xref: /openbmc/linux/net/ipv6/esp6.c (revision 9f99d983) 
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 
40 #include <linux/highmem.h>
41 
42 struct esp_skb_cb {
43 	struct xfrm_skb_cb xfrm;
44 	void *tmp;
45 };
46 
47 struct esp_output_extra {
48 	__be32 seqhi;
49 	u32 esphoff;
50 };
51 
52 #define ESP_SKB_CB(__skb) ((struct esp_skb_cb *)&((__skb)->cb[0]))
53 
54 /*
55  * Allocate an AEAD request structure with extra space for SG and IV.
56  *
57  * For alignment considerations the upper 32 bits of the sequence number are
58  * placed at the front, if present. Followed by the IV, the request and finally
59  * the SG list.
60  *
61  * TODO: Use spare space in skb for this where possible.
62  */
63 static void *esp_alloc_tmp(struct crypto_aead *aead, int nfrags, int seqihlen)
64 {
65 	unsigned int len;
66 
67 	len = seqihlen;
68 
69 	len += crypto_aead_ivsize(aead);
70 
71 	if (len) {
72 		len += crypto_aead_alignmask(aead) &
73 		       ~(crypto_tfm_ctx_alignment() - 1);
74 		len = ALIGN(len, crypto_tfm_ctx_alignment());
75 	}
76 
77 	len += sizeof(struct aead_request) + crypto_aead_reqsize(aead);
78 	len = ALIGN(len, __alignof__(struct scatterlist));
79 
80 	len += sizeof(struct scatterlist) * nfrags;
81 
82 	return kmalloc(len, GFP_ATOMIC);
83 }
84 
85 static inline void *esp_tmp_extra(void *tmp)
86 {
87 	return PTR_ALIGN(tmp, __alignof__(struct esp_output_extra));
88 }
89 
90 static inline u8 *esp_tmp_iv(struct crypto_aead *aead, void *tmp, int seqhilen)
91 {
92 	return crypto_aead_ivsize(aead) ?
93 	       PTR_ALIGN((u8 *)tmp + seqhilen,
94 			 crypto_aead_alignmask(aead) + 1) : tmp + seqhilen;
95 }
96 
97 static inline struct aead_request *esp_tmp_req(struct crypto_aead *aead, u8 *iv)
98 {
99 	struct aead_request *req;
100 
101 	req = (void *)PTR_ALIGN(iv + crypto_aead_ivsize(aead),
102 				crypto_tfm_ctx_alignment());
103 	aead_request_set_tfm(req, aead);
104 	return req;
105 }
106 
107 static inline struct scatterlist *esp_req_sg(struct crypto_aead *aead,
108 					     struct aead_request *req)
109 {
110 	return (void *)ALIGN((unsigned long)(req + 1) +
111 			     crypto_aead_reqsize(aead),
112 			     __alignof__(struct scatterlist));
113 }
114 
115 static void esp_ssg_unref(struct xfrm_state *x, void *tmp)
116 {
117 	struct crypto_aead *aead = x->data;
118 	int extralen = 0;
119 	u8 *iv;
120 	struct aead_request *req;
121 	struct scatterlist *sg;
122 
123 	if (x->props.flags & XFRM_STATE_ESN)
124 		extralen += sizeof(struct esp_output_extra);
125 
126 	iv = esp_tmp_iv(aead, tmp, extralen);
127 	req = esp_tmp_req(aead, iv);
128 
129 	/* Unref skb_frag_pages in the src scatterlist if necessary.
130 	 * Skip the first sg which comes from skb->data.
131 	 */
132 	if (req->src != req->dst)
133 		for (sg = sg_next(req->src); sg; sg = sg_next(sg))
134 			put_page(sg_page(sg));
135 }
136 
137 #ifdef CONFIG_INET6_ESPINTCP
138 struct esp_tcp_sk {
139 	struct sock *sk;
140 	struct rcu_head rcu;
141 };
142 
143 static void esp_free_tcp_sk(struct rcu_head *head)
144 {
145 	struct esp_tcp_sk *esk = container_of(head, struct esp_tcp_sk, rcu);
146 
147 	sock_put(esk->sk);
148 	kfree(esk);
149 }
150 
151 static struct sock *esp6_find_tcp_sk(struct xfrm_state *x)
152 {
153 	struct xfrm_encap_tmpl *encap = x->encap;
154 	struct esp_tcp_sk *esk;
155 	__be16 sport, dport;
156 	struct sock *nsk;
157 	struct sock *sk;
158 
159 	sk = rcu_dereference(x->encap_sk);
160 	if (sk && sk->sk_state == TCP_ESTABLISHED)
161 		return sk;
162 
163 	spin_lock_bh(&x->lock);
164 	sport = encap->encap_sport;
165 	dport = encap->encap_dport;
166 	nsk = rcu_dereference_protected(x->encap_sk,
167 					lockdep_is_held(&x->lock));
168 	if (sk && sk == nsk) {
169 		esk = kmalloc(sizeof(*esk), GFP_ATOMIC);
170 		if (!esk) {
171 			spin_unlock_bh(&x->lock);
172 			return ERR_PTR(-ENOMEM);
173 		}
174 		RCU_INIT_POINTER(x->encap_sk, NULL);
175 		esk->sk = sk;
176 		call_rcu(&esk->rcu, esp_free_tcp_sk);
177 	}
178 	spin_unlock_bh(&x->lock);
179 
180 	sk = __inet6_lookup_established(xs_net(x), &tcp_hashinfo, &x->id.daddr.in6,
181 					dport, &x->props.saddr.in6, ntohs(sport), 0, 0);
182 	if (!sk)
183 		return ERR_PTR(-ENOENT);
184 
185 	if (!tcp_is_ulp_esp(sk)) {
186 		sock_put(sk);
187 		return ERR_PTR(-EINVAL);
188 	}
189 
190 	spin_lock_bh(&x->lock);
