xref: /openbmc/linux/net/ipv4/esp4.c (revision dc6a81c3)
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 void esp_output_fill_trailer(u8 *tail, int tfclen, int plen, __u8 proto)
345 {
346 	/* Fill padding... */
347 	if (tfclen) {
348 		memset(tail, 0, tfclen);
349 		tail += tfclen;
350 	}
351 	do {
352 		int i;
353 		for (i = 0; i < plen - 2; i++)
354 			tail[i] = i + 1;
355 	} while (0);
356 	tail[plen - 2] = plen - 2;
357 	tail[plen - 1] = proto;
358 }
359 
360 static struct ip_esp_hdr *esp_output_udp_encap(struct sk_buff *skb,
361 					       int encap_type,
362 					       struct esp_info *esp,
363 					       __be16 sport,
364 					       __be16 dport)
365 {
366 	struct udphdr *uh;
367 	__be32 *udpdata32;
368 	unsigned int len;
369 
370 	len = skb->len + esp->tailen - skb_transport_offset(skb);
371 	if (len + sizeof(struct iphdr) > IP_MAX_MTU)
372 		return ERR_PTR(-EMSGSIZE);
373 
374 	uh = (struct udphdr *)esp->esph;
375 	uh->source = sport;
376 	uh->dest = dport;
377 	uh->len = htons(len);
378 	uh->check = 0;
379 
380 	*skb_mac_header(skb) = IPPROTO_UDP;
381 
382 	if (encap_type == UDP_ENCAP_ESPINUDP_NON_IKE) {
383 		udpdata32 = (__be32 *)(uh + 1);
384 		udpdata32[0] = udpdata32[1] = 0;
385 		return (struct ip_esp_hdr *)(udpdata32 + 2);
386 	}
387 
388 	return (struct ip_esp_hdr *)(uh + 1);
389 }
390 
391 #ifdef CONFIG_INET_ESPINTCP
392 static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
393 						    struct sk_buff *skb,
394 						    struct esp_info *esp)
395 {
396 	__be16 *lenp = (void *)esp->esph;
397 	struct ip_esp_hdr *esph;
398 	unsigned int len;
399 	struct sock *sk;
400 
401 	len = skb->len + esp->tailen - skb_transport_offset(skb);
402 	if (len > IP_MAX_MTU)
403 		return ERR_PTR(-EMSGSIZE);
404 
405 	rcu_read_lock();
406 	sk = esp_find_tcp_sk(x);
407 	rcu_read_unlock();
408 
409 	if (IS_ERR(sk))
410 		return ERR_CAST(sk);
411 
412 	*lenp = htons(len);
413 	esph = (struct ip_esp_hdr *)(lenp + 1);
414 
415 	return esph;
416 }
417 #else
418 static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
419 						    struct sk_buff *skb,
420 						    struct esp_info *esp)
421 {
422 	return ERR_PTR(-EOPNOTSUPP);
423 }
424 #endif
425 
426 static int esp_output_encap(struct xfrm_state *x, struct sk_buff *skb,
427 			    struct esp_info *esp)
428 {
429 	struct xfrm_encap_tmpl *encap = x->encap;
430 	struct ip_esp_hdr *esph;
431 	__be16 sport, dport;
432 	int encap_type;
433 
434 	spin_lock_bh(&x->lock);
435 	sport = encap->encap_sport;
436 	dport = encap->encap_dport;
437 	encap_type = encap->encap_type;
438 	spin_unlock_bh(&x->lock);
439 
440 	switch (encap_type) {
441 	default:
442 	case UDP_ENCAP_ESPINUDP:
443 	case UDP_ENCAP_ESPINUDP_NON_IKE:
444 		esph = esp_output_udp_encap(skb, encap_type, esp, sport, dport);
445 		break;
446 	case TCP_ENCAP_ESPINTCP:
447 		esph = esp_output_tcp_encap(x, skb, esp);
448 		break;
449 	}
450 
451 	if (IS_ERR(esph))
452 		return PTR_ERR(esph);
453 
454 	esp->esph = esph;
455 
456 	return 0;
457 }
458 
459 int esp_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
460 {
461 	u8 *tail;
462 	u8 *vaddr;
463 	int nfrags;
464 	int esph_offset;
465 	struct page *page;
466 	struct sk_buff *trailer;
467 	int tailen = esp->tailen;
468 
469 	/* this is non-NULL only with TCP/UDP Encapsulation */
470 	if (x->encap) {
471 		int err = esp_output_encap(x, skb, esp);
472 
473 		if (err < 0)
474 			return err;
475 	}
476 
477 	if (!skb_cloned(skb)) {
478 		if (tailen <= skb_tailroom(skb)) {
479 			nfrags = 1;
480 			trailer = skb;
