xref: /linux/net/ipv4/esp4.c (revision 9052e9c95d908d6c3d7570aadc8898e1d871c8bb)
1 // SPDX-License-Identifier: GPL-2.0-only
2 #define pr_fmt(fmt) "IPsec: " fmt
3 
4 #include <crypto/aead.h>
5 #include <crypto/authenc.h>
6 #include <linux/err.h>
7 #include <linux/module.h>
8 #include <net/ip.h>
9 #include <net/xfrm.h>
10 #include <net/esp.h>
11 #include <linux/scatterlist.h>
12 #include <linux/kernel.h>
13 #include <linux/pfkeyv2.h>
14 #include <linux/rtnetlink.h>
15 #include <linux/slab.h>
16 #include <linux/spinlock.h>
17 #include <linux/in6.h>
18 #include <net/icmp.h>
19 #include <net/protocol.h>
20 #include <net/udp.h>
21 #include <net/tcp.h>
22 #include <net/espintcp.h>
23 
24 #include <linux/highmem.h>
25 
26 struct esp_skb_cb {
27 	struct xfrm_skb_cb xfrm;
28 	void *tmp;
29 };
30 
31 struct esp_output_extra {
32 	__be32 seqhi;
33 	u32 esphoff;
34 };
35 
36 #define ESP_SKB_CB(__skb) ((struct esp_skb_cb *)&((__skb)->cb[0]))
37 
38 /*
39  * Allocate an AEAD request structure with extra space for SG and IV.
40  *
41  * For alignment considerations the IV is placed at the front, followed
42  * by the request and finally the SG list.
43  *
44  * TODO: Use spare space in skb for this where possible.
45  */
46 static void *esp_alloc_tmp(struct crypto_aead *aead, int nfrags, int extralen)
47 {
48 	unsigned int len;
49 
50 	len = extralen;
51 
52 	len += crypto_aead_ivsize(aead);
53 
54 	if (len) {
55 		len += crypto_aead_alignmask(aead) &
56 		       ~(crypto_tfm_ctx_alignment() - 1);
57 		len = ALIGN(len, crypto_tfm_ctx_alignment());
58 	}
59 
60 	len += sizeof(struct aead_request) + crypto_aead_reqsize(aead);
61 	len = ALIGN(len, __alignof__(struct scatterlist));
62 
63 	len += sizeof(struct scatterlist) * nfrags;
64 
65 	return kmalloc(len, GFP_ATOMIC);
66 }
67 
68 static inline void *esp_tmp_extra(void *tmp)
69 {
70 	return PTR_ALIGN(tmp, __alignof__(struct esp_output_extra));
71 }
72 
73 static inline u8 *esp_tmp_iv(struct crypto_aead *aead, void *tmp, int extralen)
74 {
75 	return crypto_aead_ivsize(aead) ?
76 	       PTR_ALIGN((u8 *)tmp + extralen,
77 			 crypto_aead_alignmask(aead) + 1) : tmp + extralen;
78 }
79 
80 static inline struct aead_request *esp_tmp_req(struct crypto_aead *aead, u8 *iv)
81 {
82 	struct aead_request *req;
83 
84 	req = (void *)PTR_ALIGN(iv + crypto_aead_ivsize(aead),
85 				crypto_tfm_ctx_alignment());
86 	aead_request_set_tfm(req, aead);
87 	return req;
88 }
89 
90 static inline struct scatterlist *esp_req_sg(struct crypto_aead *aead,
91 					     struct aead_request *req)
92 {
93 	return (void *)ALIGN((unsigned long)(req + 1) +
94 			     crypto_aead_reqsize(aead),
95 			     __alignof__(struct scatterlist));
96 }
97 
98 static void esp_ssg_unref(struct xfrm_state *x, void *tmp)
99 {
100 	struct crypto_aead *aead = x->data;
101 	int extralen = 0;
102 	u8 *iv;
103 	struct aead_request *req;
104 	struct scatterlist *sg;
105 
106 	if (x->props.flags & XFRM_STATE_ESN)
107 		extralen += sizeof(struct esp_output_extra);
108 
109 	iv = esp_tmp_iv(aead, tmp, extralen);
110 	req = esp_tmp_req(aead, iv);
111 
112 	/* Unref skb_frag_pages in the src scatterlist if necessary.
113 	 * Skip the first sg which comes from skb->data.
