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
26 struct esp_skb_cb {
27 struct xfrm_skb_cb xfrm;
28 void *tmp;
29 };
30
31 struct 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 */
esp_alloc_tmp(struct crypto_aead * aead,int nfrags,int extralen)46 static 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
esp_tmp_extra(void * tmp)68 static inline void *esp_tmp_extra(void *tmp)
69 {
70 return PTR_ALIGN(tmp, __alignof__(struct esp_output_extra));
71 }
72
esp_tmp_iv(struct crypto_aead * aead,void * tmp,int extralen)73 static 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
esp_tmp_req(struct crypto_aead * aead,u8 * iv)80 static 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
esp_req_sg(struct crypto_aead * aead,struct aead_request * req)90 static 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
esp_ssg_unref(struct xfrm_state * x,void * tmp)98 static void esp_ssg_unref(struct xfrm_state *x, void *tmp)
99 {
100 struct esp_output_extra *extra = esp_tmp_extra(tmp);
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(*extra);
109
110 extra = esp_tmp_extra(tmp);
111 iv = esp_tmp_iv(aead, tmp, extralen);
112 req = esp_tmp_req(aead, iv);
113
114 /* Unref skb_frag_pages in the src scatterlist if necessary.
115 * Skip the first sg which comes from skb->data.
116 */
117 if (req->src != req->dst)
118 for (sg = sg_next(req->src); sg; sg = sg_next(sg))
119 put_page(sg_page(sg));
120 }
121
122 #ifdef CONFIG_INET_ESPINTCP
123 struct esp_tcp_sk {
124 struct sock *sk;
125 struct rcu_head rcu;
126 };
127
esp_free_tcp_sk(struct rcu_head * head)128 static 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
esp_find_tcp_sk(struct xfrm_state * x)136 static struct sock *esp_find_tcp_sk(struct xfrm_state *x)
137 {
138 struct xfrm_encap_tmpl *encap = x->encap;
139 struct esp_tcp_sk *esk;
140 __be16 sport, dport;
141 struct sock *nsk;
142 struct sock *sk;
143
144 sk = rcu_dereference(x->encap_sk);
145 if (sk && sk->sk_state == TCP_ESTABLISHED)
146 return sk;
147
148 spin_lock_bh(&x->lock);
149 sport = encap->encap_sport;
150 dport = encap->encap_dport;
151 nsk = rcu_dereference_protected(x->encap_sk,
152 lockdep_is_held(&x->lock));
153 if (sk && sk == nsk) {
154 esk = kmalloc(sizeof(*esk), GFP_ATOMIC);
155 if (!esk) {
156 spin_unlock_bh(&x->lock);
157 return ERR_PTR(-ENOMEM);
158 }
159 RCU_INIT_POINTER(x->encap_sk, NULL);
160 esk->sk = sk;
161 call_rcu(&esk->rcu, esp_free_tcp_sk);
162 }
163 spin_unlock_bh(&x->lock);
164
165 sk = inet_lookup_established(xs_net(x), &tcp_hashinfo, x->id.daddr.a4,
166 dport, x->props.saddr.a4, sport, 0);
167 if (!sk)
168 return ERR_PTR(-ENOENT);
169
170 if (!tcp_is_ulp_esp(sk)) {
171 sock_put(sk);
172 return ERR_PTR(-EINVAL);
173 }
174
175 spin_lock_bh(&x->lock);
176 nsk = rcu_dereference_protected(x->encap_sk,
177 lockdep_is_held(&x->lock));
178 if (encap->encap_sport != sport ||
179 encap->encap_dport != dport) {
180 sock_put(sk);
181 sk = nsk ?: ERR_PTR(-EREMCHG);
182 } else if (sk == nsk) {
183 sock_put(sk);
184 } else {
185 rcu_assign_pointer(x->encap_sk, sk);
186 }
187 spin_unlock_bh(&x->lock);
188
189 return sk;
190 }
191
esp_output_tcp_finish(struct xfrm_state * x,struct sk_buff * skb)192 static int esp_output_tcp_finish(struct xfrm_state *x, struct sk_buff *skb)
193 {
194 struct sock *sk;
195 int err;
196
197 rcu_read_lock();
198
199 sk = esp_find_tcp_sk(x);
200 err = PTR_ERR_OR_ZERO(sk);
201 if (err)
202 goto out;
203
204 bh_lock_sock(sk);
205 if (sock_owned_by_user(sk))
206 err = espintcp_queue_out(sk, skb);
207 else
208 err = espintcp_push_skb(sk, skb);
209 bh_unlock_sock(sk);
210
211 out:
212 rcu_read_unlock();
213 return err;
214 }
215
esp_output_tcp_encap_cb(struct net * net,struct sock * sk,struct sk_buff * skb)216 static int esp_output_tcp_encap_cb(struct net *net, struct sock *sk,
217 struct sk_buff *skb)
218 {
219 struct dst_entry *dst = skb_dst(skb);
220 struct xfrm_state *x = dst->xfrm;
221
222 return esp_output_tcp_finish(x, skb);
223 }
224
esp_output_tail_tcp(struct xfrm_state * x,struct sk_buff * skb)225 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
226 {
227 int err;
228
229 local_bh_disable();
230 err = xfrm_trans_queue_net(xs_net(x), skb, esp_output_tcp_encap_cb);
231 local_bh_enable();
232
233 /* EINPROGRESS just happens to do the right thing. It
234 * actually means that the skb has been consumed and
235 * isn't coming back.
