1 /* Copyright (c) 2018, Mellanox Technologies All rights reserved. 2 * 3 * This software is available to you under a choice of one of two 4 * licenses. You may choose to be licensed under the terms of the GNU 5 * General Public License (GPL) Version 2, available from the file 6 * COPYING in the main directory of this source tree, or the 7 * OpenIB.org BSD license below: 8 * 9 * Redistribution and use in source and binary forms, with or 10 * without modification, are permitted provided that the following 11 * conditions are met: 12 * 13 * - Redistributions of source code must retain the above 14 * copyright notice, this list of conditions and the following 15 * disclaimer. 16 * 17 * - Redistributions in binary form must reproduce the above 18 * copyright notice, this list of conditions and the following 19 * disclaimer in the documentation and/or other materials 20 * provided with the distribution. 21 * 22 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 23 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 24 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 25 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 26 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 27 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 28 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 29 * SOFTWARE. 30 */ 31 32 #include <crypto/aead.h> 33 #include <linux/highmem.h> 34 #include <linux/module.h> 35 #include <linux/netdevice.h> 36 #include <net/dst.h> 37 #include <net/inet_connection_sock.h> 38 #include <net/tcp.h> 39 #include <net/tls.h> 40 41 /* device_offload_lock is used to synchronize tls_dev_add 42 * against NETDEV_DOWN notifications. 43 */ 44 static DECLARE_RWSEM(device_offload_lock); 45 46 static void tls_device_gc_task(struct work_struct *work); 47 48 static DECLARE_WORK(tls_device_gc_work, tls_device_gc_task); 49 static LIST_HEAD(tls_device_gc_list); 50 static LIST_HEAD(tls_device_list); 51 static DEFINE_SPINLOCK(tls_device_lock); 52 53 static void tls_device_free_ctx(struct tls_context *ctx) 54 { 55 if (ctx->tx_conf == TLS_HW) 56 kfree(tls_offload_ctx_tx(ctx)); 57 58 if (ctx->rx_conf == TLS_HW) 59 kfree(tls_offload_ctx_rx(ctx)); 60 61 kfree(ctx); 62 } 63 64 static void tls_device_gc_task(struct work_struct *work) 65 { 66 struct tls_context *ctx, *tmp; 67 unsigned long flags; 68 LIST_HEAD(gc_list); 69 70 spin_lock_irqsave(&tls_device_lock, flags); 71 list_splice_init(&tls_device_gc_list, &gc_list); 72 spin_unlock_irqrestore(&tls_device_lock, flags); 73 74 list_for_each_entry_safe(ctx, tmp, &gc_list, list) { 75 struct net_device *netdev = ctx->netdev; 76 77 if (netdev && ctx->tx_conf == TLS_HW) { 78 netdev->tlsdev_ops->tls_dev_del(netdev, ctx, 79 TLS_OFFLOAD_CTX_DIR_TX); 80 dev_put(netdev); 81 ctx->netdev = NULL; 82 } 83 84 list_del(&ctx->list); 85 tls_device_free_ctx(ctx); 86 } 87 } 88 89 static void tls_device_attach(struct tls_context *ctx, struct sock *sk, 90 struct net_device *netdev) 91 { 92 if (sk->sk_destruct != tls_device_sk_destruct) { 93 refcount_set(&ctx->refcount, 1); 94 dev_hold(netdev); 95 ctx->netdev = netdev; 96 spin_lock_irq(&tls_device_lock); 97 list_add_tail(&ctx->list, &tls_device_list); 98 spin_unlock_irq(&tls_device_lock); 99 100 ctx->sk_destruct = sk->sk_destruct; 101 sk->sk_destruct = tls_device_sk_destruct; 102 } 103 } 104 105 static void tls_device_queue_ctx_destruction(struct tls_context *ctx) 106 { 107 unsigned long flags; 108 109 spin_lock_irqsave(&tls_device_lock, flags); 110 list_move_tail(&ctx->list, &tls_device_gc_list); 