xref: /openbmc/linux/net/ipv6/seg6_iptunnel.c (revision 6d99a79c) 
1 /*
2  *  SR-IPv6 implementation
3  *
4  *  Author:
5  *  David Lebrun <david.lebrun@uclouvain.be>
6  *
7  *
8  *  This program is free software; you can redistribute it and/or
9  *        modify it under the terms of the GNU General Public License
10  *        as published by the Free Software Foundation; either version
11  *        2 of the License, or (at your option) any later version.
12  */
13 
14 #include <linux/types.h>
15 #include <linux/skbuff.h>
16 #include <linux/net.h>
17 #include <linux/module.h>
18 #include <net/ip.h>
19 #include <net/ip_tunnels.h>
20 #include <net/lwtunnel.h>
21 #include <net/netevent.h>
22 #include <net/netns/generic.h>
23 #include <net/ip6_fib.h>
24 #include <net/route.h>
25 #include <net/seg6.h>
26 #include <linux/seg6.h>
27 #include <linux/seg6_iptunnel.h>
28 #include <net/addrconf.h>
29 #include <net/ip6_route.h>
30 #include <net/dst_cache.h>
31 #ifdef CONFIG_IPV6_SEG6_HMAC
32 #include <net/seg6_hmac.h>
33 #endif
34 
35 struct seg6_lwt {
36 	struct dst_cache cache;
37 	struct seg6_iptunnel_encap tuninfo[0];
38 };
39 
40 static inline struct seg6_lwt *seg6_lwt_lwtunnel(struct lwtunnel_state *lwt)
41 {
42 	return (struct seg6_lwt *)lwt->data;
43 }
44 
45 static inline struct seg6_iptunnel_encap *
46 seg6_encap_lwtunnel(struct lwtunnel_state *lwt)
47 {
48 	return seg6_lwt_lwtunnel(lwt)->tuninfo;
49 }
50 
51 static const struct nla_policy seg6_iptunnel_policy[SEG6_IPTUNNEL_MAX + 1] = {
52 	[SEG6_IPTUNNEL_SRH]	= { .type = NLA_BINARY },
53 };
54 
55 static int nla_put_srh(struct sk_buff *skb, int attrtype,
56 		       struct seg6_iptunnel_encap *tuninfo)
57 {
58 	struct seg6_iptunnel_encap *data;
59 	struct nlattr *nla;
60 	int len;
61 
62 	len = SEG6_IPTUN_ENCAP_SIZE(tuninfo);
63 
64 	nla = nla_reserve(skb, attrtype, len);
65 	if (!nla)
66 		return -EMSGSIZE;
67 
68 	data = nla_data(nla);
69 	memcpy(data, tuninfo, len);
70 
71 	return 0;
72 }
73 
74 static void set_tun_src(struct net *net, struct net_device *dev,
75 			struct in6_addr *daddr, struct in6_addr *saddr)
76 {
77 	struct seg6_pernet_data *sdata = seg6_pernet(net);
78 	struct in6_addr *tun_src;
79 
80 	rcu_read_lock();
81 
82 	tun_src = rcu_dereference(sdata->tun_src);
83 
84 	if (!ipv6_addr_any(tun_src)) {
85 		memcpy(saddr, tun_src, sizeof(struct in6_addr));
86 	} else {
87 		ipv6_dev_get_saddr(net, dev, daddr, IPV6_PREFER_SRC_PUBLIC,
88 				   saddr);
89 	}
90 
91 	rcu_read_unlock();
92 }
93 
94 /* Compute flowlabel for outer IPv6 header */
95 static __be32 seg6_make_flowlabel(struct net *net, struct sk_buff *skb,
96 				  struct ipv6hdr *inner_hdr)
97 {
98 	int do_flowlabel = net->ipv6.sysctl.seg6_flowlabel;
99 	__be32 flowlabel = 0;
100 	u32 hash;
101 
