1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Copyright (C)2002 USAGI/WIDE Project
4 *
5 * Authors
6 *
7 *	Mitsuru KANDA @USAGI       : IPv6 Support
8 *	Kazunori MIYAZAWA @USAGI   :
9 *	Kunihiro Ishiguro <kunihiro@ipinfusion.com>
10 *
11 *	This file is derived from net/ipv4/esp.c
12 */
13
14#define pr_fmt(fmt) "IPv6: " fmt
15
16#include <crypto/aead.h>
17#include <crypto/authenc.h>
18#include <linux/err.h>
19#include <linux/module.h>
20#include <net/ip.h>
21#include <net/xfrm.h>
22#include <net/esp.h>
23#include <linux/scatterlist.h>
24#include <linux/kernel.h>
25#include <linux/pfkeyv2.h>
26#include <linux/random.h>
27#include <linux/slab.h>
28#include <linux/spinlock.h>
29#include <net/ip6_checksum.h>
30#include <net/ip6_route.h>
31#include <net/icmp.h>
32#include <net/ipv6.h>
33#include <net/protocol.h>
34#include <net/udp.h>
35#include <linux/icmpv6.h>
36#include <net/tcp.h>
37#include <net/espintcp.h>
38#include <net/inet6_hashtables.h>
39
40#include <linux/highmem.h>
41
42struct esp_skb_cb {
43	struct xfrm_skb_cb xfrm;
44	void *tmp;
45};
46
47struct esp_output_extra {
48	__be32 seqhi;
49	u32 esphoff;
50};
51
52#define ESP_SKB_CB(__skb) ((struct esp_skb_cb *)&((__skb)->cb[0]))
53
54/*
55 * Allocate an AEAD request structure with extra space for SG and IV.
56 *
57 * For alignment considerations the upper 32 bits of the sequence number are
58 * placed at the front, if present. Followed by the IV, the request and finally
59 * the SG list.
60 *
61 * TODO: Use spare space in skb for this where possible.
62 */
63static void *esp_alloc_tmp(struct crypto_aead *aead, int nfrags, int seqihlen)
64{
65	unsigned int len;
66
67	len = seqihlen;
68
69	len += crypto_aead_ivsize(aead);
70
71	if (len) {
72		len += crypto_aead_alignmask(aead) &
73		       ~(crypto_tfm_ctx_alignment() - 1);
74		len = ALIGN(len, crypto_tfm_ctx_alignment());
75	}
76
77	len += sizeof(struct aead_request) + crypto_aead_reqsize(aead);
78	len = ALIGN(len, __alignof__(struct scatterlist));
79
80	len += sizeof(struct scatterlist) * nfrags;
81
82	return kmalloc(len, GFP_ATOMIC);
83}
84
85static inline void *esp_tmp_extra(void *tmp)
86{
87	return PTR_ALIGN(tmp, __alignof__(struct esp_output_extra));
88}
89
90static inline u8 *esp_tmp_iv(struct crypto_aead *aead, void *tmp, int seqhilen)
91{
92	return crypto_aead_ivsize(aead) ?
93	       PTR_ALIGN((u8 *)tmp + seqhilen,
94			 crypto_aead_alignmask(aead) + 1) : tmp + seqhilen;
95}
96
97static inline struct aead_request *esp_tmp_req(struct crypto_aead *aead, u8 *iv)
98{
99	struct aead_request *req;
100
101	req = (void *)PTR_ALIGN(iv + crypto_aead_ivsize(aead),
102				crypto_tfm_ctx_alignment());
103	aead_request_set_tfm(req, aead);
104	return req;
105}
106
107static inline struct scatterlist *esp_req_sg(struct crypto_aead *aead,
108					     struct aead_request *req)
109{
110	return (void *)ALIGN((unsigned long)(req + 1) +
111			     crypto_aead_reqsize(aead),
112			     __alignof__(struct scatterlist));
113}
114
115static void esp_ssg_unref(struct xfrm_state *x, void *tmp, struct sk_buff *skb)
116{
117	struct crypto_aead *aead = x->data;
118	int extralen = 0;
119	u8 *iv;
120	struct aead_request *req;
121	struct scatterlist *sg;
122
123	if (x->props.flags & XFRM_STATE_ESN)
124		extralen += sizeof(struct esp_output_extra);
125
126	iv = esp_tmp_iv(aead, tmp, extralen);
127	req = esp_tmp_req(aead, iv);
128
129	/* Unref skb_frag_pages in the src scatterlist if necessary.
130	 * Skip the first sg which comes from skb->data.
