match,  u64 *attrs,
1348 		const struct nlattr **a, bool is_mask)
1349 {
1350 	if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1351 		SW_FLOW_KEY_PUT(match, phy.priority,
1352 			  nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1353 		*attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1354 	}
1355 
1356 	if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1357 		u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1358 
1359 		if (is_mask)
1360 			in_port = 0xffffffff; /* Always exact match in_port. */
1361 		else if (in_port >= DP_MAX_PORTS)
1362 			return -EINVAL;
1363 
1364 		SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1365 		*attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1366 	} else if (!is_mask) {
1367 		SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1368 	}
1369 
1370 	if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1371 		uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1372 
1373 		SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1374 		*attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1375 	}
1376 	if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
1377 		if (ovs_ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1378 					is_mask))
1379 			return -EINVAL;
1380 		*attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
1381 	}
1382 	return 0;
1383 }
1384 
1385 static int ovs_key_from_nlattrs(struct sw_flow_match *match,  u64 attrs,
1386 		const struct nlattr **a, bool is_mask)
1387 {
1388 	int err;
1389 	u64 orig_attrs = attrs;
1390 
1391 	err = metadata_from_nlattrs(match, &attrs, a, is_mask);
1392 	if (err)
1393 		return err;
1394 
1395 	if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
1396 		const struct ovs_key_ethernet *eth_key;
1397 
1398 		eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1399 		SW_FLOW_KEY_MEMCPY(match, eth.src,
1400 				eth_key->eth_src, ETH_ALEN, is_mask);
1401 		SW_FLOW_KEY_MEMCPY(match, eth.dst,
1402 				eth_key->eth_dst, ETH_ALEN, is_mask);
1403 		attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1404 	}
1405 
1406 	if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
1407 		__be16 tci;
1408 
1409 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1410 		if (!(tci & htons(VLAN_TAG_PRESENT))) {
1411 			if (is_mask)
1412 				OVS_NLERR("VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.\n");
1413 			else
1414 				OVS_NLERR("VLAN TCI does not have VLAN_TAG_PRESENT bit set.\n");
1415 
1416 			return -EINVAL;
1417 		}
1418 
1419 		SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
1420 		attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
1421 	} else if (!is_mask)
1422 		SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1423 
1424 	if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1425 		__be16 eth_type;
1426 
1427 		eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1428 		if (is_mask) {
1429 			/* Always exact match EtherType. */
1430 			eth_type = htons(0xffff);
1431 		} else if (ntohs(eth_type) < ETH_P_802_3_MIN) {
1432 			OVS_NLERR("EtherType is less than minimum (type=%x, min=%x).\n",
1433 					ntohs(eth_type), ETH_P_802_3_MIN);
1434 			return -EINVAL;
1435 		}
1436 
1437 		SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1438 		attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1439 	} else if (!is_mask) {
1440 		SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1441 	}
1442 
1443 	if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
1444 		const struct ovs_key_ipv4 *ipv4_key;
1445 
1446 		ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1447 		if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1448 			OVS_NLERR("Unknown IPv4 fragment type (value=%d, max=%d).\n",
1449 				ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1450 			return -EINVAL;
1451 		}
1452 		SW_FLOW_KEY_PUT(match, ip.proto,
1453 				ipv4_key->ipv4_proto, is_mask);
1454 		SW_FLOW_KEY_PUT(match, ip.tos,
1455 				ipv4_key->ipv4_tos, is_mask);
1456 		SW_FLOW_KEY_PUT(match, ip.ttl,
1457 				ipv4_key->ipv4_ttl, is_mask);
1458 		SW_FLOW_KEY_PUT(match, ip.frag,
1459 				ipv4_key->ipv4_frag, is_mask);
1460 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1461 				ipv4_key->ipv4_src, is_mask);
1462 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1463 				ipv4_key->ipv4_dst, is_mask);
1464 		attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1465 	}
1466 
1467 	if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
1468 		const struct ovs_key_ipv6 *ipv6_key;
1469 
1470 		ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1471 		if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1472 			OVS_NLERR("Unknown IPv6 fragment type (value=%d, max=%d).\n",
1473 				ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1474 			return -EINVAL;
1475 		}
1476 		SW_FLOW_KEY_PUT(match, ipv6.label,
1477 				ipv6_key->ipv6_label, is_mask);
1478 		SW_FLOW_KEY_PUT(match, ip.proto,
1479 				ipv6_key->ipv6_proto, is_mask);
1480 		SW_FLOW_KEY_PUT(match, ip.tos,
1481 				ipv6_key->ipv6_tclass, is_mask);
1482 		SW_FLOW_KEY_PUT(match, ip.ttl,
1483 				ipv6_key->ipv6_hlimit, is_mask);
1484 		SW_FLOW_KEY_PUT(match, ip.frag,
1485 				ipv6_key->ipv6_frag, is_mask);
1486 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1487 				ipv6_key->ipv6_src,