102 	if (do_flowlabel > 0) {
103 		hash = skb_get_hash(skb);
104 		hash = rol32(hash, 16);
105 		flowlabel = (__force __be32)hash & IPV6_FLOWLABEL_MASK;
106 	} else if (!do_flowlabel && skb->protocol == htons(ETH_P_IPV6)) {
107 		flowlabel = ip6_flowlabel(inner_hdr);
108 	}
109 	return flowlabel;
110 }
111 
112 /* encapsulate an IPv6 packet within an outer IPv6 header with a given SRH */
113 int seg6_do_srh_encap(struct sk_buff *skb, struct ipv6_sr_hdr *osrh, int proto)
114 {
115 	struct dst_entry *dst = skb_dst(skb);
116 	struct net *net = dev_net(dst->dev);
117 	struct ipv6hdr *hdr, *inner_hdr;
118 	struct ipv6_sr_hdr *isrh;
119 	int hdrlen, tot_len, err;
120 	__be32 flowlabel;
121 
122 	hdrlen = (osrh->hdrlen + 1) << 3;
123 	tot_len = hdrlen + sizeof(*hdr);
124 
125 	err = skb_cow_head(skb, tot_len + skb->mac_len);
126 	if (unlikely(err))
127 		return err;
128 
129 	inner_hdr = ipv6_hdr(skb);
130 	flowlabel = seg6_make_flowlabel(net, skb, inner_hdr);
131 
132 	skb_push(skb, tot_len);
133 	skb_reset_network_header(skb);
134 	skb_mac_header_rebuild(skb);
135 	hdr = ipv6_hdr(skb);
136 
137 	/* inherit tc, flowlabel and hlim
138 	 * hlim will be decremented in ip6_forward() afterwards and
139 	 * decapsulation will overwrite inner hlim with outer hlim
140 	 */
141 
142 	if (skb->protocol == htons(ETH_P_IPV6)) {
143 		ip6_flow_hdr(hdr, ip6_tclass(ip6_flowinfo(inner_hdr)),
144 			     flowlabel);
145 		hdr->hop_limit = inner_hdr->hop_limit;
146 	} else {
147 		ip6_flow_hdr(hdr, 0, flowlabel);
148 		hdr->hop_limit = ip6_dst_hoplimit(skb_dst(skb));
149 	}
150 
151 	hdr->nexthdr = NEXTHDR_ROUTING;
152 
153 	isrh = (void *)hdr + sizeof(*hdr);
154 	memcpy(isrh, osrh, hdrlen);
155 
156 	isrh->nexthdr = proto;
157 
158 	hdr->daddr = isrh->segments[isrh->first_segment];
159 	set_tun_src(net, dst->dev, &hdr->daddr, &hdr->saddr);
160 
161 #ifdef CONFIG_IPV6_SEG6_HMAC
162 	if (sr_has_hmac(isrh)) {
163 		err = seg6_push_hmac(net, &hdr->saddr, isrh);
164 		if (unlikely(err))
165 			return err;
166 	}
167 #endif
168 
169 	skb_postpush_rcsum(skb, hdr, tot_len);
170 
171 	return 0;
172 }
173 EXPORT_SYMBOL_GPL(seg6_do_srh_encap);
174 
175 /* insert an SRH within an IPv6 packet, just after the IPv6 header */
176 int seg6_do_srh_inline(struct sk_buff *skb, struct ipv6_sr_hdr *osrh)
177 {
178 	struct ipv6hdr *hdr, *oldhdr;
179 	struct ipv6_sr_hdr *isrh;
180 	int hdrlen, err;
181 
182 	hdrlen = (osrh->hdrlen + 1) << 3;
183 
184 	err = skb_cow_head(skb, hdrlen + skb->mac_len);
185 	if (unlikely(err))
186 		return err;
187 
188 	oldhdr = ipv6_hdr(skb);
189 
190 	skb_pull(skb, sizeof(struct ipv6hdr));
191 	skb_postpull_rcsum(skb, skb_network_header(skb),
192 			   sizeof(struct ipv6hdr));
193 
194 	skb_push(skb, sizeof(struct ipv6hdr) + hdrlen);
