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>
50 MODULE_AUTHOR("Mellanox Technologies");
51 MODULE_DESCRIPTION("Transport Layer Security Support");
52 MODULE_LICENSE("Dual BSD/GPL");
53 MODULE_ALIAS_TCP_ULP("tls");
61 #define CIPHER_SIZE_DESC(cipher) [cipher] = { \
62 .iv = cipher ## _IV_SIZE, \
63 .key = cipher ## _KEY_SIZE, \
64 .salt = cipher ## _SALT_SIZE, \
65 .tag = cipher ## _TAG_SIZE, \
66 .rec_seq = cipher ## _REC_SEQ_SIZE, \
69 const struct tls_cipher_size_desc tls_cipher_size_desc[] = {
70 CIPHER_SIZE_DESC(TLS_CIPHER_AES_GCM_128),
71 CIPHER_SIZE_DESC(TLS_CIPHER_AES_GCM_256),
72 CIPHER_SIZE_DESC(TLS_CIPHER_AES_CCM_128),
73 CIPHER_SIZE_DESC(TLS_CIPHER_CHACHA20_POLY1305),
74 CIPHER_SIZE_DESC(TLS_CIPHER_SM4_GCM),
75 CIPHER_SIZE_DESC(TLS_CIPHER_SM4_CCM),
78 static const struct proto *saved_tcpv6_prot;
79 static DEFINE_MUTEX(tcpv6_prot_mutex);
80 static const struct proto *saved_tcpv4_prot;
81 static DEFINE_MUTEX(tcpv4_prot_mutex);
82 static struct proto tls_prots[TLS_NUM_PROTS][TLS_NUM_CONFIG][TLS_NUM_CONFIG];
83 static struct proto_ops tls_proto_ops[TLS_NUM_PROTS][TLS_NUM_CONFIG][TLS_NUM_CONFIG];
84 static void build_protos(struct proto prot[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
85 const struct proto *base);
87 void update_sk_prot(struct sock *sk, struct tls_context *ctx)
89 int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4;
91 WRITE_ONCE(sk->sk_prot,
92 &tls_prots[ip_ver][ctx->tx_conf][ctx->rx_conf]);
93 WRITE_ONCE(sk->sk_socket->ops,
94 &tls_proto_ops[ip_ver][ctx->tx_conf][ctx->rx_conf]);
97 int wait_on_pending_writer(struct sock *sk, long *timeo)
100 DEFINE_WAIT_FUNC(wait, woken_wake_function);
102 add_wait_queue(sk_sleep(sk), &wait);
109 if (signal_pending(current)) {
110 rc = sock_intr_errno(*timeo);
114 if (sk_wait_event(sk, timeo, !sk->sk_write_pending, &wait))
117 remove_wait_queue(sk_sleep(sk), &wait);
121 int tls_push_sg(struct sock *sk,
122 struct tls_context *ctx,
123 struct scatterlist *sg,
127 int sendpage_flags = flags | MSG_SENDPAGE_NOTLAST;
131 int offset = first_offset;
133 size = sg->length - offset;
134 offset += sg->offset;
136 ctx->in_tcp_sendpages = true;
139 sendpage_flags = flags;
141 /* is sending application-limited? */
142 tcp_rate_check_app_limited(sk);
145 ret = do_tcp_sendpages(sk, p, offset, size, sendpage_flags);
154 offset -= sg->offset;
155 ctx->partially_sent_offset = offset;
156 ctx->partially_sent_record = (void *)sg;
157 ctx->in_tcp_sendpages = false;
162 sk_mem_uncharge(sk, sg->length);
171 ctx->in_tcp_sendpages = false;
176 static int tls_handle_open_record(struct sock *sk, int flags)
178 struct tls_context *ctx = tls_get_ctx(sk);
180 if (tls_is_pending_open_record(ctx))
181 return ctx->push_pending_record(sk, flags);
186 int tls_process_cmsg(struct sock *sk, struct msghdr *msg,
187 unsigned char *record_type)
189 struct cmsghdr *cmsg;
192 for_each_cmsghdr(cmsg, msg) {
193 if (!CMSG_OK(msg, cmsg))
195 if (cmsg->cmsg_level != SOL_TLS)
198 switch (cmsg->cmsg_type) {
199 case TLS_SET_RECORD_TYPE:
200 if (cmsg->cmsg_len < CMSG_LEN(sizeof(*record_type)))
203 if (msg->msg_flags & MSG_MORE)
206 rc = tls_handle_open_record(sk, msg->msg_flags);
210 *record_type = *(unsigned char *)CMSG_DATA(cmsg);
221 int tls_push_partial_record(struct sock *sk, struct tls_context *ctx,
224 struct scatterlist *sg;
227 sg = ctx->partially_sent_record;
228 offset = ctx->partially_sent_offset;
230 ctx->partially_sent_record = NULL;
231 return tls_push_sg(sk, ctx, sg, offset, flags);
234 void tls_free_partial_record(struct sock *sk, struct tls_context *ctx)
236 struct scatterlist *sg;
238 for (sg = ctx->partially_sent_record; sg; sg = sg_next(sg)) {
239 put_page(sg_page(sg));
240 sk_mem_uncharge(sk, sg->length);
242 ctx->partially_sent_record = NULL;
245 static void tls_write_space(struct sock *sk)
247 struct tls_context *ctx = tls_get_ctx(sk);
249 /* If in_tcp_sendpages call lower protocol write space handler
250 * to ensure we wake up any waiting operations there. For example
251 * if do_tcp_sendpages where to call sk_wait_event.
