Loading...
1/*
2 * Copyright (c) 2016-2017, Mellanox Technologies. All rights reserved.
3 * Copyright (c) 2016-2017, Dave Watson <davejwatson@fb.com>. All rights reserved.
4 *
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
10 *
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
13 * conditions are met:
14 *
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
17 * disclaimer.
18 *
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 * SOFTWARE.
32 */
33
34#include <linux/module.h>
35
36#include <net/tcp.h>
37#include <net/inet_common.h>
38#include <linux/highmem.h>
39#include <linux/netdevice.h>
40#include <linux/sched/signal.h>
41#include <linux/inetdevice.h>
42
43#include <net/tls.h>
44
45MODULE_AUTHOR("Mellanox Technologies");
46MODULE_DESCRIPTION("Transport Layer Security Support");
47MODULE_LICENSE("Dual BSD/GPL");
48
49enum {
50 TLSV4,
51 TLSV6,
52 TLS_NUM_PROTS,
53};
54
55enum {
56 TLS_BASE,
57 TLS_SW_TX,
58 TLS_SW_RX,
59 TLS_SW_RXTX,
60 TLS_HW_RECORD,
61 TLS_NUM_CONFIG,
62};
63
64static struct proto *saved_tcpv6_prot;
65static DEFINE_MUTEX(tcpv6_prot_mutex);
66static LIST_HEAD(device_list);
67static DEFINE_MUTEX(device_mutex);
68static struct proto tls_prots[TLS_NUM_PROTS][TLS_NUM_CONFIG];
69static struct proto_ops tls_sw_proto_ops;
70
71static inline void update_sk_prot(struct sock *sk, struct tls_context *ctx)
72{
73 int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4;
74
75 sk->sk_prot = &tls_prots[ip_ver][ctx->conf];
76}
77
78int wait_on_pending_writer(struct sock *sk, long *timeo)
79{
80 int rc = 0;
81 DEFINE_WAIT_FUNC(wait, woken_wake_function);
82
83 add_wait_queue(sk_sleep(sk), &wait);
84 while (1) {
85 if (!*timeo) {
86 rc = -EAGAIN;
87 break;
88 }
89
90 if (signal_pending(current)) {
91 rc = sock_intr_errno(*timeo);
92 break;
93 }
94
95 if (sk_wait_event(sk, timeo, !sk->sk_write_pending, &wait))
96 break;
97 }
98 remove_wait_queue(sk_sleep(sk), &wait);
99 return rc;
100}
101
102int tls_push_sg(struct sock *sk,
103 struct tls_context *ctx,
104 struct scatterlist *sg,
105 u16 first_offset,
106 int flags)
107{
108 int sendpage_flags = flags | MSG_SENDPAGE_NOTLAST;
109 int ret = 0;
110 struct page *p;
111 size_t size;
112 int offset = first_offset;
113
114 size = sg->length - offset;
115 offset += sg->offset;
116
117 ctx->in_tcp_sendpages = true;
118 while (1) {
119 if (sg_is_last(sg))
120 sendpage_flags = flags;
121
122 /* is sending application-limited? */
123 tcp_rate_check_app_limited(sk);
124 p = sg_page(sg);
125retry:
126 ret = do_tcp_sendpages(sk, p, offset, size, sendpage_flags);
127
128 if (ret != size) {
129 if (ret > 0) {
130 offset += ret;
131 size -= ret;
132 goto retry;
133 }
134
135 offset -= sg->offset;
136 ctx->partially_sent_offset = offset;
137 ctx->partially_sent_record = (void *)sg;
138 ctx->in_tcp_sendpages = false;
139 return ret;
140 }
141
142 put_page(p);
143 sk_mem_uncharge(sk, sg->length);
144 sg = sg_next(sg);
145 if (!sg)
146 break;
147
148 offset = sg->offset;
149 size = sg->length;
150 }
151
152 clear_bit(TLS_PENDING_CLOSED_RECORD, &ctx->flags);
153 ctx->in_tcp_sendpages = false;
154 ctx->sk_write_space(sk);
155
156 return 0;
157}
158
159static int tls_handle_open_record(struct sock *sk, int flags)
160{
161 struct tls_context *ctx = tls_get_ctx(sk);
162
163 if (tls_is_pending_open_record(ctx))
164 return ctx->push_pending_record(sk, flags);
165
166 return 0;
167}
168
169int tls_proccess_cmsg(struct sock *sk, struct msghdr *msg,
170 unsigned char *record_type)
171{
172 struct cmsghdr *cmsg;
173 int rc = -EINVAL;
174
175 for_each_cmsghdr(cmsg, msg) {
176 if (!CMSG_OK(msg, cmsg))
177 return -EINVAL;
178 if (cmsg->cmsg_level != SOL_TLS)
179 continue;
180
181 switch (cmsg->cmsg_type) {
182 case TLS_SET_RECORD_TYPE:
183 if (cmsg->cmsg_len < CMSG_LEN(sizeof(*record_type)))
184 return -EINVAL;
185
186 if (msg->msg_flags & MSG_MORE)
187 return -EINVAL;
188
189 rc = tls_handle_open_record(sk, msg->msg_flags);
190 if (rc)
191 return rc;
192
193 *record_type = *(unsigned char *)CMSG_DATA(cmsg);
194 rc = 0;
195 break;
196 default:
197 return -EINVAL;
198 }
199 }
200
201 return rc;
202}
203
204int tls_push_pending_closed_record(struct sock *sk, struct tls_context *ctx,
205 int flags, long *timeo)
206{
207 struct scatterlist *sg;
208 u16 offset;
209
210 if (!tls_is_partially_sent_record(ctx))
211 return ctx->push_pending_record(sk, flags);
212
213 sg = ctx->partially_sent_record;
214 offset = ctx->partially_sent_offset;
215
216 ctx->partially_sent_record = NULL;
217 return tls_push_sg(sk, ctx, sg, offset, flags);
218}
219
220static void tls_write_space(struct sock *sk)
221{
222 struct tls_context *ctx = tls_get_ctx(sk);
223
224 /* We are already sending pages, ignore notification */
225 if (ctx->in_tcp_sendpages)
226 return;
227
228 if (!sk->sk_write_pending && tls_is_pending_closed_record(ctx)) {
229 gfp_t sk_allocation = sk->sk_allocation;
230 int rc;
231 long timeo = 0;
232
233 sk->sk_allocation = GFP_ATOMIC;
234 rc = tls_push_pending_closed_record(sk, ctx,
235 MSG_DONTWAIT |
236 MSG_NOSIGNAL,
237 &timeo);
238 sk->sk_allocation = sk_allocation;
239
240 if (rc < 0)
241 return;
242 }
243
244 ctx->sk_write_space(sk);
245}
246
247static void tls_sk_proto_close(struct sock *sk, long timeout)
248{
249 struct tls_context *ctx = tls_get_ctx(sk);
250 long timeo = sock_sndtimeo(sk, 0);
251 void (*sk_proto_close)(struct sock *sk, long timeout);
252 bool free_ctx = false;
253
254 lock_sock(sk);
255 sk_proto_close = ctx->sk_proto_close;
256
257 if (ctx->conf == TLS_BASE || ctx->conf == TLS_HW_RECORD) {
258 free_ctx = true;
