xref: /openbmc/linux/net/ipv4/esp4.c (revision 93696d8f) 
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, struct sk_buff *skb)
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 			skb_page_unref(skb, sg_page(sg), false);
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 net *net = xs_net(x);
138 	struct esp_tcp_sk *esk;
139 	__be16 sport, dport;
140 	struct sock *nsk;
141 	struct sock *sk;
142 
143 	sk = rcu_dereference(x->encap_sk);
144 	if (sk && sk->sk_state == TCP_ESTABLISHED)
145 		return sk;
146 
147 	spin_lock_bh(&x->lock);
148 	sport = encap->encap_sport;
149 	dport = encap->encap_dport;
150 	nsk = rcu_dereference_protected(x->encap_sk,
151 					lockdep_is_held(&x->lock));
152 	if (sk && sk == nsk) {
153 		esk = kmalloc(sizeof(*esk), GFP_ATOMIC);
154 		if (!esk) {
155 			spin_unlock_bh(&x->lock);
156 			return ERR_PTR(-ENOMEM);
157 		}
158 		RCU_INIT_POINTER(x->encap_sk, NULL);
159 		esk->sk = sk;
160 		call_rcu(&esk->rcu, esp_free_tcp_sk);
161 	}
162 	spin_unlock_bh(&x->lock);
163 
164 	sk = inet_lookup_established(net, net->ipv4.tcp_death_row.hashinfo, x->id.daddr.a4,
165 				     dport, x->props.saddr.a4, sport, 0);
166 	if (!sk)
167 		return ERR_PTR(-ENOENT);
168 
169 	if (!tcp_is_ulp_esp(sk)) {
170 		sock_put(sk);
171 		return ERR_PTR(-EINVAL);
172 	}
173 
174 	spin_lock_bh(&x->lock);
175 	nsk = rcu_dereference_protected(x->encap_sk,
176 					lockdep_is_held(&x->lock));
177 	if (encap->encap_sport != sport ||
178 	    encap->encap_dport != dport) {
179 		sock_put(sk);
180 		sk = nsk ?: ERR_PTR(-EREMCHG);
181 	} else if (sk == nsk) {
182 		sock_put(sk);
183 	} else {
184 		rcu_assign_pointer(x->encap_sk, sk);
185 	}
186 	spin_unlock_bh(&x->lock);
187 
188 	return sk;
189 }
190 
191 static int esp_output_tcp_finish(struct xfrm_state *x, struct sk_buff *skb)
192 {
193 	struct sock *sk;
194 	int err;
195 
196 	rcu_read_lock();
197 
198 	sk = esp_find_tcp_sk(x);
199 	err = PTR_ERR_OR_ZERO(sk);
200 	if (err)
201 		goto out;
202 
203 	bh_lock_sock(sk);
204 	if (sock_owned_by_user(sk))
205 		err = espintcp_queue_out(sk, skb);
206 	else
207 		err = espintcp_push_skb(sk, skb);
208 	bh_unlock_sock(sk);
209 
210 out:
211 	rcu_read_unlock();
212 	return err;
213 }
214 
215 static int esp_output_tcp_encap_cb(struct net *net, struct sock *sk,
216 				   struct sk_buff *skb)
217 {
218 	struct dst_entry *dst = skb_dst(skb);
219 	struct xfrm_state *x = dst->xfrm;
220 
221 	return esp_output_tcp_finish(x, skb);
222 }
223 
224 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
225 {
226 	int err;
227 
228 	local_bh_disable();
229 	err = xfrm_trans_queue_net(xs_net(x), skb, esp_output_tcp_encap_cb);
230 	local_bh_enable();
231 
232 	/* EINPROGRESS just happens to do the right thing.  It
233 	 * actually means that the skb has been consumed and
234 	 * isn't coming back.