253 if (ctx->in_tcp_sendpages) {
254 ctx->sk_write_space(sk);
258 #ifdef CONFIG_TLS_DEVICE
259 if (ctx->tx_conf == TLS_HW)
260 tls_device_write_space(sk, ctx);
263 tls_sw_write_space(sk, ctx);
265 ctx->sk_write_space(sk);
269 * tls_ctx_free() - free TLS ULP context
270 * @sk: socket to with @ctx is attached
271 * @ctx: TLS context structure
273 * Free TLS context. If @sk is %NULL caller guarantees that the socket
274 * to which @ctx was attached has no outstanding references.
276 void tls_ctx_free(struct sock *sk, struct tls_context *ctx)
281 memzero_explicit(&ctx->crypto_send, sizeof(ctx->crypto_send));
282 memzero_explicit(&ctx->crypto_recv, sizeof(ctx->crypto_recv));
283 mutex_destroy(&ctx->tx_lock);
291 static void tls_sk_proto_cleanup(struct sock *sk,
292 struct tls_context *ctx, long timeo)
294 if (unlikely(sk->sk_write_pending) &&
295 !wait_on_pending_writer(sk, &timeo))
296 tls_handle_open_record(sk, 0);
298 /* We need these for tls_sw_fallback handling of other packets */
299 if (ctx->tx_conf == TLS_SW) {
300 kfree(ctx->tx.rec_seq);
302 tls_sw_release_resources_tx(sk);
303 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXSW);
304 } else if (ctx->tx_conf == TLS_HW) {
305 tls_device_free_resources_tx(sk);
306 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXDEVICE);
309 if (ctx->rx_conf == TLS_SW) {
310 tls_sw_release_resources_rx(sk);
311 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXSW);
312 } else if (ctx->rx_conf == TLS_HW) {
313 tls_device_offload_cleanup_rx(sk);
314 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXDEVICE);
318 static void tls_sk_proto_close(struct sock *sk, long timeout)
320 struct inet_connection_sock *icsk = inet_csk(sk);
321 struct tls_context *ctx = tls_get_ctx(sk);
322 long timeo = sock_sndtimeo(sk, 0);
325 if (ctx->tx_conf == TLS_SW)
326 tls_sw_cancel_work_tx(ctx);
329 free_ctx = ctx->tx_conf != TLS_HW && ctx->rx_conf != TLS_HW;
331 if (ctx->tx_conf != TLS_BASE || ctx->rx_conf != TLS_BASE)
332 tls_sk_proto_cleanup(sk, ctx, timeo);
334 write_lock_bh(&sk->sk_callback_lock);
336 rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
337 WRITE_ONCE(sk->sk_prot, ctx->sk_proto);
338 if (sk->sk_write_space == tls_write_space)
339 sk->sk_write_space = ctx->sk_write_space;
340 write_unlock_bh(&sk->sk_callback_lock);
342 if (ctx->tx_conf == TLS_SW)
343 tls_sw_free_ctx_tx(ctx);
344 if (ctx->rx_conf == TLS_SW || ctx->rx_conf == TLS_HW)
345 tls_sw_strparser_done(ctx);
346 if (ctx->rx_conf == TLS_SW)
347 tls_sw_free_ctx_rx(ctx);
348 ctx->sk_proto->close(sk, timeout);
351 tls_ctx_free(sk, ctx);
354 static int do_tls_getsockopt_conf(struct sock *sk, char __user *optval,
355 int __user *optlen, int tx)
358 struct tls_context *ctx = tls_get_ctx(sk);
359 struct tls_crypto_info *crypto_info;
360 struct cipher_context *cctx;
363 if (get_user(len, optlen))
366 if (!optval || (len < sizeof(*crypto_info))) {
376 /* get user crypto info */
378 crypto_info = &ctx->crypto_send.info;
