xref: /openbmc/linux/net/openvswitch/flow.c (revision 5b448065)
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  * @skb: skb containing frame to parse
298  * @key_vh: pointer to parsed vlan tag
299  * @untag_vlan: should the vlan header be removed from the frame
300  *
301  * Returns ERROR on memory error.
302  * Returns 0 if it encounters a non-vlan or incomplete packet.
303  * Returns 1 after successfully parsing vlan tag.
304  */
305 static int parse_vlan_tag(struct sk_buff *skb, struct vlan_head *key_vh,
306 			  bool untag_vlan)
307 {
308 	struct vlan_head *vh = (struct vlan_head *)skb->data;
309 
310 	if (likely(!eth_type_vlan(vh->tpid)))
311 		return 0;
312 
313 	if (unlikely(skb->len < sizeof(struct vlan_head) + sizeof(__be16)))
314 		return 0;
315 
316 	if (unlikely(!pskb_may_pull(skb, sizeof(struct vlan_head) +
317 				 sizeof(__be16))))
318 		return -ENOMEM;
319 
320 	vh = (struct vlan_head *)skb->data;
321 	key_vh->tci = vh->tci | htons(VLAN_CFI_MASK);
322 	key_vh->tpid = vh->tpid;
323 
324 	if (unlikely(untag_vlan)) {
325 		int offset = skb->data - skb_mac_header(skb);
326 		u16 tci;
327 		int err;
328 
329 		__skb_push(skb, offset);
330 		err = __skb_vlan_pop(skb, &tci);
331 		__skb_pull(skb, offset);
332 		if (err)
333 			return err;
334 		__vlan_hwaccel_put_tag(skb, key_vh->tpid, tci);
335 	} else {
336 		__skb_pull(skb, sizeof(struct vlan_head));
337 	}
338 	return 1;
339 }
340 
341 static void clear_vlan(struct sw_flow_key *key)
342 {
343 	key->eth.vlan.tci = 0;
344 	key->eth.vlan.tpid = 0;
345 	key->eth.cvlan.tci = 0;
346 	key->eth.cvlan.tpid = 0;
347 }
348 
349 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
350 {
351 	int res;
352 
353 	if (skb_vlan_tag_present(skb)) {
354 		key->eth.vlan.tci = htons(skb->vlan_tci) | htons(VLAN_CFI_MASK);
355 		key->eth.vlan.tpid = skb->vlan_proto;
356 	} else {
357 		/* Parse outer vlan tag in the non-accelerated case. */
358 		res = parse_vlan_tag(skb, &key->eth.vlan, true);
359 		if (res <= 0)
360 			return res;
361 	}
362 
363 	/* Parse inner vlan tag. */
364 	res = parse_vlan_tag(skb, &key->eth.cvlan, false);
365 	if (res <= 0)
366 		return res;
367 
368 	return 0;
369 }
370 
371 static __be16 parse_ethertype(struct sk_buff *skb)
372 {
373 	struct llc_snap_hdr {
374 		u8  dsap;  /* Always 0xAA */
375 		u8  ssap;  /* Always 0xAA */
376 		u8  ctrl;
377 		u8  oui[3];
378 		__be16 ethertype;
379 	};
380 	struct llc_snap_hdr *llc;
381 	__be16 proto;
382 
383 	proto = *(__be16 *) skb->data;
384 	__skb_pull(skb, sizeof(__be16));
385 
386 	if (eth_proto_is_802_3(proto))
387 		return proto;
388 
389 	if (skb->len < sizeof(struct llc_snap_hdr))
390 		return htons(ETH_P_802_2);
391 
392 	if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
393 		return htons(0);
394 
395 	llc = (struct llc_snap_hdr *) skb->data;
396 	if (llc->dsap != LLC_SAP_SNAP ||
397 	    llc->ssap != LLC_SAP_SNAP ||
398 	    (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
399 		return htons(ETH_P_802_2);
400 
401 	__skb_pull(skb, sizeof(struct llc_snap_hdr));
402 
403 	if (eth_proto_is_802_3(llc->ethertype))
404 		return llc->ethertype;
405 
406 	return htons(ETH_P_802_2);
407 }
408 
409 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
410 			int nh_len)
411 {
412 	struct icmp6hdr *icmp = icmp6_hdr(skb);
413 
414 	/* The ICMPv6 type and code fields use the 16-bit transport port
415 	 * fields, so we need to store them in 16-bit network byte order.
