xref: /linux/net/ipv4/esp4.c (revision 55f1b540d893da740a81200450014c45a8103f54)
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 #include <linux/skbuff_ref.h>
24 
25 #include <linux/highmem.h>
26 
27 struct esp_skb_cb {
28 	struct xfrm_skb_cb xfrm;
29 	void *tmp;
30 };
31 
32 struct esp_output_extra {
33 	__be32 seqhi;
34 	u32 esphoff;
35 };
36 
37 #define ESP_SKB_CB(__skb) ((struct esp_skb_cb *)&((__skb)->cb[0]))
38 
39 /*
40  * Allocate an AEAD request structure with extra space for SG and IV.
41  *
42  * For alignment considerations the IV is placed at the front, followed
43  * by the request and finally the SG list.
44  *
45  * TODO: Use spare space in skb for this where possible.
46  */
47 static void *esp_alloc_tmp(struct crypto_aead *aead, int nfrags, int extralen)
48 {
49 	unsigned int len;
50 
51 	len = extralen;
52 
53 	len += crypto_aead_ivsize(aead);
54 
55 	if (len) {
56 		len += crypto_aead_alignmask(aead) &
57 		       ~(crypto_tfm_ctx_alignment() - 1);
58 		len = ALIGN(len, crypto_tfm_ctx_alignment());
59 	}
60 
61 	len += sizeof(struct aead_request) + crypto_aead_reqsize(aead);
62 	len = ALIGN(len, __alignof__(struct scatterlist));
63 
64 	len += sizeof(struct scatterlist) * nfrags;
65 
66 	return kmalloc(len, GFP_ATOMIC);
67 }
68 
69 static inline void *esp_tmp_extra(void *tmp)
70 {
71 	return PTR_ALIGN(tmp, __alignof__(struct esp_output_extra));
72 }
73 
74 static inline u8 *esp_tmp_iv(struct crypto_aead *aead, void *tmp, int extralen)
75 {
76 	return crypto_aead_ivsize(aead) ?
77 	       PTR_ALIGN((u8 *)tmp + extralen,
78 			 crypto_aead_alignmask(aead) + 1) : tmp + extralen;
79 }
80 
81 static inline struct aead_request *esp_tmp_req(struct crypto_aead *aead, u8 *iv)
82 {
83 	struct aead_request *req;
84 
85 	req = (void *)PTR_ALIGN(iv + crypto_aead_ivsize(aead),
86 				crypto_tfm_ctx_alignment());
87 	aead_request_set_tfm(req, aead);
88 	return req;
89 }
90 
91 static inline struct scatterlist *esp_req_sg(struct crypto_aead *aead,
92 					     struct aead_request *req)
93 {
94 	return (void *)ALIGN((unsigned long)(req + 1) +
95 			     crypto_aead_reqsize(aead),
96 			     __alignof__(struct scatterlist));
97 }
98 
99 static void esp_ssg_unref(struct xfrm_state *x, void *tmp, struct sk_buff *skb)
100 {
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(struct esp_output_extra);
109 
110 	iv = esp_tmp_iv(aead, tmp, extralen);
111 	req = esp_tmp_req(aead, iv);
112 
113 	/* Unref skb_frag_pages in the src scatterlist if necessary.
114 	 * Skip the first sg which comes from skb->data.