1488 				sizeof(match->key->ipv6.addr.src),
1489 				is_mask);
1490 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1491 				ipv6_key->ipv6_dst,
1492 				sizeof(match->key->ipv6.addr.dst),
1493 				is_mask);
1494 
1495 		attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1496 	}
1497 
1498 	if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
1499 		const struct ovs_key_arp *arp_key;
1500 
1501 		arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1502 		if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1503 			OVS_NLERR("Unknown ARP opcode (opcode=%d).\n",
1504 				  arp_key->arp_op);
1505 			return -EINVAL;
1506 		}
1507 
1508 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1509 				arp_key->arp_sip, is_mask);
1510 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1511 			arp_key->arp_tip, is_mask);
1512 		SW_FLOW_KEY_PUT(match, ip.proto,
1513 				ntohs(arp_key->arp_op), is_mask);
1514 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1515 				arp_key->arp_sha, ETH_ALEN, is_mask);
1516 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1517 				arp_key->arp_tha, ETH_ALEN, is_mask);
1518 
1519 		attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1520 	}
1521 
1522 	if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
1523 		const struct ovs_key_tcp *tcp_key;
1524 
1525 		tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1526 		if (orig_attrs & (1 << OVS_KEY_ATTR_IPV4)) {
1527 			SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1528 					tcp_key->tcp_src, is_mask);
1529 			SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1530 					tcp_key->tcp_dst, is_mask);
1531 		} else {
1532 			SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1533 					tcp_key->tcp_src, is_mask);
1534 			SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1535 					tcp_key->tcp_dst, is_mask);
1536 		}
1537 		attrs &= ~(1 << OVS_KEY_ATTR_TCP);
1538 	}
1539 
1540 	if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
1541 		const struct ovs_key_udp *udp_key;
1542 
1543 		udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1544 		if (orig_attrs & (1 << OVS_KEY_ATTR_IPV4)) {
1545 			SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1546 					udp_key->udp_src, is_mask);
1547 			SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1548 					udp_key->udp_dst, is_mask);
1549 		} else {
1550 			SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1551 					udp_key->udp_src, is_mask);
1552 			SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1553 					udp_key->udp_dst, is_mask);
1554 		}
1555 		attrs &= ~(1 << OVS_KEY_ATTR_UDP);
1556 	}
1557 
1558 	if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
1559 		const struct ovs_key_sctp *sctp_key;
1560 
1561 		sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1562 		if (orig_attrs & (1 << OVS_KEY_ATTR_IPV4)) {
1563 			SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1564 					sctp_key->sctp_src, is_mask);
1565 			SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1566 					sctp_key->sctp_dst, is_mask);
1567 		} else {
1568 			SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1569 					sctp_key->sctp_src, is_mask);
1570 			SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1571 					sctp_key->sctp_dst, is_mask);
1572 		}
1573 		attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
1574 	}
1575 
1576 	if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1577 		const struct ovs_key_icmp *icmp_key;
1578 
1579 		icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1580 		SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1581 				htons(icmp_key->icmp_type), is_mask);
1582 		SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1583 				htons(icmp_key->icmp_code), is_mask);
1584 		attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1585 	}
1586 
1587 	if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1588 		const struct ovs_key_icmpv6 *icmpv6_key;
1589 
1590 		icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1591 		SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1592 				htons(icmpv6_key->icmpv6_type), is_mask);
1593 		SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1594 				htons(icmpv6_key->icmpv6_code), is_mask);
1595 		attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1596 	}
1597 
1598 	if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1599 		const struct ovs_key_nd *nd_key;
1600 
1601 		nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1602 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1603 			nd_key->nd_target,
1604 			sizeof(match->key->ipv6.nd.target),
1605 			is_mask);
1606 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1607 			nd_key->nd_sll, ETH_ALEN, is_mask);
1608 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1609 				nd_key->nd_tll, ETH_ALEN, is_mask);
1610 		attrs &= ~(1 << OVS_KEY_ATTR_ND);
1611 	}
1612 
1613 	if (attrs != 0)
1614 		return -EINVAL;
1615 
1616 	return 0;
1617 }
1618 
1619 /**
1620  * ovs_match_from_nlattrs - parses Netlink attributes into a flow key and
1621  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1622  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1623  * does not include any don't care bit.
