Loading...
1/*
2 * Copyright (c) 2007-2014 Nicira, Inc.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16 * 02110-1301, USA
17 */
18
19#include <linux/uaccess.h>
20#include <linux/netdevice.h>
21#include <linux/etherdevice.h>
22#include <linux/if_ether.h>
23#include <linux/if_vlan.h>
24#include <net/llc_pdu.h>
25#include <linux/kernel.h>
26#include <linux/jhash.h>
27#include <linux/jiffies.h>
28#include <linux/llc.h>
29#include <linux/module.h>
30#include <linux/in.h>
31#include <linux/rcupdate.h>
32#include <linux/cpumask.h>
33#include <linux/if_arp.h>
34#include <linux/ip.h>
35#include <linux/ipv6.h>
36#include <linux/mpls.h>
37#include <linux/sctp.h>
38#include <linux/smp.h>
39#include <linux/tcp.h>
40#include <linux/udp.h>
41#include <linux/icmp.h>
42#include <linux/icmpv6.h>
43#include <linux/rculist.h>
44#include <net/ip.h>
45#include <net/ip_tunnels.h>
46#include <net/ipv6.h>
47#include <net/mpls.h>
48#include <net/ndisc.h>
49
50#include "conntrack.h"
51#include "datapath.h"
52#include "flow.h"
53#include "flow_netlink.h"
54#include "vport.h"
55
56u64 ovs_flow_used_time(unsigned long flow_jiffies)
57{
58 struct timespec cur_ts;
59 u64 cur_ms, idle_ms;
60
61 ktime_get_ts(&cur_ts);
62 idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
63 cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
64 cur_ts.tv_nsec / NSEC_PER_MSEC;
65
66 return cur_ms - idle_ms;
67}
68
69#define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
70
71void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
72 const struct sk_buff *skb)
73{
74 struct flow_stats *stats;
75 int node = numa_node_id();
76 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 node);
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 goto unlock;
122 }
123 }
124 flow->stats_last_writer = cpu;
125 }
126 }
127
128 stats->used = jiffies;
129 stats->packet_count++;
130 stats->byte_count += len;
131 stats->tcp_flags |= tcp_flags;
132unlock:
133 spin_unlock(&stats->lock);
134}
135
136/* Must be called with rcu_read_lock or ovs_mutex. */
137void ovs_flow_stats_get(const struct sw_flow *flow,
138 struct ovs_flow_stats *ovs_stats,
139 unsigned long *used, __be16 *tcp_flags)
140{
141 int cpu;
142
143 *used = 0;
144 *tcp_flags = 0;
145 memset(ovs_stats, 0, sizeof(*ovs_stats));
146
147 /* We open code this to make sure cpu 0 is always considered */
148 for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, cpu_possible_mask)) {
149 struct flow_stats *stats = rcu_dereference_ovsl(flow->stats[cpu]);
150
151 if (stats) {
152 /* Local CPU may write on non-local stats, so we must
153 * block bottom-halves here.
154 */
155 spin_lock_bh(&stats->lock);
156 if (!*used || time_after(stats->used, *used))
157 *used = stats->used;
158 *tcp_flags |= stats->tcp_flags;
159 ovs_stats->n_packets += stats->packet_count;
160 ovs_stats->n_bytes += stats->byte_count;
161 spin_unlock_bh(&stats->lock);
162 }
163 }
164}
165
166/* Called with ovs_mutex. */
167void ovs_flow_stats_clear(struct sw_flow *flow)
168{
169 int cpu;
170
171 /* We open code this to make sure cpu 0 is always considered */
172 for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, cpu_possible_mask)) {
173 struct flow_stats *stats = ovsl_dereference(flow->stats[cpu]);
174
175 if (stats) {
176 spin_lock_bh(&stats->lock);
177 stats->used = 0;
178 stats->packet_count = 0;
179 stats->byte_count = 0;
180 stats->tcp_flags = 0;
181 spin_unlock_bh(&stats->lock);
182 }
183 }
184}
185
186static int check_header(struct sk_buff *skb, int len)
187{
188 if (unlikely(skb->len < len))
189 return -EINVAL;
190 if (unlikely(!pskb_may_pull(skb, len)))
191 return -ENOMEM;
192 return 0;
193}
194
195static bool arphdr_ok(struct sk_buff *skb)
196{
197 return pskb_may_pull(skb, skb_network_offset(skb) +
198 sizeof(struct arp_eth_header));
199}
200
201static int check_iphdr(struct sk_buff *skb)
202{
203 unsigned int nh_ofs = skb_network_offset(skb);
204 unsigned int ip_len;
205 int err;
206
207 err = check_header(skb, nh_ofs + sizeof(struct iphdr));
208 if (unlikely(err))
209 return err;
210
211 ip_len = ip_hdrlen(skb);
212 if (unlikely(ip_len < sizeof(struct iphdr) ||
213 skb->len < nh_ofs + ip_len))
214 return -EINVAL;
215
216 skb_set_transport_header(skb, nh_ofs + ip_len);
217 return 0;
218}
219
220static bool tcphdr_ok(struct sk_buff *skb)
221{
222 int th_ofs = skb_transport_offset(skb);
223 int tcp_len;
224
225 if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
226 return false;
227
228 tcp_len = tcp_hdrlen(skb);
229 if (unlikely(tcp_len < sizeof(struct tcphdr) ||
230 skb->len < th_ofs + tcp_len))
231 return false;
232
233 return true;
234}
235
236static bool udphdr_ok(struct sk_buff *skb)
237{
238 return pskb_may_pull(skb, skb_transport_offset(skb) +
239 sizeof(struct udphdr));
240}
241
242static bool sctphdr_ok(struct sk_buff *skb)
243{
244 return pskb_may_pull(skb, skb_transport_offset(skb) +
245 sizeof(struct sctphdr));
246}
247
248static bool icmphdr_ok(struct sk_buff *skb)
249{
250 return pskb_may_pull(skb, skb_transport_offset(skb) +
251 sizeof(struct icmphdr));
252}
253
254static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
255{
256 unsigned int nh_ofs = skb_network_offset(skb);
257 unsigned int nh_len;
258 int payload_ofs;
259 struct ipv6hdr *nh;
260 uint8_t nexthdr;
261 __be16 frag_off;
262 int err;
263
264 err = check_header(skb, nh_ofs + sizeof(*nh));
265 if (unlikely(err))
266 return err;
267
268 nh = ipv6_hdr(skb);
269 nexthdr = nh->nexthdr;
270 payload_ofs = (u8 *)(nh + 1) - skb->data;
271
272 key->ip.proto = NEXTHDR_NONE;
273 key->ip.tos = ipv6_get_dsfield(nh);
274 key->ip.ttl = nh->hop_limit;
275 key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
276 key->ipv6.addr.src = nh->saddr;
277 key->ipv6.addr.dst = nh->daddr;
278
279 payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
280
281 if (frag_off) {
282 if (frag_off & htons(~0x7))
283 key->ip.frag = OVS_FRAG_TYPE_LATER;
284 else
285 key->ip.frag = OVS_FRAG_TYPE_FIRST;
286 } else {
287 key->ip.frag = OVS_FRAG_TYPE_NONE;
288 }
289
290 /* Delayed handling of error in ipv6_skip_exthdr() as it
291 * always sets frag_off to a valid value which may be
292 * used to set key->ip.frag above.
