xref: /openbmc/linux/net/openvswitch/flow.c (revision 329f45bc)
1 /*
2  * Copyright (c) 2007-2014 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 <linux/uaccess.h>
20 #include <linux/netdevice.h>
21 #include <linux/etherdevice.h>
22 #include <linux/if_ether.h>
23 #include <linux/if_vlan.h>
24 #include <net/llc_pdu.h>
25 #include <linux/kernel.h>
26 #include <linux/jhash.h>
27 #include <linux/jiffies.h>
28 #include <linux/llc.h>
29 #include <linux/module.h>
30 #include <linux/in.h>
31 #include <linux/rcupdate.h>
32 #include <linux/cpumask.h>
33 #include <linux/if_arp.h>
34 #include <linux/ip.h>
35 #include <linux/ipv6.h>
36 #include <linux/mpls.h>
37 #include <linux/sctp.h>
38 #include <linux/smp.h>
39 #include <linux/tcp.h>
40 #include <linux/udp.h>
41 #include <linux/icmp.h>
42 #include <linux/icmpv6.h>
43 #include <linux/rculist.h>
44 #include <net/ip.h>
45 #include <net/ip_tunnels.h>
46 #include <net/ipv6.h>
47 #include <net/mpls.h>
48 #include <net/ndisc.h>
49 
50 #include "conntrack.h"
51 #include "datapath.h"
52 #include "flow.h"
53 #include "flow_netlink.h"
54 #include "vport.h"
55 
56 u64 ovs_flow_used_time(unsigned long flow_jiffies)
57 {
58 	struct timespec cur_ts;
59 	u64 cur_ms, idle_ms;
60 
61 	ktime_get_ts(&cur_ts);
62 	idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
63 	cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
64 		 cur_ts.tv_nsec / NSEC_PER_MSEC;
65 
66 	return cur_ms - idle_ms;
67 }
68 
69 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
70 
71 void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
72 			   const struct sk_buff *skb)
73 {
74 	struct flow_stats *stats;
75 	int node = numa_node_id();
76 	int cpu = smp_processor_id();
77 	int len = skb->len + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
78 
79 	stats = rcu_dereference(flow->stats[cpu]);
80 
81 	/* Check if already have CPU-specific stats. */
82 	if (likely(stats)) {
83 		spin_lock(&stats->lock);
84 		/* Mark if we write on the pre-allocated stats. */
85 		if (cpu == 0 && unlikely(flow->stats_last_writer != cpu))
86 			flow->stats_last_writer = cpu;
87 	} else {
88 		stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
89 		spin_lock(&stats->lock);
90 
91 		/* If the current CPU is the only writer on the
92 		 * pre-allocated stats keep using them.
93 		 */
94 		if (unlikely(flow->stats_last_writer != cpu)) {
95 			/* A previous locker may have already allocated the
96 			 * stats, so we need to check again.  If CPU-specific
97 			 * stats were already allocated, we update the pre-
98 			 * allocated stats as we have already locked them.
99 			 */
100 			if (likely(flow->stats_last_writer != -1) &&
101 			    likely(!rcu_access_pointer(flow->stats[cpu]))) {
102 				/* Try to allocate CPU-specific stats. */
103 				struct flow_stats *new_stats;
104 
105 				new_stats =
106 					kmem_cache_alloc_node(flow_stats_cache,
107 							      GFP_NOWAIT |
108 							      __GFP_THISNODE |
109 							      __GFP_NOWARN |
110 							      __GFP_NOMEMALLOC,
111 							      node);
112 				if (likely(new_stats)) {
113 					new_stats->used = jiffies;
114 					new_stats->packet_count = 1;
115 					new_stats->byte_count = len;
116 					new_stats->tcp_flags = tcp_flags;
117 					spin_lock_init(&new_stats->lock);
118 
119 					rcu_assign_pointer(flow->stats[cpu],
120 							   new_stats);
121 					goto unlock;
122 				}
123 			}
124 			flow->stats_last_writer = cpu;
125 		}
126 	}
127 
128 	stats->used = jiffies;
129 	stats->packet_count++;
130 	stats->byte_count += len;
131 	stats->tcp_flags |= tcp_flags;
132 unlock:
133 	spin_unlock(&stats->lock);
134 }
135 
136 /* Must be called with rcu_read_lock or ovs_mutex. */
137 void ovs_flow_stats_get(const struct sw_flow *flow,
138 			struct ovs_flow_stats *ovs_stats,
139 			unsigned long *used, __be16 *tcp_flags)
140 {
141 	int cpu;
142 
143 	*used = 0;
144 	*tcp_flags = 0;
145 	memset(ovs_stats, 0, sizeof(*ovs_stats));
146 
147 	/* We open code this to make sure cpu 0 is always considered */
148 	for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, cpu_possible_mask)) {
149 		struct flow_stats *stats = rcu_dereference_ovsl(flow->stats[cpu]);
150 
151 		if (stats) {
152 			/* Local CPU may write on non-local stats, so we must
153 			 * block bottom-halves here.