259 goto skip_tx_cleanup;
260 }
261
262 if (!tls_complete_pending_work(sk, ctx, 0, &timeo))
263 tls_handle_open_record(sk, 0);
264
265 if (ctx->partially_sent_record) {
266 struct scatterlist *sg = ctx->partially_sent_record;
267
268 while (1) {
269 put_page(sg_page(sg));
270 sk_mem_uncharge(sk, sg->length);
271
272 if (sg_is_last(sg))
273 break;
274 sg++;
275 }
276 }
277
278 kfree(ctx->tx.rec_seq);
279 kfree(ctx->tx.iv);
280 kfree(ctx->rx.rec_seq);
281 kfree(ctx->rx.iv);
282
283 if (ctx->conf == TLS_SW_TX ||
284 ctx->conf == TLS_SW_RX ||
285 ctx->conf == TLS_SW_RXTX) {
286 tls_sw_free_resources(sk);
287 }
288
289skip_tx_cleanup:
290 release_sock(sk);
291 sk_proto_close(sk, timeout);
292 /* free ctx for TLS_HW_RECORD, used by tcp_set_state
293 * for sk->sk_prot->unhash [tls_hw_unhash]
294 */
295 if (free_ctx)
296 kfree(ctx);
297}
298
299static int do_tls_getsockopt_tx(struct sock *sk, char __user *optval,
300 int __user *optlen)
301{
302 int rc = 0;
303 struct tls_context *ctx = tls_get_ctx(sk);
304 struct tls_crypto_info *crypto_info;
305 int len;
306
307 if (get_user(len, optlen))
308 return -EFAULT;
309
310 if (!optval || (len < sizeof(*crypto_info))) {
311 rc = -EINVAL;
312 goto out;
313 }
314
315 if (!ctx) {
316 rc = -EBUSY;
317 goto out;
318 }
319
320 /* get user crypto info */
321 crypto_info = &ctx->crypto_send;
322
323 if (!TLS_CRYPTO_INFO_READY(crypto_info)) {
324 rc = -EBUSY;
325 goto out;
326 }
327
328 if (len == sizeof(*crypto_info)) {
329 if (copy_to_user(optval, crypto_info, sizeof(*crypto_info)))
330 rc = -EFAULT;
331 goto out;
332 }
333
334 switch (crypto_info->cipher_type) {
335 case TLS_CIPHER_AES_GCM_128: {
336 struct tls12_crypto_info_aes_gcm_128 *
337 crypto_info_aes_gcm_128 =
338 container_of(crypto_info,
339 struct tls12_crypto_info_aes_gcm_128,
340 info);
341
342 if (len != sizeof(*crypto_info_aes_gcm_128)) {
343 rc = -EINVAL;
344 goto out;
345 }
346 lock_sock(sk);
347 memcpy(crypto_info_aes_gcm_128->iv,
348 ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
349 TLS_CIPHER_AES_GCM_128_IV_SIZE);
350 memcpy(crypto_info_aes_gcm_128->rec_seq, ctx->tx.rec_seq,
351 TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE);
352 release_sock(sk);
353 if (copy_to_user(optval,
354 crypto_info_aes_gcm_128,
355 sizeof(*crypto_info_aes_gcm_128)))
356 rc = -EFAULT;
357 break;
358 }
359 default:
360 rc = -EINVAL;
361 }
362
363out:
364 return rc;
365}
366
367static int do_tls_getsockopt(struct sock *sk, int optname,
368 char __user *optval, int __user *optlen)
369{
370 int rc = 0;
371
372 switch (optname) {
373 case TLS_TX:
374 rc = do_tls_getsockopt_tx(sk, optval, optlen);
375 break;
376 default:
377 rc = -ENOPROTOOPT;
378 break;
379 }
380 return rc;
381}
382
383static int tls_getsockopt(struct sock *sk, int level, int optname,
384 char __user *optval, int __user *optlen)
385{
386 struct tls_context *ctx = tls_get_ctx(sk);
387
388 if (level != SOL_TLS)
389 return ctx->getsockopt(sk, level, optname, optval, optlen);
390
391 return do_tls_getsockopt(sk, optname, optval, optlen);
392}
393
394static int do_tls_setsockopt_conf(struct sock *sk, char __user *optval,
395 unsigned int optlen, int tx)
396{
397 struct tls_crypto_info *crypto_info;
398 struct tls_context *ctx = tls_get_ctx(sk);
399 int rc = 0;
400 int conf;
401
402 if (!optval || (optlen < sizeof(*crypto_info))) {
403 rc = -EINVAL;
404 goto out;
405 }
406
407 if (tx)
408 crypto_info = &ctx->crypto_send;
409 else
410 crypto_info = &ctx->crypto_recv;
411
412 /* Currently we don't support set crypto info more than one time */
413 if (TLS_CRYPTO_INFO_READY(crypto_info)) {
414 rc = -EBUSY;
415 goto out;
416 }
417
418 rc = copy_from_user(crypto_info, optval, sizeof(*crypto_info));
419 if (rc) {
420 rc = -EFAULT;
421 goto err_crypto_info;
422 }
423
424 /* check version */
425 if (crypto_info->version != TLS_1_2_VERSION) {
426 rc = -ENOTSUPP;
427 goto err_crypto_info;
428 }
429
430 switch (crypto_info->cipher_type) {
431 case TLS_CIPHER_AES_GCM_128: {
432 if (optlen != sizeof(struct tls12_crypto_info_aes_gcm_128)) {
433 rc = -EINVAL;
434 goto err_crypto_info;
435 }
436 rc = copy_from_user(crypto_info + 1, optval + sizeof(*crypto_info),
437 optlen - sizeof(*crypto_info));
438 if (rc) {
439 rc = -EFAULT;
440 goto err_crypto_info;
441 }
442 break;
443 }
444 default:
445 rc = -EINVAL;
446 goto err_crypto_info;
447 }
448
449 /* currently SW is default, we will have ethtool in future */
450 if (tx) {
451 rc = tls_set_sw_offload(sk, ctx, 1);
452 if (ctx->conf == TLS_SW_RX)
453 conf = TLS_SW_RXTX;
454 else
455 conf = TLS_SW_TX;
456 } else {
457 rc = tls_set_sw_offload(sk, ctx, 0);
458 if (ctx->conf == TLS_SW_TX)
459 conf = TLS_SW_RXTX;
460 else
461 conf = TLS_SW_RX;
462 }
463
464 if (rc)
465 goto err_crypto_info;
466
467 ctx->conf = conf;
468 update_sk_prot(sk, ctx);
469 if (tx) {
470 ctx->sk_write_space = sk->sk_write_space;
471 sk->sk_write_space = tls_write_space;
472 } else {
473 sk->sk_socket->ops = &tls_sw_proto_ops;
474 }
475 goto out;
476
477err_crypto_info:
478 memset(crypto_info, 0, sizeof(*crypto_info));
479out:
480 return rc;
481}
482
483static int do_tls_setsockopt(struct sock *sk, int optname,
484 char __user *optval, unsigned int optlen)
485{
486 int rc = 0;
487
488 switch (optname) {
489 case TLS_TX:
490 case TLS_RX:
491 lock_sock(sk);
492 rc = do_tls_setsockopt_conf(sk, optval, optlen,
493 optname == TLS_TX);
494 release_sock(sk);
495 break;
496 default:
497 rc = -ENOPROTOOPT;
498 break;
499 }
500 return rc;
501}
502
503static int tls_setsockopt(struct sock *sk, int level, int optname,
