Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * xfrm_output.c - Common IPsec encapsulation code.
4 *
5 * Copyright (c) 2007 Herbert Xu <herbert@gondor.apana.org.au>
6 */
7
8#include <linux/errno.h>
9#include <linux/module.h>
10#include <linux/netdevice.h>
11#include <linux/netfilter.h>
12#include <linux/skbuff.h>
13#include <linux/slab.h>
14#include <linux/spinlock.h>
15#include <net/dst.h>
16#include <net/icmp.h>
17#include <net/inet_ecn.h>
18#include <net/xfrm.h>
19
20#if IS_ENABLED(CONFIG_IPV6)
21#include <net/ip6_route.h>
22#include <net/ipv6_stubs.h>
23#endif
24
25#include "xfrm_inout.h"
26
27static int xfrm_output2(struct net *net, struct sock *sk, struct sk_buff *skb);
28static int xfrm_inner_extract_output(struct xfrm_state *x, struct sk_buff *skb);
29
30static int xfrm_skb_check_space(struct sk_buff *skb)
31{
32 struct dst_entry *dst = skb_dst(skb);
33 int nhead = dst->header_len + LL_RESERVED_SPACE(dst->dev)
34 - skb_headroom(skb);
35 int ntail = dst->dev->needed_tailroom - skb_tailroom(skb);
36
37 if (nhead <= 0) {
38 if (ntail <= 0)
39 return 0;
40 nhead = 0;
41 } else if (ntail < 0)
42 ntail = 0;
43
44 return pskb_expand_head(skb, nhead, ntail, GFP_ATOMIC);
45}
46
47/* Children define the path of the packet through the
48 * Linux networking. Thus, destinations are stackable.
49 */
50
51static struct dst_entry *skb_dst_pop(struct sk_buff *skb)
52{
53 struct dst_entry *child = dst_clone(xfrm_dst_child(skb_dst(skb)));
54
55 skb_dst_drop(skb);
56 return child;
57}
58
59/* Add encapsulation header.
60 *
61 * The IP header will be moved forward to make space for the encapsulation
62 * header.
63 */
64static int xfrm4_transport_output(struct xfrm_state *x, struct sk_buff *skb)
65{
66 struct iphdr *iph = ip_hdr(skb);
67 int ihl = iph->ihl * 4;
68
69 skb_set_inner_transport_header(skb, skb_transport_offset(skb));
70
71 skb_set_network_header(skb, -x->props.header_len);
72 skb->mac_header = skb->network_header +
73 offsetof(struct iphdr, protocol);
74 skb->transport_header = skb->network_header + ihl;
75 __skb_pull(skb, ihl);
76 memmove(skb_network_header(skb), iph, ihl);
77 return 0;
78}
79
80#if IS_ENABLED(CONFIG_IPV6_MIP6)
81static int mip6_rthdr_offset(struct sk_buff *skb, u8 **nexthdr, int type)
82{
83 const unsigned char *nh = skb_network_header(skb);
84 unsigned int offset = sizeof(struct ipv6hdr);
85 unsigned int packet_len;
86 int found_rhdr = 0;
87
88 packet_len = skb_tail_pointer(skb) - nh;
89 *nexthdr = &ipv6_hdr(skb)->nexthdr;
90
91 while (offset <= packet_len) {
92 struct ipv6_opt_hdr *exthdr;
93
94 switch (**nexthdr) {
95 case NEXTHDR_HOP:
96 break;
97 case NEXTHDR_ROUTING:
98 if (type == IPPROTO_ROUTING && offset + 3 <= packet_len) {
99 struct ipv6_rt_hdr *rt;
100
101 rt = (struct ipv6_rt_hdr *)(nh + offset);
102 if (rt->type != 0)
103 return offset;
104 }
105 found_rhdr = 1;
106 break;
107 case NEXTHDR_DEST:
108 /* HAO MUST NOT appear more than once.
109 * XXX: It is better to try to find by the end of
110 * XXX: packet if HAO exists.
111 */
112 if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0) {
113 net_dbg_ratelimited("mip6: hao exists already, override\n");
114 return offset;
115 }
116
117 if (found_rhdr)
118 return offset;
119
120 break;
121 default:
122 return offset;
123 }
124
125 if (offset + sizeof(struct ipv6_opt_hdr) > packet_len)
126 return -EINVAL;
127
128 exthdr = (struct ipv6_opt_hdr *)(skb_network_header(skb) +
129 offset);
130 offset += ipv6_optlen(exthdr);
131 if (offset > IPV6_MAXPLEN)
132 return -EINVAL;
133 *nexthdr = &exthdr->nexthdr;
134 }
135
136 return -EINVAL;
137}
138#endif
139
140#if IS_ENABLED(CONFIG_IPV6)
141static int xfrm6_hdr_offset(struct xfrm_state *x, struct sk_buff *skb, u8 **prevhdr)
142{
143 switch (x->type->proto) {
144#if IS_ENABLED(CONFIG_IPV6_MIP6)
145 case IPPROTO_DSTOPTS:
146 case IPPROTO_ROUTING:
147 return mip6_rthdr_offset(skb, prevhdr, x->type->proto);
148#endif
149 default:
150 break;
151 }
152
153 return ip6_find_1stfragopt(skb, prevhdr);
154}
155#endif
156
157/* Add encapsulation header.
158 *
159 * The IP header and mutable extension headers will be moved forward to make
160 * space for the encapsulation header.
161 */
162static int xfrm6_transport_output(struct xfrm_state *x, struct sk_buff *skb)
163{
164#if IS_ENABLED(CONFIG_IPV6)
165 struct ipv6hdr *iph;
166 u8 *prevhdr;
167 int hdr_len;
168
169 iph = ipv6_hdr(skb);
170 skb_set_inner_transport_header(skb, skb_transport_offset(skb));
171
172 hdr_len = xfrm6_hdr_offset(x, skb, &prevhdr);
173 if (hdr_len < 0)
174 return hdr_len;
175 skb_set_mac_header(skb,
176 (prevhdr - x->props.header_len) - skb->data);
177 skb_set_network_header(skb, -x->props.header_len);
178 skb->transport_header = skb->network_header + hdr_len;
179 __skb_pull(skb, hdr_len);
180 memmove(ipv6_hdr(skb), iph, hdr_len);
181 return 0;
182#else
183 WARN_ON_ONCE(1);
184 return -EAFNOSUPPORT;
185#endif
186}
187
188/* Add route optimization header space.
189 *
190 * The IP header and mutable extension headers will be moved forward to make
191 * space for the route optimization header.
192 */
193static int xfrm6_ro_output(struct xfrm_state *x, struct sk_buff *skb)
194{
195#if IS_ENABLED(CONFIG_IPV6)
196 struct ipv6hdr *iph;
197 u8 *prevhdr;
198 int hdr_len;
199
200 iph = ipv6_hdr(skb);
201
202 hdr_len = xfrm6_hdr_offset(x, skb, &prevhdr);
203 if (hdr_len < 0)
204 return hdr_len;
205 skb_set_mac_header(skb,
206 (prevhdr - x->props.header_len) - skb->data);
207 skb_set_network_header(skb, -x->props.header_len);
208 skb->transport_header = skb->network_header + hdr_len;
209 __skb_pull(skb, hdr_len);
210 memmove(ipv6_hdr(skb), iph, hdr_len);
211
212 x->lastused = ktime_get_real_seconds();
213
214 return 0;
215#else
216 WARN_ON_ONCE(1);
217 return -EAFNOSUPPORT;
218#endif
219}
220
221/* Add encapsulation header.