191 	nsk = rcu_dereference_protected(x->encap_sk,
192 					lockdep_is_held(&x->lock));
193 	if (encap->encap_sport != sport ||
194 	    encap->encap_dport != dport) {
195 		sock_put(sk);
196 		sk = nsk ?: ERR_PTR(-EREMCHG);
197 	} else if (sk == nsk) {
198 		sock_put(sk);
199 	} else {
200 		rcu_assign_pointer(x->encap_sk, sk);
201 	}
202 	spin_unlock_bh(&x->lock);
203 
204 	return sk;
205 }
206 
207 static int esp_output_tcp_finish(struct xfrm_state *x, struct sk_buff *skb)
208 {
209 	struct sock *sk;
210 	int err;
211 
212 	rcu_read_lock();
213 
214 	sk = esp6_find_tcp_sk(x);
215 	err = PTR_ERR_OR_ZERO(sk);
216 	if (err)
217 		goto out;
218 
219 	bh_lock_sock(sk);
220 	if (sock_owned_by_user(sk))
221 		err = espintcp_queue_out(sk, skb);
222 	else
223 		err = espintcp_push_skb(sk, skb);
224 	bh_unlock_sock(sk);
225 
226 out:
227 	rcu_read_unlock();
228 	return err;
229 }
230 
231 static int esp_output_tcp_encap_cb(struct net *net, struct sock *sk,
232 				   struct sk_buff *skb)
233 {
234 	struct dst_entry *dst = skb_dst(skb);
235 	struct xfrm_state *x = dst->xfrm;
236 
237 	return esp_output_tcp_finish(x, skb);
238 }
239 
240 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
241 {
242 	int err;
243 
244 	local_bh_disable();
245 	err = xfrm_trans_queue_net(xs_net(x), skb, esp_output_tcp_encap_cb);
246 	local_bh_enable();
247 
248 	/* EINPROGRESS just happens to do the right thing.  It
249 	 * actually means that the skb has been consumed and
250 	 * isn't coming back.
251 	 */
252 	return err ?: -EINPROGRESS;
253 }
254 #else
255 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
256 {
257 	kfree_skb(skb);
258 
259 	return -EOPNOTSUPP;
260 }
261 #endif
262 
263 static void esp_output_encap_csum(struct sk_buff *skb)
264 {
265 	/* UDP encap with IPv6 requires a valid checksum */
266 	if (*skb_mac_header(skb) == IPPROTO_UDP) {
267 		struct udphdr *uh = udp_hdr(skb);
268 		struct ipv6hdr *ip6h = ipv6_hdr(skb);
269 		int len = ntohs(uh->len);
270 		unsigned int offset = skb_transport_offset(skb);
271 		__wsum csum = skb_checksum(skb, offset, skb->len - offset, 0);
272 
273 		uh->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
274 					    len, IPPROTO_UDP, csum);
275 		if (uh->check == 0)
276 			uh->check = CSUM_MANGLED_0;
277 	}
278 }
279 
280 static void esp_output_done(struct crypto_async_request *base, int err)
281 {
282 	struct sk_buff *skb = base->data;
283 	struct xfrm_offload *xo = xfrm_offload(skb);
284 	void *tmp;
285 	struct xfrm_state *x;
286 
287 	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
288 		struct sec_path *sp = skb_sec_path(skb);
289 
290 		x = sp->xvec[sp->len - 1];
291 	} else {
292 		x = skb_dst(skb)->xfrm;
293 	}
294 
295 	tmp = ESP_SKB_CB(skb)->tmp;
296 	esp_ssg_unref(x, tmp);
297 	kfree(tmp);
298 
299 	esp_output_encap_csum(skb);
300 
301 	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
302 		if (err) {
303 			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
304 			kfree_skb(skb);
305 			return;
306 		}
307 
308 		skb_push(skb, skb->data - skb_mac_header(skb));
309 		secpath_reset(skb);
310 		xfrm_dev_resume(skb);
311 	} else {
312 		if (!err &&
313 		    x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
314 			esp_output_tail_tcp(x, skb);
315 		else
316 			xfrm_output_resume(skb->sk, skb, err);
317 	}
318 }
319 
320 /* Move ESP header back into place. */
321 static void esp_restore_header(struct sk_buff *skb, unsigned int offset)
322 {
323 	struct ip_esp_hdr *esph = (void *)(skb->data + offset);
324 	void *tmp = ESP_SKB_CB(skb)->tmp;
325 	__be32 *seqhi = esp_tmp_extra(tmp);
326 
327 	esph->seq_no = esph->spi;
328 	esph->spi = *seqhi;
329 }
330 
331 static void esp_output_restore_header(struct sk_buff *skb)
332 {
333 	void *tmp = ESP_SKB_CB(skb)->tmp;
334 	struct esp_output_extra *extra = esp_tmp_extra(tmp);
335 
336 	esp_restore_header(skb, skb_transport_offset(skb) + extra->esphoff -
337 				sizeof(__be32));
338 }
339 
340 static struct ip_esp_hdr *esp_output_set_esn(struct sk_buff *skb,
341 					     struct xfrm_state *x,
342 					     struct ip_esp_hdr *esph,
343 					     struct esp_output_extra *extra)
344 {
345 	/* For ESN we move the header forward by 4 bytes to
346 	 * accomodate the high bits.  We will move it back after
347 	 * encryption.
