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 bool tls_free_partial_record(struct sock *sk, struct tls_context *ctx)
213 struct scatterlist *sg;
215 sg = ctx->partially_sent_record;
220 put_page(sg_page(sg));
221 sk_mem_uncharge(sk, sg->length);
227 ctx->partially_sent_record = NULL;
231 static void tls_write_space(struct sock *sk)
233 struct tls_context *ctx = tls_get_ctx(sk);
235 /* If in_tcp_sendpages call lower protocol write space handler
236 * to ensure we wake up any waiting operations there. For example
237 * if do_tcp_sendpages where to call sk_wait_event.
239 if (ctx->in_tcp_sendpages) {
240 ctx->sk_write_space(sk);
244 #ifdef CONFIG_TLS_DEVICE
245 if (ctx->tx_conf == TLS_HW)
246 tls_device_write_space(sk, ctx);
249 tls_sw_write_space(sk, ctx);
251 ctx->sk_write_space(sk);
255 * tls_ctx_free() - free TLS ULP context
256 * @sk: socket to with @ctx is attached
257 * @ctx: TLS context structure
259 * Free TLS context. If @sk is %NULL caller guarantees that the socket
260 * to which @ctx was attached has no outstanding references.
262 void tls_ctx_free(struct sock *sk, struct tls_context *ctx)
267 memzero_explicit(&ctx->crypto_send, sizeof(ctx->crypto_send));
268 memzero_explicit(&ctx->crypto_recv, sizeof(ctx->crypto_recv));
276 static void tls_sk_proto_cleanup(struct sock *sk,
277 struct tls_context *ctx, long timeo)
279 if (unlikely(sk->sk_write_pending) &&
280 !wait_on_pending_writer(sk, &timeo))
281 tls_handle_open_record(sk, 0);
283 /* We need these for tls_sw_fallback handling of other packets */
284 if (ctx->tx_conf == TLS_SW) {
285 kfree(ctx->tx.rec_seq);
287 tls_sw_release_resources_tx(sk);
288 #ifdef CONFIG_TLS_DEVICE
289 } else if (ctx->tx_conf == TLS_HW) {
290 tls_device_free_resources_tx(sk);
294 if (ctx->rx_conf == TLS_SW)
295 tls_sw_release_resources_rx(sk);
297 #ifdef CONFIG_TLS_DEVICE
298 if (ctx->rx_conf == TLS_HW)
299 tls_device_offload_cleanup_rx(sk);
303 static void tls_sk_proto_close(struct sock *sk, long timeout)
305 struct inet_connection_sock *icsk = inet_csk(sk);
306 struct tls_context *ctx = tls_get_ctx(sk);
307 long timeo = sock_sndtimeo(sk, 0);
310 if (ctx->tx_conf == TLS_SW)
311 tls_sw_cancel_work_tx(ctx);
314 free_ctx = ctx->tx_conf != TLS_HW && ctx->rx_conf != TLS_HW;
316 if (ctx->tx_conf != TLS_BASE || ctx->rx_conf != TLS_BASE)
317 tls_sk_proto_cleanup(sk, ctx, timeo);
319 write_lock_bh(&sk->sk_callback_lock);
321 rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
322 sk->sk_prot = ctx->sk_proto;
323 if (sk->sk_write_space == tls_write_space)
324 sk->sk_write_space = ctx->sk_write_space;
325 write_unlock_bh(&sk->sk_callback_lock);
327 if (ctx->tx_conf == TLS_SW)
328 tls_sw_free_ctx_tx(ctx);
329 if (ctx->rx_conf == TLS_SW || ctx->rx_conf == TLS_HW)
330 tls_sw_strparser_done(ctx);
331 if (ctx->rx_conf == TLS_SW)
332 tls_sw_free_ctx_rx(ctx);
333 ctx->sk_proto_close(sk, timeout);
336 tls_ctx_free(sk, ctx);
339 static int do_tls_getsockopt_tx(struct sock *sk, char __user *optval,
343 struct tls_context *ctx = tls_get_ctx(sk);
344 struct tls_crypto_info *crypto_info;
347 if (get_user(len, optlen))
350 if (!optval || (len < sizeof(*crypto_info))) {
360 /* get user crypto info */
361 crypto_info = &ctx->crypto_send.info;
363 if (!TLS_CRYPTO_INFO_READY(crypto_info)) {
368 if (len == sizeof(*crypto_info)) {
369 if (copy_to_user(optval, crypto_info, sizeof(*crypto_info)))
374 switch (crypto_info->cipher_type) {
375 case TLS_CIPHER_AES_GCM_128: {
376 struct tls12_crypto_info_aes_gcm_128 *
377 crypto_info_aes_gcm_128 =
