xref: /openbmc/linux/net/ipv6/esp6.c (revision b9dd2add) 
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 esp_output_extra *extra = esp_tmp_extra(tmp);
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(*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 			put_page(sg_page(sg));
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 esp_tcp_sk *esk;
156 	__be16 sport, dport;
157 	struct sock *nsk;
158 	struct sock *sk;
159 
160 	sk = rcu_dereference(x->encap_sk);
161 	if (sk && sk->sk_state == TCP_ESTABLISHED)
162 		return sk;
163 
164 	spin_lock_bh(&x->lock);
165 	sport = encap->encap_sport;
166 	dport = encap->encap_dport;
167 	nsk = rcu_dereference_protected(x->encap_sk,
168 					lockdep_is_held(&x->lock));
169 	if (sk && sk == nsk) {
170 		esk = kmalloc(sizeof(*esk), GFP_ATOMIC);
171 		if (!esk) {
172 			spin_unlock_bh(&x->lock);
173 			return ERR_PTR(-ENOMEM);
174 		}
175 		RCU_INIT_POINTER(x->encap_sk, NULL);
176 		esk->sk = sk;
177 		call_rcu(&esk->rcu, esp_free_tcp_sk);
178 	}
179 	spin_unlock_bh(&x->lock);
180 
181 	sk = __inet6_lookup_established(xs_net(x), &tcp_hashinfo, &x->id.daddr.in6,
182 					dport, &x->props.saddr.in6, ntohs(sport), 0, 0);
183 	if (!sk)
184 		return ERR_PTR(-ENOENT);
185 
186 	if (!tcp_is_ulp_esp(sk)) {
187 		sock_put(sk);
188 		return ERR_PTR(-EINVAL);
189 	}
190 
191 	spin_lock_bh(&x->lock);
192 	nsk = rcu_dereference_protected(x->encap_sk,
193 					lockdep_is_held(&x->lock));
194 	if (encap->encap_sport != sport ||
195 	    encap->encap_dport != dport) {
196 		sock_put(sk);
197 		sk = nsk ?: ERR_PTR(-EREMCHG);
198 	} else if (sk == nsk) {
199 		sock_put(sk);
200 	} else {
201 		rcu_assign_pointer(x->encap_sk, sk);
202 	}
203 	spin_unlock_bh(&x->lock);
204 
205 	return sk;
206 }
207 
208 static int esp_output_tcp_finish(struct xfrm_state *x, struct sk_buff *skb)
209 {
210 	struct sock *sk;
211 	int err;
212 
213 	rcu_read_lock();
214 
215 	sk = esp6_find_tcp_sk(x);
216 	err = PTR_ERR_OR_ZERO(sk);
217 	if (err)
218 		goto out;
219 
220 	bh_lock_sock(sk);
221 	if (sock_owned_by_user(sk))
222 		err = espintcp_queue_out(sk, skb);
223 	else
224 		err = espintcp_push_skb(sk, skb);
225 	bh_unlock_sock(sk);
226 
227 out:
228 	rcu_read_unlock();
229 	return err;
230 }
231 
232 static int esp_output_tcp_encap_cb(struct net *net, struct sock *sk,
233 				   struct sk_buff *skb)
234 {
235 	struct dst_entry *dst = skb_dst(skb);
236 	struct xfrm_state *x = dst->xfrm;
237 
238 	return esp_output_tcp_finish(x, skb);
239 }
240 
241 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
242 {
243 	int err;
244 
245 	local_bh_disable();
246 	err = xfrm_trans_queue_net(xs_net(x), skb, esp_output_tcp_encap_cb);
247 	local_bh_enable();
248 
249 	/* EINPROGRESS just happens to do the right thing.  It
250 	 * actually means that the skb has been consumed and
251 	 * isn't coming back.