236 */
237 return err ?: -EINPROGRESS;
238 }
239 #else
esp_output_tail_tcp(struct xfrm_state * x,struct sk_buff * skb)240 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
241 {
242 kfree_skb(skb);
243
244 return -EOPNOTSUPP;
245 }
246 #endif
247
esp_output_done(struct crypto_async_request * base,int err)248 static void esp_output_done(struct crypto_async_request *base, int err)
249 {
250 struct sk_buff *skb = base->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);
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, err);
283 }
284 }
285
286 /* Move ESP header back into place. */
esp_restore_header(struct sk_buff * skb,unsigned int offset)287 static 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
esp_output_restore_header(struct sk_buff * skb)297 static 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
esp_output_set_extra(struct sk_buff * skb,struct xfrm_state * x,struct ip_esp_hdr * esph,struct esp_output_extra * extra)306 static 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 * accomodate 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
esp_output_done_esn(struct crypto_async_request * base,int err)336 static void esp_output_done_esn(struct crypto_async_request *base, int err)
337 {
338 struct sk_buff *skb = base->data;
339
340 esp_output_restore_header(skb);
341 esp_output_done(base, err);
342 }
343
esp_output_udp_encap(struct sk_buff * skb,int encap_type,struct esp_info * esp,__be16 sport,__be16 dport)344 static 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 __be32 *udpdata32;
352 unsigned int len;
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 *skb_mac_header(skb) = IPPROTO_UDP;
365
366 if (encap_type == UDP_ENCAP_ESPINUDP_NON_IKE) {
367 udpdata32 = (__be32 *)(uh + 1);
368 udpdata32[0] = udpdata32[1] = 0;
369 return (struct ip_esp_hdr *)(udpdata32 + 2);
370 }
371
372 return (struct ip_esp_hdr *)(uh + 1);
373 }
374
375 #ifdef CONFIG_INET_ESPINTCP
esp_output_tcp_encap(struct xfrm_state * x,struct sk_buff * skb,struct esp_info * esp)376 static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
377 struct sk_buff *skb,
378 struct esp_info *esp)
379 {
380 __be16 *lenp = (void *)esp->esph;
381 struct ip_esp_hdr *esph;
382 unsigned int len;
383 struct sock *sk;
384
385 len = skb->len + esp->tailen - skb_transport_offset(skb);
386 if (len > IP_MAX_MTU)
387 return ERR_PTR(-EMSGSIZE);
388
389 rcu_read_lock();
390 sk = esp_find_tcp_sk(x);
391 rcu_read_unlock();
392
393 if (IS_ERR(sk))
394 return ERR_CAST(sk);
395
396 *lenp = htons(len);
397 esph = (struct ip_esp_hdr *)(lenp + 1);
398
399 return esph;
400 }
401 #else
esp_output_tcp_encap(struct xfrm_state * x,struct sk_buff * skb,struct esp_info * esp)402 static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
403 struct sk_buff *skb,
404 struct esp_info *esp)
405 {
406 return ERR_PTR(-EOPNOTSUPP);
407 }
408 #endif
409
esp_output_encap(struct xfrm_state * x,struct sk_buff * skb,struct esp_info * esp)410 static int esp_output_encap(struct xfrm_state *x, struct sk_buff *skb,
411 struct esp_info *esp)
412 {
413 struct xfrm_encap_tmpl *encap = x->encap;
414 struct ip_esp_hdr *esph;
415 __be16 sport, dport;
416 int encap_type;
417
418 spin_lock_bh(&x->lock);
419 sport = encap->encap_sport;
420 dport = encap->encap_dport;
421 encap_type = encap->encap_type;
422 spin_unlock_bh(&x->lock);
