xref: /openbmc/linux/net/ipv6/esp6.c (revision e40573a4)
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 	 * accommodate 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 (ALIGN(tailen, L1_CACHE_BYTES) > PAGE_SIZE ||
494 	    ALIGN(skb->data_len, L1_CACHE_BYTES) > PAGE_SIZE)
495 		goto cow;
496 
497 	if (!skb_cloned(skb)) {
498 		if (tailen <= skb_tailroom(skb)) {
499 			nfrags = 1;
500 			trailer = skb;
501 			tail = skb_tail_pointer(trailer);
502 
503 			goto skip_cow;
504 		} else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
505 			   && !skb_has_frag_list(skb)) {
506 			int allocsize;
507 			struct sock *sk = skb->sk;
508 			struct page_frag *pfrag = &x->xfrag;
509 
510 			esp->inplace = false;
511 
512 			allocsize = ALIGN(tailen, L1_CACHE_BYTES);
513 
514 			spin_lock_bh(&x->lock);
515 
516 			if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
517 				spin_unlock_bh(&x->lock);
518 				goto cow;
519 			}
520 
521 			page = pfrag->page;
522 			get_page(page);
523 
524 			tail = page_address(page) + pfrag->offset;
525 
526 			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
527 
528 			nfrags = skb_shinfo(skb)->nr_frags;
529 
530 			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
531 					     tailen);
532 			skb_shinfo(skb)->nr_frags = ++nfrags;
533 
534 			pfrag->offset = pfrag->offset + allocsize;
535 
536 			spin_unlock_bh(&x->lock);
537 
538 			nfrags++;
539 
540 			skb->len += tailen;
541 			skb->data_len += tailen;
542 			skb->truesize += tailen;
543 			if (sk && sk_fullsock(sk))
544 				refcount_add(tailen, &sk->sk_wmem_alloc);
545 
546 			goto out;
547 		}
548 	}
549 
550 cow:
551 	esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
552 
553 	nfrags = skb_cow_data(skb, tailen, &trailer);
554 	if (nfrags < 0)
555 		goto out;
556 	tail = skb_tail_pointer(trailer);
557 	esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
558 
559 skip_cow:
560 	esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
561 	pskb_put(skb, trailer, tailen);
562 
563 out:
564 	return nfrags;
565 }
566 EXPORT_SYMBOL_GPL(esp6_output_head);
567 
568 int esp6_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
569 {
570 	u8 *iv;
571 	int alen;
572 	void *tmp;
573 	int ivlen;
574 	int assoclen;
575 	int extralen;
576 	struct page *page;
577 	struct ip_esp_hdr *esph;
578 	struct aead_request *req;
579 	struct crypto_aead *aead;
580 	struct scatterlist *sg, *dsg;
581 	struct esp_output_extra *extra;
582 	int err = -ENOMEM;
583 
584 	assoclen = sizeof(struct ip_esp_hdr);
585 	extralen = 0;
586 
587 	if (x->props.flags & XFRM_STATE_ESN) {
588 		extralen += sizeof(*extra);
589 		assoclen += sizeof(__be32);
590 	}
591 
592 	aead = x->data;
593 	alen = crypto_aead_authsize(aead);
594 	ivlen = crypto_aead_ivsize(aead);
595 
596 	tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
597 	if (!tmp)
598 		goto error;
599 
600 	extra = esp_tmp_extra(tmp);
601 	iv = esp_tmp_iv(aead, tmp, extralen);
602 	req = esp_tmp_req(aead, iv);
603 	sg = esp_req_sg(aead, req);
604 
605 	if (esp->inplace)
606 		dsg = sg;
607 	else
608 		dsg = &sg[esp->nfrags];
609 
610 	esph = esp_output_set_esn(skb, x, esp->esph, extra);
611 	esp->esph = esph;
612 
613 	sg_init_table(sg, esp->nfrags);
614 	err = skb_to_sgvec(skb, sg,
615 		           (unsigned char *)esph - skb->data,
616 		           assoclen + ivlen + esp->clen + alen);
617 	if (unlikely(err < 0))
618 		goto error_free;
619 
620 	if (!esp->inplace) {
621 		int allocsize;
622 		struct page_frag *pfrag = &x->xfrag;
623 
624 		allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
625 
626 		spin_lock_bh(&x->lock);
627 		if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
628 			spin_unlock_bh(&x->lock);
629 			goto error_free;
630 		}
631 
