xref: /openbmc/linux/net/openvswitch/flow.c (revision 08283d30)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2007-2014 Nicira, Inc.
4  */
5 
6 #include <linux/uaccess.h>
7 #include <linux/netdevice.h>
8 #include <linux/etherdevice.h>
9 #include <linux/if_ether.h>
10 #include <linux/if_vlan.h>
11 #include <net/llc_pdu.h>
12 #include <linux/kernel.h>
13 #include <linux/jhash.h>
14 #include <linux/jiffies.h>
15 #include <linux/llc.h>
16 #include <linux/module.h>
17 #include <linux/in.h>
18 #include <linux/rcupdate.h>
19 #include <linux/cpumask.h>
20 #include <linux/if_arp.h>
21 #include <linux/ip.h>
22 #include <linux/ipv6.h>
23 #include <linux/mpls.h>
24 #include <linux/sctp.h>
25 #include <linux/smp.h>
26 #include <linux/tcp.h>
27 #include <linux/udp.h>
28 #include <linux/icmp.h>
29 #include <linux/icmpv6.h>
30 #include <linux/rculist.h>
31 #include <net/ip.h>
32 #include <net/ip_tunnels.h>
33 #include <net/ipv6.h>
34 #include <net/mpls.h>
35 #include <net/ndisc.h>
36 #include <net/nsh.h>
37 
38 #include "conntrack.h"
39 #include "datapath.h"
40 #include "flow.h"
41 #include "flow_netlink.h"
42 #include "vport.h"
43 
44 u64 ovs_flow_used_time(unsigned long flow_jiffies)
45 {
46 	struct timespec64 cur_ts;
47 	u64 cur_ms, idle_ms;
48 
49 	ktime_get_ts64(&cur_ts);
50 	idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
51 	cur_ms = (u64)(u32)cur_ts.tv_sec * MSEC_PER_SEC +
52 		 cur_ts.tv_nsec / NSEC_PER_MSEC;
53 
54 	return cur_ms - idle_ms;
55 }
56 
57 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
58 
59 void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
60 			   const struct sk_buff *skb)
61 {
62 	struct sw_flow_stats *stats;
63 	unsigned int cpu = smp_processor_id();
64 	int len = skb->len + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
65 
66 	stats = rcu_dereference(flow->stats[cpu]);
67 
68 	/* Check if already have CPU-specific stats. */
69 	if (likely(stats)) {
70 		spin_lock(&stats->lock);
71 		/* Mark if we write on the pre-allocated stats. */
72 		if (cpu == 0 && unlikely(flow->stats_last_writer != cpu))
73 			flow->stats_last_writer = cpu;
74 	} else {
75 		stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
76 		spin_lock(&stats->lock);
77 
78 		/* If the current CPU is the only writer on the
79 		 * pre-allocated stats keep using them.
80 		 */
81 		if (unlikely(flow->stats_last_writer != cpu)) {
82 			/* A previous locker may have already allocated the
83 			 * stats, so we need to check again.  If CPU-specific
84 			 * stats were already allocated, we update the pre-
85 			 * allocated stats as we have already locked them.
86 			 */
87 			if (likely(flow->stats_last_writer != -1) &&
88 			    likely(!rcu_access_pointer(flow->stats[cpu]))) {
89 				/* Try to allocate CPU-specific stats. */
90 				struct sw_flow_stats *new_stats;
91 
92 				new_stats =
93 					kmem_cache_alloc_node(flow_stats_cache,
94 							      GFP_NOWAIT |
95 							      __GFP_THISNODE |
96 							      __GFP_NOWARN |
97 							      __GFP_NOMEMALLOC,
98 							      numa_node_id());
99 				if (likely(new_stats)) {
100 					new_stats->used = jiffies;
101 					new_stats->packet_count = 1;
102 					new_stats->byte_count = len;
103 					new_stats->tcp_flags = tcp_flags;
104 					spin_lock_init(&new_stats->lock);
105 
106 					rcu_assign_pointer(flow->stats[cpu],
107 							   new_stats);
108 					cpumask_set_cpu(cpu, &flow->cpu_used_mask);
109 					goto unlock;
110 				}
111 			}
112 			flow->stats_last_writer = cpu;
113 		}
114 	}
115 
116 	stats->used = jiffies;
117 	stats->packet_count++;
118 	stats->byte_count += len;
119 	stats->tcp_flags |= tcp_flags;
120 unlock:
121 	spin_unlock(&stats->lock);
122 }
123 
124 /* Must be called with rcu_read_lock or ovs_mutex. */
125 void ovs_flow_stats_get(const struct sw_flow *flow,
126 			struct ovs_flow_stats *ovs_stats,
127 			unsigned long *used, __be16 *tcp_flags)
128 {
129 	int cpu;
130 
131 	*used = 0;
132 	*tcp_flags = 0;
133 	memset(ovs_stats, 0, sizeof(*ovs_stats));
134 
135 	/* We open code this to make sure cpu 0 is always considered */
136 	for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask)) {
137 		struct sw_flow_stats *stats = rcu_dereference_ovsl(flow->stats[cpu]);
138 
139 		if (stats) {
140 			/* Local CPU may write on non-local stats, so we must
141 			 * block bottom-halves here.
