xref: /openbmc/linux/net/openvswitch/flow.c (revision 5c816641)
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 #include <net/netfilter/nf_conntrack_zones.h>
38 
39 #include "conntrack.h"
40 #include "datapath.h"
41 #include "flow.h"
42 #include "flow_netlink.h"
43 #include "vport.h"
44 
45 u64 ovs_flow_used_time(unsigned long flow_jiffies)
46 {
47 	struct timespec64 cur_ts;
48 	u64 cur_ms, idle_ms;
49 
50 	ktime_get_ts64(&cur_ts);
51 	idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
52 	cur_ms = (u64)(u32)cur_ts.tv_sec * MSEC_PER_SEC +
53 		 cur_ts.tv_nsec / NSEC_PER_MSEC;
54 
55 	return cur_ms - idle_ms;
56 }
57 
58 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
59 
60 void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
61 			   const struct sk_buff *skb)
62 {
63 	struct sw_flow_stats *stats;
64 	unsigned int cpu = smp_processor_id();
65 	int len = skb->len + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
66 
67 	stats = rcu_dereference(flow->stats[cpu]);
68 
69 	/* Check if already have CPU-specific stats. */
70 	if (likely(stats)) {
71 		spin_lock(&stats->lock);
72 		/* Mark if we write on the pre-allocated stats. */
73 		if (cpu == 0 && unlikely(flow->stats_last_writer != cpu))
74 			flow->stats_last_writer = cpu;
75 	} else {
76 		stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
77 		spin_lock(&stats->lock);
78 
79 		/* If the current CPU is the only writer on the
80 		 * pre-allocated stats keep using them.
81 		 */
82 		if (unlikely(flow->stats_last_writer != cpu)) {
83 			/* A previous locker may have already allocated the
84 			 * stats, so we need to check again.  If CPU-specific
85 			 * stats were already allocated, we update the pre-
86 			 * allocated stats as we have already locked them.
87 			 */
88 			if (likely(flow->stats_last_writer != -1) &&
89 			    likely(!rcu_access_pointer(flow->stats[cpu]))) {
90 				/* Try to allocate CPU-specific stats. */
91 				struct sw_flow_stats *new_stats;
92 
93 				new_stats =
94 					kmem_cache_alloc_node(flow_stats_cache,
95 							      GFP_NOWAIT |
96 							      __GFP_THISNODE |
97 							      __GFP_NOWARN |
98 							      __GFP_NOMEMALLOC,
99 							      numa_node_id());
100 				if (likely(new_stats)) {
101 					new_stats->used = jiffies;
102 					new_stats->packet_count = 1;
103 					new_stats->byte_count = len;
104 					new_stats->tcp_flags = tcp_flags;
105 					spin_lock_init(&new_stats->lock);
106 
107 					rcu_assign_pointer(flow->stats[cpu],
108 							   new_stats);
109 					cpumask_set_cpu(cpu, &flow->cpu_used_mask);
110 					goto unlock;
111 				}
112 			}
113 			flow->stats_last_writer = cpu;
114 		}
115 	}
116 
117 	stats->used = jiffies;
118 	stats->packet_count++;
119 	stats->byte_count += len;
120 	stats->tcp_flags |= tcp_flags;
121 unlock:
122 	spin_unlock(&stats->lock);
123 }
124 
125 /* Must be called with rcu_read_lock or ovs_mutex. */
126 void ovs_flow_stats_get(const struct sw_flow *flow,
127 			struct ovs_flow_stats *ovs_stats,
128 			unsigned long *used, __be16 *tcp_flags)
129 {
130 	int cpu;
131 
132 	*used = 0;
133 	*tcp_flags = 0;
134 	memset(ovs_stats, 0, sizeof(*ovs_stats));
135 
136 	/* We open code this to make sure cpu 0 is always considered */
137 	for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask)) {
138 		struct sw_flow_stats *stats = rcu_dereference_ovsl(flow->stats[cpu]);
139 
140 		if (stats) {
141 			/* Local CPU may write on non-local stats, so we must
142 			 * block bottom-halves here.