348 	 */
349 	if ((x->props.flags & XFRM_STATE_ESN)) {
350 		__u32 seqhi;
351 		struct xfrm_offload *xo = xfrm_offload(skb);
352 
353 		if (xo)
354 			seqhi = xo->seq.hi;
355 		else
356 			seqhi = XFRM_SKB_CB(skb)->seq.output.hi;
357 
358 		extra->esphoff = (unsigned char *)esph -
359 				 skb_transport_header(skb);
360 		esph = (struct ip_esp_hdr *)((unsigned char *)esph - 4);
361 		extra->seqhi = esph->spi;
362 		esph->seq_no = htonl(seqhi);
363 	}
364 
365 	esph->spi = x->id.spi;
366 
367 	return esph;
368 }
369 
370 static void esp_output_done_esn(struct crypto_async_request *base, int err)
371 {
372 	struct sk_buff *skb = base->data;
373 
374 	esp_output_restore_header(skb);
375 	esp_output_done(base, err);
376 }
377 
378 static struct ip_esp_hdr *esp6_output_udp_encap(struct sk_buff *skb,
379 					       int encap_type,
380 					       struct esp_info *esp,
381 					       __be16 sport,
382 					       __be16 dport)
383 {
384 	struct udphdr *uh;
385 	__be32 *udpdata32;
386 	unsigned int len;
387 
388 	len = skb->len + esp->tailen - skb_transport_offset(skb);
389 	if (len > U16_MAX)
390 		return ERR_PTR(-EMSGSIZE);
391 
392 	uh = (struct udphdr *)esp->esph;
393 	uh->source = sport;
394 	uh->dest = dport;
395 	uh->len = htons(len);
396 	uh->check = 0;
397 
398 	*skb_mac_header(skb) = IPPROTO_UDP;
399 
400 	if (encap_type == UDP_ENCAP_ESPINUDP_NON_IKE) {
401 		udpdata32 = (__be32 *)(uh + 1);
402 		udpdata32[0] = udpdata32[1] = 0;
403 		return (struct ip_esp_hdr *)(udpdata32 + 2);
404 	}
405 
406 	return (struct ip_esp_hdr *)(uh + 1);
407 }
408 
409 #ifdef CONFIG_INET6_ESPINTCP
410 static struct ip_esp_hdr *esp6_output_tcp_encap(struct xfrm_state *x,
411 						struct sk_buff *skb,
412 						struct esp_info *esp)
413 {
414 	__be16 *lenp = (void *)esp->esph;
415 	struct ip_esp_hdr *esph;
416 	unsigned int len;
417 	struct sock *sk;
418 
419 	len = skb->len + esp->tailen - skb_transport_offset(skb);
420 	if (len > IP_MAX_MTU)
421 		return ERR_PTR(-EMSGSIZE);
422 
423 	rcu_read_lock();
424 	sk = esp6_find_tcp_sk(x);
425 	rcu_read_unlock();
426 
427 	if (IS_ERR(sk))
428 		return ERR_CAST(sk);
429 
430 	*lenp = htons(len);
431 	esph = (struct ip_esp_hdr *)(lenp + 1);
432 
433 	return esph;
434 }
435 #else
436 static struct ip_esp_hdr *esp6_output_tcp_encap(struct xfrm_state *x,
437 						struct sk_buff *skb,
438 						struct esp_info *esp)
439 {
440 	return ERR_PTR(-EOPNOTSUPP);
441 }
442 #endif
443 
444 static int esp6_output_encap(struct xfrm_state *x, struct sk_buff *skb,
445 			    struct esp_info *esp)
446 {
447 	struct xfrm_encap_tmpl *encap = x->encap;
448 	struct ip_esp_hdr *esph;
449 	__be16 sport, dport;
450 	int encap_type;
451 
452 	spin_lock_bh(&x->lock);
453 	sport = encap->encap_sport;
454 	dport = encap->encap_dport;
455 	encap_type = encap->encap_type;
456 	spin_unlock_bh(&x->lock);
457 
458 	switch (encap_type) {
459 	default:
460 	case UDP_ENCAP_ESPINUDP:
461 	case UDP_ENCAP_ESPINUDP_NON_IKE:
462 		esph = esp6_output_udp_encap(skb, encap_type, esp, sport, dport);
463 		break;
464 	case TCP_ENCAP_ESPINTCP:
465 		esph = esp6_output_tcp_encap(x, skb, esp);
466 		break;
467 	}
468 
469 	if (IS_ERR(esph))
470 		return PTR_ERR(esph);
471 
472 	esp->esph = esph;
473 
474 	return 0;
475 }
476 
477 int esp6_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
478 {
479 	u8 *tail;
480 	int nfrags;
481 	int esph_offset;
482 	struct page *page;
483 	struct sk_buff *trailer;
484 	int tailen = esp->tailen;
485 
486 	if (x->encap) {
487 		int err = esp6_output_encap(x, skb, esp);
488 
489 		if (err < 0)
490 			return err;
491 	}
492 
493 	if (!skb_cloned(skb)) {
494 		if (tailen <= skb_tailroom(skb)) {
495 			nfrags = 1;
