xref: /openbmc/linux/net/openvswitch/flow.c (revision d0e22329)
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 #include <net/nsh.h>
50 
51 #include "conntrack.h"
52 #include "datapath.h"
53 #include "flow.h"
54 #include "flow_netlink.h"
55 #include "vport.h"
56 
57 u64 ovs_flow_used_time(unsigned long flow_jiffies)
58 {
59 	struct timespec64 cur_ts;
60 	u64 cur_ms, idle_ms;
61 
62 	ktime_get_ts64(&cur_ts);
63 	idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
64 	cur_ms = (u64)(u32)cur_ts.tv_sec * MSEC_PER_SEC +
65 		 cur_ts.tv_nsec / NSEC_PER_MSEC;
66 
67 	return cur_ms - idle_ms;
68 }
69 
70 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
71 
72 void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
73 			   const struct sk_buff *skb)
74 {
75 	struct flow_stats *stats;
76 	unsigned 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 							      numa_node_id());
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 					cpumask_set_cpu(cpu, &flow->cpu_used_mask);
122 					goto unlock;
123 				}
124 			}
125 			flow->stats_last_writer = cpu;
126 		}
127 	}
128 
129 	stats->used = jiffies;
130 	stats->packet_count++;
131 	stats->byte_count += len;
132 	stats->tcp_flags |= tcp_flags;
133 unlock:
134 	spin_unlock(&stats->lock);
135 }
136 
137 /* Must be called with rcu_read_lock or ovs_mutex. */
138 void ovs_flow_stats_get(const struct sw_flow *flow,
139 			struct ovs_flow_stats *ovs_stats,
140 			unsigned long *used, __be16 *tcp_flags)
141 {
142 	int cpu;
143 
144 	*used = 0;
145 	*tcp_flags = 0;
146 	memset(ovs_stats, 0, sizeof(*ovs_stats));
147 
148 	/* We open code this to make sure cpu 0 is always considered */
149 	for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask)) {
150 		struct flow_stats *stats = rcu_dereference_ovsl(flow->stats[cpu]);
151 
152 		if (stats) {
153 			/* Local CPU may write on non-local stats, so we must
154 			 * block bottom-halves here.
155 			 */
156 			spin_lock_bh(&stats->lock);
157 			if (!*used || time_after(stats->used, *used))
158 				*used = stats->used;
159 			*tcp_flags |= stats->tcp_flags;
160 			ovs_stats->n_packets += stats->packet_count;
161 			ovs_stats->n_bytes += stats->byte_count;
162 			spin_unlock_bh(&stats->lock);
163 		}
164 	}
165 }
166 
167 /* Called with ovs_mutex. */
168 void ovs_flow_stats_clear(struct sw_flow *flow)
169 {
170 	int cpu;
171 
172 	/* We open code this to make sure cpu 0 is always considered */
173 	for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask)) {
174 		struct flow_stats *stats = ovsl_dereference(flow->stats[cpu]);
175 
176 		if (stats) {
177 			spin_lock_bh(&stats->lock);
178 			stats->used = 0;
179 			stats->packet_count = 0;
180 			stats->byte_count = 0;
181 			stats->tcp_flags = 0;
182 			spin_unlock_bh(&stats->lock);
183 		}
184 	}
185 }
186 
187 static int check_header(struct sk_buff *skb, int len)
188 {
189 	if (unlikely(skb->len < len))
190 		return -EINVAL;
191 	if (unlikely(!pskb_may_pull(skb, len)))
192 		return -ENOMEM;
193 	return 0;
194 }
195 
196 static bool arphdr_ok(struct sk_buff *skb)
197 {
198 	return pskb_may_pull(skb, skb_network_offset(skb) +