252 	 */
253 	return err ?: -EINPROGRESS;
254 }
255 #else
256 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
257 {
258 	kfree_skb(skb);
259 
260 	return -EOPNOTSUPP;
261 }
262 #endif
263 
264 static void esp_output_encap_csum(struct sk_buff *skb)
265 {
266 	/* UDP encap with IPv6 requires a valid checksum */
267 	if (*skb_mac_header(skb) == IPPROTO_UDP) {
268 		struct udphdr *uh = udp_hdr(skb);
269 		struct ipv6hdr *ip6h = ipv6_hdr(skb);
270 		int len = ntohs(uh->len);
271 		unsigned int offset = skb_transport_offset(skb);
272 		__wsum csum = skb_checksum(skb, offset, skb->len - offset, 0);
273 
274 		uh->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
275 					    len, IPPROTO_UDP, csum);
276 		if (uh->check == 0)
277 			uh->check = CSUM_MANGLED_0;
278 	}
279 }
280 
281 static void esp_output_done(struct crypto_async_request *base, int err)
282 {
283 	struct sk_buff *skb = base->data;
284 	struct xfrm_offload *xo = xfrm_offload(skb);
285 	void *tmp;
286 	struct xfrm_state *x;
287 
288 	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
289 		struct sec_path *sp = skb_sec_path(skb);
290 
291 		x = sp->xvec[sp->len - 1];
292 	} else {
293 		x = skb_dst(skb)->xfrm;
294 	}
295 
296 	tmp = ESP_SKB_CB(skb)->tmp;
297 	esp_ssg_unref(x, tmp);
298 	kfree(tmp);
299 
300 	esp_output_encap_csum(skb);
301 
302 	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
303 		if (err) {
304 			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
305 			kfree_skb(skb);
306 			return;
307 		}
308 
309 		skb_push(skb, skb->data - skb_mac_header(skb));
310 		secpath_reset(skb);
311 		xfrm_dev_resume(skb);
312 	} else {
313 		if (!err &&
314 		    x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
315 			esp_output_tail_tcp(x, skb);
316 		else
317 			xfrm_output_resume(skb, err);
318 	}
319 }
320 
321 /* Move ESP header back into place. */
322 static void esp_restore_header(struct sk_buff *skb, unsigned int offset)
323 {
324 	struct ip_esp_hdr *esph = (void *)(skb->data + offset);
325 	void *tmp = ESP_SKB_CB(skb)->tmp;
326 	__be32 *seqhi = esp_tmp_extra(tmp);
327 
328 	esph->seq_no = esph->spi;
329 	esph->spi = *seqhi;
330 }
331 
332 static void esp_output_restore_header(struct sk_buff *skb)
333 {
334 	void *tmp = ESP_SKB_CB(skb)->tmp;
335 	struct esp_output_extra *extra = esp_tmp_extra(tmp);
336 
337 	esp_restore_header(skb, skb_transport_offset(skb) + extra->esphoff -
338 				sizeof(__be32));
339 }
340 
341 static struct ip_esp_hdr *esp_output_set_esn(struct sk_buff *skb,
342 					     struct xfrm_state *x,
343 					     struct ip_esp_hdr *esph,
344 					     struct esp_output_extra *extra)
345 {
346 	/* For ESN we move the header forward by 4 bytes to
347 	 * accomodate the high bits.  We will move it back after
348 	 * encryption.
349 	 */
350 	if ((x->props.flags & XFRM_STATE_ESN)) {
351 		__u32 seqhi;
352 		struct xfrm_offload *xo = xfrm_offload(skb);
353 
354 		if (xo)
355 			seqhi = xo->seq.hi;
356 		else
357 			seqhi = XFRM_SKB_CB(skb)->seq.output.hi;
358 
359 		extra->esphoff = (unsigned char *)esph -
360 				 skb_transport_header(skb);
361 		esph = (struct ip_esp_hdr *)((unsigned char *)esph - 4);
362 		extra->seqhi = esph->spi;
363 		esph->seq_no = htonl(seqhi);
364 	}
365 
366 	esph->spi = x->id.spi;
367 
368 	return esph;
369 }
370 
371 static void esp_output_done_esn(struct crypto_async_request *base, int err)
372 {
373 	struct sk_buff *skb = base->data;
374 
375 	esp_output_restore_header(skb);
376 	esp_output_done(base, err);
377 }
378 
379 static struct ip_esp_hdr *esp6_output_udp_encap(struct sk_buff *skb,
380 					       int encap_type,
381 					       struct esp_info *esp,
382 					       __be16 sport,
383 					       __be16 dport)
384 {
385 	struct udphdr *uh;
386 	__be32 *udpdata32;
387 	unsigned int len;
388 
389 	len = skb->len + esp->tailen - skb_transport_offset(skb);
390 	if (len > U16_MAX)