504 char __user *optval, unsigned int optlen)
505{
506 struct tls_context *ctx = tls_get_ctx(sk);
507
508 if (level != SOL_TLS)
509 return ctx->setsockopt(sk, level, optname, optval, optlen);
510
511 return do_tls_setsockopt(sk, optname, optval, optlen);
512}
513
514static struct tls_context *create_ctx(struct sock *sk)
515{
516 struct inet_connection_sock *icsk = inet_csk(sk);
517 struct tls_context *ctx;
518
519 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
520 if (!ctx)
521 return NULL;
522
523 icsk->icsk_ulp_data = ctx;
524 return ctx;
525}
526
527static int tls_hw_prot(struct sock *sk)
528{
529 struct tls_context *ctx;
530 struct tls_device *dev;
531 int rc = 0;
532
533 mutex_lock(&device_mutex);
534 list_for_each_entry(dev, &device_list, dev_list) {
535 if (dev->feature && dev->feature(dev)) {
536 ctx = create_ctx(sk);
537 if (!ctx)
538 goto out;
539
540 ctx->hash = sk->sk_prot->hash;
541 ctx->unhash = sk->sk_prot->unhash;
542 ctx->sk_proto_close = sk->sk_prot->close;
543 ctx->conf = TLS_HW_RECORD;
544 update_sk_prot(sk, ctx);
545 rc = 1;
546 break;
547 }
548 }
549out:
550 mutex_unlock(&device_mutex);
551 return rc;
552}
553
554static void tls_hw_unhash(struct sock *sk)
555{
556 struct tls_context *ctx = tls_get_ctx(sk);
557 struct tls_device *dev;
558
559 mutex_lock(&device_mutex);
560 list_for_each_entry(dev, &device_list, dev_list) {
561 if (dev->unhash)
562 dev->unhash(dev, sk);
563 }
564 mutex_unlock(&device_mutex);
565 ctx->unhash(sk);
566}
567
568static int tls_hw_hash(struct sock *sk)
569{
570 struct tls_context *ctx = tls_get_ctx(sk);
571 struct tls_device *dev;
572 int err;
573
574 err = ctx->hash(sk);
575 mutex_lock(&device_mutex);
576 list_for_each_entry(dev, &device_list, dev_list) {
577 if (dev->hash)
578 err |= dev->hash(dev, sk);
579 }
580 mutex_unlock(&device_mutex);
581
582 if (err)
583 tls_hw_unhash(sk);
584 return err;
585}
586
587static void build_protos(struct proto *prot, struct proto *base)
588{
589 prot[TLS_BASE] = *base;
590 prot[TLS_BASE].setsockopt = tls_setsockopt;
591 prot[TLS_BASE].getsockopt = tls_getsockopt;
592 prot[TLS_BASE].close = tls_sk_proto_close;
593
594 prot[TLS_SW_TX] = prot[TLS_BASE];
595 prot[TLS_SW_TX].sendmsg = tls_sw_sendmsg;
596 prot[TLS_SW_TX].sendpage = tls_sw_sendpage;
597
598 prot[TLS_SW_RX] = prot[TLS_BASE];
599 prot[TLS_SW_RX].recvmsg = tls_sw_recvmsg;
600 prot[TLS_SW_RX].close = tls_sk_proto_close;
601
602 prot[TLS_SW_RXTX] = prot[TLS_SW_TX];
603 prot[TLS_SW_RXTX].recvmsg = tls_sw_recvmsg;
604 prot[TLS_SW_RXTX].close = tls_sk_proto_close;
605
606 prot[TLS_HW_RECORD] = *base;
607 prot[TLS_HW_RECORD].hash = tls_hw_hash;
608 prot[TLS_HW_RECORD].unhash = tls_hw_unhash;
609 prot[TLS_HW_RECORD].close = tls_sk_proto_close;
610}
611
612static int tls_init(struct sock *sk)
613{
614 int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4;
615 struct tls_context *ctx;
616 int rc = 0;
617
618 if (tls_hw_prot(sk))
619 goto out;
620
621 /* The TLS ulp is currently supported only for TCP sockets
622 * in ESTABLISHED state.
623 * Supporting sockets in LISTEN state will require us
624 * to modify the accept implementation to clone rather then
625 * share the ulp context.
626 */
627 if (sk->sk_state != TCP_ESTABLISHED)
628 return -ENOTSUPP;
629
630 /* allocate tls context */
631 ctx = create_ctx(sk);
632 if (!ctx) {
633 rc = -ENOMEM;
634 goto out;
635 }
636 ctx->setsockopt = sk->sk_prot->setsockopt;
637 ctx->getsockopt = sk->sk_prot->getsockopt;
638 ctx->sk_proto_close = sk->sk_prot->close;
639
640 /* Build IPv6 TLS whenever the address of tcpv6_prot changes */
641 if (ip_ver == TLSV6 &&
642 unlikely(sk->sk_prot != smp_load_acquire(&saved_tcpv6_prot))) {
643 mutex_lock(&tcpv6_prot_mutex);
644 if (likely(sk->sk_prot != saved_tcpv6_prot)) {
645 build_protos(tls_prots[TLSV6], sk->sk_prot);
646 smp_store_release(&saved_tcpv6_prot, sk->sk_prot);
647 }
648 mutex_unlock(&tcpv6_prot_mutex);
649 }
650
651 ctx->conf = TLS_BASE;
652 update_sk_prot(sk, ctx);
653out:
654 return rc;
655}
656
657void tls_register_device(struct tls_device *device)
658{
659 mutex_lock(&device_mutex);
660 list_add_tail(&device->dev_list, &device_list);
661 mutex_unlock(&device_mutex);
662}
663EXPORT_SYMBOL(tls_register_device);
664
665void tls_unregister_device(struct tls_device *device)
666{
667 mutex_lock(&device_mutex);
668 list_del(&device->dev_list);
669 mutex_unlock(&device_mutex);
670}
671EXPORT_SYMBOL(tls_unregister_device);
672
673static struct tcp_ulp_ops tcp_tls_ulp_ops __read_mostly = {
674 .name = "tls",
675 .uid = TCP_ULP_TLS,
676 .user_visible = true,
677 .owner = THIS_MODULE,
678 .init = tls_init,
679};
680
681static int __init tls_register(void)
682{
683 build_protos(tls_prots[TLSV4], &tcp_prot);
684
685 tls_sw_proto_ops = inet_stream_ops;
686 tls_sw_proto_ops.poll = tls_sw_poll;
687 tls_sw_proto_ops.splice_read = tls_sw_splice_read;
688
689 tcp_register_ulp(&tcp_tls_ulp_ops);
690
691 return 0;
692}
693
694static void __exit tls_unregister(void)
695{
696 tcp_unregister_ulp(&tcp_tls_ulp_ops);
697}
698
699module_init(tls_register);
700module_exit(tls_unregister);
1/*
2 * Copyright (c) 2016-2017, Mellanox Technologies. All rights reserved.
3 * Copyright (c) 2016-2017, Dave Watson <davejwatson@fb.com>. All rights reserved.
4 *
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
10 *
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
13 * conditions are met:
14 *
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
17 * disclaimer.
18 *
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 * SOFTWARE.