381 crypto_info = &ctx->crypto_recv.info;
385 if (!TLS_CRYPTO_INFO_READY(crypto_info)) {
390 if (len == sizeof(*crypto_info)) {
391 if (copy_to_user(optval, crypto_info, sizeof(*crypto_info)))
396 switch (crypto_info->cipher_type) {
397 case TLS_CIPHER_AES_GCM_128: {
398 struct tls12_crypto_info_aes_gcm_128 *
399 crypto_info_aes_gcm_128 =
400 container_of(crypto_info,
401 struct tls12_crypto_info_aes_gcm_128,
404 if (len != sizeof(*crypto_info_aes_gcm_128)) {
409 memcpy(crypto_info_aes_gcm_128->iv,
410 cctx->iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
411 TLS_CIPHER_AES_GCM_128_IV_SIZE);
412 memcpy(crypto_info_aes_gcm_128->rec_seq, cctx->rec_seq,
413 TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE);
415 if (copy_to_user(optval,
416 crypto_info_aes_gcm_128,
417 sizeof(*crypto_info_aes_gcm_128)))
421 case TLS_CIPHER_AES_GCM_256: {
422 struct tls12_crypto_info_aes_gcm_256 *
423 crypto_info_aes_gcm_256 =
424 container_of(crypto_info,
425 struct tls12_crypto_info_aes_gcm_256,
428 if (len != sizeof(*crypto_info_aes_gcm_256)) {
433 memcpy(crypto_info_aes_gcm_256->iv,
434 cctx->iv + TLS_CIPHER_AES_GCM_256_SALT_SIZE,
435 TLS_CIPHER_AES_GCM_256_IV_SIZE);
436 memcpy(crypto_info_aes_gcm_256->rec_seq, cctx->rec_seq,
437 TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE);
439 if (copy_to_user(optval,
440 crypto_info_aes_gcm_256,
441 sizeof(*crypto_info_aes_gcm_256)))
445 case TLS_CIPHER_AES_CCM_128: {
446 struct tls12_crypto_info_aes_ccm_128 *aes_ccm_128 =
447 container_of(crypto_info,
448 struct tls12_crypto_info_aes_ccm_128, info);
450 if (len != sizeof(*aes_ccm_128)) {
455 memcpy(aes_ccm_128->iv,
456 cctx->iv + TLS_CIPHER_AES_CCM_128_SALT_SIZE,
457 TLS_CIPHER_AES_CCM_128_IV_SIZE);
458 memcpy(aes_ccm_128->rec_seq, cctx->rec_seq,
459 TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE);
461 if (copy_to_user(optval, aes_ccm_128, sizeof(*aes_ccm_128)))
465 case TLS_CIPHER_CHACHA20_POLY1305: {
466 struct tls12_crypto_info_chacha20_poly1305 *chacha20_poly1305 =
467 container_of(crypto_info,
468 struct tls12_crypto_info_chacha20_poly1305,
471 if (len != sizeof(*chacha20_poly1305)) {
476 memcpy(chacha20_poly1305->iv,
477 cctx->iv + TLS_CIPHER_CHACHA20_POLY1305_SALT_SIZE,
478 TLS_CIPHER_CHACHA20_POLY1305_IV_SIZE);
479 memcpy(chacha20_poly1305->rec_seq, cctx->rec_seq,
480 TLS_CIPHER_CHACHA20_POLY1305_REC_SEQ_SIZE);
482 if (copy_to_user(optval, chacha20_poly1305,
483 sizeof(*chacha20_poly1305)))
487 case TLS_CIPHER_SM4_GCM: {
488 struct tls12_crypto_info_sm4_gcm *sm4_gcm_info =
489 container_of(crypto_info,
490 struct tls12_crypto_info_sm4_gcm, info);
492 if (len != sizeof(*sm4_gcm_info)) {
497 memcpy(sm4_gcm_info->iv,
498 cctx->iv + TLS_CIPHER_SM4_GCM_SALT_SIZE,
499 TLS_CIPHER_SM4_GCM_IV_SIZE);
500 memcpy(sm4_gcm_info->rec_seq, cctx->rec_seq,
501 TLS_CIPHER_SM4_GCM_REC_SEQ_SIZE);
503 if (copy_to_user(optval, sm4_gcm_info, sizeof(*sm4_gcm_info)))
507 case TLS_CIPHER_SM4_CCM: {
508 struct tls12_crypto_info_sm4_ccm *sm4_ccm_info =
