xref: /openbmc/linux/net/tls/tls_main.c (revision 72fa0f65)
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 MODULE_AUTHOR("Mellanox Technologies");
49 MODULE_DESCRIPTION("Transport Layer Security Support");
50 MODULE_LICENSE("Dual BSD/GPL");
51 MODULE_ALIAS_TCP_ULP("tls");
52 
53 enum {
54 	TLSV4,
55 	TLSV6,
56 	TLS_NUM_PROTS,
57 };
58 
59 static const struct proto *saved_tcpv6_prot;
60 static DEFINE_MUTEX(tcpv6_prot_mutex);
61 static const struct proto *saved_tcpv4_prot;
62 static DEFINE_MUTEX(tcpv4_prot_mutex);
63 static struct proto tls_prots[TLS_NUM_PROTS][TLS_NUM_CONFIG][TLS_NUM_CONFIG];
64 static struct proto_ops tls_proto_ops[TLS_NUM_PROTS][TLS_NUM_CONFIG][TLS_NUM_CONFIG];
65 static void build_protos(struct proto prot[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
66 			 const struct proto *base);
67 
68 void update_sk_prot(struct sock *sk, struct tls_context *ctx)
69 {
70 	int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4;
71 
72 	WRITE_ONCE(sk->sk_prot,
73 		   &tls_prots[ip_ver][ctx->tx_conf][ctx->rx_conf]);
74 	WRITE_ONCE(sk->sk_socket->ops,
75 		   &tls_proto_ops[ip_ver][ctx->tx_conf][ctx->rx_conf]);
76 }
77 
78 int 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 
102 int 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);
125 retry:
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 	ctx->in_tcp_sendpages = false;
153 
154 	return 0;
155 }
156 
157 static int tls_handle_open_record(struct sock *sk, int flags)
158 {
159 	struct tls_context *ctx = tls_get_ctx(sk);
160 
161 	if (tls_is_pending_open_record(ctx))
162 		return ctx->push_pending_record(sk, flags);
163 
164 	return 0;
165 }
166 
167 int tls_proccess_cmsg(struct sock *sk, struct msghdr *msg,
168 		      unsigned char *record_type)
169 {
170 	struct cmsghdr *cmsg;
171 	int rc = -EINVAL;
172 
173 	for_each_cmsghdr(cmsg, msg) {
174 		if (!CMSG_OK(msg, cmsg))
175 			return -EINVAL;
176 		if (cmsg->cmsg_level != SOL_TLS)
177 			continue;
178 
179 		switch (cmsg->cmsg_type) {
180 		case TLS_SET_RECORD_TYPE:
181 			if (cmsg->cmsg_len < CMSG_LEN(sizeof(*record_type)))
182 				return -EINVAL;
183 
184 			if (msg->msg_flags & MSG_MORE)
185 				return -EINVAL;
186 
187 			rc = tls_handle_open_record(sk, msg->msg_flags);
188 			if (rc)
189 				return rc;
190 
191 			*record_type = *(unsigned char *)CMSG_DATA(cmsg);
192 			rc = 0;
193 			break;
194 		default:
195 			return -EINVAL;
196 		}
197 	}
198 
199 	return rc;
200 }
201 
202 int tls_push_partial_record(struct sock *sk, struct tls_context *ctx,
203 			    int flags)
204 {
205 	struct scatterlist *sg;
206 	u16 offset;
207 
208 	sg = ctx->partially_sent_record;
209 	offset = ctx->partially_sent_offset;
210 
211 	ctx->partially_sent_record = NULL;
212 	return tls_push_sg(sk, ctx, sg, offset, flags);
213 }
214 
215 void tls_free_partial_record(struct sock *sk, struct tls_context *ctx)
216 {
217 	struct scatterlist *sg;
218 
219 	for (sg = ctx->partially_sent_record; sg; sg = sg_next(sg)) {
220 		put_page(sg_page(sg));
221 		sk_mem_uncharge(sk, sg->length);
222 	}
223 	ctx->partially_sent_record = NULL;
224 }
225 
226 static void tls_write_space(struct sock *sk)
227 {
228 	struct tls_context *ctx = tls_get_ctx(sk);
229 
230 	/* If in_tcp_sendpages call lower protocol write space handler
231 	 * to ensure we wake up any waiting operations there. For example
232 	 * if do_tcp_sendpages where to call sk_wait_event.
