1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * SR-IPv6 implementation
4 *
5 * Author:
6 * David Lebrun <david.lebrun@uclouvain.be>
7 */
8
9 #include <linux/types.h>
10 #include <linux/skbuff.h>
11 #include <linux/net.h>
12 #include <linux/module.h>
13 #include <net/ip.h>
14 #include <net/ip_tunnels.h>
15 #include <net/lwtunnel.h>
16 #include <net/netevent.h>
17 #include <net/netns/generic.h>
18 #include <net/ip6_fib.h>
19 #include <net/route.h>
20 #include <net/seg6.h>
21 #include <linux/seg6.h>
22 #include <linux/seg6_iptunnel.h>
23 #include <net/addrconf.h>
24 #include <net/ip6_route.h>
25 #include <net/dst_cache.h>
26 #ifdef CONFIG_IPV6_SEG6_HMAC
27 #include <net/seg6_hmac.h>
28 #endif
29 #include <linux/netfilter.h>
30
seg6_lwt_headroom(struct seg6_iptunnel_encap * tuninfo)31 static size_t seg6_lwt_headroom(struct seg6_iptunnel_encap *tuninfo)
32 {
33 int head = 0;
34
35 switch (tuninfo->mode) {
36 case SEG6_IPTUN_MODE_INLINE:
37 break;
38 case SEG6_IPTUN_MODE_ENCAP:
39 head = sizeof(struct ipv6hdr);
40 break;
41 case SEG6_IPTUN_MODE_L2ENCAP:
42 return 0;
43 }
44
45 return ((tuninfo->srh->hdrlen + 1) << 3) + head;
46 }
47
48 struct seg6_lwt {
49 struct dst_cache cache;
50 struct seg6_iptunnel_encap tuninfo[];
51 };
52
seg6_lwt_lwtunnel(struct lwtunnel_state * lwt)53 static inline struct seg6_lwt *seg6_lwt_lwtunnel(struct lwtunnel_state *lwt)
54 {
55 return (struct seg6_lwt *)lwt->data;
56 }
57
58 static inline struct seg6_iptunnel_encap *
seg6_encap_lwtunnel(struct lwtunnel_state * lwt)59 seg6_encap_lwtunnel(struct lwtunnel_state *lwt)
60 {
61 return seg6_lwt_lwtunnel(lwt)->tuninfo;
62 }
63
64 static const struct nla_policy seg6_iptunnel_policy[SEG6_IPTUNNEL_MAX + 1] = {
65 [SEG6_IPTUNNEL_SRH] = { .type = NLA_BINARY },
66 };
67
nla_put_srh(struct sk_buff * skb,int attrtype,struct seg6_iptunnel_encap * tuninfo)68 static int nla_put_srh(struct sk_buff *skb, int attrtype,
69 struct seg6_iptunnel_encap *tuninfo)
70 {
71 struct seg6_iptunnel_encap *data;
72 struct nlattr *nla;
73 int len;
74
75 len = SEG6_IPTUN_ENCAP_SIZE(tuninfo);
76
77 nla = nla_reserve(skb, attrtype, len);
78 if (!nla)
79 return -EMSGSIZE;
80
81 data = nla_data(nla);
82 memcpy(data, tuninfo, len);
83
84 return 0;
85 }
86
set_tun_src(struct net * net,struct net_device * dev,struct in6_addr * daddr,struct in6_addr * saddr)87 static void set_tun_src(struct net *net, struct net_device *dev,
88 struct in6_addr *daddr, struct in6_addr *saddr)
89 {
90 struct seg6_pernet_data *sdata = seg6_pernet(net);
91 struct in6_addr *tun_src;