496 			trailer = skb;
497 			tail = skb_tail_pointer(trailer);
498 
499 			goto skip_cow;
500 		} else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
501 			   && !skb_has_frag_list(skb)) {
502 			int allocsize;
503 			struct sock *sk = skb->sk;
504 			struct page_frag *pfrag = &x->xfrag;
505 
506 			esp->inplace = false;
507 
508 			allocsize = ALIGN(tailen, L1_CACHE_BYTES);
509 
510 			spin_lock_bh(&x->lock);
511 
512 			if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
513 				spin_unlock_bh(&x->lock);
514 				goto cow;
515 			}
516 
517 			page = pfrag->page;
518 			get_page(page);
519 
520 			tail = page_address(page) + pfrag->offset;
521 
522 			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
523 
524 			nfrags = skb_shinfo(skb)->nr_frags;
525 
526 			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
527 					     tailen);
528 			skb_shinfo(skb)->nr_frags = ++nfrags;
529 
530 			pfrag->offset = pfrag->offset + allocsize;
531 
532 			spin_unlock_bh(&x->lock);
533 
534 			nfrags++;
535 
536 			skb->len += tailen;
537 			skb->data_len += tailen;
538 			skb->truesize += tailen;
539 			if (sk && sk_fullsock(sk))
540 				refcount_add(tailen, &sk->sk_wmem_alloc);
541 
542 			goto out;
543 		}
544 	}
545 
546 cow:
547 	esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
548 
549 	nfrags = skb_cow_data(skb, tailen, &trailer);
550 	if (nfrags < 0)
551 		goto out;
552 	tail = skb_tail_pointer(trailer);
553 	esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
554 
555 skip_cow:
556 	esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
557 	pskb_put(skb, trailer, tailen);
558 
559 out:
560 	return nfrags;
561 }
562 EXPORT_SYMBOL_GPL(esp6_output_head);
563 
564 int esp6_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
565 {
566 	u8 *iv;
567 	int alen;
568 	void *tmp;
569 	int ivlen;
570 	int assoclen;
571 	int extralen;
572 	struct page *page;
573 	struct ip_esp_hdr *esph;
574 	struct aead_request *req;
575 	struct crypto_aead *aead;
576 	struct scatterlist *sg, *dsg;
577 	struct esp_output_extra *extra;
578 	int err = -ENOMEM;
579 
580 	assoclen = sizeof(struct ip_esp_hdr);
581 	extralen = 0;
582 
583 	if (x->props.flags & XFRM_STATE_ESN) {
584 		extralen += sizeof(*extra);
585 		assoclen += sizeof(__be32);
586 	}
587 
588 	aead = x->data;
589 	alen = crypto_aead_authsize(aead);
590 	ivlen = crypto_aead_ivsize(aead);
591 
592 	tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
593 	if (!tmp)
594 		goto error;
595 
596 	extra = esp_tmp_extra(tmp);
597 	iv = esp_tmp_iv(aead, tmp, extralen);
598 	req = esp_tmp_req(aead, iv);
599 	sg = esp_req_sg(aead, req);
600 
601 	if (esp->inplace)
602 		dsg = sg;
603 	else
604 		dsg = &sg[esp->nfrags];
605 
606 	esph = esp_output_set_esn(skb, x, esp->esph, extra);
607 	esp->esph = esph;
608 
609 	sg_init_table(sg, esp->nfrags);
610 	err = skb_to_sgvec(skb, sg,
611 		           (unsigned char *)esph - skb->data,
612 		           assoclen + ivlen + esp->clen + alen);
613 	if (unlikely(err < 0))
614 		goto error_free;
615 
616 	if (!esp->inplace) {
617 		int allocsize;
618 		struct page_frag *pfrag = &x->xfrag;
619 
620 		allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
621 
622 		spin_lock_bh(&x->lock);
623 		if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
624 			spin_unlock_bh(&x->lock);
625 			goto error_free;
626 		}
627 
628 		skb_shinfo(skb)->nr_frags = 1;
629 
630 		page = pfrag->page;
631 		get_page(page);
632 		/* replace page frags in skb with new page */
633 		__skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
634 		pfrag->offset = pfrag->offset + allocsize;
635 		spin_unlock_bh(&x->lock);
636 
637 		sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
