xref: /openbmc/linux/net/xfrm/xfrm_output.c (revision 9aa2cba7) 
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 
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 
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  */
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)
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)
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  */
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  */
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  */
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  */
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)
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 
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  */
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 
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 
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)
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 
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 
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 
613 static int xfrm_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
614 {
615 	return xfrm_output_resume(sk, skb, 1);
616 }
617 
618 static int xfrm_output_gso(struct net *net, struct sock *sk, struct sk_buff *skb)
619 {
620 	struct sk_buff *segs, *nskb;
621 
622 	BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET);
623 	BUILD_BUG_ON(sizeof(*IP6CB(skb)) > SKB_GSO_CB_OFFSET);
624 	segs = skb_gso_segment(skb, 0);
625 	kfree_skb(skb);
626 	if (IS_ERR(segs))
627 		return PTR_ERR(segs);
628 	if (segs == NULL)
629 		return -EINVAL;
630 
631 	skb_list_walk_safe(segs, segs, nskb) {
632 		int err;
633 
634 		skb_mark_not_on_list(segs);
635 		err = xfrm_output2(net, sk, segs);
636 
637 		if (unlikely(err)) {
638 			kfree_skb_list(nskb);
639 			return err;
640 		}
641 	}
642 
643 	return 0;
644 }
645 
646 /* For partial checksum offload, the outer header checksum is calculated
647  * by software and the inner header checksum is calculated by hardware.
648  * This requires hardware to know the inner packet type to calculate
649  * the inner header checksum. Save inner ip protocol here to avoid
650  * traversing the packet in the vendor's xmit code.
651  * For IPsec tunnel mode save the ip protocol from the IP header of the
652  * plain text packet. Otherwise If the encap type is IPIP, just save
653  * skb->inner_ipproto in any other case get the ip protocol from the IP
654  * header.
655  */
656 static void xfrm_get_inner_ipproto(struct sk_buff *skb, struct xfrm_state *x)
657 {
658 	struct xfrm_offload *xo = xfrm_offload(skb);
659 	const struct ethhdr *eth;
660 
661 	if (!xo)
662 		return;
663 
664 	if (x->outer_mode.encap == XFRM_MODE_TUNNEL) {
665 		switch (x->outer_mode.family) {
666 		case AF_INET:
667 			xo->inner_ipproto = ip_hdr(skb)->protocol;
668 			break;
669 		case AF_INET6:
670 			xo->inner_ipproto = ipv6_hdr(skb)->nexthdr;
671 			break;
672 		default:
673 			break;
674 		}
675 
676 		return;
677 	}
678 
679 	/* non-Tunnel Mode */
680 	if (!skb->encapsulation)
681 		return;
682 
683 	if (skb->inner_protocol_type == ENCAP_TYPE_IPPROTO) {
684 		xo->inner_ipproto = skb->inner_ipproto;
685 		return;
686 	}
687 
688 	if (skb->inner_protocol_type != ENCAP_TYPE_ETHER)
689 		return;
690 
691 	eth = (struct ethhdr *)skb_inner_mac_header(skb);
692 
693 	switch (ntohs(eth->h_proto)) {
694 	case ETH_P_IPV6:
695 		xo->inner_ipproto = inner_ipv6_hdr(skb)->nexthdr;
696 		break;
697 	case ETH_P_IP:
698 		xo->inner_ipproto = inner_ip_hdr(skb)->protocol;
699 		break;
700 	}
701 }
702 
703 int xfrm_output(struct sock *sk, struct sk_buff *skb)
704 {
705 	struct net *net = dev_net(skb_dst(skb)->dev);
706 	struct xfrm_state *x = skb_dst(skb)->xfrm;
707 	int err;
708 
709 	switch (x->outer_mode.family) {
710 	case AF_INET:
711 		memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
712 		IPCB(skb)->flags |= IPSKB_XFRM_TRANSFORMED;
713 		break;
714 	case AF_INET6:
715 		memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
716 
717 		IP6CB(skb)->flags |= IP6SKB_XFRM_TRANSFORMED;
718 		break;
719 	}
720 
721 	if (x->xso.type == XFRM_DEV_OFFLOAD_PACKET) {
722 		if (!xfrm_dev_offload_ok(skb, x)) {
723 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
724 			kfree_skb(skb);
725 			return -EHOSTUNREACH;
726 		}
727 
728 		return xfrm_output_resume(sk, skb, 0);
729 	}
730 
