xref: /linux/net/ipv6/esp6.c (revision 26aad08a928901296aabfbc7a33ecb951656bb98)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (C)2002 USAGI/WIDE Project
4  *
5  * Authors
6  *
7  *	Mitsuru KANDA @USAGI       : IPv6 Support
8  *	Kazunori MIYAZAWA @USAGI   :
9  *	Kunihiro Ishiguro <kunihiro@ipinfusion.com>
10  *
11  *	This file is derived from net/ipv4/esp.c
12  */
13 
14 #define pr_fmt(fmt) "IPv6: " fmt
15 
16 #include <crypto/aead.h>
17 #include <crypto/authenc.h>
18 #include <linux/err.h>
19 #include <linux/module.h>
20 #include <net/ip.h>
21 #include <net/xfrm.h>
22 #include <net/esp.h>
23 #include <linux/scatterlist.h>
24 #include <linux/kernel.h>
25 #include <linux/pfkeyv2.h>
26 #include <linux/random.h>
27 #include <linux/slab.h>
28 #include <linux/spinlock.h>
29 #include <net/ip6_checksum.h>
30 #include <net/ip6_route.h>
31 #include <net/icmp.h>
32 #include <net/ipv6.h>
33 #include <net/protocol.h>
34 #include <net/udp.h>
35 #include <linux/icmpv6.h>
36 #include <net/tcp.h>
37 #include <net/espintcp.h>
38 #include <net/inet6_hashtables.h>
39 #include <linux/skbuff_ref.h>
40 
41 #include <linux/highmem.h>
42 
43 struct esp_skb_cb {
44 	struct xfrm_skb_cb xfrm;
45 	void *tmp;
46 };
47 
48 struct esp_output_extra {
49 	__be32 seqhi;
50 	u32 esphoff;
51 };
52 
53 #define ESP_SKB_CB(__skb) ((struct esp_skb_cb *)&((__skb)->cb[0]))
54 
55 /*
56  * Allocate an AEAD request structure with extra space for SG and IV.
57  *
58  * For alignment considerations the upper 32 bits of the sequence number are
59  * placed at the front, if present. Followed by the IV, the request and finally
60  * the SG list.
61  *
62  * TODO: Use spare space in skb for this where possible.
63  */
64 static void *esp_alloc_tmp(struct crypto_aead *aead, int nfrags, int seqihlen)
65 {
66 	unsigned int len;
67 
68 	len = seqihlen;
69 
70 	len += crypto_aead_ivsize(aead);
71 
72 	if (len) {
73 		len += crypto_aead_alignmask(aead) &
74 		       ~(crypto_tfm_ctx_alignment() - 1);
75 		len = ALIGN(len, crypto_tfm_ctx_alignment());
76 	}
77 
78 	len += sizeof(struct aead_request) + crypto_aead_reqsize(aead);
79 	len = ALIGN(len, __alignof__(struct scatterlist));
80 
81 	len += sizeof(struct scatterlist) * nfrags;
82 
83 	return kmalloc(len, GFP_ATOMIC);
84 }
85 
86 static inline void *esp_tmp_extra(void *tmp)
87 {
88 	return PTR_ALIGN(tmp, __alignof__(struct esp_output_extra));
89 }
90 
91 static inline u8 *esp_tmp_iv(struct crypto_aead *aead, void *tmp, int seqhilen)
92 {
93 	return crypto_aead_ivsize(aead) ?
94 	       PTR_ALIGN((u8 *)tmp + seqhilen,
95 			 crypto_aead_alignmask(aead) + 1) : tmp + seqhilen;
96 }
97 
98 static inline struct aead_request *esp_tmp_req(struct crypto_aead *aead, u8 *iv)
99 {
100 	struct aead_request *req;
101 
102 	req = (void *)PTR_ALIGN(iv + crypto_aead_ivsize(aead),
103 				crypto_tfm_ctx_alignment());
104 	aead_request_set_tfm(req, aead);
105 	return req;
106 }
107 
108 static inline struct scatterlist *esp_req_sg(struct crypto_aead *aead,
109 					     struct aead_request *req)
110 {
111 	return (void *)ALIGN((unsigned long)(req + 1) +
112 			     crypto_aead_reqsize(aead),
113 			     __alignof__(struct scatterlist));
114 }
115 
116 static void esp_ssg_unref(struct xfrm_state *x, void *tmp, struct sk_buff *skb, bool already_unref)
117 {
118 	struct crypto_aead *aead = x->data;
119 	int extralen = 0;
120 	u8 *iv;
121 	struct aead_request *req;
122 	struct scatterlist *sg;
123 
124 	if (x->props.flags & XFRM_STATE_ESN)
125 		extralen += sizeof(struct esp_output_extra);
126 
127 	iv = esp_tmp_iv(aead, tmp, extralen);
128 	req = esp_tmp_req(aead, iv);
129 
130 	/* Unref skb_frag_pages in the src scatterlist if necessary.
131 	 * Skip the first sg which comes from skb->data.
