]> git.proxmox.com Git - mirror_ovs.git/blob - datapath/flow.c
projects / mirror_ovs.git / blob
? search:


dc935814ed049a10ed635d23edc041e317c4e268
[mirror_ovs.git] / datapath / flow.c
1 /*
2 * Copyright (c) 2007-2015 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 <linux/timekeeping.h>
45 #include <net/ip.h>
46 #include <net/ipv6.h>
47 #include <net/mpls.h>
48 #include <net/ndisc.h>
49 #include <net/nsh.h>
50
51 #include "datapath.h"
52 #include "conntrack.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 key->ip.proto = nexthdr;
282 return 0;
283 }
284 key->ip.frag = OVS_FRAG_TYPE_FIRST;
285 } else {
286 key->ip.frag = OVS_FRAG_TYPE_NONE;
287 }
288
289 /* Delayed handling of error in ipv6_find_hdr() as it
290 * always sets flags and frag_off to a valid value which may be
291 * used to set key->ip.frag above.
292 */
293 if (unlikely(nexthdr < 0))
294 return -EPROTO;
295
296 nh_len = payload_ofs - nh_ofs;
297 skb_set_transport_header(skb, nh_ofs + nh_len);
298 key->ip.proto = nexthdr;
299 return nh_len;
300 }
301
302 static bool icmp6hdr_ok(struct sk_buff *skb)
303 {
304 return pskb_may_pull(skb, skb_transport_offset(skb) +
305 sizeof(struct icmp6hdr));
306 }
307
308 /**
309 * Parse vlan tag from vlan header.
310 * Returns ERROR on memory error.
311 * Returns 0 if it encounters a non-vlan or incomplete packet.
312 * Returns 1 after successfully parsing vlan tag.
313 */
314 static int parse_vlan_tag(struct sk_buff *skb, struct vlan_head *key_vh,
315 bool untag_vlan)
316 {
317 struct vlan_head *vh = (struct vlan_head *)skb->data;
318
319 if (likely(!eth_type_vlan(vh->tpid)))
320 return 0;
321
322 if (unlikely(skb->len < sizeof(struct vlan_head) + sizeof(__be16)))
323 return 0;
324
325 if (unlikely(!pskb_may_pull(skb, sizeof(struct vlan_head) +
326 sizeof(__be16))))
327 return -ENOMEM;
328
329 vh = (struct vlan_head *)skb->data;
330 key_vh->tci = vh->tci | htons(VLAN_CFI_MASK);
331 key_vh->tpid = vh->tpid;
332
333 if (unlikely(untag_vlan)) {
334 int offset = skb->data - skb_mac_header(skb);
335 u16 tci;
336 int err;
337
338 __skb_push(skb, offset);
339 err = __skb_vlan_pop(skb, &tci);
340 __skb_pull(skb, offset);
341 if (err)
342 return err;
343 __vlan_hwaccel_put_tag(skb, key_vh->tpid, tci);
344 } else {
345 __skb_pull(skb, sizeof(struct vlan_head));
346 }
347 return 1;
348 }
349
350 static void clear_vlan(struct sw_flow_key *key)
351 {
352 key->eth.vlan.tci = 0;
353 key->eth.vlan.tpid = 0;
354 key->eth.cvlan.tci = 0;
355 key->eth.cvlan.tpid = 0;
356 }
357
358 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
359 {
360 int res;
361
362 key->eth.vlan.tci = 0;
363 key->eth.vlan.tpid = 0;
364 key->eth.cvlan.tci = 0;
365 key->eth.cvlan.tpid = 0;
366
367 if (skb_vlan_tag_present(skb)) {
368 key->eth.vlan.tci = htons(skb->vlan_tci) | htons(VLAN_CFI_MASK);
369 key->eth.vlan.tpid = skb->vlan_proto;
370 } else {
371 /* Parse outer vlan tag in the non-accelerated case. */
372 res = parse_vlan_tag(skb, &key->eth.vlan, true);
373 if (res <= 0)
374 return res;
375 }
376
377 /* Parse inner vlan tag. */
378 res = parse_vlan_tag(skb, &key->eth.cvlan, false);
379 if (res <= 0)
380 return res;
381
382 return 0;
383 }
384
385 static __be16 parse_ethertype(struct sk_buff *skb)
386 {
387 struct llc_snap_hdr {
388 u8 dsap; /* Always 0xAA */
389 u8 ssap; /* Always 0xAA */
390 u8 ctrl;
391 u8 oui[3];
392 __be16 ethertype;
393 };
394 struct llc_snap_hdr *llc;
395 __be16 proto;
396
397 proto = *(__be16 *) skb->data;
398 __skb_pull(skb, sizeof(__be16));
399
400 if (eth_proto_is_802_3(proto))
401 return proto;
402
403 if (skb->len < sizeof(struct llc_snap_hdr))
404 return htons(ETH_P_802_2);
405
406 if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
407 return htons(0);
408
409 llc = (struct llc_snap_hdr *) skb->data;
410 if (llc->dsap != LLC_SAP_SNAP ||
411 llc->ssap != LLC_SAP_SNAP ||
412 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
413 return htons(ETH_P_802_2);
414
415 __skb_pull(skb, sizeof(struct llc_snap_hdr));
416
417 if (eth_proto_is_802_3(llc->ethertype))
418 return llc->ethertype;
419
420 return htons(ETH_P_802_2);
421 }
422
423 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
424 int nh_len)
425 {
426 struct icmp6hdr *icmp = icmp6_hdr(skb);
427
428 /* The ICMPv6 type and code fields use the 16-bit transport port
429 * fields, so we need to store them in 16-bit network byte order.