115 	 */
116 	if (req->src != req->dst)
117 		for (sg = sg_next(req->src); sg; sg = sg_next(sg))
118 			skb_page_unref(page_to_netmem(sg_page(sg)),
119 				       skb->pp_recycle);
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 net *net = xs_net(x);
140 	struct esp_tcp_sk *esk;
141 	__be16 sport, dport;
142 	struct sock *nsk;
143 	struct sock *sk;
144 
145 	sk = rcu_dereference(x->encap_sk);
146 	if (sk && sk->sk_state == TCP_ESTABLISHED)
147 		return sk;
148 
149 	spin_lock_bh(&x->lock);
150 	sport = encap->encap_sport;
151 	dport = encap->encap_dport;
152 	nsk = rcu_dereference_protected(x->encap_sk,
153 					lockdep_is_held(&x->lock));
154 	if (sk && sk == nsk) {
155 		esk = kmalloc(sizeof(*esk), GFP_ATOMIC);
156 		if (!esk) {
157 			spin_unlock_bh(&x->lock);
158 			return ERR_PTR(-ENOMEM);
159 		}
160 		RCU_INIT_POINTER(x->encap_sk, NULL);
161 		esk->sk = sk;
162 		call_rcu(&esk->rcu, esp_free_tcp_sk);
163 	}
164 	spin_unlock_bh(&x->lock);
165 
166 	sk = inet_lookup_established(net, net->ipv4.tcp_death_row.hashinfo, x->id.daddr.a4,
167 				     dport, x->props.saddr.a4, sport, 0);
168 	if (!sk)
169 		return ERR_PTR(-ENOENT);
170 
171 	if (!tcp_is_ulp_esp(sk)) {
172 		sock_put(sk);
173 		return ERR_PTR(-EINVAL);
174 	}
175 
176 	spin_lock_bh(&x->lock);
177 	nsk = rcu_dereference_protected(x->encap_sk,
178 					lockdep_is_held(&x->lock));
179 	if (encap->encap_sport != sport ||
180 	    encap->encap_dport != dport) {
181 		sock_put(sk);
182 		sk = nsk ?: ERR_PTR(-EREMCHG);
183 	} else if (sk == nsk) {
184 		sock_put(sk);
185 	} else {
186 		rcu_assign_pointer(x->encap_sk, sk);
187 	}
188 	spin_unlock_bh(&x->lock);
189 
190 	return sk;
191 }
192 
193 static int esp_output_tcp_finish(struct xfrm_state *x, struct sk_buff *skb)
194 {
195 	struct sock *sk;
196 	int err;
197 
198 	rcu_read_lock();
199 
200 	sk = esp_find_tcp_sk(x);
201 	err = PTR_ERR_OR_ZERO(sk);
202 	if (err)
203 		goto out;
204 
205 	bh_lock_sock(sk);
206 	if (sock_owned_by_user(sk))
207 		err = espintcp_queue_out(sk, skb);
208 	else
209 		err = espintcp_push_skb(sk, skb);
210 	bh_unlock_sock(sk);
211 
212 out:
213 	rcu_read_unlock();
214 	return err;
215 }
216 
217 static int esp_output_tcp_encap_cb(struct net *net, struct sock *sk,
218 				   struct sk_buff *skb)
219 {
220 	struct dst_entry *dst = skb_dst(skb);
221 	struct xfrm_state *x = dst->xfrm;
222 
223 	return esp_output_tcp_finish(x, skb);
224 }
225 
226 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
227 {
228 	int err;
229 
230 	local_bh_disable();
231 	err = xfrm_trans_queue_net(xs_net(x), skb, esp_output_tcp_encap_cb);
232 	local_bh_enable();
233 
234 	/* EINPROGRESS just happens to do the right thing.  It
235 	 * actually means that the skb has been consumed and
236 	 * isn't coming back.
237 	 */
238 	return err ?: -EINPROGRESS;
239 }
240 #else
241 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
242 {
243 	WARN_ON(1);
244 	return -EOPNOTSUPP;
245 }
246 #endif
247 
248 static void esp_output_done(void *data, int err)
249 {
250 	struct sk_buff *skb = 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, skb);
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->sk, 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 	 * accommodate 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(void *data, int err)
337 {
338 	struct sk_buff *skb = data;
339 
340 	esp_output_restore_header(skb);
341 	esp_output_done(data, 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 	unsigned int len;
352 	struct xfrm_offload *xo = xfrm_offload(skb);
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 	/* For IPv4 ESP with UDP encapsulation, if xo is not null, the skb is in the crypto offload
365 	 * data path, which means that esp_output_udp_encap is called outside of the XFRM stack.