632 		skb_shinfo(skb)->nr_frags = 1;
633 
634 		page = pfrag->page;
635 		get_page(page);
636 		/* replace page frags in skb with new page */
637 		__skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
638 		pfrag->offset = pfrag->offset + allocsize;
639 		spin_unlock_bh(&x->lock);
640 
641 		sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
642 		err = skb_to_sgvec(skb, dsg,
643 			           (unsigned char *)esph - skb->data,
644 			           assoclen + ivlen + esp->clen + alen);
645 		if (unlikely(err < 0))
646 			goto error_free;
647 	}
648 
649 	if ((x->props.flags & XFRM_STATE_ESN))
650 		aead_request_set_callback(req, 0, esp_output_done_esn, skb);
651 	else
652 		aead_request_set_callback(req, 0, esp_output_done, skb);
653 
654 	aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
655 	aead_request_set_ad(req, assoclen);
656 
657 	memset(iv, 0, ivlen);
658 	memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
659 	       min(ivlen, 8));
660 
661 	ESP_SKB_CB(skb)->tmp = tmp;
662 	err = crypto_aead_encrypt(req);
663 
664 	switch (err) {
665 	case -EINPROGRESS:
666 		goto error;
667 
668 	case -ENOSPC:
669 		err = NET_XMIT_DROP;
670 		break;
671 
672 	case 0:
673 		if ((x->props.flags & XFRM_STATE_ESN))
674 			esp_output_restore_header(skb);
675 		esp_output_encap_csum(skb);
676 	}
677 
678 	if (sg != dsg)
679 		esp_ssg_unref(x, tmp);
680 
681 	if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
682 		err = esp_output_tail_tcp(x, skb);
683 
684 error_free:
685 	kfree(tmp);
686 error:
687 	return err;
688 }
689 EXPORT_SYMBOL_GPL(esp6_output_tail);
690 
691 static int esp6_output(struct xfrm_state *x, struct sk_buff *skb)
692 {
693 	int alen;
694 	int blksize;
695 	struct ip_esp_hdr *esph;
696 	struct crypto_aead *aead;
697 	struct esp_info esp;
698 
699 	esp.inplace = true;
700 
701 	esp.proto = *skb_mac_header(skb);
702 	*skb_mac_header(skb) = IPPROTO_ESP;
703 
704 	/* skb is pure payload to encrypt */
705 
706 	aead = x->data;
707 	alen = crypto_aead_authsize(aead);
708 
709 	esp.tfclen = 0;
710 	if (x->tfcpad) {
711 		struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
712 		u32 padto;
713 
714 		padto = min(x->tfcpad, xfrm_state_mtu(x, dst->child_mtu_cached));
715 		if (skb->len < padto)
716 			esp.tfclen = padto - skb->len;
717 	}
718 	blksize = ALIGN(crypto_aead_blocksize(aead), 4);
719 	esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
720 	esp.plen = esp.clen - skb->len - esp.tfclen;
721 	esp.tailen = esp.tfclen + esp.plen + alen;
722 
723 	esp.esph = ip_esp_hdr(skb);
724 
725 	esp.nfrags = esp6_output_head(x, skb, &esp);
726 	if (esp.nfrags < 0)
727 		return esp.nfrags;
728 
729 	esph = esp.esph;
730 	esph->spi = x->id.spi;
731 
732 	esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
733 	esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
734 			    ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
735 
736 	skb_push(skb, -skb_network_offset(skb));
737 
738 	return esp6_output_tail(x, skb, &esp);
739 }
740 
741 static inline int esp_remove_trailer(struct sk_buff *skb)
742 {
743 	struct xfrm_state *x = xfrm_input_state(skb);
744 	struct crypto_aead *aead = x->data;
745 	int alen, hlen, elen;
746 	int padlen, trimlen;
747 	__wsum csumdiff;
748 	u8 nexthdr[2];
749 	int ret;
750 
751 	alen = crypto_aead_authsize(aead);
752 	hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
753 	elen = skb->len - hlen;
754 
755 	ret = skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2);
756 	BUG_ON(ret);
757 
758 	ret = -EINVAL;
759 	padlen = nexthdr[0];
760 	if (padlen + 2 + alen >= elen) {
761 		net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
762 				    padlen + 2, elen - alen);
763 		goto out;
764 	}
765 
766 	trimlen = alen + padlen + 2;
