WIP FPC-III support
[linux/fpc-iii.git] / net / ipv6 / esp6.c
blob52c2f063529fbf9e33326eb6d3887f8c810b26eb
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright (C)2002 USAGI/WIDE Project
5 * Authors
7 * Mitsuru KANDA @USAGI : IPv6 Support
8 * Kazunori MIYAZAWA @USAGI :
9 * Kunihiro Ishiguro <kunihiro@ipinfusion.com>
11 * This file is derived from net/ipv4/esp.c
14 #define pr_fmt(fmt) "IPv6: " fmt
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>
40 #include <linux/highmem.h>
42 struct esp_skb_cb {
43 struct xfrm_skb_cb xfrm;
44 void *tmp;
47 struct esp_output_extra {
48 __be32 seqhi;
49 u32 esphoff;
52 #define ESP_SKB_CB(__skb) ((struct esp_skb_cb *)&((__skb)->cb[0]))
55 * Allocate an AEAD request structure with extra space for SG and IV.
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.
61 * TODO: Use spare space in skb for this where possible.
63 static void *esp_alloc_tmp(struct crypto_aead *aead, int nfrags, int seqihlen)
65 unsigned int len;
67 len = seqihlen;
69 len += crypto_aead_ivsize(aead);
71 if (len) {
72 len += crypto_aead_alignmask(aead) &
73 ~(crypto_tfm_ctx_alignment() - 1);
74 len = ALIGN(len, crypto_tfm_ctx_alignment());
77 len += sizeof(struct aead_request) + crypto_aead_reqsize(aead);
78 len = ALIGN(len, __alignof__(struct scatterlist));
80 len += sizeof(struct scatterlist) * nfrags;
82 return kmalloc(len, GFP_ATOMIC);
85 static inline void *esp_tmp_extra(void *tmp)
87 return PTR_ALIGN(tmp, __alignof__(struct esp_output_extra));
90 static inline u8 *esp_tmp_iv(struct crypto_aead *aead, void *tmp, int seqhilen)
92 return crypto_aead_ivsize(aead) ?
93 PTR_ALIGN((u8 *)tmp + seqhilen,
94 crypto_aead_alignmask(aead) + 1) : tmp + seqhilen;
97 static inline struct aead_request *esp_tmp_req(struct crypto_aead *aead, u8 *iv)
99 struct aead_request *req;
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;
107 static inline struct scatterlist *esp_req_sg(struct crypto_aead *aead,
108 struct aead_request *req)
110 return (void *)ALIGN((unsigned long)(req + 1) +
111 crypto_aead_reqsize(aead),
112 __alignof__(struct scatterlist));
115 static void esp_ssg_unref(struct xfrm_state *x, void *tmp)
117 struct esp_output_extra *extra = esp_tmp_extra(tmp);
118 struct crypto_aead *aead = x->data;
119 int extralen = 0;
120 u8 *iv;
121 struct aead_request *req;
122 struct scatterlist *sg;
124 if (x->props.flags & XFRM_STATE_ESN)
125 extralen += sizeof(*extra);
127 iv = esp_tmp_iv(aead, tmp, extralen);
128 req = esp_tmp_req(aead, iv);
130 /* Unref skb_frag_pages in the src scatterlist if necessary.
131 * Skip the first sg which comes from skb->data.
