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
30 static size_t seg6_lwt_headroom(struct seg6_iptunnel_encap *tuninfo)
31 {
32 int head = 0;
33
34 switch (tuninfo->mode) {
35 case SEG6_IPTUN_MODE_INLINE:
36 break;
37 case SEG6_IPTUN_MODE_ENCAP:
38 head = sizeof(struct ipv6hdr);
39 break;
40 case SEG6_IPTUN_MODE_L2ENCAP:
41 return 0;
42 }
43
44 return ((tuninfo->srh->hdrlen + 1) << 3) + head;
45 }
46
47 struct seg6_lwt {
48 struct dst_cache cache;
49 struct seg6_iptunnel_encap tuninfo[];
50 };
51
52 static inline struct seg6_lwt *seg6_lwt_lwtunnel(struct lwtunnel_state *lwt)
53 {
54 return (struct seg6_lwt *)lwt->data;
55 }
56
57 static inline struct seg6_iptunnel_encap *
58 seg6_encap_lwtunnel(struct lwtunnel_state *lwt)
59 {
60 return seg6_lwt_lwtunnel(lwt)->tuninfo;
61 }
62
63 static const struct nla_policy seg6_iptunnel_policy[SEG6_IPTUNNEL_MAX + 1] = {
64 [SEG6_IPTUNNEL_SRH] = { .type = NLA_BINARY },
65 };
66
67 static int nla_put_srh(struct sk_buff *skb, int attrtype,
68 struct seg6_iptunnel_encap *tuninfo)
69 {
70 struct seg6_iptunnel_encap *data;
71 struct nlattr *nla;
72 int len;
73
74 len = SEG6_IPTUN_ENCAP_SIZE(tuninfo);
75
76 nla = nla_reserve(skb, attrtype, len);
77 if (!nla)
78 return -EMSGSIZE;
79
80 data = nla_data(nla);
81 memcpy(data, tuninfo, len);
82
83 return 0;
84 }
85
86 static void set_tun_src(struct net *net, struct net_device *dev,
87 struct in6_addr *daddr, struct in6_addr *saddr)
88 {
89 struct seg6_pernet_data *sdata = seg6_pernet(net);
90 struct in6_addr *tun_src;
91
92 rcu_read_lock();
93
94 tun_src = rcu_dereference(sdata->tun_src);
95
96 if (!ipv6_addr_any(tun_src)) {
97 memcpy(saddr, tun_src, sizeof(struct in6_addr));
98 } else {
99 ipv6_dev_get_saddr(net, dev, daddr, IPV6_PREFER_SRC_PUBLIC,
100 saddr);
101 }
102
103 rcu_read_unlock();
104 }
105
106 /* Compute flowlabel for outer IPv6 header */
107 static __be32 seg6_make_flowlabel(struct net *net, struct sk_buff *skb,
108 struct ipv6hdr *inner_hdr)
109 {
110 int do_flowlabel = net->ipv6.sysctl.seg6_flowlabel;
111 __be32 flowlabel = 0;
112 u32 hash;
113
114 if (do_flowlabel > 0) {
115 hash = skb_get_hash(skb);
116 hash = rol32(hash, 16);
117 flowlabel = (__force __be32)hash & IPV6_FLOWLABEL_MASK;
118 } else if (!do_flowlabel && skb->protocol == htons(ETH_P_IPV6)) {
119 flowlabel = ip6_flowlabel(inner_hdr);
120 }
121 return flowlabel;
122 }
123
124 /* encapsulate an IPv6 packet within an outer IPv6 header with a given SRH */
125 int seg6_do_srh_encap(struct sk_buff *skb, struct ipv6_sr_hdr *osrh, int proto)
126 {
127 struct dst_entry *dst = skb_dst(skb);
128 struct net *net = dev_net(dst->dev);
129 struct ipv6hdr *hdr, *inner_hdr;
130 struct ipv6_sr_hdr *isrh;
131 int hdrlen, tot_len, err;
132 __be32 flowlabel;
133
134 hdrlen = (osrh->hdrlen + 1) << 3;
135 tot_len = hdrlen + sizeof(*hdr);
136
137 err = skb_cow_head(skb, tot_len + skb->mac_len);
138 if (unlikely(err))
139 return err;
140
141 inner_hdr = ipv6_hdr(skb);
142 flowlabel = seg6_make_flowlabel(net, skb, inner_hdr);
143
144 skb_push(skb, tot_len);
145 skb_reset_network_header(skb);
146 skb_mac_header_rebuild(skb);
147 hdr = ipv6_hdr(skb);
148
149 /* inherit tc, flowlabel and hlim