767 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
768 		csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
769 		skb->csum = csum_block_sub(skb->csum, csumdiff,
770 					   skb->len - trimlen);
771 	}
772 	pskb_trim(skb, skb->len - trimlen);
773 
774 	ret = nexthdr[1];
775 
776 out:
777 	return ret;
778 }
779 
780 int esp6_input_done2(struct sk_buff *skb, int err)
781 {
782 	struct xfrm_state *x = xfrm_input_state(skb);
783 	struct xfrm_offload *xo = xfrm_offload(skb);
784 	struct crypto_aead *aead = x->data;
785 	int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
786 	int hdr_len = skb_network_header_len(skb);
787 
788 	if (!xo || !(xo->flags & CRYPTO_DONE))
789 		kfree(ESP_SKB_CB(skb)->tmp);
790 
791 	if (unlikely(err))
792 		goto out;
793 
794 	err = esp_remove_trailer(skb);
795 	if (unlikely(err < 0))
796 		goto out;
797 
798 	if (x->encap) {
799 		const struct ipv6hdr *ip6h = ipv6_hdr(skb);
800 		int offset = skb_network_offset(skb) + sizeof(*ip6h);
801 		struct xfrm_encap_tmpl *encap = x->encap;
802 		u8 nexthdr = ip6h->nexthdr;
803 		__be16 frag_off, source;
804 		struct udphdr *uh;
805 		struct tcphdr *th;
806 
807 		offset = ipv6_skip_exthdr(skb, offset, &nexthdr, &frag_off);
808 		if (offset == -1) {
809 			err = -EINVAL;
810 			goto out;
811 		}
812 
813 		uh = (void *)(skb->data + offset);
814 		th = (void *)(skb->data + offset);
815 		hdr_len += offset;
816 
817 		switch (x->encap->encap_type) {
818 		case TCP_ENCAP_ESPINTCP:
819 			source = th->source;
820 			break;
821 		case UDP_ENCAP_ESPINUDP:
822 		case UDP_ENCAP_ESPINUDP_NON_IKE:
823 			source = uh->source;
824 			break;
825 		default:
826 			WARN_ON_ONCE(1);
827 			err = -EINVAL;
828 			goto out;
829 		}
830 
831 		/*
832 		 * 1) if the NAT-T peer's IP or port changed then
833 		 *    advertize the change to the keying daemon.
834 		 *    This is an inbound SA, so just compare
835 		 *    SRC ports.
836 		 */
837 		if (!ipv6_addr_equal(&ip6h->saddr, &x->props.saddr.in6) ||
838 		    source != encap->encap_sport) {
839 			xfrm_address_t ipaddr;
840 
841 			memcpy(&ipaddr.a6, &ip6h->saddr.s6_addr, sizeof(ipaddr.a6));
842 			km_new_mapping(x, &ipaddr, source);
843 
844 			/* XXX: perhaps add an extra
845 			 * policy check here, to see
846 			 * if we should allow or
847 			 * reject a packet from a
848 			 * different source
849 			 * address/port.
850 			 */
851 		}
852 
853 		/*
854 		 * 2) ignore UDP/TCP checksums in case
855 		 *    of NAT-T in Transport Mode, or
856 		 *    perform other post-processing fixes
857 		 *    as per draft-ietf-ipsec-udp-encaps-06,
858 		 *    section 3.1.2
859 		 */
860 		if (x->props.mode == XFRM_MODE_TRANSPORT)
861 			skb->ip_summed = CHECKSUM_UNNECESSARY;
862 	}
863 
864 	skb_postpull_rcsum(skb, skb_network_header(skb),
865 			   skb_network_header_len(skb));
866 	skb_pull_rcsum(skb, hlen);
867 	if (x->props.mode == XFRM_MODE_TUNNEL)
868 		skb_reset_transport_header(skb);
869 	else
870 		skb_set_transport_header(skb, -hdr_len);
871 
872 	/* RFC4303: Drop dummy packets without any error */
873 	if (err == IPPROTO_NONE)
874 		err = -EINVAL;
875 
876 out:
877 	return err;
878 }
879 EXPORT_SYMBOL_GPL(esp6_input_done2);
880 
881 static void esp_input_done(struct crypto_async_request *base, int err)
882 {
883 	struct sk_buff *skb = base->data;
884 
885 	xfrm_input_resume(skb, esp6_input_done2(skb, err));
886 }
887 
888 static void esp_input_restore_header(struct sk_buff *skb)
889 {
890 	esp_restore_header(skb, 0);
891 	__skb_pull(skb, 4);
892 }
893 
894 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
895 {
896 	struct xfrm_state *x = xfrm_input_state(skb);
897 
898 	/* For ESN we move the header forward by 4 bytes to
899 	 * accommodate the high bits.  We will move it back after
900 	 * decryption.
