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>
42 #include <linux/inet_diag.h>
46 #include <net/tls_toe.h>
48 MODULE_AUTHOR("Mellanox Technologies");
49 MODULE_DESCRIPTION("Transport Layer Security Support");
50 MODULE_LICENSE("Dual BSD/GPL");
51 MODULE_ALIAS_TCP_ULP("tls");
59 static struct proto *saved_tcpv6_prot;
60 static DEFINE_MUTEX(tcpv6_prot_mutex);
61 static struct proto *saved_tcpv4_prot;
62 static DEFINE_MUTEX(tcpv4_prot_mutex);
63 static struct proto tls_prots[TLS_NUM_PROTS][TLS_NUM_CONFIG][TLS_NUM_CONFIG];
64 static struct proto_ops tls_sw_proto_ops;
65 static void build_protos(struct proto prot[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
68 void update_sk_prot(struct sock *sk, struct tls_context *ctx)
70 int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4;
72 sk->sk_prot = &tls_prots[ip_ver][ctx->tx_conf][ctx->rx_conf];
75 int wait_on_pending_writer(struct sock *sk, long *timeo)
78 DEFINE_WAIT_FUNC(wait, woken_wake_function);
80 add_wait_queue(sk_sleep(sk), &wait);
87 if (signal_pending(current)) {
88 rc = sock_intr_errno(*timeo);
92 if (sk_wait_event(sk, timeo, !sk->sk_write_pending, &wait))
95 remove_wait_queue(sk_sleep(sk), &wait);
99 int tls_push_sg(struct sock *sk,
100 struct tls_context *ctx,
101 struct scatterlist *sg,
105 int sendpage_flags = flags | MSG_SENDPAGE_NOTLAST;
109 int offset = first_offset;
111 size = sg->length - offset;
112 offset += sg->offset;
114 ctx->in_tcp_sendpages = true;
117 sendpage_flags = flags;
119 /* is sending application-limited? */
120 tcp_rate_check_app_limited(sk);
123 ret = do_tcp_sendpages(sk, p, offset, size, sendpage_flags);
132 offset -= sg->offset;
133 ctx->partially_sent_offset = offset;
134 ctx->partially_sent_record = (void *)sg;
135 ctx->in_tcp_sendpages = false;
140 sk_mem_uncharge(sk, sg->length);
149 ctx->in_tcp_sendpages = false;
154 static int tls_handle_open_record(struct sock *sk, int flags)
156 struct tls_context *ctx = tls_get_ctx(sk);
158 if (tls_is_pending_open_record(ctx))
159 return ctx->push_pending_record(sk, flags);
164 int tls_proccess_cmsg(struct sock *sk, struct msghdr *msg,
165 unsigned char *record_type)
167 struct cmsghdr *cmsg;
170 for_each_cmsghdr(cmsg, msg) {
171 if (!CMSG_OK(msg, cmsg))
173 if (cmsg->cmsg_level != SOL_TLS)
176 switch (cmsg->cmsg_type) {
177 case TLS_SET_RECORD_TYPE:
178 if (cmsg->cmsg_len < CMSG_LEN(sizeof(*record_type)))
181 if (msg->msg_flags & MSG_MORE)
184 rc = tls_handle_open_record(sk, msg->msg_flags);
188 *record_type = *(unsigned char *)CMSG_DATA(cmsg);
199 int tls_push_partial_record(struct sock *sk, struct tls_context *ctx,
202 struct scatterlist *sg;
205 sg = ctx->partially_sent_record;
206 offset = ctx->partially_sent_offset;
208 ctx->partially_sent_record = NULL;
209 return tls_push_sg(sk, ctx, sg, offset, flags);
212 bool tls_free_partial_record(struct sock *sk, struct tls_context *ctx)
214 struct scatterlist *sg;
216 sg = ctx->partially_sent_record;
221 put_page(sg_page(sg));
222 sk_mem_uncharge(sk, sg->length);
228 ctx->partially_sent_record = NULL;
232 static void tls_write_space(struct sock *sk)
234 struct tls_context *ctx = tls_get_ctx(sk);
236 /* If in_tcp_sendpages call lower protocol write space handler
237 * to ensure we wake up any waiting operations there. For example
238 * if do_tcp_sendpages where to call sk_wait_event.