416 	 */
417 	key->tp.src = htons(icmp->icmp6_type);
418 	key->tp.dst = htons(icmp->icmp6_code);
419 	memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
420 
421 	if (icmp->icmp6_code == 0 &&
422 	    (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
423 	     icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
424 		int icmp_len = skb->len - skb_transport_offset(skb);
425 		struct nd_msg *nd;
426 		int offset;
427 
428 		/* In order to process neighbor discovery options, we need the
429 		 * entire packet.
430 		 */
431 		if (unlikely(icmp_len < sizeof(*nd)))
432 			return 0;
433 
434 		if (unlikely(skb_linearize(skb)))
435 			return -ENOMEM;
436 
437 		nd = (struct nd_msg *)skb_transport_header(skb);
438 		key->ipv6.nd.target = nd->target;
439 
440 		icmp_len -= sizeof(*nd);
441 		offset = 0;
442 		while (icmp_len >= 8) {
443 			struct nd_opt_hdr *nd_opt =
444 				 (struct nd_opt_hdr *)(nd->opt + offset);
445 			int opt_len = nd_opt->nd_opt_len * 8;
446 
447 			if (unlikely(!opt_len || opt_len > icmp_len))
448 				return 0;
449 
450 			/* Store the link layer address if the appropriate
451 			 * option is provided.  It is considered an error if
452 			 * the same link layer option is specified twice.
453 			 */
454 			if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
455 			    && opt_len == 8) {
456 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
457 					goto invalid;
458 				ether_addr_copy(key->ipv6.nd.sll,
459 						&nd->opt[offset+sizeof(*nd_opt)]);
460 			} else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
461 				   && opt_len == 8) {
462 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
463 					goto invalid;
464 				ether_addr_copy(key->ipv6.nd.tll,
465 						&nd->opt[offset+sizeof(*nd_opt)]);
466 			}
467 
468 			icmp_len -= opt_len;
469 			offset += opt_len;
470 		}
471 	}
472 
473 	return 0;
474 
475 invalid:
476 	memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
477 	memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
478 	memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
479 
480 	return 0;
481 }
482 
483 static int parse_nsh(struct sk_buff *skb, struct sw_flow_key *key)
484 {
485 	struct nshhdr *nh;
486 	unsigned int nh_ofs = skb_network_offset(skb);
487 	u8 version, length;
488 	int err;
489 
490 	err = check_header(skb, nh_ofs + NSH_BASE_HDR_LEN);
491 	if (unlikely(err))
492 		return err;
493 
494 	nh = nsh_hdr(skb);
495 	version = nsh_get_ver(nh);
496 	length = nsh_hdr_len(nh);
497 
498 	if (version != 0)
499 		return -EINVAL;
500 
501 	err = check_header(skb, nh_ofs + length);
502 	if (unlikely(err))
503 		return err;
504 
505 	nh = nsh_hdr(skb);
506 	key->nsh.base.flags = nsh_get_flags(nh);
507 	key->nsh.base.ttl = nsh_get_ttl(nh);
508 	key->nsh.base.mdtype = nh->mdtype;
509 	key->nsh.base.np = nh->np;
510 	key->nsh.base.path_hdr = nh->path_hdr;
511 	switch (key->nsh.base.mdtype) {
512 	case NSH_M_TYPE1:
513 		if (length != NSH_M_TYPE1_LEN)
514 			return -EINVAL;
515 		memcpy(key->nsh.context, nh->md1.context,
516 		       sizeof(nh->md1));
517 		break;
518 	case NSH_M_TYPE2:
519 		memset(key->nsh.context, 0,
520 		       sizeof(nh->md1));
521 		break;
522 	default:
523 		return -EINVAL;
524 	}
525 
526 	return 0;
527 }
528 
529 /**
530  * key_extract_l3l4 - extracts L3/L4 header information.
