Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
[linux-block.git] / net / openvswitch / flow.c
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/pkt_cls.h>
38 #include <net/netfilter/nf_conntrack_zones.h>
39
40 #include "conntrack.h"
41 #include "datapath.h"
42 #include "flow.h"
43 #include "flow_netlink.h"
44 #include "vport.h"
45
46 u64 ovs_flow_used_time(unsigned long flow_jiffies)
47 {
48         struct timespec64 cur_ts;
49         u64 cur_ms, idle_ms;
50
51         ktime_get_ts64(&cur_ts);
52         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
53         cur_ms = (u64)(u32)cur_ts.tv_sec * MSEC_PER_SEC +
54                  cur_ts.tv_nsec / NSEC_PER_MSEC;
55
56         return cur_ms - idle_ms;
57 }
58
59 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
60
61 void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
62                            const struct sk_buff *skb)
63 {
64         struct sw_flow_stats *stats;
65         unsigned int cpu = smp_processor_id();
66         int len = skb->len + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
67
68         stats = rcu_dereference(flow->stats[cpu]);
69
70         /* Check if already have CPU-specific stats. */
71         if (likely(stats)) {
72                 spin_lock(&stats->lock);
73                 /* Mark if we write on the pre-allocated stats. */
74                 if (cpu == 0 && unlikely(flow->stats_last_writer != cpu))
75                         flow->stats_last_writer = cpu;
76         } else {
77                 stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
78                 spin_lock(&stats->lock);
79
80                 /* If the current CPU is the only writer on the
81                  * pre-allocated stats keep using them.
82                  */
83                 if (unlikely(flow->stats_last_writer != cpu)) {
84                         /* A previous locker may have already allocated the
85                          * stats, so we need to check again.  If CPU-specific
86                          * stats were already allocated, we update the pre-
87                          * allocated stats as we have already locked them.
88                          */
89                         if (likely(flow->stats_last_writer != -1) &&
90                             likely(!rcu_access_pointer(flow->stats[cpu]))) {
91                                 /* Try to allocate CPU-specific stats. */
92                                 struct sw_flow_stats *new_stats;
93
94                                 new_stats =
95                                         kmem_cache_alloc_node(flow_stats_cache,
96                                                               GFP_NOWAIT |
97                                                               __GFP_THISNODE |
98                                                               __GFP_NOWARN |
99                                                               __GFP_NOMEMALLOC,
100                                                               numa_node_id());
101                                 if (likely(new_stats)) {
102                                         new_stats->used = jiffies;
103                                         new_stats->packet_count = 1;
104                                         new_stats->byte_count = len;
105                                         new_stats->tcp_flags = tcp_flags;
106                                         spin_lock_init(&new_stats->lock);
107
108                                         rcu_assign_pointer(flow->stats[cpu],
109                                                            new_stats);
110                                         cpumask_set_cpu(cpu,
111                                                         flow->cpu_used_mask);
112                                         goto unlock;
113                                 }
114                         }
115                         flow->stats_last_writer = cpu;
116                 }
117         }
118
119         stats->used = jiffies;
120         stats->packet_count++;
121         stats->byte_count += len;
122         stats->tcp_flags |= tcp_flags;
123 unlock:
124         spin_unlock(&stats->lock);
125 }
126
127 /* Must be called with rcu_read_lock or ovs_mutex. */
128 void ovs_flow_stats_get(const struct sw_flow *flow,
129                         struct ovs_flow_stats *ovs_stats,
130                         unsigned long *used, __be16 *tcp_flags)
131 {
132         int cpu;
133
134         *used = 0;
135         *tcp_flags = 0;
136         memset(ovs_stats, 0, sizeof(*ovs_stats));
137
138         /* We open code this to make sure cpu 0 is always considered */
139         for (cpu = 0; cpu < nr_cpu_ids;
140              cpu = cpumask_next(cpu, flow->cpu_used_mask)) {
141                 struct sw_flow_stats *stats = rcu_dereference_ovsl(flow->stats[cpu]);
142
143                 if (stats) {
144                         /* Local CPU may write on non-local stats, so we must
145                          * block bottom-halves here.