199 				  sizeof(struct arp_eth_header));
200 }
201 
202 static int check_iphdr(struct sk_buff *skb)
203 {
204 	unsigned int nh_ofs = skb_network_offset(skb);
205 	unsigned int ip_len;
206 	int err;
207 
208 	err = check_header(skb, nh_ofs + sizeof(struct iphdr));
209 	if (unlikely(err))
210 		return err;
211 
212 	ip_len = ip_hdrlen(skb);
213 	if (unlikely(ip_len < sizeof(struct iphdr) ||
214 		     skb->len < nh_ofs + ip_len))
215 		return -EINVAL;
216 
217 	skb_set_transport_header(skb, nh_ofs + ip_len);
218 	return 0;
219 }
220 
221 static bool tcphdr_ok(struct sk_buff *skb)
222 {
223 	int th_ofs = skb_transport_offset(skb);
224 	int tcp_len;
225 
226 	if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
227 		return false;
228 
229 	tcp_len = tcp_hdrlen(skb);
230 	if (unlikely(tcp_len < sizeof(struct tcphdr) ||
231 		     skb->len < th_ofs + tcp_len))
232 		return false;
233 
234 	return true;
235 }
236 
237 static bool udphdr_ok(struct sk_buff *skb)
238 {
239 	return pskb_may_pull(skb, skb_transport_offset(skb) +
240 				  sizeof(struct udphdr));
241 }
242 
243 static bool sctphdr_ok(struct sk_buff *skb)
244 {
245 	return pskb_may_pull(skb, skb_transport_offset(skb) +
246 				  sizeof(struct sctphdr));
247 }
248 
249 static bool icmphdr_ok(struct sk_buff *skb)
250 {
251 	return pskb_may_pull(skb, skb_transport_offset(skb) +
252 				  sizeof(struct icmphdr));
253 }
254 
255 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
256 {
257 	unsigned short frag_off;
258 	unsigned int payload_ofs = 0;
259 	unsigned int nh_ofs = skb_network_offset(skb);
260 	unsigned int nh_len;
261 	struct ipv6hdr *nh;
262 	int err, nexthdr, flags = 0;
263 
264 	err = check_header(skb, nh_ofs + sizeof(*nh));
265 	if (unlikely(err))
266 		return err;
267 
268 	nh = ipv6_hdr(skb);
269 
270 	key->ip.proto = NEXTHDR_NONE;
271 	key->ip.tos = ipv6_get_dsfield(nh);
272 	key->ip.ttl = nh->hop_limit;
273 	key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
274 	key->ipv6.addr.src = nh->saddr;
275 	key->ipv6.addr.dst = nh->daddr;
276 
277 	nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
278 	if (flags & IP6_FH_F_FRAG) {
279 		if (frag_off)
280 			key->ip.frag = OVS_FRAG_TYPE_LATER;
281 		else
282 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
283 	} else {
284 		key->ip.frag = OVS_FRAG_TYPE_NONE;
285 	}
286 
287 	/* Delayed handling of error in ipv6_find_hdr() as it
288 	 * always sets flags and frag_off to a valid value which may be
289 	 * used to set key->ip.frag above.
290 	 */
291 	if (unlikely(nexthdr < 0))
292 		return -EPROTO;
293 
294 	nh_len = payload_ofs - nh_ofs;
295 	skb_set_transport_header(skb, nh_ofs + nh_len);
296 	key->ip.proto = nexthdr;
297 	return nh_len;
298 }
299 
300 static bool icmp6hdr_ok(struct sk_buff *skb)
301 {
302 	return pskb_may_pull(skb, skb_transport_offset(skb) +
303 				  sizeof(struct icmp6hdr));
304 }
305 
306 /**
307  * Parse vlan tag from vlan header.
308  * Returns ERROR on memory error.
309  * Returns 0 if it encounters a non-vlan or incomplete packet.
310  * Returns 1 after successfully parsing vlan tag.