531  * @skb: sk_buff that contains the frame, with skb->data pointing to the
532  *       L3 header
533  * @key: output flow key
534  *
535  */
536 static int key_extract_l3l4(struct sk_buff *skb, struct sw_flow_key *key)
537 {
538 	int error;
539 
540 	/* Network layer. */
541 	if (key->eth.type == htons(ETH_P_IP)) {
542 		struct iphdr *nh;
543 		__be16 offset;
544 
545 		error = check_iphdr(skb);
546 		if (unlikely(error)) {
547 			memset(&key->ip, 0, sizeof(key->ip));
548 			memset(&key->ipv4, 0, sizeof(key->ipv4));
549 			if (error == -EINVAL) {
550 				skb->transport_header = skb->network_header;
551 				error = 0;
552 			}
553 			return error;
554 		}
555 
556 		nh = ip_hdr(skb);
557 		key->ipv4.addr.src = nh->saddr;
558 		key->ipv4.addr.dst = nh->daddr;
559 
560 		key->ip.proto = nh->protocol;
561 		key->ip.tos = nh->tos;
562 		key->ip.ttl = nh->ttl;
563 
564 		offset = nh->frag_off & htons(IP_OFFSET);
565 		if (offset) {
566 			key->ip.frag = OVS_FRAG_TYPE_LATER;
567 			memset(&key->tp, 0, sizeof(key->tp));
568 			return 0;
569 		}
570 		if (nh->frag_off & htons(IP_MF) ||
571 			skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
572 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
573 		else
574 			key->ip.frag = OVS_FRAG_TYPE_NONE;
575 
576 		/* Transport layer. */
577 		if (key->ip.proto == IPPROTO_TCP) {
578 			if (tcphdr_ok(skb)) {
579 				struct tcphdr *tcp = tcp_hdr(skb);
580 				key->tp.src = tcp->source;
581 				key->tp.dst = tcp->dest;
582 				key->tp.flags = TCP_FLAGS_BE16(tcp);
583 			} else {
584 				memset(&key->tp, 0, sizeof(key->tp));
585 			}
586 
587 		} else if (key->ip.proto == IPPROTO_UDP) {
588 			if (udphdr_ok(skb)) {
589 				struct udphdr *udp = udp_hdr(skb);
590 				key->tp.src = udp->source;
591 				key->tp.dst = udp->dest;
592 			} else {
593 				memset(&key->tp, 0, sizeof(key->tp));
594 			}
595 		} else if (key->ip.proto == IPPROTO_SCTP) {
596 			if (sctphdr_ok(skb)) {
597 				struct sctphdr *sctp = sctp_hdr(skb);
598 				key->tp.src = sctp->source;
599 				key->tp.dst = sctp->dest;
600 			} else {
601 				memset(&key->tp, 0, sizeof(key->tp));
602 			}
603 		} else if (key->ip.proto == IPPROTO_ICMP) {
604 			if (icmphdr_ok(skb)) {
605 				struct icmphdr *icmp = icmp_hdr(skb);
606 				/* The ICMP type and code fields use the 16-bit
607 				 * transport port fields, so we need to store
608 				 * them in 16-bit network byte order. */
609 				key->tp.src = htons(icmp->type);
610 				key->tp.dst = htons(icmp->code);
611 			} else {
612 				memset(&key->tp, 0, sizeof(key->tp));
613 			}
614 		}
615 
616 	} else if (key->eth.type == htons(ETH_P_ARP) ||
617 		   key->eth.type == htons(ETH_P_RARP)) {
618 		struct arp_eth_header *arp;
619 		bool arp_available = arphdr_ok(skb);
620 
621 		arp = (struct arp_eth_header *)skb_network_header(skb);
622 
623 		if (arp_available &&
624 		    arp->ar_hrd == htons(ARPHRD_ETHER) &&
625 		    arp->ar_pro == htons(ETH_P_IP) &&
626 		    arp->ar_hln == ETH_ALEN &&
627 		    arp->ar_pln == 4) {
628 
629 			/* We only match on the lower 8 bits of the opcode. */
630 			if (ntohs(arp->ar_op) <= 0xff)
631 				key->ip.proto = ntohs(arp->ar_op);
632 			else
633 				key->ip.proto = 0;
634 
635 			memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
636 			memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