146                          */
147                         spin_lock_bh(&stats->lock);
148                         if (!*used || time_after(stats->used, *used))
149                                 *used = stats->used;
150                         *tcp_flags |= stats->tcp_flags;
151                         ovs_stats->n_packets += stats->packet_count;
152                         ovs_stats->n_bytes += stats->byte_count;
153                         spin_unlock_bh(&stats->lock);
154                 }
155         }
156 }
157
158 /* Called with ovs_mutex. */
159 void ovs_flow_stats_clear(struct sw_flow *flow)
160 {
161         int cpu;
162
163         /* We open code this to make sure cpu 0 is always considered */
164         for (cpu = 0; cpu < nr_cpu_ids;
165              cpu = cpumask_next(cpu, flow->cpu_used_mask)) {
166                 struct sw_flow_stats *stats = ovsl_dereference(flow->stats[cpu]);
167
168                 if (stats) {
169                         spin_lock_bh(&stats->lock);
170                         stats->used = 0;
171                         stats->packet_count = 0;
172                         stats->byte_count = 0;
173                         stats->tcp_flags = 0;
174                         spin_unlock_bh(&stats->lock);
175                 }
176         }
177 }
178
179 static int check_header(struct sk_buff *skb, int len)
180 {
181         if (unlikely(skb->len < len))
182                 return -EINVAL;
183         if (unlikely(!pskb_may_pull(skb, len)))
184                 return -ENOMEM;
185         return 0;
186 }
187
188 static bool arphdr_ok(struct sk_buff *skb)
189 {
190         return pskb_may_pull(skb, skb_network_offset(skb) +
191                                   sizeof(struct arp_eth_header));
192 }
193
194 static int check_iphdr(struct sk_buff *skb)
195 {
196         unsigned int nh_ofs = skb_network_offset(skb);
197         unsigned int ip_len;
198         int err;
199
200         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
201         if (unlikely(err))
202                 return err;
203
204         ip_len = ip_hdrlen(skb);
205         if (unlikely(ip_len < sizeof(struct iphdr) ||
206                      skb->len < nh_ofs + ip_len))
207                 return -EINVAL;
208
209         skb_set_transport_header(skb, nh_ofs + ip_len);
210         return 0;
211 }
212
213 static bool tcphdr_ok(struct sk_buff *skb)
214 {
215         int th_ofs = skb_transport_offset(skb);
216         int tcp_len;
217
218         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
219                 return false;
220
221         tcp_len = tcp_hdrlen(skb);
222         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
223                      skb->len < th_ofs + tcp_len))
224                 return false;
225
226         return true;
227 }
228
229 static bool udphdr_ok(struct sk_buff *skb)
230 {
231         return pskb_may_pull(skb, skb_transport_offset(skb) +
232                                   sizeof(struct udphdr));
233 }
234
235 static bool sctphdr_ok(struct sk_buff *skb)
236 {
237         return pskb_may_pull(skb, skb_transport_offset(skb) +
238                                   sizeof(struct sctphdr));
239 }
240
241 static bool icmphdr_ok(struct sk_buff *skb)
242 {
243         return pskb_may_pull(skb, skb_transport_offset(skb) +
244                                   sizeof(struct icmphdr));
245 }
246
247 /**
248  * get_ipv6_ext_hdrs() - Parses packet and sets IPv6 extension header flags.
249  *
250  * @skb: buffer where extension header data starts in packet
251  * @nh: ipv6 header
252  * @ext_hdrs: flags are stored here
253  *
254  * OFPIEH12_UNREP is set if more than one of a given IPv6 extension header
255  * is unexpectedly encountered. (Two destination options headers may be
256  * expected and would not cause this bit to be set.)
257  *
258  * OFPIEH12_UNSEQ is set if IPv6 extension headers were not in the order
259  * preferred (but not required) by RFC 2460:
260  *
261  * When more than one extension header is used in the same packet, it is
262  * recommended that those headers appear in the following order:
263  *      IPv6 header
264  *      Hop-by-Hop Options header
265  *      Destination Options header
266  *      Routing header
267  *      Fragment header
268  *      Authentication header
269  *      Encapsulating Security Payload header
270  *      Destination Options header
271  *      upper-layer header
272  */
273 static void get_ipv6_ext_hdrs(struct sk_buff *skb, struct ipv6hdr *nh,
274                               u16 *ext_hdrs)
275 {
276         u8 next_type = nh->nexthdr;
277         unsigned int start = skb_network_offset(skb) + sizeof(struct ipv6hdr);
278         int dest_options_header_count = 0;
279
280         *ext_hdrs = 0;
281
282         while (ipv6_ext_hdr(next_type)) {
283                 struct ipv6_opt_hdr _hdr, *hp;
284
285                 switch (next_type) {
286                 case IPPROTO_NONE:
287                         *ext_hdrs |= OFPIEH12_NONEXT;
288                         /* stop parsing */
289                         return;
290
291                 case IPPROTO_ESP:
292                         if (*ext_hdrs & OFPIEH12_ESP)
293                                 *ext_hdrs |= OFPIEH12_UNREP;
294                         if ((*ext_hdrs & ~(OFPIEH12_HOP | OFPIEH12_DEST |
295                                            OFPIEH12_ROUTER | IPPROTO_FRAGMENT |
296                                            OFPIEH12_AUTH | OFPIEH12_UNREP)) ||
297                             dest_options_header_count >= 2) {
298                                 *ext_hdrs |= OFPIEH12_UNSEQ;
