xref: /linux/net/ipv6/esp6.c (revision 042db639bf33aee118846a87223acebe6700137d)
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 	__be32 *udpdata32;
388 	unsigned int len;
389 
390 	len = skb->len + esp->tailen - skb_transport_offset(skb);
391 	if (len > U16_MAX)
392 		return ERR_PTR(-EMSGSIZE);
393 
394 	uh = (struct udphdr *)esp->esph;
395 	uh->source = sport;
396 	uh->dest = dport;
397 	uh->len = htons(len);
398 	uh->check = 0;
399 
400 	*skb_mac_header(skb) = IPPROTO_UDP;
401 
402 	if (encap_type == UDP_ENCAP_ESPINUDP_NON_IKE) {
403 		udpdata32 = (__be32 *)(uh + 1);
404 		udpdata32[0] = udpdata32[1] = 0;
405 		return (struct ip_esp_hdr *)(udpdata32 + 2);
406 	}
407 
408 	return (struct ip_esp_hdr *)(uh + 1);
409 }
410 
411 #ifdef CONFIG_INET6_ESPINTCP
412 static struct ip_esp_hdr *esp6_output_tcp_encap(struct xfrm_state *x,
413 						struct sk_buff *skb,
414 						struct esp_info *esp)
415 {
416 	__be16 *lenp = (void *)esp->esph;
417 	struct ip_esp_hdr *esph;
418 	unsigned int len;
419 	struct sock *sk;
420 
421 	len = skb->len + esp->tailen - skb_transport_offset(skb);
422 	if (len > IP_MAX_MTU)
423 		return ERR_PTR(-EMSGSIZE);
424 
425 	rcu_read_lock();
426 	sk = esp6_find_tcp_sk(x);
427 	rcu_read_unlock();
428 
429 	if (IS_ERR(sk))
430 		return ERR_CAST(sk);
431 
432 	*lenp = htons(len);
433 	esph = (struct ip_esp_hdr *)(lenp + 1);
434 
435 	return esph;
436 }
437 #else
438 static struct ip_esp_hdr *esp6_output_tcp_encap(struct xfrm_state *x,
439 						struct sk_buff *skb,
440 						struct esp_info *esp)
441 {
442 	return ERR_PTR(-EOPNOTSUPP);
443 }
444 #endif
445 
446 static int esp6_output_encap(struct xfrm_state *x, struct sk_buff *skb,
447 			    struct esp_info *esp)
448 {
449 	struct xfrm_encap_tmpl *encap = x->encap;
450 	struct ip_esp_hdr *esph;
451 	__be16 sport, dport;
452 	int encap_type;
453 
454 	spin_lock_bh(&x->lock);
455 	sport = encap->encap_sport;
456 	dport = encap->encap_dport;
457 	encap_type = encap->encap_type;
458 	spin_unlock_bh(&x->lock);
459 
460 	switch (encap_type) {
461 	default:
462 	case UDP_ENCAP_ESPINUDP:
463 	case UDP_ENCAP_ESPINUDP_NON_IKE:
464 		esph = esp6_output_udp_encap(skb, encap_type, esp, sport, dport);
465 		break;
466 	case TCP_ENCAP_ESPINTCP:
467 		esph = esp6_output_tcp_encap(x, skb, esp);
468 		break;
469 	}
470 
471 	if (IS_ERR(esph))
472 		return PTR_ERR(esph);
473 
474 	esp->esph = esph;
475 
476 	return 0;
477 }
478 
479 int esp6_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
480 {
481 	u8 *tail;
482 	int nfrags;
483 	int esph_offset;
484 	struct page *page;
485 	struct sk_buff *trailer;
486 	int tailen = esp->tailen;
487 
488 	if (x->encap) {
489 		int err = esp6_output_encap(x, skb, esp);
490 
491 		if (err < 0)
492 			return err;
493 	}
494 
495 	if (ALIGN(tailen, L1_CACHE_BYTES) > PAGE_SIZE ||
496 	    ALIGN(skb->data_len, L1_CACHE_BYTES) > PAGE_SIZE)
497 		goto cow;
498 
499 	if (!skb_cloned(skb)) {
500 		if (tailen <= skb_tailroom(skb)) {
501 			nfrags = 1;
502 			trailer = skb;
503 			tail = skb_tail_pointer(trailer);
504 
505 			goto skip_cow;
506 		} else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
507 			   && !skb_has_frag_list(skb)) {
508 			int allocsize;
509 			struct sock *sk = skb->sk;
510 			struct page_frag *pfrag = &x->xfrag;
511 
512 			esp->inplace = false;
513 
514 			allocsize = ALIGN(tailen, L1_CACHE_BYTES);