150 * hlim will be decremented in ip6_forward() afterwards and
151 * decapsulation will overwrite inner hlim with outer hlim
152 */
153
154 if (skb->protocol == htons(ETH_P_IPV6)) {
155 ip6_flow_hdr(hdr, ip6_tclass(ip6_flowinfo(inner_hdr)),
156 flowlabel);
157 hdr->hop_limit = inner_hdr->hop_limit;
158 } else {
159 ip6_flow_hdr(hdr, 0, flowlabel);
160 hdr->hop_limit = ip6_dst_hoplimit(skb_dst(skb));
161
162 memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
163 }
164
165 hdr->nexthdr = NEXTHDR_ROUTING;
166
167 isrh = (void *)hdr + sizeof(*hdr);
168 memcpy(isrh, osrh, hdrlen);
169
170 isrh->nexthdr = proto;
171
172 hdr->daddr = isrh->segments[isrh->first_segment];
173 set_tun_src(net, dst->dev, &hdr->daddr, &hdr->saddr);
174
175 #ifdef CONFIG_IPV6_SEG6_HMAC
176 if (sr_has_hmac(isrh)) {
177 err = seg6_push_hmac(net, &hdr->saddr, isrh);
178 if (unlikely(err))
179 return err;
180 }
181 #endif
182
183 skb_postpush_rcsum(skb, hdr, tot_len);
184
185 return 0;
186 }
187 EXPORT_SYMBOL_GPL(seg6_do_srh_encap);
188
189 /* insert an SRH within an IPv6 packet, just after the IPv6 header */
190 int seg6_do_srh_inline(struct sk_buff *skb, struct ipv6_sr_hdr *osrh)
191 {
192 struct ipv6hdr *hdr, *oldhdr;
193 struct ipv6_sr_hdr *isrh;
194 int hdrlen, err;
195
196 hdrlen = (osrh->hdrlen + 1) << 3;
197
198 err = skb_cow_head(skb, hdrlen + skb->mac_len);
199 if (unlikely(err))
200 return err;
201
202 oldhdr = ipv6_hdr(skb);
203
204 skb_pull(skb, sizeof(struct ipv6hdr));
205 skb_postpull_rcsum(skb, skb_network_header(skb),
206 sizeof(struct ipv6hdr));
207
208 skb_push(skb, sizeof(struct ipv6hdr) + hdrlen);
209 skb_reset_network_header(skb);
210 skb_mac_header_rebuild(skb);
211
212 hdr = ipv6_hdr(skb);
213
214 memmove(hdr, oldhdr, sizeof(*hdr));
215
216 isrh = (void *)hdr + sizeof(*hdr);
217 memcpy(isrh, osrh, hdrlen);
218
219 isrh->nexthdr = hdr->nexthdr;
220 hdr->nexthdr = NEXTHDR_ROUTING;
221
222 isrh->segments[0] = hdr->daddr;
223 hdr->daddr = isrh->segments[isrh->first_segment];
224
225 #ifdef CONFIG_IPV6_SEG6_HMAC
226 if (sr_has_hmac(isrh)) {
227 struct net *net = dev_net(skb_dst(skb)->dev);
228
229 err = seg6_push_hmac(net, &hdr->saddr, isrh);
230 if (unlikely(err))
231 return err;
232 }
233 #endif
234
235 skb_postpush_rcsum(skb, hdr, sizeof(struct ipv6hdr) + hdrlen);
236
237 return 0;
238 }
239 EXPORT_SYMBOL_GPL(seg6_do_srh_inline);
240
241 static int seg6_do_srh(struct sk_buff *skb)
242 {
243 struct dst_entry *dst = skb_dst(skb);
244 struct seg6_iptunnel_encap *tinfo;
245 int proto, err = 0;
246
247 tinfo = seg6_encap_lwtunnel(dst->lwtstate);
248
249 switch (tinfo->mode) {
250 case SEG6_IPTUN_MODE_INLINE:
251 if (skb->protocol != htons(ETH_P_IPV6))
252 return -EINVAL;
253
254 err = seg6_do_srh_inline(skb, tinfo->srh);
255 if (err)
256 return err;
257 break;
258 case SEG6_IPTUN_MODE_ENCAP:
259 err = iptunnel_handle_offloads(skb, SKB_GSO_IPXIP6);
260 if (err)
261 return err;
262
263 if (skb->protocol == htons(ETH_P_IPV6))
264 proto = IPPROTO_IPV6;
265 else if (skb->protocol == htons(ETH_P_IP))
266 proto = IPPROTO_IPIP;
267 else
268 return -EINVAL;
269
270 err = seg6_do_srh_encap(skb, tinfo->srh, proto);
271 if (err)
272 return err;
273
274 skb_set_inner_transport_header(skb, skb_transport_offset(skb));