240 if (ctx->in_tcp_sendpages) {
241 ctx->sk_write_space(sk);
245 #ifdef CONFIG_TLS_DEVICE
246 if (ctx->tx_conf == TLS_HW)
247 tls_device_write_space(sk, ctx);
250 tls_sw_write_space(sk, ctx);
252 ctx->sk_write_space(sk);
256 * tls_ctx_free() - free TLS ULP context
257 * @sk: socket to with @ctx is attached
258 * @ctx: TLS context structure
260 * Free TLS context. If @sk is %NULL caller guarantees that the socket
261 * to which @ctx was attached has no outstanding references.
263 void tls_ctx_free(struct sock *sk, struct tls_context *ctx)
268 memzero_explicit(&ctx->crypto_send, sizeof(ctx->crypto_send));
269 memzero_explicit(&ctx->crypto_recv, sizeof(ctx->crypto_recv));
270 mutex_destroy(&ctx->tx_lock);
278 static void tls_sk_proto_cleanup(struct sock *sk,
279 struct tls_context *ctx, long timeo)
281 if (unlikely(sk->sk_write_pending) &&
282 !wait_on_pending_writer(sk, &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_release_resources_tx(sk);
290 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXSW);
291 } else if (ctx->tx_conf == TLS_HW) {
292 tls_device_free_resources_tx(sk);
293 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXDEVICE);
296 if (ctx->rx_conf == TLS_SW) {
297 tls_sw_release_resources_rx(sk);
298 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXSW);
299 } else if (ctx->rx_conf == TLS_HW) {
300 tls_device_offload_cleanup_rx(sk);
301 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXDEVICE);
305 static void tls_sk_proto_close(struct sock *sk, long timeout)
307 struct inet_connection_sock *icsk = inet_csk(sk);
308 struct tls_context *ctx = tls_get_ctx(sk);
309 long timeo = sock_sndtimeo(sk, 0);
312 if (ctx->tx_conf == TLS_SW)
313 tls_sw_cancel_work_tx(ctx);
316 free_ctx = ctx->tx_conf != TLS_HW && ctx->rx_conf != TLS_HW;
318 if (ctx->tx_conf != TLS_BASE || ctx->rx_conf != TLS_BASE)
319 tls_sk_proto_cleanup(sk, ctx, timeo);
321 write_lock_bh(&sk->sk_callback_lock);
323 rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
324 sk->sk_prot = ctx->sk_proto;
325 if (sk->sk_write_space == tls_write_space)
326 sk->sk_write_space = ctx->sk_write_space;
327 write_unlock_bh(&sk->sk_callback_lock);
329 if (ctx->tx_conf == TLS_SW)
330 tls_sw_free_ctx_tx(ctx);
331 if (ctx->rx_conf == TLS_SW || ctx->rx_conf == TLS_HW)
332 tls_sw_strparser_done(ctx);
333 if (ctx->rx_conf == TLS_SW)
334 tls_sw_free_ctx_rx(ctx);
335 ctx->sk_proto->close(sk, timeout);
338 tls_ctx_free(sk, ctx);
341 static int do_tls_getsockopt_tx(struct sock *sk, char __user *optval,
345 struct tls_context *ctx = tls_get_ctx(sk);
346 struct tls_crypto_info *crypto_info;
349 if (get_user(len, optlen))
352 if (!optval || (len < sizeof(*crypto_info))) {
362 /* get user crypto info */
363 crypto_info = &ctx->crypto_send.info;
365 if (!TLS_CRYPTO_INFO_READY(crypto_info)) {
370 if (len == sizeof(*crypto_info)) {
371 if (copy_to_user(optval, crypto_info, sizeof(*crypto_info)))
376 switch (crypto_info->cipher_type) {
377 case TLS_CIPHER_AES_GCM_128: {
378 struct tls12_crypto_info_aes_gcm_128 *
379 crypto_info_aes_gcm_128 =
380 container_of(crypto_info,
381 struct tls12_crypto_info_aes_gcm_128,
384 if (len != sizeof(*crypto_info_aes_gcm_128)) {
389 memcpy(crypto_info_aes_gcm_128->iv,
390 ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
391 TLS_CIPHER_AES_GCM_128_IV_SIZE);
392 memcpy(crypto_info_aes_gcm_128->rec_seq, ctx->tx.rec_seq,
393 TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE);
395 if (copy_to_user(optval,
396 crypto_info_aes_gcm_128,
397 sizeof(*crypto_info_aes_gcm_128)))
401 case TLS_CIPHER_AES_GCM_256: {
