2 * Copyright (c) 2016-2017, Mellanox Technologies. 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 void tls_free_partial_record(struct sock *sk, struct tls_context *ctx)
214 struct scatterlist *sg;
216 for (sg = ctx->partially_sent_record; sg; sg = sg_next(sg)) {
217 put_page(sg_page(sg));
218 sk_mem_uncharge(sk, sg->length);
220 ctx->partially_sent_record = NULL;
223 static void tls_write_space(struct sock *sk)
225 struct tls_context *ctx = tls_get_ctx(sk);
227 /* If in_tcp_sendpages call lower protocol write space handler
228 * to ensure we wake up any waiting operations there. For example
229 * if do_tcp_sendpages where to call sk_wait_event.
231 if (ctx->in_tcp_sendpages) {
232 ctx->sk_write_space(sk);
236 #ifdef CONFIG_TLS_DEVICE
237 if (ctx->tx_conf == TLS_HW)
238 tls_device_write_space(sk, ctx);
241 tls_sw_write_space(sk, ctx);
243 ctx->sk_write_space(sk);
247 * tls_ctx_free() - free TLS ULP context
248 * @sk: socket to with @ctx is attached
249 * @ctx: TLS context structure
251 * Free TLS context. If @sk is %NULL caller guarantees that the socket
252 * to which @ctx was attached has no outstanding references.
254 void tls_ctx_free(struct sock *sk, struct tls_context *ctx)
259 memzero_explicit(&ctx->crypto_send, sizeof(ctx->crypto_send));
260 memzero_explicit(&ctx->crypto_recv, sizeof(ctx->crypto_recv));
261 mutex_destroy(&ctx->tx_lock);
269 static void tls_sk_proto_cleanup(struct sock *sk,
270 struct tls_context *ctx, long timeo)
272 if (unlikely(sk->sk_write_pending) &&
273 !wait_on_pending_writer(sk, &timeo))
274 tls_handle_open_record(sk, 0);
276 /* We need these for tls_sw_fallback handling of other packets */
277 if (ctx->tx_conf == TLS_SW) {
278 kfree(ctx->tx.rec_seq);
280 tls_sw_release_resources_tx(sk);
281 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXSW);
282 } else if (ctx->tx_conf == TLS_HW) {
283 tls_device_free_resources_tx(sk);
284 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXDEVICE);
287 if (ctx->rx_conf == TLS_SW) {
288 tls_sw_release_resources_rx(sk);
289 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXSW);
290 } else if (ctx->rx_conf == TLS_HW) {
291 tls_device_offload_cleanup_rx(sk);
292 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXDEVICE);
296 static void tls_sk_proto_close(struct sock *sk, long timeout)
298 struct inet_connection_sock *icsk = inet_csk(sk);
299 struct tls_context *ctx = tls_get_ctx(sk);
300 long timeo = sock_sndtimeo(sk, 0);
303 if (ctx->tx_conf == TLS_SW)
304 tls_sw_cancel_work_tx(ctx);
307 free_ctx = ctx->tx_conf != TLS_HW && ctx->rx_conf != TLS_HW;
309 if (ctx->tx_conf != TLS_BASE || ctx->rx_conf != TLS_BASE)
310 tls_sk_proto_cleanup(sk, ctx, timeo);
312 write_lock_bh(&sk->sk_callback_lock);
314 rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
315 sk->sk_prot = ctx->sk_proto;
316 if (sk->sk_write_space == tls_write_space)
317 sk->sk_write_space = ctx->sk_write_space;
318 write_unlock_bh(&sk->sk_callback_lock);
320 if (ctx->tx_conf == TLS_SW)
321 tls_sw_free_ctx_tx(ctx);
322 if (ctx->rx_conf == TLS_SW || ctx->rx_conf == TLS_HW)
323 tls_sw_strparser_done(ctx);
324 if (ctx->rx_conf == TLS_SW)
325 tls_sw_free_ctx_rx(ctx);
326 ctx->sk_proto->close(sk, timeout);
329 tls_ctx_free(sk, ctx);
332 static int do_tls_getsockopt_tx(struct sock *sk, char __user *optval,
336 struct tls_context *ctx = tls_get_ctx(sk);
337 struct tls_crypto_info *crypto_info;
340 if (get_user(len, optlen))
343 if (!optval || (len < sizeof(*crypto_info))) {
353 /* get user crypto info */
354 crypto_info = &ctx->crypto_send.info;
356 if (!TLS_CRYPTO_INFO_READY(crypto_info)) {
361 if (len == sizeof(*crypto_info)) {
362 if (copy_to_user(optval, crypto_info, sizeof(*crypto_info)))
367 switch (crypto_info->cipher_type) {
368 case TLS_CIPHER_AES_GCM_128: {
369 struct tls12_crypto_info_aes_gcm_128 *
370 crypto_info_aes_gcm_128 =