637 			ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
638 			ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
639 		} else {
640 			memset(&key->ip, 0, sizeof(key->ip));
641 			memset(&key->ipv4, 0, sizeof(key->ipv4));
642 		}
643 	} else if (eth_p_mpls(key->eth.type)) {
644 		u8 label_count = 1;
645 
646 		memset(&key->mpls, 0, sizeof(key->mpls));
647 		skb_set_inner_network_header(skb, skb->mac_len);
648 		while (1) {
649 			__be32 lse;
650 
651 			error = check_header(skb, skb->mac_len +
652 					     label_count * MPLS_HLEN);
653 			if (unlikely(error))
654 				return 0;
655 
656 			memcpy(&lse, skb_inner_network_header(skb), MPLS_HLEN);
657 
658 			if (label_count <= MPLS_LABEL_DEPTH)
659 				memcpy(&key->mpls.lse[label_count - 1], &lse,
660 				       MPLS_HLEN);
661 
662 			skb_set_inner_network_header(skb, skb->mac_len +
663 						     label_count * MPLS_HLEN);
664 			if (lse & htonl(MPLS_LS_S_MASK))
665 				break;
666 
667 			label_count++;
668 		}
669 		if (label_count > MPLS_LABEL_DEPTH)
670 			label_count = MPLS_LABEL_DEPTH;
671 
672 		key->mpls.num_labels_mask = GENMASK(label_count - 1, 0);
673 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
674 		int nh_len;             /* IPv6 Header + Extensions */
675 
676 		nh_len = parse_ipv6hdr(skb, key);
677 		if (unlikely(nh_len < 0)) {
678 			switch (nh_len) {
679 			case -EINVAL:
680 				memset(&key->ip, 0, sizeof(key->ip));
681 				memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
682 				fallthrough;
683 			case -EPROTO:
684 				skb->transport_header = skb->network_header;
685 				error = 0;
686 				break;
687 			default:
688 				error = nh_len;
689 			}
690 			return error;
691 		}
692 
693 		if (key->ip.frag == OVS_FRAG_TYPE_LATER) {
694 			memset(&key->tp, 0, sizeof(key->tp));
695 			return 0;
696 		}
697 		if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
698 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
699 
700 		/* Transport layer. */
701 		if (key->ip.proto == NEXTHDR_TCP) {
702 			if (tcphdr_ok(skb)) {
703 				struct tcphdr *tcp = tcp_hdr(skb);
704 				key->tp.src = tcp->source;
705 				key->tp.dst = tcp->dest;
706 				key->tp.flags = TCP_FLAGS_BE16(tcp);
707 			} else {
708 				memset(&key->tp, 0, sizeof(key->tp));
709 			}
710 		} else if (key->ip.proto == NEXTHDR_UDP) {
711 			if (udphdr_ok(skb)) {
712 				struct udphdr *udp = udp_hdr(skb);
713 				key->tp.src = udp->source;
714 				key->tp.dst = udp->dest;
715 			} else {
716 				memset(&key->tp, 0, sizeof(key->tp));
717 			}
718 		} else if (key->ip.proto == NEXTHDR_SCTP) {
719 			if (sctphdr_ok(skb)) {
720 				struct sctphdr *sctp = sctp_hdr(skb);
721 				key->tp.src = sctp->source;
722 				key->tp.dst = sctp->dest;
723 			} else {
724 				memset(&key->tp, 0, sizeof(key->tp));
725 			}
726 		} else if (key->ip.proto == NEXTHDR_ICMP) {
727 			if (icmp6hdr_ok(skb)) {
728 				error = parse_icmpv6(skb, key, nh_len);
729 				if (error)
730 					return error;
731 			} else {
732 				memset(&key->tp, 0, sizeof(key->tp));
733 			}
734 		}
735 	} else if (key->eth.type == htons(ETH_P_NSH)) {
736 		error = parse_nsh(skb, key);
737 		if (error)
738 			return error;
739 	}
740 	return 0;
741 }
742 
743 /**
744  * key_extract - extracts a flow key from an Ethernet frame.
