2 * Copyright (c) 2016-2017, Mellanox Technologies. All rights reserved.
3 * Copyright (c) 2016-2017, Dave Watson <davejwatson@fb.com>. All rights reserved.
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:
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
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.
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
34 #include <linux/module.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>
45 MODULE_AUTHOR("Mellanox Technologies");
46 MODULE_DESCRIPTION("Transport Layer Security Support");
47 MODULE_LICENSE("Dual BSD/GPL");
48 MODULE_ALIAS_TCP_ULP("tls");
56 static struct proto *saved_tcpv6_prot;
57 static DEFINE_MUTEX(tcpv6_prot_mutex);
58 static struct proto *saved_tcpv4_prot;
59 static DEFINE_MUTEX(tcpv4_prot_mutex);
60 static LIST_HEAD(device_list);
61 static DEFINE_SPINLOCK(device_spinlock);
62 static struct proto tls_prots[TLS_NUM_PROTS][TLS_NUM_CONFIG][TLS_NUM_CONFIG];
63 static struct proto_ops tls_sw_proto_ops;
64 static void build_protos(struct proto prot[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
67 static void update_sk_prot(struct sock *sk, struct tls_context *ctx)
69 int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4;
71 sk->sk_prot = &tls_prots[ip_ver][ctx->tx_conf][ctx->rx_conf];
74 int wait_on_pending_writer(struct sock *sk, long *timeo)
77 DEFINE_WAIT_FUNC(wait, woken_wake_function);
79 add_wait_queue(sk_sleep(sk), &wait);
86 if (signal_pending(current)) {
87 rc = sock_intr_errno(*timeo);
91 if (sk_wait_event(sk, timeo, !sk->sk_write_pending, &wait))
94 remove_wait_queue(sk_sleep(sk), &wait);
98 int tls_push_sg(struct sock *sk,
99 struct tls_context *ctx,
100 struct scatterlist *sg,
104 int sendpage_flags = flags | MSG_SENDPAGE_NOTLAST;
108 int offset = first_offset;
110 size = sg->length - offset;
111 offset += sg->offset;
113 ctx->in_tcp_sendpages = true;
116 sendpage_flags = flags;
118 /* is sending application-limited? */
119 tcp_rate_check_app_limited(sk);
122 ret = do_tcp_sendpages(sk, p, offset, size, sendpage_flags);
131 offset -= sg->offset;
132 ctx->partially_sent_offset = offset;
133 ctx->partially_sent_record = (void *)sg;
134 ctx->in_tcp_sendpages = false;
139 sk_mem_uncharge(sk, sg->length);
148 ctx->in_tcp_sendpages = false;
153 static int tls_handle_open_record(struct sock *sk, int flags)
155 struct tls_context *ctx = tls_get_ctx(sk);
157 if (tls_is_pending_open_record(ctx))
158 return ctx->push_pending_record(sk, flags);
163 int tls_proccess_cmsg(struct sock *sk, struct msghdr *msg,
164 unsigned char *record_type)
166 struct cmsghdr *cmsg;
169 for_each_cmsghdr(cmsg, msg) {
170 if (!CMSG_OK(msg, cmsg))
172 if (cmsg->cmsg_level != SOL_TLS)
175 switch (cmsg->cmsg_type) {
176 case TLS_SET_RECORD_TYPE:
177 if (cmsg->cmsg_len < CMSG_LEN(sizeof(*record_type)))
180 if (msg->msg_flags & MSG_MORE)
183 rc = tls_handle_open_record(sk, msg->msg_flags);
187 *record_type = *(unsigned char *)CMSG_DATA(cmsg);
198 int tls_push_partial_record(struct sock *sk, struct tls_context *ctx,
201 struct scatterlist *sg;
204 sg = ctx->partially_sent_record;
205 offset = ctx->partially_sent_offset;
207 ctx->partially_sent_record = NULL;
208 return tls_push_sg(sk, ctx, sg, offset, flags);
211 static void tls_write_space(struct sock *sk)
213 struct tls_context *ctx = tls_get_ctx(sk);
215 /* If in_tcp_sendpages call lower protocol write space handler
216 * to ensure we wake up any waiting operations there. For example
217 * if do_tcp_sendpages where to call sk_wait_event.