366 	 * In this case, the mac header doesn't point to the IPv4 protocol field, so don't set it.
367 	 */
368 	if (!xo || encap_type != UDP_ENCAP_ESPINUDP)
369 		*skb_mac_header(skb) = IPPROTO_UDP;
370 
371 	return (struct ip_esp_hdr *)(uh + 1);
372 }
373 
374 #ifdef CONFIG_INET_ESPINTCP
375 static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
376 						    struct sk_buff *skb,
377 						    struct esp_info *esp)
378 {
379 	__be16 *lenp = (void *)esp->esph;
380 	struct ip_esp_hdr *esph;
381 	unsigned int len;
382 	struct sock *sk;
383 
384 	len = skb->len + esp->tailen - skb_transport_offset(skb);
385 	if (len > IP_MAX_MTU)
386 		return ERR_PTR(-EMSGSIZE);
387 
388 	rcu_read_lock();
389 	sk = esp_find_tcp_sk(x);
390 	rcu_read_unlock();
391 
392 	if (IS_ERR(sk))
393 		return ERR_CAST(sk);
394 
395 	*lenp = htons(len);
396 	esph = (struct ip_esp_hdr *)(lenp + 1);
397 
398 	return esph;
399 }
400 #else
401 static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
402 						    struct sk_buff *skb,
403 						    struct esp_info *esp)
404 {
405 	return ERR_PTR(-EOPNOTSUPP);
406 }
407 #endif
408 
409 static int esp_output_encap(struct xfrm_state *x, struct sk_buff *skb,
410 			    struct esp_info *esp)
411 {
412 	struct xfrm_encap_tmpl *encap = x->encap;
413 	struct ip_esp_hdr *esph;
414 	__be16 sport, dport;
415 	int encap_type;
416 
417 	spin_lock_bh(&x->lock);
418 	sport = encap->encap_sport;
419 	dport = encap->encap_dport;
420 	encap_type = encap->encap_type;
421 	spin_unlock_bh(&x->lock);
422 
423 	switch (encap_type) {
424 	default:
425 	case UDP_ENCAP_ESPINUDP:
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 
450 	/* this is non-NULL only with TCP/UDP Encapsulation */
451 	if (x->encap) {
452 		int err = esp_output_encap(x, skb, esp);
453 
454 		if (err < 0)
455 			return err;
456 	}
457 
458 	if (ALIGN(tailen, L1_CACHE_BYTES) > PAGE_SIZE ||
459 	    ALIGN(skb->data_len, L1_CACHE_BYTES) > PAGE_SIZE)
460 		goto cow;
461 
462 	if (!skb_cloned(skb)) {
463 		if (tailen <= skb_tailroom(skb)) {
464 			nfrags = 1;
465 			trailer = skb;
466 			tail = skb_tail_pointer(trailer);
467 
468 			goto skip_cow;
469 		} else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
470 			   && !skb_has_frag_list(skb)) {
471 			int allocsize;
472 			struct sock *sk = skb->sk;
473 			struct page_frag *pfrag = &x->xfrag;
474 
475 			esp->inplace = false;
476 
477 			allocsize = ALIGN(tailen, L1_CACHE_BYTES);
478 
479 			spin_lock_bh(&x->lock);
480 
481 			if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
482 				spin_unlock_bh(&x->lock);
483 				goto cow;
484 			}
485 
486 			page = pfrag->page;
487 			get_page(page);
488 
489 			tail = page_address(page) + pfrag->offset;
490 
491 			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
492 
493 			nfrags = skb_shinfo(skb)->nr_frags;
494 
495 			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
496 					     tailen);