901 	 */
902 	if ((x->props.flags & XFRM_STATE_ESN)) {
903 		struct ip_esp_hdr *esph = skb_push(skb, 4);
904 
905 		*seqhi = esph->spi;
906 		esph->spi = esph->seq_no;
907 		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
908 	}
909 }
910 
911 static void esp_input_done_esn(struct crypto_async_request *base, int err)
912 {
913 	struct sk_buff *skb = base->data;
914 
915 	esp_input_restore_header(skb);
916 	esp_input_done(base, err);
917 }
918 
919 static int esp6_input(struct xfrm_state *x, struct sk_buff *skb)
920 {
921 	struct crypto_aead *aead = x->data;
922 	struct aead_request *req;
923 	struct sk_buff *trailer;
924 	int ivlen = crypto_aead_ivsize(aead);
925 	int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
926 	int nfrags;
927 	int assoclen;
928 	int seqhilen;
929 	int ret = 0;
930 	void *tmp;
931 	__be32 *seqhi;
932 	u8 *iv;
933 	struct scatterlist *sg;
934 
935 	if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen)) {
936 		ret = -EINVAL;
937 		goto out;
938 	}
939 
940 	if (elen <= 0) {
941 		ret = -EINVAL;
942 		goto out;
943 	}
944 
945 	assoclen = sizeof(struct ip_esp_hdr);
946 	seqhilen = 0;
947 
948 	if (x->props.flags & XFRM_STATE_ESN) {
949 		seqhilen += sizeof(__be32);
950 		assoclen += seqhilen;
951 	}
952 
953 	if (!skb_cloned(skb)) {
954 		if (!skb_is_nonlinear(skb)) {
955 			nfrags = 1;
956 
957 			goto skip_cow;
958 		} else if (!skb_has_frag_list(skb)) {
959 			nfrags = skb_shinfo(skb)->nr_frags;
960 			nfrags++;
961 
962 			goto skip_cow;
963 		}
964 	}
965 
966 	nfrags = skb_cow_data(skb, 0, &trailer);
967 	if (nfrags < 0) {
968 		ret = -EINVAL;
969 		goto out;
970 	}
971 
972 skip_cow:
973 	ret = -ENOMEM;
974 	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
975 	if (!tmp)
976 		goto out;
977 
978 	ESP_SKB_CB(skb)->tmp = tmp;
979 	seqhi = esp_tmp_extra(tmp);
980 	iv = esp_tmp_iv(aead, tmp, seqhilen);
981 	req = esp_tmp_req(aead, iv);
982 	sg = esp_req_sg(aead, req);
983 
984 	esp_input_set_header(skb, seqhi);
985 
986 	sg_init_table(sg, nfrags);
987 	ret = skb_to_sgvec(skb, sg, 0, skb->len);
988 	if (unlikely(ret < 0)) {
989 		kfree(tmp);
990 		goto out;
991 	}
992 
993 	skb->ip_summed = CHECKSUM_NONE;
994 
995 	if ((x->props.flags & XFRM_STATE_ESN))
996 		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
997 	else
998 		aead_request_set_callback(req, 0, esp_input_done, skb);
999 
1000 	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
1001 	aead_request_set_ad(req, assoclen);
1002 
1003 	ret = crypto_aead_decrypt(req);
1004 	if (ret == -EINPROGRESS)
1005 		goto out;
1006 
1007 	if ((x->props.flags & XFRM_STATE_ESN))
1008 		esp_input_restore_header(skb);
1009 
1010 	ret = esp6_input_done2(skb, ret);
1011 
1012 out:
1013 	return ret;
1014 }
1015 
1016 static int esp6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
1017 		    u8 type, u8 code, int offset, __be32 info)
1018 {
1019 	struct net *net = dev_net(skb->dev);
1020 	const struct ipv6hdr *iph = (const struct ipv6hdr *)skb->data;
1021 	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data + offset);
1022 	struct xfrm_state *x;
1023 
1024 	if (type != ICMPV6_PKT_TOOBIG &&
1025 	    type != NDISC_REDIRECT)
1026 		return 0;
1027 
1028 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
1029 			      esph->spi, IPPROTO_ESP, AF_INET6);
1030 	if (!x)
1031 		return 0;
1032 
1033 	if (type == NDISC_REDIRECT)
