xref: /linux/net/ipv4/esp4.c (revision 26aad08a928901296aabfbc7a33ecb951656bb98)
1 // SPDX-License-Identifier: GPL-2.0-only
2 #define pr_fmt(fmt) "IPsec: " fmt
3 
4 #include <crypto/aead.h>
5 #include <crypto/authenc.h>
6 #include <linux/err.h>
7 #include <linux/module.h>
8 #include <net/ip.h>
9 #include <net/xfrm.h>
10 #include <net/esp.h>
11 #include <linux/scatterlist.h>
12 #include <linux/kernel.h>
13 #include <linux/pfkeyv2.h>
14 #include <linux/rtnetlink.h>
15 #include <linux/slab.h>
16 #include <linux/spinlock.h>
17 #include <linux/in6.h>
18 #include <net/icmp.h>
19 #include <net/protocol.h>
20 #include <net/udp.h>
21 #include <net/tcp.h>
22 #include <net/espintcp.h>
23 #include <linux/skbuff_ref.h>
24 
25 #include <linux/highmem.h>
26 
27 struct esp_skb_cb {
28 	struct xfrm_skb_cb xfrm;
29 	void *tmp;
30 };
31 
32 struct esp_output_extra {
33 	__be32 seqhi;
34 	u32 esphoff;
35 };
36 
37 #define ESP_SKB_CB(__skb) ((struct esp_skb_cb *)&((__skb)->cb[0]))
38 
39 /*
40  * Allocate an AEAD request structure with extra space for SG and IV.
41  *
42  * For alignment considerations the IV is placed at the front, followed
43  * by the request and finally the SG list.
44  *
45  * TODO: Use spare space in skb for this where possible.
46  */
47 static void *esp_alloc_tmp(struct crypto_aead *aead, int nfrags, int extralen)
48 {
49 	unsigned int len;
50 
51 	len = extralen;
52 
53 	len += crypto_aead_ivsize(aead);
54 
55 	if (len) {
56 		len += crypto_aead_alignmask(aead) &
57 		       ~(crypto_tfm_ctx_alignment() - 1);
58 		len = ALIGN(len, crypto_tfm_ctx_alignment());
59 	}
60 
61 	len += sizeof(struct aead_request) + crypto_aead_reqsize(aead);
62 	len = ALIGN(len, __alignof__(struct scatterlist));
63 
64 	len += sizeof(struct scatterlist) * nfrags;
65 
66 	return kmalloc(len, GFP_ATOMIC);
67 }
68 
69 static inline void *esp_tmp_extra(void *tmp)
70 {
71 	return PTR_ALIGN(tmp, __alignof__(struct esp_output_extra));
72 }
73 
74 static inline u8 *esp_tmp_iv(struct crypto_aead *aead, void *tmp, int extralen)
75 {
76 	return crypto_aead_ivsize(aead) ?
77 	       PTR_ALIGN((u8 *)tmp + extralen,
78 			 crypto_aead_alignmask(aead) + 1) : tmp + extralen;
79 }
80 
81 static inline struct aead_request *esp_tmp_req(struct crypto_aead *aead, u8 *iv)
82 {
83 	struct aead_request *req;
84 
85 	req = (void *)PTR_ALIGN(iv + crypto_aead_ivsize(aead),
86 				crypto_tfm_ctx_alignment());
87 	aead_request_set_tfm(req, aead);
88 	return req;
89 }
90 
91 static inline struct scatterlist *esp_req_sg(struct crypto_aead *aead,
92 					     struct aead_request *req)
93 {
94 	return (void *)ALIGN((unsigned long)(req + 1) +
95 			     crypto_aead_reqsize(aead),
96 			     __alignof__(struct scatterlist));
97 }
98 
99 static void esp_ssg_unref(struct xfrm_state *x, void *tmp, struct sk_buff *skb, bool already_unref)
100 {
101 	struct crypto_aead *aead = x->data;
102 	int extralen = 0;
103 	u8 *iv;
104 	struct aead_request *req;
105 	struct scatterlist *sg;
106 
107 	if (x->props.flags & XFRM_STATE_ESN)
108 		extralen += sizeof(struct esp_output_extra);
109 
110 	iv = esp_tmp_iv(aead, tmp, extralen);
111 	req = esp_tmp_req(aead, iv);
112 
113 	/* Unref skb_frag_pages in the src scatterlist if necessary.
114 	 * Skip the first sg which comes from skb->data.