293 */
294 if (unlikely(payload_ofs < 0))
295 return -EPROTO;
296
297 nh_len = payload_ofs - nh_ofs;
298 skb_set_transport_header(skb, nh_ofs + nh_len);
299 key->ip.proto = nexthdr;
300 return nh_len;
301}
302
303static bool icmp6hdr_ok(struct sk_buff *skb)
304{
305 return pskb_may_pull(skb, skb_transport_offset(skb) +
306 sizeof(struct icmp6hdr));
307}
308
309/**
310 * Parse vlan tag from vlan header.
311 * Returns ERROR on memory error.
312 * Returns 0 if it encounters a non-vlan or incomplete packet.
313 * Returns 1 after successfully parsing vlan tag.
314 */
315static int parse_vlan_tag(struct sk_buff *skb, struct vlan_head *key_vh,
316 bool untag_vlan)
317{
318 struct vlan_head *vh = (struct vlan_head *)skb->data;
319
320 if (likely(!eth_type_vlan(vh->tpid)))
321 return 0;
322
323 if (unlikely(skb->len < sizeof(struct vlan_head) + sizeof(__be16)))
324 return 0;
325
326 if (unlikely(!pskb_may_pull(skb, sizeof(struct vlan_head) +
327 sizeof(__be16))))
328 return -ENOMEM;
329
330 vh = (struct vlan_head *)skb->data;
331 key_vh->tci = vh->tci | htons(VLAN_TAG_PRESENT);
332 key_vh->tpid = vh->tpid;
333
334 if (unlikely(untag_vlan)) {
335 int offset = skb->data - skb_mac_header(skb);
336 u16 tci;
337 int err;
338
339 __skb_push(skb, offset);
340 err = __skb_vlan_pop(skb, &tci);
341 __skb_pull(skb, offset);
342 if (err)
343 return err;
344 __vlan_hwaccel_put_tag(skb, key_vh->tpid, tci);
345 } else {
346 __skb_pull(skb, sizeof(struct vlan_head));
347 }
348 return 1;
349}
350
351static void clear_vlan(struct sw_flow_key *key)
352{
353 key->eth.vlan.tci = 0;
354 key->eth.vlan.tpid = 0;
355 key->eth.cvlan.tci = 0;
356 key->eth.cvlan.tpid = 0;
357}
358
359static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
360{
361 int res;
362
363 if (skb_vlan_tag_present(skb)) {
364 key->eth.vlan.tci = htons(skb->vlan_tci);
365 key->eth.vlan.tpid = skb->vlan_proto;
366 } else {
367 /* Parse outer vlan tag in the non-accelerated case. */
368 res = parse_vlan_tag(skb, &key->eth.vlan, true);
369 if (res <= 0)
370 return res;
371 }
372
373 /* Parse inner vlan tag. */
374 res = parse_vlan_tag(skb, &key->eth.cvlan, false);
375 if (res <= 0)
376 return res;
377
378 return 0;
379}
380
381static __be16 parse_ethertype(struct sk_buff *skb)
382{
383 struct llc_snap_hdr {
384 u8 dsap; /* Always 0xAA */
385 u8 ssap; /* Always 0xAA */
386 u8 ctrl;
387 u8 oui[3];
388 __be16 ethertype;
389 };
390 struct llc_snap_hdr *llc;
391 __be16 proto;
392
393 proto = *(__be16 *) skb->data;
394 __skb_pull(skb, sizeof(__be16));
395
396 if (eth_proto_is_802_3(proto))
397 return proto;
398
399 if (skb->len < sizeof(struct llc_snap_hdr))
400 return htons(ETH_P_802_2);
401
402 if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
403 return htons(0);
404
405 llc = (struct llc_snap_hdr *) skb->data;
406 if (llc->dsap != LLC_SAP_SNAP ||
407 llc->ssap != LLC_SAP_SNAP ||
408 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
409 return htons(ETH_P_802_2);
410
411 __skb_pull(skb, sizeof(struct llc_snap_hdr));
412
413 if (eth_proto_is_802_3(llc->ethertype))
414 return llc->ethertype;
415
416 return htons(ETH_P_802_2);
417}
418
419static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
420 int nh_len)
421{
422 struct icmp6hdr *icmp = icmp6_hdr(skb);
423
424 /* The ICMPv6 type and code fields use the 16-bit transport port
425 * fields, so we need to store them in 16-bit network byte order.