311  */
312 static int parse_vlan_tag(struct sk_buff *skb, struct vlan_head *key_vh,
313 			  bool untag_vlan)
314 {
315 	struct vlan_head *vh = (struct vlan_head *)skb->data;
316 
317 	if (likely(!eth_type_vlan(vh->tpid)))
318 		return 0;
319 
320 	if (unlikely(skb->len < sizeof(struct vlan_head) + sizeof(__be16)))
321 		return 0;
322 
323 	if (unlikely(!pskb_may_pull(skb, sizeof(struct vlan_head) +
324 				 sizeof(__be16))))
325 		return -ENOMEM;
326 
327 	vh = (struct vlan_head *)skb->data;
328 	key_vh->tci = vh->tci | htons(VLAN_CFI_MASK);
329 	key_vh->tpid = vh->tpid;
330 
331 	if (unlikely(untag_vlan)) {
332 		int offset = skb->data - skb_mac_header(skb);
333 		u16 tci;
334 		int err;
335 
336 		__skb_push(skb, offset);
337 		err = __skb_vlan_pop(skb, &tci);
338 		__skb_pull(skb, offset);
339 		if (err)
340 			return err;
341 		__vlan_hwaccel_put_tag(skb, key_vh->tpid, tci);
342 	} else {
343 		__skb_pull(skb, sizeof(struct vlan_head));
344 	}
345 	return 1;
346 }
347 
348 static void clear_vlan(struct sw_flow_key *key)
349 {
350 	key->eth.vlan.tci = 0;
351 	key->eth.vlan.tpid = 0;
352 	key->eth.cvlan.tci = 0;
353 	key->eth.cvlan.tpid = 0;
354 }
355 
356 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
357 {
358 	int res;
359 
360 	if (skb_vlan_tag_present(skb)) {
361 		key->eth.vlan.tci = htons(skb->vlan_tci) | htons(VLAN_CFI_MASK);
362 		key->eth.vlan.tpid = skb->vlan_proto;
363 	} else {
364 		/* Parse outer vlan tag in the non-accelerated case. */
365 		res = parse_vlan_tag(skb, &key->eth.vlan, true);
366 		if (res <= 0)
367 			return res;
368 	}
369 
370 	/* Parse inner vlan tag. */
371 	res = parse_vlan_tag(skb, &key->eth.cvlan, false);
372 	if (res <= 0)
373 		return res;
374 
375 	return 0;
376 }
377 
378 static __be16 parse_ethertype(struct sk_buff *skb)
379 {
380 	struct llc_snap_hdr {
381 		u8  dsap;  /* Always 0xAA */
382 		u8  ssap;  /* Always 0xAA */
383 		u8  ctrl;
384 		u8  oui[3];
385 		__be16 ethertype;
386 	};
387 	struct llc_snap_hdr *llc;
388 	__be16 proto;
389 
390 	proto = *(__be16 *) skb->data;
391 	__skb_pull(skb, sizeof(__be16));
392 
393 	if (eth_proto_is_802_3(proto))
394 		return proto;
395 
396 	if (skb->len < sizeof(struct llc_snap_hdr))
397 		return htons(ETH_P_802_2);
398 
399 	if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
400 		return htons(0);
401 
402 	llc = (struct llc_snap_hdr *) skb->data;
403 	if (llc->dsap != LLC_SAP_SNAP ||
404 	    llc->ssap != LLC_SAP_SNAP ||
405 	    (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
406 		return htons(ETH_P_802_2);
407 
408 	__skb_pull(skb, sizeof(struct llc_snap_hdr));
409 
410 	if (eth_proto_is_802_3(llc->ethertype))
411 		return llc->ethertype;
412 
413 	return htons(ETH_P_802_2);
414 }
415 
416 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
417 			int nh_len)
418 {
419 	struct icmp6hdr *icmp = icmp6_hdr(skb);
420 
421 	/* The ICMPv6 type and code fields use the 16-bit transport port
422 	 * fields, so we need to store them in 16-bit network byte order.
423 	 */
424 	key->tp.src = htons(icmp->icmp6_type);
425 	key->tp.dst = htons(icmp->icmp6_code);
426 	memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
427 
428 	if (icmp->icmp6_code == 0 &&
429 	    (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
430 	     icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
431 		int icmp_len = skb->len - skb_transport_offset(skb);
432 		struct nd_msg *nd;
433 		int offset;
434 
435 		/* In order to process neighbor discovery options, we need the
436 		 * entire packet.
