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);
254 void tls_ctx_free(struct tls_context *ctx)
259 memzero_explicit(&ctx->crypto_send, sizeof(ctx->crypto_send));
260 memzero_explicit(&ctx->crypto_recv, sizeof(ctx->crypto_recv));
264 static void tls_sk_proto_cleanup(struct sock *sk,
265 struct tls_context *ctx, long timeo)
267 if (unlikely(sk->sk_write_pending) &&
268 !wait_on_pending_writer(sk, &timeo))
269 tls_handle_open_record(sk, 0);
271 /* We need these for tls_sw_fallback handling of other packets */
272 if (ctx->tx_conf == TLS_SW) {
273 kfree(ctx->tx.rec_seq);
275 tls_sw_release_resources_tx(sk);
276 #ifdef CONFIG_TLS_DEVICE
277 } else if (ctx->tx_conf == TLS_HW) {
278 tls_device_free_resources_tx(sk);
282 if (ctx->rx_conf == TLS_SW)
283 tls_sw_release_resources_rx(sk);
285 #ifdef CONFIG_TLS_DEVICE
286 if (ctx->rx_conf == TLS_HW)
287 tls_device_offload_cleanup_rx(sk);
291 static void tls_sk_proto_close(struct sock *sk, long timeout)
293 struct inet_connection_sock *icsk = inet_csk(sk);
294 struct tls_context *ctx = tls_get_ctx(sk);
295 long timeo = sock_sndtimeo(sk, 0);
298 if (ctx->tx_conf == TLS_SW)
299 tls_sw_cancel_work_tx(ctx);
302 free_ctx = ctx->tx_conf != TLS_HW && ctx->rx_conf != TLS_HW;
304 if (ctx->tx_conf != TLS_BASE || ctx->rx_conf != TLS_BASE)
305 tls_sk_proto_cleanup(sk, ctx, timeo);
307 write_lock_bh(&sk->sk_callback_lock);
309 icsk->icsk_ulp_data = NULL;
310 sk->sk_prot = ctx->sk_proto;
311 if (sk->sk_write_space == tls_write_space)
312 sk->sk_write_space = ctx->sk_write_space;
313 write_unlock_bh(&sk->sk_callback_lock);
315 if (ctx->tx_conf == TLS_SW)
316 tls_sw_free_ctx_tx(ctx);
317 if (ctx->rx_conf == TLS_SW || ctx->rx_conf == TLS_HW)
318 tls_sw_strparser_done(ctx);
319 if (ctx->rx_conf == TLS_SW)
320 tls_sw_free_ctx_rx(ctx);
321 ctx->sk_proto_close(sk, timeout);
327 static int do_tls_getsockopt_tx(struct sock *sk, char __user *optval,
331 struct tls_context *ctx = tls_get_ctx(sk);
332 struct tls_crypto_info *crypto_info;
335 if (get_user(len, optlen))
338 if (!optval || (len < sizeof(*crypto_info))) {
348 /* get user crypto info */
349 crypto_info = &ctx->crypto_send.info;
351 if (!TLS_CRYPTO_INFO_READY(crypto_info)) {
356 if (len == sizeof(*crypto_info)) {
357 if (copy_to_user(optval, crypto_info, sizeof(*crypto_info)))
362 switch (crypto_info->cipher_type) {
363 case TLS_CIPHER_AES_GCM_128: {
364 struct tls12_crypto_info_aes_gcm_128 *
365 crypto_info_aes_gcm_128 =
366 container_of(crypto_info,
367 struct tls12_crypto_info_aes_gcm_128,
370 if (len != sizeof(*crypto_info_aes_gcm_128)) {
375 memcpy(crypto_info_aes_gcm_128->iv,
376 ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
377 TLS_CIPHER_AES_GCM_128_IV_SIZE);
378 memcpy(crypto_info_aes_gcm_128->rec_seq, ctx->tx.rec_seq,
379 TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE);
381 if (copy_to_user(optval,
382 crypto_info_aes_gcm_128,
383 sizeof(*crypto_info_aes_gcm_128)))
387 case TLS_CIPHER_AES_GCM_256: {
388 struct tls12_crypto_info_aes_gcm_256 *
389 crypto_info_aes_gcm_256 =
390 container_of(crypto_info,
391 struct tls12_crypto_info_aes_gcm_256,
394 if (len != sizeof(*crypto_info_aes_gcm_256)) {
399 memcpy(crypto_info_aes_gcm_256->iv,
400 ctx->tx.iv + TLS_CIPHER_AES_GCM_256_SALT_SIZE,