111 112 /* schedule_work inside the spinlock 113 * to make sure tls_device_down waits for that work. 114 */ 115 schedule_work(&tls_device_gc_work); 116 117 spin_unlock_irqrestore(&tls_device_lock, flags); 118 } 119 120 /* We assume that the socket is already connected */ 121 static struct net_device *get_netdev_for_sock(struct sock *sk) 122 { 123 struct dst_entry *dst = sk_dst_get(sk); 124 struct net_device *netdev = NULL; 125 126 if (likely(dst)) { 127 netdev = dst->dev; 128 dev_hold(netdev); 129 } 130 131 dst_release(dst); 132 133 return netdev; 134 } 135 136 static void destroy_record(struct tls_record_info *record) 137 { 138 int nr_frags = record->num_frags; 139 skb_frag_t *frag; 140 141 while (nr_frags-- > 0) { 142 frag = &record->frags[nr_frags]; 143 __skb_frag_unref(frag); 144 } 145 kfree(record); 146 } 147 148 static void delete_all_records(struct tls_offload_context_tx *offload_ctx) 149 { 150 struct tls_record_info *info, *temp; 151 152 list_for_each_entry_safe(info, temp, &offload_ctx->records_list, list) { 153 list_del(&info->list); 154 destroy_record(info); 155 } 156 157 offload_ctx->retransmit_hint = NULL; 158 } 159 160 static void tls_icsk_clean_acked(struct sock *sk, u32 acked_seq) 161 { 162 struct tls_context *tls_ctx = tls_get_ctx(sk); 163 struct tls_record_info *info, *temp; 164 struct tls_offload_context_tx *ctx; 165 u64 deleted_records = 0; 166 unsigned long flags; 167 168 if (!tls_ctx) 169 return; 170 171 ctx = tls_offload_ctx_tx(tls_ctx); 172 173 spin_lock_irqsave(&ctx->lock, flags); 174 info = ctx->retransmit_hint; 175 if (info && !before(acked_seq, info->end_seq)) { 176 ctx->retransmit_hint = NULL; 177 list_del(&info->list); 178 destroy_record(info); 179 deleted_records++; 180 } 181 182 list_for_each_entry_safe(info, temp, &ctx->records_list, list) { 183 if (before(acked_seq, info->end_seq)) 184 break; 185 list_del(&info->list); 186 187 destroy_record(info); 188 deleted_records++; 189 } 190 191 ctx->unacked_record_sn += deleted_records; 192 spin_unlock_irqrestore(&ctx->lock, flags); 193 } 194 195 /* At this point, there should be no references on this 196 * socket and no in-flight SKBs associated with this 197 * socket, so it is safe to free all the resources. 198 */ 199 void tls_device_sk_destruct(struct sock *sk) 200 { 201 struct tls_context *tls_ctx = tls_get_ctx(sk); 202 struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx); 203 204 tls_ctx->sk_destruct(sk); 205 206 if (tls_ctx->tx_conf == TLS_HW) { 207 if (ctx->open_record) 208 destroy_record(ctx->open_record); 209 delete_all_records(ctx); 210 crypto_free_aead(ctx->aead_send); 211 clean_acked_data_disable(inet_csk(sk)); 212 } 213 214 if (refcount_dec_and_test(&tls_ctx->refcount)) 215 tls_device_queue_ctx_destruction(tls_ctx); 216 } 217 EXPORT_SYMBOL(tls_device_sk_destruct); 218 219 static void tls_append_frag(struct tls_record_info *record, 220 struct page_frag *pfrag, 221 int size) 222 { 223 skb_frag_t *frag; 224 225 frag = &record->frags[record->num_frags - 1]; 226 if (frag->page.p == pfrag->page && 227 frag->page_offset + frag->size == pfrag->offset) { 228 frag->size += size; 229 } else { 230 ++frag; 231 frag->page.p = pfrag->page; 232 frag->page_offset = pfrag->offset; 233 frag->size = size; 234 ++record->num_frags; 235 get_page(pfrag->page); 236 } 237 238 pfrag->offset += size; 239 record->len += size; 240 } 241 242 static int tls_push_record(struct sock *sk, 243 struct tls_context *ctx, 244 struct tls_offload_context_tx *offload_ctx, 245 