299                         }
300                         *ext_hdrs |= OFPIEH12_ESP;
301                         break;
302
303                 case IPPROTO_AH:
304                         if (*ext_hdrs & OFPIEH12_AUTH)
305                                 *ext_hdrs |= OFPIEH12_UNREP;
306                         if ((*ext_hdrs &
307                              ~(OFPIEH12_HOP | OFPIEH12_DEST | OFPIEH12_ROUTER |
308                                IPPROTO_FRAGMENT | OFPIEH12_UNREP)) ||
309                             dest_options_header_count >= 2) {
310                                 *ext_hdrs |= OFPIEH12_UNSEQ;
311                         }
312                         *ext_hdrs |= OFPIEH12_AUTH;
313                         break;
314
315                 case IPPROTO_DSTOPTS:
316                         if (dest_options_header_count == 0) {
317                                 if (*ext_hdrs &
318                                     ~(OFPIEH12_HOP | OFPIEH12_UNREP))
319                                         *ext_hdrs |= OFPIEH12_UNSEQ;
320                                 *ext_hdrs |= OFPIEH12_DEST;
321                         } else if (dest_options_header_count == 1) {
322                                 if (*ext_hdrs &
323                                     ~(OFPIEH12_HOP | OFPIEH12_DEST |
324                                       OFPIEH12_ROUTER | OFPIEH12_FRAG |
325                                       OFPIEH12_AUTH | OFPIEH12_ESP |
326                                       OFPIEH12_UNREP)) {
327                                         *ext_hdrs |= OFPIEH12_UNSEQ;
328                                 }
329                         } else {
330                                 *ext_hdrs |= OFPIEH12_UNREP;
331                         }
332                         dest_options_header_count++;
333                         break;
334
335                 case IPPROTO_FRAGMENT:
336                         if (*ext_hdrs & OFPIEH12_FRAG)
337                                 *ext_hdrs |= OFPIEH12_UNREP;
338                         if ((*ext_hdrs & ~(OFPIEH12_HOP |
339                                            OFPIEH12_DEST |
340                                            OFPIEH12_ROUTER |
341                                            OFPIEH12_UNREP)) ||
342                             dest_options_header_count >= 2) {
343                                 *ext_hdrs |= OFPIEH12_UNSEQ;
344                         }
345                         *ext_hdrs |= OFPIEH12_FRAG;
346                         break;
347
348                 case IPPROTO_ROUTING:
349                         if (*ext_hdrs & OFPIEH12_ROUTER)
350                                 *ext_hdrs |= OFPIEH12_UNREP;
351                         if ((*ext_hdrs & ~(OFPIEH12_HOP |
352                                            OFPIEH12_DEST |
353                                            OFPIEH12_UNREP)) ||
354                             dest_options_header_count >= 2) {
355                                 *ext_hdrs |= OFPIEH12_UNSEQ;
356                         }
357                         *ext_hdrs |= OFPIEH12_ROUTER;
358                         break;
359
360                 case IPPROTO_HOPOPTS:
361                         if (*ext_hdrs & OFPIEH12_HOP)
362                                 *ext_hdrs |= OFPIEH12_UNREP;
363                         /* OFPIEH12_HOP is set to 1 if a hop-by-hop IPv6
364                          * extension header is present as the first
365                          * extension header in the packet.
366                          */
367                         if (*ext_hdrs == 0)
368                                 *ext_hdrs |= OFPIEH12_HOP;
369                         else
370                                 *ext_hdrs |= OFPIEH12_UNSEQ;
371                         break;
372
373                 default:
374                         return;
375                 }
376
377                 hp = skb_header_pointer(skb, start, sizeof(_hdr), &_hdr);
378                 if (!hp)
379                         break;
380                 next_type = hp->nexthdr;
381                 start += ipv6_optlen(hp);
382         }
383 }
384
385 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
386 {
387         unsigned short frag_off;
388         unsigned int payload_ofs = 0;
389         unsigned int nh_ofs = skb_network_offset(skb);
390         unsigned int nh_len;
391         struct ipv6hdr *nh;
392         int err, nexthdr, flags = 0;
393
394         err = check_header(skb, nh_ofs + sizeof(*nh));
395         if (unlikely(err))
396                 return err;
397
398         nh = ipv6_hdr(skb);
399
400         get_ipv6_ext_hdrs(skb, nh, &key->ipv6.exthdrs);
401
402         key->ip.proto = NEXTHDR_NONE;
403         key->ip.tos = ipv6_get_dsfield(nh);
404         key->ip.ttl = nh->hop_limit;
405         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
406         key->ipv6.addr.src = nh->saddr;
407         key->ipv6.addr.dst = nh->daddr;
408
409         nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
410         if (flags & IP6_FH_F_FRAG) {
411                 if (frag_off) {
412                         key->ip.frag = OVS_FRAG_TYPE_LATER;
413                         key->ip.proto = NEXTHDR_FRAGMENT;
414                         return 0;
415                 }
416                 key->ip.frag = OVS_FRAG_TYPE_FIRST;
417         } else {
418                 key->ip.frag = OVS_FRAG_TYPE_NONE;
419         }
420
421         /* Delayed handling of error in ipv6_find_hdr() as it
422          * always sets flags and frag_off to a valid value which may be
423          * used to set key->ip.frag above.
