xref: /openbmc/linux/net/ipv6/seg6_iptunnel.c (revision 97e6ea6d) 
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 
166 	hdr->nexthdr = NEXTHDR_ROUTING;
167 
168 	isrh = (void *)hdr + sizeof(*hdr);
169 	memcpy(isrh, osrh, hdrlen);
170 
171 	isrh->nexthdr = proto;
172 
173 	hdr->daddr = isrh->segments[isrh->first_segment];
174 	set_tun_src(net, dst->dev, &hdr->daddr, &hdr->saddr);
175 
176 #ifdef CONFIG_IPV6_SEG6_HMAC
177 	if (sr_has_hmac(isrh)) {
178 		err = seg6_push_hmac(net, &hdr->saddr, isrh);
179 		if (unlikely(err))
180 			return err;
181 	}
182 #endif
183 
184 	skb_postpush_rcsum(skb, hdr, tot_len);
185 
186 	return 0;
187 }
188 EXPORT_SYMBOL_GPL(seg6_do_srh_encap);
189 
190 /* insert an SRH within an IPv6 packet, just after the IPv6 header */
191 int seg6_do_srh_inline(struct sk_buff *skb, struct ipv6_sr_hdr *osrh)
192 {
193 	struct ipv6hdr *hdr, *oldhdr;
194 	struct ipv6_sr_hdr *isrh;
195 	int hdrlen, err;
196 
197 	hdrlen = (osrh->hdrlen + 1) << 3;
198 
199 	err = skb_cow_head(skb, hdrlen + skb->mac_len);
200 	if (unlikely(err))
201 		return err;
202 
203 	oldhdr = ipv6_hdr(skb);
204 
205 	skb_pull(skb, sizeof(struct ipv6hdr));
206 	skb_postpull_rcsum(skb, skb_network_header(skb),
207 			   sizeof(struct ipv6hdr));
208 
209 	skb_push(skb, sizeof(struct ipv6hdr) + hdrlen);
210 	skb_reset_network_header(skb);
211 	skb_mac_header_rebuild(skb);
212 
213 	hdr = ipv6_hdr(skb);
214 
215 	memmove(hdr, oldhdr, sizeof(*hdr));
216 
217 	isrh = (void *)hdr + sizeof(*hdr);
218 	memcpy(isrh, osrh, hdrlen);
219 
220 	isrh->nexthdr = hdr->nexthdr;
221 	hdr->nexthdr = NEXTHDR_ROUTING;
222 
223 	isrh->segments[0] = hdr->daddr;
224 	hdr->daddr = isrh->segments[isrh->first_segment];
225 
226 #ifdef CONFIG_IPV6_SEG6_HMAC
227 	if (sr_has_hmac(isrh)) {
228 		struct net *net = dev_net(skb_dst(skb)->dev);
229 
230 		err = seg6_push_hmac(net, &hdr->saddr, isrh);
231 		if (unlikely(err))
232 			return err;
233 	}
234 #endif
235 
236 	skb_postpush_rcsum(skb, hdr, sizeof(struct ipv6hdr) + hdrlen);
237 
238 	return 0;
239 }
240 EXPORT_SYMBOL_GPL(seg6_do_srh_inline);
241 
242 static int seg6_do_srh(struct sk_buff *skb)
243 {
244 	struct dst_entry *dst = skb_dst(skb);
245 	struct seg6_iptunnel_encap *tinfo;
246 	int proto, err = 0;
247 
248 	tinfo = seg6_encap_lwtunnel(dst->lwtstate);
249 
250 	switch (tinfo->mode) {
251 	case SEG6_IPTUN_MODE_INLINE:
252 		if (skb->protocol != htons(ETH_P_IPV6))
253 			return -EINVAL;
254 
255 		err = seg6_do_srh_inline(skb, tinfo->srh);
256 		if (err)
257 			return err;
258 		break;
259 	case SEG6_IPTUN_MODE_ENCAP:
260 		err = iptunnel_handle_offloads(skb, SKB_GSO_IPXIP6);
261 		if (err)
262 			return err;
263 
264 		if (skb->protocol == htons(ETH_P_IPV6))
265 			proto = IPPROTO_IPV6;
266 		else if (skb->protocol == htons(ETH_P_IP))
267 			proto = IPPROTO_IPIP;
268 		else
269 			return -EINVAL;
270 
271 		err = seg6_do_srh_encap(skb, tinfo->srh, proto);
272 		if (err)
273 			return err;
274 
