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 * Copyright (c) 2016-2017, Lance Chao <lancerchao@fb.com>. All rights reserved. 5 * Copyright (c) 2016, Fridolin Pokorny <fridolin.pokorny@gmail.com>. All rights reserved. 6 * Copyright (c) 2016, Nikos Mavrogiannopoulos <nmav@gnutls.org>. All rights reserved. 7 * Copyright (c) 2018, Covalent IO, Inc. http://covalent.io 8 * 9 * This software is available to you under a choice of one of two 10 * licenses. You may choose to be licensed under the terms of the GNU 11 * General Public License (GPL) Version 2, available from the file 12 * COPYING in the main directory of this source tree, or the 13 * OpenIB.org BSD license below: 14 * 15 * Redistribution and use in source and binary forms, with or 16 * without modification, are permitted provided that the following 17 * conditions are met: 18 * 19 * - Redistributions of source code must retain the above 20 * copyright notice, this list of conditions and the following 21 * disclaimer. 22 * 23 * - Redistributions in binary form must reproduce the above 24 * copyright notice, this list of conditions and the following 25 * disclaimer in the documentation and/or other materials 26 * provided with the distribution. 27 * 28 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 29 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 30 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 31 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 32 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 33 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 34 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 35 * SOFTWARE. 36 */ 37 38 #include <linux/sched/signal.h> 39 #include <linux/module.h> 40 #include <crypto/aead.h> 41 42 #include <net/strparser.h> 43 #include <net/tls.h> 44 45 static int __skb_nsg(struct sk_buff *skb, int offset, int len, 46 unsigned int recursion_level) 47 { 48 int start = skb_headlen(skb); 49 int i, chunk = start - offset; 50 struct sk_buff *frag_iter; 51 int elt = 0; 52 53 if (unlikely(recursion_level >= 24)) 54 return -EMSGSIZE; 55 56 if (chunk > 0) { 57 if (chunk > len) 58 chunk = len; 59 elt++; 60 len -= chunk; 61 if (len == 0) 62 return elt; 63 offset += chunk; 64 } 65 66 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 67 int end; 68 69 WARN_ON(start > offset + len); 70 71 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 72 chunk = end - offset; 73 if (chunk > 0) { 74 if (chunk > len) 75 chunk = len; 76 elt++; 77 len -= chunk; 78 if (len == 0) 79 return elt; 80 offset += chunk; 81 } 82 start = end; 83 } 84 85 if (unlikely(skb_has_frag_list(skb))) { 86 skb_walk_frags(skb, frag_iter) { 87 int end, ret; 88 89 WARN_ON(start > offset + len); 90 91 end = start + frag_iter->len; 92 chunk = end - offset; 93 if (chunk > 0) { 94 if (chunk > len) 95 chunk = len; 96 ret = __skb_nsg(frag_iter, offset - start, chunk, 97 recursion_level + 1); 98 if (unlikely(ret < 0)) 99 return ret; 100 elt += ret; 101 len -= chunk; 102 if (len == 0) 103 return elt; 104 offset += chunk; 105 } 106 start = end; 107 } 108 } 109 BUG_ON(len); 110 return elt; 111 } 112 113 /* Return the number of scatterlist elements required to completely map the 114 * skb, or -EMSGSIZE if the recursion depth is exceeded. 115 */ 116 static int skb_nsg(struct sk_buff *skb, int offset, int len) 117 { 118 return __skb_nsg(skb, offset, len, 0); 119 } 120 121 static int padding_length(struct tls_sw_context_rx *ctx, 122 struct tls_prot_info *prot, struct sk_buff *skb) 123 { 124 struct strp_msg *rxm = strp_msg(skb); 125 int sub = 0; 126 127 /* Determine zero-padding length */ 128 if (prot->version == TLS_1_3_VERSION) { 129 char content_type = 0; 130 int err; 131 int back = 17; 132 133 while (content_type == 0) { 134 if (back > rxm->full_len - prot->prepend_size) 135 return -EBADMSG; 136 err = skb_copy_bits(skb, 137 rxm->offset + rxm->full_len - back, 138 &content_type, 1); 139 if (err) 140 return err; 141 if (content_type) 142 break; 143 sub++; 144 back++; 145 } 146 ctx->control = content_type; 147 } 148 return sub; 149 } 150 151 static void tls_decrypt_done(struct crypto_async_request *req, int err) 152 { 153 struct aead_request *aead_req = (struct aead_request *)req; 154 struct scatterlist *sgout = aead_req->dst; 155 struct scatterlist *sgin = aead_req->src; 156 struct tls_sw_context_rx *ctx; 157 struct tls_context *tls_ctx; 158 struct tls_prot_info *prot; 159 struct scatterlist *sg; 160 struct sk_buff *skb; 161 unsigned int pages; 162 int pending; 163 164 skb = (struct sk_buff *)req->data; 165 tls_ctx = tls_get_ctx(skb->sk); 166 ctx = tls_sw_ctx_rx(tls_ctx); 167 prot = &tls_ctx->prot_info; 168 169 /* Propagate if there was an err */ 170 if (err) { 171 if (err == -EBADMSG) 172 TLS_INC_STATS(sock_net(skb->sk), 173 LINUX_MIB_TLSDECRYPTERROR); 174 ctx->async_wait.err = err; 175 tls_err_abort(skb->sk, err); 176 } else { 177 struct strp_msg *rxm = strp_msg(skb); 178 int pad; 179 180 pad = padding_length(ctx, prot, skb); 181 if (pad < 0) { 182 ctx->async_wait.err = pad; 183 tls_err_abort(skb->sk, pad); 184 } else { 185 rxm->full_len -= pad; 186 rxm->offset += prot->prepend_size; 187 rxm->full_len -= prot->overhead_size; 188 } 189 } 190 191 /* After using skb->sk to propagate sk through crypto async callback 192 * we need to NULL it again. 193 */ 194 skb->sk = NULL; 195 196 197 /* Free the destination pages if skb was not decrypted inplace */ 198 if (sgout != sgin) { 199 /* Skip the first S/G entry as it points to AAD */ 200 for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) { 201 if (!sg) 202 break; 203 put_page(sg_page(sg)); 204 } 205 } 206 207 kfree(aead_req); 208 209 pending = atomic_dec_return(&ctx->decrypt_pending); 210 211 if (!pending && READ_ONCE(ctx->async_notify)) 212 complete(&ctx->async_wait.completion); 213 } 214 215 static int tls_do_decryption(struct sock *sk, 216 struct sk_buff *skb, 217 struct scatterlist *sgin, 218 struct scatterlist *sgout, 219 char *iv_recv, 220 size_t data_len, 221 struct aead_request *aead_req, 222 bool async) 223 { 224 struct tls_context *tls_ctx = tls_get_ctx(sk); 225 struct tls_prot_info *prot = &tls_ctx->prot_info; 226 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 227 int ret; 228 229 aead_request_set_tfm(aead_req, ctx->aead_recv); 230 aead_request_set_ad(aead_req, prot->aad_size); 231 aead_request_set_crypt(aead_req, sgin, sgout, 232 data_len + prot->tag_size, 233 (u8 *)iv_recv); 234 235 if (async) { 236 /* Using skb->sk to push sk through to crypto async callback 237 * handler. This allows propagating errors up to the socket 238 * if needed. It _must_ be cleared in the async handler 239 * before consume_skb is called. We _know_ skb->sk is NULL 240 * because it is a clone from strparser. 241 */ 242 skb->sk = sk; 243 aead_request_set_callback(aead_req, 244 CRYPTO_TFM_REQ_MAY_BACKLOG, 245 tls_decrypt_done, skb); 246 atomic_inc(&ctx->decrypt_pending); 247 } else { 248 aead_request_set_callback(aead_req, 249 CRYPTO_TFM_REQ_MAY_BACKLOG, 250 crypto_req_done, &ctx->async_wait); 251 } 252 253 ret = crypto_aead_decrypt(aead_req); 254 if (ret == -EINPROGRESS) { 255 if (async) 256 return ret; 257 258 ret = crypto_wait_req(ret, &ctx->async_wait); 259 } 260 261 if (async) 262 atomic_dec(&ctx->decrypt_pending); 263 264 return ret; 265 } 266 267 static void tls_trim_both_msgs(struct sock *sk, int target_size) 268 { 269 struct tls_context *tls_ctx = tls_get_ctx(sk); 270 struct tls_prot_info *prot = &tls_ctx->prot_info; 271 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 272 struct tls_rec *rec = ctx->open_rec; 273 274 sk_msg_trim(sk, &rec->msg_plaintext, target_size); 275 if (target_size > 0) 276 target_size += prot->overhead_size; 277 sk_msg_trim(sk, &rec->msg_encrypted, target_size); 278 } 279 280 static int tls_alloc_encrypted_msg(struct sock *sk, int len) 281 { 282 struct tls_context *tls_ctx = tls_get_ctx(sk); 283 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 284 struct tls_rec *rec = ctx->open_rec; 285 struct sk_msg *msg_en = &rec->msg_encrypted; 286 287 return sk_msg_alloc(sk, msg_en, len, 0); 288 } 289 290 static int tls_clone_plaintext_msg(struct sock *sk, int required) 291 { 292 struct tls_context *tls_ctx = tls_get_ctx(sk); 293 struct tls_prot_info *prot = &tls_ctx->prot_info; 294 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 295 struct tls_rec *rec = ctx->open_rec; 296 struct sk_msg *msg_pl = &rec->msg_plaintext; 297 struct sk_msg *msg_en = &rec->msg_encrypted; 298 int skip, len; 299 300 /* We add page references worth len bytes from encrypted sg 301 * at the end of plaintext sg. It is guaranteed that msg_en 302 * has enough required room (ensured by caller). 