xref: /openbmc/linux/net/ipv4/esp4.c (revision 93f5715e)
1 #define pr_fmt(fmt) "IPsec: " fmt
2 
3 #include <crypto/aead.h>
4 #include <crypto/authenc.h>
5 #include <linux/err.h>
6 #include <linux/module.h>
7 #include <net/ip.h>
8 #include <net/xfrm.h>
9 #include <net/esp.h>
10 #include <linux/scatterlist.h>
11 #include <linux/kernel.h>
12 #include <linux/pfkeyv2.h>
13 #include <linux/rtnetlink.h>
14 #include <linux/slab.h>
15 #include <linux/spinlock.h>
16 #include <linux/in6.h>
17 #include <net/icmp.h>
18 #include <net/protocol.h>
19 #include <net/udp.h>
20 
21 #include <linux/highmem.h>
22 
23 struct esp_skb_cb {
24 	struct xfrm_skb_cb xfrm;
25 	void *tmp;
26 };
27 
28 struct esp_output_extra {
29 	__be32 seqhi;
30 	u32 esphoff;
31 };
32 
33 #define ESP_SKB_CB(__skb) ((struct esp_skb_cb *)&((__skb)->cb[0]))
34 
35 static u32 esp4_get_mtu(struct xfrm_state *x, int mtu);
36 
37 /*
38  * Allocate an AEAD request structure with extra space for SG and IV.
39  *
40  * For alignment considerations the IV is placed at the front, followed
41  * by the request and finally the SG list.
42  *
43  * TODO: Use spare space in skb for this where possible.
44  */
45 static void *esp_alloc_tmp(struct crypto_aead *aead, int nfrags, int extralen)
46 {
47 	unsigned int len;
48 
49 	len = extralen;
50 
51 	len += crypto_aead_ivsize(aead);
52 
53 	if (len) {
54 		len += crypto_aead_alignmask(aead) &
55 		       ~(crypto_tfm_ctx_alignment() - 1);
56 		len = ALIGN(len, crypto_tfm_ctx_alignment());
57 	}
58 
59 	len += sizeof(struct aead_request) + crypto_aead_reqsize(aead);
60 	len = ALIGN(len, __alignof__(struct scatterlist));
61 
62 	len += sizeof(struct scatterlist) * nfrags;
63 
64 	return kmalloc(len, GFP_ATOMIC);
65 }
66 
67 static inline void *esp_tmp_extra(void *tmp)
68 {
69 	return PTR_ALIGN(tmp, __alignof__(struct esp_output_extra));
70 }
71 
72 static inline u8 *esp_tmp_iv(struct crypto_aead *aead, void *tmp, int extralen)
73 {
74 	return crypto_aead_ivsize(aead) ?
75 	       PTR_ALIGN((u8 *)tmp + extralen,
76 			 crypto_aead_alignmask(aead) + 1) : tmp + extralen;
77 }
78 
79 static inline struct aead_request *esp_tmp_req(struct crypto_aead *aead, u8 *iv)
80 {
81 	struct aead_request *req;
82 
83 	req = (void *)PTR_ALIGN(iv + crypto_aead_ivsize(aead),
84 				crypto_tfm_ctx_alignment());
85 	aead_request_set_tfm(req, aead);
86 	return req;
87 }
88 
89 static inline struct scatterlist *esp_req_sg(struct crypto_aead *aead,
90 					     struct aead_request *req)
91 {
92 	return (void *)ALIGN((unsigned long)(req + 1) +
93 			     crypto_aead_reqsize(aead),
94 			     __alignof__(struct scatterlist));
95 }
96 
97 static void esp_ssg_unref(struct xfrm_state *x, void *tmp)
98 {
99 	struct esp_output_extra *extra = esp_tmp_extra(tmp);
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(*extra);
108 
109 	extra = esp_tmp_extra(tmp);
110 	iv = esp_tmp_iv(aead, tmp, extralen);
111 	req = esp_tmp_req(aead, iv);
112 
113 	/* Unref skb_frag_pages in the src scatterlist if necessary.
114 	 * Skip the first sg which comes from skb->data.