424          */
425         if (unlikely(nexthdr < 0))
426                 return -EPROTO;
427
428         nh_len = payload_ofs - nh_ofs;
429         skb_set_transport_header(skb, nh_ofs + nh_len);
430         key->ip.proto = nexthdr;
431         return nh_len;
432 }
433
434 static bool icmp6hdr_ok(struct sk_buff *skb)
435 {
436         return pskb_may_pull(skb, skb_transport_offset(skb) +
437                                   sizeof(struct icmp6hdr));
438 }
439
440 /**
441  * parse_vlan_tag - Parse vlan tag from vlan header.
442  * @skb: skb containing frame to parse
443  * @key_vh: pointer to parsed vlan tag
444  * @untag_vlan: should the vlan header be removed from the frame
445  *
446  * Return: ERROR on memory error.
447  * %0 if it encounters a non-vlan or incomplete packet.
448  * %1 after successfully parsing vlan tag.
449  */
450 static int parse_vlan_tag(struct sk_buff *skb, struct vlan_head *key_vh,
451                           bool untag_vlan)
452 {
453         struct vlan_head *vh = (struct vlan_head *)skb->data;
454
455         if (likely(!eth_type_vlan(vh->tpid)))
456                 return 0;
457
458         if (unlikely(skb->len < sizeof(struct vlan_head) + sizeof(__be16)))
459                 return 0;
460
461         if (unlikely(!pskb_may_pull(skb, sizeof(struct vlan_head) +
462                                  sizeof(__be16))))
463                 return -ENOMEM;
464
465         vh = (struct vlan_head *)skb->data;
466         key_vh->tci = vh->tci | htons(VLAN_CFI_MASK);
467         key_vh->tpid = vh->tpid;
468
469         if (unlikely(untag_vlan)) {
470                 int offset = skb->data - skb_mac_header(skb);
471                 u16 tci;
472                 int err;
473
474                 __skb_push(skb, offset);
475                 err = __skb_vlan_pop(skb, &tci);
476                 __skb_pull(skb, offset);
477                 if (err)
478                         return err;
479                 __vlan_hwaccel_put_tag(skb, key_vh->tpid, tci);
480         } else {
481                 __skb_pull(skb, sizeof(struct vlan_head));
482         }
483         return 1;
484 }
485
486 static void clear_vlan(struct sw_flow_key *key)
487 {
488         key->eth.vlan.tci = 0;
489         key->eth.vlan.tpid = 0;
490         key->eth.cvlan.tci = 0;
491         key->eth.cvlan.tpid = 0;
492 }
493
494 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
495 {
496         int res;
497
498         if (skb_vlan_tag_present(skb)) {
499                 key->eth.vlan.tci = htons(skb->vlan_tci) | htons(VLAN_CFI_MASK);
500                 key->eth.vlan.tpid = skb->vlan_proto;
501         } else {
502                 /* Parse outer vlan tag in the non-accelerated case. */
503                 res = parse_vlan_tag(skb, &key->eth.vlan, true);
504                 if (res <= 0)
505                         return res;
506         }
507
508         /* Parse inner vlan tag. */
509         res = parse_vlan_tag(skb, &key->eth.cvlan, false);
510         if (res <= 0)
511                 return res;
512
513         return 0;
514 }
515
516 static __be16 parse_ethertype(struct sk_buff *skb)
517 {
518         struct llc_snap_hdr {
519                 u8  dsap;  /* Always 0xAA */
520                 u8  ssap;  /* Always 0xAA */
521                 u8  ctrl;
522                 u8  oui[3];
523                 __be16 ethertype;
524         };
525         struct llc_snap_hdr *llc;
526         __be16 proto;
527
528         proto = *(__be16 *) skb->data;
529         __skb_pull(skb, sizeof(__be16));
530
531         if (eth_proto_is_802_3(proto))
532                 return proto;
533
534         if (skb->len < sizeof(struct llc_snap_hdr))
535                 return htons(ETH_P_802_2);
536
537         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
538                 return htons(0);
539
540         llc = (struct llc_snap_hdr *) skb->data;
541         if (llc->dsap != LLC_SAP_SNAP ||
542             llc->ssap != LLC_SAP_SNAP ||
543             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
544                 return htons(ETH_P_802_2);
545
546         __skb_pull(skb, sizeof(struct llc_snap_hdr));
547
548         if (eth_proto_is_802_3(llc->ethertype))
549                 return llc->ethertype;
550
551         return htons(ETH_P_802_2);
552 }
553
554 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
555                         int nh_len)
556 {
557         struct icmp6hdr *icmp = icmp6_hdr(skb);
558
559         /* The ICMPv6 type and code fields use the 16-bit transport port
560          * fields, so we need to store them in 16-bit network byte order.
