xref: /openbmc/linux/net/ipv4/esp4.c (revision 6a143a7c) 
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 esp_output_extra *extra = esp_tmp_extra(tmp);
101 	struct crypto_aead *aead = x->data;
102 	int extralen = 0;
103 	u8 *iv;
104 	struct aead_request *req;
105 	struct scatterlist *sg;
106 
107 	if (x->props.flags & XFRM_STATE_ESN)
108 		extralen += sizeof(*extra);
109 
110 	extra = esp_tmp_extra(tmp);
111 	iv = esp_tmp_iv(aead, tmp, extralen);
112 	req = esp_tmp_req(aead, iv);
113 
114 	/* Unref skb_frag_pages in the src scatterlist if necessary.
115 	 * Skip the first sg which comes from skb->data.
116 	 */
117 	if (req->src != req->dst)
118 		for (sg = sg_next(req->src); sg; sg = sg_next(sg))
119 			put_page(sg_page(sg));
120 }
121 
122 #ifdef CONFIG_INET_ESPINTCP
123 struct esp_tcp_sk {
124 	struct sock *sk;
125 	struct rcu_head rcu;
126 };
127 
128 static void esp_free_tcp_sk(struct rcu_head *head)
129 {
130 	struct esp_tcp_sk *esk = container_of(head, struct esp_tcp_sk, rcu);
131 
132 	sock_put(esk->sk);
133 	kfree(esk);
134 }
135 
136 static struct sock *esp_find_tcp_sk(struct xfrm_state *x)
137 {
138 	struct xfrm_encap_tmpl *encap = x->encap;
139 	struct esp_tcp_sk *esk;
140 	__be16 sport, dport;
141 	struct sock *nsk;
142 	struct sock *sk;
143 
144 	sk = rcu_dereference(x->encap_sk);
145 	if (sk && sk->sk_state == TCP_ESTABLISHED)
146 		return sk;
147 
148 	spin_lock_bh(&x->lock);
149 	sport = encap->encap_sport;
150 	dport = encap->encap_dport;
151 	nsk = rcu_dereference_protected(x->encap_sk,
152 					lockdep_is_held(&x->lock));
153 	if (sk && sk == nsk) {
154 		esk = kmalloc(sizeof(*esk), GFP_ATOMIC);
155 		if (!esk) {
156 			spin_unlock_bh(&x->lock);
157 			return ERR_PTR(-ENOMEM);
158 		}
159 		RCU_INIT_POINTER(x->encap_sk, NULL);
160 		esk->sk = sk;
161 		call_rcu(&esk->rcu, esp_free_tcp_sk);
162 	}
163 	spin_unlock_bh(&x->lock);
164 
165 	sk = inet_lookup_established(xs_net(x), &tcp_hashinfo, x->id.daddr.a4,
166 				     dport, x->props.saddr.a4, sport, 0);
167 	if (!sk)
168 		return ERR_PTR(-ENOENT);
169 
170 	if (!tcp_is_ulp_esp(sk)) {
171 		sock_put(sk);
172 		return ERR_PTR(-EINVAL);
173 	}
174 
175 	spin_lock_bh(&x->lock);
176 	nsk = rcu_dereference_protected(x->encap_sk,
177 					lockdep_is_held(&x->lock));
178 	if (encap->encap_sport != sport ||
179 	    encap->encap_dport != dport) {
180 		sock_put(sk);
181 		sk = nsk ?: ERR_PTR(-EREMCHG);
182 	} else if (sk == nsk) {
183 		sock_put(sk);
184 	} else {
185 		rcu_assign_pointer(x->encap_sk, sk);
186 	}
187 	spin_unlock_bh(&x->lock);
188 
189 	return sk;
190 }
191 
192 static int esp_output_tcp_finish(struct xfrm_state *x, struct sk_buff *skb)
193 {
194 	struct sock *sk;
195 	int err;
196 
197 	rcu_read_lock();
198 
199 	sk = esp_find_tcp_sk(x);
200 	err = PTR_ERR_OR_ZERO(sk);
201 	if (err)
202 		goto out;
203 
204 	bh_lock_sock(sk);
205 	if (sock_owned_by_user(sk))
206 		err = espintcp_queue_out(sk, skb);
207 	else
208 		err = espintcp_push_skb(sk, skb);
209 	bh_unlock_sock(sk);
210 
211 out:
212 	rcu_read_unlock();
213 	return err;
214 }
215 
216 static int esp_output_tcp_encap_cb(struct net *net, struct sock *sk,
217 				   struct sk_buff *skb)
218 {
219 	struct dst_entry *dst = skb_dst(skb);
220 	struct xfrm_state *x = dst->xfrm;
221 
222 	return esp_output_tcp_finish(x, skb);
223 }
224 
225 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
226 {
227 	int err;
228 
229 	local_bh_disable();
230 	err = xfrm_trans_queue_net(xs_net(x), skb, esp_output_tcp_encap_cb);
231 	local_bh_enable();
232 
233 	/* EINPROGRESS just happens to do the right thing.  It
234 	 * actually means that the skb has been consumed and
235 	 * isn't coming back.
