xfrm_output.c

  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
 28static int xfrm_output2(struct net *net, struct sock *sk, struct sk_buff *skb);
 29static int xfrm_inner_extract_output(struct xfrm_state *x, struct sk_buff *skb);
 30
 31static 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
 52static 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 */
 65static 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)
 82static 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)
142static 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 */
163static 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 */
194static 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 */
224static 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 */
273static 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)
319static int xfrm6_tunnel_encap_add(struct xfrm_state *x, struct sk_buff *skb)
320{
321	struct dst_entry *dst = skb_dst(skb);
322	struct ipv6hdr *top_iph;
323	int dsfield;
324
325	skb_set_inner_network_header(skb, skb_network_offset(skb));
326	skb_set_inner_transport_header(skb, skb_transport_offset(skb));
327
328	skb_set_network_header(skb, -x->props.header_len);
329	skb->mac_header = skb->network_header +
330			  offsetof(struct ipv6hdr, nexthdr);
331	skb->transport_header = skb->network_header + sizeof(*top_iph);
332	top_iph = ipv6_hdr(skb);
333
334	top_iph->version = 6;
335
336	memcpy(top_iph->flow_lbl, XFRM_MODE_SKB_CB(skb)->flow_lbl,
337	       sizeof(top_iph->flow_lbl));
338	top_iph->nexthdr = xfrm_af2proto(skb_dst(skb)->ops->family);
339
340	if (x->props.extra_flags & XFRM_SA_XFLAG_DONT_ENCAP_DSCP)
341		dsfield = 0;
342	else
343		dsfield = XFRM_MODE_SKB_CB(skb)->tos;
344	dsfield = INET_ECN_encapsulate(dsfield, XFRM_MODE_SKB_CB(skb)->tos);
345	if (x->props.flags & XFRM_STATE_NOECN)
346		dsfield &= ~INET_ECN_MASK;
347	ipv6_change_dsfield(top_iph, 0, dsfield);
348	top_iph->hop_limit = ip6_dst_hoplimit(xfrm_dst_child(dst));
349	top_iph->saddr = *(struct in6_addr *)&x->props.saddr;
350	top_iph->daddr = *(struct in6_addr *)&x->id.daddr;
351	return 0;
352}
353
354static int xfrm6_beet_encap_add(struct xfrm_state *x, struct sk_buff *skb)
355{
356	struct ipv6hdr *top_iph;
357	struct ip_beet_phdr *ph;
358	int optlen, hdr_len;
359
360	hdr_len = 0;
361	optlen = XFRM_MODE_SKB_CB(skb)->optlen;
362	if (unlikely(optlen))
363		hdr_len += IPV4_BEET_PHMAXLEN - (optlen & 4);
364
365	skb_set_network_header(skb, -x->props.header_len - hdr_len);
366	if (x->sel.family != AF_INET6)
367		skb->network_header += IPV4_BEET_PHMAXLEN;
368	skb->mac_header = skb->network_header +
369			  offsetof(struct ipv6hdr, nexthdr);
370	skb->transport_header = skb->network_header + sizeof(*top_iph);
371	ph = __skb_pull(skb, XFRM_MODE_SKB_CB(skb)->ihl - hdr_len);
372
373	xfrm6_beet_make_header(skb);
374
375	top_iph = ipv6_hdr(skb);
376	if (unlikely(optlen)) {
377		if (WARN_ON(optlen < 0))
378			return -EINVAL;
379
380		ph->padlen = 4 - (optlen & 4);
381		ph->hdrlen = optlen / 8;
382		ph->nexthdr = top_iph->nexthdr;
383		if (ph->padlen)
384			memset(ph + 1, IPOPT_NOP, ph->padlen);
385
386		top_iph->nexthdr = IPPROTO_BEETPH;
387	}
388
389	top_iph->saddr = *(struct in6_addr *)&x->props.saddr;
390	top_iph->daddr = *(struct in6_addr *)&x->id.daddr;
391	return 0;
392}
393#endif
394
395/* Add encapsulation header.
396 *
397 * On exit, the transport header will be set to the start of the
398 * encapsulation header to be filled in by x->type->output and the mac
399 * header will be set to the nextheader (protocol for IPv4) field of the
400 * extension header directly preceding the encapsulation header, or in
401 * its absence, that of the top IP header.