1624  * @match: receives the extracted flow match information.
1625  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1626  * sequence. The fields should of the packet that triggered the creation
1627  * of this flow.
1628  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1629  * attribute specifies the mask field of the wildcarded flow.
1630  */
1631 int ovs_match_from_nlattrs(struct sw_flow_match *match,
1632 			   const struct nlattr *key,
1633 			   const struct nlattr *mask)
1634 {
1635 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1636 	const struct nlattr *encap;
1637 	u64 key_attrs = 0;
1638 	u64 mask_attrs = 0;
1639 	bool encap_valid = false;
1640 	int err;
1641 
1642 	err = parse_flow_nlattrs(key, a, &key_attrs);
1643 	if (err)
1644 		return err;
1645 
1646 	if ((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
1647 	    (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
1648 	    (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1649 		__be16 tci;
1650 
1651 		if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
1652 		      (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
1653 			OVS_NLERR("Invalid Vlan frame.\n");
1654 			return -EINVAL;
1655 		}
1656 
1657 		key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1658 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1659 		encap = a[OVS_KEY_ATTR_ENCAP];
1660 		key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1661 		encap_valid = true;
1662 
1663 		if (tci & htons(VLAN_TAG_PRESENT)) {
1664 			err = parse_flow_nlattrs(encap, a, &key_attrs);
1665 			if (err)
1666 				return err;
1667 		} else if (!tci) {
1668 			/* Corner case for truncated 802.1Q header. */
1669 			if (nla_len(encap)) {
1670 				OVS_NLERR("Truncated 802.1Q header has non-zero encap attribute.\n");
1671 				return -EINVAL;
1672 			}
1673 		} else {
1674 			OVS_NLERR("Encap attribute is set for a non-VLAN frame.\n");
1675 			return  -EINVAL;
1676 		}
1677 	}
1678 
1679 	err = ovs_key_from_nlattrs(match, key_attrs, a, false);
1680 	if (err)
1681 		return err;
1682 
1683 	if (mask) {
1684 		err = parse_flow_mask_nlattrs(mask, a, &mask_attrs);
1685 		if (err)
1686 			return err;
1687 
1688 		if (mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP)  {
1689 			__be16 eth_type = 0;
1690 			__be16 tci = 0;
1691 
1692 			if (!encap_valid) {
1693 				OVS_NLERR("Encap mask attribute is set for non-VLAN frame.\n");
1694 				return  -EINVAL;
1695 			}
1696 
1697 			mask_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1698 			if (a[OVS_KEY_ATTR_ETHERTYPE])
1699 				eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1700 
1701 			if (eth_type == htons(0xffff)) {
1702 				mask_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1703 				encap = a[OVS_KEY_ATTR_ENCAP];
1704 				err = parse_flow_mask_nlattrs(encap, a, &mask_attrs);
1705 			} else {
1706 				OVS_NLERR("VLAN frames must have an exact match on the TPID (mask=%x).\n",
1707 						ntohs(eth_type));
1708 				return -EINVAL;
1709 			}
1710 
1711 			if (a[OVS_KEY_ATTR_VLAN])
1712 				tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1713 
1714 			if (!(tci & htons(VLAN_TAG_PRESENT))) {
1715 				OVS_NLERR("VLAN tag present bit must have an exact match (tci_mask=%x).\n", ntohs(tci));
1716 				return -EINVAL;
1717 			}
1718 		}
1719 
1720 		err = ovs_key_from_nlattrs(match, mask_attrs, a, true);
1721 		if (err)
1722 			return err;
1723 	} else {
1724 		/* Populate exact match flow's key mask. */
1725 		if (match->mask)
1726 			ovs_sw_flow_mask_set(match->mask, &match->range, 0xff);
1727 	}
1728 
1729 	if (!ovs_match_validate(match, key_attrs, mask_attrs))
1730 		return -EINVAL;
1731 
1732 	return 0;
1733 }
1734 
1735 /**
1736  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1737  * @flow: Receives extracted in_port, priority, tun_key and skb_mark.
1738  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1739  * sequence.
