xref: /linux/net/ipv6/esp6.c (revision bfb921b2a9d5d1123d1d10b196a39db629ddef87)
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)
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 (req->src != req->dst)
134 		for (sg = sg_next(req->src); sg; sg = sg_next(sg))
135 			skb_page_unref(sg_page(sg), skb->pp_recycle);
136 }
137 
138 #ifdef CONFIG_INET6_ESPINTCP
139 struct esp_tcp_sk {
140 	struct sock *sk;
141 	struct rcu_head rcu;
142 };
143 
144 static void esp_free_tcp_sk(struct rcu_head *head)
145 {
146 	struct esp_tcp_sk *esk = container_of(head, struct esp_tcp_sk, rcu);
147 
148 	sock_put(esk->sk);
149 	kfree(esk);
150 }
151 
152 static struct sock *esp6_find_tcp_sk(struct xfrm_state *x)
153 {
154 	struct xfrm_encap_tmpl *encap = x->encap;
155 	struct net *net = xs_net(x);
156 	struct esp_tcp_sk *esk;
157 	__be16 sport, dport;
158 	struct sock *nsk;
159 	struct sock *sk;
160 
161 	sk = rcu_dereference(x->encap_sk);
162 	if (sk && sk->sk_state == TCP_ESTABLISHED)
163 		return sk;
164 
165 	spin_lock_bh(&x->lock);
166 	sport = encap->encap_sport;
167 	dport = encap->encap_dport;
168 	nsk = rcu_dereference_protected(x->encap_sk,
169 					lockdep_is_held(&x->lock));
170 	if (sk && sk == nsk) {
171 		esk = kmalloc(sizeof(*esk), GFP_ATOMIC);
172 		if (!esk) {
173 			spin_unlock_bh(&x->lock);
174 			return ERR_PTR(-ENOMEM);
175 		}
176 		RCU_INIT_POINTER(x->encap_sk, NULL);
177 		esk->sk = sk;
178 		call_rcu(&esk->rcu, esp_free_tcp_sk);
179 	}
180 	spin_unlock_bh(&x->lock);
181 
182 	sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo, &x->id.daddr.in6,
183 					dport, &x->props.saddr.in6, ntohs(sport), 0, 0);
184 	if (!sk)
185 		return ERR_PTR(-ENOENT);
186 
187 	if (!tcp_is_ulp_esp(sk)) {
188 		sock_put(sk);
189 		return ERR_PTR(-EINVAL);
190 	}
191 
192 	spin_lock_bh(&x->lock);
193 	nsk = rcu_dereference_protected(x->encap_sk,
194 					lockdep_is_held(&x->lock));
195 	if (encap->encap_sport != sport ||
196 	    encap->encap_dport != dport) {
197 		sock_put(sk);
198 		sk = nsk ?: ERR_PTR(-EREMCHG);
199 	} else if (sk == nsk) {
200 		sock_put(sk);
201 	} else {
202 		rcu_assign_pointer(x->encap_sk, sk);
203 	}
204 	spin_unlock_bh(&x->lock);
205 
206 	return sk;
207 }
208 
209 static int esp_output_tcp_finish(struct xfrm_state *x, struct sk_buff *skb)
210 {
211 	struct sock *sk;
212 	int err;
213 
214 	rcu_read_lock();
215 
216 	sk = esp6_find_tcp_sk(x);
217 	err = PTR_ERR_OR_ZERO(sk);
218 	if (err)
219 		goto out;
220 
221 	bh_lock_sock(sk);
222 	if (sock_owned_by_user(sk))
223 		err = espintcp_queue_out(sk, skb);
224 	else
225 		err = espintcp_push_skb(sk, skb);
226 	bh_unlock_sock(sk);
227 
228 out:
229 	rcu_read_unlock();
230 	return err;
231 }
232 
233 static int esp_output_tcp_encap_cb(struct net *net, struct sock *sk,
234 				   struct sk_buff *skb)
235 {
236 	struct dst_entry *dst = skb_dst(skb);
237 	struct xfrm_state *x = dst->xfrm;
238 
239 	return esp_output_tcp_finish(x, skb);
240 }
241 
242 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
243 {
244 	int err;
245 
246 	local_bh_disable();
247 	err = xfrm_trans_queue_net(xs_net(x), skb, esp_output_tcp_encap_cb);
248 	local_bh_enable();
249 
250 	/* EINPROGRESS just happens to do the right thing.  It
251 	 * actually means that the skb has been consumed and
252 	 * isn't coming back.