115 	 */
116 	if (req->src != req->dst)
117 		for (sg = sg_next(req->src); sg; sg = sg_next(sg))
118 			put_page(sg_page(sg));
119 }
120 
121 static void esp_output_done(struct crypto_async_request *base, int err)
122 {
123 	struct sk_buff *skb = base->data;
124 	struct xfrm_offload *xo = xfrm_offload(skb);
125 	void *tmp;
126 	struct xfrm_state *x;
127 
128 	if (xo && (xo->flags & XFRM_DEV_RESUME))
129 		x = skb->sp->xvec[skb->sp->len - 1];
130 	else
131 		x = skb_dst(skb)->xfrm;
132 
133 	tmp = ESP_SKB_CB(skb)->tmp;
134 	esp_ssg_unref(x, tmp);
135 	kfree(tmp);
136 
137 	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
138 		if (err) {
139 			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
140 			kfree_skb(skb);
141 			return;
142 		}
143 
144 		skb_push(skb, skb->data - skb_mac_header(skb));
145 		secpath_reset(skb);
146 		xfrm_dev_resume(skb);
147 	} else {
148 		xfrm_output_resume(skb, err);
149 	}
150 }
151 
152 /* Move ESP header back into place. */
153 static void esp_restore_header(struct sk_buff *skb, unsigned int offset)
154 {
155 	struct ip_esp_hdr *esph = (void *)(skb->data + offset);
156 	void *tmp = ESP_SKB_CB(skb)->tmp;
157 	__be32 *seqhi = esp_tmp_extra(tmp);
158 
159 	esph->seq_no = esph->spi;
160 	esph->spi = *seqhi;
161 }
162 
163 static void esp_output_restore_header(struct sk_buff *skb)
164 {
165 	void *tmp = ESP_SKB_CB(skb)->tmp;
166 	struct esp_output_extra *extra = esp_tmp_extra(tmp);
167 
168 	esp_restore_header(skb, skb_transport_offset(skb) + extra->esphoff -
169 				sizeof(__be32));
170 }
171 
172 static struct ip_esp_hdr *esp_output_set_extra(struct sk_buff *skb,
173 					       struct xfrm_state *x,
174 					       struct ip_esp_hdr *esph,
175 					       struct esp_output_extra *extra)
176 {
177 	/* For ESN we move the header forward by 4 bytes to
178 	 * accomodate the high bits.  We will move it back after
179 	 * encryption.
180 	 */
181 	if ((x->props.flags & XFRM_STATE_ESN)) {
182 		__u32 seqhi;
183 		struct xfrm_offload *xo = xfrm_offload(skb);
184 
185 		if (xo)
186 			seqhi = xo->seq.hi;
187 		else
188 			seqhi = XFRM_SKB_CB(skb)->seq.output.hi;
189 
190 		extra->esphoff = (unsigned char *)esph -
191 				 skb_transport_header(skb);
192 		esph = (struct ip_esp_hdr *)((unsigned char *)esph - 4);
193 		extra->seqhi = esph->spi;
194 		esph->seq_no = htonl(seqhi);
195 	}
196 
197 	esph->spi = x->id.spi;
198 
199 	return esph;
200 }
201 
202 static void esp_output_done_esn(struct crypto_async_request *base, int err)
203 {
204 	struct sk_buff *skb = base->data;
205 
206 	esp_output_restore_header(skb);
207 	esp_output_done(base, err);
208 }
209 
210 static void esp_output_fill_trailer(u8 *tail, int tfclen, int plen, __u8 proto)
211 {
212 	/* Fill padding... */
213 	if (tfclen) {
214 		memset(tail, 0, tfclen);
215 		tail += tfclen;
216 	}
217 	do {
218 		int i;
219 		for (i = 0; i < plen - 2; i++)
220 			tail[i] = i + 1;
221 	} while (0);
222 	tail[plen - 2] = plen - 2;
223 	tail[plen - 1] = proto;
224 }
225 
226 static void esp_output_udp_encap(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
227 {
228 	int encap_type;
229 	struct udphdr *uh;
230 	__be32 *udpdata32;
231 	__be16 sport, dport;
232 	struct xfrm_encap_tmpl *encap = x->encap;
233 	struct ip_esp_hdr *esph = esp->esph;
234 
235 	spin_lock_bh(&x->lock);
236 	sport = encap->encap_sport;
237 	dport = encap->encap_dport;
238 	encap_type = encap->encap_type;
239 	spin_unlock_bh(&x->lock);
240 
241 	uh = (struct udphdr *)esph;
242 	uh->source = sport;
243 	uh->dest = dport;
244 	uh->len = htons(skb->len + esp->tailen
245 		  - skb_transport_offset(skb));
246 	uh->check = 0;
247 
248 	switch (encap_type) {