402 * The value of the network header will always point to the top IP header
403 * while skb->data will point to the payload.
404 */
405static int xfrm4_prepare_output(struct xfrm_state *x, struct sk_buff *skb)
406{
407	int err;
408
409	err = xfrm_inner_extract_output(x, skb);
410	if (err)
411		return err;
412
413	IPCB(skb)->flags |= IPSKB_XFRM_TUNNEL_SIZE;
414	skb->protocol = htons(ETH_P_IP);
415
416	switch (x->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
427static int xfrm6_prepare_output(struct xfrm_state *x, struct sk_buff *skb)
428{
429#if IS_ENABLED(CONFIG_IPV6)
430	int err;
431
432	err = xfrm_inner_extract_output(x, skb);
433	if (err)
434		return err;
435
436	skb->ignore_df = 1;
437	skb->protocol = htons(ETH_P_IPV6);
438
439	switch (x->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
453static 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		if (x->mode_cbs && x->mode_cbs->prepare_output)
476			return x->mode_cbs->prepare_output(x, skb);
477		WARN_ON_ONCE(1);
478		break;
479	}
480
481	return -EOPNOTSUPP;
482}
483
484#if IS_ENABLED(CONFIG_NET_PKTGEN)
485int pktgen_xfrm_outer_mode_output(struct xfrm_state *x, struct sk_buff *skb)
486{
487	return xfrm_outer_mode_output(x, skb);
488}
489EXPORT_SYMBOL_GPL(pktgen_xfrm_outer_mode_output);
490#endif
491
492static int xfrm_output_one(struct sk_buff *skb, int err)
493{
494	struct dst_entry *dst = skb_dst(skb);
495	struct xfrm_state *x = dst->xfrm;
496	struct net *net = xs_net(x);
497
498	if (err <= 0 || x->xso.type == XFRM_DEV_OFFLOAD_PACKET)
499		goto resume;
500
501	do {
502		err = xfrm_skb_check_space(skb);
503		if (err) {
504			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
505			goto error_nolock;
506		}
507
508		skb->mark = xfrm_smark_get(skb->mark, x);
509
510		err = xfrm_outer_mode_output(x, skb);
511		if (err) {
512			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEMODEERROR);
513			goto error_nolock;
514		}
515
516		spin_lock_bh(&x->lock);
517
518		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
519			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEINVALID);
520			err = -EINVAL;
521			goto error;
522		}
523
524		err = xfrm_state_check_expire(x);
525		if (err) {
526			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEEXPIRED);
527			goto error;
528		}
529
530		err = xfrm_replay_overflow(x, skb);
531		if (err) {
532			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATESEQERROR);
533			goto error;
534		}
535
536		x->curlft.bytes += skb->len;
537		x->curlft.packets++;
538		x->lastused = ktime_get_real_seconds();
539
540		spin_unlock_bh(&x->lock);
541
542		skb_dst_force(skb);
543		if (!skb_dst(skb)) {
544			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
545			err = -EHOSTUNREACH;
546			goto error_nolock;
547		}
548
549		if (xfrm_offload(skb)) {
550			x->type_offload->encap(x, skb);
551		} else {
552			/* Inner headers are invalid now. */
553			skb->encapsulation = 0;
554
555			err = x->type->output(x, skb);
556			if (err == -EINPROGRESS)
557				goto out;
558		}
559
560resume:
561		if (err) {
562			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR);
563			goto error_nolock;
564		}
565
566		dst = skb_dst_pop(skb);
567		if (!dst) {
568			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
569			err = -EHOSTUNREACH;
