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_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 (unlikely(sk->sk_write_pending) &&
283 !wait_on_pending_writer(sk, &timeo))
284 tls_handle_open_record(sk, 0);
286 /* We need these for tls_sw_fallback handling of other packets */
287 if (ctx->tx_conf == TLS_SW) {
288 kfree(ctx->tx.rec_seq);
290 tls_sw_free_resources_tx(sk);
291 #ifdef CONFIG_TLS_DEVICE
292 } else if (ctx->tx_conf == TLS_HW) {
293 tls_device_free_resources_tx(sk);
297 if (ctx->rx_conf == TLS_SW)
298 tls_sw_free_resources_rx(sk);
300 #ifdef CONFIG_TLS_DEVICE
301 if (ctx->rx_conf == TLS_HW)
302 tls_device_offload_cleanup_rx(sk);
304 if (ctx->tx_conf != TLS_HW && ctx->rx_conf != TLS_HW) {
314 sk_proto_close(sk, timeout);
315 /* free ctx for TLS_HW_RECORD, used by tcp_set_state
316 * for sk->sk_prot->unhash [tls_hw_unhash]
322 static int do_tls_getsockopt_tx(struct sock *sk, char __user *optval,
326 struct tls_context *ctx = tls_get_ctx(sk);
327 struct tls_crypto_info *crypto_info;
330 if (get_user(len, optlen))
333 if (!optval || (len < sizeof(*crypto_info))) {
343 /* get user crypto info */
344 crypto_info = &ctx->crypto_send.info;
346 if (!TLS_CRYPTO_INFO_READY(crypto_info)) {
351 if (len == sizeof(*crypto_info)) {
352 if (copy_to_user(optval, crypto_info, sizeof(*crypto_info)))
357 switch (crypto_info->cipher_type) {
358 case TLS_CIPHER_AES_GCM_128: {
359 struct tls12_crypto_info_aes_gcm_128 *
360 crypto_info_aes_gcm_128 =
361 container_of(crypto_info,
362 struct tls12_crypto_info_aes_gcm_128,
365 if (len != sizeof(*crypto_info_aes_gcm_128)) {
370 memcpy(crypto_info_aes_gcm_128->iv,
371 ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
372 TLS_CIPHER_AES_GCM_128_IV_SIZE);
373 memcpy(crypto_info_aes_gcm_128->rec_seq, ctx->tx.rec_seq,
374 TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE);
376 if (copy_to_user(optval,
377 crypto_info_aes_gcm_128,
378 sizeof(*crypto_info_aes_gcm_128)))
382 case TLS_CIPHER_AES_GCM_256: {
383 struct tls12_crypto_info_aes_gcm_256 *
384 crypto_info_aes_gcm_256 =
385 container_of(crypto_info,
386 struct tls12_crypto_info_aes_gcm_256,
389 if (len != sizeof(*crypto_info_aes_gcm_256)) {
394 memcpy(crypto_info_aes_gcm_256->iv,
395 ctx->tx.iv + TLS_CIPHER_AES_GCM_256_SALT_SIZE,
396 TLS_CIPHER_AES_GCM_256_IV_SIZE);
397 memcpy(crypto_info_aes_gcm_256->rec_seq, ctx->tx.rec_seq,
398 TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE);
400 if (copy_to_user(optval,
401 crypto_info_aes_gcm_256,
402 sizeof(*crypto_info_aes_gcm_256)))
414 static int do_tls_getsockopt(struct sock *sk, int optname,
415 char __user *optval, int __user *optlen)
421 rc = do_tls_getsockopt_tx(sk, optval, optlen);
430 static int tls_getsockopt(struct sock *sk, int level, int optname,
431 char __user *optval, int __user *optlen)
433 struct tls_context *ctx = tls_get_ctx(sk);
435 if (level != SOL_TLS)