92
93 rcu_read_lock();
94
95 tun_src = rcu_dereference(sdata->tun_src);
96
97 if (!ipv6_addr_any(tun_src)) {
98 memcpy(saddr, tun_src, sizeof(struct in6_addr));
99 } else {
100 ipv6_dev_get_saddr(net, dev, daddr, IPV6_PREFER_SRC_PUBLIC,
101 saddr);
102 }
103
104 rcu_read_unlock();
105 }
106
107 /* Compute flowlabel for outer IPv6 header */
seg6_make_flowlabel(struct net * net,struct sk_buff * skb,struct ipv6hdr * inner_hdr)108 static __be32 seg6_make_flowlabel(struct net *net, struct sk_buff *skb,
109 struct ipv6hdr *inner_hdr)
110 {
111 int do_flowlabel = net->ipv6.sysctl.seg6_flowlabel;
112 __be32 flowlabel = 0;
113 u32 hash;
114
115 if (do_flowlabel > 0) {
116 hash = skb_get_hash(skb);
117 hash = rol32(hash, 16);
118 flowlabel = (__force __be32)hash & IPV6_FLOWLABEL_MASK;
119 } else if (!do_flowlabel && skb->protocol == htons(ETH_P_IPV6)) {
120 flowlabel = ip6_flowlabel(inner_hdr);
121 }
122 return flowlabel;
123 }
124
125 /* encapsulate an IPv6 packet within an outer IPv6 header with a given SRH */
seg6_do_srh_encap(struct sk_buff * skb,struct ipv6_sr_hdr * osrh,int proto)126 int seg6_do_srh_encap(struct sk_buff *skb, struct ipv6_sr_hdr *osrh, int proto)
127 {
128 struct dst_entry *dst = skb_dst(skb);
129 struct net *net = dev_net(dst->dev);
130 struct ipv6hdr *hdr, *inner_hdr;
131 struct ipv6_sr_hdr *isrh;
132 int hdrlen, tot_len, err;
133 __be32 flowlabel;
134
135 hdrlen = (osrh->hdrlen + 1) << 3;
136 tot_len = hdrlen + sizeof(*hdr);
137
138 err = skb_cow_head(skb, tot_len + skb->mac_len);
139 if (unlikely(err))
140 return err;
141
142 inner_hdr = ipv6_hdr(skb);
143 flowlabel = seg6_make_flowlabel(net, skb, inner_hdr);
144
145 skb_push(skb, tot_len);
146 skb_reset_network_header(skb);
147 skb_mac_header_rebuild(skb);
148 hdr = ipv6_hdr(skb);
149
150 /* inherit tc, flowlabel and hlim
151 * hlim will be decremented in ip6_forward() afterwards and
152 * decapsulation will overwrite inner hlim with outer hlim
153 */
154
155 if (skb->protocol == htons(ETH_P_IPV6)) {
156 ip6_flow_hdr(hdr, ip6_tclass(ip6_flowinfo(inner_hdr)),
157 flowlabel);
158 hdr->hop_limit = inner_hdr->hop_limit;
159 } else {
160 ip6_flow_hdr(hdr, 0, flowlabel);
161 hdr->hop_limit = ip6_dst_hoplimit(skb_dst(skb));
162
163 memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
164
165 /* the control block has been erased, so we have to set the
166 * iif once again.
167 * We read the receiving interface index directly from the
168 * skb->skb_iif as it is done in the IPv4 receiving path (i.e.:
169 * ip_rcv_core(...)).
