xref: /openbmc/linux/net/ipv4/esp4.c (revision b285d2ae) 
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 	u8 *vaddr;
447 	int nfrags;
448 	int esph_offset;
449 	struct page *page;
450 	struct sk_buff *trailer;
451 	int tailen = esp->tailen;
452 
453 	/* this is non-NULL only with TCP/UDP Encapsulation */
454 	if (x->encap) {
455 		int err = esp_output_encap(x, skb, esp);
456 
457 		if (err < 0)
458 			return err;
459 	}
460 
461 	if (!skb_cloned(skb)) {
462 		if (tailen <= skb_tailroom(skb)) {
463 			nfrags = 1;
464 			trailer = skb;
465 			tail = skb_tail_pointer(trailer);
466 
467 			goto skip_cow;
468 		} else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
469 			   && !skb_has_frag_list(skb)) {
470 			int allocsize;
471 			struct sock *sk = skb->sk;
472 			struct page_frag *pfrag = &x->xfrag;
473 
474 			esp->inplace = false;
475 
476 			allocsize = ALIGN(tailen, L1_CACHE_BYTES);
477 
478 			spin_lock_bh(&x->lock);
479 
480 			if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
481 				spin_unlock_bh(&x->lock);
482 				goto cow;
483 			}
484 
485 			page = pfrag->page;
486 			get_page(page);
487 
488 			vaddr = kmap_atomic(page);
489 
490 			tail = vaddr + pfrag->offset;
491 
492 			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
493 
494 			kunmap_atomic(vaddr);
495 
496 			nfrags = skb_shinfo(skb)->nr_frags;
497 
498 			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
499 					     tailen);
500 			skb_shinfo(skb)->nr_frags = ++nfrags;
501 
502 			pfrag->offset = pfrag->offset + allocsize;
503 
504 			spin_unlock_bh(&x->lock);
505 
506 			nfrags++;
507 
508 			skb->len += tailen;
509 			skb->data_len += tailen;
510 			skb->truesize += tailen;
511 			if (sk && sk_fullsock(sk))
512 				refcount_add(tailen, &sk->sk_wmem_alloc);
513 
514 			goto out;
515 		}
516 	}
517 
518 cow:
519 	esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
520 
521 	nfrags = skb_cow_data(skb, tailen, &trailer);
522 	if (nfrags < 0)
523 		goto out;
524 	tail = skb_tail_pointer(trailer);
525 	esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
526 
527 skip_cow:
528 	esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
529 	pskb_put(skb, trailer, tailen);
530 
531 out:
532 	return nfrags;
533 }
534 EXPORT_SYMBOL_GPL(esp_output_head);
535 
536 int esp_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
537 {
538 	u8 *iv;
539 	int alen;
540 	void *tmp;
541 	int ivlen;
542 	int assoclen;
543 	int extralen;
544 	struct page *page;
545 	struct ip_esp_hdr *esph;
546 	struct crypto_aead *aead;
547 	struct aead_request *req;
548 	struct scatterlist *sg, *dsg;
549 	struct esp_output_extra *extra;
550 	int err = -ENOMEM;
551 
552 	assoclen = sizeof(struct ip_esp_hdr);
553 	extralen = 0;
554 
555 	if (x->props.flags & XFRM_STATE_ESN) {
556 		extralen += sizeof(*extra);
557 		assoclen += sizeof(__be32);
558 	}
559 
560 	aead = x->data;
561 	alen = crypto_aead_authsize(aead);
562 	ivlen = crypto_aead_ivsize(aead);
563 
564 	tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
565 	if (!tmp)
566 		goto error;
567 
568 	extra = esp_tmp_extra(tmp);
569 	iv = esp_tmp_iv(aead, tmp, extralen);
570 	req = esp_tmp_req(aead, iv);
571 	sg = esp_req_sg(aead, req);
572 
573 	if (esp->inplace)
574 		dsg = sg;
575 	else
576 		dsg = &sg[esp->nfrags];
577 
578 	esph = esp_output_set_extra(skb, x, esp->esph, extra);
579 	esp->esph = esph;
580 
581 	sg_init_table(sg, esp->nfrags);
582 	err = skb_to_sgvec(skb, sg,
583 		           (unsigned char *)esph - skb->data,
584 		           assoclen + ivlen + esp->clen + alen);
585 	if (unlikely(err < 0))
586 		goto error_free;
587 
588 	if (!esp->inplace) {
589 		int allocsize;
590 		struct page_frag *pfrag = &x->xfrag;
591 