154 			 */
155 			spin_lock_bh(&stats->lock);
156 			if (!*used || time_after(stats->used, *used))
157 				*used = stats->used;
158 			*tcp_flags |= stats->tcp_flags;
159 			ovs_stats->n_packets += stats->packet_count;
160 			ovs_stats->n_bytes += stats->byte_count;
161 			spin_unlock_bh(&stats->lock);
162 		}
163 	}
164 }
165 
166 /* Called with ovs_mutex. */
167 void ovs_flow_stats_clear(struct sw_flow *flow)
168 {
169 	int cpu;
170 
171 	/* We open code this to make sure cpu 0 is always considered */
172 	for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, cpu_possible_mask)) {
173 		struct flow_stats *stats = ovsl_dereference(flow->stats[cpu]);
174 
175 		if (stats) {
176 			spin_lock_bh(&stats->lock);
177 			stats->used = 0;
178 			stats->packet_count = 0;
179 			stats->byte_count = 0;
180 			stats->tcp_flags = 0;
181 			spin_unlock_bh(&stats->lock);
182 		}
183 	}
184 }
185 
186 static int check_header(struct sk_buff *skb, int len)
187 {
188 	if (unlikely(skb->len < len))
189 		return -EINVAL;
190 	if (unlikely(!pskb_may_pull(skb, len)))
191 		return -ENOMEM;
192 	return 0;
193 }
194 
195 static bool arphdr_ok(struct sk_buff *skb)
196 {
197 	return pskb_may_pull(skb, skb_network_offset(skb) +
198 				  sizeof(struct arp_eth_header));
199 }
200 
201 static int check_iphdr(struct sk_buff *skb)
202 {
203 	unsigned int nh_ofs = skb_network_offset(skb);
204 	unsigned int ip_len;
205 	int err;
206 
207 	err = check_header(skb, nh_ofs + sizeof(struct iphdr));
208 	if (unlikely(err))
209 		return err;
210 
211 	ip_len = ip_hdrlen(skb);
212 	if (unlikely(ip_len < sizeof(struct iphdr) ||
213 		     skb->len < nh_ofs + ip_len))
214 		return -EINVAL;
215 
216 	skb_set_transport_header(skb, nh_ofs + ip_len);
217 	return 0;
218 }
219 
220 static bool tcphdr_ok(struct sk_buff *skb)
221 {
222 	int th_ofs = skb_transport_offset(skb);
223 	int tcp_len;
224 
225 	if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
226 		return false;
227 
228 	tcp_len = tcp_hdrlen(skb);
229 	if (unlikely(tcp_len < sizeof(struct tcphdr) ||
230 		     skb->len < th_ofs + tcp_len))
231 		return false;
232 
233 	return true;
234 }
235 
236 static bool udphdr_ok(struct sk_buff *skb)
237 {
238 	return pskb_may_pull(skb, skb_transport_offset(skb) +
239 				  sizeof(struct udphdr));
240 }
241 
242 static bool sctphdr_ok(struct sk_buff *skb)
243 {
244 	return pskb_may_pull(skb, skb_transport_offset(skb) +
245 				  sizeof(struct sctphdr));
246 }
247 
248 static bool icmphdr_ok(struct sk_buff *skb)
249 {
250 	return pskb_may_pull(skb, skb_transport_offset(skb) +
251 				  sizeof(struct icmphdr));
252 }
253 
254 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
255 {
256 	unsigned int nh_ofs = skb_network_offset(skb);
257 	unsigned int nh_len;
258 	int payload_ofs;
259 	struct ipv6hdr *nh;
260 	uint8_t nexthdr;
261 	__be16 frag_off;
262 	int err;
263 
264 	err = check_header(skb, nh_ofs + sizeof(*nh));
265 	if (unlikely(err))
266 		return err;
267 
268 	nh = ipv6_hdr(skb);
269 	nexthdr = nh->nexthdr;
270 	payload_ofs = (u8 *)(nh + 1) - skb->data;
271 
272 	key->ip.proto = NEXTHDR_NONE;
273 	key->ip.tos = ipv6_get_dsfield(nh);
274 	key->ip.ttl = nh->hop_limit;
275 	key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
276 	key->ipv6.addr.src = nh->saddr;
277 	key->ipv6.addr.dst = nh->daddr;
278 
279 	payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
280 
281 	if (frag_off) {
282 		if (frag_off & htons(~0x7))
283 			key->ip.frag = OVS_FRAG_TYPE_LATER;
284 		else
285 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
286 	} else {
287 		key->ip.frag = OVS_FRAG_TYPE_NONE;
288 	}
289 
290 	/* Delayed handling of error in ipv6_skip_exthdr() as it
291 	 * always sets frag_off to a valid value which may be
292 	 * used to set key->ip.frag above.