423
424 switch (encap_type) {
425 default:
426 case UDP_ENCAP_ESPINUDP:
427 case UDP_ENCAP_ESPINUDP_NON_IKE:
428 esph = esp_output_udp_encap(skb, encap_type, esp, sport, dport);
429 break;
430 case TCP_ENCAP_ESPINTCP:
431 esph = esp_output_tcp_encap(x, skb, esp);
432 break;
433 }
434
435 if (IS_ERR(esph))
436 return PTR_ERR(esph);
437
438 esp->esph = esph;
439
440 return 0;
441 }
442
esp_output_head(struct xfrm_state * x,struct sk_buff * skb,struct esp_info * esp)443 int esp_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
444 {
445 u8 *tail;
446 int nfrags;
447 int esph_offset;
448 struct page *page;
449 struct sk_buff *trailer;
450 int tailen = esp->tailen;
451
452 /* this is non-NULL only with TCP/UDP Encapsulation */
453 if (x->encap) {
454 int err = esp_output_encap(x, skb, esp);
455
456 if (err < 0)
457 return err;
458 }
459
460 if (ALIGN(tailen, L1_CACHE_BYTES) > PAGE_SIZE ||
461 ALIGN(skb->data_len, L1_CACHE_BYTES) > PAGE_SIZE)
462 goto cow;
463
464 if (!skb_cloned(skb)) {
465 if (tailen <= skb_tailroom(skb)) {
466 nfrags = 1;
467 trailer = skb;
468 tail = skb_tail_pointer(trailer);
469
470 goto skip_cow;
471 } else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
472 && !skb_has_frag_list(skb)) {
473 int allocsize;
474 struct sock *sk = skb->sk;
475 struct page_frag *pfrag = &x->xfrag;
476
477 esp->inplace = false;
478
479 allocsize = ALIGN(tailen, L1_CACHE_BYTES);
480
481 spin_lock_bh(&x->lock);
482
483 if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
484 spin_unlock_bh(&x->lock);
485 goto cow;
486 }
487
488 page = pfrag->page;
489 get_page(page);
490
491 tail = page_address(page) + pfrag->offset;
492
493 esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
494
495 nfrags = skb_shinfo(skb)->nr_frags;
496
497 __skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
498 tailen);
499 skb_shinfo(skb)->nr_frags = ++nfrags;
500
501 pfrag->offset = pfrag->offset + allocsize;
502
503 spin_unlock_bh(&x->lock);
504
505 nfrags++;
506
507 skb->len += tailen;
508 skb->data_len += tailen;
509 skb->truesize += tailen;
510 if (sk && sk_fullsock(sk))
511 refcount_add(tailen, &sk->sk_wmem_alloc);
512
513 goto out;
514 }
515 }
516
517 cow:
518 esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
519
520 nfrags = skb_cow_data(skb, tailen, &trailer);
521 if (nfrags < 0)
522 goto out;
523 tail = skb_tail_pointer(trailer);
524 esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
525
526 skip_cow:
527 esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
528 pskb_put(skb, trailer, tailen);
529
530 out:
531 return nfrags;
532 }
533 EXPORT_SYMBOL_GPL(esp_output_head);
534
esp_output_tail(struct xfrm_state * x,struct sk_buff * skb,struct esp_info * esp)535 int esp_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
536 {
537 u8 *iv;
538 int alen;
539 void *tmp;
540 int ivlen;
541 int assoclen;
542 int extralen;
543 struct page *page;
544 struct ip_esp_hdr *esph;
545 struct crypto_aead *aead;
546 struct aead_request *req;
547 struct scatterlist *sg, *dsg;
548 struct esp_output_extra *extra;
549 int err = -ENOMEM;
550
551 assoclen = sizeof(struct ip_esp_hdr);
552 extralen = 0;
553
554 if (x->props.flags & XFRM_STATE_ESN) {
555 extralen += sizeof(*extra);
556 assoclen += sizeof(__be32);
557 }
558
559 aead = x->data;
560 alen = crypto_aead_authsize(aead);