32 */
33
34#include <linux/module.h>
35
36#include <net/tcp.h>
37#include <net/inet_common.h>
38#include <linux/highmem.h>
39#include <linux/netdevice.h>
40#include <linux/sched/signal.h>
41#include <linux/inetdevice.h>
42#include <linux/inet_diag.h>
43
44#include <net/snmp.h>
45#include <net/tls.h>
46#include <net/tls_toe.h>
47
48#include "tls.h"
49
50MODULE_AUTHOR("Mellanox Technologies");
51MODULE_DESCRIPTION("Transport Layer Security Support");
52MODULE_LICENSE("Dual BSD/GPL");
53MODULE_ALIAS_TCP_ULP("tls");
54
55enum {
56 TLSV4,
57 TLSV6,
58 TLS_NUM_PROTS,
59};
60
61#define CHECK_CIPHER_DESC(cipher,ci) \
62 static_assert(cipher ## _IV_SIZE <= TLS_MAX_IV_SIZE); \
63 static_assert(cipher ## _SALT_SIZE <= TLS_MAX_SALT_SIZE); \
64 static_assert(cipher ## _REC_SEQ_SIZE <= TLS_MAX_REC_SEQ_SIZE); \
65 static_assert(cipher ## _TAG_SIZE == TLS_TAG_SIZE); \
66 static_assert(sizeof_field(struct ci, iv) == cipher ## _IV_SIZE); \
67 static_assert(sizeof_field(struct ci, key) == cipher ## _KEY_SIZE); \
68 static_assert(sizeof_field(struct ci, salt) == cipher ## _SALT_SIZE); \
69 static_assert(sizeof_field(struct ci, rec_seq) == cipher ## _REC_SEQ_SIZE);
70
71#define __CIPHER_DESC(ci) \
72 .iv_offset = offsetof(struct ci, iv), \
73 .key_offset = offsetof(struct ci, key), \
74 .salt_offset = offsetof(struct ci, salt), \
75 .rec_seq_offset = offsetof(struct ci, rec_seq), \
76 .crypto_info = sizeof(struct ci)
77
78#define CIPHER_DESC(cipher,ci,algname,_offloadable) [cipher - TLS_CIPHER_MIN] = { \
79 .nonce = cipher ## _IV_SIZE, \
80 .iv = cipher ## _IV_SIZE, \
81 .key = cipher ## _KEY_SIZE, \
82 .salt = cipher ## _SALT_SIZE, \
83 .tag = cipher ## _TAG_SIZE, \
84 .rec_seq = cipher ## _REC_SEQ_SIZE, \
85 .cipher_name = algname, \
86 .offloadable = _offloadable, \
87 __CIPHER_DESC(ci), \
88}
89
90#define CIPHER_DESC_NONCE0(cipher,ci,algname,_offloadable) [cipher - TLS_CIPHER_MIN] = { \
91 .nonce = 0, \
92 .iv = cipher ## _IV_SIZE, \
93 .key = cipher ## _KEY_SIZE, \
94 .salt = cipher ## _SALT_SIZE, \
95 .tag = cipher ## _TAG_SIZE, \
96 .rec_seq = cipher ## _REC_SEQ_SIZE, \
97 .cipher_name = algname, \
98 .offloadable = _offloadable, \
99 __CIPHER_DESC(ci), \
100}
101
102const struct tls_cipher_desc tls_cipher_desc[TLS_CIPHER_MAX + 1 - TLS_CIPHER_MIN] = {
103 CIPHER_DESC(TLS_CIPHER_AES_GCM_128, tls12_crypto_info_aes_gcm_128, "gcm(aes)", true),
104 CIPHER_DESC(TLS_CIPHER_AES_GCM_256, tls12_crypto_info_aes_gcm_256, "gcm(aes)", true),
105 CIPHER_DESC(TLS_CIPHER_AES_CCM_128, tls12_crypto_info_aes_ccm_128, "ccm(aes)", false),
106 CIPHER_DESC_NONCE0(TLS_CIPHER_CHACHA20_POLY1305, tls12_crypto_info_chacha20_poly1305, "rfc7539(chacha20,poly1305)", false),
107 CIPHER_DESC(TLS_CIPHER_SM4_GCM, tls12_crypto_info_sm4_gcm, "gcm(sm4)", false),
108 CIPHER_DESC(TLS_CIPHER_SM4_CCM, tls12_crypto_info_sm4_ccm, "ccm(sm4)", false),
109 CIPHER_DESC(TLS_CIPHER_ARIA_GCM_128, tls12_crypto_info_aria_gcm_128, "gcm(aria)", false),
110 CIPHER_DESC(TLS_CIPHER_ARIA_GCM_256, tls12_crypto_info_aria_gcm_256, "gcm(aria)", false),
111};
112
113CHECK_CIPHER_DESC(TLS_CIPHER_AES_GCM_128, tls12_crypto_info_aes_gcm_128);
114CHECK_CIPHER_DESC(TLS_CIPHER_AES_GCM_256, tls12_crypto_info_aes_gcm_256);
115CHECK_CIPHER_DESC(TLS_CIPHER_AES_CCM_128, tls12_crypto_info_aes_ccm_128);
116CHECK_CIPHER_DESC(TLS_CIPHER_CHACHA20_POLY1305, tls12_crypto_info_chacha20_poly1305);
117CHECK_CIPHER_DESC(TLS_CIPHER_SM4_GCM, tls12_crypto_info_sm4_gcm);
118CHECK_CIPHER_DESC(TLS_CIPHER_SM4_CCM, tls12_crypto_info_sm4_ccm);
119CHECK_CIPHER_DESC(TLS_CIPHER_ARIA_GCM_128, tls12_crypto_info_aria_gcm_128);
120CHECK_CIPHER_DESC(TLS_CIPHER_ARIA_GCM_256, tls12_crypto_info_aria_gcm_256);
121
122static const struct proto *saved_tcpv6_prot;
123static DEFINE_MUTEX(tcpv6_prot_mutex);
124static const struct proto *saved_tcpv4_prot;
125static DEFINE_MUTEX(tcpv4_prot_mutex);
126static struct proto tls_prots[TLS_NUM_PROTS][TLS_NUM_CONFIG][TLS_NUM_CONFIG];
127static struct proto_ops tls_proto_ops[TLS_NUM_PROTS][TLS_NUM_CONFIG][TLS_NUM_CONFIG];
128static void build_protos(struct proto prot[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
129 const struct proto *base);
130
131void update_sk_prot(struct sock *sk, struct tls_context *ctx)
132{
133 int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4;
134
135 WRITE_ONCE(sk->sk_prot,
136 &tls_prots[ip_ver][ctx->tx_conf][ctx->rx_conf]);
137 WRITE_ONCE(sk->sk_socket->ops,
138 &tls_proto_ops[ip_ver][ctx->tx_conf][ctx->rx_conf]);
139}
140
141int wait_on_pending_writer(struct sock *sk, long *timeo)
142{
143 DEFINE_WAIT_FUNC(wait, woken_wake_function);
144 int ret, rc = 0;
145
146 add_wait_queue(sk_sleep(sk), &wait);
147 while (1) {
148 if (!*timeo) {
149 rc = -EAGAIN;
150 break;
151 }
152
153 if (signal_pending(current)) {
154 rc = sock_intr_errno(*timeo);
155 break;
156 }
157
158 ret = sk_wait_event(sk, timeo,
159 !READ_ONCE(sk->sk_write_pending), &wait);
160 if (ret) {
161 if (ret < 0)
162 rc = ret;
163 break;
164 }
165 }
166 remove_wait_queue(sk_sleep(sk), &wait);
167 return rc;
168}
169
170int tls_push_sg(struct sock *sk,
171 struct tls_context *ctx,
172 struct scatterlist *sg,
173 u16 first_offset,
174 int flags)
175{
176 struct bio_vec bvec;
177 struct msghdr msg = {
178 .msg_flags = MSG_SPLICE_PAGES | flags,
179 };
180 int ret = 0;
181 struct page *p;
182 size_t size;
183 int offset = first_offset;
184
185 size = sg->length - offset;
186 offset += sg->offset;
187
188 ctx->splicing_pages = true;
189 while (1) {
190 /* is sending application-limited? */
191 tcp_rate_check_app_limited(sk);
192 p = sg_page(sg);
193retry:
194 bvec_set_page(&bvec, p, size, offset);
195 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1, size);
196
197 ret = tcp_sendmsg_locked(sk, &msg, size);
198
199 if (ret != size) {
200 if (ret > 0) {
201 offset += ret;
202 size -= ret;
203 goto retry;
204 }
205
206 offset -= sg->offset;
207 ctx->partially_sent_offset = offset;
208 ctx->partially_sent_record = (void *)sg;
209 ctx->splicing_pages = false;
210 return ret;
211 }
212
213 put_page(p);
214 sk_mem_uncharge(sk, sg->length);
215 sg = sg_next(sg);
216 if (!sg)
217 break;
218
219 offset = sg->offset;
220 size = sg->length;
221 }
222
223 ctx->splicing_pages = false;
224
225 return 0;
226}
227
228static int tls_handle_open_record(struct sock *sk, int flags)
229{
230 struct tls_context *ctx = tls_get_ctx(sk);
231
232 if (tls_is_pending_open_record(ctx))
233 return ctx->push_pending_record(sk, flags);
234
235 return 0;
236}
237
238int tls_process_cmsg(struct sock *sk, struct msghdr *msg,
239 unsigned char *record_type)
240{
241 struct cmsghdr *cmsg;
242 int rc = -EINVAL;
243
244 for_each_cmsghdr(cmsg, msg) {
245 if (!CMSG_OK(msg, cmsg))
246 return -EINVAL;
247 if (cmsg->cmsg_level != SOL_TLS)
248 continue;
249
250 switch (cmsg->cmsg_type) {
251 case TLS_SET_RECORD_TYPE:
252 if (cmsg->cmsg_len < CMSG_LEN(sizeof(*record_type)))
253 return -EINVAL;
254
255 if (msg->msg_flags & MSG_MORE)
256 return -EINVAL;
257
258 rc = tls_handle_open_record(sk, msg->msg_flags);
259 if (rc)
260 return rc;
261
262 *record_type = *(unsigned char *)CMSG_DATA(cmsg);
263 rc = 0;
264 break;
265 default:
266 return -EINVAL;
267 }
268 }
269
270 return rc;
271}
272
273int tls_push_partial_record(struct sock *sk, struct tls_context *ctx,
274 int flags)
275{
276 struct scatterlist *sg;
277 u16 offset;
278
279 sg = ctx->partially_sent_record;
280 offset = ctx->partially_sent_offset;
281
282 ctx->partially_sent_record = NULL;
283 return tls_push_sg(sk, ctx, sg, offset, flags);
284}
285
286void tls_free_partial_record(struct sock *sk, struct tls_context *ctx)
287{
288 struct scatterlist *sg;
289
290 for (sg = ctx->partially_sent_record; sg; sg = sg_next(sg)) {
291 put_page(sg_page(sg));
292 sk_mem_uncharge(sk, sg->length);
293 }
294 ctx->partially_sent_record = NULL;
295}
296
297static void tls_write_space(struct sock *sk)
298{
299 struct tls_context *ctx = tls_get_ctx(sk);
300
301 /* If splicing_pages call lower protocol write space handler
302 * to ensure we wake up any waiting operations there. For example
303 * if splicing pages where to call sk_wait_event.