293 	 */
294 	if (unlikely(payload_ofs < 0))
295 		return -EPROTO;
296 
297 	nh_len = payload_ofs - nh_ofs;
298 	skb_set_transport_header(skb, nh_ofs + nh_len);
299 	key->ip.proto = nexthdr;
300 	return nh_len;
301 }
302 
303 static bool icmp6hdr_ok(struct sk_buff *skb)
304 {
305 	return pskb_may_pull(skb, skb_transport_offset(skb) +
306 				  sizeof(struct icmp6hdr));
307 }
308 
309 /**
310  * Parse vlan tag from vlan header.
311  * Returns ERROR on memory error.
312  * Returns 0 if it encounters a non-vlan or incomplete packet.
313  * Returns 1 after successfully parsing vlan tag.
314  */
315 static int parse_vlan_tag(struct sk_buff *skb, struct vlan_head *key_vh)
316 {
317 	struct vlan_head *vh = (struct vlan_head *)skb->data;
318 
319 	if (likely(!eth_type_vlan(vh->tpid)))
320 		return 0;
321 
322 	if (unlikely(skb->len < sizeof(struct vlan_head) + sizeof(__be16)))
323 		return 0;
324 
325 	if (unlikely(!pskb_may_pull(skb, sizeof(struct vlan_head) +
326 				 sizeof(__be16))))
327 		return -ENOMEM;
328 
329 	vh = (struct vlan_head *)skb->data;
330 	key_vh->tci = vh->tci | htons(VLAN_TAG_PRESENT);
331 	key_vh->tpid = vh->tpid;
332 
333 	__skb_pull(skb, sizeof(struct vlan_head));
334 	return 1;
335 }
336 
337 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
338 {
339 	int res;
340 
341 	key->eth.vlan.tci = 0;
342 	key->eth.vlan.tpid = 0;
343 	key->eth.cvlan.tci = 0;
344 	key->eth.cvlan.tpid = 0;
345 
346 	if (skb_vlan_tag_present(skb)) {
347 		key->eth.vlan.tci = htons(skb->vlan_tci);
348 		key->eth.vlan.tpid = skb->vlan_proto;
349 	} else {
350 		/* Parse outer vlan tag in the non-accelerated case. */
351 		res = parse_vlan_tag(skb, &key->eth.vlan);
352 		if (res <= 0)
353 			return res;
354 	}
355 
356 	/* Parse inner vlan tag. */
357 	res = parse_vlan_tag(skb, &key->eth.cvlan);
358 	if (res <= 0)
359 		return res;
360 
361 	return 0;
362 }
363 
364 static __be16 parse_ethertype(struct sk_buff *skb)
365 {
366 	struct llc_snap_hdr {
367 		u8  dsap;  /* Always 0xAA */
368 		u8  ssap;  /* Always 0xAA */
369 		u8  ctrl;
370 		u8  oui[3];
371 		__be16 ethertype;
372 	};
373 	struct llc_snap_hdr *llc;
374 	__be16 proto;
375 
376 	proto = *(__be16 *) skb->data;
377 	__skb_pull(skb, sizeof(__be16));
378 
379 	if (eth_proto_is_802_3(proto))
380 		return proto;
381 
382 	if (skb->len < sizeof(struct llc_snap_hdr))
383 		return htons(ETH_P_802_2);
384 
385 	if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
386 		return htons(0);
387 
388 	llc = (struct llc_snap_hdr *) skb->data;
389 	if (llc->dsap != LLC_SAP_SNAP ||
390 	    llc->ssap != LLC_SAP_SNAP ||
391 	    (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
392 		return htons(ETH_P_802_2);
393 
394 	__skb_pull(skb, sizeof(struct llc_snap_hdr));
395 
396 	if (eth_proto_is_802_3(llc->ethertype))
397 		return llc->ethertype;
398 
399 	return htons(ETH_P_802_2);
400 }
401 
402 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
403 			int nh_len)
404 {
405 	struct icmp6hdr *icmp = icmp6_hdr(skb);
406 
407 	/* The ICMPv6 type and code fields use the 16-bit transport port
408 	 * fields, so we need to store them in 16-bit network byte order.