236 	 */
237 	return err ?: -EINPROGRESS;
238 }
239 #else
240 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
241 {
242 	kfree_skb(skb);
243 
244 	return -EOPNOTSUPP;
245 }
246 #endif
247 
248 static void esp_output_done(struct crypto_async_request *base, int err)
249 {
250 	struct sk_buff *skb = base->data;
251 	struct xfrm_offload *xo = xfrm_offload(skb);
252 	void *tmp;
253 	struct xfrm_state *x;
254 
255 	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
256 		struct sec_path *sp = skb_sec_path(skb);
257 
258 		x = sp->xvec[sp->len - 1];
259 	} else {
260 		x = skb_dst(skb)->xfrm;
261 	}
262 
263 	tmp = ESP_SKB_CB(skb)->tmp;
264 	esp_ssg_unref(x, tmp);
265 	kfree(tmp);
266 
267 	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
268 		if (err) {
269 			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
270 			kfree_skb(skb);
271 			return;
272 		}
273 
274 		skb_push(skb, skb->data - skb_mac_header(skb));
275 		secpath_reset(skb);
276 		xfrm_dev_resume(skb);
277 	} else {
278 		if (!err &&
279 		    x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
280 			esp_output_tail_tcp(x, skb);
281 		else
282 			xfrm_output_resume(skb, err);
283 	}
284 }
285 
286 /* Move ESP header back into place. */
287 static void esp_restore_header(struct sk_buff *skb, unsigned int offset)
288 {
289 	struct ip_esp_hdr *esph = (void *)(skb->data + offset);
290 	void *tmp = ESP_SKB_CB(skb)->tmp;
291 	__be32 *seqhi = esp_tmp_extra(tmp);
292 
293 	esph->seq_no = esph->spi;
294 	esph->spi = *seqhi;
295 }
296 
297 static void esp_output_restore_header(struct sk_buff *skb)
298 {
299 	void *tmp = ESP_SKB_CB(skb)->tmp;
300 	struct esp_output_extra *extra = esp_tmp_extra(tmp);
301 
302 	esp_restore_header(skb, skb_transport_offset(skb) + extra->esphoff -
303 				sizeof(__be32));
304 }
305 
306 static struct ip_esp_hdr *esp_output_set_extra(struct sk_buff *skb,
307 					       struct xfrm_state *x,
308 					       struct ip_esp_hdr *esph,
309 					       struct esp_output_extra *extra)
310 {
311 	/* For ESN we move the header forward by 4 bytes to
312 	 * accomodate the high bits.  We will move it back after
313 	 * encryption.
314 	 */
315 	if ((x->props.flags & XFRM_STATE_ESN)) {
316 		__u32 seqhi;
317 		struct xfrm_offload *xo = xfrm_offload(skb);
318 
319 		if (xo)
320 			seqhi = xo->seq.hi;
321 		else
322 			seqhi = XFRM_SKB_CB(skb)->seq.output.hi;
323 
324 		extra->esphoff = (unsigned char *)esph -
325 				 skb_transport_header(skb);
326 		esph = (struct ip_esp_hdr *)((unsigned char *)esph - 4);
327 		extra->seqhi = esph->spi;
328 		esph->seq_no = htonl(seqhi);
329 	}
330 
331 	esph->spi = x->id.spi;
332 
333 	return esph;
334 }
335 
336 static void esp_output_done_esn(struct crypto_async_request *base, int err)
337 {
338 	struct sk_buff *skb = base->data;
339 
340 	esp_output_restore_header(skb);
341 	esp_output_done(base, err);
342 }
343 
344 static struct ip_esp_hdr *esp_output_udp_encap(struct sk_buff *skb,
345 					       int encap_type,
346 					       struct esp_info *esp,
347 					       __be16 sport,
348 					       __be16 dport)
349 {
350 	struct udphdr *uh;
351 	__be32 *udpdata32;
352 	unsigned int len;
353 
354 	len = skb->len + esp->tailen - skb_transport_offset(skb);
355 	if (len + sizeof(struct iphdr) > IP_MAX_MTU)
356 		return ERR_PTR(-EMSGSIZE);
357 
358 	uh = (struct udphdr *)esp->esph;
359 	uh->source = sport;
360 	uh->dest = dport;
361 	uh->len = htons(len);
362 	uh->check = 0;
363 
364 	*skb_mac_header(skb) = IPPROTO_UDP;
365 
366 	if (encap_type == UDP_ENCAP_ESPINUDP_NON_IKE) {
367 		udpdata32 = (__be32 *)(uh + 1);
368 		udpdata32[0] = udpdata32[1] = 0;
369 		return (struct ip_esp_hdr *)(udpdata32 + 2);
370 	}
371 
372 	return (struct ip_esp_hdr *)(uh + 1);
373 }
374 
375 #ifdef CONFIG_INET_ESPINTCP
