xref: /linux/net/ipv6/seg6_iptunnel.c (revision 1e15510b71c99c6e49134d756df91069f7d18141)
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 	case SEG6_IPTUN_MODE_ENCAP_RED:
40 		head = sizeof(struct ipv6hdr);
41 		break;
42 	case SEG6_IPTUN_MODE_L2ENCAP:
43 	case SEG6_IPTUN_MODE_L2ENCAP_RED:
44 		return 0;
45 	}
46 
47 	return ((tuninfo->srh->hdrlen + 1) << 3) + head;
48 }
49 
50 struct seg6_lwt {
51 	struct dst_cache cache;
52 	struct seg6_iptunnel_encap tuninfo[];
53 };
54 
seg6_lwt_lwtunnel(struct lwtunnel_state * lwt)55 static inline struct seg6_lwt *seg6_lwt_lwtunnel(struct lwtunnel_state *lwt)
56 {
57 	return (struct seg6_lwt *)lwt->data;
58 }
59 
60 static inline struct seg6_iptunnel_encap *
seg6_encap_lwtunnel(struct lwtunnel_state * lwt)61 seg6_encap_lwtunnel(struct lwtunnel_state *lwt)
62 {
63 	return seg6_lwt_lwtunnel(lwt)->tuninfo;
64 }
65 
66 static const struct nla_policy seg6_iptunnel_policy[SEG6_IPTUNNEL_MAX + 1] = {
67 	[SEG6_IPTUNNEL_SRH]	= { .type = NLA_BINARY },
68 };
69 
nla_put_srh(struct sk_buff * skb,int attrtype,struct seg6_iptunnel_encap * tuninfo)70 static int nla_put_srh(struct sk_buff *skb, int attrtype,
71 		       struct seg6_iptunnel_encap *tuninfo)
72 {
73 	struct seg6_iptunnel_encap *data;
74 	struct nlattr *nla;
75 	int len;
76 
77 	len = SEG6_IPTUN_ENCAP_SIZE(tuninfo);
78 
79 	nla = nla_reserve(skb, attrtype, len);
80 	if (!nla)
81 		return -EMSGSIZE;
82 
83 	data = nla_data(nla);
84 	memcpy(data, tuninfo, len);
85 
86 	return 0;
87 }
88 
set_tun_src(struct net * net,struct net_device * dev,struct in6_addr * daddr,struct in6_addr * saddr)89 static void set_tun_src(struct net *net, struct net_device *dev,
90 			struct in6_addr *daddr, struct in6_addr *saddr)
91 {
92 	struct seg6_pernet_data *sdata = seg6_pernet(net);
93 	struct in6_addr *tun_src;
94 
95 	rcu_read_lock();
96 
97 	tun_src = rcu_dereference(sdata->tun_src);
98 
99 	if (!ipv6_addr_any(tun_src)) {
100 		memcpy(saddr, tun_src, sizeof(struct in6_addr));
101 	} else {
102 		ipv6_dev_get_saddr(net, dev, daddr, IPV6_PREFER_SRC_PUBLIC,
103 				   saddr);
104 	}
105 
106 	rcu_read_unlock();
107 }
108 
109 /* Compute flowlabel for outer IPv6 header */
seg6_make_flowlabel(struct net * net,struct sk_buff * skb,struct ipv6hdr * inner_hdr)110 static __be32 seg6_make_flowlabel(struct net *net, struct sk_buff *skb,
111 				  struct ipv6hdr *inner_hdr)
112 {
113 	int do_flowlabel = net->ipv6.sysctl.seg6_flowlabel;
114 	__be32 flowlabel = 0;
115 	u32 hash;
116 
117 	if (do_flowlabel > 0) {
118 		hash = skb_get_hash(skb);
119 		hash = rol32(hash, 16);
120 		flowlabel = (__force __be32)hash & IPV6_FLOWLABEL_MASK;