409 	 */
410 	key->tp.src = htons(icmp->icmp6_type);
411 	key->tp.dst = htons(icmp->icmp6_code);
412 	memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
413 
414 	if (icmp->icmp6_code == 0 &&
415 	    (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
416 	     icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
417 		int icmp_len = skb->len - skb_transport_offset(skb);
418 		struct nd_msg *nd;
419 		int offset;
420 
421 		/* In order to process neighbor discovery options, we need the
422 		 * entire packet.
423 		 */
424 		if (unlikely(icmp_len < sizeof(*nd)))
425 			return 0;
426 
427 		if (unlikely(skb_linearize(skb)))
428 			return -ENOMEM;
429 
430 		nd = (struct nd_msg *)skb_transport_header(skb);
431 		key->ipv6.nd.target = nd->target;
432 
433 		icmp_len -= sizeof(*nd);
434 		offset = 0;
435 		while (icmp_len >= 8) {
436 			struct nd_opt_hdr *nd_opt =
437 				 (struct nd_opt_hdr *)(nd->opt + offset);
438 			int opt_len = nd_opt->nd_opt_len * 8;
439 
440 			if (unlikely(!opt_len || opt_len > icmp_len))
441 				return 0;
442 
443 			/* Store the link layer address if the appropriate
444 			 * option is provided.  It is considered an error if
445 			 * the same link layer option is specified twice.
446 			 */
447 			if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
448 			    && opt_len == 8) {
449 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
450 					goto invalid;
451 				ether_addr_copy(key->ipv6.nd.sll,
452 						&nd->opt[offset+sizeof(*nd_opt)]);
453 			} else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
454 				   && opt_len == 8) {
455 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
456 					goto invalid;
457 				ether_addr_copy(key->ipv6.nd.tll,
458 						&nd->opt[offset+sizeof(*nd_opt)]);
459 			}
460 
461 			icmp_len -= opt_len;
462 			offset += opt_len;
463 		}
464 	}
465 
466 	return 0;
467 
468 invalid:
469 	memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
470 	memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
471 	memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
472 
473 	return 0;
474 }
475 
476 /**
477  * key_extract - extracts a flow key from an Ethernet frame.
478  * @skb: sk_buff that contains the frame, with skb->data pointing to the
479  * Ethernet header
480  * @key: output flow key
481  *
482  * The caller must ensure that skb->len >= ETH_HLEN.
483  *
484  * Returns 0 if successful, otherwise a negative errno value.
485  *
486  * Initializes @skb header pointers as follows:
487  *
488  *    - skb->mac_header: the Ethernet header.
489  *
490  *    - skb->network_header: just past the Ethernet header, or just past the
491  *      VLAN header, to the first byte of the Ethernet payload.
492  *
493  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
494  *      on output, then just past the IP header, if one is present and
495  *      of a correct length, otherwise the same as skb->network_header.
496  *      For other key->eth.type values it is left untouched.
497  */
498 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
499 {
500 	int error;
501 	struct ethhdr *eth;
502 
503 	/* Flags are always used as part of stats */
504 	key->tp.flags = 0;
505 
506 	skb_reset_mac_header(skb);
507 
508 	/* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
509 	 * header in the linear data area.
510 	 */
511 	eth = eth_hdr(skb);
512 	ether_addr_copy(key->eth.src, eth->h_source);
513 	ether_addr_copy(key->eth.dst, eth->h_dest);
514 
515 	__skb_pull(skb, 2 * ETH_ALEN);
516 	/* We are going to push all headers that we pull, so no need to
517 	 * update skb->csum here.