275 skb_set_inner_protocol(skb, skb->protocol);
276 skb->protocol = htons(ETH_P_IPV6);
277 break;
278 case SEG6_IPTUN_MODE_L2ENCAP:
279 if (!skb_mac_header_was_set(skb))
280 return -EINVAL;
281
282 if (pskb_expand_head(skb, skb->mac_len, 0, GFP_ATOMIC) < 0)
283 return -ENOMEM;
284
285 skb_mac_header_rebuild(skb);
286 skb_push(skb, skb->mac_len);
287
288 err = seg6_do_srh_encap(skb, tinfo->srh, IPPROTO_ETHERNET);
289 if (err)
290 return err;
291
292 skb->protocol = htons(ETH_P_IPV6);
293 break;
294 }
295
296 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
297 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
298
299 return 0;
300 }
301
302 static int seg6_input(struct sk_buff *skb)
303 {
304 struct dst_entry *orig_dst = skb_dst(skb);
305 struct dst_entry *dst = NULL;
306 struct seg6_lwt *slwt;
307 int err;
308
309 err = seg6_do_srh(skb);
310 if (unlikely(err)) {
311 kfree_skb(skb);
312 return err;
313 }
314
315 slwt = seg6_lwt_lwtunnel(orig_dst->lwtstate);
316
317 preempt_disable();
318 dst = dst_cache_get(&slwt->cache);
319 preempt_enable();
320
321 skb_dst_drop(skb);
322
323 if (!dst) {
324 ip6_route_input(skb);
325 dst = skb_dst(skb);
326 if (!dst->error) {
327 preempt_disable();
328 dst_cache_set_ip6(&slwt->cache, dst,
329 &ipv6_hdr(skb)->saddr);
330 preempt_enable();
331 }
332 } else {
333 skb_dst_set(skb, dst);
334 }
335
336 err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
337 if (unlikely(err))
338 return err;
339
340 return dst_input(skb);
341 }
342
343 static int seg6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
344 {
345 struct dst_entry *orig_dst = skb_dst(skb);
346 struct dst_entry *dst = NULL;
347 struct seg6_lwt *slwt;
348 int err = -EINVAL;
349
350 err = seg6_do_srh(skb);
351 if (unlikely(err))
352 goto drop;
353
354 slwt = seg6_lwt_lwtunnel(orig_dst->lwtstate);
355
356 preempt_disable();
357 dst = dst_cache_get(&slwt->cache);
358 preempt_enable();
359
360 if (unlikely(!dst)) {
361 struct ipv6hdr *hdr = ipv6_hdr(skb);
362 struct flowi6 fl6;
363
364 memset(&fl6, 0, sizeof(fl6));
365 fl6.daddr = hdr->daddr;
366 fl6.saddr = hdr->saddr;
367 fl6.flowlabel = ip6_flowinfo(hdr);
368 fl6.flowi6_mark = skb->mark;
369 fl6.flowi6_proto = hdr->nexthdr;
370
371 dst = ip6_route_output(net, NULL, &fl6);
372 if (dst->error) {
373 err = dst->error;
374 dst_release(dst);
375 goto drop;
376 }
377
378 preempt_disable();
379 dst_cache_set_ip6(&slwt->cache, dst, &fl6.saddr);
380 preempt_enable();
381 }
382
383 skb_dst_drop(skb);
384 skb_dst_set(skb, dst);
385
386 err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
387 if (unlikely(err))
388 goto drop;
389
390 return dst_output(net, sk, skb);
391 drop:
392 kfree_skb(skb);
393 return err;
394 }
395
396 static int seg6_build_state(struct net *net, struct nlattr *nla,
397 unsigned int family, const void *cfg,
398 struct lwtunnel_state **ts,
399 struct netlink_ext_ack *extack)
400 {
401 struct nlattr *tb[SEG6_IPTUNNEL_MAX + 1];
402 struct seg6_iptunnel_encap *tuninfo;
403 struct lwtunnel_state *newts;
404 int tuninfo_len, min_size;
405 struct seg6_lwt *slwt;
406 int err;
407
408 if (family != AF_INET && family != AF_INET6)
409 return -EINVAL;
410