481 			tail = skb_tail_pointer(trailer);
482 
483 			goto skip_cow;
484 		} else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
485 			   && !skb_has_frag_list(skb)) {
486 			int allocsize;
487 			struct sock *sk = skb->sk;
488 			struct page_frag *pfrag = &x->xfrag;
489 
490 			esp->inplace = false;
491 
492 			allocsize = ALIGN(tailen, L1_CACHE_BYTES);
493 
494 			spin_lock_bh(&x->lock);
495 
496 			if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
497 				spin_unlock_bh(&x->lock);
498 				goto cow;
499 			}
500 
501 			page = pfrag->page;
502 			get_page(page);
503 
504 			vaddr = kmap_atomic(page);
505 
506 			tail = vaddr + pfrag->offset;
507 
508 			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
509 
510 			kunmap_atomic(vaddr);
511 
512 			nfrags = skb_shinfo(skb)->nr_frags;
513 
514 			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
515 					     tailen);
516 			skb_shinfo(skb)->nr_frags = ++nfrags;
517 
518 			pfrag->offset = pfrag->offset + allocsize;
519 
520 			spin_unlock_bh(&x->lock);
521 
522 			nfrags++;
523 
524 			skb->len += tailen;
525 			skb->data_len += tailen;
526 			skb->truesize += tailen;
527 			if (sk && sk_fullsock(sk))
528 				refcount_add(tailen, &sk->sk_wmem_alloc);
529 
530 			goto out;
531 		}
532 	}
533 
534 cow:
535 	esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
536 
537 	nfrags = skb_cow_data(skb, tailen, &trailer);
538 	if (nfrags < 0)
539 		goto out;
540 	tail = skb_tail_pointer(trailer);
541 	esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
542 
543 skip_cow:
544 	esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
545 	pskb_put(skb, trailer, tailen);
546 
547 out:
548 	return nfrags;
549 }
550 EXPORT_SYMBOL_GPL(esp_output_head);
551 
552 int esp_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
553 {
554 	u8 *iv;
555 	int alen;
556 	void *tmp;
557 	int ivlen;
558 	int assoclen;
559 	int extralen;
560 	struct page *page;
561 	struct ip_esp_hdr *esph;
562 	struct crypto_aead *aead;
563 	struct aead_request *req;
564 	struct scatterlist *sg, *dsg;
565 	struct esp_output_extra *extra;
566 	int err = -ENOMEM;
567 
568 	assoclen = sizeof(struct ip_esp_hdr);
569 	extralen = 0;
570 
571 	if (x->props.flags & XFRM_STATE_ESN) {
572 		extralen += sizeof(*extra);
573 		assoclen += sizeof(__be32);
574 	}
575 
576 	aead = x->data;
577 	alen = crypto_aead_authsize(aead);
578 	ivlen = crypto_aead_ivsize(aead);
579 
580 	tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
581 	if (!tmp)
582 		goto error;
583 
584 	extra = esp_tmp_extra(tmp);
585 	iv = esp_tmp_iv(aead, tmp, extralen);
586 	req = esp_tmp_req(aead, iv);
587 	sg = esp_req_sg(aead, req);
588 
589 	if (esp->inplace)
590 		dsg = sg;
591 	else
592 		dsg = &sg[esp->nfrags];
593 
594 	esph = esp_output_set_extra(skb, x, esp->esph, extra);
595 	esp->esph = esph;
596 
597 	sg_init_table(sg, esp->nfrags);
598 	err = skb_to_sgvec(skb, sg,
599 		           (unsigned char *)esph - skb->data,
600 		           assoclen + ivlen + esp->clen + alen);
601 	if (unlikely(err < 0))
602 		goto error_free;
603 
604 	if (!esp->inplace) {
605 		int allocsize;
606 		struct page_frag *pfrag = &x->xfrag;
607 
608 		allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
609 
610 		spin_lock_bh(&x->lock);
611 		if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
612 			spin_unlock_bh(&x->lock);
613 			goto error_free;
614 		}
615 
616 		skb_shinfo(skb)->nr_frags = 1;
617 
618 		page = pfrag->page;
619 		get_page(page);
620 		/* replace page frags in skb with new page */
