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 write_unlock_bh(&sk->sk_callback_lock);
313 if (ctx->tx_conf == TLS_SW)
314 tls_sw_free_ctx_tx(ctx);
315 if (ctx->rx_conf == TLS_SW || ctx->rx_conf == TLS_HW)
316 tls_sw_strparser_done(ctx);
317 if (ctx->rx_conf == TLS_SW)
318 tls_sw_free_ctx_rx(ctx);
319 ctx->sk_proto_close(sk, timeout);
325 static int do_tls_getsockopt_tx(struct sock *sk, char __user *optval,
329 struct tls_context *ctx = tls_get_ctx(sk);
330 struct tls_crypto_info *crypto_info;
333 if (get_user(len, optlen))
336 if (!optval || (len < sizeof(*crypto_info))) {
346 /* get user crypto info */
347 crypto_info = &ctx->crypto_send.info;
349 if (!TLS_CRYPTO_INFO_READY(crypto_info)) {
354 if (len == sizeof(*crypto_info)) {
355 if (copy_to_user(optval, crypto_info, sizeof(*crypto_info)))
360 switch (crypto_info->cipher_type) {
361 case TLS_CIPHER_AES_GCM_128: {
362 struct tls12_crypto_info_aes_gcm_128 *
363 crypto_info_aes_gcm_128 =
364 container_of(crypto_info,
365 struct tls12_crypto_info_aes_gcm_128,
368 if (len != sizeof(*crypto_info_aes_gcm_128)) {
373 memcpy(crypto_info_aes_gcm_128->iv,
374 ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
375 TLS_CIPHER_AES_GCM_128_IV_SIZE);
376 memcpy(crypto_info_aes_gcm_128->rec_seq, ctx->tx.rec_seq,
377 TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE);
379 if (copy_to_user(optval,
380 crypto_info_aes_gcm_128,
381 sizeof(*crypto_info_aes_gcm_128)))
385 case TLS_CIPHER_AES_GCM_256: {
386 struct tls12_crypto_info_aes_gcm_256 *
387 crypto_info_aes_gcm_256 =
388 container_of(crypto_info,
389 struct tls12_crypto_info_aes_gcm_256,
392 if (len != sizeof(*crypto_info_aes_gcm_256)) {
397 memcpy(crypto_info_aes_gcm_256->iv,
398 ctx->tx.iv + TLS_CIPHER_AES_GCM_256_SALT_SIZE,
399 TLS_CIPHER_AES_GCM_256_IV_SIZE);
400 memcpy(crypto_info_aes_gcm_256->rec_seq, ctx->tx.rec_seq,
401 TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE);
403 if (copy_to_user(optval,
404 crypto_info_aes_gcm_256,
405 sizeof(*crypto_info_aes_gcm_256)))
417 static int do_tls_getsockopt(struct sock *sk, int optname,
418 char __user *optval, int __user *optlen)
424 rc = do_tls_getsockopt_tx(sk, optval, optlen);
433 static int tls_getsockopt(struct sock *sk, int level, int optname,
434 char __user *optval, int __user *optlen)
436 struct tls_context *ctx = tls_get_ctx(sk);
438 if (level != SOL_TLS)
439 return ctx->getsockopt(sk, level, optname, optval, optlen);
441 return do_tls_getsockopt(sk, optname, optval, optlen);
444 static int do_tls_setsockopt_conf(struct sock *sk, char __user *optval,
445 unsigned int optlen, int tx)
447 struct tls_crypto_info *crypto_info;
448 struct tls_crypto_info *alt_crypto_info;
449 struct tls_context *ctx = tls_get_ctx(sk);
454 if (!optval || (optlen < sizeof(*crypto_info))) {
460 crypto_info = &ctx->crypto_send.info;
461 alt_crypto_info = &ctx->crypto_recv.info;
