xref: /openbmc/linux/net/ipv4/esp4.c (revision 8047f98c)
1 // SPDX-License-Identifier: GPL-2.0-only
2 #define pr_fmt(fmt) "IPsec: " fmt
3 
4 #include <crypto/aead.h>
5 #include <crypto/authenc.h>
6 #include <linux/err.h>
7 #include <linux/module.h>
8 #include <net/ip.h>
9 #include <net/xfrm.h>
10 #include <net/esp.h>
11 #include <linux/scatterlist.h>
12 #include <linux/kernel.h>
13 #include <linux/pfkeyv2.h>
14 #include <linux/rtnetlink.h>
15 #include <linux/slab.h>
16 #include <linux/spinlock.h>
17 #include <linux/in6.h>
18 #include <net/icmp.h>
19 #include <net/protocol.h>
20 #include <net/udp.h>
21 #include <net/tcp.h>
22 #include <net/espintcp.h>
23 
24 #include <linux/highmem.h>
25 
26 struct esp_skb_cb {
27 	struct xfrm_skb_cb xfrm;
28 	void *tmp;
29 };
30 
31 struct esp_output_extra {
32 	__be32 seqhi;
33 	u32 esphoff;
34 };
35 
36 #define ESP_SKB_CB(__skb) ((struct esp_skb_cb *)&((__skb)->cb[0]))
37 
38 /*
39  * Allocate an AEAD request structure with extra space for SG and IV.
40  *
41  * For alignment considerations the IV is placed at the front, followed
42  * by the request and finally the SG list.
43  *
44  * TODO: Use spare space in skb for this where possible.
45  */
46 static void *esp_alloc_tmp(struct crypto_aead *aead, int nfrags, int extralen)
47 {
48 	unsigned int len;
49 
50 	len = extralen;
51 
52 	len += crypto_aead_ivsize(aead);
53 
54 	if (len) {
55 		len += crypto_aead_alignmask(aead) &
56 		       ~(crypto_tfm_ctx_alignment() - 1);
57 		len = ALIGN(len, crypto_tfm_ctx_alignment());
58 	}
59 
60 	len += sizeof(struct aead_request) + crypto_aead_reqsize(aead);
61 	len = ALIGN(len, __alignof__(struct scatterlist));
62 
63 	len += sizeof(struct scatterlist) * nfrags;
64 
65 	return kmalloc(len, GFP_ATOMIC);
66 }
67 
68 static inline void *esp_tmp_extra(void *tmp)
69 {
70 	return PTR_ALIGN(tmp, __alignof__(struct esp_output_extra));
71 }
72 
73 static inline u8 *esp_tmp_iv(struct crypto_aead *aead, void *tmp, int extralen)
74 {
75 	return crypto_aead_ivsize(aead) ?
76 	       PTR_ALIGN((u8 *)tmp + extralen,
77 			 crypto_aead_alignmask(aead) + 1) : tmp + extralen;
78 }
79 
80 static inline struct aead_request *esp_tmp_req(struct crypto_aead *aead, u8 *iv)
81 {
82 	struct aead_request *req;
83 
84 	req = (void *)PTR_ALIGN(iv + crypto_aead_ivsize(aead),
85 				crypto_tfm_ctx_alignment());
86 	aead_request_set_tfm(req, aead);
87 	return req;
88 }
89 
90 static inline struct scatterlist *esp_req_sg(struct crypto_aead *aead,
91 					     struct aead_request *req)
92 {
93 	return (void *)ALIGN((unsigned long)(req + 1) +
94 			     crypto_aead_reqsize(aead),
95 			     __alignof__(struct scatterlist));
96 }
97 
98 static void esp_ssg_unref(struct xfrm_state *x, void *tmp)
99 {
100 	struct crypto_aead *aead = x->data;
101 	int extralen = 0;
102 	u8 *iv;
103 	struct aead_request *req;
104 	struct scatterlist *sg;
105 
106 	if (x->props.flags & XFRM_STATE_ESN)
107 		extralen += sizeof(struct esp_output_extra);
108 
109 	iv = esp_tmp_iv(aead, tmp, extralen);
110 	req = esp_tmp_req(aead, iv);
111 
112 	/* Unref skb_frag_pages in the src scatterlist if necessary.
113 	 * Skip the first sg which comes from skb->data.