303 */ 304 len = required - msg_pl->sg.size; 305 306 /* Skip initial bytes in msg_en's data to be able to use 307 * same offset of both plain and encrypted data. 308 */ 309 skip = prot->prepend_size + msg_pl->sg.size; 310 311 return sk_msg_clone(sk, msg_pl, msg_en, skip, len); 312 } 313 314 static struct tls_rec *tls_get_rec(struct sock *sk) 315 { 316 struct tls_context *tls_ctx = tls_get_ctx(sk); 317 struct tls_prot_info *prot = &tls_ctx->prot_info; 318 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 319 struct sk_msg *msg_pl, *msg_en; 320 struct tls_rec *rec; 321 int mem_size; 322 323 mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send); 324 325 rec = kzalloc(mem_size, sk->sk_allocation); 326 if (!rec) 327 return NULL; 328 329 msg_pl = &rec->msg_plaintext; 330 msg_en = &rec->msg_encrypted; 331 332 sk_msg_init(msg_pl); 333 sk_msg_init(msg_en); 334 335 sg_init_table(rec->sg_aead_in, 2); 336 sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size); 337 sg_unmark_end(&rec->sg_aead_in[1]); 338 339 sg_init_table(rec->sg_aead_out, 2); 340 sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size); 341 sg_unmark_end(&rec->sg_aead_out[1]); 342 343 return rec; 344 } 345 346 static void tls_free_rec(struct sock *sk, struct tls_rec *rec) 347 { 348 sk_msg_free(sk, &rec->msg_encrypted); 349 sk_msg_free(sk, &rec->msg_plaintext); 350 kfree(rec); 351 } 352 353 static void tls_free_open_rec(struct sock *sk) 354 { 355 struct tls_context *tls_ctx = tls_get_ctx(sk); 356 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 357 struct tls_rec *rec = ctx->open_rec; 358 359 if (rec) { 360 tls_free_rec(sk, rec); 361 ctx->open_rec = NULL; 362 } 363 } 364 365 int tls_tx_records(struct sock *sk, int flags) 366 { 367 struct tls_context *tls_ctx = tls_get_ctx(sk); 368 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 369 struct tls_rec *rec, *tmp; 370 struct sk_msg *msg_en; 371 int tx_flags, rc = 0; 372 373 if (tls_is_partially_sent_record(tls_ctx)) { 374 rec = list_first_entry(&ctx->tx_list, 375 struct tls_rec, list); 376 377 if (flags == -1) 378 tx_flags = rec->tx_flags; 379 else 380 tx_flags = flags; 381 382 rc = tls_push_partial_record(sk, tls_ctx, tx_flags); 383 if (rc) 384 goto tx_err; 385 386 /* Full record has been transmitted. 387 * Remove the head of tx_list 388 */ 389 list_del(&rec->list); 390 sk_msg_free(sk, &rec->msg_plaintext); 391 kfree(rec); 392 } 393 394 /* Tx all ready records */ 395 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) { 396 if (READ_ONCE(rec->tx_ready)) { 397 if (flags == -1) 398 tx_flags = rec->tx_flags; 399 else 400 tx_flags = flags; 401 402 msg_en = &rec->msg_encrypted; 403 rc = tls_push_sg(sk, tls_ctx, 404 &msg_en->sg.data[msg_en->sg.curr], 405 0, tx_flags); 406 if (rc) 407 goto tx_err; 408 409 list_del(&rec->list); 410 sk_msg_free(sk, &rec->msg_plaintext); 411 kfree(rec); 412 } else { 413 break; 414 } 415 } 416 417 tx_err: 418 if (rc < 0 && rc != -EAGAIN) 419 tls_err_abort(sk, EBADMSG); 420 421 return rc; 422 } 423 424 static void tls_encrypt_done(struct crypto_async_request *req, int err) 425 { 426 struct aead_request *aead_req = (struct aead_request *)req; 427 struct sock *sk = req->data; 428 struct tls_context *tls_ctx = tls_get_ctx(sk); 429 struct tls_prot_info *prot = &tls_ctx->prot_info; 430 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 431 struct scatterlist *sge; 432 struct sk_msg *msg_en; 433 struct tls_rec *rec; 434 bool ready = false; 435 int pending; 436 437 rec = container_of(aead_req, struct tls_rec, aead_req); 438 msg_en = &rec->msg_encrypted; 439 440 sge = sk_msg_elem(msg_en, msg_en->sg.curr); 441 sge->offset -= prot->prepend_size; 442 sge->length += prot->prepend_size; 443 444 /* Check if error is previously set on socket */ 445 if (err || sk->sk_err) { 446 rec = NULL; 447 448 /* If err is already set on socket, return the same code */ 449 if (sk->sk_err) { 450 ctx->async_wait.err = sk->sk_err; 451 } else { 452 ctx->async_wait.err = err; 453 tls_err_abort(sk, err); 454 } 455 } 456 457 if (rec) { 458 struct tls_rec *first_rec; 459 460 /* Mark the record as ready for transmission */ 461 smp_store_mb(rec->tx_ready, true); 462 463 /* If received record is at head of tx_list, schedule tx */ 464 first_rec = list_first_entry(&ctx->tx_list, 465 struct tls_rec, list); 466 if (rec == first_rec) 467 ready = true; 468 } 469 470 pending = atomic_dec_return(&ctx->encrypt_pending); 471 472 if (!pending && READ_ONCE(ctx->async_notify)) 473 complete(&ctx->async_wait.completion); 474 475 if (!ready) 476 return; 477 478 /* Schedule the transmission */ 479 if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) 480 schedule_delayed_work(&ctx->tx_work.work, 1); 481 } 482 483 static int tls_do_encryption(struct sock *sk, 484 struct tls_context *tls_ctx, 485 struct tls_sw_context_tx *ctx, 486 struct aead_request *aead_req, 487 size_t data_len, u32 start) 488 { 489 struct tls_prot_info *prot = &tls_ctx->prot_info; 490 struct tls_rec *rec = ctx->open_rec; 491 struct sk_msg *msg_en = &rec->msg_encrypted; 492 struct scatterlist *sge = sk_msg_elem(msg_en, start); 493 int rc, iv_offset = 0; 494 495 /* For CCM based ciphers, first byte of IV is a constant */ 496 if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) { 497 rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE; 498 iv_offset = 1; 499 } 500 501 memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv, 502 prot->iv_size + prot->salt_size); 503 504 xor_iv_with_seq(prot->version, rec->iv_data, tls_ctx->tx.rec_seq); 505 506 sge->offset += prot->prepend_size; 507 sge->length -= prot->prepend_size; 508 509 msg_en->sg.curr = start; 510 511 aead_request_set_tfm(aead_req, ctx->aead_send); 512 aead_request_set_ad(aead_req, prot->aad_size); 513 aead_request_set_crypt(aead_req, rec->sg_aead_in, 514 rec->sg_aead_out, 515 data_len, rec->iv_data); 516 517 aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG, 518 tls_encrypt_done, sk); 519 520 /* Add the record in tx_list */ 521 list_add_tail((struct list_head *)&rec->list, &ctx->tx_list); 522 atomic_inc(&ctx->encrypt_pending); 523 524 rc = crypto_aead_encrypt(aead_req); 525 if (!rc || rc != -EINPROGRESS) { 526 atomic_dec(&ctx->encrypt_pending); 527 sge->offset -= prot->prepend_size; 528 sge->length += prot->prepend_size; 529 } 530 531 if (!rc) { 532 WRITE_ONCE(rec->tx_ready, true); 533 } else if (rc != -EINPROGRESS) { 534 list_del(&rec->list); 535 return rc; 536 } 537 538 /* Unhook the record from context if encryption is not failure */ 539 ctx->open_rec = NULL; 540 tls_advance_record_sn(sk, prot, &tls_ctx->tx); 541 return rc; 542 } 543 544 static int tls_split_open_record(struct sock *sk, struct tls_rec *from, 545 struct tls_rec **to, struct sk_msg *msg_opl, 546 struct sk_msg *msg_oen, u32 split_point, 547 u32 tx_overhead_size, u32 *orig_end) 548 { 549 u32 i, j, bytes = 0, apply = msg_opl->apply_bytes; 550 struct scatterlist *sge, *osge, *nsge; 551 u32 orig_size = msg_opl->sg.size; 552 struct scatterlist tmp = { }; 553 struct sk_msg *msg_npl; 554 struct tls_rec *new; 555 int ret; 556 557 new = tls_get_rec(sk); 