249 	default:
250 	case UDP_ENCAP_ESPINUDP:
251 		esph = (struct ip_esp_hdr *)(uh + 1);
252 		break;
253 	case UDP_ENCAP_ESPINUDP_NON_IKE:
254 		udpdata32 = (__be32 *)(uh + 1);
255 		udpdata32[0] = udpdata32[1] = 0;
256 		esph = (struct ip_esp_hdr *)(udpdata32 + 2);
257 		break;
258 	}
259 
260 	*skb_mac_header(skb) = IPPROTO_UDP;
261 	esp->esph = esph;
262 }
263 
264 int esp_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
265 {
266 	u8 *tail;
267 	u8 *vaddr;
268 	int nfrags;
269 	int esph_offset;
270 	struct page *page;
271 	struct sk_buff *trailer;
272 	int tailen = esp->tailen;
273 
274 	/* this is non-NULL only with UDP Encapsulation */
275 	if (x->encap)
276 		esp_output_udp_encap(x, skb, esp);
277 
278 	if (!skb_cloned(skb)) {
279 		if (tailen <= skb_tailroom(skb)) {
280 			nfrags = 1;
281 			trailer = skb;
282 			tail = skb_tail_pointer(trailer);
283 
284 			goto skip_cow;
285 		} else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
286 			   && !skb_has_frag_list(skb)) {
287 			int allocsize;
288 			struct sock *sk = skb->sk;
289 			struct page_frag *pfrag = &x->xfrag;
290 
291 			esp->inplace = false;
292 
293 			allocsize = ALIGN(tailen, L1_CACHE_BYTES);
294 
295 			spin_lock_bh(&x->lock);
296 
297 			if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
298 				spin_unlock_bh(&x->lock);
299 				goto cow;
300 			}
301 
302 			page = pfrag->page;
303 			get_page(page);
304 
305 			vaddr = kmap_atomic(page);
306 
307 			tail = vaddr + pfrag->offset;
308 
309 			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
310 
311 			kunmap_atomic(vaddr);
312 
313 			nfrags = skb_shinfo(skb)->nr_frags;
314 
315 			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
316 					     tailen);
317 			skb_shinfo(skb)->nr_frags = ++nfrags;
318 
319 			pfrag->offset = pfrag->offset + allocsize;
320 
321 			spin_unlock_bh(&x->lock);
322 
323 			nfrags++;
324 
325 			skb->len += tailen;
326 			skb->data_len += tailen;
327 			skb->truesize += tailen;
328 			if (sk)
329 				refcount_add(tailen, &sk->sk_wmem_alloc);
330 
331 			goto out;
332 		}
333 	}
334 
335 cow:
336 	esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
337 
338 	nfrags = skb_cow_data(skb, tailen, &trailer);
339 	if (nfrags < 0)
340 		goto out;
341 	tail = skb_tail_pointer(trailer);
342 	esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
343 
344 skip_cow:
345 	esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
346 	pskb_put(skb, trailer, tailen);
347 
348 out:
349 	return nfrags;
350 }
351 EXPORT_SYMBOL_GPL(esp_output_head);
352 
353 int esp_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
354 {
355 	u8 *iv;
356 	int alen;
357 	void *tmp;
358 	int ivlen;
359 	int assoclen;
360 	int extralen;
361 	struct page *page;
362 	struct ip_esp_hdr *esph;
363 	struct crypto_aead *aead;
364 	struct aead_request *req;
365 	struct scatterlist *sg, *dsg;
366 	struct esp_output_extra *extra;
367 	int err = -ENOMEM;
368 
369 	assoclen = sizeof(struct ip_esp_hdr);
370 	extralen = 0;
371 
372 	if (x->props.flags & XFRM_STATE_ESN) {
373 		extralen += sizeof(*extra);
374 		assoclen += sizeof(__be32);
375 	}
376 
377 	aead = x->data;
378 	alen = crypto_aead_authsize(aead);
379 	ivlen = crypto_aead_ivsize(aead);
380 
381 	tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
382 	if (!tmp)
383 		goto error;
384 
385 	extra = esp_tmp_extra(tmp);
386 	iv = esp_tmp_iv(aead, tmp, extralen);
387 	req = esp_tmp_req(aead, iv);
388 	sg = esp_req_sg(aead, req);
389 
390 	if (esp->inplace)
391 		dsg = sg;
392 	else
393 		dsg = &sg[esp->nfrags];
394 
395 	esph = esp_output_set_extra(skb, x, esp->esph, extra);
396 	esp->esph = esph;