1034 		ip6_redirect(skb, net, skb->dev->ifindex, 0,
1035 			     sock_net_uid(net, NULL));
1036 	else
1037 		ip6_update_pmtu(skb, net, info, 0, 0, sock_net_uid(net, NULL));
1038 	xfrm_state_put(x);
1039 
1040 	return 0;
1041 }
1042 
1043 static void esp6_destroy(struct xfrm_state *x)
1044 {
1045 	struct crypto_aead *aead = x->data;
1046 
1047 	if (!aead)
1048 		return;
1049 
1050 	crypto_free_aead(aead);
1051 }
1052 
1053 static int esp_init_aead(struct xfrm_state *x)
1054 {
1055 	char aead_name[CRYPTO_MAX_ALG_NAME];
1056 	struct crypto_aead *aead;
1057 	int err;
1058 
1059 	err = -ENAMETOOLONG;
1060 	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1061 		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME)
1062 		goto error;
1063 
1064 	aead = crypto_alloc_aead(aead_name, 0, 0);
1065 	err = PTR_ERR(aead);
1066 	if (IS_ERR(aead))
1067 		goto error;
1068 
1069 	x->data = aead;
1070 
1071 	err = crypto_aead_setkey(aead, x->aead->alg_key,
1072 				 (x->aead->alg_key_len + 7) / 8);
1073 	if (err)
1074 		goto error;
1075 
1076 	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1077 	if (err)
1078 		goto error;
1079 
1080 error:
1081 	return err;
1082 }
1083 
1084 static int esp_init_authenc(struct xfrm_state *x)
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 = -EINVAL;
1096 	if (!x->ealg)
1097 		goto error;
1098 
1099 	err = -ENAMETOOLONG;
1100 
1101 	if ((x->props.flags & XFRM_STATE_ESN)) {
1102 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1103 			     "%s%sauthencesn(%s,%s)%s",
1104 			     x->geniv ?: "", x->geniv ? "(" : "",
1105 			     x->aalg ? x->aalg->alg_name : "digest_null",
1106 			     x->ealg->alg_name,
1107 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1108 			goto error;
1109 	} else {
1110 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1111 			     "%s%sauthenc(%s,%s)%s",
1112 			     x->geniv ?: "", x->geniv ? "(" : "",
1113 			     x->aalg ? x->aalg->alg_name : "digest_null",
1114 			     x->ealg->alg_name,
1115 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1116 			goto error;
1117 	}
1118 
1119 	aead = crypto_alloc_aead(authenc_name, 0, 0);
1120 	err = PTR_ERR(aead);
1121 	if (IS_ERR(aead))
1122 		goto error;
1123 
1124 	x->data = aead;
1125 
1126 	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1127 		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1128 	err = -ENOMEM;
1129 	key = kmalloc(keylen, GFP_KERNEL);
1130 	if (!key)
1131 		goto error;
1132 
1133 	p = key;
1134 	rta = (void *)p;
1135 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1136 	rta->rta_len = RTA_LENGTH(sizeof(*param));
1137 	param = RTA_DATA(rta);
1138 	p += RTA_SPACE(sizeof(*param));
1139 
1140 	if (x->aalg) {
1141 		struct xfrm_algo_desc *aalg_desc;
1142 
1143 		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1144 		p += (x->aalg->alg_key_len + 7) / 8;
1145 
1146 		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1147 		BUG_ON(!aalg_desc);
1148 
1149 		err = -EINVAL;
1150 		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1151 		    crypto_aead_authsize(aead)) {
1152 			pr_info("ESP: %s digestsize %u != %u\n",
1153 				x->aalg->alg_name,
1154 				crypto_aead_authsize(aead),
1155 				aalg_desc->uinfo.auth.icv_fullbits / 8);
1156 			goto free_key;
1157 		}
1158 
1159 		err = crypto_aead_setauthsize(
1160 			aead, x->aalg->alg_trunc_len / 8);