426 */
427 key->tp.src = htons(icmp->icmp6_type);
428 key->tp.dst = htons(icmp->icmp6_code);
429 memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
430
431 if (icmp->icmp6_code == 0 &&
432 (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
433 icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
434 int icmp_len = skb->len - skb_transport_offset(skb);
435 struct nd_msg *nd;
436 int offset;
437
438 /* In order to process neighbor discovery options, we need the
439 * entire packet.
440 */
441 if (unlikely(icmp_len < sizeof(*nd)))
442 return 0;
443
444 if (unlikely(skb_linearize(skb)))
445 return -ENOMEM;
446
447 nd = (struct nd_msg *)skb_transport_header(skb);
448 key->ipv6.nd.target = nd->target;
449
450 icmp_len -= sizeof(*nd);
451 offset = 0;
452 while (icmp_len >= 8) {
453 struct nd_opt_hdr *nd_opt =
454 (struct nd_opt_hdr *)(nd->opt + offset);
455 int opt_len = nd_opt->nd_opt_len * 8;
456
457 if (unlikely(!opt_len || opt_len > icmp_len))
458 return 0;
459
460 /* Store the link layer address if the appropriate
461 * option is provided. It is considered an error if
462 * the same link layer option is specified twice.
463 */
464 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
465 && opt_len == 8) {
466 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
467 goto invalid;
468 ether_addr_copy(key->ipv6.nd.sll,
469 &nd->opt[offset+sizeof(*nd_opt)]);
470 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
471 && opt_len == 8) {
472 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
473 goto invalid;
474 ether_addr_copy(key->ipv6.nd.tll,
475 &nd->opt[offset+sizeof(*nd_opt)]);
476 }
477
478 icmp_len -= opt_len;
479 offset += opt_len;
480 }
481 }
482
483 return 0;
484
485invalid:
486 memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
487 memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
488 memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
489
490 return 0;
491}
492
493/**
494 * key_extract - extracts a flow key from an Ethernet frame.
495 * @skb: sk_buff that contains the frame, with skb->data pointing to the
496 * Ethernet header
497 * @key: output flow key
498 *
499 * The caller must ensure that skb->len >= ETH_HLEN.
500 *
501 * Returns 0 if successful, otherwise a negative errno value.
502 *
503 * Initializes @skb header fields as follows:
504 *
505 * - skb->mac_header: the L2 header.
506 *
507 * - skb->network_header: just past the L2 header, or just past the
508 * VLAN header, to the first byte of the L2 payload.
509 *
510 * - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
511 * on output, then just past the IP header, if one is present and
512 * of a correct length, otherwise the same as skb->network_header.
513 * For other key->eth.type values it is left untouched.
514 *
515 * - skb->protocol: the type of the data starting at skb->network_header.
516 * Equals to key->eth.type.
517 */
518static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
519{
520 int error;
521 struct ethhdr *eth;
522
523 /* Flags are always used as part of stats */
524 key->tp.flags = 0;
525
526 skb_reset_mac_header(skb);
527
528 /* Link layer. */
529 clear_vlan(key);
530 if (key->mac_proto == MAC_PROTO_NONE) {
531 if (unlikely(eth_type_vlan(skb->protocol)))
532 return -EINVAL;
533
534 skb_reset_network_header(skb);
535 } else {
536 eth = eth_hdr(skb);
537 ether_addr_copy(key->eth.src, eth->h_source);
538 ether_addr_copy(key->eth.dst, eth->h_dest);
539
540 __skb_pull(skb, 2 * ETH_ALEN);
541 /* We are going to push all headers that we pull, so no need to
542 * update skb->csum here.
543 */
544
545 if (unlikely(parse_vlan(skb, key)))
546 return -ENOMEM;
547
548 skb->protocol = parse_ethertype(skb);
549 if (unlikely(skb->protocol == htons(0)))
550 return -ENOMEM;
551
552 skb_reset_network_header(skb);
553 __skb_push(skb, skb->data - skb_mac_header(skb));
554 }
555 skb_reset_mac_len(skb);
556 key->eth.type = skb->protocol;
557
558 /* Network layer. */
559 if (key->eth.type == htons(ETH_P_IP)) {
560 struct iphdr *nh;
561 __be16 offset;
562
563 error = check_iphdr(skb);
564 if (unlikely(error)) {
565 memset(&key->ip, 0, sizeof(key->ip));
566 memset(&key->ipv4, 0, sizeof(key->ipv4));
567 if (error == -EINVAL) {
568 skb->transport_header = skb->network_header;
569 error = 0;
570 }
571 return error;
572 }
573
574 nh = ip_hdr(skb);
575 key->ipv4.addr.src = nh->saddr;
576 key->ipv4.addr.dst = nh->daddr;
577
578 key->ip.proto = nh->protocol;
579 key->ip.tos = nh->tos;
580 key->ip.ttl = nh->ttl;
581
582 offset = nh->frag_off & htons(IP_OFFSET);
583 if (offset) {
584 key->ip.frag = OVS_FRAG_TYPE_LATER;
585 return 0;
586 }