275 		skb_set_inner_transport_header(skb, skb_transport_offset(skb));
276 		skb_set_inner_protocol(skb, skb->protocol);
277 		skb->protocol = htons(ETH_P_IPV6);
278 		break;
279 	case SEG6_IPTUN_MODE_L2ENCAP:
280 		if (!skb_mac_header_was_set(skb))
281 			return -EINVAL;
282 
283 		if (pskb_expand_head(skb, skb->mac_len, 0, GFP_ATOMIC) < 0)
284 			return -ENOMEM;
285 
286 		skb_mac_header_rebuild(skb);
287 		skb_push(skb, skb->mac_len);
288 
289 		err = seg6_do_srh_encap(skb, tinfo->srh, IPPROTO_ETHERNET);
290 		if (err)
291 			return err;
292 
293 		skb->protocol = htons(ETH_P_IPV6);
294 		break;
295 	}
296 
297 	ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
298 	skb_set_transport_header(skb, sizeof(struct ipv6hdr));
299 	nf_reset_ct(skb);
300 
301 	return 0;
302 }
303 
304 static int seg6_input_finish(struct net *net, struct sock *sk,
305 			     struct sk_buff *skb)
306 {
307 	return dst_input(skb);
308 }
309 
310 static int seg6_input_core(struct net *net, struct sock *sk,
311 			   struct sk_buff *skb)
312 {
313 	struct dst_entry *orig_dst = skb_dst(skb);
314 	struct dst_entry *dst = NULL;
315 	struct seg6_lwt *slwt;
316 	int err;
317 
318 	err = seg6_do_srh(skb);
319 	if (unlikely(err)) {
320 		kfree_skb(skb);
321 		return err;
322 	}
323 
324 	slwt = seg6_lwt_lwtunnel(orig_dst->lwtstate);
325 
326 	preempt_disable();
327 	dst = dst_cache_get(&slwt->cache);
328 	preempt_enable();
329 
330 	skb_dst_drop(skb);
331 
332 	if (!dst) {
333 		ip6_route_input(skb);
334 		dst = skb_dst(skb);
335 		if (!dst->error) {
336 			preempt_disable();
337 			dst_cache_set_ip6(&slwt->cache, dst,
338 					  &ipv6_hdr(skb)->saddr);
339 			preempt_enable();
340 		}
341 	} else {
342 		skb_dst_set(skb, dst);
343 	}
344 
345 	err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
346 	if (unlikely(err))
347 		return err;
348 
349 	if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
350 		return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
351 			       dev_net(skb->dev), NULL, skb, NULL,
352 			       skb_dst(skb)->dev, seg6_input_finish);
353 
354 	return seg6_input_finish(dev_net(skb->dev), NULL, skb);
355 }
356 
357 static int seg6_input_nf(struct sk_buff *skb)
358 {
359 	struct net_device *dev = skb_dst(skb)->dev;
360 	struct net *net = dev_net(skb->dev);
361 
362 	switch (skb->protocol) {
363 	case htons(ETH_P_IP):
364 		return NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, NULL,
365 			       skb, NULL, dev, seg6_input_core);
366 	case htons(ETH_P_IPV6):
367 		return NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, NULL,
368 			       skb, NULL, dev, seg6_input_core);
369 	}
370 
371 	return -EINVAL;
372 }
373 
374 static int seg6_input(struct sk_buff *skb)
375 {
376 	if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
377 		return seg6_input_nf(skb);
378 
379 	return seg6_input_core(dev_net(skb->dev), NULL, skb);
380 }
381 
382 static int seg6_output_core(struct net *net, struct sock *sk,