401 TLS_CIPHER_AES_GCM_256_IV_SIZE);
402 memcpy(crypto_info_aes_gcm_256->rec_seq, ctx->tx.rec_seq,
403 TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE);
405 if (copy_to_user(optval,
406 crypto_info_aes_gcm_256,
407 sizeof(*crypto_info_aes_gcm_256)))
419 static int do_tls_getsockopt(struct sock *sk, int optname,
420 char __user *optval, int __user *optlen)
426 rc = do_tls_getsockopt_tx(sk, optval, optlen);
435 static int tls_getsockopt(struct sock *sk, int level, int optname,
436 char __user *optval, int __user *optlen)
438 struct tls_context *ctx = tls_get_ctx(sk);
440 if (level != SOL_TLS)
441 return ctx->getsockopt(sk, level, optname, optval, optlen);
443 return do_tls_getsockopt(sk, optname, optval, optlen);
446 static int do_tls_setsockopt_conf(struct sock *sk, char __user *optval,
447 unsigned int optlen, int tx)
449 struct tls_crypto_info *crypto_info;
450 struct tls_crypto_info *alt_crypto_info;
451 struct tls_context *ctx = tls_get_ctx(sk);
456 if (!optval || (optlen < sizeof(*crypto_info))) {
462 crypto_info = &ctx->crypto_send.info;
463 alt_crypto_info = &ctx->crypto_recv.info;
465 crypto_info = &ctx->crypto_recv.info;
466 alt_crypto_info = &ctx->crypto_send.info;
469 /* Currently we don't support set crypto info more than one time */
470 if (TLS_CRYPTO_INFO_READY(crypto_info)) {
475 rc = copy_from_user(crypto_info, optval, sizeof(*crypto_info));
478 goto err_crypto_info;
482 if (crypto_info->version != TLS_1_2_VERSION &&
483 crypto_info->version != TLS_1_3_VERSION) {
485 goto err_crypto_info;
488 /* Ensure that TLS version and ciphers are same in both directions */
489 if (TLS_CRYPTO_INFO_READY(alt_crypto_info)) {
490 if (alt_crypto_info->version != crypto_info->version ||
491 alt_crypto_info->cipher_type != crypto_info->cipher_type) {
493 goto err_crypto_info;
497 switch (crypto_info->cipher_type) {
498 case TLS_CIPHER_AES_GCM_128:
499 optsize = sizeof(struct tls12_crypto_info_aes_gcm_128);
501 case TLS_CIPHER_AES_GCM_256: {
502 optsize = sizeof(struct tls12_crypto_info_aes_gcm_256);
505 case TLS_CIPHER_AES_CCM_128:
506 optsize = sizeof(struct tls12_crypto_info_aes_ccm_128);
510 goto err_crypto_info;
513 if (optlen != optsize) {
515 goto err_crypto_info;
518 rc = copy_from_user(crypto_info + 1, optval + sizeof(*crypto_info),
519 optlen - sizeof(*crypto_info));
522 goto err_crypto_info;
526 #ifdef CONFIG_TLS_DEVICE
527 rc = tls_set_device_offload(sk, ctx);
533 rc = tls_set_sw_offload(sk, ctx, 1);
535 goto err_crypto_info;
539 #ifdef CONFIG_TLS_DEVICE
540 rc = tls_set_device_offload_rx(sk, ctx);
546 rc = tls_set_sw_offload(sk, ctx, 0);
548 goto err_crypto_info;
551 tls_sw_strparser_arm(sk, ctx);
558 update_sk_prot(sk, ctx);
560 ctx->sk_write_space = sk->sk_write_space;
561 sk->sk_write_space = tls_write_space;
563 sk->sk_socket->ops = &tls_sw_proto_ops;
568 memzero_explicit(crypto_info, sizeof(union tls_crypto_context));
573 static int do_tls_setsockopt(struct sock *sk, int optname,
574 char __user *optval, unsigned int optlen)
582 rc = do_tls_setsockopt_conf(sk, optval, optlen,
593 static int tls_setsockopt(struct sock *sk, int level, int optname,
594 char __user *optval, unsigned int optlen)
596 struct tls_context *ctx = tls_get_ctx(sk);
