Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2#define pr_fmt(fmt) "IPsec: " fmt
3
4#include <crypto/aead.h>
5#include <crypto/authenc.h>
6#include <linux/err.h>
7#include <linux/module.h>
8#include <net/ip.h>
9#include <net/xfrm.h>
10#include <net/esp.h>
11#include <linux/scatterlist.h>
12#include <linux/kernel.h>
13#include <linux/pfkeyv2.h>
14#include <linux/rtnetlink.h>
15#include <linux/slab.h>
16#include <linux/spinlock.h>
17#include <linux/in6.h>
18#include <net/icmp.h>
19#include <net/protocol.h>
20#include <net/udp.h>
21#include <net/tcp.h>
22#include <net/espintcp.h>
23#include <linux/skbuff_ref.h>
24
25#include <linux/highmem.h>
26
27struct esp_skb_cb {
28 struct xfrm_skb_cb xfrm;
29 void *tmp;
30};
31
32struct esp_output_extra {
33 __be32 seqhi;
34 u32 esphoff;
35};
36
37#define ESP_SKB_CB(__skb) ((struct esp_skb_cb *)&((__skb)->cb[0]))
38
39/*
40 * Allocate an AEAD request structure with extra space for SG and IV.
41 *
42 * For alignment considerations the IV is placed at the front, followed
43 * by the request and finally the SG list.
44 *
45 * TODO: Use spare space in skb for this where possible.
46 */
47static void *esp_alloc_tmp(struct crypto_aead *aead, int nfrags, int extralen)
48{
49 unsigned int len;
50
51 len = extralen;
52
53 len += crypto_aead_ivsize(aead);
54
55 if (len) {
56 len += crypto_aead_alignmask(aead) &
57 ~(crypto_tfm_ctx_alignment() - 1);
58 len = ALIGN(len, crypto_tfm_ctx_alignment());
59 }
60
61 len += sizeof(struct aead_request) + crypto_aead_reqsize(aead);
62 len = ALIGN(len, __alignof__(struct scatterlist));
63
64 len += sizeof(struct scatterlist) * nfrags;
65
66 return kmalloc(len, GFP_ATOMIC);
67}
68
69static inline void *esp_tmp_extra(void *tmp)
70{
71 return PTR_ALIGN(tmp, __alignof__(struct esp_output_extra));
72}
73
74static inline u8 *esp_tmp_iv(struct crypto_aead *aead, void *tmp, int extralen)
75{
76 return crypto_aead_ivsize(aead) ?
77 PTR_ALIGN((u8 *)tmp + extralen,
78 crypto_aead_alignmask(aead) + 1) : tmp + extralen;
79}
80
81static inline struct aead_request *esp_tmp_req(struct crypto_aead *aead, u8 *iv)
82{
83 struct aead_request *req;
84
85 req = (void *)PTR_ALIGN(iv + crypto_aead_ivsize(aead),
86 crypto_tfm_ctx_alignment());
87 aead_request_set_tfm(req, aead);
88 return req;
89}
90
91static inline struct scatterlist *esp_req_sg(struct crypto_aead *aead,
92 struct aead_request *req)
93{
94 return (void *)ALIGN((unsigned long)(req + 1) +
95 crypto_aead_reqsize(aead),
96 __alignof__(struct scatterlist));
97}
98
99static void esp_ssg_unref(struct xfrm_state *x, void *tmp, struct sk_buff *skb)
100{
101 struct crypto_aead *aead = x->data;
102 int extralen = 0;
103 u8 *iv;
104 struct aead_request *req;
105 struct scatterlist *sg;
106
107 if (x->props.flags & XFRM_STATE_ESN)
108 extralen += sizeof(struct esp_output_extra);
109
110 iv = esp_tmp_iv(aead, tmp, extralen);
111 req = esp_tmp_req(aead, iv);
112
113 /* Unref skb_frag_pages in the src scatterlist if necessary.
114 * Skip the first sg which comes from skb->data.
115 */
116 if (req->src != req->dst)
117 for (sg = sg_next(req->src); sg; sg = sg_next(sg))
118 skb_page_unref(page_to_netmem(sg_page(sg)),
119 skb->pp_recycle);
120}
121
122#ifdef CONFIG_INET_ESPINTCP
123struct esp_tcp_sk {
124 struct sock *sk;
125 struct rcu_head rcu;
126};
127
128static void esp_free_tcp_sk(struct rcu_head *head)
129{
130 struct esp_tcp_sk *esk = container_of(head, struct esp_tcp_sk, rcu);
131
132 sock_put(esk->sk);
133 kfree(esk);
134}
135
136static struct sock *esp_find_tcp_sk(struct xfrm_state *x)
137{
138 struct xfrm_encap_tmpl *encap = x->encap;
139 struct net *net = xs_net(x);
140 struct esp_tcp_sk *esk;
141 __be16 sport, dport;
142 struct sock *nsk;
143 struct sock *sk;
144
145 sk = rcu_dereference(x->encap_sk);
146 if (sk && sk->sk_state == TCP_ESTABLISHED)
147 return sk;
148
149 spin_lock_bh(&x->lock);
150 sport = encap->encap_sport;
151 dport = encap->encap_dport;
152 nsk = rcu_dereference_protected(x->encap_sk,
153 lockdep_is_held(&x->lock));
154 if (sk && sk == nsk) {
155 esk = kmalloc(sizeof(*esk), GFP_ATOMIC);
156 if (!esk) {
157 spin_unlock_bh(&x->lock);
158 return ERR_PTR(-ENOMEM);
159 }
160 RCU_INIT_POINTER(x->encap_sk, NULL);
161 esk->sk = sk;
162 call_rcu(&esk->rcu, esp_free_tcp_sk);
163 }
164 spin_unlock_bh(&x->lock);
165
166 sk = inet_lookup_established(net, net->ipv4.tcp_death_row.hashinfo, x->id.daddr.a4,
167 dport, x->props.saddr.a4, sport, 0);
168 if (!sk)
169 return ERR_PTR(-ENOENT);
170
171 if (!tcp_is_ulp_esp(sk)) {
172 sock_put(sk);
173 return ERR_PTR(-EINVAL);
174 }
175
176 spin_lock_bh(&x->lock);
177 nsk = rcu_dereference_protected(x->encap_sk,
178 lockdep_is_held(&x->lock));
179 if (encap->encap_sport != sport ||
180 encap->encap_dport != dport) {
181 sock_put(sk);
182 sk = nsk ?: ERR_PTR(-EREMCHG);
183 } else if (sk == nsk) {
184 sock_put(sk);
185 } else {
186 rcu_assign_pointer(x->encap_sk, sk);
187 }
188 spin_unlock_bh(&x->lock);
189
190 return sk;
191}
192
193static int esp_output_tcp_finish(struct xfrm_state *x, struct sk_buff *skb)
194{
195 struct sock *sk;
196 int err;
197
198 rcu_read_lock();
199
200 sk = esp_find_tcp_sk(x);
201 err = PTR_ERR_OR_ZERO(sk);
202 if (err)
203 goto out;
204
205 bh_lock_sock(sk);
206 if (sock_owned_by_user(sk))
207 err = espintcp_queue_out(sk, skb);
208 else
209 err = espintcp_push_skb(sk, skb);
210 bh_unlock_sock(sk);
211
212out:
213 rcu_read_unlock();
214 return err;
215}
216
217static int esp_output_tcp_encap_cb(struct net *net, struct sock *sk,
218 struct sk_buff *skb)
219{
220 struct dst_entry *dst = skb_dst(skb);
221 struct xfrm_state *x = dst->xfrm;
222
223 return esp_output_tcp_finish(x, skb);
224}
225
226static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
227{
228 int err;
229
230 local_bh_disable();
231 err = xfrm_trans_queue_net(xs_net(x), skb, esp_output_tcp_encap_cb);
232 local_bh_enable();
233
234 /* EINPROGRESS just happens to do the right thing. It
235 * actually means that the skb has been consumed and
236 * isn't coming back.
237 */
238 return err ?: -EINPROGRESS;
239}
240#else
241static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
242{
243 WARN_ON(1);
244 return -EOPNOTSUPP;
245}
246#endif
247
248static void esp_output_done(void *data, int err)
249{
250 struct sk_buff *skb = data;
251 struct xfrm_offload *xo = xfrm_offload(skb);
252 void *tmp;
253 struct xfrm_state *x;
254
255 if (xo && (xo->flags & XFRM_DEV_RESUME)) {
256 struct sec_path *sp = skb_sec_path(skb);
257
258 x = sp->xvec[sp->len - 1];
259 } else {
260 x = skb_dst(skb)->xfrm;
261 }
262
263 tmp = ESP_SKB_CB(skb)->tmp;
264 esp_ssg_unref(x, tmp, skb);
265 kfree(tmp);
266
267 if (xo && (xo->flags & XFRM_DEV_RESUME)) {
268 if (err) {
269 XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
270 kfree_skb(skb);
271 return;
272 }
273
274 skb_push(skb, skb->data - skb_mac_header(skb));
275 secpath_reset(skb);
276 xfrm_dev_resume(skb);
277 } else {
278 if (!err &&
279 x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
280 esp_output_tail_tcp(x, skb);
281 else
282 xfrm_output_resume(skb->sk, skb, err);
283 }
284}
285
286/* Move ESP header back into place. */
287static void esp_restore_header(struct sk_buff *skb, unsigned int offset)
288{
289 struct ip_esp_hdr *esph = (void *)(skb->data + offset);
290 void *tmp = ESP_SKB_CB(skb)->tmp;
291 __be32 *seqhi = esp_tmp_extra(tmp);
292
293 esph->seq_no = esph->spi;
294 esph->spi = *seqhi;
295}
296
297static void esp_output_restore_header(struct sk_buff *skb)
298{
299 void *tmp = ESP_SKB_CB(skb)->tmp;
300 struct esp_output_extra *extra = esp_tmp_extra(tmp);
301
302 esp_restore_header(skb, skb_transport_offset(skb) + extra->esphoff -
303 sizeof(__be32));
304}
305
306static struct ip_esp_hdr *esp_output_set_extra(struct sk_buff *skb,
307 struct xfrm_state *x,
308 struct ip_esp_hdr *esph,
309 struct esp_output_extra *extra)
310{
311 /* For ESN we move the header forward by 4 bytes to
312 * accommodate the high bits. We will move it back after
313 * encryption.