515 
516 			spin_lock_bh(&x->lock);
517 
518 			if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
519 				spin_unlock_bh(&x->lock);
520 				goto cow;
521 			}
522 
523 			page = pfrag->page;
524 			get_page(page);
525 
526 			tail = page_address(page) + pfrag->offset;
527 
528 			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
529 
530 			nfrags = skb_shinfo(skb)->nr_frags;
531 
532 			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
533 					     tailen);
534 			skb_shinfo(skb)->nr_frags = ++nfrags;
535 
536 			pfrag->offset = pfrag->offset + allocsize;
537 
538 			spin_unlock_bh(&x->lock);
539 
540 			nfrags++;
541 
542 			skb->len += tailen;
543 			skb->data_len += tailen;
544 			skb->truesize += tailen;
545 			if (sk && sk_fullsock(sk))
546 				refcount_add(tailen, &sk->sk_wmem_alloc);
547 
548 			goto out;
549 		}
550 	}
551 
552 cow:
553 	esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
554 
555 	nfrags = skb_cow_data(skb, tailen, &trailer);
556 	if (nfrags < 0)
557 		goto out;
558 	tail = skb_tail_pointer(trailer);
559 	esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
560 
561 skip_cow:
562 	esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
563 	pskb_put(skb, trailer, tailen);
564 
565 out:
566 	return nfrags;
567 }
568 EXPORT_SYMBOL_GPL(esp6_output_head);
569 
570 int esp6_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
571 {
572 	u8 *iv;
573 	int alen;
574 	void *tmp;
575 	int ivlen;
576 	int assoclen;
577 	int extralen;
578 	struct page *page;
579 	struct ip_esp_hdr *esph;
580 	struct aead_request *req;
581 	struct crypto_aead *aead;
582 	struct scatterlist *sg, *dsg;
583 	struct esp_output_extra *extra;
584 	int err = -ENOMEM;
585 
586 	assoclen = sizeof(struct ip_esp_hdr);
587 	extralen = 0;
588 
589 	if (x->props.flags & XFRM_STATE_ESN) {
590 		extralen += sizeof(*extra);
591 		assoclen += sizeof(__be32);
592 	}
593 
594 	aead = x->data;
595 	alen = crypto_aead_authsize(aead);
596 	ivlen = crypto_aead_ivsize(aead);
597 
598 	tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
599 	if (!tmp)
600 		goto error;
601 
602 	extra = esp_tmp_extra(tmp);
603 	iv = esp_tmp_iv(aead, tmp, extralen);
604 	req = esp_tmp_req(aead, iv);
605 	sg = esp_req_sg(aead, req);
606 
607 	if (esp->inplace)
608 		dsg = sg;
609 	else
610 		dsg = &sg[esp->nfrags];
611 
612 	esph = esp_output_set_esn(skb, x, esp->esph, extra);
613 	esp->esph = esph;
614 
615 	sg_init_table(sg, esp->nfrags);
616 	err = skb_to_sgvec(skb, sg,
617 		           (unsigned char *)esph - skb->data,
618 		           assoclen + ivlen + esp->clen + alen);
619 	if (unlikely(err < 0))
620 		goto error_free;
621 
622 	if (!esp->inplace) {
623 		int allocsize;
624 		struct page_frag *pfrag = &x->xfrag;
625 
626 		allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
627 
628 		spin_lock_bh(&x->lock);
629 		if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
630 			spin_unlock_bh(&x->lock);
631 			goto error_free;
632 		}
633 
634 		skb_shinfo(skb)->nr_frags = 1;
635 
636 		page = pfrag->page;
637 		get_page(page);
638 		/* replace page frags in skb with new page */
639 		__skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
640 		pfrag->offset = pfrag->offset + allocsize;
641 		spin_unlock_bh(&x->lock);
642 
643 		sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
