1 /* 2 * Copyright (c) 2016-2017, Mellanox Technologies. All rights reserved. 3 * Copyright (c) 2016-2017, Dave Watson <davejwatson@fb.com>. All rights reserved. 4 * 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: 10 * 11 * Redistribution and use in source and binary forms, with or 12 * without modification, are permitted provided that the following 13 * conditions are met: 14 * 15 * - Redistributions of source code must retain the above 16 * copyright notice, this list of conditions and the following 17 * disclaimer. 18 * 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. 23 * 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 31 * SOFTWARE. 32 */ 33 34 #include <linux/module.h> 35 36 #include <net/tcp.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> 43 44 #include <net/snmp.h> 45 #include <net/tls.h> 46 #include <net/tls_toe.h> 47 48 #include "tls.h" 49 50 MODULE_AUTHOR("Mellanox Technologies"); 51 MODULE_DESCRIPTION("Transport Layer Security Support"); 52 MODULE_LICENSE("Dual BSD/GPL"); 53 MODULE_ALIAS_TCP_ULP("tls"); 54 55 enum { 56 TLSV4, 57 TLSV6, 58 TLS_NUM_PROTS, 59 }; 60 61 #define CHECK_CIPHER_DESC(cipher,ci) \ 62 static_assert(cipher ## _IV_SIZE <= TLS_MAX_IV_SIZE); \ 63 static_assert(cipher ## _SALT_SIZE <= TLS_MAX_SALT_SIZE); \ 64 static_assert(cipher ## _REC_SEQ_SIZE <= TLS_MAX_REC_SEQ_SIZE); \ 65 static_assert(cipher ## _TAG_SIZE == TLS_TAG_SIZE); \ 66 static_assert(sizeof_field(struct ci, iv) == cipher ## _IV_SIZE); \ 67 static_assert(sizeof_field(struct ci, key) == cipher ## _KEY_SIZE); \ 68 static_assert(sizeof_field(struct ci, salt) == cipher ## _SALT_SIZE); \ 69 static_assert(sizeof_field(struct ci, rec_seq) == cipher ## _REC_SEQ_SIZE); 70 71 #define __CIPHER_DESC(ci) \ 72 .iv_offset = offsetof(struct ci, iv), \ 73 .key_offset = offsetof(struct ci, key), \ 74 .salt_offset = offsetof(struct ci, salt), \ 75 .rec_seq_offset = offsetof(struct ci, rec_seq), \ 76 .crypto_info = sizeof(struct ci) 77 78 #define CIPHER_DESC(cipher,ci,algname,_offloadable) [cipher - TLS_CIPHER_MIN] = { \ 79 .nonce = cipher ## _IV_SIZE, \ 80 .iv = cipher ## _IV_SIZE, \ 81 .key = cipher ## _KEY_SIZE, \ 82 .salt = cipher ## _SALT_SIZE, \ 83 .tag = cipher ## _TAG_SIZE, \ 84 .rec_seq = cipher ## _REC_SEQ_SIZE, \ 85 .cipher_name = algname, \ 86 .offloadable = _offloadable, \ 87 __CIPHER_DESC(ci), \ 88 } 89 90 #define CIPHER_DESC_NONCE0(cipher,ci,algname,_offloadable) [cipher - TLS_CIPHER_MIN] = { \ 91 .nonce = 0, \ 92 .iv = cipher ## _IV_SIZE, \ 93 .key = cipher ## _KEY_SIZE, \ 94 .salt = cipher ## _SALT_SIZE, \ 95 .tag = cipher ## _TAG_SIZE, \ 96 .rec_seq = cipher ## _REC_SEQ_SIZE, \ 97 .cipher_name = algname, \ 98 .offloadable = _offloadable, \ 99 __CIPHER_DESC(ci), \ 100 } 101 102 const struct tls_cipher_desc tls_cipher_desc[TLS_CIPHER_MAX + 1 - TLS_CIPHER_MIN] = { 103 CIPHER_DESC(TLS_CIPHER_AES_GCM_128, tls12_crypto_info_aes_gcm_128, "gcm(aes)", true), 104 CIPHER_DESC(TLS_CIPHER_AES_GCM_256, tls12_crypto_info_aes_gcm_256, "gcm(aes)", true), 105 CIPHER_DESC(TLS_CIPHER_AES_CCM_128, tls12_crypto_info_aes_ccm_128, "ccm(aes)", false), 106 CIPHER_DESC_NONCE0(TLS_CIPHER_CHACHA20_POLY1305, tls12_crypto_info_chacha20_poly1305, "rfc7539(chacha20,poly1305)", false), 107 CIPHER_DESC(TLS_CIPHER_SM4_GCM, tls12_crypto_info_sm4_gcm, "gcm(sm4)", false), 108 CIPHER_DESC(TLS_CIPHER_SM4_CCM, tls12_crypto_info_sm4_ccm, "ccm(sm4)", false), 109 CIPHER_DESC(TLS_CIPHER_ARIA_GCM_128, tls12_crypto_info_aria_gcm_128, "gcm(aria)", false), 110 CIPHER_DESC(TLS_CIPHER_ARIA_GCM_256, tls12_crypto_info_aria_gcm_256, "gcm(aria)", false), 111 }; 112 113 CHECK_CIPHER_DESC(TLS_CIPHER_AES_GCM_128, tls12_crypto_info_aes_gcm_128); 114 CHECK_CIPHER_DESC(TLS_CIPHER_AES_GCM_256, tls12_crypto_info_aes_gcm_256); 115 CHECK_CIPHER_DESC(TLS_CIPHER_AES_CCM_128, tls12_crypto_info_aes_ccm_128); 116 CHECK_CIPHER_DESC(TLS_CIPHER_CHACHA20_POLY1305, tls12_crypto_info_chacha20_poly1305); 117 CHECK_CIPHER_DESC(TLS_CIPHER_SM4_GCM, tls12_crypto_info_sm4_gcm); 118 CHECK_CIPHER_DESC(TLS_CIPHER_SM4_CCM, tls12_crypto_info_sm4_ccm); 119 CHECK_CIPHER_DESC(TLS_CIPHER_ARIA_GCM_128, tls12_crypto_info_aria_gcm_128); 120 