114 	 */
115 	if (req->src != req->dst)
116 		for (sg = sg_next(req->src); sg; sg = sg_next(sg))
117 			put_page(sg_page(sg));
118 }
119 
120 #ifdef CONFIG_INET_ESPINTCP
121 struct esp_tcp_sk {
122 	struct sock *sk;
123 	struct rcu_head rcu;
124 };
125 
126 static void esp_free_tcp_sk(struct rcu_head *head)
127 {
128 	struct esp_tcp_sk *esk = container_of(head, struct esp_tcp_sk, rcu);
129 
130 	sock_put(esk->sk);
131 	kfree(esk);
132 }
133 
134 static struct sock *esp_find_tcp_sk(struct xfrm_state *x)
135 {
136 	struct xfrm_encap_tmpl *encap = x->encap;
137 	struct esp_tcp_sk *esk;
138 	__be16 sport, dport;
139 	struct sock *nsk;
140 	struct sock *sk;
141 
142 	sk = rcu_dereference(x->encap_sk);
143 	if (sk && sk->sk_state == TCP_ESTABLISHED)
144 		return sk;
145 
146 	spin_lock_bh(&x->lock);
147 	sport = encap->encap_sport;
148 	dport = encap->encap_dport;
149 	nsk = rcu_dereference_protected(x->encap_sk,
150 					lockdep_is_held(&x->lock));
151 	if (sk && sk == nsk) {
152 		esk = kmalloc(sizeof(*esk), GFP_ATOMIC);
153 		if (!esk) {
154 			spin_unlock_bh(&x->lock);
155 			return ERR_PTR(-ENOMEM);
156 		}
157 		RCU_INIT_POINTER(x->encap_sk, NULL);
158 		esk->sk = sk;
159 		call_rcu(&esk->rcu, esp_free_tcp_sk);
160 	}
161 	spin_unlock_bh(&x->lock);
162 
163 	sk = inet_lookup_established(xs_net(x), &tcp_hashinfo, x->id.daddr.a4,
164 				     dport, x->props.saddr.a4, sport, 0);
165 	if (!sk)
166 		return ERR_PTR(-ENOENT);
167 
168 	if (!tcp_is_ulp_esp(sk)) {
169 		sock_put(sk);
170 		return ERR_PTR(-EINVAL);
171 	}
172 
173 	spin_lock_bh(&x->lock);
174 	nsk = rcu_dereference_protected(x->encap_sk,
175 					lockdep_is_held(&x->lock));
176 	if (encap->encap_sport != sport ||
177 	    encap->encap_dport != dport) {
178 		sock_put(sk);
179 		sk = nsk ?: ERR_PTR(-EREMCHG);
180 	} else if (sk == nsk) {
181 		sock_put(sk);
182 	} else {
183 		rcu_assign_pointer(x->encap_sk, sk);
184 	}
185 	spin_unlock_bh(&x->lock);
186 
187 	return sk;
188 }
189 
190 static int esp_output_tcp_finish(struct xfrm_state *x, struct sk_buff *skb)
191 {
192 	struct sock *sk;
193 	int err;
194 
195 	rcu_read_lock();
196 
197 	sk = esp_find_tcp_sk(x);
198 	err = PTR_ERR_OR_ZERO(sk);
199 	if (err)
200 		goto out;
201 
202 	bh_lock_sock(sk);
203 	if (sock_owned_by_user(sk))
204 		err = espintcp_queue_out(sk, skb);
205 	else
206 		err = espintcp_push_skb(sk, skb);
207 	bh_unlock_sock(sk);
208 
209 out:
210 	rcu_read_unlock();
211 	return err;
212 }
213 
214 static int esp_output_tcp_encap_cb(struct net *net, struct sock *sk,
215 				   struct sk_buff *skb)
216 {
217 	struct dst_entry *dst = skb_dst(skb);
218 	struct xfrm_state *x = dst->xfrm;
219 
220 	return esp_output_tcp_finish(x, skb);
221 }
222 
223 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
224 {
225 	int err;
226 
227 	local_bh_disable();
228 	err = xfrm_trans_queue_net(xs_net(x), skb, esp_output_tcp_encap_cb);
229 	local_bh_enable();
230 
231 	/* EINPROGRESS just happens to do the right thing.  It
232 	 * actually means that the skb has been consumed and
233 	 * isn't coming back.
234 	 */
235 	return err ?: -EINPROGRESS;
236 }
237 #else
238 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
239 {
240 	kfree_skb(skb);
241 
242 	return -EOPNOTSUPP;
243 }
244 #endif
245 
246 static void esp_output_done(struct crypto_async_request *base, int err)
247 {
248 	struct sk_buff *skb = base->data;
249 	struct xfrm_offload *xo = xfrm_offload(skb);
250 	void *tmp;
251 	struct xfrm_state *x;
252 
253 	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
254 		struct sec_path *sp = skb_sec_path(skb);
255 
256 		x = sp->xvec[sp->len - 1];
257 	} else {
258 		x = skb_dst(skb)->xfrm;
259 	}
260 
261 	tmp = ESP_SKB_CB(skb)->tmp;
262 	esp_ssg_unref(x, tmp);
263 	kfree(tmp);
264 
265 	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
266 		if (err) {
267 			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
268 			kfree_skb(skb);
269 			return;
270 		}
271 
272 		skb_push(skb, skb->data - skb_mac_header(skb));
273 		secpath_reset(skb);
274 		xfrm_dev_resume(skb);
275 	} else {
276 		if (!err &&
277 		    x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
278 			esp_output_tail_tcp(x, skb);
279 		else
280 			xfrm_output_resume(skb->sk, skb, err);
281 	}
282 }
283 
284 /* Move ESP header back into place. */
285 static void esp_restore_header(struct sk_buff *skb, unsigned int offset)
286 {
287 	struct ip_esp_hdr *esph = (void *)(skb->data + offset);
288 	void *tmp = ESP_SKB_CB(skb)->tmp;
289 	__be32 *seqhi = esp_tmp_extra(tmp);
290 
291 	esph->seq_no = esph->spi;
292 	esph->spi = *seqhi;
293 }
294 
295 static void esp_output_restore_header(struct sk_buff *skb)
296 {
297 	void *tmp = ESP_SKB_CB(skb)->tmp;
298 	struct esp_output_extra *extra = esp_tmp_extra(tmp);
299 
300 	esp_restore_header(skb, skb_transport_offset(skb) + extra->esphoff -
301 				sizeof(__be32));
302 }
303 
304 static struct ip_esp_hdr *esp_output_set_extra(struct sk_buff *skb,
305 					       struct xfrm_state *x,
306 					       struct ip_esp_hdr *esph,
307 					       struct esp_output_extra *extra)
308 {
309 	/* For ESN we move the header forward by 4 bytes to
310 	 * accommodate the high bits.  We will move it back after
311 	 * encryption.