133 if (req->src != req->dst)
134 for (sg = sg_next(req->src); sg; sg = sg_next(sg))
135 put_page(sg_page(sg));
138 #ifdef CONFIG_INET6_ESPINTCP
139 struct esp_tcp_sk {
140 struct sock *sk;
141 struct rcu_head rcu;
144 static void esp_free_tcp_sk(struct rcu_head *head)
146 struct esp_tcp_sk *esk = container_of(head, struct esp_tcp_sk, rcu);
148 sock_put(esk->sk);
149 kfree(esk);
152 static struct sock *esp6_find_tcp_sk(struct xfrm_state *x)
154 struct xfrm_encap_tmpl *encap = x->encap;
155 struct esp_tcp_sk *esk;
156 __be16 sport, dport;
157 struct sock *nsk;
158 struct sock *sk;
160 sk = rcu_dereference(x->encap_sk);
161 if (sk && sk->sk_state == TCP_ESTABLISHED)
162 return sk;
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);
175 RCU_INIT_POINTER(x->encap_sk, NULL);
176 esk->sk = sk;
177 call_rcu(&esk->rcu, esp_free_tcp_sk);
179 spin_unlock_bh(&x->lock);
181 sk = __inet6_lookup_established(xs_net(x), &tcp_hashinfo, &x->id.daddr.in6,
182 dport, &x->props.saddr.in6, ntohs(sport), 0, 0);
183 if (!sk)
184 return ERR_PTR(-ENOENT);
186 if (!tcp_is_ulp_esp(sk)) {
187 sock_put(sk);
188 return ERR_PTR(-EINVAL);
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);
203 spin_unlock_bh(&x->lock);
205 return sk;
208 static int esp_output_tcp_finish(struct xfrm_state *x, struct sk_buff *skb)
210 struct sock *sk;
211 int err;
213 rcu_read_lock();
215 sk = esp6_find_tcp_sk(x);
216 err = PTR_ERR_OR_ZERO(sk);
217 if (err)
218 goto out;
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);
227 out:
228 rcu_read_unlock();
229 return err;
232 static int esp_output_tcp_encap_cb(struct net *net, struct sock *sk,
233 struct sk_buff *skb)
235 struct dst_entry *dst = skb_dst(skb);
236 struct xfrm_state *x = dst->xfrm;
238 return esp_output_tcp_finish(x, skb);
241 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
243 int err;
245 local_bh_disable();
246 err = xfrm_trans_queue_net(xs_net(x), skb, esp_output_tcp_encap_cb);
247 local_bh_enable();
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.
253 return err ?: -EINPROGRESS;
255 #else
256 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
258 kfree_skb(skb);
260 return -EOPNOTSUPP;
262 #endif
264 static void esp_output_encap_csum(struct sk_buff *skb)
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);
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;
281 static void esp_output_done(struct crypto_async_request *base, int err)
283 struct sk_buff *skb = base->data;
284 struct xfrm_offload *xo = xfrm_offload(skb);
285 void *tmp;
286 struct xfrm_state *x;
288 if (xo && (xo->flags & XFRM_DEV_RESUME)) {
289 struct sec_path *sp = skb_sec_path(skb);
291 x = sp->xvec[sp->len - 1];
292 } else {
293 x = skb_dst(skb)->xfrm;
296 tmp = ESP_SKB_CB(skb)->tmp;
297 esp_ssg_unref(x, tmp);
298 kfree(tmp);
300 esp_output_encap_csum(skb);
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;
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, err);
321 /* Move ESP header back into place. */
322 static void esp_restore_header(struct sk_buff *skb, unsigned int offset)
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);
328 esph->seq_no = esph->spi;
329 esph->spi = *seqhi;
332 static void esp_output_restore_header(struct sk_buff *skb)
334 void *tmp = ESP_SKB_CB(skb)->tmp;
335 struct esp_output_extra *extra = esp_tmp_extra(tmp);
337 esp_restore_header(skb, skb_transport_offset(skb) + extra->esphoff -
338 sizeof(__be32));
341 static 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)
346 /* For ESN we move the header forward by 4 bytes to
347 * accomodate the high bits. We will move it back after
348 * encryption.
350 if ((x->props.flags & XFRM_STATE_ESN)) {
351 __u32 seqhi;
352 struct xfrm_offload *xo = xfrm_offload(skb);
354 if (xo)
355 seqhi = xo->seq.hi;
356 else
357 seqhi = XFRM_SKB_CB(skb)->seq.output.hi;