558 if (!new) 559 return -ENOMEM; 560 ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size + 561 tx_overhead_size, 0); 562 if (ret < 0) { 563 tls_free_rec(sk, new); 564 return ret; 565 } 566 567 *orig_end = msg_opl->sg.end; 568 i = msg_opl->sg.start; 569 sge = sk_msg_elem(msg_opl, i); 570 while (apply && sge->length) { 571 if (sge->length > apply) { 572 u32 len = sge->length - apply; 573 574 get_page(sg_page(sge)); 575 sg_set_page(&tmp, sg_page(sge), len, 576 sge->offset + apply); 577 sge->length = apply; 578 bytes += apply; 579 apply = 0; 580 } else { 581 apply -= sge->length; 582 bytes += sge->length; 583 } 584 585 sk_msg_iter_var_next(i); 586 if (i == msg_opl->sg.end) 587 break; 588 sge = sk_msg_elem(msg_opl, i); 589 } 590 591 msg_opl->sg.end = i; 592 msg_opl->sg.curr = i; 593 msg_opl->sg.copybreak = 0; 594 msg_opl->apply_bytes = 0; 595 msg_opl->sg.size = bytes; 596 597 msg_npl = &new->msg_plaintext; 598 msg_npl->apply_bytes = apply; 599 msg_npl->sg.size = orig_size - bytes; 600 601 j = msg_npl->sg.start; 602 nsge = sk_msg_elem(msg_npl, j); 603 if (tmp.length) { 604 memcpy(nsge, &tmp, sizeof(*nsge)); 605 sk_msg_iter_var_next(j); 606 nsge = sk_msg_elem(msg_npl, j); 607 } 608 609 osge = sk_msg_elem(msg_opl, i); 610 while (osge->length) { 611 memcpy(nsge, osge, sizeof(*nsge)); 612 sg_unmark_end(nsge); 613 sk_msg_iter_var_next(i); 614 sk_msg_iter_var_next(j); 615 if (i == *orig_end) 616 break; 617 osge = sk_msg_elem(msg_opl, i); 618 nsge = sk_msg_elem(msg_npl, j); 619 } 620 621 msg_npl->sg.end = j; 622 msg_npl->sg.curr = j; 623 msg_npl->sg.copybreak = 0; 624 625 *to = new; 626 return 0; 627 } 628 629 static void tls_merge_open_record(struct sock *sk, struct tls_rec *to, 630 struct tls_rec *from, u32 orig_end) 631 { 632 struct sk_msg *msg_npl = &from->msg_plaintext; 633 struct sk_msg *msg_opl = &to->msg_plaintext; 634 struct scatterlist *osge, *nsge; 635 u32 i, j; 636 637 i = msg_opl->sg.end; 638 sk_msg_iter_var_prev(i); 639 j = msg_npl->sg.start; 640 641 osge = sk_msg_elem(msg_opl, i); 642 nsge = sk_msg_elem(msg_npl, j); 643 644 if (sg_page(osge) == sg_page(nsge) && 645 osge->offset + osge->length == nsge->offset) { 646 osge->length += nsge->length; 647 put_page(sg_page(nsge)); 648 } 649 650 msg_opl->sg.end = orig_end; 651 msg_opl->sg.curr = orig_end; 652 msg_opl->sg.copybreak = 0; 653 msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size; 654 msg_opl->sg.size += msg_npl->sg.size; 655 656 sk_msg_free(sk, &to->msg_encrypted); 657 sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted); 658 659 kfree(from); 660 } 661 662 static int tls_push_record(struct sock *sk, int flags, 663 unsigned char record_type) 664 { 665 struct tls_context *tls_ctx = tls_get_ctx(sk); 666 struct tls_prot_info *prot = &tls_ctx->prot_info; 667 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 668 struct tls_rec *rec = ctx->open_rec, *tmp = NULL; 669 u32 i, split_point, uninitialized_var(orig_end); 670 struct sk_msg *msg_pl, *msg_en; 671 struct aead_request *req; 672 bool split; 673 int rc; 674 675 if (!rec) 676 return 0; 677 678 msg_pl = &rec->msg_plaintext; 679 msg_en = &rec->msg_encrypted; 680 681 split_point = msg_pl->apply_bytes; 682 split = split_point && split_point < msg_pl->sg.size; 683 if (unlikely((!split && 684 msg_pl->sg.size + 685 prot->overhead_size > msg_en->sg.size) || 686 (split && 687 split_point + 688 prot->overhead_size > msg_en->sg.size))) { 689 split = true; 690 split_point = msg_en->sg.size; 691 } 692 if (split) { 693 rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en, 694 split_point, prot->overhead_size, 695 &orig_end); 696 if (rc < 0) 697 return rc; 698 /* This can happen if above tls_split_open_record allocates 699 * a single large encryption buffer instead of two smaller 700 * ones. In this case adjust pointers and continue without 701 * split. 702 */ 703 if (!msg_pl->sg.size) { 704 tls_merge_open_record(sk, rec, tmp, orig_end); 705 msg_pl = &rec->msg_plaintext; 706 msg_en = &rec->msg_encrypted; 707 split = false; 708 } 709 sk_msg_trim(sk, msg_en, msg_pl->sg.size + 710 prot->overhead_size); 711 } 712 713 rec->tx_flags = flags; 714 req = &rec->aead_req; 715 716 i = msg_pl->sg.end; 717 sk_msg_iter_var_prev(i); 718 719 rec->content_type = record_type; 720 if (prot->version == TLS_1_3_VERSION) { 721 /* Add content type to end of message. No padding added */ 722 sg_set_buf(&rec->sg_content_type, &rec->content_type, 1); 723 sg_mark_end(&rec->sg_content_type); 724 sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1, 725 &rec->sg_content_type); 726 } else { 727 sg_mark_end(sk_msg_elem(msg_pl, i)); 728 } 729 730 if (msg_pl->sg.end < msg_pl->sg.start) { 731 sg_chain(&msg_pl->sg.data[msg_pl->sg.start], 732 MAX_SKB_FRAGS - msg_pl->sg.start + 1, 733 msg_pl->sg.data); 734 } 735 736 i = msg_pl->sg.start; 737 sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]); 738 739 i = msg_en->sg.end; 740 sk_msg_iter_var_prev(i); 741 sg_mark_end(sk_msg_elem(msg_en, i)); 742 743 i = msg_en->sg.start; 744 sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]); 745 746 tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size, 747 tls_ctx->tx.rec_seq, prot->rec_seq_size, 748 record_type, prot->version); 749 750 tls_fill_prepend(tls_ctx, 751 page_address(sg_page(&msg_en->sg.data[i])) + 752 msg_en->sg.data[i].offset, 753 msg_pl->sg.size + prot->tail_size, 754 record_type, prot->version); 755 756 tls_ctx->pending_open_record_frags = false; 757 758 rc = tls_do_encryption(sk, tls_ctx, ctx, req, 759 msg_pl->sg.size + prot->tail_size, i); 760 if (rc < 0) { 761 if (rc != -EINPROGRESS) { 762 tls_err_abort(sk, EBADMSG); 763 if (split) { 764 tls_ctx->pending_open_record_frags = true; 765 tls_merge_open_record(sk, rec, tmp, orig_end); 766 } 767 } 768 ctx->async_capable = 1; 769 return rc; 770 } else if (split) { 771 msg_pl = &tmp->msg_plaintext; 772 msg_en = &tmp->msg_encrypted; 773 sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size); 774 tls_ctx->pending_open_record_frags = true; 775 ctx->open_rec = tmp; 776 } 777 778 return tls_tx_records(sk, flags); 779 } 780 781 static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk, 782 bool full_record, u8 record_type, 783 size_t *copied, int flags) 784 { 785 struct tls_context *tls_ctx = tls_get_ctx(sk); 786 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 787 struct sk_msg msg_redir = { }; 788 struct sk_psock *psock; 789 struct sock *sk_redir; 790 struct tls_rec *rec; 791 bool enospc, policy; 792 int err = 0, send; 793 u32 delta = 0; 794 795 policy = !(flags & MSG_SENDPAGE_NOPOLICY); 796 psock = sk_psock_get(sk); 797 if (!psock || !policy) { 798 err = tls_push_record(sk, flags, record_type); 799 if (err && err != -EINPROGRESS) { 800 *copied -= sk_msg_free(sk, msg); 801 tls_free_open_rec(sk); 802 } 803 return err; 804 } 805 more_data: 806 enospc = sk_msg_full(msg); 807 if (psock->eval == __SK_NONE) { 808 delta = msg->sg.size; 809 psock->eval = sk_psock_msg_verdict(sk, psock, msg); 810 delta -= msg->sg.size; 811 } 812 if (msg->cork_bytes && msg->cork_bytes > msg->sg.size && 813 !enospc && !full_record) { 814 err = -ENOSPC; 815 goto out_err; 816 } 817 msg->cork_bytes = 0; 818 send = msg->sg.size; 819 if (msg->apply_bytes && msg->apply_bytes < send) 820 send = msg->apply_bytes; 821 822 switch (psock->eval) { 823 case __SK_PASS: 824 err = tls_push_record(sk, flags, record_type); 825 if (err && err != -EINPROGRESS) { 826 *copied -= sk_msg_free(sk, msg); 827 tls_free_open_rec(sk); 828 goto out_err; 829 } 830 break; 831 case __SK_REDIRECT: 832 sk_redir = psock->sk_redir; 833 memcpy(&msg_redir, msg, sizeof(*msg)); 834 if (msg->apply_bytes < send) 835 msg->apply_bytes = 0; 836 else 837 msg->apply_bytes -= send; 838 sk_msg_return_zero(sk, msg, send); 839 msg->sg.size -= send; 840 release_sock(sk); 841 err = tcp_bpf_sendmsg_redir(sk_redir, &msg_redir, send, flags); 842 lock_sock(sk); 843 