397 
398 	sg_init_table(sg, esp->nfrags);
399 	err = skb_to_sgvec(skb, sg,
400 		           (unsigned char *)esph - skb->data,
401 		           assoclen + ivlen + esp->clen + alen);
402 	if (unlikely(err < 0))
403 		goto error_free;
404 
405 	if (!esp->inplace) {
406 		int allocsize;
407 		struct page_frag *pfrag = &x->xfrag;
408 
409 		allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
410 
411 		spin_lock_bh(&x->lock);
412 		if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
413 			spin_unlock_bh(&x->lock);
414 			goto error_free;
415 		}
416 
417 		skb_shinfo(skb)->nr_frags = 1;
418 
419 		page = pfrag->page;
420 		get_page(page);
421 		/* replace page frags in skb with new page */
422 		__skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
423 		pfrag->offset = pfrag->offset + allocsize;
424 		spin_unlock_bh(&x->lock);
425 
426 		sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
427 		err = skb_to_sgvec(skb, dsg,
428 			           (unsigned char *)esph - skb->data,
429 			           assoclen + ivlen + esp->clen + alen);
430 		if (unlikely(err < 0))
431 			goto error_free;
432 	}
433 
434 	if ((x->props.flags & XFRM_STATE_ESN))
435 		aead_request_set_callback(req, 0, esp_output_done_esn, skb);
436 	else
437 		aead_request_set_callback(req, 0, esp_output_done, skb);
438 
439 	aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
440 	aead_request_set_ad(req, assoclen);
441 
442 	memset(iv, 0, ivlen);
443 	memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
444 	       min(ivlen, 8));
445 
446 	ESP_SKB_CB(skb)->tmp = tmp;
447 	err = crypto_aead_encrypt(req);
448 
449 	switch (err) {
450 	case -EINPROGRESS:
451 		goto error;
452 
453 	case -ENOSPC:
454 		err = NET_XMIT_DROP;
455 		break;
456 
457 	case 0:
458 		if ((x->props.flags & XFRM_STATE_ESN))
459 			esp_output_restore_header(skb);
460 	}
461 
462 	if (sg != dsg)
463 		esp_ssg_unref(x, tmp);
464 
465 error_free:
466 	kfree(tmp);
467 error:
468 	return err;
469 }
470 EXPORT_SYMBOL_GPL(esp_output_tail);
471 
472 static int esp_output(struct xfrm_state *x, struct sk_buff *skb)
473 {
474 	int alen;
475 	int blksize;
476 	struct ip_esp_hdr *esph;
477 	struct crypto_aead *aead;
478 	struct esp_info esp;
479 
480 	esp.inplace = true;
481 
482 	esp.proto = *skb_mac_header(skb);
483 	*skb_mac_header(skb) = IPPROTO_ESP;
484 
485 	/* skb is pure payload to encrypt */
486 
487 	aead = x->data;
488 	alen = crypto_aead_authsize(aead);
489 
490 	esp.tfclen = 0;
491 	if (x->tfcpad) {
492 		struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
493 		u32 padto;
494 
495 		padto = min(x->tfcpad, esp4_get_mtu(x, dst->child_mtu_cached));
496 		if (skb->len < padto)
497 			esp.tfclen = padto - skb->len;
498 	}
499 	blksize = ALIGN(crypto_aead_blocksize(aead), 4);
500 	esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
501 	esp.plen = esp.clen - skb->len - esp.tfclen;
502 	esp.tailen = esp.tfclen + esp.plen + alen;
503 
504 	esp.esph = ip_esp_hdr(skb);
505 
506 	esp.nfrags = esp_output_head(x, skb, &esp);
507 	if (esp.nfrags < 0)
508 		return esp.nfrags;
509 
510 	esph = esp.esph;
511 	esph->spi = x->id.spi;
512 
513 	esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
514 	esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
515 				 ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
516 
517 	skb_push(skb, -skb_network_offset(skb));
518 
519 	return esp_output_tail(x, skb, &esp);
520 }
521 
522 static inline int esp_remove_trailer(struct sk_buff *skb)
523 {
524 	struct xfrm_state *x = xfrm_input_state(skb);
525 	struct xfrm_offload *xo = xfrm_offload(skb);
526 	struct crypto_aead *aead = x->data;
527 	int alen, hlen, elen;
