xref: /linux/net/xfrm/xfrm_output.c (revision 376b1446153ca67e7028e6b9555d9b17477f568b)
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 #if IS_ENABLED(CONFIG_IPV6_MIP6)
81 static 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)
141 static 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  */
162 static 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  */
193 static 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 	return 0;
213 #else
214 	WARN_ON_ONCE(1);
215 	return -EAFNOSUPPORT;
216 #endif
217 }
218 
219 /* Add encapsulation header.
220  *
221  * The top IP header will be constructed per draft-nikander-esp-beet-mode-06.txt.
222  */
223 static int xfrm4_beet_encap_add(struct xfrm_state *x, struct sk_buff *skb)
224 {
225 	struct ip_beet_phdr *ph;
226 	struct iphdr *top_iph;
227 	int hdrlen, optlen;
228 
229 	hdrlen = 0;
230 	optlen = XFRM_MODE_SKB_CB(skb)->optlen;
231 	if (unlikely(optlen))
232 		hdrlen += IPV4_BEET_PHMAXLEN - (optlen & 4);
233 
234 	skb_set_network_header(skb, -x->props.header_len - hdrlen +
235 			       (XFRM_MODE_SKB_CB(skb)->ihl - sizeof(*top_iph)));
236 	if (x->sel.family != AF_INET6)
237 		skb->network_header += IPV4_BEET_PHMAXLEN;
238 	skb->mac_header = skb->network_header +
239 			  offsetof(struct iphdr, protocol);
240 	skb->transport_header = skb->network_header + sizeof(*top_iph);
241 
242 	xfrm4_beet_make_header(skb);
243 
244 	ph = __skb_pull(skb, XFRM_MODE_SKB_CB(skb)->ihl - hdrlen);
245 
246 	top_iph = ip_hdr(skb);
247 
248 	if (unlikely(optlen)) {
249 		if (WARN_ON(optlen < 0))
250 			return -EINVAL;
251 
252 		ph->padlen = 4 - (optlen & 4);
253 		ph->hdrlen = optlen / 8;
254 		ph->nexthdr = top_iph->protocol;
255 		if (ph->padlen)
256 			memset(ph + 1, IPOPT_NOP, ph->padlen);
257 
258 		top_iph->protocol = IPPROTO_BEETPH;
259 		top_iph->ihl = sizeof(struct iphdr) / 4;
260 	}
261 
262 	top_iph->saddr = x->props.saddr.a4;
263 	top_iph->daddr = x->id.daddr.a4;
264 
265 	return 0;
266 }
267 
268 /* Add encapsulation header.
269  *
270  * The top IP header will be constructed per RFC 2401.
271  */
272 static int xfrm4_tunnel_encap_add(struct xfrm_state *x, struct sk_buff *skb)
273 {
274 	bool small_ipv6 = (skb->protocol == htons(ETH_P_IPV6)) && (skb->len <= IPV6_MIN_MTU);
275 	struct dst_entry *dst = skb_dst(skb);
276 	struct iphdr *top_iph;
277 	int flags;
278 
279 	skb_set_inner_network_header(skb, skb_network_offset(skb));
280 	skb_set_inner_transport_header(skb, skb_transport_offset(skb));
281 
282 	skb_set_network_header(skb, -x->props.header_len);
283 	skb->mac_header = skb->network_header +
284 			  offsetof(struct iphdr, protocol);
285 	skb->transport_header = skb->network_header + sizeof(*top_iph);
286 	top_iph = ip_hdr(skb);
287 
288 	top_iph->ihl = 5;
289 	top_iph->version = 4;
290 
291 	top_iph->protocol = xfrm_af2proto(skb_dst(skb)->ops->family);
292 
293 	/* DS disclosing depends on XFRM_SA_XFLAG_DONT_ENCAP_DSCP */
294 	if (x->props.extra_flags & XFRM_SA_XFLAG_DONT_ENCAP_DSCP)
295 		top_iph->tos = 0;
296 	else
297 		top_iph->tos = XFRM_MODE_SKB_CB(skb)->tos;
298 	top_iph->tos = INET_ECN_encapsulate(top_iph->tos,
299 					    XFRM_MODE_SKB_CB(skb)->tos);
300 
301 	flags = x->props.flags;
302 	if (flags & XFRM_STATE_NOECN)
303 		IP_ECN_clear(top_iph);
304 
305 	top_iph->frag_off = (flags & XFRM_STATE_NOPMTUDISC) || small_ipv6 ?