621 		__skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
622 		pfrag->offset = pfrag->offset + allocsize;
623 		spin_unlock_bh(&x->lock);
624 
625 		sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
626 		err = skb_to_sgvec(skb, dsg,
627 			           (unsigned char *)esph - skb->data,
628 			           assoclen + ivlen + esp->clen + alen);
629 		if (unlikely(err < 0))
630 			goto error_free;
631 	}
632 
633 	if ((x->props.flags & XFRM_STATE_ESN))
634 		aead_request_set_callback(req, 0, esp_output_done_esn, skb);
635 	else
636 		aead_request_set_callback(req, 0, esp_output_done, skb);
637 
638 	aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
639 	aead_request_set_ad(req, assoclen);
640 
641 	memset(iv, 0, ivlen);
642 	memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
643 	       min(ivlen, 8));
644 
645 	ESP_SKB_CB(skb)->tmp = tmp;
646 	err = crypto_aead_encrypt(req);
647 
648 	switch (err) {
649 	case -EINPROGRESS:
650 		goto error;
651 
652 	case -ENOSPC:
653 		err = NET_XMIT_DROP;
654 		break;
655 
656 	case 0:
657 		if ((x->props.flags & XFRM_STATE_ESN))
658 			esp_output_restore_header(skb);
659 	}
660 
661 	if (sg != dsg)
662 		esp_ssg_unref(x, tmp);
663 
664 	if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
665 		err = esp_output_tail_tcp(x, skb);
666 
667 error_free:
668 	kfree(tmp);
669 error:
670 	return err;
671 }
672 EXPORT_SYMBOL_GPL(esp_output_tail);
673 
674 static int esp_output(struct xfrm_state *x, struct sk_buff *skb)
675 {
676 	int alen;
677 	int blksize;
678 	struct ip_esp_hdr *esph;
679 	struct crypto_aead *aead;
680 	struct esp_info esp;
681 
682 	esp.inplace = true;
683 
684 	esp.proto = *skb_mac_header(skb);
685 	*skb_mac_header(skb) = IPPROTO_ESP;
686 
687 	/* skb is pure payload to encrypt */
688 
689 	aead = x->data;
690 	alen = crypto_aead_authsize(aead);
691 
692 	esp.tfclen = 0;
693 	if (x->tfcpad) {
694 		struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
695 		u32 padto;
696 
697 		padto = min(x->tfcpad, xfrm_state_mtu(x, dst->child_mtu_cached));
698 		if (skb->len < padto)
699 			esp.tfclen = padto - skb->len;
700 	}
701 	blksize = ALIGN(crypto_aead_blocksize(aead), 4);
702 	esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
703 	esp.plen = esp.clen - skb->len - esp.tfclen;
704 	esp.tailen = esp.tfclen + esp.plen + alen;
705 
706 	esp.esph = ip_esp_hdr(skb);
707 
708 	esp.nfrags = esp_output_head(x, skb, &esp);
709 	if (esp.nfrags < 0)
710 		return esp.nfrags;
711 
712 	esph = esp.esph;
713 	esph->spi = x->id.spi;
714 
715 	esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
716 	esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
717 				 ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
718 
719 	skb_push(skb, -skb_network_offset(skb));
720 
721 	return esp_output_tail(x, skb, &esp);
722 }
723 
724 static inline int esp_remove_trailer(struct sk_buff *skb)
725 {
726 	struct xfrm_state *x = xfrm_input_state(skb);
727 	struct xfrm_offload *xo = xfrm_offload(skb);
728 	struct crypto_aead *aead = x->data;
729 	int alen, hlen, elen;
730 	int padlen, trimlen;
731 	__wsum csumdiff;
732 	u8 nexthdr[2];
733 	int ret;
734 
735 	alen = crypto_aead_authsize(aead);
736 	hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
737 	elen = skb->len - hlen;
738 
739 	if (xo && (xo->flags & XFRM_ESP_NO_TRAILER)) {
740 		ret = xo->proto;
741 		goto out;
742 	}
743 
744 	if (skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2))
745 		BUG();
746 
747 	ret = -EINVAL;
748 	padlen = nexthdr[0];
749 	if (padlen + 2 + alen >= elen) {