233 	 */
234 	if (ctx->in_tcp_sendpages) {
235 		ctx->sk_write_space(sk);
236 		return;
237 	}
238 
239 #ifdef CONFIG_TLS_DEVICE
240 	if (ctx->tx_conf == TLS_HW)
241 		tls_device_write_space(sk, ctx);
242 	else
243 #endif
244 		tls_sw_write_space(sk, ctx);
245 
246 	ctx->sk_write_space(sk);
247 }
248 
249 /**
250  * tls_ctx_free() - free TLS ULP context
251  * @sk:  socket to with @ctx is attached
252  * @ctx: TLS context structure
253  *
254  * Free TLS context. If @sk is %NULL caller guarantees that the socket
255  * to which @ctx was attached has no outstanding references.
256  */
257 void tls_ctx_free(struct sock *sk, struct tls_context *ctx)
258 {
259 	if (!ctx)
260 		return;
261 
262 	memzero_explicit(&ctx->crypto_send, sizeof(ctx->crypto_send));
263 	memzero_explicit(&ctx->crypto_recv, sizeof(ctx->crypto_recv));
264 	mutex_destroy(&ctx->tx_lock);
265 
266 	if (sk)
267 		kfree_rcu(ctx, rcu);
268 	else
269 		kfree(ctx);
270 }
271 
272 static void tls_sk_proto_cleanup(struct sock *sk,
273 				 struct tls_context *ctx, long timeo)
274 {
275 	if (unlikely(sk->sk_write_pending) &&
276 	    !wait_on_pending_writer(sk, &timeo))
277 		tls_handle_open_record(sk, 0);
278 
279 	/* We need these for tls_sw_fallback handling of other packets */
280 	if (ctx->tx_conf == TLS_SW) {
281 		kfree(ctx->tx.rec_seq);
282 		kfree(ctx->tx.iv);
283 		tls_sw_release_resources_tx(sk);
284 		TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXSW);
285 	} else if (ctx->tx_conf == TLS_HW) {
286 		tls_device_free_resources_tx(sk);
287 		TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXDEVICE);
288 	}
289 
290 	if (ctx->rx_conf == TLS_SW) {
291 		tls_sw_release_resources_rx(sk);
292 		TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXSW);
293 	} else if (ctx->rx_conf == TLS_HW) {
294 		tls_device_offload_cleanup_rx(sk);
295 		TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXDEVICE);
296 	}
297 }
298 
299 static void tls_sk_proto_close(struct sock *sk, long timeout)
300 {
301 	struct inet_connection_sock *icsk = inet_csk(sk);
302 	struct tls_context *ctx = tls_get_ctx(sk);
303 	long timeo = sock_sndtimeo(sk, 0);
304 	bool free_ctx;
305 
306 	if (ctx->tx_conf == TLS_SW)
307 		tls_sw_cancel_work_tx(ctx);
308 
309 	lock_sock(sk);
310 	free_ctx = ctx->tx_conf != TLS_HW && ctx->rx_conf != TLS_HW;
311 
312 	if (ctx->tx_conf != TLS_BASE || ctx->rx_conf != TLS_BASE)
313 		tls_sk_proto_cleanup(sk, ctx, timeo);
314 
315 	write_lock_bh(&sk->sk_callback_lock);
316 	if (free_ctx)
317 		rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
318 	WRITE_ONCE(sk->sk_prot, ctx->sk_proto);
319 	if (sk->sk_write_space == tls_write_space)
320 		sk->sk_write_space = ctx->sk_write_space;
321 	write_unlock_bh(&sk->sk_callback_lock);
322 	release_sock(sk);
323 	if (ctx->tx_conf == TLS_SW)
324 		tls_sw_free_ctx_tx(ctx);