314 */
315 if ((x->props.flags & XFRM_STATE_ESN)) {
316 __u32 seqhi;
317 struct xfrm_offload *xo = xfrm_offload(skb);
318
319 if (xo)
320 seqhi = xo->seq.hi;
321 else
322 seqhi = XFRM_SKB_CB(skb)->seq.output.hi;
323
324 extra->esphoff = (unsigned char *)esph -
325 skb_transport_header(skb);
326 esph = (struct ip_esp_hdr *)((unsigned char *)esph - 4);
327 extra->seqhi = esph->spi;
328 esph->seq_no = htonl(seqhi);
329 }
330
331 esph->spi = x->id.spi;
332
333 return esph;
334}
335
336static void esp_output_done_esn(void *data, int err)
337{
338 struct sk_buff *skb = data;
339
340 esp_output_restore_header(skb);
341 esp_output_done(data, err);
342}
343
344static struct ip_esp_hdr *esp_output_udp_encap(struct sk_buff *skb,
345 int encap_type,
346 struct esp_info *esp,
347 __be16 sport,
348 __be16 dport)
349{
350 struct udphdr *uh;
351 unsigned int len;
352 struct xfrm_offload *xo = xfrm_offload(skb);
353
354 len = skb->len + esp->tailen - skb_transport_offset(skb);
355 if (len + sizeof(struct iphdr) > IP_MAX_MTU)
356 return ERR_PTR(-EMSGSIZE);
357
358 uh = (struct udphdr *)esp->esph;
359 uh->source = sport;
360 uh->dest = dport;
361 uh->len = htons(len);
362 uh->check = 0;
363
364 /* For IPv4 ESP with UDP encapsulation, if xo is not null, the skb is in the crypto offload
365 * data path, which means that esp_output_udp_encap is called outside of the XFRM stack.
366 * In this case, the mac header doesn't point to the IPv4 protocol field, so don't set it.
367 */
368 if (!xo || encap_type != UDP_ENCAP_ESPINUDP)
369 *skb_mac_header(skb) = IPPROTO_UDP;
370
371 return (struct ip_esp_hdr *)(uh + 1);
372}
373
374#ifdef CONFIG_INET_ESPINTCP
375static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
376 struct sk_buff *skb,
377 struct esp_info *esp)
378{
379 __be16 *lenp = (void *)esp->esph;
380 struct ip_esp_hdr *esph;
381 unsigned int len;
382 struct sock *sk;
383
384 len = skb->len + esp->tailen - skb_transport_offset(skb);
385 if (len > IP_MAX_MTU)
386 return ERR_PTR(-EMSGSIZE);
387
388 rcu_read_lock();
389 sk = esp_find_tcp_sk(x);
390 rcu_read_unlock();
391
392 if (IS_ERR(sk))
393 return ERR_CAST(sk);
394
395 *lenp = htons(len);
396 esph = (struct ip_esp_hdr *)(lenp + 1);
397
398 return esph;
399}
400#else
401static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
402 struct sk_buff *skb,
403 struct esp_info *esp)
404{
405 return ERR_PTR(-EOPNOTSUPP);
406}
407#endif
408
409static int esp_output_encap(struct xfrm_state *x, struct sk_buff *skb,
410 struct esp_info *esp)
411{
412 struct xfrm_encap_tmpl *encap = x->encap;
413 struct ip_esp_hdr *esph;
414 __be16 sport, dport;
415 int encap_type;
416
417 spin_lock_bh(&x->lock);
418 sport = encap->encap_sport;
419 dport = encap->encap_dport;
420 encap_type = encap->encap_type;
421 spin_unlock_bh(&x->lock);
422
423 switch (encap_type) {
424 default:
425 case UDP_ENCAP_ESPINUDP:
426 esph = esp_output_udp_encap(skb, encap_type, esp, sport, dport);
427 break;
428 case TCP_ENCAP_ESPINTCP:
429 esph = esp_output_tcp_encap(x, skb, esp);
430 break;
431 }
432
433 if (IS_ERR(esph))
434 return PTR_ERR(esph);
435
436 esp->esph = esph;
437
438 return 0;
439}
440
441int esp_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
442{
443 u8 *tail;
444 int nfrags;
445 int esph_offset;
446 struct page *page;
447 struct sk_buff *trailer;
448 int tailen = esp->tailen;
449
450 /* this is non-NULL only with TCP/UDP Encapsulation */
451 if (x->encap) {
452 int err = esp_output_encap(x, skb, esp);
453
454 if (err < 0)
455 return err;
456 }
457
458 if (ALIGN(tailen, L1_CACHE_BYTES) > PAGE_SIZE ||
459 ALIGN(skb->data_len, L1_CACHE_BYTES) > PAGE_SIZE)
460 goto cow;
461
462 if (!skb_cloned(skb)) {
463 if (tailen <= skb_tailroom(skb)) {
464 nfrags = 1;
465 trailer = skb;
466 tail = skb_tail_pointer(trailer);
467
468 goto skip_cow;
469 } else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
470 && !skb_has_frag_list(skb)) {
471 int allocsize;
472 struct sock *sk = skb->sk;
473 struct page_frag *pfrag = &x->xfrag;
474
475 esp->inplace = false;
476
477 allocsize = ALIGN(tailen, L1_CACHE_BYTES);
478
479 spin_lock_bh(&x->lock);
480
481 if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
482 spin_unlock_bh(&x->lock);
483 goto cow;
484 }
485
486 page = pfrag->page;
487 get_page(page);
488
489 tail = page_address(page) + pfrag->offset;
490
491 esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
492
493 nfrags = skb_shinfo(skb)->nr_frags;
494
495 __skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
496 tailen);
497 skb_shinfo(skb)->nr_frags = ++nfrags;
498
499 pfrag->offset = pfrag->offset + allocsize;
500
501 spin_unlock_bh(&x->lock);
502
503 nfrags++;
504
505 skb_len_add(skb, tailen);
506 if (sk && sk_fullsock(sk))
507 refcount_add(tailen, &sk->sk_wmem_alloc);
508
509 goto out;
510 }
511 }
512
513cow:
514 esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
515
516 nfrags = skb_cow_data(skb, tailen, &trailer);
517 if (nfrags < 0)
518 goto out;
519 tail = skb_tail_pointer(trailer);
520 esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
521
522skip_cow:
523 esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
524 pskb_put(skb, trailer, tailen);
525
526out:
527 return nfrags;
528}
529EXPORT_SYMBOL_GPL(esp_output_head);
530
531int esp_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
532{
533 u8 *iv;
534 int alen;
535 void *tmp;
536 int ivlen;
537 int assoclen;
538 int extralen;
539 struct page *page;
540 struct ip_esp_hdr *esph;
541 struct crypto_aead *aead;
542 struct aead_request *req;
543 struct scatterlist *sg, *dsg;
544 struct esp_output_extra *extra;
545 int err = -ENOMEM;
546
547 assoclen = sizeof(struct ip_esp_hdr);
548 extralen = 0;
549
550 if (x->props.flags & XFRM_STATE_ESN) {
551 extralen += sizeof(*extra);
552 assoclen += sizeof(__be32);
553 }
554
555 aead = x->data;
556 alen = crypto_aead_authsize(aead);
557 ivlen = crypto_aead_ivsize(aead);
558
559 tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
560 if (!tmp)
561 goto error;
562
563 extra = esp_tmp_extra(tmp);
564 iv = esp_tmp_iv(aead, tmp, extralen);
565 req = esp_tmp_req(aead, iv);
566 sg = esp_req_sg(aead, req);
567
568 if (esp->inplace)
569 dsg = sg;
570 else
571 dsg = &sg[esp->nfrags];
572
573 esph = esp_output_set_extra(skb, x, esp->esph, extra);
574 esp->esph = esph;
575
576 sg_init_table(sg, esp->nfrags);
577 err = skb_to_sgvec(skb, sg,
578 (unsigned char *)esph - skb->data,
579 assoclen + ivlen + esp->clen + alen);
580 if (unlikely(err < 0))
581 goto error_free;
582
583 if (!esp->inplace) {
584 int allocsize;
585 struct page_frag *pfrag = &x->xfrag;
586
587 allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
588
589 spin_lock_bh(&x->lock);
590 if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
591 spin_unlock_bh(&x->lock);
592 goto error_free;
593 }
594
595 skb_shinfo(skb)->nr_frags = 1;
596
597 page = pfrag->page;
598 get_page(page);
599 /* replace page frags in skb with new page */
600 __skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