509 container_of(crypto_info,
510 struct tls12_crypto_info_sm4_ccm, info);
512 if (len != sizeof(*sm4_ccm_info)) {
517 memcpy(sm4_ccm_info->iv,
518 cctx->iv + TLS_CIPHER_SM4_CCM_SALT_SIZE,
519 TLS_CIPHER_SM4_CCM_IV_SIZE);
520 memcpy(sm4_ccm_info->rec_seq, cctx->rec_seq,
521 TLS_CIPHER_SM4_CCM_REC_SEQ_SIZE);
523 if (copy_to_user(optval, sm4_ccm_info, sizeof(*sm4_ccm_info)))
527 case TLS_CIPHER_ARIA_GCM_128: {
528 struct tls12_crypto_info_aria_gcm_128 *
529 crypto_info_aria_gcm_128 =
530 container_of(crypto_info,
531 struct tls12_crypto_info_aria_gcm_128,
534 if (len != sizeof(*crypto_info_aria_gcm_128)) {
539 memcpy(crypto_info_aria_gcm_128->iv,
540 cctx->iv + TLS_CIPHER_ARIA_GCM_128_SALT_SIZE,
541 TLS_CIPHER_ARIA_GCM_128_IV_SIZE);
542 memcpy(crypto_info_aria_gcm_128->rec_seq, cctx->rec_seq,
543 TLS_CIPHER_ARIA_GCM_128_REC_SEQ_SIZE);
545 if (copy_to_user(optval,
546 crypto_info_aria_gcm_128,
547 sizeof(*crypto_info_aria_gcm_128)))
551 case TLS_CIPHER_ARIA_GCM_256: {
552 struct tls12_crypto_info_aria_gcm_256 *
553 crypto_info_aria_gcm_256 =
554 container_of(crypto_info,
555 struct tls12_crypto_info_aria_gcm_256,
558 if (len != sizeof(*crypto_info_aria_gcm_256)) {
563 memcpy(crypto_info_aria_gcm_256->iv,
564 cctx->iv + TLS_CIPHER_ARIA_GCM_256_SALT_SIZE,
565 TLS_CIPHER_ARIA_GCM_256_IV_SIZE);
566 memcpy(crypto_info_aria_gcm_256->rec_seq, cctx->rec_seq,
567 TLS_CIPHER_ARIA_GCM_256_REC_SEQ_SIZE);
569 if (copy_to_user(optval,
570 crypto_info_aria_gcm_256,
571 sizeof(*crypto_info_aria_gcm_256)))
583 static int do_tls_getsockopt_tx_zc(struct sock *sk, char __user *optval,
586 struct tls_context *ctx = tls_get_ctx(sk);
590 if (get_user(len, optlen))
593 if (len != sizeof(value))
596 value = ctx->zerocopy_sendfile;
597 if (copy_to_user(optval, &value, sizeof(value)))
603 static int do_tls_getsockopt_no_pad(struct sock *sk, char __user *optval,
606 struct tls_context *ctx = tls_get_ctx(sk);
609 if (ctx->prot_info.version != TLS_1_3_VERSION)
612 if (get_user(len, optlen))
614 if (len < sizeof(value))
619 if (ctx->rx_conf == TLS_SW || ctx->rx_conf == TLS_HW)
620 value = ctx->rx_no_pad;
625 if (put_user(sizeof(value), optlen))
627 if (copy_to_user(optval, &value, sizeof(value)))
633 static int do_tls_getsockopt(struct sock *sk, int optname,
634 char __user *optval, int __user *optlen)
641 rc = do_tls_getsockopt_conf(sk, optval, optlen,
644 case TLS_TX_ZEROCOPY_RO:
645 rc = do_tls_getsockopt_tx_zc(sk, optval, optlen);
647 case TLS_RX_EXPECT_NO_PAD:
648 rc = do_tls_getsockopt_no_pad(sk, optval, optlen);
657 static int tls_getsockopt(struct sock *sk, int level, int optname,
658 char __user *optval, int __user *optlen)
660 struct tls_context *ctx = tls_get_ctx(sk);
662 if (level != SOL_TLS)
663 return ctx->sk_proto->getsockopt(sk, level,
664 optname, optval, optlen);
666 return do_tls_getsockopt(sk, optname, optval, optlen);
669 static int do_tls_setsockopt_conf(struct sock *sk, sockptr_t optval,
670 unsigned int optlen, int tx)
672 struct tls_crypto_info *crypto_info;