114 	 */
115 	if (req->src != req->dst)
116 		for (sg = sg_next(req->src); sg; sg = sg_next(sg))
117 			put_page(sg_page(sg));
118 }
119 
120 #ifdef CONFIG_INET_ESPINTCP
121 struct esp_tcp_sk {
122 	struct sock *sk;
123 	struct rcu_head rcu;
124 };
125 
126 static void esp_free_tcp_sk(struct rcu_head *head)
127 {
128 	struct esp_tcp_sk *esk = container_of(head, struct esp_tcp_sk, rcu);
129 
130 	sock_put(esk->sk);
131 	kfree(esk);
132 }
133 
134 static struct sock *esp_find_tcp_sk(struct xfrm_state *x)
135 {
136 	struct xfrm_encap_tmpl *encap = x->encap;
137 	struct esp_tcp_sk *esk;
138 	__be16 sport, dport;
139 	struct sock *nsk;
140 	struct sock *sk;
141 
142 	sk = rcu_dereference(x->encap_sk);
143 	if (sk && sk->sk_state == TCP_ESTABLISHED)
144 		return sk;
145 
146 	spin_lock_bh(&x->lock);
147 	sport = encap->encap_sport;
148 	dport = encap->encap_dport;
149 	nsk = rcu_dereference_protected(x->encap_sk,
150 					lockdep_is_held(&x->lock));
151 	if (sk && sk == nsk) {
152 		esk = kmalloc(sizeof(*esk), GFP_ATOMIC);
153 		if (!esk) {
154 			spin_unlock_bh(&x->lock);
155 			return ERR_PTR(-ENOMEM);
156 		}
157 		RCU_INIT_POINTER(x->encap_sk, NULL);
158 		esk->sk = sk;
159 		call_rcu(&esk->rcu, esp_free_tcp_sk);
160 	}
161 	spin_unlock_bh(&x->lock);
162 
163 	sk = inet_lookup_established(xs_net(x), &tcp_hashinfo, x->id.daddr.a4,
164 				     dport, x->props.saddr.a4, sport, 0);
165 	if (!sk)
166 		return ERR_PTR(-ENOENT);
167 
168 	if (!tcp_is_ulp_esp(sk)) {
169 		sock_put(sk);
170 		return ERR_PTR(-EINVAL);
171 	}
172 
173 	spin_lock_bh(&x->lock);
174 	nsk = rcu_dereference_protected(x->encap_sk,
175 					lockdep_is_held(&x->lock));
176 	if (encap->encap_sport != sport ||
177 	    encap->encap_dport != dport) {
178 		sock_put(sk);
179 		sk = nsk ?: ERR_PTR(-EREMCHG);
180 	} else if (sk == nsk) {
181 		sock_put(sk);
182 	} else {
183 		rcu_assign_pointer(x->encap_sk, sk);
184 	}
185 	spin_unlock_bh(&x->lock);
186 
187 	return sk;
188 }
189 
190 static int esp_output_tcp_finish(struct xfrm_state *x, struct sk_buff *skb)
191 {
192 	struct sock *sk;
193 	int err;
194 
195 	rcu_read_lock();
196 
197 	sk = esp_find_tcp_sk(x);
198 	err = PTR_ERR_OR_ZERO(sk);
199 	if (err)
200 		goto out;
201 
202 	bh_lock_sock(sk);
203 	if (sock_owned_by_user(sk))
204 		err = espintcp_queue_out(sk, skb);
205 	else
206 		err = espintcp_push_skb(sk, skb);
207 	bh_unlock_sock(sk);
208 
209 out:
210 	rcu_read_unlock();
211 	return err;
212 }
213 
214 static int esp_output_tcp_encap_cb(struct net *net, struct sock *sk,
215 				   struct sk_buff *skb)
216 {
217 	struct dst_entry *dst = skb_dst(skb);
218 	struct xfrm_state *x = dst->xfrm;
219 
220 	return esp_output_tcp_finish(x, skb);
221 }
222 
223 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
224 {
225 	int err;
226 
227 	local_bh_disable();
228 	err = xfrm_trans_queue_net(xs_net(x), skb, esp_output_tcp_encap_cb);
229 	local_bh_enable();
230 
231 	/* EINPROGRESS just happens to do the right thing.  It
232 	 * actually means that the skb has been consumed and
233 	 * isn't coming back.
234 	 */
235 	return err ?: -EINPROGRESS;
236 }
237 #else
238 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
239 {
240 	kfree_skb(skb);
241 
242 	return -EOPNOTSUPP;
243 }
244 #endif
245 
246 static void esp_output_done(struct crypto_async_request *base, int err)
247 {
248 	struct sk_buff *skb = base->data;
249 	struct xfrm_offload *xo = xfrm_offload(skb);
250 	void *tmp;
251 	struct xfrm_state *x;
252 
253 	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
254 		struct sec_path *sp = skb_sec_path(skb);
255 
256 		x = sp->xvec[sp->len - 1];
257 	} else {
258 		x = skb_dst(skb)->xfrm;
259 	}
260 
261 	tmp = ESP_SKB_CB(skb)->tmp;
262 	esp_ssg_unref(x, tmp);
263 	kfree(tmp);
264 
265 	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
266 		if (err) {
267 			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
268 			kfree_skb(skb);
269 			return;
270 		}
271 
272 		skb_push(skb, skb->data - skb_mac_header(skb));
273 		secpath_reset(skb);
274 		xfrm_dev_resume(skb);
275 	} else {
276 		if (!err &&
277 		    x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
278 			esp_output_tail_tcp(x, skb);
279 		else
280 			xfrm_output_resume(skb->sk, skb, err);
281 	}
282 }
283 
284 /* Move ESP header back into place. */
285 static void esp_restore_header(struct sk_buff *skb, unsigned int offset)
286 {
287 	struct ip_esp_hdr *esph = (void *)(skb->data + offset);
288 	void *tmp = ESP_SKB_CB(skb)->tmp;
289 	__be32 *seqhi = esp_tmp_extra(tmp);
290 
291 	esph->seq_no = esph->spi;
292 	esph->spi = *seqhi;
293 }
294 
295 static void esp_output_restore_header(struct sk_buff *skb)
296 {
297 	void *tmp = ESP_SKB_CB(skb)->tmp;
298 	struct esp_output_extra *extra = esp_tmp_extra(tmp);
299 
300 	esp_restore_header(skb, skb_transport_offset(skb) + extra->esphoff -
301 				sizeof(__be32));
302 }
303 
304 static struct ip_esp_hdr *esp_output_set_extra(struct sk_buff *skb,
305 					       struct xfrm_state *x,
306 					       struct ip_esp_hdr *esph,
307 					       struct esp_output_extra *extra)
308 {
309 	/* For ESN we move the header forward by 4 bytes to
310 	 * accommodate the high bits.  We will move it back after
311 	 * encryption.