132 	 */
133 	if (already_unref || req->src != req->dst) {
134 		struct scatterlist *src = already_unref ? esp_req_sg(aead, req) : req->src;
135 
136 		for (sg = sg_next(src); sg; sg = sg_next(sg))
137 			skb_page_unref(page_to_netmem(sg_page(sg)),
138 				       skb->pp_recycle);
139 	}
140 }
141 
142 #ifdef CONFIG_INET6_ESPINTCP
143 static struct sock *esp6_find_tcp_sk(struct xfrm_state *x)
144 {
145 	struct xfrm_encap_tmpl *encap = x->encap;
146 	struct net *net = xs_net(x);
147 	__be16 sport, dport;
148 	struct sock *sk;
149 
150 	spin_lock_bh(&x->lock);
151 	sport = encap->encap_sport;
152 	dport = encap->encap_dport;
153 	spin_unlock_bh(&x->lock);
154 
155 	sk = __inet6_lookup_established(net, &x->id.daddr.in6, dport,
156 					&x->props.saddr.in6, ntohs(sport), 0, 0);
157 	if (!sk)
158 		return ERR_PTR(-ENOENT);
159 
160 	if (!tcp_is_ulp_esp(sk)) {
161 		sock_put(sk);
162 		return ERR_PTR(-EINVAL);
163 	}
164 
165 	return sk;
166 }
167 
168 static int esp_output_tcp_finish(struct xfrm_state *x, struct sk_buff *skb)
169 {
170 	struct sock *sk;
171 	int err;
172 
173 	rcu_read_lock();
174 
175 	sk = esp6_find_tcp_sk(x);
176 	err = PTR_ERR_OR_ZERO(sk);
177 	if (err) {
178 		kfree_skb(skb);
179 		goto out;
180 	}
181 
182 	bh_lock_sock(sk);
183 	if (sock_owned_by_user(sk))
184 		err = espintcp_queue_out(sk, skb);
185 	else
186 		err = espintcp_push_skb(sk, skb);
187 	bh_unlock_sock(sk);
188 
189 	sock_put(sk);
190 
191 out:
192 	rcu_read_unlock();
193 	return err;
194 }
195 
196 static int esp_output_tcp_encap_cb(struct net *net, struct sock *sk,
197 				   struct sk_buff *skb)
198 {
199 	struct dst_entry *dst = skb_dst(skb);
200 	struct xfrm_state *x = dst->xfrm;
201 
202 	return esp_output_tcp_finish(x, skb);
203 }
204 
205 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
206 {
207 	int err;
208 
209 	local_bh_disable();
210 	err = xfrm_trans_queue_net(xs_net(x), skb, esp_output_tcp_encap_cb);
211 	local_bh_enable();
212 
213 	/* EINPROGRESS just happens to do the right thing.  It
214 	 * actually means that the skb has been consumed and
215 	 * isn't coming back.
216 	 */
217 	return err ?: -EINPROGRESS;
218 }
219 #else
220 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
221 {
222 	WARN_ON(1);
223 	return -EOPNOTSUPP;
224 }
225 #endif
226 
227 static void esp_output_encap_csum(struct sk_buff *skb)
228 {
229 	/* UDP encap with IPv6 requires a valid checksum */
230 	if (*skb_mac_header(skb) == IPPROTO_UDP) {
231 		struct udphdr *uh = udp_hdr(skb);
232 		struct ipv6hdr *ip6h = ipv6_hdr(skb);
233 		int len = ntohs(uh->len);
234 		unsigned int offset = skb_transport_offset(skb);
235 		__wsum csum = skb_checksum(skb, offset, skb->len - offset, 0);
236 
237 		uh->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
238 					    len, IPPROTO_UDP, csum);
239 		if (uh->check == 0)
240 			uh->check = CSUM_MANGLED_0;
241 	}
242 }
243 
244 static void esp_output_done(void *data, int err)
245 {
246 	struct sk_buff *skb = data;
247 	struct xfrm_offload *xo = xfrm_offload(skb);
248 	void *tmp;
249 	struct xfrm_state *x;
250 
251 	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
252 		struct sec_path *sp = skb_sec_path(skb);
253 
254 		x = sp->xvec[sp->len - 1];
255 	} else {
256 		x = skb_dst(skb)->xfrm;
257 	}
258 
259 	tmp = ESP_SKB_CB(skb)->tmp;
260 	esp_ssg_unref(x, tmp, skb, false);
261 	kfree(tmp);
262 
263 	esp_output_encap_csum(skb);
264 
265 	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
266 		if (err) {
267 			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
268 			kfree_skb(skb);
269 			return;
270 		}
271 
272 		skb_push(skb, skb->data - skb_mac_header(skb));
273 		secpath_reset(skb);
274 		xfrm_dev_resume(skb);
275 	} else {
276 		if (!err &&
277 		    x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP) {
278 			err = esp_output_tail_tcp(x, skb);
279 			if (err != -EINPROGRESS)
280 				kfree_skb(skb);
281 		} else {
282 			xfrm_output_resume(skb_to_full_sk(skb), skb, err);
283 		}
284 	}
285 }
286 
287 /* Move ESP header back into place. */
288 static void esp_restore_header(struct sk_buff *skb, unsigned int offset)
289 {
290 	struct ip_esp_hdr *esph = (void *)(skb->data + offset);
291 	void *tmp = ESP_SKB_CB(skb)->tmp;
292 	__be32 *seqhi = esp_tmp_extra(tmp);
293 
294 	esph->seq_no = esph->spi;
295 	esph->spi = *seqhi;
296 }
297 
298 static void esp_output_restore_header(struct sk_buff *skb)
299 {
300 	void *tmp = ESP_SKB_CB(skb)->tmp;
301 	struct esp_output_extra *extra = esp_tmp_extra(tmp);
302 
303 	esp_restore_header(skb, skb_transport_offset(skb) + extra->esphoff -
304 				sizeof(__be32));
305 }
306 
307 static struct ip_esp_hdr *esp_output_set_esn(struct sk_buff *skb,
308 					     struct xfrm_state *x,
309 					     struct ip_esp_hdr *esph,
310 					     struct esp_output_extra *extra)
311 {
312 	/* For ESN we move the header forward by 4 bytes to
313 	 * accommodate the high bits.  We will move it back after
314 	 * encryption.