437 		 */
438 		if (unlikely(icmp_len < sizeof(*nd)))
439 			return 0;
440 
441 		if (unlikely(skb_linearize(skb)))
442 			return -ENOMEM;
443 
444 		nd = (struct nd_msg *)skb_transport_header(skb);
445 		key->ipv6.nd.target = nd->target;
446 
447 		icmp_len -= sizeof(*nd);
448 		offset = 0;
449 		while (icmp_len >= 8) {
450 			struct nd_opt_hdr *nd_opt =
451 				 (struct nd_opt_hdr *)(nd->opt + offset);
452 			int opt_len = nd_opt->nd_opt_len * 8;
453 
454 			if (unlikely(!opt_len || opt_len > icmp_len))
455 				return 0;
456 
457 			/* Store the link layer address if the appropriate
458 			 * option is provided.  It is considered an error if
459 			 * the same link layer option is specified twice.
460 			 */
461 			if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
462 			    && opt_len == 8) {
463 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
464 					goto invalid;
465 				ether_addr_copy(key->ipv6.nd.sll,
466 						&nd->opt[offset+sizeof(*nd_opt)]);
467 			} else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
468 				   && opt_len == 8) {
469 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
470 					goto invalid;
471 				ether_addr_copy(key->ipv6.nd.tll,
472 						&nd->opt[offset+sizeof(*nd_opt)]);
473 			}
474 
475 			icmp_len -= opt_len;
476 			offset += opt_len;
477 		}
478 	}
479 
480 	return 0;
481 
482 invalid:
483 	memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
484 	memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
485 	memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
486 
487 	return 0;
488 }
489 
490 static int parse_nsh(struct sk_buff *skb, struct sw_flow_key *key)
491 {
492 	struct nshhdr *nh;
493 	unsigned int nh_ofs = skb_network_offset(skb);
494 	u8 version, length;
495 	int err;
496 
497 	err = check_header(skb, nh_ofs + NSH_BASE_HDR_LEN);
498 	if (unlikely(err))
499 		return err;
500 
501 	nh = nsh_hdr(skb);
502 	version = nsh_get_ver(nh);
503 	length = nsh_hdr_len(nh);
504 
505 	if (version != 0)
506 		return -EINVAL;
507 
508 	err = check_header(skb, nh_ofs + length);
509 	if (unlikely(err))
510 		return err;
511 
512 	nh = nsh_hdr(skb);
513 	key->nsh.base.flags = nsh_get_flags(nh);
514 	key->nsh.base.ttl = nsh_get_ttl(nh);
515 	key->nsh.base.mdtype = nh->mdtype;
516 	key->nsh.base.np = nh->np;
517 	key->nsh.base.path_hdr = nh->path_hdr;
518 	switch (key->nsh.base.mdtype) {
519 	case NSH_M_TYPE1:
520 		if (length != NSH_M_TYPE1_LEN)
521 			return -EINVAL;
522 		memcpy(key->nsh.context, nh->md1.context,
523 		       sizeof(nh->md1));
524 		break;
525 	case NSH_M_TYPE2:
526 		memset(key->nsh.context, 0,
527 		       sizeof(nh->md1));
528 		break;
529 	default:
530 		return -EINVAL;
531 	}
532 
533 	return 0;
534 }
535 
536 /**
537  * key_extract - extracts a flow key from an Ethernet frame.
538  * @skb: sk_buff that contains the frame, with skb->data pointing to the
539  * Ethernet header
540  * @key: output flow key
541  *
542  * The caller must ensure that skb->len >= ETH_HLEN.
543  *
544  * Returns 0 if successful, otherwise a negative errno value.
545  *
546  * Initializes @skb header fields as follows:
547  *
548  *    - skb->mac_header: the L2 header.
549  *
550  *    - skb->network_header: just past the L2 header, or just past the
551  *      VLAN header, to the first byte of the L2 payload.
552  *
553  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
554  *      on output, then just past the IP header, if one is present and
555  *      of a correct length, otherwise the same as skb->network_header.
556  *      For other key->eth.type values it is left untouched.
557  *
558  *    - skb->protocol: the type of the data starting at skb->network_header.
559  *      Equals to key->eth.type.
