xref: /openbmc/linux/net/ipv6/esp6.c (revision 80d0624d)
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, struct sk_buff *skb)
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 			skb_page_unref(skb, sg_page(sg), false);
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 net *net = xs_net(x);
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(net, net->ipv4.tcp_death_row.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 	WARN_ON(1);
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(void *data, int err)
281 {
282 	struct sk_buff *skb = 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, skb);
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(void *data, int err)
371 {
372 	struct sk_buff *skb = data;
373 
374 	esp_output_restore_header(skb);
375 	esp_output_done(data, 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, skb);
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 	ret = pskb_trim(skb, skb->len - trimlen);
773 	if (unlikely(ret))
774 		return ret;
775 
776 	ret = nexthdr[1];
777 
778 out:
779 	return ret;
780 }
781 
782 int esp6_input_done2(struct sk_buff *skb, int err)
783 {
784 	struct xfrm_state *x = xfrm_input_state(skb);
785 	struct xfrm_offload *xo = xfrm_offload(skb);
786 	struct crypto_aead *aead = x->data;
787 	int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
788 	int hdr_len = skb_network_header_len(skb);
789 
790 	if (!xo || !(xo->flags & CRYPTO_DONE))
791 		kfree(ESP_SKB_CB(skb)->tmp);
792 
793 	if (unlikely(err))
794 		goto out;
795 
796 	err = esp_remove_trailer(skb);
797 	if (unlikely(err < 0))
798 		goto out;
799 
800 	if (x->encap) {
801 		const struct ipv6hdr *ip6h = ipv6_hdr(skb);
802 		int offset = skb_network_offset(skb) + sizeof(*ip6h);
803 		struct xfrm_encap_tmpl *encap = x->encap;
804 		u8 nexthdr = ip6h->nexthdr;
805 		__be16 frag_off, source;
806 		struct udphdr *uh;
807 		struct tcphdr *th;
808 
809 		offset = ipv6_skip_exthdr(skb, offset, &nexthdr, &frag_off);
810 		if (offset == -1) {
811 			err = -EINVAL;
812 			goto out;
813 		}
814 
815 		uh = (void *)(skb->data + offset);
816 		th = (void *)(skb->data + offset);
817 		hdr_len += offset;
818 
819 		switch (x->encap->encap_type) {
820 		case TCP_ENCAP_ESPINTCP:
821 			source = th->source;
822 			break;
823 		case UDP_ENCAP_ESPINUDP:
824 		case UDP_ENCAP_ESPINUDP_NON_IKE:
825 			source = uh->source;
826 			break;
827 		default:
828 			WARN_ON_ONCE(1);
829 			err = -EINVAL;
830 			goto out;
831 		}
832 
833 		/*
834 		 * 1) if the NAT-T peer's IP or port changed then
835 		 *    advertise the change to the keying daemon.
836 		 *    This is an inbound SA, so just compare
837 		 *    SRC ports.
838 		 */
839 		if (!ipv6_addr_equal(&ip6h->saddr, &x->props.saddr.in6) ||
840 		    source != encap->encap_sport) {
841 			xfrm_address_t ipaddr;
842 
843 			memcpy(&ipaddr.a6, &ip6h->saddr.s6_addr, sizeof(ipaddr.a6));
844 			km_new_mapping(x, &ipaddr, source);
845 
846 			/* XXX: perhaps add an extra
847 			 * policy check here, to see
848 			 * if we should allow or
849 			 * reject a packet from a
850 			 * different source
851 			 * address/port.
852 			 */
853 		}
854 
855 		/*
856 		 * 2) ignore UDP/TCP checksums in case
857 		 *    of NAT-T in Transport Mode, or
858 		 *    perform other post-processing fixes
859 		 *    as per draft-ietf-ipsec-udp-encaps-06,
860 		 *    section 3.1.2
861 		 */
862 		if (x->props.mode == XFRM_MODE_TRANSPORT)
863 			skb->ip_summed = CHECKSUM_UNNECESSARY;
864 	}
865 
866 	skb_postpull_rcsum(skb, skb_network_header(skb),
867 			   skb_network_header_len(skb));
868 	skb_pull_rcsum(skb, hlen);
869 	if (x->props.mode == XFRM_MODE_TUNNEL)
870 		skb_reset_transport_header(skb);
871 	else
872 		skb_set_transport_header(skb, -hdr_len);
873 
874 	/* RFC4303: Drop dummy packets without any error */
875 	if (err == IPPROTO_NONE)
876 		err = -EINVAL;
877 
878 out:
879 	return err;
880 }
881 EXPORT_SYMBOL_GPL(esp6_input_done2);
882 
883 static void esp_input_done(void *data, int err)
884 {
885 	struct sk_buff *skb = data;
886 
887 	xfrm_input_resume(skb, esp6_input_done2(skb, err));
888 }
889 
890 static void esp_input_restore_header(struct sk_buff *skb)
891 {
892 	esp_restore_header(skb, 0);
893 	__skb_pull(skb, 4);
894 }
895 
896 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
897 {
898 	struct xfrm_state *x = xfrm_input_state(skb);
899 
900 	/* For ESN we move the header forward by 4 bytes to
901 	 * accommodate the high bits.  We will move it back after
902 	 * decryption.