325 	if (ctx->rx_conf == TLS_SW || ctx->rx_conf == TLS_HW)
326 		tls_sw_strparser_done(ctx);
327 	if (ctx->rx_conf == TLS_SW)
328 		tls_sw_free_ctx_rx(ctx);
329 	ctx->sk_proto->close(sk, timeout);
330 
331 	if (free_ctx)
332 		tls_ctx_free(sk, ctx);
333 }
334 
335 static int do_tls_getsockopt_conf(struct sock *sk, char __user *optval,
336 				  int __user *optlen, int tx)
337 {
338 	int rc = 0;
339 	struct tls_context *ctx = tls_get_ctx(sk);
340 	struct tls_crypto_info *crypto_info;
341 	struct cipher_context *cctx;
342 	int len;
343 
344 	if (get_user(len, optlen))
345 		return -EFAULT;
346 
347 	if (!optval || (len < sizeof(*crypto_info))) {
348 		rc = -EINVAL;
349 		goto out;
350 	}
351 
352 	if (!ctx) {
353 		rc = -EBUSY;
354 		goto out;
355 	}
356 
357 	/* get user crypto info */
358 	if (tx) {
359 		crypto_info = &ctx->crypto_send.info;
360 		cctx = &ctx->tx;
361 	} else {
362 		crypto_info = &ctx->crypto_recv.info;
363 		cctx = &ctx->rx;
364 	}
365 
366 	if (!TLS_CRYPTO_INFO_READY(crypto_info)) {
367 		rc = -EBUSY;
368 		goto out;
369 	}
370 
371 	if (len == sizeof(*crypto_info)) {
372 		if (copy_to_user(optval, crypto_info, sizeof(*crypto_info)))
373 			rc = -EFAULT;
374 		goto out;
375 	}
376 
377 	switch (crypto_info->cipher_type) {
378 	case TLS_CIPHER_AES_GCM_128: {
379 		struct tls12_crypto_info_aes_gcm_128 *
380 		  crypto_info_aes_gcm_128 =
381 		  container_of(crypto_info,
382 			       struct tls12_crypto_info_aes_gcm_128,
383 			       info);
384 
385 		if (len != sizeof(*crypto_info_aes_gcm_128)) {
386 			rc = -EINVAL;
387 			goto out;
388 		}
389 		lock_sock(sk);
390 		memcpy(crypto_info_aes_gcm_128->iv,
391 		       cctx->iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
392 		       TLS_CIPHER_AES_GCM_128_IV_SIZE);
393 		memcpy(crypto_info_aes_gcm_128->rec_seq, cctx->rec_seq,
394 		       TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE);
395 		release_sock(sk);
396 		if (copy_to_user(optval,
397 				 crypto_info_aes_gcm_128,
398 				 sizeof(*crypto_info_aes_gcm_128)))
399 			rc = -EFAULT;
400 		break;
401 	}
402 	case TLS_CIPHER_AES_GCM_256: {
403 		struct tls12_crypto_info_aes_gcm_256 *
404 		  crypto_info_aes_gcm_256 =
405 		  container_of(crypto_info,
406 			       struct tls12_crypto_info_aes_gcm_256,
407 			       info);
408 
409 		if (len != sizeof(*crypto_info_aes_gcm_256)) {
410 			rc = -EINVAL;
411 			goto out;
412 		}
413 		lock_sock(sk);
414 		memcpy(crypto_info_aes_gcm_256->iv,
415 		       cctx->iv + TLS_CIPHER_AES_GCM_256_SALT_SIZE,
416 		       TLS_CIPHER_AES_GCM_256_IV_SIZE);
417 		memcpy(crypto_info_aes_gcm_256->rec_seq, cctx->rec_seq,
418 		       TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE);
419 		release_sock(sk);
420 		if (copy_to_user(optval,
421 				 crypto_info_aes_gcm_256,
422 				 sizeof(*crypto_info_aes_gcm_256)))
423 			rc = -EFAULT;
424 		break;
425 	}