253 	 */
254 	return err ?: -EINPROGRESS;
255 }
256 #else
257 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
258 {
259 	kfree_skb(skb);
260 
261 	return -EOPNOTSUPP;
262 }
263 #endif
264 
265 static void esp_output_encap_csum(struct sk_buff *skb)
266 {
267 	/* UDP encap with IPv6 requires a valid checksum */
268 	if (*skb_mac_header(skb) == IPPROTO_UDP) {
269 		struct udphdr *uh = udp_hdr(skb);
270 		struct ipv6hdr *ip6h = ipv6_hdr(skb);
271 		int len = ntohs(uh->len);
272 		unsigned int offset = skb_transport_offset(skb);
273 		__wsum csum = skb_checksum(skb, offset, skb->len - offset, 0);
274 
275 		uh->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
276 					    len, IPPROTO_UDP, csum);
277 		if (uh->check == 0)
278 			uh->check = CSUM_MANGLED_0;
279 	}
280 }
281 
282 static void esp_output_done(void *data, int err)
283 {
284 	struct sk_buff *skb = data;
285 	struct xfrm_offload *xo = xfrm_offload(skb);
286 	void *tmp;
287 	struct xfrm_state *x;
288 
289 	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
290 		struct sec_path *sp = skb_sec_path(skb);
291 
292 		x = sp->xvec[sp->len - 1];
293 	} else {
294 		x = skb_dst(skb)->xfrm;
295 	}
296 
297 	tmp = ESP_SKB_CB(skb)->tmp;
298 	esp_ssg_unref(x, tmp, skb);
299 	kfree(tmp);
300 
301 	esp_output_encap_csum(skb);
302 
303 	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
304 		if (err) {
305 			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
306 			kfree_skb(skb);
307 			return;
308 		}
309 
310 		skb_push(skb, skb->data - skb_mac_header(skb));
311 		secpath_reset(skb);
312 		xfrm_dev_resume(skb);
313 	} else {
314 		if (!err &&
315 		    x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
316 			esp_output_tail_tcp(x, skb);
317 		else
318 			xfrm_output_resume(skb->sk, skb, err);
319 	}
320 }
321 
322 /* Move ESP header back into place. */
323 static void esp_restore_header(struct sk_buff *skb, unsigned int offset)
324 {
325 	struct ip_esp_hdr *esph = (void *)(skb->data + offset);
326 	void *tmp = ESP_SKB_CB(skb)->tmp;
327 	__be32 *seqhi = esp_tmp_extra(tmp);
328 
329 	esph->seq_no = esph->spi;
330 	esph->spi = *seqhi;
331 }
332 
333 static void esp_output_restore_header(struct sk_buff *skb)
334 {
335 	void *tmp = ESP_SKB_CB(skb)->tmp;
336 	struct esp_output_extra *extra = esp_tmp_extra(tmp);
337 
338 	esp_restore_header(skb, skb_transport_offset(skb) + extra->esphoff -
339 				sizeof(__be32));
340 }
341 
342 static struct ip_esp_hdr *esp_output_set_esn(struct sk_buff *skb,
343 					     struct xfrm_state *x,
344 					     struct ip_esp_hdr *esph,
345 					     struct esp_output_extra *extra)
346 {
347 	/* For ESN we move the header forward by 4 bytes to
348 	 * accommodate the high bits.  We will move it back after
349 	 * encryption.
350 	 */
351 	if ((x->props.flags & XFRM_STATE_ESN)) {
352 		__u32 seqhi;
353 		struct xfrm_offload *xo = xfrm_offload(skb);
354 
355 		if (xo)
356 			seqhi = xo->seq.hi;
357 		else
358 			seqhi = XFRM_SKB_CB(skb)->seq.output.hi;
359 
360 		extra->esphoff = (unsigned char *)esph -
361 				 skb_transport_header(skb);
362 		esph = (struct ip_esp_hdr *)((unsigned char *)esph - 4);
363 		extra->seqhi = esph->spi;
364 		esph->seq_no = htonl(seqhi);
365 	}
366 
367 	esph->spi = x->id.spi;
368 
369 	return esph;
370 }
371 
372 static void esp_output_done_esn(void *data, int err)
373 {
374 	struct sk_buff *skb = data;
375 
376 	esp_output_restore_header(skb);
377 	esp_output_done(data, err);
378 }
379 
380 static struct ip_esp_hdr *esp6_output_udp_encap(struct sk_buff *skb,
381 					       int encap_type,
382 					       struct esp_info *esp,
383 					       __be16 sport,
384 					       __be16 dport)
385 {
386 	struct udphdr *uh;
387 	unsigned int len;
388 
389 	len = skb->len + esp->tailen - skb_transport_offset(skb);
390 	if (len > U16_MAX)
391 		return ERR_PTR(-EMSGSIZE);
392 
393 	uh = (struct udphdr *)esp->esph;
394 	uh->source = sport;
395 	uh->dest = dport;
396 	uh->len = htons(len);