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);
366 esph->spi = x->id.spi;
368 return esph;
371 static void esp_output_done_esn(struct crypto_async_request *base, int err)
373 struct sk_buff *skb = base->data;
375 esp_output_restore_header(skb);
376 esp_output_done(base, err);
379 static 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)
385 struct udphdr *uh;
386 __be32 *udpdata32;
387 unsigned int len;
389 len = skb->len + esp->tailen - skb_transport_offset(skb);
390 if (len > U16_MAX)
391 return ERR_PTR(-EMSGSIZE);
393 uh = (struct udphdr *)esp->esph;
394 uh->source = sport;
395 uh->dest = dport;
396 uh->len = htons(len);
397 uh->check = 0;
399 *skb_mac_header(skb) = IPPROTO_UDP;
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);
407 return (struct ip_esp_hdr *)(uh + 1);
410 #ifdef CONFIG_INET6_ESPINTCP
411 static struct ip_esp_hdr *esp6_output_tcp_encap(struct xfrm_state *x,
412 struct sk_buff *skb,
413 struct esp_info *esp)
415 __be16 *lenp = (void *)esp->esph;
416 struct ip_esp_hdr *esph;
417 unsigned int len;
418 struct sock *sk;
420 len = skb->len + esp->tailen - skb_transport_offset(skb);
421 if (len > IP_MAX_MTU)
422 return ERR_PTR(-EMSGSIZE);
424 rcu_read_lock();
425 sk = esp6_find_tcp_sk(x);
426 rcu_read_unlock();
428 if (IS_ERR(sk))
429 return ERR_CAST(sk);
431 *lenp = htons(len);
432 esph = (struct ip_esp_hdr *)(lenp + 1);
434 return esph;
436 #else
437 static struct ip_esp_hdr *esp6_output_tcp_encap(struct xfrm_state *x,
438 struct sk_buff *skb,
439 struct esp_info *esp)
441 return ERR_PTR(-EOPNOTSUPP);
443 #endif
445 static int esp6_output_encap(struct xfrm_state *x, struct sk_buff *skb,
446 struct esp_info *esp)
448 struct xfrm_encap_tmpl *encap = x->encap;
449 struct ip_esp_hdr *esph;
450 __be16 sport, dport;
451 int encap_type;
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);
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;
470 if (IS_ERR(esph))
471 return PTR_ERR(esph);
473 esp->esph = esph;
475 return 0;
478 int esp6_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
480 u8 *tail;
481 u8 *vaddr;
482 int nfrags;
483 int esph_offset;
484 struct page *page;
485 struct sk_buff *trailer;
486 int tailen = esp->tailen;
488 if (x->encap) {
489 int err = esp6_output_encap(x, skb, esp);
491 if (err < 0)
492 return err;
495 if (!skb_cloned(skb)) {
496 if (tailen <= skb_tailroom(skb)) {
497 nfrags = 1;
498 trailer = skb;
499 tail = skb_tail_pointer(trailer);
501 goto skip_cow;
502 } else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
503 && !skb_has_frag_list(skb)) {
504 int allocsize;
505 struct sock *sk = skb->sk;
506 struct page_frag *pfrag = &x->xfrag;
508 esp->inplace = false;
510 allocsize = ALIGN(tailen, L1_CACHE_BYTES);
512 spin_lock_bh(&x->lock);
514 if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
515 spin_unlock_bh(&x->lock);
516 goto cow;
519 page = pfrag->page;
520 get_page(page);
522 vaddr = kmap_atomic(page);
524 tail = vaddr + pfrag->offset;
526 esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
528 kunmap_atomic(vaddr);
530 nfrags = skb_shinfo(skb)->nr_frags;
532 __skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
533 tailen);
534 skb_shinfo(skb)->nr_frags = ++nfrags;
536 pfrag->offset = pfrag->offset + allocsize;
538 spin_unlock_bh(&x->lock);
540 nfrags++;
542 skb->len += tailen;
543 skb->data_len += tailen;
544 skb->truesize += tailen;
545 if (sk && sk_fullsock(sk))
546 refcount_add(tailen, &sk->sk_wmem_alloc);
548 goto out;
552 cow:
553 esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
555 nfrags = skb_cow_data(skb, tailen, &trailer);
556 if (nfrags < 0)
557 goto out;
558 tail = skb_tail_pointer(trailer);
559 esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
561 skip_cow:
562 esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
563 pskb_put(skb, trailer, tailen);
565 out:
566 return nfrags;