587 if (nh->frag_off & htons(IP_MF) ||
588 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
589 key->ip.frag = OVS_FRAG_TYPE_FIRST;
590 else
591 key->ip.frag = OVS_FRAG_TYPE_NONE;
592
593 /* Transport layer. */
594 if (key->ip.proto == IPPROTO_TCP) {
595 if (tcphdr_ok(skb)) {
596 struct tcphdr *tcp = tcp_hdr(skb);
597 key->tp.src = tcp->source;
598 key->tp.dst = tcp->dest;
599 key->tp.flags = TCP_FLAGS_BE16(tcp);
600 } else {
601 memset(&key->tp, 0, sizeof(key->tp));
602 }
603
604 } else if (key->ip.proto == IPPROTO_UDP) {
605 if (udphdr_ok(skb)) {
606 struct udphdr *udp = udp_hdr(skb);
607 key->tp.src = udp->source;
608 key->tp.dst = udp->dest;
609 } else {
610 memset(&key->tp, 0, sizeof(key->tp));
611 }
612 } else if (key->ip.proto == IPPROTO_SCTP) {
613 if (sctphdr_ok(skb)) {
614 struct sctphdr *sctp = sctp_hdr(skb);
615 key->tp.src = sctp->source;
616 key->tp.dst = sctp->dest;
617 } else {
618 memset(&key->tp, 0, sizeof(key->tp));
619 }
620 } else if (key->ip.proto == IPPROTO_ICMP) {
621 if (icmphdr_ok(skb)) {
622 struct icmphdr *icmp = icmp_hdr(skb);
623 /* The ICMP type and code fields use the 16-bit
624 * transport port fields, so we need to store
625 * them in 16-bit network byte order. */
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 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
705 key->ip.frag = OVS_FRAG_TYPE_FIRST;
706
707 /* Transport layer. */
708 if (key->ip.proto == NEXTHDR_TCP) {
709 if (tcphdr_ok(skb)) {
710 struct tcphdr *tcp = tcp_hdr(skb);
711 key->tp.src = tcp->source;
712 key->tp.dst = tcp->dest;
713 key->tp.flags = TCP_FLAGS_BE16(tcp);
714 } else {
715 memset(&key->tp, 0, sizeof(key->tp));
716 }
717 } else if (key->ip.proto == NEXTHDR_UDP) {
718 if (udphdr_ok(skb)) {
719 struct udphdr *udp = udp_hdr(skb);
720 key->tp.src = udp->source;
721 key->tp.dst = udp->dest;
722 } else {
723 memset(&key->tp, 0, sizeof(key->tp));
724 }
725 } else if (key->ip.proto == NEXTHDR_SCTP) {
726 if (sctphdr_ok(skb)) {
727 struct sctphdr *sctp = sctp_hdr(skb);
728 key->tp.src = sctp->source;
729 key->tp.dst = sctp->dest;
730 } else {
731 memset(&key->tp, 0, sizeof(key->tp));
732 }
733 } else if (key->ip.proto == NEXTHDR_ICMP) {
734 if (icmp6hdr_ok(skb)) {
735 error = parse_icmpv6(skb, key, nh_len);
736 if (error)
737 return error;
738 } else {
739 memset(&key->tp, 0, sizeof(key->tp));
740 }
741 }
742 }
743 return 0;
744}
745
746int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
747{
748 return key_extract(skb, key);
749}
750
751static int key_extract_mac_proto(struct sk_buff *skb)
752{
753 switch (skb->dev->type) {
754 case ARPHRD_ETHER:
755 return MAC_PROTO_ETHERNET;
756 case ARPHRD_NONE:
757 if (skb->protocol == htons(ETH_P_TEB))
758 return MAC_PROTO_ETHERNET;
759 return MAC_PROTO_NONE;
760 }
761 WARN_ON_ONCE(1);
762 return -EINVAL;
763}
764
765int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
766 struct sk_buff *skb, struct sw_flow_key *key)
767{
768 int res;
769
770 /* Extract metadata from packet. */
771 if (tun_info) {
772 key->tun_proto = ip_tunnel_info_af(tun_info);
773 memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
774
775 if (tun_info->options_len) {
776 BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) *
777 8)) - 1
778 > sizeof(key->tun_opts));
779
780 ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key, tun_info->options_len),
781 tun_info);
782 key->tun_opts_len = tun_info->options_len;
783 } else {
784 key->tun_opts_len = 0;
785 }
786 } else {
787 key->tun_proto = 0;
788 key->tun_opts_len = 0;
789 memset(&key->tun_key, 0, sizeof(key->tun_key));
790 }
791
792 key->phy.priority = skb->priority;
793 key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
794 key->phy.skb_mark = skb->mark;
795 ovs_ct_fill_key(skb, key);
796 key->ovs_flow_hash = 0;
797 res = key_extract_mac_proto(skb);
798 if (res < 0)
799 return res;
800 key->mac_proto = res;
801 key->recirc_id = 0;
802
803 return key_extract(skb, key);
804}
805
806int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
807 struct sk_buff *skb,
808 struct sw_flow_key *key, bool log)
809{
810 int err;
811
812 /* Extract metadata from netlink attributes. */
813 err = ovs_nla_get_flow_metadata(net, attr, key, log);
814 if (err)
815 return err;
816
817 /* key_extract assumes that skb->protocol is set-up for
818 * layer 3 packets which is the case for other callers,
819 * in particular packets received from the network stack.
820 * Here the correct value can be set from the metadata
821 * extracted above.
822 * For L2 packet key eth type would be zero. skb protocol
823 * would be set to correct value later during key-extact.