391 		return ERR_PTR(-EMSGSIZE);
392 
393 	uh = (struct udphdr *)esp->esph;
394 	uh->source = sport;
395 	uh->dest = dport;
396 	uh->len = htons(len);
397 	uh->check = 0;
398 
399 	*skb_mac_header(skb) = IPPROTO_UDP;
400 
401 	if (encap_type == UDP_ENCAP_ESPINUDP_NON_IKE) {
402 		udpdata32 = (__be32 *)(uh + 1);
403 		udpdata32[0] = udpdata32[1] = 0;
404 		return (struct ip_esp_hdr *)(udpdata32 + 2);
405 	}
406 
407 	return (struct ip_esp_hdr *)(uh + 1);
408 }
409 
410 #ifdef CONFIG_INET6_ESPINTCP
411 static struct ip_esp_hdr *esp6_output_tcp_encap(struct xfrm_state *x,
412 						struct sk_buff *skb,
413 						struct esp_info *esp)
414 {
415 	__be16 *lenp = (void *)esp->esph;
416 	struct ip_esp_hdr *esph;
417 	unsigned int len;
418 	struct sock *sk;
419 
420 	len = skb->len + esp->tailen - skb_transport_offset(skb);
421 	if (len > IP_MAX_MTU)
422 		return ERR_PTR(-EMSGSIZE);
423 
424 	rcu_read_lock();
425 	sk = esp6_find_tcp_sk(x);
426 	rcu_read_unlock();
427 
428 	if (IS_ERR(sk))
429 		return ERR_CAST(sk);
430 
431 	*lenp = htons(len);
432 	esph = (struct ip_esp_hdr *)(lenp + 1);
433 
434 	return esph;
435 }
436 #else
437 static struct ip_esp_hdr *esp6_output_tcp_encap(struct xfrm_state *x,
438 						struct sk_buff *skb,
439 						struct esp_info *esp)
440 {
441 	return ERR_PTR(-EOPNOTSUPP);
442 }
443 #endif
444 
445 static int esp6_output_encap(struct xfrm_state *x, struct sk_buff *skb,
446 			    struct esp_info *esp)
447 {
448 	struct xfrm_encap_tmpl *encap = x->encap;
449 	struct ip_esp_hdr *esph;
450 	__be16 sport, dport;
451 	int encap_type;
452 
453 	spin_lock_bh(&x->lock);
454 	sport = encap->encap_sport;
455 	dport = encap->encap_dport;
456 	encap_type = encap->encap_type;
457 	spin_unlock_bh(&x->lock);
458 
459 	switch (encap_type) {
460 	default:
461 	case UDP_ENCAP_ESPINUDP:
462 	case UDP_ENCAP_ESPINUDP_NON_IKE:
463 		esph = esp6_output_udp_encap(skb, encap_type, esp, sport, dport);
464 		break;
465 	case TCP_ENCAP_ESPINTCP:
466 		esph = esp6_output_tcp_encap(x, skb, esp);
467 		break;
468 	}
469 
470 	if (IS_ERR(esph))
471 		return PTR_ERR(esph);
472 
473 	esp->esph = esph;
474 
475 	return 0;
476 }
477 
478 int esp6_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
479 {
480 	u8 *tail;
481 	int nfrags;
482 	int esph_offset;
483 	struct page *page;
484 	struct sk_buff *trailer;
485 	int tailen = esp->tailen;
486 
487 	if (x->encap) {
488 		int err = esp6_output_encap(x, skb, esp);
489 
490 		if (err < 0)
491 			return err;
492 	}
493 
494 	if (!skb_cloned(skb)) {
495 		if (tailen <= skb_tailroom(skb)) {
496 			nfrags = 1;
497 			trailer = skb;
498 			tail = skb_tail_pointer(trailer);
499 
500 			goto skip_cow;
501 		} else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
502 			   && !skb_has_frag_list(skb)) {
503 			int allocsize;
504 			struct sock *sk = skb->sk;
505 			struct page_frag *pfrag = &x->xfrag;
506 
507 			esp->inplace = false;
508 
509 			allocsize = ALIGN(tailen, L1_CACHE_BYTES);
510 
511 			spin_lock_bh(&x->lock);
512 
513 			if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
514 				spin_unlock_bh(&x->lock);
515 				goto cow;
516 			}
517 
518 			page = pfrag->page;
519 			get_page(page);
520 
521 			tail = page_address(page) + pfrag->offset;
522 
523 			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
524 
525 			nfrags = skb_shinfo(skb)->nr_frags;
526 
527 			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
528 					     tailen);
529 			skb_shinfo(skb)->nr_frags = ++nfrags;
530 
531 			pfrag->offset = pfrag->offset + allocsize;
532 
533 			spin_unlock_bh(&x->lock);
534 
535 			nfrags++;
536 
537 			skb->len += tailen;
538 			skb->data_len += tailen;
539 			skb->truesize += tailen;
540 			if (sk && sk_fullsock(sk))
541 				refcount_add(tailen, &sk->sk_wmem_alloc);
542 
543 			goto out;
544 		}
545 	}
546 
547 cow:
548 	esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
549 