131	 */
132	if (req->src != req->dst)
133		for (sg = sg_next(req->src); sg; sg = sg_next(sg))
134			skb_page_unref(skb, sg_page(sg), false);
135}
136
137#ifdef CONFIG_INET6_ESPINTCP
138struct esp_tcp_sk {
139	struct sock *sk;
140	struct rcu_head rcu;
141};
142
143static void esp_free_tcp_sk(struct rcu_head *head)
144{
145	struct esp_tcp_sk *esk = container_of(head, struct esp_tcp_sk, rcu);
146
147	sock_put(esk->sk);
148	kfree(esk);
149}
150
151static struct sock *esp6_find_tcp_sk(struct xfrm_state *x)
152{
153	struct xfrm_encap_tmpl *encap = x->encap;
154	struct net *net = xs_net(x);
155	struct esp_tcp_sk *esk;
156	__be16 sport, dport;
157	struct sock *nsk;
158	struct sock *sk;
159
160	sk = rcu_dereference(x->encap_sk);
161	if (sk && sk->sk_state == TCP_ESTABLISHED)
162		return sk;
163
164	spin_lock_bh(&x->lock);
165	sport = encap->encap_sport;
166	dport = encap->encap_dport;
167	nsk = rcu_dereference_protected(x->encap_sk,
168					lockdep_is_held(&x->lock));
169	if (sk && sk == nsk) {
170		esk = kmalloc(sizeof(*esk), GFP_ATOMIC);
171		if (!esk) {
172			spin_unlock_bh(&x->lock);
173			return ERR_PTR(-ENOMEM);
174		}
175		RCU_INIT_POINTER(x->encap_sk, NULL);
176		esk->sk = sk;
177		call_rcu(&esk->rcu, esp_free_tcp_sk);
178	}
179	spin_unlock_bh(&x->lock);
180
181	sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo, &x->id.daddr.in6,
182					dport, &x->props.saddr.in6, ntohs(sport), 0, 0);
183	if (!sk)
184		return ERR_PTR(-ENOENT);
185
186	if (!tcp_is_ulp_esp(sk)) {
187		sock_put(sk);
188		return ERR_PTR(-EINVAL);
189	}
190
191	spin_lock_bh(&x->lock);
192	nsk = rcu_dereference_protected(x->encap_sk,
193					lockdep_is_held(&x->lock));
194	if (encap->encap_sport != sport ||
195	    encap->encap_dport != dport) {
196		sock_put(sk);
197		sk = nsk ?: ERR_PTR(-EREMCHG);
198	} else if (sk == nsk) {
199		sock_put(sk);
200	} else {
201		rcu_assign_pointer(x->encap_sk, sk);
202	}
203	spin_unlock_bh(&x->lock);
204
205	return sk;
206}
207
208static int esp_output_tcp_finish(struct xfrm_state *x, struct sk_buff *skb)
209{
210	struct sock *sk;
211	int err;
212
213	rcu_read_lock();
214
215	sk = esp6_find_tcp_sk(x);
216	err = PTR_ERR_OR_ZERO(sk);
217	if (err)
218		goto out;
219
220	bh_lock_sock(sk);
221	if (sock_owned_by_user(sk))
222		err = espintcp_queue_out(sk, skb);
223	else
224		err = espintcp_push_skb(sk, skb);
225	bh_unlock_sock(sk);
226
227out:
228	rcu_read_unlock();
229	return err;
230}
231
232static int esp_output_tcp_encap_cb(struct net *net, struct sock *sk,
233				   struct sk_buff *skb)
234{
235	struct dst_entry *dst = skb_dst(skb);
236	struct xfrm_state *x = dst->xfrm;
237
238	return esp_output_tcp_finish(x, skb);
239}
240
241static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
242{
243	int err;
244
245	local_bh_disable();
246	err = xfrm_trans_queue_net(xs_net(x), skb, esp_output_tcp_encap_cb);
247	local_bh_enable();
248
249	/* EINPROGRESS just happens to do the right thing.  It
250	 * actually means that the skb has been consumed and
251	 * isn't coming back.
252	 */
253	return err ?: -EINPROGRESS;
254}
255#else
256static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
257{
258	kfree_skb(skb);
259
260	return -EOPNOTSUPP;
261}
262#endif
263
264static void esp_output_encap_csum(struct sk_buff *skb)
265{
266	/* UDP encap with IPv6 requires a valid checksum */
267	if (*skb_mac_header(skb) == IPPROTO_UDP) {
268		struct udphdr *uh = udp_hdr(skb);
269		struct ipv6hdr *ip6h = ipv6_hdr(skb);
270		int len = ntohs(uh->len);
271		unsigned int offset = skb_transport_offset(skb);
272		__wsum csum = skb_checksum(skb, offset, skb->len - offset, 0);
273
274		uh->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
275					    len, IPPROTO_UDP, csum);
276		if (uh->check == 0)
277			uh->check = CSUM_MANGLED_0;
278	}
279}
280
281static void esp_output_done(void *data, int err)
282{
283	struct sk_buff *skb = data;
284	struct xfrm_offload *xo = xfrm_offload(skb);
285	void *tmp;
286	struct xfrm_state *x;
287
288	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
289		struct sec_path *sp = skb_sec_path(skb);
290
291		x = sp->xvec[sp->len - 1];
292	} else {
293		x = skb_dst(skb)->xfrm;
294	}
295
296	tmp = ESP_SKB_CB(skb)->tmp;
297	esp_ssg_unref(x, tmp, skb);
298	kfree(tmp);
299
300	esp_output_encap_csum(skb);
301
302	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
303		if (err) {
304			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
305			kfree_skb(skb);
306			return;
307		}
308
309		skb_push(skb, skb->data - skb_mac_header(skb));
310		secpath_reset(skb);
311		xfrm_dev_resume(skb);
312	} else {
313		if (!err &&
314		    x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
315			esp_output_tail_tcp(x, skb);
316		else
317			xfrm_output_resume(skb->sk, skb, err);
318	}
319}
320
321/* Move ESP header back into place. */
322static void esp_restore_header(struct sk_buff *skb, unsigned int offset)
323{
324	struct ip_esp_hdr *esph = (void *)(skb->data + offset);
325	void *tmp = ESP_SKB_CB(skb)->tmp;
326	__be32 *seqhi = esp_tmp_extra(tmp);
327
328	esph->seq_no = esph->spi;
329	esph->spi = *seqhi;
330}
331
332static void esp_output_restore_header(struct sk_buff *skb)
333{
334	void *tmp = ESP_SKB_CB(skb)->tmp;
335	struct esp_output_extra *extra = esp_tmp_extra(tmp);
336
337	esp_restore_header(skb, skb_transport_offset(skb) + extra->esphoff -
338				sizeof(__be32));
339}
340
341static struct ip_esp_hdr *esp_output_set_esn(struct sk_buff *skb,
342					     struct xfrm_state *x,
343					     struct ip_esp_hdr *esph,
344					     struct esp_output_extra *extra)
345{
346	/* For ESN we move the header forward by 4 bytes to
347	 * accommodate the high bits.  We will move it back after
348	 * encryption.