601 pfrag->offset = pfrag->offset + allocsize;
602 spin_unlock_bh(&x->lock);
603
604 sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
605 err = skb_to_sgvec(skb, dsg,
606 (unsigned char *)esph - skb->data,
607 assoclen + ivlen + esp->clen + alen);
608 if (unlikely(err < 0))
609 goto error_free;
610 }
611
612 if ((x->props.flags & XFRM_STATE_ESN))
613 aead_request_set_callback(req, 0, esp_output_done_esn, skb);
614 else
615 aead_request_set_callback(req, 0, esp_output_done, skb);
616
617 aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
618 aead_request_set_ad(req, assoclen);
619
620 memset(iv, 0, ivlen);
621 memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
622 min(ivlen, 8));
623
624 ESP_SKB_CB(skb)->tmp = tmp;
625 err = crypto_aead_encrypt(req);
626
627 switch (err) {
628 case -EINPROGRESS:
629 goto error;
630
631 case -ENOSPC:
632 err = NET_XMIT_DROP;
633 break;
634
635 case 0:
636 if ((x->props.flags & XFRM_STATE_ESN))
637 esp_output_restore_header(skb);
638 }
639
640 if (sg != dsg)
641 esp_ssg_unref(x, tmp, skb);
642
643 if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
644 err = esp_output_tail_tcp(x, skb);
645
646error_free:
647 kfree(tmp);
648error:
649 return err;
650}
651EXPORT_SYMBOL_GPL(esp_output_tail);
652
653static int esp_output(struct xfrm_state *x, struct sk_buff *skb)
654{
655 int alen;
656 int blksize;
657 struct ip_esp_hdr *esph;
658 struct crypto_aead *aead;
659 struct esp_info esp;
660
661 esp.inplace = true;
662
663 esp.proto = *skb_mac_header(skb);
664 *skb_mac_header(skb) = IPPROTO_ESP;
665
666 /* skb is pure payload to encrypt */
667
668 aead = x->data;
669 alen = crypto_aead_authsize(aead);
670
671 esp.tfclen = 0;
672 if (x->tfcpad) {
673 struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
674 u32 padto;
675
676 padto = min(x->tfcpad, xfrm_state_mtu(x, dst->child_mtu_cached));
677 if (skb->len < padto)
678 esp.tfclen = padto - skb->len;
679 }
680 blksize = ALIGN(crypto_aead_blocksize(aead), 4);
681 esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
682 esp.plen = esp.clen - skb->len - esp.tfclen;
683 esp.tailen = esp.tfclen + esp.plen + alen;
684
685 esp.esph = ip_esp_hdr(skb);
686
687 esp.nfrags = esp_output_head(x, skb, &esp);
688 if (esp.nfrags < 0)
689 return esp.nfrags;
690
691 esph = esp.esph;
692 esph->spi = x->id.spi;
693
694 esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
695 esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
696 ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
697
698 skb_push(skb, -skb_network_offset(skb));
699
700 return esp_output_tail(x, skb, &esp);
701}
702
703static inline int esp_remove_trailer(struct sk_buff *skb)
704{
705 struct xfrm_state *x = xfrm_input_state(skb);
706 struct crypto_aead *aead = x->data;
707 int alen, hlen, elen;
708 int padlen, trimlen;
709 __wsum csumdiff;
710 u8 nexthdr[2];
711 int ret;
712
713 alen = crypto_aead_authsize(aead);
714 hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
715 elen = skb->len - hlen;
716
717 if (skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2))
718 BUG();
719
720 ret = -EINVAL;
721 padlen = nexthdr[0];
722 if (padlen + 2 + alen >= elen) {
723 net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
724 padlen + 2, elen - alen);
725 goto out;
726 }
727
728 trimlen = alen + padlen + 2;
729 if (skb->ip_summed == CHECKSUM_COMPLETE) {
730 csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
731 skb->csum = csum_block_sub(skb->csum, csumdiff,
732 skb->len - trimlen);
733 }
734 ret = pskb_trim(skb, skb->len - trimlen);
735 if (unlikely(ret))
736 return ret;
737
738 ret = nexthdr[1];
739
740out:
741 return ret;
742}
743
744int esp_input_done2(struct sk_buff *skb, int err)
745{
746 const struct iphdr *iph;
747 struct xfrm_state *x = xfrm_input_state(skb);
748 struct xfrm_offload *xo = xfrm_offload(skb);
749 struct crypto_aead *aead = x->data;
750 int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
751 int ihl;
752
753 if (!xo || !(xo->flags & CRYPTO_DONE))
754 kfree(ESP_SKB_CB(skb)->tmp);
755
756 if (unlikely(err))
757 goto out;
758
759 err = esp_remove_trailer(skb);
760 if (unlikely(err < 0))
761 goto out;
762
763 iph = ip_hdr(skb);
764 ihl = iph->ihl * 4;
765
766 if (x->encap) {
767 struct xfrm_encap_tmpl *encap = x->encap;
768 struct tcphdr *th = (void *)(skb_network_header(skb) + ihl);
769 struct udphdr *uh = (void *)(skb_network_header(skb) + ihl);
770 __be16 source;
771
772 switch (x->encap->encap_type) {
773 case TCP_ENCAP_ESPINTCP:
774 source = th->source;
775 break;
776 case UDP_ENCAP_ESPINUDP:
777 source = uh->source;
778 break;
779 default:
780 WARN_ON_ONCE(1);
781 err = -EINVAL;
782 goto out;
783 }
784
785 /*
786 * 1) if the NAT-T peer's IP or port changed then
787 * advertise the change to the keying daemon.
788 * This is an inbound SA, so just compare
789 * SRC ports.
790 */
791 if (iph->saddr != x->props.saddr.a4 ||
792 source != encap->encap_sport) {
793 xfrm_address_t ipaddr;
794
795 ipaddr.a4 = iph->saddr;
796 km_new_mapping(x, &ipaddr, source);
797
798 /* XXX: perhaps add an extra
799 * policy check here, to see
800 * if we should allow or
801 * reject a packet from a
802 * different source
803 * address/port.
804 */
805 }
806
807 /*
808 * 2) ignore UDP/TCP checksums in case
809 * of NAT-T in Transport Mode, or
810 * perform other post-processing fixes
811 * as per draft-ietf-ipsec-udp-encaps-06,
812 * section 3.1.2
813 */
814 if (x->props.mode == XFRM_MODE_TRANSPORT)
815 skb->ip_summed = CHECKSUM_UNNECESSARY;
816 }
817
818 skb_pull_rcsum(skb, hlen);
819 if (x->props.mode == XFRM_MODE_TUNNEL)
820 skb_reset_transport_header(skb);
821 else
822 skb_set_transport_header(skb, -ihl);
823
824 /* RFC4303: Drop dummy packets without any error */
825 if (err == IPPROTO_NONE)
826 err = -EINVAL;
827
828out:
829 return err;
830}
831EXPORT_SYMBOL_GPL(esp_input_done2);
832
833static void esp_input_done(void *data, int err)
834{
835 struct sk_buff *skb = data;
836
837 xfrm_input_resume(skb, esp_input_done2(skb, err));
838}
839
840static void esp_input_restore_header(struct sk_buff *skb)
841{
842 esp_restore_header(skb, 0);
843 __skb_pull(skb, 4);
844}
845
846static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
847{
848 struct xfrm_state *x = xfrm_input_state(skb);
849 struct ip_esp_hdr *esph;
850
851 /* For ESN we move the header forward by 4 bytes to
852 * accommodate the high bits. We will move it back after
853 * decryption.
854 */
855 if ((x->props.flags & XFRM_STATE_ESN)) {
856 esph = skb_push(skb, 4);
857 *seqhi = esph->spi;
858 esph->spi = esph->seq_no;
859 esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
860 }
861}
862
863static void esp_input_done_esn(void *data, int err)
864{
865 struct sk_buff *skb = data;
866
867 esp_input_restore_header(skb);
868 esp_input_done(data, err);
869}
870
871/*
872 * Note: detecting truncated vs. non-truncated authentication data is very
873 * expensive, so we only support truncated data, which is the recommended
874 * and common case.