426 	default:
427 		rc = -EINVAL;
428 	}
429 
430 out:
431 	return rc;
432 }
433 
434 static int do_tls_getsockopt(struct sock *sk, int optname,
435 			     char __user *optval, int __user *optlen)
436 {
437 	int rc = 0;
438 
439 	switch (optname) {
440 	case TLS_TX:
441 	case TLS_RX:
442 		rc = do_tls_getsockopt_conf(sk, optval, optlen,
443 					    optname == TLS_TX);
444 		break;
445 	default:
446 		rc = -ENOPROTOOPT;
447 		break;
448 	}
449 	return rc;
450 }
451 
452 static int tls_getsockopt(struct sock *sk, int level, int optname,
453 			  char __user *optval, int __user *optlen)
454 {
455 	struct tls_context *ctx = tls_get_ctx(sk);
456 
457 	if (level != SOL_TLS)
458 		return ctx->sk_proto->getsockopt(sk, level,
459 						 optname, optval, optlen);
460 
461 	return do_tls_getsockopt(sk, optname, optval, optlen);
462 }
463 
464 static int do_tls_setsockopt_conf(struct sock *sk, sockptr_t optval,
465 				  unsigned int optlen, int tx)
466 {
467 	struct tls_crypto_info *crypto_info;
468 	struct tls_crypto_info *alt_crypto_info;
469 	struct tls_context *ctx = tls_get_ctx(sk);
470 	size_t optsize;
471 	int rc = 0;
472 	int conf;
473 
474 	if (sockptr_is_null(optval) || (optlen < sizeof(*crypto_info))) {
475 		rc = -EINVAL;
476 		goto out;
477 	}
478 
479 	if (tx) {
480 		crypto_info = &ctx->crypto_send.info;
481 		alt_crypto_info = &ctx->crypto_recv.info;
482 	} else {
483 		crypto_info = &ctx->crypto_recv.info;
484 		alt_crypto_info = &ctx->crypto_send.info;
485 	}
486 
487 	/* Currently we don't support set crypto info more than one time */
488 	if (TLS_CRYPTO_INFO_READY(crypto_info)) {
489 		rc = -EBUSY;
490 		goto out;
491 	}
492 
493 	rc = copy_from_sockptr(crypto_info, optval, sizeof(*crypto_info));
494 	if (rc) {
495 		rc = -EFAULT;
496 		goto err_crypto_info;
497 	}
498 
499 	/* check version */
500 	if (crypto_info->version != TLS_1_2_VERSION &&
501 	    crypto_info->version != TLS_1_3_VERSION) {
502 		rc = -EINVAL;
503 		goto err_crypto_info;
504 	}
505 
506 	/* Ensure that TLS version and ciphers are same in both directions */
507 	if (TLS_CRYPTO_INFO_READY(alt_crypto_info)) {
508 		if (alt_crypto_info->version != crypto_info->version ||
509 		    alt_crypto_info->cipher_type != crypto_info->cipher_type) {
510 			rc = -EINVAL;
511 			goto err_crypto_info;
512 		}
513 	}
514 
515 	switch (crypto_info->cipher_type) {
516 	case TLS_CIPHER_AES_GCM_128:
517 		optsize = sizeof(struct tls12_crypto_info_aes_gcm_128);
518 		break;
519 	case TLS_CIPHER_AES_GCM_256: {
520 		optsize = sizeof(struct tls12_crypto_info_aes_gcm_256);
521 		break;
522 	}
523 	case TLS_CIPHER_AES_CCM_128:
524 		optsize = sizeof(struct tls12_crypto_info_aes_ccm_128);
525 		break;
526 	case TLS_CIPHER_CHACHA20_POLY1305:
527 		optsize = sizeof(struct tls12_crypto_info_chacha20_poly1305);
528 		break;
529 	default:
530 		rc = -EINVAL;