170 */
171 IP6CB(skb)->iif = skb->skb_iif;
172 }
173
174 hdr->nexthdr = NEXTHDR_ROUTING;
175
176 isrh = (void *)hdr + sizeof(*hdr);
177 memcpy(isrh, osrh, hdrlen);
178
179 isrh->nexthdr = proto;
180
181 hdr->daddr = isrh->segments[isrh->first_segment];
182 set_tun_src(net, dst->dev, &hdr->daddr, &hdr->saddr);
183
184 #ifdef CONFIG_IPV6_SEG6_HMAC
185 if (sr_has_hmac(isrh)) {
186 err = seg6_push_hmac(net, &hdr->saddr, isrh);
187 if (unlikely(err))
188 return err;
189 }
190 #endif
191
192 skb_postpush_rcsum(skb, hdr, tot_len);
193
194 return 0;
195 }
196 EXPORT_SYMBOL_GPL(seg6_do_srh_encap);
197
198 /* insert an SRH within an IPv6 packet, just after the IPv6 header */
seg6_do_srh_inline(struct sk_buff * skb,struct ipv6_sr_hdr * osrh)199 int seg6_do_srh_inline(struct sk_buff *skb, struct ipv6_sr_hdr *osrh)
200 {
201 struct ipv6hdr *hdr, *oldhdr;
202 struct ipv6_sr_hdr *isrh;
203 int hdrlen, err;
204
205 hdrlen = (osrh->hdrlen + 1) << 3;
206
207 err = skb_cow_head(skb, hdrlen + skb->mac_len);
208 if (unlikely(err))
209 return err;
210
211 oldhdr = ipv6_hdr(skb);
212
213 skb_pull(skb, sizeof(struct ipv6hdr));
214 skb_postpull_rcsum(skb, skb_network_header(skb),
215 sizeof(struct ipv6hdr));
216
217 skb_push(skb, sizeof(struct ipv6hdr) + hdrlen);
218 skb_reset_network_header(skb);
219 skb_mac_header_rebuild(skb);
220
221 hdr = ipv6_hdr(skb);
222
223 memmove(hdr, oldhdr, sizeof(*hdr));
224
225 isrh = (void *)hdr + sizeof(*hdr);
226 memcpy(isrh, osrh, hdrlen);
227
228 isrh->nexthdr = hdr->nexthdr;
229 hdr->nexthdr = NEXTHDR_ROUTING;
230
231 isrh->segments[0] = hdr->daddr;
232 hdr->daddr = isrh->segments[isrh->first_segment];
233
234 #ifdef CONFIG_IPV6_SEG6_HMAC
235 if (sr_has_hmac(isrh)) {
236 struct net *net = dev_net(skb_dst(skb)->dev);
237
238 err = seg6_push_hmac(net, &hdr->saddr, isrh);
239 if (unlikely(err))
240 return err;
241 }
242 #endif
243
244 skb_postpush_rcsum(skb, hdr, sizeof(struct ipv6hdr) + hdrlen);
245
246 return 0;
247 }
248 EXPORT_SYMBOL_GPL(seg6_do_srh_inline);
249
seg6_do_srh(struct sk_buff * skb)250 static int seg6_do_srh(struct sk_buff *skb)
251 {
252 struct dst_entry *dst = skb_dst(skb);
253 struct seg6_iptunnel_encap *tinfo;
254 int proto, err = 0;
255
256 tinfo = seg6_encap_lwtunnel(dst->lwtstate);
257
258 switch (tinfo->mode) {
259 case SEG6_IPTUN_MODE_INLINE:
260 if (skb->protocol != htons(ETH_P_IPV6))
261 return -EINVAL;
262
263 err = seg6_do_srh_inline(skb, tinfo->srh);
264 if (err)
265 return err;
266 break;
267 case SEG6_IPTUN_MODE_ENCAP:
268 err = iptunnel_handle_offloads(skb, SKB_GSO_IPXIP6);
269 if (err)
270 return err;
271
272 if (skb->protocol == htons(ETH_P_IPV6))
273 proto = IPPROTO_IPV6;
274 else if (skb->protocol == htons(ETH_P_IP))
275 proto = IPPROTO_IPIP;
276 else
277 return -EINVAL;
278
279 err = seg6_do_srh_encap(skb, tinfo->srh, proto);
280 if (err)
281 return err;
282
283 skb_set_inner_transport_header(skb, skb_transport_offset(skb));
284 skb_set_inner_protocol(skb, skb->protocol);
285 skb->protocol = htons(ETH_P_IPV6);
286 break;