561 ivlen = crypto_aead_ivsize(aead);
562
563 tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
564 if (!tmp)
565 goto error;
566
567 extra = esp_tmp_extra(tmp);
568 iv = esp_tmp_iv(aead, tmp, extralen);
569 req = esp_tmp_req(aead, iv);
570 sg = esp_req_sg(aead, req);
571
572 if (esp->inplace)
573 dsg = sg;
574 else
575 dsg = &sg[esp->nfrags];
576
577 esph = esp_output_set_extra(skb, x, esp->esph, extra);
578 esp->esph = esph;
579
580 sg_init_table(sg, esp->nfrags);
581 err = skb_to_sgvec(skb, sg,
582 (unsigned char *)esph - skb->data,
583 assoclen + ivlen + esp->clen + alen);
584 if (unlikely(err < 0))
585 goto error_free;
586
587 if (!esp->inplace) {
588 int allocsize;
589 struct page_frag *pfrag = &x->xfrag;
590
591 allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
592
593 spin_lock_bh(&x->lock);
594 if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
595 spin_unlock_bh(&x->lock);
596 goto error_free;
597 }
598
599 skb_shinfo(skb)->nr_frags = 1;
600
601 page = pfrag->page;
602 get_page(page);
603 /* replace page frags in skb with new page */
604 __skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
605 pfrag->offset = pfrag->offset + allocsize;
606 spin_unlock_bh(&x->lock);
607
608 sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
609 err = skb_to_sgvec(skb, dsg,
610 (unsigned char *)esph - skb->data,
611 assoclen + ivlen + esp->clen + alen);
612 if (unlikely(err < 0))
613 goto error_free;
614 }
615
616 if ((x->props.flags & XFRM_STATE_ESN))
617 aead_request_set_callback(req, 0, esp_output_done_esn, skb);
618 else
619 aead_request_set_callback(req, 0, esp_output_done, skb);
620
621 aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
622 aead_request_set_ad(req, assoclen);
623
624 memset(iv, 0, ivlen);
625 memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
626 min(ivlen, 8));
627
628 ESP_SKB_CB(skb)->tmp = tmp;
629 err = crypto_aead_encrypt(req);
630
631 switch (err) {
632 case -EINPROGRESS:
633 goto error;
634
635 case -ENOSPC:
636 err = NET_XMIT_DROP;
637 break;
638
639 case 0:
640 if ((x->props.flags & XFRM_STATE_ESN))
641 esp_output_restore_header(skb);
642 }
643
644 if (sg != dsg)
645 esp_ssg_unref(x, tmp);
646
647 if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
648 err = esp_output_tail_tcp(x, skb);
649
650 error_free:
651 kfree(tmp);
652 error:
653 return err;
654 }
655 EXPORT_SYMBOL_GPL(esp_output_tail);
656
esp_output(struct xfrm_state * x,struct sk_buff * skb)657 static int esp_output(struct xfrm_state *x, struct sk_buff *skb)
658 {
659 int alen;
660 int blksize;
661 struct ip_esp_hdr *esph;
662 struct crypto_aead *aead;
663 struct esp_info esp;
664
665 esp.inplace = true;
666
667 esp.proto = *skb_mac_header(skb);
668 *skb_mac_header(skb) = IPPROTO_ESP;
669
670 /* skb is pure payload to encrypt */
671
672 aead = x->data;
673 alen = crypto_aead_authsize(aead);
674
675 esp.tfclen = 0;
676 if (x->tfcpad) {
677 struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
678 u32 padto;
679
680 padto = min(x->tfcpad, xfrm_state_mtu(x, dst->child_mtu_cached));
681 if (skb->len < padto)
682 esp.tfclen = padto - skb->len;
683 }
684 blksize = ALIGN(crypto_aead_blocksize(aead), 4);
685 esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
686 esp.plen = esp.clen - skb->len - esp.tfclen;
687 esp.tailen = esp.tfclen + esp.plen + alen;
688
