xref: /openbmc/linux/net/ipv6/seg6_iptunnel.c (revision 4fc4dca8) 
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 		memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
151 	}
152 
153 	hdr->nexthdr = NEXTHDR_ROUTING;
154 
155 	isrh = (void *)hdr + sizeof(*hdr);
156 	memcpy(isrh, osrh, hdrlen);
157 
158 	isrh->nexthdr = proto;
159 
160 	hdr->daddr = isrh->segments[isrh->first_segment];
161 	set_tun_src(net, dst->dev, &hdr->daddr, &hdr->saddr);
162 
163 #ifdef CONFIG_IPV6_SEG6_HMAC
164 	if (sr_has_hmac(isrh)) {
165 		err = seg6_push_hmac(net, &hdr->saddr, isrh);
166 		if (unlikely(err))
167 			return err;
168 	}
169 #endif
170 
171 	skb_postpush_rcsum(skb, hdr, tot_len);
172 
173 	return 0;
174 }
175 EXPORT_SYMBOL_GPL(seg6_do_srh_encap);
176 
177 /* insert an SRH within an IPv6 packet, just after the IPv6 header */
178 int seg6_do_srh_inline(struct sk_buff *skb, struct ipv6_sr_hdr *osrh)
179 {
180 	struct ipv6hdr *hdr, *oldhdr;
181 	struct ipv6_sr_hdr *isrh;
182 	int hdrlen, err;
183 
184 	hdrlen = (osrh->hdrlen + 1) << 3;
185 
186 	err = skb_cow_head(skb, hdrlen + skb->mac_len);
187 	if (unlikely(err))
188 		return err;
189 
190 	oldhdr = ipv6_hdr(skb);
191 
192 	skb_pull(skb, sizeof(struct ipv6hdr));
193 	skb_postpull_rcsum(skb, skb_network_header(skb),
194 			   sizeof(struct ipv6hdr));
195 
196 	skb_push(skb, sizeof(struct ipv6hdr) + hdrlen);
197 	skb_reset_network_header(skb);
198 	skb_mac_header_rebuild(skb);
199 
200 	hdr = ipv6_hdr(skb);
201 
202 	memmove(hdr, oldhdr, sizeof(*hdr));
203 
204 	isrh = (void *)hdr + sizeof(*hdr);
205 	memcpy(isrh, osrh, hdrlen);
206 
207 	isrh->nexthdr = hdr->nexthdr;
208 	hdr->nexthdr = NEXTHDR_ROUTING;
209 
210 	isrh->segments[0] = hdr->daddr;
211 	hdr->daddr = isrh->segments[isrh->first_segment];
212 
213 #ifdef CONFIG_IPV6_SEG6_HMAC
214 	if (sr_has_hmac(isrh)) {
215 		struct net *net = dev_net(skb_dst(skb)->dev);
216 
217 		err = seg6_push_hmac(net, &hdr->saddr, isrh);
218 		if (unlikely(err))
219 			return err;
220 	}
221 #endif
222 
223 	skb_postpush_rcsum(skb, hdr, sizeof(struct ipv6hdr) + hdrlen);
224 
225 	return 0;
226 }
227 EXPORT_SYMBOL_GPL(seg6_do_srh_inline);
228 
229 static int seg6_do_srh(struct sk_buff *skb)
230 {
231 	struct dst_entry *dst = skb_dst(skb);
232 	struct seg6_iptunnel_encap *tinfo;
233 	int proto, err = 0;
234 
235 	tinfo = seg6_encap_lwtunnel(dst->lwtstate);
236 
237 	switch (tinfo->mode) {
238 	case SEG6_IPTUN_MODE_INLINE:
239 		if (skb->protocol != htons(ETH_P_IPV6))
240 			return -EINVAL;
241 
242 		err = seg6_do_srh_inline(skb, tinfo->srh);
243 		if (err)
244 			return err;
245 		break;
246 	case SEG6_IPTUN_MODE_ENCAP:
247 		err = iptunnel_handle_offloads(skb, SKB_GSO_IPXIP6);
248 		if (err)
249 			return err;
250 
251 		if (skb->protocol == htons(ETH_P_IPV6))
252 			proto = IPPROTO_IPV6;
253 		else if (skb->protocol == htons(ETH_P_IP))
254 			proto = IPPROTO_IPIP;
255 		else
256 			return -EINVAL;
257 
258 		err = seg6_do_srh_encap(skb, tinfo->srh, proto);
259 		if (err)
260 			return err;
261 
262 		skb_set_inner_transport_header(skb, skb_transport_offset(skb));
263 		skb_set_inner_protocol(skb, skb->protocol);
264 		skb->protocol = htons(ETH_P_IPV6);
265 		break;
266 	case SEG6_IPTUN_MODE_L2ENCAP:
267 		if (!skb_mac_header_was_set(skb))
268 			return -EINVAL;
269 
270 		if (pskb_expand_head(skb, skb->mac_len, 0, GFP_ATOMIC) < 0)
271 			return -ENOMEM;
272 
273 		skb_mac_header_rebuild(skb);
274 		skb_push(skb, skb->mac_len);
275 
276 		err = seg6_do_srh_encap(skb, tinfo->srh, NEXTHDR_NONE);
277 		if (err)
278 			return err;
279 
280 		skb->protocol = htons(ETH_P_IPV6);
281 		break;
282 	}
283 
284 	ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
285 	skb_set_transport_header(skb, sizeof(struct ipv6hdr));
286 
287 	return 0;
288 }
289 
290 static int seg6_input(struct sk_buff *skb)
291 {
292 	struct dst_entry *orig_dst = skb_dst(skb);
293 	struct dst_entry *dst = NULL;
294 	struct seg6_lwt *slwt;
295 	int err;
296 
297 	err = seg6_do_srh(skb);
298 	if (unlikely(err)) {
299 		kfree_skb(skb);
300 		return err;
301 	}
302 
303 	slwt = seg6_lwt_lwtunnel(orig_dst->lwtstate);
304 
305 	preempt_disable();
306 	dst = dst_cache_get(&slwt->cache);
307 	preempt_enable();
308 
309 	skb_dst_drop(skb);
310 
311 	if (!dst) {
312 		ip6_route_input(skb);
313 		dst = skb_dst(skb);
314 		if (!dst->error) {
315 			preempt_disable();
316 			dst_cache_set_ip6(&slwt->cache, dst,
317 					  &ipv6_hdr(skb)->saddr);
318 			preempt_enable();
319 		}
320 	} else {
321 		skb_dst_set(skb, dst);
322 	}
323 
324 	err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
325 	if (unlikely(err))
326 		return err;
327 
328 	return dst_input(skb);
329 }
330 
331 static int seg6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
332 {
333 	struct dst_entry *orig_dst = skb_dst(skb);
334 	struct dst_entry *dst = NULL;
335 	struct seg6_lwt *slwt;
336 	int err = -EINVAL;
337 
338 	err = seg6_do_srh(skb);
339 	if (unlikely(err))
340 		goto drop;
341 
342 	slwt = seg6_lwt_lwtunnel(orig_dst->lwtstate);
343 
344 	preempt_disable();
345 	dst = dst_cache_get(&slwt->cache);
346 	preempt_enable();
347 
348 	if (unlikely(!dst)) {
349 		struct ipv6hdr *hdr = ipv6_hdr(skb);
350 		struct flowi6 fl6;
351 
352 		memset(&fl6, 0, sizeof(fl6));
353 		fl6.daddr = hdr->daddr;
354 		fl6.saddr = hdr->saddr;
355 		fl6.flowlabel = ip6_flowinfo(hdr);
356 		fl6.flowi6_mark = skb->mark;
357 		fl6.flowi6_proto = hdr->nexthdr;
358 
359 		dst = ip6_route_output(net, NULL, &fl6);
360 		if (dst->error) {
361 			err = dst->error;
362 			dst_release(dst);
363 			goto drop;
364 		}
365 
366 		preempt_disable();
367 		dst_cache_set_ip6(&slwt->cache, dst, &fl6.saddr);
368 		preempt_enable();
369 	}
370 
371 	skb_dst_drop(skb);
372 	skb_dst_set(skb, dst);
373 
374 	err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
375 	if (unlikely(err))
376 		goto drop;
377 
378 	return dst_output(net, sk, skb);
379 drop:
380 	kfree_skb(skb);
381 	return err;
382 }
383 
384 static int seg6_build_state(struct nlattr *nla,
385 			    unsigned int family, const void *cfg,
386 			    struct lwtunnel_state **ts,
387 			    struct netlink_ext_ack *extack)
388 {
389 	struct nlattr *tb[SEG6_IPTUNNEL_MAX + 1];
390 	struct seg6_iptunnel_encap *tuninfo;
391 	struct lwtunnel_state *newts;
392 	int tuninfo_len, min_size;
393 	struct seg6_lwt *slwt;
394 	int err;
395 
396 	if (family != AF_INET && family != AF_INET6)
397 		return -EINVAL;
398 