673 struct tls_crypto_info *alt_crypto_info;
674 struct tls_context *ctx = tls_get_ctx(sk);
679 if (sockptr_is_null(optval) || (optlen < sizeof(*crypto_info)))
683 crypto_info = &ctx->crypto_send.info;
684 alt_crypto_info = &ctx->crypto_recv.info;
686 crypto_info = &ctx->crypto_recv.info;
687 alt_crypto_info = &ctx->crypto_send.info;
690 /* Currently we don't support set crypto info more than one time */
691 if (TLS_CRYPTO_INFO_READY(crypto_info))
694 rc = copy_from_sockptr(crypto_info, optval, sizeof(*crypto_info));
697 goto err_crypto_info;
701 if (crypto_info->version != TLS_1_2_VERSION &&
702 crypto_info->version != TLS_1_3_VERSION) {
704 goto err_crypto_info;
707 /* Ensure that TLS version and ciphers are same in both directions */
708 if (TLS_CRYPTO_INFO_READY(alt_crypto_info)) {
709 if (alt_crypto_info->version != crypto_info->version ||
710 alt_crypto_info->cipher_type != crypto_info->cipher_type) {
712 goto err_crypto_info;
716 switch (crypto_info->cipher_type) {
717 case TLS_CIPHER_AES_GCM_128:
718 optsize = sizeof(struct tls12_crypto_info_aes_gcm_128);
720 case TLS_CIPHER_AES_GCM_256: {
721 optsize = sizeof(struct tls12_crypto_info_aes_gcm_256);
724 case TLS_CIPHER_AES_CCM_128:
725 optsize = sizeof(struct tls12_crypto_info_aes_ccm_128);
727 case TLS_CIPHER_CHACHA20_POLY1305:
728 optsize = sizeof(struct tls12_crypto_info_chacha20_poly1305);
730 case TLS_CIPHER_SM4_GCM:
731 optsize = sizeof(struct tls12_crypto_info_sm4_gcm);
733 case TLS_CIPHER_SM4_CCM:
734 optsize = sizeof(struct tls12_crypto_info_sm4_ccm);
736 case TLS_CIPHER_ARIA_GCM_128:
737 if (crypto_info->version != TLS_1_2_VERSION) {
739 goto err_crypto_info;
741 optsize = sizeof(struct tls12_crypto_info_aria_gcm_128);
743 case TLS_CIPHER_ARIA_GCM_256:
744 if (crypto_info->version != TLS_1_2_VERSION) {
746 goto err_crypto_info;
748 optsize = sizeof(struct tls12_crypto_info_aria_gcm_256);
752 goto err_crypto_info;
755 if (optlen != optsize) {
757 goto err_crypto_info;
760 rc = copy_from_sockptr_offset(crypto_info + 1, optval,
761 sizeof(*crypto_info),
762 optlen - sizeof(*crypto_info));
765 goto err_crypto_info;
769 rc = tls_set_device_offload(sk, ctx);
772 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSTXDEVICE);
773 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXDEVICE);
775 rc = tls_set_sw_offload(sk, ctx, 1);
777 goto err_crypto_info;
778 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSTXSW);
779 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXSW);
783 rc = tls_set_device_offload_rx(sk, ctx);
786 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXDEVICE);
787 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXDEVICE);
789 rc = tls_set_sw_offload(sk, ctx, 0);
791 goto err_crypto_info;
792 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXSW);
793 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXSW);
796 tls_sw_strparser_arm(sk, ctx);
803 update_sk_prot(sk, ctx);
805 ctx->sk_write_space = sk->sk_write_space;