312 	 */
313 	if ((x->props.flags & XFRM_STATE_ESN)) {
314 		__u32 seqhi;
315 		struct xfrm_offload *xo = xfrm_offload(skb);
316 
317 		if (xo)
318 			seqhi = xo->seq.hi;
319 		else
320 			seqhi = XFRM_SKB_CB(skb)->seq.output.hi;
321 
322 		extra->esphoff = (unsigned char *)esph -
323 				 skb_transport_header(skb);
324 		esph = (struct ip_esp_hdr *)((unsigned char *)esph - 4);
325 		extra->seqhi = esph->spi;
326 		esph->seq_no = htonl(seqhi);
327 	}
328 
329 	esph->spi = x->id.spi;
330 
331 	return esph;
332 }
333 
334 static void esp_output_done_esn(struct crypto_async_request *base, int err)
335 {
336 	struct sk_buff *skb = base->data;
337 
338 	esp_output_restore_header(skb);
339 	esp_output_done(base, err);
340 }
341 
342 static struct ip_esp_hdr *esp_output_udp_encap(struct sk_buff *skb,
343 					       int encap_type,
344 					       struct esp_info *esp,
345 					       __be16 sport,
346 					       __be16 dport)
347 {
348 	struct udphdr *uh;
349 	__be32 *udpdata32;
350 	unsigned int len;
351 
352 	len = skb->len + esp->tailen - skb_transport_offset(skb);
353 	if (len + sizeof(struct iphdr) > IP_MAX_MTU)
354 		return ERR_PTR(-EMSGSIZE);
355 
356 	uh = (struct udphdr *)esp->esph;
357 	uh->source = sport;
358 	uh->dest = dport;
359 	uh->len = htons(len);
360 	uh->check = 0;
361 
362 	*skb_mac_header(skb) = IPPROTO_UDP;
363 
364 	if (encap_type == UDP_ENCAP_ESPINUDP_NON_IKE) {
365 		udpdata32 = (__be32 *)(uh + 1);
366 		udpdata32[0] = udpdata32[1] = 0;
367 		return (struct ip_esp_hdr *)(udpdata32 + 2);
368 	}
369 
370 	return (struct ip_esp_hdr *)(uh + 1);
371 }
372 
373 #ifdef CONFIG_INET_ESPINTCP
374 static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
375 						    struct sk_buff *skb,
376 						    struct esp_info *esp)
377 {
378 	__be16 *lenp = (void *)esp->esph;
379 	struct ip_esp_hdr *esph;
380 	unsigned int len;
381 	struct sock *sk;
382 
383 	len = skb->len + esp->tailen - skb_transport_offset(skb);
384 	if (len > IP_MAX_MTU)
385 		return ERR_PTR(-EMSGSIZE);
386 
387 	rcu_read_lock();
388 	sk = esp_find_tcp_sk(x);
389 	rcu_read_unlock();
390 
391 	if (IS_ERR(sk))
392 		return ERR_CAST(sk);
393 
394 	*lenp = htons(len);
395 	esph = (struct ip_esp_hdr *)(lenp + 1);
396 
397 	return esph;
398 }
399 #else
400 static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
401 						    struct sk_buff *skb,
402 						    struct esp_info *esp)
403 {
404 	return ERR_PTR(-EOPNOTSUPP);
405 }
406 #endif
407 
408 static int esp_output_encap(struct xfrm_state *x, struct sk_buff *skb,
409 			    struct esp_info *esp)
410 {
411 	struct xfrm_encap_tmpl *encap = x->encap;
412 	struct ip_esp_hdr *esph;
413 	__be16 sport, dport;
414 	int encap_type;
415 
416 	spin_lock_bh(&x->lock);
417 	sport = encap->encap_sport;
418 	dport = encap->encap_dport;
419 	encap_type = encap->encap_type;
420 	spin_unlock_bh(&x->lock);
421 
422 	switch (encap_type) {
423 	default:
424 	case UDP_ENCAP_ESPINUDP:
425 	case UDP_ENCAP_ESPINUDP_NON_IKE:
426 		esph = esp_output_udp_encap(skb, encap_type, esp, sport, dport);
427 		break;
428 	case TCP_ENCAP_ESPINTCP:
429 		esph = esp_output_tcp_encap(x, skb, esp);
430 		break;
431 	}
432 
433 	if (IS_ERR(esph))
434 		return PTR_ERR(esph);
435 
436 	esp->esph = esph;
437 
438 	return 0;
439 }
440 
441 int esp_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
442 {
443 	u8 *tail;
444 	int nfrags;
445 	int esph_offset;
446 	struct page *page;
447 	struct sk_buff *trailer;
448 	int tailen = esp->tailen;
449 	unsigned int allocsz;
450 
451 	/* this is non-NULL only with TCP/UDP Encapsulation */
452 	if (x->encap) {
453 		int err = esp_output_encap(x, skb, esp);
454 
455 		if (err < 0)
456 			return err;
457 	}
458 
459 	allocsz = ALIGN(skb->data_len + tailen, L1_CACHE_BYTES);
460 	if (allocsz > ESP_SKB_FRAG_MAXSIZE)
461 		goto cow;
462 
463 	if (!skb_cloned(skb)) {
464 		if (tailen <= skb_tailroom(skb)) {
465 			nfrags = 1;
466 			trailer = skb;
467 			tail = skb_tail_pointer(trailer);
468 
469 			goto skip_cow;
470 		} else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
471 			   && !skb_has_frag_list(skb)) {
472 			int allocsize;
473 			struct sock *sk = skb->sk;
474 			struct page_frag *pfrag = &x->xfrag;
475 
476 			esp->inplace = false;