349	 */
350	if ((x->props.flags & XFRM_STATE_ESN)) {
351		__u32 seqhi;
352		struct xfrm_offload *xo = xfrm_offload(skb);
353
354		if (xo)
355			seqhi = xo->seq.hi;
356		else
357			seqhi = XFRM_SKB_CB(skb)->seq.output.hi;
358
359		extra->esphoff = (unsigned char *)esph -
360				 skb_transport_header(skb);
361		esph = (struct ip_esp_hdr *)((unsigned char *)esph - 4);
362		extra->seqhi = esph->spi;
363		esph->seq_no = htonl(seqhi);
364	}
365
366	esph->spi = x->id.spi;
367
368	return esph;
369}
370
371static void esp_output_done_esn(void *data, int err)
372{
373	struct sk_buff *skb = data;
374
375	esp_output_restore_header(skb);
376	esp_output_done(data, err);
377}
378
379static struct ip_esp_hdr *esp6_output_udp_encap(struct sk_buff *skb,
380					       int encap_type,
381					       struct esp_info *esp,
382					       __be16 sport,
383					       __be16 dport)
384{
385	struct udphdr *uh;
386	__be32 *udpdata32;
387	unsigned int len;
388
389	len = skb->len + esp->tailen - skb_transport_offset(skb);
390	if (len > U16_MAX)
391		return ERR_PTR(-EMSGSIZE);
392
393	uh = (struct udphdr *)esp->esph;
394	uh->source = sport;
395	uh->dest = dport;
396	uh->len = htons(len);
397	uh->check = 0;
398
399	*skb_mac_header(skb) = IPPROTO_UDP;
400
401	if (encap_type == UDP_ENCAP_ESPINUDP_NON_IKE) {
402		udpdata32 = (__be32 *)(uh + 1);
403		udpdata32[0] = udpdata32[1] = 0;
404		return (struct ip_esp_hdr *)(udpdata32 + 2);
405	}
406
407	return (struct ip_esp_hdr *)(uh + 1);
408}
409
410#ifdef CONFIG_INET6_ESPINTCP
411static struct ip_esp_hdr *esp6_output_tcp_encap(struct xfrm_state *x,
412						struct sk_buff *skb,
413						struct esp_info *esp)
414{
415	__be16 *lenp = (void *)esp->esph;
416	struct ip_esp_hdr *esph;
417	unsigned int len;
418	struct sock *sk;
419
420	len = skb->len + esp->tailen - skb_transport_offset(skb);
421	if (len > IP_MAX_MTU)
422		return ERR_PTR(-EMSGSIZE);
423
424	rcu_read_lock();
425	sk = esp6_find_tcp_sk(x);
426	rcu_read_unlock();
427
428	if (IS_ERR(sk))
429		return ERR_CAST(sk);
430
431	*lenp = htons(len);
432	esph = (struct ip_esp_hdr *)(lenp + 1);
433
434	return esph;
435}
436#else
437static struct ip_esp_hdr *esp6_output_tcp_encap(struct xfrm_state *x,
438						struct sk_buff *skb,
439						struct esp_info *esp)
440{
441	return ERR_PTR(-EOPNOTSUPP);
442}
443#endif
444
445static int esp6_output_encap(struct xfrm_state *x, struct sk_buff *skb,
446			    struct esp_info *esp)
447{
448	struct xfrm_encap_tmpl *encap = x->encap;
449	struct ip_esp_hdr *esph;
450	__be16 sport, dport;
451	int encap_type;
452
453	spin_lock_bh(&x->lock);
454	sport = encap->encap_sport;
455	dport = encap->encap_dport;
456	encap_type = encap->encap_type;
457	spin_unlock_bh(&x->lock);
458
459	switch (encap_type) {
460	default:
461	case UDP_ENCAP_ESPINUDP:
462	case UDP_ENCAP_ESPINUDP_NON_IKE:
463		esph = esp6_output_udp_encap(skb, encap_type, esp, sport, dport);
464		break;
465	case TCP_ENCAP_ESPINTCP:
466		esph = esp6_output_tcp_encap(x, skb, esp);
467		break;
468	}
469
470	if (IS_ERR(esph))
471		return PTR_ERR(esph);
472
473	esp->esph = esph;
474
475	return 0;
476}
477
478int esp6_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
479{
480	u8 *tail;
481	int nfrags;
482	int esph_offset;
483	struct page *page;
484	struct sk_buff *trailer;
485	int tailen = esp->tailen;
486
487	if (x->encap) {
488		int err = esp6_output_encap(x, skb, esp);
489
490		if (err < 0)
491			return err;
492	}
493
494	if (ALIGN(tailen, L1_CACHE_BYTES) > PAGE_SIZE ||
495	    ALIGN(skb->data_len, L1_CACHE_BYTES) > PAGE_SIZE)
496		goto cow;
497
498	if (!skb_cloned(skb)) {
499		if (tailen <= skb_tailroom(skb)) {
500			nfrags = 1;
501			trailer = skb;
502			tail = skb_tail_pointer(trailer);
503
504			goto skip_cow;
505		} else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
506			   && !skb_has_frag_list(skb)) {
507			int allocsize;
508			struct sock *sk = skb->sk;
509			struct page_frag *pfrag = &x->xfrag;
510
511			esp->inplace = false;
512
513			allocsize = ALIGN(tailen, L1_CACHE_BYTES);
514
515			spin_lock_bh(&x->lock);
516
517			if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
