xref: /openbmc/linux/net/xfrm/xfrm_output.c (revision babbdf5b) 
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 
27 static int xfrm_output2(struct net *net, struct sock *sk, struct sk_buff *skb);
28 static int xfrm_inner_extract_output(struct xfrm_state *x, struct sk_buff *skb);
29 
30 static 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 
51 static 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  */
64 static 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  */
85 static 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  */
116 static 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  */
148 static 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  */
197 static 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)
242 static 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 
277 static 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  */
328 static 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 
350 static 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 
376 static 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)
406 int pktgen_xfrm_outer_mode_output(struct xfrm_state *x, struct sk_buff *skb)
407 {
408 	return xfrm_outer_mode_output(x, skb);
409 }
410 EXPORT_SYMBOL_GPL(pktgen_xfrm_outer_mode_output);
411 #endif
412 
413 static 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 
480 resume:
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 
498 error:
499 	spin_unlock_bh(&x->lock);
500 error_nolock:
501 	kfree_skb(skb);
502 out:
503 	return err;
504 }
505 
506 int xfrm_output_resume(struct sock *sk, 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, sk, skb);
514 		if (unlikely(err != 1))
515 			goto out;
516 
517 		if (!skb_dst(skb)->xfrm)
518 			return dst_output(net, sk, skb);
519 
520 		err = nf_hook(skb_dst(skb)->ops->family,
521 			      NF_INET_POST_ROUTING, net, 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 
530 out:
531 	return err;
532 }
533 EXPORT_SYMBOL_GPL(xfrm_output_resume);
534 
535 static int xfrm_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
536 {
537 	return xfrm_output_resume(sk, skb, 1);
538 }
539 
540 static 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 
568 int 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 
627 out:
628 	return xfrm_output2(net, sk, skb);
629 }
630 EXPORT_SYMBOL_GPL(xfrm_output);
631 
632 static 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 	}
655 out:
656 	return ret;
657 }
658 
659 static int xfrm4_extract_output(struct xfrm_state *x, struct sk_buff *skb)
660 {
661 	int err;
662 
663 	if (x->outer_mode.encap == XFRM_MODE_BEET &&
664 	    ip_is_fragment(ip_hdr(skb))) {
665 		net_warn_ratelimited("BEET mode doesn't support inner IPv4 fragments\n");
666 		return -EAFNOSUPPORT;
667 	}
668 
669 	err = xfrm4_tunnel_check_size(skb);
670 	if (err)
671 		return err;
672 
673 	XFRM_MODE_SKB_CB(skb)->protocol = ip_hdr(skb)->protocol;
674 
675 	xfrm4_extract_header(skb);
676 	return 0;
677 }
678 
679 #if IS_ENABLED(CONFIG_IPV6)
680 static int xfrm6_tunnel_check_size(struct sk_buff *skb)
681 {
682 	int mtu, ret = 0;
683 	struct dst_entry *dst = skb_dst(skb);
684 
685 	if (skb->ignore_df)
686 		goto out;
687 
688 	mtu = dst_mtu(dst);
689 	if (mtu < IPV6_MIN_MTU)
690 		mtu = IPV6_MIN_MTU;
691 
692 	if ((!skb_is_gso(skb) && skb->len > mtu) ||
693 	    (skb_is_gso(skb) &&
694 	     !skb_gso_validate_network_len(skb, ip6_skb_dst_mtu(skb)))) {
695 		skb->dev = dst->dev;
696 		skb->protocol = htons(ETH_P_IPV6);
697 
698 		if (xfrm6_local_dontfrag(skb->sk))
699 			ipv6_stub->xfrm6_local_rxpmtu(skb, mtu);
700 		else if (skb->sk)
701 			xfrm_local_error(skb, mtu);
702 		else
703 			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
704 		ret = -EMSGSIZE;
705 	}
706 out:
707 	return ret;
708 }
709 #endif
710 
711 static int xfrm6_extract_output(struct xfrm_state *x, struct sk_buff *skb)
712 {
713 #if IS_ENABLED(CONFIG_IPV6)
714 	unsigned int ptr = 0;
715 	int err;
716 
717 	if (x->outer_mode.encap == XFRM_MODE_BEET &&
718 	    ipv6_find_hdr(skb, &ptr, NEXTHDR_FRAGMENT, NULL, NULL) >= 0) {
719 		net_warn_ratelimited("BEET mode doesn't support inner IPv6 fragments\n");
720 		return -EAFNOSUPPORT;
721 	}
722 
723 	err = xfrm6_tunnel_check_size(skb);
724 	if (err)
725 		return err;
726 
727 	XFRM_MODE_SKB_CB(skb)->protocol = ipv6_hdr(skb)->nexthdr;
728 
729 	xfrm6_extract_header(skb);
730 	return 0;
731 #else
732 	WARN_ON_ONCE(1);
733 	return -EAFNOSUPPORT;
734 #endif
735 }
736 
737 static int xfrm_inner_extract_output(struct xfrm_state *x, struct sk_buff *skb)
738 {
739 	const struct xfrm_mode *inner_mode;
740 
741 	if (x->sel.family == AF_UNSPEC)
742 		inner_mode = xfrm_ip2inner_mode(x,
743 				xfrm_af2proto(skb_dst(skb)->ops->family));
744 	else
745 		inner_mode = &x->inner_mode;
746 
747 	if (inner_mode == NULL)
748 		return -EAFNOSUPPORT;
749 
750 	switch (inner_mode->family) {
751 	case AF_INET:
752 		return xfrm4_extract_output(x, skb);
753 	case AF_INET6:
754 		return xfrm6_extract_output(x, skb);
755 	}
756 
757 	return -EAFNOSUPPORT;
758 }
759 
760 void xfrm_local_error(struct sk_buff *skb, int mtu)
761 {
762 	unsigned int proto;
763 	struct xfrm_state_afinfo *afinfo;
764 
765 	if (skb->protocol == htons(ETH_P_IP))
766 		proto = AF_INET;
767 	else if (skb->protocol == htons(ETH_P_IPV6) &&
768 		 skb->sk->sk_family == AF_INET6)
769 		proto = AF_INET6;
770 	else
771 		return;
772 
773 	afinfo = xfrm_state_get_afinfo(proto);
774 	if (afinfo) {
775 		afinfo->local_error(skb, mtu);
776 		rcu_read_unlock();
777 	}
778 }
779 EXPORT_SYMBOL_GPL(xfrm_local_error);
780