371 container_of(crypto_info,
372 struct tls12_crypto_info_aes_gcm_128,
375 if (len != sizeof(*crypto_info_aes_gcm_128)) {
380 memcpy(crypto_info_aes_gcm_128->iv,
381 ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
382 TLS_CIPHER_AES_GCM_128_IV_SIZE);
383 memcpy(crypto_info_aes_gcm_128->rec_seq, ctx->tx.rec_seq,
384 TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE);
386 if (copy_to_user(optval,
387 crypto_info_aes_gcm_128,
388 sizeof(*crypto_info_aes_gcm_128)))
392 case TLS_CIPHER_AES_GCM_256: {
393 struct tls12_crypto_info_aes_gcm_256 *
394 crypto_info_aes_gcm_256 =
395 container_of(crypto_info,
396 struct tls12_crypto_info_aes_gcm_256,
399 if (len != sizeof(*crypto_info_aes_gcm_256)) {
404 memcpy(crypto_info_aes_gcm_256->iv,
405 ctx->tx.iv + TLS_CIPHER_AES_GCM_256_SALT_SIZE,
406 TLS_CIPHER_AES_GCM_256_IV_SIZE);
407 memcpy(crypto_info_aes_gcm_256->rec_seq, ctx->tx.rec_seq,
408 TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE);
410 if (copy_to_user(optval,
411 crypto_info_aes_gcm_256,
412 sizeof(*crypto_info_aes_gcm_256)))
424 static int do_tls_getsockopt(struct sock *sk, int optname,
425 char __user *optval, int __user *optlen)
431 rc = do_tls_getsockopt_tx(sk, optval, optlen);
440 static int tls_getsockopt(struct sock *sk, int level, int optname,
441 char __user *optval, int __user *optlen)
443 struct tls_context *ctx = tls_get_ctx(sk);
445 if (level != SOL_TLS)
446 return ctx->sk_proto->getsockopt(sk, level,
447 optname, optval, optlen);
449 return do_tls_getsockopt(sk, optname, optval, optlen);
452 static int do_tls_setsockopt_conf(struct sock *sk, char __user *optval,
453 unsigned int optlen, int tx)
455 struct tls_crypto_info *crypto_info;
456 struct tls_crypto_info *alt_crypto_info;
457 struct tls_context *ctx = tls_get_ctx(sk);
462 if (!optval || (optlen < sizeof(*crypto_info))) {
468 crypto_info = &ctx->crypto_send.info;
469 alt_crypto_info = &ctx->crypto_recv.info;
471 crypto_info = &ctx->crypto_recv.info;
472 alt_crypto_info = &ctx->crypto_send.info;
475 /* Currently we don't support set crypto info more than one time */
476 if (TLS_CRYPTO_INFO_READY(crypto_info)) {
481 rc = copy_from_user(crypto_info, optval, sizeof(*crypto_info));
484 goto err_crypto_info;
488 if (crypto_info->version != TLS_1_2_VERSION &&
489 crypto_info->version != TLS_1_3_VERSION) {
491 goto err_crypto_info;
494 /* Ensure that TLS version and ciphers are same in both directions */
495 if (TLS_CRYPTO_INFO_READY(alt_crypto_info)) {
496 if (alt_crypto_info->version != crypto_info->version ||
497 alt_crypto_info->cipher_type != crypto_info->cipher_type) {
499 goto err_crypto_info;
503 switch (crypto_info->cipher_type) {
504 case TLS_CIPHER_AES_GCM_128:
505 optsize = sizeof(struct tls12_crypto_info_aes_gcm_128);
507 case TLS_CIPHER_AES_GCM_256: {
508 optsize = sizeof(struct tls12_crypto_info_aes_gcm_256);
511 case TLS_CIPHER_AES_CCM_128:
512 optsize = sizeof(struct tls12_crypto_info_aes_ccm_128);
516 goto err_crypto_info;
519 if (optlen != optsize) {
521 goto err_crypto_info;
524 rc = copy_from_user(crypto_info + 1, optval + sizeof(*crypto_info),
525 optlen - sizeof(*crypto_info));
528 goto err_crypto_info;
532 rc = tls_set_device_offload(sk, ctx);
535 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSTXDEVICE);
536 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXDEVICE);
538 rc = tls_set_sw_offload(sk, ctx, 1);
540 goto err_crypto_info;
541 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSTXSW);
542 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXSW);
546 rc = tls_set_device_offload_rx(sk, ctx);
549 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXDEVICE);
550 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXDEVICE);
552 rc = tls_set_sw_offload(sk, ctx, 0);
554 goto err_crypto_info;
555 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXSW);