121 	} else if (!do_flowlabel && skb->protocol == htons(ETH_P_IPV6)) {
122 		flowlabel = ip6_flowlabel(inner_hdr);
123 	}
124 	return flowlabel;
125 }
126 
__seg6_do_srh_encap(struct sk_buff * skb,struct ipv6_sr_hdr * osrh,int proto,struct dst_entry * cache_dst)127 static int __seg6_do_srh_encap(struct sk_buff *skb, struct ipv6_sr_hdr *osrh,
128 			       int proto, struct dst_entry *cache_dst)
129 {
130 	struct dst_entry *dst = skb_dst(skb);
131 	struct net *net = dev_net(dst->dev);
132 	struct ipv6hdr *hdr, *inner_hdr;
133 	struct ipv6_sr_hdr *isrh;
134 	int hdrlen, tot_len, err;
135 	__be32 flowlabel;
136 
137 	hdrlen = (osrh->hdrlen + 1) << 3;
138 	tot_len = hdrlen + sizeof(*hdr);
139 
140 	err = skb_cow_head(skb, tot_len + dst_dev_overhead(cache_dst, skb));
141 	if (unlikely(err))
142 		return err;
143 
144 	inner_hdr = ipv6_hdr(skb);
145 	flowlabel = seg6_make_flowlabel(net, skb, inner_hdr);
146 
147 	skb_push(skb, tot_len);
148 	skb_reset_network_header(skb);
149 	skb_mac_header_rebuild(skb);
150 	hdr = ipv6_hdr(skb);
151 
152 	/* inherit tc, flowlabel and hlim
153 	 * hlim will be decremented in ip6_forward() afterwards and
154 	 * decapsulation will overwrite inner hlim with outer hlim
155 	 */
156 
157 	if (skb->protocol == htons(ETH_P_IPV6)) {
158 		ip6_flow_hdr(hdr, ip6_tclass(ip6_flowinfo(inner_hdr)),
159 			     flowlabel);
160 		hdr->hop_limit = inner_hdr->hop_limit;
161 	} else {
162 		ip6_flow_hdr(hdr, 0, flowlabel);
163 		hdr->hop_limit = ip6_dst_hoplimit(skb_dst(skb));
164 
165 		memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
166 
167 		/* the control block has been erased, so we have to set the
168 		 * iif once again.
169 		 * We read the receiving interface index directly from the
170 		 * skb->skb_iif as it is done in the IPv4 receiving path (i.e.:
171 		 * ip_rcv_core(...)).
172 		 */
173 		IP6CB(skb)->iif = skb->skb_iif;
174 	}
175 
176 	hdr->nexthdr = NEXTHDR_ROUTING;
177 
178 	isrh = (void *)hdr + sizeof(*hdr);
179 	memcpy(isrh, osrh, hdrlen);
180 
181 	isrh->nexthdr = proto;
182 
183 	hdr->daddr = isrh->segments[isrh->first_segment];
184 	set_tun_src(net, dst->dev, &hdr->daddr, &hdr->saddr);
185 
186 #ifdef CONFIG_IPV6_SEG6_HMAC
187 	if (sr_has_hmac(isrh)) {
188 		err = seg6_push_hmac(net, &hdr->saddr, isrh);
189 		if (unlikely(err))
190 			return err;
191 	}
192 #endif
193 
194 	hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
195 
196 	skb_postpush_rcsum(skb, hdr, tot_len);
197 
198 	return 0;
199 }
200 
201 /* 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)202 int seg6_do_srh_encap(struct sk_buff *skb, struct ipv6_sr_hdr *osrh, int proto)
203 {
204 	return __seg6_do_srh_encap(skb, osrh, proto, NULL);