397 	uh->check = 0;
398 
399 	*skb_mac_header(skb) = IPPROTO_UDP;
400 
401 	return (struct ip_esp_hdr *)(uh + 1);
402 }
403 
404 #ifdef CONFIG_INET6_ESPINTCP
405 static struct ip_esp_hdr *esp6_output_tcp_encap(struct xfrm_state *x,
406 						struct sk_buff *skb,
407 						struct esp_info *esp)
408 {
409 	__be16 *lenp = (void *)esp->esph;
410 	struct ip_esp_hdr *esph;
411 	unsigned int len;
412 	struct sock *sk;
413 
414 	len = skb->len + esp->tailen - skb_transport_offset(skb);
415 	if (len > IP_MAX_MTU)
416 		return ERR_PTR(-EMSGSIZE);
417 
418 	rcu_read_lock();
419 	sk = esp6_find_tcp_sk(x);
420 	rcu_read_unlock();
421 
422 	if (IS_ERR(sk))
423 		return ERR_CAST(sk);
424 
425 	*lenp = htons(len);
426 	esph = (struct ip_esp_hdr *)(lenp + 1);
427 
428 	return esph;
429 }
430 #else
431 static struct ip_esp_hdr *esp6_output_tcp_encap(struct xfrm_state *x,
432 						struct sk_buff *skb,
433 						struct esp_info *esp)
434 {
435 	return ERR_PTR(-EOPNOTSUPP);
436 }
437 #endif
438 
439 static int esp6_output_encap(struct xfrm_state *x, struct sk_buff *skb,
440 			    struct esp_info *esp)
441 {
442 	struct xfrm_encap_tmpl *encap = x->encap;
443 	struct ip_esp_hdr *esph;
444 	__be16 sport, dport;
445 	int encap_type;
446 
447 	spin_lock_bh(&x->lock);
448 	sport = encap->encap_sport;
449 	dport = encap->encap_dport;
450 	encap_type = encap->encap_type;
451 	spin_unlock_bh(&x->lock);
452 
453 	switch (encap_type) {
454 	default:
455 	case UDP_ENCAP_ESPINUDP:
456 		esph = esp6_output_udp_encap(skb, encap_type, esp, sport, dport);
457 		break;
458 	case TCP_ENCAP_ESPINTCP:
459 		esph = esp6_output_tcp_encap(x, skb, esp);
460 		break;
461 	}
462 
463 	if (IS_ERR(esph))
464 		return PTR_ERR(esph);
465 
466 	esp->esph = esph;
467 
468 	return 0;
469 }
470 
471 int esp6_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
472 {
473 	u8 *tail;
474 	int nfrags;
475 	int esph_offset;
476 	struct page *page;
477 	struct sk_buff *trailer;
478 	int tailen = esp->tailen;
479 
480 	if (x->encap) {
481 		int err = esp6_output_encap(x, skb, esp);
482 
483 		if (err < 0)
484 			return err;
485 	}
486 
487 	if (ALIGN(tailen, L1_CACHE_BYTES) > PAGE_SIZE ||
488 	    ALIGN(skb->data_len, L1_CACHE_BYTES) > PAGE_SIZE)
489 		goto cow;
490 
491 	if (!skb_cloned(skb)) {
492 		if (tailen <= skb_tailroom(skb)) {
493 			nfrags = 1;
494 			trailer = skb;
495 			tail = skb_tail_pointer(trailer);
496 
497 			goto skip_cow;
498 		} else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
499 			   && !skb_has_frag_list(skb)) {
500 			int allocsize;
501 			struct sock *sk = skb->sk;
502 			struct page_frag *pfrag = &x->xfrag;
503 
504 			esp->inplace = false;
505 
506 			allocsize = ALIGN(tailen, L1_CACHE_BYTES);
507 
508 			spin_lock_bh(&x->lock);
509 
510 			if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
511 				spin_unlock_bh(&x->lock);
512 				goto cow;
513 			}
514 
515 			page = pfrag->page;
516 			get_page(page);
517 
518 			tail = page_address(page) + pfrag->offset;
519 
520 			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
521 
522 			nfrags = skb_shinfo(skb)->nr_frags;
523 
524 			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
525 					     tailen);
526 			skb_shinfo(skb)->nr_frags = ++nfrags;
527 
528 			pfrag->offset = pfrag->offset + allocsize;
529 
530 			spin_unlock_bh(&x->lock);
531 
532 			nfrags++;
533 
534 			skb->len += tailen;
535 			skb->data_len += tailen;
536 			skb->truesize += tailen;
537 			if (sk && sk_fullsock(sk))
538 				refcount_add(tailen, &sk->sk_wmem_alloc);
539 
540 			goto out;
541 		}
542 	}
543 
544 cow:
545 	esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
546 
547 	nfrags = skb_cow_data(skb, tailen, &trailer);
548 	if (nfrags < 0)
549 		goto out;
550 	tail = skb_tail_pointer(trailer);
551 	esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
552 
553 skip_cow:
554 	esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
555 	pskb_put(skb, trailer, tailen);