556 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXSW);
559 tls_sw_strparser_arm(sk, ctx);
566 update_sk_prot(sk, ctx);
568 ctx->sk_write_space = sk->sk_write_space;
569 sk->sk_write_space = tls_write_space;
571 sk->sk_socket->ops = &tls_sw_proto_ops;
576 memzero_explicit(crypto_info, sizeof(union tls_crypto_context));
581 static int do_tls_setsockopt(struct sock *sk, int optname,
582 char __user *optval, unsigned int optlen)
590 rc = do_tls_setsockopt_conf(sk, optval, optlen,
601 static int tls_setsockopt(struct sock *sk, int level, int optname,
602 char __user *optval, unsigned int optlen)
604 struct tls_context *ctx = tls_get_ctx(sk);
606 if (level != SOL_TLS)
607 return ctx->sk_proto->setsockopt(sk, level, optname, optval,
610 return do_tls_setsockopt(sk, optname, optval, optlen);
613 struct tls_context *tls_ctx_create(struct sock *sk)
615 struct inet_connection_sock *icsk = inet_csk(sk);
616 struct tls_context *ctx;
618 ctx = kzalloc(sizeof(*ctx), GFP_ATOMIC);
622 mutex_init(&ctx->tx_lock);
623 rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
624 ctx->sk_proto = sk->sk_prot;
628 static void tls_build_proto(struct sock *sk)
630 int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4;
632 /* Build IPv6 TLS whenever the address of tcpv6 _prot changes */
633 if (ip_ver == TLSV6 &&
634 unlikely(sk->sk_prot != smp_load_acquire(&saved_tcpv6_prot))) {
635 mutex_lock(&tcpv6_prot_mutex);
636 if (likely(sk->sk_prot != saved_tcpv6_prot)) {
637 build_protos(tls_prots[TLSV6], sk->sk_prot);
638 smp_store_release(&saved_tcpv6_prot, sk->sk_prot);
640 mutex_unlock(&tcpv6_prot_mutex);
643 if (ip_ver == TLSV4 &&
644 unlikely(sk->sk_prot != smp_load_acquire(&saved_tcpv4_prot))) {
645 mutex_lock(&tcpv4_prot_mutex);
646 if (likely(sk->sk_prot != saved_tcpv4_prot)) {
647 build_protos(tls_prots[TLSV4], sk->sk_prot);
648 smp_store_release(&saved_tcpv4_prot, sk->sk_prot);
650 mutex_unlock(&tcpv4_prot_mutex);
654 static void build_protos(struct proto prot[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
657 prot[TLS_BASE][TLS_BASE] = *base;
658 prot[TLS_BASE][TLS_BASE].setsockopt = tls_setsockopt;
659 prot[TLS_BASE][TLS_BASE].getsockopt = tls_getsockopt;
660 prot[TLS_BASE][TLS_BASE].close = tls_sk_proto_close;
662 prot[TLS_SW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
663 prot[TLS_SW][TLS_BASE].sendmsg = tls_sw_sendmsg;
664 prot[TLS_SW][TLS_BASE].sendpage = tls_sw_sendpage;
666 prot[TLS_BASE][TLS_SW] = prot[TLS_BASE][TLS_BASE];
667 prot[TLS_BASE][TLS_SW].recvmsg = tls_sw_recvmsg;
668 prot[TLS_BASE][TLS_SW].stream_memory_read = tls_sw_stream_read;
669 prot[TLS_BASE][TLS_SW].close = tls_sk_proto_close;
671 prot[TLS_SW][TLS_SW] = prot[TLS_SW][TLS_BASE];
672 prot[TLS_SW][TLS_SW].recvmsg = tls_sw_recvmsg;
673 prot[TLS_SW][TLS_SW].stream_memory_read = tls_sw_stream_read;
674 prot[TLS_SW][TLS_SW].close = tls_sk_proto_close;
676 #ifdef CONFIG_TLS_DEVICE
677 prot[TLS_HW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
678 prot[TLS_HW][TLS_BASE].sendmsg = tls_device_sendmsg;
679 prot[TLS_HW][TLS_BASE].sendpage = tls_device_sendpage;
681 prot[TLS_HW][TLS_SW] = prot[TLS_BASE][TLS_SW];
682 prot[TLS_HW][TLS_SW].sendmsg = tls_device_sendmsg;
683 prot[TLS_HW][TLS_SW].sendpage = tls_device_sendpage;
685 prot[TLS_BASE][TLS_HW] = prot[TLS_BASE][TLS_SW];
687 prot[TLS_SW][TLS_HW] = prot[TLS_SW][TLS_SW];
689 prot[TLS_HW][TLS_HW] = prot[TLS_HW][TLS_SW];
691 #ifdef CONFIG_TLS_TOE
692 prot[TLS_HW_RECORD][TLS_HW_RECORD] = *base;
693 prot[TLS_HW_RECORD][TLS_HW_RECORD].hash = tls_toe_hash;
694 prot[TLS_HW_RECORD][TLS_HW_RECORD].unhash = tls_toe_unhash;
698 static int tls_init(struct sock *sk)
700 struct tls_context *ctx;
705 #ifdef CONFIG_TLS_TOE
706 if (tls_toe_bypass(sk))
710 /* The TLS ulp is currently supported only for TCP sockets
711 * in ESTABLISHED state.