142 			 */
143 			spin_lock_bh(&stats->lock);
144 			if (!*used || time_after(stats->used, *used))
145 				*used = stats->used;
146 			*tcp_flags |= stats->tcp_flags;
147 			ovs_stats->n_packets += stats->packet_count;
148 			ovs_stats->n_bytes += stats->byte_count;
149 			spin_unlock_bh(&stats->lock);
150 		}
151 	}
152 }
153 
154 /* Called with ovs_mutex. */
155 void ovs_flow_stats_clear(struct sw_flow *flow)
156 {
157 	int cpu;
158 
159 	/* We open code this to make sure cpu 0 is always considered */
160 	for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask)) {
161 		struct sw_flow_stats *stats = ovsl_dereference(flow->stats[cpu]);
162 
163 		if (stats) {
164 			spin_lock_bh(&stats->lock);
165 			stats->used = 0;
166 			stats->packet_count = 0;
167 			stats->byte_count = 0;
168 			stats->tcp_flags = 0;
169 			spin_unlock_bh(&stats->lock);
170 		}
171 	}
172 }
173 
174 static int check_header(struct sk_buff *skb, int len)
175 {
176 	if (unlikely(skb->len < len))
177 		return -EINVAL;
178 	if (unlikely(!pskb_may_pull(skb, len)))
179 		return -ENOMEM;
180 	return 0;
181 }
182 
183 static bool arphdr_ok(struct sk_buff *skb)
184 {
185 	return pskb_may_pull(skb, skb_network_offset(skb) +
186 				  sizeof(struct arp_eth_header));
187 }
188 
189 static int check_iphdr(struct sk_buff *skb)
190 {
191 	unsigned int nh_ofs = skb_network_offset(skb);
192 	unsigned int ip_len;
193 	int err;
194 
195 	err = check_header(skb, nh_ofs + sizeof(struct iphdr));
196 	if (unlikely(err))
197 		return err;
198 
199 	ip_len = ip_hdrlen(skb);
200 	if (unlikely(ip_len < sizeof(struct iphdr) ||
201 		     skb->len < nh_ofs + ip_len))
202 		return -EINVAL;
203 
204 	skb_set_transport_header(skb, nh_ofs + ip_len);
205 	return 0;
206 }
207 
208 static bool tcphdr_ok(struct sk_buff *skb)
209 {
210 	int th_ofs = skb_transport_offset(skb);
211 	int tcp_len;
212 
213 	if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
214 		return false;
215 
216 	tcp_len = tcp_hdrlen(skb);
217 	if (unlikely(tcp_len < sizeof(struct tcphdr) ||
218 		     skb->len < th_ofs + tcp_len))
219 		return false;
220 
221 	return true;
222 }
223 
224 static bool udphdr_ok(struct sk_buff *skb)
225 {
226 	return pskb_may_pull(skb, skb_transport_offset(skb) +
227 				  sizeof(struct udphdr));
228 }
229 
230 static bool sctphdr_ok(struct sk_buff *skb)
231 {
232 	return pskb_may_pull(skb, skb_transport_offset(skb) +
233 				  sizeof(struct sctphdr));
234 }
235 
236 static bool icmphdr_ok(struct sk_buff *skb)
237 {
238 	return pskb_may_pull(skb, skb_transport_offset(skb) +
239 				  sizeof(struct icmphdr));
240 }
241 
242 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
243 {
244 	unsigned short frag_off;
245 	unsigned int payload_ofs = 0;
246 	unsigned int nh_ofs = skb_network_offset(skb);
247 	unsigned int nh_len;
248 	struct ipv6hdr *nh;
249 	int err, nexthdr, flags = 0;