struct tls_record_info *record, 246 struct page_frag *pfrag, 247 int flags, 248 unsigned char record_type) 249 { 250 struct tls_prot_info *prot = &ctx->prot_info; 251 struct tcp_sock *tp = tcp_sk(sk); 252 struct page_frag dummy_tag_frag; 253 skb_frag_t *frag; 254 int i; 255 256 /* fill prepend */ 257 frag = &record->frags[0]; 258 tls_fill_prepend(ctx, 259 skb_frag_address(frag), 260 record->len - prot->prepend_size, 261 record_type, 262 ctx->crypto_send.info.version); 263 264 /* HW doesn't care about the data in the tag, because it fills it. */ 265 dummy_tag_frag.page = skb_frag_page(frag); 266 dummy_tag_frag.offset = 0; 267 268 tls_append_frag(record, &dummy_tag_frag, prot->tag_size); 269 record->end_seq = tp->write_seq + record->len; 270 spin_lock_irq(&offload_ctx->lock); 271 list_add_tail(&record->list, &offload_ctx->records_list); 272 spin_unlock_irq(&offload_ctx->lock); 273 offload_ctx->open_record = NULL; 274 tls_advance_record_sn(sk, &ctx->tx, ctx->crypto_send.info.version); 275 276 for (i = 0; i < record->num_frags; i++) { 277 frag = &record->frags[i]; 278 sg_unmark_end(&offload_ctx->sg_tx_data[i]); 279 sg_set_page(&offload_ctx->sg_tx_data[i], skb_frag_page(frag), 280 frag->size, frag->page_offset); 281 sk_mem_charge(sk, frag->size); 282 get_page(skb_frag_page(frag)); 283 } 284 sg_mark_end(&offload_ctx->sg_tx_data[record->num_frags - 1]); 285 286 /* all ready, send */ 287 return tls_push_sg(sk, ctx, offload_ctx->sg_tx_data, 0, flags); 288 } 289 290 static int tls_create_new_record(struct tls_offload_context_tx *offload_ctx, 291 struct page_frag *pfrag, 292 size_t prepend_size) 293 { 294 struct tls_record_info *record; 295 skb_frag_t *frag; 296 297 record = kmalloc(sizeof(*record), GFP_KERNEL); 298 if (!record) 299 return -ENOMEM; 300 301 frag = &record->frags[0]; 302 __skb_frag_set_page(frag, pfrag->page); 303 frag->page_offset = pfrag->offset; 304 skb_frag_size_set(frag, prepend_size); 305 306 get_page(pfrag->page); 307 pfrag->offset += prepend_size; 308 309 record->num_frags = 1; 310 record->len = prepend_size; 311 offload_ctx->open_record = record; 312 return 0; 313 } 314 315 static int tls_do_allocation(struct sock *sk, 316 struct tls_offload_context_tx *offload_ctx, 317 struct page_frag *pfrag, 318 size_t prepend_size) 319 { 320 int ret; 321 322 if (!offload_ctx->open_record) { 323 if (unlikely(!skb_page_frag_refill(prepend_size, pfrag, 324 sk->sk_allocation))) { 325 sk->sk_prot->enter_memory_pressure(sk); 326 sk_stream_moderate_sndbuf(sk); 327 return -ENOMEM; 328 } 329 330 ret = tls_create_new_record(offload_ctx, pfrag, prepend_size); 331 if (ret) 332 return ret; 333 334 if (pfrag->size > pfrag->offset) 335 return 0; 336 } 337 338 if (!sk_page_frag_refill(sk, pfrag)) 339 return -ENOMEM; 340 341 return 0; 342 } 343 344 static int tls_push_data(struct sock *sk, 345 struct iov_iter *msg_iter, 346 size_t size, int flags, 347 unsigned char record_type) 348 { 349 struct tls_context *tls_ctx = tls_get_ctx(sk); 350 struct tls_prot_info *prot = &tls_ctx->prot_info; 351 struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx); 352 int tls_push_record_flags = flags | MSG_SENDPAGE_NOTLAST; 353 int more = flags & (MSG_SENDPAGE_NOTLAST | MSG_MORE); 354 struct tls_record_info *record = ctx->open_record; 355 struct page_frag *pfrag; 356 size_t orig_size = size; 357 u32 max_open_record_len; 358 int copy, rc = 0; 359 bool done = false; 360 long timeo; 361 362 if (flags & 363 ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_SENDPAGE_NOTLAST)) 