712 * Supporting sockets in LISTEN state will require us
713 * to modify the accept implementation to clone rather then
714 * share the ulp context.
716 if (sk->sk_state != TCP_ESTABLISHED)
719 /* allocate tls context */
720 write_lock_bh(&sk->sk_callback_lock);
721 ctx = tls_ctx_create(sk);
727 ctx->tx_conf = TLS_BASE;
728 ctx->rx_conf = TLS_BASE;
729 update_sk_prot(sk, ctx);
731 write_unlock_bh(&sk->sk_callback_lock);
735 static void tls_update(struct sock *sk, struct proto *p,
736 void (*write_space)(struct sock *sk))
738 struct tls_context *ctx;
740 ctx = tls_get_ctx(sk);
742 ctx->sk_write_space = write_space;
746 sk->sk_write_space = write_space;
750 static int tls_get_info(const struct sock *sk, struct sk_buff *skb)
752 u16 version, cipher_type;
753 struct tls_context *ctx;
754 struct nlattr *start;
757 start = nla_nest_start_noflag(skb, INET_ULP_INFO_TLS);
762 ctx = rcu_dereference(inet_csk(sk)->icsk_ulp_data);
767 version = ctx->prot_info.version;
769 err = nla_put_u16(skb, TLS_INFO_VERSION, version);
773 cipher_type = ctx->prot_info.cipher_type;
775 err = nla_put_u16(skb, TLS_INFO_CIPHER, cipher_type);
779 err = nla_put_u16(skb, TLS_INFO_TXCONF, tls_user_config(ctx, true));
783 err = nla_put_u16(skb, TLS_INFO_RXCONF, tls_user_config(ctx, false));
788 nla_nest_end(skb, start);
793 nla_nest_cancel(skb, start);
797 static size_t tls_get_info_size(const struct sock *sk)
801 size += nla_total_size(0) + /* INET_ULP_INFO_TLS */
802 nla_total_size(sizeof(u16)) + /* TLS_INFO_VERSION */
803 nla_total_size(sizeof(u16)) + /* TLS_INFO_CIPHER */
804 nla_total_size(sizeof(u16)) + /* TLS_INFO_RXCONF */
805 nla_total_size(sizeof(u16)) + /* TLS_INFO_TXCONF */
811 static int __net_init tls_init_net(struct net *net)
815 net->mib.tls_statistics = alloc_percpu(struct linux_tls_mib);
816 if (!net->mib.tls_statistics)
819 err = tls_proc_init(net);
825 free_percpu(net->mib.tls_statistics);
829 static void __net_exit tls_exit_net(struct net *net)
832 free_percpu(net->mib.tls_statistics);
835 static struct pernet_operations tls_proc_ops = {
836 .init = tls_init_net,
837 .exit = tls_exit_net,
840 static struct tcp_ulp_ops tcp_tls_ulp_ops __read_mostly = {
842 .owner = THIS_MODULE,
844 .update = tls_update,
845 .get_info = tls_get_info,
846 .get_info_size = tls_get_info_size,
849 static int __init tls_register(void)
853 err = register_pernet_subsys(&tls_proc_ops);
857 tls_sw_proto_ops = inet_stream_ops;
858 tls_sw_proto_ops.splice_read = tls_sw_splice_read;
859 tls_sw_proto_ops.sendpage_locked = tls_sw_sendpage_locked,
862 tcp_register_ulp(&tcp_tls_ulp_ops);
867 static void __exit tls_unregister(void)
869 tcp_unregister_ulp(&tcp_tls_ulp_ops);
870 tls_device_cleanup();
871 unregister_pernet_subsys(&tls_proc_ops);
874 module_init(tls_register);
875 module_exit(tls_unregister);