222 *
223 * The top IP header will be constructed per draft-nikander-esp-beet-mode-06.txt.
224 */
225static int xfrm4_beet_encap_add(struct xfrm_state *x, struct sk_buff *skb)
226{
227 struct ip_beet_phdr *ph;
228 struct iphdr *top_iph;
229 int hdrlen, optlen;
230
231 hdrlen = 0;
232 optlen = XFRM_MODE_SKB_CB(skb)->optlen;
233 if (unlikely(optlen))
234 hdrlen += IPV4_BEET_PHMAXLEN - (optlen & 4);
235
236 skb_set_network_header(skb, -x->props.header_len - hdrlen +
237 (XFRM_MODE_SKB_CB(skb)->ihl - sizeof(*top_iph)));
238 if (x->sel.family != AF_INET6)
239 skb->network_header += IPV4_BEET_PHMAXLEN;
240 skb->mac_header = skb->network_header +
241 offsetof(struct iphdr, protocol);
242 skb->transport_header = skb->network_header + sizeof(*top_iph);
243
244 xfrm4_beet_make_header(skb);
245
246 ph = __skb_pull(skb, XFRM_MODE_SKB_CB(skb)->ihl - hdrlen);
247
248 top_iph = ip_hdr(skb);
249
250 if (unlikely(optlen)) {
251 if (WARN_ON(optlen < 0))
252 return -EINVAL;
253
254 ph->padlen = 4 - (optlen & 4);
255 ph->hdrlen = optlen / 8;
256 ph->nexthdr = top_iph->protocol;
257 if (ph->padlen)
258 memset(ph + 1, IPOPT_NOP, ph->padlen);
259
260 top_iph->protocol = IPPROTO_BEETPH;
261 top_iph->ihl = sizeof(struct iphdr) / 4;
262 }
263
264 top_iph->saddr = x->props.saddr.a4;
265 top_iph->daddr = x->id.daddr.a4;
266
267 return 0;
268}
269
270/* Add encapsulation header.
271 *
272 * The top IP header will be constructed per RFC 2401.
273 */
274static int xfrm4_tunnel_encap_add(struct xfrm_state *x, struct sk_buff *skb)
275{
276 struct dst_entry *dst = skb_dst(skb);
277 struct iphdr *top_iph;
278 int flags;
279
280 skb_set_inner_network_header(skb, skb_network_offset(skb));
281 skb_set_inner_transport_header(skb, skb_transport_offset(skb));
282
283 skb_set_network_header(skb, -x->props.header_len);
284 skb->mac_header = skb->network_header +
285 offsetof(struct iphdr, protocol);
286 skb->transport_header = skb->network_header + sizeof(*top_iph);
287 top_iph = ip_hdr(skb);
288
289 top_iph->ihl = 5;
290 top_iph->version = 4;
291
292 top_iph->protocol = xfrm_af2proto(skb_dst(skb)->ops->family);
293
294 /* DS disclosing depends on XFRM_SA_XFLAG_DONT_ENCAP_DSCP */
295 if (x->props.extra_flags & XFRM_SA_XFLAG_DONT_ENCAP_DSCP)
296 top_iph->tos = 0;
297 else
298 top_iph->tos = XFRM_MODE_SKB_CB(skb)->tos;
299 top_iph->tos = INET_ECN_encapsulate(top_iph->tos,
300 XFRM_MODE_SKB_CB(skb)->tos);
301
302 flags = x->props.flags;
303 if (flags & XFRM_STATE_NOECN)
304 IP_ECN_clear(top_iph);
305
306 top_iph->frag_off = (flags & XFRM_STATE_NOPMTUDISC) ?
307 0 : (XFRM_MODE_SKB_CB(skb)->frag_off & htons(IP_DF));
308
309 top_iph->ttl = ip4_dst_hoplimit(xfrm_dst_child(dst));
310
311 top_iph->saddr = x->props.saddr.a4;
312 top_iph->daddr = x->id.daddr.a4;
313 ip_select_ident(dev_net(dst->dev), skb, NULL);
314
315 return 0;
316}
317
318#if IS_ENABLED(CONFIG_IPV6)
319static int xfrm6_tunnel_encap_add(struct xfrm_state *x, struct sk_buff *skb)
320{
321 struct dst_entry *dst = skb_dst(skb);
322 struct ipv6hdr *top_iph;
323 int dsfield;
324
325 skb_set_inner_network_header(skb, skb_network_offset(skb));
326 skb_set_inner_transport_header(skb, skb_transport_offset(skb));
327
328 skb_set_network_header(skb, -x->props.header_len);
329 skb->mac_header = skb->network_header +
330 offsetof(struct ipv6hdr, nexthdr);
331 skb->transport_header = skb->network_header + sizeof(*top_iph);
332 top_iph = ipv6_hdr(skb);
333
334 top_iph->version = 6;
335
336 memcpy(top_iph->flow_lbl, XFRM_MODE_SKB_CB(skb)->flow_lbl,
337 sizeof(top_iph->flow_lbl));
338 top_iph->nexthdr = xfrm_af2proto(skb_dst(skb)->ops->family);
339
340 if (x->props.extra_flags & XFRM_SA_XFLAG_DONT_ENCAP_DSCP)
341 dsfield = 0;
342 else
343 dsfield = XFRM_MODE_SKB_CB(skb)->tos;
344 dsfield = INET_ECN_encapsulate(dsfield, XFRM_MODE_SKB_CB(skb)->tos);
345 if (x->props.flags & XFRM_STATE_NOECN)
346 dsfield &= ~INET_ECN_MASK;
347 ipv6_change_dsfield(top_iph, 0, dsfield);
348 top_iph->hop_limit = ip6_dst_hoplimit(xfrm_dst_child(dst));
349 top_iph->saddr = *(struct in6_addr *)&x->props.saddr;
350 top_iph->daddr = *(struct in6_addr *)&x->id.daddr;
351 return 0;
352}
353
354static int xfrm6_beet_encap_add(struct xfrm_state *x, struct sk_buff *skb)
355{
356 struct ipv6hdr *top_iph;
357 struct ip_beet_phdr *ph;
358 int optlen, hdr_len;
359
360 hdr_len = 0;
361 optlen = XFRM_MODE_SKB_CB(skb)->optlen;
362 if (unlikely(optlen))
363 hdr_len += IPV4_BEET_PHMAXLEN - (optlen & 4);
364
365 skb_set_network_header(skb, -x->props.header_len - hdr_len);
366 if (x->sel.family != AF_INET6)
367 skb->network_header += IPV4_BEET_PHMAXLEN;
368 skb->mac_header = skb->network_header +
369 offsetof(struct ipv6hdr, nexthdr);
370 skb->transport_header = skb->network_header + sizeof(*top_iph);
371 ph = __skb_pull(skb, XFRM_MODE_SKB_CB(skb)->ihl - hdr_len);
372
373 xfrm6_beet_make_header(skb);
374
375 top_iph = ipv6_hdr(skb);
376 if (unlikely(optlen)) {
377 if (WARN_ON(optlen < 0))
378 return -EINVAL;
379
380 ph->padlen = 4 - (optlen & 4);
381 ph->hdrlen = optlen / 8;
382 ph->nexthdr = top_iph->nexthdr;
383 if (ph->padlen)
384 memset(ph + 1, IPOPT_NOP, ph->padlen);
385
386 top_iph->nexthdr = IPPROTO_BEETPH;
387 }
388
389 top_iph->saddr = *(struct in6_addr *)&x->props.saddr;
390 top_iph->daddr = *(struct in6_addr *)&x->id.daddr;
391 return 0;
392}
393#endif
394
395/* Add encapsulation header.