528 	int padlen, trimlen;
529 	__wsum csumdiff;
530 	u8 nexthdr[2];
531 	int ret;
532 
533 	alen = crypto_aead_authsize(aead);
534 	hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
535 	elen = skb->len - hlen;
536 
537 	if (xo && (xo->flags & XFRM_ESP_NO_TRAILER)) {
538 		ret = xo->proto;
539 		goto out;
540 	}
541 
542 	if (skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2))
543 		BUG();
544 
545 	ret = -EINVAL;
546 	padlen = nexthdr[0];
547 	if (padlen + 2 + alen >= elen) {
548 		net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
549 				    padlen + 2, elen - alen);
550 		goto out;
551 	}
552 
553 	trimlen = alen + padlen + 2;
554 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
555 		csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
556 		skb->csum = csum_block_sub(skb->csum, csumdiff,
557 					   skb->len - trimlen);
558 	}
559 	pskb_trim(skb, skb->len - trimlen);
560 
561 	ret = nexthdr[1];
562 
563 out:
564 	return ret;
565 }
566 
567 int esp_input_done2(struct sk_buff *skb, int err)
568 {
569 	const struct iphdr *iph;
570 	struct xfrm_state *x = xfrm_input_state(skb);
571 	struct xfrm_offload *xo = xfrm_offload(skb);
572 	struct crypto_aead *aead = x->data;
573 	int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
574 	int ihl;
575 
576 	if (!xo || (xo && !(xo->flags & CRYPTO_DONE)))
577 		kfree(ESP_SKB_CB(skb)->tmp);
578 
579 	if (unlikely(err))
580 		goto out;
581 
582 	err = esp_remove_trailer(skb);
583 	if (unlikely(err < 0))
584 		goto out;
585 
586 	iph = ip_hdr(skb);
587 	ihl = iph->ihl * 4;
588 
589 	if (x->encap) {
590 		struct xfrm_encap_tmpl *encap = x->encap;
591 		struct udphdr *uh = (void *)(skb_network_header(skb) + ihl);
592 
593 		/*
594 		 * 1) if the NAT-T peer's IP or port changed then
595 		 *    advertize the change to the keying daemon.
596 		 *    This is an inbound SA, so just compare
597 		 *    SRC ports.
598 		 */
599 		if (iph->saddr != x->props.saddr.a4 ||
600 		    uh->source != encap->encap_sport) {
601 			xfrm_address_t ipaddr;
602 
603 			ipaddr.a4 = iph->saddr;
604 			km_new_mapping(x, &ipaddr, uh->source);
605 
606 			/* XXX: perhaps add an extra
607 			 * policy check here, to see
608 			 * if we should allow or
609 			 * reject a packet from a
610 			 * different source
611 			 * address/port.
612 			 */
613 		}
614 
615 		/*
616 		 * 2) ignore UDP/TCP checksums in case
617 		 *    of NAT-T in Transport Mode, or
618 		 *    perform other post-processing fixes
619 		 *    as per draft-ietf-ipsec-udp-encaps-06,
620 		 *    section 3.1.2
621 		 */
622 		if (x->props.mode == XFRM_MODE_TRANSPORT)
623 			skb->ip_summed = CHECKSUM_UNNECESSARY;
624 	}
625 
626 	skb_pull_rcsum(skb, hlen);
627 	if (x->props.mode == XFRM_MODE_TUNNEL)
628 		skb_reset_transport_header(skb);
629 	else
630 		skb_set_transport_header(skb, -ihl);
631 
632 	/* RFC4303: Drop dummy packets without any error */
633 	if (err == IPPROTO_NONE)
634 		err = -EINVAL;
635 
636 out:
637 	return err;
638 }
639 EXPORT_SYMBOL_GPL(esp_input_done2);
640 
641 static void esp_input_done(struct crypto_async_request *base, int err)
642 {
643 	struct sk_buff *skb = base->data;
644 
645 	xfrm_input_resume(skb, esp_input_done2(skb, err));
646 }
647 
648 static void esp_input_restore_header(struct sk_buff *skb)
649 {
650 	esp_restore_header(skb, 0);
651 	__skb_pull(skb, 4);
652 }
653 
654 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
655 {
656 	struct xfrm_state *x = xfrm_input_state(skb);
657 	struct ip_esp_hdr *esph;
658 
659 	/* For ESN we move the header forward by 4 bytes to
660 	 * accomodate the high bits.  We will move it back after
661 	 * decryption.