CHECK_CIPHER_DESC(TLS_CIPHER_ARIA_GCM_256, tls12_crypto_info_aria_gcm_256); 121 122 static const struct proto *saved_tcpv6_prot; 123 static DEFINE_MUTEX(tcpv6_prot_mutex); 124 static const struct proto *saved_tcpv4_prot; 125 static DEFINE_MUTEX(tcpv4_prot_mutex); 126 static struct proto tls_prots[TLS_NUM_PROTS][TLS_NUM_CONFIG][TLS_NUM_CONFIG]; 127 static struct proto_ops tls_proto_ops[TLS_NUM_PROTS][TLS_NUM_CONFIG][TLS_NUM_CONFIG]; 128 static void build_protos(struct proto prot[TLS_NUM_CONFIG][TLS_NUM_CONFIG], 129 const struct proto *base); 130 131 void update_sk_prot(struct sock *sk, struct tls_context *ctx) 132 { 133 int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4; 134 135 WRITE_ONCE(sk->sk_prot, 136 &tls_prots[ip_ver][ctx->tx_conf][ctx->rx_conf]); 137 WRITE_ONCE(sk->sk_socket->ops, 138 &tls_proto_ops[ip_ver][ctx->tx_conf][ctx->rx_conf]); 139 } 140 141 int wait_on_pending_writer(struct sock *sk, long *timeo) 142 { 143 DEFINE_WAIT_FUNC(wait, woken_wake_function); 144 int ret, rc = 0; 145 146 add_wait_queue(sk_sleep(sk), &wait); 147 while (1) { 148 if (!*timeo) { 149 rc = -EAGAIN; 150 break; 151 } 152 153 if (signal_pending(current)) { 154 rc = sock_intr_errno(*timeo); 155 break; 156 } 157 158 ret = sk_wait_event(sk, timeo, 159 !READ_ONCE(sk->sk_write_pending), &wait); 160 if (ret) { 161 if (ret < 0) 162 rc = ret; 163 break; 164 } 165 } 166 remove_wait_queue(sk_sleep(sk), &wait); 167 return rc; 168 } 169 170 int tls_push_sg(struct sock *sk, 171 struct tls_context *ctx, 172 struct scatterlist *sg, 173 u16 first_offset, 174 int flags) 175 { 176 struct bio_vec bvec; 177 struct msghdr msg = { 178 .msg_flags = MSG_SPLICE_PAGES | flags, 179 }; 180 int ret = 0; 181 struct page *p; 182 size_t size; 183 int offset = first_offset; 184 185 size = sg->length - offset; 186 offset += sg->offset; 187 188 ctx->splicing_pages = true; 189 while (1) { 190 /* is sending application-limited? */ 191 tcp_rate_check_app_limited(sk); 192 p = sg_page(sg); 193 retry: 194 bvec_set_page(&bvec, p, size, offset); 195 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1, size); 196 197 ret = tcp_sendmsg_locked(sk, &msg, size); 198 199 if (ret != size) { 200 if (ret > 0) { 201 offset += ret; 202 size -= ret; 203 goto retry; 204 } 205 206 offset -= sg->offset; 207 ctx->partially_sent_offset = offset; 208 ctx->partially_sent_record = (void *)sg; 209 ctx->splicing_pages = false; 210 return ret; 211 } 212 213 put_page(p); 214 sk_mem_uncharge(sk, sg->length); 215 sg = sg_next(sg); 216 if (!sg) 217 break; 218 219 offset = sg->offset; 220 size = sg->length; 221 } 222 223 ctx->splicing_pages = false; 224 225 return 0; 226 } 227 228 static int tls_handle_open_record(struct sock *sk, int flags) 229 { 230 struct tls_context *ctx = tls_get_ctx(sk); 231 232 if (tls_is_pending_open_record(ctx)) 233 return ctx->push_pending_record(sk, flags); 234 235 return 0; 236 } 237 238 int tls_process_cmsg(struct sock *sk, struct msghdr *msg, 239 unsigned char *record_type) 240 { 241 struct cmsghdr *cmsg; 242 int rc = -EINVAL; 243 244 for_each_cmsghdr(cmsg, msg) { 245 if (!CMSG_OK(msg, cmsg)) 246 return -EINVAL; 247 if (cmsg->cmsg_level != SOL_TLS) 248 continue; 249 250 switch (cmsg->cmsg_type) { 251 case TLS_SET_RECORD_TYPE: 252 if (cmsg->cmsg_len < CMSG_LEN(sizeof(*record_type))) 253 return -EINVAL; 254 255 if (msg->msg_flags & MSG_MORE) 256 return -EINVAL; 257 258 *record_type = *(unsigned char *)CMSG_DATA(cmsg); 259 260 rc = tls_handle_open_record(sk, msg->msg_flags); 261 break; 262 default: 263 return -EINVAL; 264 } 265 } 266 267 return rc; 268 } 269 270 int tls_push_partial_record(struct sock *sk, struct tls_context *ctx, 271 int flags) 272 { 273 struct scatterlist *sg; 274 u16 offset; 275 276 sg = ctx->partially_sent_record; 277 offset = ctx->partially_sent_offset; 278 279 ctx->partially_sent_record = NULL; 280 return tls_push_sg(sk, ctx, sg, offset, flags); 281 } 282 283 void tls_free_partial_record(struct sock *sk, struct tls_context *ctx) 284 { 285 struct scatterlist *sg; 286 287 for (sg = ctx->partially_sent_record; sg; sg = sg_next(sg)) { 288 put_page(sg_page(sg)); 289 sk_mem_uncharge(sk, sg->length); 290 } 291 ctx->partially_sent_record = NULL; 292 } 293 294 static void tls_write_space(struct sock *sk) 295 { 296 struct tls_context *ctx = tls_get_ctx(sk); 297 298 /* If splicing_pages call lower protocol write space handler 299 * to ensure we wake up any waiting operations there. For example 300 * if splicing pages where to call sk_wait_event. 