1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * SR-IPv6 implementation
6 * David Lebrun <david.lebrun@uclouvain.be>
9 #include <linux/types.h>
10 #include <linux/skbuff.h>
11 #include <linux/net.h>
12 #include <linux/module.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>
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>
29 #include <linux/netfilter.h>
31 static size_t seg6_lwt_headroom(struct seg6_iptunnel_encap *tuninfo)
35 switch (tuninfo->mode) {
36 case SEG6_IPTUN_MODE_INLINE:
38 case SEG6_IPTUN_MODE_ENCAP:
39 case SEG6_IPTUN_MODE_ENCAP_RED:
40 head = sizeof(struct ipv6hdr);
42 case SEG6_IPTUN_MODE_L2ENCAP:
43 case SEG6_IPTUN_MODE_L2ENCAP_RED:
47 return ((tuninfo->srh->hdrlen + 1) << 3) + head;
51 struct dst_cache cache;
52 struct seg6_iptunnel_encap tuninfo[];
55 static inline struct seg6_lwt *seg6_lwt_lwtunnel(struct lwtunnel_state *lwt)
57 return (struct seg6_lwt *)lwt->data;
60 static inline struct seg6_iptunnel_encap *
61 seg6_encap_lwtunnel(struct lwtunnel_state *lwt)
63 return seg6_lwt_lwtunnel(lwt)->tuninfo;
66 static const struct nla_policy seg6_iptunnel_policy[SEG6_IPTUNNEL_MAX + 1] = {
67 [SEG6_IPTUNNEL_SRH] = { .type = NLA_BINARY },
70 static int nla_put_srh(struct sk_buff *skb, int attrtype,
71 struct seg6_iptunnel_encap *tuninfo)
73 struct seg6_iptunnel_encap *data;
77 len = SEG6_IPTUN_ENCAP_SIZE(tuninfo);
79 nla = nla_reserve(skb, attrtype, len);
84 memcpy(data, tuninfo, len);
89 static void set_tun_src(struct net *net, struct net_device *dev,
90 struct in6_addr *daddr, struct in6_addr *saddr)
92 struct seg6_pernet_data *sdata = seg6_pernet(net);
93 struct in6_addr *tun_src;
97 tun_src = rcu_dereference(sdata->tun_src);
99 if (!ipv6_addr_any(tun_src)) {
100 memcpy(saddr, tun_src, sizeof(struct in6_addr));
102 ipv6_dev_get_saddr(net, dev, daddr, IPV6_PREFER_SRC_PUBLIC,
109 /* Compute flowlabel for outer IPv6 header */
110 static __be32 seg6_make_flowlabel(struct net *net, struct sk_buff *skb,
111 struct ipv6hdr *inner_hdr)
113 int do_flowlabel = net->ipv6.sysctl.seg6_flowlabel;
114 __be32 flowlabel = 0;
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);
127 /* encapsulate an IPv6 packet within an outer IPv6 header with a given SRH */
128 int seg6_do_srh_encap(struct sk_buff *skb, struct ipv6_sr_hdr *osrh, int proto)
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;
137 hdrlen = (osrh->hdrlen + 1) << 3;
138 tot_len = hdrlen + sizeof(*hdr);
140 err = skb_cow_head(skb, tot_len + skb->mac_len);
144 inner_hdr = ipv6_hdr(skb);
145 flowlabel = seg6_make_flowlabel(net, skb, inner_hdr);
147 skb_push(skb, tot_len);
148 skb_reset_network_header(skb);
149 skb_mac_header_rebuild(skb);
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
157 if (skb->protocol == htons(ETH_P_IPV6)) {
158 ip6_flow_hdr(hdr, ip6_tclass(ip6_flowinfo(inner_hdr)),
160 hdr->hop_limit = inner_hdr->hop_limit;
162 ip6_flow_hdr(hdr, 0, flowlabel);
163 hdr->hop_limit = ip6_dst_hoplimit(skb_dst(skb));
165 memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