306 		0 : (XFRM_MODE_SKB_CB(skb)->frag_off & htons(IP_DF));
307 
308 	top_iph->ttl = ip4_dst_hoplimit(xfrm_dst_child(dst));
309 
310 	top_iph->saddr = x->props.saddr.a4;
311 	top_iph->daddr = x->id.daddr.a4;
312 	ip_select_ident(dev_net(dst->dev), skb, NULL);
313 
314 	return 0;
315 }
316 
317 #if IS_ENABLED(CONFIG_IPV6)
318 static int xfrm6_tunnel_encap_add(struct xfrm_state *x, struct sk_buff *skb)
319 {
320 	struct dst_entry *dst = skb_dst(skb);
321 	struct ipv6hdr *top_iph;
322 	int dsfield;
323 
324 	skb_set_inner_network_header(skb, skb_network_offset(skb));
325 	skb_set_inner_transport_header(skb, skb_transport_offset(skb));
326 
327 	skb_set_network_header(skb, -x->props.header_len);
328 	skb->mac_header = skb->network_header +
329 			  offsetof(struct ipv6hdr, nexthdr);
330 	skb->transport_header = skb->network_header + sizeof(*top_iph);
331 	top_iph = ipv6_hdr(skb);
332 
333 	top_iph->version = 6;
334 
335 	memcpy(top_iph->flow_lbl, XFRM_MODE_SKB_CB(skb)->flow_lbl,
336 	       sizeof(top_iph->flow_lbl));
337 	top_iph->nexthdr = xfrm_af2proto(skb_dst(skb)->ops->family);
338 
339 	if (x->props.extra_flags & XFRM_SA_XFLAG_DONT_ENCAP_DSCP)
340 		dsfield = 0;
341 	else
342 		dsfield = XFRM_MODE_SKB_CB(skb)->tos;
343 	dsfield = INET_ECN_encapsulate(dsfield, XFRM_MODE_SKB_CB(skb)->tos);
344 	if (x->props.flags & XFRM_STATE_NOECN)
345 		dsfield &= ~INET_ECN_MASK;
346 	ipv6_change_dsfield(top_iph, 0, dsfield);
347 	top_iph->hop_limit = ip6_dst_hoplimit(xfrm_dst_child(dst));
348 	top_iph->saddr = *(struct in6_addr *)&x->props.saddr;
349 	top_iph->daddr = *(struct in6_addr *)&x->id.daddr;
350 	return 0;
351 }
352 
353 static int xfrm6_beet_encap_add(struct xfrm_state *x, struct sk_buff *skb)
354 {
355 	struct ipv6hdr *top_iph;
356 	struct ip_beet_phdr *ph;
357 	int optlen, hdr_len;
358 
359 	hdr_len = 0;
360 	optlen = XFRM_MODE_SKB_CB(skb)->optlen;
361 	if (unlikely(optlen))
362 		hdr_len += IPV4_BEET_PHMAXLEN - (optlen & 4);
363 
364 	skb_set_network_header(skb, -x->props.header_len - hdr_len);
365 	if (x->sel.family != AF_INET6)
366 		skb->network_header += IPV4_BEET_PHMAXLEN;
367 	skb->mac_header = skb->network_header +
368 			  offsetof(struct ipv6hdr, nexthdr);
369 	skb->transport_header = skb->network_header + sizeof(*top_iph);
370 	ph = __skb_pull(skb, XFRM_MODE_SKB_CB(skb)->ihl - hdr_len);
371 
372 	xfrm6_beet_make_header(skb);
373 
374 	top_iph = ipv6_hdr(skb);
375 	if (unlikely(optlen)) {
376 		if (WARN_ON(optlen < 0))
377 			return -EINVAL;
378 
379 		ph->padlen = 4 - (optlen & 4);
380 		ph->hdrlen = optlen / 8;
381 		ph->nexthdr = top_iph->nexthdr;
382 		if (ph->padlen)
383 			memset(ph + 1, IPOPT_NOP, ph->padlen);
384 
385 		top_iph->nexthdr = IPPROTO_BEETPH;
386 	}
387 
388 	top_iph->saddr = *(struct in6_addr *)&x->props.saddr;
389 	top_iph->daddr = *(struct in6_addr *)&x->id.daddr;
390 	return 0;
391 }
392 #endif
393 
394 /* Add encapsulation header.