731 	secpath_reset(skb);
732 
733 	if (xfrm_dev_offload_ok(skb, x)) {
734 		struct sec_path *sp;
735 
736 		sp = secpath_set(skb);
737 		if (!sp) {
738 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
739 			kfree_skb(skb);
740 			return -ENOMEM;
741 		}
742 
743 		sp->olen++;
744 		sp->xvec[sp->len++] = x;
745 		xfrm_state_hold(x);
746 
747 		xfrm_get_inner_ipproto(skb, x);
748 		skb->encapsulation = 1;
749 
750 		if (skb_is_gso(skb)) {
751 			if (skb->inner_protocol)
752 				return xfrm_output_gso(net, sk, skb);
753 
754 			skb_shinfo(skb)->gso_type |= SKB_GSO_ESP;
755 			goto out;
756 		}
757 
758 		if (x->xso.dev && x->xso.dev->features & NETIF_F_HW_ESP_TX_CSUM)
759 			goto out;
760 	} else {
761 		if (skb_is_gso(skb))
762 			return xfrm_output_gso(net, sk, skb);
763 	}
764 
765 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
766 		err = skb_checksum_help(skb);
767 		if (err) {
768 			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
769 			kfree_skb(skb);
770 			return err;
771 		}
772 	}
773 
774 out:
775 	return xfrm_output2(net, sk, skb);
776 }
777 EXPORT_SYMBOL_GPL(xfrm_output);
778 
779 static int xfrm4_tunnel_check_size(struct sk_buff *skb)
780 {
781 	int mtu, ret = 0;
782 
783 	if (IPCB(skb)->flags & IPSKB_XFRM_TUNNEL_SIZE)
784 		goto out;
785 
786 	if (!(ip_hdr(skb)->frag_off & htons(IP_DF)) || skb->ignore_df)
787 		goto out;
788 
789 	mtu = dst_mtu(skb_dst(skb));
790 	if ((!skb_is_gso(skb) && skb->len > mtu) ||
791 	    (skb_is_gso(skb) &&
792 	     !skb_gso_validate_network_len(skb, ip_skb_dst_mtu(skb->sk, skb)))) {
793 		skb->protocol = htons(ETH_P_IP);
794 
795 		if (skb->sk)
796 			xfrm_local_error(skb, mtu);
797 		else
798 			icmp_send(skb, ICMP_DEST_UNREACH,
799 				  ICMP_FRAG_NEEDED, htonl(mtu));
800 		ret = -EMSGSIZE;
801 	}
802 out:
803 	return ret;
804 }
805 
806 static int xfrm4_extract_output(struct xfrm_state *x, struct sk_buff *skb)
807 {
808 	int err;
809 
810 	if (x->outer_mode.encap == XFRM_MODE_BEET &&
811 	    ip_is_fragment(ip_hdr(skb))) {
812 		net_warn_ratelimited("BEET mode doesn't support inner IPv4 fragments\n");
813 		return -EAFNOSUPPORT;
814 	}
815 
816 	err = xfrm4_tunnel_check_size(skb);
817 	if (err)
818 		return err;
819 
820 	XFRM_MODE_SKB_CB(skb)->protocol = ip_hdr(skb)->protocol;
821 
822 	xfrm4_extract_header(skb);
823 	return 0;
824 }
825 
826 #if IS_ENABLED(CONFIG_IPV6)
827 static int xfrm6_tunnel_check_size(struct sk_buff *skb)
828 {
829 	int mtu, ret = 0;
830 	struct dst_entry *dst = skb_dst(skb);
831 
832 	if (skb->ignore_df)
833 		goto out;
834 
835 	mtu = dst_mtu(dst);
836 	if (mtu < IPV6_MIN_MTU)
837 		mtu = IPV6_MIN_MTU;
838 
839 	if ((!skb_is_gso(skb) && skb->len > mtu) ||
840 	    (skb_is_gso(skb) &&
841 	     !skb_gso_validate_network_len(skb, ip6_skb_dst_mtu(skb)))) {
842 		skb->dev = dst->dev;
843 		skb->protocol = htons(ETH_P_IPV6);
844 
845 		if (xfrm6_local_dontfrag(skb->sk))
846 			ipv6_stub->xfrm6_local_rxpmtu(skb, mtu);
847 		else if (skb->sk)
848 			xfrm_local_error(skb, mtu);
849 		else
850 			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
851 		ret = -EMSGSIZE;
852 	}
853 out:
854 	return ret;
855 }
856 #endif
857 
858 static int xfrm6_extract_output(struct xfrm_state *x, struct sk_buff *skb)
859 {
860 #if IS_ENABLED(CONFIG_IPV6)
861 	int err;
862 
863 	err = xfrm6_tunnel_check_size(skb);
864 	if (err)
865 		return err;
866 
867 	XFRM_MODE_SKB_CB(skb)->protocol = ipv6_hdr(skb)->nexthdr;
868 
869 	xfrm6_extract_header(skb);
870 	return 0;
871 #else
872 	WARN_ON_ONCE(1);
873 	return -EAFNOSUPPORT;
874 #endif
875 }
876 
877 static int xfrm_inner_extract_output(struct xfrm_state *x, struct sk_buff *skb)
878 {
879 	switch (skb->protocol) {
880 	case htons(ETH_P_IP):
881 		return xfrm4_extract_output(x, skb);
882 	case htons(ETH_P_IPV6):
883 		return xfrm6_extract_output(x, skb);
884 	}
885 
886 	return -EAFNOSUPPORT;
887 }
888 
889 void xfrm_local_error(struct sk_buff *skb, int mtu)
890 {
891 	unsigned int proto;
892 	struct xfrm_state_afinfo *afinfo;
893 
894 	if (skb->protocol == htons(ETH_P_IP))
895 		proto = AF_INET;
896 	else if (skb->protocol == htons(ETH_P_IPV6) &&
897 		 skb->sk->sk_family == AF_INET6)
898 		proto = AF_INET6;
899 	else
900 		return;
901 
902 	afinfo = xfrm_state_get_afinfo(proto);
903 	if (afinfo) {
904 		afinfo->local_error(skb, mtu);
905 		rcu_read_unlock();
906 	}
907 }
908 EXPORT_SYMBOL_GPL(xfrm_local_error);
909