364 return -ENOTSUPP; 365 366 if (sk->sk_err) 367 return -sk->sk_err; 368 369 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); 370 if (tls_is_partially_sent_record(tls_ctx)) { 371 rc = tls_push_partial_record(sk, tls_ctx, flags); 372 if (rc < 0) 373 return rc; 374 } 375 376 pfrag = sk_page_frag(sk); 377 378 /* TLS_HEADER_SIZE is not counted as part of the TLS record, and 379 * we need to leave room for an authentication tag. 380 */ 381 max_open_record_len = TLS_MAX_PAYLOAD_SIZE + 382 prot->prepend_size; 383 do { 384 rc = tls_do_allocation(sk, ctx, pfrag, 385 prot->prepend_size); 386 if (rc) { 387 rc = sk_stream_wait_memory(sk, &timeo); 388 if (!rc) 389 continue; 390 391 record = ctx->open_record; 392 if (!record) 393 break; 394 handle_error: 395 if (record_type != TLS_RECORD_TYPE_DATA) { 396 /* avoid sending partial 397 * record with type != 398 * application_data 399 */ 400 size = orig_size; 401 destroy_record(record); 402 ctx->open_record = NULL; 403 } else if (record->len > prot->prepend_size) { 404 goto last_record; 405 } 406 407 break; 408 } 409 410 record = ctx->open_record; 411 copy = min_t(size_t, size, (pfrag->size - pfrag->offset)); 412 copy = min_t(size_t, copy, (max_open_record_len - record->len)); 413 414 if (copy_from_iter_nocache(page_address(pfrag->page) + 415 pfrag->offset, 416 copy, msg_iter) != copy) { 417 rc = -EFAULT; 418 goto handle_error; 419 } 420 tls_append_frag(record, pfrag, copy); 421 422 size -= copy; 423 if (!size) { 424 last_record: 425 tls_push_record_flags = flags; 426 if (more) { 427 tls_ctx->pending_open_record_frags = 428 !!record->num_frags; 429 break; 430 } 431 432 done = true; 433 } 434 435 if (done || record->len >= max_open_record_len || 436 (record->num_frags >= MAX_SKB_FRAGS - 1)) { 437 rc = tls_push_record(sk, 438 tls_ctx, 439 ctx, 440 record, 441 pfrag, 442 tls_push_record_flags, 443 record_type); 444 if (rc < 0) 445 break; 446 } 447 } while (!done); 448 449 if (orig_size - size > 0) 450 rc = orig_size - size; 451 452 return rc; 453 } 454 455 int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size) 456 { 457 unsigned char record_type = TLS_RECORD_TYPE_DATA; 458 int rc; 459 460 lock_sock(sk); 461 462 if (unlikely(msg->msg_controllen)) { 463 rc = tls_proccess_cmsg(sk, msg, &record_type); 464 if (rc) 465 goto out; 466 } 467 468 rc = tls_push_data(sk, &msg->msg_iter, size, 469 msg->msg_flags, record_type); 470 471 out: 472 release_sock(sk); 473 return rc; 474 } 475 476 int tls_device_sendpage(struct sock *sk, struct page *page, 477 int offset, size_t size, int flags) 478 { 479 struct iov_iter msg_iter; 480 char *kaddr = kmap(page); 481 struct kvec iov; 482 int rc; 483 484 if (flags & MSG_SENDPAGE_NOTLAST) 485 flags |= MSG_MORE; 486 487 lock_sock(sk); 488 489 if (flags & MSG_OOB) { 490 rc = -ENOTSUPP; 491 goto out; 492 } 493 494 iov.iov_base = kaddr + offset; 495 iov.iov_len = size; 496 iov_iter_kvec(&msg_iter, WRITE, &iov, 1, size); 497 rc = tls_push_data(sk, &msg_iter, size, 498 flags, TLS_RECORD_TYPE_DATA); 499 kunmap(page); 500 501 out: 502 release_sock(sk); 503 return rc; 504 } 505 506 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context, 507 u32 seq, u64 *p_record_sn) 508 { 509 u64 record_sn = context->hint_record_sn; 510 struct tls_record_info *info; 511 512 info = context->retransmit_hint; 513 if (!info || 514 before(seq, info->end_seq - info->len)) { 515 /* if retransmit_hint is irrelevant start 516 * from the beggining of the list 517 */ 518 info = list_first_entry(&context->records_list, 