518				spin_unlock_bh(&x->lock);
519				goto cow;
520			}
521
522			page = pfrag->page;
523			get_page(page);
524
525			tail = page_address(page) + pfrag->offset;
526
527			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
528
529			nfrags = skb_shinfo(skb)->nr_frags;
530
531			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
532					     tailen);
533			skb_shinfo(skb)->nr_frags = ++nfrags;
534
535			pfrag->offset = pfrag->offset + allocsize;
536
537			spin_unlock_bh(&x->lock);
538
539			nfrags++;
540
541			skb->len += tailen;
542			skb->data_len += tailen;
543			skb->truesize += tailen;
544			if (sk && sk_fullsock(sk))
545				refcount_add(tailen, &sk->sk_wmem_alloc);
546
547			goto out;
548		}
549	}
550
551cow:
552	esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
553
554	nfrags = skb_cow_data(skb, tailen, &trailer);
555	if (nfrags < 0)
556		goto out;
557	tail = skb_tail_pointer(trailer);
558	esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
559
560skip_cow:
561	esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
562	pskb_put(skb, trailer, tailen);
563
564out:
565	return nfrags;
566}
567EXPORT_SYMBOL_GPL(esp6_output_head);
568
569int esp6_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
570{
571	u8 *iv;
572	int alen;
573	void *tmp;
574	int ivlen;
575	int assoclen;
576	int extralen;
577	struct page *page;
578	struct ip_esp_hdr *esph;
579	struct aead_request *req;
580	struct crypto_aead *aead;
581	struct scatterlist *sg, *dsg;
582	struct esp_output_extra *extra;
583	int err = -ENOMEM;
584
585	assoclen = sizeof(struct ip_esp_hdr);
586	extralen = 0;
587
588	if (x->props.flags & XFRM_STATE_ESN) {
589		extralen += sizeof(*extra);
590		assoclen += sizeof(__be32);
591	}
592
593	aead = x->data;
594	alen = crypto_aead_authsize(aead);
595	ivlen = crypto_aead_ivsize(aead);
596
597	tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
598	if (!tmp)
599		goto error;
600
601	extra = esp_tmp_extra(tmp);
602	iv = esp_tmp_iv(aead, tmp, extralen);
603	req = esp_tmp_req(aead, iv);
604	sg = esp_req_sg(aead, req);
605
606	if (esp->inplace)
607		dsg = sg;
608	else
609		dsg = &sg[esp->nfrags];
610
611	esph = esp_output_set_esn(skb, x, esp->esph, extra);
612	esp->esph = esph;
613
614	sg_init_table(sg, esp->nfrags);
615	err = skb_to_sgvec(skb, sg,
616		           (unsigned char *)esph - skb->data,
617		           assoclen + ivlen + esp->clen + alen);
618	if (unlikely(err < 0))
619		goto error_free;
620
621	if (!esp->inplace) {
622		int allocsize;
623		struct page_frag *pfrag = &x->xfrag;
624
625		allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
626
627		spin_lock_bh(&x->lock);
628		if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
629			spin_unlock_bh(&x->lock);
630			goto error_free;
631		}
632
633		skb_shinfo(skb)->nr_frags = 1;
634
635		page = pfrag->page;
636		get_page(page);
637		/* replace page frags in skb with new page */
638		__skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
639		pfrag->offset = pfrag->offset + allocsize;
640		spin_unlock_bh(&x->lock);
641
642		sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
643		err = skb_to_sgvec(skb, dsg,
644			           (unsigned char *)esph - skb->data,
645			           assoclen + ivlen + esp->clen + alen);
646		if (unlikely(err < 0))
647			goto error_free;
648	}
649
650	if ((x->props.flags & XFRM_STATE_ESN))
651		aead_request_set_callback(req, 0, esp_output_done_esn, skb);
652	else
653		aead_request_set_callback(req, 0, esp_output_done, skb);
654
655	aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
656	aead_request_set_ad(req, assoclen);
657
658	memset(iv, 0, ivlen);
659	memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
660	       min(ivlen, 8));
661
662	ESP_SKB_CB(skb)->tmp = tmp;
663	err = crypto_aead_encrypt(req);
664
665	switch (err) {
666	case -EINPROGRESS:
667		goto error;
668
669	case -ENOSPC:
670		err = NET_XMIT_DROP;
671		break;
672
673	case 0:
674		if ((x->props.flags & XFRM_STATE_ESN))
675			esp_output_restore_header(skb);
676		esp_output_encap_csum(skb);
677	}
678
679	if (sg != dsg)
680		esp_ssg_unref(x, tmp, skb);
681