235 	 */
236 	return err ?: -EINPROGRESS;
237 }
238 #else
239 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
240 {
241 	WARN_ON(1);
242 	return -EOPNOTSUPP;
243 }
244 #endif
245 
246 static void esp_output_done(void *data, int err)
247 {
248 	struct sk_buff *skb = 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, skb);
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(void *data, int err)
335 {
336 	struct sk_buff *skb = data;
337 
338 	esp_output_restore_header(skb);
339 	esp_output_done(data, 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 (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 		case UDP_ENCAP_ESPINUDP_NON_IKE:
778 			source = uh->source;
779 			break;
780 		default:
781 			WARN_ON_ONCE(1);
782 			err = -EINVAL;
783 			goto out;
784 		}
785 
786 		/*
787 		 * 1) if the NAT-T peer's IP or port changed then
788 		 *    advertise the change to the keying daemon.
789 		 *    This is an inbound SA, so just compare
790 		 *    SRC ports.
791 		 */
792 		if (iph->saddr != x->props.saddr.a4 ||
793 		    source != encap->encap_sport) {
794 			xfrm_address_t ipaddr;
795 
796 			ipaddr.a4 = iph->saddr;
797 			km_new_mapping(x, &ipaddr, source);
798 
799 			/* XXX: perhaps add an extra
800 			 * policy check here, to see
801 			 * if we should allow or
802 			 * reject a packet from a
803 			 * different source
804 			 * address/port.
805 			 */
806 		}
807 
808 		/*
809 		 * 2) ignore UDP/TCP checksums in case
810 		 *    of NAT-T in Transport Mode, or
811 		 *    perform other post-processing fixes
812 		 *    as per draft-ietf-ipsec-udp-encaps-06,
813 		 *    section 3.1.2
814 		 */
815 		if (x->props.mode == XFRM_MODE_TRANSPORT)
816 			skb->ip_summed = CHECKSUM_UNNECESSARY;
817 	}
818 
819 	skb_pull_rcsum(skb, hlen);
820 	if (x->props.mode == XFRM_MODE_TUNNEL)
821 		skb_reset_transport_header(skb);
822 	else
823 		skb_set_transport_header(skb, -ihl);
824 
825 	/* RFC4303: Drop dummy packets without any error */
826 	if (err == IPPROTO_NONE)
827 		err = -EINVAL;
828 
829 out:
830 	return err;
831 }
832 EXPORT_SYMBOL_GPL(esp_input_done2);
833 
834 static void esp_input_done(void *data, int err)
835 {
836 	struct sk_buff *skb = data;
837 
838 	xfrm_input_resume(skb, esp_input_done2(skb, err));
839 }
840 
841 static void esp_input_restore_header(struct sk_buff *skb)
842 {
843 	esp_restore_header(skb, 0);
844 	__skb_pull(skb, 4);
845 }
846 
847 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
848 {
849 	struct xfrm_state *x = xfrm_input_state(skb);
850 	struct ip_esp_hdr *esph;
851 
852 	/* For ESN we move the header forward by 4 bytes to
853 	 * accommodate the high bits.  We will move it back after
854 	 * decryption.
855 	 */
856 	if ((x->props.flags & XFRM_STATE_ESN)) {
857 		esph = skb_push(skb, 4);
858 		*seqhi = esph->spi;
859 		esph->spi = esph->seq_no;
860 		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
861 	}
862 }
863 
864 static void esp_input_done_esn(void *data, int err)
865 {
866 	struct sk_buff *skb = data;
867 
868 	esp_input_restore_header(skb);
869 	esp_input_done(data, err);
870 }
871 
872 /*
873  * Note: detecting truncated vs. non-truncated authentication data is very
874  * expensive, so we only support truncated data, which is the recommended
875  * and common case.