143 			 */
144 			spin_lock_bh(&stats->lock);
145 			if (!*used || time_after(stats->used, *used))
146 				*used = stats->used;
147 			*tcp_flags |= stats->tcp_flags;
148 			ovs_stats->n_packets += stats->packet_count;
149 			ovs_stats->n_bytes += stats->byte_count;
150 			spin_unlock_bh(&stats->lock);
151 		}
152 	}
153 }
154 
155 /* Called with ovs_mutex. */
156 void ovs_flow_stats_clear(struct sw_flow *flow)
157 {
158 	int cpu;
159 
160 	/* We open code this to make sure cpu 0 is always considered */
161 	for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask)) {
162 		struct sw_flow_stats *stats = ovsl_dereference(flow->stats[cpu]);
163 
164 		if (stats) {
165 			spin_lock_bh(&stats->lock);
166 			stats->used = 0;
167 			stats->packet_count = 0;
168 			stats->byte_count = 0;
169 			stats->tcp_flags = 0;
170 			spin_unlock_bh(&stats->lock);
171 		}
172 	}
173 }
174 
175 static int check_header(struct sk_buff *skb, int len)
176 {
177 	if (unlikely(skb->len < len))
178 		return -EINVAL;
179 	if (unlikely(!pskb_may_pull(skb, len)))
180 		return -ENOMEM;
181 	return 0;
182 }
183 
184 static bool arphdr_ok(struct sk_buff *skb)
185 {
186 	return pskb_may_pull(skb, skb_network_offset(skb) +
187 				  sizeof(struct arp_eth_header));
188 }
189 
190 static int check_iphdr(struct sk_buff *skb)
191 {
192 	unsigned int nh_ofs = skb_network_offset(skb);
193 	unsigned int ip_len;
194 	int err;
195 
196 	err = check_header(skb, nh_ofs + sizeof(struct iphdr));
197 	if (unlikely(err))
198 		return err;
199 
200 	ip_len = ip_hdrlen(skb);
201 	if (unlikely(ip_len < sizeof(struct iphdr) ||
202 		     skb->len < nh_ofs + ip_len))
203 		return -EINVAL;
204 
205 	skb_set_transport_header(skb, nh_ofs + ip_len);
206 	return 0;
207 }
208 
209 static bool tcphdr_ok(struct sk_buff *skb)
210 {
211 	int th_ofs = skb_transport_offset(skb);
212 	int tcp_len;
213 
214 	if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
215 		return false;
216 
217 	tcp_len = tcp_hdrlen(skb);
218 	if (unlikely(tcp_len < sizeof(struct tcphdr) ||
219 		     skb->len < th_ofs + tcp_len))
220 		return false;
221 
222 	return true;
223 }
224 
225 static bool udphdr_ok(struct sk_buff *skb)
226 {
227 	return pskb_may_pull(skb, skb_transport_offset(skb) +
228 				  sizeof(struct udphdr));
229 }
230 
231 static bool sctphdr_ok(struct sk_buff *skb)
232 {
233 	return pskb_may_pull(skb, skb_transport_offset(skb) +
234 				  sizeof(struct sctphdr));
235 }
236 
237 static bool icmphdr_ok(struct sk_buff *skb)
238 {
239 	return pskb_may_pull(skb, skb_transport_offset(skb) +
240 				  sizeof(struct icmphdr));
241 }
242 
243 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
244 {
245 	unsigned short frag_off;
246 	unsigned int payload_ofs = 0;
247 	unsigned int nh_ofs = skb_network_offset(skb);
248 	unsigned int nh_len;
249 	struct ipv6hdr *nh;
250 	int err, nexthdr, flags = 0;
251 
252 	err = check_header(skb, nh_ofs + sizeof(*nh));
253 	if (unlikely(err))
254 		return err;
255 
256 	nh = ipv6_hdr(skb);
257 
258 	key->ip.proto = NEXTHDR_NONE;
259 	key->ip.tos = ipv6_get_dsfield(nh);
260 	key->ip.ttl = nh->hop_limit;
261 	key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
262 	key->ipv6.addr.src = nh->saddr;
263 	key->ipv6.addr.dst = nh->daddr;
264 
265 	nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
266 	if (flags & IP6_FH_F_FRAG) {
267 		if (frag_off) {
268 			key->ip.frag = OVS_FRAG_TYPE_LATER;
269 			key->ip.proto = nexthdr;
270 			return 0;
271 		}
272 		key->ip.frag = OVS_FRAG_TYPE_FIRST;
273 	} else {
274 		key->ip.frag = OVS_FRAG_TYPE_NONE;
275 	}
276 
277 	/* Delayed handling of error in ipv6_find_hdr() as it
278 	 * always sets flags and frag_off to a valid value which may be
279 	 * used to set key->ip.frag above.