519 struct tls_record_info, list); 520 record_sn = context->unacked_record_sn; 521 } 522 523 list_for_each_entry_from(info, &context->records_list, list) { 524 if (before(seq, info->end_seq)) { 525 if (!context->retransmit_hint || 526 after(info->end_seq, 527 context->retransmit_hint->end_seq)) { 528 context->hint_record_sn = record_sn; 529 context->retransmit_hint = info; 530 } 531 *p_record_sn = record_sn; 532 return info; 533 } 534 record_sn++; 535 } 536 537 return NULL; 538 } 539 EXPORT_SYMBOL(tls_get_record); 540 541 static int tls_device_push_pending_record(struct sock *sk, int flags) 542 { 543 struct iov_iter msg_iter; 544 545 iov_iter_kvec(&msg_iter, WRITE, NULL, 0, 0); 546 return tls_push_data(sk, &msg_iter, 0, flags, TLS_RECORD_TYPE_DATA); 547 } 548 549 void tls_device_write_space(struct sock *sk, struct tls_context *ctx) 550 { 551 int rc = 0; 552 553 if (!sk->sk_write_pending && tls_is_partially_sent_record(ctx)) { 554 gfp_t sk_allocation = sk->sk_allocation; 555 556 sk->sk_allocation = GFP_ATOMIC; 557 rc = tls_push_partial_record(sk, ctx, 558 MSG_DONTWAIT | MSG_NOSIGNAL); 559 sk->sk_allocation = sk_allocation; 560 } 561 562 if (!rc) 563 ctx->sk_write_space(sk); 564 } 565 566 void handle_device_resync(struct sock *sk, u32 seq, u64 rcd_sn) 567 { 568 struct tls_context *tls_ctx = tls_get_ctx(sk); 569 struct net_device *netdev = tls_ctx->netdev; 570 struct tls_offload_context_rx *rx_ctx; 571 u32 is_req_pending; 572 s64 resync_req; 573 u32 req_seq; 574 575 if (tls_ctx->rx_conf != TLS_HW) 576 return; 577 578 rx_ctx = tls_offload_ctx_rx(tls_ctx); 579 resync_req = atomic64_read(&rx_ctx->resync_req); 580 req_seq = ntohl(resync_req >> 32) - ((u32)TLS_HEADER_SIZE - 1); 581 is_req_pending = resync_req; 582 583 if (unlikely(is_req_pending) && req_seq == seq && 584 atomic64_try_cmpxchg(&rx_ctx->resync_req, &resync_req, 0)) 585 netdev->tlsdev_ops->tls_dev_resync_rx(netdev, sk, 586 seq + TLS_HEADER_SIZE - 1, 587 rcd_sn); 588 } 589 590 static int tls_device_reencrypt(struct sock *sk, struct sk_buff *skb) 591 { 592 struct strp_msg *rxm = strp_msg(skb); 593 int err = 0, offset = rxm->offset, copy, nsg; 594 struct sk_buff *skb_iter, *unused; 595 struct scatterlist sg[1]; 596 char *orig_buf, *buf; 597 598 orig_buf = kmalloc(rxm->full_len + TLS_HEADER_SIZE + 599 TLS_CIPHER_AES_GCM_128_IV_SIZE, sk->sk_allocation); 600 if (!orig_buf) 601 return -ENOMEM; 602 buf = orig_buf; 603 604 nsg = skb_cow_data(skb, 0, &unused); 605 if (unlikely(nsg < 0)) { 606 err = nsg; 607 goto free_buf; 608 } 609 610 sg_init_table(sg, 1); 611 sg_set_buf(&sg[0], buf, 612 rxm->full_len + TLS_HEADER_SIZE + 613 TLS_CIPHER_AES_GCM_128_IV_SIZE); 614 skb_copy_bits(skb, offset, buf, 615 TLS_HEADER_SIZE + TLS_CIPHER_AES_GCM_128_IV_SIZE); 616 617 /* We are interested only in the decrypted data not the auth */ 618 err = decrypt_skb(sk, skb, sg); 619 if (err != -EBADMSG) 620 goto free_buf; 621 else 622 err = 0; 623 624 copy = min_t(int, skb_pagelen(skb) - offset, 625 rxm->full_len - TLS_CIPHER_AES_GCM_128_TAG_SIZE); 626 627 if (skb->decrypted) 628 skb_store_bits(skb, offset, buf, copy); 629 630 offset += copy; 631 buf += copy; 632 633 skb_walk_frags(skb, skb_iter) { 634 copy = min_t(int, skb_iter->len, 635 rxm->full_len - offset + rxm->offset - 636 TLS_CIPHER_AES_GCM_128_TAG_SIZE); 637 638 if (skb_iter->decrypted) 639 skb_store_bits(skb_iter, offset, buf, copy); 640 641 offset += copy; 642 buf += copy; 643 } 644 645 free_buf: 646 kfree(orig_buf); 647 return err; 648 } 649 650 int