644 		err = skb_to_sgvec(skb, dsg,
645 			           (unsigned char *)esph - skb->data,
646 			           assoclen + ivlen + esp->clen + alen);
647 		if (unlikely(err < 0))
648 			goto error_free;
649 	}
650 
651 	if ((x->props.flags & XFRM_STATE_ESN))
652 		aead_request_set_callback(req, 0, esp_output_done_esn, skb);
653 	else
654 		aead_request_set_callback(req, 0, esp_output_done, skb);
655 
656 	aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
657 	aead_request_set_ad(req, assoclen);
658 
659 	memset(iv, 0, ivlen);
660 	memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
661 	       min(ivlen, 8));
662 
663 	ESP_SKB_CB(skb)->tmp = tmp;
664 	err = crypto_aead_encrypt(req);
665 
666 	switch (err) {
667 	case -EINPROGRESS:
668 		goto error;
669 
670 	case -ENOSPC:
671 		err = NET_XMIT_DROP;
672 		break;
673 
674 	case 0:
675 		if ((x->props.flags & XFRM_STATE_ESN))
676 			esp_output_restore_header(skb);
677 		esp_output_encap_csum(skb);
678 	}
679 
680 	if (sg != dsg)
681 		esp_ssg_unref(x, tmp, skb);
682 
683 	if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
684 		err = esp_output_tail_tcp(x, skb);
685 
686 error_free:
687 	kfree(tmp);
688 error:
689 	return err;
690 }
691 EXPORT_SYMBOL_GPL(esp6_output_tail);
692 
693 static int esp6_output(struct xfrm_state *x, struct sk_buff *skb)
694 {
695 	int alen;
696 	int blksize;
697 	struct ip_esp_hdr *esph;
698 	struct crypto_aead *aead;
699 	struct esp_info esp;
700 
701 	esp.inplace = true;
702 
703 	esp.proto = *skb_mac_header(skb);
704 	*skb_mac_header(skb) = IPPROTO_ESP;
705 
706 	/* skb is pure payload to encrypt */
707 
708 	aead = x->data;
709 	alen = crypto_aead_authsize(aead);
710 
711 	esp.tfclen = 0;
712 	if (x->tfcpad) {
713 		struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
714 		u32 padto;
715 
716 		padto = min(x->tfcpad, xfrm_state_mtu(x, dst->child_mtu_cached));
717 		if (skb->len < padto)
718 			esp.tfclen = padto - skb->len;
719 	}
720 	blksize = ALIGN(crypto_aead_blocksize(aead), 4);
721 	esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
722 	esp.plen = esp.clen - skb->len - esp.tfclen;
723 	esp.tailen = esp.tfclen + esp.plen + alen;
724 
725 	esp.esph = ip_esp_hdr(skb);
726 
727 	esp.nfrags = esp6_output_head(x, skb, &esp);
728 	if (esp.nfrags < 0)
729 		return esp.nfrags;
730 
731 	esph = esp.esph;
732 	esph->spi = x->id.spi;
733 
734 	esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
735 	esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
736 			    ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
737 
738 	skb_push(skb, -skb_network_offset(skb));
739 
740 	return esp6_output_tail(x, skb, &esp);
741 }
742 
743 static inline int esp_remove_trailer(struct sk_buff *skb)
744 {
745 	struct xfrm_state *x = xfrm_input_state(skb);
746 	struct crypto_aead *aead = x->data;
747 	int alen, hlen, elen;
748 	int padlen, trimlen;
749 	__wsum csumdiff;
750 	u8 nexthdr[2];
751 	int ret;
752 
753 	alen = crypto_aead_authsize(aead);
754 	hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
755 	elen = skb->len - hlen;
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 	ret = pskb_trim(skb, skb->len - trimlen);
775 	if (unlikely(ret))
776 		return ret;
777 
778 	ret = nexthdr[1];
779 
780 out:
781 	return ret;
782 }
783 
784 int esp6_input_done2(struct sk_buff *skb, int err)
785 {