463 crypto_info = &ctx->crypto_recv.info;
464 alt_crypto_info = &ctx->crypto_send.info;
467 /* Currently we don't support set crypto info more than one time */
468 if (TLS_CRYPTO_INFO_READY(crypto_info)) {
473 rc = copy_from_user(crypto_info, optval, sizeof(*crypto_info));
476 goto err_crypto_info;
480 if (crypto_info->version != TLS_1_2_VERSION &&
481 crypto_info->version != TLS_1_3_VERSION) {
483 goto err_crypto_info;
486 /* Ensure that TLS version and ciphers are same in both directions */
487 if (TLS_CRYPTO_INFO_READY(alt_crypto_info)) {
488 if (alt_crypto_info->version != crypto_info->version ||
489 alt_crypto_info->cipher_type != crypto_info->cipher_type) {
491 goto err_crypto_info;
495 switch (crypto_info->cipher_type) {
496 case TLS_CIPHER_AES_GCM_128:
497 optsize = sizeof(struct tls12_crypto_info_aes_gcm_128);
499 case TLS_CIPHER_AES_GCM_256: {
500 optsize = sizeof(struct tls12_crypto_info_aes_gcm_256);
503 case TLS_CIPHER_AES_CCM_128:
504 optsize = sizeof(struct tls12_crypto_info_aes_ccm_128);
508 goto err_crypto_info;
511 if (optlen != optsize) {
513 goto err_crypto_info;
516 rc = copy_from_user(crypto_info + 1, optval + sizeof(*crypto_info),
517 optlen - sizeof(*crypto_info));
520 goto err_crypto_info;
524 #ifdef CONFIG_TLS_DEVICE
525 rc = tls_set_device_offload(sk, ctx);
531 rc = tls_set_sw_offload(sk, ctx, 1);
533 goto err_crypto_info;
537 #ifdef CONFIG_TLS_DEVICE
538 rc = tls_set_device_offload_rx(sk, ctx);
544 rc = tls_set_sw_offload(sk, ctx, 0);
546 goto err_crypto_info;
549 tls_sw_strparser_arm(sk, ctx);
556 update_sk_prot(sk, ctx);
558 ctx->sk_write_space = sk->sk_write_space;
559 sk->sk_write_space = tls_write_space;
561 sk->sk_socket->ops = &tls_sw_proto_ops;
566 memzero_explicit(crypto_info, sizeof(union tls_crypto_context));
571 static int do_tls_setsockopt(struct sock *sk, int optname,
572 char __user *optval, unsigned int optlen)
580 rc = do_tls_setsockopt_conf(sk, optval, optlen,
591 static int tls_setsockopt(struct sock *sk, int level, int optname,
592 char __user *optval, unsigned int optlen)
594 struct tls_context *ctx = tls_get_ctx(sk);
596 if (level != SOL_TLS)
597 return ctx->setsockopt(sk, level, optname, optval, optlen);
599 return do_tls_setsockopt(sk, optname, optval, optlen);
602 static struct tls_context *create_ctx(struct sock *sk)
604 struct inet_connection_sock *icsk = inet_csk(sk);
605 struct tls_context *ctx;
607 ctx = kzalloc(sizeof(*ctx), GFP_ATOMIC);
611 icsk->icsk_ulp_data = ctx;
612 ctx->setsockopt = sk->sk_prot->setsockopt;
613 ctx->getsockopt = sk->sk_prot->getsockopt;
614 ctx->sk_proto_close = sk->sk_prot->close;
615 ctx->unhash = sk->sk_prot->unhash;
619 static void tls_build_proto(struct sock *sk)
621 int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4;
623 /* Build IPv6 TLS whenever the address of tcpv6 _prot changes */