378 container_of(crypto_info,
379 struct tls12_crypto_info_aes_gcm_128,
382 if (len != sizeof(*crypto_info_aes_gcm_128)) {
387 memcpy(crypto_info_aes_gcm_128->iv,
388 ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
389 TLS_CIPHER_AES_GCM_128_IV_SIZE);
390 memcpy(crypto_info_aes_gcm_128->rec_seq, ctx->tx.rec_seq,
391 TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE);
393 if (copy_to_user(optval,
394 crypto_info_aes_gcm_128,
395 sizeof(*crypto_info_aes_gcm_128)))
399 case TLS_CIPHER_AES_GCM_256: {
400 struct tls12_crypto_info_aes_gcm_256 *
401 crypto_info_aes_gcm_256 =
402 container_of(crypto_info,
403 struct tls12_crypto_info_aes_gcm_256,
406 if (len != sizeof(*crypto_info_aes_gcm_256)) {
411 memcpy(crypto_info_aes_gcm_256->iv,
412 ctx->tx.iv + TLS_CIPHER_AES_GCM_256_SALT_SIZE,
413 TLS_CIPHER_AES_GCM_256_IV_SIZE);
414 memcpy(crypto_info_aes_gcm_256->rec_seq, ctx->tx.rec_seq,
415 TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE);
417 if (copy_to_user(optval,
418 crypto_info_aes_gcm_256,
419 sizeof(*crypto_info_aes_gcm_256)))
431 static int do_tls_getsockopt(struct sock *sk, int optname,
432 char __user *optval, int __user *optlen)
438 rc = do_tls_getsockopt_tx(sk, optval, optlen);
447 static int tls_getsockopt(struct sock *sk, int level, int optname,
448 char __user *optval, int __user *optlen)
450 struct tls_context *ctx = tls_get_ctx(sk);
452 if (level != SOL_TLS)
453 return ctx->getsockopt(sk, level, optname, optval, optlen);
455 return do_tls_getsockopt(sk, optname, optval, optlen);
458 static int do_tls_setsockopt_conf(struct sock *sk, char __user *optval,
459 unsigned int optlen, int tx)
461 struct tls_crypto_info *crypto_info;
462 struct tls_crypto_info *alt_crypto_info;
463 struct tls_context *ctx = tls_get_ctx(sk);
468 if (!optval || (optlen < sizeof(*crypto_info))) {
474 crypto_info = &ctx->crypto_send.info;
475 alt_crypto_info = &ctx->crypto_recv.info;
477 crypto_info = &ctx->crypto_recv.info;
478 alt_crypto_info = &ctx->crypto_send.info;
481 /* Currently we don't support set crypto info more than one time */
482 if (TLS_CRYPTO_INFO_READY(crypto_info)) {
487 rc = copy_from_user(crypto_info, optval, sizeof(*crypto_info));
490 goto err_crypto_info;
494 if (crypto_info->version != TLS_1_2_VERSION &&
495 crypto_info->version != TLS_1_3_VERSION) {
497 goto err_crypto_info;
500 /* Ensure that TLS version and ciphers are same in both directions */
501 if (TLS_CRYPTO_INFO_READY(alt_crypto_info)) {
502 if (alt_crypto_info->version != crypto_info->version ||
503 alt_crypto_info->cipher_type != crypto_info->cipher_type) {
505 goto err_crypto_info;
509 switch (crypto_info->cipher_type) {
510 case TLS_CIPHER_AES_GCM_128:
511 optsize = sizeof(struct tls12_crypto_info_aes_gcm_128);
513 case TLS_CIPHER_AES_GCM_256: {
514 optsize = sizeof(struct tls12_crypto_info_aes_gcm_256);
517 case TLS_CIPHER_AES_CCM_128:
518 optsize = sizeof(struct tls12_crypto_info_aes_ccm_128);
522 goto err_crypto_info;
525 if (optlen != optsize) {
527 goto err_crypto_info;
530 rc = copy_from_user(crypto_info + 1, optval + sizeof(*crypto_info),
531 optlen - sizeof(*crypto_info));
534 goto err_crypto_info;
538 #ifdef CONFIG_TLS_DEVICE
539 rc = tls_set_device_offload(sk, ctx);
545 rc = tls_set_sw_offload(sk, ctx, 1);
547 goto err_crypto_info;
551 #ifdef CONFIG_TLS_DEVICE
552 rc = tls_set_device_offload_rx(sk, ctx);
558 rc = tls_set_sw_offload(sk, ctx, 0);
560 goto err_crypto_info;
563 tls_sw_strparser_arm(sk, ctx);
570 update_sk_prot(sk, ctx);
572 ctx->sk_write_space = sk->sk_write_space;