786 	struct xfrm_state *x = xfrm_input_state(skb);
787 	struct xfrm_offload *xo = xfrm_offload(skb);
788 	struct crypto_aead *aead = x->data;
789 	int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
790 	int hdr_len = skb_network_header_len(skb);
791 
792 	if (!xo || !(xo->flags & CRYPTO_DONE))
793 		kfree(ESP_SKB_CB(skb)->tmp);
794 
795 	if (unlikely(err))
796 		goto out;
797 
798 	err = esp_remove_trailer(skb);
799 	if (unlikely(err < 0))
800 		goto out;
801 
802 	if (x->encap) {
803 		const struct ipv6hdr *ip6h = ipv6_hdr(skb);
804 		int offset = skb_network_offset(skb) + sizeof(*ip6h);
805 		struct xfrm_encap_tmpl *encap = x->encap;
806 		u8 nexthdr = ip6h->nexthdr;
807 		__be16 frag_off, source;
808 		struct udphdr *uh;
809 		struct tcphdr *th;
810 
811 		offset = ipv6_skip_exthdr(skb, offset, &nexthdr, &frag_off);
812 		if (offset == -1) {
813 			err = -EINVAL;
814 			goto out;
815 		}
816 
817 		uh = (void *)(skb->data + offset);
818 		th = (void *)(skb->data + offset);
819 		hdr_len += offset;
820 
821 		switch (x->encap->encap_type) {
822 		case TCP_ENCAP_ESPINTCP:
823 			source = th->source;
824 			break;
825 		case UDP_ENCAP_ESPINUDP:
826 		case UDP_ENCAP_ESPINUDP_NON_IKE:
827 			source = uh->source;
828 			break;
829 		default:
830 			WARN_ON_ONCE(1);
831 			err = -EINVAL;
832 			goto out;
833 		}
834 
835 		/*
836 		 * 1) if the NAT-T peer's IP or port changed then
837 		 *    advertise the change to the keying daemon.
838 		 *    This is an inbound SA, so just compare
839 		 *    SRC ports.
840 		 */
841 		if (!ipv6_addr_equal(&ip6h->saddr, &x->props.saddr.in6) ||
842 		    source != encap->encap_sport) {
843 			xfrm_address_t ipaddr;
844 
845 			memcpy(&ipaddr.a6, &ip6h->saddr.s6_addr, sizeof(ipaddr.a6));
846 			km_new_mapping(x, &ipaddr, source);
847 
848 			/* XXX: perhaps add an extra
849 			 * policy check here, to see
850 			 * if we should allow or
851 			 * reject a packet from a
852 			 * different source
853 			 * address/port.
854 			 */
855 		}
856 
857 		/*
858 		 * 2) ignore UDP/TCP checksums in case
859 		 *    of NAT-T in Transport Mode, or
860 		 *    perform other post-processing fixes
861 		 *    as per draft-ietf-ipsec-udp-encaps-06,
862 		 *    section 3.1.2
863 		 */
864 		if (x->props.mode == XFRM_MODE_TRANSPORT)
865 			skb->ip_summed = CHECKSUM_UNNECESSARY;
866 	}
867 
868 	skb_postpull_rcsum(skb, skb_network_header(skb),
869 			   skb_network_header_len(skb));
870 	skb_pull_rcsum(skb, hlen);
871 	if (x->props.mode == XFRM_MODE_TUNNEL)
872 		skb_reset_transport_header(skb);
873 	else
874 		skb_set_transport_header(skb, -hdr_len);
875 
876 	/* RFC4303: Drop dummy packets without any error */
877 	if (err == IPPROTO_NONE)
878 		err = -EINVAL;
879 
880 out:
881 	return err;
882 }
883 EXPORT_SYMBOL_GPL(esp6_input_done2);
884 
885 static void esp_input_done(void *data, int err)
886 {
887 	struct sk_buff *skb = data;
888 
889 	xfrm_input_resume(skb, esp6_input_done2(skb, err));
890 }
891 
892 static void esp_input_restore_header(struct sk_buff *skb)
893 {
894 	esp_restore_header(skb, 0);
895 	__skb_pull(skb, 4);
896 }
897 
898 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
899 {
900 	struct xfrm_state *x = xfrm_input_state(skb);
901 
902 	/* For ESN we move the header forward by 4 bytes to
903 	 * accommodate the high bits.  We will move it back after
904 	 * decryption.
