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/gso.h>
17 #include <net/icmp.h>
18 #include <net/inet_ecn.h>
19 #include <net/xfrm.h>
20
21 #if IS_ENABLED(CONFIG_IPV6)
22 #include <net/ip6_route.h>
23 #include <net/ipv6_stubs.h>
24 #endif
25
26 #include "xfrm_inout.h"
27
28 static int xfrm_output2(struct net *net, struct sock *sk, struct sk_buff *skb);
29 static int xfrm_inner_extract_output(struct xfrm_state *x, struct sk_buff *skb);
30
xfrm_skb_check_space(struct sk_buff * skb)31 static int xfrm_skb_check_space(struct sk_buff *skb)
32 {
33 struct dst_entry *dst = skb_dst(skb);
34 int nhead = dst->header_len + LL_RESERVED_SPACE(dst->dev)
35 - skb_headroom(skb);
36 int ntail = dst->dev->needed_tailroom - skb_tailroom(skb);
37
38 if (nhead <= 0) {
39 if (ntail <= 0)
40 return 0;
41 nhead = 0;
42 } else if (ntail < 0)
43 ntail = 0;
44
45 return pskb_expand_head(skb, nhead, ntail, GFP_ATOMIC);
46 }
47
48 /* Children define the path of the packet through the
49 * Linux networking. Thus, destinations are stackable.
50 */
51
skb_dst_pop(struct sk_buff * skb)52 static struct dst_entry *skb_dst_pop(struct sk_buff *skb)
53 {
54 struct dst_entry *child = dst_clone(xfrm_dst_child(skb_dst(skb)));
55
56 skb_dst_drop(skb);
57 return child;
58 }
59
60 /* Add encapsulation header.
61 *
62 * The IP header will be moved forward to make space for the encapsulation
63 * header.
64 */
xfrm4_transport_output(struct xfrm_state * x,struct sk_buff * skb)65 static int xfrm4_transport_output(struct xfrm_state *x, struct sk_buff *skb)
66 {
67 struct iphdr *iph = ip_hdr(skb);
68 int ihl = iph->ihl * 4;
69
70 skb_set_inner_transport_header(skb, skb_transport_offset(skb));
71
72 skb_set_network_header(skb, -x->props.header_len);
73 skb->mac_header = skb->network_header +
74 offsetof(struct iphdr, protocol);
75 skb->transport_header = skb->network_header + ihl;
76 __skb_pull(skb, ihl);
77 memmove(skb_network_header(skb), iph, ihl);
78 return 0;
79 }
80
81 #if IS_ENABLED(CONFIG_IPV6_MIP6)
mip6_rthdr_offset(struct sk_buff * skb,u8 ** nexthdr,int type)82 static int mip6_rthdr_offset(struct sk_buff *skb, u8 **nexthdr, int type)
83 {
84 const unsigned char *nh = skb_network_header(skb);
85 unsigned int offset = sizeof(struct ipv6hdr);
86 unsigned int packet_len;
87 int found_rhdr = 0;
88
89 packet_len = skb_tail_pointer(skb) - nh;
90 *nexthdr = &ipv6_hdr(skb)->nexthdr;
91
92 while (offset <= packet_len) {
93 struct ipv6_opt_hdr *exthdr;
94
95 switch (**nexthdr) {
96 case NEXTHDR_HOP:
97 break;
98 case NEXTHDR_ROUTING:
99 if (type == IPPROTO_ROUTING && offset + 3 <= packet_len) {
100 struct ipv6_rt_hdr *rt;
101
102 rt = (struct ipv6_rt_hdr *)(nh + offset);
103 if (rt->type != 0)
104 return offset;
105 }
106 found_rhdr = 1;
107 break;
108 case NEXTHDR_DEST:
109 /* HAO MUST NOT appear more than once.
110 * XXX: It is better to try to find by the end of
111 * XXX: packet if HAO exists.