402 struct tls12_crypto_info_aes_gcm_256 *
403 crypto_info_aes_gcm_256 =
404 container_of(crypto_info,
405 struct tls12_crypto_info_aes_gcm_256,
408 if (len != sizeof(*crypto_info_aes_gcm_256)) {
413 memcpy(crypto_info_aes_gcm_256->iv,
414 ctx->tx.iv + TLS_CIPHER_AES_GCM_256_SALT_SIZE,
415 TLS_CIPHER_AES_GCM_256_IV_SIZE);
416 memcpy(crypto_info_aes_gcm_256->rec_seq, ctx->tx.rec_seq,
417 TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE);
419 if (copy_to_user(optval,
420 crypto_info_aes_gcm_256,
421 sizeof(*crypto_info_aes_gcm_256)))
433 static int do_tls_getsockopt(struct sock *sk, int optname,
434 char __user *optval, int __user *optlen)
440 rc = do_tls_getsockopt_tx(sk, optval, optlen);
449 static int tls_getsockopt(struct sock *sk, int level, int optname,
450 char __user *optval, int __user *optlen)
452 struct tls_context *ctx = tls_get_ctx(sk);
454 if (level != SOL_TLS)
455 return ctx->sk_proto->getsockopt(sk, level,
456 optname, optval, optlen);
458 return do_tls_getsockopt(sk, optname, optval, optlen);
461 static int do_tls_setsockopt_conf(struct sock *sk, char __user *optval,
462 unsigned int optlen, int tx)
464 struct tls_crypto_info *crypto_info;
465 struct tls_crypto_info *alt_crypto_info;
466 struct tls_context *ctx = tls_get_ctx(sk);
471 if (!optval || (optlen < sizeof(*crypto_info))) {
477 crypto_info = &ctx->crypto_send.info;
478 alt_crypto_info = &ctx->crypto_recv.info;
480 crypto_info = &ctx->crypto_recv.info;
481 alt_crypto_info = &ctx->crypto_send.info;
484 /* Currently we don't support set crypto info more than one time */
485 if (TLS_CRYPTO_INFO_READY(crypto_info)) {
490 rc = copy_from_user(crypto_info, optval, sizeof(*crypto_info));
493 goto err_crypto_info;
497 if (crypto_info->version != TLS_1_2_VERSION &&
498 crypto_info->version != TLS_1_3_VERSION) {
500 goto err_crypto_info;
503 /* Ensure that TLS version and ciphers are same in both directions */
504 if (TLS_CRYPTO_INFO_READY(alt_crypto_info)) {
505 if (alt_crypto_info->version != crypto_info->version ||
506 alt_crypto_info->cipher_type != crypto_info->cipher_type) {
508 goto err_crypto_info;
512 switch (crypto_info->cipher_type) {
513 case TLS_CIPHER_AES_GCM_128:
514 optsize = sizeof(struct tls12_crypto_info_aes_gcm_128);
516 case TLS_CIPHER_AES_GCM_256: {
517 optsize = sizeof(struct tls12_crypto_info_aes_gcm_256);
520 case TLS_CIPHER_AES_CCM_128:
521 optsize = sizeof(struct tls12_crypto_info_aes_ccm_128);
525 goto err_crypto_info;
528 if (optlen != optsize) {
530 goto err_crypto_info;
533 rc = copy_from_user(crypto_info + 1, optval + sizeof(*crypto_info),
534 optlen - sizeof(*crypto_info));
537 goto err_crypto_info;
541 rc = tls_set_device_offload(sk, ctx);
544 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSTXDEVICE);
545 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXDEVICE);
547 rc = tls_set_sw_offload(sk, ctx, 1);
549 goto err_crypto_info;
550 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSTXSW);
551 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXSW);
555 rc = tls_set_device_offload_rx(sk, ctx);
558 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXDEVICE);
559 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXDEVICE);
561 rc = tls_set_sw_offload(sk, ctx, 0);
563 goto err_crypto_info;
564 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXSW);
565 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXSW);
568 tls_sw_strparser_arm(sk, ctx);
575 update_sk_prot(sk, ctx);
577 ctx->sk_write_space = sk->sk_write_space;
578 sk->sk_write_space = tls_write_space;
580 sk->sk_socket->ops = &tls_sw_proto_ops;