876  */
877 static int esp_input(struct xfrm_state *x, struct sk_buff *skb)
878 {
879 	struct crypto_aead *aead = x->data;
880 	struct aead_request *req;
881 	struct sk_buff *trailer;
882 	int ivlen = crypto_aead_ivsize(aead);
883 	int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
884 	int nfrags;
885 	int assoclen;
886 	int seqhilen;
887 	__be32 *seqhi;
888 	void *tmp;
889 	u8 *iv;
890 	struct scatterlist *sg;
891 	int err = -EINVAL;
892 
893 	if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen))
894 		goto out;
895 
896 	if (elen <= 0)
897 		goto out;
898 
899 	assoclen = sizeof(struct ip_esp_hdr);
900 	seqhilen = 0;
901 
902 	if (x->props.flags & XFRM_STATE_ESN) {
903 		seqhilen += sizeof(__be32);
904 		assoclen += seqhilen;
905 	}
906 
907 	if (!skb_cloned(skb)) {
908 		if (!skb_is_nonlinear(skb)) {
909 			nfrags = 1;
910 
911 			goto skip_cow;
912 		} else if (!skb_has_frag_list(skb)) {
913 			nfrags = skb_shinfo(skb)->nr_frags;
914 			nfrags++;
915 
916 			goto skip_cow;
917 		}
918 	}
919 
920 	err = skb_cow_data(skb, 0, &trailer);
921 	if (err < 0)
922 		goto out;
923 
924 	nfrags = err;
925 
926 skip_cow:
927 	err = -ENOMEM;
928 	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
929 	if (!tmp)
930 		goto out;
931 
932 	ESP_SKB_CB(skb)->tmp = tmp;
933 	seqhi = esp_tmp_extra(tmp);
934 	iv = esp_tmp_iv(aead, tmp, seqhilen);
935 	req = esp_tmp_req(aead, iv);
936 	sg = esp_req_sg(aead, req);
937 
938 	esp_input_set_header(skb, seqhi);
939 
940 	sg_init_table(sg, nfrags);
941 	err = skb_to_sgvec(skb, sg, 0, skb->len);
942 	if (unlikely(err < 0)) {
943 		kfree(tmp);
944 		goto out;
945 	}
946 
947 	skb->ip_summed = CHECKSUM_NONE;
948 
949 	if ((x->props.flags & XFRM_STATE_ESN))
950 		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
951 	else
952 		aead_request_set_callback(req, 0, esp_input_done, skb);
953 
954 	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
955 	aead_request_set_ad(req, assoclen);
956 
957 	err = crypto_aead_decrypt(req);
958 	if (err == -EINPROGRESS)
959 		goto out;
960 
961 	if ((x->props.flags & XFRM_STATE_ESN))
962 		esp_input_restore_header(skb);
963 
964 	err = esp_input_done2(skb, err);
965 
966 out:
967 	return err;
968 }
969 
970 static int esp4_err(struct sk_buff *skb, u32 info)
971 {
972 	struct net *net = dev_net(skb->dev);
973 	const struct iphdr *iph = (const struct iphdr *)skb->data;
974 	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data+(iph->ihl<<2));
975 	struct xfrm_state *x;
976 
977 	switch (icmp_hdr(skb)->type) {
978 	case ICMP_DEST_UNREACH:
979 		if (icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
980 			return 0;
981 		break;
982 	case ICMP_REDIRECT:
983 		break;
984 	default:
985 		return 0;
986 	}
987 
988 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
989 			      esph->spi, IPPROTO_ESP, AF_INET);
990 	if (!x)
991 		return 0;
992 
993 	if (icmp_hdr(skb)->type == ICMP_DEST_UNREACH)
994 		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ESP);
995 	else
996 		ipv4_redirect(skb, net, 0, IPPROTO_ESP);
997 	xfrm_state_put(x);
998 
999 	return 0;
1000 }
1001 
1002 static void esp_destroy(struct xfrm_state *x)
1003 {
1004 	struct crypto_aead *aead = x->data;
1005 
1006 	if (!aead)
1007 		return;
1008 
1009 	crypto_free_aead(aead);
1010 }
1011 
1012 static int esp_init_aead(struct xfrm_state *x, struct netlink_ext_ack *extack)
1013 {