436 return ctx->getsockopt(sk, level, optname, optval, optlen);
438 return do_tls_getsockopt(sk, optname, optval, optlen);
441 static int do_tls_setsockopt_conf(struct sock *sk, char __user *optval,
442 unsigned int optlen, int tx)
444 struct tls_crypto_info *crypto_info;
445 struct tls_crypto_info *alt_crypto_info;
446 struct tls_context *ctx = tls_get_ctx(sk);
451 if (!optval || (optlen < sizeof(*crypto_info))) {
457 crypto_info = &ctx->crypto_send.info;
458 alt_crypto_info = &ctx->crypto_recv.info;
460 crypto_info = &ctx->crypto_recv.info;
461 alt_crypto_info = &ctx->crypto_send.info;
464 /* Currently we don't support set crypto info more than one time */
465 if (TLS_CRYPTO_INFO_READY(crypto_info)) {
470 rc = copy_from_user(crypto_info, optval, sizeof(*crypto_info));
473 goto err_crypto_info;
477 if (crypto_info->version != TLS_1_2_VERSION &&
478 crypto_info->version != TLS_1_3_VERSION) {
480 goto err_crypto_info;
483 /* Ensure that TLS version and ciphers are same in both directions */
484 if (TLS_CRYPTO_INFO_READY(alt_crypto_info)) {
485 if (alt_crypto_info->version != crypto_info->version ||
486 alt_crypto_info->cipher_type != crypto_info->cipher_type) {
488 goto err_crypto_info;
492 switch (crypto_info->cipher_type) {
493 case TLS_CIPHER_AES_GCM_128:
494 optsize = sizeof(struct tls12_crypto_info_aes_gcm_128);
496 case TLS_CIPHER_AES_GCM_256: {
497 optsize = sizeof(struct tls12_crypto_info_aes_gcm_256);
500 case TLS_CIPHER_AES_CCM_128:
501 optsize = sizeof(struct tls12_crypto_info_aes_ccm_128);
505 goto err_crypto_info;
508 if (optlen != optsize) {
510 goto err_crypto_info;
513 rc = copy_from_user(crypto_info + 1, optval + sizeof(*crypto_info),
514 optlen - sizeof(*crypto_info));
517 goto err_crypto_info;
521 #ifdef CONFIG_TLS_DEVICE
522 rc = tls_set_device_offload(sk, ctx);
528 rc = tls_set_sw_offload(sk, ctx, 1);
532 #ifdef CONFIG_TLS_DEVICE
533 rc = tls_set_device_offload_rx(sk, ctx);
539 rc = tls_set_sw_offload(sk, ctx, 0);
545 goto err_crypto_info;
551 update_sk_prot(sk, ctx);
553 ctx->sk_write_space = sk->sk_write_space;
554 sk->sk_write_space = tls_write_space;
556 sk->sk_socket->ops = &tls_sw_proto_ops;
561 memzero_explicit(crypto_info, sizeof(union tls_crypto_context));
566 static int do_tls_setsockopt(struct sock *sk, int optname,
567 char __user *optval, unsigned int optlen)
575 rc = do_tls_setsockopt_conf(sk, optval, optlen,
586 static int tls_setsockopt(struct sock *sk, int level, int optname,
587 char __user *optval, unsigned int optlen)
589 struct tls_context *ctx = tls_get_ctx(sk);
591 if (level != SOL_TLS)
592 return ctx->setsockopt(sk, level, optname, optval, optlen);
594 return do_tls_setsockopt(sk, optname, optval, optlen);
597 static struct tls_context *create_ctx(struct sock *sk)
599 struct inet_connection_sock *icsk = inet_csk(sk);
600 struct tls_context *ctx;
602 ctx = kzalloc(sizeof(*ctx), GFP_ATOMIC);
606 icsk->icsk_ulp_data = ctx;