585 memzero_explicit(crypto_info, sizeof(union tls_crypto_context));
590 static int do_tls_setsockopt(struct sock *sk, int optname,
591 char __user *optval, unsigned int optlen)
599 rc = do_tls_setsockopt_conf(sk, optval, optlen,
610 static int tls_setsockopt(struct sock *sk, int level, int optname,
611 char __user *optval, unsigned int optlen)
613 struct tls_context *ctx = tls_get_ctx(sk);
615 if (level != SOL_TLS)
616 return ctx->sk_proto->setsockopt(sk, level, optname, optval,
619 return do_tls_setsockopt(sk, optname, optval, optlen);
622 struct tls_context *tls_ctx_create(struct sock *sk)
624 struct inet_connection_sock *icsk = inet_csk(sk);
625 struct tls_context *ctx;
627 ctx = kzalloc(sizeof(*ctx), GFP_ATOMIC);
631 mutex_init(&ctx->tx_lock);
632 rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
633 ctx->sk_proto = sk->sk_prot;
637 static void tls_build_proto(struct sock *sk)
639 int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4;
641 /* Build IPv6 TLS whenever the address of tcpv6 _prot changes */
642 if (ip_ver == TLSV6 &&
643 unlikely(sk->sk_prot != smp_load_acquire(&saved_tcpv6_prot))) {
644 mutex_lock(&tcpv6_prot_mutex);
645 if (likely(sk->sk_prot != saved_tcpv6_prot)) {
646 build_protos(tls_prots[TLSV6], sk->sk_prot);
647 smp_store_release(&saved_tcpv6_prot, sk->sk_prot);
649 mutex_unlock(&tcpv6_prot_mutex);
652 if (ip_ver == TLSV4 &&
653 unlikely(sk->sk_prot != smp_load_acquire(&saved_tcpv4_prot))) {
654 mutex_lock(&tcpv4_prot_mutex);
655 if (likely(sk->sk_prot != saved_tcpv4_prot)) {
656 build_protos(tls_prots[TLSV4], sk->sk_prot);
657 smp_store_release(&saved_tcpv4_prot, sk->sk_prot);
659 mutex_unlock(&tcpv4_prot_mutex);
663 static void build_protos(struct proto prot[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
666 prot[TLS_BASE][TLS_BASE] = *base;
667 prot[TLS_BASE][TLS_BASE].setsockopt = tls_setsockopt;
668 prot[TLS_BASE][TLS_BASE].getsockopt = tls_getsockopt;
669 prot[TLS_BASE][TLS_BASE].close = tls_sk_proto_close;
671 prot[TLS_SW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
672 prot[TLS_SW][TLS_BASE].sendmsg = tls_sw_sendmsg;
673 prot[TLS_SW][TLS_BASE].sendpage = tls_sw_sendpage;
675 prot[TLS_BASE][TLS_SW] = prot[TLS_BASE][TLS_BASE];
676 prot[TLS_BASE][TLS_SW].recvmsg = tls_sw_recvmsg;
677 prot[TLS_BASE][TLS_SW].stream_memory_read = tls_sw_stream_read;
678 prot[TLS_BASE][TLS_SW].close = tls_sk_proto_close;
680 prot[TLS_SW][TLS_SW] = prot[TLS_SW][TLS_BASE];
681 prot[TLS_SW][TLS_SW].recvmsg = tls_sw_recvmsg;
682 prot[TLS_SW][TLS_SW].stream_memory_read = tls_sw_stream_read;
683 prot[TLS_SW][TLS_SW].close = tls_sk_proto_close;
685 #ifdef CONFIG_TLS_DEVICE
686 prot[TLS_HW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
687 prot[TLS_HW][TLS_BASE].sendmsg = tls_device_sendmsg;
688 prot[TLS_HW][TLS_BASE].sendpage = tls_device_sendpage;
690 prot[TLS_HW][TLS_SW] = prot[TLS_BASE][TLS_SW];
691 prot[TLS_HW][TLS_SW].sendmsg = tls_device_sendmsg;
692 prot[TLS_HW][TLS_SW].sendpage = tls_device_sendpage;
694 prot[TLS_BASE][TLS_HW] = prot[TLS_BASE][TLS_SW];
696 prot[TLS_SW][TLS_HW] = prot[TLS_SW][TLS_SW];
698 prot[TLS_HW][TLS_HW] = prot[TLS_HW][TLS_SW];
700 #ifdef CONFIG_TLS_TOE
701 prot[TLS_HW_RECORD][TLS_HW_RECORD] = *base;
702 prot[TLS_HW_RECORD][TLS_HW_RECORD].hash = tls_toe_hash;
703 prot[TLS_HW_RECORD][TLS_HW_RECORD].unhash = tls_toe_unhash;
707 static int tls_init(struct sock *sk)
709 struct tls_context *ctx;
714 #ifdef CONFIG_TLS_TOE
715 if (tls_toe_bypass(sk))
719 /* The TLS ulp is currently supported only for TCP sockets
720 * in ESTABLISHED state.