592 		allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
593 
594 		spin_lock_bh(&x->lock);
595 		if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
596 			spin_unlock_bh(&x->lock);
597 			goto error_free;
598 		}
599 
600 		skb_shinfo(skb)->nr_frags = 1;
601 
602 		page = pfrag->page;
603 		get_page(page);
604 		/* replace page frags in skb with new page */
605 		__skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
606 		pfrag->offset = pfrag->offset + allocsize;
607 		spin_unlock_bh(&x->lock);
608 
609 		sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
610 		err = skb_to_sgvec(skb, dsg,
611 			           (unsigned char *)esph - skb->data,
612 			           assoclen + ivlen + esp->clen + alen);
613 		if (unlikely(err < 0))
614 			goto error_free;
615 	}
616 
617 	if ((x->props.flags & XFRM_STATE_ESN))
618 		aead_request_set_callback(req, 0, esp_output_done_esn, skb);
619 	else
620 		aead_request_set_callback(req, 0, esp_output_done, skb);
621 
622 	aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
623 	aead_request_set_ad(req, assoclen);
624 
625 	memset(iv, 0, ivlen);
626 	memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
627 	       min(ivlen, 8));
628 
629 	ESP_SKB_CB(skb)->tmp = tmp;
630 	err = crypto_aead_encrypt(req);
631 
632 	switch (err) {
633 	case -EINPROGRESS:
634 		goto error;
635 
636 	case -ENOSPC:
637 		err = NET_XMIT_DROP;
638 		break;
639 
640 	case 0:
641 		if ((x->props.flags & XFRM_STATE_ESN))
642 			esp_output_restore_header(skb);
643 	}
644 
645 	if (sg != dsg)
646 		esp_ssg_unref(x, tmp);
647 
648 	if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
649 		err = esp_output_tail_tcp(x, skb);
650 
651 error_free:
652 	kfree(tmp);
653 error:
654 	return err;
655 }
656 EXPORT_SYMBOL_GPL(esp_output_tail);
657 
658 static int esp_output(struct xfrm_state *x, struct sk_buff *skb)
659 {
660 	int alen;
661 	int blksize;
662 	struct ip_esp_hdr *esph;
663 	struct crypto_aead *aead;
664 	struct esp_info esp;
665 
666 	esp.inplace = true;
667 
668 	esp.proto = *skb_mac_header(skb);
669 	*skb_mac_header(skb) = IPPROTO_ESP;
670 
671 	/* skb is pure payload to encrypt */
672 
673 	aead = x->data;
674 	alen = crypto_aead_authsize(aead);
675 
676 	esp.tfclen = 0;
677 	if (x->tfcpad) {
678 		struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
679 		u32 padto;
680 
681 		padto = min(x->tfcpad, xfrm_state_mtu(x, dst->child_mtu_cached));
682 		if (skb->len < padto)
683 			esp.tfclen = padto - skb->len;
684 	}
685 	blksize = ALIGN(crypto_aead_blocksize(aead), 4);
686 	esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
687 	esp.plen = esp.clen - skb->len - esp.tfclen;
688 	esp.tailen = esp.tfclen + esp.plen + alen;
689 
690 	esp.esph = ip_esp_hdr(skb);
691 
692 	esp.nfrags = esp_output_head(x, skb, &esp);
693 	if (esp.nfrags < 0)
694 		return esp.nfrags;
695 
696 	esph = esp.esph;
697 	esph->spi = x->id.spi;
698 
699 	esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
700 	esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
701 				 ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
702 
703 	skb_push(skb, -skb_network_offset(skb));
704 
705 	return esp_output_tail(x, skb, &esp);
706 }
707 
708 static inline int esp_remove_trailer(struct sk_buff *skb)
709 {
710 	struct xfrm_state *x = xfrm_input_state(skb);
711 	struct xfrm_offload *xo = xfrm_offload(skb);
712 	struct crypto_aead *aead = x->data;
713 	int alen, hlen, elen;
714 	int padlen, trimlen;
715 	__wsum csumdiff;
716 	u8 nexthdr[2];
717 	int ret;
718 
719 	alen = crypto_aead_authsize(aead);
720 	hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
721 	elen = skb->len - hlen;
722 
723 	if (xo && (xo->flags & XFRM_ESP_NO_TRAILER)) {
724 		ret = xo->proto;
725 		goto out;
