xref: /openbmc/linux/net/ipv6/seg6_iptunnel.c (revision 5fb859f7) 
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 
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 
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 *
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 
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 
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 */
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 */
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 	hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
193 
194 	skb_postpush_rcsum(skb, hdr, tot_len);
195 
196 	return 0;
197 }
198 EXPORT_SYMBOL_GPL(seg6_do_srh_encap);
199 
200 /* insert an SRH within an IPv6 packet, just after the IPv6 header */
201 int seg6_do_srh_inline(struct sk_buff *skb, struct ipv6_sr_hdr *osrh)
202 {
203 	struct ipv6hdr *hdr, *oldhdr;
204 	struct ipv6_sr_hdr *isrh;
205 	int hdrlen, err;
206 
207 	hdrlen = (osrh->hdrlen + 1) << 3;
208 
209 	err = skb_cow_head(skb, hdrlen + skb->mac_len);
210 	if (unlikely(err))
211 		return err;
212 
213 	oldhdr = ipv6_hdr(skb);
214 
215 	skb_pull(skb, sizeof(struct ipv6hdr));
216 	skb_postpull_rcsum(skb, skb_network_header(skb),
217 			   sizeof(struct ipv6hdr));
218 
219 	skb_push(skb, sizeof(struct ipv6hdr) + hdrlen);
220 	skb_reset_network_header(skb);
221 	skb_mac_header_rebuild(skb);
222 
223 	hdr = ipv6_hdr(skb);
224 
225 	memmove(hdr, oldhdr, sizeof(*hdr));
226 
227 	isrh = (void *)hdr + sizeof(*hdr);
228 	memcpy(isrh, osrh, hdrlen);
229 
230 	isrh->nexthdr = hdr->nexthdr;
231 	hdr->nexthdr = NEXTHDR_ROUTING;
232 
233 	isrh->segments[0] = hdr->daddr;
234 	hdr->daddr = isrh->segments[isrh->first_segment];
235 
236 #ifdef CONFIG_IPV6_SEG6_HMAC
237 	if (sr_has_hmac(isrh)) {
238 		struct net *net = dev_net(skb_dst(skb)->dev);
239 
240 		err = seg6_push_hmac(net, &hdr->saddr, isrh);
241 		if (unlikely(err))
242 			return err;
243 	}
244 #endif
245 
246 	hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
247 
248 	skb_postpush_rcsum(skb, hdr, sizeof(struct ipv6hdr) + hdrlen);
249 
250 	return 0;
251 }
252 EXPORT_SYMBOL_GPL(seg6_do_srh_inline);
253 
254 static int seg6_do_srh(struct sk_buff *skb)
255 {
256 	struct dst_entry *dst = skb_dst(skb);
257 	struct seg6_iptunnel_encap *tinfo;
258 	int proto, err = 0;
259 
260 	tinfo = seg6_encap_lwtunnel(dst->lwtstate);
261 
262 	switch (tinfo->mode) {
263 	case SEG6_IPTUN_MODE_INLINE:
264 		if (skb->protocol != htons(ETH_P_IPV6))
265 			return -EINVAL;
266 
267 		err = seg6_do_srh_inline(skb, tinfo->srh);
268 		if (err)
269 			return err;
270 		break;
271 	case SEG6_IPTUN_MODE_ENCAP:
272 		err = iptunnel_handle_offloads(skb, SKB_GSO_IPXIP6);
273 		if (err)
274 			return err;
275 
276 		if (skb->protocol == htons(ETH_P_IPV6))
277 			proto = IPPROTO_IPV6;
278 		else if (skb->protocol == htons(ETH_P_IP))
279 			proto = IPPROTO_IPIP;
280 		else
281 			return -EINVAL;
282 
283 		err = seg6_do_srh_encap(skb, tinfo->srh, proto);