556 
557 out:
558 	return nfrags;
559 }
560 EXPORT_SYMBOL_GPL(esp6_output_head);
561 
562 int esp6_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
563 {
564 	u8 *iv;
565 	int alen;
566 	void *tmp;
567 	int ivlen;
568 	int assoclen;
569 	int extralen;
570 	struct page *page;
571 	struct ip_esp_hdr *esph;
572 	struct aead_request *req;
573 	struct crypto_aead *aead;
574 	struct scatterlist *sg, *dsg;
575 	struct esp_output_extra *extra;
576 	int err = -ENOMEM;
577 
578 	assoclen = sizeof(struct ip_esp_hdr);
579 	extralen = 0;
580 
581 	if (x->props.flags & XFRM_STATE_ESN) {
582 		extralen += sizeof(*extra);
583 		assoclen += sizeof(__be32);
584 	}
585 
586 	aead = x->data;
587 	alen = crypto_aead_authsize(aead);
588 	ivlen = crypto_aead_ivsize(aead);
589 
590 	tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
591 	if (!tmp)
592 		goto error;
593 
594 	extra = esp_tmp_extra(tmp);
595 	iv = esp_tmp_iv(aead, tmp, extralen);
596 	req = esp_tmp_req(aead, iv);
597 	sg = esp_req_sg(aead, req);
598 
599 	if (esp->inplace)
600 		dsg = sg;
601 	else
602 		dsg = &sg[esp->nfrags];
603 
604 	esph = esp_output_set_esn(skb, x, esp->esph, extra);
605 	esp->esph = esph;
606 
607 	sg_init_table(sg, esp->nfrags);
608 	err = skb_to_sgvec(skb, sg,
609 		           (unsigned char *)esph - skb->data,
610 		           assoclen + ivlen + esp->clen + alen);
611 	if (unlikely(err < 0))
612 		goto error_free;
613 
614 	if (!esp->inplace) {
615 		int allocsize;
616 		struct page_frag *pfrag = &x->xfrag;
617 
618 		allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
619 
620 		spin_lock_bh(&x->lock);
621 		if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
622 			spin_unlock_bh(&x->lock);
623 			goto error_free;
624 		}
625 
626 		skb_shinfo(skb)->nr_frags = 1;
627 
628 		page = pfrag->page;
629 		get_page(page);
630 		/* replace page frags in skb with new page */
631 		__skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
632 		pfrag->offset = pfrag->offset + allocsize;
633 		spin_unlock_bh(&x->lock);
634 
635 		sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
636 		err = skb_to_sgvec(skb, dsg,
637 			           (unsigned char *)esph - skb->data,
638 			           assoclen + ivlen + esp->clen + alen);
639 		if (unlikely(err < 0))
640 			goto error_free;
641 	}
642 
643 	if ((x->props.flags & XFRM_STATE_ESN))
644 		aead_request_set_callback(req, 0, esp_output_done_esn, skb);
645 	else
646 		aead_request_set_callback(req, 0, esp_output_done, skb);
647 
648 	aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
649 	aead_request_set_ad(req, assoclen);
650 
651 	memset(iv, 0, ivlen);
652 	memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
653 	       min(ivlen, 8));
654 
655 	ESP_SKB_CB(skb)->tmp = tmp;
656 	err = crypto_aead_encrypt(req);
657 
658 	switch (err) {
659 	case -EINPROGRESS:
660 		goto error;
661 
662 	case -ENOSPC:
663 		err = NET_XMIT_DROP;
664 		break;
665 
666 	case 0:
667 		if ((x->props.flags & XFRM_STATE_ESN))
668 			esp_output_restore_header(skb);
669 		esp_output_encap_csum(skb);
670 	}
671 
672 	if (sg != dsg)
673 		esp_ssg_unref(x, tmp, skb);
674 
675 	if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
676 		err = esp_output_tail_tcp(x, skb);
677 
678 error_free:
679 	kfree(tmp);
680 error:
681 	return err;
682 }
683 EXPORT_SYMBOL_GPL(esp6_output_tail);
684 
685 static int esp6_output(struct xfrm_state *x, struct sk_buff *skb)
686 {
687 	int alen;
688 	int blksize;
689 	struct ip_esp_hdr *esph;
690 	struct crypto_aead *aead;
691 	struct esp_info esp;
692 
693 	esp.inplace = true;
694 
695 	esp.proto = *skb_mac_header(skb);
696 	*skb_mac_header(skb) = IPPROTO_ESP;
697 
698 	/* skb is pure payload to encrypt */
699 
700 	aead = x->data;
701 	alen = crypto_aead_authsize(aead);
702 
703 	esp.tfclen = 0;
704 	if (x->tfcpad) {
705 		struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
706 		u32 padto;
707 