315 	 */
316 	if ((x->props.flags & XFRM_STATE_ESN)) {
317 		__u32 seqhi;
318 		struct xfrm_offload *xo = xfrm_offload(skb);
319 
320 		if (xo)
321 			seqhi = xo->seq.hi;
322 		else
323 			seqhi = XFRM_SKB_CB(skb)->seq.output.hi;
324 
325 		extra->esphoff = (unsigned char *)esph -
326 				 skb_transport_header(skb);
327 		esph = (struct ip_esp_hdr *)((unsigned char *)esph - 4);
328 		extra->seqhi = esph->spi;
329 		esph->seq_no = htonl(seqhi);
330 	}
331 
332 	esph->spi = x->id.spi;
333 
334 	return esph;
335 }
336 
337 static void esp_output_done_esn(void *data, int err)
338 {
339 	struct sk_buff *skb = data;
340 
341 	esp_output_restore_header(skb);
342 	esp_output_done(data, err);
343 }
344 
345 static struct ip_esp_hdr *esp6_output_udp_encap(struct sk_buff *skb,
346 					       int encap_type,
347 					       struct esp_info *esp,
348 					       __be16 sport,
349 					       __be16 dport)
350 {
351 	struct udphdr *uh;
352 	unsigned int len;
353 
354 	len = skb->len + esp->tailen - skb_transport_offset(skb);
355 	if (len > U16_MAX)
356 		return ERR_PTR(-EMSGSIZE);
357 
358 	uh = (struct udphdr *)esp->esph;
359 	uh->source = sport;
360 	uh->dest = dport;
361 	uh->len = htons(len);
362 	uh->check = 0;
363 
364 	*skb_mac_header(skb) = IPPROTO_UDP;
365 
366 	return (struct ip_esp_hdr *)(uh + 1);
367 }
368 
369 #ifdef CONFIG_INET6_ESPINTCP
370 static struct ip_esp_hdr *esp6_output_tcp_encap(struct xfrm_state *x,
371 						struct sk_buff *skb,
372 						struct esp_info *esp)
373 {
374 	__be16 *lenp = (void *)esp->esph;
375 	struct ip_esp_hdr *esph;
376 	unsigned int len;
377 	struct sock *sk;
378 
379 	len = skb->len + esp->tailen - skb_transport_offset(skb);
380 	if (len > IP_MAX_MTU)
381 		return ERR_PTR(-EMSGSIZE);
382 
383 	rcu_read_lock();
384 	sk = esp6_find_tcp_sk(x);
385 	rcu_read_unlock();
386 
387 	if (IS_ERR(sk))
388 		return ERR_CAST(sk);
389 
390 	sock_put(sk);
391 
392 	*lenp = htons(len);
393 	esph = (struct ip_esp_hdr *)(lenp + 1);
394 
395 	return esph;
396 }
397 #else
398 static struct ip_esp_hdr *esp6_output_tcp_encap(struct xfrm_state *x,
399 						struct sk_buff *skb,
400 						struct esp_info *esp)
401 {
402 	return ERR_PTR(-EOPNOTSUPP);
403 }
404 #endif
405 
406 static int esp6_output_encap(struct xfrm_state *x, struct sk_buff *skb,
407 			    struct esp_info *esp)
408 {
409 	struct xfrm_encap_tmpl *encap = x->encap;
410 	struct ip_esp_hdr *esph;
411 	__be16 sport, dport;
412 	int encap_type;
413 
414 	spin_lock_bh(&x->lock);
415 	sport = encap->encap_sport;
416 	dport = encap->encap_dport;
417 	encap_type = encap->encap_type;
418 	spin_unlock_bh(&x->lock);
419 
420 	switch (encap_type) {
421 	default:
422 	case UDP_ENCAP_ESPINUDP:
423 		esph = esp6_output_udp_encap(skb, encap_type, esp, sport, dport);
424 		break;
425 	case TCP_ENCAP_ESPINTCP:
426 		esph = esp6_output_tcp_encap(x, skb, esp);
427 		break;
428 	}
429 
430 	if (IS_ERR(esph))
431 		return PTR_ERR(esph);
432 
433 	esp->esph = esph;
434 
435 	return 0;
436 }
437 
438 int esp6_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
439 {
440 	u8 *tail;
441 	int nfrags;
442 	int esph_offset;
443 	struct page *page;
444 	struct sk_buff *trailer;
445 	int tailen = esp->tailen;
446 
447 	if (x->encap) {
448 		int err = esp6_output_encap(x, skb, esp);
449 
450 		if (err < 0)
451 			return err;
452 	}
453 
454 	if (ALIGN(skb->data_len + tailen, L1_CACHE_BYTES) >
455 	    PAGE_SIZE)
456 		goto cow;
457 
458 	if (!skb_cloned(skb)) {
459 		if (tailen <= skb_tailroom(skb)) {
460 			nfrags = 1;
461 			trailer = skb;
462 			tail = skb_tail_pointer(trailer);
463 
464 			goto skip_cow;
465 		} else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
466 			   && !skb_has_frag_list(skb)) {
467 			int allocsize;
468 			struct sock *sk = skb->sk;
469 			struct page_frag *pfrag = &x->xfrag;
470 
471 			esp->inplace = false;
472 
473 			allocsize = ALIGN(tailen, L1_CACHE_BYTES);
474 
475 			spin_lock_bh(&x->lock);
476 
477 			if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
478 				spin_unlock_bh(&x->lock);
479 				goto cow;
480 			}
481 
482 			page = pfrag->page;
483 			get_page(page);
484 
485 			tail = page_address(page) + pfrag->offset;
486 