301 */ 302 if (ctx->splicing_pages) { 303 ctx->sk_write_space(sk); 304 return; 305 } 306 307 #ifdef CONFIG_TLS_DEVICE 308 if (ctx->tx_conf == TLS_HW) 309 tls_device_write_space(sk, ctx); 310 else 311 #endif 312 tls_sw_write_space(sk, ctx); 313 314 ctx->sk_write_space(sk); 315 } 316 317 /** 318 * tls_ctx_free() - free TLS ULP context 319 * @sk: socket to with @ctx is attached 320 * @ctx: TLS context structure 321 * 322 * Free TLS context. If @sk is %NULL caller guarantees that the socket 323 * to which @ctx was attached has no outstanding references. 324 */ 325 void tls_ctx_free(struct sock *sk, struct tls_context *ctx) 326 { 327 if (!ctx) 328 return; 329 330 memzero_explicit(&ctx->crypto_send, sizeof(ctx->crypto_send)); 331 memzero_explicit(&ctx->crypto_recv, sizeof(ctx->crypto_recv)); 332 mutex_destroy(&ctx->tx_lock); 333 334 if (sk) 335 kfree_rcu(ctx, rcu); 336 else 337 kfree(ctx); 338 } 339 340 static void tls_sk_proto_cleanup(struct sock *sk, 341 struct tls_context *ctx, long timeo) 342 { 343 if (unlikely(sk->sk_write_pending) && 344 !wait_on_pending_writer(sk, &timeo)) 345 tls_handle_open_record(sk, 0); 346 347 /* We need these for tls_sw_fallback handling of other packets */ 348 if (ctx->tx_conf == TLS_SW) { 349 tls_sw_release_resources_tx(sk); 350 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXSW); 351 } else if (ctx->tx_conf == TLS_HW) { 352 tls_device_free_resources_tx(sk); 353 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXDEVICE); 354 } 355 356 if (ctx->rx_conf == TLS_SW) { 357 tls_sw_release_resources_rx(sk); 358 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXSW); 359 } else if (ctx->rx_conf == TLS_HW) { 360 tls_device_offload_cleanup_rx(sk); 361 TLS_DEC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXDEVICE); 362 } 363 } 364 365 static void tls_sk_proto_close(struct sock *sk, long timeout) 366 { 367 struct inet_connection_sock *icsk = inet_csk(sk); 368 struct tls_context *ctx = tls_get_ctx(sk); 369 long timeo = sock_sndtimeo(sk, 0); 370 bool free_ctx; 371 372 if (ctx->tx_conf == TLS_SW) 373 tls_sw_cancel_work_tx(ctx); 374 375 lock_sock(sk); 376 free_ctx = ctx->tx_conf != TLS_HW && ctx->rx_conf != TLS_HW; 377 378 if (ctx->tx_conf != TLS_BASE || ctx->rx_conf != TLS_BASE) 379 tls_sk_proto_cleanup(sk, ctx, timeo); 380 381 write_lock_bh(&sk->sk_callback_lock); 382 if (free_ctx) 383 rcu_assign_pointer(icsk->icsk_ulp_data, NULL); 384 WRITE_ONCE(sk->sk_prot, ctx->sk_proto); 385 if (sk->sk_write_space == tls_write_space) 386 sk->sk_write_space = ctx->sk_write_space; 387 write_unlock_bh(&sk->sk_callback_lock); 388 release_sock(sk); 389 if (ctx->tx_conf == TLS_SW) 390 tls_sw_free_ctx_tx(ctx); 391 if (ctx->rx_conf == TLS_SW || ctx->rx_conf == TLS_HW) 392 tls_sw_strparser_done(ctx); 393 if (ctx->rx_conf == TLS_SW) 394 tls_sw_free_ctx_rx(ctx); 395 ctx->sk_proto->close(sk, timeout); 396 397 if (free_ctx) 398 tls_ctx_free(sk, ctx); 399 } 400 401 static __poll_t tls_sk_poll(struct file *file, struct socket *sock, 402 struct poll_table_struct *wait) 403 { 404 struct tls_sw_context_rx *ctx; 405 struct tls_context *tls_ctx; 406 struct sock *sk = sock->sk; 407 struct sk_psock *psock; 408 __poll_t mask = 0; 409 u8 shutdown; 410 int state; 411 412 mask = tcp_poll(file, sock, wait); 413 414 state = inet_sk_state_load(sk); 415 shutdown = READ_ONCE(sk->sk_shutdown); 416 if (unlikely(state != TCP_ESTABLISHED || shutdown & RCV_SHUTDOWN)) 417 return mask; 418 419 tls_ctx = tls_get_ctx(sk); 420 ctx = tls_sw_ctx_rx(tls_ctx); 421 psock = sk_psock_get(sk); 422 423 if ((skb_queue_empty_lockless(&ctx->rx_list) && 424 !tls_strp_msg_ready(ctx) && 425 sk_psock_queue_empty(psock)) || 426 READ_ONCE(ctx->key_update_pending)) 427 mask &= ~(EPOLLIN | EPOLLRDNORM); 428 429 if (psock) 430 sk_psock_put(sk, psock); 431 432 return mask; 433 } 434 435 static