219 if (ctx->in_tcp_sendpages) {
220 ctx->sk_write_space(sk);
224 #ifdef CONFIG_TLS_DEVICE
225 if (ctx->tx_conf == TLS_HW)
226 tls_device_write_space(sk, ctx);
229 tls_sw_write_space(sk, ctx);
231 ctx->sk_write_space(sk);
234 static void tls_ctx_free(struct tls_context *ctx)
239 memzero_explicit(&ctx->crypto_send, sizeof(ctx->crypto_send));
240 memzero_explicit(&ctx->crypto_recv, sizeof(ctx->crypto_recv));
244 static void tls_sk_proto_close(struct sock *sk, long timeout)
246 struct tls_context *ctx = tls_get_ctx(sk);
247 long timeo = sock_sndtimeo(sk, 0);
248 void (*sk_proto_close)(struct sock *sk, long timeout);
249 bool free_ctx = false;
252 sk_proto_close = ctx->sk_proto_close;
254 if (ctx->tx_conf == TLS_HW_RECORD && ctx->rx_conf == TLS_HW_RECORD)
255 goto skip_tx_cleanup;
257 if (ctx->tx_conf == TLS_BASE && ctx->rx_conf == TLS_BASE) {
259 goto skip_tx_cleanup;
262 if (!tls_complete_pending_work(sk, ctx, 0, &timeo))
263 tls_handle_open_record(sk, 0);
265 /* We need these for tls_sw_fallback handling of other packets */
266 if (ctx->tx_conf == TLS_SW) {
267 kfree(ctx->tx.rec_seq);
269 tls_sw_free_resources_tx(sk);
272 if (ctx->rx_conf == TLS_SW) {
273 kfree(ctx->rx.rec_seq);
275 tls_sw_free_resources_rx(sk);
278 #ifdef CONFIG_TLS_DEVICE
279 if (ctx->rx_conf == TLS_HW)
280 tls_device_offload_cleanup_rx(sk);
282 if (ctx->tx_conf != TLS_HW && ctx->rx_conf != TLS_HW) {
292 sk_proto_close(sk, timeout);
293 /* free ctx for TLS_HW_RECORD, used by tcp_set_state
294 * for sk->sk_prot->unhash [tls_hw_unhash]
300 static int do_tls_getsockopt_tx(struct sock *sk, char __user *optval,
304 struct tls_context *ctx = tls_get_ctx(sk);
305 struct tls_crypto_info *crypto_info;
308 if (get_user(len, optlen))
311 if (!optval || (len < sizeof(*crypto_info))) {
321 /* get user crypto info */
322 crypto_info = &ctx->crypto_send.info;
324 if (!TLS_CRYPTO_INFO_READY(crypto_info)) {
329 if (len == sizeof(*crypto_info)) {
330 if (copy_to_user(optval, crypto_info, sizeof(*crypto_info)))
335 switch (crypto_info->cipher_type) {
336 case TLS_CIPHER_AES_GCM_128: {
337 struct tls12_crypto_info_aes_gcm_128 *
338 crypto_info_aes_gcm_128 =
339 container_of(crypto_info,
340 struct tls12_crypto_info_aes_gcm_128,
343 if (len != sizeof(*crypto_info_aes_gcm_128)) {
348 memcpy(crypto_info_aes_gcm_128->iv,
349 ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
350 TLS_CIPHER_AES_GCM_128_IV_SIZE);
351 memcpy(crypto_info_aes_gcm_128->rec_seq, ctx->tx.rec_seq,
352 TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE);
354 if (copy_to_user(optval,
355 crypto_info_aes_gcm_128,
356 sizeof(*crypto_info_aes_gcm_128)))
360 case TLS_CIPHER_AES_GCM_256: {
361 struct tls12_crypto_info_aes_gcm_256 *
362 crypto_info_aes_gcm_256 =
363 container_of(crypto_info,
364 struct tls12_crypto_info_aes_gcm_256,
367 if (len != sizeof(*crypto_info_aes_gcm_256)) {
372 memcpy(crypto_info_aes_gcm_256->iv,
373 ctx->tx.iv + TLS_CIPHER_AES_GCM_256_SALT_SIZE,
374 TLS_CIPHER_AES_GCM_256_IV_SIZE);
375 memcpy(crypto_info_aes_gcm_256->rec_seq, ctx->tx.rec_seq,