598 if (level != SOL_TLS)
599 return ctx->setsockopt(sk, level, optname, optval, optlen);
601 return do_tls_setsockopt(sk, optname, optval, optlen);
604 static struct tls_context *create_ctx(struct sock *sk)
606 struct inet_connection_sock *icsk = inet_csk(sk);
607 struct tls_context *ctx;
609 ctx = kzalloc(sizeof(*ctx), GFP_ATOMIC);
613 icsk->icsk_ulp_data = ctx;
614 ctx->setsockopt = sk->sk_prot->setsockopt;
615 ctx->getsockopt = sk->sk_prot->getsockopt;
616 ctx->sk_proto_close = sk->sk_prot->close;
617 ctx->unhash = sk->sk_prot->unhash;
621 static void tls_build_proto(struct sock *sk)
623 int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4;
625 /* Build IPv6 TLS whenever the address of tcpv6 _prot changes */
626 if (ip_ver == TLSV6 &&
627 unlikely(sk->sk_prot != smp_load_acquire(&saved_tcpv6_prot))) {
628 mutex_lock(&tcpv6_prot_mutex);
629 if (likely(sk->sk_prot != saved_tcpv6_prot)) {
630 build_protos(tls_prots[TLSV6], sk->sk_prot);
631 smp_store_release(&saved_tcpv6_prot, sk->sk_prot);
633 mutex_unlock(&tcpv6_prot_mutex);
636 if (ip_ver == TLSV4 &&
637 unlikely(sk->sk_prot != smp_load_acquire(&saved_tcpv4_prot))) {
638 mutex_lock(&tcpv4_prot_mutex);
639 if (likely(sk->sk_prot != saved_tcpv4_prot)) {
640 build_protos(tls_prots[TLSV4], sk->sk_prot);
641 smp_store_release(&saved_tcpv4_prot, sk->sk_prot);
643 mutex_unlock(&tcpv4_prot_mutex);
647 static void tls_hw_sk_destruct(struct sock *sk)
649 struct tls_context *ctx = tls_get_ctx(sk);
650 struct inet_connection_sock *icsk = inet_csk(sk);
652 ctx->sk_destruct(sk);
655 icsk->icsk_ulp_data = NULL;
658 static int tls_hw_prot(struct sock *sk)
660 struct tls_context *ctx;
661 struct tls_device *dev;
664 spin_lock_bh(&device_spinlock);
665 list_for_each_entry(dev, &device_list, dev_list) {
666 if (dev->feature && dev->feature(dev)) {
667 ctx = create_ctx(sk);
671 spin_unlock_bh(&device_spinlock);
673 ctx->hash = sk->sk_prot->hash;
674 ctx->unhash = sk->sk_prot->unhash;
675 ctx->sk_proto_close = sk->sk_prot->close;
676 ctx->sk_destruct = sk->sk_destruct;
677 sk->sk_destruct = tls_hw_sk_destruct;
678 ctx->rx_conf = TLS_HW_RECORD;
679 ctx->tx_conf = TLS_HW_RECORD;
680 update_sk_prot(sk, ctx);
681 spin_lock_bh(&device_spinlock);
687 spin_unlock_bh(&device_spinlock);
691 static void tls_hw_unhash(struct sock *sk)
693 struct tls_context *ctx = tls_get_ctx(sk);
694 struct tls_device *dev;
696 spin_lock_bh(&device_spinlock);
697 list_for_each_entry(dev, &device_list, dev_list) {
699 kref_get(&dev->kref);
700 spin_unlock_bh(&device_spinlock);
701 dev->unhash(dev, sk);
702 kref_put(&dev->kref, dev->release);
703 spin_lock_bh(&device_spinlock);
706 spin_unlock_bh(&device_spinlock);
710 static int tls_hw_hash(struct sock *sk)
712 struct tls_context *ctx = tls_get_ctx(sk);
713 struct tls_device *dev;
717 spin_lock_bh(&device_spinlock);
718 list_for_each_entry(dev, &device_list, dev_list) {
720 kref_get(&dev->kref);
721 spin_unlock_bh(&device_spinlock);
722 err |= dev->hash(dev, sk);
723 kref_put(&dev->kref, dev->release);
724 spin_lock_bh(&device_spinlock);
727 spin_unlock_bh(&device_spinlock);
734 static void build_protos(struct proto prot[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
737 prot[TLS_BASE][TLS_BASE] = *base;