607 ctx->setsockopt = sk->sk_prot->setsockopt;
608 ctx->getsockopt = sk->sk_prot->getsockopt;
609 ctx->sk_proto_close = sk->sk_prot->close;
613 static void tls_build_proto(struct sock *sk)
615 int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4;
617 /* Build IPv6 TLS whenever the address of tcpv6 _prot changes */
618 if (ip_ver == TLSV6 &&
619 unlikely(sk->sk_prot != smp_load_acquire(&saved_tcpv6_prot))) {
620 mutex_lock(&tcpv6_prot_mutex);
621 if (likely(sk->sk_prot != saved_tcpv6_prot)) {
622 build_protos(tls_prots[TLSV6], sk->sk_prot);
623 smp_store_release(&saved_tcpv6_prot, sk->sk_prot);
625 mutex_unlock(&tcpv6_prot_mutex);
628 if (ip_ver == TLSV4 &&
629 unlikely(sk->sk_prot != smp_load_acquire(&saved_tcpv4_prot))) {
630 mutex_lock(&tcpv4_prot_mutex);
631 if (likely(sk->sk_prot != saved_tcpv4_prot)) {
632 build_protos(tls_prots[TLSV4], sk->sk_prot);
633 smp_store_release(&saved_tcpv4_prot, sk->sk_prot);
635 mutex_unlock(&tcpv4_prot_mutex);
639 static void tls_hw_sk_destruct(struct sock *sk)
641 struct tls_context *ctx = tls_get_ctx(sk);
642 struct inet_connection_sock *icsk = inet_csk(sk);
644 ctx->sk_destruct(sk);
647 icsk->icsk_ulp_data = NULL;
650 static int tls_hw_prot(struct sock *sk)
652 struct tls_context *ctx;
653 struct tls_device *dev;
656 spin_lock_bh(&device_spinlock);
657 list_for_each_entry(dev, &device_list, dev_list) {
658 if (dev->feature && dev->feature(dev)) {
659 ctx = create_ctx(sk);
663 spin_unlock_bh(&device_spinlock);
665 ctx->hash = sk->sk_prot->hash;
666 ctx->unhash = sk->sk_prot->unhash;
667 ctx->sk_proto_close = sk->sk_prot->close;
668 ctx->sk_destruct = sk->sk_destruct;
669 sk->sk_destruct = tls_hw_sk_destruct;
670 ctx->rx_conf = TLS_HW_RECORD;
671 ctx->tx_conf = TLS_HW_RECORD;
672 update_sk_prot(sk, ctx);
673 spin_lock_bh(&device_spinlock);
679 spin_unlock_bh(&device_spinlock);
683 static void tls_hw_unhash(struct sock *sk)
685 struct tls_context *ctx = tls_get_ctx(sk);
686 struct tls_device *dev;
688 spin_lock_bh(&device_spinlock);
689 list_for_each_entry(dev, &device_list, dev_list) {
691 kref_get(&dev->kref);
692 spin_unlock_bh(&device_spinlock);
693 dev->unhash(dev, sk);
694 kref_put(&dev->kref, dev->release);
695 spin_lock_bh(&device_spinlock);
698 spin_unlock_bh(&device_spinlock);
702 static int tls_hw_hash(struct sock *sk)
704 struct tls_context *ctx = tls_get_ctx(sk);
705 struct tls_device *dev;
709 spin_lock_bh(&device_spinlock);
710 list_for_each_entry(dev, &device_list, dev_list) {
712 kref_get(&dev->kref);
713 spin_unlock_bh(&device_spinlock);
714 err |= dev->hash(dev, sk);
715 kref_put(&dev->kref, dev->release);
716 spin_lock_bh(&device_spinlock);
719 spin_unlock_bh(&device_spinlock);
726 static void build_protos(struct proto prot[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
729 prot[TLS_BASE][TLS_BASE] = *base;
730 prot[TLS_BASE][TLS_BASE].setsockopt = tls_setsockopt;
731 prot[TLS_BASE][TLS_BASE].getsockopt = tls_getsockopt;