167 /* the control block has been erased, so we have to set the
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.:
173 IP6CB(skb)->iif = skb->skb_iif;
176 hdr->nexthdr = NEXTHDR_ROUTING;
178 isrh = (void *)hdr + sizeof(*hdr);
179 memcpy(isrh, osrh, hdrlen);
181 isrh->nexthdr = proto;
183 hdr->daddr = isrh->segments[isrh->first_segment];
184 set_tun_src(net, dst->dev, &hdr->daddr, &hdr->saddr);
186 #ifdef CONFIG_IPV6_SEG6_HMAC
187 if (sr_has_hmac(isrh)) {
188 err = seg6_push_hmac(net, &hdr->saddr, isrh);
194 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
196 skb_postpush_rcsum(skb, hdr, tot_len);
200 EXPORT_SYMBOL_GPL(seg6_do_srh_encap);
202 /* encapsulate an IPv6 packet within an outer IPv6 header with reduced SRH */
203 static int seg6_do_srh_encap_red(struct sk_buff *skb,
204 struct ipv6_sr_hdr *osrh, int proto)
206 __u8 first_seg = osrh->first_segment;
207 struct dst_entry *dst = skb_dst(skb);
208 struct net *net = dev_net(dst->dev);
209 struct ipv6hdr *hdr, *inner_hdr;
210 int hdrlen = ipv6_optlen(osrh);
211 int red_tlv_offset, tlv_offset;
212 struct ipv6_sr_hdr *isrh;
213 bool skip_srh = false;
220 red_hdrlen = hdrlen - sizeof(struct in6_addr);
222 /* NOTE: if tag/flags and/or other TLVs are introduced in the
223 * seg6_iptunnel infrastructure, they should be considered when
224 * deciding to skip the SRH.
226 skip_srh = !sr_has_hmac(osrh);
228 red_hdrlen = skip_srh ? 0 : hdrlen;
231 tot_len = red_hdrlen + sizeof(struct ipv6hdr);
233 err = skb_cow_head(skb, tot_len + skb->mac_len);
237 inner_hdr = ipv6_hdr(skb);
238 flowlabel = seg6_make_flowlabel(net, skb, inner_hdr);
240 skb_push(skb, tot_len);
241 skb_reset_network_header(skb);
242 skb_mac_header_rebuild(skb);
245 /* based on seg6_do_srh_encap() */
246 if (skb->protocol == htons(ETH_P_IPV6)) {
247 ip6_flow_hdr(hdr, ip6_tclass(ip6_flowinfo(inner_hdr)),
249 hdr->hop_limit = inner_hdr->hop_limit;
251 ip6_flow_hdr(hdr, 0, flowlabel);
252 hdr->hop_limit = ip6_dst_hoplimit(skb_dst(skb));
254 memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
255 IP6CB(skb)->iif = skb->skb_iif;
258 /* no matter if we have to skip the SRH or not, the first segment
259 * always comes in the pushed IPv6 header.
261 hdr->daddr = osrh->segments[first_seg];
264 hdr->nexthdr = proto;
266 set_tun_src(net, dst->dev, &hdr->daddr, &hdr->saddr);
270 /* we cannot skip the SRH, slow path */
272 hdr->nexthdr = NEXTHDR_ROUTING;
273 isrh = (void *)hdr + sizeof(struct ipv6hdr);
275 if (unlikely(!first_seg)) {
276 /* this is a very rare case; we have only one SID but
277 * we cannot skip the SRH since we are carrying some
280 memcpy(isrh, osrh, hdrlen);
284 tlv_offset = sizeof(*osrh) + (first_seg + 1) * sizeof(struct in6_addr);
285 red_tlv_offset = tlv_offset - sizeof(struct in6_addr);
287 memcpy(isrh, osrh, red_tlv_offset);
289 tlvs_len = hdrlen - tlv_offset;
290 if (unlikely(tlvs_len > 0)) {
291 const void *s = (const void *)osrh + tlv_offset;
292 void *d = (void *)isrh + red_tlv_offset;
294 memcpy(d, s, tlvs_len);