int do_tls_getsockopt_conf(struct sock *sk, char __user *optval, 436 int __user *optlen, int tx) 437 { 438 int rc = 0; 439 const struct tls_cipher_desc *cipher_desc; 440 struct tls_context *ctx = tls_get_ctx(sk); 441 struct tls_crypto_info *crypto_info; 442 struct cipher_context *cctx; 443 int len; 444 445 if (get_user(len, optlen)) 446 return -EFAULT; 447 448 if (!optval || (len < sizeof(*crypto_info))) { 449 rc = -EINVAL; 450 goto out; 451 } 452 453 if (!ctx) { 454 rc = -EBUSY; 455 goto out; 456 } 457 458 /* get user crypto info */ 459 if (tx) { 460 crypto_info = &ctx->crypto_send.info; 461 cctx = &ctx->tx; 462 } else { 463 crypto_info = &ctx->crypto_recv.info; 464 cctx = &ctx->rx; 465 } 466 467 if (!TLS_CRYPTO_INFO_READY(crypto_info)) { 468 rc = -EBUSY; 469 goto out; 470 } 471 472 if (len == sizeof(*crypto_info)) { 473 if (copy_to_user(optval, crypto_info, sizeof(*crypto_info))) 474 rc = -EFAULT; 475 goto out; 476 } 477 478 cipher_desc = get_cipher_desc(crypto_info->cipher_type); 479 if (!cipher_desc || len != cipher_desc->crypto_info) { 480 rc = -EINVAL; 481 goto out; 482 } 483 484 memcpy(crypto_info_iv(crypto_info, cipher_desc), 485 cctx->iv + cipher_desc->salt, cipher_desc->iv); 486 memcpy(crypto_info_rec_seq(crypto_info, cipher_desc), 487 cctx->rec_seq, cipher_desc->rec_seq); 488 489 if (copy_to_user(optval, crypto_info, cipher_desc->crypto_info)) 490 rc = -EFAULT; 491 492 out: 493 return rc; 494 } 495 496 static int do_tls_getsockopt_tx_zc(struct sock *sk, char __user *optval, 497 int __user *optlen) 498 { 499 struct tls_context *ctx = tls_get_ctx(sk); 500 unsigned int value; 501 int len; 502 503 if (get_user(len, optlen)) 504 return -EFAULT; 505 506 if (len != sizeof(value)) 507 return -EINVAL; 508 509 value = ctx->zerocopy_sendfile; 510 if (copy_to_user(optval, &value, sizeof(value))) 511 return -EFAULT; 512 513 return 0; 514 } 515 516 static int do_tls_getsockopt_no_pad(struct sock *sk, char __user *optval, 517 int __user *optlen) 518 { 519 struct tls_context *ctx = tls_get_ctx(sk); 520 int value, len; 521 522 if (ctx->prot_info.version != TLS_1_3_VERSION) 523 return -EINVAL; 524 525 if (get_user(len, optlen)) 526 return -EFAULT; 527 if (len < sizeof(value)) 528 return -EINVAL; 529 530 value = -EINVAL; 531 if (ctx->rx_conf == TLS_SW || ctx->rx_conf == TLS_HW) 532 value = ctx->rx_no_pad; 533 if (value < 0) 534 return value; 535 536 if (put_user(sizeof(value), optlen)) 537 return -EFAULT; 538 if (copy_to_user(optval, &value, sizeof(value))) 539 return -EFAULT; 540 541 return 0; 542 } 543 544 static int do_tls_getsockopt(struct sock *sk, int optname, 545 char __user *optval, int __user *optlen) 546 { 547 int rc = 0; 548 549 lock_sock(sk); 550 551 switch (optname) { 552 case TLS_TX: 553 case TLS_RX: 554 rc = do_tls_getsockopt_conf(sk, optval, optlen, 555 optname == TLS_TX); 556 break; 557 case TLS_TX_ZEROCOPY_RO: 558 rc = do_tls_getsockopt_tx_zc(sk, optval, optlen); 559 break; 560 case TLS_RX_EXPECT_NO_PAD: 561 rc = do_tls_getsockopt_no_pad(sk, optval, optlen); 562 break; 563 default: 564 rc = -ENOPROTOOPT; 565 break; 566 } 567 568 release_sock(sk); 569 570 return rc; 571 } 572 573 static int tls_getsockopt(struct sock *sk, int level, int optname, 574 char __user *optval, int __user *optlen) 575 { 576 struct tls_context *ctx = tls_get_ctx(sk); 577 578 if (level != SOL_TLS) 579 return ctx->sk_proto->getsockopt(sk, level, 580 optname, optval, optlen); 581 582 return do_tls_getsockopt(sk, optname, optval, optlen); 583 } 584 585 static int validate_crypto_info(const struct tls_crypto_info *crypto_info, 586 const struct tls_crypto_info *alt_crypto_info) 587 { 588 if (crypto_info->version != TLS_1_2_VERSION && 589 crypto_info->version != TLS_1_3_VERSION) 590 return -EINVAL; 591 592 switch (crypto_info->cipher_type) { 593 case TLS_CIPHER_ARIA_GCM_128: 594 case TLS_CIPHER_ARIA_GCM_256: 595 if (crypto_info->version != TLS_1_2_VERSION) 596 return -EINVAL; 597 break; 598 } 599 600 /* Ensure that TLS version and ciphers are same in both directions */ 601 if (TLS_CRYPTO_INFO_READY(alt_crypto_info)) { 602 if (alt_crypto_info->version != crypto_info->version || 603 