195 	skb_reset_network_header(skb);
196 	skb_mac_header_rebuild(skb);
197 
198 	hdr = ipv6_hdr(skb);
199 
200 	memmove(hdr, oldhdr, sizeof(*hdr));
201 
202 	isrh = (void *)hdr + sizeof(*hdr);
203 	memcpy(isrh, osrh, hdrlen);
204 
205 	isrh->nexthdr = hdr->nexthdr;
206 	hdr->nexthdr = NEXTHDR_ROUTING;
207 
208 	isrh->segments[0] = hdr->daddr;
209 	hdr->daddr = isrh->segments[isrh->first_segment];
210 
211 #ifdef CONFIG_IPV6_SEG6_HMAC
212 	if (sr_has_hmac(isrh)) {
213 		struct net *net = dev_net(skb_dst(skb)->dev);
214 
215 		err = seg6_push_hmac(net, &hdr->saddr, isrh);
216 		if (unlikely(err))
217 			return err;
218 	}
219 #endif
220 
221 	skb_postpush_rcsum(skb, hdr, sizeof(struct ipv6hdr) + hdrlen);
222 
223 	return 0;
224 }
225 EXPORT_SYMBOL_GPL(seg6_do_srh_inline);
226 
227 static int seg6_do_srh(struct sk_buff *skb)
228 {
229 	struct dst_entry *dst = skb_dst(skb);
230 	struct seg6_iptunnel_encap *tinfo;
231 	int proto, err = 0;
232 
233 	tinfo = seg6_encap_lwtunnel(dst->lwtstate);
234 
235 	switch (tinfo->mode) {
236 	case SEG6_IPTUN_MODE_INLINE:
237 		if (skb->protocol != htons(ETH_P_IPV6))
238 			return -EINVAL;
239 
240 		err = seg6_do_srh_inline(skb, tinfo->srh);
241 		if (err)
242 			return err;
243 		break;
244 	case SEG6_IPTUN_MODE_ENCAP:
245 		err = iptunnel_handle_offloads(skb, SKB_GSO_IPXIP6);
246 		if (err)
247 			return err;
248 
249 		if (skb->protocol == htons(ETH_P_IPV6))
250 			proto = IPPROTO_IPV6;
251 		else if (skb->protocol == htons(ETH_P_IP))
252 			proto = IPPROTO_IPIP;
253 		else
254 			return -EINVAL;
255 
256 		err = seg6_do_srh_encap(skb, tinfo->srh, proto);
257 		if (err)
258 			return err;
259 
260 		skb_set_inner_transport_header(skb, skb_transport_offset(skb));
261 		skb_set_inner_protocol(skb, skb->protocol);
262 		skb->protocol = htons(ETH_P_IPV6);
263 		break;
264 	case SEG6_IPTUN_MODE_L2ENCAP:
265 		if (!skb_mac_header_was_set(skb))
266 			return -EINVAL;
267 
268 		if (pskb_expand_head(skb, skb->mac_len, 0, GFP_ATOMIC) < 0)
269 			return -ENOMEM;
270 
271 		skb_mac_header_rebuild(skb);
272 		skb_push(skb, skb->mac_len);
273 
274 		err = seg6_do_srh_encap(skb, tinfo->srh, NEXTHDR_NONE);
275 		if (err)
276 			return err;
277 
278 		skb->protocol = htons(ETH_P_IPV6);
279 		break;
280 	}
281 
282 	ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
283 	skb_set_transport_header(skb, sizeof(struct ipv6hdr));
284 
285 	return 0;
286 }
287 
288 static int seg6_input(struct sk_buff *skb)
289 {
290 	struct dst_entry *orig_dst = skb_dst(skb);
291 	struct dst_entry *dst = NULL;
292 	struct seg6_lwt *slwt;
293 	int err;
294 
295 	err = seg6_do_srh(skb);
296 	if (unlikely(err)) {
297 		kfree_skb(skb);
298 		return err;
299 	}
300 
301 	slwt = seg6_lwt_lwtunnel(orig_dst->lwtstate);
302 