if (err < 0) { 844 *copied -= sk_msg_free_nocharge(sk, &msg_redir); 845 msg->sg.size = 0; 846 } 847 if (msg->sg.size == 0) 848 tls_free_open_rec(sk); 849 break; 850 case __SK_DROP: 851 default: 852 sk_msg_free_partial(sk, msg, send); 853 if (msg->apply_bytes < send) 854 msg->apply_bytes = 0; 855 else 856 msg->apply_bytes -= send; 857 if (msg->sg.size == 0) 858 tls_free_open_rec(sk); 859 *copied -= (send + delta); 860 err = -EACCES; 861 } 862 863 if (likely(!err)) { 864 bool reset_eval = !ctx->open_rec; 865 866 rec = ctx->open_rec; 867 if (rec) { 868 msg = &rec->msg_plaintext; 869 if (!msg->apply_bytes) 870 reset_eval = true; 871 } 872 if (reset_eval) { 873 psock->eval = __SK_NONE; 874 if (psock->sk_redir) { 875 sock_put(psock->sk_redir); 876 psock->sk_redir = NULL; 877 } 878 } 879 if (rec) 880 goto more_data; 881 } 882 out_err: 883 sk_psock_put(sk, psock); 884 return err; 885 } 886 887 static int tls_sw_push_pending_record(struct sock *sk, int flags) 888 { 889 struct tls_context *tls_ctx = tls_get_ctx(sk); 890 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 891 struct tls_rec *rec = ctx->open_rec; 892 struct sk_msg *msg_pl; 893 size_t copied; 894 895 if (!rec) 896 return 0; 897 898 msg_pl = &rec->msg_plaintext; 899 copied = msg_pl->sg.size; 900 if (!copied) 901 return 0; 902 903 return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA, 904 &copied, flags); 905 } 906 907 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size) 908 { 909 long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT); 910 struct tls_context *tls_ctx = tls_get_ctx(sk); 911 struct tls_prot_info *prot = &tls_ctx->prot_info; 912 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 913 bool async_capable = ctx->async_capable; 914 unsigned char record_type = TLS_RECORD_TYPE_DATA; 915 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter); 916 bool eor = !(msg->msg_flags & MSG_MORE); 917 size_t try_to_copy, copied = 0; 918 struct sk_msg *msg_pl, *msg_en; 919 struct tls_rec *rec; 920 int required_size; 921 int num_async = 0; 922 bool full_record; 923 int record_room; 924 int num_zc = 0; 925 int orig_size; 926 int ret = 0; 927 928 if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL)) 929 return -EOPNOTSUPP; 930 931 mutex_lock(&tls_ctx->tx_lock); 932 lock_sock(sk); 933 934 if (unlikely(msg->msg_controllen)) { 935 ret = tls_proccess_cmsg(sk, msg, &record_type); 936 if (ret) { 937 if (ret == -EINPROGRESS) 938 num_async++; 939 else if (ret != -EAGAIN) 940 goto send_end; 941 } 942 } 943 944 while (msg_data_left(msg)) { 945 if (sk->sk_err) { 946 ret = -sk->sk_err; 947 goto send_end; 948 } 949 950 if (ctx->open_rec) 951 rec = ctx->open_rec; 952 else 953 rec = ctx->open_rec = tls_get_rec(sk); 954 if (!rec) { 955 ret = -ENOMEM; 956 goto send_end; 957 } 958 959 msg_pl = &rec->msg_plaintext; 960 msg_en = &rec->msg_encrypted; 961 962 orig_size = msg_pl->sg.size; 963 full_record = false; 964 try_to_copy = msg_data_left(msg); 965 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size; 966 if (try_to_copy >= record_room) { 967 try_to_copy = record_room; 968 full_record = true; 969 } 970 971 required_size = msg_pl->sg.size + try_to_copy + 972 prot->overhead_size; 973 974 if (!sk_stream_memory_free(sk)) 975 goto wait_for_sndbuf; 976 977 alloc_encrypted: 978 ret = tls_alloc_encrypted_msg(sk, required_size); 979 if (ret) { 980 if (ret != -ENOSPC) 981 goto wait_for_memory; 982 983 /* Adjust try_to_copy according to the amount that was 984 * actually allocated. The difference is due 985 * to max sg elements limit 986 */ 987 try_to_copy -= required_size - msg_en->sg.size; 988 full_record = true; 989 } 990 991 if (!is_kvec && (full_record || eor) && !async_capable) { 992 u32 first = msg_pl->sg.end; 993 994 ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter, 995 msg_pl, try_to_copy); 996 if (ret) 997 goto fallback_to_reg_send; 998 999 num_zc++; 1000 copied += try_to_copy; 1001 1002 sk_msg_sg_copy_set(msg_pl, first); 1003 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record, 1004 record_type, &copied, 1005 msg->msg_flags); 1006 if (ret) { 1007 if (ret == -EINPROGRESS) 1008 num_async++; 1009 else if (ret == -ENOMEM) 1010 goto wait_for_memory; 1011 else if (ctx->open_rec && ret == -ENOSPC) 1012 goto rollback_iter; 1013 else if (ret != -EAGAIN) 1014 goto send_end; 1015 } 1016 continue; 1017 rollback_iter: 1018 copied -= try_to_copy; 1019 sk_msg_sg_copy_clear(msg_pl, first); 1020 iov_iter_revert(&msg->msg_iter, 1021 msg_pl->sg.size - orig_size); 1022 fallback_to_reg_send: 1023 sk_msg_trim(sk, msg_pl, orig_size); 1024 } 1025 1026 required_size = msg_pl->sg.size + try_to_copy; 1027 1028 ret = tls_clone_plaintext_msg(sk, required_size); 1029 if (ret) { 1030 if (ret != -ENOSPC) 1031 goto send_end; 1032 1033 /* Adjust try_to_copy according to the amount that was 1034 * actually allocated. The difference is due 1035 * to max sg elements limit 1036 */ 1037 try_to_copy -= required_size - msg_pl->sg.size; 1038 full_record = true; 1039 sk_msg_trim(sk, msg_en, 1040 msg_pl->sg.size + prot->overhead_size); 1041 } 1042 1043 if (try_to_copy) { 1044 ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter, 1045 msg_pl, try_to_copy); 1046 if (ret < 0) 1047 goto trim_sgl; 1048 } 1049 1050 /* Open records defined only if successfully copied, otherwise 1051 * we would trim the sg but not reset the open record frags. 1052 */ 1053 tls_ctx->pending_open_record_frags = true; 1054 copied += try_to_copy; 1055 if (full_record || eor) { 1056 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record, 1057 record_type, &copied, 1058 msg->msg_flags); 1059 if (ret) { 1060 if (ret == -EINPROGRESS) 1061 num_async++; 1062 else if (ret == -ENOMEM) 1063 goto wait_for_memory; 1064 else if (ret != -EAGAIN) { 1065 if (ret == -ENOSPC) 1066 ret = 0; 1067 goto send_end; 1068 } 1069 } 1070 } 1071 1072 continue; 1073 1074 wait_for_sndbuf: 1075 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1076 wait_for_memory: 1077 ret = sk_stream_wait_memory(sk, &timeo); 1078 if (ret) { 1079 trim_sgl: 1080 if (ctx->open_rec) 1081 tls_trim_both_msgs(sk, orig_size); 1082 goto send_end; 1083 } 1084 1085 if (ctx->open_rec && msg_en->sg.size < required_size) 1086 goto alloc_encrypted; 1087 } 1088 1089 if (!num_async) { 1090 goto send_end; 1091 } else if (num_zc) { 1092 /* Wait for pending encryptions to get completed */ 1093 smp_store_mb(ctx->async_notify, true); 1094 1095 if (atomic_read(&ctx->encrypt_pending)) 1096 crypto_wait_req(-EINPROGRESS, &ctx->async_wait); 1097 else 1098 reinit_completion(&ctx->async_wait.completion); 1099 1100 WRITE_ONCE(ctx->async_notify, false); 1101 1102 if (ctx->async_wait.err) { 1103 ret = ctx->async_wait.err; 1104 copied = 0; 1105 } 1106 } 1107 1108 /* Transmit if any encryptions have completed */ 1109 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) { 1110 cancel_delayed_work(&ctx->tx_work.work); 1111 tls_tx_records(sk, msg->msg_flags); 1112 } 1113 1114 send_end: 1115 ret = sk_stream_error(sk, msg->msg_flags, ret); 1116 1117 release_sock(sk); 1118 mutex_unlock(&tls_ctx->tx_lock); 1119 return copied ? copied : ret; 1120 } 1121 1122 static int tls_sw_do_sendpage(struct sock *sk, struct page *page, 1123 int offset, size_t size, int flags) 1124 { 1125 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); 1126 struct tls_context *tls_ctx = tls_get_ctx(sk); 1127 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 1128 struct tls_prot_info *prot = &tls_ctx->prot_info; 1129 unsigned char record_type = TLS_RECORD_TYPE_DATA; 1130 struct sk_msg *msg_pl; 1131 struct tls_rec *rec; 1132 int num_async = 0; 1133 size_t copied = 0; 1134 bool full_record; 1135 int record_room; 1136 int ret = 0; 1137 bool eor; 1138 1139 eor = !