112 */
113 if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0) {
114 net_dbg_ratelimited("mip6: hao exists already, override\n");
115 return offset;
116 }
117
118 if (found_rhdr)
119 return offset;
120
121 break;
122 default:
123 return offset;
124 }
125
126 if (offset + sizeof(struct ipv6_opt_hdr) > packet_len)
127 return -EINVAL;
128
129 exthdr = (struct ipv6_opt_hdr *)(skb_network_header(skb) +
130 offset);
131 offset += ipv6_optlen(exthdr);
132 if (offset > IPV6_MAXPLEN)
133 return -EINVAL;
134 *nexthdr = &exthdr->nexthdr;
135 }
136
137 return -EINVAL;
138 }
139 #endif
140
141 #if IS_ENABLED(CONFIG_IPV6)
xfrm6_hdr_offset(struct xfrm_state * x,struct sk_buff * skb,u8 ** prevhdr)142 static int xfrm6_hdr_offset(struct xfrm_state *x, struct sk_buff *skb, u8 **prevhdr)
143 {
144 switch (x->type->proto) {
145 #if IS_ENABLED(CONFIG_IPV6_MIP6)
146 case IPPROTO_DSTOPTS:
147 case IPPROTO_ROUTING:
148 return mip6_rthdr_offset(skb, prevhdr, x->type->proto);
149 #endif
150 default:
151 break;
152 }
153
154 return ip6_find_1stfragopt(skb, prevhdr);
155 }
156 #endif
157
158 /* Add encapsulation header.
159 *
160 * The IP header and mutable extension headers will be moved forward to make
161 * space for the encapsulation header.
162 */
xfrm6_transport_output(struct xfrm_state * x,struct sk_buff * skb)163 static int xfrm6_transport_output(struct xfrm_state *x, struct sk_buff *skb)
164 {
165 #if IS_ENABLED(CONFIG_IPV6)
166 struct ipv6hdr *iph;
167 u8 *prevhdr;
168 int hdr_len;
169
170 iph = ipv6_hdr(skb);
171 skb_set_inner_transport_header(skb, skb_transport_offset(skb));
172
173 hdr_len = xfrm6_hdr_offset(x, skb, &prevhdr);
174 if (hdr_len < 0)
175 return hdr_len;
176 skb_set_mac_header(skb,
177 (prevhdr - x->props.header_len) - skb->data);
178 skb_set_network_header(skb, -x->props.header_len);
179 skb->transport_header = skb->network_header + hdr_len;
180 __skb_pull(skb, hdr_len);
181 memmove(ipv6_hdr(skb), iph, hdr_len);
182 return 0;
183 #else
184 WARN_ON_ONCE(1);
185 return -EAFNOSUPPORT;
186 #endif
187 }
188
189 /* Add route optimization header space.
190 *
191 * The IP header and mutable extension headers will be moved forward to make
192 * space for the route optimization header.
193 */
xfrm6_ro_output(struct xfrm_state * x,struct sk_buff * skb)194 static int xfrm6_ro_output(struct xfrm_state *x, struct sk_buff *skb)
195 {
196 #if IS_ENABLED(CONFIG_IPV6)
197 struct ipv6hdr *iph;
198 u8 *prevhdr;
199 int hdr_len;
200
201 iph = ipv6_hdr(skb);
202
203 hdr_len = xfrm6_hdr_offset(x, skb, &prevhdr);
204 if (hdr_len < 0)
205 return hdr_len;
206 skb_set_mac_header(skb,
207 (prevhdr - x->props.header_len) - skb->data);
208 skb_set_network_header(skb, -x->props.header_len);
209 skb->transport_header = skb->network_header + hdr_len;
210 __skb_pull(skb, hdr_len);
211 memmove(ipv6_hdr(skb), iph, hdr_len);
212
213 return 0;
214 #else
215 WARN_ON_ONCE(1);
216 return -EAFNOSUPPORT;
217 #endif
218 }
219
220 /* Add encapsulation header.
221 *
222 * The top IP header will be constructed per draft-nikander-esp-beet-mode-06.txt.
223 */
xfrm4_beet_encap_add(struct xfrm_state * x,struct sk_buff * skb)224 static int xfrm4_beet_encap_add(struct xfrm_state *x, struct sk_buff *skb)
225 {
226 struct ip_beet_phdr *ph;
227 struct iphdr *top_iph;
228 int hdrlen, optlen;
229
230 hdrlen = 0;
231 optlen = XFRM_MODE_SKB_CB(skb)->optlen;
232 if (unlikely(optlen))
233 hdrlen += IPV4_BEET_PHMAXLEN - (optlen & 4);
234
235 skb_set_network_header(skb, -x->props.header_len - hdrlen +
236 (XFRM_MODE_SKB_CB(skb)->ihl - sizeof(*top_iph)));
237 if (x->sel.family != AF_INET6)
238 skb->network_header += IPV4_BEET_PHMAXLEN;
239 skb->mac_header = skb->network_header +
240 offsetof(struct iphdr, protocol);
241 skb->transport_header = skb->network_header + sizeof(*top_iph);
242
243 xfrm4_beet_make_header(skb);
244
245 ph = __skb_pull(skb, XFRM_MODE_SKB_CB(skb)->ihl - hdrlen);
246
247 top_iph = ip_hdr(skb);
248
249 if (unlikely(optlen)) {
250 if (WARN_ON(optlen < 0))
251 return -EINVAL;
252
253 ph->padlen = 4 - (optlen & 4);
254 ph->hdrlen = optlen / 8;
255 ph->nexthdr = top_iph->protocol;
256 if (ph->padlen)
257 memset(ph + 1, IPOPT_NOP, ph->padlen);
258
259 top_iph->protocol = IPPROTO_BEETPH;
260 top_iph->ihl = sizeof(struct iphdr) / 4;
261 }
262
263 top_iph->saddr = x->props.saddr.a4;
264 top_iph->daddr = x->id.daddr.a4;
265
266 return 0;
267 }
268
269 /* Add encapsulation header.