1740  *
1741  * This parses a series of Netlink attributes that form a flow key, which must
1742  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1743  * get the metadata, that is, the parts of the flow key that cannot be
1744  * extracted from the packet itself.
1745  */
1746 
1747 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow,
1748 		const struct nlattr *attr)
1749 {
1750 	struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1751 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1752 	u64 attrs = 0;
1753 	int err;
1754 	struct sw_flow_match match;
1755 
1756 	flow->key.phy.in_port = DP_MAX_PORTS;
1757 	flow->key.phy.priority = 0;
1758 	flow->key.phy.skb_mark = 0;
1759 	memset(tun_key, 0, sizeof(flow->key.tun_key));
1760 
1761 	err = parse_flow_nlattrs(attr, a, &attrs);
1762 	if (err)
1763 		return -EINVAL;
1764 
1765 	memset(&match, 0, sizeof(match));
1766 	match.key = &flow->key;
1767 
1768 	err = metadata_from_nlattrs(&match, &attrs, a, false);
1769 	if (err)
1770 		return err;
1771 
1772 	return 0;
1773 }
1774 
1775 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey,
1776 		const struct sw_flow_key *output, struct sk_buff *skb)
1777 {
1778 	struct ovs_key_ethernet *eth_key;
1779 	struct nlattr *nla, *encap;
1780 	bool is_mask = (swkey != output);
1781 
1782 	if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1783 		goto nla_put_failure;
1784 
1785 	if ((swkey->tun_key.ipv4_dst || is_mask) &&
1786 	    ovs_ipv4_tun_to_nlattr(skb, &swkey->tun_key, &output->tun_key))
1787 		goto nla_put_failure;
1788 
1789 	if (swkey->phy.in_port == DP_MAX_PORTS) {
1790 		if (is_mask && (output->phy.in_port == 0xffff))
1791 			if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1792 				goto nla_put_failure;
1793 	} else {
1794 		u16 upper_u16;
1795 		upper_u16 = !is_mask ? 0 : 0xffff;
1796 
1797 		if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1798 				(upper_u16 << 16) | output->phy.in_port))
1799 			goto nla_put_failure;
1800 	}
1801 
1802 	if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1803 		goto nla_put_failure;
1804 
1805 	nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1806 	if (!nla)
1807 		goto nla_put_failure;
1808 
1809 	eth_key = nla_data(nla);
1810 	memcpy(eth_key->eth_src, output->eth.src, ETH_ALEN);
1811 	memcpy(eth_key->eth_dst, output->eth.dst, ETH_ALEN);
1812 
1813 	if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1814 		__be16 eth_type;
1815 		eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1816 		if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1817 		    nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1818 			goto nla_put_failure;
1819 		encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1820 		if (!swkey->eth.tci)
1821 			goto unencap;
1822 	} else
1823 		encap = NULL;
1824 
1825 	if (swkey->eth.type == htons(ETH_P_802_2)) {
1826 		/*
1827 		 * Ethertype 802.2 is represented in the netlink with omitted
1828 		 * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1829 		 * 0xffff in the mask attribute.  Ethertype can also
1830 		 * be wildcarded.