531 		goto err_crypto_info;
532 	}
533 
534 	if (optlen != optsize) {
535 		rc = -EINVAL;
536 		goto err_crypto_info;
537 	}
538 
539 	rc = copy_from_sockptr_offset(crypto_info + 1, optval,
540 				      sizeof(*crypto_info),
541 				      optlen - sizeof(*crypto_info));
542 	if (rc) {
543 		rc = -EFAULT;
544 		goto err_crypto_info;
545 	}
546 
547 	if (tx) {
548 		rc = tls_set_device_offload(sk, ctx);
549 		conf = TLS_HW;
550 		if (!rc) {
551 			TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSTXDEVICE);
552 			TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXDEVICE);
553 		} else {
554 			rc = tls_set_sw_offload(sk, ctx, 1);
555 			if (rc)
556 				goto err_crypto_info;
557 			TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSTXSW);
558 			TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXSW);
559 			conf = TLS_SW;
560 		}
561 	} else {
562 		rc = tls_set_device_offload_rx(sk, ctx);
563 		conf = TLS_HW;
564 		if (!rc) {
565 			TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXDEVICE);
566 			TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXDEVICE);
567 		} else {
568 			rc = tls_set_sw_offload(sk, ctx, 0);
569 			if (rc)
570 				goto err_crypto_info;
571 			TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXSW);
572 			TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXSW);
573 			conf = TLS_SW;
574 		}
575 		tls_sw_strparser_arm(sk, ctx);
576 	}
577 
578 	if (tx)
579 		ctx->tx_conf = conf;
580 	else
581 		ctx->rx_conf = conf;
582 	update_sk_prot(sk, ctx);
583 	if (tx) {
584 		ctx->sk_write_space = sk->sk_write_space;
585 		sk->sk_write_space = tls_write_space;
586 	}
587 	goto out;
588 
589 err_crypto_info:
590 	memzero_explicit(crypto_info, sizeof(union tls_crypto_context));
591 out:
592 	return rc;
593 }
594 
595 static int do_tls_setsockopt(struct sock *sk, int optname, sockptr_t optval,
596 			     unsigned int optlen)
597 {
598 	int rc = 0;
599 
600 	switch (optname) {
601 	case TLS_TX:
602 	case TLS_RX:
603 		lock_sock(sk);
604 		rc = do_tls_setsockopt_conf(sk, optval, optlen,
605 					    optname == TLS_TX);
606 		release_sock(sk);
607 		break;
608 	default:
609 		rc = -ENOPROTOOPT;
610 		break;
611 	}
612 	return rc;
613 }
614 
615 static int tls_setsockopt(struct sock *sk, int level, int optname,
616 			  sockptr_t optval, unsigned int optlen)
617 {
618 	struct tls_context *ctx = tls_get_ctx(sk);
619 
620 	if (level != SOL_TLS)
621 		return ctx->sk_proto->setsockopt(sk, level, optname, optval,
622 						 optlen);
623 
624 	return do_tls_setsockopt(sk, optname, optval, optlen);
625 }
626 
627 struct tls_context *tls_ctx_create(struct sock *sk)
628 {
629 	struct inet_connection_sock *icsk = inet_csk(sk);
630 	struct tls_context *ctx;
631 
632 	ctx = kzalloc(sizeof(*ctx), GFP_ATOMIC);
633 	if (!ctx)
634 		return NULL;
635 
636 	mutex_init(&ctx->tx_lock);