568 EXPORT_SYMBOL_GPL(esp6_output_head);
570 int esp6_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
572 u8 *iv;
573 int alen;
574 void *tmp;
575 int ivlen;
576 int assoclen;
577 int extralen;
578 struct page *page;
579 struct ip_esp_hdr *esph;
580 struct aead_request *req;
581 struct crypto_aead *aead;
582 struct scatterlist *sg, *dsg;
583 struct esp_output_extra *extra;
584 int err = -ENOMEM;
586 assoclen = sizeof(struct ip_esp_hdr);
587 extralen = 0;
589 if (x->props.flags & XFRM_STATE_ESN) {
590 extralen += sizeof(*extra);
591 assoclen += sizeof(__be32);
594 aead = x->data;
595 alen = crypto_aead_authsize(aead);
596 ivlen = crypto_aead_ivsize(aead);
598 tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
599 if (!tmp)
600 goto error;
602 extra = esp_tmp_extra(tmp);
603 iv = esp_tmp_iv(aead, tmp, extralen);
604 req = esp_tmp_req(aead, iv);
605 sg = esp_req_sg(aead, req);
607 if (esp->inplace)
608 dsg = sg;
609 else
610 dsg = &sg[esp->nfrags];
612 esph = esp_output_set_esn(skb, x, esp->esph, extra);
613 esp->esph = esph;
615 sg_init_table(sg, esp->nfrags);
616 err = skb_to_sgvec(skb, sg,
617 (unsigned char *)esph - skb->data,
618 assoclen + ivlen + esp->clen + alen);
619 if (unlikely(err < 0))
620 goto error_free;
622 if (!esp->inplace) {
623 int allocsize;
624 struct page_frag *pfrag = &x->xfrag;
626 allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
628 spin_lock_bh(&x->lock);
629 if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
630 spin_unlock_bh(&x->lock);
631 goto error_free;
634 skb_shinfo(skb)->nr_frags = 1;
636 page = pfrag->page;
637 get_page(page);
638 /* replace page frags in skb with new page */
639 __skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
640 pfrag->offset = pfrag->offset + allocsize;
641 spin_unlock_bh(&x->lock);
643 sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
644 err = skb_to_sgvec(skb, dsg,
645 (unsigned char *)esph - skb->data,
646 assoclen + ivlen + esp->clen + alen);
647 if (unlikely(err < 0))
648 goto error_free;
651 if ((x->props.flags & XFRM_STATE_ESN))
652 aead_request_set_callback(req, 0, esp_output_done_esn, skb);
653 else
654 aead_request_set_callback(req, 0, esp_output_done, skb);
656 aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
657 aead_request_set_ad(req, assoclen);
659 memset(iv, 0, ivlen);
660 memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
661 min(ivlen, 8));
663 ESP_SKB_CB(skb)->tmp = tmp;
664 err = crypto_aead_encrypt(req);
666 switch (err) {
667 case -EINPROGRESS:
668 goto error;
670 case -ENOSPC:
671 err = NET_XMIT_DROP;
672 break;
674 case 0:
675 if ((x->props.flags & XFRM_STATE_ESN))
676 esp_output_restore_header(skb);
677 esp_output_encap_csum(skb);
680 if (sg != dsg)
681 esp_ssg_unref(x, tmp);
683 if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
684 err = esp_output_tail_tcp(x, skb);
686 error_free:
687 kfree(tmp);
688 error:
689 return err;
691 EXPORT_SYMBOL_GPL(esp6_output_tail);
693 static int esp6_output(struct xfrm_state *x, struct sk_buff *skb)
695 int alen;
696 int blksize;
697 struct ip_esp_hdr *esph;
698 struct crypto_aead *aead;
699 struct esp_info esp;
701 esp.inplace = true;
703 esp.proto = *skb_mac_header(skb);
704 *skb_mac_header(skb) = IPPROTO_ESP;
706 /* skb is pure payload to encrypt */
708 aead = x->data;
709 alen = crypto_aead_authsize(aead);
711 esp.tfclen = 0;
712 if (x->tfcpad) {
713 struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
714 u32 padto;
716 padto = min(x->tfcpad, xfrm_state_mtu(x, dst->child_mtu_cached));
717 if (skb->len < padto)
718 esp.tfclen = padto - skb->len;
720 blksize = ALIGN(crypto_aead_blocksize(aead), 4);
721 esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
722 esp.plen = esp.clen - skb->len - esp.tfclen;
723 esp.tailen = esp.tfclen + esp.plen + alen;
725 esp.esph = ip_esp_hdr(skb);