824 */
825
826 skb->protocol = key->eth.type;
827 return key_extract(skb, key);
828}
1/*
2 * Copyright (c) 2007-2014 Nicira, Inc.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16 * 02110-1301, USA
17 */
18
19#include <linux/uaccess.h>
20#include <linux/netdevice.h>
21#include <linux/etherdevice.h>
22#include <linux/if_ether.h>
23#include <linux/if_vlan.h>
24#include <net/llc_pdu.h>
25#include <linux/kernel.h>
26#include <linux/jhash.h>
27#include <linux/jiffies.h>
28#include <linux/llc.h>
29#include <linux/module.h>
30#include <linux/in.h>
31#include <linux/rcupdate.h>
32#include <linux/cpumask.h>
33#include <linux/if_arp.h>
34#include <linux/ip.h>
35#include <linux/ipv6.h>
36#include <linux/mpls.h>
37#include <linux/sctp.h>
38#include <linux/smp.h>
39#include <linux/tcp.h>
40#include <linux/udp.h>
41#include <linux/icmp.h>
42#include <linux/icmpv6.h>
43#include <linux/rculist.h>
44#include <net/ip.h>
45#include <net/ip_tunnels.h>
46#include <net/ipv6.h>
47#include <net/mpls.h>
48#include <net/ndisc.h>
49#include <net/nsh.h>
50
51#include "conntrack.h"
52#include "datapath.h"
53#include "flow.h"
54#include "flow_netlink.h"
55#include "vport.h"
56
57u64 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
72void 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;
133unlock:
134 spin_unlock(&stats->lock);
135}
136
137/* Must be called with rcu_read_lock or ovs_mutex. */
138void 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. */
168void 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
187static 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
196static 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
202static 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
221static 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
237static bool udphdr_ok(struct sk_buff *skb)
238{
239 return pskb_may_pull(skb, skb_transport_offset(skb) +
240 sizeof(struct udphdr));
241}
242
243static bool sctphdr_ok(struct sk_buff *skb)
244{
245 return pskb_may_pull(skb, skb_transport_offset(skb) +
246 sizeof(struct sctphdr));
247}
248
249static bool icmphdr_ok(struct sk_buff *skb)
250{
251 return pskb_may_pull(skb, skb_transport_offset(skb) +
252 sizeof(struct icmphdr));
253}
254
255static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
256{
257 unsigned int nh_ofs = skb_network_offset(skb);
258 unsigned int nh_len;
259 int payload_ofs;
260 struct ipv6hdr *nh;
261 uint8_t nexthdr;
262 __be16 frag_off;
263 int err;
264
265 err = check_header(skb, nh_ofs + sizeof(*nh));
266 if (unlikely(err))
267 return err;
268
269 nh = ipv6_hdr(skb);
270 nexthdr = nh->nexthdr;
271 payload_ofs = (u8 *)(nh + 1) - skb->data;
272
273 key->ip.proto = NEXTHDR_NONE;
274 key->ip.tos = ipv6_get_dsfield(nh);
275 key->ip.ttl = nh->hop_limit;
276 key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
277 key->ipv6.addr.src = nh->saddr;
278 key->ipv6.addr.dst = nh->daddr;
279
280 payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
281
282 if (frag_off) {
283 if (frag_off & htons(~0x7))
284 key->ip.frag = OVS_FRAG_TYPE_LATER;
285 else
286 key->ip.frag = OVS_FRAG_TYPE_FIRST;
287 } else {
288 key->ip.frag = OVS_FRAG_TYPE_NONE;
289 }
290
291 /* Delayed handling of error in ipv6_skip_exthdr() as it
292 * always sets frag_off to a valid value which may be
293 * used to set key->ip.frag above.
294 */
295 if (unlikely(payload_ofs < 0))
296 return -EPROTO;
297
298 nh_len = payload_ofs - nh_ofs;
299 skb_set_transport_header(skb, nh_ofs + nh_len);
300 key->ip.proto = nexthdr;
301 return nh_len;
302}
303
304static bool icmp6hdr_ok(struct sk_buff *skb)
305{
306 return pskb_may_pull(skb, skb_transport_offset(skb) +
307 sizeof(struct icmp6hdr));
308}
309
310/**
311 * Parse vlan tag from vlan header.
312 * Returns ERROR on memory error.
313 * Returns 0 if it encounters a non-vlan or incomplete packet.
314 * Returns 1 after successfully parsing vlan tag.
315 */
316static int parse_vlan_tag(struct sk_buff *skb, struct vlan_head *key_vh,
317 bool untag_vlan)
318{
319 struct vlan_head *vh = (struct vlan_head *)skb->data;
320
321 if (likely(!eth_type_vlan(vh->tpid)))
322 return 0;
323
324 if (unlikely(skb->len < sizeof(struct vlan_head) + sizeof(__be16)))
325 return 0;
326
327 if (unlikely(!pskb_may_pull(skb, sizeof(struct vlan_head) +
328 sizeof(__be16))))
329 return -ENOMEM;
330
331 vh = (struct vlan_head *)skb->data;
332 key_vh->tci = vh->tci | htons(VLAN_TAG_PRESENT);
333 key_vh->tpid = vh->tpid;
334
335 if (unlikely(untag_vlan)) {
336 int offset = skb->data - skb_mac_header(skb);
337 u16 tci;
338 int err;
339
340 __skb_push(skb, offset);
341 err = __skb_vlan_pop(skb, &tci);
342 __skb_pull(skb, offset);
343 if (err)
344 return err;
345 __vlan_hwaccel_put_tag(skb, key_vh->tpid, tci);
346 } else {
347 __skb_pull(skb, sizeof(struct vlan_head));
348 }
349 return 1;
350}
351
352static void clear_vlan(struct sw_flow_key *key)
353{
354 key->eth.vlan.tci = 0;
355 key->eth.vlan.tpid = 0;
356 key->eth.cvlan.tci = 0;
357 key->eth.cvlan.tpid = 0;
358}
359
360static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
361{
362 int res;
363
364 if (skb_vlan_tag_present(skb)) {
365 key->eth.vlan.tci = htons(skb->vlan_tci);
366 key->eth.vlan.tpid = skb->vlan_proto;
367 } else {
368 /* Parse outer vlan tag in the non-accelerated case. */
369 res = parse_vlan_tag(skb, &key->eth.vlan, true);
370 if (res <= 0)
371 return res;
372 }
373
374 /* Parse inner vlan tag. */
375 res = parse_vlan_tag(skb, &key->eth.cvlan, false);
376 if (res <= 0)
377 return res;
378
379 return 0;
380}
381
382static __be16 parse_ethertype(struct sk_buff *skb)
383{
384 struct llc_snap_hdr {
385 u8 dsap; /* Always 0xAA */
386 u8 ssap; /* Always 0xAA */
387 u8 ctrl;
388 u8 oui[3];
389 __be16 ethertype;
390 };
391 struct llc_snap_hdr *llc;
392 __be16 proto;
393
394 proto = *(__be16 *) skb->data;
395 __skb_pull(skb, sizeof(__be16));
396
397 if (eth_proto_is_802_3(proto))
398 return proto;
399
400 if (skb->len < sizeof(struct llc_snap_hdr))
401 return htons(ETH_P_802_2);
402
403 if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
404 return htons(0);
405
406 llc = (struct llc_snap_hdr *) skb->data;
407 if (llc->dsap != LLC_SAP_SNAP ||
408 llc->ssap != LLC_SAP_SNAP ||
409 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
410 return htons(ETH_P_802_2);
411
412 __skb_pull(skb, sizeof(struct llc_snap_hdr));
413
414 if (eth_proto_is_802_3(llc->ethertype))
415 return llc->ethertype;
416
417 return htons(ETH_P_802_2);
418}
419
420static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
421 int nh_len)
422{
423 struct icmp6hdr *icmp = icmp6_hdr(skb);
424
425 /* The ICMPv6 type and code fields use the 16-bit transport port
426 * fields, so we need to store them in 16-bit network byte order.