726 	}
727 
728 	if (skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2))
729 		BUG();
730 
731 	ret = -EINVAL;
732 	padlen = nexthdr[0];
733 	if (padlen + 2 + alen >= elen) {
734 		net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
735 				    padlen + 2, elen - alen);
736 		goto out;
737 	}
738 
739 	trimlen = alen + padlen + 2;
740 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
741 		csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
742 		skb->csum = csum_block_sub(skb->csum, csumdiff,
743 					   skb->len - trimlen);
744 	}
745 	pskb_trim(skb, skb->len - trimlen);
746 
747 	ret = nexthdr[1];
748 
749 out:
750 	return ret;
751 }
752 
753 int esp_input_done2(struct sk_buff *skb, int err)
754 {
755 	const struct iphdr *iph;
756 	struct xfrm_state *x = xfrm_input_state(skb);
757 	struct xfrm_offload *xo = xfrm_offload(skb);
758 	struct crypto_aead *aead = x->data;
759 	int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
760 	int ihl;
761 
762 	if (!xo || (xo && !(xo->flags & CRYPTO_DONE)))
763 		kfree(ESP_SKB_CB(skb)->tmp);
764 
765 	if (unlikely(err))
766 		goto out;
767 
768 	err = esp_remove_trailer(skb);
769 	if (unlikely(err < 0))
770 		goto out;
771 
772 	iph = ip_hdr(skb);
773 	ihl = iph->ihl * 4;
774 
775 	if (x->encap) {
776 		struct xfrm_encap_tmpl *encap = x->encap;
777 		struct tcphdr *th = (void *)(skb_network_header(skb) + ihl);
778 		struct udphdr *uh = (void *)(skb_network_header(skb) + ihl);
779 		__be16 source;
780 
781 		switch (x->encap->encap_type) {
782 		case TCP_ENCAP_ESPINTCP:
783 			source = th->source;
784 			break;
785 		case UDP_ENCAP_ESPINUDP:
786 		case UDP_ENCAP_ESPINUDP_NON_IKE:
787 			source = uh->source;
788 			break;
789 		default:
790 			WARN_ON_ONCE(1);
791 			err = -EINVAL;
792 			goto out;
793 		}
794 
795 		/*
796 		 * 1) if the NAT-T peer's IP or port changed then
797 		 *    advertize the change to the keying daemon.
798 		 *    This is an inbound SA, so just compare
799 		 *    SRC ports.
800 		 */
801 		if (iph->saddr != x->props.saddr.a4 ||
802 		    source != encap->encap_sport) {
803 			xfrm_address_t ipaddr;
804 
805 			ipaddr.a4 = iph->saddr;
806 			km_new_mapping(x, &ipaddr, source);
807 
808 			/* XXX: perhaps add an extra
809 			 * policy check here, to see
810 			 * if we should allow or
811 			 * reject a packet from a
812 			 * different source
813 			 * address/port.
814 			 */
815 		}
816 
817 		/*
818 		 * 2) ignore UDP/TCP checksums in case
819 		 *    of NAT-T in Transport Mode, or
820 		 *    perform other post-processing fixes
821 		 *    as per draft-ietf-ipsec-udp-encaps-06,
822 		 *    section 3.1.2
823 		 */
824 		if (x->props.mode == XFRM_MODE_TRANSPORT)
825 			skb->ip_summed = CHECKSUM_UNNECESSARY;
826 	}
827 
828 	skb_pull_rcsum(skb, hlen);
829 	if (x->props.mode == XFRM_MODE_TUNNEL)
830 		skb_reset_transport_header(skb);
831 	else
832 		skb_set_transport_header(skb, -ihl);
833 
834 	/* RFC4303: Drop dummy packets without any error */
835 	if (err == IPPROTO_NONE)
836 		err = -EINVAL;
837 
838 out:
839 	return err;
840 }
841 EXPORT_SYMBOL_GPL(esp_input_done2);
842 
843 static void esp_input_done(struct crypto_async_request *base, int err)
844 {
845 	struct sk_buff *skb = base->data;
846 
847 	xfrm_input_resume(skb, esp_input_done2(skb, err));
848 }
849 
850 static void esp_input_restore_header(struct sk_buff *skb)
851 {
852 	esp_restore_header(skb, 0);
853 	__skb_pull(skb, 4);
854 }
855 
856 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
857 {
858 	struct xfrm_state *x = xfrm_input_state(skb);
859 	struct ip_esp_hdr *esph;
860 
861 	/* For ESN we move the header forward by 4 bytes to
862 	 * accomodate the high bits.  We will move it back after
863 	 * decryption.