303 	preempt_disable();
304 	dst = dst_cache_get(&slwt->cache);
305 	preempt_enable();
306 
307 	skb_dst_drop(skb);
308 
309 	if (!dst) {
310 		ip6_route_input(skb);
311 		dst = skb_dst(skb);
312 		if (!dst->error) {
313 			preempt_disable();
314 			dst_cache_set_ip6(&slwt->cache, dst,
315 					  &ipv6_hdr(skb)->saddr);
316 			preempt_enable();
317 		}
318 	} else {
319 		skb_dst_set(skb, dst);
320 	}
321 
322 	err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
323 	if (unlikely(err))
324 		return err;
325 
326 	return dst_input(skb);
327 }
328 
329 static int seg6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
330 {
331 	struct dst_entry *orig_dst = skb_dst(skb);
332 	struct dst_entry *dst = NULL;
333 	struct seg6_lwt *slwt;
334 	int err = -EINVAL;
335 
336 	err = seg6_do_srh(skb);
337 	if (unlikely(err))
338 		goto drop;
339 
340 	slwt = seg6_lwt_lwtunnel(orig_dst->lwtstate);
341 
342 	preempt_disable();
343 	dst = dst_cache_get(&slwt->cache);
344 	preempt_enable();
345 
346 	if (unlikely(!dst)) {
347 		struct ipv6hdr *hdr = ipv6_hdr(skb);
348 		struct flowi6 fl6;
349 
350 		fl6.daddr = hdr->daddr;
351 		fl6.saddr = hdr->saddr;
352 		fl6.flowlabel = ip6_flowinfo(hdr);
353 		fl6.flowi6_mark = skb->mark;
354 		fl6.flowi6_proto = hdr->nexthdr;
355 
356 		dst = ip6_route_output(net, NULL, &fl6);
357 		if (dst->error) {
358 			err = dst->error;
359 			dst_release(dst);
360 			goto drop;
361 		}
362 
363 		preempt_disable();
364 		dst_cache_set_ip6(&slwt->cache, dst, &fl6.saddr);
365 		preempt_enable();
366 	}
367 
368 	skb_dst_drop(skb);
369 	skb_dst_set(skb, dst);
370 
371 	err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
372 	if (unlikely(err))
373 		goto drop;
374 
375 	return dst_output(net, sk, skb);
376 drop:
377 	kfree_skb(skb);
378 	return err;
379 }
380 
381 static int seg6_build_state(struct nlattr *nla,
382 			    unsigned int family, const void *cfg,
383 			    struct lwtunnel_state **ts,
384 			    struct netlink_ext_ack *extack)
385 {
386 	struct nlattr *tb[SEG6_IPTUNNEL_MAX + 1];
387 	struct seg6_iptunnel_encap *tuninfo;
388 	struct lwtunnel_state *newts;
389 	int tuninfo_len, min_size;
390 	struct seg6_lwt *slwt;
391 	int err;
392 
393 	if (family != AF_INET && family != AF_INET6)
394 		return -EINVAL;
395 
396 	err = nla_parse_nested(tb, SEG6_IPTUNNEL_MAX, nla,
397 			       seg6_iptunnel_policy, extack);
398 
399 	if (err < 0)
400 		return err;
401 
402 	if (!tb[SEG6_IPTUNNEL_SRH])
403 		return -EINVAL;
404 
405 	tuninfo = nla_data(tb[SEG6_IPTUNNEL_SRH]);
406 	tuninfo_len = nla_len(tb[SEG6_IPTUNNEL_SRH]);
407 
408 	/* tuninfo must contain at least the iptunnel encap structure,
409 	 * the SRH and one segment
410 	 */
411 	min_size = sizeof(*tuninfo) + sizeof(struct ipv6_sr_hdr) +