477 
478 			allocsize = ALIGN(tailen, L1_CACHE_BYTES);
479 
480 			spin_lock_bh(&x->lock);
481 
482 			if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
483 				spin_unlock_bh(&x->lock);
484 				goto cow;
485 			}
486 
487 			page = pfrag->page;
488 			get_page(page);
489 
490 			tail = page_address(page) + pfrag->offset;
491 
492 			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
493 
494 			nfrags = skb_shinfo(skb)->nr_frags;
495 
496 			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
497 					     tailen);
498 			skb_shinfo(skb)->nr_frags = ++nfrags;
499 
500 			pfrag->offset = pfrag->offset + allocsize;
501 
502 			spin_unlock_bh(&x->lock);
503 
504 			nfrags++;
505 
506 			skb->len += tailen;
507 			skb->data_len += tailen;
508 			skb->truesize += tailen;
509 			if (sk && sk_fullsock(sk))
510 				refcount_add(tailen, &sk->sk_wmem_alloc);
511 
512 			goto out;
513 		}
514 	}
515 
516 cow:
517 	esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
518 
519 	nfrags = skb_cow_data(skb, tailen, &trailer);
520 	if (nfrags < 0)
521 		goto out;
522 	tail = skb_tail_pointer(trailer);
523 	esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
524 
525 skip_cow:
526 	esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
527 	pskb_put(skb, trailer, tailen);
528 
529 out:
530 	return nfrags;
531 }
532 EXPORT_SYMBOL_GPL(esp_output_head);
533 
534 int esp_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
535 {
536 	u8 *iv;
537 	int alen;
538 	void *tmp;
539 	int ivlen;
540 	int assoclen;
541 	int extralen;
542 	struct page *page;
543 	struct ip_esp_hdr *esph;
544 	struct crypto_aead *aead;
545 	struct aead_request *req;
546 	struct scatterlist *sg, *dsg;
547 	struct esp_output_extra *extra;
548 	int err = -ENOMEM;
549 
550 	assoclen = sizeof(struct ip_esp_hdr);
551 	extralen = 0;
552 
553 	if (x->props.flags & XFRM_STATE_ESN) {
554 		extralen += sizeof(*extra);
555 		assoclen += sizeof(__be32);
556 	}
557 
558 	aead = x->data;
559 	alen = crypto_aead_authsize(aead);
560 	ivlen = crypto_aead_ivsize(aead);
561 
562 	tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
563 	if (!tmp)
564 		goto error;
565 
566 	extra = esp_tmp_extra(tmp);
567 	iv = esp_tmp_iv(aead, tmp, extralen);
568 	req = esp_tmp_req(aead, iv);
569 	sg = esp_req_sg(aead, req);
570 
571 	if (esp->inplace)
572 		dsg = sg;
573 	else
574 		dsg = &sg[esp->nfrags];
575 
576 	esph = esp_output_set_extra(skb, x, esp->esph, extra);
577 	esp->esph = esph;
578 
579 	sg_init_table(sg, esp->nfrags);
580 	err = skb_to_sgvec(skb, sg,
581 		           (unsigned char *)esph - skb->data,
582 		           assoclen + ivlen + esp->clen + alen);
583 	if (unlikely(err < 0))
584 		goto error_free;
585 
586 	if (!esp->inplace) {
587 		int allocsize;
588 		struct page_frag *pfrag = &x->xfrag;
589 
590 		allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
591 
592 		spin_lock_bh(&x->lock);
593 		if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
594 			spin_unlock_bh(&x->lock);
595 			goto error_free;
596 		}
597 
598 		skb_shinfo(skb)->nr_frags = 1;
599 
600 		page = pfrag->page;
601 		get_page(page);
602 		/* replace page frags in skb with new page */
603 		__skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
604 		pfrag->offset = pfrag->offset + allocsize;
605 		spin_unlock_bh(&x->lock);
606 
607 		sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
608 		err = skb_to_sgvec(skb, dsg,
609 			           (unsigned char *)esph - skb->data,
610 			           assoclen + ivlen + esp->clen + alen);
611 		if (unlikely(err < 0))
612 			goto error_free;
613 	}
614 
615 	if ((x->props.flags & XFRM_STATE_ESN))
616 		aead_request_set_callback(req, 0, esp_output_done_esn, skb);
617 	else
618 		aead_request_set_callback(req, 0, esp_output_done, skb);
619 
620 	aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
621 	aead_request_set_ad(req, assoclen);
622 
623 	memset(iv, 0, ivlen);
624 	memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
625 	       min(ivlen, 8));
626 
627 	ESP_SKB_CB(skb)->tmp = tmp;
628 	err = crypto_aead_encrypt(req);
629 
630 	switch (err) {
631 	case -EINPROGRESS:
632 		goto error;
633 
634 	case -ENOSPC:
635 		err = NET_XMIT_DROP;
636 		break;
637 
638 	case 0:
639 		if ((x->props.flags & XFRM_STATE_ESN))
640 			esp_output_restore_header(skb);
641 	}
642 
643 	if (sg != dsg)
644 		esp_ssg_unref(x, tmp);
645 
646 	if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
647 		err = esp_output_tail_tcp(x, skb);
648 
649 error_free:
650 	kfree(tmp);
651 error:
652 	return err;
653 }
654 EXPORT_SYMBOL_GPL(esp_output_tail);
655 
656 static int esp_output(struct xfrm_state *x, struct sk_buff *skb)
657 {
658 	int alen;
659 	int blksize;
660 	struct ip_esp_hdr *esph;
661 	struct crypto_aead *aead;
662 	struct esp_info esp;
663 
664 	esp.inplace = true;
665 
666 	esp.proto = *skb_mac_header(skb);
667 	*skb_mac_header(skb) = IPPROTO_ESP;
668 
669 	/* skb is pure payload to encrypt */
670 
671 	aead = x->data;
672 	alen = crypto_aead_authsize(aead);
673 
674 	esp.tfclen = 0;
675 	if (x->tfcpad) {
676 		struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
677 		u32 padto;
678 
679 		padto = min(x->tfcpad, xfrm_state_mtu(x, dst->child_mtu_cached));
680 		if (skb->len < padto)
681 			esp.tfclen = padto - skb->len;
682 	}
683 	blksize = ALIGN(crypto_aead_blocksize(aead), 4);
684 	esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
685 	esp.plen = esp.clen - skb->len - esp.tfclen;
686 	esp.tailen = esp.tfclen + esp.plen + alen;
687 
688 	esp.esph = ip_esp_hdr(skb);
689 
690 	esp.nfrags = esp_output_head(x, skb, &esp);
691 	if (esp.nfrags < 0)
692 		return esp.nfrags;
693 
694 	esph = esp.esph;
695 	esph->spi = x->id.spi;
696 
697 	esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
698 	esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
699 				 ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
700 
701 	skb_push(skb, -skb_network_offset(skb));
702 
703 	return esp_output_tail(x, skb, &esp);
704 }
705 
706 static inline int esp_remove_trailer(struct sk_buff *skb)
707 {
708 	struct xfrm_state *x = xfrm_input_state(skb);
709 	struct xfrm_offload *xo = xfrm_offload(skb);
710 	struct crypto_aead *aead = x->data;
711 	int alen, hlen, elen;
712 	int padlen, trimlen;
713 	__wsum csumdiff;
714 	u8 nexthdr[2];
715 	int ret;
716 
717 	alen = crypto_aead_authsize(aead);
718 	hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
719 	elen = skb->len - hlen;
720 
721 	if (xo && (xo->flags & XFRM_ESP_NO_TRAILER)) {
722 		ret = xo->proto;
723 		goto out;
724 	}
725 
726 	if (skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2))
727 		BUG();
728 
729 	ret = -EINVAL;
730 	padlen = nexthdr[0];
731 	if (padlen + 2 + alen >= elen) {
732 		net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
733 				    padlen + 2, elen - alen);
734 		goto out;
735 	}
736 
737 	trimlen = alen + padlen + 2;
738 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
739 		csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
740 		skb->csum = csum_block_sub(skb->csum, csumdiff,
741 					   skb->len - trimlen);
742 	}
743 	pskb_trim(skb, skb->len - trimlen);
744 
745 	ret = nexthdr[1];
746 
747 out:
748 	return ret;
749 }
750 
751 int esp_input_done2(struct sk_buff *skb, int err)
752 {
753 	const struct iphdr *iph;
754 	struct xfrm_state *x = xfrm_input_state(skb);
755 	struct xfrm_offload *xo = xfrm_offload(skb);
756 	struct crypto_aead *aead = x->data;
757 	int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
758 	int ihl;
759 
760 	if (!xo || !(xo->flags & CRYPTO_DONE))
761 		kfree(ESP_SKB_CB(skb)->tmp);
762 
763 	if (unlikely(err))
764 		goto out;
765 
766 	err = esp_remove_trailer(skb);
767 	if (unlikely(err < 0))
768 		goto out;
769 
770 	iph = ip_hdr(skb);
771 	ihl = iph->ihl * 4;
772 
773 	if (x->encap) {
774 		struct xfrm_encap_tmpl *encap = x->encap;
775 		struct tcphdr *th = (void *)(skb_network_header(skb) + ihl);
776 		struct udphdr *uh = (void *)(skb_network_header(skb) + ihl);
777 		__be16 source;
778 
779 		switch (x->encap->encap_type) {
780 		case TCP_ENCAP_ESPINTCP:
781 			source = th->source;
782 			break;
783 		case UDP_ENCAP_ESPINUDP:
784 		case UDP_ENCAP_ESPINUDP_NON_IKE:
785 			source = uh->source;
786 			break;
787 		default:
788 			WARN_ON_ONCE(1);
789 			err = -EINVAL;
790 			goto out;
791 		}
792 
793 		/*
794 		 * 1) if the NAT-T peer's IP or port changed then
795 		 *    advertize the change to the keying daemon.