270 *
271 * The top IP header will be constructed per RFC 2401.
272 */
xfrm4_tunnel_encap_add(struct xfrm_state * x,struct sk_buff * skb)273 static int xfrm4_tunnel_encap_add(struct xfrm_state *x, struct sk_buff *skb)
274 {
275 bool small_ipv6 = (skb->protocol == htons(ETH_P_IPV6)) && (skb->len <= IPV6_MIN_MTU);
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) || small_ipv6 ?
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)
xfrm6_tunnel_encap_add(struct xfrm_state * x,struct sk_buff * skb)319 static 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
xfrm6_beet_encap_add(struct xfrm_state * x,struct sk_buff * skb)354 static 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 */
xfrm4_prepare_output(struct xfrm_state * x,struct sk_buff * skb)405 static 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->props.mode) {
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
xfrm6_prepare_output(struct xfrm_state * x,struct sk_buff * skb)427 static 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->props.mode) {
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
xfrm_outer_mode_output(struct xfrm_state * x,struct sk_buff * skb)453 static int xfrm_outer_mode_output(struct xfrm_state *x, struct sk_buff *skb)
454 {
455 switch (x->props.mode) {
456 case XFRM_MODE_BEET:
457 case XFRM_MODE_TUNNEL:
458 if (x->props.family == AF_INET)
459 return xfrm4_prepare_output(x, skb);
460 if (x->props.family == AF_INET6)
461 return xfrm6_prepare_output(x, skb);
462 break;
463 case XFRM_MODE_TRANSPORT:
464 if (x->props.family == AF_INET)
465 return xfrm4_transport_output(x, skb);
466 if (x->props.family == AF_INET6)
467 return xfrm6_transport_output(x, skb);
468 break;
469 case XFRM_MODE_ROUTEOPTIMIZATION:
470 if (x->props.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)
pktgen_xfrm_outer_mode_output(struct xfrm_state * x,struct sk_buff * skb)483 int pktgen_xfrm_outer_mode_output(struct xfrm_state *x, struct sk_buff *skb)
484 {
485 return xfrm_outer_mode_output(x, skb);
486 }
487 EXPORT_SYMBOL_GPL(pktgen_xfrm_outer_mode_output);
488 #endif
489
xfrm_output_one(struct sk_buff * skb,int err)490 static 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 || x->xso.type == XFRM_DEV_OFFLOAD_PACKET)
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 x->lastused = ktime_get_real_seconds();
537
538 spin_unlock_bh(&x->lock);
539
540 skb_dst_force(skb);
541 if (!skb_dst(skb)) {
542 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
543 err = -EHOSTUNREACH;
544 goto error_nolock;
545 }
546
547 if (xfrm_offload(skb)) {
548 x->type_offload->encap(x, skb);
549 } else {
550 /* Inner headers are invalid now. */
551 skb->encapsulation = 0;
552
553 err = x->type->output(x, skb);
554 if (err == -EINPROGRESS)
555 goto out;
556 }
557
558 resume:
559 if (err) {
560 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR);
561 goto error_nolock;
562 }
563
564 dst = skb_dst_pop(skb);
565 if (!dst) {
566 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
567 err = -EHOSTUNREACH;
568 goto error_nolock;
569 }
570 skb_dst_set(skb, dst);
571 x = dst->xfrm;
572 } while (x && !(x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL));
573
574 return 0;
575
576 error:
577 spin_unlock_bh(&x->lock);
578 error_nolock:
579 kfree_skb(skb);
580 out:
581 return err;
582 }
583
xfrm_output_resume(struct sock * sk,struct sk_buff * skb,int err)584 int xfrm_output_resume(struct sock *sk, struct sk_buff *skb, int err)
585 {
586 struct net *net = xs_net(skb_dst(skb)->xfrm);
587
588 while (likely((err = xfrm_output_one(skb, err)) == 0)) {
589 nf_reset_ct(skb);
590
591 err = skb_dst(skb)->ops->local_out(net, sk, skb);
592 if (unlikely(err != 1))
593 goto out;
594
595 if (!skb_dst(skb)->xfrm)
596 return dst_output(net, sk, skb);
597
598 err = nf_hook(skb_dst(skb)->ops->family,
599 NF_INET_POST_ROUTING, net, sk, skb,
600 NULL, skb_dst(skb)->dev, xfrm_output2);
601 if (unlikely(err != 1))
602 goto out;
603 }
604
605 if (err == -EINPROGRESS)
606 err = 0;
607
608 out:
609 return err;
610 }
611 EXPORT_SYMBOL_GPL(xfrm_output_resume);
612
xfrm_dev_direct_output(struct sock * sk,struct xfrm_state * x,struct sk_buff * skb)613 static int xfrm_dev_direct_output(struct sock *sk, struct xfrm_state *x,
614 struct sk_buff *skb)
615 {
616 struct dst_entry *dst = skb_dst(skb);
617 struct net *net = xs_net(x);
618 int err;
619
620 dst = skb_dst_pop(skb);
621 if (!dst) {
622 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
623 kfree_skb(skb);
624 return -EHOSTUNREACH;
625 }
626 skb_dst_set(skb, dst);
627 nf_reset_ct(skb);
628
629 err = skb_dst(skb)->ops->local_out(net, sk, skb);
630 if (unlikely(err != 1)) {
631 kfree_skb(skb);
632 return err;
633 }
634
635 /* In transport mode, network destination is
636 * directly reachable, while in tunnel mode,
637 * inner packet network may not be. In packet
638 * offload type, HW is responsible for hard
639 * header packet mangling so directly xmit skb
640 * to netdevice.
641 */
642 skb->dev = x->xso.dev;
643 __skb_push(skb, skb->dev->hard_header_len);
644 return dev_queue_xmit(skb);
645 }
646
xfrm_output2(struct net * net,struct sock * sk,struct sk_buff * skb)647 static int xfrm_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
648 {
649 return xfrm_output_resume(sk, skb, 1);
650 }
651
xfrm_output_gso(struct net * net,struct sock * sk,struct sk_buff * skb)652 static int xfrm_output_gso(struct net *net, struct sock *sk, struct sk_buff *skb)
653 {
654 struct sk_buff *segs, *nskb;
655
656 BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET);
657 BUILD_BUG_ON(sizeof(*IP6CB(skb)) > SKB_GSO_CB_OFFSET);
658 segs = skb_gso_segment(skb, 0);
659 kfree_skb(skb);
660 if (IS_ERR(segs))
661 return PTR_ERR(segs);
662 if (segs == NULL)
663 return -EINVAL;
664
665 skb_list_walk_safe(segs, segs, nskb) {
666 int err;
667
668 skb_mark_not_on_list(segs);
669 err = xfrm_output2(net, sk, segs);
670
671 if (unlikely(err)) {
672 kfree_skb_list(nskb);
673 return err;
674 }
675 }
676
677 return 0;
678 }
679
680 /* For partial checksum offload, the outer header checksum is calculated
681 * by software and the inner header checksum is calculated by hardware.
682 * This requires hardware to know the inner packet type to calculate
683 * the inner header checksum. Save inner ip protocol here to avoid
684 * traversing the packet in the vendor's xmit code.
685 * For IPsec tunnel mode save the ip protocol from the IP header of the
686 * plain text packet. Otherwise If the encap type is IPIP, just save
687 * skb->inner_ipproto in any other case get the ip protocol from the IP
688 * header.