875 */
876static int esp_input(struct xfrm_state *x, struct sk_buff *skb)
877{
878 struct crypto_aead *aead = x->data;
879 struct aead_request *req;
880 struct sk_buff *trailer;
881 int ivlen = crypto_aead_ivsize(aead);
882 int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
883 int nfrags;
884 int assoclen;
885 int seqhilen;
886 __be32 *seqhi;
887 void *tmp;
888 u8 *iv;
889 struct scatterlist *sg;
890 int err = -EINVAL;
891
892 if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen))
893 goto out;
894
895 if (elen <= 0)
896 goto out;
897
898 assoclen = sizeof(struct ip_esp_hdr);
899 seqhilen = 0;
900
901 if (x->props.flags & XFRM_STATE_ESN) {
902 seqhilen += sizeof(__be32);
903 assoclen += seqhilen;
904 }
905
906 if (!skb_cloned(skb)) {
907 if (!skb_is_nonlinear(skb)) {
908 nfrags = 1;
909
910 goto skip_cow;
911 } else if (!skb_has_frag_list(skb)) {
912 nfrags = skb_shinfo(skb)->nr_frags;
913 nfrags++;
914
915 goto skip_cow;
916 }
917 }
918
919 err = skb_cow_data(skb, 0, &trailer);
920 if (err < 0)
921 goto out;
922
923 nfrags = err;
924
925skip_cow:
926 err = -ENOMEM;
927 tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
928 if (!tmp)
929 goto out;
930
931 ESP_SKB_CB(skb)->tmp = tmp;
932 seqhi = esp_tmp_extra(tmp);
933 iv = esp_tmp_iv(aead, tmp, seqhilen);
934 req = esp_tmp_req(aead, iv);
935 sg = esp_req_sg(aead, req);
936
937 esp_input_set_header(skb, seqhi);
938
939 sg_init_table(sg, nfrags);
940 err = skb_to_sgvec(skb, sg, 0, skb->len);
941 if (unlikely(err < 0)) {
942 kfree(tmp);
943 goto out;
944 }
945
946 skb->ip_summed = CHECKSUM_NONE;
947
948 if ((x->props.flags & XFRM_STATE_ESN))
949 aead_request_set_callback(req, 0, esp_input_done_esn, skb);
950 else
951 aead_request_set_callback(req, 0, esp_input_done, skb);
952
953 aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
954 aead_request_set_ad(req, assoclen);
955
956 err = crypto_aead_decrypt(req);
957 if (err == -EINPROGRESS)
958 goto out;
959
960 if ((x->props.flags & XFRM_STATE_ESN))
961 esp_input_restore_header(skb);
962
963 err = esp_input_done2(skb, err);
964
965out:
966 return err;
967}
968
969static int esp4_err(struct sk_buff *skb, u32 info)
970{
971 struct net *net = dev_net(skb->dev);
972 const struct iphdr *iph = (const struct iphdr *)skb->data;
973 struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data+(iph->ihl<<2));
974 struct xfrm_state *x;
975
976 switch (icmp_hdr(skb)->type) {
977 case ICMP_DEST_UNREACH:
978 if (icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
979 return 0;
980 break;
981 case ICMP_REDIRECT:
982 break;
983 default:
984 return 0;
985 }
986
987 x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
988 esph->spi, IPPROTO_ESP, AF_INET);
989 if (!x)
990 return 0;
991
992 if (icmp_hdr(skb)->type == ICMP_DEST_UNREACH)
993 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ESP);
994 else
995 ipv4_redirect(skb, net, 0, IPPROTO_ESP);
996 xfrm_state_put(x);
997
998 return 0;
999}
1000
1001static void esp_destroy(struct xfrm_state *x)
1002{
1003 struct crypto_aead *aead = x->data;
1004
1005 if (!aead)
1006 return;
1007
1008 crypto_free_aead(aead);
1009}
1010
1011static int esp_init_aead(struct xfrm_state *x, struct netlink_ext_ack *extack)
1012{
1013 char aead_name[CRYPTO_MAX_ALG_NAME];
1014 struct crypto_aead *aead;
1015 int err;
1016
1017 if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1018 x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME) {
1019 NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1020 return -ENAMETOOLONG;
1021 }
1022
1023 aead = crypto_alloc_aead(aead_name, 0, 0);
1024 err = PTR_ERR(aead);
1025 if (IS_ERR(aead))
1026 goto error;
1027
1028 x->data = aead;
1029
1030 err = crypto_aead_setkey(aead, x->aead->alg_key,
1031 (x->aead->alg_key_len + 7) / 8);
1032 if (err)
1033 goto error;
1034
1035 err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1036 if (err)
1037 goto error;
1038
1039 return 0;
1040
1041error:
1042 NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1043 return err;
1044}
1045
1046static int esp_init_authenc(struct xfrm_state *x,
1047 struct netlink_ext_ack *extack)
1048{
1049 struct crypto_aead *aead;
1050 struct crypto_authenc_key_param *param;
1051 struct rtattr *rta;
1052 char *key;
1053 char *p;
1054 char authenc_name[CRYPTO_MAX_ALG_NAME];
1055 unsigned int keylen;
1056 int err;
1057
1058 err = -ENAMETOOLONG;
1059
1060 if ((x->props.flags & XFRM_STATE_ESN)) {
1061 if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1062 "%s%sauthencesn(%s,%s)%s",
1063 x->geniv ?: "", x->geniv ? "(" : "",
1064 x->aalg ? x->aalg->alg_name : "digest_null",
1065 x->ealg->alg_name,
1066 x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1067 NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1068 goto error;
1069 }
1070 } else {
1071 if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1072 "%s%sauthenc(%s,%s)%s",
1073 x->geniv ?: "", x->geniv ? "(" : "",
1074 x->aalg ? x->aalg->alg_name : "digest_null",
1075 x->ealg->alg_name,
1076 x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1077 NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1078 goto error;
1079 }
1080 }
1081
1082 aead = crypto_alloc_aead(authenc_name, 0, 0);
1083 err = PTR_ERR(aead);
1084 if (IS_ERR(aead)) {
1085 NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1086 goto error;
1087 }
1088
1089 x->data = aead;
1090
1091 keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1092 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1093 err = -ENOMEM;
1094 key = kmalloc(keylen, GFP_KERNEL);
1095 if (!key)
1096 goto error;
1097
1098 p = key;
1099 rta = (void *)p;
1100 rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1101 rta->rta_len = RTA_LENGTH(sizeof(*param));
1102 param = RTA_DATA(rta);
1103 p += RTA_SPACE(sizeof(*param));
1104
1105 if (x->aalg) {
1106 struct xfrm_algo_desc *aalg_desc;
1107
1108 memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1109 p += (x->aalg->alg_key_len + 7) / 8;
1110
1111 aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1112 BUG_ON(!aalg_desc);
1113
1114 err = -EINVAL;
1115 if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1116 crypto_aead_authsize(aead)) {
1117 NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1118 goto free_key;
1119 }
1120
1121 err = crypto_aead_setauthsize(
1122 aead, x->aalg->alg_trunc_len / 8);
1123 if (err) {
1124 NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1125 goto free_key;
1126 }
1127 }
1128
1129 param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1130 memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1131
1132 err = crypto_aead_setkey(aead, key, keylen);
1133
1134free_key:
1135 kfree_sensitive(key);
1136
1137error:
1138 return err;
1139}
1140
1141static int esp_init_state(struct xfrm_state *x, struct netlink_ext_ack *extack)
1142{
1143 struct crypto_aead *aead;
1144 u32 align;
1145 int err;
1146
1147 x->data = NULL;
1148
1149 if (x->aead) {
1150 err = esp_init_aead(x, extack);
1151 } else if (x->ealg) {
1152 err = esp_init_authenc(x, extack);
1153 } else {
1154 NL_SET_ERR_MSG(extack, "ESP: AEAD or CRYPT must be provided");
1155 err = -EINVAL;
1156 }
1157
1158 if (err)
1159 goto error;
1160
1161 aead = x->data;
1162
1163 x->props.header_len = sizeof(struct ip_esp_hdr) +
1164 crypto_aead_ivsize(aead);
1165 if (x->props.mode == XFRM_MODE_TUNNEL)
1166 x->props.header_len += sizeof(struct iphdr);
1167 else if (x->props.mode == XFRM_MODE_BEET && x->sel.family != AF_INET6)
1168 x->props.header_len += IPV4_BEET_PHMAXLEN;
1169 if (x->encap) {
1170 struct xfrm_encap_tmpl *encap = x->encap;
1171
1172 switch (encap->encap_type) {
1173 default:
1174 NL_SET_ERR_MSG(extack, "Unsupported encapsulation type for ESP");
1175 err = -EINVAL;
1176 goto error;
1177 case UDP_ENCAP_ESPINUDP:
1178 x->props.header_len += sizeof(struct udphdr);
1179 break;
1180#ifdef CONFIG_INET_ESPINTCP
1181 case TCP_ENCAP_ESPINTCP:
1182 /* only the length field, TCP encap is done by
1183 * the socket
1184 */
1185 x->props.header_len += 2;
1186 break;
1187#endif
1188 }
1189 }
1190
1191 align = ALIGN(crypto_aead_blocksize(aead), 4);
1192 x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1193
1194error:
1195 return err;
1196}
1197
1198static int esp4_rcv_cb(struct sk_buff *skb, int err)
1199{
1200 return 0;
1201}
1202
1203static const struct xfrm_type esp_type =
1204{
1205 .owner = THIS_MODULE,
1206 .proto = IPPROTO_ESP,
1207 .flags = XFRM_TYPE_REPLAY_PROT,
1208 .init_state = esp_init_state,
1209 .destructor = esp_destroy,
1210 .input = esp_input,
1211 .output = esp_output,
1212};
1213
1214static struct xfrm4_protocol esp4_protocol = {
1215 .handler = xfrm4_rcv,
1216 .input_handler = xfrm_input,
1217 .cb_handler = esp4_rcv_cb,
1218 .err_handler = esp4_err,
1219 .priority = 0,
1220};
1221
1222static int __init esp4_init(void)
1223{
1224 if (xfrm_register_type(&esp_type, AF_INET) < 0) {
1225 pr_info("%s: can't add xfrm type\n", __func__);
1226 return -EAGAIN;
1227 }
1228 if (xfrm4_protocol_register(&esp4_protocol, IPPROTO_ESP) < 0) {
1229 pr_info("%s: can't add protocol\n", __func__);
1230 xfrm_unregister_type(&esp_type, AF_INET);
1231 return -EAGAIN;
1232 }
1233 return 0;
1234}
1235
1236static void __exit esp4_fini(void)
1237{
1238 if (xfrm4_protocol_deregister(&esp4_protocol, IPPROTO_ESP) < 0)
1239 pr_info("%s: can't remove protocol\n", __func__);
1240 xfrm_unregister_type(&esp_type, AF_INET);
1241}
1242
1243module_init(esp4_init);
1244module_exit(esp4_fini);
1245MODULE_DESCRIPTION("IPv4 ESP transformation library");
1246MODULE_LICENSE("GPL");
1247MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_ESP);
1// SPDX-License-Identifier: GPL-2.0-only
2#define pr_fmt(fmt) "IPsec: " fmt
3
4#include <crypto/aead.h>
5#include <crypto/authenc.h>
6#include <linux/err.h>
7#include <linux/module.h>
8#include <net/ip.h>
9#include <net/xfrm.h>
10#include <net/esp.h>
11#include <linux/scatterlist.h>
12#include <linux/kernel.h>
13#include <linux/pfkeyv2.h>
14#include <linux/rtnetlink.h>
15#include <linux/slab.h>
16#include <linux/spinlock.h>
17#include <linux/in6.h>
18#include <net/icmp.h>
19#include <net/protocol.h>
20#include <net/udp.h>
21#include <net/tcp.h>
22#include <net/espintcp.h>
23
24#include <linux/highmem.h>
25
26struct esp_skb_cb {
27 struct xfrm_skb_cb xfrm;
28 void *tmp;
29};
30
31struct esp_output_extra {
32 __be32 seqhi;
33 u32 esphoff;
34};
35
36#define ESP_SKB_CB(__skb) ((struct esp_skb_cb *)&((__skb)->cb[0]))
37
38/*
39 * Allocate an AEAD request structure with extra space for SG and IV.