396 *
397 * On exit, the transport header will be set to the start of the
398 * encapsulation header to be filled in by x->type->output and the mac
399 * header will be set to the nextheader (protocol for IPv4) field of the
400 * extension header directly preceding the encapsulation header, or in
401 * its absence, that of the top IP header.
402 * The value of the network header will always point to the top IP header
403 * while skb->data will point to the payload.
404 */
405static int xfrm4_prepare_output(struct xfrm_state *x, struct sk_buff *skb)
406{
407 int err;
408
409 err = xfrm_inner_extract_output(x, skb);
410 if (err)
411 return err;
412
413 IPCB(skb)->flags |= IPSKB_XFRM_TUNNEL_SIZE;
414 skb->protocol = htons(ETH_P_IP);
415
416 switch (x->outer_mode.encap) {
417 case XFRM_MODE_BEET:
418 return xfrm4_beet_encap_add(x, skb);
419 case XFRM_MODE_TUNNEL:
420 return xfrm4_tunnel_encap_add(x, skb);
421 }
422
423 WARN_ON_ONCE(1);
424 return -EOPNOTSUPP;
425}
426
427static int xfrm6_prepare_output(struct xfrm_state *x, struct sk_buff *skb)
428{
429#if IS_ENABLED(CONFIG_IPV6)
430 int err;
431
432 err = xfrm_inner_extract_output(x, skb);
433 if (err)
434 return err;
435
436 skb->ignore_df = 1;
437 skb->protocol = htons(ETH_P_IPV6);
438
439 switch (x->outer_mode.encap) {
440 case XFRM_MODE_BEET:
441 return xfrm6_beet_encap_add(x, skb);
442 case XFRM_MODE_TUNNEL:
443 return xfrm6_tunnel_encap_add(x, skb);
444 default:
445 WARN_ON_ONCE(1);
446 return -EOPNOTSUPP;
447 }
448#endif
449 WARN_ON_ONCE(1);
450 return -EAFNOSUPPORT;
451}
452
453static int xfrm_outer_mode_output(struct xfrm_state *x, struct sk_buff *skb)
454{
455 switch (x->outer_mode.encap) {
456 case XFRM_MODE_BEET:
457 case XFRM_MODE_TUNNEL:
458 if (x->outer_mode.family == AF_INET)
459 return xfrm4_prepare_output(x, skb);
460 if (x->outer_mode.family == AF_INET6)
461 return xfrm6_prepare_output(x, skb);
462 break;
463 case XFRM_MODE_TRANSPORT:
464 if (x->outer_mode.family == AF_INET)
465 return xfrm4_transport_output(x, skb);
466 if (x->outer_mode.family == AF_INET6)
467 return xfrm6_transport_output(x, skb);
468 break;
469 case XFRM_MODE_ROUTEOPTIMIZATION:
470 if (x->outer_mode.family == AF_INET6)
471 return xfrm6_ro_output(x, skb);
472 WARN_ON_ONCE(1);
473 break;
474 default:
475 WARN_ON_ONCE(1);
476 break;
477 }
478
479 return -EOPNOTSUPP;
480}
481
482#if IS_ENABLED(CONFIG_NET_PKTGEN)
483int pktgen_xfrm_outer_mode_output(struct xfrm_state *x, struct sk_buff *skb)
484{
485 return xfrm_outer_mode_output(x, skb);
486}
487EXPORT_SYMBOL_GPL(pktgen_xfrm_outer_mode_output);
488#endif
489
490static int xfrm_output_one(struct sk_buff *skb, int err)
491{
492 struct dst_entry *dst = skb_dst(skb);
493 struct xfrm_state *x = dst->xfrm;
494 struct net *net = xs_net(x);
495
496 if (err <= 0)
497 goto resume;
498
499 do {
500 err = xfrm_skb_check_space(skb);
501 if (err) {
502 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
503 goto error_nolock;
504 }
505
506 skb->mark = xfrm_smark_get(skb->mark, x);
507
508 err = xfrm_outer_mode_output(x, skb);
509 if (err) {
510 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEMODEERROR);
511 goto error_nolock;
512 }
513
514 spin_lock_bh(&x->lock);
515
516 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
517 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEINVALID);
518 err = -EINVAL;
519 goto error;
520 }
521
522 err = xfrm_state_check_expire(x);
523 if (err) {
524 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEEXPIRED);
525 goto error;
526 }
527
528 err = xfrm_replay_overflow(x, skb);
529 if (err) {
530 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATESEQERROR);
531 goto error;
532 }
533
534 x->curlft.bytes += skb->len;
535 x->curlft.packets++;
536
537 spin_unlock_bh(&x->lock);
538
539 skb_dst_force(skb);
540 if (!skb_dst(skb)) {
541 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
542 err = -EHOSTUNREACH;
543 goto error_nolock;
544 }
545
546 if (xfrm_offload(skb)) {
547 x->type_offload->encap(x, skb);
548 } else {
549 /* Inner headers are invalid now. */
550 skb->encapsulation = 0;
551
552 err = x->type->output(x, skb);
553 if (err == -EINPROGRESS)
554 goto out;
555 }
556
557resume:
558 if (err) {
559 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR);
560 goto error_nolock;
561 }
562
563 dst = skb_dst_pop(skb);
564 if (!dst) {
565 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
566 err = -EHOSTUNREACH;
567 goto error_nolock;
568 }
569 skb_dst_set(skb, dst);
570 x = dst->xfrm;
571 } while (x && !(x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL));
572
573 return 0;
574
575error:
576 spin_unlock_bh(&x->lock);
577error_nolock:
578 kfree_skb(skb);
579out:
580 return err;
581}
582
583int xfrm_output_resume(struct sock *sk, struct sk_buff *skb, int err)
584{
585 struct net *net = xs_net(skb_dst(skb)->xfrm);
586
587 while (likely((err = xfrm_output_one(skb, err)) == 0)) {
588 nf_reset_ct(skb);
589
590 err = skb_dst(skb)->ops->local_out(net, sk, skb);
591 if (unlikely(err != 1))
592 goto out;
593
594 if (!skb_dst(skb)->xfrm)
595 return dst_output(net, sk, skb);
596
597 err = nf_hook(skb_dst(skb)->ops->family,
598 NF_INET_POST_ROUTING, net, sk, skb,
599 NULL, skb_dst(skb)->dev, xfrm_output2);
600 if (unlikely(err != 1))
601 goto out;
602 }
603
604 if (err == -EINPROGRESS)
605 err = 0;
606
607out:
608 return err;
609}
610EXPORT_SYMBOL_GPL(xfrm_output_resume);
611
612static int xfrm_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
613{
614 return xfrm_output_resume(sk, skb, 1);
615}
616
617static int xfrm_output_gso(struct net *net, struct sock *sk, struct sk_buff *skb)
618{
619 struct sk_buff *segs, *nskb;
620
621 BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET);
622 BUILD_BUG_ON(sizeof(*IP6CB(skb)) > SKB_GSO_CB_OFFSET);
623 segs = skb_gso_segment(skb, 0);
624 kfree_skb(skb);
625 if (IS_ERR(segs))
626 return PTR_ERR(segs);
627 if (segs == NULL)
628 return -EINVAL;
629
630 skb_list_walk_safe(segs, segs, nskb) {
631 int err;
632
633 skb_mark_not_on_list(segs);
634 err = xfrm_output2(net, sk, segs);
635
636 if (unlikely(err)) {
637 kfree_skb_list(nskb);
638 return err;
639 }
640 }
641
642 return 0;
643}
644
645/* For partial checksum offload, the outer header checksum is calculated
646 * by software and the inner header checksum is calculated by hardware.