250 
251 	err = check_header(skb, nh_ofs + sizeof(*nh));
252 	if (unlikely(err))
253 		return err;
254 
255 	nh = ipv6_hdr(skb);
256 
257 	key->ip.proto = NEXTHDR_NONE;
258 	key->ip.tos = ipv6_get_dsfield(nh);
259 	key->ip.ttl = nh->hop_limit;
260 	key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
261 	key->ipv6.addr.src = nh->saddr;
262 	key->ipv6.addr.dst = nh->daddr;
263 
264 	nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
265 	if (flags & IP6_FH_F_FRAG) {
266 		if (frag_off) {
267 			key->ip.frag = OVS_FRAG_TYPE_LATER;
268 			key->ip.proto = nexthdr;
269 			return 0;
270 		}
271 		key->ip.frag = OVS_FRAG_TYPE_FIRST;
272 	} else {
273 		key->ip.frag = OVS_FRAG_TYPE_NONE;
274 	}
275 
276 	/* Delayed handling of error in ipv6_find_hdr() as it
277 	 * always sets flags and frag_off to a valid value which may be
278 	 * used to set key->ip.frag above.
279 	 */
280 	if (unlikely(nexthdr < 0))
281 		return -EPROTO;
282 
283 	nh_len = payload_ofs - nh_ofs;
284 	skb_set_transport_header(skb, nh_ofs + nh_len);
285 	key->ip.proto = nexthdr;
286 	return nh_len;
287 }
288 
289 static bool icmp6hdr_ok(struct sk_buff *skb)
290 {
291 	return pskb_may_pull(skb, skb_transport_offset(skb) +
292 				  sizeof(struct icmp6hdr));
293 }
294 
295 /**
296  * Parse vlan tag from vlan header.
297  * Returns ERROR on memory error.
298  * Returns 0 if it encounters a non-vlan or incomplete packet.
299  * Returns 1 after successfully parsing vlan tag.
300  */
301 static int parse_vlan_tag(struct sk_buff *skb, struct vlan_head *key_vh,
302 			  bool untag_vlan)
303 {
304 	struct vlan_head *vh = (struct vlan_head *)skb->data;
305 
306 	if (likely(!eth_type_vlan(vh->tpid)))
307 		return 0;
308 
309 	if (unlikely(skb->len < sizeof(struct vlan_head) + sizeof(__be16)))
310 		return 0;
311 
312 	if (unlikely(!pskb_may_pull(skb, sizeof(struct vlan_head) +
313 				 sizeof(__be16))))
314 		return -ENOMEM;
315 
316 	vh = (struct vlan_head *)skb->data;
317 	key_vh->tci = vh->tci | htons(VLAN_CFI_MASK);
318 	key_vh->tpid = vh->tpid;
319 
320 	if (unlikely(untag_vlan)) {
321 		int offset = skb->data - skb_mac_header(skb);
322 		u16 tci;
323 		int err;
324 
325 		__skb_push(skb, offset);
326 		err = __skb_vlan_pop(skb, &tci);
327 		__skb_pull(skb, offset);
328 		if (err)
329 			return err;
330 		__vlan_hwaccel_put_tag(skb, key_vh->tpid, tci);
331 	} else {
332 		__skb_pull(skb, sizeof(struct vlan_head));
333 	}
334 	return 1;
335 }
336 
337 static void clear_vlan(struct sw_flow_key *key)
338 {
339 	key->eth.vlan.tci = 0;
340 	key->eth.vlan.tpid = 0;
341 	key->eth.cvlan.tci = 0;
342 	key->eth.cvlan.tpid = 0;
343 }
344 
345 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
346 {
347 	int res;
348 
349 	if (skb_vlan_tag_present(skb)) {
350 		key->eth.vlan.tci = htons(skb->vlan_tci) | htons(VLAN_CFI_MASK);
351 		key->eth.vlan.tpid = skb->vlan_proto;