806 sk->sk_write_space = tls_write_space;
808 struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(ctx);
810 tls_strp_check_rcv(&rx_ctx->strp);
815 memzero_explicit(crypto_info, sizeof(union tls_crypto_context));
819 static int do_tls_setsockopt_tx_zc(struct sock *sk, sockptr_t optval,
822 struct tls_context *ctx = tls_get_ctx(sk);
825 if (sockptr_is_null(optval) || optlen != sizeof(value))
828 if (copy_from_sockptr(&value, optval, sizeof(value)))
834 ctx->zerocopy_sendfile = value;
839 static int do_tls_setsockopt_no_pad(struct sock *sk, sockptr_t optval,
842 struct tls_context *ctx = tls_get_ctx(sk);
846 if (ctx->prot_info.version != TLS_1_3_VERSION ||
847 sockptr_is_null(optval) || optlen < sizeof(val))
850 rc = copy_from_sockptr(&val, optval, sizeof(val));
855 rc = check_zeroed_sockptr(optval, sizeof(val), optlen - sizeof(val));
857 return rc == 0 ? -EINVAL : rc;
861 if (ctx->rx_conf == TLS_SW || ctx->rx_conf == TLS_HW) {
862 ctx->rx_no_pad = val;
863 tls_update_rx_zc_capable(ctx);
871 static int do_tls_setsockopt(struct sock *sk, int optname, sockptr_t optval,
880 rc = do_tls_setsockopt_conf(sk, optval, optlen,
884 case TLS_TX_ZEROCOPY_RO:
886 rc = do_tls_setsockopt_tx_zc(sk, optval, optlen);
889 case TLS_RX_EXPECT_NO_PAD:
890 rc = do_tls_setsockopt_no_pad(sk, optval, optlen);
899 static int tls_setsockopt(struct sock *sk, int level, int optname,
900 sockptr_t optval, unsigned int optlen)
902 struct tls_context *ctx = tls_get_ctx(sk);
904 if (level != SOL_TLS)
905 return ctx->sk_proto->setsockopt(sk, level, optname, optval,
908 return do_tls_setsockopt(sk, optname, optval, optlen);
911 struct tls_context *tls_ctx_create(struct sock *sk)
913 struct inet_connection_sock *icsk = inet_csk(sk);
914 struct tls_context *ctx;
916 ctx = kzalloc(sizeof(*ctx), GFP_ATOMIC);
920 mutex_init(&ctx->tx_lock);
921 rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
922 ctx->sk_proto = READ_ONCE(sk->sk_prot);
927 static void build_proto_ops(struct proto_ops ops[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
928 const struct proto_ops *base)
930 ops[TLS_BASE][TLS_BASE] = *base;
932 ops[TLS_SW ][TLS_BASE] = ops[TLS_BASE][TLS_BASE];
933 ops[TLS_SW ][TLS_BASE].sendpage_locked = tls_sw_sendpage_locked;
935 ops[TLS_BASE][TLS_SW ] = ops[TLS_BASE][TLS_BASE];
936 ops[TLS_BASE][TLS_SW ].splice_read = tls_sw_splice_read;
938 ops[TLS_SW ][TLS_SW ] = ops[TLS_SW ][TLS_BASE];
939 ops[TLS_SW ][TLS_SW ].splice_read = tls_sw_splice_read;
941 #ifdef CONFIG_TLS_DEVICE
942 ops[TLS_HW ][TLS_BASE] = ops[TLS_BASE][TLS_BASE];
943 ops[TLS_HW ][TLS_BASE].sendpage_locked = NULL;
945 ops[TLS_HW ][TLS_SW ] = ops[TLS_BASE][TLS_SW ];
946 ops[TLS_HW ][TLS_SW ].sendpage_locked = NULL;
948 ops[TLS_BASE][TLS_HW ] = ops[TLS_BASE][TLS_SW ];
950 ops[TLS_SW ][TLS_HW ] = ops[TLS_SW ][TLS_SW ];
952 ops[TLS_HW ][TLS_HW ] = ops[TLS_HW ][TLS_SW ];
953 ops[TLS_HW ][TLS_HW ].sendpage_locked = NULL;
955 #ifdef CONFIG_TLS_TOE
956 ops[TLS_HW_RECORD][TLS_HW_RECORD] = *base;