1831 		 */
1832 		if (is_mask && output->eth.type)
1833 			if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1834 						output->eth.type))
1835 				goto nla_put_failure;
1836 		goto unencap;
1837 	}
1838 
1839 	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1840 		goto nla_put_failure;
1841 
1842 	if (swkey->eth.type == htons(ETH_P_IP)) {
1843 		struct ovs_key_ipv4 *ipv4_key;
1844 
1845 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1846 		if (!nla)
1847 			goto nla_put_failure;
1848 		ipv4_key = nla_data(nla);
1849 		ipv4_key->ipv4_src = output->ipv4.addr.src;
1850 		ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1851 		ipv4_key->ipv4_proto = output->ip.proto;
1852 		ipv4_key->ipv4_tos = output->ip.tos;
1853 		ipv4_key->ipv4_ttl = output->ip.ttl;
1854 		ipv4_key->ipv4_frag = output->ip.frag;
1855 	} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1856 		struct ovs_key_ipv6 *ipv6_key;
1857 
1858 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1859 		if (!nla)
1860 			goto nla_put_failure;
1861 		ipv6_key = nla_data(nla);
1862 		memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1863 				sizeof(ipv6_key->ipv6_src));
1864 		memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1865 				sizeof(ipv6_key->ipv6_dst));
1866 		ipv6_key->ipv6_label = output->ipv6.label;
1867 		ipv6_key->ipv6_proto = output->ip.proto;
1868 		ipv6_key->ipv6_tclass = output->ip.tos;
1869 		ipv6_key->ipv6_hlimit = output->ip.ttl;
1870 		ipv6_key->ipv6_frag = output->ip.frag;
1871 	} else if (swkey->eth.type == htons(ETH_P_ARP) ||
1872 		   swkey->eth.type == htons(ETH_P_RARP)) {
1873 		struct ovs_key_arp *arp_key;
1874 
1875 		nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1876 		if (!nla)
1877 			goto nla_put_failure;
1878 		arp_key = nla_data(nla);
1879 		memset(arp_key, 0, sizeof(struct ovs_key_arp));
1880 		arp_key->arp_sip = output->ipv4.addr.src;
1881 		arp_key->arp_tip = output->ipv4.addr.dst;
1882 		arp_key->arp_op = htons(output->ip.proto);
1883 		memcpy(arp_key->arp_sha, output->ipv4.arp.sha, ETH_ALEN);
1884 		memcpy(arp_key->arp_tha, output->ipv4.arp.tha, ETH_ALEN);
1885 	}
1886 
1887 	if ((swkey->eth.type == htons(ETH_P_IP) ||
1888 	     swkey->eth.type == htons(ETH_P_IPV6)) &&
1889 	     swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1890 
1891 		if (swkey->ip.proto == IPPROTO_TCP) {
1892 			struct ovs_key_tcp *tcp_key;
1893 
1894 			nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1895 			if (!nla)
1896 				goto nla_put_failure;
1897 			tcp_key = nla_data(nla);
1898 			if (swkey->eth.type == htons(ETH_P_IP)) {
1899 				tcp_key->tcp_src = output->ipv4.tp.src;
1900 				tcp_key->tcp_dst = output->ipv4.tp.dst;
1901 			} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1902 				tcp_key->tcp_src = output->ipv6.tp.src;
1903 				tcp_key->tcp_dst = output->ipv6.tp.dst;
1904 			}
1905 		} else if (swkey->ip.proto == IPPROTO_UDP) {
1906 			struct ovs_key_udp *udp_key;
1907 
1908 			nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1909 			if (!nla)
1910 				goto nla_put_failure;
1911 			udp_key = nla_data(nla);
1912 			if (swkey->eth.type == htons(ETH_P_IP)) {
1913 				udp_key->udp_src = output->ipv4.tp.src;
1914 				udp_key->udp_dst = output->ipv4.tp.dst;
1915 			} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1916 				udp_key->udp_src = output->ipv6.tp.src;
1917 				udp_key->udp_dst = output->ipv6.tp.dst;
1918 			}
1919 		} else if (swkey->ip.proto == IPPROTO_SCTP) {
1920 			struct ovs_key_sctp *sctp_key;
1921 
1922 			nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1923 			if (!nla)
1924 				goto nla_put_failure;
1925 			sctp_key = nla_data(nla);
1926 			if (swkey->eth.type == htons(ETH_P_IP)) {
1927 				sctp_key->sctp_src = swkey->ipv4.tp.src;
1928 				sctp_key->sctp_dst = swkey->ipv4.tp.dst;
1929 			} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1930 				sctp_key->sctp_src = swkey->ipv6.tp.src;
1931 				sctp_key->sctp_dst = swkey->ipv6.tp.dst;
1932 			}
1933 		} else if (swkey->eth.type == htons(ETH_P_IP) &&
1934 			   swkey->ip.proto == IPPROTO_ICMP) {
1935 			struct ovs_key_icmp *icmp_key;
1936 
1937 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1938 			if (!nla)
1939 				goto nla_put_failure;
1940 			icmp_key = nla_data(nla);
1941 			icmp_key->icmp_type = ntohs(output->ipv4.tp.src);
1942 			icmp_key->icmp_code = ntohs(output->ipv4.tp.dst);
1943 		} else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1944 			   swkey->ip.proto == IPPROTO_ICMPV6) {
1945 			struct ovs_key_icmpv6 *icmpv6_key;
1946 
1947 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1948 						sizeof(*icmpv6_key));
1949 			if (!nla)
1950 				goto nla_put_failure;
1951 			icmpv6_key = nla_data(nla);
1952 			icmpv6_key->icmpv6_type = ntohs(output->ipv6.tp.src);
1953 			icmpv6_key->icmpv6_code = ntohs(output->ipv6.tp.dst);
1954 
1955 			if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1956 			    icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1957 				struct ovs_key_nd *nd_key;
1958 
1959 				nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1960 				if (!nla)
1961 					goto nla_put_failure;
1962 				nd_key = nla_data(nla);
1963 				memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1964 							sizeof(nd_key->nd_target));
1965 				memcpy(nd_key->nd_sll, output->ipv6.nd.sll, ETH_ALEN);
1966 				memcpy(nd_key->nd_tll, output->ipv6.nd.tll, ETH_ALEN);
1967 			}
1968 		}
1969 	}
1970 
1971 unencap:
1972 	if (encap)
1973 		nla_nest_end(skb, encap);
1974 
1975 	return 0;
1976 
1977 nla_put_failure:
1978 	return -EMSGSIZE;
1979 }
1980 
1981 /* Initializes the flow module.