399 	err = nla_parse_nested_deprecated(tb, SEG6_IPTUNNEL_MAX, nla,
400 					  seg6_iptunnel_policy, extack);
401 
402 	if (err < 0)
403 		return err;
404 
405 	if (!tb[SEG6_IPTUNNEL_SRH])
406 		return -EINVAL;
407 
408 	tuninfo = nla_data(tb[SEG6_IPTUNNEL_SRH]);
409 	tuninfo_len = nla_len(tb[SEG6_IPTUNNEL_SRH]);
410 
411 	/* tuninfo must contain at least the iptunnel encap structure,
412 	 * the SRH and one segment
413 	 */
414 	min_size = sizeof(*tuninfo) + sizeof(struct ipv6_sr_hdr) +
415 		   sizeof(struct in6_addr);
416 	if (tuninfo_len < min_size)
417 		return -EINVAL;
418 
419 	switch (tuninfo->mode) {
420 	case SEG6_IPTUN_MODE_INLINE:
421 		if (family != AF_INET6)
422 			return -EINVAL;
423 
424 		break;
425 	case SEG6_IPTUN_MODE_ENCAP:
426 		break;
427 	case SEG6_IPTUN_MODE_L2ENCAP:
428 		break;
429 	default:
430 		return -EINVAL;
431 	}
432 
433 	/* verify that SRH is consistent */
434 	if (!seg6_validate_srh(tuninfo->srh, tuninfo_len - sizeof(*tuninfo)))
435 		return -EINVAL;
436 
437 	newts = lwtunnel_state_alloc(tuninfo_len + sizeof(*slwt));
438 	if (!newts)
439 		return -ENOMEM;
440 
441 	slwt = seg6_lwt_lwtunnel(newts);
442 
443 	err = dst_cache_init(&slwt->cache, GFP_ATOMIC);
444 	if (err) {
445 		kfree(newts);
446 		return err;
447 	}
448 
449 	memcpy(&slwt->tuninfo, tuninfo, tuninfo_len);
450 
451 	newts->type = LWTUNNEL_ENCAP_SEG6;
452 	newts->flags |= LWTUNNEL_STATE_INPUT_REDIRECT;
453 
454 	if (tuninfo->mode != SEG6_IPTUN_MODE_L2ENCAP)
455 		newts->flags |= LWTUNNEL_STATE_OUTPUT_REDIRECT;
456 
457 	newts->headroom = seg6_lwt_headroom(tuninfo);
458 
459 	*ts = newts;
460 
461 	return 0;
462 }
463 
464 static void seg6_destroy_state(struct lwtunnel_state *lwt)
465 {
466 	dst_cache_destroy(&seg6_lwt_lwtunnel(lwt)->cache);
467 }
468 
469 static int seg6_fill_encap_info(struct sk_buff *skb,
470 				struct lwtunnel_state *lwtstate)
471 {
472 	struct seg6_iptunnel_encap *tuninfo = seg6_encap_lwtunnel(lwtstate);
473 
474 	if (nla_put_srh(skb, SEG6_IPTUNNEL_SRH, tuninfo))
475 		return -EMSGSIZE;
476 
477 	return 0;
478 }
479 
480 static int seg6_encap_nlsize(struct lwtunnel_state *lwtstate)
481 {
482 	struct seg6_iptunnel_encap *tuninfo = seg6_encap_lwtunnel(lwtstate);
483 
484 	return nla_total_size(SEG6_IPTUN_ENCAP_SIZE(tuninfo));
485 }
486 
487 static int seg6_encap_cmp(struct lwtunnel_state *a, struct lwtunnel_state *b)
488 {
489 	struct seg6_iptunnel_encap *a_hdr = seg6_encap_lwtunnel(a);
490 	struct seg6_iptunnel_encap *b_hdr = seg6_encap_lwtunnel(b);
491 	int len = SEG6_IPTUN_ENCAP_SIZE(a_hdr);
492 
493 	if (len != SEG6_IPTUN_ENCAP_SIZE(b_hdr))
494 		return 1;
495 
496 	return memcmp(a_hdr, b_hdr, len);
497 }
498 
499 static const struct lwtunnel_encap_ops seg6_iptun_ops = {
500 	.build_state = seg6_build_state,
501 	.destroy_state = seg6_destroy_state,
502 	.output = seg6_output,
503 	.input = seg6_input,
504 	.fill_encap = seg6_fill_encap_info,
505 	.get_encap_size = seg6_encap_nlsize,
506 	.cmp_encap = seg6_encap_cmp,
507 	.owner = THIS_MODULE,
508 };
509 
510 int __init seg6_iptunnel_init(void)
511 {
512 	return lwtunnel_encap_add_ops(&seg6_iptun_ops, LWTUNNEL_ENCAP_SEG6);
513 }
514 
515 void seg6_iptunnel_exit(void)
516 {
517 	lwtunnel_encap_del_ops(&seg6_iptun_ops, LWTUNNEL_ENCAP_SEG6);
518 }
519