497 			skb_shinfo(skb)->nr_frags = ++nfrags;
498 
499 			pfrag->offset = pfrag->offset + allocsize;
500 
501 			spin_unlock_bh(&x->lock);
502 
503 			nfrags++;
504 
505 			skb_len_add(skb, 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, skb);
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 crypto_aead *aead = x->data;
707 	int alen, hlen, elen;
708 	int padlen, trimlen;
709 	__wsum csumdiff;
710 	u8 nexthdr[2];
711 	int ret;
712 
713 	alen = crypto_aead_authsize(aead);
714 	hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
715 	elen = skb->len - hlen;
716 
717 	if (skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2))
718 		BUG();
719 
720 	ret = -EINVAL;
721 	padlen = nexthdr[0];
722 	if (padlen + 2 + alen >= elen) {
723 		net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
724 				    padlen + 2, elen - alen);
725 		goto out;
726 	}
727 
728 	trimlen = alen + padlen + 2;
729 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
730 		csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
731 		skb->csum = csum_block_sub(skb->csum, csumdiff,
732 					   skb->len - trimlen);
733 	}
734 	ret = pskb_trim(skb, skb->len - trimlen);
735 	if (unlikely(ret))
736 		return ret;
737 
738 	ret = nexthdr[1];
739 
740 out:
741 	return ret;
742 }
743 
744 int esp_input_done2(struct sk_buff *skb, int err)
745 {
746 	const struct iphdr *iph;
747 	struct xfrm_state *x = xfrm_input_state(skb);
748 	struct xfrm_offload *xo = xfrm_offload(skb);
749 	struct crypto_aead *aead = x->data;
750 	int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
751 	int ihl;
752 
753 	if (!xo || !(xo->flags & CRYPTO_DONE))
754 		kfree(ESP_SKB_CB(skb)->tmp);
755 
756 	if (unlikely(err))
757 		goto out;
758 
759 	err = esp_remove_trailer(skb);
760 	if (unlikely(err < 0))
761 		goto out;
762 
763 	iph = ip_hdr(skb);
764 	ihl = iph->ihl * 4;
765 
766 	if (x->encap) {
767 		struct xfrm_encap_tmpl *encap = x->encap;
768 		struct tcphdr *th = (void *)(skb_network_header(skb) + ihl);
769 		struct udphdr *uh = (void *)(skb_network_header(skb) + ihl);
770 		__be16 source;
771 
772 		switch (x->encap->encap_type) {
773 		case TCP_ENCAP_ESPINTCP:
774 			source = th->source;
775 			break;
776 		case UDP_ENCAP_ESPINUDP:
777 			source = uh->source;
778 			break;
779 		default:
780 			WARN_ON_ONCE(1);
781 			err = -EINVAL;
782 			goto out;
783 		}
784 
785 		/*
786 		 * 1) if the NAT-T peer's IP or port changed then
787 		 *    advertise the change to the keying daemon.
788 		 *    This is an inbound SA, so just compare
789 		 *    SRC ports.
790 		 */
791 		if (iph->saddr != x->props.saddr.a4 ||
792 		    source != encap->encap_sport) {
793 			xfrm_address_t ipaddr;
794 
795 			ipaddr.a4 = iph->saddr;
796 			km_new_mapping(x, &ipaddr, source);
797 
798 			/* XXX: perhaps add an extra
799 			 * policy check here, to see
800 			 * if we should allow or
801 			 * reject a packet from a
802 			 * different source
803 			 * address/port.