637 	rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
638 	ctx->sk_proto = READ_ONCE(sk->sk_prot);
639 	ctx->sk = sk;
640 	return ctx;
641 }
642 
643 static void build_proto_ops(struct proto_ops ops[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
644 			    const struct proto_ops *base)
645 {
646 	ops[TLS_BASE][TLS_BASE] = *base;
647 
648 	ops[TLS_SW  ][TLS_BASE] = ops[TLS_BASE][TLS_BASE];
649 	ops[TLS_SW  ][TLS_BASE].sendpage_locked	= tls_sw_sendpage_locked;
650 
651 	ops[TLS_BASE][TLS_SW  ] = ops[TLS_BASE][TLS_BASE];
652 	ops[TLS_BASE][TLS_SW  ].splice_read	= tls_sw_splice_read;
653 
654 	ops[TLS_SW  ][TLS_SW  ] = ops[TLS_SW  ][TLS_BASE];
655 	ops[TLS_SW  ][TLS_SW  ].splice_read	= tls_sw_splice_read;
656 
657 #ifdef CONFIG_TLS_DEVICE
658 	ops[TLS_HW  ][TLS_BASE] = ops[TLS_BASE][TLS_BASE];
659 	ops[TLS_HW  ][TLS_BASE].sendpage_locked	= NULL;
660 
661 	ops[TLS_HW  ][TLS_SW  ] = ops[TLS_BASE][TLS_SW  ];
662 	ops[TLS_HW  ][TLS_SW  ].sendpage_locked	= NULL;
663 
664 	ops[TLS_BASE][TLS_HW  ] = ops[TLS_BASE][TLS_SW  ];
665 
666 	ops[TLS_SW  ][TLS_HW  ] = ops[TLS_SW  ][TLS_SW  ];
667 
668 	ops[TLS_HW  ][TLS_HW  ] = ops[TLS_HW  ][TLS_SW  ];
669 	ops[TLS_HW  ][TLS_HW  ].sendpage_locked	= NULL;
670 #endif
671 #ifdef CONFIG_TLS_TOE
672 	ops[TLS_HW_RECORD][TLS_HW_RECORD] = *base;
673 #endif
674 }
675 
676 static void tls_build_proto(struct sock *sk)
677 {
678 	int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4;
679 	struct proto *prot = READ_ONCE(sk->sk_prot);
680 
681 	/* Build IPv6 TLS whenever the address of tcpv6 _prot changes */
682 	if (ip_ver == TLSV6 &&
683 	    unlikely(prot != smp_load_acquire(&saved_tcpv6_prot))) {
684 		mutex_lock(&tcpv6_prot_mutex);
685 		if (likely(prot != saved_tcpv6_prot)) {
686 			build_protos(tls_prots[TLSV6], prot);
687 			build_proto_ops(tls_proto_ops[TLSV6],
688 					sk->sk_socket->ops);
689 			smp_store_release(&saved_tcpv6_prot, prot);
690 		}
691 		mutex_unlock(&tcpv6_prot_mutex);
692 	}
693 
694 	if (ip_ver == TLSV4 &&
695 	    unlikely(prot != smp_load_acquire(&saved_tcpv4_prot))) {
696 		mutex_lock(&tcpv4_prot_mutex);
697 		if (likely(prot != saved_tcpv4_prot)) {
698 			build_protos(tls_prots[TLSV4], prot);
699 			build_proto_ops(tls_proto_ops[TLSV4],
700 					sk->sk_socket->ops);
701 			smp_store_release(&saved_tcpv4_prot, prot);
702 		}
703 		mutex_unlock(&tcpv4_prot_mutex);
704 	}
705 }
706 
707 static void build_protos(struct proto prot[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
708 			 const struct proto *base)
709 {
710 	prot[TLS_BASE][TLS_BASE] = *base;
711 	prot[TLS_BASE][TLS_BASE].setsockopt	= tls_setsockopt;
712 	prot[TLS_BASE][TLS_BASE].getsockopt	= tls_getsockopt;
713 	prot[TLS_BASE][TLS_BASE].close		= tls_sk_proto_close;
714 
715 	prot[TLS_SW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
716 	prot[TLS_SW][TLS_BASE].sendmsg		= tls_sw_sendmsg;