395  *
396  * On exit, the transport header will be set to the start of the
397  * encapsulation header to be filled in by x->type->output and the mac
398  * header will be set to the nextheader (protocol for IPv4) field of the
399  * extension header directly preceding the encapsulation header, or in
400  * its absence, that of the top IP header.
401  * The value of the network header will always point to the top IP header
402  * while skb->data will point to the payload.
403  */
404 static int xfrm4_prepare_output(struct xfrm_state *x, struct sk_buff *skb)
405 {
406 	int err;
407 
408 	err = xfrm_inner_extract_output(x, skb);
409 	if (err)
410 		return err;
411 
412 	IPCB(skb)->flags |= IPSKB_XFRM_TUNNEL_SIZE;
413 	skb->protocol = htons(ETH_P_IP);
414 
415 	switch (x->outer_mode.encap) {
416 	case XFRM_MODE_BEET:
417 		return xfrm4_beet_encap_add(x, skb);
418 	case XFRM_MODE_TUNNEL:
419 		return xfrm4_tunnel_encap_add(x, skb);
420 	}
421 
422 	WARN_ON_ONCE(1);
423 	return -EOPNOTSUPP;
424 }
425 
426 static int xfrm6_prepare_output(struct xfrm_state *x, struct sk_buff *skb)
427 {
428 #if IS_ENABLED(CONFIG_IPV6)
429 	int err;
430 
431 	err = xfrm_inner_extract_output(x, skb);
432 	if (err)
433 		return err;
434 
435 	skb->ignore_df = 1;
436 	skb->protocol = htons(ETH_P_IPV6);
437 
438 	switch (x->outer_mode.encap) {
439 	case XFRM_MODE_BEET:
440 		return xfrm6_beet_encap_add(x, skb);
441 	case XFRM_MODE_TUNNEL:
442 		return xfrm6_tunnel_encap_add(x, skb);
443 	default:
444 		WARN_ON_ONCE(1);
445 		return -EOPNOTSUPP;
446 	}
447 #endif
448 	WARN_ON_ONCE(1);
449 	return -EAFNOSUPPORT;
450 }
451 
452 static int xfrm_outer_mode_output(struct xfrm_state *x, struct sk_buff *skb)
453 {
454 	switch (x->outer_mode.encap) {
455 	case XFRM_MODE_BEET:
456 	case XFRM_MODE_TUNNEL:
457 		if (x->outer_mode.family == AF_INET)
458 			return xfrm4_prepare_output(x, skb);
459 		if (x->outer_mode.family == AF_INET6)
460 			return xfrm6_prepare_output(x, skb);
461 		break;
462 	case XFRM_MODE_TRANSPORT:
463 		if (x->outer_mode.family == AF_INET)
464 			return xfrm4_transport_output(x, skb);
465 		if (x->outer_mode.family == AF_INET6)
466 			return xfrm6_transport_output(x, skb);
467 		break;
468 	case XFRM_MODE_ROUTEOPTIMIZATION:
469 		if (x->outer_mode.family == AF_INET6)
470 			return xfrm6_ro_output(x, skb);
471 		WARN_ON_ONCE(1);
472 		break;
473 	default:
474 		WARN_ON_ONCE(1);
475 		break;
476 	}
477 
478 	return -EOPNOTSUPP;
479 }
480 
481 #if IS_ENABLED(CONFIG_NET_PKTGEN)
482 int pktgen_xfrm_outer_mode_output(struct xfrm_state *x, struct sk_buff *skb)
483 {
484 	return xfrm_outer_mode_output(x, skb);
485 }
486 EXPORT_SYMBOL_GPL(pktgen_xfrm_outer_mode_output);
487 #endif
488 
489 static int xfrm_output_one(struct sk_buff *skb, int err)
490 {
491 	struct dst_entry *dst = skb_dst(skb);
492 	struct xfrm_state *x = dst->xfrm;