804 			 */
805 		}
806 
807 		/*
808 		 * 2) ignore UDP/TCP checksums in case
809 		 *    of NAT-T in Transport Mode, or
810 		 *    perform other post-processing fixes
811 		 *    as per draft-ietf-ipsec-udp-encaps-06,
812 		 *    section 3.1.2
813 		 */
814 		if (x->props.mode == XFRM_MODE_TRANSPORT)
815 			skb->ip_summed = CHECKSUM_UNNECESSARY;
816 	}
817 
818 	skb_pull_rcsum(skb, hlen);
819 	if (x->props.mode == XFRM_MODE_TUNNEL)
820 		skb_reset_transport_header(skb);
821 	else
822 		skb_set_transport_header(skb, -ihl);
823 
824 	/* RFC4303: Drop dummy packets without any error */
825 	if (err == IPPROTO_NONE)
826 		err = -EINVAL;
827 
828 out:
829 	return err;
830 }
831 EXPORT_SYMBOL_GPL(esp_input_done2);
832 
833 static void esp_input_done(void *data, int err)
834 {
835 	struct sk_buff *skb = data;
836 
837 	xfrm_input_resume(skb, esp_input_done2(skb, err));
838 }
839 
840 static void esp_input_restore_header(struct sk_buff *skb)
841 {
842 	esp_restore_header(skb, 0);
843 	__skb_pull(skb, 4);
844 }
845 
846 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
847 {
848 	struct xfrm_state *x = xfrm_input_state(skb);
849 	struct ip_esp_hdr *esph;
850 
851 	/* For ESN we move the header forward by 4 bytes to
852 	 * accommodate the high bits.  We will move it back after
853 	 * decryption.
854 	 */
855 	if ((x->props.flags & XFRM_STATE_ESN)) {
856 		esph = skb_push(skb, 4);
857 		*seqhi = esph->spi;
858 		esph->spi = esph->seq_no;
859 		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
860 	}
861 }
862 
863 static void esp_input_done_esn(void *data, int err)
864 {
865 	struct sk_buff *skb = data;
866 
867 	esp_input_restore_header(skb);
868 	esp_input_done(data, err);
869 }
870 
871 /*
872  * Note: detecting truncated vs. non-truncated authentication data is very
873  * expensive, so we only support truncated data, which is the recommended
874  * and common case.
875  */
876 static int esp_input(struct xfrm_state *x, struct sk_buff *skb)
877 {
878 	struct crypto_aead *aead = x->data;
879 	struct aead_request *req;
880 	struct sk_buff *trailer;
881 	int ivlen = crypto_aead_ivsize(aead);
882 	int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
883 	int nfrags;
884 	int assoclen;
885 	int seqhilen;
886 	__be32 *seqhi;
887 	void *tmp;
888 	u8 *iv;
889 	struct scatterlist *sg;
890 	int err = -EINVAL;
891 
892 	if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen))
893 		goto out;
894 
895 	if (elen <= 0)
896 		goto out;
897 
898 	assoclen = sizeof(struct ip_esp_hdr);
899 	seqhilen = 0;
900 
901 	if (x->props.flags & XFRM_STATE_ESN) {
902 		seqhilen += sizeof(__be32);
903 		assoclen += seqhilen;
904 	}
905 
906 	if (!skb_cloned(skb)) {
907 		if (!skb_is_nonlinear(skb)) {
908 			nfrags = 1;
909 
910 			goto skip_cow;
911 		} else if (!skb_has_frag_list(skb)) {
912 			nfrags = skb_shinfo(skb)->nr_frags;
913 			nfrags++;
914 
915 			goto skip_cow;
916 		}
917 	}
918 
919 	err = skb_cow_data(skb, 0, &trailer);
920 	if (err < 0)
921 		goto out;
922 
923 	nfrags = err;
924 
925 skip_cow:
926 	err = -ENOMEM;
927 	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
928 	if (!tmp)
929 		goto out;
930 
931 	ESP_SKB_CB(skb)->tmp = tmp;
932 	seqhi = esp_tmp_extra(tmp);
933 	iv = esp_tmp_iv(aead, tmp, seqhilen);
934 	req = esp_tmp_req(aead, iv);
935 	sg = esp_req_sg(aead, req);
936 
937 	esp_input_set_header(skb, seqhi);
938 
939 	sg_init_table(sg, nfrags);
940 	err = skb_to_sgvec(skb, sg, 0, skb->len);