312 	 */
313 	if ((x->props.flags & XFRM_STATE_ESN)) {
314 		__u32 seqhi;
315 		struct xfrm_offload *xo = xfrm_offload(skb);
316 
317 		if (xo)
318 			seqhi = xo->seq.hi;
319 		else
320 			seqhi = XFRM_SKB_CB(skb)->seq.output.hi;
321 
322 		extra->esphoff = (unsigned char *)esph -
323 				 skb_transport_header(skb);
324 		esph = (struct ip_esp_hdr *)((unsigned char *)esph - 4);
325 		extra->seqhi = esph->spi;
326 		esph->seq_no = htonl(seqhi);
327 	}
328 
329 	esph->spi = x->id.spi;
330 
331 	return esph;
332 }
333 
334 static void esp_output_done_esn(struct crypto_async_request *base, int err)
335 {
336 	struct sk_buff *skb = base->data;
337 
338 	esp_output_restore_header(skb);
339 	esp_output_done(base, err);
340 }
341 
342 static struct ip_esp_hdr *esp_output_udp_encap(struct sk_buff *skb,
343 					       int encap_type,
344 					       struct esp_info *esp,
345 					       __be16 sport,
346 					       __be16 dport)
347 {
348 	struct udphdr *uh;
349 	__be32 *udpdata32;
350 	unsigned int len;
351 
352 	len = skb->len + esp->tailen - skb_transport_offset(skb);
353 	if (len + sizeof(struct iphdr) > IP_MAX_MTU)
354 		return ERR_PTR(-EMSGSIZE);
355 
356 	uh = (struct udphdr *)esp->esph;
357 	uh->source = sport;
358 	uh->dest = dport;
359 	uh->len = htons(len);
360 	uh->check = 0;
361 
362 	*skb_mac_header(skb) = IPPROTO_UDP;
363 
364 	if (encap_type == UDP_ENCAP_ESPINUDP_NON_IKE) {
365 		udpdata32 = (__be32 *)(uh + 1);
366 		udpdata32[0] = udpdata32[1] = 0;
367 		return (struct ip_esp_hdr *)(udpdata32 + 2);
368 	}
369 
370 	return (struct ip_esp_hdr *)(uh + 1);
371 }
372 
373 #ifdef CONFIG_INET_ESPINTCP
374 static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
375 						    struct sk_buff *skb,
376 						    struct esp_info *esp)
377 {
378 	__be16 *lenp = (void *)esp->esph;
379 	struct ip_esp_hdr *esph;
380 	unsigned int len;
381 	struct sock *sk;
382 
383 	len = skb->len + esp->tailen - skb_transport_offset(skb);
384 	if (len > IP_MAX_MTU)
385 		return ERR_PTR(-EMSGSIZE);
386 
387 	rcu_read_lock();
388 	sk = esp_find_tcp_sk(x);
389 	rcu_read_unlock();
390 
391 	if (IS_ERR(sk))
392 		return ERR_CAST(sk);
393 
394 	*lenp = htons(len);
395 	esph = (struct ip_esp_hdr *)(lenp + 1);
396 
397 	return esph;
398 }
399 #else
400 static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
401 						    struct sk_buff *skb,
402 						    struct esp_info *esp)
403 {
404 	return ERR_PTR(-EOPNOTSUPP);
405 }
406 #endif
407 
408 static int esp_output_encap(struct xfrm_state *x, struct sk_buff *skb,
409 			    struct esp_info *esp)
410 {
411 	struct xfrm_encap_tmpl *encap = x->encap;
412 	struct ip_esp_hdr *esph;
413 	__be16 sport, dport;
414 	int encap_type;
415 
416 	spin_lock_bh(&x->lock);
417 	sport = encap->encap_sport;
418 	dport = encap->encap_dport;
419 	encap_type = encap->encap_type;
420 	spin_unlock_bh(&x->lock);
421 
422 	switch (encap_type) {
423 	default:
424 	case UDP_ENCAP_ESPINUDP:
425 	case UDP_ENCAP_ESPINUDP_NON_IKE:
426 		esph = esp_output_udp_encap(skb, encap_type, esp, sport, dport);
427 		break;
428 	case TCP_ENCAP_ESPINTCP:
429 		esph = esp_output_tcp_encap(x, skb, esp);
430 		break;
431 	}
432 
433 	if (IS_ERR(esph))
434 		return PTR_ERR(esph);
435 
436 	esp->esph = esph;
437 
438 	return 0;
439 }
440 
441 int esp_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
442 {
443 	u8 *tail;
444 	int nfrags;
445 	int esph_offset;
446 	struct page *page;
447 	struct sk_buff *trailer;
448 	int tailen = esp->tailen;
449 
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 (!skb_cloned(skb)) {
459 		if (tailen <= skb_tailroom(skb)) {
460 			nfrags = 1;
461 			trailer = skb;
462 			tail = skb_tail_pointer(trailer);
463 
464 			goto skip_cow;
465 		} else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
466 			   && !skb_has_frag_list(skb)) {
467 			int allocsize;
468 			struct sock *sk = skb->sk;
469 			struct page_frag *pfrag = &x->xfrag;
470 
471 			esp->inplace = false;
472 
473 			allocsize = ALIGN(tailen, L1_CACHE_BYTES);
474 
475 			spin_lock_bh(&x->lock);
476 
477 			if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
478 				spin_unlock_bh(&x->lock);