732 prot[TLS_BASE][TLS_BASE].close = tls_sk_proto_close;
734 prot[TLS_SW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
735 prot[TLS_SW][TLS_BASE].sendmsg = tls_sw_sendmsg;
736 prot[TLS_SW][TLS_BASE].sendpage = tls_sw_sendpage;
738 prot[TLS_BASE][TLS_SW] = prot[TLS_BASE][TLS_BASE];
739 prot[TLS_BASE][TLS_SW].recvmsg = tls_sw_recvmsg;
740 prot[TLS_BASE][TLS_SW].stream_memory_read = tls_sw_stream_read;
741 prot[TLS_BASE][TLS_SW].close = tls_sk_proto_close;
743 prot[TLS_SW][TLS_SW] = prot[TLS_SW][TLS_BASE];
744 prot[TLS_SW][TLS_SW].recvmsg = tls_sw_recvmsg;
745 prot[TLS_SW][TLS_SW].stream_memory_read = tls_sw_stream_read;
746 prot[TLS_SW][TLS_SW].close = tls_sk_proto_close;
748 #ifdef CONFIG_TLS_DEVICE
749 prot[TLS_HW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
750 prot[TLS_HW][TLS_BASE].sendmsg = tls_device_sendmsg;
751 prot[TLS_HW][TLS_BASE].sendpage = tls_device_sendpage;
753 prot[TLS_HW][TLS_SW] = prot[TLS_BASE][TLS_SW];
754 prot[TLS_HW][TLS_SW].sendmsg = tls_device_sendmsg;
755 prot[TLS_HW][TLS_SW].sendpage = tls_device_sendpage;
757 prot[TLS_BASE][TLS_HW] = prot[TLS_BASE][TLS_SW];
759 prot[TLS_SW][TLS_HW] = prot[TLS_SW][TLS_SW];
761 prot[TLS_HW][TLS_HW] = prot[TLS_HW][TLS_SW];
764 prot[TLS_HW_RECORD][TLS_HW_RECORD] = *base;
765 prot[TLS_HW_RECORD][TLS_HW_RECORD].hash = tls_hw_hash;
766 prot[TLS_HW_RECORD][TLS_HW_RECORD].unhash = tls_hw_unhash;
767 prot[TLS_HW_RECORD][TLS_HW_RECORD].close = tls_sk_proto_close;
770 static int tls_init(struct sock *sk)
772 struct tls_context *ctx;
778 /* The TLS ulp is currently supported only for TCP sockets
779 * in ESTABLISHED state.
780 * Supporting sockets in LISTEN state will require us
781 * to modify the accept implementation to clone rather then
782 * share the ulp context.
784 if (sk->sk_state != TCP_ESTABLISHED)
787 /* allocate tls context */
788 ctx = create_ctx(sk);
795 ctx->tx_conf = TLS_BASE;
796 ctx->rx_conf = TLS_BASE;
797 update_sk_prot(sk, ctx);
802 void tls_register_device(struct tls_device *device)
804 spin_lock_bh(&device_spinlock);
805 list_add_tail(&device->dev_list, &device_list);
806 spin_unlock_bh(&device_spinlock);
808 EXPORT_SYMBOL(tls_register_device);
810 void tls_unregister_device(struct tls_device *device)
812 spin_lock_bh(&device_spinlock);
813 list_del(&device->dev_list);
814 spin_unlock_bh(&device_spinlock);
816 EXPORT_SYMBOL(tls_unregister_device);
818 static struct tcp_ulp_ops tcp_tls_ulp_ops __read_mostly = {
820 .owner = THIS_MODULE,
824 static int __init tls_register(void)
826 tls_sw_proto_ops = inet_stream_ops;
827 tls_sw_proto_ops.splice_read = tls_sw_splice_read;
829 #ifdef CONFIG_TLS_DEVICE
832 tcp_register_ulp(&tcp_tls_ulp_ops);
837 static void __exit tls_unregister(void)
839 tcp_unregister_ulp(&tcp_tls_ulp_ops);
840 #ifdef CONFIG_TLS_DEVICE
841 tls_device_cleanup();
845 module_init(tls_register);
846 module_exit(tls_unregister);