638 		err = skb_to_sgvec(skb, dsg,
639 			           (unsigned char *)esph - skb->data,
640 			           assoclen + ivlen + esp->clen + alen);
641 		if (unlikely(err < 0))
642 			goto error_free;
643 	}
644 
645 	if ((x->props.flags & XFRM_STATE_ESN))
646 		aead_request_set_callback(req, 0, esp_output_done_esn, skb);
647 	else
648 		aead_request_set_callback(req, 0, esp_output_done, skb);
649 
650 	aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
651 	aead_request_set_ad(req, assoclen);
652 
653 	memset(iv, 0, ivlen);
654 	memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
655 	       min(ivlen, 8));
656 
657 	ESP_SKB_CB(skb)->tmp = tmp;
658 	err = crypto_aead_encrypt(req);
659 
660 	switch (err) {
661 	case -EINPROGRESS:
662 		goto error;
663 
664 	case -ENOSPC:
665 		err = NET_XMIT_DROP;
666 		break;
667 
668 	case 0:
669 		if ((x->props.flags & XFRM_STATE_ESN))
670 			esp_output_restore_header(skb);
671 		esp_output_encap_csum(skb);
672 	}
673 
674 	if (sg != dsg)
675 		esp_ssg_unref(x, tmp);
676 
677 	if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
678 		err = esp_output_tail_tcp(x, skb);
679 
680 error_free:
681 	kfree(tmp);
682 error:
683 	return err;
684 }
685 EXPORT_SYMBOL_GPL(esp6_output_tail);
686 
687 static int esp6_output(struct xfrm_state *x, struct sk_buff *skb)
688 {
689 	int alen;
690 	int blksize;
691 	struct ip_esp_hdr *esph;
692 	struct crypto_aead *aead;
693 	struct esp_info esp;
694 
695 	esp.inplace = true;
696 
697 	esp.proto = *skb_mac_header(skb);
698 	*skb_mac_header(skb) = IPPROTO_ESP;
699 
700 	/* skb is pure payload to encrypt */
701 
702 	aead = x->data;
703 	alen = crypto_aead_authsize(aead);
704 
705 	esp.tfclen = 0;
706 	if (x->tfcpad) {
707 		struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
708 		u32 padto;
709 
710 		padto = min(x->tfcpad, __xfrm_state_mtu(x, dst->child_mtu_cached));
711 		if (skb->len < padto)
712 			esp.tfclen = padto - skb->len;
713 	}
714 	blksize = ALIGN(crypto_aead_blocksize(aead), 4);
715 	esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
716 	esp.plen = esp.clen - skb->len - esp.tfclen;
717 	esp.tailen = esp.tfclen + esp.plen + alen;
718 
719 	esp.esph = ip_esp_hdr(skb);
720 
721 	esp.nfrags = esp6_output_head(x, skb, &esp);
722 	if (esp.nfrags < 0)
723 		return esp.nfrags;
724 
725 	esph = esp.esph;
726 	esph->spi = x->id.spi;
727 
728 	esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
729 	esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
730 			    ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
731 
732 	skb_push(skb, -skb_network_offset(skb));
733 
734 	return esp6_output_tail(x, skb, &esp);
735 }
736 
737 static inline int esp_remove_trailer(struct sk_buff *skb)
738 {
739 	struct xfrm_state *x = xfrm_input_state(skb);
740 	struct xfrm_offload *xo = xfrm_offload(skb);
741 	struct crypto_aead *aead = x->data;
742 	int alen, hlen, elen;
743 	int padlen, trimlen;
744 	__wsum csumdiff;
745 	u8 nexthdr[2];
746 	int ret;
747 
748 	alen = crypto_aead_authsize(aead);
749 	hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
750 	elen = skb->len - hlen;
751 
752 	if (xo && (xo->flags & XFRM_ESP_NO_TRAILER)) {
753 		ret = xo->proto;
754 		goto out;
755 	}
756 
757 	ret = skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2);
758 	BUG_ON(ret);
759 
760 	ret = -EINVAL;
761 	padlen = nexthdr[0];
762 	if (padlen + 2 + alen >= elen) {
763 		net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
764 				    padlen + 2, elen - alen);
765 		goto out;
766 	}
767 
768 	trimlen = alen + padlen + 2;
769 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
770 		csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