376 TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE);
378 if (copy_to_user(optval,
379 crypto_info_aes_gcm_256,
380 sizeof(*crypto_info_aes_gcm_256)))
392 static int do_tls_getsockopt(struct sock *sk, int optname,
393 char __user *optval, int __user *optlen)
399 rc = do_tls_getsockopt_tx(sk, optval, optlen);
408 static int tls_getsockopt(struct sock *sk, int level, int optname,
409 char __user *optval, int __user *optlen)
411 struct tls_context *ctx = tls_get_ctx(sk);
413 if (level != SOL_TLS)
414 return ctx->getsockopt(sk, level, optname, optval, optlen);
416 return do_tls_getsockopt(sk, optname, optval, optlen);
419 static int do_tls_setsockopt_conf(struct sock *sk, char __user *optval,
420 unsigned int optlen, int tx)
422 struct tls_crypto_info *crypto_info;
423 struct tls_crypto_info *alt_crypto_info;
424 struct tls_context *ctx = tls_get_ctx(sk);
429 if (!optval || (optlen < sizeof(*crypto_info))) {
435 crypto_info = &ctx->crypto_send.info;
436 alt_crypto_info = &ctx->crypto_recv.info;
438 crypto_info = &ctx->crypto_recv.info;
439 alt_crypto_info = &ctx->crypto_send.info;
442 /* Currently we don't support set crypto info more than one time */
443 if (TLS_CRYPTO_INFO_READY(crypto_info)) {
448 rc = copy_from_user(crypto_info, optval, sizeof(*crypto_info));
451 goto err_crypto_info;
455 if (crypto_info->version != TLS_1_2_VERSION &&
456 crypto_info->version != TLS_1_3_VERSION) {
458 goto err_crypto_info;
461 /* Ensure that TLS version and ciphers are same in both directions */
462 if (TLS_CRYPTO_INFO_READY(alt_crypto_info)) {
463 if (alt_crypto_info->version != crypto_info->version ||
464 alt_crypto_info->cipher_type != crypto_info->cipher_type) {
466 goto err_crypto_info;
470 switch (crypto_info->cipher_type) {
471 case TLS_CIPHER_AES_GCM_128:
472 optsize = sizeof(struct tls12_crypto_info_aes_gcm_128);
474 case TLS_CIPHER_AES_GCM_256: {
475 optsize = sizeof(struct tls12_crypto_info_aes_gcm_256);
478 case TLS_CIPHER_AES_CCM_128:
479 optsize = sizeof(struct tls12_crypto_info_aes_ccm_128);
483 goto err_crypto_info;
486 if (optlen != optsize) {
488 goto err_crypto_info;
491 rc = copy_from_user(crypto_info + 1, optval + sizeof(*crypto_info),
492 optlen - sizeof(*crypto_info));
495 goto err_crypto_info;
499 #ifdef CONFIG_TLS_DEVICE
500 rc = tls_set_device_offload(sk, ctx);
506 rc = tls_set_sw_offload(sk, ctx, 1);
510 #ifdef CONFIG_TLS_DEVICE
511 rc = tls_set_device_offload_rx(sk, ctx);
517 rc = tls_set_sw_offload(sk, ctx, 0);
523 goto err_crypto_info;
529 update_sk_prot(sk, ctx);
531 ctx->sk_write_space = sk->sk_write_space;
532 sk->sk_write_space = tls_write_space;
534 sk->sk_socket->ops = &tls_sw_proto_ops;
539 memzero_explicit(crypto_info, sizeof(union tls_crypto_context));
544 static int do_tls_setsockopt(struct sock *sk, int optname,
545 char __user *optval, unsigned int optlen)
553 rc = do_tls_setsockopt_conf(sk, optval, optlen,
564 static int tls_setsockopt(struct sock *sk, int level, int optname,
565 char __user *optval, unsigned int optlen)
567 struct tls_context *ctx = tls_get_ctx(sk);
569 if (level != SOL_TLS)