1982  * Returns zero if successful or a negative error code. */
1983 int ovs_flow_init(void)
1984 {
1985 	BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long));
1986 	BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
1987 
1988 	flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1989 					0, NULL);
1990 	if (flow_cache == NULL)
1991 		return -ENOMEM;
1992 
1993 	return 0;
1994 }
1995 
1996 /* Uninitializes the flow module. */
1997 void ovs_flow_exit(void)
1998 {
1999 	kmem_cache_destroy(flow_cache);
2000 }
2001 
2002 struct sw_flow_mask *ovs_sw_flow_mask_alloc(void)
2003 {
2004 	struct sw_flow_mask *mask;
2005 
2006 	mask = kmalloc(sizeof(*mask), GFP_KERNEL);
2007 	if (mask)
2008 		mask->ref_count = 0;
2009 
2010 	return mask;
2011 }
2012 
2013 void ovs_sw_flow_mask_add_ref(struct sw_flow_mask *mask)
2014 {
2015 	mask->ref_count++;
2016 }
2017 
2018 void ovs_sw_flow_mask_del_ref(struct sw_flow_mask *mask, bool deferred)
2019 {
2020 	if (!mask)
2021 		return;
2022 
2023 	BUG_ON(!mask->ref_count);
2024 	mask->ref_count--;
2025 
2026 	if (!mask->ref_count) {
2027 		list_del_rcu(&mask->list);
2028 		if (deferred)
2029 			kfree_rcu(mask, rcu);
2030 		else
2031 			kfree(mask);
2032 	}
2033 }
2034 
2035 static bool ovs_sw_flow_mask_equal(const struct sw_flow_mask *a,
2036 		const struct sw_flow_mask *b)
2037 {
2038 	u8 *a_ = (u8 *)&a->key + a->range.start;
2039 	u8 *b_ = (u8 *)&b->key + b->range.start;
2040 
2041 	return  (a->range.end == b->range.end)
2042 		&& (a->range.start == b->range.start)
2043 		&& (memcmp(a_, b_, range_n_bytes(&a->range)) == 0);
2044 }
2045 
2046 struct sw_flow_mask *ovs_sw_flow_mask_find(const struct flow_table *tbl,
2047                                            const struct sw_flow_mask *mask)
2048 {
2049 	struct list_head *ml;
2050 
2051 	list_for_each(ml, tbl->mask_list) {
2052 		struct sw_flow_mask *m;
2053 		m = container_of(ml, struct sw_flow_mask, list);
2054 		if (ovs_sw_flow_mask_equal(mask, m))
2055 			return m;
2056 	}
2057 
2058 	return NULL;
2059 }
2060 
2061 /**
2062  * add a new mask into the mask list.
2063  * The caller needs to make sure that 'mask' is not the same
2064  * as any masks that are already on the list.
2065  */
2066 void ovs_sw_flow_mask_insert(struct flow_table *tbl, struct sw_flow_mask *mask)
2067 {
2068 	list_add_rcu(&mask->list, tbl->mask_list);
2069 }
2070 
2071 /**
2072  * Set 'range' fields in the mask to the value of 'val'.
2073  */
2074 static void ovs_sw_flow_mask_set(struct sw_flow_mask *mask,
2075 		struct sw_flow_key_range *range, u8 val)
2076 {
2077 	u8 *m = (u8 *)&mask->key + range->start;
2078 
2079 	mask->range = *range;
2080 	memset(m, val, range_n_bytes(range));
2081 }
2082