727 esp.nfrags = esp6_output_head(x, skb, &esp);
728 if (esp.nfrags < 0)
729 return esp.nfrags;
731 esph = esp.esph;
732 esph->spi = x->id.spi;
734 esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
735 esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
736 ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
738 skb_push(skb, -skb_network_offset(skb));
740 return esp6_output_tail(x, skb, &esp);
743 static inline int esp_remove_trailer(struct sk_buff *skb)
745 struct xfrm_state *x = xfrm_input_state(skb);
746 struct xfrm_offload *xo = xfrm_offload(skb);
747 struct crypto_aead *aead = x->data;
748 int alen, hlen, elen;
749 int padlen, trimlen;
750 __wsum csumdiff;
751 u8 nexthdr[2];
752 int ret;
754 alen = crypto_aead_authsize(aead);
755 hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
756 elen = skb->len - hlen;
758 if (xo && (xo->flags & XFRM_ESP_NO_TRAILER)) {
759 ret = xo->proto;
760 goto out;
763 ret = skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2);
764 BUG_ON(ret);
766 ret = -EINVAL;
767 padlen = nexthdr[0];
768 if (padlen + 2 + alen >= elen) {
769 net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
770 padlen + 2, elen - alen);
771 goto out;
774 trimlen = alen + padlen + 2;
775 if (skb->ip_summed == CHECKSUM_COMPLETE) {
776 csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
777 skb->csum = csum_block_sub(skb->csum, csumdiff,
778 skb->len - trimlen);
780 pskb_trim(skb, skb->len - trimlen);
782 ret = nexthdr[1];
784 out:
785 return ret;
788 int esp6_input_done2(struct sk_buff *skb, int err)
790 struct xfrm_state *x = xfrm_input_state(skb);
791 struct xfrm_offload *xo = xfrm_offload(skb);
792 struct crypto_aead *aead = x->data;
793 int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
794 int hdr_len = skb_network_header_len(skb);
796 if (!xo || (xo && !(xo->flags & CRYPTO_DONE)))
797 kfree(ESP_SKB_CB(skb)->tmp);
799 if (unlikely(err))
800 goto out;
802 err = esp_remove_trailer(skb);
803 if (unlikely(err < 0))
804 goto out;
806 if (x->encap) {
807 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
808 int offset = skb_network_offset(skb) + sizeof(*ip6h);
809 struct xfrm_encap_tmpl *encap = x->encap;
810 u8 nexthdr = ip6h->nexthdr;
811 __be16 frag_off, source;
812 struct udphdr *uh;
813 struct tcphdr *th;
815 offset = ipv6_skip_exthdr(skb, offset, &nexthdr, &frag_off);
816 uh = (void *)(skb->data + offset);
817 th = (void *)(skb->data + offset);
818 hdr_len += offset;
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;
835 * 1) if the NAT-T peer's IP or port changed then
836 * advertize the change to the keying daemon.
837 * This is an inbound SA, so just compare
838 * SRC ports.
840 if (!ipv6_addr_equal(&ip6h->saddr, &x->props.saddr.in6) ||
841 source != encap->encap_sport) {
842 xfrm_address_t ipaddr;
844 memcpy(&ipaddr.a6, &ip6h->saddr.s6_addr, sizeof(ipaddr.a6));
845 km_new_mapping(x, &ipaddr, source);
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.
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
863 if (x->props.mode == XFRM_MODE_TRANSPORT)
864 skb->ip_summed = CHECKSUM_UNNECESSARY;
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);
875 /* RFC4303: Drop dummy packets without any error */
876 if (err == IPPROTO_NONE)
877 err = -EINVAL;
879 out:
880 return err;
882 EXPORT_SYMBOL_GPL(esp6_input_done2);
884 static void esp_input_done(struct crypto_async_request *base, int err)
886 struct sk_buff *skb = base->data;
888 xfrm_input_resume(skb, esp6_input_done2(skb, err));
891 static void esp_input_restore_header(struct sk_buff *skb)
893 esp_restore_header(skb, 0);
894 __skb_pull(skb, 4);
897 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
899 struct xfrm_state *x = xfrm_input_state(skb);
901 /* For ESN we move the header forward by 4 bytes to
902 * accomodate the high bits. We will move it back after
903 * decryption.