960 static void tls_build_proto(struct sock *sk)
962 int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4;
963 struct proto *prot = READ_ONCE(sk->sk_prot);
965 /* Build IPv6 TLS whenever the address of tcpv6 _prot changes */
966 if (ip_ver == TLSV6 &&
967 unlikely(prot != smp_load_acquire(&saved_tcpv6_prot))) {
968 mutex_lock(&tcpv6_prot_mutex);
969 if (likely(prot != saved_tcpv6_prot)) {
970 build_protos(tls_prots[TLSV6], prot);
971 build_proto_ops(tls_proto_ops[TLSV6],
973 smp_store_release(&saved_tcpv6_prot, prot);
975 mutex_unlock(&tcpv6_prot_mutex);
978 if (ip_ver == TLSV4 &&
979 unlikely(prot != smp_load_acquire(&saved_tcpv4_prot))) {
980 mutex_lock(&tcpv4_prot_mutex);
981 if (likely(prot != saved_tcpv4_prot)) {
982 build_protos(tls_prots[TLSV4], prot);
983 build_proto_ops(tls_proto_ops[TLSV4],
985 smp_store_release(&saved_tcpv4_prot, prot);
987 mutex_unlock(&tcpv4_prot_mutex);
991 static void build_protos(struct proto prot[TLS_NUM_CONFIG][TLS_NUM_CONFIG],
992 const struct proto *base)
994 prot[TLS_BASE][TLS_BASE] = *base;
995 prot[TLS_BASE][TLS_BASE].setsockopt = tls_setsockopt;
996 prot[TLS_BASE][TLS_BASE].getsockopt = tls_getsockopt;
997 prot[TLS_BASE][TLS_BASE].close = tls_sk_proto_close;
999 prot[TLS_SW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
1000 prot[TLS_SW][TLS_BASE].sendmsg = tls_sw_sendmsg;
1001 prot[TLS_SW][TLS_BASE].sendpage = tls_sw_sendpage;
1003 prot[TLS_BASE][TLS_SW] = prot[TLS_BASE][TLS_BASE];
1004 prot[TLS_BASE][TLS_SW].recvmsg = tls_sw_recvmsg;
1005 prot[TLS_BASE][TLS_SW].sock_is_readable = tls_sw_sock_is_readable;
1006 prot[TLS_BASE][TLS_SW].close = tls_sk_proto_close;
1008 prot[TLS_SW][TLS_SW] = prot[TLS_SW][TLS_BASE];
1009 prot[TLS_SW][TLS_SW].recvmsg = tls_sw_recvmsg;
1010 prot[TLS_SW][TLS_SW].sock_is_readable = tls_sw_sock_is_readable;
1011 prot[TLS_SW][TLS_SW].close = tls_sk_proto_close;
1013 #ifdef CONFIG_TLS_DEVICE
1014 prot[TLS_HW][TLS_BASE] = prot[TLS_BASE][TLS_BASE];
1015 prot[TLS_HW][TLS_BASE].sendmsg = tls_device_sendmsg;
1016 prot[TLS_HW][TLS_BASE].sendpage = tls_device_sendpage;
1018 prot[TLS_HW][TLS_SW] = prot[TLS_BASE][TLS_SW];
1019 prot[TLS_HW][TLS_SW].sendmsg = tls_device_sendmsg;
1020 prot[TLS_HW][TLS_SW].sendpage = tls_device_sendpage;
1022 prot[TLS_BASE][TLS_HW] = prot[TLS_BASE][TLS_SW];
1024 prot[TLS_SW][TLS_HW] = prot[TLS_SW][TLS_SW];
1026 prot[TLS_HW][TLS_HW] = prot[TLS_HW][TLS_SW];
1028 #ifdef CONFIG_TLS_TOE
1029 prot[TLS_HW_RECORD][TLS_HW_RECORD] = *base;
1030 prot[TLS_HW_RECORD][TLS_HW_RECORD].hash = tls_toe_hash;
1031 prot[TLS_HW_RECORD][TLS_HW_RECORD].unhash = tls_toe_unhash;
1035 static int tls_init(struct sock *sk)
1037 struct tls_context *ctx;
1040 tls_build_proto(sk);
1042 #ifdef CONFIG_TLS_TOE
1043 if (tls_toe_bypass(sk))
1047 /* The TLS ulp is currently supported only for TCP sockets
1048 * in ESTABLISHED state.
1049 * Supporting sockets in LISTEN state will require us
1050 * to modify the accept implementation to clone rather then
1051 * share the ulp context.