750 		net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
751 				    padlen + 2, elen - alen);
752 		goto out;
753 	}
754 
755 	trimlen = alen + padlen + 2;
756 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
757 		csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
758 		skb->csum = csum_block_sub(skb->csum, csumdiff,
759 					   skb->len - trimlen);
760 	}
761 	pskb_trim(skb, skb->len - trimlen);
762 
763 	ret = nexthdr[1];
764 
765 out:
766 	return ret;
767 }
768 
769 int esp_input_done2(struct sk_buff *skb, int err)
770 {
771 	const struct iphdr *iph;
772 	struct xfrm_state *x = xfrm_input_state(skb);
773 	struct xfrm_offload *xo = xfrm_offload(skb);
774 	struct crypto_aead *aead = x->data;
775 	int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
776 	int ihl;
777 
778 	if (!xo || (xo && !(xo->flags & CRYPTO_DONE)))
779 		kfree(ESP_SKB_CB(skb)->tmp);
780 
781 	if (unlikely(err))
782 		goto out;
783 
784 	err = esp_remove_trailer(skb);
785 	if (unlikely(err < 0))
786 		goto out;
787 
788 	iph = ip_hdr(skb);
789 	ihl = iph->ihl * 4;
790 
791 	if (x->encap) {
792 		struct xfrm_encap_tmpl *encap = x->encap;
793 		struct tcphdr *th = (void *)(skb_network_header(skb) + ihl);
794 		struct udphdr *uh = (void *)(skb_network_header(skb) + ihl);
795 		__be16 source;
796 
797 		switch (x->encap->encap_type) {
798 		case TCP_ENCAP_ESPINTCP:
799 			source = th->source;
800 			break;
801 		case UDP_ENCAP_ESPINUDP:
802 		case UDP_ENCAP_ESPINUDP_NON_IKE:
803 			source = uh->source;
804 			break;
805 		default:
806 			WARN_ON_ONCE(1);
807 			err = -EINVAL;
808 			goto out;
809 		}
810 
811 		/*
812 		 * 1) if the NAT-T peer's IP or port changed then
813 		 *    advertize the change to the keying daemon.
814 		 *    This is an inbound SA, so just compare
815 		 *    SRC ports.
816 		 */
817 		if (iph->saddr != x->props.saddr.a4 ||
818 		    source != encap->encap_sport) {
819 			xfrm_address_t ipaddr;
820 
821 			ipaddr.a4 = iph->saddr;
822 			km_new_mapping(x, &ipaddr, source);
823 
824 			/* XXX: perhaps add an extra
825 			 * policy check here, to see
826 			 * if we should allow or
827 			 * reject a packet from a
828 			 * different source
829 			 * address/port.
830 			 */
831 		}
832 
833 		/*
834 		 * 2) ignore UDP/TCP checksums in case
835 		 *    of NAT-T in Transport Mode, or
836 		 *    perform other post-processing fixes
837 		 *    as per draft-ietf-ipsec-udp-encaps-06,
838 		 *    section 3.1.2
839 		 */
840 		if (x->props.mode == XFRM_MODE_TRANSPORT)
841 			skb->ip_summed = CHECKSUM_UNNECESSARY;
842 	}
843 
844 	skb_pull_rcsum(skb, hlen);
845 	if (x->props.mode == XFRM_MODE_TUNNEL)
846 		skb_reset_transport_header(skb);
847 	else
848 		skb_set_transport_header(skb, -ihl);
849 
850 	/* RFC4303: Drop dummy packets without any error */
851 	if (err == IPPROTO_NONE)
852 		err = -EINVAL;
853 
854 out:
855 	return err;
856 }
857 EXPORT_SYMBOL_GPL(esp_input_done2);
858 
859 static void esp_input_done(struct crypto_async_request *base, int err)
860 {
861 	struct sk_buff *skb = base->data;
862 
863 	xfrm_input_resume(skb, esp_input_done2(skb, err));
864 }
865 
866 static void esp_input_restore_header(struct sk_buff *skb)
867 {
868 	esp_restore_header(skb, 0);
869 	__skb_pull(skb, 4);
870 }
871 
872 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
873 {
874 	struct xfrm_state *x = xfrm_input_state(skb);
875 	struct ip_esp_hdr *esph;
876 
877 	/* For ESN we move the header forward by 4 bytes to
878 	 * accomodate the high bits.  We will move it back after
879 	 * decryption.