905 if ((x->props.flags & XFRM_STATE_ESN)) {
906 struct ip_esp_hdr *esph = skb_push(skb, 4);
908 *seqhi = esph->spi;
909 esph->spi = esph->seq_no;
910 esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
914 static void esp_input_done_esn(struct crypto_async_request *base, int err)
916 struct sk_buff *skb = base->data;
918 esp_input_restore_header(skb);
919 esp_input_done(base, err);
922 static int esp6_input(struct xfrm_state *x, struct sk_buff *skb)
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;
938 if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen)) {
939 ret = -EINVAL;
940 goto out;
943 if (elen <= 0) {
944 ret = -EINVAL;
945 goto out;
948 assoclen = sizeof(struct ip_esp_hdr);
949 seqhilen = 0;
951 if (x->props.flags & XFRM_STATE_ESN) {
952 seqhilen += sizeof(__be32);
953 assoclen += seqhilen;
956 if (!skb_cloned(skb)) {
957 if (!skb_is_nonlinear(skb)) {
958 nfrags = 1;
960 goto skip_cow;
961 } else if (!skb_has_frag_list(skb)) {
962 nfrags = skb_shinfo(skb)->nr_frags;
963 nfrags++;
965 goto skip_cow;
969 nfrags = skb_cow_data(skb, 0, &trailer);
970 if (nfrags < 0) {
971 ret = -EINVAL;
972 goto out;
975 skip_cow:
976 ret = -ENOMEM;
977 tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
978 if (!tmp)
979 goto out;
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);
987 esp_input_set_header(skb, seqhi);
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;
996 skb->ip_summed = CHECKSUM_NONE;
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);
1003 aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
1004 aead_request_set_ad(req, assoclen);
1006 ret = crypto_aead_decrypt(req);
1007 if (ret == -EINPROGRESS)
1008 goto out;
1010 if ((x->props.flags & XFRM_STATE_ESN))
1011 esp_input_restore_header(skb);
1013 ret = esp6_input_done2(skb, ret);
1015 out:
1016 return ret;
1019 static int esp6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
1020 u8 type, u8 code, int offset, __be32 info)
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;
1027 if (type != ICMPV6_PKT_TOOBIG &&
1028 type != NDISC_REDIRECT)
1029 return 0;
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;
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);
1043 return 0;
1046 static void esp6_destroy(struct xfrm_state *x)
1048 struct crypto_aead *aead = x->data;
1050 if (!aead)
1051 return;
1053 crypto_free_aead(aead);
1056 static int esp_init_aead(struct xfrm_state *x)
1058 char aead_name[CRYPTO_MAX_ALG_NAME];
1059 struct crypto_aead *aead;
1060 int err;
1062 err = -ENAMETOOLONG;
1063 if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1064 x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME)
1065 goto error;
1067 aead = crypto_alloc_aead(aead_name, 0, 0);
1068 err = PTR_ERR(aead);
1069 if (IS_ERR(aead))
1070 goto error;
1072 x->data = aead;
1074 err = crypto_aead_setkey(aead, x->aead->alg_key,
1075 (x->aead->alg_key_len + 7) / 8);
1076 if (err)
1077 goto error;
1079 err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1080 if (err)
1081 goto error;
1083 error:
1084 return err;
1087 static int esp_init_authenc(struct xfrm_state *x)
1089 struct crypto_aead *aead;
1090 struct crypto_authenc_key_param *param;
1091 struct rtattr *rta;
1092 char *key;
1093 char *p;
1094 char authenc_name[CRYPTO_MAX_ALG_NAME];
1095 unsigned int keylen;
1096 int err;
1098 err = -EINVAL;
1099 if (!x->ealg)
1100 goto error;
1102 err = -ENAMETOOLONG;
1104 if ((x->props.flags & XFRM_STATE_ESN)) {
1105 if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1106 "%s%sauthencesn(%s,%s)%s",
1107 x->geniv ?: "", x->geniv ? "(" : "",
1108 x->aalg ? x->aalg->alg_name : "digest_null",
1109 x->ealg->alg_name,
1110 x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1111 goto error;
1112 } else {
1113 if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1114 "%s%sauthenc(%s,%s)%s",
1115 x->geniv ?: "", x->geniv ? "(" : "",
1116 x->aalg ? x->aalg->alg_name : "digest_null",
1117 x->ealg->alg_name,
1118 x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1119 goto error;
1122 aead = crypto_alloc_aead(authenc_name, 0, 0);
1123 err = PTR_ERR(aead);