662 	 */
663 	if ((x->props.flags & XFRM_STATE_ESN)) {
664 		esph = skb_push(skb, 4);
665 		*seqhi = esph->spi;
666 		esph->spi = esph->seq_no;
667 		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
668 	}
669 }
670 
671 static void esp_input_done_esn(struct crypto_async_request *base, int err)
672 {
673 	struct sk_buff *skb = base->data;
674 
675 	esp_input_restore_header(skb);
676 	esp_input_done(base, err);
677 }
678 
679 /*
680  * Note: detecting truncated vs. non-truncated authentication data is very
681  * expensive, so we only support truncated data, which is the recommended
682  * and common case.
683  */
684 static int esp_input(struct xfrm_state *x, struct sk_buff *skb)
685 {
686 	struct ip_esp_hdr *esph;
687 	struct crypto_aead *aead = x->data;
688 	struct aead_request *req;
689 	struct sk_buff *trailer;
690 	int ivlen = crypto_aead_ivsize(aead);
691 	int elen = skb->len - sizeof(*esph) - ivlen;
692 	int nfrags;
693 	int assoclen;
694 	int seqhilen;
695 	__be32 *seqhi;
696 	void *tmp;
697 	u8 *iv;
698 	struct scatterlist *sg;
699 	int err = -EINVAL;
700 
701 	if (!pskb_may_pull(skb, sizeof(*esph) + ivlen))
702 		goto out;
703 
704 	if (elen <= 0)
705 		goto out;
706 
707 	assoclen = sizeof(*esph);
708 	seqhilen = 0;
709 
710 	if (x->props.flags & XFRM_STATE_ESN) {
711 		seqhilen += sizeof(__be32);
712 		assoclen += seqhilen;
713 	}
714 
715 	if (!skb_cloned(skb)) {
716 		if (!skb_is_nonlinear(skb)) {
717 			nfrags = 1;
718 
719 			goto skip_cow;
720 		} else if (!skb_has_frag_list(skb)) {
721 			nfrags = skb_shinfo(skb)->nr_frags;
722 			nfrags++;
723 
724 			goto skip_cow;
725 		}
726 	}
727 
728 	err = skb_cow_data(skb, 0, &trailer);
729 	if (err < 0)
730 		goto out;
731 
732 	nfrags = err;
733 
734 skip_cow:
735 	err = -ENOMEM;
736 	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
737 	if (!tmp)
738 		goto out;
739 
740 	ESP_SKB_CB(skb)->tmp = tmp;
741 	seqhi = esp_tmp_extra(tmp);
742 	iv = esp_tmp_iv(aead, tmp, seqhilen);
743 	req = esp_tmp_req(aead, iv);
744 	sg = esp_req_sg(aead, req);
745 
746 	esp_input_set_header(skb, seqhi);
747 
748 	sg_init_table(sg, nfrags);
749 	err = skb_to_sgvec(skb, sg, 0, skb->len);
750 	if (unlikely(err < 0)) {
751 		kfree(tmp);
752 		goto out;
753 	}
754 
755 	skb->ip_summed = CHECKSUM_NONE;
756 
757 	if ((x->props.flags & XFRM_STATE_ESN))
758 		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
759 	else
760 		aead_request_set_callback(req, 0, esp_input_done, skb);
761 
762 	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
763 	aead_request_set_ad(req, assoclen);
764 
765 	err = crypto_aead_decrypt(req);
766 	if (err == -EINPROGRESS)
767 		goto out;
768 
769 	if ((x->props.flags & XFRM_STATE_ESN))
770 		esp_input_restore_header(skb);
771 
772 	err = esp_input_done2(skb, err);
773 
774 out:
775 	return err;
776 }
777 
778 static u32 esp4_get_mtu(struct xfrm_state *x, int mtu)
779 {
780 	struct crypto_aead *aead = x->data;
781 	u32 blksize = ALIGN(crypto_aead_blocksize(aead), 4);
782 	unsigned int net_adj;
783 
784 	switch (x->props.mode) {
785 	case XFRM_MODE_TRANSPORT:
786 	case XFRM_MODE_BEET:
787 		net_adj = sizeof(struct iphdr);
788 		break;
789 	case XFRM_MODE_TUNNEL:
790 		net_adj = 0;
791 		break;
792 	default:
793 		BUG();
794 	}
795 
796 	return ((mtu - x->props.header_len - crypto_aead_authsize(aead) -
797 		 net_adj) & ~(blksize - 1)) + net_adj - 2;
798 }
799 
800 static int esp4_err(struct sk_buff *skb, u32 info)
801 {
802 	struct net *net = dev_net(skb->dev);