tls_device_decrypted(struct sock *sk, struct sk_buff *skb) 651 { 652 struct tls_context *tls_ctx = tls_get_ctx(sk); 653 struct tls_offload_context_rx *ctx = tls_offload_ctx_rx(tls_ctx); 654 int is_decrypted = skb->decrypted; 655 int is_encrypted = !is_decrypted; 656 struct sk_buff *skb_iter; 657 658 /* Skip if it is already decrypted */ 659 if (ctx->sw.decrypted) 660 return 0; 661 662 /* Check if all the data is decrypted already */ 663 skb_walk_frags(skb, skb_iter) { 664 is_decrypted &= skb_iter->decrypted; 665 is_encrypted &= !skb_iter->decrypted; 666 } 667 668 ctx->sw.decrypted |= is_decrypted; 669 670 /* Return immedeatly if the record is either entirely plaintext or 671 * entirely ciphertext. Otherwise handle reencrypt partially decrypted 672 * record. 673 */ 674 return (is_encrypted || is_decrypted) ? 0 : 675 tls_device_reencrypt(sk, skb); 676 } 677 678 int tls_set_device_offload(struct sock *sk, struct tls_context *ctx) 679 { 680 u16 nonce_size, tag_size, iv_size, rec_seq_size; 681 struct tls_context *tls_ctx = tls_get_ctx(sk); 682 struct tls_prot_info *prot = &tls_ctx->prot_info; 683 struct tls_record_info *start_marker_record; 684 struct tls_offload_context_tx *offload_ctx; 685 struct tls_crypto_info *crypto_info; 686 struct net_device *netdev; 687 char *iv, *rec_seq; 688 struct sk_buff *skb; 689 int rc = -EINVAL; 690 __be64 rcd_sn; 691 692 if (!ctx) 693 goto out; 694 695 if (ctx->priv_ctx_tx) { 696 rc = -EEXIST; 697 goto out; 698 } 699 700 start_marker_record = kmalloc(sizeof(*start_marker_record), GFP_KERNEL); 701 if (!start_marker_record) { 702 rc = -ENOMEM; 703 goto out; 704 } 705 706 offload_ctx = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_TX, GFP_KERNEL); 707 if (!offload_ctx) { 708 rc = -ENOMEM; 709 goto free_marker_record; 710 } 711 712 crypto_info = &ctx->crypto_send.info; 713 switch (crypto_info->cipher_type) { 714 case TLS_CIPHER_AES_GCM_128: 715 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE; 716 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE; 717 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE; 718 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv; 719 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE; 720 rec_seq = 721 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq; 722 break; 723 default: 724 rc = -EINVAL; 725 goto free_offload_ctx; 726 } 727 728 prot->prepend_size = TLS_HEADER_SIZE + nonce_size; 729 prot->tag_size = tag_size; 730 prot->overhead_size = prot->prepend_size + prot->tag_size; 731 prot->iv_size = iv_size; 732 ctx->tx.iv = kmalloc(iv_size + TLS_CIPHER_AES_GCM_128_SALT_SIZE, 733 GFP_KERNEL); 734 if (!ctx->tx.iv) { 735 rc = -ENOMEM; 736 goto free_offload_ctx; 737 } 738 739 memcpy(ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv, iv_size); 740 741 prot->rec_seq_size = rec_seq_size; 742 ctx->tx.rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL); 743 if (!ctx->tx.rec_seq) { 744 rc = -ENOMEM; 745 goto free_iv; 746 } 747 748 rc = tls_sw_fallback_init(sk, offload_ctx, crypto_info); 749 if (rc) 750 goto free_rec_seq; 751 752 /* start at rec_seq - 1 to account for the start marker record */ 753 memcpy(&rcd_sn, ctx->tx.rec_seq, sizeof(rcd_sn)); 754 offload_ctx->unacked_record_sn = be64_to_cpu(rcd_sn) - 1; 755 756 start_marker_record->end_seq = tcp_sk(sk)->write_seq; 757 start_marker_record->len = 0; 758 start_marker_record->num_frags = 0; 759 760 INIT_LIST_HEAD(&offload_ctx->records_list); 761 list_add_tail(&start_marker_record->list, &offload_ctx->records_list); 762 spin_lock_init(&offload_ctx->lock); 763 sg_init_table(offload_ctx->sg_tx_data, 764 ARRAY_SIZE(offload_ctx->sg_tx_data)); 765 766 clean_acked_data_enable(inet_csk(sk), &tls_icsk_clean_acked); 767 ctx->push_pending_record = tls_device_push_pending_record; 768 769 /* TLS offload is greatly simplified if we don't send 770 * SKBs where only part of the payload needs to be encrypted. 