430 */
431 key->tp.src = htons(icmp->icmp6_type);
432 key->tp.dst = htons(icmp->icmp6_code);
433 memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
434
435 if (icmp->icmp6_code == 0 &&
436 (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
437 icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
438 int icmp_len = skb->len - skb_transport_offset(skb);
439 struct nd_msg *nd;
440 int offset;
441
442 /* In order to process neighbor discovery options, we need the
443 * entire packet.
444 */
445 if (unlikely(icmp_len < sizeof(*nd)))
446 return 0;
447
448 if (unlikely(skb_linearize(skb)))
449 return -ENOMEM;
450
451 nd = (struct nd_msg *)skb_transport_header(skb);
452 key->ipv6.nd.target = nd->target;
453
454 icmp_len -= sizeof(*nd);
455 offset = 0;
456 while (icmp_len >= 8) {
457 struct nd_opt_hdr *nd_opt =
458 (struct nd_opt_hdr *)(nd->opt + offset);
459 int opt_len = nd_opt->nd_opt_len * 8;
460
461 if (unlikely(!opt_len || opt_len > icmp_len))
462 return 0;
463
464 /* Store the link layer address if the appropriate
465 * option is provided. It is considered an error if
466 * the same link layer option is specified twice.
467 */
468 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
469 && opt_len == 8) {
470 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
471 goto invalid;
472 ether_addr_copy(key->ipv6.nd.sll,
473 &nd->opt[offset+sizeof(*nd_opt)]);
474 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
475 && opt_len == 8) {
476 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
477 goto invalid;
478 ether_addr_copy(key->ipv6.nd.tll,
479 &nd->opt[offset+sizeof(*nd_opt)]);
480 }
481
482 icmp_len -= opt_len;
483 offset += opt_len;
484 }
485 }
486
487 return 0;
488
489 invalid:
490 memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
491 memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
492 memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
493
494 return 0;
495 }
496
497 static int parse_nsh(struct sk_buff *skb, struct sw_flow_key *key)
498 {
499 struct nshhdr *nh;
500 unsigned int nh_ofs = skb_network_offset(skb);
501 u8 version, length;
502 int err;
503
504 err = check_header(skb, nh_ofs + NSH_BASE_HDR_LEN);
505 if (unlikely(err))
506 return err;
507
508 nh = nsh_hdr(skb);
509 version = nsh_get_ver(nh);
510 length = nsh_hdr_len(nh);
511
512 if (version != 0)
513 return -EINVAL;
514
515 err = check_header(skb, nh_ofs + length);
516 if (unlikely(err))
517 return err;
518
519 nh = nsh_hdr(skb);
520 key->nsh.base.flags = nsh_get_flags(nh);
521 key->nsh.base.ttl = nsh_get_ttl(nh);
522 key->nsh.base.mdtype = nh->mdtype;
523 key->nsh.base.np = nh->np;
524 key->nsh.base.path_hdr = nh->path_hdr;
525 switch (key->nsh.base.mdtype) {
526 case NSH_M_TYPE1:
527 if (length != NSH_M_TYPE1_LEN)
528 return -EINVAL;
529 memcpy(key->nsh.context, nh->md1.context,
530 sizeof(nh->md1));
531 break;
532 case NSH_M_TYPE2:
533 memset(key->nsh.context, 0,
534 sizeof(nh->md1));
535 break;
536 default:
537 return -EINVAL;
538 }
539
540 return 0;
541 }
542
543 /**
544 * key_extract_l3l4 - extracts L3/L4 header information.