376 static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
377 						    struct sk_buff *skb,
378 						    struct esp_info *esp)
379 {
380 	__be16 *lenp = (void *)esp->esph;
381 	struct ip_esp_hdr *esph;
382 	unsigned int len;
383 	struct sock *sk;
384 
385 	len = skb->len + esp->tailen - skb_transport_offset(skb);
386 	if (len > IP_MAX_MTU)
387 		return ERR_PTR(-EMSGSIZE);
388 
389 	rcu_read_lock();
390 	sk = esp_find_tcp_sk(x);
391 	rcu_read_unlock();
392 
393 	if (IS_ERR(sk))
394 		return ERR_CAST(sk);
395 
396 	*lenp = htons(len);
397 	esph = (struct ip_esp_hdr *)(lenp + 1);
398 
399 	return esph;
400 }
401 #else
402 static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
403 						    struct sk_buff *skb,
404 						    struct esp_info *esp)
405 {
406 	return ERR_PTR(-EOPNOTSUPP);
407 }
408 #endif
409 
410 static int esp_output_encap(struct xfrm_state *x, struct sk_buff *skb,
411 			    struct esp_info *esp)
412 {
413 	struct xfrm_encap_tmpl *encap = x->encap;
414 	struct ip_esp_hdr *esph;
415 	__be16 sport, dport;
416 	int encap_type;
417 
418 	spin_lock_bh(&x->lock);
419 	sport = encap->encap_sport;
420 	dport = encap->encap_dport;
421 	encap_type = encap->encap_type;
422 	spin_unlock_bh(&x->lock);
423 
424 	switch (encap_type) {
425 	default:
426 	case UDP_ENCAP_ESPINUDP:
427 	case UDP_ENCAP_ESPINUDP_NON_IKE:
428 		esph = esp_output_udp_encap(skb, encap_type, esp, sport, dport);
429 		break;
430 	case TCP_ENCAP_ESPINTCP:
431 		esph = esp_output_tcp_encap(x, skb, esp);
432 		break;
433 	}
434 
435 	if (IS_ERR(esph))
436 		return PTR_ERR(esph);
437 
438 	esp->esph = esph;
439 
440 	return 0;
441 }
442 
443 int esp_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
444 {
445 	u8 *tail;
446 	int nfrags;
447 	int esph_offset;
448 	struct page *page;
449 	struct sk_buff *trailer;
450 	int tailen = esp->tailen;
451 
452 	/* this is non-NULL only with TCP/UDP Encapsulation */
453 	if (x->encap) {
454 		int err = esp_output_encap(x, skb, esp);
455 
456 		if (err < 0)
457 			return err;
458 	}
459 
460 	if (!skb_cloned(skb)) {
461 		if (tailen <= skb_tailroom(skb)) {
462 			nfrags = 1;
463 			trailer = skb;
464 			tail = skb_tail_pointer(trailer);
465 
466 			goto skip_cow;
467 		} else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
468 			   && !skb_has_frag_list(skb)) {
469 			int allocsize;
470 			struct sock *sk = skb->sk;
471 			struct page_frag *pfrag = &x->xfrag;
472 
473 			esp->inplace = false;
474 
475 			allocsize = ALIGN(tailen, L1_CACHE_BYTES);
476 
477 			spin_lock_bh(&x->lock);
478 
479 			if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
480 				spin_unlock_bh(&x->lock);
481 				goto cow;
482 			}
483 
484 			page = pfrag->page;
485 			get_page(page);
486 
487 			tail = page_address(page) + pfrag->offset;
488 
489 			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
490 
491 			nfrags = skb_shinfo(skb)->nr_frags;
492 
493 			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
494 					     tailen);
495 			skb_shinfo(skb)->nr_frags = ++nfrags;
496 
497 			pfrag->offset = pfrag->offset + allocsize;
498 
499 			spin_unlock_bh(&x->lock);
500 
501 			nfrags++;
502 
503 			skb->len += tailen;
504 			skb->data_len += tailen;
505 			skb->truesize += tailen;
506 			if (sk && sk_fullsock(sk))
507 				refcount_add(tailen, &sk->sk_wmem_alloc);
508 
509 			goto out;
510 		}
511 	}
512 
513 cow:
514 	esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
515 
516 	nfrags = skb_cow_data(skb, tailen, &trailer);
517 	if (nfrags < 0)
518 		goto out;
519 	tail = skb_tail_pointer(trailer);
520 	esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
521 
522 skip_cow:
523 	esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
524 	pskb_put(skb, trailer, tailen);
525 
526 out:
527 	return nfrags;
528 }
529 EXPORT_SYMBOL_GPL(esp_output_head);