708 		padto = min(x->tfcpad, xfrm_state_mtu(x, dst->child_mtu_cached));
709 		if (skb->len < padto)
710 			esp.tfclen = padto - skb->len;
711 	}
712 	blksize = ALIGN(crypto_aead_blocksize(aead), 4);
713 	esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
714 	esp.plen = esp.clen - skb->len - esp.tfclen;
715 	esp.tailen = esp.tfclen + esp.plen + alen;
716 
717 	esp.esph = ip_esp_hdr(skb);
718 
719 	esp.nfrags = esp6_output_head(x, skb, &esp);
720 	if (esp.nfrags < 0)
721 		return esp.nfrags;
722 
723 	esph = esp.esph;
724 	esph->spi = x->id.spi;
725 
726 	esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
727 	esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
728 			    ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
729 
730 	skb_push(skb, -skb_network_offset(skb));
731 
732 	return esp6_output_tail(x, skb, &esp);
733 }
734 
735 static inline int esp_remove_trailer(struct sk_buff *skb)
736 {
737 	struct xfrm_state *x = xfrm_input_state(skb);
738 	struct crypto_aead *aead = x->data;
739 	int alen, hlen, elen;
740 	int padlen, trimlen;
741 	__wsum csumdiff;
742 	u8 nexthdr[2];
743 	int ret;
744 
745 	alen = crypto_aead_authsize(aead);
746 	hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
747 	elen = skb->len - hlen;
748 
749 	ret = skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2);
750 	BUG_ON(ret);
751 
752 	ret = -EINVAL;
753 	padlen = nexthdr[0];
754 	if (padlen + 2 + alen >= elen) {
755 		net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
756 				    padlen + 2, elen - alen);
757 		goto out;
758 	}
759 
760 	trimlen = alen + padlen + 2;
761 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
762 		csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
763 		skb->csum = csum_block_sub(skb->csum, csumdiff,
764 					   skb->len - trimlen);
765 	}
766 	ret = pskb_trim(skb, skb->len - trimlen);
767 	if (unlikely(ret))
768 		return ret;
769 
770 	ret = nexthdr[1];
771 
772 out:
773 	return ret;
774 }
775 
776 int esp6_input_done2(struct sk_buff *skb, int err)
777 {
778 	struct xfrm_state *x = xfrm_input_state(skb);
779 	struct xfrm_offload *xo = xfrm_offload(skb);
780 	struct crypto_aead *aead = x->data;
781 	int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
782 	int hdr_len = skb_network_header_len(skb);
783 
784 	if (!xo || !(xo->flags & CRYPTO_DONE))
785 		kfree(ESP_SKB_CB(skb)->tmp);
786 
787 	if (unlikely(err))
788 		goto out;
789 
790 	err = esp_remove_trailer(skb);
791 	if (unlikely(err < 0))
792 		goto out;
793 
794 	if (x->encap) {
795 		const struct ipv6hdr *ip6h = ipv6_hdr(skb);
796 		int offset = skb_network_offset(skb) + sizeof(*ip6h);
797 		struct xfrm_encap_tmpl *encap = x->encap;
798 		u8 nexthdr = ip6h->nexthdr;
799 		__be16 frag_off, source;
800 		struct udphdr *uh;
801 		struct tcphdr *th;
802 
803 		offset = ipv6_skip_exthdr(skb, offset, &nexthdr, &frag_off);
804 		if (offset == -1) {
805 			err = -EINVAL;
806 			goto out;
807 		}
808 
809 		uh = (void *)(skb->data + offset);
810 		th = (void *)(skb->data + offset);
811 		hdr_len += offset;
812 
813 		switch (x->encap->encap_type) {
814 		case TCP_ENCAP_ESPINTCP:
815 			source = th->source;
816 			break;
817 		case UDP_ENCAP_ESPINUDP:
818 			source = uh->source;
819 			break;
820 		default:
821 			WARN_ON_ONCE(1);
822 			err = -EINVAL;
823 			goto out;
824 		}
825 
826 		/*
827 		 * 1) if the NAT-T peer's IP or port changed then
828 		 *    advertise the change to the keying daemon.
829 		 *    This is an inbound SA, so just compare
830 		 *    SRC ports.
831 		 */
832 		if (!ipv6_addr_equal(&ip6h->saddr, &x->props.saddr.in6) ||
833 		    source != encap->encap_sport) {
834 			xfrm_address_t ipaddr;
835 
836 			memcpy(&ipaddr.a6, &ip6h->saddr.s6_addr, sizeof(ipaddr.a6));
837 			km_new_mapping(x, &ipaddr, source);
838 
839 			/* XXX: perhaps add an extra
840 			 * policy check here, to see
841 			 * if we should allow or
842 			 * reject a packet from a
843 			 * different source
844 			 * address/port.