570			goto error_nolock;
571		}
572		skb_dst_set(skb, dst);
573		x = dst->xfrm;
574	} while (x && !(x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL));
575
576	return 0;
577
578error:
579	spin_unlock_bh(&x->lock);
580error_nolock:
581	kfree_skb(skb);
582out:
583	return err;
584}
585
586int xfrm_output_resume(struct sock *sk, struct sk_buff *skb, int err)
587{
588	struct net *net = xs_net(skb_dst(skb)->xfrm);
589
590	while (likely((err = xfrm_output_one(skb, err)) == 0)) {
591		nf_reset_ct(skb);
592
593		err = skb_dst(skb)->ops->local_out(net, sk, skb);
594		if (unlikely(err != 1))
595			goto out;
596
597		if (!skb_dst(skb)->xfrm)
598			return dst_output(net, sk, skb);
599
600		err = nf_hook(skb_dst(skb)->ops->family,
601			      NF_INET_POST_ROUTING, net, sk, skb,
602			      NULL, skb_dst(skb)->dev, xfrm_output2);
603		if (unlikely(err != 1))
604			goto out;
605	}
606
607	if (err == -EINPROGRESS)
608		err = 0;
609
610out:
611	return err;
612}
613EXPORT_SYMBOL_GPL(xfrm_output_resume);
614
615static int xfrm_dev_direct_output(struct sock *sk, struct xfrm_state *x,
616				  struct sk_buff *skb)
617{
618	struct dst_entry *dst = skb_dst(skb);
619	struct net *net = xs_net(x);
620	int err;
621
622	dst = skb_dst_pop(skb);
623	if (!dst) {
624		XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
625		kfree_skb(skb);
626		return -EHOSTUNREACH;
627	}
628	skb_dst_set(skb, dst);
629	nf_reset_ct(skb);
630
631	err = skb_dst(skb)->ops->local_out(net, sk, skb);
632	if (unlikely(err != 1)) {
633		kfree_skb(skb);
634		return err;
635	}
636
637	/* In transport mode, network destination is
638	 * directly reachable, while in tunnel mode,
639	 * inner packet network may not be. In packet
640	 * offload type, HW is responsible for hard
641	 * header packet mangling so directly xmit skb
642	 * to netdevice.
643	 */
644	skb->dev = x->xso.dev;
645	__skb_push(skb, skb->dev->hard_header_len);
646	return dev_queue_xmit(skb);
647}
648
649static int xfrm_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
650{
651	return xfrm_output_resume(sk, skb, 1);
652}
653
654static int xfrm_output_gso(struct net *net, struct sock *sk, struct sk_buff *skb)
655{
656	struct sk_buff *segs, *nskb;
657
658	BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET);
659	BUILD_BUG_ON(sizeof(*IP6CB(skb)) > SKB_GSO_CB_OFFSET);
660	segs = skb_gso_segment(skb, 0);
661	kfree_skb(skb);
662	if (IS_ERR(segs))
663		return PTR_ERR(segs);
664	if (segs == NULL)
665		return -EINVAL;
666
667	skb_list_walk_safe(segs, segs, nskb) {
668		int err;
669
670		skb_mark_not_on_list(segs);
671		err = xfrm_output2(net, sk, segs);
672
673		if (unlikely(err)) {
674			kfree_skb_list(nskb);
675			return err;
676		}
677	}
678
679	return 0;
680}
681
682/* For partial checksum offload, the outer header checksum is calculated
683 * by software and the inner header checksum is calculated by hardware.
684 * This requires hardware to know the inner packet type to calculate
685 * the inner header checksum. Save inner ip protocol here to avoid
686 * traversing the packet in the vendor's xmit code.
687 * For IPsec tunnel mode save the ip protocol from the IP header of the
688 * plain text packet. Otherwise If the encap type is IPIP, just save
689 * skb->inner_ipproto in any other case get the ip protocol from the IP
690 * header.