570 return ctx->setsockopt(sk, level, optname, optval, optlen);
572 return do_tls_setsockopt(sk, optname, optval, optlen);
575 static struct tls_context *create_ctx(struct sock *sk)
577 struct inet_connection_sock *icsk = inet_csk(sk);
578 struct tls_context *ctx;
580 ctx = kzalloc(sizeof(*ctx), GFP_ATOMIC);
584 icsk->icsk_ulp_data = ctx;
585 ctx->setsockopt = sk->sk_prot->setsockopt;
586 ctx->getsockopt = sk->sk_prot->getsockopt;
587 ctx->sk_proto_close = sk->sk_prot->close;
591 static void tls_build_proto(struct sock *sk)
593 int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4;
595 /* Build IPv6 TLS whenever the address of tcpv6 _prot changes */
596 if (ip_ver == TLSV6 &&
597 unlikely(sk->sk_prot != smp_load_acquire(&saved_tcpv6_prot))) {
598 mutex_lock(&tcpv6_prot_mutex);
599 if (likely(sk->sk_prot != saved_tcpv6_prot)) {
600 build_protos(tls_prots[TLSV6], sk->sk_prot);
601 smp_store_release(&saved_tcpv6_prot, sk->sk_prot);
603 mutex_unlock(&tcpv6_prot_mutex);
606 if (ip_ver == TLSV4 &&
607 unlikely(sk->sk_prot != smp_load_acquire(&saved_tcpv4_prot))) {
608 mutex_lock(&tcpv4_prot_mutex);
609 if (likely(sk->sk_prot != saved_tcpv4_prot)) {
610 build_protos(tls_prots[TLSV4], sk->sk_prot);
611 smp_store_release(&saved_tcpv4_prot, sk->sk_prot);
613 mutex_unlock(&tcpv4_prot_mutex);
617 static void tls_hw_sk_destruct(struct sock *sk)
619 struct tls_context *ctx = tls_get_ctx(sk);
620 struct inet_connection_sock *icsk = inet_csk(sk);
622 ctx->sk_destruct(sk);
625 icsk->icsk_ulp_data = NULL;
628 static int tls_hw_prot(struct sock *sk)
630 struct tls_context *ctx;
631 struct tls_device *dev;
634 spin_lock_bh(&device_spinlock);
635 list_for_each_entry(dev, &device_list, dev_list) {
636 if (dev->feature && dev->feature(dev)) {
637 ctx = create_ctx(sk);
641 spin_unlock_bh(&device_spinlock);
643 ctx->hash = sk->sk_prot->hash;
644 ctx->unhash = sk->sk_prot->unhash;
645 ctx->sk_proto_close = sk->sk_prot->close;
646 ctx->sk_destruct = sk->sk_destruct;
647 sk->sk_destruct = tls_hw_sk_destruct;
648 ctx->rx_conf = TLS_HW_RECORD;
649 ctx->tx_conf = TLS_HW_RECORD;
650 update_sk_prot(sk, ctx);
651 spin_lock_bh(&device_spinlock);
657 spin_unlock_bh(&device_spinlock);
661 static void tls_hw_unhash(struct sock *sk)
663 struct tls_context *ctx = tls_get_ctx(sk);
664 struct tls_device *dev;
666 spin_lock_bh(&device_spinlock);
667 list_for_each_entry(dev, &device_list, dev_list) {
669 kref_get(&dev->kref);
670 spin_unlock_bh(&device_spinlock);
671 dev->unhash(dev, sk);
672 kref_put(&dev->kref, dev->release);
673 spin_lock_bh(&device_spinlock);
676 spin_unlock_bh(&device_spinlock);
680 static int tls_hw_hash(struct sock *sk)
682 struct tls_context *ctx = tls_get_ctx(sk);
683 struct tls_device *dev;
687 spin_lock_bh(&device_spinlock);
688 list_for_each_entry(dev, &device_list, dev_list) {
690 kref_get(&dev->kref);
691 spin_unlock_bh(&device_spinlock);
692 err |= dev->hash(dev, sk);
693 kref_put(&dev->kref, dev->release);
694 spin_lock_bh(&device_spinlock);
697 spin_unlock_bh(&device_spinlock);
704 static void build_protos(struct proto prot[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