682	if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
683		err = esp_output_tail_tcp(x, skb);
684
685error_free:
686	kfree(tmp);
687error:
688	return err;
689}
690EXPORT_SYMBOL_GPL(esp6_output_tail);
691
692static int esp6_output(struct xfrm_state *x, struct sk_buff *skb)
693{
694	int alen;
695	int blksize;
696	struct ip_esp_hdr *esph;
697	struct crypto_aead *aead;
698	struct esp_info esp;
699
700	esp.inplace = true;
701
702	esp.proto = *skb_mac_header(skb);
703	*skb_mac_header(skb) = IPPROTO_ESP;
704
705	/* skb is pure payload to encrypt */
706
707	aead = x->data;
708	alen = crypto_aead_authsize(aead);
709
710	esp.tfclen = 0;
711	if (x->tfcpad) {
712		struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
713		u32 padto;
714
715		padto = min(x->tfcpad, xfrm_state_mtu(x, dst->child_mtu_cached));
716		if (skb->len < padto)
717			esp.tfclen = padto - skb->len;
718	}
719	blksize = ALIGN(crypto_aead_blocksize(aead), 4);
720	esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
721	esp.plen = esp.clen - skb->len - esp.tfclen;
722	esp.tailen = esp.tfclen + esp.plen + alen;
723
724	esp.esph = ip_esp_hdr(skb);
725
726	esp.nfrags = esp6_output_head(x, skb, &esp);
727	if (esp.nfrags < 0)
728		return esp.nfrags;
729
730	esph = esp.esph;
731	esph->spi = x->id.spi;
732
733	esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
734	esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
735			    ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
736
737	skb_push(skb, -skb_network_offset(skb));
738
739	return esp6_output_tail(x, skb, &esp);
740}
741
742static inline int esp_remove_trailer(struct sk_buff *skb)
743{
744	struct xfrm_state *x = xfrm_input_state(skb);
745	struct crypto_aead *aead = x->data;
746	int alen, hlen, elen;
747	int padlen, trimlen;
748	__wsum csumdiff;
749	u8 nexthdr[2];
750	int ret;
751
752	alen = crypto_aead_authsize(aead);
753	hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
754	elen = skb->len - hlen;
755
756	ret = skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2);
757	BUG_ON(ret);
758
759	ret = -EINVAL;
760	padlen = nexthdr[0];
761	if (padlen + 2 + alen >= elen) {
762		net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
763				    padlen + 2, elen - alen);
764		goto out;
765	}
766
767	trimlen = alen + padlen + 2;
768	if (skb->ip_summed == CHECKSUM_COMPLETE) {
769		csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
770		skb->csum = csum_block_sub(skb->csum, csumdiff,
771					   skb->len - trimlen);
772	}
773	ret = pskb_trim(skb, skb->len - trimlen);
774	if (unlikely(ret))
775		return ret;
776
777	ret = nexthdr[1];
778
779out:
780	return ret;
781}
782
783int esp6_input_done2(struct sk_buff *skb, int err)
784{
785	struct xfrm_state *x = xfrm_input_state(skb);
786	struct xfrm_offload *xo = xfrm_offload(skb);
787	struct crypto_aead *aead = x->data;
788	int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
789	int hdr_len = skb_network_header_len(skb);
790
791	if (!xo || !(xo->flags & CRYPTO_DONE))
792		kfree(ESP_SKB_CB(skb)->tmp);
793
794	if (unlikely(err))
795		goto out;
796
797	err = esp_remove_trailer(skb);
798	if (unlikely(err < 0))
799		goto out;
800
801	if (x->encap) {
802		const struct ipv6hdr *ip6h = ipv6_hdr(skb);
803		int offset = skb_network_offset(skb) + sizeof(*ip6h);
804		struct xfrm_encap_tmpl *encap = x->encap;
805		u8 nexthdr = ip6h->nexthdr;
806		__be16 frag_off, source;
807		struct udphdr *uh;
808		struct tcphdr *th;
809
810		offset = ipv6_skip_exthdr(skb, offset, &nexthdr, &frag_off);
811		if (offset == -1) {
812			err = -EINVAL;
813			goto out;
814		}
815
816		uh = (void *)(skb->data + offset);
817		th = (void *)(skb->data + offset);
818		hdr_len += offset;
819
820		switch (x->encap->encap_type) {
821		case TCP_ENCAP_ESPINTCP:
822			source = th->source;
823			break;
824		case UDP_ENCAP_ESPINUDP:
825		case UDP_ENCAP_ESPINUDP_NON_IKE:
826			source = uh->source;
827			break;
828		default:
829			WARN_ON_ONCE(1);
830			err = -EINVAL;
831			goto out;
832		}
833
834		/*
835		 * 1) if the NAT-T peer's IP or port changed then
836		 *    advertise the change to the keying daemon.