545 * @skb: sk_buff that contains the frame, with skb->data pointing to the
546 * L3 header
547 * @key: output flow key
548 */
549 static int key_extract_l3l4(struct sk_buff *skb, struct sw_flow_key *key)
550 {
551 int error;
552
553 /* Network layer. */
554 if (key->eth.type == htons(ETH_P_IP)) {
555 struct iphdr *nh;
556 __be16 offset;
557
558 error = check_iphdr(skb);
559 if (unlikely(error)) {
560 memset(&key->ip, 0, sizeof(key->ip));
561 memset(&key->ipv4, 0, sizeof(key->ipv4));
562 if (error == -EINVAL) {
563 skb->transport_header = skb->network_header;
564 error = 0;
565 }
566 return error;
567 }
568
569 nh = ip_hdr(skb);
570 key->ipv4.addr.src = nh->saddr;
571 key->ipv4.addr.dst = nh->daddr;
572
573 key->ip.proto = nh->protocol;
574 key->ip.tos = nh->tos;
575 key->ip.ttl = nh->ttl;
576
577 offset = nh->frag_off & htons(IP_OFFSET);
578 if (offset) {
579 key->ip.frag = OVS_FRAG_TYPE_LATER;
580 return 0;
581 }
582 #ifdef HAVE_SKB_GSO_UDP
583 if (nh->frag_off & htons(IP_MF) ||
584 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
585 #else
586 if (nh->frag_off & htons(IP_MF))
587 #endif
588 key->ip.frag = OVS_FRAG_TYPE_FIRST;
589 else
590 key->ip.frag = OVS_FRAG_TYPE_NONE;
591
592 /* Transport layer. */
593 if (key->ip.proto == IPPROTO_TCP) {
594 if (tcphdr_ok(skb)) {
595 struct tcphdr *tcp = tcp_hdr(skb);
596 key->tp.src = tcp->source;
597 key->tp.dst = tcp->dest;
598 key->tp.flags = TCP_FLAGS_BE16(tcp);
599 } else {
600 memset(&key->tp, 0, sizeof(key->tp));
601 }
602
603 } else if (key->ip.proto == IPPROTO_UDP) {
604 if (udphdr_ok(skb)) {
605 struct udphdr *udp = udp_hdr(skb);
606 key->tp.src = udp->source;
607 key->tp.dst = udp->dest;
608 } else {
609 memset(&key->tp, 0, sizeof(key->tp));
610 }
611 } else if (key->ip.proto == IPPROTO_SCTP) {
612 if (sctphdr_ok(skb)) {
613 struct sctphdr *sctp = sctp_hdr(skb);
614 key->tp.src = sctp->source;
615 key->tp.dst = sctp->dest;
616 } else {
617 memset(&key->tp, 0, sizeof(key->tp));
618 }
619 } else if (key->ip.proto == IPPROTO_ICMP) {
620 if (icmphdr_ok(skb)) {
621 struct icmphdr *icmp = icmp_hdr(skb);
622 /* The ICMP type and code fields use the 16-bit
623 * transport port fields, so we need to store
624 * them in 16-bit network byte order.