903 	 */
904 	if ((x->props.flags & XFRM_STATE_ESN)) {
905 		struct ip_esp_hdr *esph = skb_push(skb, 4);
906 
907 		*seqhi = esph->spi;
908 		esph->spi = esph->seq_no;
909 		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
910 	}
911 }
912 
913 static void esp_input_done_esn(void *data, int err)
914 {
915 	struct sk_buff *skb = data;
916 
917 	esp_input_restore_header(skb);
918 	esp_input_done(data, err);
919 }
920 
921 static int esp6_input(struct xfrm_state *x, struct sk_buff *skb)
922 {
923 	struct crypto_aead *aead = x->data;
924 	struct aead_request *req;
925 	struct sk_buff *trailer;
926 	int ivlen = crypto_aead_ivsize(aead);
927 	int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
928 	int nfrags;
929 	int assoclen;
930 	int seqhilen;
931 	int ret = 0;
932 	void *tmp;
933 	__be32 *seqhi;
934 	u8 *iv;
935 	struct scatterlist *sg;
936 
937 	if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen)) {
938 		ret = -EINVAL;
939 		goto out;
940 	}
941 
942 	if (elen <= 0) {
943 		ret = -EINVAL;
944 		goto out;
945 	}
946 
947 	assoclen = sizeof(struct ip_esp_hdr);
948 	seqhilen = 0;
949 
950 	if (x->props.flags & XFRM_STATE_ESN) {
951 		seqhilen += sizeof(__be32);
952 		assoclen += seqhilen;
953 	}
954 
955 	if (!skb_cloned(skb)) {
956 		if (!skb_is_nonlinear(skb)) {
957 			nfrags = 1;
958 
959 			goto skip_cow;
960 		} else if (!skb_has_frag_list(skb)) {
961 			nfrags = skb_shinfo(skb)->nr_frags;
962 			nfrags++;
963 
964 			goto skip_cow;
965 		}
966 	}
967 
968 	nfrags = skb_cow_data(skb, 0, &trailer);
969 	if (nfrags < 0) {
970 		ret = -EINVAL;
971 		goto out;
972 	}
973 
974 skip_cow:
975 	ret = -ENOMEM;
976 	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
977 	if (!tmp)
978 		goto out;
979 
980 	ESP_SKB_CB(skb)->tmp = tmp;
981 	seqhi = esp_tmp_extra(tmp);
982 	iv = esp_tmp_iv(aead, tmp, seqhilen);
983 	req = esp_tmp_req(aead, iv);
984 	sg = esp_req_sg(aead, req);
985 
986 	esp_input_set_header(skb, seqhi);
987 
988 	sg_init_table(sg, nfrags);
989 	ret = skb_to_sgvec(skb, sg, 0, skb->len);
990 	if (unlikely(ret < 0)) {
991 		kfree(tmp);
992 		goto out;
993 	}
994 
995 	skb->ip_summed = CHECKSUM_NONE;
996 
997 	if ((x->props.flags & XFRM_STATE_ESN))
998 		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
999 	else
1000 		aead_request_set_callback(req, 0, esp_input_done, skb);
1001 
1002 	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
1003 	aead_request_set_ad(req, assoclen);
1004 
1005 	ret = crypto_aead_decrypt(req);
1006 	if (ret == -EINPROGRESS)
1007 		goto out;
1008 
1009 	if ((x->props.flags & XFRM_STATE_ESN))
1010 		esp_input_restore_header(skb);
1011 
1012 	ret = esp6_input_done2(skb, ret);
1013 
1014 out:
1015 	return ret;
1016 }
1017 
1018 static int esp6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
1019 		    u8 type, u8 code, int offset, __be32 info)
1020 {
1021 	struct net *net = dev_net(skb->dev);
1022 	const struct ipv6hdr *iph = (const struct ipv6hdr *)skb->data;
1023 	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data + offset);
1024 	struct xfrm_state *x;
1025 
1026 	if (type != ICMPV6_PKT_TOOBIG &&
1027 	    type != NDISC_REDIRECT)
1028 		return 0;
1029 
1030 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
1031 			      esph->spi, IPPROTO_ESP, AF_INET6);
1032 	if (!x)
1033 		return 0;
1034 
1035 	if (type == NDISC_REDIRECT)
1036 		ip6_redirect(skb, net, skb->dev->ifindex, 0,
1037 			     sock_net_uid(net, NULL));
1038 	else
1039 		ip6_update_pmtu(skb, net, info, 0, 0, sock_net_uid(net, NULL));
1040 	xfrm_state_put(x);
1041 
1042 	return 0;
1043 }
1044 
1045 static void esp6_destroy(struct xfrm_state *x)