40 *
41 * For alignment considerations the IV is placed at the front, followed
42 * by the request and finally the SG list.
43 *
44 * TODO: Use spare space in skb for this where possible.
45 */
46static void *esp_alloc_tmp(struct crypto_aead *aead, int nfrags, int extralen)
47{
48 unsigned int len;
49
50 len = extralen;
51
52 len += crypto_aead_ivsize(aead);
53
54 if (len) {
55 len += crypto_aead_alignmask(aead) &
56 ~(crypto_tfm_ctx_alignment() - 1);
57 len = ALIGN(len, crypto_tfm_ctx_alignment());
58 }
59
60 len += sizeof(struct aead_request) + crypto_aead_reqsize(aead);
61 len = ALIGN(len, __alignof__(struct scatterlist));
62
63 len += sizeof(struct scatterlist) * nfrags;
64
65 return kmalloc(len, GFP_ATOMIC);
66}
67
68static inline void *esp_tmp_extra(void *tmp)
69{
70 return PTR_ALIGN(tmp, __alignof__(struct esp_output_extra));
71}
72
73static inline u8 *esp_tmp_iv(struct crypto_aead *aead, void *tmp, int extralen)
74{
75 return crypto_aead_ivsize(aead) ?
76 PTR_ALIGN((u8 *)tmp + extralen,
77 crypto_aead_alignmask(aead) + 1) : tmp + extralen;
78}
79
80static inline struct aead_request *esp_tmp_req(struct crypto_aead *aead, u8 *iv)
81{
82 struct aead_request *req;
83
84 req = (void *)PTR_ALIGN(iv + crypto_aead_ivsize(aead),
85 crypto_tfm_ctx_alignment());
86 aead_request_set_tfm(req, aead);
87 return req;
88}
89
90static inline struct scatterlist *esp_req_sg(struct crypto_aead *aead,
91 struct aead_request *req)
92{
93 return (void *)ALIGN((unsigned long)(req + 1) +
94 crypto_aead_reqsize(aead),
95 __alignof__(struct scatterlist));
96}
97
98static void esp_ssg_unref(struct xfrm_state *x, void *tmp)
99{
100 struct crypto_aead *aead = x->data;
101 int extralen = 0;
102 u8 *iv;
103 struct aead_request *req;
104 struct scatterlist *sg;
105
106 if (x->props.flags & XFRM_STATE_ESN)
107 extralen += sizeof(struct esp_output_extra);
108
109 iv = esp_tmp_iv(aead, tmp, extralen);
110 req = esp_tmp_req(aead, iv);
111
112 /* Unref skb_frag_pages in the src scatterlist if necessary.
113 * Skip the first sg which comes from skb->data.
114 */
115 if (req->src != req->dst)
116 for (sg = sg_next(req->src); sg; sg = sg_next(sg))
117 put_page(sg_page(sg));
118}
119
120#ifdef CONFIG_INET_ESPINTCP
121struct esp_tcp_sk {
122 struct sock *sk;
123 struct rcu_head rcu;
124};
125
126static void esp_free_tcp_sk(struct rcu_head *head)
127{
128 struct esp_tcp_sk *esk = container_of(head, struct esp_tcp_sk, rcu);
129
130 sock_put(esk->sk);
131 kfree(esk);
132}
133
134static struct sock *esp_find_tcp_sk(struct xfrm_state *x)
135{
136 struct xfrm_encap_tmpl *encap = x->encap;
137 struct net *net = xs_net(x);
138 struct esp_tcp_sk *esk;
139 __be16 sport, dport;
140 struct sock *nsk;
141 struct sock *sk;
142
143 sk = rcu_dereference(x->encap_sk);
144 if (sk && sk->sk_state == TCP_ESTABLISHED)
145 return sk;
146
147 spin_lock_bh(&x->lock);
148 sport = encap->encap_sport;
149 dport = encap->encap_dport;
150 nsk = rcu_dereference_protected(x->encap_sk,
151 lockdep_is_held(&x->lock));
152 if (sk && sk == nsk) {
153 esk = kmalloc(sizeof(*esk), GFP_ATOMIC);
154 if (!esk) {
155 spin_unlock_bh(&x->lock);
156 return ERR_PTR(-ENOMEM);
157 }
158 RCU_INIT_POINTER(x->encap_sk, NULL);
159 esk->sk = sk;
160 call_rcu(&esk->rcu, esp_free_tcp_sk);
161 }
162 spin_unlock_bh(&x->lock);
163
164 sk = inet_lookup_established(net, net->ipv4.tcp_death_row.hashinfo, x->id.daddr.a4,
165 dport, x->props.saddr.a4, sport, 0);
166 if (!sk)
167 return ERR_PTR(-ENOENT);
168
169 if (!tcp_is_ulp_esp(sk)) {
170 sock_put(sk);
171 return ERR_PTR(-EINVAL);
172 }
173
174 spin_lock_bh(&x->lock);
175 nsk = rcu_dereference_protected(x->encap_sk,
176 lockdep_is_held(&x->lock));
177 if (encap->encap_sport != sport ||
178 encap->encap_dport != dport) {
179 sock_put(sk);
180 sk = nsk ?: ERR_PTR(-EREMCHG);
181 } else if (sk == nsk) {
182 sock_put(sk);
183 } else {
184 rcu_assign_pointer(x->encap_sk, sk);
185 }
186 spin_unlock_bh(&x->lock);
187
188 return sk;
189}
190
191static int esp_output_tcp_finish(struct xfrm_state *x, struct sk_buff *skb)
192{
193 struct sock *sk;
194 int err;
195
196 rcu_read_lock();
197
198 sk = esp_find_tcp_sk(x);
199 err = PTR_ERR_OR_ZERO(sk);
200 if (err)
201 goto out;
202
203 bh_lock_sock(sk);
204 if (sock_owned_by_user(sk))
205 err = espintcp_queue_out(sk, skb);
206 else
207 err = espintcp_push_skb(sk, skb);
208 bh_unlock_sock(sk);
209
210out:
211 rcu_read_unlock();
212 return err;
213}
214
215static int esp_output_tcp_encap_cb(struct net *net, struct sock *sk,
216 struct sk_buff *skb)
217{
218 struct dst_entry *dst = skb_dst(skb);
219 struct xfrm_state *x = dst->xfrm;
220
221 return esp_output_tcp_finish(x, skb);
222}
223
224static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
225{
226 int err;
227
228 local_bh_disable();
229 err = xfrm_trans_queue_net(xs_net(x), skb, esp_output_tcp_encap_cb);
230 local_bh_enable();
231
232 /* EINPROGRESS just happens to do the right thing. It
233 * actually means that the skb has been consumed and
234 * isn't coming back.
235 */
236 return err ?: -EINPROGRESS;
237}
238#else
239static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
240{
241 kfree_skb(skb);
242
243 return -EOPNOTSUPP;
244}
245#endif
246
247static void esp_output_done(void *data, int err)
248{
249 struct sk_buff *skb = data;
250 struct xfrm_offload *xo = xfrm_offload(skb);
251 void *tmp;
252 struct xfrm_state *x;
253
254 if (xo && (xo->flags & XFRM_DEV_RESUME)) {
255 struct sec_path *sp = skb_sec_path(skb);
256
257 x = sp->xvec[sp->len - 1];
258 } else {
259 x = skb_dst(skb)->xfrm;
260 }
261
262 tmp = ESP_SKB_CB(skb)->tmp;
263 esp_ssg_unref(x, tmp);
264 kfree(tmp);
265
266 if (xo && (xo->flags & XFRM_DEV_RESUME)) {
267 if (err) {
268 XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
269 kfree_skb(skb);
270 return;
271 }
272
273 skb_push(skb, skb->data - skb_mac_header(skb));
274 secpath_reset(skb);
275 xfrm_dev_resume(skb);
276 } else {
277 if (!err &&
278 x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
279 esp_output_tail_tcp(x, skb);
280 else
281 xfrm_output_resume(skb->sk, skb, err);
282 }
283}
284
285/* Move ESP header back into place. */
286static void esp_restore_header(struct sk_buff *skb, unsigned int offset)
287{
288 struct ip_esp_hdr *esph = (void *)(skb->data + offset);
289 void *tmp = ESP_SKB_CB(skb)->tmp;
290 __be32 *seqhi = esp_tmp_extra(tmp);
291
292 esph->seq_no = esph->spi;
293 esph->spi = *seqhi;
294}
295
296static void esp_output_restore_header(struct sk_buff *skb)
297{
298 void *tmp = ESP_SKB_CB(skb)->tmp;
299 struct esp_output_extra *extra = esp_tmp_extra(tmp);
300
301 esp_restore_header(skb, skb_transport_offset(skb) + extra->esphoff -
302 sizeof(__be32));
303}
304
305static struct ip_esp_hdr *esp_output_set_extra(struct sk_buff *skb,
306 struct xfrm_state *x,
307 struct ip_esp_hdr *esph,
308 struct esp_output_extra *extra)
309{
310 /* For ESN we move the header forward by 4 bytes to
311 * accommodate the high bits. We will move it back after
312 * encryption.