771 		skb->csum = csum_block_sub(skb->csum, csumdiff,
772 					   skb->len - trimlen);
773 	}
774 	pskb_trim(skb, skb->len - trimlen);
775 
776 	ret = nexthdr[1];
777 
778 out:
779 	return ret;
780 }
781 
782 int esp6_input_done2(struct sk_buff *skb, int err)
783 {
784 	struct xfrm_state *x = xfrm_input_state(skb);
785 	struct xfrm_offload *xo = xfrm_offload(skb);
786 	struct crypto_aead *aead = x->data;
787 	int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
788 	int hdr_len = skb_network_header_len(skb);
789 
790 	if (!xo || !(xo->flags & CRYPTO_DONE))
791 		kfree(ESP_SKB_CB(skb)->tmp);
792 
793 	if (unlikely(err))
794 		goto out;
795 
796 	err = esp_remove_trailer(skb);
797 	if (unlikely(err < 0))
798 		goto out;
799 
800 	if (x->encap) {
801 		const struct ipv6hdr *ip6h = ipv6_hdr(skb);
802 		int offset = skb_network_offset(skb) + sizeof(*ip6h);
803 		struct xfrm_encap_tmpl *encap = x->encap;
804 		u8 nexthdr = ip6h->nexthdr;
805 		__be16 frag_off, source;
806 		struct udphdr *uh;
807 		struct tcphdr *th;
808 
809 		offset = ipv6_skip_exthdr(skb, offset, &nexthdr, &frag_off);
810 
811 		if (offset < 0) {
812 			err = -EINVAL;
813 			goto out;
814 		}
815 
816 		uh = (void *)(skb->data + offset);
817 		th = (void *)(skb->data + offset);
818 		hdr_len += offset;
819 
820 		switch (x->encap->encap_type) {
821 		case TCP_ENCAP_ESPINTCP:
822 			source = th->source;
823 			break;
824 		case UDP_ENCAP_ESPINUDP:
825 		case UDP_ENCAP_ESPINUDP_NON_IKE:
826 			source = uh->source;
827 			break;
828 		default:
829 			WARN_ON_ONCE(1);
830 			err = -EINVAL;
831 			goto out;
832 		}
833 
834 		/*
835 		 * 1) if the NAT-T peer's IP or port changed then
836 		 *    advertize the change to the keying daemon.
837 		 *    This is an inbound SA, so just compare
838 		 *    SRC ports.
839 		 */
840 		if (!ipv6_addr_equal(&ip6h->saddr, &x->props.saddr.in6) ||
841 		    source != encap->encap_sport) {
842 			xfrm_address_t ipaddr;
843 
844 			memcpy(&ipaddr.a6, &ip6h->saddr.s6_addr, sizeof(ipaddr.a6));
845 			km_new_mapping(x, &ipaddr, source);
846 
847 			/* XXX: perhaps add an extra
848 			 * policy check here, to see
849 			 * if we should allow or
850 			 * reject a packet from a
851 			 * different source
852 			 * address/port.
853 			 */
854 		}
855 
856 		/*
857 		 * 2) ignore UDP/TCP checksums in case
858 		 *    of NAT-T in Transport Mode, or
859 		 *    perform other post-processing fixes
860 		 *    as per draft-ietf-ipsec-udp-encaps-06,
861 		 *    section 3.1.2
862 		 */
863 		if (x->props.mode == XFRM_MODE_TRANSPORT)
864 			skb->ip_summed = CHECKSUM_UNNECESSARY;
865 	}
866 
867 	skb_postpull_rcsum(skb, skb_network_header(skb),
868 			   skb_network_header_len(skb));
869 	skb_pull_rcsum(skb, hlen);
870 	if (x->props.mode == XFRM_MODE_TUNNEL)
871 		skb_reset_transport_header(skb);
872 	else
873 		skb_set_transport_header(skb, -hdr_len);
874 
875 	/* RFC4303: Drop dummy packets without any error */
876 	if (err == IPPROTO_NONE)
877 		err = -EINVAL;
878 
879 out:
880 	return err;
881 }
882 EXPORT_SYMBOL_GPL(esp6_input_done2);
883 
884 static void esp_input_done(struct crypto_async_request *base, int err)
885 {
886 	struct sk_buff *skb = base->data;
887 
888 	xfrm_input_resume(skb, esp6_input_done2(skb, err));
889 }
890 
891 static void esp_input_restore_header(struct sk_buff *skb)
892 {
893 	esp_restore_header(skb, 0);
894 	__skb_pull(skb, 4);
895 }
896 
897 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
898 {
899 	struct xfrm_state *x = xfrm_input_state(skb);
900 
901 	/* For ESN we move the header forward by 4 bytes to
902 	 * accomodate the high bits.  We will move it back after
903 	 * decryption.