864 	 */
865 	if ((x->props.flags & XFRM_STATE_ESN)) {
866 		esph = skb_push(skb, 4);
867 		*seqhi = esph->spi;
868 		esph->spi = esph->seq_no;
869 		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
870 	}
871 }
872 
873 static void esp_input_done_esn(struct crypto_async_request *base, int err)
874 {
875 	struct sk_buff *skb = base->data;
876 
877 	esp_input_restore_header(skb);
878 	esp_input_done(base, err);
879 }
880 
881 /*
882  * Note: detecting truncated vs. non-truncated authentication data is very
883  * expensive, so we only support truncated data, which is the recommended
884  * and common case.
885  */
886 static int esp_input(struct xfrm_state *x, struct sk_buff *skb)
887 {
888 	struct crypto_aead *aead = x->data;
889 	struct aead_request *req;
890 	struct sk_buff *trailer;
891 	int ivlen = crypto_aead_ivsize(aead);
892 	int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
893 	int nfrags;
894 	int assoclen;
895 	int seqhilen;
896 	__be32 *seqhi;
897 	void *tmp;
898 	u8 *iv;
899 	struct scatterlist *sg;
900 	int err = -EINVAL;
901 
902 	if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen))
903 		goto out;
904 
905 	if (elen <= 0)
906 		goto out;
907 
908 	assoclen = sizeof(struct ip_esp_hdr);
909 	seqhilen = 0;
910 
911 	if (x->props.flags & XFRM_STATE_ESN) {
912 		seqhilen += sizeof(__be32);
913 		assoclen += seqhilen;
914 	}
915 
916 	if (!skb_cloned(skb)) {
917 		if (!skb_is_nonlinear(skb)) {
918 			nfrags = 1;
919 
920 			goto skip_cow;
921 		} else if (!skb_has_frag_list(skb)) {
922 			nfrags = skb_shinfo(skb)->nr_frags;
923 			nfrags++;
924 
925 			goto skip_cow;
926 		}
927 	}
928 
929 	err = skb_cow_data(skb, 0, &trailer);
930 	if (err < 0)
931 		goto out;
932 
933 	nfrags = err;
934 
935 skip_cow:
936 	err = -ENOMEM;
937 	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
938 	if (!tmp)
939 		goto out;
940 
941 	ESP_SKB_CB(skb)->tmp = tmp;
942 	seqhi = esp_tmp_extra(tmp);
943 	iv = esp_tmp_iv(aead, tmp, seqhilen);
944 	req = esp_tmp_req(aead, iv);
945 	sg = esp_req_sg(aead, req);
946 
947 	esp_input_set_header(skb, seqhi);
948 
949 	sg_init_table(sg, nfrags);
950 	err = skb_to_sgvec(skb, sg, 0, skb->len);
951 	if (unlikely(err < 0)) {
952 		kfree(tmp);
953 		goto out;
954 	}
955 
956 	skb->ip_summed = CHECKSUM_NONE;
957 
958 	if ((x->props.flags & XFRM_STATE_ESN))
959 		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
960 	else
961 		aead_request_set_callback(req, 0, esp_input_done, skb);
962 
963 	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
964 	aead_request_set_ad(req, assoclen);
965 
966 	err = crypto_aead_decrypt(req);
967 	if (err == -EINPROGRESS)
968 		goto out;
969 
970 	if ((x->props.flags & XFRM_STATE_ESN))
971 		esp_input_restore_header(skb);
972 
973 	err = esp_input_done2(skb, err);
974 
975 out:
976 	return err;
977 }
978 
979 static int esp4_err(struct sk_buff *skb, u32 info)
980 {
981 	struct net *net = dev_net(skb->dev);
982 	const struct iphdr *iph = (const struct iphdr *)skb->data;
983 	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data+(iph->ihl<<2));
984 	struct xfrm_state *x;
985 
986 	switch (icmp_hdr(skb)->type) {
987 	case ICMP_DEST_UNREACH:
988 		if (icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
989 			return 0;
990 	case ICMP_REDIRECT:
991 		break;
992 	default:
993 		return 0;
994 	}
995 
996 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