560  */
561 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
562 {
563 	int error;
564 	struct ethhdr *eth;
565 
566 	/* Flags are always used as part of stats */
567 	key->tp.flags = 0;
568 
569 	skb_reset_mac_header(skb);
570 
571 	/* Link layer. */
572 	clear_vlan(key);
573 	if (ovs_key_mac_proto(key) == MAC_PROTO_NONE) {
574 		if (unlikely(eth_type_vlan(skb->protocol)))
575 			return -EINVAL;
576 
577 		skb_reset_network_header(skb);
578 		key->eth.type = skb->protocol;
579 	} else {
580 		eth = eth_hdr(skb);
581 		ether_addr_copy(key->eth.src, eth->h_source);
582 		ether_addr_copy(key->eth.dst, eth->h_dest);
583 
584 		__skb_pull(skb, 2 * ETH_ALEN);
585 		/* We are going to push all headers that we pull, so no need to
586 		* update skb->csum here.
587 		*/
588 
589 		if (unlikely(parse_vlan(skb, key)))
590 			return -ENOMEM;
591 
592 		key->eth.type = parse_ethertype(skb);
593 		if (unlikely(key->eth.type == htons(0)))
594 			return -ENOMEM;
595 
596 		/* Multiple tagged packets need to retain TPID to satisfy
597 		 * skb_vlan_pop(), which will later shift the ethertype into
598 		 * skb->protocol.
599 		 */
600 		if (key->eth.cvlan.tci & htons(VLAN_CFI_MASK))
601 			skb->protocol = key->eth.cvlan.tpid;
602 		else
603 			skb->protocol = key->eth.type;
604 
605 		skb_reset_network_header(skb);
606 		__skb_push(skb, skb->data - skb_mac_header(skb));
607 	}
608 	skb_reset_mac_len(skb);
609 
610 	/* Network layer. */
611 	if (key->eth.type == htons(ETH_P_IP)) {
612 		struct iphdr *nh;
613 		__be16 offset;
614 
615 		error = check_iphdr(skb);
616 		if (unlikely(error)) {
617 			memset(&key->ip, 0, sizeof(key->ip));
618 			memset(&key->ipv4, 0, sizeof(key->ipv4));
619 			if (error == -EINVAL) {
620 				skb->transport_header = skb->network_header;
621 				error = 0;
622 			}
623 			return error;
624 		}
625 
626 		nh = ip_hdr(skb);
627 		key->ipv4.addr.src = nh->saddr;
628 		key->ipv4.addr.dst = nh->daddr;
629 
630 		key->ip.proto = nh->protocol;
631 		key->ip.tos = nh->tos;
632 		key->ip.ttl = nh->ttl;
633 
634 		offset = nh->frag_off & htons(IP_OFFSET);
635 		if (offset) {
636 			key->ip.frag = OVS_FRAG_TYPE_LATER;
637 			return 0;
638 		}
639 		if (nh->frag_off & htons(IP_MF) ||
640 			skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
641 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
642 		else
643 			key->ip.frag = OVS_FRAG_TYPE_NONE;
644 
645 		/* Transport layer. */
646 		if (key->ip.proto == IPPROTO_TCP) {
647 			if (tcphdr_ok(skb)) {
648 				struct tcphdr *tcp = tcp_hdr(skb);
649 				key->tp.src = tcp->source;
650 				key->tp.dst = tcp->dest;
651 				key->tp.flags = TCP_FLAGS_BE16(tcp);
652 			} else {
653 				memset(&key->tp, 0, sizeof(key->tp));
654 			}
655 
656 		} else if (key->ip.proto == IPPROTO_UDP) {
657 			if (udphdr_ok(skb)) {
658 				struct udphdr *udp = udp_hdr(skb);
659 				key->tp.src = udp->source;
660 				key->tp.dst = udp->dest;
661 			} else {
662 				memset(&key->tp, 0, sizeof(key->tp));
663 			}
664 		} else if (key->ip.proto == IPPROTO_SCTP) {
665 			if (sctphdr_ok(skb)) {
666 				struct sctphdr *sctp = sctp_hdr(skb);
667 				key->tp.src = sctp->source;
668 				key->tp.dst = sctp->dest;
669 			} else {
670 				memset(&key->tp, 0, sizeof(key->tp));
671 			}
672 		} else if (key->ip.proto == IPPROTO_ICMP) {
673 			if (icmphdr_ok(skb)) {
674 				struct icmphdr *icmp = icmp_hdr(skb);
675 				/* The ICMP type and code fields use the 16-bit
676 				 * transport port fields, so we need to store
677 				 * them in 16-bit network byte order. */
678 				key->tp.src = htons(icmp->type);