647 * This requires hardware to know the inner packet type to calculate
648 * the inner header checksum. Save inner ip protocol here to avoid
649 * traversing the packet in the vendor's xmit code.
650 * If the encap type is IPIP, just save skb->inner_ipproto. Otherwise,
651 * get the ip protocol from the IP header.
652 */
653static void xfrm_get_inner_ipproto(struct sk_buff *skb)
654{
655 struct xfrm_offload *xo = xfrm_offload(skb);
656 const struct ethhdr *eth;
657
658 if (!xo)
659 return;
660
661 if (skb->inner_protocol_type == ENCAP_TYPE_IPPROTO) {
662 xo->inner_ipproto = skb->inner_ipproto;
663 return;
664 }
665
666 if (skb->inner_protocol_type != ENCAP_TYPE_ETHER)
667 return;
668
669 eth = (struct ethhdr *)skb_inner_mac_header(skb);
670
671 switch (ntohs(eth->h_proto)) {
672 case ETH_P_IPV6:
673 xo->inner_ipproto = inner_ipv6_hdr(skb)->nexthdr;
674 break;
675 case ETH_P_IP:
676 xo->inner_ipproto = inner_ip_hdr(skb)->protocol;
677 break;
678 }
679}
680
681int xfrm_output(struct sock *sk, struct sk_buff *skb)
682{
683 struct net *net = dev_net(skb_dst(skb)->dev);
684 struct xfrm_state *x = skb_dst(skb)->xfrm;
685 int err;
686
687 switch (x->outer_mode.family) {
688 case AF_INET:
689 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
690 IPCB(skb)->flags |= IPSKB_XFRM_TRANSFORMED;
691 break;
692 case AF_INET6:
693 memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
694
695 IP6CB(skb)->flags |= IP6SKB_XFRM_TRANSFORMED;
696 break;
697 }
698
699 secpath_reset(skb);
700
701 if (xfrm_dev_offload_ok(skb, x)) {
702 struct sec_path *sp;
703
704 sp = secpath_set(skb);
705 if (!sp) {
706 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
707 kfree_skb(skb);
708 return -ENOMEM;
709 }
710
711 sp->olen++;
712 sp->xvec[sp->len++] = x;
713 xfrm_state_hold(x);
714
715 if (skb->encapsulation)
716 xfrm_get_inner_ipproto(skb);
717 skb->encapsulation = 1;
718
719 if (skb_is_gso(skb)) {
720 if (skb->inner_protocol)
721 return xfrm_output_gso(net, sk, skb);
722
723 skb_shinfo(skb)->gso_type |= SKB_GSO_ESP;
724 goto out;
725 }
726
727 if (x->xso.dev && x->xso.dev->features & NETIF_F_HW_ESP_TX_CSUM)
728 goto out;
729 } else {
730 if (skb_is_gso(skb))
731 return xfrm_output_gso(net, sk, skb);
732 }
733
734 if (skb->ip_summed == CHECKSUM_PARTIAL) {
735 err = skb_checksum_help(skb);
736 if (err) {
737 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
738 kfree_skb(skb);
739 return err;
740 }
741 }
742
743out:
744 return xfrm_output2(net, sk, skb);
745}
746EXPORT_SYMBOL_GPL(xfrm_output);
747
748static int xfrm4_tunnel_check_size(struct sk_buff *skb)
749{
750 int mtu, ret = 0;
751
752 if (IPCB(skb)->flags & IPSKB_XFRM_TUNNEL_SIZE)
753 goto out;
754
755 if (!(ip_hdr(skb)->frag_off & htons(IP_DF)) || skb->ignore_df)
756 goto out;
757
758 mtu = dst_mtu(skb_dst(skb));
759 if ((!skb_is_gso(skb) && skb->len > mtu) ||
760 (skb_is_gso(skb) &&
761 !skb_gso_validate_network_len(skb, ip_skb_dst_mtu(skb->sk, skb)))) {
762 skb->protocol = htons(ETH_P_IP);
763
764 if (skb->sk)
765 xfrm_local_error(skb, mtu);
766 else
767 icmp_send(skb, ICMP_DEST_UNREACH,
768 ICMP_FRAG_NEEDED, htonl(mtu));
769 ret = -EMSGSIZE;
770 }
771out:
772 return ret;
773}
774
775static int xfrm4_extract_output(struct xfrm_state *x, struct sk_buff *skb)
776{
777 int err;
778
779 if (x->outer_mode.encap == XFRM_MODE_BEET &&
780 ip_is_fragment(ip_hdr(skb))) {
781 net_warn_ratelimited("BEET mode doesn't support inner IPv4 fragments\n");
782 return -EAFNOSUPPORT;
783 }
784
785 err = xfrm4_tunnel_check_size(skb);
786 if (err)
787 return err;
788
789 XFRM_MODE_SKB_CB(skb)->protocol = ip_hdr(skb)->protocol;
790
791 xfrm4_extract_header(skb);
792 return 0;
793}
794
795#if IS_ENABLED(CONFIG_IPV6)
796static int xfrm6_tunnel_check_size(struct sk_buff *skb)
797{
798 int mtu, ret = 0;
799 struct dst_entry *dst = skb_dst(skb);
800
801 if (skb->ignore_df)
802 goto out;
803
804 mtu = dst_mtu(dst);
805 if (mtu < IPV6_MIN_MTU)
806 mtu = IPV6_MIN_MTU;
807
808 if ((!skb_is_gso(skb) && skb->len > mtu) ||
809 (skb_is_gso(skb) &&
810 !skb_gso_validate_network_len(skb, ip6_skb_dst_mtu(skb)))) {
811 skb->dev = dst->dev;
812 skb->protocol = htons(ETH_P_IPV6);
813
814 if (xfrm6_local_dontfrag(skb->sk))
815 ipv6_stub->xfrm6_local_rxpmtu(skb, mtu);
816 else if (skb->sk)
817 xfrm_local_error(skb, mtu);
818 else
819 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
820 ret = -EMSGSIZE;
821 }
822out:
823 return ret;
824}
825#endif
826
827static int xfrm6_extract_output(struct xfrm_state *x, struct sk_buff *skb)
828{
829#if IS_ENABLED(CONFIG_IPV6)
830 int err;
831
832 err = xfrm6_tunnel_check_size(skb);
833 if (err)
834 return err;
835
836 XFRM_MODE_SKB_CB(skb)->protocol = ipv6_hdr(skb)->nexthdr;
837
838 xfrm6_extract_header(skb);
839 return 0;
840#else
841 WARN_ON_ONCE(1);
842 return -EAFNOSUPPORT;
843#endif
844}
845
846static int xfrm_inner_extract_output(struct xfrm_state *x, struct sk_buff *skb)