530 
531 int esp_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
532 {
533 	u8 *iv;
534 	int alen;
535 	void *tmp;
536 	int ivlen;
537 	int assoclen;
538 	int extralen;
539 	struct page *page;
540 	struct ip_esp_hdr *esph;
541 	struct crypto_aead *aead;
542 	struct aead_request *req;
543 	struct scatterlist *sg, *dsg;
544 	struct esp_output_extra *extra;
545 	int err = -ENOMEM;
546 
547 	assoclen = sizeof(struct ip_esp_hdr);
548 	extralen = 0;
549 
550 	if (x->props.flags & XFRM_STATE_ESN) {
551 		extralen += sizeof(*extra);
552 		assoclen += sizeof(__be32);
553 	}
554 
555 	aead = x->data;
556 	alen = crypto_aead_authsize(aead);
557 	ivlen = crypto_aead_ivsize(aead);
558 
559 	tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
560 	if (!tmp)
561 		goto error;
562 
563 	extra = esp_tmp_extra(tmp);
564 	iv = esp_tmp_iv(aead, tmp, extralen);
565 	req = esp_tmp_req(aead, iv);
566 	sg = esp_req_sg(aead, req);
567 
568 	if (esp->inplace)
569 		dsg = sg;
570 	else
571 		dsg = &sg[esp->nfrags];
572 
573 	esph = esp_output_set_extra(skb, x, esp->esph, extra);
574 	esp->esph = esph;
575 
576 	sg_init_table(sg, esp->nfrags);
577 	err = skb_to_sgvec(skb, sg,
578 		           (unsigned char *)esph - skb->data,
579 		           assoclen + ivlen + esp->clen + alen);
580 	if (unlikely(err < 0))
581 		goto error_free;
582 
583 	if (!esp->inplace) {
584 		int allocsize;
585 		struct page_frag *pfrag = &x->xfrag;
586 
587 		allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
588 
589 		spin_lock_bh(&x->lock);
590 		if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
591 			spin_unlock_bh(&x->lock);
592 			goto error_free;
593 		}
594 
595 		skb_shinfo(skb)->nr_frags = 1;
596 
597 		page = pfrag->page;
598 		get_page(page);
599 		/* replace page frags in skb with new page */
600 		__skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
601 		pfrag->offset = pfrag->offset + allocsize;
602 		spin_unlock_bh(&x->lock);
603 
604 		sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
605 		err = skb_to_sgvec(skb, dsg,
606 			           (unsigned char *)esph - skb->data,
607 			           assoclen + ivlen + esp->clen + alen);
608 		if (unlikely(err < 0))
609 			goto error_free;
610 	}
611 
612 	if ((x->props.flags & XFRM_STATE_ESN))
613 		aead_request_set_callback(req, 0, esp_output_done_esn, skb);
614 	else
615 		aead_request_set_callback(req, 0, esp_output_done, skb);
616 
617 	aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
618 	aead_request_set_ad(req, assoclen);
619 
620 	memset(iv, 0, ivlen);
621 	memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
622 	       min(ivlen, 8));
623 
624 	ESP_SKB_CB(skb)->tmp = tmp;
625 	err = crypto_aead_encrypt(req);
626 
627 	switch (err) {
628 	case -EINPROGRESS:
629 		goto error;
630 
631 	case -ENOSPC:
632 		err = NET_XMIT_DROP;
633 		break;
634 
635 	case 0:
636 		if ((x->props.flags & XFRM_STATE_ESN))
637 			esp_output_restore_header(skb);
638 	}
639 
640 	if (sg != dsg)
641 		esp_ssg_unref(x, tmp);
642 
643 	if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
644 		err = esp_output_tail_tcp(x, skb);
645 
646 error_free:
647 	kfree(tmp);
648 error:
649 	return err;
650 }
651 EXPORT_SYMBOL_GPL(esp_output_tail);
652 
653 static int esp_output(struct xfrm_state *x, struct sk_buff *skb)
654 {
655 	int alen;
656 	int blksize;
657 	struct ip_esp_hdr *esph;
658 	struct crypto_aead *aead;
659 	struct esp_info esp;
660 
661 	esp.inplace = true;
662 
663 	esp.proto = *skb_mac_header(skb);
664 	*skb_mac_header(skb) = IPPROTO_ESP;
665 
666 	/* skb is pure payload to encrypt */
667 
668 	aead = x->data;
669 	alen = crypto_aead_authsize(aead);
670 
671 	esp.tfclen = 0;
672 	if (x->tfcpad) {
673 		struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
674 		u32 padto;