313 */
314 if ((x->props.flags & XFRM_STATE_ESN)) {
315 __u32 seqhi;
316 struct xfrm_offload *xo = xfrm_offload(skb);
317
318 if (xo)
319 seqhi = xo->seq.hi;
320 else
321 seqhi = XFRM_SKB_CB(skb)->seq.output.hi;
322
323 extra->esphoff = (unsigned char *)esph -
324 skb_transport_header(skb);
325 esph = (struct ip_esp_hdr *)((unsigned char *)esph - 4);
326 extra->seqhi = esph->spi;
327 esph->seq_no = htonl(seqhi);
328 }
329
330 esph->spi = x->id.spi;
331
332 return esph;
333}
334
335static void esp_output_done_esn(void *data, int err)
336{
337 struct sk_buff *skb = data;
338
339 esp_output_restore_header(skb);
340 esp_output_done(data, err);
341}
342
343static struct ip_esp_hdr *esp_output_udp_encap(struct sk_buff *skb,
344 int encap_type,
345 struct esp_info *esp,
346 __be16 sport,
347 __be16 dport)
348{
349 struct udphdr *uh;
350 __be32 *udpdata32;
351 unsigned int len;
352
353 len = skb->len + esp->tailen - skb_transport_offset(skb);
354 if (len + sizeof(struct iphdr) > IP_MAX_MTU)
355 return ERR_PTR(-EMSGSIZE);
356
357 uh = (struct udphdr *)esp->esph;
358 uh->source = sport;
359 uh->dest = dport;
360 uh->len = htons(len);
361 uh->check = 0;
362
363 *skb_mac_header(skb) = IPPROTO_UDP;
364
365 if (encap_type == UDP_ENCAP_ESPINUDP_NON_IKE) {
366 udpdata32 = (__be32 *)(uh + 1);
367 udpdata32[0] = udpdata32[1] = 0;
368 return (struct ip_esp_hdr *)(udpdata32 + 2);
369 }
370
371 return (struct ip_esp_hdr *)(uh + 1);
372}
373
374#ifdef CONFIG_INET_ESPINTCP
375static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
376 struct sk_buff *skb,
377 struct esp_info *esp)
378{
379 __be16 *lenp = (void *)esp->esph;
380 struct ip_esp_hdr *esph;
381 unsigned int len;
382 struct sock *sk;
383
384 len = skb->len + esp->tailen - skb_transport_offset(skb);
385 if (len > IP_MAX_MTU)
386 return ERR_PTR(-EMSGSIZE);
387
388 rcu_read_lock();
389 sk = esp_find_tcp_sk(x);
390 rcu_read_unlock();
391
392 if (IS_ERR(sk))
393 return ERR_CAST(sk);
394
395 *lenp = htons(len);
396 esph = (struct ip_esp_hdr *)(lenp + 1);
397
398 return esph;
399}
400#else
401static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
402 struct sk_buff *skb,
403 struct esp_info *esp)
404{
405 return ERR_PTR(-EOPNOTSUPP);
406}
407#endif
408
409static int esp_output_encap(struct xfrm_state *x, struct sk_buff *skb,
410 struct esp_info *esp)
411{
412 struct xfrm_encap_tmpl *encap = x->encap;
413 struct ip_esp_hdr *esph;
414 __be16 sport, dport;
415 int encap_type;
416
417 spin_lock_bh(&x->lock);
418 sport = encap->encap_sport;
419 dport = encap->encap_dport;
420 encap_type = encap->encap_type;
421 spin_unlock_bh(&x->lock);
422
423 switch (encap_type) {
424 default:
425 case UDP_ENCAP_ESPINUDP:
426 case UDP_ENCAP_ESPINUDP_NON_IKE:
427 esph = esp_output_udp_encap(skb, encap_type, esp, sport, dport);
428 break;
429 case TCP_ENCAP_ESPINTCP:
430 esph = esp_output_tcp_encap(x, skb, esp);
431 break;
432 }
433
434 if (IS_ERR(esph))
435 return PTR_ERR(esph);
436
437 esp->esph = esph;
438
439 return 0;
440}
441
442int esp_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
443{
444 u8 *tail;
445 int nfrags;
446 int esph_offset;
447 struct page *page;
448 struct sk_buff *trailer;
449 int tailen = esp->tailen;
450
451 /* this is non-NULL only with TCP/UDP Encapsulation */
452 if (x->encap) {
453 int err = esp_output_encap(x, skb, esp);
454
455 if (err < 0)
456 return err;
457 }
458
459 if (ALIGN(tailen, L1_CACHE_BYTES) > PAGE_SIZE ||
460 ALIGN(skb->data_len, L1_CACHE_BYTES) > PAGE_SIZE)
461 goto cow;
462
463 if (!skb_cloned(skb)) {
464 if (tailen <= skb_tailroom(skb)) {
465 nfrags = 1;
466 trailer = skb;
467 tail = skb_tail_pointer(trailer);
468
469 goto skip_cow;
470 } else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
471 && !skb_has_frag_list(skb)) {
472 int allocsize;
473 struct sock *sk = skb->sk;
474 struct page_frag *pfrag = &x->xfrag;
475
476 esp->inplace = false;
477
478 allocsize = ALIGN(tailen, L1_CACHE_BYTES);
479
480 spin_lock_bh(&x->lock);
481
482 if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
483 spin_unlock_bh(&x->lock);
484 goto cow;
485 }
486
487 page = pfrag->page;
488 get_page(page);
489
490 tail = page_address(page) + pfrag->offset;
491
492 esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
493
494 nfrags = skb_shinfo(skb)->nr_frags;
495
496 __skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
497 tailen);
498 skb_shinfo(skb)->nr_frags = ++nfrags;
499
500 pfrag->offset = pfrag->offset + allocsize;
501
502 spin_unlock_bh(&x->lock);
503
504 nfrags++;
505
506 skb_len_add(skb, tailen);
507 if (sk && sk_fullsock(sk))
508 refcount_add(tailen, &sk->sk_wmem_alloc);
509
510 goto out;
511 }
512 }
513
514cow:
515 esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
516
517 nfrags = skb_cow_data(skb, tailen, &trailer);
518 if (nfrags < 0)
519 goto out;
520 tail = skb_tail_pointer(trailer);
521 esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
522
523skip_cow:
524 esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
525 pskb_put(skb, trailer, tailen);
526
527out:
528 return nfrags;
529}
530EXPORT_SYMBOL_GPL(esp_output_head);
531
532int esp_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
533{
534 u8 *iv;
535 int alen;
536 void *tmp;
537 int ivlen;
538 int assoclen;
539 int extralen;
540 struct page *page;
541 struct ip_esp_hdr *esph;
542 struct crypto_aead *aead;
543 struct aead_request *req;
544 struct scatterlist *sg, *dsg;
545 struct esp_output_extra *extra;
546 int err = -ENOMEM;
547
548 assoclen = sizeof(struct ip_esp_hdr);
549 extralen = 0;
550
551 if (x->props.flags & XFRM_STATE_ESN) {
552 extralen += sizeof(*extra);
553 assoclen += sizeof(__be32);
554 }
555
556 aead = x->data;
557 alen = crypto_aead_authsize(aead);
558 ivlen = crypto_aead_ivsize(aead);
559
560 tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
561 if (!tmp)
562 goto error;
563
564 extra = esp_tmp_extra(tmp);
565 iv = esp_tmp_iv(aead, tmp, extralen);
566 req = esp_tmp_req(aead, iv);
567 sg = esp_req_sg(aead, req);
568
569 if (esp->inplace)
570 dsg = sg;
571 else
572 dsg = &sg[esp->nfrags];
573
574 esph = esp_output_set_extra(skb, x, esp->esph, extra);
575 esp->esph = esph;
576
577 sg_init_table(sg, esp->nfrags);
578 err = skb_to_sgvec(skb, sg,
579 (unsigned char *)esph - skb->data,
580 assoclen + ivlen + esp->clen + alen);
581 if (unlikely(err < 0))
582 goto error_free;
583
584 if (!esp->inplace) {
585 int allocsize;
586 struct page_frag *pfrag = &x->xfrag;
587
588 allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
589
590 spin_lock_bh(&x->lock);
591 if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
592 spin_unlock_bh(&x->lock);
593 goto error_free;
594 }
595
596 skb_shinfo(skb)->nr_frags = 1;
597
598 page = pfrag->page;
599 get_page(page);
600 /* replace page frags in skb with new page */
601 __skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
602 pfrag->offset = pfrag->offset + allocsize;
603 spin_unlock_bh(&x->lock);
604