689 esp.esph = ip_esp_hdr(skb);
690
691 esp.nfrags = esp_output_head(x, skb, &esp);
692 if (esp.nfrags < 0)
693 return esp.nfrags;
694
695 esph = esp.esph;
696 esph->spi = x->id.spi;
697
698 esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
699 esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
700 ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
701
702 skb_push(skb, -skb_network_offset(skb));
703
704 return esp_output_tail(x, skb, &esp);
705 }
706
esp_remove_trailer(struct sk_buff * skb)707 static inline int esp_remove_trailer(struct sk_buff *skb)
708 {
709 struct xfrm_state *x = xfrm_input_state(skb);
710 struct xfrm_offload *xo = xfrm_offload(skb);
711 struct crypto_aead *aead = x->data;
712 int alen, hlen, elen;
713 int padlen, trimlen;
714 __wsum csumdiff;
715 u8 nexthdr[2];
716 int ret;
717
718 alen = crypto_aead_authsize(aead);
719 hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
720 elen = skb->len - hlen;
721
722 if (xo && (xo->flags & XFRM_ESP_NO_TRAILER)) {
723 ret = xo->proto;
724 goto out;
725 }
726
727 if (skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2))
728 BUG();
729
730 ret = -EINVAL;
731 padlen = nexthdr[0];
732 if (padlen + 2 + alen >= elen) {
733 net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
734 padlen + 2, elen - alen);
735 goto out;
736 }
737
738 trimlen = alen + padlen + 2;
739 if (skb->ip_summed == CHECKSUM_COMPLETE) {
740 csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
741 skb->csum = csum_block_sub(skb->csum, csumdiff,
742 skb->len - trimlen);
743 }
744 ret = pskb_trim(skb, skb->len - trimlen);
745 if (unlikely(ret))
746 return ret;
747
748 ret = nexthdr[1];
749
750 out:
751 return ret;
752 }
753
esp_input_done2(struct sk_buff * skb,int err)754 int esp_input_done2(struct sk_buff *skb, int err)
755 {
756 const struct iphdr *iph;
757 struct xfrm_state *x = xfrm_input_state(skb);
758 struct xfrm_offload *xo = xfrm_offload(skb);
759 struct crypto_aead *aead = x->data;
760 int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
761 int ihl;
762
763 if (!xo || (xo && !(xo->flags & CRYPTO_DONE)))
764 kfree(ESP_SKB_CB(skb)->tmp);
765
766 if (unlikely(err))
767 goto out;
768
769 err = esp_remove_trailer(skb);
770 if (unlikely(err < 0))
771 goto out;
772
773 iph = ip_hdr(skb);
774 ihl = iph->ihl * 4;
775
776 if (x->encap) {
777 struct xfrm_encap_tmpl *encap = x->encap;
778 struct tcphdr *th = (void *)(skb_network_header(skb) + ihl);
779 struct udphdr *uh = (void *)(skb_network_header(skb) + ihl);
780 __be16 source;
781
782 switch (x->encap->encap_type) {
783 case TCP_ENCAP_ESPINTCP:
784 source = th->source;
785 break;
786 case UDP_ENCAP_ESPINUDP:
787 case UDP_ENCAP_ESPINUDP_NON_IKE:
788 source = uh->source;
789 break;
790 default:
791 WARN_ON_ONCE(1);
792 err = -EINVAL;
793 goto out;
794 }
795
796 /*
797 * 1) if the NAT-T peer's IP or port changed then
798 * advertize the change to the keying daemon.
799 * This is an inbound SA, so just compare
800 * SRC ports.
801 */
802 if (iph->saddr != x->props.saddr.a4 ||
803 source != encap->encap_sport) {
804 xfrm_address_t ipaddr;
805
806 ipaddr.a4 = iph->saddr;
807 km_new_mapping(x, &ipaddr, source);
808
809 /* XXX: perhaps add an extra
810 * policy check here, to see
811 * if we should allow or
812 * reject a packet from a
813 * different source
814 * address/port.