561          */
562         key->tp.src = htons(icmp->icmp6_type);
563         key->tp.dst = htons(icmp->icmp6_code);
564
565         if (icmp->icmp6_code == 0 &&
566             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
567              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
568                 int icmp_len = skb->len - skb_transport_offset(skb);
569                 struct nd_msg *nd;
570                 int offset;
571
572                 memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
573
574                 /* In order to process neighbor discovery options, we need the
575                  * entire packet.
576                  */
577                 if (unlikely(icmp_len < sizeof(*nd)))
578                         return 0;
579
580                 if (unlikely(skb_linearize(skb)))
581                         return -ENOMEM;
582
583                 nd = (struct nd_msg *)skb_transport_header(skb);
584                 key->ipv6.nd.target = nd->target;
585
586                 icmp_len -= sizeof(*nd);
587                 offset = 0;
588                 while (icmp_len >= 8) {
589                         struct nd_opt_hdr *nd_opt =
590                                  (struct nd_opt_hdr *)(nd->opt + offset);
591                         int opt_len = nd_opt->nd_opt_len * 8;
592
593                         if (unlikely(!opt_len || opt_len > icmp_len))
594                                 return 0;
595
596                         /* Store the link layer address if the appropriate
597                          * option is provided.  It is considered an error if
598                          * the same link layer option is specified twice.
599                          */
600                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
601                             && opt_len == 8) {
602                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
603                                         goto invalid;
604                                 ether_addr_copy(key->ipv6.nd.sll,
605                                                 &nd->opt[offset+sizeof(*nd_opt)]);
606                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
607                                    && opt_len == 8) {
608                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
609                                         goto invalid;
610                                 ether_addr_copy(key->ipv6.nd.tll,
611                                                 &nd->opt[offset+sizeof(*nd_opt)]);
612                         }
613
614                         icmp_len -= opt_len;
615                         offset += opt_len;
616                 }
617         }
618
619         return 0;
620
621 invalid:
622         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
623         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
624         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
625
626         return 0;
627 }
628
629 static int parse_nsh(struct sk_buff *skb, struct sw_flow_key *key)
630 {
631         struct nshhdr *nh;
632         unsigned int nh_ofs = skb_network_offset(skb);
633         u8 version, length;
634         int err;
635
636         err = check_header(skb, nh_ofs + NSH_BASE_HDR_LEN);
637         if (unlikely(err))
638                 return err;
639
640         nh = nsh_hdr(skb);
641         version = nsh_get_ver(nh);
642         length = nsh_hdr_len(nh);
643
644         if (version != 0)
645                 return -EINVAL;
646
647         err = check_header(skb, nh_ofs + length);
648         if (unlikely(err))
649                 return err;
650
651         nh = nsh_hdr(skb);
652         key->nsh.base.flags = nsh_get_flags(nh);
653         key->nsh.base.ttl = nsh_get_ttl(nh);
654         key->nsh.base.mdtype = nh->mdtype;
655         key->nsh.base.np = nh->np;
656         key->nsh.base.path_hdr = nh->path_hdr;
657         switch (key->nsh.base.mdtype) {
658         case NSH_M_TYPE1:
659                 if (length != NSH_M_TYPE1_LEN)
660                         return -EINVAL;
661                 memcpy(key->nsh.context, nh->md1.context,
662                        sizeof(nh->md1));
663                 break;
664         case NSH_M_TYPE2:
665                 memset(key->nsh.context, 0,
666                        sizeof(nh->md1));
667                 break;
668         default:
669                 return -EINVAL;
670         }
671
672         return 0;
673 }
674
675 /**
676  * key_extract_l3l4 - extracts L3/L4 header information.
677  * @skb: sk_buff that contains the frame, with skb->data pointing to the
678  *       L3 header
679  * @key: output flow key
680  *
681  * Return: %0 if successful, otherwise a negative errno value.