625 */
626 key->tp.src = htons(icmp->type);
627 key->tp.dst = htons(icmp->code);
628 } else {
629 memset(&key->tp, 0, sizeof(key->tp));
630 }
631 }
632
633 } else if (key->eth.type == htons(ETH_P_ARP) ||
634 key->eth.type == htons(ETH_P_RARP)) {
635 struct arp_eth_header *arp;
636 bool arp_available = arphdr_ok(skb);
637
638 arp = (struct arp_eth_header *)skb_network_header(skb);
639
640 if (arp_available &&
641 arp->ar_hrd == htons(ARPHRD_ETHER) &&
642 arp->ar_pro == htons(ETH_P_IP) &&
643 arp->ar_hln == ETH_ALEN &&
644 arp->ar_pln == 4) {
645
646 /* We only match on the lower 8 bits of the opcode. */
647 if (ntohs(arp->ar_op) <= 0xff)
648 key->ip.proto = ntohs(arp->ar_op);
649 else
650 key->ip.proto = 0;
651
652 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
653 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
654 ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
655 ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
656 } else {
657 memset(&key->ip, 0, sizeof(key->ip));
658 memset(&key->ipv4, 0, sizeof(key->ipv4));
659 }
660 } else if (eth_p_mpls(key->eth.type)) {
661 size_t stack_len = MPLS_HLEN;
662
663 skb_set_inner_network_header(skb, skb->mac_len);
664 while (1) {
665 __be32 lse;
666
667 error = check_header(skb, skb->mac_len + stack_len);
668 if (unlikely(error))
669 return 0;
670
671 memcpy(&lse, skb_inner_network_header(skb), MPLS_HLEN);
672
673 if (stack_len == MPLS_HLEN)
674 memcpy(&key->mpls.top_lse, &lse, MPLS_HLEN);
675
676 skb_set_inner_network_header(skb, skb->mac_len + stack_len);
677 if (lse & htonl(MPLS_LS_S_MASK))
678 break;
679
680 stack_len += MPLS_HLEN;
681 }
682 } else if (key->eth.type == htons(ETH_P_IPV6)) {
683 int nh_len; /* IPv6 Header + Extensions */
684
685 nh_len = parse_ipv6hdr(skb, key);
686 if (unlikely(nh_len < 0)) {
687 switch (nh_len) {
688 case -EINVAL:
689 memset(&key->ip, 0, sizeof(key->ip));
690 memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
691 /* fall-through */
692 case -EPROTO:
693 skb->transport_header = skb->network_header;
694 error = 0;
695 break;
696 default:
697 error = nh_len;
698 }
699 return error;
700 }
701
702 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
703 return 0;
704 #ifdef HAVE_SKB_GSO_UDP
705 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
706 key->ip.frag = OVS_FRAG_TYPE_FIRST;
707
708 #endif
709 /* Transport layer. */
710 if (key->ip.proto == NEXTHDR_TCP) {
711 if (tcphdr_ok(skb)) {
712 struct tcphdr *tcp = tcp_hdr(skb);
713 key->tp.src = tcp->source;
714 key->tp.dst = tcp->dest;
715 key->tp.flags = TCP_FLAGS_BE16(tcp);
716 } else {
717 memset(&key->tp, 0, sizeof(key->tp));
718 }
719 } else if (key->ip.proto == NEXTHDR_UDP) {
720 if (udphdr_ok(skb)) {
721 struct udphdr *udp = udp_hdr(skb);
722 key->tp.src = udp->source;
723 key->tp.dst = udp->dest;
724 } else {
725 memset(&key->tp, 0, sizeof(key->tp));
726 }
727 } else if (key->ip.proto == NEXTHDR_SCTP) {
728 if (sctphdr_ok(skb)) {
729 struct sctphdr *sctp = sctp_hdr(skb);
730 key->tp.src = sctp->source;
731 key->tp.dst = sctp->dest;
732 } else {
733 memset(&key->tp, 0, sizeof(key->tp));
734 }
735 } else if (key->ip.proto == NEXTHDR_ICMP) {
736 if (icmp6hdr_ok(skb)) {
737 error = parse_icmpv6(skb, key, nh_len);
738 if (error)
739 return error;
740 } else {
741 memset(&key->tp, 0, sizeof(key->tp));
742 }
743 }
744 } else if (key->eth.type == htons(ETH_P_NSH)) {
745 error = parse_nsh(skb, key);
746 if (error)
747 return error;
748 }
749 return 0;
750 }
751
752 /**
753 * key_extract - extracts a flow key from an Ethernet frame.
754 * @skb: sk_buff that contains the frame, with skb->data pointing to the
755 * Ethernet header
756 * @key: output flow key
757 *
758 * The caller must ensure that skb->len >= ETH_HLEN.
759 *
760 * Returns 0 if successful, otherwise a negative errno value.
761 *
762 * Initializes @skb header fields as follows:
763 *
764 * - skb->mac_header: the L2 header.
765 *
766 * - skb->network_header: just past the L2 header, or just past the
767 * VLAN header, to the first byte of the L2 payload.
768 *
769 * - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
770 * on output, then just past the IP header, if one is present and
771 * of a correct length, otherwise the same as skb->network_header.
772 * For other key->eth.type values it is left untouched.
773 *
774 * - skb->protocol: the type of the data starting at skb->network_header.
775 * Equals to key->eth.type.