815 */
816 }
817
818 /*
819 * 2) ignore UDP/TCP checksums in case
820 * of NAT-T in Transport Mode, or
821 * perform other post-processing fixes
822 * as per draft-ietf-ipsec-udp-encaps-06,
823 * section 3.1.2
824 */
825 if (x->props.mode == XFRM_MODE_TRANSPORT)
826 skb->ip_summed = CHECKSUM_UNNECESSARY;
827 }
828
829 skb_pull_rcsum(skb, hlen);
830 if (x->props.mode == XFRM_MODE_TUNNEL)
831 skb_reset_transport_header(skb);
832 else
833 skb_set_transport_header(skb, -ihl);
834
835 /* RFC4303: Drop dummy packets without any error */
836 if (err == IPPROTO_NONE)
837 err = -EINVAL;
838
839 out:
840 return err;
841 }
842 EXPORT_SYMBOL_GPL(esp_input_done2);
843
esp_input_done(struct crypto_async_request * base,int err)844 static void esp_input_done(struct crypto_async_request *base, int err)
845 {
846 struct sk_buff *skb = base->data;
847
848 xfrm_input_resume(skb, esp_input_done2(skb, err));
849 }
850
esp_input_restore_header(struct sk_buff * skb)851 static void esp_input_restore_header(struct sk_buff *skb)
852 {
853 esp_restore_header(skb, 0);
854 __skb_pull(skb, 4);
855 }
856
esp_input_set_header(struct sk_buff * skb,__be32 * seqhi)857 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
858 {
859 struct xfrm_state *x = xfrm_input_state(skb);
860 struct ip_esp_hdr *esph;
861
862 /* For ESN we move the header forward by 4 bytes to
863 * accomodate the high bits. We will move it back after
864 * decryption.
865 */
866 if ((x->props.flags & XFRM_STATE_ESN)) {
867 esph = skb_push(skb, 4);
868 *seqhi = esph->spi;
869 esph->spi = esph->seq_no;
870 esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
871 }
872 }
873
esp_input_done_esn(struct crypto_async_request * base,int err)874 static void esp_input_done_esn(struct crypto_async_request *base, int err)
875 {
876 struct sk_buff *skb = base->data;
877
878 esp_input_restore_header(skb);
879 esp_input_done(base, err);
880 }
881
882 /*
883 * Note: detecting truncated vs. non-truncated authentication data is very
884 * expensive, so we only support truncated data, which is the recommended
885 * and common case.
886 */
esp_input(struct xfrm_state * x,struct sk_buff * skb)887 static int esp_input(struct xfrm_state *x, struct sk_buff *skb)
888 {
889 struct crypto_aead *aead = x->data;
890 struct aead_request *req;
891 struct sk_buff *trailer;
892 int ivlen = crypto_aead_ivsize(aead);
893 int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
894 int nfrags;
895 int assoclen;
896 int seqhilen;
897 __be32 *seqhi;
898 void *tmp;
899 u8 *iv;
900 struct scatterlist *sg;
901 int err = -EINVAL;
902
903 if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen))
904 goto out;
905
906 if (elen <= 0)
907 goto out;
908
909 assoclen = sizeof(struct ip_esp_hdr);
910 seqhilen = 0;
911
912 if (x->props.flags & XFRM_STATE_ESN) {
913 seqhilen += sizeof(__be32);
914 assoclen += seqhilen;
915 }
916
917 if (!skb_cloned(skb)) {
918 if (!skb_is_nonlinear(skb)) {
919 nfrags = 1;
920
921 goto skip_cow;
922 } else if (!skb_has_frag_list(skb)) {
923 nfrags = skb_shinfo(skb)->nr_frags;
924 nfrags++;
925
926 goto skip_cow;
927 }
928 }
929
930 err = skb_cow_data(skb, 0, &trailer);
931 if (err < 0)
932 goto out;
933
934 nfrags = err;
935
936 skip_cow:
937 err = -ENOMEM;
938 tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
939 if (!tmp)
940 goto out;
941
942 ESP_SKB_CB(skb)->tmp = tmp;
943 seqhi = esp_tmp_extra(tmp);
944 iv = esp_tmp_iv(aead, tmp, seqhilen);
945 req = esp_tmp_req(aead, iv);
946 sg = esp_req_sg(aead, req);
947
948 esp_input_set_header(skb, seqhi);
949
950 sg_init_table(sg, nfrags);
951 err = skb_to_sgvec(skb, sg, 0, skb->len);
952 if (unlikely(err < 0)) {
953 kfree(tmp);
954 goto out;
955 }
956