alt_crypto_info->cipher_type != crypto_info->cipher_type) 604 return -EINVAL; 605 } 606 607 return 0; 608 } 609 610 static int do_tls_setsockopt_conf(struct sock *sk, sockptr_t optval, 611 unsigned int optlen, int tx) 612 { 613 struct tls_crypto_info *crypto_info, *alt_crypto_info; 614 struct tls_crypto_info *old_crypto_info = NULL; 615 struct tls_context *ctx = tls_get_ctx(sk); 616 const struct tls_cipher_desc *cipher_desc; 617 union tls_crypto_context *crypto_ctx; 618 union tls_crypto_context tmp = {}; 619 bool update = false; 620 int rc = 0; 621 int conf; 622 623 if (sockptr_is_null(optval) || (optlen < sizeof(*crypto_info))) 624 return -EINVAL; 625 626 if (tx) { 627 crypto_ctx = &ctx->crypto_send; 628 alt_crypto_info = &ctx->crypto_recv.info; 629 } else { 630 crypto_ctx = &ctx->crypto_recv; 631 alt_crypto_info = &ctx->crypto_send.info; 632 } 633 634 crypto_info = &crypto_ctx->info; 635 636 if (TLS_CRYPTO_INFO_READY(crypto_info)) { 637 /* Currently we only support setting crypto info more 638 * than one time for TLS 1.3 639 */ 640 if (crypto_info->version != TLS_1_3_VERSION) { 641 TLS_INC_STATS(sock_net(sk), tx ? LINUX_MIB_TLSTXREKEYERROR 642 : LINUX_MIB_TLSRXREKEYERROR); 643 return -EBUSY; 644 } 645 646 update = true; 647 old_crypto_info = crypto_info; 648 crypto_info = &tmp.info; 649 crypto_ctx = &tmp; 650 } 651 652 rc = copy_from_sockptr(crypto_info, optval, sizeof(*crypto_info)); 653 if (rc) { 654 rc = -EFAULT; 655 goto err_crypto_info; 656 } 657 658 if (update) { 659 /* Ensure that TLS version and ciphers are not modified */ 660 if (crypto_info->version != old_crypto_info->version || 661 crypto_info->cipher_type != old_crypto_info->cipher_type) 662 rc = -EINVAL; 663 } else { 664 rc = validate_crypto_info(crypto_info, alt_crypto_info); 665 } 666 if (rc) 667 goto err_crypto_info; 668 669 cipher_desc = get_cipher_desc(crypto_info->cipher_type); 670 if (!cipher_desc) { 671 rc = -EINVAL; 672 goto err_crypto_info; 673 } 674 675 if (optlen != cipher_desc->crypto_info) { 676 rc = -EINVAL; 677 goto err_crypto_info; 678 } 679 680 rc = copy_from_sockptr_offset(crypto_info + 1, optval, 681 sizeof(*crypto_info), 682 optlen - sizeof(*crypto_info)); 683 if (rc) { 684 rc = -EFAULT; 685 goto err_crypto_info; 686 } 687 688 if (tx) { 689 rc = tls_set_device_offload(sk); 690 conf = TLS_HW; 691 if (!rc) { 692 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSTXDEVICE); 693 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXDEVICE); 694 } else { 695 rc = tls_set_sw_offload(sk, 1, 696 update ? crypto_info : NULL); 697 if (rc) 698 goto err_crypto_info; 699 700 if (update) { 701 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSTXREKEYOK); 702 } else { 703 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSTXSW); 704 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRTXSW); 705 } 706 conf = TLS_SW; 707 } 708 } else { 709 rc = tls_set_device_offload_rx(sk, ctx); 710 conf = TLS_HW; 711 if (!rc) { 712 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXDEVICE); 713 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXDEVICE); 714 } else { 715 rc = tls_set_sw_offload(sk, 0, 716 update ? crypto_info : NULL); 717 if (rc) 718 goto err_crypto_info; 719 720 if (update) { 721 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXREKEYOK); 722 } else { 723 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXSW); 724 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSCURRRXSW); 725 } 726 conf = TLS_SW; 727 } 728 if (!update) 729 tls_sw_strparser_arm(sk, ctx); 730 } 731 732 if (tx) 733 ctx->tx_conf = conf; 734 else 735 ctx->rx_conf = conf; 736 update_sk_prot(sk, ctx); 737 738 if (update) 739 return 0; 740 741 if (tx) { 742 ctx->sk_write_space = sk->sk_write_space; 743 sk->sk_write_space = tls_write_space; 744 } else { 745 struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(ctx); 746 747 tls_strp_check_rcv(&rx_ctx->strp); 748 } 749 return 0; 750 751 err_crypto_info: 752 if (update) { 753 TLS_INC_STATS(sock_net(sk), tx ? LINUX_MIB_TLSTXREKEYERROR 754 : LINUX_MIB_TLSRXREKEYERROR); 755 } 756 memzero_explicit(crypto_ctx, sizeof(*crypto_ctx)); 757 return rc; 758 } 759 760 static int