905 	 */
906 	if ((x->props.flags & XFRM_STATE_ESN)) {
907 		struct ip_esp_hdr *esph = skb_push(skb, 4);
908 
909 		*seqhi = esph->spi;
910 		esph->spi = esph->seq_no;
911 		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
912 	}
913 }
914 
915 static void esp_input_done_esn(void *data, int err)
916 {
917 	struct sk_buff *skb = data;
918 
919 	esp_input_restore_header(skb);
920 	esp_input_done(data, err);
921 }
922 
923 static int esp6_input(struct xfrm_state *x, struct sk_buff *skb)
924 {
925 	struct crypto_aead *aead = x->data;
926 	struct aead_request *req;
927 	struct sk_buff *trailer;
928 	int ivlen = crypto_aead_ivsize(aead);
929 	int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
930 	int nfrags;
931 	int assoclen;
932 	int seqhilen;
933 	int ret = 0;
934 	void *tmp;
935 	__be32 *seqhi;
936 	u8 *iv;
937 	struct scatterlist *sg;
938 
939 	if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen)) {
940 		ret = -EINVAL;
941 		goto out;
942 	}
943 
944 	if (elen <= 0) {
945 		ret = -EINVAL;
946 		goto out;
947 	}
948 
949 	assoclen = sizeof(struct ip_esp_hdr);
950 	seqhilen = 0;
951 
952 	if (x->props.flags & XFRM_STATE_ESN) {
953 		seqhilen += sizeof(__be32);
954 		assoclen += seqhilen;
955 	}
956 
957 	if (!skb_cloned(skb)) {
958 		if (!skb_is_nonlinear(skb)) {
959 			nfrags = 1;
960 
961 			goto skip_cow;
962 		} else if (!skb_has_frag_list(skb)) {
963 			nfrags = skb_shinfo(skb)->nr_frags;
964 			nfrags++;
965 
966 			goto skip_cow;
967 		}
968 	}
969 
970 	nfrags = skb_cow_data(skb, 0, &trailer);
971 	if (nfrags < 0) {
972 		ret = -EINVAL;
973 		goto out;
974 	}
975 
976 skip_cow:
977 	ret = -ENOMEM;
978 	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
979 	if (!tmp)
980 		goto out;
981 
982 	ESP_SKB_CB(skb)->tmp = tmp;
983 	seqhi = esp_tmp_extra(tmp);
984 	iv = esp_tmp_iv(aead, tmp, seqhilen);
985 	req = esp_tmp_req(aead, iv);
986 	sg = esp_req_sg(aead, req);
987 
988 	esp_input_set_header(skb, seqhi);
989 
990 	sg_init_table(sg, nfrags);
991 	ret = skb_to_sgvec(skb, sg, 0, skb->len);
992 	if (unlikely(ret < 0)) {
993 		kfree(tmp);
994 		goto out;
995 	}
996 
997 	skb->ip_summed = CHECKSUM_NONE;
998 
999 	if ((x->props.flags & XFRM_STATE_ESN))
1000 		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
1001 	else
1002 		aead_request_set_callback(req, 0, esp_input_done, skb);
1003 
1004 	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
1005 	aead_request_set_ad(req, assoclen);
1006 
1007 	ret = crypto_aead_decrypt(req);
1008 	if (ret == -EINPROGRESS)
1009 		goto out;
1010 
1011 	if ((x->props.flags & XFRM_STATE_ESN))
1012 		esp_input_restore_header(skb);
1013 
1014 	ret = esp6_input_done2(skb, ret);
1015 
1016 out:
1017 	return ret;
1018 }
1019 
1020 static int esp6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
1021 		    u8 type, u8 code, int offset, __be32 info)
1022 {
1023 	struct net *net = dev_net(skb->dev);
1024 	const struct ipv6hdr *iph = (const struct ipv6hdr *)skb->data;
1025 	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data + offset);
1026 	struct xfrm_state *x;
1027 
1028 	if (type != ICMPV6_PKT_TOOBIG &&
1029 	    type != NDISC_REDIRECT)
1030 		return 0;
1031 
1032 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
1033 			      esph->spi, IPPROTO_ESP, AF_INET6);
1034 	if (!x)
1035 		return 0;
1036 
1037 	if (type == NDISC_REDIRECT)
1038 		ip6_redirect(skb, net, skb->dev->ifindex, 0,
1039 			     sock_net_uid(net, NULL));
1040 	else
1041 		ip6_update_pmtu(skb, net, info, 0, 0, sock_net_uid(net, NULL));
1042 	xfrm_state_put(x);
1043 
1044 	return 0;
1045 }
1046 
1047 static void esp6_destroy(struct xfrm_state *x)