1014 	char aead_name[CRYPTO_MAX_ALG_NAME];
1015 	struct crypto_aead *aead;
1016 	int err;
1017 
1018 	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1019 		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME) {
1020 		NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1021 		return -ENAMETOOLONG;
1022 	}
1023 
1024 	aead = crypto_alloc_aead(aead_name, 0, 0);
1025 	err = PTR_ERR(aead);
1026 	if (IS_ERR(aead))
1027 		goto error;
1028 
1029 	x->data = aead;
1030 
1031 	err = crypto_aead_setkey(aead, x->aead->alg_key,
1032 				 (x->aead->alg_key_len + 7) / 8);
1033 	if (err)
1034 		goto error;
1035 
1036 	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1037 	if (err)
1038 		goto error;
1039 
1040 	return 0;
1041 
1042 error:
1043 	NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1044 	return err;
1045 }
1046 
1047 static int esp_init_authenc(struct xfrm_state *x,
1048 			    struct netlink_ext_ack *extack)
1049 {
1050 	struct crypto_aead *aead;
1051 	struct crypto_authenc_key_param *param;
1052 	struct rtattr *rta;
1053 	char *key;
1054 	char *p;
1055 	char authenc_name[CRYPTO_MAX_ALG_NAME];
1056 	unsigned int keylen;
1057 	int err;
1058 
1059 	err = -ENAMETOOLONG;
1060 
1061 	if ((x->props.flags & XFRM_STATE_ESN)) {
1062 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1063 			     "%s%sauthencesn(%s,%s)%s",
1064 			     x->geniv ?: "", x->geniv ? "(" : "",
1065 			     x->aalg ? x->aalg->alg_name : "digest_null",
1066 			     x->ealg->alg_name,
1067 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1068 			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1069 			goto error;
1070 		}
1071 	} else {
1072 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1073 			     "%s%sauthenc(%s,%s)%s",
1074 			     x->geniv ?: "", x->geniv ? "(" : "",
1075 			     x->aalg ? x->aalg->alg_name : "digest_null",
1076 			     x->ealg->alg_name,
1077 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1078 			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1079 			goto error;
1080 		}
1081 	}
1082 
1083 	aead = crypto_alloc_aead(authenc_name, 0, 0);
1084 	err = PTR_ERR(aead);
1085 	if (IS_ERR(aead)) {
1086 		NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1087 		goto error;
1088 	}
1089 
1090 	x->data = aead;
1091 
1092 	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1093 		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1094 	err = -ENOMEM;
1095 	key = kmalloc(keylen, GFP_KERNEL);
1096 	if (!key)
1097 		goto error;
1098 
1099 	p = key;
1100 	rta = (void *)p;
1101 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1102 	rta->rta_len = RTA_LENGTH(sizeof(*param));
1103 	param = RTA_DATA(rta);
1104 	p += RTA_SPACE(sizeof(*param));
1105 
1106 	if (x->aalg) {
1107 		struct xfrm_algo_desc *aalg_desc;
1108 
1109 		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1110 		p += (x->aalg->alg_key_len + 7) / 8;
1111 
1112 		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1113 		BUG_ON(!aalg_desc);
1114 
1115 		err = -EINVAL;
1116 		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1117 		    crypto_aead_authsize(aead)) {
1118 			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1119 			goto free_key;
1120 		}
1121 
1122 		err = crypto_aead_setauthsize(
1123 			aead, x->aalg->alg_trunc_len / 8);
1124 		if (err) {
1125 			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1126 			goto free_key;
1127 		}
1128 	}