do_tls_setsockopt_tx_zc(struct sock *sk, sockptr_t optval, 761 unsigned int optlen) 762 { 763 struct tls_context *ctx = tls_get_ctx(sk); 764 unsigned int value; 765 766 if (sockptr_is_null(optval) || optlen != sizeof(value)) 767 return -EINVAL; 768 769 if (copy_from_sockptr(&value, optval, sizeof(value))) 770 return -EFAULT; 771 772 if (value > 1) 773 return -EINVAL; 774 775 ctx->zerocopy_sendfile = value; 776 777 return 0; 778 } 779 780 static int do_tls_setsockopt_no_pad(struct sock *sk, sockptr_t optval, 781 unsigned int optlen) 782 { 783 struct tls_context *ctx = tls_get_ctx(sk); 784 u32 val; 785 int rc; 786 787 if (ctx->prot_info.version != TLS_1_3_VERSION || 788 sockptr_is_null(optval) || optlen < sizeof(val)) 789 return -EINVAL; 790 791 rc = copy_from_sockptr(&val, optval, sizeof(val)); 792 if (rc) 793 return -EFAULT; 794 if (val > 1) 795 return -EINVAL; 796 rc = check_zeroed_sockptr(optval, sizeof(val), optlen - sizeof(val)); 797 if (rc < 1) 798 return rc == 0 ? -EINVAL : rc; 799 800 lock_sock(sk); 801 rc = -EINVAL; 802 if (ctx->rx_conf == TLS_SW || ctx->rx_conf == TLS_HW) { 803 ctx->rx_no_pad = val; 804 tls_update_rx_zc_capable(ctx); 805 rc = 0; 806 } 807 release_sock(sk); 808 809 return rc; 810 } 811 812 static int do_tls_setsockopt(struct sock *sk, int optname, sockptr_t optval, 813 unsigned int optlen) 814 { 815 int rc = 0; 816 817 switch (optname) { 818 case TLS_TX: 819 case TLS_RX: 820 lock_sock(sk); 821 rc = do_tls_setsockopt_conf(sk, optval, optlen, 822 optname == TLS_TX); 823 release_sock(sk); 824 break; 825 case TLS_TX_ZEROCOPY_RO: 826 lock_sock(sk); 827 rc = do_tls_setsockopt_tx_zc(sk, optval, optlen); 828 release_sock(sk); 829 break; 830 case TLS_RX_EXPECT_NO_PAD: 831 rc = do_tls_setsockopt_no_pad(sk, optval, optlen); 832 break; 833 default: 834 rc = -ENOPROTOOPT; 835 break; 836 } 837 return rc; 838 } 839 840 static int tls_setsockopt(struct sock *sk, int level, int optname, 841 sockptr_t optval, unsigned int optlen) 842 { 843 struct tls_context *ctx = tls_get_ctx(sk); 844 845 if (level != SOL_TLS) 846 return ctx->sk_proto->setsockopt(sk, level, optname, optval, 847 optlen); 848 849 return do_tls_setsockopt(sk, optname, optval, optlen); 850 } 851 852 static int tls_disconnect(struct sock *sk, int flags) 853 { 854 return -EOPNOTSUPP; 855 } 856 857 struct tls_context *tls_ctx_create(struct sock *sk) 858 { 859 struct inet_connection_sock *icsk = inet_csk(sk); 860 struct tls_context *ctx; 861 862 ctx = kzalloc(sizeof(*ctx), GFP_ATOMIC); 863 if (!ctx) 864 return NULL; 865 866 mutex_init(&ctx->tx_lock); 867 ctx->sk_proto = READ_ONCE(sk->sk_prot); 868 ctx->sk = sk; 869 /* Release semantic of rcu_assign_pointer() ensures that 870 * ctx->sk_proto is visible before changing sk->sk_prot in 871 * update_sk_prot(), and prevents reading uninitialized value in 872 * tls_{getsockopt, setsockopt}. Note that we do not need a 873 * read barrier in tls_{getsockopt,setsockopt} as there is an 874 * address dependency between sk->sk_proto->{getsockopt,setsockopt} 875 * and ctx->sk_proto. 876 */ 877 rcu_assign_pointer(icsk->icsk_ulp_data, ctx); 878 return ctx; 879 } 880 881 static void build_proto_ops(struct proto_ops ops[TLS_NUM_CONFIG][TLS_NUM_CONFIG], 882 const struct proto_ops *base) 883 { 884 ops[TLS_BASE][TLS_BASE] = *base; 885 886 ops[TLS_SW ][TLS_BASE] = ops[TLS_BASE][TLS_BASE]; 887 ops[TLS_SW ][TLS_BASE].splice_eof = tls_sw_splice_eof; 888 889 ops[TLS_BASE][TLS_SW ] = ops[TLS_BASE][TLS_BASE]; 890 ops[TLS_BASE][TLS_SW ].splice_read = tls_sw_splice_read; 891 ops[TLS_BASE][TLS_SW ].poll = tls_sk_poll; 892 ops[TLS_BASE][TLS_SW ].read_sock = tls_sw_read_sock; 893 894 ops[TLS_SW ][TLS_SW ] = ops[TLS_SW ][TLS_BASE]; 895 ops[TLS_SW ][TLS_SW ].splice_read = tls_sw_splice_read; 896 ops[TLS_SW ][TLS_SW ].poll = tls_sk_poll; 897 ops[TLS_SW ][TLS_SW ].read_sock = tls_sw_read_sock; 898 899 #ifdef CONFIG_TLS_DEVICE 900 ops[TLS_HW ][TLS_BASE] = ops[TLS_BASE][TLS_BASE]; 901 902 ops[TLS_HW ][TLS_SW ] = ops[TLS_BASE][TLS_SW ]; 903 904 ops[TLS_BASE][TLS_HW ] = ops[TLS_BASE][TLS_SW ]; 905 906 ops[TLS_SW ][TLS_HW ] = ops[TLS_SW ][TLS_SW ]; 907 908 ops[TLS_HW ][TLS_HW ] = ops[TLS_HW ][TLS_SW ]; 909 #endif 910 #ifdef CONFIG_TLS_TOE 911 ops[TLS_HW_RECORD][TLS_HW_RECORD] = *base; 912 #endif 913 } 914 915 static void tls_build_proto(struct sock *sk) 916 { 917 int ip_ver = sk->sk_family == AF_INET6 ? TLSV6 : TLSV4; 918 struct proto *prot = READ_ONCE(sk->sk_prot); 919 920 /* Build IPv6 TLS whenever the address of tcpv6 _prot changes */ 921 if (ip_ver == TLSV6 && 922 unlikely(prot != smp_load_acquire(&saved_tcpv6_prot))) { 923 mutex_lock(&tcpv6_prot_mutex); 924 if (likely(prot != saved_tcpv6_prot)) { 925 build_protos(tls_prots[TLSV6], prot); 926 build_proto_ops(tls_proto_ops[TLSV6], 927 sk->sk_socket->ops); 928 smp_store_release(&saved_tcpv6_prot, prot); 929 } 930 mutex_unlock(&tcpv6_prot_mutex); 931 } 932 933 if (ip_ver == TLSV4 && 934 unlikely(prot != smp_load_acquire(&saved_tcpv4_prot))) { 935 mutex_lock(&tcpv4_prot_mutex); 936 if (likely(prot != saved_tcpv4_prot)) { 937 build_protos(tls_prots[TLSV4], prot); 938 build_proto_ops(tls_proto_ops[TLSV4], 939 sk->sk_socket->ops); 940 smp_store_release(&saved_tcpv4_prot, prot); 941 } 942 mutex_unlock(&tcpv4_prot_mutex); 943 } 944 } 945 946 static void build_protos(struct proto prot[TLS_NUM_CONFIG][TLS_NUM_CONFIG], 947 const struct proto *base) 948 { 949 prot[TLS_BASE][TLS_BASE] = *base; 950 prot[TLS_BASE][TLS_BASE].setsockopt = tls_setsockopt; 951 prot[TLS_BASE][TLS_BASE].getsockopt = tls_getsockopt; 952 prot[TLS_BASE][TLS_BASE].disconnect = tls_disconnect; 953 prot[TLS_BASE][TLS_BASE].close = tls_sk_proto_close; 954 955 prot[TLS_SW][TLS_BASE] = prot[TLS_BASE][TLS_BASE]; 956 prot[TLS_SW][TLS_BASE].sendmsg = tls_sw_sendmsg; 957 prot[TLS_SW][TLS_BASE].splice_eof = tls_sw_splice_eof; 958 959 prot[TLS_BASE][TLS_SW] = prot[TLS_BASE][TLS_BASE]; 960 prot[TLS_BASE][TLS_SW].recvmsg = tls_sw_recvmsg; 961 prot[TLS_BASE][TLS_SW].sock_is_readable = tls_sw_sock_is_readable; 962 prot[TLS_BASE][TLS_SW].close = tls_sk_proto_close; 963 964 prot[TLS_SW][TLS_SW] = prot[TLS_SW][TLS_BASE]; 965 prot[TLS_SW][TLS_SW].recvmsg = tls_sw_recvmsg; 966 prot[TLS_SW][TLS_SW].sock_is_readable = tls_sw_sock_is_readable; 967 prot[TLS_SW][TLS_SW].close = tls_sk_proto_close; 968 969 #ifdef CONFIG_TLS_DEVICE 970 prot[TLS_HW][TLS_BASE] = prot[TLS_BASE][TLS_BASE]; 971 prot[TLS_HW][TLS_BASE].sendmsg = tls_device_sendmsg; 972 prot[TLS_HW][TLS_BASE].splice_eof = tls_device_splice_eof; 973 974 prot[TLS_HW][TLS_SW] = prot[TLS_BASE][TLS_SW]; 975 prot[TLS_HW][TLS_SW].sendmsg = tls_device_sendmsg; 976 prot[TLS_HW][TLS_SW].splice_eof = tls_device_splice_eof; 977 978 prot[TLS_BASE][TLS_HW] = prot[TLS_BASE][TLS_SW]; 979 980 prot[TLS_SW][TLS_HW] = prot[TLS_SW][TLS_SW]; 981 982 prot[TLS_HW][TLS_HW] = prot[TLS_HW][TLS_SW]; 983 #endif 984 #ifdef CONFIG_TLS_TOE 985 prot[TLS_HW_RECORD][TLS_HW_RECORD] = *base; 986 prot[TLS_HW_RECORD][TLS_HW_RECORD].hash = tls_toe_hash; 987 prot[TLS_HW_RECORD][TLS_HW_RECORD].unhash = tls_toe_unhash; 988 #endif 989 } 990 991 static int tls_init(struct sock *sk) 992 { 993 struct tls_context *ctx; 994 int rc = 0; 995 996 tls_build_proto(sk); 997 998 #ifdef CONFIG_TLS_TOE 999 if (tls_toe_bypass(sk)) 1000 return 0; 1001 #endif 1002 1003 /* The TLS ulp is currently supported only for TCP sockets 1004 * in ESTABLISHED state. 1005 * Supporting sockets in LISTEN state will require us 1006 * to modify the accept implementation to clone rather then 1007 * share the ulp context. 1008 */ 1009 if (sk->sk_state != TCP_ESTABLISHED) 1010 return -ENOTCONN; 1011 1012 /* allocate tls context */ 1013 write_lock_bh(&sk->sk_callback_lock); 1014 ctx = tls_ctx_create(sk); 1015 if (!ctx) { 1016 rc = -ENOMEM; 1017 goto out; 1018 } 1019 1020 ctx->tx_conf = TLS_BASE; 1021 ctx->rx_conf = TLS_BASE; 1022 update_sk_prot(sk, ctx); 1023 out: 1024 write_unlock_bh(&sk->sk_callback_lock); 1025 return rc; 1026 } 1027 1028 static void tls_update(struct sock *sk, struct proto *p, 1029 void (*write_space)(struct sock *sk)) 1030 { 1031 struct tls_context *ctx; 1032 1033 WARN_ON_ONCE(sk->sk_prot == p); 1034 1035 ctx = tls_get_ctx(sk); 1036 if (likely(ctx)) { 1037 ctx->sk_write_space = write_space; 1038 ctx->sk_proto = p; 1039 } else { 1040 /* Pairs with lockless read in sk_clone_lock(). */ 