479 				goto cow;
480 			}
481 
482 			page = pfrag->page;
483 			get_page(page);
484 
485 			tail = page_address(page) + pfrag->offset;
486 
487 			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
488 
489 			nfrags = skb_shinfo(skb)->nr_frags;
490 
491 			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
492 					     tailen);
493 			skb_shinfo(skb)->nr_frags = ++nfrags;
494 
495 			pfrag->offset = pfrag->offset + allocsize;
496 
497 			spin_unlock_bh(&x->lock);
498 
499 			nfrags++;
500 
501 			skb->len += tailen;
502 			skb->data_len += tailen;
503 			skb->truesize += tailen;
504 			if (sk && sk_fullsock(sk))
505 				refcount_add(tailen, &sk->sk_wmem_alloc);
506 
507 			goto out;
508 		}
509 	}
510 
511 cow:
512 	esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
513 
514 	nfrags = skb_cow_data(skb, tailen, &trailer);
515 	if (nfrags < 0)
516 		goto out;
517 	tail = skb_tail_pointer(trailer);
518 	esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
519 
520 skip_cow:
521 	esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
522 	pskb_put(skb, trailer, tailen);
523 
524 out:
525 	return nfrags;
526 }
527 EXPORT_SYMBOL_GPL(esp_output_head);
528 
529 int esp_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
530 {
531 	u8 *iv;
532 	int alen;
533 	void *tmp;
534 	int ivlen;
535 	int assoclen;
536 	int extralen;
537 	struct page *page;
538 	struct ip_esp_hdr *esph;
539 	struct crypto_aead *aead;
540 	struct aead_request *req;
541 	struct scatterlist *sg, *dsg;
542 	struct esp_output_extra *extra;
543 	int err = -ENOMEM;
544 
545 	assoclen = sizeof(struct ip_esp_hdr);
546 	extralen = 0;
547 
548 	if (x->props.flags & XFRM_STATE_ESN) {
549 		extralen += sizeof(*extra);
550 		assoclen += sizeof(__be32);
551 	}
552 
553 	aead = x->data;
554 	alen = crypto_aead_authsize(aead);
555 	ivlen = crypto_aead_ivsize(aead);
556 
557 	tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
558 	if (!tmp)
559 		goto error;
560 
561 	extra = esp_tmp_extra(tmp);
562 	iv = esp_tmp_iv(aead, tmp, extralen);
563 	req = esp_tmp_req(aead, iv);
564 	sg = esp_req_sg(aead, req);
565 
566 	if (esp->inplace)
567 		dsg = sg;
568 	else
569 		dsg = &sg[esp->nfrags];
570 
571 	esph = esp_output_set_extra(skb, x, esp->esph, extra);
572 	esp->esph = esph;
573 
574 	sg_init_table(sg, esp->nfrags);
575 	err = skb_to_sgvec(skb, sg,
576 		           (unsigned char *)esph - skb->data,
577 		           assoclen + ivlen + esp->clen + alen);
578 	if (unlikely(err < 0))
579 		goto error_free;
580 
581 	if (!esp->inplace) {
582 		int allocsize;
583 		struct page_frag *pfrag = &x->xfrag;
584 
585 		allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
586 
587 		spin_lock_bh(&x->lock);
588 		if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
589 			spin_unlock_bh(&x->lock);
590 			goto error_free;
591 		}
592 
593 		skb_shinfo(skb)->nr_frags = 1;
594 
595 		page = pfrag->page;
596 		get_page(page);
597 		/* replace page frags in skb with new page */
598 		__skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
599 		pfrag->offset = pfrag->offset + allocsize;
600 		spin_unlock_bh(&x->lock);
601 
602 		sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
603 		err = skb_to_sgvec(skb, dsg,
604 			           (unsigned char *)esph - skb->data,
605 			           assoclen + ivlen + esp->clen + alen);
606 		if (unlikely(err < 0))
607 			goto error_free;
608 	}
609 
610 	if ((x->props.flags & XFRM_STATE_ESN))
611 		aead_request_set_callback(req, 0, esp_output_done_esn, skb);
612 	else
613 		aead_request_set_callback(req, 0, esp_output_done, skb);
614 
615 	aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
616 	aead_request_set_ad(req, assoclen);
617 
618 	memset(iv, 0, ivlen);
619 	memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
620 	       min(ivlen, 8));
621 
622 	ESP_SKB_CB(skb)->tmp = tmp;
623 	err = crypto_aead_encrypt(req);
624 
625 	switch (err) {
626 	case -EINPROGRESS:
627 		goto error;
628 
629 	case -ENOSPC:
630 		err = NET_XMIT_DROP;
631 		break;
632 
633 	case 0:
634 		if ((x->props.flags & XFRM_STATE_ESN))
635 			esp_output_restore_header(skb);
636 	}
637 