845 			 */
846 		}
847 
848 		/*
849 		 * 2) ignore UDP/TCP checksums in case
850 		 *    of NAT-T in Transport Mode, or
851 		 *    perform other post-processing fixes
852 		 *    as per draft-ietf-ipsec-udp-encaps-06,
853 		 *    section 3.1.2
854 		 */
855 		if (x->props.mode == XFRM_MODE_TRANSPORT)
856 			skb->ip_summed = CHECKSUM_UNNECESSARY;
857 	}
858 
859 	skb_postpull_rcsum(skb, skb_network_header(skb),
860 			   skb_network_header_len(skb));
861 	skb_pull_rcsum(skb, hlen);
862 	if (x->props.mode == XFRM_MODE_TUNNEL)
863 		skb_reset_transport_header(skb);
864 	else
865 		skb_set_transport_header(skb, -hdr_len);
866 
867 	/* RFC4303: Drop dummy packets without any error */
868 	if (err == IPPROTO_NONE)
869 		err = -EINVAL;
870 
871 out:
872 	return err;
873 }
874 EXPORT_SYMBOL_GPL(esp6_input_done2);
875 
876 static void esp_input_done(void *data, int err)
877 {
878 	struct sk_buff *skb = data;
879 
880 	xfrm_input_resume(skb, esp6_input_done2(skb, err));
881 }
882 
883 static void esp_input_restore_header(struct sk_buff *skb)
884 {
885 	esp_restore_header(skb, 0);
886 	__skb_pull(skb, 4);
887 }
888 
889 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
890 {
891 	struct xfrm_state *x = xfrm_input_state(skb);
892 
893 	/* For ESN we move the header forward by 4 bytes to
894 	 * accommodate the high bits.  We will move it back after
895 	 * decryption.
896 	 */
897 	if ((x->props.flags & XFRM_STATE_ESN)) {
898 		struct ip_esp_hdr *esph = skb_push(skb, 4);
899 
900 		*seqhi = esph->spi;
901 		esph->spi = esph->seq_no;
902 		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
903 	}
904 }
905 
906 static void esp_input_done_esn(void *data, int err)
907 {
908 	struct sk_buff *skb = data;
909 
910 	esp_input_restore_header(skb);
911 	esp_input_done(data, err);
912 }
913 
914 static int esp6_input(struct xfrm_state *x, struct sk_buff *skb)
915 {
916 	struct crypto_aead *aead = x->data;
917 	struct aead_request *req;
918 	struct sk_buff *trailer;
919 	int ivlen = crypto_aead_ivsize(aead);
920 	int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
921 	int nfrags;
922 	int assoclen;
923 	int seqhilen;
924 	int ret = 0;
925 	void *tmp;
926 	__be32 *seqhi;
927 	u8 *iv;
928 	struct scatterlist *sg;
929 
930 	if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen)) {
931 		ret = -EINVAL;
932 		goto out;
933 	}
934 
935 	if (elen <= 0) {
936 		ret = -EINVAL;
937 		goto out;
938 	}
939 
940 	assoclen = sizeof(struct ip_esp_hdr);
941 	seqhilen = 0;
942 
943 	if (x->props.flags & XFRM_STATE_ESN) {
944 		seqhilen += sizeof(__be32);
945 		assoclen += seqhilen;
946 	}
947 
948 	if (!skb_cloned(skb)) {
949 		if (!skb_is_nonlinear(skb)) {
950 			nfrags = 1;
951 
952 			goto skip_cow;
953 		} else if (!skb_has_frag_list(skb)) {
954 			nfrags = skb_shinfo(skb)->nr_frags;
955 			nfrags++;
956 
957 			goto skip_cow;
958 		}
959 	}
960 
961 	nfrags = skb_cow_data(skb, 0, &trailer);
962 	if (nfrags < 0) {
963 		ret = -EINVAL;
964 		goto out;
965 	}
966 
967 skip_cow:
968 	ret = -ENOMEM;
969 	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
970 	if (!tmp)
971 		goto out;
972 
973 	ESP_SKB_CB(skb)->tmp = tmp;
974 	seqhi = esp_tmp_extra(tmp);
975 	iv = esp_tmp_iv(aead, tmp, seqhilen);
976 	req = esp_tmp_req(aead, iv);
977 	sg = esp_req_sg(aead, req);
978 
979 	esp_input_set_header(skb, seqhi);
980 
981 	sg_init_table(sg, nfrags);
982 	ret = skb_to_sgvec(skb, sg, 0, skb->len);
983 	if (unlikely(ret < 0)) {
984 		kfree(tmp);
985 		goto out;
986 	}
987 
988 	skb->ip_summed = CHECKSUM_NONE;
989 
990 	if ((x->props.flags & XFRM_STATE_ESN))
991 		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
992 	else
993 		aead_request_set_callback(req, 0, esp_input_done, skb);
994 
995 	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
996 	aead_request_set_ad(req, assoclen);
997 
998 	ret = crypto_aead_decrypt(req);
999 	if (ret == -EINPROGRESS)
1000 		goto out;
1001 
1002 	if ((x->props.flags & XFRM_STATE_ESN))
1003 		esp_input_restore_header(skb);
1004 
1005 	ret = esp6_input_done2(skb, ret);