287 case SEG6_IPTUN_MODE_L2ENCAP:
288 if (!skb_mac_header_was_set(skb))
289 return -EINVAL;
290
291 if (pskb_expand_head(skb, skb->mac_len, 0, GFP_ATOMIC) < 0)
292 return -ENOMEM;
293
294 skb_mac_header_rebuild(skb);
295 skb_push(skb, skb->mac_len);
296
297 err = seg6_do_srh_encap(skb, tinfo->srh, IPPROTO_ETHERNET);
298 if (err)
299 return err;
300
301 skb->protocol = htons(ETH_P_IPV6);
302 break;
303 }
304
305 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
306 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
307 nf_reset_ct(skb);
308
309 return 0;
310 }
311
seg6_input_finish(struct net * net,struct sock * sk,struct sk_buff * skb)312 static int seg6_input_finish(struct net *net, struct sock *sk,
313 struct sk_buff *skb)
314 {
315 return dst_input(skb);
316 }
317
seg6_input_core(struct net * net,struct sock * sk,struct sk_buff * skb)318 static int seg6_input_core(struct net *net, struct sock *sk,
319 struct sk_buff *skb)
320 {
321 struct dst_entry *orig_dst = skb_dst(skb);
322 struct dst_entry *dst = NULL;
323 struct seg6_lwt *slwt;
324 int err;
325
326 err = seg6_do_srh(skb);
327 if (unlikely(err)) {
328 kfree_skb(skb);
329 return err;
330 }
331
332 slwt = seg6_lwt_lwtunnel(orig_dst->lwtstate);
333
334 preempt_disable();
335 dst = dst_cache_get(&slwt->cache);
336 preempt_enable();
337
338 skb_dst_drop(skb);
339
340 if (!dst) {
341 ip6_route_input(skb);
342 dst = skb_dst(skb);
343 if (!dst->error) {
344 preempt_disable();
345 dst_cache_set_ip6(&slwt->cache, dst,
346 &ipv6_hdr(skb)->saddr);
347 preempt_enable();
348 }
349 } else {
350 skb_dst_set(skb, dst);
351 }
352
353 err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
354 if (unlikely(err))
355 return err;
356
357 if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
358 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
359 dev_net(skb->dev), NULL, skb, NULL,
360 skb_dst(skb)->dev, seg6_input_finish);
361
362 return seg6_input_finish(dev_net(skb->dev), NULL, skb);
363 }
364
seg6_input_nf(struct sk_buff * skb)365 static int seg6_input_nf(struct sk_buff *skb)
366 {
367 struct net_device *dev = skb_dst(skb)->dev;
368 struct net *net = dev_net(skb->dev);
369
370 switch (skb->protocol) {
371 case htons(ETH_P_IP):
372 return NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, NULL,
373 skb, NULL, dev, seg6_input_core);
374 case htons(ETH_P_IPV6):
375 return NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, NULL,
376 skb, NULL, dev, seg6_input_core);
377 }
378
379 return -EINVAL;
380 }
381
seg6_input(struct sk_buff * skb)382 static int seg6_input(struct sk_buff *skb)
383 {
384 if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
385 return seg6_input_nf(skb);
386
387 return seg6_input_core(dev_net(skb->dev), NULL, skb);
388 }
389
seg6_output_core(struct net * net,struct sock * sk,struct sk_buff * skb)390 static int seg6_output_core(struct net *net, struct sock *sk,
391 struct sk_buff *skb)
392 {
393 struct dst_entry *orig_dst = skb_dst(skb);
394 struct dst_entry *dst = NULL;
395 struct seg6_lwt *slwt;
396 int err;