638 	if (sg != dsg)
639 		esp_ssg_unref(x, tmp);
640 
641 	if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
642 		err = esp_output_tail_tcp(x, skb);
643 
644 error_free:
645 	kfree(tmp);
646 error:
647 	return err;
648 }
649 EXPORT_SYMBOL_GPL(esp_output_tail);
650 
651 static int esp_output(struct xfrm_state *x, struct sk_buff *skb)
652 {
653 	int alen;
654 	int blksize;
655 	struct ip_esp_hdr *esph;
656 	struct crypto_aead *aead;
657 	struct esp_info esp;
658 
659 	esp.inplace = true;
660 
661 	esp.proto = *skb_mac_header(skb);
662 	*skb_mac_header(skb) = IPPROTO_ESP;
663 
664 	/* skb is pure payload to encrypt */
665 
666 	aead = x->data;
667 	alen = crypto_aead_authsize(aead);
668 
669 	esp.tfclen = 0;
670 	if (x->tfcpad) {
671 		struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
672 		u32 padto;
673 
674 		padto = min(x->tfcpad, __xfrm_state_mtu(x, dst->child_mtu_cached));
675 		if (skb->len < padto)
676 			esp.tfclen = padto - skb->len;
677 	}
678 	blksize = ALIGN(crypto_aead_blocksize(aead), 4);
679 	esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
680 	esp.plen = esp.clen - skb->len - esp.tfclen;
681 	esp.tailen = esp.tfclen + esp.plen + alen;
682 
683 	esp.esph = ip_esp_hdr(skb);
684 
685 	esp.nfrags = esp_output_head(x, skb, &esp);
686 	if (esp.nfrags < 0)
687 		return esp.nfrags;
688 
689 	esph = esp.esph;
690 	esph->spi = x->id.spi;
691 
692 	esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
693 	esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
694 				 ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
695 
696 	skb_push(skb, -skb_network_offset(skb));
697 
698 	return esp_output_tail(x, skb, &esp);
699 }
700 
701 static inline int esp_remove_trailer(struct sk_buff *skb)
702 {
703 	struct xfrm_state *x = xfrm_input_state(skb);
704 	struct xfrm_offload *xo = xfrm_offload(skb);
705 	struct crypto_aead *aead = x->data;
706 	int alen, hlen, elen;
707 	int padlen, trimlen;
708 	__wsum csumdiff;
709 	u8 nexthdr[2];
710 	int ret;
711 
712 	alen = crypto_aead_authsize(aead);
713 	hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
714 	elen = skb->len - hlen;
715 
716 	if (xo && (xo->flags & XFRM_ESP_NO_TRAILER)) {
717 		ret = xo->proto;
718 		goto out;
719 	}
720 
721 	if (skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2))
722 		BUG();
723 
724 	ret = -EINVAL;
725 	padlen = nexthdr[0];
726 	if (padlen + 2 + alen >= elen) {
727 		net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
728 				    padlen + 2, elen - alen);
729 		goto out;
730 	}
731 
732 	trimlen = alen + padlen + 2;
733 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
734 		csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
735 		skb->csum = csum_block_sub(skb->csum, csumdiff,
736 					   skb->len - trimlen);
737 	}
738 	pskb_trim(skb, skb->len - trimlen);
739 
740 	ret = nexthdr[1];
741 
742 out:
743 	return ret;
744 }
745 
746 int esp_input_done2(struct sk_buff *skb, int err)
747 {
748 	const struct iphdr *iph;
749 	struct xfrm_state *x = xfrm_input_state(skb);
750 	struct xfrm_offload *xo = xfrm_offload(skb);
751 	struct crypto_aead *aead = x->data;
752 	int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
753 	int ihl;
754 
755 	if (!xo || !(xo->flags & CRYPTO_DONE))
756 		kfree(ESP_SKB_CB(skb)->tmp);
757 
758 	if (unlikely(err))
759 		goto out;
760 
761 	err = esp_remove_trailer(skb);
762 	if (unlikely(err < 0))
763 		goto out;
764 
765 	iph = ip_hdr(skb);
766 	ihl = iph->ihl * 4;
767 
768 	if (x->encap) {
769 		struct xfrm_encap_tmpl *encap = x->encap;
770 		struct tcphdr *th = (void *)(skb_network_header(skb) + ihl);
771 		struct udphdr *uh = (void *)(skb_network_header(skb) + ihl);
772 		__be16 source;
773 
774 		switch (x->encap->encap_type) {
775 		case TCP_ENCAP_ESPINTCP:
776 			source = th->source;
777 			break;
778 		case UDP_ENCAP_ESPINUDP:
779 		case UDP_ENCAP_ESPINUDP_NON_IKE:
780 			source = uh->source;
781 			break;
782 		default:
783 			WARN_ON_ONCE(1);
784 			err = -EINVAL;
785 			goto out;
786 		}
787 
788 		/*
789 		 * 1) if the NAT-T peer's IP or port changed then
790 		 *    advertize the change to the keying daemon.