115 	 */
116 	if (already_unref || req->src != req->dst) {
117 		struct scatterlist *src = already_unref ? esp_req_sg(aead, req) : req->src;
118 
119 		for (sg = sg_next(src); sg; sg = sg_next(sg))
120 			skb_page_unref(page_to_netmem(sg_page(sg)),
121 				       skb->pp_recycle);
122 	}
123 }
124 
125 #ifdef CONFIG_INET_ESPINTCP
126 static struct sock *esp_find_tcp_sk(struct xfrm_state *x)
127 {
128 	struct xfrm_encap_tmpl *encap = x->encap;
129 	struct net *net = xs_net(x);
130 	__be16 sport, dport;
131 	struct sock *sk;
132 
133 	spin_lock_bh(&x->lock);
134 	sport = encap->encap_sport;
135 	dport = encap->encap_dport;
136 	spin_unlock_bh(&x->lock);
137 
138 	sk = inet_lookup_established(net, x->id.daddr.a4, dport,
139 				     x->props.saddr.a4, sport, 0);
140 	if (!sk)
141 		return ERR_PTR(-ENOENT);
142 
143 	if (!tcp_is_ulp_esp(sk)) {
144 		sock_put(sk);
145 		return ERR_PTR(-EINVAL);
146 	}
147 
148 	return sk;
149 }
150 
151 static int esp_output_tcp_finish(struct xfrm_state *x, struct sk_buff *skb)
152 {
153 	struct sock *sk;
154 	int err;
155 
156 	rcu_read_lock();
157 
158 	sk = esp_find_tcp_sk(x);
159 	err = PTR_ERR_OR_ZERO(sk);
160 	if (err) {
161 		kfree_skb(skb);
162 		goto out;
163 	}
164 
165 	bh_lock_sock(sk);
166 	if (sock_owned_by_user(sk))
167 		err = espintcp_queue_out(sk, skb);
168 	else
169 		err = espintcp_push_skb(sk, skb);
170 	bh_unlock_sock(sk);
171 
172 	sock_put(sk);
173 
174 out:
175 	rcu_read_unlock();
176 	return err;
177 }
178 
179 static int esp_output_tcp_encap_cb(struct net *net, struct sock *sk,
180 				   struct sk_buff *skb)
181 {
182 	struct dst_entry *dst = skb_dst(skb);
183 	struct xfrm_state *x = dst->xfrm;
184 
185 	return esp_output_tcp_finish(x, skb);
186 }
187 
188 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
189 {
190 	int err;
191 
192 	local_bh_disable();
193 	err = xfrm_trans_queue_net(xs_net(x), skb, esp_output_tcp_encap_cb);
194 	local_bh_enable();
195 
196 	/* EINPROGRESS just happens to do the right thing.  It
197 	 * actually means that the skb has been consumed and
198 	 * isn't coming back.
199 	 */
200 	return err ?: -EINPROGRESS;
201 }
202 #else
203 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
204 {
205 	WARN_ON(1);
206 	return -EOPNOTSUPP;
207 }
208 #endif
209 
210 static void esp_output_done(void *data, int err)
211 {
212 	struct sk_buff *skb = data;
213 	struct xfrm_offload *xo = xfrm_offload(skb);
214 	void *tmp;
215 	struct xfrm_state *x;
216 
217 	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
218 		struct sec_path *sp = skb_sec_path(skb);
219 
220 		x = sp->xvec[sp->len - 1];
221 	} else {
222 		x = skb_dst(skb)->xfrm;
223 	}
224 
225 	tmp = ESP_SKB_CB(skb)->tmp;
226 	esp_ssg_unref(x, tmp, skb, false);
227 	kfree(tmp);
228 
229 	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
230 		if (err) {
231 			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
232 			kfree_skb(skb);
233 			return;
234 		}
235 
236 		skb_push(skb, skb->data - skb_mac_header(skb));
237 		secpath_reset(skb);
238 		xfrm_dev_resume(skb);
239 	} else {
240 		if (!err &&
241 		    x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP) {
242 			err = esp_output_tail_tcp(x, skb);
243 			if (err != -EINPROGRESS)
244 				kfree_skb(skb);
245 		} else {
246 			xfrm_output_resume(skb_to_full_sk(skb), skb, err);
247 		}
248 	}
249 }
250 
251 /* Move ESP header back into place. */
252 static void esp_restore_header(struct sk_buff *skb, unsigned int offset)
253 {
254 	struct ip_esp_hdr *esph = (void *)(skb->data + offset);
255 	void *tmp = ESP_SKB_CB(skb)->tmp;
256 	__be32 *seqhi = esp_tmp_extra(tmp);
257 
258 	esph->seq_no = esph->spi;
259 	esph->spi = *seqhi;
260 }
261 
262 static void esp_output_restore_header(struct sk_buff *skb)
263 {
264 	void *tmp = ESP_SKB_CB(skb)->tmp;
265 	struct esp_output_extra *extra = esp_tmp_extra(tmp);
266 
267 	esp_restore_header(skb, skb_transport_offset(skb) + extra->esphoff -
268 				sizeof(__be32));
269 }
270 
271 static struct ip_esp_hdr *esp_output_set_extra(struct sk_buff *skb,
272 					       struct xfrm_state *x,
273 					       struct ip_esp_hdr *esph,
274 					       struct esp_output_extra *extra)
275 {
276 	/* For ESN we move the header forward by 4 bytes to
277 	 * accommodate the high bits.  We will move it back after
278 	 * encryption.