745  * @skb: sk_buff that contains the frame, with skb->data pointing to the
746  * Ethernet header
747  * @key: output flow key
748  *
749  * The caller must ensure that skb->len >= ETH_HLEN.
750  *
751  * Returns 0 if successful, otherwise a negative errno value.
752  *
753  * Initializes @skb header fields as follows:
754  *
755  *    - skb->mac_header: the L2 header.
756  *
757  *    - skb->network_header: just past the L2 header, or just past the
758  *      VLAN header, to the first byte of the L2 payload.
759  *
760  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
761  *      on output, then just past the IP header, if one is present and
762  *      of a correct length, otherwise the same as skb->network_header.
763  *      For other key->eth.type values it is left untouched.
764  *
765  *    - skb->protocol: the type of the data starting at skb->network_header.
766  *      Equals to key->eth.type.
767  */
768 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
769 {
770 	struct ethhdr *eth;
771 
772 	/* Flags are always used as part of stats */
773 	key->tp.flags = 0;
774 
775 	skb_reset_mac_header(skb);
776 
777 	/* Link layer. */
778 	clear_vlan(key);
779 	if (ovs_key_mac_proto(key) == MAC_PROTO_NONE) {
780 		if (unlikely(eth_type_vlan(skb->protocol)))
781 			return -EINVAL;
782 
783 		skb_reset_network_header(skb);
784 		key->eth.type = skb->protocol;
785 	} else {
786 		eth = eth_hdr(skb);
787 		ether_addr_copy(key->eth.src, eth->h_source);
788 		ether_addr_copy(key->eth.dst, eth->h_dest);
789 
790 		__skb_pull(skb, 2 * ETH_ALEN);
791 		/* We are going to push all headers that we pull, so no need to
792 		 * update skb->csum here.
793 		 */
794 
795 		if (unlikely(parse_vlan(skb, key)))
796 			return -ENOMEM;
797 
798 		key->eth.type = parse_ethertype(skb);
799 		if (unlikely(key->eth.type == htons(0)))
800 			return -ENOMEM;
801 
802 		/* Multiple tagged packets need to retain TPID to satisfy
803 		 * skb_vlan_pop(), which will later shift the ethertype into
804 		 * skb->protocol.
805 		 */
806 		if (key->eth.cvlan.tci & htons(VLAN_CFI_MASK))
807 			skb->protocol = key->eth.cvlan.tpid;
808 		else
809 			skb->protocol = key->eth.type;
810 
811 		skb_reset_network_header(skb);
812 		__skb_push(skb, skb->data - skb_mac_header(skb));
813 	}
814 
815 	skb_reset_mac_len(skb);
816 
817 	/* Fill out L3/L4 key info, if any */
818 	return key_extract_l3l4(skb, key);
819 }
820 
821 /* In the case of conntrack fragment handling it expects L3 headers,
822  * add a helper.