427 */
428 key->tp.src = htons(icmp->icmp6_type);
429 key->tp.dst = htons(icmp->icmp6_code);
430 memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
431
432 if (icmp->icmp6_code == 0 &&
433 (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
434 icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
435 int icmp_len = skb->len - skb_transport_offset(skb);
436 struct nd_msg *nd;
437 int offset;
438
439 /* In order to process neighbor discovery options, we need the
440 * entire packet.
441 */
442 if (unlikely(icmp_len < sizeof(*nd)))
443 return 0;
444
445 if (unlikely(skb_linearize(skb)))
446 return -ENOMEM;
447
448 nd = (struct nd_msg *)skb_transport_header(skb);
449 key->ipv6.nd.target = nd->target;
450
451 icmp_len -= sizeof(*nd);
452 offset = 0;
453 while (icmp_len >= 8) {
454 struct nd_opt_hdr *nd_opt =
455 (struct nd_opt_hdr *)(nd->opt + offset);
456 int opt_len = nd_opt->nd_opt_len * 8;
457
458 if (unlikely(!opt_len || opt_len > icmp_len))
459 return 0;
460
461 /* Store the link layer address if the appropriate
462 * option is provided. It is considered an error if
463 * the same link layer option is specified twice.
464 */
465 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
466 && opt_len == 8) {
467 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
468 goto invalid;
469 ether_addr_copy(key->ipv6.nd.sll,
470 &nd->opt[offset+sizeof(*nd_opt)]);
471 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
472 && opt_len == 8) {
473 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
474 goto invalid;
475 ether_addr_copy(key->ipv6.nd.tll,
476 &nd->opt[offset+sizeof(*nd_opt)]);
477 }
478
479 icmp_len -= opt_len;
480 offset += opt_len;
481 }
482 }
483
484 return 0;
485
486invalid:
487 memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
488 memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
489 memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
490
491 return 0;
492}
493
494static int parse_nsh(struct sk_buff *skb, struct sw_flow_key *key)
495{
496 struct nshhdr *nh;
497 unsigned int nh_ofs = skb_network_offset(skb);
498 u8 version, length;
499 int err;
500
501 err = check_header(skb, nh_ofs + NSH_BASE_HDR_LEN);
502 if (unlikely(err))
503 return err;
504
505 nh = nsh_hdr(skb);
506 version = nsh_get_ver(nh);
507 length = nsh_hdr_len(nh);
508
509 if (version != 0)
510 return -EINVAL;
511
512 err = check_header(skb, nh_ofs + length);
513 if (unlikely(err))
514 return err;
515
516 nh = nsh_hdr(skb);
517 key->nsh.base.flags = nsh_get_flags(nh);
518 key->nsh.base.ttl = nsh_get_ttl(nh);
519 key->nsh.base.mdtype = nh->mdtype;
520 key->nsh.base.np = nh->np;
521 key->nsh.base.path_hdr = nh->path_hdr;
522 switch (key->nsh.base.mdtype) {
523 case NSH_M_TYPE1:
524 if (length != NSH_M_TYPE1_LEN)
525 return -EINVAL;
526 memcpy(key->nsh.context, nh->md1.context,
527 sizeof(nh->md1));
528 break;
529 case NSH_M_TYPE2:
530 memset(key->nsh.context, 0,
531 sizeof(nh->md1));
532 break;
533 default:
534 return -EINVAL;
535 }
536
537 return 0;
538}
539
540/**
541 * key_extract - extracts a flow key from an Ethernet frame.
542 * @skb: sk_buff that contains the frame, with skb->data pointing to the
543 * Ethernet header
544 * @key: output flow key
545 *
546 * The caller must ensure that skb->len >= ETH_HLEN.
547 *
548 * Returns 0 if successful, otherwise a negative errno value.
549 *
550 * Initializes @skb header fields as follows:
551 *
552 * - skb->mac_header: the L2 header.
553 *
554 * - skb->network_header: just past the L2 header, or just past the
555 * VLAN header, to the first byte of the L2 payload.
556 *
557 * - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
558 * on output, then just past the IP header, if one is present and
559 * of a correct length, otherwise the same as skb->network_header.
560 * For other key->eth.type values it is left untouched.
561 *
562 * - skb->protocol: the type of the data starting at skb->network_header.
563 * Equals to key->eth.type.