1046 {
1047 	struct crypto_aead *aead = x->data;
1048 
1049 	if (!aead)
1050 		return;
1051 
1052 	crypto_free_aead(aead);
1053 }
1054 
1055 static int esp_init_aead(struct xfrm_state *x, struct netlink_ext_ack *extack)
1056 {
1057 	char aead_name[CRYPTO_MAX_ALG_NAME];
1058 	struct crypto_aead *aead;
1059 	int err;
1060 
1061 	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1062 		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME) {
1063 		NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1064 		return -ENAMETOOLONG;
1065 	}
1066 
1067 	aead = crypto_alloc_aead(aead_name, 0, 0);
1068 	err = PTR_ERR(aead);
1069 	if (IS_ERR(aead))
1070 		goto error;
1071 
1072 	x->data = aead;
1073 
1074 	err = crypto_aead_setkey(aead, x->aead->alg_key,
1075 				 (x->aead->alg_key_len + 7) / 8);
1076 	if (err)
1077 		goto error;
1078 
1079 	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1080 	if (err)
1081 		goto error;
1082 
1083 	return 0;
1084 
1085 error:
1086 	NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1087 	return err;
1088 }
1089 
1090 static int esp_init_authenc(struct xfrm_state *x,
1091 			    struct netlink_ext_ack *extack)
1092 {
1093 	struct crypto_aead *aead;
1094 	struct crypto_authenc_key_param *param;
1095 	struct rtattr *rta;
1096 	char *key;
1097 	char *p;
1098 	char authenc_name[CRYPTO_MAX_ALG_NAME];
1099 	unsigned int keylen;
1100 	int err;
1101 
1102 	err = -ENAMETOOLONG;
1103 
1104 	if ((x->props.flags & XFRM_STATE_ESN)) {
1105 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1106 			     "%s%sauthencesn(%s,%s)%s",
1107 			     x->geniv ?: "", x->geniv ? "(" : "",
1108 			     x->aalg ? x->aalg->alg_name : "digest_null",
1109 			     x->ealg->alg_name,
1110 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1111 			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1112 			goto error;
1113 		}
1114 	} else {
1115 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1116 			     "%s%sauthenc(%s,%s)%s",
1117 			     x->geniv ?: "", x->geniv ? "(" : "",
1118 			     x->aalg ? x->aalg->alg_name : "digest_null",
1119 			     x->ealg->alg_name,
1120 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1121 			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1122 			goto error;
1123 		}
1124 	}
1125 
1126 	aead = crypto_alloc_aead(authenc_name, 0, 0);
1127 	err = PTR_ERR(aead);
1128 	if (IS_ERR(aead)) {
1129 		NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1130 		goto error;
1131 	}
1132 
1133 	x->data = aead;
1134 
1135 	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1136 		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1137 	err = -ENOMEM;
1138 	key = kmalloc(keylen, GFP_KERNEL);
1139 	if (!key)
1140 		goto error;
1141 
1142 	p = key;
1143 	rta = (void *)p;
1144 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1145 	rta->rta_len = RTA_LENGTH(sizeof(*param));
1146 	param = RTA_DATA(rta);
1147 	p += RTA_SPACE(sizeof(*param));
1148 
1149 	if (x->aalg) {
1150 		struct xfrm_algo_desc *aalg_desc;
1151 
1152 		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1153 		p += (x->aalg->alg_key_len + 7) / 8;
1154 
1155 		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1156 		BUG_ON(!aalg_desc);
1157 
1158 		err = -EINVAL;
1159 		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1160 		    crypto_aead_authsize(aead)) {
1161 			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1162 			goto free_key;
1163 		}
1164 
1165 		err = crypto_aead_setauthsize(
1166 			aead, x->aalg->alg_trunc_len / 8);
1167 		if (err) {
1168 			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1169 			goto free_key;