675 
676 		padto = min(x->tfcpad, xfrm_state_mtu(x, dst->child_mtu_cached));
677 		if (skb->len < padto)
678 			esp.tfclen = padto - skb->len;
679 	}
680 	blksize = ALIGN(crypto_aead_blocksize(aead), 4);
681 	esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
682 	esp.plen = esp.clen - skb->len - esp.tfclen;
683 	esp.tailen = esp.tfclen + esp.plen + alen;
684 
685 	esp.esph = ip_esp_hdr(skb);
686 
687 	esp.nfrags = esp_output_head(x, skb, &esp);
688 	if (esp.nfrags < 0)
689 		return esp.nfrags;
690 
691 	esph = esp.esph;
692 	esph->spi = x->id.spi;
693 
694 	esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
695 	esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
696 				 ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
697 
698 	skb_push(skb, -skb_network_offset(skb));
699 
700 	return esp_output_tail(x, skb, &esp);
701 }
702 
703 static inline int esp_remove_trailer(struct sk_buff *skb)
704 {
705 	struct xfrm_state *x = xfrm_input_state(skb);
706 	struct xfrm_offload *xo = xfrm_offload(skb);
707 	struct crypto_aead *aead = x->data;
708 	int alen, hlen, elen;
709 	int padlen, trimlen;
710 	__wsum csumdiff;
711 	u8 nexthdr[2];
712 	int ret;
713 
714 	alen = crypto_aead_authsize(aead);
715 	hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
716 	elen = skb->len - hlen;
717 
718 	if (xo && (xo->flags & XFRM_ESP_NO_TRAILER)) {
719 		ret = xo->proto;
720 		goto out;
721 	}
722 
723 	if (skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2))
724 		BUG();
725 
726 	ret = -EINVAL;
727 	padlen = nexthdr[0];
728 	if (padlen + 2 + alen >= elen) {
729 		net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
730 				    padlen + 2, elen - alen);
731 		goto out;
732 	}
733 
734 	trimlen = alen + padlen + 2;
735 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
736 		csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
737 		skb->csum = csum_block_sub(skb->csum, csumdiff,
738 					   skb->len - trimlen);
739 	}
740 	pskb_trim(skb, skb->len - trimlen);
741 
742 	ret = nexthdr[1];
743 
744 out:
745 	return ret;
746 }
747 
748 int esp_input_done2(struct sk_buff *skb, int err)
749 {
750 	const struct iphdr *iph;
751 	struct xfrm_state *x = xfrm_input_state(skb);
752 	struct xfrm_offload *xo = xfrm_offload(skb);
753 	struct crypto_aead *aead = x->data;
754 	int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
755 	int ihl;
756 
757 	if (!xo || (xo && !(xo->flags & CRYPTO_DONE)))
758 		kfree(ESP_SKB_CB(skb)->tmp);
759 
760 	if (unlikely(err))
761 		goto out;
762 
763 	err = esp_remove_trailer(skb);
764 	if (unlikely(err < 0))
765 		goto out;
766 
767 	iph = ip_hdr(skb);
768 	ihl = iph->ihl * 4;
769 
770 	if (x->encap) {
771 		struct xfrm_encap_tmpl *encap = x->encap;
772 		struct tcphdr *th = (void *)(skb_network_header(skb) + ihl);
773 		struct udphdr *uh = (void *)(skb_network_header(skb) + ihl);
774 		__be16 source;
775 
776 		switch (x->encap->encap_type) {
777 		case TCP_ENCAP_ESPINTCP:
778 			source = th->source;
779 			break;
780 		case UDP_ENCAP_ESPINUDP:
781 		case UDP_ENCAP_ESPINUDP_NON_IKE:
782 			source = uh->source;
783 			break;
784 		default:
785 			WARN_ON_ONCE(1);
786 			err = -EINVAL;
787 			goto out;
788 		}
789 
790 		/*
791 		 * 1) if the NAT-T peer's IP or port changed then
792 		 *    advertize the change to the keying daemon.
793 		 *    This is an inbound SA, so just compare
794 		 *    SRC ports.
795 		 */
796 		if (iph->saddr != x->props.saddr.a4 ||
797 		    source != encap->encap_sport) {
798 			xfrm_address_t ipaddr;
799 
800 			ipaddr.a4 = iph->saddr;
801 			km_new_mapping(x, &ipaddr, source);
802 
803 			/* XXX: perhaps add an extra
804 			 * policy check here, to see
805 			 * if we should allow or
806 			 * reject a packet from a
807 			 * different source
808 			 * address/port.