297 --isrh->first_segment;
301 isrh->nexthdr = proto;
302 set_tun_src(net, dst->dev, &hdr->daddr, &hdr->saddr);
304 #ifdef CONFIG_IPV6_SEG6_HMAC
305 if (unlikely(!skip_srh && sr_has_hmac(isrh))) {
306 err = seg6_push_hmac(net, &hdr->saddr, isrh);
313 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
315 skb_postpush_rcsum(skb, hdr, tot_len);
320 /* insert an SRH within an IPv6 packet, just after the IPv6 header */
321 int seg6_do_srh_inline(struct sk_buff *skb, struct ipv6_sr_hdr *osrh)
323 struct ipv6hdr *hdr, *oldhdr;
324 struct ipv6_sr_hdr *isrh;
327 hdrlen = (osrh->hdrlen + 1) << 3;
329 err = skb_cow_head(skb, hdrlen + skb->mac_len);
333 oldhdr = ipv6_hdr(skb);
335 skb_pull(skb, sizeof(struct ipv6hdr));
336 skb_postpull_rcsum(skb, skb_network_header(skb),
337 sizeof(struct ipv6hdr));
339 skb_push(skb, sizeof(struct ipv6hdr) + hdrlen);
340 skb_reset_network_header(skb);
341 skb_mac_header_rebuild(skb);
345 memmove(hdr, oldhdr, sizeof(*hdr));
347 isrh = (void *)hdr + sizeof(*hdr);
348 memcpy(isrh, osrh, hdrlen);
350 isrh->nexthdr = hdr->nexthdr;
351 hdr->nexthdr = NEXTHDR_ROUTING;
353 isrh->segments[0] = hdr->daddr;
354 hdr->daddr = isrh->segments[isrh->first_segment];
356 #ifdef CONFIG_IPV6_SEG6_HMAC
357 if (sr_has_hmac(isrh)) {
358 struct net *net = dev_net(skb_dst(skb)->dev);
360 err = seg6_push_hmac(net, &hdr->saddr, isrh);
366 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
368 skb_postpush_rcsum(skb, hdr, sizeof(struct ipv6hdr) + hdrlen);
372 EXPORT_SYMBOL_GPL(seg6_do_srh_inline);
374 static int seg6_do_srh(struct sk_buff *skb)
376 struct dst_entry *dst = skb_dst(skb);
377 struct seg6_iptunnel_encap *tinfo;
380 tinfo = seg6_encap_lwtunnel(dst->lwtstate);
382 switch (tinfo->mode) {
383 case SEG6_IPTUN_MODE_INLINE:
384 if (skb->protocol != htons(ETH_P_IPV6))
387 err = seg6_do_srh_inline(skb, tinfo->srh);
391 case SEG6_IPTUN_MODE_ENCAP:
392 case SEG6_IPTUN_MODE_ENCAP_RED:
393 err = iptunnel_handle_offloads(skb, SKB_GSO_IPXIP6);
397 if (skb->protocol == htons(ETH_P_IPV6))
398 proto = IPPROTO_IPV6;
399 else if (skb->protocol == htons(ETH_P_IP))
400 proto = IPPROTO_IPIP;
404 if (tinfo->mode == SEG6_IPTUN_MODE_ENCAP)
405 err = seg6_do_srh_encap(skb, tinfo->srh, proto);
407 err = seg6_do_srh_encap_red(skb, tinfo->srh, proto);
412 skb_set_inner_transport_header(skb, skb_transport_offset(skb));
413 skb_set_inner_protocol(skb, skb->protocol);
414 skb->protocol = htons(ETH_P_IPV6);
416 case SEG6_IPTUN_MODE_L2ENCAP:
417 case SEG6_IPTUN_MODE_L2ENCAP_RED:
418 if (!skb_mac_header_was_set(skb))
421 if (pskb_expand_head(skb, skb->mac_len, 0, GFP_ATOMIC) < 0)
424 skb_mac_header_rebuild(skb);
425 skb_push(skb, skb->mac_len);
427 if (tinfo->mode == SEG6_IPTUN_MODE_L2ENCAP)
428 err = seg6_do_srh_encap(skb, tinfo->srh,
431 err = seg6_do_srh_encap_red(skb, tinfo->srh,
437 skb->protocol = htons(ETH_P_IPV6);
441 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
447 static int seg6_input_finish(struct net *net, struct sock *sk,
450 return dst_input(skb);