564 */
565static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
566{
567 int error;
568 struct ethhdr *eth;
569
570 /* Flags are always used as part of stats */
571 key->tp.flags = 0;
572
573 skb_reset_mac_header(skb);
574
575 /* Link layer. */
576 clear_vlan(key);
577 if (ovs_key_mac_proto(key) == MAC_PROTO_NONE) {
578 if (unlikely(eth_type_vlan(skb->protocol)))
579 return -EINVAL;
580
581 skb_reset_network_header(skb);
582 key->eth.type = skb->protocol;
583 } else {
584 eth = eth_hdr(skb);
585 ether_addr_copy(key->eth.src, eth->h_source);
586 ether_addr_copy(key->eth.dst, eth->h_dest);
587
588 __skb_pull(skb, 2 * ETH_ALEN);
589 /* We are going to push all headers that we pull, so no need to
590 * update skb->csum here.
591 */
592
593 if (unlikely(parse_vlan(skb, key)))
594 return -ENOMEM;
595
596 key->eth.type = parse_ethertype(skb);
597 if (unlikely(key->eth.type == htons(0)))
598 return -ENOMEM;
599
600 /* Multiple tagged packets need to retain TPID to satisfy
601 * skb_vlan_pop(), which will later shift the ethertype into
602 * skb->protocol.
603 */
604 if (key->eth.cvlan.tci & htons(VLAN_TAG_PRESENT))
605 skb->protocol = key->eth.cvlan.tpid;
606 else
607 skb->protocol = key->eth.type;
608
609 skb_reset_network_header(skb);
610 __skb_push(skb, skb->data - skb_mac_header(skb));
611 }
612 skb_reset_mac_len(skb);
613
614 /* Network layer. */
615 if (key->eth.type == htons(ETH_P_IP)) {
616 struct iphdr *nh;
617 __be16 offset;
618
619 error = check_iphdr(skb);
620 if (unlikely(error)) {
621 memset(&key->ip, 0, sizeof(key->ip));
622 memset(&key->ipv4, 0, sizeof(key->ipv4));
623 if (error == -EINVAL) {
624 skb->transport_header = skb->network_header;
625 error = 0;
626 }
627 return error;
628 }
629
630 nh = ip_hdr(skb);
631 key->ipv4.addr.src = nh->saddr;
632 key->ipv4.addr.dst = nh->daddr;
633
634 key->ip.proto = nh->protocol;
635 key->ip.tos = nh->tos;
636 key->ip.ttl = nh->ttl;
637
638 offset = nh->frag_off & htons(IP_OFFSET);
639 if (offset) {
640 key->ip.frag = OVS_FRAG_TYPE_LATER;
641 return 0;
642 }
643 if (nh->frag_off & htons(IP_MF) ||
644 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
645 key->ip.frag = OVS_FRAG_TYPE_FIRST;
646 else
647 key->ip.frag = OVS_FRAG_TYPE_NONE;
648
649 /* Transport layer. */
650 if (key->ip.proto == IPPROTO_TCP) {
651 if (tcphdr_ok(skb)) {
652 struct tcphdr *tcp = tcp_hdr(skb);
653 key->tp.src = tcp->source;
654 key->tp.dst = tcp->dest;
655 key->tp.flags = TCP_FLAGS_BE16(tcp);
656 } else {
657 memset(&key->tp, 0, sizeof(key->tp));
658 }
659
660 } else if (key->ip.proto == IPPROTO_UDP) {
661 if (udphdr_ok(skb)) {
662 struct udphdr *udp = udp_hdr(skb);
663 key->tp.src = udp->source;
664 key->tp.dst = udp->dest;
665 } else {
666 memset(&key->tp, 0, sizeof(key->tp));
667 }
668 } else if (key->ip.proto == IPPROTO_SCTP) {
669 if (sctphdr_ok(skb)) {
670 struct sctphdr *sctp = sctp_hdr(skb);
671 key->tp.src = sctp->source;
672 key->tp.dst = sctp->dest;
673 } else {
674 memset(&key->tp, 0, sizeof(key->tp));
675 }
676 } else if (key->ip.proto == IPPROTO_ICMP) {
677 if (icmphdr_ok(skb)) {
678 struct icmphdr *icmp = icmp_hdr(skb);
679 /* The ICMP type and code fields use the 16-bit
680 * transport port fields, so we need to store
681 * them in 16-bit network byte order. */
682 key->tp.src = htons(icmp->type);
683 key->tp.dst = htons(icmp->code);
684 } else {
685 memset(&key->tp, 0, sizeof(key->tp));
686 }
687 }
688
689 } else if (key->eth.type == htons(ETH_P_ARP) ||
690 key->eth.type == htons(ETH_P_RARP)) {
691 struct arp_eth_header *arp;
692 bool arp_available = arphdr_ok(skb);
693
694 arp = (struct arp_eth_header *)skb_network_header(skb);
695
696 if (arp_available &&
697 arp->ar_hrd == htons(ARPHRD_ETHER) &&
698 arp->ar_pro == htons(ETH_P_IP) &&
699 arp->ar_hln == ETH_ALEN &&
700 arp->ar_pln == 4) {
701
702 /* We only match on the lower 8 bits of the opcode. */
703 if (ntohs(arp->ar_op) <= 0xff)
704 key->ip.proto = ntohs(arp->ar_op);
705 else
706 key->ip.proto = 0;
707
708 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
709 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
710 ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
711 ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
712 } else {
713 memset(&key->ip, 0, sizeof(key->ip));
714 memset(&key->ipv4, 0, sizeof(key->ipv4));
715 }
716 } else if (eth_p_mpls(key->eth.type)) {
717 size_t stack_len = MPLS_HLEN;
718
719 skb_set_inner_network_header(skb, skb->mac_len);
720 while (1) {
721 __be32 lse;
722
723 error = check_header(skb, skb->mac_len + stack_len);
724 if (unlikely(error))
725 return 0;
726
727 memcpy(&lse, skb_inner_network_header(skb), MPLS_HLEN);
728
729 if (stack_len == MPLS_HLEN)
730 memcpy(&key->mpls.top_lse, &lse, MPLS_HLEN);
731
732 skb_set_inner_network_header(skb, skb->mac_len + stack_len);
733 if (lse & htonl(MPLS_LS_S_MASK))
734 break;
735
736 stack_len += MPLS_HLEN;
737 }
738 } else if (key->eth.type == htons(ETH_P_IPV6)) {