487 			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
488 
489 			nfrags = skb_shinfo(skb)->nr_frags;
490 
491 			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
492 					     tailen);
493 			skb_shinfo(skb)->nr_frags = ++nfrags;
494 
495 			pfrag->offset = pfrag->offset + allocsize;
496 
497 			spin_unlock_bh(&x->lock);
498 
499 			nfrags++;
500 
501 			skb->len += tailen;
502 			skb->data_len += tailen;
503 			skb->truesize += tailen;
504 			if (sk && sk_fullsock(sk))
505 				refcount_add(tailen, &sk->sk_wmem_alloc);
506 
507 			goto out;
508 		}
509 	}
510 
511 cow:
512 	esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
513 
514 	nfrags = skb_cow_data(skb, tailen, &trailer);
515 	if (nfrags < 0)
516 		goto out;
517 	tail = skb_tail_pointer(trailer);
518 	esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
519 
520 skip_cow:
521 	esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
522 	pskb_put(skb, trailer, tailen);
523 
524 out:
525 	return nfrags;
526 }
527 EXPORT_SYMBOL_GPL(esp6_output_head);
528 
529 int esp6_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
530 {
531 	u8 *iv;
532 	int alen;
533 	void *tmp;
534 	int ivlen;
535 	int assoclen;
536 	int extralen;
537 	struct page *page;
538 	struct ip_esp_hdr *esph;
539 	struct aead_request *req;
540 	struct crypto_aead *aead;
541 	struct scatterlist *sg, *dsg;
542 	struct esp_output_extra *extra;
543 	int err = -ENOMEM;
544 
545 	assoclen = sizeof(struct ip_esp_hdr);
546 	extralen = 0;
547 
548 	if (x->props.flags & XFRM_STATE_ESN) {
549 		extralen += sizeof(*extra);
550 		assoclen += sizeof(__be32);
551 	}
552 
553 	aead = x->data;
554 	alen = crypto_aead_authsize(aead);
555 	ivlen = crypto_aead_ivsize(aead);
556 
557 	tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
558 	if (!tmp)
559 		goto error;
560 
561 	extra = esp_tmp_extra(tmp);
562 	iv = esp_tmp_iv(aead, tmp, extralen);
563 	req = esp_tmp_req(aead, iv);
564 	sg = esp_req_sg(aead, req);
565 
566 	if (esp->inplace)
567 		dsg = sg;
568 	else
569 		dsg = &sg[esp->nfrags];
570 
571 	esph = esp_output_set_esn(skb, x, esp->esph, extra);
572 	esp->esph = esph;
573 
574 	sg_init_table(sg, esp->nfrags);
575 	err = skb_to_sgvec(skb, sg,
576 		           (unsigned char *)esph - skb->data,
577 		           assoclen + ivlen + esp->clen + alen);
578 	if (unlikely(err < 0))
579 		goto error_free;
580 
581 	if (!esp->inplace) {
582 		int allocsize;
583 		struct page_frag *pfrag = &x->xfrag;
584 
585 		allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
586 
587 		spin_lock_bh(&x->lock);
588 		if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
589 			spin_unlock_bh(&x->lock);
590 			goto error_free;
591 		}
592 
593 		skb_shinfo(skb)->nr_frags = 1;
594 
595 		page = pfrag->page;
596 		get_page(page);
597 		/* replace page frags in skb with new page */
598 		__skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
599 		pfrag->offset = pfrag->offset + allocsize;
600 		spin_unlock_bh(&x->lock);
601 
602 		sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
603 		err = skb_to_sgvec(skb, dsg,
604 			           (unsigned char *)esph - skb->data,
605 			           assoclen + ivlen + esp->clen + alen);
606 		if (unlikely(err < 0)) {
607 			esp_ssg_unref(x, tmp, skb, true);
608 			goto error_free;
609 		}
610 	}
611 
612 	if ((x->props.flags & XFRM_STATE_ESN))
613 		aead_request_set_callback(req, 0, esp_output_done_esn, skb);
614 	else
615 		aead_request_set_callback(req, 0, esp_output_done, skb);
616 
617 	aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
618 	aead_request_set_ad(req, assoclen);
619 
620 	memset(iv, 0, ivlen);
621 	memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
622 	       min(ivlen, 8));
623 
624 	ESP_SKB_CB(skb)->tmp = tmp;
625 	err = crypto_aead_encrypt(req);
626 
627 	switch (err) {
628 	case -EINPROGRESS:
629 		goto error;
630 
631 	case -ENOSPC:
632 		err = NET_XMIT_DROP;
633 		break;
634 
635 	case 0:
636 		if ((x->props.flags & XFRM_STATE_ESN))
637 			esp_output_restore_header(skb);
638 		esp_output_encap_csum(skb);
639 	}
640 
641 	if (sg != dsg)
642 		esp_ssg_unref(x, tmp, skb, false);
643 
644 	if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