284 		if (err)
285 			return err;
286 
287 		skb_set_inner_transport_header(skb, skb_transport_offset(skb));
288 		skb_set_inner_protocol(skb, skb->protocol);
289 		skb->protocol = htons(ETH_P_IPV6);
290 		break;
291 	case SEG6_IPTUN_MODE_L2ENCAP:
292 		if (!skb_mac_header_was_set(skb))
293 			return -EINVAL;
294 
295 		if (pskb_expand_head(skb, skb->mac_len, 0, GFP_ATOMIC) < 0)
296 			return -ENOMEM;
297 
298 		skb_mac_header_rebuild(skb);
299 		skb_push(skb, skb->mac_len);
300 
301 		err = seg6_do_srh_encap(skb, tinfo->srh, IPPROTO_ETHERNET);
302 		if (err)
303 			return err;
304 
305 		skb->protocol = htons(ETH_P_IPV6);
306 		break;
307 	}
308 
309 	skb_set_transport_header(skb, sizeof(struct ipv6hdr));
310 	nf_reset_ct(skb);
311 
312 	return 0;
313 }
314 
315 static int seg6_input_finish(struct net *net, struct sock *sk,
316 			     struct sk_buff *skb)
317 {
318 	return dst_input(skb);
319 }
320 
321 static int seg6_input_core(struct net *net, struct sock *sk,
322 			   struct sk_buff *skb)
323 {
324 	struct dst_entry *orig_dst = skb_dst(skb);
325 	struct dst_entry *dst = NULL;
326 	struct seg6_lwt *slwt;
327 	int err;
328 
329 	err = seg6_do_srh(skb);
330 	if (unlikely(err)) {
331 		kfree_skb(skb);
332 		return err;
333 	}
334 
335 	slwt = seg6_lwt_lwtunnel(orig_dst->lwtstate);
336 
337 	preempt_disable();
338 	dst = dst_cache_get(&slwt->cache);
339 	preempt_enable();
340 
341 	skb_dst_drop(skb);
342 
343 	if (!dst) {
344 		ip6_route_input(skb);
345 		dst = skb_dst(skb);
346 		if (!dst->error) {
347 			preempt_disable();
348 			dst_cache_set_ip6(&slwt->cache, dst,
349 					  &ipv6_hdr(skb)->saddr);
350 			preempt_enable();
351 		}
352 	} else {
353 		skb_dst_set(skb, dst);
354 	}
355 
356 	err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
357 	if (unlikely(err))
358 		return err;
359 
360 	if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
361 		return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
362 			       dev_net(skb->dev), NULL, skb, NULL,
363 			       skb_dst(skb)->dev, seg6_input_finish);
364 
365 	return seg6_input_finish(dev_net(skb->dev), NULL, skb);
366 }
367 
368 static int seg6_input_nf(struct sk_buff *skb)
369 {
370 	struct net_device *dev = skb_dst(skb)->dev;
371 	struct net *net = dev_net(skb->dev);
372 
373 	switch (skb->protocol) {
374 	case htons(ETH_P_IP):
375 		return NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, NULL,
376 			       skb, NULL, dev, seg6_input_core);
377 	case htons(ETH_P_IPV6):
378 		return NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, NULL,
379 			       skb, NULL, dev, seg6_input_core);
380 	}
381 
382 	return -EINVAL;
383 }
384 
385 static int seg6_input(struct sk_buff *skb)
386 {
387 	if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
388 		return seg6_input_nf(skb);
389 
390 	return seg6_input_core(dev_net(skb->dev), NULL, skb);
391 }
392 
393 static int seg6_output_core(struct net *net, struct sock *sk,