550 	nfrags = skb_cow_data(skb, tailen, &trailer);
551 	if (nfrags < 0)
552 		goto out;
553 	tail = skb_tail_pointer(trailer);
554 	esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
555 
556 skip_cow:
557 	esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
558 	pskb_put(skb, trailer, tailen);
559 
560 out:
561 	return nfrags;
562 }
563 EXPORT_SYMBOL_GPL(esp6_output_head);
564 
565 int esp6_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
566 {
567 	u8 *iv;
568 	int alen;
569 	void *tmp;
570 	int ivlen;
571 	int assoclen;
572 	int extralen;
573 	struct page *page;
574 	struct ip_esp_hdr *esph;
575 	struct aead_request *req;
576 	struct crypto_aead *aead;
577 	struct scatterlist *sg, *dsg;
578 	struct esp_output_extra *extra;
579 	int err = -ENOMEM;
580 
581 	assoclen = sizeof(struct ip_esp_hdr);
582 	extralen = 0;
583 
584 	if (x->props.flags & XFRM_STATE_ESN) {
585 		extralen += sizeof(*extra);
586 		assoclen += sizeof(__be32);
587 	}
588 
589 	aead = x->data;
590 	alen = crypto_aead_authsize(aead);
591 	ivlen = crypto_aead_ivsize(aead);
592 
593 	tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
594 	if (!tmp)
595 		goto error;
596 
597 	extra = esp_tmp_extra(tmp);
598 	iv = esp_tmp_iv(aead, tmp, extralen);
599 	req = esp_tmp_req(aead, iv);
600 	sg = esp_req_sg(aead, req);
601 
602 	if (esp->inplace)
603 		dsg = sg;
604 	else
605 		dsg = &sg[esp->nfrags];
606 
607 	esph = esp_output_set_esn(skb, x, esp->esph, extra);
608 	esp->esph = esph;
609 
610 	sg_init_table(sg, esp->nfrags);
611 	err = skb_to_sgvec(skb, sg,
612 		           (unsigned char *)esph - skb->data,
613 		           assoclen + ivlen + esp->clen + alen);
614 	if (unlikely(err < 0))
615 		goto error_free;
616 
617 	if (!esp->inplace) {
618 		int allocsize;
619 		struct page_frag *pfrag = &x->xfrag;
620 
621 		allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
622 
623 		spin_lock_bh(&x->lock);
624 		if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
625 			spin_unlock_bh(&x->lock);
626 			goto error_free;
627 		}
628 
629 		skb_shinfo(skb)->nr_frags = 1;
630 
631 		page = pfrag->page;
632 		get_page(page);
633 		/* replace page frags in skb with new page */
634 		__skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
635 		pfrag->offset = pfrag->offset + allocsize;
636 		spin_unlock_bh(&x->lock);
637 
638 		sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
639 		err = skb_to_sgvec(skb, dsg,
640 			           (unsigned char *)esph - skb->data,
641 			           assoclen + ivlen + esp->clen + alen);
642 		if (unlikely(err < 0))
643 			goto error_free;
644 	}
645 
646 	if ((x->props.flags & XFRM_STATE_ESN))
647 		aead_request_set_callback(req, 0, esp_output_done_esn, skb);
648 	else
649 		aead_request_set_callback(req, 0, esp_output_done, skb);
650 
651 	aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
652 	aead_request_set_ad(req, assoclen);
653 
654 	memset(iv, 0, ivlen);
655 	memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
656 	       min(ivlen, 8));
657 
658 	ESP_SKB_CB(skb)->tmp = tmp;
659 	err = crypto_aead_encrypt(req);
660 
661 	switch (err) {
662 	case -EINPROGRESS:
663 		goto error;
664 
665 	case -ENOSPC:
666 		err = NET_XMIT_DROP;
667 		break;
668 
669 	case 0:
670 		if ((x->props.flags & XFRM_STATE_ESN))
671 			esp_output_restore_header(skb);
672 		esp_output_encap_csum(skb);
673 	}
674 
675 	if (sg != dsg)
676 		esp_ssg_unref(x, tmp);
677 
678 	if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
679 		err = esp_output_tail_tcp(x, skb);
680 
681 error_free:
682 	kfree(tmp);
683 error:
684 	return err;
685 }
686 EXPORT_SYMBOL_GPL(esp6_output_tail);
687 
688 static int esp6_output(struct xfrm_state *x, struct sk_buff *skb)
689 {
690 	int alen;
691 	int blksize;
692 	struct ip_esp_hdr *esph;
693 	struct crypto_aead *aead;
694 	struct esp_info esp;
695 
696 	esp.inplace = true;
697 
698 	esp.proto = *skb_mac_header(skb);