397
398 err = seg6_do_srh(skb);
399 if (unlikely(err))
400 goto drop;
401
402 slwt = seg6_lwt_lwtunnel(orig_dst->lwtstate);
403
404 preempt_disable();
405 dst = dst_cache_get(&slwt->cache);
406 preempt_enable();
407
408 if (unlikely(!dst)) {
409 struct ipv6hdr *hdr = ipv6_hdr(skb);
410 struct flowi6 fl6;
411
412 memset(&fl6, 0, sizeof(fl6));
413 fl6.daddr = hdr->daddr;
414 fl6.saddr = hdr->saddr;
415 fl6.flowlabel = ip6_flowinfo(hdr);
416 fl6.flowi6_mark = skb->mark;
417 fl6.flowi6_proto = hdr->nexthdr;
418
419 dst = ip6_route_output(net, NULL, &fl6);
420 if (dst->error) {
421 err = dst->error;
422 dst_release(dst);
423 goto drop;
424 }
425
426 preempt_disable();
427 dst_cache_set_ip6(&slwt->cache, dst, &fl6.saddr);
428 preempt_enable();
429 }
430
431 skb_dst_drop(skb);
432 skb_dst_set(skb, dst);
433
434 err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
435 if (unlikely(err))
436 goto drop;
437
438 if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
439 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, skb,
440 NULL, skb_dst(skb)->dev, dst_output);
441
442 return dst_output(net, sk, skb);
443 drop:
444 kfree_skb(skb);
445 return err;
446 }
447
seg6_output_nf(struct net * net,struct sock * sk,struct sk_buff * skb)448 static int seg6_output_nf(struct net *net, struct sock *sk, struct sk_buff *skb)
449 {
450 struct net_device *dev = skb_dst(skb)->dev;
451
452 switch (skb->protocol) {
453 case htons(ETH_P_IP):
454 return NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb,
455 NULL, dev, seg6_output_core);
456 case htons(ETH_P_IPV6):
457 return NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb,
458 NULL, dev, seg6_output_core);
459 }
460
461 return -EINVAL;
462 }
463
seg6_output(struct net * net,struct sock * sk,struct sk_buff * skb)464 static int seg6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
465 {
466 if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
467 return seg6_output_nf(net, sk, skb);
468
469 return seg6_output_core(net, sk, skb);
470 }
471
seg6_build_state(struct net * net,struct nlattr * nla,unsigned int family,const void * cfg,struct lwtunnel_state ** ts,struct netlink_ext_ack * extack)472 static int seg6_build_state(struct net *net, struct nlattr *nla,
473 unsigned int family, const void *cfg,
474 struct lwtunnel_state **ts,
475 struct netlink_ext_ack *extack)
476 {
477 struct nlattr *tb[SEG6_IPTUNNEL_MAX + 1];
478 struct seg6_iptunnel_encap *tuninfo;
479 struct lwtunnel_state *newts;
480 int tuninfo_len, min_size;
481 struct seg6_lwt *slwt;
482 int err;
483
484 if (family != AF_INET && family != AF_INET6)
485 return -EINVAL;
486
487 err = nla_parse_nested_deprecated(tb, SEG6_IPTUNNEL_MAX, nla,
488 seg6_iptunnel_policy, extack);
489
490 if (err < 0)
491 return err;
492
493 if (!tb[SEG6_IPTUNNEL_SRH])
494 return -EINVAL;
495
496 tuninfo = nla_data(tb[SEG6_IPTUNNEL_SRH]);
497 tuninfo_len = nla_len(tb[SEG6_IPTUNNEL_SRH]);
498
499 /* tuninfo must contain at least the iptunnel encap structure,
500 * the SRH and one segment
501 */