280 	 */
281 	if (unlikely(nexthdr < 0))
282 		return -EPROTO;
283 
284 	nh_len = payload_ofs - nh_ofs;
285 	skb_set_transport_header(skb, nh_ofs + nh_len);
286 	key->ip.proto = nexthdr;
287 	return nh_len;
288 }
289 
290 static bool icmp6hdr_ok(struct sk_buff *skb)
291 {
292 	return pskb_may_pull(skb, skb_transport_offset(skb) +
293 				  sizeof(struct icmp6hdr));
294 }
295 
296 /**
297  * parse_vlan_tag - Parse vlan tag from vlan header.
298  * @skb: skb containing frame to parse
299  * @key_vh: pointer to parsed vlan tag
300  * @untag_vlan: should the vlan header be removed from the frame
301  *
302  * Return: ERROR on memory error.
303  * %0 if it encounters a non-vlan or incomplete packet.
304  * %1 after successfully parsing vlan tag.
305  */
306 static int parse_vlan_tag(struct sk_buff *skb, struct vlan_head *key_vh,
307 			  bool untag_vlan)
308 {
309 	struct vlan_head *vh = (struct vlan_head *)skb->data;
310 
311 	if (likely(!eth_type_vlan(vh->tpid)))
312 		return 0;
313 
314 	if (unlikely(skb->len < sizeof(struct vlan_head) + sizeof(__be16)))
315 		return 0;
316 
317 	if (unlikely(!pskb_may_pull(skb, sizeof(struct vlan_head) +
318 				 sizeof(__be16))))
319 		return -ENOMEM;
320 
321 	vh = (struct vlan_head *)skb->data;
322 	key_vh->tci = vh->tci | htons(VLAN_CFI_MASK);
323 	key_vh->tpid = vh->tpid;
324 
325 	if (unlikely(untag_vlan)) {
326 		int offset = skb->data - skb_mac_header(skb);
327 		u16 tci;
328 		int err;
329 
330 		__skb_push(skb, offset);
331 		err = __skb_vlan_pop(skb, &tci);
332 		__skb_pull(skb, offset);
333 		if (err)
334 			return err;
335 		__vlan_hwaccel_put_tag(skb, key_vh->tpid, tci);
336 	} else {
337 		__skb_pull(skb, sizeof(struct vlan_head));
338 	}
339 	return 1;
340 }
341 
342 static void clear_vlan(struct sw_flow_key *key)
343 {
344 	key->eth.vlan.tci = 0;
345 	key->eth.vlan.tpid = 0;
346 	key->eth.cvlan.tci = 0;
347 	key->eth.cvlan.tpid = 0;
348 }
349 
350 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
351 {
352 	int res;
353 
354 	if (skb_vlan_tag_present(skb)) {
355 		key->eth.vlan.tci = htons(skb->vlan_tci) | htons(VLAN_CFI_MASK);
356 		key->eth.vlan.tpid = skb->vlan_proto;
357 	} else {
358 		/* Parse outer vlan tag in the non-accelerated case. */
359 		res = parse_vlan_tag(skb, &key->eth.vlan, true);
360 		if (res <= 0)
361 			return res;
362 	}
363 
364 	/* Parse inner vlan tag. */
365 	res = parse_vlan_tag(skb, &key->eth.cvlan, false);
366 	if (res <= 0)
367 		return res;
368 
369 	return 0;
370 }
371 
372 static __be16 parse_ethertype(struct sk_buff *skb)
373 {
374 	struct llc_snap_hdr {
375 		u8  dsap;  /* Always 0xAA */
376 		u8  ssap;  /* Always 0xAA */
377 		u8  ctrl;
378 		u8  oui[3];
379 		__be16 ethertype;
380 	};
381 	struct llc_snap_hdr *llc;
382 	__be16 proto;
383 
384 	proto = *(__be16 *) skb->data;
385 	__skb_pull(skb, sizeof(__be16));
386 
387 	if (eth_proto_is_802_3(proto))
388 		return proto;
389 
390 	if (skb->len < sizeof(struct llc_snap_hdr))
391 		return htons(ETH_P_802_2);
392 
393 	if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
394 		return htons(0);
395 
396 	llc = (struct llc_snap_hdr *) skb->data;
397 	if (llc->dsap != LLC_SAP_SNAP ||
398 	    llc->ssap != LLC_SAP_SNAP ||
399 	    (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
400 		return htons(ETH_P_802_2);
401 
402 	__skb_pull(skb, sizeof(struct llc_snap_hdr));
403 
404 	if (eth_proto_is_802_3(llc->ethertype))
405 		return llc->ethertype;
406 
407 	return htons(ETH_P_802_2);
408 }
409 
410 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
411 			int nh_len)
412 {
413 	struct icmp6hdr *icmp = icmp6_hdr(skb);
414 
415 	/* The ICMPv6 type and code fields use the 16-bit transport port
416 	 * fields, so we need to store them in 16-bit network byte order.