699 	*skb_mac_header(skb) = IPPROTO_ESP;
700 
701 	/* skb is pure payload to encrypt */
702 
703 	aead = x->data;
704 	alen = crypto_aead_authsize(aead);
705 
706 	esp.tfclen = 0;
707 	if (x->tfcpad) {
708 		struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
709 		u32 padto;
710 
711 		padto = min(x->tfcpad, xfrm_state_mtu(x, dst->child_mtu_cached));
712 		if (skb->len < padto)
713 			esp.tfclen = padto - skb->len;
714 	}
715 	blksize = ALIGN(crypto_aead_blocksize(aead), 4);
716 	esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
717 	esp.plen = esp.clen - skb->len - esp.tfclen;
718 	esp.tailen = esp.tfclen + esp.plen + alen;
719 
720 	esp.esph = ip_esp_hdr(skb);
721 
722 	esp.nfrags = esp6_output_head(x, skb, &esp);
723 	if (esp.nfrags < 0)
724 		return esp.nfrags;
725 
726 	esph = esp.esph;
727 	esph->spi = x->id.spi;
728 
729 	esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
730 	esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
731 			    ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
732 
733 	skb_push(skb, -skb_network_offset(skb));
734 
735 	return esp6_output_tail(x, skb, &esp);
736 }
737 
738 static inline int esp_remove_trailer(struct sk_buff *skb)
739 {
740 	struct xfrm_state *x = xfrm_input_state(skb);
741 	struct xfrm_offload *xo = xfrm_offload(skb);
742 	struct crypto_aead *aead = x->data;
743 	int alen, hlen, elen;
744 	int padlen, trimlen;
745 	__wsum csumdiff;
746 	u8 nexthdr[2];
747 	int ret;
748 
749 	alen = crypto_aead_authsize(aead);
750 	hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
751 	elen = skb->len - hlen;
752 
753 	if (xo && (xo->flags & XFRM_ESP_NO_TRAILER)) {
754 		ret = xo->proto;
755 		goto out;
756 	}
757 
758 	ret = skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2);
759 	BUG_ON(ret);
760 
761 	ret = -EINVAL;
762 	padlen = nexthdr[0];
763 	if (padlen + 2 + alen >= elen) {
764 		net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
765 				    padlen + 2, elen - alen);
766 		goto out;
767 	}
768 
769 	trimlen = alen + padlen + 2;
770 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
771 		csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
772 		skb->csum = csum_block_sub(skb->csum, csumdiff,
773 					   skb->len - trimlen);
774 	}
775 	pskb_trim(skb, skb->len - trimlen);
776 
777 	ret = nexthdr[1];
778 
779 out:
780 	return ret;
781 }
782 
783 int esp6_input_done2(struct sk_buff *skb, int err)
784 {
785 	struct xfrm_state *x = xfrm_input_state(skb);
786 	struct xfrm_offload *xo = xfrm_offload(skb);
787 	struct crypto_aead *aead = x->data;
788 	int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
789 	int hdr_len = skb_network_header_len(skb);
790 
791 	if (!xo || (xo && !(xo->flags & CRYPTO_DONE)))
792 		kfree(ESP_SKB_CB(skb)->tmp);
793 
794 	if (unlikely(err))
795 		goto out;
796 
797 	err = esp_remove_trailer(skb);
798 	if (unlikely(err < 0))
799 		goto out;
800 
801 	if (x->encap) {
802 		const struct ipv6hdr *ip6h = ipv6_hdr(skb);
803 		int offset = skb_network_offset(skb) + sizeof(*ip6h);
804 		struct xfrm_encap_tmpl *encap = x->encap;
805 		u8 nexthdr = ip6h->nexthdr;
806 		__be16 frag_off, source;
807 		struct udphdr *uh;
808 		struct tcphdr *th;
809 
810 		offset = ipv6_skip_exthdr(skb, offset, &nexthdr, &frag_off);
811 		uh = (void *)(skb->data + offset);
812 		th = (void *)(skb->data + offset);
813 		hdr_len += offset;
814 
815 		switch (x->encap->encap_type) {
816 		case TCP_ENCAP_ESPINTCP:
817 			source = th->source;
818 			break;
819 		case UDP_ENCAP_ESPINUDP:
820 		case UDP_ENCAP_ESPINUDP_NON_IKE:
821 			source = uh->source;
822 			break;
823 		default:
824 			WARN_ON_ONCE(1);
825 			err = -EINVAL;
826 			goto out;
827 		}
828 
829 		/*
830 		 * 1) if the NAT-T peer's IP or port changed then
831 		 *    advertize the change to the keying daemon.
832 		 *    This is an inbound SA, so just compare
833 		 *    SRC ports.