997 			      esph->spi, IPPROTO_ESP, AF_INET);
998 	if (!x)
999 		return 0;
1000 
1001 	if (icmp_hdr(skb)->type == ICMP_DEST_UNREACH)
1002 		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ESP);
1003 	else
1004 		ipv4_redirect(skb, net, 0, IPPROTO_ESP);
1005 	xfrm_state_put(x);
1006 
1007 	return 0;
1008 }
1009 
1010 static void esp_destroy(struct xfrm_state *x)
1011 {
1012 	struct crypto_aead *aead = x->data;
1013 
1014 	if (!aead)
1015 		return;
1016 
1017 	crypto_free_aead(aead);
1018 }
1019 
1020 static int esp_init_aead(struct xfrm_state *x)
1021 {
1022 	char aead_name[CRYPTO_MAX_ALG_NAME];
1023 	struct crypto_aead *aead;
1024 	int err;
1025 
1026 	err = -ENAMETOOLONG;
1027 	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1028 		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME)
1029 		goto error;
1030 
1031 	aead = crypto_alloc_aead(aead_name, 0, 0);
1032 	err = PTR_ERR(aead);
1033 	if (IS_ERR(aead))
1034 		goto error;
1035 
1036 	x->data = aead;
1037 
1038 	err = crypto_aead_setkey(aead, x->aead->alg_key,
1039 				 (x->aead->alg_key_len + 7) / 8);
1040 	if (err)
1041 		goto error;
1042 
1043 	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1044 	if (err)
1045 		goto error;
1046 
1047 error:
1048 	return err;
1049 }
1050 
1051 static int esp_init_authenc(struct xfrm_state *x)
1052 {
1053 	struct crypto_aead *aead;
1054 	struct crypto_authenc_key_param *param;
1055 	struct rtattr *rta;
1056 	char *key;
1057 	char *p;
1058 	char authenc_name[CRYPTO_MAX_ALG_NAME];
1059 	unsigned int keylen;
1060 	int err;
1061 
1062 	err = -EINVAL;
1063 	if (!x->ealg)
1064 		goto error;
1065 
1066 	err = -ENAMETOOLONG;
1067 
1068 	if ((x->props.flags & XFRM_STATE_ESN)) {
1069 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1070 			     "%s%sauthencesn(%s,%s)%s",
1071 			     x->geniv ?: "", x->geniv ? "(" : "",
1072 			     x->aalg ? x->aalg->alg_name : "digest_null",
1073 			     x->ealg->alg_name,
1074 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1075 			goto error;
1076 	} else {
1077 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1078 			     "%s%sauthenc(%s,%s)%s",
1079 			     x->geniv ?: "", x->geniv ? "(" : "",
1080 			     x->aalg ? x->aalg->alg_name : "digest_null",
1081 			     x->ealg->alg_name,
1082 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1083 			goto error;
1084 	}
1085 
1086 	aead = crypto_alloc_aead(authenc_name, 0, 0);
1087 	err = PTR_ERR(aead);
1088 	if (IS_ERR(aead))
1089 		goto error;
1090 
1091 	x->data = aead;
1092 
1093 	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1094 		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1095 	err = -ENOMEM;
1096 	key = kmalloc(keylen, GFP_KERNEL);
1097 	if (!key)
1098 		goto error;
1099 
1100 	p = key;
1101 	rta = (void *)p;
1102 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1103 	rta->rta_len = RTA_LENGTH(sizeof(*param));
1104 	param = RTA_DATA(rta);
1105 	p += RTA_SPACE(sizeof(*param));
1106 
1107 	if (x->aalg) {
1108 		struct xfrm_algo_desc *aalg_desc;
1109 
1110 		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1111 		p += (x->aalg->alg_key_len + 7) / 8;
1112 
1113 		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1114 		BUG_ON(!aalg_desc);
1115 
1116 		err = -EINVAL;
1117 		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1118 		    crypto_aead_authsize(aead)) {
1119 			pr_info("ESP: %s digestsize %u != %hu\n",
1120 				x->aalg->alg_name,