645 		err = esp_output_tail_tcp(x, skb);
646 
647 error_free:
648 	kfree(tmp);
649 error:
650 	return err;
651 }
652 EXPORT_SYMBOL_GPL(esp6_output_tail);
653 
654 static int esp6_output(struct xfrm_state *x, struct sk_buff *skb)
655 {
656 	int alen;
657 	int blksize;
658 	struct ip_esp_hdr *esph;
659 	struct crypto_aead *aead;
660 	struct esp_info esp;
661 
662 	esp.inplace = true;
663 
664 	esp.proto = *skb_mac_header(skb);
665 	*skb_mac_header(skb) = IPPROTO_ESP;
666 
667 	/* skb is pure payload to encrypt */
668 
669 	aead = x->data;
670 	alen = crypto_aead_authsize(aead);
671 
672 	esp.tfclen = 0;
673 	if (x->tfcpad) {
674 		struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
675 		u32 padto;
676 
677 		padto = min(x->tfcpad, xfrm_state_mtu(x, dst->child_mtu_cached));
678 		if (skb->len < padto)
679 			esp.tfclen = padto - skb->len;
680 	}
681 	blksize = ALIGN(crypto_aead_blocksize(aead), 4);
682 	esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
683 	esp.plen = esp.clen - skb->len - esp.tfclen;
684 	esp.tailen = esp.tfclen + esp.plen + alen;
685 
686 	esp.esph = ip_esp_hdr(skb);
687 
688 	esp.nfrags = esp6_output_head(x, skb, &esp);
689 	if (esp.nfrags < 0)
690 		return esp.nfrags;
691 
692 	esph = esp.esph;
693 	esph->spi = x->id.spi;
694 
695 	esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
696 	esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
697 			    ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
698 
699 	skb_push(skb, -skb_network_offset(skb));
700 
701 	return esp6_output_tail(x, skb, &esp);
702 }
703 
704 static inline int esp_remove_trailer(struct sk_buff *skb)
705 {
706 	struct xfrm_state *x = xfrm_input_state(skb);
707 	struct crypto_aead *aead = x->data;
708 	int alen, hlen, elen;
709 	int padlen, trimlen;
710 	__wsum csumdiff;
711 	u8 nexthdr[2];
712 	int ret;
713 
714 	alen = crypto_aead_authsize(aead);
715 	hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
716 	elen = skb->len - hlen;
717 
718 	ret = skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2);
719 	BUG_ON(ret);
720 
721 	ret = -EINVAL;
722 	padlen = nexthdr[0];
723 	if (padlen + 2 + alen >= elen) {
724 		net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
725 				    padlen + 2, elen - alen);
726 		goto out;
727 	}
728 
729 	trimlen = alen + padlen + 2;
730 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
731 		csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
732 		skb->csum = csum_block_sub(skb->csum, csumdiff,
733 					   skb->len - trimlen);
734 	}
735 	ret = pskb_trim(skb, skb->len - trimlen);
736 	if (unlikely(ret))
737 		return ret;
738 
739 	ret = nexthdr[1];
740 
741 out:
742 	return ret;
743 }
744 
745 int esp6_input_done2(struct sk_buff *skb, int err)
746 {
747 	struct xfrm_state *x = xfrm_input_state(skb);
748 	struct xfrm_offload *xo = xfrm_offload(skb);
749 	struct crypto_aead *aead = x->data;
750 	int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
751 	int hdr_len = skb_network_header_len(skb);
752 
753 	if (!xo || !(xo->flags & CRYPTO_DONE))
754 		kfree(ESP_SKB_CB(skb)->tmp);
755 
756 	if (unlikely(err))
757 		goto out;
758 
759 	err = esp_remove_trailer(skb);
760 	if (unlikely(err < 0))
761 		goto out;
762 
763 	if (x->encap) {
764 		const struct ipv6hdr *ip6h = ipv6_hdr(skb);
765 		int offset = skb_network_offset(skb) + sizeof(*ip6h);
766 		struct xfrm_encap_tmpl *encap = x->encap;
767 		u8 nexthdr = ip6h->nexthdr;
768 		__be16 frag_off, source;
769 		struct udphdr *uh;
770 		struct tcphdr *th;
771 
772 		offset = ipv6_skip_exthdr(skb, offset, &nexthdr, &frag_off);
773 		if (offset == -1) {
774 			err = -EINVAL;
775 			goto out;
776 		}
777 
778 		uh = (void *)(skb->data + offset);
779 		th = (void *)(skb->data + offset);
780 		hdr_len += offset;
781 
782 		switch (x->encap->encap_type) {
783 		case TCP_ENCAP_ESPINTCP:
784 			source = th->source;
785 			break;
786 		case UDP_ENCAP_ESPINUDP:
787 			source = uh->source;
788 			break;
789 		default:
790 			WARN_ON_ONCE(1);
791 			err = -EINVAL;
792 			goto out;
793 		}
794 
795 		/*
796 		 * 1) if the NAT-T peer's IP or port changed then
797 		 *    advertise the change to the keying daemon.