352 	} else {
353 		/* Parse outer vlan tag in the non-accelerated case. */
354 		res = parse_vlan_tag(skb, &key->eth.vlan, true);
355 		if (res <= 0)
356 			return res;
357 	}
358 
359 	/* Parse inner vlan tag. */
360 	res = parse_vlan_tag(skb, &key->eth.cvlan, false);
361 	if (res <= 0)
362 		return res;
363 
364 	return 0;
365 }
366 
367 static __be16 parse_ethertype(struct sk_buff *skb)
368 {
369 	struct llc_snap_hdr {
370 		u8  dsap;  /* Always 0xAA */
371 		u8  ssap;  /* Always 0xAA */
372 		u8  ctrl;
373 		u8  oui[3];
374 		__be16 ethertype;
375 	};
376 	struct llc_snap_hdr *llc;
377 	__be16 proto;
378 
379 	proto = *(__be16 *) skb->data;
380 	__skb_pull(skb, sizeof(__be16));
381 
382 	if (eth_proto_is_802_3(proto))
383 		return proto;
384 
385 	if (skb->len < sizeof(struct llc_snap_hdr))
386 		return htons(ETH_P_802_2);
387 
388 	if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
389 		return htons(0);
390 
391 	llc = (struct llc_snap_hdr *) skb->data;
392 	if (llc->dsap != LLC_SAP_SNAP ||
393 	    llc->ssap != LLC_SAP_SNAP ||
394 	    (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
395 		return htons(ETH_P_802_2);
396 
397 	__skb_pull(skb, sizeof(struct llc_snap_hdr));
398 
399 	if (eth_proto_is_802_3(llc->ethertype))
400 		return llc->ethertype;
401 
402 	return htons(ETH_P_802_2);
403 }
404 
405 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
406 			int nh_len)
407 {
408 	struct icmp6hdr *icmp = icmp6_hdr(skb);
409 
410 	/* The ICMPv6 type and code fields use the 16-bit transport port
411 	 * fields, so we need to store them in 16-bit network byte order.
412 	 */
413 	key->tp.src = htons(icmp->icmp6_type);
414 	key->tp.dst = htons(icmp->icmp6_code);
415 	memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
416 
417 	if (icmp->icmp6_code == 0 &&
418 	    (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
419 	     icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
420 		int icmp_len = skb->len - skb_transport_offset(skb);
421 		struct nd_msg *nd;
422 		int offset;
423 
424 		/* In order to process neighbor discovery options, we need the
425 		 * entire packet.
426 		 */
427 		if (unlikely(icmp_len < sizeof(*nd)))
428 			return 0;
429 
430 		if (unlikely(skb_linearize(skb)))
431 			return -ENOMEM;
432 
433 		nd = (struct nd_msg *)skb_transport_header(skb);
434 		key->ipv6.nd.target = nd->target;
435 
436 		icmp_len -= sizeof(*nd);
437 		offset = 0;
438 		while (icmp_len >= 8) {
439 			struct nd_opt_hdr *nd_opt =
440 				 (struct nd_opt_hdr *)(nd->opt + offset);
441 			int opt_len = nd_opt->nd_opt_len * 8;
442 
443 			if (unlikely(!opt_len || opt_len > icmp_len))
444 				return 0;
445 
446 			/* Store the link layer address if the appropriate
447 			 * option is provided.  It is considered an error if
448 			 * the same link layer option is specified twice.