383 			    struct sk_buff *skb)
384 {
385 	struct dst_entry *orig_dst = skb_dst(skb);
386 	struct dst_entry *dst = NULL;
387 	struct seg6_lwt *slwt;
388 	int err;
389 
390 	err = seg6_do_srh(skb);
391 	if (unlikely(err))
392 		goto drop;
393 
394 	slwt = seg6_lwt_lwtunnel(orig_dst->lwtstate);
395 
396 	preempt_disable();
397 	dst = dst_cache_get(&slwt->cache);
398 	preempt_enable();
399 
400 	if (unlikely(!dst)) {
401 		struct ipv6hdr *hdr = ipv6_hdr(skb);
402 		struct flowi6 fl6;
403 
404 		memset(&fl6, 0, sizeof(fl6));
405 		fl6.daddr = hdr->daddr;
406 		fl6.saddr = hdr->saddr;
407 		fl6.flowlabel = ip6_flowinfo(hdr);
408 		fl6.flowi6_mark = skb->mark;
409 		fl6.flowi6_proto = hdr->nexthdr;
410 
411 		dst = ip6_route_output(net, NULL, &fl6);
412 		if (dst->error) {
413 			err = dst->error;
414 			dst_release(dst);
415 			goto drop;
416 		}
417 
418 		preempt_disable();
419 		dst_cache_set_ip6(&slwt->cache, dst, &fl6.saddr);
420 		preempt_enable();
421 	}
422 
423 	skb_dst_drop(skb);
424 	skb_dst_set(skb, dst);
425 
426 	err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
427 	if (unlikely(err))
428 		goto drop;
429 
430 	if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
431 		return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, skb,
432 			       NULL, skb_dst(skb)->dev, dst_output);
433 
434 	return dst_output(net, sk, skb);
435 drop:
436 	kfree_skb(skb);
437 	return err;
438 }
439 
440 static int seg6_output_nf(struct net *net, struct sock *sk, struct sk_buff *skb)
441 {
442 	struct net_device *dev = skb_dst(skb)->dev;
443 
444 	switch (skb->protocol) {
445 	case htons(ETH_P_IP):
446 		return NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb,
447 			       NULL, dev, seg6_output_core);
448 	case htons(ETH_P_IPV6):
449 		return NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb,
450 			       NULL, dev, seg6_output_core);
451 	}
452 
453 	return -EINVAL;
454 }
455 
456 static int seg6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
457 {
458 	if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
459 		return seg6_output_nf(net, sk, skb);
460 
461 	return seg6_output_core(net, sk, skb);
462 }
463 
464 static int seg6_build_state(struct net *net, struct nlattr *nla,
465 			    unsigned int family, const void *cfg,
466 			    struct lwtunnel_state **ts,
467 			    struct netlink_ext_ack *extack)
468 {
469 	struct nlattr *tb[SEG6_IPTUNNEL_MAX + 1];
470 	struct seg6_iptunnel_encap *tuninfo;
471 	struct lwtunnel_state *newts;
472 	int tuninfo_len, min_size;
473 	struct seg6_lwt *slwt;
474 	int err;
475 
476 	if (family != AF_INET && family != AF_INET6)
477 		return -EINVAL;
478 
479 	err = nla_parse_nested_deprecated(tb, SEG6_IPTUNNEL_MAX, nla,
480 					  seg6_iptunnel_policy, extack);
481 
482 	if (err < 0)
483 		return err;
484 
485 	if (!tb[SEG6_IPTUNNEL_SRH])
486 		return -EINVAL;
487 
488 	tuninfo = nla_data(tb[SEG6_IPTUNNEL_SRH]);
489 	tuninfo_len = nla_len(tb[SEG6_IPTUNNEL_SRH]);