796 		 *    This is an inbound SA, so just compare
797 		 *    SRC ports.
798 		 */
799 		if (iph->saddr != x->props.saddr.a4 ||
800 		    source != encap->encap_sport) {
801 			xfrm_address_t ipaddr;
802 
803 			ipaddr.a4 = iph->saddr;
804 			km_new_mapping(x, &ipaddr, source);
805 
806 			/* XXX: perhaps add an extra
807 			 * policy check here, to see
808 			 * if we should allow or
809 			 * reject a packet from a
810 			 * different source
811 			 * address/port.
812 			 */
813 		}
814 
815 		/*
816 		 * 2) ignore UDP/TCP checksums in case
817 		 *    of NAT-T in Transport Mode, or
818 		 *    perform other post-processing fixes
819 		 *    as per draft-ietf-ipsec-udp-encaps-06,
820 		 *    section 3.1.2
821 		 */
822 		if (x->props.mode == XFRM_MODE_TRANSPORT)
823 			skb->ip_summed = CHECKSUM_UNNECESSARY;
824 	}
825 
826 	skb_pull_rcsum(skb, hlen);
827 	if (x->props.mode == XFRM_MODE_TUNNEL)
828 		skb_reset_transport_header(skb);
829 	else
830 		skb_set_transport_header(skb, -ihl);
831 
832 	/* RFC4303: Drop dummy packets without any error */
833 	if (err == IPPROTO_NONE)
834 		err = -EINVAL;
835 
836 out:
837 	return err;
838 }
839 EXPORT_SYMBOL_GPL(esp_input_done2);
840 
841 static void esp_input_done(struct crypto_async_request *base, int err)
842 {
843 	struct sk_buff *skb = base->data;
844 
845 	xfrm_input_resume(skb, esp_input_done2(skb, err));
846 }
847 
848 static void esp_input_restore_header(struct sk_buff *skb)
849 {
850 	esp_restore_header(skb, 0);
851 	__skb_pull(skb, 4);
852 }
853 
854 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
855 {
856 	struct xfrm_state *x = xfrm_input_state(skb);
857 	struct ip_esp_hdr *esph;
858 
859 	/* For ESN we move the header forward by 4 bytes to
860 	 * accommodate the high bits.  We will move it back after
861 	 * decryption.
862 	 */
863 	if ((x->props.flags & XFRM_STATE_ESN)) {
864 		esph = skb_push(skb, 4);
865 		*seqhi = esph->spi;
866 		esph->spi = esph->seq_no;
867 		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
868 	}
869 }
870 
871 static void esp_input_done_esn(struct crypto_async_request *base, int err)
872 {
873 	struct sk_buff *skb = base->data;
874 
875 	esp_input_restore_header(skb);
876 	esp_input_done(base, err);
877 }
878 
879 /*
880  * Note: detecting truncated vs. non-truncated authentication data is very
881  * expensive, so we only support truncated data, which is the recommended
882  * and common case.