304 */
305 if (ctx->splicing_pages) {
306 ctx->sk_write_space(sk);
307 return;
308 }
309
310#ifdef CONFIG_TLS_DEVICE
311 if (ctx->tx_conf == TLS_HW)
312 tls_device_write_space(sk, ctx);
313 else
314#endif
315 tls_sw_write_space(sk, ctx);
316
317 ctx->sk_write_space(sk);
318}
319
320/**
321 * tls_ctx_free() - free TLS ULP context
322 * @sk: socket to with @ctx is attached
323 * @ctx: TLS context structure
324 *
325 * Free TLS context. If @sk is %NULL caller guarantees that the socket
326 * to which @ctx was attached has no outstanding references.
327 */
328void tls_ctx_free(struct sock *sk, struct tls_context *ctx)
329{
330 if (!ctx)
331 return;
332
333 memzero_explicit(&ctx->crypto_send, sizeof(ctx->crypto_send));
334 memzero_explicit(&ctx->crypto_recv, sizeof(ctx->crypto_recv));
335 mutex_destroy(&ctx->tx_lock);
336
337 if (sk)
338 kfree_rcu(ctx, rcu);
339 else
340 kfree(ctx);
341}
342
343static void tls_sk_proto_cleanup(struct sock *sk,
344 struct tls_context *ctx, long timeo)
345{
346 if (unlikely(sk->sk_write_pending) &&
347 !wait_on_pending_writer(sk, &timeo))
348 tls_handle_open_record(sk, 0);
349
350 /* We need these for tls_sw_fallback handling of other packets */
351 if (ctx->tx_conf == TLS_SW) {
352 tls_sw_release_resources_tx(sk);
353 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXSW);
354 } else if (ctx->tx_conf == TLS_HW) {
355 tls_device_free_resources_tx(sk);
356 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXDEVICE);
357 }
358
359 if (ctx->rx_conf == TLS_SW) {
360 tls_sw_release_resources_rx(sk);
361 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXSW);
362 } else if (ctx->rx_conf == TLS_HW) {
363 tls_device_offload_cleanup_rx(sk);
364 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXDEVICE);
365 }
366}
367
368static void tls_sk_proto_close(struct sock *sk, long timeout)
369{
370 struct inet_connection_sock *icsk = inet_csk(sk);
371 struct tls_context *ctx = tls_get_ctx(sk);
372 long timeo = sock_sndtimeo(sk, 0);
373 bool free_ctx;
374
375 if (ctx->tx_conf == TLS_SW)
376 tls_sw_cancel_work_tx(ctx);
377
378 lock_sock(sk);
379 free_ctx = ctx->tx_conf != TLS_HW && ctx->rx_conf != TLS_HW;
380
381 if (ctx->tx_conf != TLS_BASE || ctx->rx_conf != TLS_BASE)
382 tls_sk_proto_cleanup(sk, ctx, timeo);
383
384 write_lock_bh(&sk->sk_callback_lock);
385 if (free_ctx)
386 rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
387 WRITE_ONCE(sk->sk_prot, ctx->sk_proto);
388 if (sk->sk_write_space == tls_write_space)
389 sk->sk_write_space = ctx->sk_write_space;
390 write_unlock_bh(&sk->sk_callback_lock);
391 release_sock(sk);
392 if (ctx->tx_conf == TLS_SW)
393 tls_sw_free_ctx_tx(ctx);
394 if (ctx->rx_conf == TLS_SW || ctx->rx_conf == TLS_HW)
395 tls_sw_strparser_done(ctx);
396 if (ctx->rx_conf == TLS_SW)
397 tls_sw_free_ctx_rx(ctx);
398 ctx->sk_proto->close(sk, timeout);
399
400 if (free_ctx)
401 tls_ctx_free(sk, ctx);
402}
403
404static __poll_t tls_sk_poll(struct file *file, struct socket *sock,
405 struct poll_table_struct *wait)
406{
407 struct tls_sw_context_rx *ctx;
408 struct tls_context *tls_ctx;
409 struct sock *sk = sock->sk;
410 struct sk_psock *psock;
411 __poll_t mask = 0;
412 u8 shutdown;
413 int state;
414
415 mask = tcp_poll(file, sock, wait);
416
417 state = inet_sk_state_load(sk);
418 shutdown = READ_ONCE(sk->sk_shutdown);
419 if (unlikely(state != TCP_ESTABLISHED || shutdown & RCV_SHUTDOWN))
420 return mask;
421
422 tls_ctx = tls_get_ctx(sk);
423 ctx = tls_sw_ctx_rx(tls_ctx);
424 psock = sk_psock_get(sk);
425
426 if (skb_queue_empty_lockless(&ctx->rx_list) &&
427 !tls_strp_msg_ready(ctx) &&
428 sk_psock_queue_empty(psock))
429 mask &= ~(EPOLLIN | EPOLLRDNORM);
430
431 if (psock)
432 sk_psock_put(sk, psock);
433
434 return mask;
435}
436
437static int do_tls_getsockopt_conf(struct sock *sk, char __user *optval,
438 int __user *optlen, int tx)
439{
440 int rc = 0;
441 const struct tls_cipher_desc *cipher_desc;
442 struct tls_context *ctx = tls_get_ctx(sk);
443 struct tls_crypto_info *crypto_info;
444 struct cipher_context *cctx;
445 int len;
446
447 if (get_user(len, optlen))
448 return -EFAULT;
449
450 if (!optval || (len < sizeof(*crypto_info))) {
451 rc = -EINVAL;
452 goto out;
453 }
454
455 if (!ctx) {
456 rc = -EBUSY;
457 goto out;
458 }
459
460 /* get user crypto info */
461 if (tx) {
462 crypto_info = &ctx->crypto_send.info;
463 cctx = &ctx->tx;
464 } else {
465 crypto_info = &ctx->crypto_recv.info;
466 cctx = &ctx->rx;
467 }
468
469 if (!TLS_CRYPTO_INFO_READY(crypto_info)) {
470 rc = -EBUSY;
471 goto out;
472 }
473
474 if (len == sizeof(*crypto_info)) {
475 if (copy_to_user(optval, crypto_info, sizeof(*crypto_info)))
476 rc = -EFAULT;
477 goto out;
478 }
479
480 cipher_desc = get_cipher_desc(crypto_info->cipher_type);
481 if (!cipher_desc || len != cipher_desc->crypto_info) {
482 rc = -EINVAL;
483 goto out;
484 }
485
486 memcpy(crypto_info_iv(crypto_info, cipher_desc),
487 cctx->iv + cipher_desc->salt, cipher_desc->iv);
488 memcpy(crypto_info_rec_seq(crypto_info, cipher_desc),
489 cctx->rec_seq, cipher_desc->rec_seq);
490
491 if (copy_to_user(optval, crypto_info, cipher_desc->crypto_info))
492 rc = -EFAULT;
493
494out:
495 return rc;
496}
497
498static int do_tls_getsockopt_tx_zc(struct sock *sk, char __user *optval,
499 int __user *optlen)
500{
501 struct tls_context *ctx = tls_get_ctx(sk);
502 unsigned int value;
503 int len;
504
505 if (get_user(len, optlen))
506 return -EFAULT;
507
508 if (len != sizeof(value))
509 return -EINVAL;
510
511 value = ctx->zerocopy_sendfile;
512 if (copy_to_user(optval, &value, sizeof(value)))
513 return -EFAULT;
514
515 return 0;
516}
517
518static int do_tls_getsockopt_no_pad(struct sock *sk, char __user *optval,
519 int __user *optlen)
520{
521 struct tls_context *ctx = tls_get_ctx(sk);
522 int value, len;
523
524 if (ctx->prot_info.version != TLS_1_3_VERSION)
525 return -EINVAL;
526
527 if (get_user(len, optlen))
528 return -EFAULT;
529 if (len < sizeof(value))
530 return -EINVAL;
531
532 value = -EINVAL;
533 if (ctx->rx_conf == TLS_SW || ctx->rx_conf == TLS_HW)