417 	 */
418 	key->tp.src = htons(icmp->icmp6_type);
419 	key->tp.dst = htons(icmp->icmp6_code);
420 	memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
421 
422 	if (icmp->icmp6_code == 0 &&
423 	    (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
424 	     icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
425 		int icmp_len = skb->len - skb_transport_offset(skb);
426 		struct nd_msg *nd;
427 		int offset;
428 
429 		/* In order to process neighbor discovery options, we need the
430 		 * entire packet.
431 		 */
432 		if (unlikely(icmp_len < sizeof(*nd)))
433 			return 0;
434 
435 		if (unlikely(skb_linearize(skb)))
436 			return -ENOMEM;
437 
438 		nd = (struct nd_msg *)skb_transport_header(skb);
439 		key->ipv6.nd.target = nd->target;
440 
441 		icmp_len -= sizeof(*nd);
442 		offset = 0;
443 		while (icmp_len >= 8) {
444 			struct nd_opt_hdr *nd_opt =
445 				 (struct nd_opt_hdr *)(nd->opt + offset);
446 			int opt_len = nd_opt->nd_opt_len * 8;
447 
448 			if (unlikely(!opt_len || opt_len > icmp_len))
449 				return 0;
450 
451 			/* Store the link layer address if the appropriate
452 			 * option is provided.  It is considered an error if
453 			 * the same link layer option is specified twice.
454 			 */
455 			if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
456 			    && opt_len == 8) {
457 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
458 					goto invalid;
459 				ether_addr_copy(key->ipv6.nd.sll,
460 						&nd->opt[offset+sizeof(*nd_opt)]);
461 			} else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
462 				   && opt_len == 8) {
463 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
464 					goto invalid;
465 				ether_addr_copy(key->ipv6.nd.tll,
466 						&nd->opt[offset+sizeof(*nd_opt)]);
467 			}
468 
469 			icmp_len -= opt_len;
470 			offset += opt_len;
471 		}
472 	}
473 
474 	return 0;
475 
476 invalid:
477 	memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
478 	memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
479 	memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
480 
481 	return 0;
482 }
483 
484 static int parse_nsh(struct sk_buff *skb, struct sw_flow_key *key)
485 {
486 	struct nshhdr *nh;
487 	unsigned int nh_ofs = skb_network_offset(skb);
488 	u8 version, length;
489 	int err;
490 
491 	err = check_header(skb, nh_ofs + NSH_BASE_HDR_LEN);
492 	if (unlikely(err))
493 		return err;
494 
495 	nh = nsh_hdr(skb);
496 	version = nsh_get_ver(nh);
497 	length = nsh_hdr_len(nh);
498 
499 	if (version != 0)
500 		return -EINVAL;
501 
502 	err = check_header(skb, nh_ofs + length);
503 	if (unlikely(err))
504 		return err;
505 
506 	nh = nsh_hdr(skb);
507 	key->nsh.base.flags = nsh_get_flags(nh);
508 	key->nsh.base.ttl = nsh_get_ttl(nh);
509 	key->nsh.base.mdtype = nh->mdtype;
510 	key->nsh.base.np = nh->np;
511 	key->nsh.base.path_hdr = nh->path_hdr;
512 	switch (key->nsh.base.mdtype) {
513 	case NSH_M_TYPE1:
514 		if (length != NSH_M_TYPE1_LEN)
515 			return -EINVAL;
516 		memcpy(key->nsh.context, nh->md1.context,
517 		       sizeof(nh->md1));
518 		break;
519 	case NSH_M_TYPE2:
520 		memset(key->nsh.context, 0,
521 		       sizeof(nh->md1));
522 		break;
523 	default:
524 		return -EINVAL;
525 	}
526 
527 	return 0;
528 }
529 
530 /**
531  * key_extract_l3l4 - extracts L3/L4 header information.