847{
848 const struct xfrm_mode *inner_mode;
849
850 if (x->sel.family == AF_UNSPEC)
851 inner_mode = xfrm_ip2inner_mode(x,
852 xfrm_af2proto(skb_dst(skb)->ops->family));
853 else
854 inner_mode = &x->inner_mode;
855
856 if (inner_mode == NULL)
857 return -EAFNOSUPPORT;
858
859 switch (inner_mode->family) {
860 case AF_INET:
861 return xfrm4_extract_output(x, skb);
862 case AF_INET6:
863 return xfrm6_extract_output(x, skb);
864 }
865
866 return -EAFNOSUPPORT;
867}
868
869void xfrm_local_error(struct sk_buff *skb, int mtu)
870{
871 unsigned int proto;
872 struct xfrm_state_afinfo *afinfo;
873
874 if (skb->protocol == htons(ETH_P_IP))
875 proto = AF_INET;
876 else if (skb->protocol == htons(ETH_P_IPV6) &&
877 skb->sk->sk_family == AF_INET6)
878 proto = AF_INET6;
879 else
880 return;
881
882 afinfo = xfrm_state_get_afinfo(proto);
883 if (afinfo) {
884 afinfo->local_error(skb, mtu);
885 rcu_read_unlock();
886 }
887}
888EXPORT_SYMBOL_GPL(xfrm_local_error);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * xfrm_output.c - Common IPsec encapsulation code.
4 *
5 * Copyright (c) 2007 Herbert Xu <herbert@gondor.apana.org.au>
6 */
7
8#include <linux/errno.h>
9#include <linux/module.h>
10#include <linux/netdevice.h>
11#include <linux/netfilter.h>
12#include <linux/skbuff.h>
13#include <linux/slab.h>
14#include <linux/spinlock.h>
15#include <net/dst.h>
16#include <net/icmp.h>
17#include <net/inet_ecn.h>
18#include <net/xfrm.h>
19
20#if IS_ENABLED(CONFIG_IPV6)
21#include <net/ip6_route.h>
22#include <net/ipv6_stubs.h>
23#endif
24
25#include "xfrm_inout.h"
26
27static int xfrm_output2(struct net *net, struct sock *sk, struct sk_buff *skb);
28static int xfrm_inner_extract_output(struct xfrm_state *x, struct sk_buff *skb);
29
30static int xfrm_skb_check_space(struct sk_buff *skb)
31{
32 struct dst_entry *dst = skb_dst(skb);
33 int nhead = dst->header_len + LL_RESERVED_SPACE(dst->dev)
34 - skb_headroom(skb);
35 int ntail = dst->dev->needed_tailroom - skb_tailroom(skb);
36
37 if (nhead <= 0) {
38 if (ntail <= 0)
39 return 0;
40 nhead = 0;
41 } else if (ntail < 0)
42 ntail = 0;
43
44 return pskb_expand_head(skb, nhead, ntail, GFP_ATOMIC);
45}
46
47/* Children define the path of the packet through the
48 * Linux networking. Thus, destinations are stackable.
49 */
50
51static struct dst_entry *skb_dst_pop(struct sk_buff *skb)
52{
53 struct dst_entry *child = dst_clone(xfrm_dst_child(skb_dst(skb)));
54
55 skb_dst_drop(skb);
56 return child;
57}
58
59/* Add encapsulation header.
60 *
61 * The IP header will be moved forward to make space for the encapsulation
62 * header.
63 */
64static int xfrm4_transport_output(struct xfrm_state *x, struct sk_buff *skb)
65{
66 struct iphdr *iph = ip_hdr(skb);
67 int ihl = iph->ihl * 4;
68
69 skb_set_inner_transport_header(skb, skb_transport_offset(skb));
70
71 skb_set_network_header(skb, -x->props.header_len);
72 skb->mac_header = skb->network_header +
73 offsetof(struct iphdr, protocol);
74 skb->transport_header = skb->network_header + ihl;
75 __skb_pull(skb, ihl);
76 memmove(skb_network_header(skb), iph, ihl);
77 return 0;
78}
79
80/* Add encapsulation header.
81 *
82 * The IP header and mutable extension headers will be moved forward to make
83 * space for the encapsulation header.
84 */
85static int xfrm6_transport_output(struct xfrm_state *x, struct sk_buff *skb)
86{
87#if IS_ENABLED(CONFIG_IPV6)
88 struct ipv6hdr *iph;
89 u8 *prevhdr;
90 int hdr_len;
91
92 iph = ipv6_hdr(skb);
93 skb_set_inner_transport_header(skb, skb_transport_offset(skb));
94
95 hdr_len = x->type->hdr_offset(x, skb, &prevhdr);
96 if (hdr_len < 0)
97 return hdr_len;
98 skb_set_mac_header(skb,
99 (prevhdr - x->props.header_len) - skb->data);
100 skb_set_network_header(skb, -x->props.header_len);
101 skb->transport_header = skb->network_header + hdr_len;
102 __skb_pull(skb, hdr_len);
103 memmove(ipv6_hdr(skb), iph, hdr_len);
104 return 0;
105#else
106 WARN_ON_ONCE(1);
107 return -EAFNOSUPPORT;
108#endif
109}
110
111/* Add route optimization header space.
112 *
113 * The IP header and mutable extension headers will be moved forward to make
114 * space for the route optimization header.
115 */
116static int xfrm6_ro_output(struct xfrm_state *x, struct sk_buff *skb)
117{
118#if IS_ENABLED(CONFIG_IPV6)
119 struct ipv6hdr *iph;
120 u8 *prevhdr;
121 int hdr_len;
122
123 iph = ipv6_hdr(skb);
124
125 hdr_len = x->type->hdr_offset(x, skb, &prevhdr);
126 if (hdr_len < 0)
127 return hdr_len;
128 skb_set_mac_header(skb,
129 (prevhdr - x->props.header_len) - skb->data);
130 skb_set_network_header(skb, -x->props.header_len);
131 skb->transport_header = skb->network_header + hdr_len;
132 __skb_pull(skb, hdr_len);
133 memmove(ipv6_hdr(skb), iph, hdr_len);
134
135 x->lastused = ktime_get_real_seconds();
136
137 return 0;
138#else
139 WARN_ON_ONCE(1);
140 return -EAFNOSUPPORT;
141#endif
142}
143
144/* Add encapsulation header.