534 value = ctx->rx_no_pad;
535 if (value < 0)
536 return value;
537
538 if (put_user(sizeof(value), optlen))
539 return -EFAULT;
540 if (copy_to_user(optval, &value, sizeof(value)))
541 return -EFAULT;
542
543 return 0;
544}
545
546static int do_tls_getsockopt(struct sock *sk, int optname,
547 char __user *optval, int __user *optlen)
548{
549 int rc = 0;
550
551 lock_sock(sk);
552
553 switch (optname) {
554 case TLS_TX:
555 case TLS_RX:
556 rc = do_tls_getsockopt_conf(sk, optval, optlen,
557 optname == TLS_TX);
558 break;
559 case TLS_TX_ZEROCOPY_RO:
560 rc = do_tls_getsockopt_tx_zc(sk, optval, optlen);
561 break;
562 case TLS_RX_EXPECT_NO_PAD:
563 rc = do_tls_getsockopt_no_pad(sk, optval, optlen);
564 break;
565 default:
566 rc = -ENOPROTOOPT;
567 break;
568 }
569
570 release_sock(sk);
571
572 return rc;
573}
574
575static int tls_getsockopt(struct sock *sk, int level, int optname,
576 char __user *optval, int __user *optlen)
577{
578 struct tls_context *ctx = tls_get_ctx(sk);
579
580 if (level != SOL_TLS)
581 return ctx->sk_proto->getsockopt(sk, level,
582 optname, optval, optlen);
583
584 return do_tls_getsockopt(sk, optname, optval, optlen);
585}
586
587static int validate_crypto_info(const struct tls_crypto_info *crypto_info,
588 const struct tls_crypto_info *alt_crypto_info)
589{
590 if (crypto_info->version != TLS_1_2_VERSION &&
591 crypto_info->version != TLS_1_3_VERSION)
592 return -EINVAL;
593
594 switch (crypto_info->cipher_type) {
595 case TLS_CIPHER_ARIA_GCM_128:
596 case TLS_CIPHER_ARIA_GCM_256:
597 if (crypto_info->version != TLS_1_2_VERSION)
598 return -EINVAL;
599 break;
600 }
601
602 /* Ensure that TLS version and ciphers are same in both directions */
603 if (TLS_CRYPTO_INFO_READY(alt_crypto_info)) {
604 if (alt_crypto_info->version != crypto_info->version ||
605 alt_crypto_info->cipher_type != crypto_info->cipher_type)
606 return -EINVAL;
607 }
608
609 return 0;
610}
611
612static int do_tls_setsockopt_conf(struct sock *sk, sockptr_t optval,
613 unsigned int optlen, int tx)
614{
615 struct tls_crypto_info *crypto_info;
616 struct tls_crypto_info *alt_crypto_info;
617 struct tls_context *ctx = tls_get_ctx(sk);
618 const struct tls_cipher_desc *cipher_desc;
619 int rc = 0;
620 int conf;
621
622 if (sockptr_is_null(optval) || (optlen < sizeof(*crypto_info)))
623 return -EINVAL;
624
625 if (tx) {
626 crypto_info = &ctx->crypto_send.info;
627 alt_crypto_info = &ctx->crypto_recv.info;
628 } else {
629 crypto_info = &ctx->crypto_recv.info;
630 alt_crypto_info = &ctx->crypto_send.info;
631 }
632
633 /* Currently we don't support set crypto info more than one time */
634 if (TLS_CRYPTO_INFO_READY(crypto_info))
635 return -EBUSY;
636
637 rc = copy_from_sockptr(crypto_info, optval, sizeof(*crypto_info));
638 if (rc) {
639 rc = -EFAULT;
640 goto err_crypto_info;
641 }
642
643 rc = validate_crypto_info(crypto_info, alt_crypto_info);
644 if (rc)
645 goto err_crypto_info;
646
647 cipher_desc = get_cipher_desc(crypto_info->cipher_type);
648 if (!cipher_desc) {
649 rc = -EINVAL;
650 goto err_crypto_info;
651 }
652
653 if (optlen != cipher_desc->crypto_info) {
654 rc = -EINVAL;
655 goto err_crypto_info;
656 }
657
658 rc = copy_from_sockptr_offset(crypto_info + 1, optval,
659 sizeof(*crypto_info),
660 optlen - sizeof(*crypto_info));
661 if (rc) {
662 rc = -EFAULT;
663 goto err_crypto_info;
664 }
665
666 if (tx) {
667 rc = tls_set_device_offload(sk);
668 conf = TLS_HW;
669 if (!rc) {
670 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSTXDEVICE);
671 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXDEVICE);
672 } else {
673 rc = tls_set_sw_offload(sk, 1);
674 if (rc)
675 goto err_crypto_info;
676 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSTXSW);
677 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXSW);
678 conf = TLS_SW;
679 }
680 } else {
681 rc = tls_set_device_offload_rx(sk, ctx);
682 conf = TLS_HW;
683 if (!rc) {
684 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXDEVICE);
685 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXDEVICE);
686 } else {
687 rc = tls_set_sw_offload(sk, 0);
688 if (rc)
689 goto err_crypto_info;
690 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXSW);
691 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXSW);
692 conf = TLS_SW;
693 }
694 tls_sw_strparser_arm(sk, ctx);
695 }
696
697 if (tx)
698 ctx->tx_conf = conf;
699 else
700 ctx->rx_conf = conf;
701 update_sk_prot(sk, ctx);
702 if (tx) {
703 ctx->sk_write_space = sk->sk_write_space;
704 sk->sk_write_space = tls_write_space;
705 } else {
706 struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(ctx);
707
708 tls_strp_check_rcv(&rx_ctx->strp);
709 }
710 return 0;
711
712err_crypto_info:
713 memzero_explicit(crypto_info, sizeof(union tls_crypto_context));
714 return rc;
715}
716
717static int do_tls_setsockopt_tx_zc(struct sock *sk, sockptr_t optval,
718 unsigned int optlen)
719{
720 struct tls_context *ctx = tls_get_ctx(sk);
721 unsigned int value;
722
723 if (sockptr_is_null(optval) || optlen != sizeof(value))
724 return -EINVAL;
725
726 if (copy_from_sockptr(&value, optval, sizeof(value)))
727 return -EFAULT;
728
729 if (value > 1)
730 return -EINVAL;
731
732 ctx->zerocopy_sendfile = value;
733
734 return 0;
735}
736
737static int do_tls_setsockopt_no_pad(struct sock *sk, sockptr_t optval,
738 unsigned int optlen)
739{
740 struct tls_context *ctx = tls_get_ctx(sk);
741 u32 val;
742 int rc;
743
744 if (ctx->prot_info.version != TLS_1_3_VERSION ||
745 sockptr_is_null(optval) || optlen < sizeof(val))
746 return -EINVAL;
747
748 rc = copy_from_sockptr(&val, optval, sizeof(val));
749 if (rc)
750 return -EFAULT;
751 if (val > 1)
752 return -EINVAL;
753 rc = check_zeroed_sockptr(optval, sizeof(val), optlen - sizeof(val));
754 if (rc < 1)
755 return rc == 0 ? -EINVAL : rc;
756
757 lock_sock(sk);
758 rc = -EINVAL;
759 if (ctx->rx_conf == TLS_SW || ctx->rx_conf == TLS_HW) {
760 ctx->rx_no_pad = val;
761 tls_update_rx_zc_capable(ctx);
762 rc = 0;
763 }
764 release_sock(sk);
765