691 */
692static void xfrm_get_inner_ipproto(struct sk_buff *skb, struct xfrm_state *x)
693{
694	struct xfrm_offload *xo = xfrm_offload(skb);
695	const struct ethhdr *eth;
696
697	if (!xo)
698		return;
699
700	if (x->outer_mode.encap == XFRM_MODE_TUNNEL) {
701		switch (skb_dst(skb)->ops->family) {
702		case AF_INET:
703			xo->inner_ipproto = ip_hdr(skb)->protocol;
704			break;
705		case AF_INET6:
706			xo->inner_ipproto = ipv6_hdr(skb)->nexthdr;
707			break;
708		default:
709			break;
710		}
711
712		return;
713	}
714	if (x->outer_mode.encap == XFRM_MODE_IPTFS) {
715		xo->inner_ipproto = IPPROTO_AGGFRAG;
716		return;
717	}
718
719	/* non-Tunnel Mode */
720	if (!skb->encapsulation)
721		return;
722
723	if (skb->inner_protocol_type == ENCAP_TYPE_IPPROTO) {
724		xo->inner_ipproto = skb->inner_ipproto;
725		return;
726	}
727
728	if (skb->inner_protocol_type != ENCAP_TYPE_ETHER)
729		return;
730
731	eth = (struct ethhdr *)skb_inner_mac_header(skb);
732
733	switch (ntohs(eth->h_proto)) {
734	case ETH_P_IPV6:
735		xo->inner_ipproto = inner_ipv6_hdr(skb)->nexthdr;
736		break;
737	case ETH_P_IP:
738		xo->inner_ipproto = inner_ip_hdr(skb)->protocol;
739		break;
740	}
741}
742
743int xfrm_output(struct sock *sk, struct sk_buff *skb)
744{
745	struct net *net = dev_net(skb_dst(skb)->dev);
746	struct xfrm_state *x = skb_dst(skb)->xfrm;
747	int family;
748	int err;
749
750	family = (x->xso.type != XFRM_DEV_OFFLOAD_PACKET) ? x->outer_mode.family
751		: skb_dst(skb)->ops->family;
752
753	switch (family) {
754	case AF_INET:
755		memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
756		IPCB(skb)->flags |= IPSKB_XFRM_TRANSFORMED;
757		break;
758	case AF_INET6:
759		memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
760
761		IP6CB(skb)->flags |= IP6SKB_XFRM_TRANSFORMED;
762		break;
763	}
764
765	if (x->xso.type == XFRM_DEV_OFFLOAD_PACKET) {
766		if (!xfrm_dev_offload_ok(skb, x)) {
767			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
768			kfree_skb(skb);
769			return -EHOSTUNREACH;
770		}
771
772		/* Exclusive direct xmit for tunnel mode, as
773		 * some filtering or matching rules may apply
774		 * in transport mode.
775		 * Locally generated packets also require
776		 * the normal XFRM path for L2 header setup,
777		 * as the hardware needs the L2 header to match
778		 * for encryption, so skip direct output as well.
779		 */
780		if (x->props.mode == XFRM_MODE_TUNNEL && !skb->sk)
781			return xfrm_dev_direct_output(sk, x, skb);
782
783		return xfrm_output_resume(sk, skb, 0);
784	}
785
786	secpath_reset(skb);
787
788	if (xfrm_dev_offload_ok(skb, x)) {
789		struct sec_path *sp;
790
791		sp = secpath_set(skb);
792		if (!sp) {
793			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
794			kfree_skb(skb);
795			return -ENOMEM;
796		}
797
798		sp->olen++;
799		sp->xvec[sp->len++] = x;
800		xfrm_state_hold(x);
801
802		xfrm_get_inner_ipproto(skb, x);
803		skb->encapsulation = 1;
804
805		if (skb_is_gso(skb)) {
806			if (skb->inner_protocol && x->props.mode == XFRM_MODE_TUNNEL)
807				return xfrm_output_gso(net, sk, skb);
808
809			skb_shinfo(skb)->gso_type |= SKB_GSO_ESP;
810			goto out;
811		}
812
813		if (x->xso.dev && x->xso.dev->features & NETIF_F_HW_ESP_TX_CSUM)
814			goto out;
815	} else {
816		if (skb_is_gso(skb))
817			return xfrm_output_gso(net, sk, skb);
818	}
819
820	if (skb->ip_summed == CHECKSUM_PARTIAL) {
821		err = skb_checksum_help(skb);
822		if (err) {
823			XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