957 skb->ip_summed = CHECKSUM_NONE;
958
959 if ((x->props.flags & XFRM_STATE_ESN))
960 aead_request_set_callback(req, 0, esp_input_done_esn, skb);
961 else
962 aead_request_set_callback(req, 0, esp_input_done, skb);
963
964 aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
965 aead_request_set_ad(req, assoclen);
966
967 err = crypto_aead_decrypt(req);
968 if (err == -EINPROGRESS)
969 goto out;
970
971 if ((x->props.flags & XFRM_STATE_ESN))
972 esp_input_restore_header(skb);
973
974 err = esp_input_done2(skb, err);
975
976 out:
977 return err;
978 }
979
esp4_err(struct sk_buff * skb,u32 info)980 static int esp4_err(struct sk_buff *skb, u32 info)
981 {
982 struct net *net = dev_net(skb->dev);
983 const struct iphdr *iph = (const struct iphdr *)skb->data;
984 struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data+(iph->ihl<<2));
985 struct xfrm_state *x;
986
987 switch (icmp_hdr(skb)->type) {
988 case ICMP_DEST_UNREACH:
989 if (icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
990 return 0;
991 case ICMP_REDIRECT:
992 break;
993 default:
994 return 0;
995 }
996
997 x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
998 esph->spi, IPPROTO_ESP, AF_INET);
999 if (!x)
1000 return 0;
1001
1002 if (icmp_hdr(skb)->type == ICMP_DEST_UNREACH)
1003 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ESP);
1004 else
1005 ipv4_redirect(skb, net, 0, IPPROTO_ESP);
1006 xfrm_state_put(x);
1007
1008 return 0;
1009 }
1010
esp_destroy(struct xfrm_state * x)1011 static void esp_destroy(struct xfrm_state *x)
1012 {
1013 struct crypto_aead *aead = x->data;
1014
1015 if (!aead)
1016 return;
1017
1018 crypto_free_aead(aead);
1019 }
1020
esp_init_aead(struct xfrm_state * x)1021 static int esp_init_aead(struct xfrm_state *x)
1022 {
1023 char aead_name[CRYPTO_MAX_ALG_NAME];
1024 struct crypto_aead *aead;
1025 int err;
1026
1027 err = -ENAMETOOLONG;
1028 if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1029 x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME)
1030 goto error;
1031
1032 aead = crypto_alloc_aead(aead_name, 0, 0);
1033 err = PTR_ERR(aead);
1034 if (IS_ERR(aead))
1035 goto error;
1036
1037 x->data = aead;
1038
1039 err = crypto_aead_setkey(aead, x->aead->alg_key,
1040 (x->aead->alg_key_len + 7) / 8);
1041 if (err)
1042 goto error;
1043
1044 err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1045 if (err)
1046 goto error;
1047
1048 error:
1049 return err;
1050 }
1051
esp_init_authenc(struct xfrm_state * x)1052 static int esp_init_authenc(struct xfrm_state *x)
1053 {
1054 struct crypto_aead *aead;
1055 struct crypto_authenc_key_param *param;
1056 struct rtattr *rta;
1057 char *key;
1058 char *p;
1059 char authenc_name[CRYPTO_MAX_ALG_NAME];
1060 unsigned int keylen;
1061 int err;
1062
1063 err = -EINVAL;
1064 if (!x->ealg)
1065 goto error;
1066
1067 err = -ENAMETOOLONG;
1068
1069 if ((x->props.flags & XFRM_STATE_ESN)) {
1070 if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1071 "%s%sauthencesn(%s,%s)%s",
1072 x->geniv ?: "", x->geniv ? "(" : "",
1073 x->aalg ? x->aalg->alg_name : "digest_null",
1074 x->ealg->alg_name,
1075 x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1076 goto error;
1077 } else {
1078 if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1079 "%s%sauthenc(%s,%s)%s",
1080 x->geniv ?: "", x->geniv ? "(" : "",
1081 x->aalg ? x->aalg->alg_name : "digest_null",
1082 x->ealg->alg_name,
1083 x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME)
1084 goto error;
1085 }
1086
1087 aead = crypto_alloc_aead(authenc_name, 0, 0);
1088 err = PTR_ERR(aead);
1089 if (IS_ERR(aead))
1090 goto error;
1091
1092 x->data = aead;
1093
1094 keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1095 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1096 err = -ENOMEM;