453 static int seg6_input_core(struct net *net, struct sock *sk,
456 struct dst_entry *orig_dst = skb_dst(skb);
457 struct dst_entry *dst = NULL;
458 struct seg6_lwt *slwt;
461 err = seg6_do_srh(skb);
467 slwt = seg6_lwt_lwtunnel(orig_dst->lwtstate);
470 dst = dst_cache_get(&slwt->cache);
476 ip6_route_input(skb);
480 dst_cache_set_ip6(&slwt->cache, dst,
481 &ipv6_hdr(skb)->saddr);
485 skb_dst_set(skb, dst);
488 err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
492 if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
493 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
494 dev_net(skb->dev), NULL, skb, NULL,
495 skb_dst(skb)->dev, seg6_input_finish);
497 return seg6_input_finish(dev_net(skb->dev), NULL, skb);
500 static int seg6_input_nf(struct sk_buff *skb)
502 struct net_device *dev = skb_dst(skb)->dev;
503 struct net *net = dev_net(skb->dev);
505 switch (skb->protocol) {
506 case htons(ETH_P_IP):
507 return NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, NULL,
508 skb, NULL, dev, seg6_input_core);
509 case htons(ETH_P_IPV6):
510 return NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, NULL,
511 skb, NULL, dev, seg6_input_core);
517 static int seg6_input(struct sk_buff *skb)
519 if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
520 return seg6_input_nf(skb);
522 return seg6_input_core(dev_net(skb->dev), NULL, skb);
525 static int seg6_output_core(struct net *net, struct sock *sk,
528 struct dst_entry *orig_dst = skb_dst(skb);
529 struct dst_entry *dst = NULL;
530 struct seg6_lwt *slwt;
533 err = seg6_do_srh(skb);
537 slwt = seg6_lwt_lwtunnel(orig_dst->lwtstate);
540 dst = dst_cache_get(&slwt->cache);
543 if (unlikely(!dst)) {
544 struct ipv6hdr *hdr = ipv6_hdr(skb);
547 memset(&fl6, 0, sizeof(fl6));
548 fl6.daddr = hdr->daddr;
549 fl6.saddr = hdr->saddr;
550 fl6.flowlabel = ip6_flowinfo(hdr);
551 fl6.flowi6_mark = skb->mark;
552 fl6.flowi6_proto = hdr->nexthdr;
554 dst = ip6_route_output(net, NULL, &fl6);
562 dst_cache_set_ip6(&slwt->cache, dst, &fl6.saddr);
567 skb_dst_set(skb, dst);
569 err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
573 if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
574 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, skb,
575 NULL, skb_dst(skb)->dev, dst_output);
577 return dst_output(net, sk, skb);
583 static int seg6_output_nf(struct net *net, struct sock *sk, struct sk_buff *skb)
585 struct net_device *dev = skb_dst(skb)->dev;
587 switch (skb->protocol) {
588 case htons(ETH_P_IP):
589 return NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb,
590 NULL, dev, seg6_output_core);
591 case htons(ETH_P_IPV6):
592 return NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb,
593 NULL, dev, seg6_output_core);
599 static int seg6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
601 if (static_branch_unlikely(&nf_hooks_lwtunnel_enabled))
602 return seg6_output_nf(net, sk, skb);
604 return seg6_output_core(net, sk, skb);
607 static int seg6_build_state(struct net *net, struct nlattr *nla,