834 		 */
835 		if (!ipv6_addr_equal(&ip6h->saddr, &x->props.saddr.in6) ||
836 		    source != encap->encap_sport) {
837 			xfrm_address_t ipaddr;
838 
839 			memcpy(&ipaddr.a6, &ip6h->saddr.s6_addr, sizeof(ipaddr.a6));
840 			km_new_mapping(x, &ipaddr, source);
841 
842 			/* XXX: perhaps add an extra
843 			 * policy check here, to see
844 			 * if we should allow or
845 			 * reject a packet from a
846 			 * different source
847 			 * address/port.
848 			 */
849 		}
850 
851 		/*
852 		 * 2) ignore UDP/TCP checksums in case
853 		 *    of NAT-T in Transport Mode, or
854 		 *    perform other post-processing fixes
855 		 *    as per draft-ietf-ipsec-udp-encaps-06,
856 		 *    section 3.1.2
857 		 */
858 		if (x->props.mode == XFRM_MODE_TRANSPORT)
859 			skb->ip_summed = CHECKSUM_UNNECESSARY;
860 	}
861 
862 	skb_postpull_rcsum(skb, skb_network_header(skb),
863 			   skb_network_header_len(skb));
864 	skb_pull_rcsum(skb, hlen);
865 	if (x->props.mode == XFRM_MODE_TUNNEL)
866 		skb_reset_transport_header(skb);
867 	else
868 		skb_set_transport_header(skb, -hdr_len);
869 
870 	/* RFC4303: Drop dummy packets without any error */
871 	if (err == IPPROTO_NONE)
872 		err = -EINVAL;
873 
874 out:
875 	return err;
876 }
877 EXPORT_SYMBOL_GPL(esp6_input_done2);
878 
879 static void esp_input_done(struct crypto_async_request *base, int err)
880 {
881 	struct sk_buff *skb = base->data;
882 
883 	xfrm_input_resume(skb, esp6_input_done2(skb, err));
884 }
885 
886 static void esp_input_restore_header(struct sk_buff *skb)
887 {
888 	esp_restore_header(skb, 0);
889 	__skb_pull(skb, 4);
890 }
891 
892 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
893 {
894 	struct xfrm_state *x = xfrm_input_state(skb);
895 
896 	/* For ESN we move the header forward by 4 bytes to
897 	 * accomodate the high bits.  We will move it back after
898 	 * decryption.
899 	 */
900 	if ((x->props.flags & XFRM_STATE_ESN)) {
901 		struct ip_esp_hdr *esph = skb_push(skb, 4);
902 
903 		*seqhi = esph->spi;
904 		esph->spi = esph->seq_no;
905 		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
906 	}
907 }
908 
909 static void esp_input_done_esn(struct crypto_async_request *base, int err)
910 {
911 	struct sk_buff *skb = base->data;
912 
913 	esp_input_restore_header(skb);
914 	esp_input_done(base, err);
915 }
916 
917 static int esp6_input(struct xfrm_state *x, struct sk_buff *skb)
918 {
919 	struct crypto_aead *aead = x->data;
920 	struct aead_request *req;
921 	struct sk_buff *trailer;
922 	int ivlen = crypto_aead_ivsize(aead);
923 	int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
924 	int nfrags;
925 	int assoclen;
926 	int seqhilen;
927 	int ret = 0;
928 	void *tmp;
929 	__be32 *seqhi;
930 	u8 *iv;
931 	struct scatterlist *sg;
932 
933 	if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen)) {
934 		ret = -EINVAL;
935 		goto out;
936 	}
937 
938 	if (elen <= 0) {
939 		ret = -EINVAL;
940 		goto out;
941 	}
942 
943 	assoclen = sizeof(struct ip_esp_hdr);
944 	seqhilen = 0;
945 
946 	if (x->props.flags & XFRM_STATE_ESN) {
947 		seqhilen += sizeof(__be32);
948 		assoclen += seqhilen;
949 	}
950 
951 	if (!skb_cloned(skb)) {
952 		if (!skb_is_nonlinear(skb)) {
953 			nfrags = 1;
954 
955 			goto skip_cow;
956 		} else if (!skb_has_frag_list(skb)) {
957 			nfrags = skb_shinfo(skb)->nr_frags;
958 			nfrags++;
959 
960 			goto skip_cow;
961 		}
962 	}
963 
964 	nfrags = skb_cow_data(skb, 0, &trailer);
965 	if (nfrags < 0) {
966 		ret = -EINVAL;
967 		goto out;
968 	}
969 
970 skip_cow:
971 	ret = -ENOMEM;
972 	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
973 	if (!tmp)
974 		goto out;
975 
976 	ESP_SKB_CB(skb)->tmp = tmp;
977 	seqhi = esp_tmp_extra(tmp);
978 	iv = esp_tmp_iv(aead, tmp, seqhilen);