803 	const struct iphdr *iph = (const struct iphdr *)skb->data;
804 	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data+(iph->ihl<<2));
805 	struct xfrm_state *x;
806 
807 	switch (icmp_hdr(skb)->type) {
808 	case ICMP_DEST_UNREACH:
809 		if (icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
810 			return 0;
811 	case ICMP_REDIRECT:
812 		break;
813 	default:
814 		return 0;
815 	}
816 
817 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
818 			      esph->spi, IPPROTO_ESP, AF_INET);
819 	if (!x)
820 		return 0;
821 
822 	if (icmp_hdr(skb)->type == ICMP_DEST_UNREACH)
823 		ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ESP, 0);
824 	else
825 		ipv4_redirect(skb, net, 0, 0, IPPROTO_ESP, 0);
826 	xfrm_state_put(x);
827 
828 	return 0;
829 }
830 
831 static void esp_destroy(struct xfrm_state *x)
832 {
833 	struct crypto_aead *aead = x->data;
834 
835 	if (!aead)
836 		return;
837 
838 	crypto_free_aead(aead);
839 }
840 
841 static int esp_init_aead(struct xfrm_state *x)
842 {
843 	char aead_name[CRYPTO_MAX_ALG_NAME];
844 	struct crypto_aead *aead;
845 	int err;
846 
847 	err = -ENAMETOOLONG;
848 	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
849 		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME)
850 		goto error;
851 
852 	aead = crypto_alloc_aead(aead_name, 0, 0);
853 	err = PTR_ERR(aead);
854 	if (IS_ERR(aead))
855 		goto error;
856 
857 	x->data = aead;
858 
859 	err = crypto_aead_setkey(aead, x->aead->alg_key,
860 				 (x->aead->alg_key_len + 7) / 8);
861 	if (err)
862 		goto error;
863 
864 	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
865 	if (err)
866 		goto error;
867 
868 error:
869 	return err;
870 }
871 
872 static int esp_init_authenc(struct xfrm_state *x)
873 {
874 	struct crypto_aead *aead;
875 	struct crypto_authenc_key_param *param;
876 	struct rtattr *rta;
877 	char *key;
878 	char *p;
879 	char authenc_name[CRYPTO_MAX_ALG_NAME];
880 	unsigned int keylen;
881 	int err;
882 
883 	err = -EINVAL;
884 	if (!x->ealg)
885 		goto error;
886 
887 	err = -ENAMETOOLONG;
888 
889 	if ((x->props.flags & XFRM_STATE_ESN)) {
890 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
891 			     "%s%sauthencesn(%s,%s)%s",
892 			     x->geniv ?: "", x->geniv ? "(" : "",
893 			     x->aalg ? x->aalg->alg_name : "digest_null",
894 			     x->ealg->alg_name,
895 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
896 			goto error;
897 	} else {
898 		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
899 			     "%s%sauthenc(%s,%s)%s",
900 			     x->geniv ?: "", x->geniv ? "(" : "",
901 			     x->aalg ? x->aalg->alg_name : "digest_null",
902 			     x->ealg->alg_name,
903 			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
904 			goto error;
905 	}
906 
907 	aead = crypto_alloc_aead(authenc_name, 0, 0);
908 	err = PTR_ERR(aead);
909 	if (IS_ERR(aead))
910 		goto error;
911 
912 	x->data = aead;
913 
914 	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
915 		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
916 	err = -ENOMEM;
917 	key = kmalloc(keylen, GFP_KERNEL);
918 	if (!key)
919 		goto error;
920 
921 	p = key;
922 	rta = (void *)p;
923 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
924 	rta->rta_len = RTA_LENGTH(sizeof(*param));
925 	param = RTA_DATA(rta);
926 	p += RTA_SPACE(sizeof(*param));
927 
928 	if (x->aalg) {
929 		struct xfrm_algo_desc *aalg_desc;
930 
931 		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
932 		p += (x->aalg->alg_key_len + 7) / 8;
933 