941 	if (unlikely(err < 0)) {
942 		kfree(tmp);
943 		goto out;
944 	}
945 
946 	skb->ip_summed = CHECKSUM_NONE;
947 
948 	if ((x->props.flags & XFRM_STATE_ESN))
949 		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
950 	else
951 		aead_request_set_callback(req, 0, esp_input_done, skb);
952 
953 	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
954 	aead_request_set_ad(req, assoclen);
955 
956 	err = crypto_aead_decrypt(req);
957 	if (err == -EINPROGRESS)
958 		goto out;
959 
960 	if ((x->props.flags & XFRM_STATE_ESN))
961 		esp_input_restore_header(skb);
962 
963 	err = esp_input_done2(skb, err);
964 
965 out:
966 	return err;
967 }
968 
969 static int esp4_err(struct sk_buff *skb, u32 info)
970 {
971 	struct net *net = dev_net(skb->dev);
972 	const struct iphdr *iph = (const struct iphdr *)skb->data;
973 	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data+(iph->ihl<<2));
974 	struct xfrm_state *x;
975 
976 	switch (icmp_hdr(skb)->type) {
977 	case ICMP_DEST_UNREACH:
978 		if (icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
979 			return 0;
980 		break;
981 	case ICMP_REDIRECT:
982 		break;
983 	default:
984 		return 0;
985 	}
986 
987 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
988 			      esph->spi, IPPROTO_ESP, AF_INET);
989 	if (!x)
990 		return 0;
991 
992 	if (icmp_hdr(skb)->type == ICMP_DEST_UNREACH)
993 		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ESP);
994 	else
995 		ipv4_redirect(skb, net, 0, IPPROTO_ESP);
996 	xfrm_state_put(x);
997 
998 	return 0;
999 }
1000 
1001 static void esp_destroy(struct xfrm_state *x)
1002 {
1003 	struct crypto_aead *aead = x->data;
1004 
1005 	if (!aead)
1006 		return;
1007 
1008 	crypto_free_aead(aead);
1009 }
1010 
1011 static int esp_init_aead(struct xfrm_state *x, struct netlink_ext_ack *extack)
1012 {
1013 	char aead_name[CRYPTO_MAX_ALG_NAME];
1014 	struct crypto_aead *aead;
1015 	int err;
1016 
1017 	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1018 		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME) {
1019 		NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1020 		return -ENAMETOOLONG;
1021 	}
1022 
1023 	aead = crypto_alloc_aead(aead_name, 0, 0);
1024 	err = PTR_ERR(aead);
1025 	if (IS_ERR(aead))
1026 		goto error;
1027 
1028 	x->data = aead;
1029 
1030 	err = crypto_aead_setkey(aead, x->aead->alg_key,
1031 				 (x->aead->alg_key_len + 7) / 8);
1032 	if (err)
1033 		goto error;
1034 
1035 	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1036 	if (err)
1037 		goto error;
1038 
1039 	return 0;
1040 
1041 error:
1042 	NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1043 	return err;
1044 }
1045 
1046 static int esp_init_authenc(struct xfrm_state *x,
1047 			    struct netlink_ext_ack *extack)
1048 {
1049 	struct crypto_aead *aead;
1050 	struct crypto_authenc_key_param *param;
1051 	struct rtattr *rta;
1052 	char *key;
1053 	char *p;
1054 	char authenc_name[CRYPTO_MAX_ALG_NAME];
1055 	unsigned int keylen;
1056 	int err;
1057 
1058 	err = -ENAMETOOLONG;
1059 
1060 	if ((x->props.flags & XFRM_STATE_ESN)) {
1061 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1062 			     "%s%sauthencesn(%s,%s)%s",
1063 			     x->geniv ?: "", x->geniv ? "(" : "",
1064 			     x->aalg ? x->aalg->alg_name : "digest_null",
1065 			     x->ealg->alg_name,
1066 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1067 			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1068 			goto error;