979 	req = esp_tmp_req(aead, iv);
980 	sg = esp_req_sg(aead, req);
981 
982 	esp_input_set_header(skb, seqhi);
983 
984 	sg_init_table(sg, nfrags);
985 	ret = skb_to_sgvec(skb, sg, 0, skb->len);
986 	if (unlikely(ret < 0)) {
987 		kfree(tmp);
988 		goto out;
989 	}
990 
991 	skb->ip_summed = CHECKSUM_NONE;
992 
993 	if ((x->props.flags & XFRM_STATE_ESN))
994 		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
995 	else
996 		aead_request_set_callback(req, 0, esp_input_done, skb);
997 
998 	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
999 	aead_request_set_ad(req, assoclen);
1000 
1001 	ret = crypto_aead_decrypt(req);
1002 	if (ret == -EINPROGRESS)
1003 		goto out;
1004 
1005 	if ((x->props.flags & XFRM_STATE_ESN))
1006 		esp_input_restore_header(skb);
1007 
1008 	ret = esp6_input_done2(skb, ret);
1009 
1010 out:
1011 	return ret;
1012 }
1013 
1014 static int esp6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
1015 		    u8 type, u8 code, int offset, __be32 info)
1016 {
1017 	struct net *net = dev_net(skb->dev);
1018 	const struct ipv6hdr *iph = (const struct ipv6hdr *)skb->data;
1019 	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data + offset);
1020 	struct xfrm_state *x;
1021 
1022 	if (type != ICMPV6_PKT_TOOBIG &&
1023 	    type != NDISC_REDIRECT)
1024 		return 0;
1025 
1026 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
1027 			      esph->spi, IPPROTO_ESP, AF_INET6);
1028 	if (!x)
1029 		return 0;
1030 
1031 	if (type == NDISC_REDIRECT)
1032 		ip6_redirect(skb, net, skb->dev->ifindex, 0,
1033 			     sock_net_uid(net, NULL));
1034 	else
1035 		ip6_update_pmtu(skb, net, info, 0, 0, sock_net_uid(net, NULL));
1036 	xfrm_state_put(x);
1037 
1038 	return 0;
1039 }
1040 
1041 static void esp6_destroy(struct xfrm_state *x)
1042 {
1043 	struct crypto_aead *aead = x->data;
1044 
1045 	if (!aead)
1046 		return;
1047 
1048 	crypto_free_aead(aead);
1049 }
1050 
1051 static int esp_init_aead(struct xfrm_state *x)
1052 {
1053 	char aead_name[CRYPTO_MAX_ALG_NAME];
1054 	struct crypto_aead *aead;
1055 	int err;
1056 
1057 	err = -ENAMETOOLONG;
1058 	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1059 		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME)
1060 		goto error;
1061 
1062 	aead = crypto_alloc_aead(aead_name, 0, 0);
1063 	err = PTR_ERR(aead);
1064 	if (IS_ERR(aead))
1065 		goto error;
1066 
1067 	x->data = aead;
1068 
1069 	err = crypto_aead_setkey(aead, x->aead->alg_key,
1070 				 (x->aead->alg_key_len + 7) / 8);
1071 	if (err)
1072 		goto error;
1073 
1074 	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1075 	if (err)
1076 		goto error;
1077 
1078 error:
1079 	return err;
1080 }
1081 
1082 static int esp_init_authenc(struct xfrm_state *x)
1083 {
1084 	struct crypto_aead *aead;
1085 	struct crypto_authenc_key_param *param;
1086 	struct rtattr *rta;
1087 	char *key;
1088 	char *p;
1089 	char authenc_name[CRYPTO_MAX_ALG_NAME];
1090 	unsigned int keylen;
1091 	int err;
1092 
1093 	err = -EINVAL;
1094 	if (!x->ealg)
1095 		goto error;
1096 
1097 	err = -ENAMETOOLONG;
1098 
1099 	if ((x->props.flags & XFRM_STATE_ESN)) {
1100 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1101 			     "%s%sauthencesn(%s,%s)%s",
1102 			     x->geniv ?: "", x->geniv ? "(" : "",
1103 			     x->aalg ? x->aalg->alg_name : "digest_null",
1104 			     x->ealg->alg_name,
1105 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1106 			goto error;
1107 	} else {
1108 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1109 			     "%s%sauthenc(%s,%s)%s",
1110 			     x->geniv ?: "", x->geniv ? "(" : "",
1111 			     x->aalg ? x->aalg->alg_name : "digest_null",
1112 			     x->ealg->alg_name,
1113 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1114 			goto error;