883  */
884 static int esp_input(struct xfrm_state *x, struct sk_buff *skb)
885 {
886 	struct crypto_aead *aead = x->data;
887 	struct aead_request *req;
888 	struct sk_buff *trailer;
889 	int ivlen = crypto_aead_ivsize(aead);
890 	int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
891 	int nfrags;
892 	int assoclen;
893 	int seqhilen;
894 	__be32 *seqhi;
895 	void *tmp;
896 	u8 *iv;
897 	struct scatterlist *sg;
898 	int err = -EINVAL;
899 
900 	if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen))
901 		goto out;
902 
903 	if (elen <= 0)
904 		goto out;
905 
906 	assoclen = sizeof(struct ip_esp_hdr);
907 	seqhilen = 0;
908 
909 	if (x->props.flags & XFRM_STATE_ESN) {
910 		seqhilen += sizeof(__be32);
911 		assoclen += seqhilen;
912 	}
913 
914 	if (!skb_cloned(skb)) {
915 		if (!skb_is_nonlinear(skb)) {
916 			nfrags = 1;
917 
918 			goto skip_cow;
919 		} else if (!skb_has_frag_list(skb)) {
920 			nfrags = skb_shinfo(skb)->nr_frags;
921 			nfrags++;
922 
923 			goto skip_cow;
924 		}
925 	}
926 
927 	err = skb_cow_data(skb, 0, &trailer);
928 	if (err < 0)
929 		goto out;
930 
931 	nfrags = err;
932 
933 skip_cow:
934 	err = -ENOMEM;
935 	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
936 	if (!tmp)
937 		goto out;
938 
939 	ESP_SKB_CB(skb)->tmp = tmp;
940 	seqhi = esp_tmp_extra(tmp);
941 	iv = esp_tmp_iv(aead, tmp, seqhilen);
942 	req = esp_tmp_req(aead, iv);
943 	sg = esp_req_sg(aead, req);
944 
945 	esp_input_set_header(skb, seqhi);
946 
947 	sg_init_table(sg, nfrags);
948 	err = skb_to_sgvec(skb, sg, 0, skb->len);
949 	if (unlikely(err < 0)) {
950 		kfree(tmp);
951 		goto out;
952 	}
953 
954 	skb->ip_summed = CHECKSUM_NONE;
955 
956 	if ((x->props.flags & XFRM_STATE_ESN))
957 		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
958 	else
959 		aead_request_set_callback(req, 0, esp_input_done, skb);
960 
961 	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
962 	aead_request_set_ad(req, assoclen);
963 
964 	err = crypto_aead_decrypt(req);
965 	if (err == -EINPROGRESS)
966 		goto out;
967 
968 	if ((x->props.flags & XFRM_STATE_ESN))
969 		esp_input_restore_header(skb);
970 
971 	err = esp_input_done2(skb, err);
972 
973 out:
974 	return err;
975 }
976 
977 static int esp4_err(struct sk_buff *skb, u32 info)
978 {
979 	struct net *net = dev_net(skb->dev);
980 	const struct iphdr *iph = (const struct iphdr *)skb->data;
981 	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data+(iph->ihl<<2));
982 	struct xfrm_state *x;
983 
984 	switch (icmp_hdr(skb)->type) {
985 	case ICMP_DEST_UNREACH:
986 		if (icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
987 			return 0;
988 		break;
989 	case ICMP_REDIRECT:
990 		break;
991 	default:
992 		return 0;
993 	}
994 
995 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
996 			      esph->spi, IPPROTO_ESP, AF_INET);
997 	if (!x)
998 		return 0;
999 
1000 	if (icmp_hdr(skb)->type == ICMP_DEST_UNREACH)
1001 		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ESP);
1002 	else
1003 		ipv4_redirect(skb, net, 0, IPPROTO_ESP);
1004 	xfrm_state_put(x);
1005 
1006 	return 0;
1007 }
1008 
1009 static void esp_destroy(struct xfrm_state *x)
1010 {
1011 	struct crypto_aead *aead = x->data;
1012 
1013 	if (!aead)
1014 		return;
1015 
1016 	crypto_free_aead(aead);
1017 }
1018 
1019 static int esp_init_aead(struct xfrm_state *x)
1020 {
1021 	char aead_name[CRYPTO_MAX_ALG_NAME];
1022 	struct crypto_aead *aead;
1023 	int err;
1024 
1025 	err = -ENAMETOOLONG;
1026 	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1027 		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME)
1028 		goto error;
1029 
1030 	aead = crypto_alloc_aead(aead_name, 0, 0);
1031 	err = PTR_ERR(aead);
1032 	if (IS_ERR(aead))
1033 		goto error;
1034 
1035 	x->data = aead;
1036 
1037 	err = crypto_aead_setkey(aead, x->aead->alg_key,
1038 				 (x->aead->alg_key_len + 7) / 8);
1039 	if (err)
1040 		goto error;
1041 
1042 	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1043 	if (err)
1044 		goto error;
1045 
1046 error:
1047 	return err;
1048 }
1049 
1050 static int esp_init_authenc(struct xfrm_state *x)
1051 {
1052 	struct crypto_aead *aead;
1053 	struct crypto_authenc_key_param *param;
1054 	struct rtattr *rta;
1055 	char *key;
1056 	char *p;
1057 	char authenc_name[CRYPTO_MAX_ALG_NAME];
1058 	unsigned int keylen;
1059 	int err;
1060 
1061 	err = -EINVAL;
1062 	if (!x->ealg)
1063 		goto error;
1064 
1065 	err = -ENAMETOOLONG;
1066 
1067 	if ((x->props.flags & XFRM_STATE_ESN)) {
1068 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1069 			     "%s%sauthencesn(%s,%s)%s",
1070 			     x->geniv ?: "", x->geniv ? "(" : "",
1071 			     x->aalg ? x->aalg->alg_name : "digest_null",
1072 			     x->ealg->alg_name,
1073 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1074 			goto error;