739 int nh_len; /* IPv6 Header + Extensions */
740
741 nh_len = parse_ipv6hdr(skb, key);
742 if (unlikely(nh_len < 0)) {
743 switch (nh_len) {
744 case -EINVAL:
745 memset(&key->ip, 0, sizeof(key->ip));
746 memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
747 /* fall-through */
748 case -EPROTO:
749 skb->transport_header = skb->network_header;
750 error = 0;
751 break;
752 default:
753 error = nh_len;
754 }
755 return error;
756 }
757
758 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
759 return 0;
760 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
761 key->ip.frag = OVS_FRAG_TYPE_FIRST;
762
763 /* Transport layer. */
764 if (key->ip.proto == NEXTHDR_TCP) {
765 if (tcphdr_ok(skb)) {
766 struct tcphdr *tcp = tcp_hdr(skb);
767 key->tp.src = tcp->source;
768 key->tp.dst = tcp->dest;
769 key->tp.flags = TCP_FLAGS_BE16(tcp);
770 } else {
771 memset(&key->tp, 0, sizeof(key->tp));
772 }
773 } else if (key->ip.proto == NEXTHDR_UDP) {
774 if (udphdr_ok(skb)) {
775 struct udphdr *udp = udp_hdr(skb);
776 key->tp.src = udp->source;
777 key->tp.dst = udp->dest;
778 } else {
779 memset(&key->tp, 0, sizeof(key->tp));
780 }
781 } else if (key->ip.proto == NEXTHDR_SCTP) {
782 if (sctphdr_ok(skb)) {
783 struct sctphdr *sctp = sctp_hdr(skb);
784 key->tp.src = sctp->source;
785 key->tp.dst = sctp->dest;
786 } else {
787 memset(&key->tp, 0, sizeof(key->tp));
788 }
789 } else if (key->ip.proto == NEXTHDR_ICMP) {
790 if (icmp6hdr_ok(skb)) {
791 error = parse_icmpv6(skb, key, nh_len);
792 if (error)
793 return error;
794 } else {
795 memset(&key->tp, 0, sizeof(key->tp));
796 }
797 }
798 } else if (key->eth.type == htons(ETH_P_NSH)) {
799 error = parse_nsh(skb, key);
800 if (error)
801 return error;
802 }
803 return 0;
804}
805
806int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
807{
808 int res;
809
810 res = key_extract(skb, key);
811 if (!res)
812 key->mac_proto &= ~SW_FLOW_KEY_INVALID;
813
814 return res;
815}
816
817static int key_extract_mac_proto(struct sk_buff *skb)
818{
819 switch (skb->dev->type) {
820 case ARPHRD_ETHER:
821 return MAC_PROTO_ETHERNET;
822 case ARPHRD_NONE:
823 if (skb->protocol == htons(ETH_P_TEB))
824 return MAC_PROTO_ETHERNET;
825 return MAC_PROTO_NONE;
826 }
827 WARN_ON_ONCE(1);
828 return -EINVAL;
829}
830
831int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
832 struct sk_buff *skb, struct sw_flow_key *key)
833{
834 int res, err;
835
836 /* Extract metadata from packet. */
837 if (tun_info) {
838 key->tun_proto = ip_tunnel_info_af(tun_info);
839 memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
840
841 if (tun_info->options_len) {
842 BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) *
843 8)) - 1
844 > sizeof(key->tun_opts));
845
846 ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key, tun_info->options_len),
847 tun_info);
848 key->tun_opts_len = tun_info->options_len;
849 } else {
850 key->tun_opts_len = 0;
851 }
852 } else {
853 key->tun_proto = 0;
854 key->tun_opts_len = 0;
855 memset(&key->tun_key, 0, sizeof(key->tun_key));
856 }
857
858 key->phy.priority = skb->priority;
859 key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
860 key->phy.skb_mark = skb->mark;
861 key->ovs_flow_hash = 0;
862 res = key_extract_mac_proto(skb);
863 if (res < 0)
864 return res;
865 key->mac_proto = res;
866 key->recirc_id = 0;
867
868 err = key_extract(skb, key);
869 if (!err)
870 ovs_ct_fill_key(skb, key); /* Must be after key_extract(). */
871 return err;
872}
873
874int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
875 struct sk_buff *skb,
876 struct sw_flow_key *key, bool log)
877{
878 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
879 u64 attrs = 0;
880 int err;
881
882 err = parse_flow_nlattrs(attr, a, &attrs, log);
883 if (err)
884 return -EINVAL;
885
886 /* Extract metadata from netlink attributes. */
887 err = ovs_nla_get_flow_metadata(net, a, attrs, key, log);
888 if (err)
889 return err;
890
891 /* key_extract assumes that skb->protocol is set-up for
892 * layer 3 packets which is the case for other callers,
893 * in particular packets received from the network stack.
894 * Here the correct value can be set from the metadata
895 * extracted above.
896 * For L2 packet key eth type would be zero. skb protocol
897 * would be set to correct value later during key-extact.
898 */
899
900 skb->protocol = key->eth.type;
901 err = key_extract(skb, key);
902 if (err)
903 return err;
904
905 /* Check that we have conntrack original direction tuple metadata only
906 * for packets for which it makes sense. Otherwise the key may be
907 * corrupted due to overlapping key fields.
908 */
909 if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4) &&
910 key->eth.type != htons(ETH_P_IP))
911 return -EINVAL;
912 if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6) &&
913 (key->eth.type != htons(ETH_P_IPV6) ||
914 sw_flow_key_is_nd(key)))
915 return -EINVAL;
916
917 return 0;
918}