279 	 */
280 	if ((x->props.flags & XFRM_STATE_ESN)) {
281 		__u32 seqhi;
282 		struct xfrm_offload *xo = xfrm_offload(skb);
283 
284 		if (xo)
285 			seqhi = xo->seq.hi;
286 		else
287 			seqhi = XFRM_SKB_CB(skb)->seq.output.hi;
288 
289 		extra->esphoff = (unsigned char *)esph -
290 				 skb_transport_header(skb);
291 		esph = (struct ip_esp_hdr *)((unsigned char *)esph - 4);
292 		extra->seqhi = esph->spi;
293 		esph->seq_no = htonl(seqhi);
294 	}
295 
296 	esph->spi = x->id.spi;
297 
298 	return esph;
299 }
300 
301 static void esp_output_done_esn(void *data, int err)
302 {
303 	struct sk_buff *skb = data;
304 
305 	esp_output_restore_header(skb);
306 	esp_output_done(data, err);
307 }
308 
309 static struct ip_esp_hdr *esp_output_udp_encap(struct sk_buff *skb,
310 					       int encap_type,
311 					       struct esp_info *esp,
312 					       __be16 sport,
313 					       __be16 dport)
314 {
315 	struct udphdr *uh;
316 	unsigned int len;
317 	struct xfrm_offload *xo = xfrm_offload(skb);
318 
319 	len = skb->len + esp->tailen - skb_transport_offset(skb);
320 	if (len + sizeof(struct iphdr) > IP_MAX_MTU)
321 		return ERR_PTR(-EMSGSIZE);
322 
323 	uh = (struct udphdr *)esp->esph;
324 	uh->source = sport;
325 	uh->dest = dport;
326 	uh->len = htons(len);
327 	uh->check = 0;
328 
329 	/* For IPv4 ESP with UDP encapsulation, if xo is not null, the skb is in the crypto offload
330 	 * data path, which means that esp_output_udp_encap is called outside of the XFRM stack.
331 	 * In this case, the mac header doesn't point to the IPv4 protocol field, so don't set it.
332 	 */
333 	if (!xo || encap_type != UDP_ENCAP_ESPINUDP)
334 		*skb_mac_header(skb) = IPPROTO_UDP;
335 
336 	return (struct ip_esp_hdr *)(uh + 1);
337 }
338 
339 #ifdef CONFIG_INET_ESPINTCP
340 static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
341 						    struct sk_buff *skb,
342 						    struct esp_info *esp)
343 {
344 	__be16 *lenp = (void *)esp->esph;
345 	struct ip_esp_hdr *esph;
346 	unsigned int len;
347 	struct sock *sk;
348 
349 	len = skb->len + esp->tailen - skb_transport_offset(skb);
350 	if (len > IP_MAX_MTU)
351 		return ERR_PTR(-EMSGSIZE);
352 
353 	rcu_read_lock();
354 	sk = esp_find_tcp_sk(x);
355 	rcu_read_unlock();
356 
357 	if (IS_ERR(sk))
358 		return ERR_CAST(sk);
359 
360 	sock_put(sk);
361 
362 	*lenp = htons(len);
363 	esph = (struct ip_esp_hdr *)(lenp + 1);
364 
365 	return esph;
366 }
367 #else
368 static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
369 						    struct sk_buff *skb,
370 						    struct esp_info *esp)
371 {
372 	return ERR_PTR(-EOPNOTSUPP);
373 }
374 #endif
375 
376 static int esp_output_encap(struct xfrm_state *x, struct sk_buff *skb,
377 			    struct esp_info *esp)
378 {
379 	struct xfrm_encap_tmpl *encap = x->encap;
380 	struct ip_esp_hdr *esph;
381 	__be16 sport, dport;
382 	int encap_type;
383 
384 	spin_lock_bh(&x->lock);
385 	sport = encap->encap_sport;
386 	dport = encap->encap_dport;
387 	encap_type = encap->encap_type;
388 	spin_unlock_bh(&x->lock);
389 
390 	switch (encap_type) {
391 	default:
392 	case UDP_ENCAP_ESPINUDP:
393 		esph = esp_output_udp_encap(skb, encap_type, esp, sport, dport);
394 		break;
395 	case TCP_ENCAP_ESPINTCP:
396 		esph = esp_output_tcp_encap(x, skb, esp);
397 		break;
398 	}
399 
400 	if (IS_ERR(esph))
401 		return PTR_ERR(esph);
402 
403 	esp->esph = esph;
404 
405 	return 0;
406 }
407 
408 int esp_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
409 {
410 	u8 *tail;
411 	int nfrags;
412 	int esph_offset;
413 	struct page *page;
414 	struct sk_buff *trailer;
415 	int tailen = esp->tailen;
416 
417 	/* this is non-NULL only with TCP/UDP Encapsulation */
418 	if (x->encap) {
419 		int err = esp_output_encap(x, skb, esp);
420 
421 		if (err < 0)
422 			return err;
423 	}
424 
425 	if (ALIGN(skb->data_len + tailen, L1_CACHE_BYTES) >
426 	    PAGE_SIZE)
427 		goto cow;
428 
429 	if (!skb_cloned(skb)) {
430 		if (tailen <= skb_tailroom(skb)) {
431 			nfrags = 1;
432 			trailer = skb;
433 			tail = skb_tail_pointer(trailer);
434 
435 			goto skip_cow;
436 		} else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
437 			   && !skb_has_frag_list(skb)) {
438 			int allocsize;