1115 	}
1116 
1117 	aead = crypto_alloc_aead(authenc_name, 0, 0);
1118 	err = PTR_ERR(aead);
1119 	if (IS_ERR(aead))
1120 		goto error;
1121 
1122 	x->data = aead;
1123 
1124 	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1125 		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1126 	err = -ENOMEM;
1127 	key = kmalloc(keylen, GFP_KERNEL);
1128 	if (!key)
1129 		goto error;
1130 
1131 	p = key;
1132 	rta = (void *)p;
1133 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1134 	rta->rta_len = RTA_LENGTH(sizeof(*param));
1135 	param = RTA_DATA(rta);
1136 	p += RTA_SPACE(sizeof(*param));
1137 
1138 	if (x->aalg) {
1139 		struct xfrm_algo_desc *aalg_desc;
1140 
1141 		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1142 		p += (x->aalg->alg_key_len + 7) / 8;
1143 
1144 		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1145 		BUG_ON(!aalg_desc);
1146 
1147 		err = -EINVAL;
1148 		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1149 		    crypto_aead_authsize(aead)) {
1150 			pr_info("ESP: %s digestsize %u != %hu\n",
1151 				x->aalg->alg_name,
1152 				crypto_aead_authsize(aead),
1153 				aalg_desc->uinfo.auth.icv_fullbits / 8);
1154 			goto free_key;
1155 		}
1156 
1157 		err = crypto_aead_setauthsize(
1158 			aead, x->aalg->alg_trunc_len / 8);
1159 		if (err)
1160 			goto free_key;
1161 	}
1162 
1163 	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1164 	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1165 
1166 	err = crypto_aead_setkey(aead, key, keylen);
1167 
1168 free_key:
1169 	kfree(key);
1170 
1171 error:
1172 	return err;
1173 }
1174 
1175 static int esp6_init_state(struct xfrm_state *x)
1176 {
1177 	struct crypto_aead *aead;
1178 	u32 align;
1179 	int err;
1180 
1181 	x->data = NULL;
1182 
1183 	if (x->aead)
1184 		err = esp_init_aead(x);
1185 	else
1186 		err = esp_init_authenc(x);
1187 
1188 	if (err)
1189 		goto error;
1190 
1191 	aead = x->data;
1192 
1193 	x->props.header_len = sizeof(struct ip_esp_hdr) +
1194 			      crypto_aead_ivsize(aead);
1195 	switch (x->props.mode) {
1196 	case XFRM_MODE_BEET:
1197 		if (x->sel.family != AF_INET6)
1198 			x->props.header_len += IPV4_BEET_PHMAXLEN +
1199 					       (sizeof(struct ipv6hdr) - sizeof(struct iphdr));
1200 		break;
1201 	default:
1202 	case XFRM_MODE_TRANSPORT:
1203 		break;
1204 	case XFRM_MODE_TUNNEL:
1205 		x->props.header_len += sizeof(struct ipv6hdr);
1206 		break;
1207 	}
1208 
1209 	if (x->encap) {
1210 		struct xfrm_encap_tmpl *encap = x->encap;
1211 
1212 		switch (encap->encap_type) {
1213 		default:
1214 			err = -EINVAL;
1215 			goto error;
1216 		case UDP_ENCAP_ESPINUDP:
1217 			x->props.header_len += sizeof(struct udphdr);
1218 			break;
1219 		case UDP_ENCAP_ESPINUDP_NON_IKE:
1220 			x->props.header_len += sizeof(struct udphdr) + 2 * sizeof(u32);
1221 			break;
1222 #ifdef CONFIG_INET6_ESPINTCP
1223 		case TCP_ENCAP_ESPINTCP:
1224 			/* only the length field, TCP encap is done by
1225 			 * the socket
1226 			 */
1227 			x->props.header_len += 2;
1228 			break;
1229 #endif
1230 		}
1231 	}
1232 
1233 	align = ALIGN(crypto_aead_blocksize(aead), 4);
1234 	x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1235 
1236 error:
1237 	return err;
1238 }
1239 
1240 static int esp6_rcv_cb(struct sk_buff *skb, int err)
1241 {
1242 	return 0;
1243 }
1244 
1245 static const struct xfrm_type esp6_type = {
1246 	.description	= "ESP6",
1247 	.owner		= THIS_MODULE,
1248 	.proto		= IPPROTO_ESP,
1249 	.flags		= XFRM_TYPE_REPLAY_PROT,
1250 	.init_state	= esp6_init_state,
1251 	.destructor	= esp6_destroy,
1252 	.input		= esp6_input,
1253 	.output		= esp6_output,
1254 	.hdr_offset	= xfrm6_find_1stfragopt,
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