412 		   sizeof(struct in6_addr);
413 	if (tuninfo_len < min_size)
414 		return -EINVAL;
415 
416 	switch (tuninfo->mode) {
417 	case SEG6_IPTUN_MODE_INLINE:
418 		if (family != AF_INET6)
419 			return -EINVAL;
420 
421 		break;
422 	case SEG6_IPTUN_MODE_ENCAP:
423 		break;
424 	case SEG6_IPTUN_MODE_L2ENCAP:
425 		break;
426 	default:
427 		return -EINVAL;
428 	}
429 
430 	/* verify that SRH is consistent */
431 	if (!seg6_validate_srh(tuninfo->srh, tuninfo_len - sizeof(*tuninfo)))
432 		return -EINVAL;
433 
434 	newts = lwtunnel_state_alloc(tuninfo_len + sizeof(*slwt));
435 	if (!newts)
436 		return -ENOMEM;
437 
438 	slwt = seg6_lwt_lwtunnel(newts);
439 
440 	err = dst_cache_init(&slwt->cache, GFP_ATOMIC);
441 	if (err) {
442 		kfree(newts);
443 		return err;
444 	}
445 
446 	memcpy(&slwt->tuninfo, tuninfo, tuninfo_len);
447 
448 	newts->type = LWTUNNEL_ENCAP_SEG6;
449 	newts->flags |= LWTUNNEL_STATE_INPUT_REDIRECT;
450 
451 	if (tuninfo->mode != SEG6_IPTUN_MODE_L2ENCAP)
452 		newts->flags |= LWTUNNEL_STATE_OUTPUT_REDIRECT;
453 
454 	newts->headroom = seg6_lwt_headroom(tuninfo);
455 
456 	*ts = newts;
457 
458 	return 0;
459 }
460 
461 static void seg6_destroy_state(struct lwtunnel_state *lwt)
462 {
463 	dst_cache_destroy(&seg6_lwt_lwtunnel(lwt)->cache);
464 }
465 
466 static int seg6_fill_encap_info(struct sk_buff *skb,
467 				struct lwtunnel_state *lwtstate)
468 {
469 	struct seg6_iptunnel_encap *tuninfo = seg6_encap_lwtunnel(lwtstate);
470 
471 	if (nla_put_srh(skb, SEG6_IPTUNNEL_SRH, tuninfo))
472 		return -EMSGSIZE;
473 
474 	return 0;
475 }
476 
477 static int seg6_encap_nlsize(struct lwtunnel_state *lwtstate)
478 {
479 	struct seg6_iptunnel_encap *tuninfo = seg6_encap_lwtunnel(lwtstate);
480 
481 	return nla_total_size(SEG6_IPTUN_ENCAP_SIZE(tuninfo));
482 }
483 
484 static int seg6_encap_cmp(struct lwtunnel_state *a, struct lwtunnel_state *b)
485 {
486 	struct seg6_iptunnel_encap *a_hdr = seg6_encap_lwtunnel(a);
487 	struct seg6_iptunnel_encap *b_hdr = seg6_encap_lwtunnel(b);
488 	int len = SEG6_IPTUN_ENCAP_SIZE(a_hdr);
489 
490 	if (len != SEG6_IPTUN_ENCAP_SIZE(b_hdr))
491 		return 1;
492 
493 	return memcmp(a_hdr, b_hdr, len);
494 }
495 
496 static const struct lwtunnel_encap_ops seg6_iptun_ops = {
497 	.build_state = seg6_build_state,
498 	.destroy_state = seg6_destroy_state,
499 	.output = seg6_output,
500 	.input = seg6_input,
501 	.fill_encap = seg6_fill_encap_info,
502 	.get_encap_size = seg6_encap_nlsize,
503 	.cmp_encap = seg6_encap_cmp,
504 	.owner = THIS_MODULE,
505 };
506 
507 int __init seg6_iptunnel_init(void)
508 {
509 	return lwtunnel_encap_add_ops(&seg6_iptun_ops, LWTUNNEL_ENCAP_SEG6);
510 }
511 
512 void seg6_iptunnel_exit(void)
513 {
514 	lwtunnel_encap_del_ops(&seg6_iptun_ops, LWTUNNEL_ENCAP_SEG6);
515 }
516