493 	struct net *net = xs_net(x);
494 
495 	if (err <= 0 || x->xso.type == XFRM_DEV_OFFLOAD_PACKET)
496 		goto resume;
497 
498 	do {
499 		err = xfrm_skb_check_space(skb);
500 		if (err) {
501 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
502 			goto error_nolock;
503 		}
504 
505 		skb->mark = xfrm_smark_get(skb->mark, x);
506 
507 		err = xfrm_outer_mode_output(x, skb);
508 		if (err) {
509 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEMODEERROR);
510 			goto error_nolock;
511 		}
512 
513 		spin_lock_bh(&x->lock);
514 
515 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
516 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEINVALID);
517 			err = -EINVAL;
518 			goto error;
519 		}
520 
521 		err = xfrm_state_check_expire(x);
522 		if (err) {
523 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEEXPIRED);
524 			goto error;
525 		}
526 
527 		err = xfrm_replay_overflow(x, skb);
528 		if (err) {
529 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATESEQERROR);
530 			goto error;
531 		}
532 
533 		x->curlft.bytes += skb->len;
534 		x->curlft.packets++;
535 		x->lastused = ktime_get_real_seconds();
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 
557 resume:
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 
575 error:
576 	spin_unlock_bh(&x->lock);
577 error_nolock:
578 	kfree_skb(skb);
579 out:
580 	return err;
581 }
582 
583 int 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 
607 out:
608 	return err;
609 }
610 EXPORT_SYMBOL_GPL(xfrm_output_resume);
611 
612 static int xfrm_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
613 {
614 	return xfrm_output_resume(sk, skb, 1);
615 }
616 
617 static 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  * For IPsec tunnel mode save the ip protocol from the IP header of the
651  * plain text packet. Otherwise If the encap type is IPIP, just save
652  * skb->inner_ipproto in any other case get the ip protocol from the IP
653  * header.
654  */
655 static void xfrm_get_inner_ipproto(struct sk_buff *skb, struct xfrm_state *x)
656 {
657 	struct xfrm_offload *xo = xfrm_offload(skb);
658 	const struct ethhdr *eth;
659 
660 	if (!xo)
661 		return;
662 
663 	if (x->outer_mode.encap == XFRM_MODE_TUNNEL) {
664 		switch (x->outer_mode.family) {
665 		case AF_INET:
666 			xo->inner_ipproto = ip_hdr(skb)->protocol;
667 			break;
668 		case AF_INET6:
669 			xo->inner_ipproto = ipv6_hdr(skb)->nexthdr;
670 			break;
671 		default:
672 			break;
673 		}
674 
675 		return;
676 	}
677 
678 	/* non-Tunnel Mode */
679 	if (!skb->encapsulation)
680 		return;
681 
682 	if (skb->inner_protocol_type == ENCAP_TYPE_IPPROTO) {
683 		xo->inner_ipproto = skb->inner_ipproto;
684 		return;
685 	}
686 
687 	if (skb->inner_protocol_type != ENCAP_TYPE_ETHER)
688 		return;
689 
690 	eth = (struct ethhdr *)skb_inner_mac_header(skb);
691 