679 				key->tp.dst = htons(icmp->code);
680 			} else {
681 				memset(&key->tp, 0, sizeof(key->tp));
682 			}
683 		}
684 
685 	} else if (key->eth.type == htons(ETH_P_ARP) ||
686 		   key->eth.type == htons(ETH_P_RARP)) {
687 		struct arp_eth_header *arp;
688 		bool arp_available = arphdr_ok(skb);
689 
690 		arp = (struct arp_eth_header *)skb_network_header(skb);
691 
692 		if (arp_available &&
693 		    arp->ar_hrd == htons(ARPHRD_ETHER) &&
694 		    arp->ar_pro == htons(ETH_P_IP) &&
695 		    arp->ar_hln == ETH_ALEN &&
696 		    arp->ar_pln == 4) {
697 
698 			/* We only match on the lower 8 bits of the opcode. */
699 			if (ntohs(arp->ar_op) <= 0xff)
700 				key->ip.proto = ntohs(arp->ar_op);
701 			else
702 				key->ip.proto = 0;
703 
704 			memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
705 			memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
706 			ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
707 			ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
708 		} else {
709 			memset(&key->ip, 0, sizeof(key->ip));
710 			memset(&key->ipv4, 0, sizeof(key->ipv4));
711 		}
712 	} else if (eth_p_mpls(key->eth.type)) {
713 		size_t stack_len = MPLS_HLEN;
714 
715 		skb_set_inner_network_header(skb, skb->mac_len);
716 		while (1) {
717 			__be32 lse;
718 
719 			error = check_header(skb, skb->mac_len + stack_len);
720 			if (unlikely(error))
721 				return 0;
722 
723 			memcpy(&lse, skb_inner_network_header(skb), MPLS_HLEN);
724 
725 			if (stack_len == MPLS_HLEN)
726 				memcpy(&key->mpls.top_lse, &lse, MPLS_HLEN);
727 
728 			skb_set_inner_network_header(skb, skb->mac_len + stack_len);
729 			if (lse & htonl(MPLS_LS_S_MASK))
730 				break;
731 
732 			stack_len += MPLS_HLEN;
733 		}
734 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
735 		int nh_len;             /* IPv6 Header + Extensions */
736 
737 		nh_len = parse_ipv6hdr(skb, key);
738 		if (unlikely(nh_len < 0)) {
739 			switch (nh_len) {
740 			case -EINVAL:
741 				memset(&key->ip, 0, sizeof(key->ip));
742 				memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
743 				/* fall-through */
744 			case -EPROTO:
745 				skb->transport_header = skb->network_header;
746 				error = 0;
747 				break;
748 			default:
749 				error = nh_len;
750 			}
751 			return error;
752 		}
753 
754 		if (key->ip.frag == OVS_FRAG_TYPE_LATER)
755 			return 0;
756 		if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
757 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
758 
759 		/* Transport layer. */
760 		if (key->ip.proto == NEXTHDR_TCP) {
761 			if (tcphdr_ok(skb)) {
762 				struct tcphdr *tcp = tcp_hdr(skb);
763 				key->tp.src = tcp->source;
764 				key->tp.dst = tcp->dest;
765 				key->tp.flags = TCP_FLAGS_BE16(tcp);
766 			} else {
767 				memset(&key->tp, 0, sizeof(key->tp));
768 			}
769 		} else if (key->ip.proto == NEXTHDR_UDP) {
770 			if (udphdr_ok(skb)) {
771 				struct udphdr *udp = udp_hdr(skb);
772 				key->tp.src = udp->source;
773 				key->tp.dst = udp->dest;
774 			} else {
775 				memset(&key->tp, 0, sizeof(key->tp));
776 			}
777 		} else if (key->ip.proto == NEXTHDR_SCTP) {
778 			if (sctphdr_ok(skb)) {
779 				struct sctphdr *sctp = sctp_hdr(skb);
780 				key->tp.src = sctp->source;
781 				key->tp.dst = sctp->dest;
782 			} else {
783 				memset(&key->tp, 0, sizeof(key->tp));
784 			}
785 		} else if (key->ip.proto == NEXTHDR_ICMP) {
786 			if (icmp6hdr_ok(skb)) {
787 				error = parse_icmpv6(skb, key, nh_len);
788 				if (error)
789 					return error;
790 			} else {
791 				memset(&key->tp, 0, sizeof(key->tp));