798 		 *    This is an inbound SA, so just compare
799 		 *    SRC ports.
800 		 */
801 		if (!ipv6_addr_equal(&ip6h->saddr, &x->props.saddr.in6) ||
802 		    source != encap->encap_sport) {
803 			xfrm_address_t ipaddr;
804 
805 			memcpy(&ipaddr.a6, &ip6h->saddr.s6_addr, sizeof(ipaddr.a6));
806 			km_new_mapping(x, &ipaddr, source);
807 
808 			/* XXX: perhaps add an extra
809 			 * policy check here, to see
810 			 * if we should allow or
811 			 * reject a packet from a
812 			 * different source
813 			 * address/port.
814 			 */
815 		}
816 
817 		/*
818 		 * 2) ignore UDP/TCP checksums in case
819 		 *    of NAT-T in Transport Mode, or
820 		 *    perform other post-processing fixes
821 		 *    as per draft-ietf-ipsec-udp-encaps-06,
822 		 *    section 3.1.2
823 		 */
824 		if (x->props.mode == XFRM_MODE_TRANSPORT)
825 			skb->ip_summed = CHECKSUM_UNNECESSARY;
826 	}
827 
828 	skb_postpull_rcsum(skb, skb_network_header(skb),
829 			   skb_network_header_len(skb));
830 	skb_pull_rcsum(skb, hlen);
831 	if (x->props.mode == XFRM_MODE_TUNNEL ||
832 	    x->props.mode == XFRM_MODE_IPTFS)
833 		skb_reset_transport_header(skb);
834 	else
835 		skb_set_transport_header(skb, -hdr_len);
836 
837 	/* RFC4303: Drop dummy packets without any error */
838 	if (err == IPPROTO_NONE)
839 		err = -EINVAL;
840 
841 out:
842 	return err;
843 }
844 EXPORT_SYMBOL_GPL(esp6_input_done2);
845 
846 static void esp_input_done(void *data, int err)
847 {
848 	struct sk_buff *skb = data;
849 
850 	xfrm_input_resume(skb, esp6_input_done2(skb, err));
851 }
852 
853 static void esp_input_restore_header(struct sk_buff *skb)
854 {
855 	esp_restore_header(skb, 0);
856 	__skb_pull(skb, 4);
857 }
858 
859 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
860 {
861 	struct xfrm_state *x = xfrm_input_state(skb);
862 
863 	/* For ESN we move the header forward by 4 bytes to
864 	 * accommodate the high bits.  We will move it back after
865 	 * decryption.
866 	 */
867 	if ((x->props.flags & XFRM_STATE_ESN)) {
868 		struct ip_esp_hdr *esph = skb_push(skb, 4);
869 
870 		*seqhi = esph->spi;
871 		esph->spi = esph->seq_no;
872 		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
873 	}
874 }
875 
876 static void esp_input_done_esn(void *data, int err)
877 {
878 	struct sk_buff *skb = data;
879 
880 	esp_input_restore_header(skb);
881 	esp_input_done(data, err);
882 }
883 
884 static int esp6_input(struct xfrm_state *x, struct sk_buff *skb)
885 {
886 	struct crypto_aead *aead = x->data;
887 	struct aead_request *req;
888 	struct sk_buff *trailer;
889 	int ivlen = crypto_aead_ivsize(aead);
890 	int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
891 	int nfrags;
892 	int assoclen;
893 	int seqhilen;
894 	int ret = 0;
895 	void *tmp;
896 	__be32 *seqhi;
897 	u8 *iv;
898 	struct scatterlist *sg;
899 
900 	if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen)) {
901 		ret = -EINVAL;
902 		goto out;
903 	}
904 
905 	if (elen <= 0) {
906 		ret = -EINVAL;
907 		goto out;
908 	}
909 
910 	assoclen = sizeof(struct ip_esp_hdr);
911 	seqhilen = 0;
912 
913 	if (x->props.flags & XFRM_STATE_ESN) {
914 		seqhilen += sizeof(__be32);
915 		assoclen += seqhilen;
916 	}
917 
918 	if (!skb_cloned(skb)) {
919 		if (!skb_is_nonlinear(skb)) {
920 			nfrags = 1;
921 
922 			goto skip_cow;
923 		} else if (!skb_has_frag_list(skb) &&
924 			   !skb_has_shared_frag(skb)) {
925 			nfrags = skb_shinfo(skb)->nr_frags;
926 			nfrags++;
927 
928 			goto skip_cow;
929 		}
930 	}
931 
932 	nfrags = skb_cow_data(skb, 0, &trailer);
933 	if (nfrags < 0) {
934 		ret = -EINVAL;
935 		goto out;
936 	}
937 
938 skip_cow:
939 	ret = -ENOMEM;
940 	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
941 	if (!tmp)
942 		goto out;
943 
944 	ESP_SKB_CB(skb)->tmp = tmp;
945 	seqhi = esp_tmp_extra(tmp);
946 	iv = esp_tmp_iv(aead, tmp, seqhilen);
947 	req = esp_tmp_req(aead, iv);
948 	sg = esp_req_sg(aead, req);
949 
950 	esp_input_set_header(skb, seqhi);
951 
952 	sg_init_table(sg, nfrags);
953 	ret = skb_to_sgvec(skb, sg, 0, skb->len);
954 	if (unlikely(ret < 0)) {
955 		kfree(tmp);
956 		goto out;
957 	}
958 
959 	skb->ip_summed = CHECKSUM_NONE;
960 
961 	if ((x->props.flags & XFRM_STATE_ESN))
962 		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