904 	 */
905 	if ((x->props.flags & XFRM_STATE_ESN)) {
906 		struct ip_esp_hdr *esph = skb_push(skb, 4);
907 
908 		*seqhi = esph->spi;
909 		esph->spi = esph->seq_no;
910 		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
911 	}
912 }
913 
914 static void esp_input_done_esn(struct crypto_async_request *base, int err)
915 {
916 	struct sk_buff *skb = base->data;
917 
918 	esp_input_restore_header(skb);
919 	esp_input_done(base, err);
920 }
921 
922 static int esp6_input(struct xfrm_state *x, struct sk_buff *skb)
923 {
924 	struct crypto_aead *aead = x->data;
925 	struct aead_request *req;
926 	struct sk_buff *trailer;
927 	int ivlen = crypto_aead_ivsize(aead);
928 	int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
929 	int nfrags;
930 	int assoclen;
931 	int seqhilen;
932 	int ret = 0;
933 	void *tmp;
934 	__be32 *seqhi;
935 	u8 *iv;
936 	struct scatterlist *sg;
937 
938 	if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen)) {
939 		ret = -EINVAL;
940 		goto out;
941 	}
942 
943 	if (elen <= 0) {
944 		ret = -EINVAL;
945 		goto out;
946 	}
947 
948 	assoclen = sizeof(struct ip_esp_hdr);
949 	seqhilen = 0;
950 
951 	if (x->props.flags & XFRM_STATE_ESN) {
952 		seqhilen += sizeof(__be32);
953 		assoclen += seqhilen;
954 	}
955 
956 	if (!skb_cloned(skb)) {
957 		if (!skb_is_nonlinear(skb)) {
958 			nfrags = 1;
959 
960 			goto skip_cow;
961 		} else if (!skb_has_frag_list(skb)) {
962 			nfrags = skb_shinfo(skb)->nr_frags;
963 			nfrags++;
964 
965 			goto skip_cow;
966 		}
967 	}
968 
969 	nfrags = skb_cow_data(skb, 0, &trailer);
970 	if (nfrags < 0) {
971 		ret = -EINVAL;
972 		goto out;
973 	}
974 
975 skip_cow:
976 	ret = -ENOMEM;
977 	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
978 	if (!tmp)
979 		goto out;
980 
981 	ESP_SKB_CB(skb)->tmp = tmp;
982 	seqhi = esp_tmp_extra(tmp);
983 	iv = esp_tmp_iv(aead, tmp, seqhilen);
984 	req = esp_tmp_req(aead, iv);
985 	sg = esp_req_sg(aead, req);
986 
987 	esp_input_set_header(skb, seqhi);
988 
989 	sg_init_table(sg, nfrags);
990 	ret = skb_to_sgvec(skb, sg, 0, skb->len);
991 	if (unlikely(ret < 0)) {
992 		kfree(tmp);
993 		goto out;
994 	}
995 
996 	skb->ip_summed = CHECKSUM_NONE;
997 
998 	if ((x->props.flags & XFRM_STATE_ESN))
999 		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
1000 	else
1001 		aead_request_set_callback(req, 0, esp_input_done, skb);
1002 
1003 	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
1004 	aead_request_set_ad(req, assoclen);
1005 
1006 	ret = crypto_aead_decrypt(req);
1007 	if (ret == -EINPROGRESS)
1008 		goto out;
1009 
1010 	if ((x->props.flags & XFRM_STATE_ESN))
1011 		esp_input_restore_header(skb);
1012 
1013 	ret = esp6_input_done2(skb, ret);
1014 
1015 out:
1016 	return ret;
1017 }
1018 
1019 static int esp6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
1020 		    u8 type, u8 code, int offset, __be32 info)
1021 {
1022 	struct net *net = dev_net(skb->dev);
1023 	const struct ipv6hdr *iph = (const struct ipv6hdr *)skb->data;
1024 	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data + offset);
1025 	struct xfrm_state *x;
1026 
1027 	if (type != ICMPV6_PKT_TOOBIG &&
1028 	    type != NDISC_REDIRECT)
1029 		return 0;
1030 
1031 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
1032 			      esph->spi, IPPROTO_ESP, AF_INET6);
1033 	if (!x)
1034 		return 0;
1035 
1036 	if (type == NDISC_REDIRECT)
1037 		ip6_redirect(skb, net, skb->dev->ifindex, 0,
1038 			     sock_net_uid(net, NULL));
1039 	else
1040 		ip6_update_pmtu(skb, net, info, 0, 0, sock_net_uid(net, NULL));
1041 	xfrm_state_put(x);
1042 
1043 	return 0;
1044 }
1045 
1046 static void esp6_destroy(struct xfrm_state *x)
1047 {
1048 	struct crypto_aead *aead = x->data;
1049 
1050 	if (!aead)
1051 		return;
1052 
1053 	crypto_free_aead(aead);
1054 }
1055 
1056 static int esp_init_aead(struct xfrm_state *x)
1057 {
1058 	char aead_name[CRYPTO_MAX_ALG_NAME];
1059 	struct crypto_aead *aead;
1060 	int err;
1061 
1062 	err = -ENAMETOOLONG;
1063 	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1064 		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME)
1065 		goto error;
1066 
1067 	aead = crypto_alloc_aead(aead_name, 0, 0);
1068 	err = PTR_ERR(aead);
1069 	if (IS_ERR(aead))
1070 		goto error;
1071 
1072 	x->data = aead;
1073 
