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 static 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_close(struct sock *sk, long timeout)
266 struct tls_context *ctx = tls_get_ctx(sk);
267 long timeo = sock_sndtimeo(sk, 0);
268 void (*sk_proto_close)(struct sock *sk, long timeout);
269 bool free_ctx = false;
272 sk_proto_close = ctx->sk_proto_close;
274 if (ctx->tx_conf == TLS_HW_RECORD && ctx->rx_conf == TLS_HW_RECORD)
275 goto skip_tx_cleanup;
277 if (ctx->tx_conf == TLS_BASE && ctx->rx_conf == TLS_BASE) {
279 goto skip_tx_cleanup;
282 if (!tls_complete_pending_work(sk, ctx, 0, &timeo))
283 tls_handle_open_record(sk, 0);
285 /* We need these for tls_sw_fallback handling of other packets */
286 if (ctx->tx_conf == TLS_SW) {
287 kfree(ctx->tx.rec_seq);
289 tls_sw_free_resources_tx(sk);
290 #ifdef CONFIG_TLS_DEVICE
291 } else if (ctx->tx_conf == TLS_HW) {
292 tls_device_free_resources_tx(sk);
296 if (ctx->rx_conf == TLS_SW)
297 tls_sw_free_resources_rx(sk);
299 #ifdef CONFIG_TLS_DEVICE
300 if (ctx->rx_conf == TLS_HW)
301 tls_device_offload_cleanup_rx(sk);
303 if (ctx->tx_conf != TLS_HW && ctx->rx_conf != TLS_HW) {
313 sk_proto_close(sk, timeout);
314 /* free ctx for TLS_HW_RECORD, used by tcp_set_state
315 * for sk->sk_prot->unhash [tls_hw_unhash]
321 static int do_tls_getsockopt_tx(struct sock *sk, char __user *optval,
325 struct tls_context *ctx = tls_get_ctx(sk);
326 struct tls_crypto_info *crypto_info;
329 if (get_user(len, optlen))
332 if (!optval || (len < sizeof(*crypto_info))) {
342 /* get user crypto info */
343 crypto_info = &ctx->crypto_send.info;
345 if (!TLS_CRYPTO_INFO_READY(crypto_info)) {
350 if (len == sizeof(*crypto_info)) {
351 if (copy_to_user(optval, crypto_info, sizeof(*crypto_info)))
356 switch (crypto_info->cipher_type) {
357 case TLS_CIPHER_AES_GCM_128: {
358 struct tls12_crypto_info_aes_gcm_128 *
359 crypto_info_aes_gcm_128 =
360 container_of(crypto_info,
361 struct tls12_crypto_info_aes_gcm_128,
364 if (len != sizeof(*crypto_info_aes_gcm_128)) {
369 memcpy(crypto_info_aes_gcm_128->iv,
370 ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
371 TLS_CIPHER_AES_GCM_128_IV_SIZE);
372 memcpy(crypto_info_aes_gcm_128->rec_seq, ctx->tx.rec_seq,
373 TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE);
375 if (copy_to_user(optval,
376 crypto_info_aes_gcm_128,
377 sizeof(*crypto_info_aes_gcm_128)))
381 case TLS_CIPHER_AES_GCM_256: {
382 struct tls12_crypto_info_aes_gcm_256 *
383 crypto_info_aes_gcm_256 =
384 container_of(crypto_info,
385 struct tls12_crypto_info_aes_gcm_256,
388 if (len != sizeof(*crypto_info_aes_gcm_256)) {
393 memcpy(crypto_info_aes_gcm_256->iv,
394 ctx->tx.iv + TLS_CIPHER_AES_GCM_256_SALT_SIZE,
395 TLS_CIPHER_AES_GCM_256_IV_SIZE);
396 memcpy(crypto_info_aes_gcm_256->rec_seq, ctx->tx.rec_seq,
397 TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE);
399 if (copy_to_user(optval,
400 crypto_info_aes_gcm_256,
401 sizeof(*crypto_info_aes_gcm_256)))
413 static int do_tls_getsockopt(struct sock *sk, int optname,