792 			}
793 		}
794 	} else if (key->eth.type == htons(ETH_P_NSH)) {
795 		error = parse_nsh(skb, key);
796 		if (error)
797 			return error;
798 	}
799 	return 0;
800 }
801 
802 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
803 {
804 	int res;
805 
806 	res = key_extract(skb, key);
807 	if (!res)
808 		key->mac_proto &= ~SW_FLOW_KEY_INVALID;
809 
810 	return res;
811 }
812 
813 static int key_extract_mac_proto(struct sk_buff *skb)
814 {
815 	switch (skb->dev->type) {
816 	case ARPHRD_ETHER:
817 		return MAC_PROTO_ETHERNET;
818 	case ARPHRD_NONE:
819 		if (skb->protocol == htons(ETH_P_TEB))
820 			return MAC_PROTO_ETHERNET;
821 		return MAC_PROTO_NONE;
822 	}
823 	WARN_ON_ONCE(1);
824 	return -EINVAL;
825 }
826 
827 int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
828 			 struct sk_buff *skb, struct sw_flow_key *key)
829 {
830 	int res, err;
831 
832 	/* Extract metadata from packet. */
833 	if (tun_info) {
834 		key->tun_proto = ip_tunnel_info_af(tun_info);
835 		memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
836 
837 		if (tun_info->options_len) {
838 			BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) *
839 						   8)) - 1
840 					> sizeof(key->tun_opts));
841 
842 			ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key, tun_info->options_len),
843 						tun_info);
844 			key->tun_opts_len = tun_info->options_len;
845 		} else {
846 			key->tun_opts_len = 0;
847 		}
848 	} else  {
849 		key->tun_proto = 0;
850 		key->tun_opts_len = 0;
851 		memset(&key->tun_key, 0, sizeof(key->tun_key));
852 	}
853 
854 	key->phy.priority = skb->priority;
855 	key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
856 	key->phy.skb_mark = skb->mark;
857 	key->ovs_flow_hash = 0;
858 	res = key_extract_mac_proto(skb);
859 	if (res < 0)
860 		return res;
861 	key->mac_proto = res;
862 	key->recirc_id = 0;
863 
864 	err = key_extract(skb, key);
865 	if (!err)
866 		ovs_ct_fill_key(skb, key);   /* Must be after key_extract(). */
867 	return err;
868 }
869 
870 int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
871 				   struct sk_buff *skb,
872 				   struct sw_flow_key *key, bool log)
873 {
874 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
875 	u64 attrs = 0;
876 	int err;
877 
878 	err = parse_flow_nlattrs(attr, a, &attrs, log);
879 	if (err)
880 		return -EINVAL;
881 
882 	/* Extract metadata from netlink attributes. */
883 	err = ovs_nla_get_flow_metadata(net, a, attrs, key, log);
884 	if (err)
885 		return err;
886 
887 	/* key_extract assumes that skb->protocol is set-up for
888 	 * layer 3 packets which is the case for other callers,
889 	 * in particular packets received from the network stack.
890 	 * Here the correct value can be set from the metadata
891 	 * extracted above.
892 	 * For L2 packet key eth type would be zero. skb protocol
893 	 * would be set to correct value later during key-extact.
894 	 */
895 
896 	skb->protocol = key->eth.type;
897 	err = key_extract(skb, key);
898 	if (err)
899 		return err;
900 
901 	/* Check that we have conntrack original direction tuple metadata only
902 	 * for packets for which it makes sense.  Otherwise the key may be
903 	 * corrupted due to overlapping key fields.
904 	 */
905 	if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4) &&
906 	    key->eth.type != htons(ETH_P_IP))
907 		return -EINVAL;
908 	if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6) &&
909 	    (key->eth.type != htons(ETH_P_IPV6) ||
910 	     sw_flow_key_is_nd(key)))
911 		return -EINVAL;
912 
913 	return 0;
914 }
915