689 */
xfrm_get_inner_ipproto(struct sk_buff * skb,struct xfrm_state * x)690 static void xfrm_get_inner_ipproto(struct sk_buff *skb, struct xfrm_state *x)
691 {
692 struct xfrm_offload *xo = xfrm_offload(skb);
693 const struct ethhdr *eth;
694
695 if (!xo)
696 return;
697
698 if (x->outer_mode.encap == XFRM_MODE_TUNNEL) {
699 switch (x->outer_mode.family) {
700 case AF_INET:
701 xo->inner_ipproto = ip_hdr(skb)->protocol;
702 break;
703 case AF_INET6:
704 xo->inner_ipproto = ipv6_hdr(skb)->nexthdr;
705 break;
706 default:
707 break;
708 }
709
710 return;
711 }
712
713 /* non-Tunnel Mode */
714 if (!skb->encapsulation)
715 return;
716
717 if (skb->inner_protocol_type == ENCAP_TYPE_IPPROTO) {
718 xo->inner_ipproto = skb->inner_ipproto;
719 return;
720 }
721
722 if (skb->inner_protocol_type != ENCAP_TYPE_ETHER)
723 return;
724
725 eth = (struct ethhdr *)skb_inner_mac_header(skb);
726
727 switch (ntohs(eth->h_proto)) {
728 case ETH_P_IPV6:
729 xo->inner_ipproto = inner_ipv6_hdr(skb)->nexthdr;
730 break;
731 case ETH_P_IP:
732 xo->inner_ipproto = inner_ip_hdr(skb)->protocol;
733 break;
734 }
735 }
736
xfrm_output(struct sock * sk,struct sk_buff * skb)737 int xfrm_output(struct sock *sk, struct sk_buff *skb)
738 {
739 struct net *net = dev_net(skb_dst(skb)->dev);
740 struct xfrm_state *x = skb_dst(skb)->xfrm;
741 int family;
742 int err;
743
744 family = (x->xso.type != XFRM_DEV_OFFLOAD_PACKET) ? x->outer_mode.family
745 : skb_dst(skb)->ops->family;
746
747 switch (family) {
748 case AF_INET:
749 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
750 IPCB(skb)->flags |= IPSKB_XFRM_TRANSFORMED;
751 break;
752 case AF_INET6:
753 memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
754
755 IP6CB(skb)->flags |= IP6SKB_XFRM_TRANSFORMED;
756 break;
757 }
758
759 if (x->xso.type == XFRM_DEV_OFFLOAD_PACKET) {
760 if (!xfrm_dev_offload_ok(skb, x)) {
761 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
762 kfree_skb(skb);
763 return -EHOSTUNREACH;
764 }
765
766 /* Exclusive direct xmit for tunnel mode, as
767 * some filtering or matching rules may apply
768 * in transport mode.
769 */
770 if (x->props.mode == XFRM_MODE_TUNNEL)
771 return xfrm_dev_direct_output(sk, x, skb);
772
773 return xfrm_output_resume(sk, skb, 0);
774 }
775
776 secpath_reset(skb);
777
778 if (xfrm_dev_offload_ok(skb, x)) {
779 struct sec_path *sp;
780
781 sp = secpath_set(skb);
782 if (!sp) {
783 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
784 kfree_skb(skb);
785 return -ENOMEM;
786 }
787
788 sp->olen++;
789 sp->xvec[sp->len++] = x;
790 xfrm_state_hold(x);
791
792 xfrm_get_inner_ipproto(skb, x);
793 skb->encapsulation = 1;
794
795 if (skb_is_gso(skb)) {
796 if (skb->inner_protocol && x->props.mode == XFRM_MODE_TUNNEL)
797 return xfrm_output_gso(net, sk, skb);
798
799 skb_shinfo(skb)->gso_type |= SKB_GSO_ESP;
800 goto out;
801 }
802
803 if (x->xso.dev && x->xso.dev->features & NETIF_F_HW_ESP_TX_CSUM)
804 goto out;
805 } else {
806 if (skb_is_gso(skb))
807 return xfrm_output_gso(net, sk, skb);
808 }
809
810 if (skb->ip_summed == CHECKSUM_PARTIAL) {
811 err = skb_checksum_help(skb);
812 if (err) {
813 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
814 kfree_skb(skb);
815 return err;
816 }
817 }
818
819 out:
820 return xfrm_output2(net, sk, skb);
821 }
822 EXPORT_SYMBOL_GPL(xfrm_output);
823
xfrm4_tunnel_check_size(struct sk_buff * skb)824 static int xfrm4_tunnel_check_size(struct sk_buff *skb)