(flags & (MSG_MORE | MSG_SENDPAGE_NOTLAST)); 1140 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 1141 1142 /* Call the sk_stream functions to manage the sndbuf mem. */ 1143 while (size > 0) { 1144 size_t copy, required_size; 1145 1146 if (sk->sk_err) { 1147 ret = -sk->sk_err; 1148 goto sendpage_end; 1149 } 1150 1151 if (ctx->open_rec) 1152 rec = ctx->open_rec; 1153 else 1154 rec = ctx->open_rec = tls_get_rec(sk); 1155 if (!rec) { 1156 ret = -ENOMEM; 1157 goto sendpage_end; 1158 } 1159 1160 msg_pl = &rec->msg_plaintext; 1161 1162 full_record = false; 1163 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size; 1164 copy = size; 1165 if (copy >= record_room) { 1166 copy = record_room; 1167 full_record = true; 1168 } 1169 1170 required_size = msg_pl->sg.size + copy + prot->overhead_size; 1171 1172 if (!sk_stream_memory_free(sk)) 1173 goto wait_for_sndbuf; 1174 alloc_payload: 1175 ret = tls_alloc_encrypted_msg(sk, required_size); 1176 if (ret) { 1177 if (ret != -ENOSPC) 1178 goto wait_for_memory; 1179 1180 /* Adjust copy according to the amount that was 1181 * actually allocated. The difference is due 1182 * to max sg elements limit 1183 */ 1184 copy -= required_size - msg_pl->sg.size; 1185 full_record = true; 1186 } 1187 1188 sk_msg_page_add(msg_pl, page, copy, offset); 1189 sk_mem_charge(sk, copy); 1190 1191 offset += copy; 1192 size -= copy; 1193 copied += copy; 1194 1195 tls_ctx->pending_open_record_frags = true; 1196 if (full_record || eor || sk_msg_full(msg_pl)) { 1197 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record, 1198 record_type, &copied, flags); 1199 if (ret) { 1200 if (ret == -EINPROGRESS) 1201 num_async++; 1202 else if (ret == -ENOMEM) 1203 goto wait_for_memory; 1204 else if (ret != -EAGAIN) { 1205 if (ret == -ENOSPC) 1206 ret = 0; 1207 goto sendpage_end; 1208 } 1209 } 1210 } 1211 continue; 1212 wait_for_sndbuf: 1213 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1214 wait_for_memory: 1215 ret = sk_stream_wait_memory(sk, &timeo); 1216 if (ret) { 1217 if (ctx->open_rec) 1218 tls_trim_both_msgs(sk, msg_pl->sg.size); 1219 goto sendpage_end; 1220 } 1221 1222 if (ctx->open_rec) 1223 goto alloc_payload; 1224 } 1225 1226 if (num_async) { 1227 /* Transmit if any encryptions have completed */ 1228 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) { 1229 cancel_delayed_work(&ctx->tx_work.work); 1230 tls_tx_records(sk, flags); 1231 } 1232 } 1233 sendpage_end: 1234 ret = sk_stream_error(sk, flags, ret); 1235 return copied ? copied : ret; 1236 } 1237 1238 int tls_sw_sendpage_locked(struct sock *sk, struct page *page, 1239 int offset, size_t size, int flags) 1240 { 1241 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | 1242 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY | 1243 MSG_NO_SHARED_FRAGS)) 1244 return -EOPNOTSUPP; 1245 1246 return tls_sw_do_sendpage(sk, page, offset, size, flags); 1247 } 1248 1249 int tls_sw_sendpage(struct sock *sk, struct page *page, 1250 int offset, size_t size, int flags) 1251 { 1252 struct tls_context *tls_ctx = tls_get_ctx(sk); 1253 int ret; 1254 1255 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | 1256 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY)) 1257 return -EOPNOTSUPP; 1258 1259 mutex_lock(&tls_ctx->tx_lock); 1260 lock_sock(sk); 1261 ret = tls_sw_do_sendpage(sk, page, offset, size, flags); 1262 release_sock(sk); 1263 mutex_unlock(&tls_ctx->tx_lock); 1264 return ret; 1265 } 1266 1267 static struct sk_buff *tls_wait_data(struct sock *sk, struct sk_psock *psock, 1268 int flags, long timeo, int *err) 1269 { 1270 struct tls_context *tls_ctx = tls_get_ctx(sk); 1271 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 1272 struct sk_buff *skb; 1273 DEFINE_WAIT_FUNC(wait, woken_wake_function); 1274 1275 while (!(skb = ctx->recv_pkt) && sk_psock_queue_empty(psock)) { 1276 if (sk->sk_err) { 1277 *err = sock_error(sk); 1278 return NULL; 1279 } 1280 1281 if (sk->sk_shutdown & RCV_SHUTDOWN) 1282 return NULL; 1283 1284 if (sock_flag(sk, SOCK_DONE)) 1285 return NULL; 1286 1287 if ((flags & MSG_DONTWAIT) || !timeo) { 1288 *err = -EAGAIN; 1289 return NULL; 1290 } 1291 1292 add_wait_queue(sk_sleep(sk), &wait); 1293 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 1294 sk_wait_event(sk, &timeo, 1295 ctx->recv_pkt != skb || 1296 !sk_psock_queue_empty(psock), 1297 &wait); 1298 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 1299 remove_wait_queue(sk_sleep(sk), &wait); 1300 1301 /* Handle signals */ 1302 if (signal_pending(current)) { 1303 *err = sock_intr_errno(timeo); 1304 return NULL; 1305 } 1306 } 1307 1308 return skb; 1309 } 1310 1311 static int tls_setup_from_iter(struct sock *sk, struct iov_iter *from, 1312 int length, int *pages_used, 1313 unsigned int *size_used, 1314 struct scatterlist *to, 1315 int to_max_pages) 1316 { 1317 int rc = 0, i = 0, num_elem = *pages_used, maxpages; 1318 struct page *pages[MAX_SKB_FRAGS]; 1319 unsigned int size = *size_used; 1320 ssize_t copied, use; 1321 size_t offset; 1322 1323 while (length > 0) { 1324 i = 0; 1325 maxpages = to_max_pages - num_elem; 1326 if (maxpages == 0) { 1327 rc = -EFAULT; 1328 goto out; 1329 } 1330 copied = iov_iter_get_pages(from, pages, 1331 length, 1332 maxpages, &offset); 1333 if (copied <= 0) { 1334 rc = -EFAULT; 1335 goto out; 1336 } 1337 1338 iov_iter_advance(from, copied); 1339 1340 length -= copied; 1341 size += copied; 1342 while (copied) { 1343 use = min_t(int, copied, PAGE_SIZE - offset); 1344 1345 sg_set_page(&to[num_elem], 1346 pages[i], use, offset); 1347 sg_unmark_end(&to[num_elem]); 1348 /* We do not uncharge memory from this API */ 1349 1350 offset = 0; 1351 copied -= use; 1352 1353 i++; 1354 num_elem++; 1355 } 1356 } 1357 /* Mark the end in the last sg entry if newly added */ 1358 if (num_elem > *pages_used) 1359 sg_mark_end(&to[num_elem - 1]); 1360 out: 1361 if (rc) 1362 iov_iter_revert(from, size - *size_used); 1363 *size_used = size; 1364 *pages_used = num_elem; 1365 1366 return rc; 1367 } 1368 1369 /* This function decrypts the input skb into either out_iov or in out_sg 1370 * or in skb buffers itself. The input parameter 'zc' indicates if 1371 * zero-copy mode needs to be tried or not. With zero-copy mode, either 1372 * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are 1373 * NULL, then the decryption happens inside skb buffers itself, i.e. 1374 * zero-copy gets disabled and 'zc' is updated. 1375 */ 1376 1377 static int decrypt_internal(struct sock *sk, struct sk_buff *skb, 1378 struct iov_iter *out_iov, 1379 struct scatterlist *out_sg, 1380 int *chunk, bool *zc, bool async) 1381 { 1382 struct tls_context *tls_ctx = tls_get_ctx(sk); 1383 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 1384 struct tls_prot_info *prot = &tls_ctx->prot_info; 1385 struct strp_msg *rxm = strp_msg(skb); 1386 int n_sgin, n_sgout, nsg, mem_size, aead_size, err, pages = 0; 1387 struct aead_request *aead_req; 1388 struct sk_buff *unused; 1389 u8 *aad, *iv, *mem = NULL; 1390 struct scatterlist *sgin = NULL; 1391 struct scatterlist *sgout = NULL; 1392 const int data_len = rxm->full_len - prot->overhead_size + 1393 prot->tail_size; 1394 int iv_offset = 0; 1395 1396 if (*zc && (out_iov || out_sg)) { 1397 if (out_iov) 1398 n_sgout = iov_iter_npages(out_iov, INT_MAX) + 1; 1399 else 1400 n_sgout = sg_nents(out_sg); 1401 n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size, 1402 rxm->full_len - prot->prepend_size); 1403 } else { 1404 n_sgout = 0; 1405 *zc = false; 1406 n_sgin = skb_cow_data(skb, 0, &unused); 1407 } 1408 1409 if (n_sgin < 1) 1410 return -EBADMSG; 1411 1412 /* Increment to accommodate AAD */ 1413 n_sgin = n_sgin + 1; 1414 1415 nsg = n_sgin + n_sgout; 1416 1417 aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv); 1418 mem_size = aead_size + (nsg * sizeof(struct scatterlist)); 1419 mem_size = mem_size + prot->aad_size; 1420 mem_size = mem_size + crypto_aead_ivsize(ctx->aead_recv); 1421 1422 /* Allocate a single block of memory which contains 1423 * aead_req || sgin[] || sgout[] || aad || iv. 1424 * This order achieves correct alignment for aead_req, sgin, sgout. 