682  */
683 static int key_extract_l3l4(struct sk_buff *skb, struct sw_flow_key *key)
684 {
685         int error;
686
687         /* Network layer. */
688         if (key->eth.type == htons(ETH_P_IP)) {
689                 struct iphdr *nh;
690                 __be16 offset;
691
692                 error = check_iphdr(skb);
693                 if (unlikely(error)) {
694                         memset(&key->ip, 0, sizeof(key->ip));
695                         memset(&key->ipv4, 0, sizeof(key->ipv4));
696                         if (error == -EINVAL) {
697                                 skb->transport_header = skb->network_header;
698                                 error = 0;
699                         }
700                         return error;
701                 }
702
703                 nh = ip_hdr(skb);
704                 key->ipv4.addr.src = nh->saddr;
705                 key->ipv4.addr.dst = nh->daddr;
706
707                 key->ip.proto = nh->protocol;
708                 key->ip.tos = nh->tos;
709                 key->ip.ttl = nh->ttl;
710
711                 offset = nh->frag_off & htons(IP_OFFSET);
712                 if (offset) {
713                         key->ip.frag = OVS_FRAG_TYPE_LATER;
714                         memset(&key->tp, 0, sizeof(key->tp));
715                         return 0;
716                 }
717                 if (nh->frag_off & htons(IP_MF) ||
718                         skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
719                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
720                 else
721                         key->ip.frag = OVS_FRAG_TYPE_NONE;
722
723                 /* Transport layer. */
724                 if (key->ip.proto == IPPROTO_TCP) {
725                         if (tcphdr_ok(skb)) {
726                                 struct tcphdr *tcp = tcp_hdr(skb);
727                                 key->tp.src = tcp->source;
728                                 key->tp.dst = tcp->dest;
729                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
730                         } else {
731                                 memset(&key->tp, 0, sizeof(key->tp));
732                         }
733
734                 } else if (key->ip.proto == IPPROTO_UDP) {
735                         if (udphdr_ok(skb)) {
736                                 struct udphdr *udp = udp_hdr(skb);
737                                 key->tp.src = udp->source;
738                                 key->tp.dst = udp->dest;
739                         } else {
740                                 memset(&key->tp, 0, sizeof(key->tp));
741                         }
742                 } else if (key->ip.proto == IPPROTO_SCTP) {
743                         if (sctphdr_ok(skb)) {
744                                 struct sctphdr *sctp = sctp_hdr(skb);
745                                 key->tp.src = sctp->source;
746                                 key->tp.dst = sctp->dest;
747                         } else {
748                                 memset(&key->tp, 0, sizeof(key->tp));
749                         }
750                 } else if (key->ip.proto == IPPROTO_ICMP) {
751                         if (icmphdr_ok(skb)) {
752                                 struct icmphdr *icmp = icmp_hdr(skb);
753                                 /* The ICMP type and code fields use the 16-bit
754                                  * transport port fields, so we need to store
755                                  * them in 16-bit network byte order. */
756                                 key->tp.src = htons(icmp->type);
757                                 key->tp.dst = htons(icmp->code);
758                         } else {
759                                 memset(&key->tp, 0, sizeof(key->tp));
760                         }
761                 }
762
763         } else if (key->eth.type == htons(ETH_P_ARP) ||
764                    key->eth.type == htons(ETH_P_RARP)) {
765                 struct arp_eth_header *arp;
766                 bool arp_available = arphdr_ok(skb);
767
768                 arp = (struct arp_eth_header *)skb_network_header(skb);
769
770                 if (arp_available &&
771                     arp->ar_hrd == htons(ARPHRD_ETHER) &&
772                     arp->ar_pro == htons(ETH_P_IP) &&
773                     arp->ar_hln == ETH_ALEN &&
774                     arp->ar_pln == 4) {
775
776                         /* We only match on the lower 8 bits of the opcode. */
777                         if (ntohs(arp->ar_op) <= 0xff)
778                                 key->ip.proto = ntohs(arp->ar_op);
779                         else
780                                 key->ip.proto = 0;
781
782                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
783                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
784                         ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
785                         ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
786                 } else {
787                         memset(&key->ip, 0, sizeof(key->ip));
788                         memset(&key->ipv4, 0, sizeof(key->ipv4));
789                 }
790         } else if (eth_p_mpls(key->eth.type)) {
791                 u8 label_count = 1;
792
793                 memset(&key->mpls, 0, sizeof(key->mpls));
794                 skb_set_inner_network_header(skb, skb->mac_len);
795                 while (1) {
796                         __be32 lse;
797
798                         error = check_header(skb, skb->mac_len +
799                                              label_count * MPLS_HLEN);
800                         if (unlikely(error))
801                                 return 0;
802
803                         memcpy(&lse, skb_inner_network_header(skb), MPLS_HLEN);
804
805                         if (label_count <= MPLS_LABEL_DEPTH)
806                                 memcpy(&key->mpls.lse[label_count - 1], &lse,
807                                        MPLS_HLEN);
808
809                         skb_set_inner_network_header(skb, skb->mac_len +
810                                                      label_count * MPLS_HLEN);
811                         if (lse & htonl(MPLS_LS_S_MASK))
812                                 break;
813
814                         label_count++;
815                 }
816                 if (label_count > MPLS_LABEL_DEPTH)
817                         label_count = MPLS_LABEL_DEPTH;
818