1048 {
1049 	struct crypto_aead *aead = x->data;
1050 
1051 	if (!aead)
1052 		return;
1053 
1054 	crypto_free_aead(aead);
1055 }
1056 
1057 static int esp_init_aead(struct xfrm_state *x, struct netlink_ext_ack *extack)
1058 {
1059 	char aead_name[CRYPTO_MAX_ALG_NAME];
1060 	struct crypto_aead *aead;
1061 	int err;
1062 
1063 	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1064 		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME) {
1065 		NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1066 		return -ENAMETOOLONG;
1067 	}
1068 
1069 	aead = crypto_alloc_aead(aead_name, 0, 0);
1070 	err = PTR_ERR(aead);
1071 	if (IS_ERR(aead))
1072 		goto error;
1073 
1074 	x->data = aead;
1075 
1076 	err = crypto_aead_setkey(aead, x->aead->alg_key,
1077 				 (x->aead->alg_key_len + 7) / 8);
1078 	if (err)
1079 		goto error;
1080 
1081 	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1082 	if (err)
1083 		goto error;
1084 
1085 	return 0;
1086 
1087 error:
1088 	NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1089 	return err;
1090 }
1091 
1092 static int esp_init_authenc(struct xfrm_state *x,
1093 			    struct netlink_ext_ack *extack)
1094 {
1095 	struct crypto_aead *aead;
1096 	struct crypto_authenc_key_param *param;
1097 	struct rtattr *rta;
1098 	char *key;
1099 	char *p;
1100 	char authenc_name[CRYPTO_MAX_ALG_NAME];
1101 	unsigned int keylen;
1102 	int err;
1103 
1104 	err = -ENAMETOOLONG;
1105 
1106 	if ((x->props.flags & XFRM_STATE_ESN)) {
1107 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1108 			     "%s%sauthencesn(%s,%s)%s",
1109 			     x->geniv ?: "", x->geniv ? "(" : "",
1110 			     x->aalg ? x->aalg->alg_name : "digest_null",
1111 			     x->ealg->alg_name,
1112 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1113 			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1114 			goto error;
1115 		}
1116 	} else {
1117 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1118 			     "%s%sauthenc(%s,%s)%s",
1119 			     x->geniv ?: "", x->geniv ? "(" : "",
1120 			     x->aalg ? x->aalg->alg_name : "digest_null",
1121 			     x->ealg->alg_name,
1122 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1123 			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1124 			goto error;
1125 		}
1126 	}
1127 
1128 	aead = crypto_alloc_aead(authenc_name, 0, 0);
1129 	err = PTR_ERR(aead);
1130 	if (IS_ERR(aead)) {
1131 		NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1132 		goto error;
1133 	}
1134 
1135 	x->data = aead;
1136 
1137 	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1138 		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1139 	err = -ENOMEM;
1140 	key = kmalloc(keylen, GFP_KERNEL);
1141 	if (!key)
1142 		goto error;
1143 
1144 	p = key;
1145 	rta = (void *)p;
1146 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1147 	rta->rta_len = RTA_LENGTH(sizeof(*param));
1148 	param = RTA_DATA(rta);
1149 	p += RTA_SPACE(sizeof(*param));
1150 
1151 	if (x->aalg) {
1152 		struct xfrm_algo_desc *aalg_desc;
1153 
1154 		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1155 		p += (x->aalg->alg_key_len + 7) / 8;
1156 
1157 		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1158 		BUG_ON(!aalg_desc);
1159 
1160 		err = -EINVAL;
1161 		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1162 		    crypto_aead_authsize(aead)) {
1163 			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1164 			goto free_key;
1165 		}
1166 
1167 		err = crypto_aead_setauthsize(
1168 			aead, x->aalg->alg_trunc_len / 8);
1169 		if (err) {
1170 			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1171 			goto free_key;
1172 		}