449 			 */
450 			if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
451 			    && opt_len == 8) {
452 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
453 					goto invalid;
454 				ether_addr_copy(key->ipv6.nd.sll,
455 						&nd->opt[offset+sizeof(*nd_opt)]);
456 			} else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
457 				   && opt_len == 8) {
458 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
459 					goto invalid;
460 				ether_addr_copy(key->ipv6.nd.tll,
461 						&nd->opt[offset+sizeof(*nd_opt)]);
462 			}
463 
464 			icmp_len -= opt_len;
465 			offset += opt_len;
466 		}
467 	}
468 
469 	return 0;
470 
471 invalid:
472 	memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
473 	memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
474 	memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
475 
476 	return 0;
477 }
478 
479 static int parse_nsh(struct sk_buff *skb, struct sw_flow_key *key)
480 {
481 	struct nshhdr *nh;
482 	unsigned int nh_ofs = skb_network_offset(skb);
483 	u8 version, length;
484 	int err;
485 
486 	err = check_header(skb, nh_ofs + NSH_BASE_HDR_LEN);
487 	if (unlikely(err))
488 		return err;
489 
490 	nh = nsh_hdr(skb);
491 	version = nsh_get_ver(nh);
492 	length = nsh_hdr_len(nh);
493 
494 	if (version != 0)
495 		return -EINVAL;
496 
497 	err = check_header(skb, nh_ofs + length);
498 	if (unlikely(err))
499 		return err;
500 
501 	nh = nsh_hdr(skb);
502 	key->nsh.base.flags = nsh_get_flags(nh);
503 	key->nsh.base.ttl = nsh_get_ttl(nh);
504 	key->nsh.base.mdtype = nh->mdtype;
505 	key->nsh.base.np = nh->np;
506 	key->nsh.base.path_hdr = nh->path_hdr;
507 	switch (key->nsh.base.mdtype) {
508 	case NSH_M_TYPE1:
509 		if (length != NSH_M_TYPE1_LEN)
510 			return -EINVAL;
511 		memcpy(key->nsh.context, nh->md1.context,
512 		       sizeof(nh->md1));
513 		break;
514 	case NSH_M_TYPE2:
515 		memset(key->nsh.context, 0,
516 		       sizeof(nh->md1));
517 		break;
518 	default:
519 		return -EINVAL;
520 	}
521 
522 	return 0;
523 }
524 
525 /**
526  * key_extract_l3l4 - extracts L3/L4 header information.
527  * @skb: sk_buff that contains the frame, with skb->data pointing to the
528  *       L3 header
529  * @key: output flow key
530  *
531  */
532 static int key_extract_l3l4(struct sk_buff *skb, struct sw_flow_key *key)
533 {
534 	int error;
535 
536 	/* Network layer. */
537 	if (key->eth.type == htons(ETH_P_IP)) {
538 		struct iphdr *nh;
539 		__be16 offset;
540 
541 		error = check_iphdr(skb);
542 		if (unlikely(error)) {
543 			memset(&key->ip, 0, sizeof(key->ip));
544 			memset(&key->ipv4, 0, sizeof(key->ipv4));
545 			if (error == -EINVAL) {
546 				skb->transport_header = skb->network_header;
547 				error = 0;
548 			}
549 			return error;
550 		}
551 
552 		nh = ip_hdr(skb);
553 		key->ipv4.addr.src = nh->saddr;
554 		key->ipv4.addr.dst = nh->daddr;
555 
556 		key->ip.proto = nh->protocol;
557 		key->ip.tos = nh->tos;
558 		key->ip.ttl = nh->ttl;
559 
560 		offset = nh->frag_off & htons(IP_OFFSET);
561 		if (offset) {
562 			key->ip.frag = OVS_FRAG_TYPE_LATER;
563 			memset(&key->tp, 0, sizeof(key->tp));
564 			return 0;
565 		}
566 		if (nh->frag_off & htons(IP_MF) ||
567 			skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
568 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
569 		else
570 			key->ip.frag = OVS_FRAG_TYPE_NONE;
571 
572 		/* Transport layer. */
573 		if (key->ip.proto == IPPROTO_TCP) {
574 			if (tcphdr_ok(skb)) {
575 				struct tcphdr *tcp = tcp_hdr(skb);
576 				key->tp.src = tcp->source;
577 				key->tp.dst = tcp->dest;
578 				key->tp.flags = TCP_FLAGS_BE16(tcp);
579 			} else {