721 * Supporting sockets in LISTEN state will require us
722 * to modify the accept implementation to clone rather then
723 * share the ulp context.
725 if (sk->sk_state != TCP_ESTABLISHED)
728 /* allocate tls context */
729 write_lock_bh(&sk->sk_callback_lock);
730 ctx = tls_ctx_create(sk);
736 ctx->tx_conf = TLS_BASE;
737 ctx->rx_conf = TLS_BASE;
738 update_sk_prot(sk, ctx);
740 write_unlock_bh(&sk->sk_callback_lock);
744 static void tls_update(struct sock *sk, struct proto *p)
746 struct tls_context *ctx;
748 ctx = tls_get_ctx(sk);
755 static int tls_get_info(const struct sock *sk, struct sk_buff *skb)
757 u16 version, cipher_type;
758 struct tls_context *ctx;
759 struct nlattr *start;
762 start = nla_nest_start_noflag(skb, INET_ULP_INFO_TLS);
767 ctx = rcu_dereference(inet_csk(sk)->icsk_ulp_data);
772 version = ctx->prot_info.version;
774 err = nla_put_u16(skb, TLS_INFO_VERSION, version);
778 cipher_type = ctx->prot_info.cipher_type;
780 err = nla_put_u16(skb, TLS_INFO_CIPHER, cipher_type);
784 err = nla_put_u16(skb, TLS_INFO_TXCONF, tls_user_config(ctx, true));
788 err = nla_put_u16(skb, TLS_INFO_RXCONF, tls_user_config(ctx, false));
793 nla_nest_end(skb, start);
798 nla_nest_cancel(skb, start);
802 static size_t tls_get_info_size(const struct sock *sk)
806 size += nla_total_size(0) + /* INET_ULP_INFO_TLS */
807 nla_total_size(sizeof(u16)) + /* TLS_INFO_VERSION */
808 nla_total_size(sizeof(u16)) + /* TLS_INFO_CIPHER */
809 nla_total_size(sizeof(u16)) + /* TLS_INFO_RXCONF */
810 nla_total_size(sizeof(u16)) + /* TLS_INFO_TXCONF */
816 static int __net_init tls_init_net(struct net *net)
820 net->mib.tls_statistics = alloc_percpu(struct linux_tls_mib);
821 if (!net->mib.tls_statistics)
824 err = tls_proc_init(net);
830 free_percpu(net->mib.tls_statistics);
834 static void __net_exit tls_exit_net(struct net *net)
837 free_percpu(net->mib.tls_statistics);
840 static struct pernet_operations tls_proc_ops = {
841 .init = tls_init_net,
842 .exit = tls_exit_net,
845 static struct tcp_ulp_ops tcp_tls_ulp_ops __read_mostly = {
847 .owner = THIS_MODULE,
849 .update = tls_update,
850 .get_info = tls_get_info,
851 .get_info_size = tls_get_info_size,
854 static int __init tls_register(void)
858 err = register_pernet_subsys(&tls_proc_ops);
862 tls_sw_proto_ops = inet_stream_ops;
863 tls_sw_proto_ops.splice_read = tls_sw_splice_read;
864 tls_sw_proto_ops.sendpage_locked = tls_sw_sendpage_locked,
867 tcp_register_ulp(&tcp_tls_ulp_ops);
872 static void __exit tls_unregister(void)
874 tcp_unregister_ulp(&tcp_tls_ulp_ops);
875 tls_device_cleanup();
876 unregister_pernet_subsys(&tls_proc_ops);
879 module_init(tls_register);
880 module_exit(tls_unregister);