1074 	err = crypto_aead_setkey(aead, x->aead->alg_key,
1075 				 (x->aead->alg_key_len + 7) / 8);
1076 	if (err)
1077 		goto error;
1078 
1079 	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1080 	if (err)
1081 		goto error;
1082 
1083 error:
1084 	return err;
1085 }
1086 
1087 static int esp_init_authenc(struct xfrm_state *x)
1088 {
1089 	struct crypto_aead *aead;
1090 	struct crypto_authenc_key_param *param;
1091 	struct rtattr *rta;
1092 	char *key;
1093 	char *p;
1094 	char authenc_name[CRYPTO_MAX_ALG_NAME];
1095 	unsigned int keylen;
1096 	int err;
1097 
1098 	err = -EINVAL;
1099 	if (!x->ealg)
1100 		goto error;
1101 
1102 	err = -ENAMETOOLONG;
1103 
1104 	if ((x->props.flags & XFRM_STATE_ESN)) {
1105 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1106 			     "%s%sauthencesn(%s,%s)%s",
1107 			     x->geniv ?: "", x->geniv ? "(" : "",
1108 			     x->aalg ? x->aalg->alg_name : "digest_null",
1109 			     x->ealg->alg_name,
1110 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1111 			goto error;
1112 	} else {
1113 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1114 			     "%s%sauthenc(%s,%s)%s",
1115 			     x->geniv ?: "", x->geniv ? "(" : "",
1116 			     x->aalg ? x->aalg->alg_name : "digest_null",
1117 			     x->ealg->alg_name,
1118 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1119 			goto error;
1120 	}
1121 
1122 	aead = crypto_alloc_aead(authenc_name, 0, 0);
1123 	err = PTR_ERR(aead);
1124 	if (IS_ERR(aead))
1125 		goto error;
1126 
1127 	x->data = aead;
1128 
1129 	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1130 		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1131 	err = -ENOMEM;
1132 	key = kmalloc(keylen, GFP_KERNEL);
1133 	if (!key)
1134 		goto error;
1135 
1136 	p = key;
1137 	rta = (void *)p;
1138 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1139 	rta->rta_len = RTA_LENGTH(sizeof(*param));
1140 	param = RTA_DATA(rta);
1141 	p += RTA_SPACE(sizeof(*param));
1142 
1143 	if (x->aalg) {
1144 		struct xfrm_algo_desc *aalg_desc;
1145 
1146 		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1147 		p += (x->aalg->alg_key_len + 7) / 8;
1148 
1149 		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1150 		BUG_ON(!aalg_desc);
1151 
1152 		err = -EINVAL;
1153 		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1154 		    crypto_aead_authsize(aead)) {
1155 			pr_info("ESP: %s digestsize %u != %u\n",
1156 				x->aalg->alg_name,
1157 				crypto_aead_authsize(aead),
1158 				aalg_desc->uinfo.auth.icv_fullbits / 8);
1159 			goto free_key;
1160 		}
1161 
1162 		err = crypto_aead_setauthsize(
1163 			aead, x->aalg->alg_trunc_len / 8);
1164 		if (err)
1165 			goto free_key;
1166 	}
1167 
1168 	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1169 	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1170 
1171 	err = crypto_aead_setkey(aead, key, keylen);
1172 
1173 free_key:
1174 	kfree(key);
1175 
1176 error:
1177 	return err;
1178 }
1179 
1180 static int esp6_init_state(struct xfrm_state *x)
1181 {
1182 	struct crypto_aead *aead;
1183 	u32 align;
1184 	int err;
1185 
1186 	x->data = NULL;
1187 
1188 	if (x->aead)
1189 		err = esp_init_aead(x);
1190 	else
1191 		err = esp_init_authenc(x);
1192 
1193 	if (err)
1194 		goto error;
1195 
1196 	aead = x->data;
1197 
1198 	x->props.header_len = sizeof(struct ip_esp_hdr) +
1199 			      crypto_aead_ivsize(aead);
1200 	switch (x->props.mode) {
1201 	case XFRM_MODE_BEET:
1202 		if (x->sel.family != AF_INET6)
1203 			x->props.header_len += IPV4_BEET_PHMAXLEN +
1204 					       (sizeof(struct ipv6hdr) - sizeof(struct iphdr));
1205 		break;
1206 	default:
1207 	case XFRM_MODE_TRANSPORT:
1208 		break;
1209 	case XFRM_MODE_TUNNEL:
1210 		x->props.header_len += sizeof(struct ipv6hdr);
1211 		break;
1212 	}
1213 
1214 	if (x->encap) {
1215 		struct xfrm_encap_tmpl *encap = x->encap;
1216 
1217 		switch (encap->encap_type) {
1218 		default:
1219 			err = -EINVAL;
1220 			goto error;
1221 		case UDP_ENCAP_ESPINUDP:
1222 			x->props.header_len += sizeof(struct udphdr);
1223 			break;
1224 		case UDP_ENCAP_ESPINUDP_NON_IKE:
1225 			x->props.header_len += sizeof(struct udphdr) + 2 * sizeof(u32);
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_LICENSE("GPL");
1294 MODULE_ALIAS_XFRM_TYPE(AF_INET6, XFRM_PROTO_ESP);
1295