394 			    struct sk_buff *skb)
395 {
396 	struct dst_entry *orig_dst = skb_dst(skb);
397 	struct dst_entry *dst = NULL;
398 	struct seg6_lwt *slwt;
399 	int err;
400 
401 	err = seg6_do_srh(skb);
402 	if (unlikely(err))
403 		goto drop;
404 
405 	slwt = seg6_lwt_lwtunnel(orig_dst->lwtstate);
406 
407 	preempt_disable();
408 	dst = dst_cache_get(&slwt->cache);
409 	preempt_enable();
410 
411 	if (unlikely(!dst)) {
412 		struct ipv6hdr *hdr = ipv6_hdr(skb);
413 		struct flowi6 fl6;
414 
415 		memset(&fl6, 0, sizeof(fl6));
416 		fl6.daddr = hdr->daddr;
417 		fl6.saddr = hdr->saddr;
418 		fl6.flowlabel = ip6_flowinfo(hdr);
419 		fl6.flowi6_mark = skb->mark;
420 		fl6.flowi6_proto = hdr->nexthdr;
421 
422 		dst = ip6_route_output(net, NULL, &fl6);
423 		if (dst->error) {
424 			err = dst->error;
425 			dst_release(dst);
426 			goto drop;
427 		}
428 
429 		preempt_disable();
430 		dst_cache_set_ip6(&slwt->cache, dst, &fl6.saddr);
431 		preempt_enable();
432 	}
433 
434 	skb_dst_drop(skb);
435 	skb_dst_set(skb, dst);
436 
437 	err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
438 	if (unlikely(err))
439 		goto drop;
440 
441 	if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
442 		return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, skb,
443 			       NULL, skb_dst(skb)->dev, dst_output);
444 
445 	return dst_output(net, sk, skb);
446 drop:
447 	kfree_skb(skb);
448 	return err;
449 }
450 
451 static int seg6_output_nf(struct net *net, struct sock *sk, struct sk_buff *skb)
452 {
453 	struct net_device *dev = skb_dst(skb)->dev;
454 
455 	switch (skb->protocol) {
456 	case htons(ETH_P_IP):
457 		return NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb,
458 			       NULL, dev, seg6_output_core);
459 	case htons(ETH_P_IPV6):
460 		return NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb,
461 			       NULL, dev, seg6_output_core);
462 	}
463 
464 	return -EINVAL;
465 }
466 
467 static int seg6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
468 {
469 	if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
470 		return seg6_output_nf(net, sk, skb);
471 
472 	return seg6_output_core(net, sk, skb);
473 }
474 
475 static int seg6_build_state(struct net *net, struct nlattr *nla,
476 			    unsigned int family, const void *cfg,
477 			    struct lwtunnel_state **ts,
478 			    struct netlink_ext_ack *extack)
479 {
480 	struct nlattr *tb[SEG6_IPTUNNEL_MAX + 1];
481 	struct seg6_iptunnel_encap *tuninfo;
482 	struct lwtunnel_state *newts;
483 	int tuninfo_len, min_size;
484 	struct seg6_lwt *slwt;
485 	int err;
486 
487 	if (family != AF_INET && family != AF_INET6)
488 		return -EINVAL;
489 
490 	err = nla_parse_nested_deprecated(tb, SEG6_IPTUNNEL_MAX, nla,
491 					  seg6_iptunnel_policy, extack);
492 
493 	if (err < 0)
494 		return err;
495 
496 	if (!tb[SEG6_IPTUNNEL_SRH])
497 		return -EINVAL;
498 
499 	tuninfo = nla_data(tb[SEG6_IPTUNNEL_SRH]);
500 	tuninfo_len = nla_len(tb[SEG6_IPTUNNEL_SRH]);