532  * @skb: sk_buff that contains the frame, with skb->data pointing to the
533  *       L3 header
534  * @key: output flow key
535  *
536  * Return: %0 if successful, otherwise a negative errno value.
537  */
538 static int key_extract_l3l4(struct sk_buff *skb, struct sw_flow_key *key)
539 {
540 	int error;
541 
542 	/* Network layer. */
543 	if (key->eth.type == htons(ETH_P_IP)) {
544 		struct iphdr *nh;
545 		__be16 offset;
546 
547 		error = check_iphdr(skb);
548 		if (unlikely(error)) {
549 			memset(&key->ip, 0, sizeof(key->ip));
550 			memset(&key->ipv4, 0, sizeof(key->ipv4));
551 			if (error == -EINVAL) {
552 				skb->transport_header = skb->network_header;
553 				error = 0;
554 			}
555 			return error;
556 		}
557 
558 		nh = ip_hdr(skb);
559 		key->ipv4.addr.src = nh->saddr;
560 		key->ipv4.addr.dst = nh->daddr;
561 
562 		key->ip.proto = nh->protocol;
563 		key->ip.tos = nh->tos;
564 		key->ip.ttl = nh->ttl;
565 
566 		offset = nh->frag_off & htons(IP_OFFSET);
567 		if (offset) {
568 			key->ip.frag = OVS_FRAG_TYPE_LATER;
569 			memset(&key->tp, 0, sizeof(key->tp));
570 			return 0;
571 		}
572 		if (nh->frag_off & htons(IP_MF) ||
573 			skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
574 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
575 		else
576 			key->ip.frag = OVS_FRAG_TYPE_NONE;
577 
578 		/* Transport layer. */
579 		if (key->ip.proto == IPPROTO_TCP) {
580 			if (tcphdr_ok(skb)) {
581 				struct tcphdr *tcp = tcp_hdr(skb);
582 				key->tp.src = tcp->source;
583 				key->tp.dst = tcp->dest;
584 				key->tp.flags = TCP_FLAGS_BE16(tcp);
585 			} else {
586 				memset(&key->tp, 0, sizeof(key->tp));
587 			}
588 
589 		} else if (key->ip.proto == IPPROTO_UDP) {
590 			if (udphdr_ok(skb)) {
591 				struct udphdr *udp = udp_hdr(skb);
592 				key->tp.src = udp->source;
593 				key->tp.dst = udp->dest;
594 			} else {
595 				memset(&key->tp, 0, sizeof(key->tp));
596 			}
597 		} else if (key->ip.proto == IPPROTO_SCTP) {
598 			if (sctphdr_ok(skb)) {
599 				struct sctphdr *sctp = sctp_hdr(skb);
600 				key->tp.src = sctp->source;
601 				key->tp.dst = sctp->dest;
602 			} else {
603 				memset(&key->tp, 0, sizeof(key->tp));
604 			}
605 		} else if (key->ip.proto == IPPROTO_ICMP) {
606 			if (icmphdr_ok(skb)) {
607 				struct icmphdr *icmp = icmp_hdr(skb);
608 				/* The ICMP type and code fields use the 16-bit
609 				 * transport port fields, so we need to store
610 				 * them in 16-bit network byte order. */
611 				key->tp.src = htons(icmp->type);
612 				key->tp.dst = htons(icmp->code);
613 			} else {
614 				memset(&key->tp, 0, sizeof(key->tp));
615 			}
616 		}
617 
618 	} else if (key->eth.type == htons(ETH_P_ARP) ||
619 		   key->eth.type == htons(ETH_P_RARP)) {
620 		struct arp_eth_header *arp;
621 		bool arp_available = arphdr_ok(skb);
622 
623 		arp = (struct arp_eth_header *)skb_network_header(skb);
624 
625 		if (arp_available &&
626 		    arp->ar_hrd == htons(ARPHRD_ETHER) &&
627 		    arp->ar_pro == htons(ETH_P_IP) &&
628 		    arp->ar_hln == ETH_ALEN &&
629 		    arp->ar_pln == 4) {
630 
631 			/* We only match on the lower 8 bits of the opcode. */
632 			if (ntohs(arp->ar_op) <= 0xff)
633 				key->ip.proto = ntohs(arp->ar_op);
634 			else
635 				key->ip.proto = 0;
636 
637 			memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