963 	else
964 		aead_request_set_callback(req, 0, esp_input_done, skb);
965 
966 	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
967 	aead_request_set_ad(req, assoclen);
968 
969 	ret = crypto_aead_decrypt(req);
970 	if (ret == -EINPROGRESS)
971 		goto out;
972 
973 	if ((x->props.flags & XFRM_STATE_ESN))
974 		esp_input_restore_header(skb);
975 
976 	ret = esp6_input_done2(skb, ret);
977 
978 out:
979 	return ret;
980 }
981 
982 static int esp6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
983 		    u8 type, u8 code, int offset, __be32 info)
984 {
985 	struct net *net = dev_net(skb->dev);
986 	const struct ipv6hdr *iph = (const struct ipv6hdr *)skb->data;
987 	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data + offset);
988 	struct xfrm_state *x;
989 
990 	if (type != ICMPV6_PKT_TOOBIG &&
991 	    type != NDISC_REDIRECT)
992 		return 0;
993 
994 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
995 			      esph->spi, IPPROTO_ESP, AF_INET6);
996 	if (!x)
997 		return 0;
998 
999 	if (type == NDISC_REDIRECT)
1000 		ip6_redirect(skb, net, skb->dev->ifindex, 0,
1001 			     sock_net_uid(net, NULL));
1002 	else
1003 		ip6_update_pmtu(skb, net, info, 0, 0, sock_net_uid(net, NULL));
1004 	xfrm_state_put(x);
1005 
1006 	return 0;
1007 }
1008 
1009 static void esp6_destroy(struct xfrm_state *x)
1010 {
1011 	struct crypto_aead *aead = x->data;
1012 
1013 	if (!aead)
1014 		return;
1015 
1016 	crypto_free_aead(aead);
1017 }
1018 
1019 static int esp_init_aead(struct xfrm_state *x, struct netlink_ext_ack *extack)
1020 {
1021 	char aead_name[CRYPTO_MAX_ALG_NAME];
1022 	struct crypto_aead *aead;
1023 	int err;
1024 
1025 	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1026 		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME) {
1027 		NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1028 		return -ENAMETOOLONG;
1029 	}
1030 
1031 	aead = crypto_alloc_aead(aead_name, 0, 0);
1032 	err = PTR_ERR(aead);
1033 	if (IS_ERR(aead))
1034 		goto error;
1035 
1036 	x->data = aead;
1037 
1038 	err = crypto_aead_setkey(aead, x->aead->alg_key,
1039 				 (x->aead->alg_key_len + 7) / 8);
1040 	if (err)
1041 		goto error;
1042 
1043 	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1044 	if (err)
1045 		goto error;
1046 
1047 	return 0;
1048 
1049 error:
1050 	NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1051 	return err;
1052 }
1053 
1054 static int esp_init_authenc(struct xfrm_state *x,
1055 			    struct netlink_ext_ack *extack)
1056 {
1057 	struct crypto_aead *aead;
1058 	struct crypto_authenc_key_param *param;
1059 	struct rtattr *rta;
1060 	char *key;
1061 	char *p;
1062 	char authenc_name[CRYPTO_MAX_ALG_NAME];
1063 	unsigned int keylen;
1064 	int err;
1065 
1066 	err = -ENAMETOOLONG;
1067 
1068 	if ((x->props.flags & XFRM_STATE_ESN)) {
1069 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1070 			     "%s%sauthencesn(%s,%s)%s",
1071 			     x->geniv ?: "", x->geniv ? "(" : "",
1072 			     x->aalg ? x->aalg->alg_name : "digest_null",
1073 			     x->ealg->alg_name,
1074 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1075 			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1076 			goto error;
1077 		}
1078 	} else {
1079 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1080 			     "%s%sauthenc(%s,%s)%s",
1081 			     x->geniv ?: "", x->geniv ? "(" : "",
1082 			     x->aalg ? x->aalg->alg_name : "digest_null",
1083 			     x->ealg->alg_name,
1084 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1085 			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1086 			goto error;
1087 		}
1088 	}
1089 
1090 	aead = crypto_alloc_aead(authenc_name, 0, 0);
1091 	err = PTR_ERR(aead);
1092 	if (IS_ERR(aead)) {
1093 		NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1094 		goto error;
1095 	}
1096 
1097 	x->data = aead;
1098 
1099 	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1100 		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1101 	err = -ENOMEM;
1102 	key = kmalloc(keylen, GFP_KERNEL);
1103 	if (!key)
1104 		goto error;
1105 
1106 	p = key;
1107 	rta = (void *)p;
1108 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1109 	rta->rta_len = RTA_LENGTH(sizeof(*param));
1110 	param = RTA_DATA(rta);