934 		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
935 		BUG_ON(!aalg_desc);
936 
937 		err = -EINVAL;
938 		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
939 		    crypto_aead_authsize(aead)) {
940 			pr_info("ESP: %s digestsize %u != %hu\n",
941 				x->aalg->alg_name,
942 				crypto_aead_authsize(aead),
943 				aalg_desc->uinfo.auth.icv_fullbits / 8);
944 			goto free_key;
945 		}
946 
947 		err = crypto_aead_setauthsize(
948 			aead, x->aalg->alg_trunc_len / 8);
949 		if (err)
950 			goto free_key;
951 	}
952 
953 	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
954 	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
955 
956 	err = crypto_aead_setkey(aead, key, keylen);
957 
958 free_key:
959 	kfree(key);
960 
961 error:
962 	return err;
963 }
964 
965 static int esp_init_state(struct xfrm_state *x)
966 {
967 	struct crypto_aead *aead;
968 	u32 align;
969 	int err;
970 
971 	x->data = NULL;
972 
973 	if (x->aead)
974 		err = esp_init_aead(x);
975 	else
976 		err = esp_init_authenc(x);
977 
978 	if (err)
979 		goto error;
980 
981 	aead = x->data;
982 
983 	x->props.header_len = sizeof(struct ip_esp_hdr) +
984 			      crypto_aead_ivsize(aead);
985 	if (x->props.mode == XFRM_MODE_TUNNEL)
986 		x->props.header_len += sizeof(struct iphdr);
987 	else if (x->props.mode == XFRM_MODE_BEET && x->sel.family != AF_INET6)
988 		x->props.header_len += IPV4_BEET_PHMAXLEN;
989 	if (x->encap) {
990 		struct xfrm_encap_tmpl *encap = x->encap;
991 
992 		switch (encap->encap_type) {
993 		default:
994 			err = -EINVAL;
995 			goto error;
996 		case UDP_ENCAP_ESPINUDP:
997 			x->props.header_len += sizeof(struct udphdr);
998 			break;
999 		case UDP_ENCAP_ESPINUDP_NON_IKE:
1000 			x->props.header_len += sizeof(struct udphdr) + 2 * sizeof(u32);
1001 			break;
1002 		}
1003 	}
1004 
1005 	align = ALIGN(crypto_aead_blocksize(aead), 4);
1006 	x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1007 
1008 error:
1009 	return err;
1010 }
1011 
1012 static int esp4_rcv_cb(struct sk_buff *skb, int err)
1013 {
1014 	return 0;
1015 }
1016 
1017 static const struct xfrm_type esp_type =
1018 {
1019 	.description	= "ESP4",
1020 	.owner		= THIS_MODULE,
1021 	.proto	     	= IPPROTO_ESP,
1022 	.flags		= XFRM_TYPE_REPLAY_PROT,
1023 	.init_state	= esp_init_state,
1024 	.destructor	= esp_destroy,
1025 	.get_mtu	= esp4_get_mtu,
1026 	.input		= esp_input,
1027 	.output		= esp_output,
1028 };
1029 
1030 static struct xfrm4_protocol esp4_protocol = {
1031 	.handler	=	xfrm4_rcv,
1032 	.input_handler	=	xfrm_input,
1033 	.cb_handler	=	esp4_rcv_cb,
1034 	.err_handler	=	esp4_err,
1035 	.priority	=	0,
1036 };
1037 
1038 static int __init esp4_init(void)
1039 {
1040 	if (xfrm_register_type(&esp_type, AF_INET) < 0) {
1041 		pr_info("%s: can't add xfrm type\n", __func__);
1042 		return -EAGAIN;
1043 	}
1044 	if (xfrm4_protocol_register(&esp4_protocol, IPPROTO_ESP) < 0) {
1045 		pr_info("%s: can't add protocol\n", __func__);
1046 		xfrm_unregister_type(&esp_type, AF_INET);
1047 		return -EAGAIN;
1048 	}
1049 	return 0;
1050 }
1051 
1052 static void __exit esp4_fini(void)
1053 {
1054 	if (xfrm4_protocol_deregister(&esp4_protocol, IPPROTO_ESP) < 0)
1055 		pr_info("%s: can't remove protocol\n", __func__);
1056 	if (xfrm_unregister_type(&esp_type, AF_INET) < 0)
1057 		pr_info("%s: can't remove xfrm type\n", __func__);
1058 }
1059 
1060 module_init(esp4_init);
1061 module_exit(esp4_fini);
1062 MODULE_LICENSE("GPL");
1063 MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_ESP);
1064