205 }
206 EXPORT_SYMBOL_GPL(seg6_do_srh_encap);
207 
208 /* encapsulate an IPv6 packet within an outer IPv6 header with reduced SRH */
seg6_do_srh_encap_red(struct sk_buff * skb,struct ipv6_sr_hdr * osrh,int proto,struct dst_entry * cache_dst)209 static int seg6_do_srh_encap_red(struct sk_buff *skb,
210 				 struct ipv6_sr_hdr *osrh, int proto,
211 				 struct dst_entry *cache_dst)
212 {
213 	__u8 first_seg = osrh->first_segment;
214 	struct dst_entry *dst = skb_dst(skb);
215 	struct net *net = dev_net(dst->dev);
216 	struct ipv6hdr *hdr, *inner_hdr;
217 	int hdrlen = ipv6_optlen(osrh);
218 	int red_tlv_offset, tlv_offset;
219 	struct ipv6_sr_hdr *isrh;
220 	bool skip_srh = false;
221 	__be32 flowlabel;
222 	int tot_len, err;
223 	int red_hdrlen;
224 	int tlvs_len;
225 
226 	if (first_seg > 0) {
227 		red_hdrlen = hdrlen - sizeof(struct in6_addr);
228 	} else {
229 		/* NOTE: if tag/flags and/or other TLVs are introduced in the
230 		 * seg6_iptunnel infrastructure, they should be considered when
231 		 * deciding to skip the SRH.
232 		 */
233 		skip_srh = !sr_has_hmac(osrh);
234 
235 		red_hdrlen = skip_srh ? 0 : hdrlen;
236 	}
237 
238 	tot_len = red_hdrlen + sizeof(struct ipv6hdr);
239 
240 	err = skb_cow_head(skb, tot_len + dst_dev_overhead(cache_dst, skb));
241 	if (unlikely(err))
242 		return err;
243 
244 	inner_hdr = ipv6_hdr(skb);
245 	flowlabel = seg6_make_flowlabel(net, skb, inner_hdr);
246 
247 	skb_push(skb, tot_len);
248 	skb_reset_network_header(skb);
249 	skb_mac_header_rebuild(skb);
250 	hdr = ipv6_hdr(skb);
251 
252 	/* based on seg6_do_srh_encap() */
253 	if (skb->protocol == htons(ETH_P_IPV6)) {
254 		ip6_flow_hdr(hdr, ip6_tclass(ip6_flowinfo(inner_hdr)),
255 			     flowlabel);
256 		hdr->hop_limit = inner_hdr->hop_limit;
257 	} else {
258 		ip6_flow_hdr(hdr, 0, flowlabel);
259 		hdr->hop_limit = ip6_dst_hoplimit(skb_dst(skb));
260 
261 		memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
262 		IP6CB(skb)->iif = skb->skb_iif;
263 	}
264 
265 	/* no matter if we have to skip the SRH or not, the first segment
266 	 * always comes in the pushed IPv6 header.
267 	 */
268 	hdr->daddr = osrh->segments[first_seg];
269 
270 	if (skip_srh) {
271 		hdr->nexthdr = proto;
272 
273 		set_tun_src(net, dst->dev, &hdr->daddr, &hdr->saddr);
274 		goto out;
275 	}
276 
277 	/* we cannot skip the SRH, slow path */
278 
279 	hdr->nexthdr = NEXTHDR_ROUTING;
280 	isrh = (void *)hdr + sizeof(struct ipv6hdr);
281 
282 	if (unlikely(!first_seg)) {
283 		/* this is a very rare case; we have only one SID but
284 		 * we cannot skip the SRH since we are carrying some
285 		 * other info.