837		 *    This is an inbound SA, so just compare
838		 *    SRC ports.
839		 */
840		if (!ipv6_addr_equal(&ip6h->saddr, &x->props.saddr.in6) ||
841		    source != encap->encap_sport) {
842			xfrm_address_t ipaddr;
843
844			memcpy(&ipaddr.a6, &ip6h->saddr.s6_addr, sizeof(ipaddr.a6));
845			km_new_mapping(x, &ipaddr, source);
846
847			/* XXX: perhaps add an extra
848			 * policy check here, to see
849			 * if we should allow or
850			 * reject a packet from a
851			 * different source
852			 * address/port.
853			 */
854		}
855
856		/*
857		 * 2) ignore UDP/TCP checksums in case
858		 *    of NAT-T in Transport Mode, or
859		 *    perform other post-processing fixes
860		 *    as per draft-ietf-ipsec-udp-encaps-06,
861		 *    section 3.1.2
862		 */
863		if (x->props.mode == XFRM_MODE_TRANSPORT)
864			skb->ip_summed = CHECKSUM_UNNECESSARY;
865	}
866
867	skb_postpull_rcsum(skb, skb_network_header(skb),
868			   skb_network_header_len(skb));
869	skb_pull_rcsum(skb, hlen);
870	if (x->props.mode == XFRM_MODE_TUNNEL)
871		skb_reset_transport_header(skb);
872	else
873		skb_set_transport_header(skb, -hdr_len);
874
875	/* RFC4303: Drop dummy packets without any error */
876	if (err == IPPROTO_NONE)
877		err = -EINVAL;
878
879out:
880	return err;
881}
882EXPORT_SYMBOL_GPL(esp6_input_done2);
883
884static void esp_input_done(void *data, int err)
885{
886	struct sk_buff *skb = data;
887
888	xfrm_input_resume(skb, esp6_input_done2(skb, err));
889}
890
891static void esp_input_restore_header(struct sk_buff *skb)
892{
893	esp_restore_header(skb, 0);
894	__skb_pull(skb, 4);
895}
896
897static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
898{
899	struct xfrm_state *x = xfrm_input_state(skb);
900
901	/* For ESN we move the header forward by 4 bytes to
902	 * accommodate the high bits.  We will move it back after
903	 * decryption.
904	 */
905	if ((x->props.flags & XFRM_STATE_ESN)) {
906		struct ip_esp_hdr *esph = skb_push(skb, 4);
907
908		*seqhi = esph->spi;
909		esph->spi = esph->seq_no;
910		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
911	}
912}
913
914static void esp_input_done_esn(void *data, int err)
915{
916	struct sk_buff *skb = data;
917
918	esp_input_restore_header(skb);
919	esp_input_done(data, err);
920}
921
922static int esp6_input(struct xfrm_state *x, struct sk_buff *skb)
923{
924	struct crypto_aead *aead = x->data;
925	struct aead_request *req;
926	struct sk_buff *trailer;
927	int ivlen = crypto_aead_ivsize(aead);
928	int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
929	int nfrags;
930	int assoclen;
931	int seqhilen;
932	int ret = 0;
933	void *tmp;
934	__be32 *seqhi;
935	u8 *iv;
936	struct scatterlist *sg;
937
938	if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen)) {
939		ret = -EINVAL;
940		goto out;
941	}
942
943	if (elen <= 0) {
944		ret = -EINVAL;
945		goto out;
946	}
947
948	assoclen = sizeof(struct ip_esp_hdr);
949	seqhilen = 0;
950
951	if (x->props.flags & XFRM_STATE_ESN) {
952		seqhilen += sizeof(__be32);
953		assoclen += seqhilen;
954	}
955
956	if (!skb_cloned(skb)) {
957		if (!skb_is_nonlinear(skb)) {
958			nfrags = 1;
959
960			goto skip_cow;
961		} else if (!skb_has_frag_list(skb)) {
962			nfrags = skb_shinfo(skb)->nr_frags;
963			nfrags++;
964
965			goto skip_cow;
966		}
967	}
968
969	nfrags = skb_cow_data(skb, 0, &trailer);
970	if (nfrags < 0) {
971		ret = -EINVAL;
972		goto out;
973	}
974
975skip_cow:
976	ret = -ENOMEM;
977	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
978	if (!tmp)
979		goto out;
980
981	ESP_SKB_CB(skb)->tmp = tmp;
982	seqhi = esp_tmp_extra(tmp);
983	iv = esp_tmp_iv(aead, tmp, seqhilen);
984	req = esp_tmp_req(aead, iv);
985	sg = esp_req_sg(aead, req);
986
987	esp_input_set_header(skb, seqhi);
988
989	sg_init_table(sg, nfrags);
990	ret = skb_to_sgvec(skb, sg, 0, skb->len);
991	if (unlikely(ret < 0)) {
992		kfree(tmp);
993		goto out;
994	}
995
996	skb->ip_summed = CHECKSUM_NONE;
997
998	if ((x->props.flags & XFRM_STATE_ESN))
999		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
1000	else
1001		aead_request_set_callback(req, 0, esp_input_done, skb);
1002