1097 key = kmalloc(keylen, GFP_KERNEL);
1098 if (!key)
1099 goto error;
1100
1101 p = key;
1102 rta = (void *)p;
1103 rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1104 rta->rta_len = RTA_LENGTH(sizeof(*param));
1105 param = RTA_DATA(rta);
1106 p += RTA_SPACE(sizeof(*param));
1107
1108 if (x->aalg) {
1109 struct xfrm_algo_desc *aalg_desc;
1110
1111 memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1112 p += (x->aalg->alg_key_len + 7) / 8;
1113
1114 aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1115 BUG_ON(!aalg_desc);
1116
1117 err = -EINVAL;
1118 if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1119 crypto_aead_authsize(aead)) {
1120 pr_info("ESP: %s digestsize %u != %hu\n",
1121 x->aalg->alg_name,
1122 crypto_aead_authsize(aead),
1123 aalg_desc->uinfo.auth.icv_fullbits / 8);
1124 goto free_key;
1125 }
1126
1127 err = crypto_aead_setauthsize(
1128 aead, x->aalg->alg_trunc_len / 8);
1129 if (err)
1130 goto free_key;
1131 }
1132
1133 param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1134 memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1135
1136 err = crypto_aead_setkey(aead, key, keylen);
1137
1138 free_key:
1139 kfree_sensitive(key);
1140
1141 error:
1142 return err;
1143 }
1144
esp_init_state(struct xfrm_state * x)1145 static int esp_init_state(struct xfrm_state *x)
1146 {
1147 struct crypto_aead *aead;
1148 u32 align;
1149 int err;
1150
1151 x->data = NULL;
1152
1153 if (x->aead)
1154 err = esp_init_aead(x);
1155 else
1156 err = esp_init_authenc(x);
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 err = -EINVAL;
1175 goto error;
1176 case UDP_ENCAP_ESPINUDP:
1177 x->props.header_len += sizeof(struct udphdr);
1178 break;
1179 case UDP_ENCAP_ESPINUDP_NON_IKE:
1180 x->props.header_len += sizeof(struct udphdr) + 2 * sizeof(u32);
1181 break;
1182 #ifdef CONFIG_INET_ESPINTCP
1183 case TCP_ENCAP_ESPINTCP:
1184 /* only the length field, TCP encap is done by
1185 * the socket
1186 */
1187 x->props.header_len += 2;
1188 break;
1189 #endif
1190 }
1191 }
1192
1193 align = ALIGN(crypto_aead_blocksize(aead), 4);
1194 x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1195
1196 error:
1197 return err;
1198 }
1199
esp4_rcv_cb(struct sk_buff * skb,int err)1200 static int esp4_rcv_cb(struct sk_buff *skb, int err)
1201 {
1202 return 0;
1203 }
1204
1205 static const struct xfrm_type esp_type =
1206 {
1207 .description = "ESP4",
1208 .owner = THIS_MODULE,
1209 .proto = IPPROTO_ESP,
1210 .flags = XFRM_TYPE_REPLAY_PROT,
1211 .init_state = esp_init_state,
1212 .destructor = esp_destroy,
1213 .input = esp_input,
1214 .output = esp_output,
1215 };
1216
1217 static struct xfrm4_protocol esp4_protocol = {
1218 .handler = xfrm4_rcv,
1219 .input_handler = xfrm_input,
1220 .cb_handler = esp4_rcv_cb,
1221 .err_handler = esp4_err,
1222 .priority = 0,
1223 };
1224
esp4_init(void)1225 static int __init esp4_init(void)
1226 {
1227 if (xfrm_register_type(&esp_type, AF_INET) < 0) {
1228 pr_info("%s: can't add xfrm type\n", __func__);
1229 return -EAGAIN;
1230 }
1231 if (xfrm4_protocol_register(&esp4_protocol, IPPROTO_ESP) < 0) {
1232 pr_info("%s: can't add protocol\n", __func__);
1233 xfrm_unregister_type(&esp_type, AF_INET);
1234 return -EAGAIN;
1235 }
1236 return 0;
1237 }
1238
esp4_fini(void)1239 static void __exit esp4_fini(void)
1240 {
1241 if (xfrm4_protocol_deregister(&esp4_protocol, IPPROTO_ESP) < 0)
1242 pr_info("%s: can't remove protocol\n", __func__);
1243 xfrm_unregister_type(&esp_type, AF_INET);
1244 }
1245
1246 module_init(esp4_init);
1247 module_exit(esp4_fini);
1248 MODULE_LICENSE("GPL");
1249 MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_ESP);
1250