573 sk->sk_write_space = tls_write_space;
575 sk->sk_socket->ops = &tls_sw_proto_ops;
580 memzero_explicit(crypto_info, sizeof(union tls_crypto_context));
585 static int do_tls_setsockopt(struct sock *sk, int optname,
586 char __user *optval, unsigned int optlen)
594 rc = do_tls_setsockopt_conf(sk, optval, optlen,
605 static int tls_setsockopt(struct sock *sk, int level, int optname,
606 char __user *optval, unsigned int optlen)
608 struct tls_context *ctx = tls_get_ctx(sk);
610 if (level != SOL_TLS)
611 return ctx->setsockopt(sk, level, optname, optval, optlen);
613 return do_tls_setsockopt(sk, optname, optval, optlen);
616 static struct tls_context *create_ctx(struct sock *sk)
618 struct inet_connection_sock *icsk = inet_csk(sk);
619 struct tls_context *ctx;
621 ctx = kzalloc(sizeof(*ctx), GFP_ATOMIC);
625 rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
626 ctx->setsockopt = sk->sk_prot->setsockopt;
627 ctx->getsockopt = sk->sk_prot->getsockopt;
628 ctx->sk_proto_close = sk->sk_prot->close;
629 ctx->unhash = sk->sk_prot->unhash;
633 static void tls_build_proto(struct sock *sk)
635 int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4;
637 /* Build IPv6 TLS whenever the address of tcpv6 _prot changes */
638 if (ip_ver == TLSV6 &&
639 unlikely(sk->sk_prot != smp_load_acquire(&saved_tcpv6_prot))) {
640 mutex_lock(&tcpv6_prot_mutex);
641 if (likely(sk->sk_prot != saved_tcpv6_prot)) {
642 build_protos(tls_prots[TLSV6], sk->sk_prot);
643 smp_store_release(&saved_tcpv6_prot, sk->sk_prot);
645 mutex_unlock(&tcpv6_prot_mutex);
648 if (ip_ver == TLSV4 &&
649 unlikely(sk->sk_prot != smp_load_acquire(&saved_tcpv4_prot))) {
650 mutex_lock(&tcpv4_prot_mutex);
651 if (likely(sk->sk_prot != saved_tcpv4_prot)) {
652 build_protos(tls_prots[TLSV4], sk->sk_prot);
653 smp_store_release(&saved_tcpv4_prot, sk->sk_prot);
655 mutex_unlock(&tcpv4_prot_mutex);
659 static void tls_hw_sk_destruct(struct sock *sk)
661 struct tls_context *ctx = tls_get_ctx(sk);
662 struct inet_connection_sock *icsk = inet_csk(sk);
664 ctx->sk_destruct(sk);
666 rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
667 tls_ctx_free(sk, ctx);
670 static int tls_hw_prot(struct sock *sk)
672 struct tls_context *ctx;
673 struct tls_device *dev;
676 spin_lock_bh(&device_spinlock);
677 list_for_each_entry(dev, &device_list, dev_list) {
678 if (dev->feature && dev->feature(dev)) {
679 ctx = create_ctx(sk);
683 spin_unlock_bh(&device_spinlock);
685 ctx->hash = sk->sk_prot->hash;
686 ctx->unhash = sk->sk_prot->unhash;
687 ctx->sk_proto_close = sk->sk_prot->close;
688 ctx->sk_destruct = sk->sk_destruct;
689 sk->sk_destruct = tls_hw_sk_destruct;
690 ctx->rx_conf = TLS_HW_RECORD;
691 ctx->tx_conf = TLS_HW_RECORD;
692 update_sk_prot(sk, ctx);
693 spin_lock_bh(&device_spinlock);
699 spin_unlock_bh(&device_spinlock);
703 static void tls_hw_unhash(struct sock *sk)
705 struct tls_context *ctx = tls_get_ctx(sk);
706 struct tls_device *dev;
708 spin_lock_bh(&device_spinlock);
709 list_for_each_entry(dev, &device_list, dev_list) {
711 kref_get(&dev->kref);
712 spin_unlock_bh(&device_spinlock);
713 dev->unhash(dev, sk);
714 kref_put(&dev->kref, dev->release);
715 spin_lock_bh(&device_spinlock);
718 spin_unlock_bh(&device_spinlock);
722 static int tls_hw_hash(struct sock *sk)
724 struct tls_context *ctx = tls_get_ctx(sk);
725 struct tls_device *dev;
729 spin_lock_bh(&device_spinlock);
730 list_for_each_entry(dev, &device_list, dev_list) {
732 kref_get(&dev->kref);
733 spin_unlock_bh(&device_spinlock);