1006 
1007 out:
1008 	return ret;
1009 }
1010 
1011 static int esp6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
1012 		    u8 type, u8 code, int offset, __be32 info)
1013 {
1014 	struct net *net = dev_net(skb->dev);
1015 	const struct ipv6hdr *iph = (const struct ipv6hdr *)skb->data;
1016 	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data + offset);
1017 	struct xfrm_state *x;
1018 
1019 	if (type != ICMPV6_PKT_TOOBIG &&
1020 	    type != NDISC_REDIRECT)
1021 		return 0;
1022 
1023 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
1024 			      esph->spi, IPPROTO_ESP, AF_INET6);
1025 	if (!x)
1026 		return 0;
1027 
1028 	if (type == NDISC_REDIRECT)
1029 		ip6_redirect(skb, net, skb->dev->ifindex, 0,
1030 			     sock_net_uid(net, NULL));
1031 	else
1032 		ip6_update_pmtu(skb, net, info, 0, 0, sock_net_uid(net, NULL));
1033 	xfrm_state_put(x);
1034 
1035 	return 0;
1036 }
1037 
1038 static void esp6_destroy(struct xfrm_state *x)
1039 {
1040 	struct crypto_aead *aead = x->data;
1041 
1042 	if (!aead)
1043 		return;
1044 
1045 	crypto_free_aead(aead);
1046 }
1047 
1048 static int esp_init_aead(struct xfrm_state *x, struct netlink_ext_ack *extack)
1049 {
1050 	char aead_name[CRYPTO_MAX_ALG_NAME];
1051 	struct crypto_aead *aead;
1052 	int err;
1053 
1054 	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1055 		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME) {
1056 		NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1057 		return -ENAMETOOLONG;
1058 	}
1059 
1060 	aead = crypto_alloc_aead(aead_name, 0, 0);
1061 	err = PTR_ERR(aead);
1062 	if (IS_ERR(aead))
1063 		goto error;
1064 
1065 	x->data = aead;
1066 
1067 	err = crypto_aead_setkey(aead, x->aead->alg_key,
1068 				 (x->aead->alg_key_len + 7) / 8);
1069 	if (err)
1070 		goto error;
1071 
1072 	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1073 	if (err)
1074 		goto error;
1075 
1076 	return 0;
1077 
1078 error:
1079 	NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1080 	return err;
1081 }
1082 
1083 static int esp_init_authenc(struct xfrm_state *x,
1084 			    struct netlink_ext_ack *extack)
1085 {
1086 	struct crypto_aead *aead;
1087 	struct crypto_authenc_key_param *param;
1088 	struct rtattr *rta;
1089 	char *key;
1090 	char *p;
1091 	char authenc_name[CRYPTO_MAX_ALG_NAME];
1092 	unsigned int keylen;
1093 	int err;
1094 
1095 	err = -ENAMETOOLONG;
1096 
1097 	if ((x->props.flags & XFRM_STATE_ESN)) {
1098 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1099 			     "%s%sauthencesn(%s,%s)%s",
1100 			     x->geniv ?: "", x->geniv ? "(" : "",
1101 			     x->aalg ? x->aalg->alg_name : "digest_null",
1102 			     x->ealg->alg_name,
1103 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1104 			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1105 			goto error;
1106 		}
1107 	} else {
1108 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1109 			     "%s%sauthenc(%s,%s)%s",
1110 			     x->geniv ?: "", x->geniv ? "(" : "",
1111 			     x->aalg ? x->aalg->alg_name : "digest_null",
1112 			     x->ealg->alg_name,
1113 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1114 			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1115 			goto error;
1116 		}
1117 	}
1118 
1119 	aead = crypto_alloc_aead(authenc_name, 0, 0);
1120 	err = PTR_ERR(aead);
1121 	if (IS_ERR(aead)) {
1122 		NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1123 		goto error;
1124 	}
1125 
1126 	x->data = aead;
1127 
1128 	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1129 		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1130 	err = -ENOMEM;
1131 	key = kmalloc(keylen, GFP_KERNEL);
1132 	if (!key)
1133 		goto error;
1134 
1135 	p = key;
1136 	rta = (void *)p;
1137 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1138 	rta->rta_len = RTA_LENGTH(sizeof(*param));
1139 	param = RTA_DATA(rta);
1140 	p += RTA_SPACE(sizeof(*param));
1141 
1142 	if (x->aalg) {
1143 		struct xfrm_algo_desc *aalg_desc;
1144 
1145 		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1146 		p += (x->aalg->alg_key_len + 7) / 8;