738 prot[TLS_BASE][TLS_BASE].setsockopt = tls_setsockopt;
739 prot[TLS_BASE][TLS_BASE].getsockopt = tls_getsockopt;
740 prot[TLS_BASE][TLS_BASE].close = tls_sk_proto_close;
742 prot[TLS_SW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
743 prot[TLS_SW][TLS_BASE].sendmsg = tls_sw_sendmsg;
744 prot[TLS_SW][TLS_BASE].sendpage = tls_sw_sendpage;
746 prot[TLS_BASE][TLS_SW] = prot[TLS_BASE][TLS_BASE];
747 prot[TLS_BASE][TLS_SW].recvmsg = tls_sw_recvmsg;
748 prot[TLS_BASE][TLS_SW].stream_memory_read = tls_sw_stream_read;
749 prot[TLS_BASE][TLS_SW].close = tls_sk_proto_close;
751 prot[TLS_SW][TLS_SW] = prot[TLS_SW][TLS_BASE];
752 prot[TLS_SW][TLS_SW].recvmsg = tls_sw_recvmsg;
753 prot[TLS_SW][TLS_SW].stream_memory_read = tls_sw_stream_read;
754 prot[TLS_SW][TLS_SW].close = tls_sk_proto_close;
756 #ifdef CONFIG_TLS_DEVICE
757 prot[TLS_HW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
758 prot[TLS_HW][TLS_BASE].sendmsg = tls_device_sendmsg;
759 prot[TLS_HW][TLS_BASE].sendpage = tls_device_sendpage;
761 prot[TLS_HW][TLS_SW] = prot[TLS_BASE][TLS_SW];
762 prot[TLS_HW][TLS_SW].sendmsg = tls_device_sendmsg;
763 prot[TLS_HW][TLS_SW].sendpage = tls_device_sendpage;
765 prot[TLS_BASE][TLS_HW] = prot[TLS_BASE][TLS_SW];
767 prot[TLS_SW][TLS_HW] = prot[TLS_SW][TLS_SW];
769 prot[TLS_HW][TLS_HW] = prot[TLS_HW][TLS_SW];
772 prot[TLS_HW_RECORD][TLS_HW_RECORD] = *base;
773 prot[TLS_HW_RECORD][TLS_HW_RECORD].hash = tls_hw_hash;
774 prot[TLS_HW_RECORD][TLS_HW_RECORD].unhash = tls_hw_unhash;
777 static int tls_init(struct sock *sk)
779 struct tls_context *ctx;
785 /* The TLS ulp is currently supported only for TCP sockets
786 * in ESTABLISHED state.
787 * Supporting sockets in LISTEN state will require us
788 * to modify the accept implementation to clone rather then
789 * share the ulp context.
791 if (sk->sk_state != TCP_ESTABLISHED)
796 /* allocate tls context */
797 write_lock_bh(&sk->sk_callback_lock);
798 ctx = create_ctx(sk);
804 ctx->tx_conf = TLS_BASE;
805 ctx->rx_conf = TLS_BASE;
806 ctx->sk_proto = sk->sk_prot;
807 update_sk_prot(sk, ctx);
809 write_unlock_bh(&sk->sk_callback_lock);
813 static void tls_update(struct sock *sk, struct proto *p)
815 struct tls_context *ctx;
817 ctx = tls_get_ctx(sk);
819 ctx->sk_proto_close = p->close;
826 void tls_register_device(struct tls_device *device)
828 spin_lock_bh(&device_spinlock);
829 list_add_tail(&device->dev_list, &device_list);
830 spin_unlock_bh(&device_spinlock);
832 EXPORT_SYMBOL(tls_register_device);
834 void tls_unregister_device(struct tls_device *device)
836 spin_lock_bh(&device_spinlock);
837 list_del(&device->dev_list);
838 spin_unlock_bh(&device_spinlock);
840 EXPORT_SYMBOL(tls_unregister_device);
842 static struct tcp_ulp_ops tcp_tls_ulp_ops __read_mostly = {
844 .owner = THIS_MODULE,
846 .update = tls_update,
849 static int __init tls_register(void)
851 tls_sw_proto_ops = inet_stream_ops;
852 tls_sw_proto_ops.splice_read = tls_sw_splice_read;
854 #ifdef CONFIG_TLS_DEVICE
857 tcp_register_ulp(&tcp_tls_ulp_ops);
862 static void __exit tls_unregister(void)
864 tcp_unregister_ulp(&tcp_tls_ulp_ops);
865 #ifdef CONFIG_TLS_DEVICE
866 tls_device_cleanup();
870 module_init(tls_register);
871 module_exit(tls_unregister);