145 *
146 * The top IP header will be constructed per draft-nikander-esp-beet-mode-06.txt.
147 */
148static int xfrm4_beet_encap_add(struct xfrm_state *x, struct sk_buff *skb)
149{
150 struct ip_beet_phdr *ph;
151 struct iphdr *top_iph;
152 int hdrlen, optlen;
153
154 hdrlen = 0;
155 optlen = XFRM_MODE_SKB_CB(skb)->optlen;
156 if (unlikely(optlen))
157 hdrlen += IPV4_BEET_PHMAXLEN - (optlen & 4);
158
159 skb_set_network_header(skb, -x->props.header_len - hdrlen +
160 (XFRM_MODE_SKB_CB(skb)->ihl - sizeof(*top_iph)));
161 if (x->sel.family != AF_INET6)
162 skb->network_header += IPV4_BEET_PHMAXLEN;
163 skb->mac_header = skb->network_header +
164 offsetof(struct iphdr, protocol);
165 skb->transport_header = skb->network_header + sizeof(*top_iph);
166
167 xfrm4_beet_make_header(skb);
168
169 ph = __skb_pull(skb, XFRM_MODE_SKB_CB(skb)->ihl - hdrlen);
170
171 top_iph = ip_hdr(skb);
172
173 if (unlikely(optlen)) {
174 if (WARN_ON(optlen < 0))
175 return -EINVAL;
176
177 ph->padlen = 4 - (optlen & 4);
178 ph->hdrlen = optlen / 8;
179 ph->nexthdr = top_iph->protocol;
180 if (ph->padlen)
181 memset(ph + 1, IPOPT_NOP, ph->padlen);
182
183 top_iph->protocol = IPPROTO_BEETPH;
184 top_iph->ihl = sizeof(struct iphdr) / 4;
185 }
186
187 top_iph->saddr = x->props.saddr.a4;
188 top_iph->daddr = x->id.daddr.a4;
189
190 return 0;
191}
192
193/* Add encapsulation header.
194 *
195 * The top IP header will be constructed per RFC 2401.
196 */
197static int xfrm4_tunnel_encap_add(struct xfrm_state *x, struct sk_buff *skb)
198{
199 struct dst_entry *dst = skb_dst(skb);
200 struct iphdr *top_iph;
201 int flags;
202
203 skb_set_inner_network_header(skb, skb_network_offset(skb));
204 skb_set_inner_transport_header(skb, skb_transport_offset(skb));
205
206 skb_set_network_header(skb, -x->props.header_len);
207 skb->mac_header = skb->network_header +
208 offsetof(struct iphdr, protocol);
209 skb->transport_header = skb->network_header + sizeof(*top_iph);
210 top_iph = ip_hdr(skb);
211
212 top_iph->ihl = 5;
213 top_iph->version = 4;
214
215 top_iph->protocol = xfrm_af2proto(skb_dst(skb)->ops->family);
216
217 /* DS disclosing depends on XFRM_SA_XFLAG_DONT_ENCAP_DSCP */
218 if (x->props.extra_flags & XFRM_SA_XFLAG_DONT_ENCAP_DSCP)
219 top_iph->tos = 0;
220 else
221 top_iph->tos = XFRM_MODE_SKB_CB(skb)->tos;
222 top_iph->tos = INET_ECN_encapsulate(top_iph->tos,
223 XFRM_MODE_SKB_CB(skb)->tos);
224
225 flags = x->props.flags;
226 if (flags & XFRM_STATE_NOECN)
227 IP_ECN_clear(top_iph);
228
229 top_iph->frag_off = (flags & XFRM_STATE_NOPMTUDISC) ?
230 0 : (XFRM_MODE_SKB_CB(skb)->frag_off & htons(IP_DF));
231
232 top_iph->ttl = ip4_dst_hoplimit(xfrm_dst_child(dst));
233
234 top_iph->saddr = x->props.saddr.a4;
235 top_iph->daddr = x->id.daddr.a4;
236 ip_select_ident(dev_net(dst->dev), skb, NULL);
237
238 return 0;
239}
240
241#if IS_ENABLED(CONFIG_IPV6)
242static int xfrm6_tunnel_encap_add(struct xfrm_state *x, struct sk_buff *skb)
243{
244 struct dst_entry *dst = skb_dst(skb);
245 struct ipv6hdr *top_iph;
246 int dsfield;
247
248 skb_set_inner_network_header(skb, skb_network_offset(skb));
249 skb_set_inner_transport_header(skb, skb_transport_offset(skb));
250
251 skb_set_network_header(skb, -x->props.header_len);
252 skb->mac_header = skb->network_header +
253 offsetof(struct ipv6hdr, nexthdr);
254 skb->transport_header = skb->network_header + sizeof(*top_iph);
255 top_iph = ipv6_hdr(skb);
256
257 top_iph->version = 6;
258
259 memcpy(top_iph->flow_lbl, XFRM_MODE_SKB_CB(skb)->flow_lbl,
260 sizeof(top_iph->flow_lbl));
261 top_iph->nexthdr = xfrm_af2proto(skb_dst(skb)->ops->family);
262
263 if (x->props.extra_flags & XFRM_SA_XFLAG_DONT_ENCAP_DSCP)
264 dsfield = 0;
265 else
266 dsfield = XFRM_MODE_SKB_CB(skb)->tos;
267 dsfield = INET_ECN_encapsulate(dsfield, XFRM_MODE_SKB_CB(skb)->tos);
268 if (x->props.flags & XFRM_STATE_NOECN)
269 dsfield &= ~INET_ECN_MASK;
270 ipv6_change_dsfield(top_iph, 0, dsfield);
271 top_iph->hop_limit = ip6_dst_hoplimit(xfrm_dst_child(dst));
272 top_iph->saddr = *(struct in6_addr *)&x->props.saddr;
273 top_iph->daddr = *(struct in6_addr *)&x->id.daddr;
274 return 0;
275}
276
277static int xfrm6_beet_encap_add(struct xfrm_state *x, struct sk_buff *skb)
278{
279 struct ipv6hdr *top_iph;
280 struct ip_beet_phdr *ph;
281 int optlen, hdr_len;
282
283 hdr_len = 0;
284 optlen = XFRM_MODE_SKB_CB(skb)->optlen;
285 if (unlikely(optlen))
286 hdr_len += IPV4_BEET_PHMAXLEN - (optlen & 4);
287
288 skb_set_network_header(skb, -x->props.header_len - hdr_len);
289 if (x->sel.family != AF_INET6)
290 skb->network_header += IPV4_BEET_PHMAXLEN;
291 skb->mac_header = skb->network_header +
292 offsetof(struct ipv6hdr, nexthdr);
293 skb->transport_header = skb->network_header + sizeof(*top_iph);
294 ph = __skb_pull(skb, XFRM_MODE_SKB_CB(skb)->ihl - hdr_len);
295
296 xfrm6_beet_make_header(skb);
297
298 top_iph = ipv6_hdr(skb);
299 if (unlikely(optlen)) {
300 if (WARN_ON(optlen < 0))
301 return -EINVAL;
302
303 ph->padlen = 4 - (optlen & 4);
304 ph->hdrlen = optlen / 8;
305 ph->nexthdr = top_iph->nexthdr;
306 if (ph->padlen)
307 memset(ph + 1, IPOPT_NOP, ph->padlen);
308
309 top_iph->nexthdr = IPPROTO_BEETPH;
310 }
311
312 top_iph->saddr = *(struct in6_addr *)&x->props.saddr;
313 top_iph->daddr = *(struct in6_addr *)&x->id.daddr;
314 return 0;
315}
316#endif
317
318/* Add encapsulation header.