1147 
1148 		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1149 		BUG_ON(!aalg_desc);
1150 
1151 		err = -EINVAL;
1152 		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1153 		    crypto_aead_authsize(aead)) {
1154 			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1155 			goto free_key;
1156 		}
1157 
1158 		err = crypto_aead_setauthsize(
1159 			aead, x->aalg->alg_trunc_len / 8);
1160 		if (err) {
1161 			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1162 			goto free_key;
1163 		}
1164 	}
1165 
1166 	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1167 	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1168 
1169 	err = crypto_aead_setkey(aead, key, keylen);
1170 
1171 free_key:
1172 	kfree(key);
1173 
1174 error:
1175 	return err;
1176 }
1177 
1178 static int esp6_init_state(struct xfrm_state *x, struct netlink_ext_ack *extack)
1179 {
1180 	struct crypto_aead *aead;
1181 	u32 align;
1182 	int err;
1183 
1184 	x->data = NULL;
1185 
1186 	if (x->aead) {
1187 		err = esp_init_aead(x, extack);
1188 	} else if (x->ealg) {
1189 		err = esp_init_authenc(x, extack);
1190 	} else {
1191 		NL_SET_ERR_MSG(extack, "ESP: AEAD or CRYPT must be provided");
1192 		err = -EINVAL;
1193 	}
1194 
1195 	if (err)
1196 		goto error;
1197 
1198 	aead = x->data;
1199 
1200 	x->props.header_len = sizeof(struct ip_esp_hdr) +
1201 			      crypto_aead_ivsize(aead);
1202 	switch (x->props.mode) {
1203 	case XFRM_MODE_BEET:
1204 		if (x->sel.family != AF_INET6)
1205 			x->props.header_len += IPV4_BEET_PHMAXLEN +
1206 					       (sizeof(struct ipv6hdr) - sizeof(struct iphdr));
1207 		break;
1208 	default:
1209 	case XFRM_MODE_TRANSPORT:
1210 		break;
1211 	case XFRM_MODE_TUNNEL:
1212 		x->props.header_len += sizeof(struct ipv6hdr);
1213 		break;
1214 	}
1215 
1216 	if (x->encap) {
1217 		struct xfrm_encap_tmpl *encap = x->encap;
1218 
1219 		switch (encap->encap_type) {
1220 		default:
1221 			NL_SET_ERR_MSG(extack, "Unsupported encapsulation type for ESP");
1222 			err = -EINVAL;
1223 			goto error;
1224 		case UDP_ENCAP_ESPINUDP:
1225 			x->props.header_len += sizeof(struct udphdr);
1226 			break;
1227 #ifdef CONFIG_INET6_ESPINTCP
1228 		case TCP_ENCAP_ESPINTCP:
1229 			/* only the length field, TCP encap is done by
1230 			 * the socket
1231 			 */
1232 			x->props.header_len += 2;
1233 			break;
1234 #endif
1235 		}
1236 	}
1237 
1238 	align = ALIGN(crypto_aead_blocksize(aead), 4);
1239 	x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1240 
1241 error:
1242 	return err;
1243 }
1244 
1245 static int esp6_rcv_cb(struct sk_buff *skb, int err)
1246 {
1247 	return 0;
1248 }
1249 
1250 static const struct xfrm_type esp6_type = {
1251 	.owner		= THIS_MODULE,
1252 	.proto		= IPPROTO_ESP,
1253 	.flags		= XFRM_TYPE_REPLAY_PROT,
1254 	.init_state	= esp6_init_state,
1255 	.destructor	= esp6_destroy,
1256 	.input		= esp6_input,
1257 	.output		= esp6_output,
1258 };
1259 
1260 static struct xfrm6_protocol esp6_protocol = {
1261 	.handler	=	xfrm6_rcv,
1262 	.input_handler	=	xfrm_input,
1263 	.cb_handler	=	esp6_rcv_cb,
1264 	.err_handler	=	esp6_err,
1265 	.priority	=	0,
1266 };
1267 
1268 static int __init esp6_init(void)
1269 {
1270 	if (xfrm_register_type(&esp6_type, AF_INET6) < 0) {
1271 		pr_info("%s: can't add xfrm type\n", __func__);
1272 		return -EAGAIN;
1273 	}
1274 	if (xfrm6_protocol_register(&esp6_protocol, IPPROTO_ESP) < 0) {
1275 		pr_info("%s: can't add protocol\n", __func__);
1276 		xfrm_unregister_type(&esp6_type, AF_INET6);
1277 		return -EAGAIN;
1278 	}
1279 
1280 	return 0;
1281 }
1282 
1283 static void __exit esp6_fini(void)
1284 {
1285 	if (xfrm6_protocol_deregister(&esp6_protocol, IPPROTO_ESP) < 0)
1286 		pr_info("%s: can't remove protocol\n", __func__);
1287 	xfrm_unregister_type(&esp6_type, AF_INET6);
1288 }
1289 
1290 module_init(esp6_init);
1291 module_exit(esp6_fini);
1292 
1293 MODULE_DESCRIPTION("IPv6 ESP transformation helpers");
1294 MODULE_LICENSE("GPL");
1295 MODULE_ALIAS_XFRM_TYPE(AF_INET6, XFRM_PROTO_ESP);
1296