809 			 */
810 		}
811 
812 		/*
813 		 * 2) ignore UDP/TCP checksums in case
814 		 *    of NAT-T in Transport Mode, or
815 		 *    perform other post-processing fixes
816 		 *    as per draft-ietf-ipsec-udp-encaps-06,
817 		 *    section 3.1.2
818 		 */
819 		if (x->props.mode == XFRM_MODE_TRANSPORT)
820 			skb->ip_summed = CHECKSUM_UNNECESSARY;
821 	}
822 
823 	skb_pull_rcsum(skb, hlen);
824 	if (x->props.mode == XFRM_MODE_TUNNEL)
825 		skb_reset_transport_header(skb);
826 	else
827 		skb_set_transport_header(skb, -ihl);
828 
829 	/* RFC4303: Drop dummy packets without any error */
830 	if (err == IPPROTO_NONE)
831 		err = -EINVAL;
832 
833 out:
834 	return err;
835 }
836 EXPORT_SYMBOL_GPL(esp_input_done2);
837 
838 static void esp_input_done(struct crypto_async_request *base, int err)
839 {
840 	struct sk_buff *skb = base->data;
841 
842 	xfrm_input_resume(skb, esp_input_done2(skb, err));
843 }
844 
845 static void esp_input_restore_header(struct sk_buff *skb)
846 {
847 	esp_restore_header(skb, 0);
848 	__skb_pull(skb, 4);
849 }
850 
851 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
852 {
853 	struct xfrm_state *x = xfrm_input_state(skb);
854 	struct ip_esp_hdr *esph;
855 
856 	/* For ESN we move the header forward by 4 bytes to
857 	 * accomodate the high bits.  We will move it back after
858 	 * decryption.
859 	 */
860 	if ((x->props.flags & XFRM_STATE_ESN)) {
861 		esph = skb_push(skb, 4);
862 		*seqhi = esph->spi;
863 		esph->spi = esph->seq_no;
864 		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
865 	}
866 }
867 
868 static void esp_input_done_esn(struct crypto_async_request *base, int err)
869 {
870 	struct sk_buff *skb = base->data;
871 
872 	esp_input_restore_header(skb);
873 	esp_input_done(base, err);
874 }
875 
876 /*
877  * Note: detecting truncated vs. non-truncated authentication data is very
878  * expensive, so we only support truncated data, which is the recommended
879  * and common case.
880  */
881 static int esp_input(struct xfrm_state *x, struct sk_buff *skb)
882 {
883 	struct crypto_aead *aead = x->data;
884 	struct aead_request *req;
885 	struct sk_buff *trailer;
886 	int ivlen = crypto_aead_ivsize(aead);
887 	int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
888 	int nfrags;
889 	int assoclen;
890 	int seqhilen;
891 	__be32 *seqhi;
892 	void *tmp;
893 	u8 *iv;
894 	struct scatterlist *sg;
895 	int err = -EINVAL;
896 
897 	if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen))
898 		goto out;
899 
900 	if (elen <= 0)
901 		goto out;
902 
903 	assoclen = sizeof(struct ip_esp_hdr);
904 	seqhilen = 0;
905 
906 	if (x->props.flags & XFRM_STATE_ESN) {
907 		seqhilen += sizeof(__be32);
908 		assoclen += seqhilen;
909 	}
910 
911 	if (!skb_cloned(skb)) {
912 		if (!skb_is_nonlinear(skb)) {
913 			nfrags = 1;
914 
915 			goto skip_cow;
916 		} else if (!skb_has_frag_list(skb)) {
917 			nfrags = skb_shinfo(skb)->nr_frags;
918 			nfrags++;
919 
920 			goto skip_cow;
921 		}
922 	}
923 
924 	err = skb_cow_data(skb, 0, &trailer);
925 	if (err < 0)
926 		goto out;
927 
928 	nfrags = err;
929 
930 skip_cow:
931 	err = -ENOMEM;
932 	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
933 	if (!tmp)
934 		goto out;
935 
936 	ESP_SKB_CB(skb)->tmp = tmp;
937 	seqhi = esp_tmp_extra(tmp);
938 	iv = esp_tmp_iv(aead, tmp, seqhilen);
939 	req = esp_tmp_req(aead, iv);
940 	sg = esp_req_sg(aead, req);
941 
942 	esp_input_set_header(skb, seqhi);
943 
944 	sg_init_table(sg, nfrags);
945 	err = skb_to_sgvec(skb, sg, 0, skb->len);
946 	if (unlikely(err < 0)) {
947 		kfree(tmp);
948 		goto out;
949 	}
950 
951 	skb->ip_summed = CHECKSUM_NONE;
952 
953 	if ((x->props.flags & XFRM_STATE_ESN))
954 		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
955 	else
956 		aead_request_set_callback(req, 0, esp_input_done, skb);
957 
958 	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