1170 		}
1171 	}
1172 
1173 	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1174 	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1175 
1176 	err = crypto_aead_setkey(aead, key, keylen);
1177 
1178 free_key:
1179 	kfree(key);
1180 
1181 error:
1182 	return err;
1183 }
1184 
1185 static int esp6_init_state(struct xfrm_state *x, struct netlink_ext_ack *extack)
1186 {
1187 	struct crypto_aead *aead;
1188 	u32 align;
1189 	int err;
1190 
1191 	x->data = NULL;
1192 
1193 	if (x->aead) {
1194 		err = esp_init_aead(x, extack);
1195 	} else if (x->ealg) {
1196 		err = esp_init_authenc(x, extack);
1197 	} else {
1198 		NL_SET_ERR_MSG(extack, "ESP: AEAD or CRYPT must be provided");
1199 		err = -EINVAL;
1200 	}
1201 
1202 	if (err)
1203 		goto error;
1204 
1205 	aead = x->data;
1206 
1207 	x->props.header_len = sizeof(struct ip_esp_hdr) +
1208 			      crypto_aead_ivsize(aead);
1209 	switch (x->props.mode) {
1210 	case XFRM_MODE_BEET:
1211 		if (x->sel.family != AF_INET6)
1212 			x->props.header_len += IPV4_BEET_PHMAXLEN +
1213 					       (sizeof(struct ipv6hdr) - sizeof(struct iphdr));
1214 		break;
1215 	default:
1216 	case XFRM_MODE_TRANSPORT:
1217 		break;
1218 	case XFRM_MODE_TUNNEL:
1219 		x->props.header_len += sizeof(struct ipv6hdr);
1220 		break;
1221 	}
1222 
1223 	if (x->encap) {
1224 		struct xfrm_encap_tmpl *encap = x->encap;
1225 
1226 		switch (encap->encap_type) {
1227 		default:
1228 			NL_SET_ERR_MSG(extack, "Unsupported encapsulation type for ESP");
1229 			err = -EINVAL;
1230 			goto error;
1231 		case UDP_ENCAP_ESPINUDP:
1232 			x->props.header_len += sizeof(struct udphdr);
1233 			break;
1234 		case UDP_ENCAP_ESPINUDP_NON_IKE:
1235 			x->props.header_len += sizeof(struct udphdr) + 2 * sizeof(u32);
1236 			break;
1237 #ifdef CONFIG_INET6_ESPINTCP
1238 		case TCP_ENCAP_ESPINTCP:
1239 			/* only the length field, TCP encap is done by
1240 			 * the socket
1241 			 */
1242 			x->props.header_len += 2;
1243 			break;
1244 #endif
1245 		}
1246 	}
1247 
1248 	align = ALIGN(crypto_aead_blocksize(aead), 4);
1249 	x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1250 
1251 error:
1252 	return err;
1253 }
1254 
1255 static int esp6_rcv_cb(struct sk_buff *skb, int err)
1256 {
1257 	return 0;
1258 }
1259 
1260 static const struct xfrm_type esp6_type = {
1261 	.owner		= THIS_MODULE,
1262 	.proto		= IPPROTO_ESP,
1263 	.flags		= XFRM_TYPE_REPLAY_PROT,
1264 	.init_state	= esp6_init_state,
1265 	.destructor	= esp6_destroy,
1266 	.input		= esp6_input,
1267 	.output		= esp6_output,
1268 };
1269 
1270 static struct xfrm6_protocol esp6_protocol = {
1271 	.handler	=	xfrm6_rcv,
1272 	.input_handler	=	xfrm_input,
1273 	.cb_handler	=	esp6_rcv_cb,
1274 	.err_handler	=	esp6_err,
1275 	.priority	=	0,
1276 };
1277 
1278 static int __init esp6_init(void)
1279 {
1280 	if (xfrm_register_type(&esp6_type, AF_INET6) < 0) {
1281 		pr_info("%s: can't add xfrm type\n", __func__);
1282 		return -EAGAIN;
1283 	}
1284 	if (xfrm6_protocol_register(&esp6_protocol, IPPROTO_ESP) < 0) {
1285 		pr_info("%s: can't add protocol\n", __func__);
1286 		xfrm_unregister_type(&esp6_type, AF_INET6);
1287 		return -EAGAIN;
1288 	}
1289 
1290 	return 0;
1291 }
1292 
1293 static void __exit esp6_fini(void)
1294 {
1295 	if (xfrm6_protocol_deregister(&esp6_protocol, IPPROTO_ESP) < 0)
1296 		pr_info("%s: can't remove protocol\n", __func__);
1297 	xfrm_unregister_type(&esp6_type, AF_INET6);
1298 }
1299 
1300 module_init(esp6_init);
1301 module_exit(esp6_fini);
1302 
1303 MODULE_LICENSE("GPL");
1304 MODULE_ALIAS_XFRM_TYPE(AF_INET6, XFRM_PROTO_ESP);
1305