502 min_size = sizeof(*tuninfo) + sizeof(struct ipv6_sr_hdr) +
503 sizeof(struct in6_addr);
504 if (tuninfo_len < min_size)
505 return -EINVAL;
506
507 switch (tuninfo->mode) {
508 case SEG6_IPTUN_MODE_INLINE:
509 if (family != AF_INET6)
510 return -EINVAL;
511
512 break;
513 case SEG6_IPTUN_MODE_ENCAP:
514 break;
515 case SEG6_IPTUN_MODE_L2ENCAP:
516 break;
517 default:
518 return -EINVAL;
519 }
520
521 /* verify that SRH is consistent */
522 if (!seg6_validate_srh(tuninfo->srh, tuninfo_len - sizeof(*tuninfo), false))
523 return -EINVAL;
524
525 newts = lwtunnel_state_alloc(tuninfo_len + sizeof(*slwt));
526 if (!newts)
527 return -ENOMEM;
528
529 slwt = seg6_lwt_lwtunnel(newts);
530
531 err = dst_cache_init(&slwt->cache, GFP_ATOMIC);
532 if (err) {
533 kfree(newts);
534 return err;
535 }
536
537 memcpy(&slwt->tuninfo, tuninfo, tuninfo_len);
538
539 newts->type = LWTUNNEL_ENCAP_SEG6;
540 newts->flags |= LWTUNNEL_STATE_INPUT_REDIRECT;
541
542 if (tuninfo->mode != SEG6_IPTUN_MODE_L2ENCAP)
543 newts->flags |= LWTUNNEL_STATE_OUTPUT_REDIRECT;
544
545 newts->headroom = seg6_lwt_headroom(tuninfo);
546
547 *ts = newts;
548
549 return 0;
550 }
551
seg6_destroy_state(struct lwtunnel_state * lwt)552 static void seg6_destroy_state(struct lwtunnel_state *lwt)
553 {
554 dst_cache_destroy(&seg6_lwt_lwtunnel(lwt)->cache);
555 }
556
seg6_fill_encap_info(struct sk_buff * skb,struct lwtunnel_state * lwtstate)557 static int seg6_fill_encap_info(struct sk_buff *skb,
558 struct lwtunnel_state *lwtstate)
559 {
560 struct seg6_iptunnel_encap *tuninfo = seg6_encap_lwtunnel(lwtstate);
561
562 if (nla_put_srh(skb, SEG6_IPTUNNEL_SRH, tuninfo))
563 return -EMSGSIZE;
564
565 return 0;
566 }
567
seg6_encap_nlsize(struct lwtunnel_state * lwtstate)568 static int seg6_encap_nlsize(struct lwtunnel_state *lwtstate)
569 {
570 struct seg6_iptunnel_encap *tuninfo = seg6_encap_lwtunnel(lwtstate);
571
572 return nla_total_size(SEG6_IPTUN_ENCAP_SIZE(tuninfo));
573 }
574
seg6_encap_cmp(struct lwtunnel_state * a,struct lwtunnel_state * b)575 static int seg6_encap_cmp(struct lwtunnel_state *a, struct lwtunnel_state *b)
576 {
577 struct seg6_iptunnel_encap *a_hdr = seg6_encap_lwtunnel(a);
578 struct seg6_iptunnel_encap *b_hdr = seg6_encap_lwtunnel(b);
579 int len = SEG6_IPTUN_ENCAP_SIZE(a_hdr);
580
581 if (len != SEG6_IPTUN_ENCAP_SIZE(b_hdr))
582 return 1;
583
584 return memcmp(a_hdr, b_hdr, len);
585 }
586
587 static const struct lwtunnel_encap_ops seg6_iptun_ops = {
588 .build_state = seg6_build_state,
589 .destroy_state = seg6_destroy_state,
590 .output = seg6_output,
591 .input = seg6_input,
592 .fill_encap = seg6_fill_encap_info,
593 .get_encap_size = seg6_encap_nlsize,
594 .cmp_encap = seg6_encap_cmp,
595 .owner = THIS_MODULE,
596 };
597
seg6_iptunnel_init(void)598 int __init seg6_iptunnel_init(void)
599 {
600 return lwtunnel_encap_add_ops(&seg6_iptun_ops, LWTUNNEL_ENCAP_SEG6);
601 }
602
seg6_iptunnel_exit(void)603 void seg6_iptunnel_exit(void)
604 {
605 lwtunnel_encap_del_ops(&seg6_iptun_ops, LWTUNNEL_ENCAP_SEG6);
606 }
607