1111 	p += RTA_SPACE(sizeof(*param));
1112 
1113 	if (x->aalg) {
1114 		struct xfrm_algo_desc *aalg_desc;
1115 
1116 		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1117 		p += (x->aalg->alg_key_len + 7) / 8;
1118 
1119 		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1120 		BUG_ON(!aalg_desc);
1121 
1122 		err = -EINVAL;
1123 		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1124 		    crypto_aead_authsize(aead)) {
1125 			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1126 			goto free_key;
1127 		}
1128 
1129 		err = crypto_aead_setauthsize(
1130 			aead, x->aalg->alg_trunc_len / 8);
1131 		if (err) {
1132 			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1133 			goto free_key;
1134 		}
1135 	}
1136 
1137 	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1138 	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1139 
1140 	err = crypto_aead_setkey(aead, key, keylen);
1141 
1142 free_key:
1143 	kfree(key);
1144 
1145 error:
1146 	return err;
1147 }
1148 
1149 static int esp6_init_state(struct xfrm_state *x, struct netlink_ext_ack *extack)
1150 {
1151 	struct crypto_aead *aead;
1152 	u32 align;
1153 	int err;
1154 
1155 	x->data = NULL;
1156 
1157 	if (x->aead) {
1158 		err = esp_init_aead(x, extack);
1159 	} else if (x->ealg) {
1160 		err = esp_init_authenc(x, extack);
1161 	} else {
1162 		NL_SET_ERR_MSG(extack, "ESP: AEAD or CRYPT must be provided");
1163 		err = -EINVAL;
1164 	}
1165 
1166 	if (err)
1167 		goto error;
1168 
1169 	aead = x->data;
1170 
1171 	x->props.header_len = sizeof(struct ip_esp_hdr) +
1172 			      crypto_aead_ivsize(aead);
1173 	switch (x->props.mode) {
1174 	case XFRM_MODE_BEET:
1175 		if (x->sel.family != AF_INET6)
1176 			x->props.header_len += IPV4_BEET_PHMAXLEN +
1177 					       (sizeof(struct ipv6hdr) - sizeof(struct iphdr));
1178 		break;
1179 	default:
1180 	case XFRM_MODE_TRANSPORT:
1181 		break;
1182 	case XFRM_MODE_TUNNEL:
1183 		x->props.header_len += sizeof(struct ipv6hdr);
1184 		break;
1185 	}
1186 
1187 	if (x->encap) {
1188 		struct xfrm_encap_tmpl *encap = x->encap;
1189 
1190 		switch (encap->encap_type) {
1191 		default:
1192 			NL_SET_ERR_MSG(extack, "Unsupported encapsulation type for ESP");
1193 			err = -EINVAL;
1194 			goto error;
1195 		case UDP_ENCAP_ESPINUDP:
1196 			x->props.header_len += sizeof(struct udphdr);
1197 			break;
1198 #ifdef CONFIG_INET6_ESPINTCP
1199 		case TCP_ENCAP_ESPINTCP:
1200 			/* only the length field, TCP encap is done by
1201 			 * the socket
1202 			 */
1203 			x->props.header_len += 2;
1204 			break;
1205 #endif
1206 		}
1207 	}
1208 
1209 	align = ALIGN(crypto_aead_blocksize(aead), 4);
1210 	x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1211 
1212 error:
1213 	return err;
1214 }
1215 
1216 static int esp6_rcv_cb(struct sk_buff *skb, int err)
1217 {
1218 	return 0;
1219 }
1220 
1221 static const struct xfrm_type esp6_type = {
1222 	.owner		= THIS_MODULE,
1223 	.proto		= IPPROTO_ESP,
1224 	.flags		= XFRM_TYPE_REPLAY_PROT,
1225 	.init_state	= esp6_init_state,
1226 	.destructor	= esp6_destroy,
1227 	.input		= esp6_input,
1228 	.output		= esp6_output,
1229 };
1230 
1231 static struct xfrm6_protocol esp6_protocol = {
1232 	.handler	=	xfrm6_rcv,
1233 	.input_handler	=	xfrm_input,
1234 	.cb_handler	=	esp6_rcv_cb,
1235 	.err_handler	=	esp6_err,
1236 	.priority	=	0,
1237 };
1238 
1239 static int __init esp6_init(void)
1240 {
1241 	if (xfrm_register_type(&esp6_type, AF_INET6) < 0) {
1242 		pr_info("%s: can't add xfrm type\n", __func__);
1243 		return -EAGAIN;
1244 	}
1245 	if (xfrm6_protocol_register(&esp6_protocol, IPPROTO_ESP) < 0) {
1246 		pr_info("%s: can't add protocol\n", __func__);
1247 		xfrm_unregister_type(&esp6_type, AF_INET6);
1248 		return -EAGAIN;
1249 	}
1250 
1251 	return 0;
1252 }
1253 
1254 static void __exit esp6_fini(void)
1255 {
1256 	if (xfrm6_protocol_deregister(&esp6_protocol, IPPROTO_ESP) < 0)
1257 		pr_info("%s: can't remove protocol\n", __func__);
1258 	xfrm_unregister_type(&esp6_type, AF_INET6);
1259 }
1260 
1261 module_init(esp6_init);
1262 module_exit(esp6_fini);
1263 
1264 MODULE_DESCRIPTION("IPv6 ESP transformation helpers");
1265 MODULE_LICENSE("GPL");
1266 MODULE_ALIAS_XFRM_TYPE(AF_INET6, XFRM_PROTO_ESP);
1267