624 if (ip_ver == TLSV6 &&
625 unlikely(sk->sk_prot != smp_load_acquire(&saved_tcpv6_prot))) {
626 mutex_lock(&tcpv6_prot_mutex);
627 if (likely(sk->sk_prot != saved_tcpv6_prot)) {
628 build_protos(tls_prots[TLSV6], sk->sk_prot);
629 smp_store_release(&saved_tcpv6_prot, sk->sk_prot);
631 mutex_unlock(&tcpv6_prot_mutex);
634 if (ip_ver == TLSV4 &&
635 unlikely(sk->sk_prot != smp_load_acquire(&saved_tcpv4_prot))) {
636 mutex_lock(&tcpv4_prot_mutex);
637 if (likely(sk->sk_prot != saved_tcpv4_prot)) {
638 build_protos(tls_prots[TLSV4], sk->sk_prot);
639 smp_store_release(&saved_tcpv4_prot, sk->sk_prot);
641 mutex_unlock(&tcpv4_prot_mutex);
645 static void tls_hw_sk_destruct(struct sock *sk)
647 struct tls_context *ctx = tls_get_ctx(sk);
648 struct inet_connection_sock *icsk = inet_csk(sk);
650 ctx->sk_destruct(sk);
653 icsk->icsk_ulp_data = NULL;
656 static int tls_hw_prot(struct sock *sk)
658 struct tls_context *ctx;
659 struct tls_device *dev;
662 spin_lock_bh(&device_spinlock);
663 list_for_each_entry(dev, &device_list, dev_list) {
664 if (dev->feature && dev->feature(dev)) {
665 ctx = create_ctx(sk);
669 spin_unlock_bh(&device_spinlock);
671 ctx->hash = sk->sk_prot->hash;
672 ctx->unhash = sk->sk_prot->unhash;
673 ctx->sk_proto_close = sk->sk_prot->close;
674 ctx->sk_destruct = sk->sk_destruct;
675 sk->sk_destruct = tls_hw_sk_destruct;
676 ctx->rx_conf = TLS_HW_RECORD;
677 ctx->tx_conf = TLS_HW_RECORD;
678 update_sk_prot(sk, ctx);
679 spin_lock_bh(&device_spinlock);
685 spin_unlock_bh(&device_spinlock);
689 static void tls_hw_unhash(struct sock *sk)
691 struct tls_context *ctx = tls_get_ctx(sk);
692 struct tls_device *dev;
694 spin_lock_bh(&device_spinlock);
695 list_for_each_entry(dev, &device_list, dev_list) {
697 kref_get(&dev->kref);
698 spin_unlock_bh(&device_spinlock);
699 dev->unhash(dev, sk);
700 kref_put(&dev->kref, dev->release);
701 spin_lock_bh(&device_spinlock);
704 spin_unlock_bh(&device_spinlock);
708 static int tls_hw_hash(struct sock *sk)
710 struct tls_context *ctx = tls_get_ctx(sk);
711 struct tls_device *dev;
715 spin_lock_bh(&device_spinlock);
716 list_for_each_entry(dev, &device_list, dev_list) {
718 kref_get(&dev->kref);
719 spin_unlock_bh(&device_spinlock);
720 err |= dev->hash(dev, sk);
721 kref_put(&dev->kref, dev->release);
722 spin_lock_bh(&device_spinlock);
725 spin_unlock_bh(&device_spinlock);
732 static void build_protos(struct proto prot[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
735 prot[TLS_BASE][TLS_BASE] = *base;
736 prot[TLS_BASE][TLS_BASE].setsockopt = tls_setsockopt;
737 prot[TLS_BASE][TLS_BASE].getsockopt = tls_getsockopt;
738 prot[TLS_BASE][TLS_BASE].close = tls_sk_proto_close;
740 prot[TLS_SW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
741 prot[TLS_SW][TLS_BASE].sendmsg = tls_sw_sendmsg;
742 prot[TLS_SW][TLS_BASE].sendpage = tls_sw_sendpage;