319 *
320 * On exit, the transport header will be set to the start of the
321 * encapsulation header to be filled in by x->type->output and the mac
322 * header will be set to the nextheader (protocol for IPv4) field of the
323 * extension header directly preceding the encapsulation header, or in
324 * its absence, that of the top IP header.
325 * The value of the network header will always point to the top IP header
326 * while skb->data will point to the payload.
327 */
328static int xfrm4_prepare_output(struct xfrm_state *x, struct sk_buff *skb)
329{
330 int err;
331
332 err = xfrm_inner_extract_output(x, skb);
333 if (err)
334 return err;
335
336 IPCB(skb)->flags |= IPSKB_XFRM_TUNNEL_SIZE;
337 skb->protocol = htons(ETH_P_IP);
338
339 switch (x->outer_mode.encap) {
340 case XFRM_MODE_BEET:
341 return xfrm4_beet_encap_add(x, skb);
342 case XFRM_MODE_TUNNEL:
343 return xfrm4_tunnel_encap_add(x, skb);
344 }
345
346 WARN_ON_ONCE(1);
347 return -EOPNOTSUPP;
348}
349
350static int xfrm6_prepare_output(struct xfrm_state *x, struct sk_buff *skb)
351{
352#if IS_ENABLED(CONFIG_IPV6)
353 int err;
354
355 err = xfrm_inner_extract_output(x, skb);
356 if (err)
357 return err;
358
359 skb->ignore_df = 1;
360 skb->protocol = htons(ETH_P_IPV6);
361
362 switch (x->outer_mode.encap) {
363 case XFRM_MODE_BEET:
364 return xfrm6_beet_encap_add(x, skb);
365 case XFRM_MODE_TUNNEL:
366 return xfrm6_tunnel_encap_add(x, skb);
367 default:
368 WARN_ON_ONCE(1);
369 return -EOPNOTSUPP;
370 }
371#endif
372 WARN_ON_ONCE(1);
373 return -EAFNOSUPPORT;
374}
375
376static int xfrm_outer_mode_output(struct xfrm_state *x, struct sk_buff *skb)
377{
378 switch (x->outer_mode.encap) {
379 case XFRM_MODE_BEET:
380 case XFRM_MODE_TUNNEL:
381 if (x->outer_mode.family == AF_INET)
382 return xfrm4_prepare_output(x, skb);
383 if (x->outer_mode.family == AF_INET6)
384 return xfrm6_prepare_output(x, skb);
385 break;
386 case XFRM_MODE_TRANSPORT:
387 if (x->outer_mode.family == AF_INET)
388 return xfrm4_transport_output(x, skb);
389 if (x->outer_mode.family == AF_INET6)
390 return xfrm6_transport_output(x, skb);
391 break;
392 case XFRM_MODE_ROUTEOPTIMIZATION:
393 if (x->outer_mode.family == AF_INET6)
394 return xfrm6_ro_output(x, skb);
395 WARN_ON_ONCE(1);
396 break;
397 default:
398 WARN_ON_ONCE(1);
399 break;
400 }
401
402 return -EOPNOTSUPP;
403}
404
405#if IS_ENABLED(CONFIG_NET_PKTGEN)
406int pktgen_xfrm_outer_mode_output(struct xfrm_state *x, struct sk_buff *skb)
407{
408 return xfrm_outer_mode_output(x, skb);
409}
410EXPORT_SYMBOL_GPL(pktgen_xfrm_outer_mode_output);
411#endif
412
413static int xfrm_output_one(struct sk_buff *skb, int err)
414{
415 struct dst_entry *dst = skb_dst(skb);
416 struct xfrm_state *x = dst->xfrm;
417 struct net *net = xs_net(x);
418
419 if (err <= 0)
420 goto resume;
421
422 do {
423 err = xfrm_skb_check_space(skb);
424 if (err) {
425 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
426 goto error_nolock;
427 }
428
429 skb->mark = xfrm_smark_get(skb->mark, x);
430
431 err = xfrm_outer_mode_output(x, skb);
432 if (err) {
433 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEMODEERROR);
434 goto error_nolock;
435 }
436
437 spin_lock_bh(&x->lock);
438
439 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
440 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEINVALID);
441 err = -EINVAL;
442 goto error;
443 }
444
445 err = xfrm_state_check_expire(x);
446 if (err) {
447 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEEXPIRED);
448 goto error;
449 }
450
451 err = x->repl->overflow(x, skb);
452 if (err) {
453 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATESEQERROR);
454 goto error;
455 }
456
457 x->curlft.bytes += skb->len;
458 x->curlft.packets++;
459
460 spin_unlock_bh(&x->lock);
461
462 skb_dst_force(skb);
463 if (!skb_dst(skb)) {
464 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
465 err = -EHOSTUNREACH;
466 goto error_nolock;
467 }
468
469 if (xfrm_offload(skb)) {
470 x->type_offload->encap(x, skb);
471 } else {
472 /* Inner headers are invalid now. */
473 skb->encapsulation = 0;
474
475 err = x->type->output(x, skb);
476 if (err == -EINPROGRESS)
477 goto out;
478 }
479
480resume:
481 if (err) {
482 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR);
483 goto error_nolock;
484 }
485
486 dst = skb_dst_pop(skb);
487 if (!dst) {
488 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
489 err = -EHOSTUNREACH;
490 goto error_nolock;
491 }
492 skb_dst_set(skb, dst);
493 x = dst->xfrm;
494 } while (x && !(x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL));
495
496 return 0;
497
498error:
499 spin_unlock_bh(&x->lock);
500error_nolock:
501 kfree_skb(skb);
502out:
503 return err;
504}
505
506int xfrm_output_resume(struct sk_buff *skb, int err)
507{
508 struct net *net = xs_net(skb_dst(skb)->xfrm);
509
510 while (likely((err = xfrm_output_one(skb, err)) == 0)) {
511 nf_reset_ct(skb);
512
513 err = skb_dst(skb)->ops->local_out(net, skb->sk, skb);
514 if (unlikely(err != 1))
515 goto out;
516
517 if (!skb_dst(skb)->xfrm)
518 return dst_output(net, skb->sk, skb);
519
520 err = nf_hook(skb_dst(skb)->ops->family,
521 NF_INET_POST_ROUTING, net, skb->sk, skb,
522 NULL, skb_dst(skb)->dev, xfrm_output2);
523 if (unlikely(err != 1))
524 goto out;
525 }
526
527 if (err == -EINPROGRESS)
528 err = 0;
529
530out:
531 return err;
532}
533EXPORT_SYMBOL_GPL(xfrm_output_resume);
534