825 {
826 int mtu, ret = 0;
827
828 if (IPCB(skb)->flags & IPSKB_XFRM_TUNNEL_SIZE)
829 goto out;
830
831 if (!(ip_hdr(skb)->frag_off & htons(IP_DF)) || skb->ignore_df)
832 goto out;
833
834 mtu = dst_mtu(skb_dst(skb));
835 if ((!skb_is_gso(skb) && skb->len > mtu) ||
836 (skb_is_gso(skb) &&
837 !skb_gso_validate_network_len(skb, ip_skb_dst_mtu(skb->sk, skb)))) {
838 skb->protocol = htons(ETH_P_IP);
839
840 if (skb->sk)
841 xfrm_local_error(skb, mtu);
842 else
843 icmp_send(skb, ICMP_DEST_UNREACH,
844 ICMP_FRAG_NEEDED, htonl(mtu));
845 ret = -EMSGSIZE;
846 }
847 out:
848 return ret;
849 }
850
xfrm4_extract_output(struct xfrm_state * x,struct sk_buff * skb)851 static int xfrm4_extract_output(struct xfrm_state *x, struct sk_buff *skb)
852 {
853 int err;
854
855 if (x->outer_mode.encap == XFRM_MODE_BEET &&
856 ip_is_fragment(ip_hdr(skb))) {
857 net_warn_ratelimited("BEET mode doesn't support inner IPv4 fragments\n");
858 return -EAFNOSUPPORT;
859 }
860
861 err = xfrm4_tunnel_check_size(skb);
862 if (err)
863 return err;
864
865 XFRM_MODE_SKB_CB(skb)->protocol = ip_hdr(skb)->protocol;
866
867 xfrm4_extract_header(skb);
868 return 0;
869 }
870
871 #if IS_ENABLED(CONFIG_IPV6)
xfrm6_tunnel_check_size(struct sk_buff * skb)872 static int xfrm6_tunnel_check_size(struct sk_buff *skb)
873 {
874 int mtu, ret = 0;
875 struct dst_entry *dst = skb_dst(skb);
876
877 if (skb->ignore_df)
878 goto out;
879
880 mtu = dst_mtu(dst);
881 if (mtu < IPV6_MIN_MTU)
882 mtu = IPV6_MIN_MTU;
883
884 if ((!skb_is_gso(skb) && skb->len > mtu) ||
885 (skb_is_gso(skb) &&
886 !skb_gso_validate_network_len(skb, ip6_skb_dst_mtu(skb)))) {
887 skb->dev = dst->dev;
888 skb->protocol = htons(ETH_P_IPV6);
889
890 if (xfrm6_local_dontfrag(skb->sk))
891 ipv6_stub->xfrm6_local_rxpmtu(skb, mtu);
892 else if (skb->sk)
893 xfrm_local_error(skb, mtu);
894 else
895 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
896 ret = -EMSGSIZE;
897 }
898 out:
899 return ret;
900 }
901 #endif
902
xfrm6_extract_output(struct xfrm_state * x,struct sk_buff * skb)903 static int xfrm6_extract_output(struct xfrm_state *x, struct sk_buff *skb)
904 {
905 #if IS_ENABLED(CONFIG_IPV6)
906 int err;
907
908 err = xfrm6_tunnel_check_size(skb);
909 if (err)
910 return err;
911
912 XFRM_MODE_SKB_CB(skb)->protocol = ipv6_hdr(skb)->nexthdr;
913
914 xfrm6_extract_header(skb);
915 return 0;
916 #else
917 WARN_ON_ONCE(1);
918 return -EAFNOSUPPORT;
919 #endif
920 }
921
xfrm_inner_extract_output(struct xfrm_state * x,struct sk_buff * skb)922 static int xfrm_inner_extract_output(struct xfrm_state *x, struct sk_buff *skb)
923 {
924 switch (skb->protocol) {
925 case htons(ETH_P_IP):
926 return xfrm4_extract_output(x, skb);
927 case htons(ETH_P_IPV6):
928 return xfrm6_extract_output(x, skb);
929 }
930
931 return -EAFNOSUPPORT;
932 }
933
xfrm_local_error(struct sk_buff * skb,int mtu)934 void xfrm_local_error(struct sk_buff *skb, int mtu)
935 {
936 unsigned int proto;
937 struct xfrm_state_afinfo *afinfo;
938
939 if (skb->protocol == htons(ETH_P_IP))
940 proto = AF_INET;
941 else if (skb->protocol == htons(ETH_P_IPV6) &&
942 skb->sk->sk_family == AF_INET6)
943 proto = AF_INET6;
944 else
945 return;
946
947 afinfo = xfrm_state_get_afinfo(proto);
948 if (afinfo) {
949 afinfo->local_error(skb, mtu);
950 rcu_read_unlock();
951 }
952 }
953 EXPORT_SYMBOL_GPL(xfrm_local_error);
954