439 			struct sock *sk = skb->sk;
440 			struct page_frag *pfrag = &x->xfrag;
441 
442 			esp->inplace = false;
443 
444 			allocsize = ALIGN(tailen, L1_CACHE_BYTES);
445 
446 			spin_lock_bh(&x->lock);
447 
448 			if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
449 				spin_unlock_bh(&x->lock);
450 				goto cow;
451 			}
452 
453 			page = pfrag->page;
454 			get_page(page);
455 
456 			tail = page_address(page) + pfrag->offset;
457 
458 			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
459 
460 			nfrags = skb_shinfo(skb)->nr_frags;
461 
462 			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
463 					     tailen);
464 			skb_shinfo(skb)->nr_frags = ++nfrags;
465 
466 			pfrag->offset = pfrag->offset + allocsize;
467 
468 			spin_unlock_bh(&x->lock);
469 
470 			nfrags++;
471 
472 			skb_len_add(skb, tailen);
473 			if (sk && sk_fullsock(sk))
474 				refcount_add(tailen, &sk->sk_wmem_alloc);
475 
476 			goto out;
477 		}
478 	}
479 
480 cow:
481 	esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
482 
483 	nfrags = skb_cow_data(skb, tailen, &trailer);
484 	if (nfrags < 0)
485 		goto out;
486 	tail = skb_tail_pointer(trailer);
487 	esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
488 
489 skip_cow:
490 	esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
491 	pskb_put(skb, trailer, tailen);
492 
493 out:
494 	return nfrags;
495 }
496 EXPORT_SYMBOL_GPL(esp_output_head);
497 
498 int esp_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
499 {
500 	u8 *iv;
501 	int alen;
502 	void *tmp;
503 	int ivlen;
504 	int assoclen;
505 	int extralen;
506 	struct page *page;
507 	struct ip_esp_hdr *esph;
508 	struct crypto_aead *aead;
509 	struct aead_request *req;
510 	struct scatterlist *sg, *dsg;
511 	struct esp_output_extra *extra;
512 	int err = -ENOMEM;
513 
514 	assoclen = sizeof(struct ip_esp_hdr);
515 	extralen = 0;
516 
517 	if (x->props.flags & XFRM_STATE_ESN) {
518 		extralen += sizeof(*extra);
519 		assoclen += sizeof(__be32);
520 	}
521 
522 	aead = x->data;
523 	alen = crypto_aead_authsize(aead);
524 	ivlen = crypto_aead_ivsize(aead);
525 
526 	tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
527 	if (!tmp)
528 		goto error;
529 
530 	extra = esp_tmp_extra(tmp);
531 	iv = esp_tmp_iv(aead, tmp, extralen);
532 	req = esp_tmp_req(aead, iv);
533 	sg = esp_req_sg(aead, req);
534 
535 	if (esp->inplace)
536 		dsg = sg;
537 	else
538 		dsg = &sg[esp->nfrags];
539 
540 	esph = esp_output_set_extra(skb, x, esp->esph, extra);
541 	esp->esph = esph;
542 
543 	sg_init_table(sg, esp->nfrags);
544 	err = skb_to_sgvec(skb, sg,
545 		           (unsigned char *)esph - skb->data,
546 		           assoclen + ivlen + esp->clen + alen);
547 	if (unlikely(err < 0))
548 		goto error_free;
549 
550 	if (!esp->inplace) {
551 		int allocsize;
552 		struct page_frag *pfrag = &x->xfrag;
553 
554 		allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
555 
556 		spin_lock_bh(&x->lock);
557 		if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
558 			spin_unlock_bh(&x->lock);
559 			goto error_free;
560 		}
561 
562 		skb_shinfo(skb)->nr_frags = 1;
563 
564 		page = pfrag->page;
565 		get_page(page);
566 		/* replace page frags in skb with new page */
567 		__skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
568 		pfrag->offset = pfrag->offset + allocsize;
569 		spin_unlock_bh(&x->lock);
570 
571 		sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
572 		err = skb_to_sgvec(skb, dsg,
573 			           (unsigned char *)esph - skb->data,
574 			           assoclen + ivlen + esp->clen + alen);
575 		if (unlikely(err < 0)) {
576 			esp_ssg_unref(x, tmp, skb, true);
577 			goto error_free;
578 		}
579 	}
580 
581 	if ((x->props.flags & XFRM_STATE_ESN))
582 		aead_request_set_callback(req, 0, esp_output_done_esn, skb);
583 	else
584 		aead_request_set_callback(req, 0, esp_output_done, skb);
585 
586 	aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
587 	aead_request_set_ad(req, assoclen);
588 
589 	memset(iv, 0, ivlen);
590 	memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
591 	       min(ivlen, 8));
592 
593 	ESP_SKB_CB(skb)->tmp = tmp;
594 	err = crypto_aead_encrypt(req);
595 
596 	switch (err) {
597 	case -EINPROGRESS:
598 		goto error;
599 
600 	case -ENOSPC:
601 		err = NET_XMIT_DROP;
602 		break;
603 
604 	case 0:
605 		if ((x->props.flags & XFRM_STATE_ESN))
606 			esp_output_restore_header(skb);
607 	}
608 
609 	if (sg != dsg)
610 		esp_ssg_unref(x, tmp, skb, false);
611 
612 	if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
613 		err = esp_output_tail_tcp(x, skb);
614 
615 error_free:
616 	kfree(tmp);
617 error:
618 	return err;
619 }
620 EXPORT_SYMBOL_GPL(esp_output_tail);
621 
622 static int esp_output(struct xfrm_state *x, struct sk_buff *skb)
623 {
624 	int alen;
625 	int blksize;
626 	struct ip_esp_hdr *esph;
627 	struct crypto_aead *aead;
628 	struct esp_info esp;
629 
630 	esp.inplace = true;
631 
632 	esp.proto = *skb_mac_header(skb);
633 	*skb_mac_header(skb) = IPPROTO_ESP;
634 
635 	/* skb is pure payload to encrypt */
636 
637 	aead = x->data;
638 	alen = crypto_aead_authsize(aead);
639 
640 	esp.tfclen = 0;
641 	if (x->tfcpad) {
642 		struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
643 		u32 padto;
644 
645 		padto = min(x->tfcpad, xfrm_state_mtu(x, dst->child_mtu_cached));
646 		if (skb->len < padto)
647 			esp.tfclen = padto - skb->len;
648 	}
649 	blksize = ALIGN(crypto_aead_blocksize(aead), 4);
650 	esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
651 	esp.plen = esp.clen - skb->len - esp.tfclen;
652 	esp.tailen = esp.tfclen + esp.plen + alen;
653 
654 	esp.esph = ip_esp_hdr(skb);
655 
656 	esp.nfrags = esp_output_head(x, skb, &esp);
657 	if (esp.nfrags < 0)
658 		return esp.nfrags;
659 
660 	esph = esp.esph;
661 	esph->spi = x->id.spi;
662 
663 	esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
664 	esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
665 				 ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
666 
667 	skb_push(skb, -skb_network_offset(skb));
668 
669 	return esp_output_tail(x, skb, &esp);
670 }
671 
672 static inline int esp_remove_trailer(struct sk_buff *skb)
673 {
674 	struct xfrm_state *x = xfrm_input_state(skb);
675 	struct crypto_aead *aead = x->data;
676 	int alen, hlen, elen;
677 	int padlen, trimlen;
678 	__wsum csumdiff;
679 	u8 nexthdr[2];
680 	int ret;
681 
682 	alen = crypto_aead_authsize(aead);
683 	hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
684 	elen = skb->len - hlen;
685 
686 	if (skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2))
687 		BUG();
688 
689 	ret = -EINVAL;
690 	padlen = nexthdr[0];
691 	if (padlen + 2 + alen >= elen) {
692 		net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
693 				    padlen + 2, elen - alen);
694 		goto out;
695 	}
696 
697 	trimlen = alen + padlen + 2;
698 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
699 		csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
700 		skb->csum = csum_block_sub(skb->csum, csumdiff,
701 					   skb->len - trimlen);
702 	}
703 	ret = pskb_trim(skb, skb->len - trimlen);
704 	if (unlikely(ret))
705 		return ret;
706 
707 	ret = nexthdr[1];
708 
709 out:
710 	return ret;
711 }
712 
713 int esp_input_done2(struct sk_buff *skb, int err)
714 {
715 	const struct iphdr *iph;
716 	struct xfrm_state *x = xfrm_input_state(skb);
717 	struct xfrm_offload *xo = xfrm_offload(skb);
718 	struct crypto_aead *aead = x->data;
719 	int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
720 	int ihl;
721 
722 	if (!xo || !(xo->flags & CRYPTO_DONE))
723 		kfree(ESP_SKB_CB(skb)->tmp);
724 
725 	if (unlikely(err))
726 		goto out;
727 
728 	err = esp_remove_trailer(skb);
729 	if (unlikely(err < 0))
730 		goto out;
731 
732 	iph = ip_hdr(skb);
733 	ihl = iph->ihl * 4;
734 
735 	if (x->encap) {
736 		struct xfrm_encap_tmpl *encap = x->encap;
737 		struct tcphdr *th = (void *)(skb_network_header(skb) + ihl);
738 		struct udphdr *uh = (void *)(skb_network_header(skb) + ihl);
739 		__be16 source;
740 
741 		switch (x->encap->encap_type) {
742 		case TCP_ENCAP_ESPINTCP:
743 			source = th->source;
744 			break;
745 		case UDP_ENCAP_ESPINUDP:
746 			source = uh->source;
747 			break;
748 		default:
749 			WARN_ON_ONCE(1);
750 			err = -EINVAL;
751 			goto out;
752 		}
753 
754 		/*
755 		 * 1) if the NAT-T peer's IP or port changed then
756 		 *    advertise the change to the keying daemon.