580 				memset(&key->tp, 0, sizeof(key->tp));
581 			}
582 
583 		} else if (key->ip.proto == IPPROTO_UDP) {
584 			if (udphdr_ok(skb)) {
585 				struct udphdr *udp = udp_hdr(skb);
586 				key->tp.src = udp->source;
587 				key->tp.dst = udp->dest;
588 			} else {
589 				memset(&key->tp, 0, sizeof(key->tp));
590 			}
591 		} else if (key->ip.proto == IPPROTO_SCTP) {
592 			if (sctphdr_ok(skb)) {
593 				struct sctphdr *sctp = sctp_hdr(skb);
594 				key->tp.src = sctp->source;
595 				key->tp.dst = sctp->dest;
596 			} else {
597 				memset(&key->tp, 0, sizeof(key->tp));
598 			}
599 		} else if (key->ip.proto == IPPROTO_ICMP) {
600 			if (icmphdr_ok(skb)) {
601 				struct icmphdr *icmp = icmp_hdr(skb);
602 				/* The ICMP type and code fields use the 16-bit
603 				 * transport port fields, so we need to store
604 				 * them in 16-bit network byte order. */
605 				key->tp.src = htons(icmp->type);
606 				key->tp.dst = htons(icmp->code);
607 			} else {
608 				memset(&key->tp, 0, sizeof(key->tp));
609 			}
610 		}
611 
612 	} else if (key->eth.type == htons(ETH_P_ARP) ||
613 		   key->eth.type == htons(ETH_P_RARP)) {
614 		struct arp_eth_header *arp;
615 		bool arp_available = arphdr_ok(skb);
616 
617 		arp = (struct arp_eth_header *)skb_network_header(skb);
618 
619 		if (arp_available &&
620 		    arp->ar_hrd == htons(ARPHRD_ETHER) &&
621 		    arp->ar_pro == htons(ETH_P_IP) &&
622 		    arp->ar_hln == ETH_ALEN &&
623 		    arp->ar_pln == 4) {
624 
625 			/* We only match on the lower 8 bits of the opcode. */
626 			if (ntohs(arp->ar_op) <= 0xff)
627 				key->ip.proto = ntohs(arp->ar_op);
628 			else
629 				key->ip.proto = 0;
630 
631 			memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
632 			memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
633 			ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
634 			ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
635 		} else {
636 			memset(&key->ip, 0, sizeof(key->ip));
637 			memset(&key->ipv4, 0, sizeof(key->ipv4));
638 		}
639 	} else if (eth_p_mpls(key->eth.type)) {
640 		size_t stack_len = MPLS_HLEN;
641 
642 		skb_set_inner_network_header(skb, skb->mac_len);
643 		while (1) {
644 			__be32 lse;
645 
646 			error = check_header(skb, skb->mac_len + stack_len);
647 			if (unlikely(error))
648 				return 0;
649 
650 			memcpy(&lse, skb_inner_network_header(skb), MPLS_HLEN);
651 
652 			if (stack_len == MPLS_HLEN)
653 				memcpy(&key->mpls.top_lse, &lse, MPLS_HLEN);
654 
655 			skb_set_inner_network_header(skb, skb->mac_len + stack_len);
656 			if (lse & htonl(MPLS_LS_S_MASK))
657 				break;
658 
659 			stack_len += MPLS_HLEN;
660 		}
661 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
662 		int nh_len;             /* IPv6 Header + Extensions */
663 
664 		nh_len = parse_ipv6hdr(skb, key);
665 		if (unlikely(nh_len < 0)) {
666 			switch (nh_len) {
667 			case -EINVAL:
668 				memset(&key->ip, 0, sizeof(key->ip));
669 				memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
670 				/* fall-through */
671 			case -EPROTO:
672 				skb->transport_header = skb->network_header;
673 				error = 0;
674 				break;
675 			default:
676 				error = nh_len;
677 			}
678 			return error;
679 		}
680 
681 		if (key->ip.frag == OVS_FRAG_TYPE_LATER) {
682 			memset(&key->tp, 0, sizeof(key->tp));
683 			return 0;
684 		}
685 		if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
686 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
687 
688 		/* Transport layer. */
689 		if (key->ip.proto == NEXTHDR_TCP) {
690 			if (tcphdr_ok(skb)) {
691 				struct tcphdr *tcp = tcp_hdr(skb);
692 				key->tp.src = tcp->source;
693 				key->tp.dst = tcp->dest;