692 	switch (ntohs(eth->h_proto)) {
693 	case ETH_P_IPV6:
694 		xo->inner_ipproto = inner_ipv6_hdr(skb)->nexthdr;
695 		break;
696 	case ETH_P_IP:
697 		xo->inner_ipproto = inner_ip_hdr(skb)->protocol;
698 		break;
699 	}
700 }
701 
702 int xfrm_output(struct sock *sk, struct sk_buff *skb)
703 {
704 	struct net *net = dev_net(skb_dst(skb)->dev);
705 	struct xfrm_state *x = skb_dst(skb)->xfrm;
706 	int err;
707 
708 	switch (x->outer_mode.family) {
709 	case AF_INET:
710 		memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
711 		IPCB(skb)->flags |= IPSKB_XFRM_TRANSFORMED;
712 		break;
713 	case AF_INET6:
714 		memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
715 
716 		IP6CB(skb)->flags |= IP6SKB_XFRM_TRANSFORMED;
717 		break;
718 	}
719 
720 	if (x->xso.type == XFRM_DEV_OFFLOAD_PACKET) {
721 		if (!xfrm_dev_offload_ok(skb, x)) {
722 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
723 			kfree_skb(skb);
724 			return -EHOSTUNREACH;
725 		}
726 
727 		return xfrm_output_resume(sk, skb, 0);
728 	}
729 
730 	secpath_reset(skb);
731 
732 	if (xfrm_dev_offload_ok(skb, x)) {
733 		struct sec_path *sp;
734 
735 		sp = secpath_set(skb);
736 		if (!sp) {
737 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
738 			kfree_skb(skb);
739 			return -ENOMEM;
740 		}
741 
742 		sp->olen++;
743 		sp->xvec[sp->len++] = x;
744 		xfrm_state_hold(x);
745 
746 		xfrm_get_inner_ipproto(skb, x);
747 		skb->encapsulation = 1;
748 
749 		if (skb_is_gso(skb)) {
750 			if (skb->inner_protocol)
751 				return xfrm_output_gso(net, sk, skb);
752 
753 			skb_shinfo(skb)->gso_type |= SKB_GSO_ESP;
754 			goto out;
755 		}
756 
757 		if (x->xso.dev && x->xso.dev->features & NETIF_F_HW_ESP_TX_CSUM)
758 			goto out;
759 	} else {
760 		if (skb_is_gso(skb))
761 			return xfrm_output_gso(net, sk, skb);
762 	}
763 
764 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
765 		err = skb_checksum_help(skb);
766 		if (err) {
767 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
768 			kfree_skb(skb);
769 			return err;
770 		}
771 	}
772 
773 out:
774 	return xfrm_output2(net, sk, skb);
775 }
776 EXPORT_SYMBOL_GPL(xfrm_output);
777 
778 static int xfrm4_tunnel_check_size(struct sk_buff *skb)
779 {
780 	int mtu, ret = 0;
781 
782 	if (IPCB(skb)->flags & IPSKB_XFRM_TUNNEL_SIZE)
783 		goto out;
784 
785 	if (!(ip_hdr(skb)->frag_off & htons(IP_DF)) || skb->ignore_df)
786 		goto out;
787 
788 	mtu = dst_mtu(skb_dst(skb));
789 	if ((!skb_is_gso(skb) && skb->len > mtu) ||
790 	    (skb_is_gso(skb) &&
791 	     !skb_gso_validate_network_len(skb, ip_skb_dst_mtu(skb->sk, skb)))) {
792 		skb->protocol = htons(ETH_P_IP);
793 
794 		if (skb->sk)
795 			xfrm_local_error(skb, mtu);
796 		else
797 			icmp_send(skb, ICMP_DEST_UNREACH,
798 				  ICMP_FRAG_NEEDED, htonl(mtu));
799 		ret = -EMSGSIZE;
800 	}
801 out:
802 	return ret;
803 }
804 