791 		 *    This is an inbound SA, so just compare
792 		 *    SRC ports.
793 		 */
794 		if (iph->saddr != x->props.saddr.a4 ||
795 		    source != encap->encap_sport) {
796 			xfrm_address_t ipaddr;
797 
798 			ipaddr.a4 = iph->saddr;
799 			km_new_mapping(x, &ipaddr, source);
800 
801 			/* XXX: perhaps add an extra
802 			 * policy check here, to see
803 			 * if we should allow or
804 			 * reject a packet from a
805 			 * different source
806 			 * address/port.
807 			 */
808 		}
809 
810 		/*
811 		 * 2) ignore UDP/TCP checksums in case
812 		 *    of NAT-T in Transport Mode, or
813 		 *    perform other post-processing fixes
814 		 *    as per draft-ietf-ipsec-udp-encaps-06,
815 		 *    section 3.1.2
816 		 */
817 		if (x->props.mode == XFRM_MODE_TRANSPORT)
818 			skb->ip_summed = CHECKSUM_UNNECESSARY;
819 	}
820 
821 	skb_pull_rcsum(skb, hlen);
822 	if (x->props.mode == XFRM_MODE_TUNNEL)
823 		skb_reset_transport_header(skb);
824 	else
825 		skb_set_transport_header(skb, -ihl);
826 
827 	/* RFC4303: Drop dummy packets without any error */
828 	if (err == IPPROTO_NONE)
829 		err = -EINVAL;
830 
831 out:
832 	return err;
833 }
834 EXPORT_SYMBOL_GPL(esp_input_done2);
835 
836 static void esp_input_done(struct crypto_async_request *base, int err)
837 {
838 	struct sk_buff *skb = base->data;
839 
840 	xfrm_input_resume(skb, esp_input_done2(skb, err));
841 }
842 
843 static void esp_input_restore_header(struct sk_buff *skb)
844 {
845 	esp_restore_header(skb, 0);
846 	__skb_pull(skb, 4);
847 }
848 
849 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
850 {
851 	struct xfrm_state *x = xfrm_input_state(skb);
852 	struct ip_esp_hdr *esph;
853 
854 	/* For ESN we move the header forward by 4 bytes to
855 	 * accommodate the high bits.  We will move it back after
856 	 * decryption.
857 	 */
858 	if ((x->props.flags & XFRM_STATE_ESN)) {
859 		esph = skb_push(skb, 4);
860 		*seqhi = esph->spi;
861 		esph->spi = esph->seq_no;
862 		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
863 	}
864 }
865 
866 static void esp_input_done_esn(struct crypto_async_request *base, int err)
867 {
868 	struct sk_buff *skb = base->data;
869 
870 	esp_input_restore_header(skb);
871 	esp_input_done(base, err);
872 }
873 
874 /*
875  * Note: detecting truncated vs. non-truncated authentication data is very
876  * expensive, so we only support truncated data, which is the recommended
877  * and common case.