608 unsigned int family, const void *cfg,
609 struct lwtunnel_state **ts,
610 struct netlink_ext_ack *extack)
612 struct nlattr *tb[SEG6_IPTUNNEL_MAX + 1];
613 struct seg6_iptunnel_encap *tuninfo;
614 struct lwtunnel_state *newts;
615 int tuninfo_len, min_size;
616 struct seg6_lwt *slwt;
619 if (family != AF_INET && family != AF_INET6)
622 err = nla_parse_nested_deprecated(tb, SEG6_IPTUNNEL_MAX, nla,
623 seg6_iptunnel_policy, extack);
628 if (!tb[SEG6_IPTUNNEL_SRH])
631 tuninfo = nla_data(tb[SEG6_IPTUNNEL_SRH]);
632 tuninfo_len = nla_len(tb[SEG6_IPTUNNEL_SRH]);
634 /* tuninfo must contain at least the iptunnel encap structure,
635 * the SRH and one segment
637 min_size = sizeof(*tuninfo) + sizeof(struct ipv6_sr_hdr) +
638 sizeof(struct in6_addr);
639 if (tuninfo_len < min_size)
642 switch (tuninfo->mode) {
643 case SEG6_IPTUN_MODE_INLINE:
644 if (family != AF_INET6)
648 case SEG6_IPTUN_MODE_ENCAP:
650 case SEG6_IPTUN_MODE_L2ENCAP:
652 case SEG6_IPTUN_MODE_ENCAP_RED:
654 case SEG6_IPTUN_MODE_L2ENCAP_RED:
660 /* verify that SRH is consistent */
661 if (!seg6_validate_srh(tuninfo->srh, tuninfo_len - sizeof(*tuninfo), false))
664 newts = lwtunnel_state_alloc(tuninfo_len + sizeof(*slwt));
668 slwt = seg6_lwt_lwtunnel(newts);
670 err = dst_cache_init(&slwt->cache, GFP_ATOMIC);
676 memcpy(&slwt->tuninfo, tuninfo, tuninfo_len);
678 newts->type = LWTUNNEL_ENCAP_SEG6;
679 newts->flags |= LWTUNNEL_STATE_INPUT_REDIRECT;
681 if (tuninfo->mode != SEG6_IPTUN_MODE_L2ENCAP)
682 newts->flags |= LWTUNNEL_STATE_OUTPUT_REDIRECT;
684 newts->headroom = seg6_lwt_headroom(tuninfo);
691 static void seg6_destroy_state(struct lwtunnel_state *lwt)
693 dst_cache_destroy(&seg6_lwt_lwtunnel(lwt)->cache);
696 static int seg6_fill_encap_info(struct sk_buff *skb,
697 struct lwtunnel_state *lwtstate)
699 struct seg6_iptunnel_encap *tuninfo = seg6_encap_lwtunnel(lwtstate);
701 if (nla_put_srh(skb, SEG6_IPTUNNEL_SRH, tuninfo))
707 static int seg6_encap_nlsize(struct lwtunnel_state *lwtstate)
709 struct seg6_iptunnel_encap *tuninfo = seg6_encap_lwtunnel(lwtstate);
711 return nla_total_size(SEG6_IPTUN_ENCAP_SIZE(tuninfo));
714 static int seg6_encap_cmp(struct lwtunnel_state *a, struct lwtunnel_state *b)
716 struct seg6_iptunnel_encap *a_hdr = seg6_encap_lwtunnel(a);
717 struct seg6_iptunnel_encap *b_hdr = seg6_encap_lwtunnel(b);
718 int len = SEG6_IPTUN_ENCAP_SIZE(a_hdr);
720 if (len != SEG6_IPTUN_ENCAP_SIZE(b_hdr))
723 return memcmp(a_hdr, b_hdr, len);
726 static const struct lwtunnel_encap_ops seg6_iptun_ops = {
727 .build_state = seg6_build_state,
728 .destroy_state = seg6_destroy_state,
729 .output = seg6_output,
731 .fill_encap = seg6_fill_encap_info,
732 .get_encap_size = seg6_encap_nlsize,
733 .cmp_encap = seg6_encap_cmp,
734 .owner = THIS_MODULE,
737 int __init seg6_iptunnel_init(void)
739 return lwtunnel_encap_add_ops(&seg6_iptun_ops, LWTUNNEL_ENCAP_SEG6);
742 void seg6_iptunnel_exit(void)
744 lwtunnel_encap_del_ops(&seg6_iptun_ops, LWTUNNEL_ENCAP_SEG6);