1425 */ 1426 mem = kmalloc(mem_size, sk->sk_allocation); 1427 if (!mem) 1428 return -ENOMEM; 1429 1430 /* Segment the allocated memory */ 1431 aead_req = (struct aead_request *)mem; 1432 sgin = (struct scatterlist *)(mem + aead_size); 1433 sgout = sgin + n_sgin; 1434 aad = (u8 *)(sgout + n_sgout); 1435 iv = aad + prot->aad_size; 1436 1437 /* For CCM based ciphers, first byte of nonce+iv is always '2' */ 1438 if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) { 1439 iv[0] = 2; 1440 iv_offset = 1; 1441 } 1442 1443 /* Prepare IV */ 1444 err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE, 1445 iv + iv_offset + prot->salt_size, 1446 prot->iv_size); 1447 if (err < 0) { 1448 kfree(mem); 1449 return err; 1450 } 1451 if (prot->version == TLS_1_3_VERSION) 1452 memcpy(iv + iv_offset, tls_ctx->rx.iv, 1453 crypto_aead_ivsize(ctx->aead_recv)); 1454 else 1455 memcpy(iv + iv_offset, tls_ctx->rx.iv, prot->salt_size); 1456 1457 xor_iv_with_seq(prot->version, iv, tls_ctx->rx.rec_seq); 1458 1459 /* Prepare AAD */ 1460 tls_make_aad(aad, rxm->full_len - prot->overhead_size + 1461 prot->tail_size, 1462 tls_ctx->rx.rec_seq, prot->rec_seq_size, 1463 ctx->control, prot->version); 1464 1465 /* Prepare sgin */ 1466 sg_init_table(sgin, n_sgin); 1467 sg_set_buf(&sgin[0], aad, prot->aad_size); 1468 err = skb_to_sgvec(skb, &sgin[1], 1469 rxm->offset + prot->prepend_size, 1470 rxm->full_len - prot->prepend_size); 1471 if (err < 0) { 1472 kfree(mem); 1473 return err; 1474 } 1475 1476 if (n_sgout) { 1477 if (out_iov) { 1478 sg_init_table(sgout, n_sgout); 1479 sg_set_buf(&sgout[0], aad, prot->aad_size); 1480 1481 *chunk = 0; 1482 err = tls_setup_from_iter(sk, out_iov, data_len, 1483 &pages, chunk, &sgout[1], 1484 (n_sgout - 1)); 1485 if (err < 0) 1486 goto fallback_to_reg_recv; 1487 } else if (out_sg) { 1488 memcpy(sgout, out_sg, n_sgout * sizeof(*sgout)); 1489 } else { 1490 goto fallback_to_reg_recv; 1491 } 1492 } else { 1493 fallback_to_reg_recv: 1494 sgout = sgin; 1495 pages = 0; 1496 *chunk = data_len; 1497 *zc = false; 1498 } 1499 1500 /* Prepare and submit AEAD request */ 1501 err = tls_do_decryption(sk, skb, sgin, sgout, iv, 1502 data_len, aead_req, async); 1503 if (err == -EINPROGRESS) 1504 return err; 1505 1506 /* Release the pages in case iov was mapped to pages */ 1507 for (; pages > 0; pages--) 1508 put_page(sg_page(&sgout[pages])); 1509 1510 kfree(mem); 1511 return err; 1512 } 1513 1514 static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb, 1515 struct iov_iter *dest, int *chunk, bool *zc, 1516 bool async) 1517 { 1518 struct tls_context *tls_ctx = tls_get_ctx(sk); 1519 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 1520 struct tls_prot_info *prot = &tls_ctx->prot_info; 1521 struct strp_msg *rxm = strp_msg(skb); 1522 int pad, err = 0; 1523 1524 if (!ctx->decrypted) { 1525 if (tls_ctx->rx_conf == TLS_HW) { 1526 err = tls_device_decrypted(sk, tls_ctx, skb, rxm); 1527 if (err < 0) 1528 return err; 1529 } 1530 1531 /* Still not decrypted after tls_device */ 1532 if (!ctx->decrypted) { 1533 err = decrypt_internal(sk, skb, dest, NULL, chunk, zc, 1534 async); 1535 if (err < 0) { 1536 if (err == -EINPROGRESS) 1537 tls_advance_record_sn(sk, prot, 1538 &tls_ctx->rx); 1539 else if (err == -EBADMSG) 1540 TLS_INC_STATS(sock_net(sk), 1541 LINUX_MIB_TLSDECRYPTERROR); 1542 return err; 1543 } 1544 } else { 1545 *zc = false; 1546 } 1547 1548 pad = padding_length(ctx, prot, skb); 1549 if (pad < 0) 1550 return pad; 1551 1552 rxm->full_len -= pad; 1553 rxm->offset += prot->prepend_size; 1554 rxm->full_len -= prot->overhead_size; 1555 tls_advance_record_sn(sk, prot, &tls_ctx->rx); 1556 ctx->decrypted = 1; 1557 ctx->saved_data_ready(sk); 1558 } else { 1559 *zc = false; 1560 } 1561 1562 return err; 1563 } 1564 1565 int decrypt_skb(struct sock *sk, struct sk_buff *skb, 1566 struct scatterlist *sgout) 1567 { 1568 bool zc = true; 1569 int chunk; 1570 1571 return decrypt_internal(sk, skb, NULL, sgout, &chunk, &zc, false); 1572 } 1573 1574 static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb, 1575 unsigned int len) 1576 { 1577 struct tls_context *tls_ctx = tls_get_ctx(sk); 1578 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 1579 1580 if (skb) { 1581 struct strp_msg *rxm = strp_msg(skb); 1582 1583 if (len < rxm->full_len) { 1584 rxm->offset += len; 1585 rxm->full_len -= len; 1586 return false; 1587 } 1588 consume_skb(skb); 1589 } 1590 1591 /* Finished with message */ 1592 ctx->recv_pkt = NULL; 1593 __strp_unpause(&ctx->strp); 1594 1595 return true; 1596 } 1597 1598 /* This function traverses the rx_list in tls receive context to copies the 1599 * decrypted records into the buffer provided by caller zero copy is not 1600 * true. Further, the records are removed from the rx_list if it is not a peek 1601 * case and the record has been consumed completely. 1602 */ 1603 static int process_rx_list(struct tls_sw_context_rx *ctx, 1604 struct msghdr *msg, 1605 u8 *control, 1606 bool *cmsg, 1607 size_t skip, 1608 size_t len, 1609 bool zc, 1610 bool is_peek) 1611 { 1612 struct sk_buff *skb = skb_peek(&ctx->rx_list); 1613 u8 ctrl = *control; 1614 u8 msgc = *cmsg; 1615 struct tls_msg *tlm; 1616 ssize_t copied = 0; 1617 1618 /* Set the record type in 'control' if caller didn't pass it */ 1619 if (!ctrl && skb) { 1620 tlm = tls_msg(skb); 1621 ctrl = tlm->control; 1622 } 1623 1624 while (skip && skb) { 1625 struct strp_msg *rxm = strp_msg(skb); 1626 tlm = tls_msg(skb); 1627 1628 /* Cannot process a record of different type */ 1629 if (ctrl != tlm->control) 1630 return 0; 1631 1632 if (skip < rxm->full_len) 1633 break; 1634 1635 skip = skip - rxm->full_len; 1636 skb = skb_peek_next(skb, &ctx->rx_list); 1637 } 1638 1639 while (len && skb) { 1640 struct sk_buff *next_skb; 1641 struct strp_msg *rxm = strp_msg(skb); 1642 int chunk = min_t(unsigned int, rxm->full_len - skip, len); 1643 1644 tlm = tls_msg(skb); 1645 1646 /* Cannot process a record of different type */ 1647 if (ctrl != tlm->control) 1648 return 0; 1649 1650 /* Set record type if not already done. For a non-data record, 1651 * do not proceed if record type could not be copied. 1652 */ 1653 if (!msgc) { 1654 int cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE, 1655 sizeof(ctrl), &ctrl); 1656 msgc = true; 1657 if (ctrl != TLS_RECORD_TYPE_DATA) { 1658 if (cerr || msg->msg_flags & MSG_CTRUNC) 1659 return -EIO; 1660 1661 *cmsg = msgc; 1662 } 1663 } 1664 1665 if (!zc || (rxm->full_len - skip) > len) { 1666 int err = skb_copy_datagram_msg(skb, rxm->offset + skip, 1667 msg, chunk); 1668 if (err < 0) 1669 return err; 1670 } 1671 1672 len = len - chunk; 1673 copied = copied + chunk; 1674 1675 /* Consume the data from record if it is non-peek case*/ 1676 if (!is_peek) { 1677 rxm->offset = rxm->offset + chunk; 1678 rxm->full_len = rxm->full_len - chunk; 1679 1680 /* Return if there is unconsumed data in the record */ 1681 if (rxm->full_len - skip) 1682 break; 1683 } 1684 1685 /* The remaining skip-bytes must lie in 1st record in rx_list. 1686 * So from the 2nd record, 'skip' should be 0. 