717 	prot[TLS_SW][TLS_BASE].sendpage		= tls_sw_sendpage;
718 
719 	prot[TLS_BASE][TLS_SW] = prot[TLS_BASE][TLS_BASE];
720 	prot[TLS_BASE][TLS_SW].recvmsg		  = tls_sw_recvmsg;
721 	prot[TLS_BASE][TLS_SW].sock_is_readable   = tls_sw_sock_is_readable;
722 	prot[TLS_BASE][TLS_SW].close		  = tls_sk_proto_close;
723 
724 	prot[TLS_SW][TLS_SW] = prot[TLS_SW][TLS_BASE];
725 	prot[TLS_SW][TLS_SW].recvmsg		= tls_sw_recvmsg;
726 	prot[TLS_SW][TLS_SW].sock_is_readable   = tls_sw_sock_is_readable;
727 	prot[TLS_SW][TLS_SW].close		= tls_sk_proto_close;
728 
729 #ifdef CONFIG_TLS_DEVICE
730 	prot[TLS_HW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
731 	prot[TLS_HW][TLS_BASE].sendmsg		= tls_device_sendmsg;
732 	prot[TLS_HW][TLS_BASE].sendpage		= tls_device_sendpage;
733 
734 	prot[TLS_HW][TLS_SW] = prot[TLS_BASE][TLS_SW];
735 	prot[TLS_HW][TLS_SW].sendmsg		= tls_device_sendmsg;
736 	prot[TLS_HW][TLS_SW].sendpage		= tls_device_sendpage;
737 
738 	prot[TLS_BASE][TLS_HW] = prot[TLS_BASE][TLS_SW];
739 
740 	prot[TLS_SW][TLS_HW] = prot[TLS_SW][TLS_SW];
741 
742 	prot[TLS_HW][TLS_HW] = prot[TLS_HW][TLS_SW];
743 #endif
744 #ifdef CONFIG_TLS_TOE
745 	prot[TLS_HW_RECORD][TLS_HW_RECORD] = *base;
746 	prot[TLS_HW_RECORD][TLS_HW_RECORD].hash		= tls_toe_hash;
747 	prot[TLS_HW_RECORD][TLS_HW_RECORD].unhash	= tls_toe_unhash;
748 #endif
749 }
750 
751 static int tls_init(struct sock *sk)
752 {
753 	struct tls_context *ctx;
754 	int rc = 0;
755 
756 	tls_build_proto(sk);
757 
758 #ifdef CONFIG_TLS_TOE
759 	if (tls_toe_bypass(sk))
760 		return 0;
761 #endif
762 
763 	/* The TLS ulp is currently supported only for TCP sockets
764 	 * in ESTABLISHED state.
765 	 * Supporting sockets in LISTEN state will require us
766 	 * to modify the accept implementation to clone rather then
767 	 * share the ulp context.
768 	 */
769 	if (sk->sk_state != TCP_ESTABLISHED)
770 		return -ENOTCONN;
771 
772 	/* allocate tls context */
773 	write_lock_bh(&sk->sk_callback_lock);
774 	ctx = tls_ctx_create(sk);
775 	if (!ctx) {
776 		rc = -ENOMEM;
777 		goto out;
778 	}
779 
780 	ctx->tx_conf = TLS_BASE;
781 	ctx->rx_conf = TLS_BASE;
782 	update_sk_prot(sk, ctx);
783 out:
784 	write_unlock_bh(&sk->sk_callback_lock);
785 	return rc;
786 }
787 
788 static void tls_update(struct sock *sk, struct proto *p,
789 		       void (*write_space)(struct sock *sk))
790 {
791 	struct tls_context *ctx;
792 
793 	WARN_ON_ONCE(sk->sk_prot == p);
794 
795 	ctx = tls_get_ctx(sk);
796 	if (likely(ctx)) {
797 		ctx->sk_write_space = write_space;
798 		ctx->sk_proto = p;
799 	} else {
800 		/* Pairs with lockless read in sk_clone_lock(). */
801 		WRITE_ONCE(sk->sk_prot, p);
802 		sk->sk_write_space = write_space;
803 	}
804 }
805 
806 static int tls_get_info(const struct sock *sk, struct sk_buff *skb)
807 {
808 	u16 version, cipher_type;
809 	struct tls_context *ctx;
810 	struct nlattr *start;