878  */
879 static int esp_input(struct xfrm_state *x, struct sk_buff *skb)
880 {
881 	struct crypto_aead *aead = x->data;
882 	struct aead_request *req;
883 	struct sk_buff *trailer;
884 	int ivlen = crypto_aead_ivsize(aead);
885 	int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
886 	int nfrags;
887 	int assoclen;
888 	int seqhilen;
889 	__be32 *seqhi;
890 	void *tmp;
891 	u8 *iv;
892 	struct scatterlist *sg;
893 	int err = -EINVAL;
894 
895 	if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen))
896 		goto out;
897 
898 	if (elen <= 0)
899 		goto out;
900 
901 	assoclen = sizeof(struct ip_esp_hdr);
902 	seqhilen = 0;
903 
904 	if (x->props.flags & XFRM_STATE_ESN) {
905 		seqhilen += sizeof(__be32);
906 		assoclen += seqhilen;
907 	}
908 
909 	if (!skb_cloned(skb)) {
910 		if (!skb_is_nonlinear(skb)) {
911 			nfrags = 1;
912 
913 			goto skip_cow;
914 		} else if (!skb_has_frag_list(skb)) {
915 			nfrags = skb_shinfo(skb)->nr_frags;
916 			nfrags++;
917 
918 			goto skip_cow;
919 		}
920 	}
921 
922 	err = skb_cow_data(skb, 0, &trailer);
923 	if (err < 0)
924 		goto out;
925 
926 	nfrags = err;
927 
928 skip_cow:
929 	err = -ENOMEM;
930 	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
931 	if (!tmp)
932 		goto out;
933 
934 	ESP_SKB_CB(skb)->tmp = tmp;
935 	seqhi = esp_tmp_extra(tmp);
936 	iv = esp_tmp_iv(aead, tmp, seqhilen);
937 	req = esp_tmp_req(aead, iv);
938 	sg = esp_req_sg(aead, req);
939 
940 	esp_input_set_header(skb, seqhi);
941 
942 	sg_init_table(sg, nfrags);
943 	err = skb_to_sgvec(skb, sg, 0, skb->len);
944 	if (unlikely(err < 0)) {
945 		kfree(tmp);
946 		goto out;
947 	}
948 
949 	skb->ip_summed = CHECKSUM_NONE;
950 
951 	if ((x->props.flags & XFRM_STATE_ESN))
952 		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
953 	else
954 		aead_request_set_callback(req, 0, esp_input_done, skb);
955 
956 	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
957 	aead_request_set_ad(req, assoclen);
958 
959 	err = crypto_aead_decrypt(req);
960 	if (err == -EINPROGRESS)
961 		goto out;
962 
963 	if ((x->props.flags & XFRM_STATE_ESN))
964 		esp_input_restore_header(skb);
965 
966 	err = esp_input_done2(skb, err);
967 
968 out:
969 	return err;
970 }
971 
972 static int esp4_err(struct sk_buff *skb, u32 info)
973 {
974 	struct net *net = dev_net(skb->dev);
975 	const struct iphdr *iph = (const struct iphdr *)skb->data;
976 	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data+(iph->ihl<<2));
977 	struct xfrm_state *x;
978 
979 	switch (icmp_hdr(skb)->type) {
980 	case ICMP_DEST_UNREACH:
981 		if (icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
982 			return 0;
983 		break;
984 	case ICMP_REDIRECT:
985 		break;
986 	default:
987 		return 0;
988 	}
989 
990 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
991 			      esph->spi, IPPROTO_ESP, AF_INET);
992 	if (!x)
993 		return 0;
994 
995 	if (icmp_hdr(skb)->type == ICMP_DEST_UNREACH)
996 		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ESP);
997 	else
998 		ipv4_redirect(skb, net, 0, IPPROTO_ESP);
999 	xfrm_state_put(x);
1000 
1001 	return 0;
1002 }
1003 
1004 static void esp_destroy(struct xfrm_state *x)
1005 {
1006 	struct crypto_aead *aead = x->data;
1007 
1008 	if (!aead)
1009 		return;
1010 
1011 	crypto_free_aead(aead);
1012 }
1013 
1014 static int esp_init_aead(struct xfrm_state *x)
1015 {
1016 	char aead_name[CRYPTO_MAX_ALG_NAME];
1017 	struct crypto_aead *aead;
1018 	int err;
1019 
1020 	err = -ENAMETOOLONG;
1021 	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1022 		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME)
1023 		goto error;
1024 
1025 	aead = crypto_alloc_aead(aead_name, 0, 0);
1026 	err = PTR_ERR(aead);
1027 	if (IS_ERR(aead))
1028 		goto error;
1029 
1030 	x->data = aead;
1031 
1032 	err = crypto_aead_setkey(aead, x->aead->alg_key,
1033 				 (x->aead->alg_key_len + 7) / 8);
1034 	if (err)
1035 		goto error;
1036 
1037 	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1038 	if (err)
1039 		goto error;
1040 
1041 error:
1042 	return err;
1043 }
1044 
1045 static int esp_init_authenc(struct xfrm_state *x)
1046 {
1047 	struct crypto_aead *aead;
1048 	struct crypto_authenc_key_param *param;
1049 	struct rtattr *rta;
1050 	char *key;
1051 	char *p;
1052 	char authenc_name[CRYPTO_MAX_ALG_NAME];
1053 	unsigned int keylen;
1054 	int err;
1055 
1056 	err = -EINVAL;
1057 	if (!x->ealg)
1058 		goto error;
1059 
1060 	err = -ENAMETOOLONG;
1061 
1062 	if ((x->props.flags & XFRM_STATE_ESN)) {
1063 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1064 			     "%s%sauthencesn(%s,%s)%s",
1065 			     x->geniv ?: "", x->geniv ? "(" : "",