1121 				crypto_aead_authsize(aead),
1122 				aalg_desc->uinfo.auth.icv_fullbits / 8);
1123 			goto free_key;
1124 		}
1125 
1126 		err = crypto_aead_setauthsize(
1127 			aead, x->aalg->alg_trunc_len / 8);
1128 		if (err)
1129 			goto free_key;
1130 	}
1131 
1132 	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1133 	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1134 
1135 	err = crypto_aead_setkey(aead, key, keylen);
1136 
1137 free_key:
1138 	kfree(key);
1139 
1140 error:
1141 	return err;
1142 }
1143 
1144 static int esp_init_state(struct xfrm_state *x)
1145 {
1146 	struct crypto_aead *aead;
1147 	u32 align;
1148 	int err;
1149 
1150 	x->data = NULL;
1151 
1152 	if (x->aead)
1153 		err = esp_init_aead(x);
1154 	else
1155 		err = esp_init_authenc(x);
1156 
1157 	if (err)
1158 		goto error;
1159 
1160 	aead = x->data;
1161 
1162 	x->props.header_len = sizeof(struct ip_esp_hdr) +
1163 			      crypto_aead_ivsize(aead);
1164 	if (x->props.mode == XFRM_MODE_TUNNEL)
1165 		x->props.header_len += sizeof(struct iphdr);
1166 	else if (x->props.mode == XFRM_MODE_BEET && x->sel.family != AF_INET6)
1167 		x->props.header_len += IPV4_BEET_PHMAXLEN;
1168 	if (x->encap) {
1169 		struct xfrm_encap_tmpl *encap = x->encap;
1170 
1171 		switch (encap->encap_type) {
1172 		default:
1173 			err = -EINVAL;
1174 			goto error;
1175 		case UDP_ENCAP_ESPINUDP:
1176 			x->props.header_len += sizeof(struct udphdr);
1177 			break;
1178 		case UDP_ENCAP_ESPINUDP_NON_IKE:
1179 			x->props.header_len += sizeof(struct udphdr) + 2 * sizeof(u32);
1180 			break;
1181 #ifdef CONFIG_INET_ESPINTCP
1182 		case TCP_ENCAP_ESPINTCP:
1183 			/* only the length field, TCP encap is done by
1184 			 * the socket
1185 			 */
1186 			x->props.header_len += 2;
1187 			break;
1188 #endif
1189 		}
1190 	}
1191 
1192 	align = ALIGN(crypto_aead_blocksize(aead), 4);
1193 	x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1194 
1195 error:
1196 	return err;
1197 }
1198 
1199 static int esp4_rcv_cb(struct sk_buff *skb, int err)
1200 {
1201 	return 0;
1202 }
1203 
1204 static const struct xfrm_type esp_type =
1205 {
1206 	.description	= "ESP4",
1207 	.owner		= THIS_MODULE,
1208 	.proto	     	= IPPROTO_ESP,
1209 	.flags		= XFRM_TYPE_REPLAY_PROT,
1210 	.init_state	= esp_init_state,
1211 	.destructor	= esp_destroy,
1212 	.input		= esp_input,
1213 	.output		= esp_output,
1214 };
1215 
1216 static struct xfrm4_protocol esp4_protocol = {
1217 	.handler	=	xfrm4_rcv,
1218 	.input_handler	=	xfrm_input,
1219 	.cb_handler	=	esp4_rcv_cb,
1220 	.err_handler	=	esp4_err,
1221 	.priority	=	0,
1222 };
1223 
1224 static int __init esp4_init(void)
1225 {
1226 	if (xfrm_register_type(&esp_type, AF_INET) < 0) {
1227 		pr_info("%s: can't add xfrm type\n", __func__);
1228 		return -EAGAIN;
1229 	}
1230 	if (xfrm4_protocol_register(&esp4_protocol, IPPROTO_ESP) < 0) {
1231 		pr_info("%s: can't add protocol\n", __func__);
1232 		xfrm_unregister_type(&esp_type, AF_INET);
1233 		return -EAGAIN;
1234 	}
1235 	return 0;
1236 }
1237 
1238 static void __exit esp4_fini(void)
1239 {
1240 	if (xfrm4_protocol_deregister(&esp4_protocol, IPPROTO_ESP) < 0)
1241 		pr_info("%s: can't remove protocol\n", __func__);
1242 	xfrm_unregister_type(&esp_type, AF_INET);
1243 }
1244 
1245 module_init(esp4_init);
1246 module_exit(esp4_fini);
1247 MODULE_LICENSE("GPL");
1248 MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_ESP);
1249