880 	 */
881 	if ((x->props.flags & XFRM_STATE_ESN)) {
882 		esph = skb_push(skb, 4);
883 		*seqhi = esph->spi;
884 		esph->spi = esph->seq_no;
885 		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
886 	}
887 }
888 
889 static void esp_input_done_esn(struct crypto_async_request *base, int err)
890 {
891 	struct sk_buff *skb = base->data;
892 
893 	esp_input_restore_header(skb);
894 	esp_input_done(base, err);
895 }
896 
897 /*
898  * Note: detecting truncated vs. non-truncated authentication data is very
899  * expensive, so we only support truncated data, which is the recommended
900  * and common case.
901  */
902 static int esp_input(struct xfrm_state *x, struct sk_buff *skb)
903 {
904 	struct crypto_aead *aead = x->data;
905 	struct aead_request *req;
906 	struct sk_buff *trailer;
907 	int ivlen = crypto_aead_ivsize(aead);
908 	int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
909 	int nfrags;
910 	int assoclen;
911 	int seqhilen;
912 	__be32 *seqhi;
913 	void *tmp;
914 	u8 *iv;
915 	struct scatterlist *sg;
916 	int err = -EINVAL;
917 
918 	if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen))
919 		goto out;
920 
921 	if (elen <= 0)
922 		goto out;
923 
924 	assoclen = sizeof(struct ip_esp_hdr);
925 	seqhilen = 0;
926 
927 	if (x->props.flags & XFRM_STATE_ESN) {
928 		seqhilen += sizeof(__be32);
929 		assoclen += seqhilen;
930 	}
931 
932 	if (!skb_cloned(skb)) {
933 		if (!skb_is_nonlinear(skb)) {
934 			nfrags = 1;
935 
936 			goto skip_cow;
937 		} else if (!skb_has_frag_list(skb)) {
938 			nfrags = skb_shinfo(skb)->nr_frags;
939 			nfrags++;
940 
941 			goto skip_cow;
942 		}
943 	}
944 
945 	err = skb_cow_data(skb, 0, &trailer);
946 	if (err < 0)
947 		goto out;
948 
949 	nfrags = err;
950 
951 skip_cow:
952 	err = -ENOMEM;
953 	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
954 	if (!tmp)
955 		goto out;
956 
957 	ESP_SKB_CB(skb)->tmp = tmp;
958 	seqhi = esp_tmp_extra(tmp);
959 	iv = esp_tmp_iv(aead, tmp, seqhilen);
960 	req = esp_tmp_req(aead, iv);
961 	sg = esp_req_sg(aead, req);
962 
963 	esp_input_set_header(skb, seqhi);
964 
965 	sg_init_table(sg, nfrags);
966 	err = skb_to_sgvec(skb, sg, 0, skb->len);
967 	if (unlikely(err < 0)) {
968 		kfree(tmp);
969 		goto out;
970 	}
971 
972 	skb->ip_summed = CHECKSUM_NONE;
973 
974 	if ((x->props.flags & XFRM_STATE_ESN))
975 		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
976 	else
977 		aead_request_set_callback(req, 0, esp_input_done, skb);
978 
979 	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
980 	aead_request_set_ad(req, assoclen);
981 
982 	err = crypto_aead_decrypt(req);
983 	if (err == -EINPROGRESS)
984 		goto out;
985 
986 	if ((x->props.flags & XFRM_STATE_ESN))
987 		esp_input_restore_header(skb);
988 
989 	err = esp_input_done2(skb, err);
990 
991 out:
992 	return err;
993 }
994 
995 static int esp4_err(struct sk_buff *skb, u32 info)
996 {
997 	struct net *net = dev_net(skb->dev);
998 	const struct iphdr *iph = (const struct iphdr *)skb->data;
999 	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data+(iph->ihl<<2));
1000 	struct xfrm_state *x;
1001 
1002 	switch (icmp_hdr(skb)->type) {
1003 	case ICMP_DEST_UNREACH:
1004 		if (icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
1005 			return 0;
1006 	case ICMP_REDIRECT:
1007 		break;
1008 	default:
1009 		return 0;
1010 	}
1011 
1012 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