771 * So mark the last skb in the write queue as end of record. 772 */ 773 skb = tcp_write_queue_tail(sk); 774 if (skb) 775 TCP_SKB_CB(skb)->eor = 1; 776 777 /* We support starting offload on multiple sockets 778 * concurrently, so we only need a read lock here. 779 * This lock must precede get_netdev_for_sock to prevent races between 780 * NETDEV_DOWN and setsockopt. 781 */ 782 down_read(&device_offload_lock); 783 netdev = get_netdev_for_sock(sk); 784 if (!netdev) { 785 pr_err_ratelimited("%s: netdev not found\n", __func__); 786 rc = -EINVAL; 787 goto release_lock; 788 } 789 790 if (!(netdev->features & NETIF_F_HW_TLS_TX)) { 791 rc = -ENOTSUPP; 792 goto release_netdev; 793 } 794 795 /* Avoid offloading if the device is down 796 * We don't want to offload new flows after 797 * the NETDEV_DOWN event 798 */ 799 if (!(netdev->flags & IFF_UP)) { 800 rc = -EINVAL; 801 goto release_netdev; 802 } 803 804 ctx->priv_ctx_tx = offload_ctx; 805 rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_TX, 806 &ctx->crypto_send.info, 807 tcp_sk(sk)->write_seq); 808 if (rc) 809 goto release_netdev; 810 811 tls_device_attach(ctx, sk, netdev); 812 813 /* following this assignment tls_is_sk_tx_device_offloaded 814 * will return true and the context might be accessed 815 * by the netdev's xmit function. 816 */ 817 smp_store_release(&sk->sk_validate_xmit_skb, tls_validate_xmit_skb); 818 dev_put(netdev); 819 up_read(&device_offload_lock); 820 goto out; 821 822 release_netdev: 823 dev_put(netdev); 824 release_lock: 825 up_read(&device_offload_lock); 826 clean_acked_data_disable(inet_csk(sk)); 827 crypto_free_aead(offload_ctx->aead_send); 828 free_rec_seq: 829 kfree(ctx->tx.rec_seq); 830 free_iv: 831 kfree(ctx->tx.iv); 832 free_offload_ctx: 833 kfree(offload_ctx); 834 ctx->priv_ctx_tx = NULL; 835 free_marker_record: 836 kfree(start_marker_record); 837 out: 838 return rc; 839 } 840 841 int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx) 842 { 843 struct tls_offload_context_rx *context; 844 struct net_device *netdev; 845 int rc = 0; 846 847 /* We support starting offload on multiple sockets 848 * concurrently, so we only need a read lock here. 849 * This lock must precede get_netdev_for_sock to prevent races between 850 * NETDEV_DOWN and setsockopt. 851 */ 852 down_read(&device_offload_lock); 853 netdev = get_netdev_for_sock(sk); 854 if (!netdev) { 855 pr_err_ratelimited("%s: netdev not found\n", __func__); 856 rc = -EINVAL; 857 goto release_lock; 858 } 859 860 if (!(netdev->features & NETIF_F_HW_TLS_RX)) { 861 pr_err_ratelimited("%s: netdev %s with no TLS offload\n", 862 __func__, netdev->name); 863 rc = -ENOTSUPP; 864 goto release_netdev; 865 } 866 867 /* Avoid offloading if the device is down 868 * We don't want to offload new flows after 869 * the NETDEV_DOWN event 870 */ 871 if (!