605 sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
606 err = skb_to_sgvec(skb, dsg,
607 (unsigned char *)esph - skb->data,
608 assoclen + ivlen + esp->clen + alen);
609 if (unlikely(err < 0))
610 goto error_free;
611 }
612
613 if ((x->props.flags & XFRM_STATE_ESN))
614 aead_request_set_callback(req, 0, esp_output_done_esn, skb);
615 else
616 aead_request_set_callback(req, 0, esp_output_done, skb);
617
618 aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
619 aead_request_set_ad(req, assoclen);
620
621 memset(iv, 0, ivlen);
622 memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
623 min(ivlen, 8));
624
625 ESP_SKB_CB(skb)->tmp = tmp;
626 err = crypto_aead_encrypt(req);
627
628 switch (err) {
629 case -EINPROGRESS:
630 goto error;
631
632 case -ENOSPC:
633 err = NET_XMIT_DROP;
634 break;
635
636 case 0:
637 if ((x->props.flags & XFRM_STATE_ESN))
638 esp_output_restore_header(skb);
639 }
640
641 if (sg != dsg)
642 esp_ssg_unref(x, tmp);
643
644 if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
645 err = esp_output_tail_tcp(x, skb);
646
647error_free:
648 kfree(tmp);
649error:
650 return err;
651}
652EXPORT_SYMBOL_GPL(esp_output_tail);
653
654static int esp_output(struct xfrm_state *x, struct sk_buff *skb)
655{
656 int alen;
657 int blksize;
658 struct ip_esp_hdr *esph;
659 struct crypto_aead *aead;
660 struct esp_info esp;
661
662 esp.inplace = true;
663
664 esp.proto = *skb_mac_header(skb);
665 *skb_mac_header(skb) = IPPROTO_ESP;
666
667 /* skb is pure payload to encrypt */
668
669 aead = x->data;
670 alen = crypto_aead_authsize(aead);
671
672 esp.tfclen = 0;
673 if (x->tfcpad) {
674 struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
675 u32 padto;
676
677 padto = min(x->tfcpad, xfrm_state_mtu(x, dst->child_mtu_cached));
678 if (skb->len < padto)
679 esp.tfclen = padto - skb->len;
680 }
681 blksize = ALIGN(crypto_aead_blocksize(aead), 4);
682 esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
683 esp.plen = esp.clen - skb->len - esp.tfclen;
684 esp.tailen = esp.tfclen + esp.plen + alen;
685
686 esp.esph = ip_esp_hdr(skb);
687
688 esp.nfrags = esp_output_head(x, skb, &esp);
689 if (esp.nfrags < 0)
690 return esp.nfrags;
691
692 esph = esp.esph;
693 esph->spi = x->id.spi;
694
695 esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
696 esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
697 ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
698
699 skb_push(skb, -skb_network_offset(skb));
700
701 return esp_output_tail(x, skb, &esp);
702}
703
704static inline int esp_remove_trailer(struct sk_buff *skb)
705{
706 struct xfrm_state *x = xfrm_input_state(skb);
707 struct crypto_aead *aead = x->data;
708 int alen, hlen, elen;
709 int padlen, trimlen;
710 __wsum csumdiff;
711 u8 nexthdr[2];
712 int ret;
713
714 alen = crypto_aead_authsize(aead);
715 hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
716 elen = skb->len - hlen;
717
718 if (skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2))
719 BUG();
720
721 ret = -EINVAL;
722 padlen = nexthdr[0];
723 if (padlen + 2 + alen >= elen) {
724 net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
725 padlen + 2, elen - alen);
726 goto out;
727 }
728
729 trimlen = alen + padlen + 2;
730 if (skb->ip_summed == CHECKSUM_COMPLETE) {
731 csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
732 skb->csum = csum_block_sub(skb->csum, csumdiff,
733 skb->len - trimlen);
734 }
735 ret = pskb_trim(skb, skb->len - trimlen);
736 if (unlikely(ret))
737 return ret;
738
739 ret = nexthdr[1];
740
741out:
742 return ret;
743}
744
745int esp_input_done2(struct sk_buff *skb, int err)
746{
747 const struct iphdr *iph;
748 struct xfrm_state *x = xfrm_input_state(skb);
749 struct xfrm_offload *xo = xfrm_offload(skb);
750 struct crypto_aead *aead = x->data;
751 int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
752 int ihl;
753
754 if (!xo || !(xo->flags & CRYPTO_DONE))
755 kfree(ESP_SKB_CB(skb)->tmp);
756
757 if (unlikely(err))
758 goto out;
759
760 err = esp_remove_trailer(skb);
761 if (unlikely(err < 0))
762 goto out;
763
764 iph = ip_hdr(skb);
765 ihl = iph->ihl * 4;
766
767 if (x->encap) {
768 struct xfrm_encap_tmpl *encap = x->encap;
769 struct tcphdr *th = (void *)(skb_network_header(skb) + ihl);
770 struct udphdr *uh = (void *)(skb_network_header(skb) + ihl);
771 __be16 source;
772
773 switch (x->encap->encap_type) {
774 case TCP_ENCAP_ESPINTCP:
775 source = th->source;
776 break;
777 case UDP_ENCAP_ESPINUDP:
778 case UDP_ENCAP_ESPINUDP_NON_IKE:
779 source = uh->source;
780 break;
781 default:
782 WARN_ON_ONCE(1);
783 err = -EINVAL;
784 goto out;
785 }
786
787 /*
788 * 1) if the NAT-T peer's IP or port changed then
789 * advertise the change to the keying daemon.
790 * This is an inbound SA, so just compare
791 * SRC ports.
792 */
793 if (iph->saddr != x->props.saddr.a4 ||
794 source != encap->encap_sport) {
795 xfrm_address_t ipaddr;
796
797 ipaddr.a4 = iph->saddr;
798 km_new_mapping(x, &ipaddr, source);
799
800 /* XXX: perhaps add an extra
801 * policy check here, to see
802 * if we should allow or
803 * reject a packet from a
804 * different source
805 * address/port.
806 */
807 }
808
809 /*
810 * 2) ignore UDP/TCP checksums in case
811 * of NAT-T in Transport Mode, or
812 * perform other post-processing fixes
813 * as per draft-ietf-ipsec-udp-encaps-06,
814 * section 3.1.2
815 */
816 if (x->props.mode == XFRM_MODE_TRANSPORT)
817 skb->ip_summed = CHECKSUM_UNNECESSARY;
818 }
819
820 skb_pull_rcsum(skb, hlen);
821 if (x->props.mode == XFRM_MODE_TUNNEL)
822 skb_reset_transport_header(skb);
823 else
824 skb_set_transport_header(skb, -ihl);
825
826 /* RFC4303: Drop dummy packets without any error */
827 if (err == IPPROTO_NONE)
828 err = -EINVAL;
829
830out:
831 return err;
832}
833EXPORT_SYMBOL_GPL(esp_input_done2);
834
835static void esp_input_done(void *data, int err)
836{
837 struct sk_buff *skb = data;
838
839 xfrm_input_resume(skb, esp_input_done2(skb, err));
840}
841
842static void esp_input_restore_header(struct sk_buff *skb)
843{
844 esp_restore_header(skb, 0);
845 __skb_pull(skb, 4);
846}
847
848static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
849{
850 struct xfrm_state *x = xfrm_input_state(skb);
851 struct ip_esp_hdr *esph;
852
853 /* For ESN we move the header forward by 4 bytes to
854 * accommodate the high bits. We will move it back after
855 * decryption.
856 */
857 if ((x->props.flags & XFRM_STATE_ESN)) {
858 esph = skb_push(skb, 4);
859 *seqhi = esph->spi;
860 esph->spi = esph->seq_no;
861 esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
862 }
863}
864
865static void esp_input_done_esn(void *data, int err)
866{
867 struct sk_buff *skb = data;
868
869 esp_input_restore_header(skb);
870 esp_input_done(data, err);
871}
872
873/*
874 * Note: detecting truncated vs. non-truncated authentication data is very
875 * expensive, so we only support truncated data, which is the recommended
876 * and common case.