823  */
824 int ovs_flow_key_update_l3l4(struct sk_buff *skb, struct sw_flow_key *key)
825 {
826 	return key_extract_l3l4(skb, key);
827 }
828 
829 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
830 {
831 	int res;
832 
833 	res = key_extract(skb, key);
834 	if (!res)
835 		key->mac_proto &= ~SW_FLOW_KEY_INVALID;
836 
837 	return res;
838 }
839 
840 static int key_extract_mac_proto(struct sk_buff *skb)
841 {
842 	switch (skb->dev->type) {
843 	case ARPHRD_ETHER:
844 		return MAC_PROTO_ETHERNET;
845 	case ARPHRD_NONE:
846 		if (skb->protocol == htons(ETH_P_TEB))
847 			return MAC_PROTO_ETHERNET;
848 		return MAC_PROTO_NONE;
849 	}
850 	WARN_ON_ONCE(1);
851 	return -EINVAL;
852 }
853 
854 int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
855 			 struct sk_buff *skb, struct sw_flow_key *key)
856 {
857 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
858 	struct tc_skb_ext *tc_ext;
859 #endif
860 	bool post_ct = false;
861 	int res, err;
862 
863 	/* Extract metadata from packet. */
864 	if (tun_info) {
865 		key->tun_proto = ip_tunnel_info_af(tun_info);
866 		memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
867 
868 		if (tun_info->options_len) {
869 			BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) *
870 						   8)) - 1
871 					> sizeof(key->tun_opts));
872 
873 			ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key, tun_info->options_len),
874 						tun_info);
875 			key->tun_opts_len = tun_info->options_len;
876 		} else {
877 			key->tun_opts_len = 0;
878 		}
879 	} else  {
880 		key->tun_proto = 0;
881 		key->tun_opts_len = 0;
882 		memset(&key->tun_key, 0, sizeof(key->tun_key));
883 	}
884 
885 	key->phy.priority = skb->priority;
886 	key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
887 	key->phy.skb_mark = skb->mark;
888 	key->ovs_flow_hash = 0;
889 	res = key_extract_mac_proto(skb);
890 	if (res < 0)
891 		return res;
892 	key->mac_proto = res;
893 
894 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
895 	if (static_branch_unlikely(&tc_recirc_sharing_support)) {
896 		tc_ext = skb_ext_find(skb, TC_SKB_EXT);
897 		key->recirc_id = tc_ext ? tc_ext->chain : 0;
898 		OVS_CB(skb)->mru = tc_ext ? tc_ext->mru : 0;
899 		post_ct = tc_ext ? tc_ext->post_ct : false;
900 	} else {
901 		key->recirc_id = 0;
902 	}
903 #else
904 	key->recirc_id = 0;
905 #endif
906 
907 	err = key_extract(skb, key);
908 	if (!err)
909 		ovs_ct_fill_key(skb, key, post_ct);   /* Must be after key_extract(). */
910 	return err;
911 }
912 
913 int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
914 				   struct sk_buff *skb,
915 				   struct sw_flow_key *key, bool log)
916 {
917 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
918 	u64 attrs = 0;
919 	int err;
920 
921 	err = parse_flow_nlattrs(attr, a, &attrs, log);
922 	if (err)
923 		return -EINVAL;
924 
925 	/* Extract metadata from netlink attributes. */
926 	err = ovs_nla_get_flow_metadata(net, a, attrs, key, log);
927 	if (err)
928 		return err;
929 
930 	/* key_extract assumes that skb->protocol is set-up for
931 	 * layer 3 packets which is the case for other callers,
932 	 * in particular packets received from the network stack.
933 	 * Here the correct value can be set from the metadata
934 	 * extracted above.
935 	 * For L2 packet key eth type would be zero. skb protocol
936 	 * would be set to correct value later during key-extact.
937 	 */
938 
939 	skb->protocol = key->eth.type;
940 	err = key_extract(skb, key);
941 	if (err)
942 		return err;
943 
944 	/* Check that we have conntrack original direction tuple metadata only
945 	 * for packets for which it makes sense.  Otherwise the key may be
946 	 * corrupted due to overlapping key fields.
947 	 */
948 	if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4) &&
949 	    key->eth.type != htons(ETH_P_IP))
950 		return -EINVAL;
951 	if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6) &&
952 	    (key->eth.type != htons(ETH_P_IPV6) ||
953 	     sw_flow_key_is_nd(key)))
954 		return -EINVAL;
955 
956 	return 0;
957 }
958