766 return rc;
767}
768
769static int do_tls_setsockopt(struct sock *sk, int optname, sockptr_t optval,
770 unsigned int optlen)
771{
772 int rc = 0;
773
774 switch (optname) {
775 case TLS_TX:
776 case TLS_RX:
777 lock_sock(sk);
778 rc = do_tls_setsockopt_conf(sk, optval, optlen,
779 optname == TLS_TX);
780 release_sock(sk);
781 break;
782 case TLS_TX_ZEROCOPY_RO:
783 lock_sock(sk);
784 rc = do_tls_setsockopt_tx_zc(sk, optval, optlen);
785 release_sock(sk);
786 break;
787 case TLS_RX_EXPECT_NO_PAD:
788 rc = do_tls_setsockopt_no_pad(sk, optval, optlen);
789 break;
790 default:
791 rc = -ENOPROTOOPT;
792 break;
793 }
794 return rc;
795}
796
797static int tls_setsockopt(struct sock *sk, int level, int optname,
798 sockptr_t optval, unsigned int optlen)
799{
800 struct tls_context *ctx = tls_get_ctx(sk);
801
802 if (level != SOL_TLS)
803 return ctx->sk_proto->setsockopt(sk, level, optname, optval,
804 optlen);
805
806 return do_tls_setsockopt(sk, optname, optval, optlen);
807}
808
809struct tls_context *tls_ctx_create(struct sock *sk)
810{
811 struct inet_connection_sock *icsk = inet_csk(sk);
812 struct tls_context *ctx;
813
814 ctx = kzalloc(sizeof(*ctx), GFP_ATOMIC);
815 if (!ctx)
816 return NULL;
817
818 mutex_init(&ctx->tx_lock);
819 ctx->sk_proto = READ_ONCE(sk->sk_prot);
820 ctx->sk = sk;
821 /* Release semantic of rcu_assign_pointer() ensures that
822 * ctx->sk_proto is visible before changing sk->sk_prot in
823 * update_sk_prot(), and prevents reading uninitialized value in
824 * tls_{getsockopt, setsockopt}. Note that we do not need a
825 * read barrier in tls_{getsockopt,setsockopt} as there is an
826 * address dependency between sk->sk_proto->{getsockopt,setsockopt}
827 * and ctx->sk_proto.
828 */
829 rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
830 return ctx;
831}
832
833static void build_proto_ops(struct proto_ops ops[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
834 const struct proto_ops *base)
835{
836 ops[TLS_BASE][TLS_BASE] = *base;
837
838 ops[TLS_SW ][TLS_BASE] = ops[TLS_BASE][TLS_BASE];
839 ops[TLS_SW ][TLS_BASE].splice_eof = tls_sw_splice_eof;
840
841 ops[TLS_BASE][TLS_SW ] = ops[TLS_BASE][TLS_BASE];
842 ops[TLS_BASE][TLS_SW ].splice_read = tls_sw_splice_read;
843 ops[TLS_BASE][TLS_SW ].poll = tls_sk_poll;
844 ops[TLS_BASE][TLS_SW ].read_sock = tls_sw_read_sock;
845
846 ops[TLS_SW ][TLS_SW ] = ops[TLS_SW ][TLS_BASE];
847 ops[TLS_SW ][TLS_SW ].splice_read = tls_sw_splice_read;
848 ops[TLS_SW ][TLS_SW ].poll = tls_sk_poll;
849 ops[TLS_SW ][TLS_SW ].read_sock = tls_sw_read_sock;
850
851#ifdef CONFIG_TLS_DEVICE
852 ops[TLS_HW ][TLS_BASE] = ops[TLS_BASE][TLS_BASE];
853
854 ops[TLS_HW ][TLS_SW ] = ops[TLS_BASE][TLS_SW ];
855
856 ops[TLS_BASE][TLS_HW ] = ops[TLS_BASE][TLS_SW ];
857
858 ops[TLS_SW ][TLS_HW ] = ops[TLS_SW ][TLS_SW ];
859
860 ops[TLS_HW ][TLS_HW ] = ops[TLS_HW ][TLS_SW ];
861#endif
862#ifdef CONFIG_TLS_TOE
863 ops[TLS_HW_RECORD][TLS_HW_RECORD] = *base;
864#endif
865}
866
867static void tls_build_proto(struct sock *sk)
868{
869 int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4;
870 struct proto *prot = READ_ONCE(sk->sk_prot);
871
872 /* Build IPv6 TLS whenever the address of tcpv6 _prot changes */
873 if (ip_ver == TLSV6 &&
874 unlikely(prot != smp_load_acquire(&saved_tcpv6_prot))) {
875 mutex_lock(&tcpv6_prot_mutex);
876 if (likely(prot != saved_tcpv6_prot)) {
877 build_protos(tls_prots[TLSV6], prot);
878 build_proto_ops(tls_proto_ops[TLSV6],
879 sk->sk_socket->ops);
880 smp_store_release(&saved_tcpv6_prot, prot);
881 }
882 mutex_unlock(&tcpv6_prot_mutex);
883 }
884
885 if (ip_ver == TLSV4 &&
886 unlikely(prot != smp_load_acquire(&saved_tcpv4_prot))) {
887 mutex_lock(&tcpv4_prot_mutex);
888 if (likely(prot != saved_tcpv4_prot)) {
889 build_protos(tls_prots[TLSV4], prot);
890 build_proto_ops(tls_proto_ops[TLSV4],
891 sk->sk_socket->ops);
892 smp_store_release(&saved_tcpv4_prot, prot);
893 }
894 mutex_unlock(&tcpv4_prot_mutex);
895 }
896}
897
898static void build_protos(struct proto prot[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
899 const struct proto *base)
900{
901 prot[TLS_BASE][TLS_BASE] = *base;
902 prot[TLS_BASE][TLS_BASE].setsockopt = tls_setsockopt;
903 prot[TLS_BASE][TLS_BASE].getsockopt = tls_getsockopt;
904 prot[TLS_BASE][TLS_BASE].close = tls_sk_proto_close;
905
906 prot[TLS_SW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
907 prot[TLS_SW][TLS_BASE].sendmsg = tls_sw_sendmsg;
908 prot[TLS_SW][TLS_BASE].splice_eof = tls_sw_splice_eof;
909
910 prot[TLS_BASE][TLS_SW] = prot[TLS_BASE][TLS_BASE];
911 prot[TLS_BASE][TLS_SW].recvmsg = tls_sw_recvmsg;
912 prot[TLS_BASE][TLS_SW].sock_is_readable = tls_sw_sock_is_readable;
913 prot[TLS_BASE][TLS_SW].close = tls_sk_proto_close;
914
915 prot[TLS_SW][TLS_SW] = prot[TLS_SW][TLS_BASE];
916 prot[TLS_SW][TLS_SW].recvmsg = tls_sw_recvmsg;
917 prot[TLS_SW][TLS_SW].sock_is_readable = tls_sw_sock_is_readable;
918 prot[TLS_SW][TLS_SW].close = tls_sk_proto_close;
919
920#ifdef CONFIG_TLS_DEVICE
921 prot[TLS_HW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
922 prot[TLS_HW][TLS_BASE].sendmsg = tls_device_sendmsg;
923 prot[TLS_HW][TLS_BASE].splice_eof = tls_device_splice_eof;
924
925 prot[TLS_HW][TLS_SW] = prot[TLS_BASE][TLS_SW];
926 prot[TLS_HW][TLS_SW].sendmsg = tls_device_sendmsg;
927 prot[TLS_HW][TLS_SW].splice_eof = tls_device_splice_eof;
928
929 prot[TLS_BASE][TLS_HW] = prot[TLS_BASE][TLS_SW];
930
931 prot[TLS_SW][TLS_HW] = prot[TLS_SW][TLS_SW];
932
933 prot[TLS_HW][TLS_HW] = prot[TLS_HW][TLS_SW];
934#endif
935#ifdef CONFIG_TLS_TOE
936 prot[TLS_HW_RECORD][TLS_HW_RECORD] = *base;
937 prot[TLS_HW_RECORD][TLS_HW_RECORD].hash = tls_toe_hash;
938 prot[TLS_HW_RECORD][TLS_HW_RECORD].unhash = tls_toe_unhash;
939#endif
940}
941
942static int tls_init(struct sock *sk)
943{
944 struct tls_context *ctx;
945 int rc = 0;
946
947 tls_build_proto(sk);
948
949#ifdef CONFIG_TLS_TOE
950 if (tls_toe_bypass(sk))
951 return 0;
952#endif
953
954 /* The TLS ulp is currently supported only for TCP sockets
955 * in ESTABLISHED state.