1003	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
1004	aead_request_set_ad(req, assoclen);
1005
1006	ret = crypto_aead_decrypt(req);
1007	if (ret == -EINPROGRESS)
1008		goto out;
1009
1010	if ((x->props.flags & XFRM_STATE_ESN))
1011		esp_input_restore_header(skb);
1012
1013	ret = esp6_input_done2(skb, ret);
1014
1015out:
1016	return ret;
1017}
1018
1019static int esp6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
1020		    u8 type, u8 code, int offset, __be32 info)
1021{
1022	struct net *net = dev_net(skb->dev);
1023	const struct ipv6hdr *iph = (const struct ipv6hdr *)skb->data;
1024	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data + offset);
1025	struct xfrm_state *x;
1026
1027	if (type != ICMPV6_PKT_TOOBIG &&
1028	    type != NDISC_REDIRECT)
1029		return 0;
1030
1031	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
1032			      esph->spi, IPPROTO_ESP, AF_INET6);
1033	if (!x)
1034		return 0;
1035
1036	if (type == NDISC_REDIRECT)
1037		ip6_redirect(skb, net, skb->dev->ifindex, 0,
1038			     sock_net_uid(net, NULL));
1039	else
1040		ip6_update_pmtu(skb, net, info, 0, 0, sock_net_uid(net, NULL));
1041	xfrm_state_put(x);
1042
1043	return 0;
1044}
1045
1046static void esp6_destroy(struct xfrm_state *x)
1047{
1048	struct crypto_aead *aead = x->data;
1049
1050	if (!aead)
1051		return;
1052
1053	crypto_free_aead(aead);
1054}
1055
1056static int esp_init_aead(struct xfrm_state *x, struct netlink_ext_ack *extack)
1057{
1058	char aead_name[CRYPTO_MAX_ALG_NAME];
1059	struct crypto_aead *aead;
1060	int err;
1061
1062	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1063		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME) {
1064		NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1065		return -ENAMETOOLONG;
1066	}
1067
1068	aead = crypto_alloc_aead(aead_name, 0, 0);
1069	err = PTR_ERR(aead);
1070	if (IS_ERR(aead))
1071		goto error;
1072
1073	x->data = aead;
1074
1075	err = crypto_aead_setkey(aead, x->aead->alg_key,
1076				 (x->aead->alg_key_len + 7) / 8);
1077	if (err)
1078		goto error;
1079
1080	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1081	if (err)
1082		goto error;
1083
1084	return 0;
1085
1086error:
1087	NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1088	return err;
1089}
1090
1091static int esp_init_authenc(struct xfrm_state *x,
1092			    struct netlink_ext_ack *extack)
1093{
1094	struct crypto_aead *aead;
1095	struct crypto_authenc_key_param *param;
1096	struct rtattr *rta;
1097	char *key;
1098	char *p;
1099	char authenc_name[CRYPTO_MAX_ALG_NAME];
1100	unsigned int keylen;
1101	int err;
1102
1103	err = -ENAMETOOLONG;
1104
1105	if ((x->props.flags & XFRM_STATE_ESN)) {
1106		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1107			     "%s%sauthencesn(%s,%s)%s",
1108			     x->geniv ?: "", x->geniv ? "(" : "",
1109			     x->aalg ? x->aalg->alg_name : "digest_null",
1110			     x->ealg->alg_name,
1111			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1112			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1113			goto error;
1114		}
1115	} else {
1116		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1117			     "%s%sauthenc(%s,%s)%s",
1118			     x->geniv ?: "", x->geniv ? "(" : "",
1119			     x->aalg ? x->aalg->alg_name : "digest_null",
1120			     x->ealg->alg_name,
1121			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1122			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1123			goto error;
1124		}
1125	}
1126
1127	aead = crypto_alloc_aead(authenc_name, 0, 0);
1128	err = PTR_ERR(aead);
1129	if (IS_ERR(aead)) {
1130		NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1131		goto error;
1132	}
1133
1134	x->data = aead;
1135
1136	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1137		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1138	err = -ENOMEM;
1139	key = kmalloc(keylen, GFP_KERNEL);
1140	if (!key)
1141		goto error;
1142
1143	p = key;
1144	rta = (void *)p;
1145	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1146	rta->rta_len = RTA_LENGTH(sizeof(*param));
1147	param = RTA_DATA(rta);
1148	p += RTA_SPACE(sizeof(*param));
1149
1150	if (x->aalg) {