824			kfree_skb(skb);
825			return err;
826		}
827	}
828
829out:
830	return xfrm_output2(net, sk, skb);
831}
832EXPORT_SYMBOL_GPL(xfrm_output);
833
834int xfrm4_tunnel_check_size(struct sk_buff *skb)
835{
836	int mtu, ret = 0;
837
838	if (IPCB(skb)->flags & IPSKB_XFRM_TUNNEL_SIZE)
839		goto out;
840
841	if (!(ip_hdr(skb)->frag_off & htons(IP_DF)) || skb->ignore_df)
842		goto out;
843
844	mtu = dst_mtu(skb_dst(skb));
845	if ((!skb_is_gso(skb) && skb->len > mtu) ||
846	    (skb_is_gso(skb) &&
847	     !skb_gso_validate_network_len(skb, ip_skb_dst_mtu(skb->sk, skb)))) {
848		skb->protocol = htons(ETH_P_IP);
849
850		if (skb->sk && sk_fullsock(skb->sk))
851			xfrm_local_error(skb, mtu);
852		else
853			icmp_send(skb, ICMP_DEST_UNREACH,
854				  ICMP_FRAG_NEEDED, htonl(mtu));
855		ret = -EMSGSIZE;
856	}
857out:
858	return ret;
859}
860EXPORT_SYMBOL_GPL(xfrm4_tunnel_check_size);
861
862static int xfrm4_extract_output(struct xfrm_state *x, struct sk_buff *skb)
863{
864	int err;
865
866	if (x->outer_mode.encap == XFRM_MODE_BEET &&
867	    ip_is_fragment(ip_hdr(skb))) {
868		net_warn_ratelimited("BEET mode doesn't support inner IPv4 fragments\n");
869		return -EAFNOSUPPORT;
870	}
871
872	err = xfrm4_tunnel_check_size(skb);
873	if (err)
874		return err;
875
876	XFRM_MODE_SKB_CB(skb)->protocol = ip_hdr(skb)->protocol;
877
878	xfrm4_extract_header(skb);
879	return 0;
880}
881
882#if IS_ENABLED(CONFIG_IPV6)
883int xfrm6_tunnel_check_size(struct sk_buff *skb)
884{
885	int mtu, ret = 0;
886	struct dst_entry *dst = skb_dst(skb);
887	struct sock *sk = skb_to_full_sk(skb);
888
889	if (skb->ignore_df)
890		goto out;
891
892	mtu = dst_mtu(dst);
893	if (mtu < IPV6_MIN_MTU)
894		mtu = IPV6_MIN_MTU;
895
896	if ((!skb_is_gso(skb) && skb->len > mtu) ||
897	    (skb_is_gso(skb) &&
898	     !skb_gso_validate_network_len(skb, ip6_skb_dst_mtu(skb)))) {
899		skb->dev = dst->dev;
900		skb->protocol = htons(ETH_P_IPV6);
901
902		if (xfrm6_local_dontfrag(sk))
903			ipv6_stub->xfrm6_local_rxpmtu(skb, mtu);
904		else if (sk)
905			xfrm_local_error(skb, mtu);
906		else
907			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
908		ret = -EMSGSIZE;
909	}
910out:
911	return ret;
912}
913EXPORT_SYMBOL_GPL(xfrm6_tunnel_check_size);
914#endif
915
916static int xfrm6_extract_output(struct xfrm_state *x, struct sk_buff *skb)
917{
918#if IS_ENABLED(CONFIG_IPV6)
919	int err;
920
921	err = xfrm6_tunnel_check_size(skb);
922	if (err)
923		return err;
924
925	XFRM_MODE_SKB_CB(skb)->protocol = ipv6_hdr(skb)->nexthdr;
926
927	xfrm6_extract_header(skb);
928	return 0;
929#else
930	WARN_ON_ONCE(1);
931	return -EAFNOSUPPORT;
932#endif
933}
934
935static int xfrm_inner_extract_output(struct xfrm_state *x, struct sk_buff *skb)
936{
937	switch (skb->protocol) {
938	case htons(ETH_P_IP):
939		return xfrm4_extract_output(x, skb);
940	case htons(ETH_P_IPV6):
941		return xfrm6_extract_output(x, skb);
942	}
943
944	return -EAFNOSUPPORT;
945}
946
947void xfrm_local_error(struct sk_buff *skb, int mtu)
948{
949	unsigned int proto;
950	struct xfrm_state_afinfo *afinfo;
951
952	if (skb->protocol == htons(ETH_P_IP))
953		proto = AF_INET;
954	else if (skb->protocol == htons(ETH_P_IPV6) &&
955		 skb->sk->sk_family == AF_INET6)
956		proto = AF_INET6;
957	else
958		return;
959
960	afinfo = xfrm_state_get_afinfo(proto);
961	if (afinfo) {
962		afinfo->local_error(skb, mtu);
963		rcu_read_unlock();
964	}
965}
966EXPORT_SYMBOL_GPL(xfrm_local_error);