1053 if (sk->sk_state != TCP_ESTABLISHED)
1056 /* allocate tls context */
1057 write_lock_bh(&sk->sk_callback_lock);
1058 ctx = tls_ctx_create(sk);
1064 ctx->tx_conf = TLS_BASE;
1065 ctx->rx_conf = TLS_BASE;
1066 update_sk_prot(sk, ctx);
1068 write_unlock_bh(&sk->sk_callback_lock);
1072 static void tls_update(struct sock *sk, struct proto *p,
1073 void (*write_space)(struct sock *sk))
1075 struct tls_context *ctx;
1077 WARN_ON_ONCE(sk->sk_prot == p);
1079 ctx = tls_get_ctx(sk);
1081 ctx->sk_write_space = write_space;
1084 /* Pairs with lockless read in sk_clone_lock(). */
1085 WRITE_ONCE(sk->sk_prot, p);
1086 sk->sk_write_space = write_space;
1090 static u16 tls_user_config(struct tls_context *ctx, bool tx)
1092 u16 config = tx ? ctx->tx_conf : ctx->rx_conf;
1096 return TLS_CONF_BASE;
1102 return TLS_CONF_HW_RECORD;
1107 static int tls_get_info(const struct sock *sk, struct sk_buff *skb)
1109 u16 version, cipher_type;
1110 struct tls_context *ctx;
1111 struct nlattr *start;
1114 start = nla_nest_start_noflag(skb, INET_ULP_INFO_TLS);
1119 ctx = rcu_dereference(inet_csk(sk)->icsk_ulp_data);
1124 version = ctx->prot_info.version;
1126 err = nla_put_u16(skb, TLS_INFO_VERSION, version);
1130 cipher_type = ctx->prot_info.cipher_type;
1132 err = nla_put_u16(skb, TLS_INFO_CIPHER, cipher_type);
1136 err = nla_put_u16(skb, TLS_INFO_TXCONF, tls_user_config(ctx, true));
1140 err = nla_put_u16(skb, TLS_INFO_RXCONF, tls_user_config(ctx, false));
1144 if (ctx->tx_conf == TLS_HW && ctx->zerocopy_sendfile) {
1145 err = nla_put_flag(skb, TLS_INFO_ZC_RO_TX);
1149 if (ctx->rx_no_pad) {
1150 err = nla_put_flag(skb, TLS_INFO_RX_NO_PAD);
1156 nla_nest_end(skb, start);
1161 nla_nest_cancel(skb, start);
1165 static size_t tls_get_info_size(const struct sock *sk)
1169 size += nla_total_size(0) + /* INET_ULP_INFO_TLS */
1170 nla_total_size(sizeof(u16)) + /* TLS_INFO_VERSION */
1171 nla_total_size(sizeof(u16)) + /* TLS_INFO_CIPHER */
1172 nla_total_size(sizeof(u16)) + /* TLS_INFO_RXCONF */
1173 nla_total_size(sizeof(u16)) + /* TLS_INFO_TXCONF */
1174 nla_total_size(0) + /* TLS_INFO_ZC_RO_TX */
1175 nla_total_size(0) + /* TLS_INFO_RX_NO_PAD */
1181 static int __net_init tls_init_net(struct net *net)
1185 net->mib.tls_statistics = alloc_percpu(struct linux_tls_mib);
1186 if (!net->mib.tls_statistics)
1189 err = tls_proc_init(net);
1191 goto err_free_stats;
1195 free_percpu(net->mib.tls_statistics);
1199 static void __net_exit tls_exit_net(struct net *net)
1202 free_percpu(net->mib.tls_statistics);
1205 static struct pernet_operations tls_proc_ops = {
1206 .init = tls_init_net,
1207 .exit = tls_exit_net,
1210 static struct tcp_ulp_ops tcp_tls_ulp_ops __read_mostly = {
1212 .owner = THIS_MODULE,
1214 .update = tls_update,
1215 .get_info = tls_get_info,
1216 .get_info_size = tls_get_info_size,
1219 static int __init tls_register(void)
1223 err = register_pernet_subsys(&tls_proc_ops);
1227 err = tls_strp_dev_init();
1231 err = tls_device_init();
1235 tcp_register_ulp(&tcp_tls_ulp_ops);
1239 tls_strp_dev_exit();
1241 unregister_pernet_subsys(&tls_proc_ops);
1245 static void __exit tls_unregister(void)
1247 tcp_unregister_ulp(&tcp_tls_ulp_ops);
1248 tls_strp_dev_exit();
1249 tls_device_cleanup();
1250 unregister_pernet_subsys(&tls_proc_ops);
1253 module_init(tls_register);
1254 module_exit(tls_unregister);