1687 */ 1688 skip = 0; 1689 1690 if (msg) 1691 msg->msg_flags |= MSG_EOR; 1692 1693 next_skb = skb_peek_next(skb, &ctx->rx_list); 1694 1695 if (!is_peek) { 1696 skb_unlink(skb, &ctx->rx_list); 1697 consume_skb(skb); 1698 } 1699 1700 skb = next_skb; 1701 } 1702 1703 *control = ctrl; 1704 return copied; 1705 } 1706 1707 int tls_sw_recvmsg(struct sock *sk, 1708 struct msghdr *msg, 1709 size_t len, 1710 int nonblock, 1711 int flags, 1712 int *addr_len) 1713 { 1714 struct tls_context *tls_ctx = tls_get_ctx(sk); 1715 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 1716 struct tls_prot_info *prot = &tls_ctx->prot_info; 1717 struct sk_psock *psock; 1718 unsigned char control = 0; 1719 ssize_t decrypted = 0; 1720 struct strp_msg *rxm; 1721 struct tls_msg *tlm; 1722 struct sk_buff *skb; 1723 ssize_t copied = 0; 1724 bool cmsg = false; 1725 int target, err = 0; 1726 long timeo; 1727 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter); 1728 bool is_peek = flags & MSG_PEEK; 1729 int num_async = 0; 1730 1731 flags |= nonblock; 1732 1733 if (unlikely(flags & MSG_ERRQUEUE)) 1734 return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR); 1735 1736 psock = sk_psock_get(sk); 1737 lock_sock(sk); 1738 1739 /* Process pending decrypted records. It must be non-zero-copy */ 1740 err = process_rx_list(ctx, msg, &control, &cmsg, 0, len, false, 1741 is_peek); 1742 if (err < 0) { 1743 tls_err_abort(sk, err); 1744 goto end; 1745 } else { 1746 copied = err; 1747 } 1748 1749 if (len <= copied) 1750 goto recv_end; 1751 1752 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); 1753 len = len - copied; 1754 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1755 1756 while (len && (decrypted + copied < target || ctx->recv_pkt)) { 1757 bool retain_skb = false; 1758 bool zc = false; 1759 int to_decrypt; 1760 int chunk = 0; 1761 bool async_capable; 1762 bool async = false; 1763 1764 skb = tls_wait_data(sk, psock, flags, timeo, &err); 1765 if (!skb) { 1766 if (psock) { 1767 int ret = __tcp_bpf_recvmsg(sk, psock, 1768 msg, len, flags); 1769 1770 if (ret > 0) { 1771 decrypted += ret; 1772 len -= ret; 1773 continue; 1774 } 1775 } 1776 goto recv_end; 1777 } else { 1778 tlm = tls_msg(skb); 1779 if (prot->version == TLS_1_3_VERSION) 1780 tlm->control = 0; 1781 else 1782 tlm->control = ctx->control; 1783 } 1784 1785 rxm = strp_msg(skb); 1786 1787 to_decrypt = rxm->full_len - prot->overhead_size; 1788 1789 if (to_decrypt <= len && !is_kvec && !is_peek && 1790 ctx->control == TLS_RECORD_TYPE_DATA && 1791 prot->version != TLS_1_3_VERSION) 1792 zc = true; 1793 1794 /* Do not use async mode if record is non-data */ 1795 if (ctx->control == TLS_RECORD_TYPE_DATA) 1796 async_capable = ctx->async_capable; 1797 else 1798 async_capable = false; 1799 1800 err = decrypt_skb_update(sk, skb, &msg->msg_iter, 1801 &chunk, &zc, async_capable); 1802 if (err < 0 && err != -EINPROGRESS) { 1803 tls_err_abort(sk, EBADMSG); 1804 goto recv_end; 1805 } 1806 1807 if (err == -EINPROGRESS) { 1808 async = true; 1809 num_async++; 1810 } else if (prot->version == TLS_1_3_VERSION) { 1811 tlm->control = ctx->control; 1812 } 1813 1814 /* If the type of records being processed is not known yet, 1815 * set it to record type just dequeued. If it is already known, 1816 * but does not match the record type just dequeued, go to end. 1817 * We always get record type here since for tls1.2, record type 1818 * is known just after record is dequeued from stream parser. 1819 * For tls1.3, we disable async. 1820 */ 1821 1822 if (!control) 1823 control = tlm->control; 1824 else if (control != tlm->control) 1825 goto recv_end; 1826 1827 if (!cmsg) { 1828 int cerr; 1829 1830 cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE, 1831 sizeof(control), &control); 1832 cmsg = true; 1833 if (control != TLS_RECORD_TYPE_DATA) { 1834 if (cerr || msg->msg_flags & MSG_CTRUNC) { 1835 err = -EIO; 1836 goto recv_end; 1837 } 1838 } 1839 } 1840 1841 if (async) 1842 goto pick_next_record; 1843 1844 if (!zc) { 1845 if (rxm->full_len > len) { 1846 retain_skb = true; 1847 chunk = len; 1848 } else { 1849 chunk = rxm->full_len; 1850 } 1851 1852 err = skb_copy_datagram_msg(skb, rxm->offset, 1853 msg, chunk); 1854 if (err < 0) 1855 goto recv_end; 1856 1857 if (!is_peek) { 1858 rxm->offset = rxm->offset + chunk; 1859 rxm->full_len = rxm->full_len - chunk; 1860 } 1861 } 1862 1863 pick_next_record: 1864 if (chunk > len) 1865 chunk = len; 1866 1867 decrypted += chunk; 1868 len -= chunk; 1869 1870 /* For async or peek case, queue the current skb */ 1871 if (async || is_peek || retain_skb) { 1872 skb_queue_tail(&ctx->rx_list, skb); 1873 skb = NULL; 1874 } 1875 1876 if (tls_sw_advance_skb(sk, skb, chunk)) { 1877 /* Return full control message to 1878 * userspace before trying to parse 1879 * another message type 1880 */ 1881 msg->msg_flags |= MSG_EOR; 1882 if (ctx->control != TLS_RECORD_TYPE_DATA) 1883 goto recv_end; 1884 } else { 1885 break; 1886 } 1887 } 1888 1889 recv_end: 1890 if (num_async) { 1891 /* Wait for all previously submitted records to be decrypted */ 1892 smp_store_mb(ctx->async_notify, true); 1893 if (atomic_read(&ctx->decrypt_pending)) { 1894 err = crypto_wait_req(-EINPROGRESS, &ctx->async_wait); 1895 if (err) { 1896 /* one of async decrypt failed */ 1897 tls_err_abort(sk, err); 1898 copied = 0; 1899 decrypted = 0; 1900 goto end; 1901 } 1902 } else { 1903 reinit_completion(&ctx->async_wait.completion); 1904 } 1905 WRITE_ONCE(ctx->async_notify, false); 1906 1907 /* Drain records from the rx_list & copy if required */ 1908 if (is_peek || is_kvec) 1909 err = process_rx_list(ctx, msg, &control, &cmsg, copied, 1910 decrypted, false, is_peek); 1911 else 1912 err = process_rx_list(ctx, msg, &control, &cmsg, 0, 1913 decrypted, true, is_peek); 1914 if (err < 0) { 1915 tls_err_abort(sk, err); 1916 copied = 0; 1917 goto end; 1918 } 1919 } 1920 1921 copied += decrypted; 1922 1923 end: 1924 release_sock(sk); 1925 if (psock) 1926 sk_psock_put(sk, psock); 1927 return copied ? : err; 1928 } 1929 1930 ssize_t tls_sw_splice_read(struct socket *sock, loff_t *ppos, 1931 struct pipe_inode_info *pipe, 1932 size_t len, unsigned int flags) 1933 { 1934 struct tls_context *tls_ctx = tls_get_ctx(sock->sk); 1935 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 1936 struct strp_msg *rxm = NULL; 1937 struct sock *sk = sock->sk; 1938 struct sk_buff *skb; 1939 ssize_t copied = 0; 1940 int err = 0; 1941 long timeo; 1942 int chunk; 1943 bool zc = false; 1944 1945 lock_sock(sk); 1946 1947 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1948 1949 skb = tls_wait_data(sk, NULL, flags, timeo, &err); 1950 if (!skb) 1951 goto splice_read_end; 1952 1953 if (!ctx->decrypted) { 1954 err = decrypt_skb_update(sk, skb, NULL, &chunk, &zc, false); 1955 1956 /* splice does not support reading control messages */ 1957 if (ctx->control != TLS_RECORD_TYPE_DATA) { 1958 err = -EINVAL; 1959 goto splice_read_end; 1960 } 1961 1962 if (err < 0) { 1963 tls_err_abort(sk, EBADMSG); 1964 goto splice_read_end; 1965 } 1966 ctx->decrypted = 1; 1967 } 1968 rxm = strp_msg(skb); 1969 1970 chunk = min_t(unsigned int, rxm->full_len, len); 1971 copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags); 1972 if (copied < 0) 1973 goto splice_read_end; 1974 1975 if (likely(!(flags & MSG_PEEK))) 1976 tls_sw_advance_skb(sk, skb, copied); 1977 1978 splice_read_end: 1979 release_sock(sk); 1980 return copied ? : err; 1981 } 1982 1983 bool tls_sw_stream_read(const struct sock *sk) 1984 { 1985 struct tls_context *tls_ctx = tls_get_ctx(sk); 1986 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 1987 bool ingress_empty = true; 1988 struct sk_psock *psock; 1989 1990 rcu_read_lock(); 1991 psock = sk_psock(sk); 1992 if (psock) 1993 ingress_empty = list_empty(&psock->ingress_msg); 1994 rcu_read_unlock(); 1995 1996 return !ingress_empty || ctx->recv_pkt || 1997 !skb_queue_empty(&ctx->rx_list); 1998 } 1999 2000 static int tls_read_size(struct strparser *strp, struct sk_buff *skb) 2001 { 2002 struct tls_context *tls_ctx = tls_get_ctx(strp->sk); 2003 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2004 struct tls_prot_info *prot = &tls_ctx->prot_info; 2005 char header[TLS_HEADER_SIZE + MAX_IV_SIZE]; 2006 struct strp_msg *rxm = strp_msg(skb); 2007 size_t cipher_overhead; 2008 size_t data_len = 0; 2009 int ret; 2010 2011 /* Verify that we have a full TLS header, or wait for more data */ 2012 if (rxm->offset + prot->prepend_size > skb->len) 2013 return 0; 2014 2015 /* Sanity-check size of on-stack buffer. */ 2016 if (WARN_ON(prot->prepend_size > sizeof(header))) { 2017 ret = -EINVAL; 2018 goto read_failure; 2019 } 2020 2021 /* Linearize header