757 		 *    This is an inbound SA, so just compare
758 		 *    SRC ports.
759 		 */
760 		if (iph->saddr != x->props.saddr.a4 ||
761 		    source != encap->encap_sport) {
762 			xfrm_address_t ipaddr;
763 
764 			ipaddr.a4 = iph->saddr;
765 			km_new_mapping(x, &ipaddr, source);
766 
767 			/* XXX: perhaps add an extra
768 			 * policy check here, to see
769 			 * if we should allow or
770 			 * reject a packet from a
771 			 * different source
772 			 * address/port.
773 			 */
774 		}
775 
776 		/*
777 		 * 2) ignore UDP/TCP checksums in case
778 		 *    of NAT-T in Transport Mode, or
779 		 *    perform other post-processing fixes
780 		 *    as per draft-ietf-ipsec-udp-encaps-06,
781 		 *    section 3.1.2
782 		 */
783 		if (x->props.mode == XFRM_MODE_TRANSPORT)
784 			skb->ip_summed = CHECKSUM_UNNECESSARY;
785 	}
786 
787 	skb_pull_rcsum(skb, hlen);
788 	if (x->props.mode == XFRM_MODE_TUNNEL ||
789 	    x->props.mode == XFRM_MODE_IPTFS)
790 		skb_reset_transport_header(skb);
791 	else
792 		skb_set_transport_header(skb, -ihl);
793 
794 	/* RFC4303: Drop dummy packets without any error */
795 	if (err == IPPROTO_NONE)
796 		err = -EINVAL;
797 
798 out:
799 	return err;
800 }
801 EXPORT_SYMBOL_GPL(esp_input_done2);
802 
803 static void esp_input_done(void *data, int err)
804 {
805 	struct sk_buff *skb = data;
806 
807 	xfrm_input_resume(skb, esp_input_done2(skb, err));
808 }
809 
810 static void esp_input_restore_header(struct sk_buff *skb)
811 {
812 	esp_restore_header(skb, 0);
813 	__skb_pull(skb, 4);
814 }
815 
816 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
817 {
818 	struct xfrm_state *x = xfrm_input_state(skb);
819 	struct ip_esp_hdr *esph;
820 
821 	/* For ESN we move the header forward by 4 bytes to
822 	 * accommodate the high bits.  We will move it back after
823 	 * decryption.
824 	 */
825 	if ((x->props.flags & XFRM_STATE_ESN)) {
826 		esph = skb_push(skb, 4);
827 		*seqhi = esph->spi;
828 		esph->spi = esph->seq_no;
829 		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
830 	}
831 }
832 
833 static void esp_input_done_esn(void *data, int err)
834 {
835 	struct sk_buff *skb = data;
836 
837 	esp_input_restore_header(skb);
838 	esp_input_done(data, err);
839 }
840 
841 /*
842  * Note: detecting truncated vs. non-truncated authentication data is very
843  * expensive, so we only support truncated data, which is the recommended
844  * and common case.
845  */
846 static int esp_input(struct xfrm_state *x, struct sk_buff *skb)
847 {
848 	struct crypto_aead *aead = x->data;
849 	struct aead_request *req;
850 	struct sk_buff *trailer;
851 	int ivlen = crypto_aead_ivsize(aead);
852 	int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
853 	int nfrags;
854 	int assoclen;
855 	int seqhilen;
856 	__be32 *seqhi;
857 	void *tmp;
858 	u8 *iv;
859 	struct scatterlist *sg;
860 	int err = -EINVAL;
861 
862 	if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen))
863 		goto out;
864 
865 	if (elen <= 0)
866 		goto out;
867 
868 	assoclen = sizeof(struct ip_esp_hdr);
869 	seqhilen = 0;
870 
871 	if (x->props.flags & XFRM_STATE_ESN) {
872 		seqhilen += sizeof(__be32);
873 		assoclen += seqhilen;
874 	}
875 
876 	if (!skb_cloned(skb)) {
877 		if (!skb_is_nonlinear(skb)) {
878 			nfrags = 1;
879 
880 			goto skip_cow;
881 		} else if (!skb_has_frag_list(skb) &&
882 			   !skb_has_shared_frag(skb)) {
883 			nfrags = skb_shinfo(skb)->nr_frags;
884 			nfrags++;
885 
886 			goto skip_cow;
887 		}
888 	}
889 
890 	err = skb_cow_data(skb, 0, &trailer);
891 	if (err < 0)
892 		goto out;
893 
894 	nfrags = err;
895 
896 skip_cow:
897 	err = -ENOMEM;
898 	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
899 	if (!tmp)
900 		goto out;
901 
902 	ESP_SKB_CB(skb)->tmp = tmp;
903 	seqhi = esp_tmp_extra(tmp);
904 	iv = esp_tmp_iv(aead, tmp, seqhilen);
905 	req = esp_tmp_req(aead, iv);
906 	sg = esp_req_sg(aead, req);
907 
908 	esp_input_set_header(skb, seqhi);
909 
910 	sg_init_table(sg, nfrags);
911 	err = skb_to_sgvec(skb, sg, 0, skb->len);
912 	if (unlikely(err < 0)) {
913 		kfree(tmp);
914 		goto out;
915 	}
916 
917 	skb->ip_summed = CHECKSUM_NONE;
918 
919 	if ((x->props.flags & XFRM_STATE_ESN))