518 	 */
519 
520 	if (unlikely(parse_vlan(skb, key)))
521 		return -ENOMEM;
522 
523 	key->eth.type = parse_ethertype(skb);
524 	if (unlikely(key->eth.type == htons(0)))
525 		return -ENOMEM;
526 
527 	skb_reset_network_header(skb);
528 	skb_reset_mac_len(skb);
529 	__skb_push(skb, skb->data - skb_mac_header(skb));
530 
531 	/* Network layer. */
532 	if (key->eth.type == htons(ETH_P_IP)) {
533 		struct iphdr *nh;
534 		__be16 offset;
535 
536 		error = check_iphdr(skb);
537 		if (unlikely(error)) {
538 			memset(&key->ip, 0, sizeof(key->ip));
539 			memset(&key->ipv4, 0, sizeof(key->ipv4));
540 			if (error == -EINVAL) {
541 				skb->transport_header = skb->network_header;
542 				error = 0;
543 			}
544 			return error;
545 		}
546 
547 		nh = ip_hdr(skb);
548 		key->ipv4.addr.src = nh->saddr;
549 		key->ipv4.addr.dst = nh->daddr;
550 
551 		key->ip.proto = nh->protocol;
552 		key->ip.tos = nh->tos;
553 		key->ip.ttl = nh->ttl;
554 
555 		offset = nh->frag_off & htons(IP_OFFSET);
556 		if (offset) {
557 			key->ip.frag = OVS_FRAG_TYPE_LATER;
558 			return 0;
559 		}
560 		if (nh->frag_off & htons(IP_MF) ||
561 			skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
562 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
563 		else
564 			key->ip.frag = OVS_FRAG_TYPE_NONE;
565 
566 		/* Transport layer. */
567 		if (key->ip.proto == IPPROTO_TCP) {
568 			if (tcphdr_ok(skb)) {
569 				struct tcphdr *tcp = tcp_hdr(skb);
570 				key->tp.src = tcp->source;
571 				key->tp.dst = tcp->dest;
572 				key->tp.flags = TCP_FLAGS_BE16(tcp);
573 			} else {
574 				memset(&key->tp, 0, sizeof(key->tp));
575 			}
576 
577 		} else if (key->ip.proto == IPPROTO_UDP) {
578 			if (udphdr_ok(skb)) {
579 				struct udphdr *udp = udp_hdr(skb);
580 				key->tp.src = udp->source;
581 				key->tp.dst = udp->dest;
582 			} else {
583 				memset(&key->tp, 0, sizeof(key->tp));
584 			}
585 		} else if (key->ip.proto == IPPROTO_SCTP) {
586 			if (sctphdr_ok(skb)) {
587 				struct sctphdr *sctp = sctp_hdr(skb);
588 				key->tp.src = sctp->source;
589 				key->tp.dst = sctp->dest;
590 			} else {
591 				memset(&key->tp, 0, sizeof(key->tp));
592 			}
593 		} else if (key->ip.proto == IPPROTO_ICMP) {
594 			if (icmphdr_ok(skb)) {
595 				struct icmphdr *icmp = icmp_hdr(skb);
596 				/* The ICMP type and code fields use the 16-bit
597 				 * transport port fields, so we need to store
598 				 * them in 16-bit network byte order. */
599 				key->tp.src = htons(icmp->type);
600 				key->tp.dst = htons(icmp->code);
601 			} else {
602 				memset(&key->tp, 0, sizeof(key->tp));
603 			}
604 		}
605 
606 	} else if (key->eth.type == htons(ETH_P_ARP) ||
607 		   key->eth.type == htons(ETH_P_RARP)) {
608 		struct arp_eth_header *arp;
609 		bool arp_available = arphdr_ok(skb);
610 
611 		arp = (struct arp_eth_header *)skb_network_header(skb);
612 
613 		if (arp_available &&
614 		    arp->ar_hrd == htons(ARPHRD_ETHER) &&
615 		    arp->ar_pro == htons(ETH_P_IP) &&
616 		    arp->ar_hln == ETH_ALEN &&
617 		    arp->ar_pln == 4) {
618 
619 			/* We only match on the lower 8 bits of the opcode. */
620 			if (ntohs(arp->ar_op) <= 0xff)
621 				key->ip.proto = ntohs(arp->ar_op);
622 			else
623 				key->ip.proto = 0;
624 
625 			memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
626 			memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
627 			ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
628 			ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
629 		} else {
630 			memset(&key->ip, 0, sizeof(key->ip));
631 			memset(&key->ipv4, 0, sizeof(key->ipv4));
632 		}
633 	} else if (eth_p_mpls(key->eth.type)) {
634 		size_t stack_len = MPLS_HLEN;
635 
636 		skb_set_inner_network_header(skb, skb->mac_len);
637 		while (1) {
638 			__be32 lse;
639 