694 				key->tp.flags = TCP_FLAGS_BE16(tcp);
695 			} else {
696 				memset(&key->tp, 0, sizeof(key->tp));
697 			}
698 		} else if (key->ip.proto == NEXTHDR_UDP) {
699 			if (udphdr_ok(skb)) {
700 				struct udphdr *udp = udp_hdr(skb);
701 				key->tp.src = udp->source;
702 				key->tp.dst = udp->dest;
703 			} else {
704 				memset(&key->tp, 0, sizeof(key->tp));
705 			}
706 		} else if (key->ip.proto == NEXTHDR_SCTP) {
707 			if (sctphdr_ok(skb)) {
708 				struct sctphdr *sctp = sctp_hdr(skb);
709 				key->tp.src = sctp->source;
710 				key->tp.dst = sctp->dest;
711 			} else {
712 				memset(&key->tp, 0, sizeof(key->tp));
713 			}
714 		} else if (key->ip.proto == NEXTHDR_ICMP) {
715 			if (icmp6hdr_ok(skb)) {
716 				error = parse_icmpv6(skb, key, nh_len);
717 				if (error)
718 					return error;
719 			} else {
720 				memset(&key->tp, 0, sizeof(key->tp));
721 			}
722 		}
723 	} else if (key->eth.type == htons(ETH_P_NSH)) {
724 		error = parse_nsh(skb, key);
725 		if (error)
726 			return error;
727 	}
728 	return 0;
729 }
730 
731 /**
732  * key_extract - extracts a flow key from an Ethernet frame.
733  * @skb: sk_buff that contains the frame, with skb->data pointing to the
734  * Ethernet header
735  * @key: output flow key
736  *
737  * The caller must ensure that skb->len >= ETH_HLEN.
738  *
739  * Returns 0 if successful, otherwise a negative errno value.
740  *
741  * Initializes @skb header fields as follows:
742  *
743  *    - skb->mac_header: the L2 header.
744  *
745  *    - skb->network_header: just past the L2 header, or just past the
746  *      VLAN header, to the first byte of the L2 payload.
747  *
748  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
749  *      on output, then just past the IP header, if one is present and
750  *      of a correct length, otherwise the same as skb->network_header.
751  *      For other key->eth.type values it is left untouched.
752  *
753  *    - skb->protocol: the type of the data starting at skb->network_header.
754  *      Equals to key->eth.type.
755  */
756 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
757 {
758 	struct ethhdr *eth;
759 
760 	/* Flags are always used as part of stats */
761 	key->tp.flags = 0;
762 
763 	skb_reset_mac_header(skb);
764 
765 	/* Link layer. */
766 	clear_vlan(key);
767 	if (ovs_key_mac_proto(key) == MAC_PROTO_NONE) {
768 		if (unlikely(eth_type_vlan(skb->protocol)))
769 			return -EINVAL;
770 
771 		skb_reset_network_header(skb);
772 		key->eth.type = skb->protocol;
773 	} else {
774 		eth = eth_hdr(skb);
775 		ether_addr_copy(key->eth.src, eth->h_source);
776 		ether_addr_copy(key->eth.dst, eth->h_dest);
777 
778 		__skb_pull(skb, 2 * ETH_ALEN);
779 		/* We are going to push all headers that we pull, so no need to
780 		 * update skb->csum here.
781 		 */
782 
783 		if (unlikely(parse_vlan(skb, key)))
784 			return -ENOMEM;
785 
786 		key->eth.type = parse_ethertype(skb);
787 		if (unlikely(key->eth.type == htons(0)))
788 			return -ENOMEM;
789 
790 		/* Multiple tagged packets need to retain TPID to satisfy
791 		 * skb_vlan_pop(), which will later shift the ethertype into
792 		 * skb->protocol.
793 		 */
794 		if (key->eth.cvlan.tci & htons(VLAN_CFI_MASK))
795 			skb->protocol = key->eth.cvlan.tpid;
796 		else
797 			skb->protocol = key->eth.type;
798 
799 		skb_reset_network_header(skb);
800 		__skb_push(skb, skb->data - skb_mac_header(skb));
801 	}
802 
803 	skb_reset_mac_len(skb);
804 
805 	/* Fill out L3/L4 key info, if any */
806 	return key_extract_l3l4(skb, key);
807 }
808 
809 /* In the case of conntrack fragment handling it expects L3 headers,
810  * add a helper.