286 		 */
287 		memcpy(isrh, osrh, hdrlen);
288 		goto srcaddr;
289 	}
290 
291 	tlv_offset = sizeof(*osrh) + (first_seg + 1) * sizeof(struct in6_addr);
292 	red_tlv_offset = tlv_offset - sizeof(struct in6_addr);
293 
294 	memcpy(isrh, osrh, red_tlv_offset);
295 
296 	tlvs_len = hdrlen - tlv_offset;
297 	if (unlikely(tlvs_len > 0)) {
298 		const void *s = (const void *)osrh + tlv_offset;
299 		void *d = (void *)isrh + red_tlv_offset;
300 
301 		memcpy(d, s, tlvs_len);
302 	}
303 
304 	--isrh->first_segment;
305 	isrh->hdrlen -= 2;
306 
307 srcaddr:
308 	isrh->nexthdr = proto;
309 	set_tun_src(net, dst->dev, &hdr->daddr, &hdr->saddr);
310 
311 #ifdef CONFIG_IPV6_SEG6_HMAC
312 	if (unlikely(!skip_srh && sr_has_hmac(isrh))) {
313 		err = seg6_push_hmac(net, &hdr->saddr, isrh);
314 		if (unlikely(err))
315 			return err;
316 	}
317 #endif
318 
319 out:
320 	hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
321 
322 	skb_postpush_rcsum(skb, hdr, tot_len);
323 
324 	return 0;
325 }
326 
__seg6_do_srh_inline(struct sk_buff * skb,struct ipv6_sr_hdr * osrh,struct dst_entry * cache_dst)327 static int __seg6_do_srh_inline(struct sk_buff *skb, struct ipv6_sr_hdr *osrh,
328 				struct dst_entry *cache_dst)
329 {
330 	struct ipv6hdr *hdr, *oldhdr;
331 	struct ipv6_sr_hdr *isrh;
332 	int hdrlen, err;
333 
334 	hdrlen = (osrh->hdrlen + 1) << 3;
335 
336 	err = skb_cow_head(skb, hdrlen + dst_dev_overhead(cache_dst, skb));
337 	if (unlikely(err))
338 		return err;
339 
340 	oldhdr = ipv6_hdr(skb);
341 
342 	skb_pull(skb, sizeof(struct ipv6hdr));
343 	skb_postpull_rcsum(skb, skb_network_header(skb),
344 			   sizeof(struct ipv6hdr));
345 
346 	skb_push(skb, sizeof(struct ipv6hdr) + hdrlen);
347 	skb_reset_network_header(skb);
348 	skb_mac_header_rebuild(skb);
349 
350 	hdr = ipv6_hdr(skb);
351 
352 	memmove(hdr, oldhdr, sizeof(*hdr));
353 
354 	isrh = (void *)hdr + sizeof(*hdr);
355 	memcpy(isrh, osrh, hdrlen);
356 
357 	isrh->nexthdr = hdr->nexthdr;
358 	hdr->nexthdr = NEXTHDR_ROUTING;
359 
360 	isrh->segments[0] = hdr->daddr;
361 	hdr->daddr = isrh->segments[isrh->first_segment];
362 
363 #ifdef CONFIG_IPV6_SEG6_HMAC
364 	if (sr_has_hmac(isrh)) {
365 		struct net *net = dev_net(skb_dst(skb)->dev);
366 
367 		err = seg6_push_hmac(net, &hdr->saddr, isrh);
368 		if (unlikely(err))
369 			return err;
370 	}
371 #endif
372 
373 	hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
374 
375 	skb_postpush_rcsum(skb, hdr, sizeof(struct ipv6hdr) + hdrlen);
376 
377 	return 0;
378 }
379 
seg6_do_srh(struct sk_buff * skb,struct dst_entry * cache_dst)380 static int seg6_do_srh(struct sk_buff *skb, struct dst_entry *cache_dst)
381 {
382 	struct dst_entry *dst = skb_dst(skb);
383 	struct seg6_iptunnel_encap *tinfo;
384 	int proto, err = 0;
385 
386 	tinfo = seg6_encap_lwtunnel(dst->lwtstate);
387 
388 	switch (tinfo->mode) {
389 	case SEG6_IPTUN_MODE_INLINE:
390 		if (skb->protocol != htons(ETH_P_IPV6))
391 			return -EINVAL;
392 
393 		err = __seg6_do_srh_inline(skb, tinfo->srh, cache_dst);
394 		if (err)
395 			return err;
396 		break;
397 	case SEG6_IPTUN_MODE_ENCAP:
398 	case SEG6_IPTUN_MODE_ENCAP_RED:
399 		err = iptunnel_handle_offloads(skb, SKB_GSO_IPXIP6);
400 		if (err)
401 			return err;
402 
403 		if (skb->protocol == htons(ETH_P_IPV6))
404 			proto = IPPROTO_IPV6;
405 		else if (skb->protocol == htons(ETH_P_IP))
406 			proto = IPPROTO_IPIP;
407 		else
408 			return -EINVAL;
409 
410 		if (tinfo->mode == SEG6_IPTUN_MODE_ENCAP)
411 			err = __seg6_do_srh_encap(skb, tinfo->srh,
412 						  proto, cache_dst);
413 		else
414 			err = seg6_do_srh_encap_red(skb, tinfo->srh,
415 						    proto, cache_dst);
416 
417 		if (err)
418 			return err;
419 
420 		skb_set_inner_transport_header(skb, skb_transport_offset(skb));
421 		skb_set_inner_protocol(skb, skb->protocol);
422 		skb->protocol = htons(ETH_P_IPV6);
423 		break;
424 	case SEG6_IPTUN_MODE_L2ENCAP:
425 	case SEG6_IPTUN_MODE_L2ENCAP_RED:
426 		if (!skb_mac_header_was_set(skb))
427 			return -EINVAL;
428 
429 		if (pskb_expand_head(skb, skb->mac_len, 0, GFP_ATOMIC) < 0)
430 			return -ENOMEM;
431 
432 		skb_mac_header_rebuild(skb);
433 		skb_push(skb, skb->mac_len);
434 
435 		if (tinfo->mode == SEG6_IPTUN_MODE_L2ENCAP)
436 			err = __seg6_do_srh_encap(skb, tinfo->srh,
437 						  IPPROTO_ETHERNET,
438 						  cache_dst);
439 		else
440 			err = seg6_do_srh_encap_red(skb, tinfo->srh,
441 						    IPPROTO_ETHERNET,
442 						    cache_dst);
443 
444 		if (err)
445 			return err;
446 
447 		skb->protocol = htons(ETH_P_IPV6);
448 		break;
449 	}
450 
451 	skb_set_transport_header(skb, sizeof(struct ipv6hdr));
452 	nf_reset_ct(skb);
453 
454 	return 0;
455 }
456 
457 /* 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)458 int seg6_do_srh_inline(struct sk_buff *skb, struct ipv6_sr_hdr *osrh)
459 {
460 	return __seg6_do_srh_inline(skb, osrh, NULL);
461 }
462 EXPORT_SYMBOL_GPL(seg6_do_srh_inline);
463 
seg6_input_finish(struct net * net,struct sock * sk,struct sk_buff * skb)464 static int seg6_input_finish(struct net *net, struct sock *sk,
465 			     struct sk_buff *skb)
466 {
467 	return dst_input(skb);
468 }
469 
seg6_input_core(struct net * net,struct sock * sk,struct sk_buff * skb)470 static int seg6_input_core(struct net *net, struct sock *sk,
471 			   struct sk_buff *skb)
472 {
473 	struct dst_entry *orig_dst = skb_dst(skb);
474 	struct dst_entry *dst = NULL;
475 	struct lwtunnel_state *lwtst;
476 	struct seg6_lwt *slwt;
477 	int err;
478 
479 	/* We cannot dereference "orig_dst" once ip6_route_input() or
480 	 * skb_dst_drop() is called. However, in order to detect a dst loop, we
481 	 * need the address of its lwtstate. So, save the address of lwtstate
482 	 * now and use it later as a comparison.