959 	aead_request_set_ad(req, assoclen);
960 
961 	err = crypto_aead_decrypt(req);
962 	if (err == -EINPROGRESS)
963 		goto out;
964 
965 	if ((x->props.flags & XFRM_STATE_ESN))
966 		esp_input_restore_header(skb);
967 
968 	err = esp_input_done2(skb, err);
969 
970 out:
971 	return err;
972 }
973 
974 static int esp4_err(struct sk_buff *skb, u32 info)
975 {
976 	struct net *net = dev_net(skb->dev);
977 	const struct iphdr *iph = (const struct iphdr *)skb->data;
978 	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data+(iph->ihl<<2));
979 	struct xfrm_state *x;
980 
981 	switch (icmp_hdr(skb)->type) {
982 	case ICMP_DEST_UNREACH:
983 		if (icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
984 			return 0;
985 	case ICMP_REDIRECT:
986 		break;
987 	default:
988 		return 0;
989 	}
990 
991 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
992 			      esph->spi, IPPROTO_ESP, AF_INET);
993 	if (!x)
994 		return 0;
995 
996 	if (icmp_hdr(skb)->type == ICMP_DEST_UNREACH)
997 		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ESP);
998 	else
999 		ipv4_redirect(skb, net, 0, IPPROTO_ESP);
1000 	xfrm_state_put(x);
1001 
1002 	return 0;
1003 }
1004 
1005 static void esp_destroy(struct xfrm_state *x)
1006 {
1007 	struct crypto_aead *aead = x->data;
1008 
1009 	if (!aead)
1010 		return;
1011 
1012 	crypto_free_aead(aead);
1013 }
1014 
1015 static int esp_init_aead(struct xfrm_state *x)
1016 {
1017 	char aead_name[CRYPTO_MAX_ALG_NAME];
1018 	struct crypto_aead *aead;
1019 	int err;
1020 
1021 	err = -ENAMETOOLONG;
1022 	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1023 		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME)
1024 		goto error;
1025 
1026 	aead = crypto_alloc_aead(aead_name, 0, 0);
1027 	err = PTR_ERR(aead);
1028 	if (IS_ERR(aead))
1029 		goto error;
1030 
1031 	x->data = aead;
1032 
1033 	err = crypto_aead_setkey(aead, x->aead->alg_key,
1034 				 (x->aead->alg_key_len + 7) / 8);
1035 	if (err)
1036 		goto error;
1037 
1038 	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1039 	if (err)
1040 		goto error;
1041 
1042 error:
1043 	return err;
1044 }
1045 
1046 static int esp_init_authenc(struct xfrm_state *x)
1047 {
1048 	struct crypto_aead *aead;
1049 	struct crypto_authenc_key_param *param;
1050 	struct rtattr *rta;
1051 	char *key;
1052 	char *p;
1053 	char authenc_name[CRYPTO_MAX_ALG_NAME];
1054 	unsigned int keylen;
1055 	int err;
1056 
1057 	err = -EINVAL;
1058 	if (!x->ealg)
1059 		goto error;
1060 
1061 	err = -ENAMETOOLONG;
1062 
1063 	if ((x->props.flags & XFRM_STATE_ESN)) {
1064 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1065 			     "%s%sauthencesn(%s,%s)%s",
1066 			     x->geniv ?: "", x->geniv ? "(" : "",
1067 			     x->aalg ? x->aalg->alg_name : "digest_null",
1068 			     x->ealg->alg_name,
1069 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1070 			goto error;
1071 	} else {
1072 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1073 			     "%s%sauthenc(%s,%s)%s",
1074 			     x->geniv ?: "", x->geniv ? "(" : "",
1075 			     x->aalg ? x->aalg->alg_name : "digest_null",
1076 			     x->ealg->alg_name,
1077 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1078 			goto error;
1079 	}
1080 
1081 	aead = crypto_alloc_aead(authenc_name, 0, 0);
1082 	err = PTR_ERR(aead);
1083 	if (IS_ERR(aead))
1084 		goto error;
1085 
1086 	x->data = aead;
1087 
1088 	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1089 		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1090 	err = -ENOMEM;
1091 	key = kmalloc(keylen, GFP_KERNEL);
1092 	if (!key)
1093 		goto error;
1094 
1095 	p = key;
1096 	rta = (void *)p;
1097 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1098 	rta->rta_len = RTA_LENGTH(sizeof(*param));
1099 	param = RTA_DATA(rta);
1100 	p += RTA_SPACE(sizeof(*param));
1101 
1102 	if (x->aalg) {