1161 		if (err)
1162 			goto free_key;
1163 	}
1164 
1165 	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1166 	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1167 
1168 	err = crypto_aead_setkey(aead, key, keylen);
1169 
1170 free_key:
1171 	kfree(key);
1172 
1173 error:
1174 	return err;
1175 }
1176 
1177 static int esp6_init_state(struct xfrm_state *x)
1178 {
1179 	struct crypto_aead *aead;
1180 	u32 align;
1181 	int err;
1182 
1183 	x->data = NULL;
1184 
1185 	if (x->aead)
1186 		err = esp_init_aead(x);
1187 	else
1188 		err = esp_init_authenc(x);
1189 
1190 	if (err)
1191 		goto error;
1192 
1193 	aead = x->data;
1194 
1195 	x->props.header_len = sizeof(struct ip_esp_hdr) +
1196 			      crypto_aead_ivsize(aead);
1197 	switch (x->props.mode) {
1198 	case XFRM_MODE_BEET:
1199 		if (x->sel.family != AF_INET6)
1200 			x->props.header_len += IPV4_BEET_PHMAXLEN +
1201 					       (sizeof(struct ipv6hdr) - sizeof(struct iphdr));
1202 		break;
1203 	default:
1204 	case XFRM_MODE_TRANSPORT:
1205 		break;
1206 	case XFRM_MODE_TUNNEL:
1207 		x->props.header_len += sizeof(struct ipv6hdr);
1208 		break;
1209 	}
1210 
1211 	if (x->encap) {
1212 		struct xfrm_encap_tmpl *encap = x->encap;
1213 
1214 		switch (encap->encap_type) {
1215 		default:
1216 			err = -EINVAL;
1217 			goto error;
1218 		case UDP_ENCAP_ESPINUDP:
1219 			x->props.header_len += sizeof(struct udphdr);
1220 			break;
1221 		case UDP_ENCAP_ESPINUDP_NON_IKE:
1222 			x->props.header_len += sizeof(struct udphdr) + 2 * sizeof(u32);
1223 			break;
1224 #ifdef CONFIG_INET6_ESPINTCP
1225 		case TCP_ENCAP_ESPINTCP:
1226 			/* only the length field, TCP encap is done by
1227 			 * the socket
1228 			 */
1229 			x->props.header_len += 2;
1230 			break;
1231 #endif
1232 		}
1233 	}
1234 
1235 	align = ALIGN(crypto_aead_blocksize(aead), 4);
1236 	x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1237 
1238 error:
1239 	return err;
1240 }
1241 
1242 static int esp6_rcv_cb(struct sk_buff *skb, int err)
1243 {
1244 	return 0;
1245 }
1246 
1247 static const struct xfrm_type esp6_type = {
1248 	.owner		= THIS_MODULE,
1249 	.proto		= IPPROTO_ESP,
1250 	.flags		= XFRM_TYPE_REPLAY_PROT,
1251 	.init_state	= esp6_init_state,
1252 	.destructor	= esp6_destroy,
1253 	.input		= esp6_input,
1254 	.output		= esp6_output,
1255 };
1256 
1257 static struct xfrm6_protocol esp6_protocol = {
1258 	.handler	=	xfrm6_rcv,
1259 	.input_handler	=	xfrm_input,
1260 	.cb_handler	=	esp6_rcv_cb,
1261 	.err_handler	=	esp6_err,
1262 	.priority	=	0,
1263 };
1264 
1265 static int __init esp6_init(void)
1266 {
1267 	if (xfrm_register_type(&esp6_type, AF_INET6) < 0) {
1268 		pr_info("%s: can't add xfrm type\n", __func__);
1269 		return -EAGAIN;
1270 	}
1271 	if (xfrm6_protocol_register(&esp6_protocol, IPPROTO_ESP) < 0) {
1272 		pr_info("%s: can't add protocol\n", __func__);
1273 		xfrm_unregister_type(&esp6_type, AF_INET6);
1274 		return -EAGAIN;
1275 	}
1276 
1277 	return 0;
1278 }
1279 
1280 static void __exit esp6_fini(void)
1281 {
1282 	if (xfrm6_protocol_deregister(&esp6_protocol, IPPROTO_ESP) < 0)
1283 		pr_info("%s: can't remove protocol\n", __func__);
1284 	xfrm_unregister_type(&esp6_type, AF_INET6);
1285 }
1286 
1287 module_init(esp6_init);
1288 module_exit(esp6_fini);
1289 
1290 MODULE_LICENSE("GPL");
1291 MODULE_ALIAS_XFRM_TYPE(AF_INET6, XFRM_PROTO_ESP);
1292