638 			memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
639 			ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
640 			ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
641 		} else {
642 			memset(&key->ip, 0, sizeof(key->ip));
643 			memset(&key->ipv4, 0, sizeof(key->ipv4));
644 		}
645 	} else if (eth_p_mpls(key->eth.type)) {
646 		u8 label_count = 1;
647 
648 		memset(&key->mpls, 0, sizeof(key->mpls));
649 		skb_set_inner_network_header(skb, skb->mac_len);
650 		while (1) {
651 			__be32 lse;
652 
653 			error = check_header(skb, skb->mac_len +
654 					     label_count * MPLS_HLEN);
655 			if (unlikely(error))
656 				return 0;
657 
658 			memcpy(&lse, skb_inner_network_header(skb), MPLS_HLEN);
659 
660 			if (label_count <= MPLS_LABEL_DEPTH)
661 				memcpy(&key->mpls.lse[label_count - 1], &lse,
662 				       MPLS_HLEN);
663 
664 			skb_set_inner_network_header(skb, skb->mac_len +
665 						     label_count * MPLS_HLEN);
666 			if (lse & htonl(MPLS_LS_S_MASK))
667 				break;
668 
669 			label_count++;
670 		}
671 		if (label_count > MPLS_LABEL_DEPTH)
672 			label_count = MPLS_LABEL_DEPTH;
673 
674 		key->mpls.num_labels_mask = GENMASK(label_count - 1, 0);
675 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
676 		int nh_len;             /* IPv6 Header + Extensions */
677 
678 		nh_len = parse_ipv6hdr(skb, key);
679 		if (unlikely(nh_len < 0)) {
680 			switch (nh_len) {
681 			case -EINVAL:
682 				memset(&key->ip, 0, sizeof(key->ip));
683 				memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
684 				fallthrough;
685 			case -EPROTO:
686 				skb->transport_header = skb->network_header;
687 				error = 0;
688 				break;
689 			default:
690 				error = nh_len;
691 			}
692 			return error;
693 		}
694 
695 		if (key->ip.frag == OVS_FRAG_TYPE_LATER) {
696 			memset(&key->tp, 0, sizeof(key->tp));
697 			return 0;
698 		}
699 		if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
700 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
701 
702 		/* Transport layer. */
703 		if (key->ip.proto == NEXTHDR_TCP) {
704 			if (tcphdr_ok(skb)) {
705 				struct tcphdr *tcp = tcp_hdr(skb);
706 				key->tp.src = tcp->source;
707 				key->tp.dst = tcp->dest;
708 				key->tp.flags = TCP_FLAGS_BE16(tcp);
709 			} else {
710 				memset(&key->tp, 0, sizeof(key->tp));
711 			}
712 		} else if (key->ip.proto == NEXTHDR_UDP) {
713 			if (udphdr_ok(skb)) {
714 				struct udphdr *udp = udp_hdr(skb);
715 				key->tp.src = udp->source;
716 				key->tp.dst = udp->dest;
717 			} else {
718 				memset(&key->tp, 0, sizeof(key->tp));
719 			}
720 		} else if (key->ip.proto == NEXTHDR_SCTP) {
721 			if (sctphdr_ok(skb)) {
722 				struct sctphdr *sctp = sctp_hdr(skb);
723 				key->tp.src = sctp->source;
724 				key->tp.dst = sctp->dest;
725 			} else {
726 				memset(&key->tp, 0, sizeof(key->tp));
727 			}
728 		} else if (key->ip.proto == NEXTHDR_ICMP) {
729 			if (icmp6hdr_ok(skb)) {
730 				error = parse_icmpv6(skb, key, nh_len);
731 				if (error)
732 					return error;
733 			} else {
734 				memset(&key->tp, 0, sizeof(key->tp));
735 			}
736 		}
737 	} else if (key->eth.type == htons(ETH_P_NSH)) {
738 		error = parse_nsh(skb, key);
739 		if (error)
740 			return error;
741 	}
742 	return 0;
743 }
744 
745 /**
746  * key_extract - extracts a flow key from an Ethernet frame.