1124 if (IS_ERR(aead))
1125 goto error;
1127 x->data = aead;
1129 keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1130 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1131 err = -ENOMEM;
1132 key = kmalloc(keylen, GFP_KERNEL);
1133 if (!key)
1134 goto error;
1136 p = key;
1137 rta = (void *)p;
1138 rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1139 rta->rta_len = RTA_LENGTH(sizeof(*param));
1140 param = RTA_DATA(rta);
1141 p += RTA_SPACE(sizeof(*param));
1143 if (x->aalg) {
1144 struct xfrm_algo_desc *aalg_desc;
1146 memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1147 p += (x->aalg->alg_key_len + 7) / 8;
1149 aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1150 BUG_ON(!aalg_desc);
1152 err = -EINVAL;
1153 if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1154 crypto_aead_authsize(aead)) {
1155 pr_info("ESP: %s digestsize %u != %hu\n",
1156 x->aalg->alg_name,
1157 crypto_aead_authsize(aead),
1158 aalg_desc->uinfo.auth.icv_fullbits / 8);
1159 goto free_key;
1162 err = crypto_aead_setauthsize(
1163 aead, x->aalg->alg_trunc_len / 8);
1164 if (err)
1165 goto free_key;
1168 param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1169 memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1171 err = crypto_aead_setkey(aead, key, keylen);
1173 free_key:
1174 kfree(key);
1176 error:
1177 return err;
1180 static int esp6_init_state(struct xfrm_state *x)
1182 struct crypto_aead *aead;
1183 u32 align;
1184 int err;
1186 x->data = NULL;
1188 if (x->aead)
1189 err = esp_init_aead(x);
1190 else
1191 err = esp_init_authenc(x);
1193 if (err)
1194 goto error;
1196 aead = x->data;
1198 x->props.header_len = sizeof(struct ip_esp_hdr) +
1199 crypto_aead_ivsize(aead);
1200 switch (x->props.mode) {
1201 case XFRM_MODE_BEET:
1202 if (x->sel.family != AF_INET6)
1203 x->props.header_len += IPV4_BEET_PHMAXLEN +
1204 (sizeof(struct ipv6hdr) - sizeof(struct iphdr));
1205 break;
1206 default:
1207 case XFRM_MODE_TRANSPORT:
1208 break;
1209 case XFRM_MODE_TUNNEL:
1210 x->props.header_len += sizeof(struct ipv6hdr);
1211 break;
1214 if (x->encap) {
1215 struct xfrm_encap_tmpl *encap = x->encap;
1217 switch (encap->encap_type) {
1218 default:
1219 err = -EINVAL;
1220 goto error;
1221 case UDP_ENCAP_ESPINUDP:
1222 x->props.header_len += sizeof(struct udphdr);
1223 break;
1224 case UDP_ENCAP_ESPINUDP_NON_IKE:
1225 x->props.header_len += sizeof(struct udphdr) + 2 * sizeof(u32);
1226 break;
1227 #ifdef CONFIG_INET6_ESPINTCP
1228 case TCP_ENCAP_ESPINTCP:
1229 /* only the length field, TCP encap is done by
1230 * the socket
1232 x->props.header_len += 2;
1233 break;
1234 #endif
1238 align = ALIGN(crypto_aead_blocksize(aead), 4);
1239 x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1241 error:
1242 return err;
1245 static int esp6_rcv_cb(struct sk_buff *skb, int err)
1247 return 0;
1250 static const struct xfrm_type esp6_type = {
1251 .description = "ESP6",
1252 .owner = THIS_MODULE,
1253 .proto = IPPROTO_ESP,
1254 .flags = XFRM_TYPE_REPLAY_PROT,
1255 .init_state = esp6_init_state,
1256 .destructor = esp6_destroy,
1257 .input = esp6_input,
1258 .output = esp6_output,
1259 .hdr_offset = xfrm6_find_1stfragopt,
1262 static struct xfrm6_protocol esp6_protocol = {
1263 .handler = xfrm6_rcv,
1264 .input_handler = xfrm_input,
1265 .cb_handler = esp6_rcv_cb,
1266 .err_handler = esp6_err,
1267 .priority = 0,
1270 static int __init esp6_init(void)
1272 if (xfrm_register_type(&esp6_type, AF_INET6) < 0) {
1273 pr_info("%s: can't add xfrm type\n", __func__);
1274 return -EAGAIN;
1276 if (xfrm6_protocol_register(&esp6_protocol, IPPROTO_ESP) < 0) {
1277 pr_info("%s: can't add protocol\n", __func__);
1278 xfrm_unregister_type(&esp6_type, AF_INET6);
1279 return -EAGAIN;
1282 return 0;
1285 static void __exit esp6_fini(void)
1287 if (xfrm6_protocol_deregister(&esp6_protocol, IPPROTO_ESP) < 0)
1288 pr_info("%s: can't remove protocol\n", __func__);
1289 xfrm_unregister_type(&esp6_type, AF_INET6);
1292 module_init(esp6_init);
1293 module_exit(esp6_fini);
1295 MODULE_LICENSE("GPL");
1296 MODULE_ALIAS_XFRM_TYPE(AF_INET6, XFRM_PROTO_ESP);