414 char __user *optval, int __user *optlen)
420 rc = do_tls_getsockopt_tx(sk, optval, optlen);
429 static int tls_getsockopt(struct sock *sk, int level, int optname,
430 char __user *optval, int __user *optlen)
432 struct tls_context *ctx = tls_get_ctx(sk);
434 if (level != SOL_TLS)
435 return ctx->getsockopt(sk, level, optname, optval, optlen);
437 return do_tls_getsockopt(sk, optname, optval, optlen);
440 static int do_tls_setsockopt_conf(struct sock *sk, char __user *optval,
441 unsigned int optlen, int tx)
443 struct tls_crypto_info *crypto_info;
444 struct tls_crypto_info *alt_crypto_info;
445 struct tls_context *ctx = tls_get_ctx(sk);
450 if (!optval || (optlen < sizeof(*crypto_info))) {
456 crypto_info = &ctx->crypto_send.info;
457 alt_crypto_info = &ctx->crypto_recv.info;
459 crypto_info = &ctx->crypto_recv.info;
460 alt_crypto_info = &ctx->crypto_send.info;
463 /* Currently we don't support set crypto info more than one time */
464 if (TLS_CRYPTO_INFO_READY(crypto_info)) {
469 rc = copy_from_user(crypto_info, optval, sizeof(*crypto_info));
472 goto err_crypto_info;
476 if (crypto_info->version != TLS_1_2_VERSION &&
477 crypto_info->version != TLS_1_3_VERSION) {
479 goto err_crypto_info;
482 /* Ensure that TLS version and ciphers are same in both directions */
483 if (TLS_CRYPTO_INFO_READY(alt_crypto_info)) {
484 if (alt_crypto_info->version != crypto_info->version ||
485 alt_crypto_info->cipher_type != crypto_info->cipher_type) {
487 goto err_crypto_info;
491 switch (crypto_info->cipher_type) {
492 case TLS_CIPHER_AES_GCM_128:
493 case TLS_CIPHER_AES_GCM_256: {
494 optsize = crypto_info->cipher_type == TLS_CIPHER_AES_GCM_128 ?
495 sizeof(struct tls12_crypto_info_aes_gcm_128) :
496 sizeof(struct tls12_crypto_info_aes_gcm_256);
497 if (optlen != optsize) {
499 goto err_crypto_info;
501 rc = copy_from_user(crypto_info + 1, optval + sizeof(*crypto_info),
502 optlen - sizeof(*crypto_info));
505 goto err_crypto_info;
511 goto err_crypto_info;
515 #ifdef CONFIG_TLS_DEVICE
516 rc = tls_set_device_offload(sk, ctx);
522 rc = tls_set_sw_offload(sk, ctx, 1);
526 #ifdef CONFIG_TLS_DEVICE
527 rc = tls_set_device_offload_rx(sk, ctx);
533 rc = tls_set_sw_offload(sk, ctx, 0);
539 goto err_crypto_info;
545 update_sk_prot(sk, ctx);
547 ctx->sk_write_space = sk->sk_write_space;
548 sk->sk_write_space = tls_write_space;
550 sk->sk_socket->ops = &tls_sw_proto_ops;
555 memzero_explicit(crypto_info, sizeof(union tls_crypto_context));
560 static int do_tls_setsockopt(struct sock *sk, int optname,
561 char __user *optval, unsigned int optlen)
569 rc = do_tls_setsockopt_conf(sk, optval, optlen,
580 static int tls_setsockopt(struct sock *sk, int level, int optname,
581 char __user *optval, unsigned int optlen)
583 struct tls_context *ctx = tls_get_ctx(sk);
585 if (level != SOL_TLS)
586 return ctx->setsockopt(sk, level, optname, optval, optlen);
588 return do_tls_setsockopt(sk, optname, optval, optlen);
591 static struct tls_context *create_ctx(struct sock *sk)