811 	int err;
812 
813 	start = nla_nest_start_noflag(skb, INET_ULP_INFO_TLS);
814 	if (!start)
815 		return -EMSGSIZE;
816 
817 	rcu_read_lock();
818 	ctx = rcu_dereference(inet_csk(sk)->icsk_ulp_data);
819 	if (!ctx) {
820 		err = 0;
821 		goto nla_failure;
822 	}
823 	version = ctx->prot_info.version;
824 	if (version) {
825 		err = nla_put_u16(skb, TLS_INFO_VERSION, version);
826 		if (err)
827 			goto nla_failure;
828 	}
829 	cipher_type = ctx->prot_info.cipher_type;
830 	if (cipher_type) {
831 		err = nla_put_u16(skb, TLS_INFO_CIPHER, cipher_type);
832 		if (err)
833 			goto nla_failure;
834 	}
835 	err = nla_put_u16(skb, TLS_INFO_TXCONF, tls_user_config(ctx, true));
836 	if (err)
837 		goto nla_failure;
838 
839 	err = nla_put_u16(skb, TLS_INFO_RXCONF, tls_user_config(ctx, false));
840 	if (err)
841 		goto nla_failure;
842 
843 	rcu_read_unlock();
844 	nla_nest_end(skb, start);
845 	return 0;
846 
847 nla_failure:
848 	rcu_read_unlock();
849 	nla_nest_cancel(skb, start);
850 	return err;
851 }
852 
853 static size_t tls_get_info_size(const struct sock *sk)
854 {
855 	size_t size = 0;
856 
857 	size += nla_total_size(0) +		/* INET_ULP_INFO_TLS */
858 		nla_total_size(sizeof(u16)) +	/* TLS_INFO_VERSION */
859 		nla_total_size(sizeof(u16)) +	/* TLS_INFO_CIPHER */
860 		nla_total_size(sizeof(u16)) +	/* TLS_INFO_RXCONF */
861 		nla_total_size(sizeof(u16)) +	/* TLS_INFO_TXCONF */
862 		0;
863 
864 	return size;
865 }
866 
867 static int __net_init tls_init_net(struct net *net)
868 {
869 	int err;
870 
871 	net->mib.tls_statistics = alloc_percpu(struct linux_tls_mib);
872 	if (!net->mib.tls_statistics)
873 		return -ENOMEM;
874 
875 	err = tls_proc_init(net);
876 	if (err)
877 		goto err_free_stats;
878 
879 	return 0;
880 err_free_stats:
881 	free_percpu(net->mib.tls_statistics);
882 	return err;
883 }
884 
885 static void __net_exit tls_exit_net(struct net *net)
886 {
887 	tls_proc_fini(net);
888 	free_percpu(net->mib.tls_statistics);
889 }
890 
891 static struct pernet_operations tls_proc_ops = {
892 	.init = tls_init_net,
893 	.exit = tls_exit_net,
894 };
895 
896 static struct tcp_ulp_ops tcp_tls_ulp_ops __read_mostly = {
897 	.name			= "tls",
898 	.owner			= THIS_MODULE,
899 	.init			= tls_init,
900 	.update			= tls_update,
901 	.get_info		= tls_get_info,
902 	.get_info_size		= tls_get_info_size,
903 };
904 
905 static int __init tls_register(void)
906 {
907 	int err;
908 
909 	err = register_pernet_subsys(&tls_proc_ops);
910 	if (err)
911 		return err;
912 
913 	tls_device_init();
914 	tcp_register_ulp(&tcp_tls_ulp_ops);
915 
916 	return 0;
917 }
918 
919 static void __exit tls_unregister(void)
920 {
921 	tcp_unregister_ulp(&tcp_tls_ulp_ops);
922 	tls_device_cleanup();
923 	unregister_pernet_subsys(&tls_proc_ops);
924 }
925 
926 module_init(tls_register);
927 module_exit(tls_unregister);
928