819                 key->mpls.num_labels_mask = GENMASK(label_count - 1, 0);
820         } else if (key->eth.type == htons(ETH_P_IPV6)) {
821                 int nh_len;             /* IPv6 Header + Extensions */
822
823                 nh_len = parse_ipv6hdr(skb, key);
824                 if (unlikely(nh_len < 0)) {
825                         switch (nh_len) {
826                         case -EINVAL:
827                                 memset(&key->ip, 0, sizeof(key->ip));
828                                 memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
829                                 fallthrough;
830                         case -EPROTO:
831                                 skb->transport_header = skb->network_header;
832                                 error = 0;
833                                 break;
834                         default:
835                                 error = nh_len;
836                         }
837                         return error;
838                 }
839
840                 if (key->ip.frag == OVS_FRAG_TYPE_LATER) {
841                         memset(&key->tp, 0, sizeof(key->tp));
842                         return 0;
843                 }
844                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
845                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
846
847                 /* Transport layer. */
848                 if (key->ip.proto == NEXTHDR_TCP) {
849                         if (tcphdr_ok(skb)) {
850                                 struct tcphdr *tcp = tcp_hdr(skb);
851                                 key->tp.src = tcp->source;
852                                 key->tp.dst = tcp->dest;
853                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
854                         } else {
855                                 memset(&key->tp, 0, sizeof(key->tp));
856                         }
857                 } else if (key->ip.proto == NEXTHDR_UDP) {
858                         if (udphdr_ok(skb)) {
859                                 struct udphdr *udp = udp_hdr(skb);
860                                 key->tp.src = udp->source;
861                                 key->tp.dst = udp->dest;
862                         } else {
863                                 memset(&key->tp, 0, sizeof(key->tp));
864                         }
865                 } else if (key->ip.proto == NEXTHDR_SCTP) {
866                         if (sctphdr_ok(skb)) {
867                                 struct sctphdr *sctp = sctp_hdr(skb);
868                                 key->tp.src = sctp->source;
869                                 key->tp.dst = sctp->dest;
870                         } else {
871                                 memset(&key->tp, 0, sizeof(key->tp));
872                         }
873                 } else if (key->ip.proto == NEXTHDR_ICMP) {
874                         if (icmp6hdr_ok(skb)) {
875                                 error = parse_icmpv6(skb, key, nh_len);
876                                 if (error)
877                                         return error;
878                         } else {
879                                 memset(&key->tp, 0, sizeof(key->tp));
880                         }
881                 }
882         } else if (key->eth.type == htons(ETH_P_NSH)) {
883                 error = parse_nsh(skb, key);
884                 if (error)
885                         return error;
886         }
887         return 0;
888 }
889
890 /**
891  * key_extract - extracts a flow key from an Ethernet frame.
892  * @skb: sk_buff that contains the frame, with skb->data pointing to the
893  * Ethernet header
894  * @key: output flow key
895  *
896  * The caller must ensure that skb->len >= ETH_HLEN.
897  *
898  * Initializes @skb header fields as follows:
899  *
900  *    - skb->mac_header: the L2 header.
901  *
902  *    - skb->network_header: just past the L2 header, or just past the
903  *      VLAN header, to the first byte of the L2 payload.
904  *
905  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
906  *      on output, then just past the IP header, if one is present and
907  *      of a correct length, otherwise the same as skb->network_header.
908  *      For other key->eth.type values it is left untouched.
909  *
910  *    - skb->protocol: the type of the data starting at skb->network_header.
911  *      Equals to key->eth.type.
912  *
913  * Return: %0 if successful, otherwise a negative errno value.
914  */
915 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
916 {
917         struct ethhdr *eth;
918
919         /* Flags are always used as part of stats */
920         key->tp.flags = 0;
921
922         skb_reset_mac_header(skb);
923
924         /* Link layer. */
925         clear_vlan(key);
926         if (ovs_key_mac_proto(key) == MAC_PROTO_NONE) {
927                 if (unlikely(eth_type_vlan(skb->protocol)))
928                         return -EINVAL;
929
930                 skb_reset_network_header(skb);
931                 key->eth.type = skb->protocol;
932         } else {
933                 eth = eth_hdr(skb);
934                 ether_addr_copy(key->eth.src, eth->h_source);
935                 ether_addr_copy(key->eth.dst, eth->h_dest);
936
937                 __skb_pull(skb, 2 * ETH_ALEN);
938                 /* We are going to push all headers that we pull, so no need to
939                  * update skb->csum here.
940                  */
941
942                 if (unlikely(parse_vlan(skb, key)))
943                         return -ENOMEM;
944
945                 key->eth.type = parse_ethertype(skb);
946                 if (unlikely(key->eth.type == htons(0)))
947                         return -ENOMEM;
948
949                 /* Multiple tagged packets need to retain TPID to satisfy
950                  * skb_vlan_pop(), which will later shift the ethertype into
951                  * skb->protocol.
952                  */
953                 if (key->eth.cvlan.tci & htons(VLAN_CFI_MASK))
954                         skb->protocol = key->eth.cvlan.tpid;
955                 else
956                         skb->protocol = key->eth.type;
957
958                 skb_reset_network_header(skb);
959                 __skb_push(skb, skb->data - skb_mac_header(skb));
960         }
961
962         skb_reset_mac_len(skb);
963
964         /* Fill out L3/L4 key info, if any */
965         return key_extract_l3l4(skb, key);
966 }
967
968 /* In the case of conntrack fragment handling it expects L3 headers,
969  * add a helper.