1129 
1130 	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1131 	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1132 
1133 	err = crypto_aead_setkey(aead, key, keylen);
1134 
1135 free_key:
1136 	kfree_sensitive(key);
1137 
1138 error:
1139 	return err;
1140 }
1141 
1142 static int esp_init_state(struct xfrm_state *x, struct netlink_ext_ack *extack)
1143 {
1144 	struct crypto_aead *aead;
1145 	u32 align;
1146 	int err;
1147 
1148 	x->data = NULL;
1149 
1150 	if (x->aead) {
1151 		err = esp_init_aead(x, extack);
1152 	} else if (x->ealg) {
1153 		err = esp_init_authenc(x, extack);
1154 	} else {
1155 		NL_SET_ERR_MSG(extack, "ESP: AEAD or CRYPT must be provided");
1156 		err = -EINVAL;
1157 	}
1158 
1159 	if (err)
1160 		goto error;
1161 
1162 	aead = x->data;
1163 
1164 	x->props.header_len = sizeof(struct ip_esp_hdr) +
1165 			      crypto_aead_ivsize(aead);
1166 	if (x->props.mode == XFRM_MODE_TUNNEL)
1167 		x->props.header_len += sizeof(struct iphdr);
1168 	else if (x->props.mode == XFRM_MODE_BEET && x->sel.family != AF_INET6)
1169 		x->props.header_len += IPV4_BEET_PHMAXLEN;
1170 	if (x->encap) {
1171 		struct xfrm_encap_tmpl *encap = x->encap;
1172 
1173 		switch (encap->encap_type) {
1174 		default:
1175 			NL_SET_ERR_MSG(extack, "Unsupported encapsulation type for ESP");
1176 			err = -EINVAL;
1177 			goto error;
1178 		case UDP_ENCAP_ESPINUDP:
1179 			x->props.header_len += sizeof(struct udphdr);
1180 			break;
1181 		case UDP_ENCAP_ESPINUDP_NON_IKE:
1182 			x->props.header_len += sizeof(struct udphdr) + 2 * sizeof(u32);
1183 			break;
1184 #ifdef CONFIG_INET_ESPINTCP
1185 		case TCP_ENCAP_ESPINTCP:
1186 			/* only the length field, TCP encap is done by
1187 			 * the socket
1188 			 */
1189 			x->props.header_len += 2;
1190 			break;
1191 #endif
1192 		}
1193 	}
1194 
1195 	align = ALIGN(crypto_aead_blocksize(aead), 4);
1196 	x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1197 
1198 error:
1199 	return err;
1200 }
1201 
1202 static int esp4_rcv_cb(struct sk_buff *skb, int err)
1203 {
1204 	return 0;
1205 }
1206 
1207 static const struct xfrm_type esp_type =
1208 {
1209 	.owner		= THIS_MODULE,
1210 	.proto	     	= IPPROTO_ESP,
1211 	.flags		= XFRM_TYPE_REPLAY_PROT,
1212 	.init_state	= esp_init_state,
1213 	.destructor	= esp_destroy,
1214 	.input		= esp_input,
1215 	.output		= esp_output,
1216 };
1217 
1218 static struct xfrm4_protocol esp4_protocol = {
1219 	.handler	=	xfrm4_rcv,
1220 	.input_handler	=	xfrm_input,
1221 	.cb_handler	=	esp4_rcv_cb,
1222 	.err_handler	=	esp4_err,
1223 	.priority	=	0,
1224 };
1225 
1226 static int __init esp4_init(void)
1227 {
1228 	if (xfrm_register_type(&esp_type, AF_INET) < 0) {
1229 		pr_info("%s: can't add xfrm type\n", __func__);
1230 		return -EAGAIN;
1231 	}
1232 	if (xfrm4_protocol_register(&esp4_protocol, IPPROTO_ESP) < 0) {
1233 		pr_info("%s: can't add protocol\n", __func__);
1234 		xfrm_unregister_type(&esp_type, AF_INET);
1235 		return -EAGAIN;
1236 	}
1237 	return 0;
1238 }
1239 
1240 static void __exit esp4_fini(void)
1241 {
1242 	if (xfrm4_protocol_deregister(&esp4_protocol, IPPROTO_ESP) < 0)
1243 		pr_info("%s: can't remove protocol\n", __func__);
1244 	xfrm_unregister_type(&esp_type, AF_INET);
1245 }
1246 
1247 module_init(esp4_init);
1248 module_exit(esp4_fini);
1249 MODULE_LICENSE("GPL");
1250 MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_ESP);
1251