877 */
878static int esp_input(struct xfrm_state *x, struct sk_buff *skb)
879{
880 struct crypto_aead *aead = x->data;
881 struct aead_request *req;
882 struct sk_buff *trailer;
883 int ivlen = crypto_aead_ivsize(aead);
884 int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
885 int nfrags;
886 int assoclen;
887 int seqhilen;
888 __be32 *seqhi;
889 void *tmp;
890 u8 *iv;
891 struct scatterlist *sg;
892 int err = -EINVAL;
893
894 if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen))
895 goto out;
896
897 if (elen <= 0)
898 goto out;
899
900 assoclen = sizeof(struct ip_esp_hdr);
901 seqhilen = 0;
902
903 if (x->props.flags & XFRM_STATE_ESN) {
904 seqhilen += sizeof(__be32);
905 assoclen += seqhilen;
906 }
907
908 if (!skb_cloned(skb)) {
909 if (!skb_is_nonlinear(skb)) {
910 nfrags = 1;
911
912 goto skip_cow;
913 } else if (!skb_has_frag_list(skb)) {
914 nfrags = skb_shinfo(skb)->nr_frags;
915 nfrags++;
916
917 goto skip_cow;
918 }
919 }
920
921 err = skb_cow_data(skb, 0, &trailer);
922 if (err < 0)
923 goto out;
924
925 nfrags = err;
926
927skip_cow:
928 err = -ENOMEM;
929 tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
930 if (!tmp)
931 goto out;
932
933 ESP_SKB_CB(skb)->tmp = tmp;
934 seqhi = esp_tmp_extra(tmp);
935 iv = esp_tmp_iv(aead, tmp, seqhilen);
936 req = esp_tmp_req(aead, iv);
937 sg = esp_req_sg(aead, req);
938
939 esp_input_set_header(skb, seqhi);
940
941 sg_init_table(sg, nfrags);
942 err = skb_to_sgvec(skb, sg, 0, skb->len);
943 if (unlikely(err < 0)) {
944 kfree(tmp);
945 goto out;
946 }
947
948 skb->ip_summed = CHECKSUM_NONE;
949
950 if ((x->props.flags & XFRM_STATE_ESN))
951 aead_request_set_callback(req, 0, esp_input_done_esn, skb);
952 else
953 aead_request_set_callback(req, 0, esp_input_done, skb);
954
955 aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
956 aead_request_set_ad(req, assoclen);
957
958 err = crypto_aead_decrypt(req);
959 if (err == -EINPROGRESS)
960 goto out;
961
962 if ((x->props.flags & XFRM_STATE_ESN))
963 esp_input_restore_header(skb);
964
965 err = esp_input_done2(skb, err);
966
967out:
968 return err;
969}
970
971static int esp4_err(struct sk_buff *skb, u32 info)
972{
973 struct net *net = dev_net(skb->dev);
974 const struct iphdr *iph = (const struct iphdr *)skb->data;
975 struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data+(iph->ihl<<2));
976 struct xfrm_state *x;
977
978 switch (icmp_hdr(skb)->type) {
979 case ICMP_DEST_UNREACH:
980 if (icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
981 return 0;
982 break;
983 case ICMP_REDIRECT:
984 break;
985 default:
986 return 0;
987 }
988
989 x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
990 esph->spi, IPPROTO_ESP, AF_INET);
991 if (!x)
992 return 0;
993
994 if (icmp_hdr(skb)->type == ICMP_DEST_UNREACH)
995 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ESP);
996 else
997 ipv4_redirect(skb, net, 0, IPPROTO_ESP);
998 xfrm_state_put(x);
999
1000 return 0;
1001}
1002
1003static void esp_destroy(struct xfrm_state *x)
1004{
1005 struct crypto_aead *aead = x->data;
1006
1007 if (!aead)
1008 return;
1009
1010 crypto_free_aead(aead);
1011}
1012
1013static int esp_init_aead(struct xfrm_state *x, struct netlink_ext_ack *extack)
1014{
1015 char aead_name[CRYPTO_MAX_ALG_NAME];
1016 struct crypto_aead *aead;
1017 int err;
1018
1019 if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1020 x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME) {
1021 NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1022 return -ENAMETOOLONG;
1023 }
1024
1025 aead = crypto_alloc_aead(aead_name, 0, 0);
1026 err = PTR_ERR(aead);
1027 if (IS_ERR(aead))
1028 goto error;
1029
1030 x->data = aead;
1031
1032 err = crypto_aead_setkey(aead, x->aead->alg_key,
1033 (x->aead->alg_key_len + 7) / 8);
1034 if (err)
1035 goto error;
1036
1037 err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1038 if (err)
1039 goto error;
1040
1041 return 0;
1042
1043error:
1044 NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1045 return err;
1046}
1047
1048static int esp_init_authenc(struct xfrm_state *x,
1049 struct netlink_ext_ack *extack)
1050{
1051 struct crypto_aead *aead;
1052 struct crypto_authenc_key_param *param;
1053 struct rtattr *rta;
1054 char *key;
1055 char *p;
1056 char authenc_name[CRYPTO_MAX_ALG_NAME];
1057 unsigned int keylen;
1058 int err;
1059
1060 err = -ENAMETOOLONG;
1061
1062 if ((x->props.flags & XFRM_STATE_ESN)) {
1063 if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1064 "%s%sauthencesn(%s,%s)%s",
1065 x->geniv ?: "", x->geniv ? "(" : "",
1066 x->aalg ? x->aalg->alg_name : "digest_null",
1067 x->ealg->alg_name,
1068 x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1069 NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1070 goto error;
1071 }
1072 } else {
1073 if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1074 "%s%sauthenc(%s,%s)%s",
1075 x->geniv ?: "", x->geniv ? "(" : "",
1076 x->aalg ? x->aalg->alg_name : "digest_null",
1077 x->ealg->alg_name,
1078 x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1079 NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1080 goto error;
1081 }
1082 }
1083
1084 aead = crypto_alloc_aead(authenc_name, 0, 0);
1085 err = PTR_ERR(aead);
1086 if (IS_ERR(aead)) {
1087 NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1088 goto error;
1089 }
1090
1091 x->data = aead;
1092
1093 keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1094 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1095 err = -ENOMEM;
1096 key = kmalloc(keylen, GFP_KERNEL);
1097 if (!key)
1098 goto error;
1099
1100 p = key;
1101 rta = (void *)p;
1102 rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1103 rta->rta_len = RTA_LENGTH(sizeof(*param));
1104 param = RTA_DATA(rta);
1105 p += RTA_SPACE(sizeof(*param));
1106
1107 if (x->aalg) {
1108 struct xfrm_algo_desc *aalg_desc;
1109
1110 memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1111 p += (x->aalg->alg_key_len + 7) / 8;
1112
1113 aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1114 BUG_ON(!aalg_desc);
1115
1116 err = -EINVAL;
1117 if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1118 crypto_aead_authsize(aead)) {
1119 NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1120 goto free_key;
1121 }
1122
1123 err = crypto_aead_setauthsize(
1124 aead, x->aalg->alg_trunc_len / 8);
1125 if (err) {
1126 NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1127 goto free_key;
1128 }
1129 }
1130
1131 param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1132 memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1133
1134 err = crypto_aead_setkey(aead, key, keylen);
1135
1136free_key:
1137 kfree_sensitive(key);
1138
1139error:
1140 return err;
1141}
1142
1143static int esp_init_state(struct xfrm_state *x, struct netlink_ext_ack *extack)
1144{
1145 struct crypto_aead *aead;
1146 u32 align;
1147 int err;
1148
1149 x->data = NULL;
1150
1151 if (x->aead) {
1152 err = esp_init_aead(x, extack);
1153 } else if (x->ealg) {
1154 err = esp_init_authenc(x, extack);
1155 } else {
1156 NL_SET_ERR_MSG(extack, "ESP: AEAD or CRYPT must be provided");
1157 err = -EINVAL;
1158 }
1159
1160 if (err)
1161 goto error;
1162
1163 aead = x->data;
1164
1165 x->props.header_len = sizeof(struct ip_esp_hdr) +
1166 crypto_aead_ivsize(aead);
1167 if (x->props.mode == XFRM_MODE_TUNNEL)
1168 x->props.header_len += sizeof(struct iphdr);
1169 else if (x->props.mode == XFRM_MODE_BEET && x->sel.family != AF_INET6)
1170 x->props.header_len += IPV4_BEET_PHMAXLEN;
1171 if (x->encap) {
1172 struct xfrm_encap_tmpl *encap = x->encap;
1173
1174 switch (encap->encap_type) {
1175 default:
1176 NL_SET_ERR_MSG(extack, "Unsupported encapsulation type for ESP");
1177 err = -EINVAL;
1178 goto error;
1179 case UDP_ENCAP_ESPINUDP:
1180 x->props.header_len += sizeof(struct udphdr);
1181 break;
1182 case UDP_ENCAP_ESPINUDP_NON_IKE:
1183 x->props.header_len += sizeof(struct udphdr) + 2 * sizeof(u32);
1184 break;
1185#ifdef CONFIG_INET_ESPINTCP
1186 case TCP_ENCAP_ESPINTCP:
1187 /* only the length field, TCP encap is done by
1188 * the socket
1189 */
1190 x->props.header_len += 2;
1191 break;
1192#endif
1193 }
1194 }
1195
1196 align = ALIGN(crypto_aead_blocksize(aead), 4);
1197 x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1198
1199error:
1200 return err;
1201}
1202
1203static int esp4_rcv_cb(struct sk_buff *skb, int err)
1204{
1205 return 0;
1206}
1207
1208static const struct xfrm_type esp_type =
1209{
1210 .owner = THIS_MODULE,
1211 .proto = IPPROTO_ESP,
1212 .flags = XFRM_TYPE_REPLAY_PROT,
1213 .init_state = esp_init_state,
1214 .destructor = esp_destroy,
1215 .input = esp_input,
1216 .output = esp_output,
1217};
1218
1219static struct xfrm4_protocol esp4_protocol = {
1220 .handler = xfrm4_rcv,
1221 .input_handler = xfrm_input,
1222 .cb_handler = esp4_rcv_cb,
1223 .err_handler = esp4_err,
1224 .priority = 0,
1225};
1226
1227static int __init esp4_init(void)
1228{
1229 if (xfrm_register_type(&esp_type, AF_INET) < 0) {
1230 pr_info("%s: can't add xfrm type\n", __func__);
1231 return -EAGAIN;
1232 }
1233 if (xfrm4_protocol_register(&esp4_protocol, IPPROTO_ESP) < 0) {
1234 pr_info("%s: can't add protocol\n", __func__);
1235 xfrm_unregister_type(&esp_type, AF_INET);
1236 return -EAGAIN;
1237 }
1238 return 0;
1239}
1240
1241static void __exit esp4_fini(void)
1242{
1243 if (xfrm4_protocol_deregister(&esp4_protocol, IPPROTO_ESP) < 0)
1244 pr_info("%s: can't remove protocol\n", __func__);
1245 xfrm_unregister_type(&esp_type, AF_INET);
1246}
1247
1248module_init(esp4_init);
1249module_exit(esp4_fini);
1250MODULE_DESCRIPTION("IPv4 ESP transformation library");
1251MODULE_LICENSE("GPL");
1252MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_ESP);