483 	 */
484 	lwtst = orig_dst->lwtstate;
485 
486 	slwt = seg6_lwt_lwtunnel(lwtst);
487 
488 	local_bh_disable();
489 	dst = dst_cache_get(&slwt->cache);
490 	local_bh_enable();
491 
492 	err = seg6_do_srh(skb, dst);
493 	if (unlikely(err)) {
494 		dst_release(dst);
495 		goto drop;
496 	}
497 
498 	if (!dst) {
499 		ip6_route_input(skb);
500 		dst = skb_dst(skb);
501 
502 		/* cache only if we don't create a dst reference loop */
503 		if (!dst->error && lwtst != dst->lwtstate) {
504 			local_bh_disable();
505 			dst_cache_set_ip6(&slwt->cache, dst,
506 					  &ipv6_hdr(skb)->saddr);
507 			local_bh_enable();
508 		}
509 
510 		err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
511 		if (unlikely(err))
512 			goto drop;
513 	} else {
514 		skb_dst_drop(skb);
515 		skb_dst_set(skb, dst);
516 	}
517 
518 	if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
519 		return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
520 			       dev_net(skb->dev), NULL, skb, NULL,
521 			       skb_dst(skb)->dev, seg6_input_finish);
522 
523 	return seg6_input_finish(dev_net(skb->dev), NULL, skb);
524 drop:
525 	kfree_skb(skb);
526 	return err;
527 }
528 
seg6_input_nf(struct sk_buff * skb)529 static int seg6_input_nf(struct sk_buff *skb)
530 {
531 	struct net_device *dev = skb_dst(skb)->dev;
532 	struct net *net = dev_net(skb->dev);
533 
534 	switch (skb->protocol) {
535 	case htons(ETH_P_IP):
536 		return NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, NULL,
537 			       skb, NULL, dev, seg6_input_core);
538 	case htons(ETH_P_IPV6):
539 		return NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, NULL,
540 			       skb, NULL, dev, seg6_input_core);
541 	}
542 
543 	return -EINVAL;
544 }
545 
seg6_input(struct sk_buff * skb)546 static int seg6_input(struct sk_buff *skb)
547 {
548 	if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
549 		return seg6_input_nf(skb);
550 
551 	return seg6_input_core(dev_net(skb->dev), NULL, skb);
552 }
553 
seg6_output_core(struct net * net,struct sock * sk,struct sk_buff * skb)554 static int seg6_output_core(struct net *net, struct sock *sk,
555 			    struct sk_buff *skb)
556 {
557 	struct dst_entry *orig_dst = skb_dst(skb);
558 	struct dst_entry *dst = NULL;
559 	struct seg6_lwt *slwt;
560 	int err;
561 
562 	slwt = seg6_lwt_lwtunnel(orig_dst->lwtstate);
563 
564 	local_bh_disable();
565 	dst = dst_cache_get(&slwt->cache);
566 	local_bh_enable();
567 
568 	err = seg6_do_srh(skb, dst);
569 	if (unlikely(err))
570 		goto drop;
571 
572 	if (unlikely(!dst)) {
573 		struct ipv6hdr *hdr = ipv6_hdr(skb);
574 		struct flowi6 fl6;
575 
576 		memset(&fl6, 0, sizeof(fl6));
577 		fl6.daddr = hdr->daddr;
578 		fl6.saddr = hdr->saddr;
579 		fl6.flowlabel = ip6_flowinfo(hdr);
580 		fl6.flowi6_mark = skb->mark;
581 		fl6.flowi6_proto = hdr->nexthdr;
582 
583 		dst = ip6_route_output(net, NULL, &fl6);
584 		if (dst->error) {
585 			err = dst->error;
586 			goto drop;
587 		}