1069 		}
1070 	} else {
1071 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1072 			     "%s%sauthenc(%s,%s)%s",
1073 			     x->geniv ?: "", x->geniv ? "(" : "",
1074 			     x->aalg ? x->aalg->alg_name : "digest_null",
1075 			     x->ealg->alg_name,
1076 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1077 			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1078 			goto error;
1079 		}
1080 	}
1081 
1082 	aead = crypto_alloc_aead(authenc_name, 0, 0);
1083 	err = PTR_ERR(aead);
1084 	if (IS_ERR(aead)) {
1085 		NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1086 		goto error;
1087 	}
1088 
1089 	x->data = aead;
1090 
1091 	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1092 		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1093 	err = -ENOMEM;
1094 	key = kmalloc(keylen, GFP_KERNEL);
1095 	if (!key)
1096 		goto error;
1097 
1098 	p = key;
1099 	rta = (void *)p;
1100 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1101 	rta->rta_len = RTA_LENGTH(sizeof(*param));
1102 	param = RTA_DATA(rta);
1103 	p += RTA_SPACE(sizeof(*param));
1104 
1105 	if (x->aalg) {
1106 		struct xfrm_algo_desc *aalg_desc;
1107 
1108 		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1109 		p += (x->aalg->alg_key_len + 7) / 8;
1110 
1111 		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1112 		BUG_ON(!aalg_desc);
1113 
1114 		err = -EINVAL;
1115 		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1116 		    crypto_aead_authsize(aead)) {
1117 			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1118 			goto free_key;
1119 		}
1120 
1121 		err = crypto_aead_setauthsize(
1122 			aead, x->aalg->alg_trunc_len / 8);
1123 		if (err) {
1124 			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1125 			goto free_key;
1126 		}
1127 	}
1128 
1129 	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1130 	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1131 
1132 	err = crypto_aead_setkey(aead, key, keylen);
1133 
1134 free_key:
1135 	kfree_sensitive(key);
1136 
1137 error:
1138 	return err;
1139 }
1140 
1141 static int esp_init_state(struct xfrm_state *x, struct netlink_ext_ack *extack)
1142 {
1143 	struct crypto_aead *aead;
1144 	u32 align;
1145 	int err;
1146 
1147 	x->data = NULL;
1148 
1149 	if (x->aead) {
1150 		err = esp_init_aead(x, extack);
1151 	} else if (x->ealg) {
1152 		err = esp_init_authenc(x, extack);
1153 	} else {
1154 		NL_SET_ERR_MSG(extack, "ESP: AEAD or CRYPT must be provided");
1155 		err = -EINVAL;
1156 	}
1157 
1158 	if (err)
1159 		goto error;
1160 
1161 	aead = x->data;
1162 
1163 	x->props.header_len = sizeof(struct ip_esp_hdr) +
1164 			      crypto_aead_ivsize(aead);
1165 	if (x->props.mode == XFRM_MODE_TUNNEL)
1166 		x->props.header_len += sizeof(struct iphdr);
1167 	else if (x->props.mode == XFRM_MODE_BEET && x->sel.family != AF_INET6)
1168 		x->props.header_len += IPV4_BEET_PHMAXLEN;
1169 	if (x->encap) {
1170 		struct xfrm_encap_tmpl *encap = x->encap;
1171 
1172 		switch (encap->encap_type) {
1173 		default:
1174 			NL_SET_ERR_MSG(extack, "Unsupported encapsulation type for ESP");
1175 			err = -EINVAL;
1176 			goto error;
1177 		case UDP_ENCAP_ESPINUDP:
1178 			x->props.header_len += sizeof(struct udphdr);
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_DESCRIPTION("IPv4 ESP transformation library");
1246 MODULE_LICENSE("GPL");
1247 MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_ESP);
1248