747  * @skb: sk_buff that contains the frame, with skb->data pointing to the
748  * Ethernet header
749  * @key: output flow key
750  *
751  * The caller must ensure that skb->len >= ETH_HLEN.
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  * Return: %0 if successful, otherwise a negative errno value.
769  */
770 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
771 {
772 	struct ethhdr *eth;
773 
774 	/* Flags are always used as part of stats */
775 	key->tp.flags = 0;
776 
777 	skb_reset_mac_header(skb);
778 
779 	/* Link layer. */
780 	clear_vlan(key);
781 	if (ovs_key_mac_proto(key) == MAC_PROTO_NONE) {
782 		if (unlikely(eth_type_vlan(skb->protocol)))
783 			return -EINVAL;
784 
785 		skb_reset_network_header(skb);
786 		key->eth.type = skb->protocol;
787 	} else {
788 		eth = eth_hdr(skb);
789 		ether_addr_copy(key->eth.src, eth->h_source);
790 		ether_addr_copy(key->eth.dst, eth->h_dest);
791 
792 		__skb_pull(skb, 2 * ETH_ALEN);
793 		/* We are going to push all headers that we pull, so no need to
794 		 * update skb->csum here.
795 		 */
796 
797 		if (unlikely(parse_vlan(skb, key)))
798 			return -ENOMEM;
799 
800 		key->eth.type = parse_ethertype(skb);
801 		if (unlikely(key->eth.type == htons(0)))
802 			return -ENOMEM;
803 
804 		/* Multiple tagged packets need to retain TPID to satisfy
805 		 * skb_vlan_pop(), which will later shift the ethertype into
806 		 * skb->protocol.
807 		 */
808 		if (key->eth.cvlan.tci & htons(VLAN_CFI_MASK))
809 			skb->protocol = key->eth.cvlan.tpid;
810 		else
811 			skb->protocol = key->eth.type;
812 
813 		skb_reset_network_header(skb);
814 		__skb_push(skb, skb->data - skb_mac_header(skb));
815 	}
816 
817 	skb_reset_mac_len(skb);
818 
819 	/* Fill out L3/L4 key info, if any */
820 	return key_extract_l3l4(skb, key);
821 }
822 
823 /* In the case of conntrack fragment handling it expects L3 headers,
824  * add a helper.