588 
589 		/* cache only if we don't create a dst reference loop */
590 		if (orig_dst->lwtstate != dst->lwtstate) {
591 			local_bh_disable();
592 			dst_cache_set_ip6(&slwt->cache, dst, &fl6.saddr);
593 			local_bh_enable();
594 		}
595 
596 		err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
597 		if (unlikely(err))
598 			goto drop;
599 	}
600 
601 	skb_dst_drop(skb);
602 	skb_dst_set(skb, dst);
603 
604 	if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
605 		return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, skb,
606 			       NULL, skb_dst(skb)->dev, dst_output);
607 
608 	return dst_output(net, sk, skb);
609 drop:
610 	dst_release(dst);
611 	kfree_skb(skb);
612 	return err;
613 }
614 
seg6_output_nf(struct net * net,struct sock * sk,struct sk_buff * skb)615 static int seg6_output_nf(struct net *net, struct sock *sk, struct sk_buff *skb)
616 {
617 	struct net_device *dev = skb_dst(skb)->dev;
618 
619 	switch (skb->protocol) {
620 	case htons(ETH_P_IP):
621 		return NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb,
622 			       NULL, dev, seg6_output_core);
623 	case htons(ETH_P_IPV6):
624 		return NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb,
625 			       NULL, dev, seg6_output_core);
626 	}
627 
628 	return -EINVAL;
629 }
630 
seg6_output(struct net * net,struct sock * sk,struct sk_buff * skb)631 static int seg6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
632 {
633 	if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
634 		return seg6_output_nf(net, sk, skb);
635 
636 	return seg6_output_core(net, sk, skb);
637 }
638 
seg6_build_state(struct net * net,struct nlattr * nla,unsigned int family,const void * cfg,struct lwtunnel_state ** ts,struct netlink_ext_ack * extack)639 static int seg6_build_state(struct net *net, struct nlattr *nla,
640 			    unsigned int family, const void *cfg,
641 			    struct lwtunnel_state **ts,
642 			    struct netlink_ext_ack *extack)
643 {
644 	struct nlattr *tb[SEG6_IPTUNNEL_MAX + 1];
645 	struct seg6_iptunnel_encap *tuninfo;
646 	struct lwtunnel_state *newts;
647 	int tuninfo_len, min_size;
648 	struct seg6_lwt *slwt;
649 	int err;
650 
651 	if (family != AF_INET && family != AF_INET6)
652 		return -EINVAL;
653 
654 	err = nla_parse_nested_deprecated(tb, SEG6_IPTUNNEL_MAX, nla,
655 					  seg6_iptunnel_policy, extack);
656 
657 	if (err < 0)
658 		return err;
659 
660 	if (!tb[SEG6_IPTUNNEL_SRH])
661 		return -EINVAL;
662 
663 	tuninfo = nla_data(tb[SEG6_IPTUNNEL_SRH]);
664 	tuninfo_len = nla_len(tb[SEG6_IPTUNNEL_SRH]);
665 
666 	/* tuninfo must contain at least the iptunnel encap structure,
667 	 * the SRH and one segment
668 	 */
669 	min_size = sizeof(*tuninfo) + sizeof(struct ipv6_sr_hdr) +
670 		   sizeof(struct in6_addr);
671 	if (tuninfo_len < min_size)
672 		return -EINVAL;
673 
674 	switch (tuninfo->mode) {
675 	case SEG6_IPTUN_MODE_INLINE:
676 		if (family != AF_INET6)
677 			return -EINVAL;
678 