640 			error = check_header(skb, skb->mac_len + stack_len);
641 			if (unlikely(error))
642 				return 0;
643 
644 			memcpy(&lse, skb_inner_network_header(skb), MPLS_HLEN);
645 
646 			if (stack_len == MPLS_HLEN)
647 				memcpy(&key->mpls.top_lse, &lse, MPLS_HLEN);
648 
649 			skb_set_inner_network_header(skb, skb->mac_len + stack_len);
650 			if (lse & htonl(MPLS_LS_S_MASK))
651 				break;
652 
653 			stack_len += MPLS_HLEN;
654 		}
655 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
656 		int nh_len;             /* IPv6 Header + Extensions */
657 
658 		nh_len = parse_ipv6hdr(skb, key);
659 		if (unlikely(nh_len < 0)) {
660 			switch (nh_len) {
661 			case -EINVAL:
662 				memset(&key->ip, 0, sizeof(key->ip));
663 				memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
664 				/* fall-through */
665 			case -EPROTO:
666 				skb->transport_header = skb->network_header;
667 				error = 0;
668 				break;
669 			default:
670 				error = nh_len;
671 			}
672 			return error;
673 		}
674 
675 		if (key->ip.frag == OVS_FRAG_TYPE_LATER)
676 			return 0;
677 		if (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 == NEXTHDR_TCP) {
682 			if (tcphdr_ok(skb)) {
683 				struct tcphdr *tcp = tcp_hdr(skb);
684 				key->tp.src = tcp->source;
685 				key->tp.dst = tcp->dest;
686 				key->tp.flags = TCP_FLAGS_BE16(tcp);
687 			} else {
688 				memset(&key->tp, 0, sizeof(key->tp));
689 			}
690 		} else if (key->ip.proto == NEXTHDR_UDP) {
691 			if (udphdr_ok(skb)) {
692 				struct udphdr *udp = udp_hdr(skb);
693 				key->tp.src = udp->source;
694 				key->tp.dst = udp->dest;
695 			} else {
696 				memset(&key->tp, 0, sizeof(key->tp));
697 			}
698 		} else if (key->ip.proto == NEXTHDR_SCTP) {
699 			if (sctphdr_ok(skb)) {
700 				struct sctphdr *sctp = sctp_hdr(skb);
701 				key->tp.src = sctp->source;
702 				key->tp.dst = sctp->dest;
703 			} else {
704 				memset(&key->tp, 0, sizeof(key->tp));
705 			}
706 		} else if (key->ip.proto == NEXTHDR_ICMP) {
707 			if (icmp6hdr_ok(skb)) {
708 				error = parse_icmpv6(skb, key, nh_len);
709 				if (error)
710 					return error;
711 			} else {
712 				memset(&key->tp, 0, sizeof(key->tp));
713 			}
714 		}
715 	}
716 	return 0;
717 }
718 
719 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
720 {
721 	return key_extract(skb, key);
722 }
723 
724 int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
725 			 struct sk_buff *skb, struct sw_flow_key *key)
726 {
727 	/* Extract metadata from packet. */
728 	if (tun_info) {
729 		key->tun_proto = ip_tunnel_info_af(tun_info);
730 		memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
731 
732 		if (tun_info->options_len) {
733 			BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) *
734 						   8)) - 1
735 					> sizeof(key->tun_opts));
736 
737 			ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key, tun_info->options_len),
738 						tun_info);
739 			key->tun_opts_len = tun_info->options_len;
740 		} else {
741 			key->tun_opts_len = 0;
742 		}
743 	} else  {
744 		key->tun_proto = 0;
745 		key->tun_opts_len = 0;
746 		memset(&key->tun_key, 0, sizeof(key->tun_key));
747 	}
748 
749 	key->phy.priority = skb->priority;
750 	key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
751 	key->phy.skb_mark = skb->mark;
752 	ovs_ct_fill_key(skb, key);
753 	key->ovs_flow_hash = 0;
754 	key->mac_proto = MAC_PROTO_ETHERNET;
755 	key->recirc_id = 0;
756 
757 	return key_extract(skb, key);
758 }
759 
760 int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
761 				   struct sk_buff *skb,
762 				   struct sw_flow_key *key, bool log)
763 {
764 	int err;
765 
766 	/* Extract metadata from netlink attributes. */
767 	err = ovs_nla_get_flow_metadata(net, attr, key, log);
768 	if (err)
769 		return err;
770 
771 	return key_extract(skb, key);
772 }
773