1066 			     x->aalg ? x->aalg->alg_name : "digest_null",
1067 			     x->ealg->alg_name,
1068 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1069 			goto error;
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 			goto error;
1078 	}
1079 
1080 	aead = crypto_alloc_aead(authenc_name, 0, 0);
1081 	err = PTR_ERR(aead);
1082 	if (IS_ERR(aead))
1083 		goto error;
1084 
1085 	x->data = aead;
1086 
1087 	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1088 		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1089 	err = -ENOMEM;
1090 	key = kmalloc(keylen, GFP_KERNEL);
1091 	if (!key)
1092 		goto error;
1093 
1094 	p = key;
1095 	rta = (void *)p;
1096 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1097 	rta->rta_len = RTA_LENGTH(sizeof(*param));
1098 	param = RTA_DATA(rta);
1099 	p += RTA_SPACE(sizeof(*param));
1100 
1101 	if (x->aalg) {
1102 		struct xfrm_algo_desc *aalg_desc;
1103 
1104 		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1105 		p += (x->aalg->alg_key_len + 7) / 8;
1106 
1107 		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1108 		BUG_ON(!aalg_desc);
1109 
1110 		err = -EINVAL;
1111 		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1112 		    crypto_aead_authsize(aead)) {
1113 			pr_info("ESP: %s digestsize %u != %hu\n",
1114 				x->aalg->alg_name,
1115 				crypto_aead_authsize(aead),
1116 				aalg_desc->uinfo.auth.icv_fullbits / 8);
1117 			goto free_key;
1118 		}
1119 
1120 		err = crypto_aead_setauthsize(
1121 			aead, x->aalg->alg_trunc_len / 8);
1122 		if (err)
1123 			goto free_key;
1124 	}
1125 
1126 	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1127 	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1128 
1129 	err = crypto_aead_setkey(aead, key, keylen);
1130 
1131 free_key:
1132 	kfree(key);
1133 
1134 error:
1135 	return err;
1136 }
1137 
1138 static int esp_init_state(struct xfrm_state *x)
1139 {
1140 	struct crypto_aead *aead;
1141 	u32 align;
1142 	int err;
1143 
1144 	x->data = NULL;
1145 
1146 	if (x->aead)
1147 		err = esp_init_aead(x);
1148 	else
1149 		err = esp_init_authenc(x);
1150 
1151 	if (err)
1152 		goto error;
1153 
1154 	aead = x->data;
1155 
1156 	x->props.header_len = sizeof(struct ip_esp_hdr) +
1157 			      crypto_aead_ivsize(aead);
1158 	if (x->props.mode == XFRM_MODE_TUNNEL)
1159 		x->props.header_len += sizeof(struct iphdr);
1160 	else if (x->props.mode == XFRM_MODE_BEET && x->sel.family != AF_INET6)
1161 		x->props.header_len += IPV4_BEET_PHMAXLEN;
1162 	if (x->encap) {
1163 		struct xfrm_encap_tmpl *encap = x->encap;
1164 
1165 		switch (encap->encap_type) {
1166 		default:
1167 			err = -EINVAL;
1168 			goto error;
1169 		case UDP_ENCAP_ESPINUDP:
1170 			x->props.header_len += sizeof(struct udphdr);
1171 			break;
1172 		case UDP_ENCAP_ESPINUDP_NON_IKE:
1173 			x->props.header_len += sizeof(struct udphdr) + 2 * sizeof(u32);
1174 			break;
1175 #ifdef CONFIG_INET_ESPINTCP
1176 		case TCP_ENCAP_ESPINTCP:
1177 			/* only the length field, TCP encap is done by
1178 			 * the socket
1179 			 */
1180 			x->props.header_len += 2;
1181 			break;
1182 #endif
1183 		}
1184 	}
1185 
1186 	align = ALIGN(crypto_aead_blocksize(aead), 4);
1187 	x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1188 
1189 error:
1190 	return err;
1191 }
1192 
1193 static int esp4_rcv_cb(struct sk_buff *skb, int err)
1194 {
1195 	return 0;
1196 }
1197 
1198 static const struct xfrm_type esp_type =
1199 {
1200 	.owner		= THIS_MODULE,
1201 	.proto	     	= IPPROTO_ESP,
1202 	.flags		= XFRM_TYPE_REPLAY_PROT,
1203 	.init_state	= esp_init_state,
1204 	.destructor	= esp_destroy,
1205 	.input		= esp_input,
1206 	.output		= esp_output,
1207 };
1208 
1209 static struct xfrm4_protocol esp4_protocol = {
1210 	.handler	=	xfrm4_rcv,
1211 	.input_handler	=	xfrm_input,
1212 	.cb_handler	=	esp4_rcv_cb,
1213 	.err_handler	=	esp4_err,
1214 	.priority	=	0,
1215 };
1216 
1217 static int __init esp4_init(void)
1218 {
1219 	if (xfrm_register_type(&esp_type, AF_INET) < 0) {
1220 		pr_info("%s: can't add xfrm type\n", __func__);
1221 		return -EAGAIN;
1222 	}
1223 	if (xfrm4_protocol_register(&esp4_protocol, IPPROTO_ESP) < 0) {
1224 		pr_info("%s: can't add protocol\n", __func__);
1225 		xfrm_unregister_type(&esp_type, AF_INET);
1226 		return -EAGAIN;
1227 	}
1228 	return 0;
1229 }
1230 
1231 static void __exit esp4_fini(void)
1232 {
1233 	if (xfrm4_protocol_deregister(&esp4_protocol, IPPROTO_ESP) < 0)
1234 		pr_info("%s: can't remove protocol\n", __func__);
1235 	xfrm_unregister_type(&esp_type, AF_INET);
1236 }
1237 
1238 module_init(esp4_init);
1239 module_exit(esp4_fini);
1240 MODULE_LICENSE("GPL");
1241 MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_ESP);
1242