535static int xfrm_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
536{
537 return xfrm_output_resume(skb, 1);
538}
539
540static int xfrm_output_gso(struct net *net, struct sock *sk, struct sk_buff *skb)
541{
542 struct sk_buff *segs, *nskb;
543
544 BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET);
545 BUILD_BUG_ON(sizeof(*IP6CB(skb)) > SKB_GSO_CB_OFFSET);
546 segs = skb_gso_segment(skb, 0);
547 kfree_skb(skb);
548 if (IS_ERR(segs))
549 return PTR_ERR(segs);
550 if (segs == NULL)
551 return -EINVAL;
552
553 skb_list_walk_safe(segs, segs, nskb) {
554 int err;
555
556 skb_mark_not_on_list(segs);
557 err = xfrm_output2(net, sk, segs);
558
559 if (unlikely(err)) {
560 kfree_skb_list(nskb);
561 return err;
562 }
563 }
564
565 return 0;
566}
567
568int xfrm_output(struct sock *sk, struct sk_buff *skb)
569{
570 struct net *net = dev_net(skb_dst(skb)->dev);
571 struct xfrm_state *x = skb_dst(skb)->xfrm;
572 int err;
573
574 switch (x->outer_mode.family) {
575 case AF_INET:
576 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
577 IPCB(skb)->flags |= IPSKB_XFRM_TRANSFORMED;
578 break;
579 case AF_INET6:
580 memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
581
582 IP6CB(skb)->flags |= IP6SKB_XFRM_TRANSFORMED;
583 break;
584 }
585
586 secpath_reset(skb);
587
588 if (xfrm_dev_offload_ok(skb, x)) {
589 struct sec_path *sp;
590
591 sp = secpath_set(skb);
592 if (!sp) {
593 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
594 kfree_skb(skb);
595 return -ENOMEM;
596 }
597 skb->encapsulation = 1;
598
599 sp->olen++;
600 sp->xvec[sp->len++] = x;
601 xfrm_state_hold(x);
602
603 if (skb_is_gso(skb)) {
604 if (skb->inner_protocol)
605 return xfrm_output_gso(net, sk, skb);
606
607 skb_shinfo(skb)->gso_type |= SKB_GSO_ESP;
608 goto out;
609 }
610
611 if (x->xso.dev && x->xso.dev->features & NETIF_F_HW_ESP_TX_CSUM)
612 goto out;
613 } else {
614 if (skb_is_gso(skb))
615 return xfrm_output_gso(net, sk, skb);
616 }
617
618 if (skb->ip_summed == CHECKSUM_PARTIAL) {
619 err = skb_checksum_help(skb);
620 if (err) {
621 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
622 kfree_skb(skb);
623 return err;
624 }
625 }
626
627out:
628 return xfrm_output2(net, sk, skb);
629}
630EXPORT_SYMBOL_GPL(xfrm_output);
631
632static int xfrm4_tunnel_check_size(struct sk_buff *skb)
633{
634 int mtu, ret = 0;
635
636 if (IPCB(skb)->flags & IPSKB_XFRM_TUNNEL_SIZE)
637 goto out;
638
639 if (!(ip_hdr(skb)->frag_off & htons(IP_DF)) || skb->ignore_df)
640 goto out;
641
642 mtu = dst_mtu(skb_dst(skb));
643 if ((!skb_is_gso(skb) && skb->len > mtu) ||
644 (skb_is_gso(skb) &&
645 !skb_gso_validate_network_len(skb, ip_skb_dst_mtu(skb->sk, skb)))) {
646 skb->protocol = htons(ETH_P_IP);
647
648 if (skb->sk)
649 xfrm_local_error(skb, mtu);
650 else
651 icmp_send(skb, ICMP_DEST_UNREACH,
652 ICMP_FRAG_NEEDED, htonl(mtu));
653 ret = -EMSGSIZE;
654 }
655out:
656 return ret;
657}
658
659static int xfrm4_extract_output(struct xfrm_state *x, struct sk_buff *skb)
660{
661 int err;
662
663 err = xfrm4_tunnel_check_size(skb);
664 if (err)
665 return err;
666
667 XFRM_MODE_SKB_CB(skb)->protocol = ip_hdr(skb)->protocol;
668
669 xfrm4_extract_header(skb);
670 return 0;
671}
672
673#if IS_ENABLED(CONFIG_IPV6)
674static int xfrm6_tunnel_check_size(struct sk_buff *skb)
675{
676 int mtu, ret = 0;
677 struct dst_entry *dst = skb_dst(skb);
678
679 if (skb->ignore_df)
680 goto out;
681
682 mtu = dst_mtu(dst);
683 if (mtu < IPV6_MIN_MTU)
684 mtu = IPV6_MIN_MTU;
685
686 if ((!skb_is_gso(skb) && skb->len > mtu) ||
687 (skb_is_gso(skb) &&
688 !skb_gso_validate_network_len(skb, ip6_skb_dst_mtu(skb)))) {
689 skb->dev = dst->dev;
690 skb->protocol = htons(ETH_P_IPV6);
691
692 if (xfrm6_local_dontfrag(skb->sk))
693 ipv6_stub->xfrm6_local_rxpmtu(skb, mtu);
694 else if (skb->sk)
695 xfrm_local_error(skb, mtu);
696 else
697 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
698 ret = -EMSGSIZE;
699 }
700out:
701 return ret;
702}
703#endif
704
705static int xfrm6_extract_output(struct xfrm_state *x, struct sk_buff *skb)
706{
707#if IS_ENABLED(CONFIG_IPV6)
708 int err;
709
710 err = xfrm6_tunnel_check_size(skb);
711 if (err)
712 return err;
713
714 XFRM_MODE_SKB_CB(skb)->protocol = ipv6_hdr(skb)->nexthdr;
715
716 xfrm6_extract_header(skb);
717 return 0;
718#else
719 WARN_ON_ONCE(1);
720 return -EAFNOSUPPORT;
721#endif
722}
723
724static int xfrm_inner_extract_output(struct xfrm_state *x, struct sk_buff *skb)
725{
726 const struct xfrm_mode *inner_mode;
727
728 if (x->sel.family == AF_UNSPEC)
729 inner_mode = xfrm_ip2inner_mode(x,
730 xfrm_af2proto(skb_dst(skb)->ops->family));
731 else
732 inner_mode = &x->inner_mode;
733
734 if (inner_mode == NULL)
735 return -EAFNOSUPPORT;
736
737 switch (inner_mode->family) {
738 case AF_INET:
739 return xfrm4_extract_output(x, skb);
740 case AF_INET6:
741 return xfrm6_extract_output(x, skb);
742 }
743
744 return -EAFNOSUPPORT;
745}
746
747void xfrm_local_error(struct sk_buff *skb, int mtu)
748{
749 unsigned int proto;
750 struct xfrm_state_afinfo *afinfo;
751
752 if (skb->protocol == htons(ETH_P_IP))
753 proto = AF_INET;
754 else if (skb->protocol == htons(ETH_P_IPV6) &&
755 skb->sk->sk_family == AF_INET6)
756 proto = AF_INET6;
757 else
758 return;
759
760 afinfo = xfrm_state_get_afinfo(proto);
761 if (afinfo) {
762 afinfo->local_error(skb, mtu);
763 rcu_read_unlock();
764 }
765}
766EXPORT_SYMBOL_GPL(xfrm_local_error);