825  */
826 int ovs_flow_key_update_l3l4(struct sk_buff *skb, struct sw_flow_key *key)
827 {
828 	return key_extract_l3l4(skb, key);
829 }
830 
831 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
832 {
833 	int res;
834 
835 	res = key_extract(skb, key);
836 	if (!res)
837 		key->mac_proto &= ~SW_FLOW_KEY_INVALID;
838 
839 	return res;
840 }
841 
842 static int key_extract_mac_proto(struct sk_buff *skb)
843 {
844 	switch (skb->dev->type) {
845 	case ARPHRD_ETHER:
846 		return MAC_PROTO_ETHERNET;
847 	case ARPHRD_NONE:
848 		if (skb->protocol == htons(ETH_P_TEB))
849 			return MAC_PROTO_ETHERNET;
850 		return MAC_PROTO_NONE;
851 	}
852 	WARN_ON_ONCE(1);
853 	return -EINVAL;
854 }
855 
856 int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
857 			 struct sk_buff *skb, struct sw_flow_key *key)
858 {
859 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
860 	struct tc_skb_ext *tc_ext;
861 #endif
862 	bool post_ct = false, post_ct_snat = false, post_ct_dnat = false;
863 	int res, err;
864 	u16 zone = 0;
865 
866 	/* Extract metadata from packet. */
867 	if (tun_info) {
868 		key->tun_proto = ip_tunnel_info_af(tun_info);
869 		memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
870 
871 		if (tun_info->options_len) {
872 			BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) *
873 						   8)) - 1
874 					> sizeof(key->tun_opts));
875 
876 			ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key, tun_info->options_len),
877 						tun_info);
878 			key->tun_opts_len = tun_info->options_len;
879 		} else {
880 			key->tun_opts_len = 0;
881 		}
882 	} else  {
883 		key->tun_proto = 0;
884 		key->tun_opts_len = 0;
885 		memset(&key->tun_key, 0, sizeof(key->tun_key));
886 	}
887 
888 	key->phy.priority = skb->priority;
889 	key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
890 	key->phy.skb_mark = skb->mark;
891 	key->ovs_flow_hash = 0;
892 	res = key_extract_mac_proto(skb);
893 	if (res < 0)
894 		return res;
895 	key->mac_proto = res;
896 
897 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
898 	if (static_branch_unlikely(&tc_recirc_sharing_support)) {
899 		tc_ext = skb_ext_find(skb, TC_SKB_EXT);
900 		key->recirc_id = tc_ext ? tc_ext->chain : 0;
901 		OVS_CB(skb)->mru = tc_ext ? tc_ext->mru : 0;
902 		post_ct = tc_ext ? tc_ext->post_ct : false;
903 		post_ct_snat = post_ct ? tc_ext->post_ct_snat : false;
904 		post_ct_dnat = post_ct ? tc_ext->post_ct_dnat : false;
905 		zone = post_ct ? tc_ext->zone : 0;
906 	} else {
907 		key->recirc_id = 0;
908 	}
909 #else
910 	key->recirc_id = 0;
911 #endif
912 
913 	err = key_extract(skb, key);
914 	if (!err) {
915 		ovs_ct_fill_key(skb, key, post_ct);   /* Must be after key_extract(). */
916 		if (post_ct) {
917 			if (!skb_get_nfct(skb)) {
918 				key->ct_zone = zone;
919 			} else {
920 				if (!post_ct_dnat)
921 					key->ct_state &= ~OVS_CS_F_DST_NAT;
922 				if (!post_ct_snat)
923 					key->ct_state &= ~OVS_CS_F_SRC_NAT;
924 			}
925 		}
926 	}
927 	return err;
928 }
929 
930 int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
931 				   struct sk_buff *skb,
932 				   struct sw_flow_key *key, bool log)
933 {
934 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
935 	u64 attrs = 0;
936 	int err;
937 
938 	err = parse_flow_nlattrs(attr, a, &attrs, log);
939 	if (err)
940 		return -EINVAL;
941 
942 	/* Extract metadata from netlink attributes. */
943 	err = ovs_nla_get_flow_metadata(net, a, attrs, key, log);
944 	if (err)
945 		return err;
946 
947 	/* key_extract assumes that skb->protocol is set-up for
948 	 * layer 3 packets which is the case for other callers,
949 	 * in particular packets received from the network stack.
950 	 * Here the correct value can be set from the metadata
951 	 * extracted above.
952 	 * For L2 packet key eth type would be zero. skb protocol
953 	 * would be set to correct value later during key-extact.
954 	 */
955 
956 	skb->protocol = key->eth.type;
957 	err = key_extract(skb, key);
958 	if (err)
959 		return err;
960 
961 	/* Check that we have conntrack original direction tuple metadata only
962 	 * for packets for which it makes sense.  Otherwise the key may be
963 	 * corrupted due to overlapping key fields.
964 	 */
965 	if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4) &&
966 	    key->eth.type != htons(ETH_P_IP))
967 		return -EINVAL;
968 	if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6) &&
969 	    (key->eth.type != htons(ETH_P_IPV6) ||
970 	     sw_flow_key_is_nd(key)))
971 		return -EINVAL;
972 
973 	return 0;
974 }
975