490 
491 	/* tuninfo must contain at least the iptunnel encap structure,
492 	 * the SRH and one segment
493 	 */
494 	min_size = sizeof(*tuninfo) + sizeof(struct ipv6_sr_hdr) +
495 		   sizeof(struct in6_addr);
496 	if (tuninfo_len < min_size)
497 		return -EINVAL;
498 
499 	switch (tuninfo->mode) {
500 	case SEG6_IPTUN_MODE_INLINE:
501 		if (family != AF_INET6)
502 			return -EINVAL;
503 
504 		break;
505 	case SEG6_IPTUN_MODE_ENCAP:
506 		break;
507 	case SEG6_IPTUN_MODE_L2ENCAP:
508 		break;
509 	default:
510 		return -EINVAL;
511 	}
512 
513 	/* verify that SRH is consistent */
514 	if (!seg6_validate_srh(tuninfo->srh, tuninfo_len - sizeof(*tuninfo), false))
515 		return -EINVAL;
516 
517 	newts = lwtunnel_state_alloc(tuninfo_len + sizeof(*slwt));
518 	if (!newts)
519 		return -ENOMEM;
520 
521 	slwt = seg6_lwt_lwtunnel(newts);
522 
523 	err = dst_cache_init(&slwt->cache, GFP_ATOMIC);
524 	if (err) {
525 		kfree(newts);
526 		return err;
527 	}
528 
529 	memcpy(&slwt->tuninfo, tuninfo, tuninfo_len);
530 
531 	newts->type = LWTUNNEL_ENCAP_SEG6;
532 	newts->flags |= LWTUNNEL_STATE_INPUT_REDIRECT;
533 
534 	if (tuninfo->mode != SEG6_IPTUN_MODE_L2ENCAP)
535 		newts->flags |= LWTUNNEL_STATE_OUTPUT_REDIRECT;
536 
537 	newts->headroom = seg6_lwt_headroom(tuninfo);
538 
539 	*ts = newts;
540 
541 	return 0;
542 }
543 
544 static void seg6_destroy_state(struct lwtunnel_state *lwt)
545 {
546 	dst_cache_destroy(&seg6_lwt_lwtunnel(lwt)->cache);
547 }
548 
549 static int seg6_fill_encap_info(struct sk_buff *skb,
550 				struct lwtunnel_state *lwtstate)
551 {
552 	struct seg6_iptunnel_encap *tuninfo = seg6_encap_lwtunnel(lwtstate);
553 
554 	if (nla_put_srh(skb, SEG6_IPTUNNEL_SRH, tuninfo))
555 		return -EMSGSIZE;
556 
557 	return 0;
558 }
559 
560 static int seg6_encap_nlsize(struct lwtunnel_state *lwtstate)
561 {
562 	struct seg6_iptunnel_encap *tuninfo = seg6_encap_lwtunnel(lwtstate);
563 
564 	return nla_total_size(SEG6_IPTUN_ENCAP_SIZE(tuninfo));
565 }
566 
567 static int seg6_encap_cmp(struct lwtunnel_state *a, struct lwtunnel_state *b)
568 {
569 	struct seg6_iptunnel_encap *a_hdr = seg6_encap_lwtunnel(a);
570 	struct seg6_iptunnel_encap *b_hdr = seg6_encap_lwtunnel(b);
571 	int len = SEG6_IPTUN_ENCAP_SIZE(a_hdr);
572 
573 	if (len != SEG6_IPTUN_ENCAP_SIZE(b_hdr))
574 		return 1;
575 
576 	return memcmp(a_hdr, b_hdr, len);
577 }
578 
579 static const struct lwtunnel_encap_ops seg6_iptun_ops = {
580 	.build_state = seg6_build_state,
581 	.destroy_state = seg6_destroy_state,
582 	.output = seg6_output,
583 	.input = seg6_input,
584 	.fill_encap = seg6_fill_encap_info,
585 	.get_encap_size = seg6_encap_nlsize,
586 	.cmp_encap = seg6_encap_cmp,
587 	.owner = THIS_MODULE,
588 };
589 
590 int __init seg6_iptunnel_init(void)
591 {
592 	return lwtunnel_encap_add_ops(&seg6_iptun_ops, LWTUNNEL_ENCAP_SEG6);
593 }
594 
595 void seg6_iptunnel_exit(void)
596 {
597 	lwtunnel_encap_del_ops(&seg6_iptun_ops, LWTUNNEL_ENCAP_SEG6);
598 }
599