1013 			      esph->spi, IPPROTO_ESP, AF_INET);
1014 	if (!x)
1015 		return 0;
1016 
1017 	if (icmp_hdr(skb)->type == ICMP_DEST_UNREACH)
1018 		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ESP);
1019 	else
1020 		ipv4_redirect(skb, net, 0, IPPROTO_ESP);
1021 	xfrm_state_put(x);
1022 
1023 	return 0;
1024 }
1025 
1026 static void esp_destroy(struct xfrm_state *x)
1027 {
1028 	struct crypto_aead *aead = x->data;
1029 
1030 	if (!aead)
1031 		return;
1032 
1033 	crypto_free_aead(aead);
1034 }
1035 
1036 static int esp_init_aead(struct xfrm_state *x)
1037 {
1038 	char aead_name[CRYPTO_MAX_ALG_NAME];
1039 	struct crypto_aead *aead;
1040 	int err;
1041 
1042 	err = -ENAMETOOLONG;
1043 	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1044 		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME)
1045 		goto error;
1046 
1047 	aead = crypto_alloc_aead(aead_name, 0, 0);
1048 	err = PTR_ERR(aead);
1049 	if (IS_ERR(aead))
1050 		goto error;
1051 
1052 	x->data = aead;
1053 
1054 	err = crypto_aead_setkey(aead, x->aead->alg_key,
1055 				 (x->aead->alg_key_len + 7) / 8);
1056 	if (err)
1057 		goto error;
1058 
1059 	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1060 	if (err)
1061 		goto error;
1062 
1063 error:
1064 	return err;
1065 }
1066 
1067 static int esp_init_authenc(struct xfrm_state *x)
1068 {
1069 	struct crypto_aead *aead;
1070 	struct crypto_authenc_key_param *param;
1071 	struct rtattr *rta;
1072 	char *key;
1073 	char *p;
1074 	char authenc_name[CRYPTO_MAX_ALG_NAME];
1075 	unsigned int keylen;
1076 	int err;
1077 
1078 	err = -EINVAL;
1079 	if (!x->ealg)
1080 		goto error;
1081 
1082 	err = -ENAMETOOLONG;
1083 
1084 	if ((x->props.flags & XFRM_STATE_ESN)) {
1085 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1086 			     "%s%sauthencesn(%s,%s)%s",
1087 			     x->geniv ?: "", x->geniv ? "(" : "",
1088 			     x->aalg ? x->aalg->alg_name : "digest_null",
1089 			     x->ealg->alg_name,
1090 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1091 			goto error;
1092 	} else {
1093 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1094 			     "%s%sauthenc(%s,%s)%s",
1095 			     x->geniv ?: "", x->geniv ? "(" : "",
1096 			     x->aalg ? x->aalg->alg_name : "digest_null",
1097 			     x->ealg->alg_name,
1098 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1099 			goto error;
1100 	}
1101 
1102 	aead = crypto_alloc_aead(authenc_name, 0, 0);
1103 	err = PTR_ERR(aead);
1104 	if (IS_ERR(aead))
1105 		goto error;
1106 
1107 	x->data = aead;
1108 
1109 	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1110 		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1111 	err = -ENOMEM;
1112 	key = kmalloc(keylen, GFP_KERNEL);
1113 	if (!key)
1114 		goto error;
1115 
1116 	p = key;
1117 	rta = (void *)p;
1118 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1119 	rta->rta_len = RTA_LENGTH(sizeof(*param));
1120 	param = RTA_DATA(rta);
1121 	p += RTA_SPACE(sizeof(*param));
1122 
1123 	if (x->aalg) {
1124 		struct xfrm_algo_desc *aalg_desc;
1125 
1126 		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1127 		p += (x->aalg->alg_key_len + 7) / 8;
1128 
1129 		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1130 		BUG_ON(!aalg_desc);
1131 
1132 		err = -EINVAL;
1133 		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1134 		    crypto_aead_authsize(aead)) {
1135 			pr_info("ESP: %s digestsize %u != %hu\n",
1136 				x->aalg->alg_name,