811  */
812 int ovs_flow_key_update_l3l4(struct sk_buff *skb, struct sw_flow_key *key)
813 {
814 	return key_extract_l3l4(skb, key);
815 }
816 
817 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
818 {
819 	int res;
820 
821 	res = key_extract(skb, key);
822 	if (!res)
823 		key->mac_proto &= ~SW_FLOW_KEY_INVALID;
824 
825 	return res;
826 }
827 
828 static int key_extract_mac_proto(struct sk_buff *skb)
829 {
830 	switch (skb->dev->type) {
831 	case ARPHRD_ETHER:
832 		return MAC_PROTO_ETHERNET;
833 	case ARPHRD_NONE:
834 		if (skb->protocol == htons(ETH_P_TEB))
835 			return MAC_PROTO_ETHERNET;
836 		return MAC_PROTO_NONE;
837 	}
838 	WARN_ON_ONCE(1);
839 	return -EINVAL;
840 }
841 
842 int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
843 			 struct sk_buff *skb, struct sw_flow_key *key)
844 {
845 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
846 	struct tc_skb_ext *tc_ext;
847 #endif
848 	int res, err;
849 
850 	/* Extract metadata from packet. */
851 	if (tun_info) {
852 		key->tun_proto = ip_tunnel_info_af(tun_info);
853 		memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
854 
855 		if (tun_info->options_len) {
856 			BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) *
857 						   8)) - 1
858 					> sizeof(key->tun_opts));
859 
860 			ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key, tun_info->options_len),
861 						tun_info);
862 			key->tun_opts_len = tun_info->options_len;
863 		} else {
864 			key->tun_opts_len = 0;
865 		}
866 	} else  {
867 		key->tun_proto = 0;
868 		key->tun_opts_len = 0;
869 		memset(&key->tun_key, 0, sizeof(key->tun_key));
870 	}
871 
872 	key->phy.priority = skb->priority;
873 	key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
874 	key->phy.skb_mark = skb->mark;
875 	key->ovs_flow_hash = 0;
876 	res = key_extract_mac_proto(skb);
877 	if (res < 0)
878 		return res;
879 	key->mac_proto = res;
880 
881 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
882 	if (static_branch_unlikely(&tc_recirc_sharing_support)) {
883 		tc_ext = skb_ext_find(skb, TC_SKB_EXT);
884 		key->recirc_id = tc_ext ? tc_ext->chain : 0;
885 	} else {
886 		key->recirc_id = 0;
887 	}
888 #else
889 	key->recirc_id = 0;
890 #endif
891 
892 	err = key_extract(skb, key);
893 	if (!err)
894 		ovs_ct_fill_key(skb, key);   /* Must be after key_extract(). */
895 	return err;
896 }
897 
898 int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
899 				   struct sk_buff *skb,
900 				   struct sw_flow_key *key, bool log)
901 {
902 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
903 	u64 attrs = 0;
904 	int err;
905 
906 	err = parse_flow_nlattrs(attr, a, &attrs, log);
907 	if (err)
908 		return -EINVAL;
909 
910 	/* Extract metadata from netlink attributes. */
911 	err = ovs_nla_get_flow_metadata(net, a, attrs, key, log);
912 	if (err)
913 		return err;
914 
915 	/* key_extract assumes that skb->protocol is set-up for
916 	 * layer 3 packets which is the case for other callers,
917 	 * in particular packets received from the network stack.
918 	 * Here the correct value can be set from the metadata
919 	 * extracted above.
920 	 * For L2 packet key eth type would be zero. skb protocol
921 	 * would be set to correct value later during key-extact.
922 	 */
923 
924 	skb->protocol = key->eth.type;
925 	err = key_extract(skb, key);
926 	if (err)
927 		return err;
928 
929 	/* Check that we have conntrack original direction tuple metadata only
930 	 * for packets for which it makes sense.  Otherwise the key may be
931 	 * corrupted due to overlapping key fields.
932 	 */
933 	if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4) &&
934 	    key->eth.type != htons(ETH_P_IP))
935 		return -EINVAL;
936 	if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6) &&
937 	    (key->eth.type != htons(ETH_P_IPV6) ||
938 	     sw_flow_key_is_nd(key)))
939 		return -EINVAL;
940 
941 	return 0;
942 }
943