1173 	}
1174 
1175 	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1176 	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1177 
1178 	err = crypto_aead_setkey(aead, key, keylen);
1179 
1180 free_key:
1181 	kfree(key);
1182 
1183 error:
1184 	return err;
1185 }
1186 
1187 static int esp6_init_state(struct xfrm_state *x, struct netlink_ext_ack *extack)
1188 {
1189 	struct crypto_aead *aead;
1190 	u32 align;
1191 	int err;
1192 
1193 	x->data = NULL;
1194 
1195 	if (x->aead) {
1196 		err = esp_init_aead(x, extack);
1197 	} else if (x->ealg) {
1198 		err = esp_init_authenc(x, extack);
1199 	} else {
1200 		NL_SET_ERR_MSG(extack, "ESP: AEAD or CRYPT must be provided");
1201 		err = -EINVAL;
1202 	}
1203 
1204 	if (err)
1205 		goto error;
1206 
1207 	aead = x->data;
1208 
1209 	x->props.header_len = sizeof(struct ip_esp_hdr) +
1210 			      crypto_aead_ivsize(aead);
1211 	switch (x->props.mode) {
1212 	case XFRM_MODE_BEET:
1213 		if (x->sel.family != AF_INET6)
1214 			x->props.header_len += IPV4_BEET_PHMAXLEN +
1215 					       (sizeof(struct ipv6hdr) - sizeof(struct iphdr));
1216 		break;
1217 	default:
1218 	case XFRM_MODE_TRANSPORT:
1219 		break;
1220 	case XFRM_MODE_TUNNEL:
1221 		x->props.header_len += sizeof(struct ipv6hdr);
1222 		break;
1223 	}
1224 
1225 	if (x->encap) {
1226 		struct xfrm_encap_tmpl *encap = x->encap;
1227 
1228 		switch (encap->encap_type) {
1229 		default:
1230 			NL_SET_ERR_MSG(extack, "Unsupported encapsulation type for ESP");
1231 			err = -EINVAL;
1232 			goto error;
1233 		case UDP_ENCAP_ESPINUDP:
1234 			x->props.header_len += sizeof(struct udphdr);
1235 			break;
1236 		case UDP_ENCAP_ESPINUDP_NON_IKE:
1237 			x->props.header_len += sizeof(struct udphdr) + 2 * sizeof(u32);
1238 			break;
1239 #ifdef CONFIG_INET6_ESPINTCP
1240 		case TCP_ENCAP_ESPINTCP:
1241 			/* only the length field, TCP encap is done by
1242 			 * the socket
1243 			 */
1244 			x->props.header_len += 2;
1245 			break;
1246 #endif
1247 		}
1248 	}
1249 
1250 	align = ALIGN(crypto_aead_blocksize(aead), 4);
1251 	x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1252 
1253 error:
1254 	return err;
1255 }
1256 
1257 static int esp6_rcv_cb(struct sk_buff *skb, int err)
1258 {
1259 	return 0;
1260 }
1261 
1262 static const struct xfrm_type esp6_type = {
1263 	.owner		= THIS_MODULE,
1264 	.proto		= IPPROTO_ESP,
1265 	.flags		= XFRM_TYPE_REPLAY_PROT,
1266 	.init_state	= esp6_init_state,
1267 	.destructor	= esp6_destroy,
1268 	.input		= esp6_input,
1269 	.output		= esp6_output,
1270 };
1271 
1272 static struct xfrm6_protocol esp6_protocol = {
1273 	.handler	=	xfrm6_rcv,
1274 	.input_handler	=	xfrm_input,
1275 	.cb_handler	=	esp6_rcv_cb,
1276 	.err_handler	=	esp6_err,
1277 	.priority	=	0,
1278 };
1279 
1280 static int __init esp6_init(void)
1281 {
1282 	if (xfrm_register_type(&esp6_type, AF_INET6) < 0) {
1283 		pr_info("%s: can't add xfrm type\n", __func__);
1284 		return -EAGAIN;
1285 	}
1286 	if (xfrm6_protocol_register(&esp6_protocol, IPPROTO_ESP) < 0) {
1287 		pr_info("%s: can't add protocol\n", __func__);
1288 		xfrm_unregister_type(&esp6_type, AF_INET6);
1289 		return -EAGAIN;
1290 	}
1291 
1292 	return 0;
1293 }
1294 
1295 static void __exit esp6_fini(void)
1296 {
1297 	if (xfrm6_protocol_deregister(&esp6_protocol, IPPROTO_ESP) < 0)
1298 		pr_info("%s: can't remove protocol\n", __func__);
1299 	xfrm_unregister_type(&esp6_type, AF_INET6);
1300 }
1301 
1302 module_init(esp6_init);
1303 module_exit(esp6_fini);
1304 
1305 MODULE_DESCRIPTION("IPv6 ESP transformation helpers");
1306 MODULE_LICENSE("GPL");
1307 MODULE_ALIAS_XFRM_TYPE(AF_INET6, XFRM_PROTO_ESP);
1308