679 		break;
680 	case SEG6_IPTUN_MODE_ENCAP:
681 		break;
682 	case SEG6_IPTUN_MODE_L2ENCAP:
683 		break;
684 	case SEG6_IPTUN_MODE_ENCAP_RED:
685 		break;
686 	case SEG6_IPTUN_MODE_L2ENCAP_RED:
687 		break;
688 	default:
689 		return -EINVAL;
690 	}
691 
692 	/* verify that SRH is consistent */
693 	if (!seg6_validate_srh(tuninfo->srh, tuninfo_len - sizeof(*tuninfo), false))
694 		return -EINVAL;
695 
696 	newts = lwtunnel_state_alloc(tuninfo_len + sizeof(*slwt));
697 	if (!newts)
698 		return -ENOMEM;
699 
700 	slwt = seg6_lwt_lwtunnel(newts);
701 
702 	err = dst_cache_init(&slwt->cache, GFP_ATOMIC);
703 	if (err) {
704 		kfree(newts);
705 		return err;
706 	}
707 
708 	memcpy(&slwt->tuninfo, tuninfo, tuninfo_len);
709 
710 	newts->type = LWTUNNEL_ENCAP_SEG6;
711 	newts->flags |= LWTUNNEL_STATE_INPUT_REDIRECT;
712 
713 	if (tuninfo->mode != SEG6_IPTUN_MODE_L2ENCAP)
714 		newts->flags |= LWTUNNEL_STATE_OUTPUT_REDIRECT;
715 
716 	newts->headroom = seg6_lwt_headroom(tuninfo);
717 
718 	*ts = newts;
719 
720 	return 0;
721 }
722 
seg6_destroy_state(struct lwtunnel_state * lwt)723 static void seg6_destroy_state(struct lwtunnel_state *lwt)
724 {
725 	dst_cache_destroy(&seg6_lwt_lwtunnel(lwt)->cache);
726 }
727 
seg6_fill_encap_info(struct sk_buff * skb,struct lwtunnel_state * lwtstate)728 static int seg6_fill_encap_info(struct sk_buff *skb,
729 				struct lwtunnel_state *lwtstate)
730 {
731 	struct seg6_iptunnel_encap *tuninfo = seg6_encap_lwtunnel(lwtstate);
732 
733 	if (nla_put_srh(skb, SEG6_IPTUNNEL_SRH, tuninfo))
734 		return -EMSGSIZE;
735 
736 	return 0;
737 }
738 
seg6_encap_nlsize(struct lwtunnel_state * lwtstate)739 static int seg6_encap_nlsize(struct lwtunnel_state *lwtstate)
740 {
741 	struct seg6_iptunnel_encap *tuninfo = seg6_encap_lwtunnel(lwtstate);
742 
743 	return nla_total_size(SEG6_IPTUN_ENCAP_SIZE(tuninfo));
744 }
745 
seg6_encap_cmp(struct lwtunnel_state * a,struct lwtunnel_state * b)746 static int seg6_encap_cmp(struct lwtunnel_state *a, struct lwtunnel_state *b)
747 {
748 	struct seg6_iptunnel_encap *a_hdr = seg6_encap_lwtunnel(a);
749 	struct seg6_iptunnel_encap *b_hdr = seg6_encap_lwtunnel(b);
750 	int len = SEG6_IPTUN_ENCAP_SIZE(a_hdr);
751 
752 	if (len != SEG6_IPTUN_ENCAP_SIZE(b_hdr))
753 		return 1;
754 
755 	return memcmp(a_hdr, b_hdr, len);
756 }
757 
758 static const struct lwtunnel_encap_ops seg6_iptun_ops = {
759 	.build_state = seg6_build_state,
760 	.destroy_state = seg6_destroy_state,
761 	.output = seg6_output,
762 	.input = seg6_input,
763 	.fill_encap = seg6_fill_encap_info,
764 	.get_encap_size = seg6_encap_nlsize,
765 	.cmp_encap = seg6_encap_cmp,
766 	.owner = THIS_MODULE,
767 };
768 
seg6_iptunnel_init(void)769 int __init seg6_iptunnel_init(void)
770 {
771 	return lwtunnel_encap_add_ops(&seg6_iptun_ops, LWTUNNEL_ENCAP_SEG6);
772 }
773 
seg6_iptunnel_exit(void)774 void seg6_iptunnel_exit(void)
775 {
776 	lwtunnel_encap_del_ops(&seg6_iptun_ops, LWTUNNEL_ENCAP_SEG6);
777 }
778