1137 				crypto_aead_authsize(aead),
1138 				aalg_desc->uinfo.auth.icv_fullbits / 8);
1139 			goto free_key;
1140 		}
1141 
1142 		err = crypto_aead_setauthsize(
1143 			aead, x->aalg->alg_trunc_len / 8);
1144 		if (err)
1145 			goto free_key;
1146 	}
1147 
1148 	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1149 	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1150 
1151 	err = crypto_aead_setkey(aead, key, keylen);
1152 
1153 free_key:
1154 	kfree(key);
1155 
1156 error:
1157 	return err;
1158 }
1159 
1160 static int esp_init_state(struct xfrm_state *x)
1161 {
1162 	struct crypto_aead *aead;
1163 	u32 align;
1164 	int err;
1165 
1166 	x->data = NULL;
1167 
1168 	if (x->aead)
1169 		err = esp_init_aead(x);
1170 	else
1171 		err = esp_init_authenc(x);
1172 
1173 	if (err)
1174 		goto error;
1175 
1176 	aead = x->data;
1177 
1178 	x->props.header_len = sizeof(struct ip_esp_hdr) +
1179 			      crypto_aead_ivsize(aead);
1180 	if (x->props.mode == XFRM_MODE_TUNNEL)
1181 		x->props.header_len += sizeof(struct iphdr);
1182 	else if (x->props.mode == XFRM_MODE_BEET && x->sel.family != AF_INET6)
1183 		x->props.header_len += IPV4_BEET_PHMAXLEN;
1184 	if (x->encap) {
1185 		struct xfrm_encap_tmpl *encap = x->encap;
1186 
1187 		switch (encap->encap_type) {
1188 		default:
1189 			err = -EINVAL;
1190 			goto error;
1191 		case UDP_ENCAP_ESPINUDP:
1192 			x->props.header_len += sizeof(struct udphdr);
1193 			break;
1194 		case UDP_ENCAP_ESPINUDP_NON_IKE:
1195 			x->props.header_len += sizeof(struct udphdr) + 2 * sizeof(u32);
1196 			break;
1197 #ifdef CONFIG_INET_ESPINTCP
1198 		case TCP_ENCAP_ESPINTCP:
1199 			/* only the length field, TCP encap is done by
1200 			 * the socket
1201 			 */
1202 			x->props.header_len += 2;
1203 			break;
1204 #endif
1205 		}
1206 	}
1207 
1208 	align = ALIGN(crypto_aead_blocksize(aead), 4);
1209 	x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1210 
1211 error:
1212 	return err;
1213 }
1214 
1215 static int esp4_rcv_cb(struct sk_buff *skb, int err)
1216 {
1217 	return 0;
1218 }
1219 
1220 static const struct xfrm_type esp_type =
1221 {
1222 	.description	= "ESP4",
1223 	.owner		= THIS_MODULE,
1224 	.proto	     	= IPPROTO_ESP,
1225 	.flags		= XFRM_TYPE_REPLAY_PROT,
1226 	.init_state	= esp_init_state,
1227 	.destructor	= esp_destroy,
1228 	.input		= esp_input,
1229 	.output		= esp_output,
1230 };
1231 
1232 static struct xfrm4_protocol esp4_protocol = {
1233 	.handler	=	xfrm4_rcv,
1234 	.input_handler	=	xfrm_input,
1235 	.cb_handler	=	esp4_rcv_cb,
1236 	.err_handler	=	esp4_err,
1237 	.priority	=	0,
1238 };
1239 
1240 static int __init esp4_init(void)
1241 {
1242 	if (xfrm_register_type(&esp_type, AF_INET) < 0) {
1243 		pr_info("%s: can't add xfrm type\n", __func__);
1244 		return -EAGAIN;
1245 	}
1246 	if (xfrm4_protocol_register(&esp4_protocol, IPPROTO_ESP) < 0) {
1247 		pr_info("%s: can't add protocol\n", __func__);
1248 		xfrm_unregister_type(&esp_type, AF_INET);
1249 		return -EAGAIN;
1250 	}
1251 	return 0;
1252 }
1253 
1254 static void __exit esp4_fini(void)
1255 {
1256 	if (xfrm4_protocol_deregister(&esp4_protocol, IPPROTO_ESP) < 0)
1257 		pr_info("%s: can't remove protocol\n", __func__);
1258 	xfrm_unregister_type(&esp_type, AF_INET);
1259 }
1260 
1261 module_init(esp4_init);
1262 module_exit(esp4_fini);
1263 MODULE_LICENSE("GPL");
1264 MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_ESP);
1265