1103 		struct xfrm_algo_desc *aalg_desc;
1104 
1105 		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1106 		p += (x->aalg->alg_key_len + 7) / 8;
1107 
1108 		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1109 		BUG_ON(!aalg_desc);
1110 
1111 		err = -EINVAL;
1112 		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1113 		    crypto_aead_authsize(aead)) {
1114 			pr_info("ESP: %s digestsize %u != %hu\n",
1115 				x->aalg->alg_name,
1116 				crypto_aead_authsize(aead),
1117 				aalg_desc->uinfo.auth.icv_fullbits / 8);
1118 			goto free_key;
1119 		}
1120 
1121 		err = crypto_aead_setauthsize(
1122 			aead, x->aalg->alg_trunc_len / 8);
1123 		if (err)
1124 			goto free_key;
1125 	}
1126 
1127 	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1128 	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1129 
1130 	err = crypto_aead_setkey(aead, key, keylen);
1131 
1132 free_key:
1133 	kfree(key);
1134 
1135 error:
1136 	return err;
1137 }
1138 
1139 static int esp_init_state(struct xfrm_state *x)
1140 {
1141 	struct crypto_aead *aead;
1142 	u32 align;
1143 	int err;
1144 
1145 	x->data = NULL;
1146 
1147 	if (x->aead)
1148 		err = esp_init_aead(x);
1149 	else
1150 		err = esp_init_authenc(x);
1151 
1152 	if (err)
1153 		goto error;
1154 
1155 	aead = x->data;
1156 
1157 	x->props.header_len = sizeof(struct ip_esp_hdr) +
1158 			      crypto_aead_ivsize(aead);
1159 	if (x->props.mode == XFRM_MODE_TUNNEL)
1160 		x->props.header_len += sizeof(struct iphdr);
1161 	else if (x->props.mode == XFRM_MODE_BEET && x->sel.family != AF_INET6)
1162 		x->props.header_len += IPV4_BEET_PHMAXLEN;
1163 	if (x->encap) {
1164 		struct xfrm_encap_tmpl *encap = x->encap;
1165 
1166 		switch (encap->encap_type) {
1167 		default:
1168 			err = -EINVAL;
1169 			goto error;
1170 		case UDP_ENCAP_ESPINUDP:
1171 			x->props.header_len += sizeof(struct udphdr);
1172 			break;
1173 		case UDP_ENCAP_ESPINUDP_NON_IKE:
1174 			x->props.header_len += sizeof(struct udphdr) + 2 * sizeof(u32);
1175 			break;
1176 #ifdef CONFIG_INET_ESPINTCP
1177 		case TCP_ENCAP_ESPINTCP:
1178 			/* only the length field, TCP encap is done by
1179 			 * the socket
1180 			 */
1181 			x->props.header_len += 2;
1182 			break;
1183 #endif
1184 		}
1185 	}
1186 
1187 	align = ALIGN(crypto_aead_blocksize(aead), 4);
1188 	x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1189 
1190 error:
1191 	return err;
1192 }
1193 
1194 static int esp4_rcv_cb(struct sk_buff *skb, int err)
1195 {
1196 	return 0;
1197 }
1198 
1199 static const struct xfrm_type esp_type =
1200 {
1201 	.description	= "ESP4",
1202 	.owner		= THIS_MODULE,
1203 	.proto	     	= IPPROTO_ESP,
1204 	.flags		= XFRM_TYPE_REPLAY_PROT,
1205 	.init_state	= esp_init_state,
1206 	.destructor	= esp_destroy,
1207 	.input		= esp_input,
1208 	.output		= esp_output,
1209 };
1210 
1211 static struct xfrm4_protocol esp4_protocol = {
1212 	.handler	=	xfrm4_rcv,
1213 	.input_handler	=	xfrm_input,
1214 	.cb_handler	=	esp4_rcv_cb,
1215 	.err_handler	=	esp4_err,
1216 	.priority	=	0,
1217 };
1218 
1219 static int __init esp4_init(void)
1220 {
1221 	if (xfrm_register_type(&esp_type, AF_INET) < 0) {
1222 		pr_info("%s: can't add xfrm type\n", __func__);
1223 		return -EAGAIN;
1224 	}
1225 	if (xfrm4_protocol_register(&esp4_protocol, IPPROTO_ESP) < 0) {
1226 		pr_info("%s: can't add protocol\n", __func__);
1227 		xfrm_unregister_type(&esp_type, AF_INET);
1228 		return -EAGAIN;
1229 	}
1230 	return 0;
1231 }
1232 
1233 static void __exit esp4_fini(void)
1234 {
1235 	if (xfrm4_protocol_deregister(&esp4_protocol, IPPROTO_ESP) < 0)
1236 		pr_info("%s: can't remove protocol\n", __func__);
1237 	xfrm_unregister_type(&esp_type, AF_INET);
1238 }
1239 
1240 module_init(esp4_init);
1241 module_exit(esp4_fini);
1242 MODULE_LICENSE("GPL");
1243 MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_ESP);
1244