734 err |= dev->hash(dev, sk);
735 kref_put(&dev->kref, dev->release);
736 spin_lock_bh(&device_spinlock);
739 spin_unlock_bh(&device_spinlock);
746 static void build_protos(struct proto prot[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
749 prot[TLS_BASE][TLS_BASE] = *base;
750 prot[TLS_BASE][TLS_BASE].setsockopt = tls_setsockopt;
751 prot[TLS_BASE][TLS_BASE].getsockopt = tls_getsockopt;
752 prot[TLS_BASE][TLS_BASE].close = tls_sk_proto_close;
754 prot[TLS_SW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
755 prot[TLS_SW][TLS_BASE].sendmsg = tls_sw_sendmsg;
756 prot[TLS_SW][TLS_BASE].sendpage = tls_sw_sendpage;
758 prot[TLS_BASE][TLS_SW] = prot[TLS_BASE][TLS_BASE];
759 prot[TLS_BASE][TLS_SW].recvmsg = tls_sw_recvmsg;
760 prot[TLS_BASE][TLS_SW].stream_memory_read = tls_sw_stream_read;
761 prot[TLS_BASE][TLS_SW].close = tls_sk_proto_close;
763 prot[TLS_SW][TLS_SW] = prot[TLS_SW][TLS_BASE];
764 prot[TLS_SW][TLS_SW].recvmsg = tls_sw_recvmsg;
765 prot[TLS_SW][TLS_SW].stream_memory_read = tls_sw_stream_read;
766 prot[TLS_SW][TLS_SW].close = tls_sk_proto_close;
768 #ifdef CONFIG_TLS_DEVICE
769 prot[TLS_HW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
770 prot[TLS_HW][TLS_BASE].sendmsg = tls_device_sendmsg;
771 prot[TLS_HW][TLS_BASE].sendpage = tls_device_sendpage;
773 prot[TLS_HW][TLS_SW] = prot[TLS_BASE][TLS_SW];
774 prot[TLS_HW][TLS_SW].sendmsg = tls_device_sendmsg;
775 prot[TLS_HW][TLS_SW].sendpage = tls_device_sendpage;
777 prot[TLS_BASE][TLS_HW] = prot[TLS_BASE][TLS_SW];
779 prot[TLS_SW][TLS_HW] = prot[TLS_SW][TLS_SW];
781 prot[TLS_HW][TLS_HW] = prot[TLS_HW][TLS_SW];
784 prot[TLS_HW_RECORD][TLS_HW_RECORD] = *base;
785 prot[TLS_HW_RECORD][TLS_HW_RECORD].hash = tls_hw_hash;
786 prot[TLS_HW_RECORD][TLS_HW_RECORD].unhash = tls_hw_unhash;
789 static int tls_init(struct sock *sk)
791 struct tls_context *ctx;
797 /* The TLS ulp is currently supported only for TCP sockets
798 * in ESTABLISHED state.
799 * Supporting sockets in LISTEN state will require us
800 * to modify the accept implementation to clone rather then
801 * share the ulp context.
803 if (sk->sk_state != TCP_ESTABLISHED)
808 /* allocate tls context */
809 write_lock_bh(&sk->sk_callback_lock);
810 ctx = create_ctx(sk);
816 ctx->tx_conf = TLS_BASE;
817 ctx->rx_conf = TLS_BASE;
818 ctx->sk_proto = sk->sk_prot;
819 update_sk_prot(sk, ctx);
821 write_unlock_bh(&sk->sk_callback_lock);
825 static void tls_update(struct sock *sk, struct proto *p)
827 struct tls_context *ctx;
829 ctx = tls_get_ctx(sk);
831 ctx->sk_proto_close = p->close;
838 void tls_register_device(struct tls_device *device)
840 spin_lock_bh(&device_spinlock);
841 list_add_tail(&device->dev_list, &device_list);
842 spin_unlock_bh(&device_spinlock);
844 EXPORT_SYMBOL(tls_register_device);
846 void tls_unregister_device(struct tls_device *device)
848 spin_lock_bh(&device_spinlock);
849 list_del(&device->dev_list);
850 spin_unlock_bh(&device_spinlock);
852 EXPORT_SYMBOL(tls_unregister_device);
854 static struct tcp_ulp_ops tcp_tls_ulp_ops __read_mostly = {
856 .owner = THIS_MODULE,
858 .update = tls_update,
861 static int __init tls_register(void)
863 tls_sw_proto_ops = inet_stream_ops;
864 tls_sw_proto_ops.splice_read = tls_sw_splice_read;
866 #ifdef CONFIG_TLS_DEVICE
869 tcp_register_ulp(&tcp_tls_ulp_ops);
874 static void __exit tls_unregister(void)
876 tcp_unregister_ulp(&tcp_tls_ulp_ops);
877 #ifdef CONFIG_TLS_DEVICE
878 tls_device_cleanup();
882 module_init(tls_register);
883 module_exit(tls_unregister);