970  */
971 int ovs_flow_key_update_l3l4(struct sk_buff *skb, struct sw_flow_key *key)
972 {
973         return key_extract_l3l4(skb, key);
974 }
975
976 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
977 {
978         int res;
979
980         res = key_extract(skb, key);
981         if (!res)
982                 key->mac_proto &= ~SW_FLOW_KEY_INVALID;
983
984         return res;
985 }
986
987 static int key_extract_mac_proto(struct sk_buff *skb)
988 {
989         switch (skb->dev->type) {
990         case ARPHRD_ETHER:
991                 return MAC_PROTO_ETHERNET;
992         case ARPHRD_NONE:
993                 if (skb->protocol == htons(ETH_P_TEB))
994                         return MAC_PROTO_ETHERNET;
995                 return MAC_PROTO_NONE;
996         }
997         WARN_ON_ONCE(1);
998         return -EINVAL;
999 }
1000
1001 int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
1002                          struct sk_buff *skb, struct sw_flow_key *key)
1003 {
1004 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1005         struct tc_skb_ext *tc_ext;
1006 #endif
1007         bool post_ct = false, post_ct_snat = false, post_ct_dnat = false;
1008         int res, err;
1009         u16 zone = 0;
1010
1011         /* Extract metadata from packet. */
1012         if (tun_info) {
1013                 key->tun_proto = ip_tunnel_info_af(tun_info);
1014                 memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
1015
1016                 if (tun_info->options_len) {
1017                         BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) *
1018                                                    8)) - 1
1019                                         > sizeof(key->tun_opts));
1020
1021                         ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key, tun_info->options_len),
1022                                                 tun_info);
1023                         key->tun_opts_len = tun_info->options_len;
1024                 } else {
1025                         key->tun_opts_len = 0;
1026                 }
1027         } else  {
1028                 key->tun_proto = 0;
1029                 key->tun_opts_len = 0;
1030                 memset(&key->tun_key, 0, sizeof(key->tun_key));
1031         }
1032
1033         key->phy.priority = skb->priority;
1034         key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
1035         key->phy.skb_mark = skb->mark;
1036         key->ovs_flow_hash = 0;
1037         res = key_extract_mac_proto(skb);
1038         if (res < 0)
1039                 return res;
1040         key->mac_proto = res;
1041
1042 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1043         if (tc_skb_ext_tc_enabled()) {
1044                 tc_ext = skb_ext_find(skb, TC_SKB_EXT);
1045                 key->recirc_id = tc_ext && !tc_ext->act_miss ?
1046                                  tc_ext->chain : 0;
1047                 OVS_CB(skb)->mru = tc_ext ? tc_ext->mru : 0;
1048                 post_ct = tc_ext ? tc_ext->post_ct : false;
1049                 post_ct_snat = post_ct ? tc_ext->post_ct_snat : false;
1050                 post_ct_dnat = post_ct ? tc_ext->post_ct_dnat : false;
1051                 zone = post_ct ? tc_ext->zone : 0;
1052         } else {
1053                 key->recirc_id = 0;
1054         }
1055 #else
1056         key->recirc_id = 0;
1057 #endif
1058
1059         err = key_extract(skb, key);
1060         if (!err) {
1061                 ovs_ct_fill_key(skb, key, post_ct);   /* Must be after key_extract(). */
1062                 if (post_ct) {
1063                         if (!skb_get_nfct(skb)) {
1064                                 key->ct_zone = zone;
1065                         } else {
1066                                 if (!post_ct_dnat)
1067                                         key->ct_state &= ~OVS_CS_F_DST_NAT;
1068                                 if (!post_ct_snat)
1069                                         key->ct_state &= ~OVS_CS_F_SRC_NAT;
1070                         }
1071                 }
1072         }
1073         return err;
1074 }
1075
1076 int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
1077                                    struct sk_buff *skb,
1078                                    struct sw_flow_key *key, bool log)
1079 {
1080         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1081         u64 attrs = 0;
1082         int err;
1083
1084         err = parse_flow_nlattrs(attr, a, &attrs, log);
1085         if (err)
1086                 return -EINVAL;
1087
1088         /* Extract metadata from netlink attributes. */
1089         err = ovs_nla_get_flow_metadata(net, a, attrs, key, log);
1090         if (err)
1091                 return err;
1092
1093         /* key_extract assumes that skb->protocol is set-up for
1094          * layer 3 packets which is the case for other callers,
1095          * in particular packets received from the network stack.
1096          * Here the correct value can be set from the metadata
1097          * extracted above.
1098          * For L2 packet key eth type would be zero. skb protocol
1099          * would be set to correct value later during key-extact.
1100          */
1101
1102         skb->protocol = key->eth.type;
1103         err = key_extract(skb, key);
1104         if (err)
1105                 return err;
1106
1107         /* Check that we have conntrack original direction tuple metadata only
1108          * for packets for which it makes sense.  Otherwise the key may be
1109          * corrupted due to overlapping key fields.
1110          */
1111         if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4) &&
1112             key->eth.type != htons(ETH_P_IP))
1113                 return -EINVAL;
1114         if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6) &&
1115             (key->eth.type != htons(ETH_P_IPV6) ||
1116              sw_flow_key_is_nd(key)))
1117                 return -EINVAL;
1118
1119         return 0;
1120 }