776 */
777 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
778 {
779 struct ethhdr *eth;
780
781 /* Flags are always used as part of stats */
782 key->tp.flags = 0;
783
784 skb_reset_mac_header(skb);
785
786 /* Link layer. */
787 clear_vlan(key);
788 if (ovs_key_mac_proto(key) == MAC_PROTO_NONE) {
789 if (unlikely(eth_type_vlan(skb->protocol)))
790 return -EINVAL;
791
792 skb_reset_network_header(skb);
793 key->eth.type = skb->protocol;
794 } else {
795 eth = eth_hdr(skb);
796 ether_addr_copy(key->eth.src, eth->h_source);
797 ether_addr_copy(key->eth.dst, eth->h_dest);
798
799 __skb_pull(skb, 2 * ETH_ALEN);
800 /* We are going to push all headers that we pull, so no need to
801 * update skb->csum here.
802 */
803
804 if (unlikely(parse_vlan(skb, key)))
805 return -ENOMEM;
806
807 key->eth.type = parse_ethertype(skb);
808 if (unlikely(key->eth.type == htons(0)))
809 return -ENOMEM;
810
811 /* Multiple tagged packets need to retain TPID to satisfy
812 * skb_vlan_pop(), which will later shift the ethertype into
813 * skb->protocol.
814 */
815 if (key->eth.cvlan.tci & htons(VLAN_CFI_MASK))
816 skb->protocol = key->eth.cvlan.tpid;
817 else
818 skb->protocol = key->eth.type;
819
820 skb_reset_network_header(skb);
821 __skb_push(skb, skb->data - skb_mac_header(skb));
822 }
823
824 skb_reset_mac_len(skb);
825
826 /* Fill out L3/L4 key info, if any */
827 return key_extract_l3l4(skb, key);
828 }
829
830 /* In the case of conntrack fragment handling it expects L3 headers,
831 * add a helper.
832 */
833 int ovs_flow_key_update_l3l4(struct sk_buff *skb, struct sw_flow_key *key)
834 {
835 return key_extract_l3l4(skb, key);
836 }
837
838 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
839 {
840 int res;
841
842 res = key_extract(skb, key);
843 if (!res)
844 key->mac_proto &= ~SW_FLOW_KEY_INVALID;
845
846 return res;
847 }
848
849 static int key_extract_mac_proto(struct sk_buff *skb)
850 {
851 switch (skb->dev->type) {
852 case ARPHRD_ETHER:
853 return MAC_PROTO_ETHERNET;
854 case ARPHRD_NONE:
855 if (skb->protocol == htons(ETH_P_TEB))
856 return MAC_PROTO_ETHERNET;
857 return MAC_PROTO_NONE;
858 }
859 WARN_ON_ONCE(1);
860 return -EINVAL;
861 }
862
863 int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
864 struct sk_buff *skb, struct sw_flow_key *key)
865 {
866 int res, err;
867
868 /* Extract metadata from packet. */
869 if (tun_info) {
870 key->tun_proto = ip_tunnel_info_af(tun_info);
871 memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
872 BUILD_BUG_ON(((1 << (sizeof(tun_info->options_len) * 8)) - 1) >
873 sizeof(key->tun_opts));
874
875 if (tun_info->options_len) {
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 key->recirc_id = 0;
897
898 err = key_extract(skb, key);
899 if (!err)
900 ovs_ct_fill_key(skb, key); /* Must be after key_extract(). */
901 return err;
902 }
903
904 int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
905 struct sk_buff *skb,
906 struct sw_flow_key *key, bool log)
907 {
908 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
909 u64 attrs = 0;
910 int err;
911
912 err = parse_flow_nlattrs(attr, a, &attrs, log);
913 if (err)
914 return -EINVAL;
915
916 /* Extract metadata from netlink attributes. */
917 err = ovs_nla_get_flow_metadata(net, a, attrs, key, log);
918 if (err)
919 return err;
920
921 /* key_extract assumes that skb->protocol is set-up for
922 * layer 3 packets which is the case for other callers,
923 * in particular packets received from the network stack.
924 * Here the correct value can be set from the metadata
925 * extracted above.
926 * For L2 packet key eth type would be zero. skb protocol
927 * would be set to correct value later during key-extact.
928 */
929
930 skb->protocol = key->eth.type;
931 err = key_extract(skb, key);
932 if (err)
933 return err;
934
935 /* Check that we have conntrack original direction tuple metadata only
936 * for packets for which it makes sense. Otherwise the key may be
937 * corrupted due to overlapping key fields.
938 */
939 if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4) &&
940 key->eth.type != htons(ETH_P_IP))
941 return -EINVAL;
942 if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6) &&
943 (key->eth.type != htons(ETH_P_IPV6) ||
944 sw_flow_key_is_nd(key)))
945 return -EINVAL;
946
947 return 0;
948 }
openvswitch mirror