1075 	} else {
1076 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1077 			     "%s%sauthenc(%s,%s)%s",
1078 			     x->geniv ?: "", x->geniv ? "(" : "",
1079 			     x->aalg ? x->aalg->alg_name : "digest_null",
1080 			     x->ealg->alg_name,
1081 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1082 			goto error;
1083 	}
1084 
1085 	aead = crypto_alloc_aead(authenc_name, 0, 0);
1086 	err = PTR_ERR(aead);
1087 	if (IS_ERR(aead))
1088 		goto error;
1089 
1090 	x->data = aead;
1091 
1092 	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1093 		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1094 	err = -ENOMEM;
1095 	key = kmalloc(keylen, GFP_KERNEL);
1096 	if (!key)
1097 		goto error;
1098 
1099 	p = key;
1100 	rta = (void *)p;
1101 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1102 	rta->rta_len = RTA_LENGTH(sizeof(*param));
1103 	param = RTA_DATA(rta);
1104 	p += RTA_SPACE(sizeof(*param));
1105 
1106 	if (x->aalg) {
1107 		struct xfrm_algo_desc *aalg_desc;
1108 
1109 		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1110 		p += (x->aalg->alg_key_len + 7) / 8;
1111 
1112 		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1113 		BUG_ON(!aalg_desc);
1114 
1115 		err = -EINVAL;
1116 		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1117 		    crypto_aead_authsize(aead)) {
1118 			pr_info("ESP: %s digestsize %u != %hu\n",
1119 				x->aalg->alg_name,
1120 				crypto_aead_authsize(aead),
1121 				aalg_desc->uinfo.auth.icv_fullbits / 8);
1122 			goto free_key;
1123 		}
1124 
1125 		err = crypto_aead_setauthsize(
1126 			aead, x->aalg->alg_trunc_len / 8);
1127 		if (err)
1128 			goto free_key;
1129 	}
1130 
1131 	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1132 	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1133 
1134 	err = crypto_aead_setkey(aead, key, keylen);
1135 
1136 free_key:
1137 	kfree(key);
1138 
1139 error:
1140 	return err;
1141 }
1142 
1143 static int esp_init_state(struct xfrm_state *x)
1144 {
1145 	struct crypto_aead *aead;
1146 	u32 align;
1147 	int err;
1148 
1149 	x->data = NULL;
1150 
1151 	if (x->aead)
1152 		err = esp_init_aead(x);
1153 	else
1154 		err = esp_init_authenc(x);
1155 
1156 	if (err)
1157 		goto error;
1158 
1159 	aead = x->data;
1160 
1161 	x->props.header_len = sizeof(struct ip_esp_hdr) +
1162 			      crypto_aead_ivsize(aead);
1163 	if (x->props.mode == XFRM_MODE_TUNNEL)
1164 		x->props.header_len += sizeof(struct iphdr);
1165 	else if (x->props.mode == XFRM_MODE_BEET && x->sel.family != AF_INET6)
1166 		x->props.header_len += IPV4_BEET_PHMAXLEN;
1167 	if (x->encap) {
1168 		struct xfrm_encap_tmpl *encap = x->encap;
1169 
1170 		switch (encap->encap_type) {
1171 		default:
1172 			err = -EINVAL;
1173 			goto error;
1174 		case UDP_ENCAP_ESPINUDP:
1175 			x->props.header_len += sizeof(struct udphdr);
1176 			break;
1177 		case UDP_ENCAP_ESPINUDP_NON_IKE:
1178 			x->props.header_len += sizeof(struct udphdr) + 2 * sizeof(u32);
1179 			break;
1180 #ifdef CONFIG_INET_ESPINTCP
1181 		case TCP_ENCAP_ESPINTCP:
1182 			/* only the length field, TCP encap is done by
1183 			 * the socket
1184 			 */
1185 			x->props.header_len += 2;
1186 			break;
1187 #endif
1188 		}
1189 	}
1190 
1191 	align = ALIGN(crypto_aead_blocksize(aead), 4);
1192 	x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1193 
1194 error:
1195 	return err;
1196 }
1197 
1198 static int esp4_rcv_cb(struct sk_buff *skb, int err)
1199 {
1200 	return 0;
1201 }
1202 
1203 static const struct xfrm_type esp_type =
1204 {
1205 	.owner		= THIS_MODULE,
1206 	.proto	     	= IPPROTO_ESP,
1207 	.flags		= XFRM_TYPE_REPLAY_PROT,
1208 	.init_state	= esp_init_state,
1209 	.destructor	= esp_destroy,
1210 	.input		= esp_input,
1211 	.output		= esp_output,
1212 };
1213 
1214 static struct xfrm4_protocol esp4_protocol = {
1215 	.handler	=	xfrm4_rcv,
1216 	.input_handler	=	xfrm_input,
1217 	.cb_handler	=	esp4_rcv_cb,
1218 	.err_handler	=	esp4_err,
1219 	.priority	=	0,
1220 };
1221 
1222 static int __init esp4_init(void)
1223 {
1224 	if (xfrm_register_type(&esp_type, AF_INET) < 0) {
1225 		pr_info("%s: can't add xfrm type\n", __func__);
1226 		return -EAGAIN;
1227 	}
1228 	if (xfrm4_protocol_register(&esp4_protocol, IPPROTO_ESP) < 0) {
1229 		pr_info("%s: can't add protocol\n", __func__);
1230 		xfrm_unregister_type(&esp_type, AF_INET);
1231 		return -EAGAIN;
1232 	}
1233 	return 0;
1234 }
1235 
1236 static void __exit esp4_fini(void)
1237 {
1238 	if (xfrm4_protocol_deregister(&esp4_protocol, IPPROTO_ESP) < 0)
1239 		pr_info("%s: can't remove protocol\n", __func__);
1240 	xfrm_unregister_type(&esp_type, AF_INET);
1241 }
1242 
1243 module_init(esp4_init);
1244 module_exit(esp4_fini);
1245 MODULE_LICENSE("GPL");
1246 MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_ESP);
1247