593 struct inet_connection_sock *icsk = inet_csk(sk);
594 struct tls_context *ctx;
596 ctx = kzalloc(sizeof(*ctx), GFP_ATOMIC);
600 icsk->icsk_ulp_data = ctx;
601 ctx->setsockopt = sk->sk_prot->setsockopt;
602 ctx->getsockopt = sk->sk_prot->getsockopt;
603 ctx->sk_proto_close = sk->sk_prot->close;
607 static void tls_build_proto(struct sock *sk)
609 int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4;
611 /* Build IPv6 TLS whenever the address of tcpv6 _prot changes */
612 if (ip_ver == TLSV6 &&
613 unlikely(sk->sk_prot != smp_load_acquire(&saved_tcpv6_prot))) {
614 mutex_lock(&tcpv6_prot_mutex);
615 if (likely(sk->sk_prot != saved_tcpv6_prot)) {
616 build_protos(tls_prots[TLSV6], sk->sk_prot);
617 smp_store_release(&saved_tcpv6_prot, sk->sk_prot);
619 mutex_unlock(&tcpv6_prot_mutex);
622 if (ip_ver == TLSV4 &&
623 unlikely(sk->sk_prot != smp_load_acquire(&saved_tcpv4_prot))) {
624 mutex_lock(&tcpv4_prot_mutex);
625 if (likely(sk->sk_prot != saved_tcpv4_prot)) {
626 build_protos(tls_prots[TLSV4], sk->sk_prot);
627 smp_store_release(&saved_tcpv4_prot, sk->sk_prot);
629 mutex_unlock(&tcpv4_prot_mutex);
633 static void tls_hw_sk_destruct(struct sock *sk)
635 struct tls_context *ctx = tls_get_ctx(sk);
636 struct inet_connection_sock *icsk = inet_csk(sk);
638 ctx->sk_destruct(sk);
641 icsk->icsk_ulp_data = NULL;
644 static int tls_hw_prot(struct sock *sk)
646 struct tls_context *ctx;
647 struct tls_device *dev;
650 spin_lock_bh(&device_spinlock);
651 list_for_each_entry(dev, &device_list, dev_list) {
652 if (dev->feature && dev->feature(dev)) {
653 ctx = create_ctx(sk);
657 spin_unlock_bh(&device_spinlock);
659 ctx->hash = sk->sk_prot->hash;
660 ctx->unhash = sk->sk_prot->unhash;
661 ctx->sk_proto_close = sk->sk_prot->close;
662 ctx->sk_destruct = sk->sk_destruct;
663 sk->sk_destruct = tls_hw_sk_destruct;
664 ctx->rx_conf = TLS_HW_RECORD;
665 ctx->tx_conf = TLS_HW_RECORD;
666 update_sk_prot(sk, ctx);
667 spin_lock_bh(&device_spinlock);
673 spin_unlock_bh(&device_spinlock);
677 static void tls_hw_unhash(struct sock *sk)
679 struct tls_context *ctx = tls_get_ctx(sk);
680 struct tls_device *dev;
682 spin_lock_bh(&device_spinlock);
683 list_for_each_entry(dev, &device_list, dev_list) {
685 kref_get(&dev->kref);
686 spin_unlock_bh(&device_spinlock);
687 dev->unhash(dev, sk);
688 kref_put(&dev->kref, dev->release);
689 spin_lock_bh(&device_spinlock);
692 spin_unlock_bh(&device_spinlock);
696 static int tls_hw_hash(struct sock *sk)
698 struct tls_context *ctx = tls_get_ctx(sk);
699 struct tls_device *dev;
703 spin_lock_bh(&device_spinlock);
704 list_for_each_entry(dev, &device_list, dev_list) {
706 kref_get(&dev->kref);
707 spin_unlock_bh(&device_spinlock);
708 err |= dev->hash(dev, sk);
709 kref_put(&dev->kref, dev->release);
710 spin_lock_bh(&device_spinlock);
713 spin_unlock_bh(&device_spinlock);
720 static void build_protos(struct proto prot[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
723 prot[TLS_BASE][TLS_BASE] = *base;