501 
502 	/* tuninfo must contain at least the iptunnel encap structure,
503 	 * the SRH and one segment
504 	 */
505 	min_size = sizeof(*tuninfo) + sizeof(struct ipv6_sr_hdr) +
506 		   sizeof(struct in6_addr);
507 	if (tuninfo_len < min_size)
508 		return -EINVAL;
509 
510 	switch (tuninfo->mode) {
511 	case SEG6_IPTUN_MODE_INLINE:
512 		if (family != AF_INET6)
513 			return -EINVAL;
514 
515 		break;
516 	case SEG6_IPTUN_MODE_ENCAP:
517 		break;
518 	case SEG6_IPTUN_MODE_L2ENCAP:
519 		break;
520 	default:
521 		return -EINVAL;
522 	}
523 
524 	/* verify that SRH is consistent */
525 	if (!seg6_validate_srh(tuninfo->srh, tuninfo_len - sizeof(*tuninfo), false))
526 		return -EINVAL;
527 
528 	newts = lwtunnel_state_alloc(tuninfo_len + sizeof(*slwt));
529 	if (!newts)
530 		return -ENOMEM;
531 
532 	slwt = seg6_lwt_lwtunnel(newts);
533 
534 	err = dst_cache_init(&slwt->cache, GFP_ATOMIC);
535 	if (err) {
536 		kfree(newts);
537 		return err;
538 	}
539 
540 	memcpy(&slwt->tuninfo, tuninfo, tuninfo_len);
541 
542 	newts->type = LWTUNNEL_ENCAP_SEG6;
543 	newts->flags |= LWTUNNEL_STATE_INPUT_REDIRECT;
544 
545 	if (tuninfo->mode != SEG6_IPTUN_MODE_L2ENCAP)
546 		newts->flags |= LWTUNNEL_STATE_OUTPUT_REDIRECT;
547 
548 	newts->headroom = seg6_lwt_headroom(tuninfo);
549 
550 	*ts = newts;
551 
552 	return 0;
553 }
554 
555 static void seg6_destroy_state(struct lwtunnel_state *lwt)
556 {
557 	dst_cache_destroy(&seg6_lwt_lwtunnel(lwt)->cache);
558 }
559 
560 static int seg6_fill_encap_info(struct sk_buff *skb,
561 				struct lwtunnel_state *lwtstate)
562 {
563 	struct seg6_iptunnel_encap *tuninfo = seg6_encap_lwtunnel(lwtstate);
564 
565 	if (nla_put_srh(skb, SEG6_IPTUNNEL_SRH, tuninfo))
566 		return -EMSGSIZE;
567 
568 	return 0;
569 }
570 
571 static int seg6_encap_nlsize(struct lwtunnel_state *lwtstate)
572 {
573 	struct seg6_iptunnel_encap *tuninfo = seg6_encap_lwtunnel(lwtstate);
574 
575 	return nla_total_size(SEG6_IPTUN_ENCAP_SIZE(tuninfo));
576 }
577 
578 static int seg6_encap_cmp(struct lwtunnel_state *a, struct lwtunnel_state *b)
579 {
580 	struct seg6_iptunnel_encap *a_hdr = seg6_encap_lwtunnel(a);
581 	struct seg6_iptunnel_encap *b_hdr = seg6_encap_lwtunnel(b);
582 	int len = SEG6_IPTUN_ENCAP_SIZE(a_hdr);
583 
584 	if (len != SEG6_IPTUN_ENCAP_SIZE(b_hdr))
585 		return 1;
586 
587 	return memcmp(a_hdr, b_hdr, len);
588 }
589 
590 static const struct lwtunnel_encap_ops seg6_iptun_ops = {
591 	.build_state = seg6_build_state,
592 	.destroy_state = seg6_destroy_state,
593 	.output = seg6_output,
594 	.input = seg6_input,
595 	.fill_encap = seg6_fill_encap_info,
596 	.get_encap_size = seg6_encap_nlsize,
597 	.cmp_encap = seg6_encap_cmp,
598 	.owner = THIS_MODULE,
599 };
600 
601 int __init seg6_iptunnel_init(void)
602 {
603 	return lwtunnel_encap_add_ops(&seg6_iptun_ops, LWTUNNEL_ENCAP_SEG6);
604 }
605 
606 void seg6_iptunnel_exit(void)
607 {
608 	lwtunnel_encap_del_ops(&seg6_iptun_ops, LWTUNNEL_ENCAP_SEG6);
609 }
610