920 		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
921 	else
922 		aead_request_set_callback(req, 0, esp_input_done, skb);
923 
924 	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
925 	aead_request_set_ad(req, assoclen);
926 
927 	err = crypto_aead_decrypt(req);
928 	if (err == -EINPROGRESS)
929 		goto out;
930 
931 	if ((x->props.flags & XFRM_STATE_ESN))
932 		esp_input_restore_header(skb);
933 
934 	err = esp_input_done2(skb, err);
935 
936 out:
937 	return err;
938 }
939 
940 static int esp4_err(struct sk_buff *skb, u32 info)
941 {
942 	struct net *net = dev_net(skb->dev);
943 	const struct iphdr *iph = (const struct iphdr *)skb->data;
944 	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data+(iph->ihl<<2));
945 	struct xfrm_state *x;
946 
947 	switch (icmp_hdr(skb)->type) {
948 	case ICMP_DEST_UNREACH:
949 		if (icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
950 			return 0;
951 		break;
952 	case ICMP_REDIRECT:
953 		break;
954 	default:
955 		return 0;
956 	}
957 
958 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
959 			      esph->spi, IPPROTO_ESP, AF_INET);
960 	if (!x)
961 		return 0;
962 
963 	if (icmp_hdr(skb)->type == ICMP_DEST_UNREACH)
964 		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ESP);
965 	else
966 		ipv4_redirect(skb, net, 0, IPPROTO_ESP);
967 	xfrm_state_put(x);
968 
969 	return 0;
970 }
971 
972 static void esp_destroy(struct xfrm_state *x)
973 {
974 	struct crypto_aead *aead = x->data;
975 
976 	if (!aead)
977 		return;
978 
979 	crypto_free_aead(aead);
980 }
981 
982 static int esp_init_aead(struct xfrm_state *x, struct netlink_ext_ack *extack)
983 {
984 	char aead_name[CRYPTO_MAX_ALG_NAME];
985 	struct crypto_aead *aead;
986 	int err;
987 
988 	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
989 		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME) {
990 		NL_SET_ERR_MSG(extack, "Algorithm name is too long");
991 		return -ENAMETOOLONG;
992 	}
993 
994 	aead = crypto_alloc_aead(aead_name, 0, 0);
995 	err = PTR_ERR(aead);
996 	if (IS_ERR(aead))
997 		goto error;
998 
999 	x->data = aead;
1000 
1001 	err = crypto_aead_setkey(aead, x->aead->alg_key,
1002 				 (x->aead->alg_key_len + 7) / 8);
1003 	if (err)
1004 		goto error;
1005 
1006 	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1007 	if (err)
1008 		goto error;
1009 
1010 	return 0;
1011 
1012 error:
1013 	NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1014 	return err;
1015 }
1016 
1017 static int esp_init_authenc(struct xfrm_state *x,
1018 			    struct netlink_ext_ack *extack)
1019 {
1020 	struct crypto_aead *aead;
1021 	struct crypto_authenc_key_param *param;
1022 	struct rtattr *rta;
1023 	char *key;
1024 	char *p;
1025 	char authenc_name[CRYPTO_MAX_ALG_NAME];
1026 	unsigned int keylen;
1027 	int err;
1028 
1029 	err = -ENAMETOOLONG;
1030 
1031 	if ((x->props.flags & XFRM_STATE_ESN)) {
1032 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1033 			     "%s%sauthencesn(%s,%s)%s",
1034 			     x->geniv ?: "", x->geniv ? "(" : "",
1035 			     x->aalg ? x->aalg->alg_name : "digest_null",
1036 			     x->ealg->alg_name,
1037 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1038 			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1039 			goto error;
1040 		}
1041 	} else {
1042 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1043 			     "%s%sauthenc(%s,%s)%s",
1044 			     x->geniv ?: "", x->geniv ? "(" : "",
1045 			     x->aalg ? x->aalg->alg_name : "digest_null",
1046 			     x->ealg->alg_name,
1047 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1048 			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1049 			goto error;
1050 		}
1051 	}
1052 
1053 	aead = crypto_alloc_aead(authenc_name, 0, 0);
1054 	err = PTR_ERR(aead);
1055 	if (IS_ERR(aead)) {
1056 		NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1057 		goto error;
1058 	}
1059 
1060 	x->data = aead;
1061 
1062 	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1063 		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1064 	err = -ENOMEM;
1065 	key = kmalloc(keylen, GFP_KERNEL);
1066 	if (!key)
1067 		goto error;
1068 
1069 	p = key;
1070 	rta = (void *)p;