805 static int xfrm4_extract_output(struct xfrm_state *x, struct sk_buff *skb)
806 {
807 	int err;
808 
809 	if (x->outer_mode.encap == XFRM_MODE_BEET &&
810 	    ip_is_fragment(ip_hdr(skb))) {
811 		net_warn_ratelimited("BEET mode doesn't support inner IPv4 fragments\n");
812 		return -EAFNOSUPPORT;
813 	}
814 
815 	err = xfrm4_tunnel_check_size(skb);
816 	if (err)
817 		return err;
818 
819 	XFRM_MODE_SKB_CB(skb)->protocol = ip_hdr(skb)->protocol;
820 
821 	xfrm4_extract_header(skb);
822 	return 0;
823 }
824 
825 #if IS_ENABLED(CONFIG_IPV6)
826 static int xfrm6_tunnel_check_size(struct sk_buff *skb)
827 {
828 	int mtu, ret = 0;
829 	struct dst_entry *dst = skb_dst(skb);
830 
831 	if (skb->ignore_df)
832 		goto out;
833 
834 	mtu = dst_mtu(dst);
835 	if (mtu < IPV6_MIN_MTU)
836 		mtu = IPV6_MIN_MTU;
837 
838 	if ((!skb_is_gso(skb) && skb->len > mtu) ||
839 	    (skb_is_gso(skb) &&
840 	     !skb_gso_validate_network_len(skb, ip6_skb_dst_mtu(skb)))) {
841 		skb->dev = dst->dev;
842 		skb->protocol = htons(ETH_P_IPV6);
843 
844 		if (xfrm6_local_dontfrag(skb->sk))
845 			ipv6_stub->xfrm6_local_rxpmtu(skb, mtu);
846 		else if (skb->sk)
847 			xfrm_local_error(skb, mtu);
848 		else
849 			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
850 		ret = -EMSGSIZE;
851 	}
852 out:
853 	return ret;
854 }
855 #endif
856 
857 static int xfrm6_extract_output(struct xfrm_state *x, struct sk_buff *skb)
858 {
859 #if IS_ENABLED(CONFIG_IPV6)
860 	int err;
861 
862 	err = xfrm6_tunnel_check_size(skb);
863 	if (err)
864 		return err;
865 
866 	XFRM_MODE_SKB_CB(skb)->protocol = ipv6_hdr(skb)->nexthdr;
867 
868 	xfrm6_extract_header(skb);
869 	return 0;
870 #else
871 	WARN_ON_ONCE(1);
872 	return -EAFNOSUPPORT;
873 #endif
874 }
875 
876 static int xfrm_inner_extract_output(struct xfrm_state *x, struct sk_buff *skb)
877 {
878 	const struct xfrm_mode *inner_mode;
879 
880 	if (x->sel.family == AF_UNSPEC)
881 		inner_mode = xfrm_ip2inner_mode(x,
882 				xfrm_af2proto(skb_dst(skb)->ops->family));
883 	else
884 		inner_mode = &x->inner_mode;
885 
886 	if (inner_mode == NULL)
887 		return -EAFNOSUPPORT;
888 
889 	switch (inner_mode->family) {
890 	case AF_INET:
891 		return xfrm4_extract_output(x, skb);
892 	case AF_INET6:
893 		return xfrm6_extract_output(x, skb);
894 	}
895 
896 	return -EAFNOSUPPORT;
897 }
898 
899 void xfrm_local_error(struct sk_buff *skb, int mtu)
900 {
901 	unsigned int proto;
902 	struct xfrm_state_afinfo *afinfo;
903 
904 	if (skb->protocol == htons(ETH_P_IP))
905 		proto = AF_INET;
906 	else if (skb->protocol == htons(ETH_P_IPV6) &&
907 		 skb->sk->sk_family == AF_INET6)
908 		proto = AF_INET6;
909 	else
910 		return;
911 
912 	afinfo = xfrm_state_get_afinfo(proto);
913 	if (afinfo) {
914 		afinfo->local_error(skb, mtu);
915 		rcu_read_unlock();
916 	}
917 }
918 EXPORT_SYMBOL_GPL(xfrm_local_error);
919