744 prot[TLS_BASE][TLS_SW] = prot[TLS_BASE][TLS_BASE];
745 prot[TLS_BASE][TLS_SW].recvmsg = tls_sw_recvmsg;
746 prot[TLS_BASE][TLS_SW].stream_memory_read = tls_sw_stream_read;
747 prot[TLS_BASE][TLS_SW].close = tls_sk_proto_close;
749 prot[TLS_SW][TLS_SW] = prot[TLS_SW][TLS_BASE];
750 prot[TLS_SW][TLS_SW].recvmsg = tls_sw_recvmsg;
751 prot[TLS_SW][TLS_SW].stream_memory_read = tls_sw_stream_read;
752 prot[TLS_SW][TLS_SW].close = tls_sk_proto_close;
754 #ifdef CONFIG_TLS_DEVICE
755 prot[TLS_HW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
756 prot[TLS_HW][TLS_BASE].sendmsg = tls_device_sendmsg;
757 prot[TLS_HW][TLS_BASE].sendpage = tls_device_sendpage;
759 prot[TLS_HW][TLS_SW] = prot[TLS_BASE][TLS_SW];
760 prot[TLS_HW][TLS_SW].sendmsg = tls_device_sendmsg;
761 prot[TLS_HW][TLS_SW].sendpage = tls_device_sendpage;
763 prot[TLS_BASE][TLS_HW] = prot[TLS_BASE][TLS_SW];
765 prot[TLS_SW][TLS_HW] = prot[TLS_SW][TLS_SW];
767 prot[TLS_HW][TLS_HW] = prot[TLS_HW][TLS_SW];
770 prot[TLS_HW_RECORD][TLS_HW_RECORD] = *base;
771 prot[TLS_HW_RECORD][TLS_HW_RECORD].hash = tls_hw_hash;
772 prot[TLS_HW_RECORD][TLS_HW_RECORD].unhash = tls_hw_unhash;
775 static int tls_init(struct sock *sk)
777 struct tls_context *ctx;
783 /* The TLS ulp is currently supported only for TCP sockets
784 * in ESTABLISHED state.
785 * Supporting sockets in LISTEN state will require us
786 * to modify the accept implementation to clone rather then
787 * share the ulp context.
789 if (sk->sk_state != TCP_ESTABLISHED)
794 /* allocate tls context */
795 write_lock_bh(&sk->sk_callback_lock);
796 ctx = create_ctx(sk);
802 ctx->tx_conf = TLS_BASE;
803 ctx->rx_conf = TLS_BASE;
804 ctx->sk_proto = sk->sk_prot;
805 update_sk_prot(sk, ctx);
807 write_unlock_bh(&sk->sk_callback_lock);
811 static void tls_update(struct sock *sk, struct proto *p)
813 struct tls_context *ctx;
815 ctx = tls_get_ctx(sk);
817 ctx->sk_proto_close = p->close;
824 void tls_register_device(struct tls_device *device)
826 spin_lock_bh(&device_spinlock);
827 list_add_tail(&device->dev_list, &device_list);
828 spin_unlock_bh(&device_spinlock);
830 EXPORT_SYMBOL(tls_register_device);
832 void tls_unregister_device(struct tls_device *device)
834 spin_lock_bh(&device_spinlock);
835 list_del(&device->dev_list);
836 spin_unlock_bh(&device_spinlock);
838 EXPORT_SYMBOL(tls_unregister_device);
840 static struct tcp_ulp_ops tcp_tls_ulp_ops __read_mostly = {
842 .owner = THIS_MODULE,
844 .update = tls_update,
847 static int __init tls_register(void)
849 tls_sw_proto_ops = inet_stream_ops;
850 tls_sw_proto_ops.splice_read = tls_sw_splice_read;
852 #ifdef CONFIG_TLS_DEVICE
855 tcp_register_ulp(&tcp_tls_ulp_ops);
860 static void __exit tls_unregister(void)
862 tcp_unregister_ulp(&tcp_tls_ulp_ops);
863 #ifdef CONFIG_TLS_DEVICE
864 tls_device_cleanup();
868 module_init(tls_register);
869 module_exit(tls_unregister);