1041 WRITE_ONCE(sk->sk_prot, p); 1042 sk->sk_write_space = write_space; 1043 } 1044 } 1045 1046 static u16 tls_user_config(struct tls_context *ctx, bool tx) 1047 { 1048 u16 config = tx ? ctx->tx_conf : ctx->rx_conf; 1049 1050 switch (config) { 1051 case TLS_BASE: 1052 return TLS_CONF_BASE; 1053 case TLS_SW: 1054 return TLS_CONF_SW; 1055 case TLS_HW: 1056 return TLS_CONF_HW; 1057 case TLS_HW_RECORD: 1058 return TLS_CONF_HW_RECORD; 1059 } 1060 return 0; 1061 } 1062 1063 static int tls_get_info(struct sock *sk, struct sk_buff *skb, bool net_admin) 1064 { 1065 u16 version, cipher_type; 1066 struct tls_context *ctx; 1067 struct nlattr *start; 1068 int err; 1069 1070 start = nla_nest_start_noflag(skb, INET_ULP_INFO_TLS); 1071 if (!start) 1072 return -EMSGSIZE; 1073 1074 rcu_read_lock(); 1075 ctx = rcu_dereference(inet_csk(sk)->icsk_ulp_data); 1076 if (!ctx) { 1077 err = 0; 1078 goto nla_failure; 1079 } 1080 version = ctx->prot_info.version; 1081 if (version) { 1082 err = nla_put_u16(skb, TLS_INFO_VERSION, version); 1083 if (err) 1084 goto nla_failure; 1085 } 1086 cipher_type = ctx->prot_info.cipher_type; 1087 if (cipher_type) { 1088 err = nla_put_u16(skb, TLS_INFO_CIPHER, cipher_type); 1089 if (err) 1090 goto nla_failure; 1091 } 1092 err = nla_put_u16(skb, TLS_INFO_TXCONF, tls_user_config(ctx, true)); 1093 if (err) 1094 goto nla_failure; 1095 1096 err = nla_put_u16(skb, TLS_INFO_RXCONF, tls_user_config(ctx, false)); 1097 if (err) 1098 goto nla_failure; 1099 1100 if (ctx->tx_conf == TLS_HW && ctx->zerocopy_sendfile) { 1101 err = nla_put_flag(skb, TLS_INFO_ZC_RO_TX); 1102 if (err) 1103 goto nla_failure; 1104 } 1105 if (ctx->rx_no_pad) { 1106 err = nla_put_flag(skb, TLS_INFO_RX_NO_PAD); 1107 if (err) 1108 goto nla_failure; 1109 } 1110 1111 rcu_read_unlock(); 1112 nla_nest_end(skb, start); 1113 return 0; 1114 1115 nla_failure: 1116 rcu_read_unlock(); 1117 nla_nest_cancel(skb, start); 1118 return err; 1119 } 1120 1121 static size_t tls_get_info_size(const struct sock *sk, bool net_admin) 1122 { 1123 size_t size = 0; 1124 1125 size += nla_total_size(0) + /* INET_ULP_INFO_TLS */ 1126 nla_total_size(sizeof(u16)) + /* TLS_INFO_VERSION */ 1127 nla_total_size(sizeof(u16)) + /* TLS_INFO_CIPHER */ 1128 nla_total_size(sizeof(u16)) + /* TLS_INFO_RXCONF */ 1129 nla_total_size(sizeof(u16)) + /* TLS_INFO_TXCONF */ 1130 nla_total_size(0) + /* TLS_INFO_ZC_RO_TX */ 1131 nla_total_size(0) + /* TLS_INFO_RX_NO_PAD */ 1132 0; 1133 1134 return size; 1135 } 1136 1137 static int __net_init tls_init_net(struct net *net) 1138 { 1139 int err; 1140 1141 net->mib.tls_statistics = alloc_percpu(struct linux_tls_mib); 1142 if (!net->mib.tls_statistics) 1143 return -ENOMEM; 1144 1145 err = tls_proc_init(net); 1146 if (err) 1147 goto err_free_stats; 1148 1149 return 0; 1150 err_free_stats: 1151 free_percpu(net->mib.tls_statistics); 1152 return err; 1153 } 1154 1155 static void __net_exit tls_exit_net(struct net *net) 1156 { 1157 tls_proc_fini(net); 1158 free_percpu(net->mib.tls_statistics); 1159 } 1160 1161 static struct pernet_operations tls_proc_ops = { 1162 .init = tls_init_net, 1163 .exit = tls_exit_net, 1164 }; 1165 1166 static struct tcp_ulp_ops tcp_tls_ulp_ops __read_mostly = { 1167 .name = "tls", 1168 .owner = THIS_MODULE, 1169 .init = tls_init, 1170 .update = tls_update, 1171 .get_info = tls_get_info, 1172 .get_info_size = tls_get_info_size, 1173 }; 1174 1175 static int __init tls_register(void) 1176 { 1177 int err; 1178 1179 err = register_pernet_subsys(&tls_proc_ops); 1180 if (err) 1181 return err; 1182 1183 err = tls_strp_dev_init(); 1184 if (err) 1185 goto err_pernet; 1186 1187 err = tls_device_init(); 1188 if (err) 1189 goto err_strp; 1190 1191 tcp_register_ulp(&tcp_tls_ulp_ops); 1192 1193 return 0; 1194 err_strp: 1195 tls_strp_dev_exit(); 1196 err_pernet: 1197 unregister_pernet_subsys(&tls_proc_ops); 1198 return err; 1199 } 1200 1201 static void __exit tls_unregister(void) 1202 { 1203 tcp_unregister_ulp(&tcp_tls_ulp_ops); 1204 tls_strp_dev_exit(); 1205 tls_device_cleanup(); 1206 unregister_pernet_subsys(&tls_proc_ops); 1207 } 1208 1209 module_init(tls_register); 1210 module_exit(tls_unregister); 1211