724 prot[TLS_BASE][TLS_BASE].setsockopt = tls_setsockopt;
725 prot[TLS_BASE][TLS_BASE].getsockopt = tls_getsockopt;
726 prot[TLS_BASE][TLS_BASE].close = tls_sk_proto_close;
728 prot[TLS_SW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
729 prot[TLS_SW][TLS_BASE].sendmsg = tls_sw_sendmsg;
730 prot[TLS_SW][TLS_BASE].sendpage = tls_sw_sendpage;
732 prot[TLS_BASE][TLS_SW] = prot[TLS_BASE][TLS_BASE];
733 prot[TLS_BASE][TLS_SW].recvmsg = tls_sw_recvmsg;
734 prot[TLS_BASE][TLS_SW].stream_memory_read = tls_sw_stream_read;
735 prot[TLS_BASE][TLS_SW].close = tls_sk_proto_close;
737 prot[TLS_SW][TLS_SW] = prot[TLS_SW][TLS_BASE];
738 prot[TLS_SW][TLS_SW].recvmsg = tls_sw_recvmsg;
739 prot[TLS_SW][TLS_SW].stream_memory_read = tls_sw_stream_read;
740 prot[TLS_SW][TLS_SW].close = tls_sk_proto_close;
742 #ifdef CONFIG_TLS_DEVICE
743 prot[TLS_HW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
744 prot[TLS_HW][TLS_BASE].sendmsg = tls_device_sendmsg;
745 prot[TLS_HW][TLS_BASE].sendpage = tls_device_sendpage;
747 prot[TLS_HW][TLS_SW] = prot[TLS_BASE][TLS_SW];
748 prot[TLS_HW][TLS_SW].sendmsg = tls_device_sendmsg;
749 prot[TLS_HW][TLS_SW].sendpage = tls_device_sendpage;
751 prot[TLS_BASE][TLS_HW] = prot[TLS_BASE][TLS_SW];
753 prot[TLS_SW][TLS_HW] = prot[TLS_SW][TLS_SW];
755 prot[TLS_HW][TLS_HW] = prot[TLS_HW][TLS_SW];
758 prot[TLS_HW_RECORD][TLS_HW_RECORD] = *base;
759 prot[TLS_HW_RECORD][TLS_HW_RECORD].hash = tls_hw_hash;
760 prot[TLS_HW_RECORD][TLS_HW_RECORD].unhash = tls_hw_unhash;
761 prot[TLS_HW_RECORD][TLS_HW_RECORD].close = tls_sk_proto_close;
764 static int tls_init(struct sock *sk)
766 struct tls_context *ctx;
772 /* The TLS ulp is currently supported only for TCP sockets
773 * in ESTABLISHED state.
774 * Supporting sockets in LISTEN state will require us
775 * to modify the accept implementation to clone rather then
776 * share the ulp context.
778 if (sk->sk_state != TCP_ESTABLISHED)
781 /* allocate tls context */
782 ctx = create_ctx(sk);
789 ctx->tx_conf = TLS_BASE;
790 ctx->rx_conf = TLS_BASE;
791 update_sk_prot(sk, ctx);
796 void tls_register_device(struct tls_device *device)
798 spin_lock_bh(&device_spinlock);
799 list_add_tail(&device->dev_list, &device_list);
800 spin_unlock_bh(&device_spinlock);
802 EXPORT_SYMBOL(tls_register_device);
804 void tls_unregister_device(struct tls_device *device)
806 spin_lock_bh(&device_spinlock);
807 list_del(&device->dev_list);
808 spin_unlock_bh(&device_spinlock);
810 EXPORT_SYMBOL(tls_unregister_device);
812 static struct tcp_ulp_ops tcp_tls_ulp_ops __read_mostly = {
814 .owner = THIS_MODULE,
818 static int __init tls_register(void)
820 tls_sw_proto_ops = inet_stream_ops;
821 tls_sw_proto_ops.splice_read = tls_sw_splice_read;
823 #ifdef CONFIG_TLS_DEVICE
826 tcp_register_ulp(&tcp_tls_ulp_ops);
831 static void __exit tls_unregister(void)
833 tcp_unregister_ulp(&tcp_tls_ulp_ops);
834 #ifdef CONFIG_TLS_DEVICE
835 tls_device_cleanup();
839 module_init(tls_register);
840 module_exit(tls_unregister);