1071 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1072 	rta->rta_len = RTA_LENGTH(sizeof(*param));
1073 	param = RTA_DATA(rta);
1074 	p += RTA_SPACE(sizeof(*param));
1075 
1076 	if (x->aalg) {
1077 		struct xfrm_algo_desc *aalg_desc;
1078 
1079 		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1080 		p += (x->aalg->alg_key_len + 7) / 8;
1081 
1082 		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1083 		BUG_ON(!aalg_desc);
1084 
1085 		err = -EINVAL;
1086 		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1087 		    crypto_aead_authsize(aead)) {
1088 			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1089 			goto free_key;
1090 		}
1091 
1092 		err = crypto_aead_setauthsize(
1093 			aead, x->aalg->alg_trunc_len / 8);
1094 		if (err) {
1095 			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1096 			goto free_key;
1097 		}
1098 	}
1099 
1100 	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1101 	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1102 
1103 	err = crypto_aead_setkey(aead, key, keylen);
1104 
1105 free_key:
1106 	kfree_sensitive(key);
1107 
1108 error:
1109 	return err;
1110 }
1111 
1112 static int esp_init_state(struct xfrm_state *x, struct netlink_ext_ack *extack)
1113 {
1114 	struct crypto_aead *aead;
1115 	u32 align;
1116 	int err;
1117 
1118 	x->data = NULL;
1119 
1120 	if (x->aead) {
1121 		err = esp_init_aead(x, extack);
1122 	} else if (x->ealg) {
1123 		err = esp_init_authenc(x, extack);
1124 	} else {
1125 		NL_SET_ERR_MSG(extack, "ESP: AEAD or CRYPT must be provided");
1126 		err = -EINVAL;
1127 	}
1128 
1129 	if (err)
1130 		goto error;
1131 
1132 	aead = x->data;
1133 
1134 	x->props.header_len = sizeof(struct ip_esp_hdr) +
1135 			      crypto_aead_ivsize(aead);
1136 	if (x->props.mode == XFRM_MODE_TUNNEL)
1137 		x->props.header_len += sizeof(struct iphdr);
1138 	else if (x->props.mode == XFRM_MODE_BEET && x->sel.family != AF_INET6)
1139 		x->props.header_len += IPV4_BEET_PHMAXLEN;
1140 	if (x->encap) {
1141 		struct xfrm_encap_tmpl *encap = x->encap;
1142 
1143 		switch (encap->encap_type) {
1144 		default:
1145 			NL_SET_ERR_MSG(extack, "Unsupported encapsulation type for ESP");
1146 			err = -EINVAL;
1147 			goto error;
1148 		case UDP_ENCAP_ESPINUDP:
1149 			x->props.header_len += sizeof(struct udphdr);
1150 			break;
1151 #ifdef CONFIG_INET_ESPINTCP
1152 		case TCP_ENCAP_ESPINTCP:
1153 			/* only the length field, TCP encap is done by
1154 			 * the socket
1155 			 */
1156 			x->props.header_len += 2;
1157 			break;
1158 #endif
1159 		}
1160 	}
1161 
1162 	align = ALIGN(crypto_aead_blocksize(aead), 4);
1163 	x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1164 
1165 error:
1166 	return err;
1167 }
1168 
1169 static int esp4_rcv_cb(struct sk_buff *skb, int err)
1170 {
1171 	return 0;
1172 }
1173 
1174 static const struct xfrm_type esp_type =
1175 {
1176 	.owner		= THIS_MODULE,
1177 	.proto	     	= IPPROTO_ESP,
1178 	.flags		= XFRM_TYPE_REPLAY_PROT,
1179 	.init_state	= esp_init_state,
1180 	.destructor	= esp_destroy,
1181 	.input		= esp_input,
1182 	.output		= esp_output,
1183 };
1184 
1185 static struct xfrm4_protocol esp4_protocol = {
1186 	.handler	=	xfrm4_rcv,
1187 	.input_handler	=	xfrm_input,
1188 	.cb_handler	=	esp4_rcv_cb,
1189 	.err_handler	=	esp4_err,
1190 	.priority	=	0,
1191 };
1192 
1193 static int __init esp4_init(void)
1194 {
1195 	if (xfrm_register_type(&esp_type, AF_INET) < 0) {
1196 		pr_info("%s: can't add xfrm type\n", __func__);
1197 		return -EAGAIN;
1198 	}
1199 	if (xfrm4_protocol_register(&esp4_protocol, IPPROTO_ESP) < 0) {
1200 		pr_info("%s: can't add protocol\n", __func__);
1201 		xfrm_unregister_type(&esp_type, AF_INET);
1202 		return -EAGAIN;
1203 	}
1204 	return 0;
1205 }
1206 
1207 static void __exit esp4_fini(void)
1208 {
1209 	if (xfrm4_protocol_deregister(&esp4_protocol, IPPROTO_ESP) < 0)
1210 		pr_info("%s: can't remove protocol\n", __func__);
1211 	xfrm_unregister_type(&esp_type, AF_INET);
1212 }
1213 
1214 module_init(esp4_init);
1215 module_exit(esp4_fini);
1216 MODULE_DESCRIPTION("IPv4 ESP transformation library");
1217 MODULE_LICENSE("GPL");
1218 MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_ESP);
1219