707 prot[TLS_BASE][TLS_BASE] = *base;
708 prot[TLS_BASE][TLS_BASE].setsockopt = tls_setsockopt;
709 prot[TLS_BASE][TLS_BASE].getsockopt = tls_getsockopt;
710 prot[TLS_BASE][TLS_BASE].close = tls_sk_proto_close;
712 prot[TLS_SW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
713 prot[TLS_SW][TLS_BASE].sendmsg = tls_sw_sendmsg;
714 prot[TLS_SW][TLS_BASE].sendpage = tls_sw_sendpage;
716 prot[TLS_BASE][TLS_SW] = prot[TLS_BASE][TLS_BASE];
717 prot[TLS_BASE][TLS_SW].recvmsg = tls_sw_recvmsg;
718 prot[TLS_BASE][TLS_SW].stream_memory_read = tls_sw_stream_read;
719 prot[TLS_BASE][TLS_SW].close = tls_sk_proto_close;
721 prot[TLS_SW][TLS_SW] = prot[TLS_SW][TLS_BASE];
722 prot[TLS_SW][TLS_SW].recvmsg = tls_sw_recvmsg;
723 prot[TLS_SW][TLS_SW].stream_memory_read = tls_sw_stream_read;
724 prot[TLS_SW][TLS_SW].close = tls_sk_proto_close;
726 #ifdef CONFIG_TLS_DEVICE
727 prot[TLS_HW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
728 prot[TLS_HW][TLS_BASE].sendmsg = tls_device_sendmsg;
729 prot[TLS_HW][TLS_BASE].sendpage = tls_device_sendpage;
731 prot[TLS_HW][TLS_SW] = prot[TLS_BASE][TLS_SW];
732 prot[TLS_HW][TLS_SW].sendmsg = tls_device_sendmsg;
733 prot[TLS_HW][TLS_SW].sendpage = tls_device_sendpage;
735 prot[TLS_BASE][TLS_HW] = prot[TLS_BASE][TLS_SW];
737 prot[TLS_SW][TLS_HW] = prot[TLS_SW][TLS_SW];
739 prot[TLS_HW][TLS_HW] = prot[TLS_HW][TLS_SW];
742 prot[TLS_HW_RECORD][TLS_HW_RECORD] = *base;
743 prot[TLS_HW_RECORD][TLS_HW_RECORD].hash = tls_hw_hash;
744 prot[TLS_HW_RECORD][TLS_HW_RECORD].unhash = tls_hw_unhash;
745 prot[TLS_HW_RECORD][TLS_HW_RECORD].close = tls_sk_proto_close;
748 static int tls_init(struct sock *sk)
750 struct tls_context *ctx;
756 /* The TLS ulp is currently supported only for TCP sockets
757 * in ESTABLISHED state.
758 * Supporting sockets in LISTEN state will require us
759 * to modify the accept implementation to clone rather then
760 * share the ulp context.
762 if (sk->sk_state != TCP_ESTABLISHED)
765 /* allocate tls context */
766 ctx = create_ctx(sk);
773 ctx->tx_conf = TLS_BASE;
774 ctx->rx_conf = TLS_BASE;
775 update_sk_prot(sk, ctx);
780 void tls_register_device(struct tls_device *device)
782 spin_lock_bh(&device_spinlock);
783 list_add_tail(&device->dev_list, &device_list);
784 spin_unlock_bh(&device_spinlock);
786 EXPORT_SYMBOL(tls_register_device);
788 void tls_unregister_device(struct tls_device *device)
790 spin_lock_bh(&device_spinlock);
791 list_del(&device->dev_list);
792 spin_unlock_bh(&device_spinlock);
794 EXPORT_SYMBOL(tls_unregister_device);
796 static struct tcp_ulp_ops tcp_tls_ulp_ops __read_mostly = {
798 .owner = THIS_MODULE,
802 static int __init tls_register(void)
804 tls_sw_proto_ops = inet_stream_ops;
805 tls_sw_proto_ops.splice_read = tls_sw_splice_read;
807 #ifdef CONFIG_TLS_DEVICE
810 tcp_register_ulp(&tcp_tls_ulp_ops);
815 static void __exit tls_unregister(void)
817 tcp_unregister_ulp(&tcp_tls_ulp_ops);
818 #ifdef CONFIG_TLS_DEVICE
819 tls_device_cleanup();
823 module_init(tls_register);
824 module_exit(tls_unregister);