411 err = nla_parse_nested_deprecated(tb, SEG6_IPTUNNEL_MAX, nla,
412 seg6_iptunnel_policy, extack);
413
414 if (err < 0)
415 return err;
416
417 if (!tb[SEG6_IPTUNNEL_SRH])
418 return -EINVAL;
419
420 tuninfo = nla_data(tb[SEG6_IPTUNNEL_SRH]);
421 tuninfo_len = nla_len(tb[SEG6_IPTUNNEL_SRH]);
422
423 /* tuninfo must contain at least the iptunnel encap structure,
424 * the SRH and one segment
425 */
426 min_size = sizeof(*tuninfo) + sizeof(struct ipv6_sr_hdr) +
427 sizeof(struct in6_addr);
428 if (tuninfo_len < min_size)
429 return -EINVAL;
430
431 switch (tuninfo->mode) {
432 case SEG6_IPTUN_MODE_INLINE:
433 if (family != AF_INET6)
434 return -EINVAL;
435
436 break;
437 case SEG6_IPTUN_MODE_ENCAP:
438 break;
439 case SEG6_IPTUN_MODE_L2ENCAP:
440 break;
441 default:
442 return -EINVAL;
443 }
444
445 /* verify that SRH is consistent */
446 if (!seg6_validate_srh(tuninfo->srh, tuninfo_len - sizeof(*tuninfo), false))
447 return -EINVAL;
448
449 newts = lwtunnel_state_alloc(tuninfo_len + sizeof(*slwt));
450 if (!newts)
451 return -ENOMEM;
452
453 slwt = seg6_lwt_lwtunnel(newts);
454
455 err = dst_cache_init(&slwt->cache, GFP_ATOMIC);
456 if (err) {
457 kfree(newts);
458 return err;
459 }
460
461 memcpy(&slwt->tuninfo, tuninfo, tuninfo_len);
462
463 newts->type = LWTUNNEL_ENCAP_SEG6;
464 newts->flags |= LWTUNNEL_STATE_INPUT_REDIRECT;
465
466 if (tuninfo->mode != SEG6_IPTUN_MODE_L2ENCAP)
467 newts->flags |= LWTUNNEL_STATE_OUTPUT_REDIRECT;
468
469 newts->headroom = seg6_lwt_headroom(tuninfo);
470
471 *ts = newts;
472
473 return 0;
474 }
475
476 static void seg6_destroy_state(struct lwtunnel_state *lwt)
477 {
478 dst_cache_destroy(&seg6_lwt_lwtunnel(lwt)->cache);
479 }
480
481 static int seg6_fill_encap_info(struct sk_buff *skb,
482 struct lwtunnel_state *lwtstate)
483 {
484 struct seg6_iptunnel_encap *tuninfo = seg6_encap_lwtunnel(lwtstate);
485
486 if (nla_put_srh(skb, SEG6_IPTUNNEL_SRH, tuninfo))
487 return -EMSGSIZE;
488
489 return 0;
490 }
491
492 static int seg6_encap_nlsize(struct lwtunnel_state *lwtstate)
493 {
494 struct seg6_iptunnel_encap *tuninfo = seg6_encap_lwtunnel(lwtstate);
495
496 return nla_total_size(SEG6_IPTUN_ENCAP_SIZE(tuninfo));
497 }
498
499 static int seg6_encap_cmp(struct lwtunnel_state *a, struct lwtunnel_state *b)
500 {
501 struct seg6_iptunnel_encap *a_hdr = seg6_encap_lwtunnel(a);
502 struct seg6_iptunnel_encap *b_hdr = seg6_encap_lwtunnel(b);
503 int len = SEG6_IPTUN_ENCAP_SIZE(a_hdr);
504
505 if (len != SEG6_IPTUN_ENCAP_SIZE(b_hdr))
506 return 1;
507
508 return memcmp(a_hdr, b_hdr, len);
509 }
510
511 static const struct lwtunnel_encap_ops seg6_iptun_ops = {
512 .build_state = seg6_build_state,
513 .destroy_state = seg6_destroy_state,
514 .output = seg6_output,
515 .input = seg6_input,
516 .fill_encap = seg6_fill_encap_info,
517 .get_encap_size = seg6_encap_nlsize,
518 .cmp_encap = seg6_encap_cmp,
519 .owner = THIS_MODULE,
520 };
521
522 int __init seg6_iptunnel_init(void)
523 {
524 return lwtunnel_encap_add_ops(&seg6_iptun_ops, LWTUNNEL_ENCAP_SEG6);
525 }
526
527 void seg6_iptunnel_exit(void)
528 {
529 lwtunnel_encap_del_ops(&seg6_iptun_ops, LWTUNNEL_ENCAP_SEG6);
530 }
531
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.