956 * Supporting sockets in LISTEN state will require us
957 * to modify the accept implementation to clone rather then
958 * share the ulp context.
959 */
960 if (sk->sk_state != TCP_ESTABLISHED)
961 return -ENOTCONN;
962
963 /* allocate tls context */
964 write_lock_bh(&sk->sk_callback_lock);
965 ctx = tls_ctx_create(sk);
966 if (!ctx) {
967 rc = -ENOMEM;
968 goto out;
969 }
970
971 ctx->tx_conf = TLS_BASE;
972 ctx->rx_conf = TLS_BASE;
973 update_sk_prot(sk, ctx);
974out:
975 write_unlock_bh(&sk->sk_callback_lock);
976 return rc;
977}
978
979static void tls_update(struct sock *sk, struct proto *p,
980 void (*write_space)(struct sock *sk))
981{
982 struct tls_context *ctx;
983
984 WARN_ON_ONCE(sk->sk_prot == p);
985
986 ctx = tls_get_ctx(sk);
987 if (likely(ctx)) {
988 ctx->sk_write_space = write_space;
989 ctx->sk_proto = p;
990 } else {
991 /* Pairs with lockless read in sk_clone_lock(). */
992 WRITE_ONCE(sk->sk_prot, p);
993 sk->sk_write_space = write_space;
994 }
995}
996
997static u16 tls_user_config(struct tls_context *ctx, bool tx)
998{
999 u16 config = tx ? ctx->tx_conf : ctx->rx_conf;
1000
1001 switch (config) {
1002 case TLS_BASE:
1003 return TLS_CONF_BASE;
1004 case TLS_SW:
1005 return TLS_CONF_SW;
1006 case TLS_HW:
1007 return TLS_CONF_HW;
1008 case TLS_HW_RECORD:
1009 return TLS_CONF_HW_RECORD;
1010 }
1011 return 0;
1012}
1013
1014static int tls_get_info(struct sock *sk, struct sk_buff *skb)
1015{
1016 u16 version, cipher_type;
1017 struct tls_context *ctx;
1018 struct nlattr *start;
1019 int err;
1020
1021 start = nla_nest_start_noflag(skb, INET_ULP_INFO_TLS);
1022 if (!start)
1023 return -EMSGSIZE;
1024
1025 rcu_read_lock();
1026 ctx = rcu_dereference(inet_csk(sk)->icsk_ulp_data);
1027 if (!ctx) {
1028 err = 0;
1029 goto nla_failure;
1030 }
1031 version = ctx->prot_info.version;
1032 if (version) {
1033 err = nla_put_u16(skb, TLS_INFO_VERSION, version);
1034 if (err)
1035 goto nla_failure;
1036 }
1037 cipher_type = ctx->prot_info.cipher_type;
1038 if (cipher_type) {
1039 err = nla_put_u16(skb, TLS_INFO_CIPHER, cipher_type);
1040 if (err)
1041 goto nla_failure;
1042 }
1043 err = nla_put_u16(skb, TLS_INFO_TXCONF, tls_user_config(ctx, true));
1044 if (err)
1045 goto nla_failure;
1046
1047 err = nla_put_u16(skb, TLS_INFO_RXCONF, tls_user_config(ctx, false));
1048 if (err)
1049 goto nla_failure;
1050
1051 if (ctx->tx_conf == TLS_HW && ctx->zerocopy_sendfile) {
1052 err = nla_put_flag(skb, TLS_INFO_ZC_RO_TX);
1053 if (err)
1054 goto nla_failure;
1055 }
1056 if (ctx->rx_no_pad) {
1057 err = nla_put_flag(skb, TLS_INFO_RX_NO_PAD);
1058 if (err)
1059 goto nla_failure;
1060 }
1061
1062 rcu_read_unlock();
1063 nla_nest_end(skb, start);
1064 return 0;
1065
1066nla_failure:
1067 rcu_read_unlock();
1068 nla_nest_cancel(skb, start);
1069 return err;
1070}
1071
1072static size_t tls_get_info_size(const struct sock *sk)
1073{
1074 size_t size = 0;
1075
1076 size += nla_total_size(0) + /* INET_ULP_INFO_TLS */
1077 nla_total_size(sizeof(u16)) + /* TLS_INFO_VERSION */
1078 nla_total_size(sizeof(u16)) + /* TLS_INFO_CIPHER */
1079 nla_total_size(sizeof(u16)) + /* TLS_INFO_RXCONF */
1080 nla_total_size(sizeof(u16)) + /* TLS_INFO_TXCONF */
1081 nla_total_size(0) + /* TLS_INFO_ZC_RO_TX */
1082 nla_total_size(0) + /* TLS_INFO_RX_NO_PAD */
1083 0;
1084
1085 return size;
1086}
1087
1088static int __net_init tls_init_net(struct net *net)
1089{
1090 int err;
1091
1092 net->mib.tls_statistics = alloc_percpu(struct linux_tls_mib);
1093 if (!net->mib.tls_statistics)
1094 return -ENOMEM;
1095
1096 err = tls_proc_init(net);
1097 if (err)
1098 goto err_free_stats;
1099
1100 return 0;
1101err_free_stats:
1102 free_percpu(net->mib.tls_statistics);
1103 return err;
1104}
1105
1106static void __net_exit tls_exit_net(struct net *net)
1107{
1108 tls_proc_fini(net);
1109 free_percpu(net->mib.tls_statistics);
1110}
1111
1112static struct pernet_operations tls_proc_ops = {
1113 .init = tls_init_net,
1114 .exit = tls_exit_net,
1115};
1116
1117static struct tcp_ulp_ops tcp_tls_ulp_ops __read_mostly = {
1118 .name = "tls",
1119 .owner = THIS_MODULE,
1120 .init = tls_init,
1121 .update = tls_update,
1122 .get_info = tls_get_info,
1123 .get_info_size = tls_get_info_size,
1124};
1125
1126static int __init tls_register(void)
1127{
1128 int err;
1129
1130 err = register_pernet_subsys(&tls_proc_ops);
1131 if (err)
1132 return err;
1133
1134 err = tls_strp_dev_init();
1135 if (err)
1136 goto err_pernet;
1137
1138 err = tls_device_init();
1139 if (err)
1140 goto err_strp;
1141
1142 tcp_register_ulp(&tcp_tls_ulp_ops);
1143
1144 return 0;
1145err_strp:
1146 tls_strp_dev_exit();
1147err_pernet:
1148 unregister_pernet_subsys(&tls_proc_ops);
1149 return err;
1150}
1151
1152static void __exit tls_unregister(void)
1153{
1154 tcp_unregister_ulp(&tcp_tls_ulp_ops);
1155 tls_strp_dev_exit();
1156 tls_device_cleanup();
1157 unregister_pernet_subsys(&tls_proc_ops);
1158}
1159
1160module_init(tls_register);
1161module_exit(tls_unregister);