1151		struct xfrm_algo_desc *aalg_desc;
1152
1153		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1154		p += (x->aalg->alg_key_len + 7) / 8;
1155
1156		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1157		BUG_ON(!aalg_desc);
1158
1159		err = -EINVAL;
1160		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1161		    crypto_aead_authsize(aead)) {
1162			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1163			goto free_key;
1164		}
1165
1166		err = crypto_aead_setauthsize(
1167			aead, x->aalg->alg_trunc_len / 8);
1168		if (err) {
1169			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1170			goto free_key;
1171		}
1172	}
1173
1174	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1175	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1176
1177	err = crypto_aead_setkey(aead, key, keylen);
1178
1179free_key:
1180	kfree(key);
1181
1182error:
1183	return err;
1184}
1185
1186static int esp6_init_state(struct xfrm_state *x, struct netlink_ext_ack *extack)
1187{
1188	struct crypto_aead *aead;
1189	u32 align;
1190	int err;
1191
1192	x->data = NULL;
1193
1194	if (x->aead) {
1195		err = esp_init_aead(x, extack);
1196	} else if (x->ealg) {
1197		err = esp_init_authenc(x, extack);
1198	} else {
1199		NL_SET_ERR_MSG(extack, "ESP: AEAD or CRYPT must be provided");
1200		err = -EINVAL;
1201	}
1202
1203	if (err)
1204		goto error;
1205
1206	aead = x->data;
1207
1208	x->props.header_len = sizeof(struct ip_esp_hdr) +
1209			      crypto_aead_ivsize(aead);
1210	switch (x->props.mode) {
1211	case XFRM_MODE_BEET:
1212		if (x->sel.family != AF_INET6)
1213			x->props.header_len += IPV4_BEET_PHMAXLEN +
1214					       (sizeof(struct ipv6hdr) - sizeof(struct iphdr));
1215		break;
1216	default:
1217	case XFRM_MODE_TRANSPORT:
1218		break;
1219	case XFRM_MODE_TUNNEL:
1220		x->props.header_len += sizeof(struct ipv6hdr);
1221		break;
1222	}
1223
1224	if (x->encap) {
1225		struct xfrm_encap_tmpl *encap = x->encap;
1226
1227		switch (encap->encap_type) {
1228		default:
1229			NL_SET_ERR_MSG(extack, "Unsupported encapsulation type for ESP");
1230			err = -EINVAL;
1231			goto error;
1232		case UDP_ENCAP_ESPINUDP:
1233			x->props.header_len += sizeof(struct udphdr);
1234			break;
1235		case UDP_ENCAP_ESPINUDP_NON_IKE:
1236			x->props.header_len += sizeof(struct udphdr) + 2 * sizeof(u32);
1237			break;
1238#ifdef CONFIG_INET6_ESPINTCP
1239		case TCP_ENCAP_ESPINTCP:
1240			/* only the length field, TCP encap is done by
1241			 * the socket
1242			 */
1243			x->props.header_len += 2;
1244			break;
1245#endif
1246		}
1247	}
1248
1249	align = ALIGN(crypto_aead_blocksize(aead), 4);
1250	x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1251
1252error:
1253	return err;
1254}
1255
1256static int esp6_rcv_cb(struct sk_buff *skb, int err)
1257{
1258	return 0;
1259}
1260
1261static const struct xfrm_type esp6_type = {
1262	.owner		= THIS_MODULE,
1263	.proto		= IPPROTO_ESP,
1264	.flags		= XFRM_TYPE_REPLAY_PROT,
1265	.init_state	= esp6_init_state,
1266	.destructor	= esp6_destroy,
1267	.input		= esp6_input,
1268	.output		= esp6_output,
1269};
1270
1271static struct xfrm6_protocol esp6_protocol = {
1272	.handler	=	xfrm6_rcv,
1273	.input_handler	=	xfrm_input,
1274	.cb_handler	=	esp6_rcv_cb,
1275	.err_handler	=	esp6_err,
1276	.priority	=	0,
1277};
1278
1279static int __init esp6_init(void)
1280{
1281	if (xfrm_register_type(&esp6_type, AF_INET6) < 0) {
1282		pr_info("%s: can't add xfrm type\n", __func__);
1283		return -EAGAIN;
1284	}
1285	if (xfrm6_protocol_register(&esp6_protocol, IPPROTO_ESP) < 0) {
1286		pr_info("%s: can't add protocol\n", __func__);
1287		xfrm_unregister_type(&esp6_type, AF_INET6);
1288		return -EAGAIN;
1289	}
1290
1291	return 0;
1292}
1293
1294static void __exit esp6_fini(void)
1295{
1296	if (xfrm6_protocol_deregister(&esp6_protocol, IPPROTO_ESP) < 0)
1297		pr_info("%s: can't remove protocol\n", __func__);
1298	xfrm_unregister_type(&esp6_type, AF_INET6);
1299}
1300
1301module_init(esp6_init);
1302module_exit(esp6_fini);
1303
1304MODULE_DESCRIPTION("IPv6 ESP transformation helpers");
1305MODULE_LICENSE("GPL");
1306MODULE_ALIAS_XFRM_TYPE(AF_INET6, XFRM_PROTO_ESP);
1307