(netdev->flags & IFF_UP)) { 872 rc = -EINVAL; 873 goto release_netdev; 874 } 875 876 context = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_RX, GFP_KERNEL); 877 if (!context) { 878 rc = -ENOMEM; 879 goto release_netdev; 880 } 881 882 ctx->priv_ctx_rx = context; 883 rc = tls_set_sw_offload(sk, ctx, 0); 884 if (rc) 885 goto release_ctx; 886 887 rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_RX, 888 &ctx->crypto_recv.info, 889 tcp_sk(sk)->copied_seq); 890 if (rc) { 891 pr_err_ratelimited("%s: The netdev has refused to offload this socket\n", 892 __func__); 893 goto free_sw_resources; 894 } 895 896 tls_device_attach(ctx, sk, netdev); 897 goto release_netdev; 898 899 free_sw_resources: 900 tls_sw_free_resources_rx(sk); 901 release_ctx: 902 ctx->priv_ctx_rx = NULL; 903 release_netdev: 904 dev_put(netdev); 905 release_lock: 906 up_read(&device_offload_lock); 907 return rc; 908 } 909 910 void tls_device_offload_cleanup_rx(struct sock *sk) 911 { 912 struct tls_context *tls_ctx = tls_get_ctx(sk); 913 struct net_device *netdev; 914 915 down_read(&device_offload_lock); 916 netdev = tls_ctx->netdev; 917 if (!netdev) 918 goto out; 919 920 if (!(netdev->features & NETIF_F_HW_TLS_RX)) { 921 pr_err_ratelimited("%s: device is missing NETIF_F_HW_TLS_RX cap\n", 922 __func__); 923 goto out; 924 } 925 926 netdev->tlsdev_ops->tls_dev_del(netdev, tls_ctx, 927 TLS_OFFLOAD_CTX_DIR_RX); 928 929 if (tls_ctx->tx_conf != TLS_HW) { 930 dev_put(netdev); 931 tls_ctx->netdev = NULL; 932 } 933 out: 934 up_read(&device_offload_lock); 935 kfree(tls_ctx->rx.rec_seq); 936 kfree(tls_ctx->rx.iv); 937 tls_sw_release_resources_rx(sk); 938 } 939 940 static int tls_device_down(struct net_device *netdev) 941 { 942 struct tls_context *ctx, *tmp; 943 unsigned long flags; 944 LIST_HEAD(list); 945 946 /* Request a write lock to block new offload attempts */ 947 down_write(&device_offload_lock); 948 949 spin_lock_irqsave(&tls_device_lock, flags); 950 list_for_each_entry_safe(ctx, tmp, &tls_device_list, list) { 951 if (ctx->netdev != netdev || 952 !refcount_inc_not_zero(&ctx->refcount)) 953 continue; 954 955 list_move(&ctx->list, &list); 956 } 957 spin_unlock_irqrestore(&tls_device_lock, flags); 958 959 list_for_each_entry_safe(ctx, tmp, &list, list) { 960 if (ctx->tx_conf == TLS_HW) 961 netdev->tlsdev_ops->tls_dev_del(netdev, ctx, 962 TLS_OFFLOAD_CTX_DIR_TX); 963 if (ctx->rx_conf == TLS_HW) 964 netdev->tlsdev_ops->tls_dev_del(netdev, ctx, 965 TLS_OFFLOAD_CTX_DIR_RX); 966 ctx->netdev = NULL; 967 dev_put(netdev); 968 list_del_init(&ctx->list); 969 970 if (refcount_dec_and_test(&ctx->refcount)) 971 tls_device_free_ctx(ctx); 972 } 973 974 up_write(&device_offload_lock); 975 976 flush_work(&tls_device_gc_work); 977 978 return NOTIFY_DONE; 979 } 980 981 static int tls_dev_event(struct notifier_block *this, unsigned long event, 982 void *ptr) 983 { 984 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 985 986 if (!(dev->features & (NETIF_F_HW_TLS_RX | NETIF_F_HW_TLS_TX))) 987 return NOTIFY_DONE; 988 989 switch (event) { 990 case NETDEV_REGISTER: 991 case NETDEV_FEAT_CHANGE: 992 if ((dev->features & NETIF_F_HW_TLS_RX) && 993 !dev->tlsdev_ops->tls_dev_resync_rx) 994 return NOTIFY_BAD; 995 996 if (dev->tlsdev_ops && 997 dev->tlsdev_ops->tls_dev_add && 998 dev->tlsdev_ops->tls_dev_del) 999 return NOTIFY_DONE; 1000 else 1001 return NOTIFY_BAD; 1002 case NETDEV_DOWN: 1003 return tls_device_down(dev); 1004 } 1005 return NOTIFY_DONE; 1006 } 1007 1008 static struct notifier_block tls_dev_notifier = { 1009 .notifier_call = tls_dev_event, 1010 }; 1011 1012 void __init tls_device_init(void) 1013 { 1014 register_netdevice_notifier(&tls_dev_notifier); 1015 } 1016 1017 void __exit tls_device_cleanup(void) 1018 { 1019 unregister_netdevice_notifier(&tls_dev_notifier); 1020 flush_work(&tls_device_gc_work); 1021 } 1022