to local buffer */ 2022 ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size); 2023 2024 if (ret < 0) 2025 goto read_failure; 2026 2027 ctx->control = header[0]; 2028 2029 data_len = ((header[4] & 0xFF) | (header[3] << 8)); 2030 2031 cipher_overhead = prot->tag_size; 2032 if (prot->version != TLS_1_3_VERSION) 2033 cipher_overhead += prot->iv_size; 2034 2035 if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead + 2036 prot->tail_size) { 2037 ret = -EMSGSIZE; 2038 goto read_failure; 2039 } 2040 if (data_len < cipher_overhead) { 2041 ret = -EBADMSG; 2042 goto read_failure; 2043 } 2044 2045 /* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */ 2046 if (header[1] != TLS_1_2_VERSION_MINOR || 2047 header[2] != TLS_1_2_VERSION_MAJOR) { 2048 ret = -EINVAL; 2049 goto read_failure; 2050 } 2051 2052 tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE, 2053 TCP_SKB_CB(skb)->seq + rxm->offset); 2054 return data_len + TLS_HEADER_SIZE; 2055 2056 read_failure: 2057 tls_err_abort(strp->sk, ret); 2058 2059 return ret; 2060 } 2061 2062 static void tls_queue(struct strparser *strp, struct sk_buff *skb) 2063 { 2064 struct tls_context *tls_ctx = tls_get_ctx(strp->sk); 2065 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2066 2067 ctx->decrypted = 0; 2068 2069 ctx->recv_pkt = skb; 2070 strp_pause(strp); 2071 2072 ctx->saved_data_ready(strp->sk); 2073 } 2074 2075 static void tls_data_ready(struct sock *sk) 2076 { 2077 struct tls_context *tls_ctx = tls_get_ctx(sk); 2078 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2079 struct sk_psock *psock; 2080 2081 strp_data_ready(&ctx->strp); 2082 2083 psock = sk_psock_get(sk); 2084 if (psock && !list_empty(&psock->ingress_msg)) { 2085 ctx->saved_data_ready(sk); 2086 sk_psock_put(sk, psock); 2087 } 2088 } 2089 2090 void tls_sw_cancel_work_tx(struct tls_context *tls_ctx) 2091 { 2092 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 2093 2094 set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask); 2095 set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask); 2096 cancel_delayed_work_sync(&ctx->tx_work.work); 2097 } 2098 2099 void tls_sw_release_resources_tx(struct sock *sk) 2100 { 2101 struct tls_context *tls_ctx = tls_get_ctx(sk); 2102 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 2103 struct tls_rec *rec, *tmp; 2104 2105 /* Wait for any pending async encryptions to complete */ 2106 smp_store_mb(ctx->async_notify, true); 2107 if (atomic_read(&ctx->encrypt_pending)) 2108 crypto_wait_req(-EINPROGRESS, &ctx->async_wait); 2109 2110 tls_tx_records(sk, -1); 2111 2112 /* Free up un-sent records in tx_list. First, free 2113 * the partially sent record if any at head of tx_list. 2114 */ 2115 if (tls_ctx->partially_sent_record) { 2116 tls_free_partial_record(sk, tls_ctx); 2117 rec = list_first_entry(&ctx->tx_list, 2118 struct tls_rec, list); 2119 list_del(&rec->list); 2120 sk_msg_free(sk, &rec->msg_plaintext); 2121 kfree(rec); 2122 } 2123 2124 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) { 2125 list_del(&rec->list); 2126 sk_msg_free(sk, &rec->msg_encrypted); 2127 sk_msg_free(sk, &rec->msg_plaintext); 2128 kfree(rec); 2129 } 2130 2131 crypto_free_aead(ctx->aead_send); 2132 tls_free_open_rec(sk); 2133 } 2134 2135 void tls_sw_free_ctx_tx(struct tls_context *tls_ctx) 2136 { 2137 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 2138 2139 kfree(ctx); 2140 } 2141 2142 void tls_sw_release_resources_rx(struct sock *sk) 2143 { 2144 struct tls_context *tls_ctx = tls_get_ctx(sk); 2145 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2146 2147 kfree(tls_ctx->rx.rec_seq); 2148 kfree(tls_ctx->rx.iv); 2149 2150 if (ctx->aead_recv) { 2151 kfree_skb(ctx->recv_pkt); 2152 ctx->recv_pkt = NULL; 2153 skb_queue_purge(&ctx->rx_list); 2154 crypto_free_aead(ctx->aead_recv); 2155 strp_stop(&ctx->strp); 2156 /* If tls_sw_strparser_arm() was not called (cleanup paths) 2157 * we still want to strp_stop(), but sk->sk_data_ready was 2158 * never swapped. 2159 */ 2160 if (ctx->saved_data_ready) { 2161 write_lock_bh(&sk->sk_callback_lock); 2162 sk->sk_data_ready = ctx->saved_data_ready; 2163 write_unlock_bh(&sk->sk_callback_lock); 2164 } 2165 } 2166 } 2167 2168 void tls_sw_strparser_done(struct tls_context *tls_ctx) 2169 { 2170 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2171 2172 strp_done(&ctx->strp); 2173 } 2174 2175 void tls_sw_free_ctx_rx(struct tls_context *tls_ctx) 2176 { 2177 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2178 2179 kfree(ctx); 2180 } 2181 2182 void tls_sw_free_resources_rx(struct sock *sk) 2183 { 2184 struct tls_context *tls_ctx = tls_get_ctx(sk); 2185 2186 tls_sw_release_resources_rx(sk); 2187 tls_sw_free_ctx_rx(tls_ctx); 2188 } 2189 2190 /* The work handler to transmitt the encrypted records in tx_list */ 2191 static void tx_work_handler(struct work_struct *work) 2192 { 2193 struct delayed_work *delayed_work = to_delayed_work(work); 2194 struct tx_work *tx_work = container_of(delayed_work, 2195 struct tx_work, work); 2196 struct sock *sk = tx_work->sk; 2197 struct tls_context *tls_ctx = tls_get_ctx(sk); 2198 struct tls_sw_context_tx *ctx; 2199 2200 if (unlikely(!tls_ctx)) 2201 return; 2202 2203 ctx = tls_sw_ctx_tx(tls_ctx); 2204 if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask)) 2205 return; 2206 2207 if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) 2208 return; 2209 mutex_lock(&tls_ctx->tx_lock); 2210 lock_sock(sk); 2211 tls_tx_records(sk, -1); 2212 release_sock(sk); 2213 mutex_unlock(&tls_ctx->tx_lock); 2214 } 2215 2216 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx) 2217 { 2218 struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx); 2219 2220 /* Schedule the transmission if tx list is ready */ 2221 if (is_tx_ready(tx_ctx) && 2222 !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask)) 2223 schedule_delayed_work(&tx_ctx->tx_work.work, 0); 2224 } 2225 2226 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx) 2227 { 2228 struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx); 2229 2230 write_lock_bh(&sk->sk_callback_lock); 2231 rx_ctx->saved_data_ready = sk->sk_data_ready; 2232 sk->sk_data_ready = tls_data_ready; 2233 write_unlock_bh(&sk->sk_callback_lock); 2234 2235 strp_check_rcv(&rx_ctx->strp); 2236 } 2237 2238 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx) 2239 { 2240 struct tls_context *tls_ctx = tls_get_ctx(sk); 2241 struct tls_prot_info *prot = &tls_ctx->prot_info; 2242 struct tls_crypto_info *crypto_info; 2243 struct tls12_crypto_info_aes_gcm_128 *gcm_128_info; 2244 struct tls12_crypto_info_aes_gcm_256 *gcm_256_info; 2245 struct tls12_crypto_info_aes_ccm_128 *ccm_128_info; 2246 struct tls_sw_context_tx *sw_ctx_tx = NULL; 2247 struct tls_sw_context_rx *sw_ctx_rx = NULL; 2248 struct cipher_context *cctx; 2249 struct crypto_aead **aead; 2250 struct strp_callbacks cb; 2251 u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size; 2252 struct crypto_tfm *tfm; 2253 char *iv, *rec_seq, *key, *salt, *cipher_name; 2254 size_t keysize; 2255 int rc = 0; 2256 2257 if (!ctx) { 2258 rc = -EINVAL; 2259 goto out; 2260 } 2261 2262 if (tx) { 2263 if (!ctx->priv_ctx_tx) { 2264 sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL); 2265 if (!sw_ctx_tx) { 2266 rc = -ENOMEM; 2267 goto out; 2268 } 2269 ctx->priv_ctx_tx = sw_ctx_tx; 2270 } else { 2271 sw_ctx_tx = 2272 (struct tls_sw_context_tx *)ctx->priv_ctx_tx; 2273 } 2274 } else { 2275 if (!ctx->priv_ctx_rx) { 2276 sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL); 2277 if (!sw_ctx_rx) { 2278 rc = -ENOMEM; 2279 goto out; 2280 } 2281 ctx->priv_ctx_rx = sw_ctx_rx; 2282 } else { 2283 sw_ctx_rx = 2284 (struct tls_sw_context_rx *)ctx->priv_ctx_rx; 2285 } 2286 } 2287 2288 if (tx) { 2289 crypto_init_wait(&sw_ctx_tx->async_wait); 2290 crypto_info = &ctx->crypto_send.info; 2291 cctx = &ctx->tx; 2292 aead = &sw_ctx_tx->aead_send; 2293 INIT_LIST_HEAD(&sw_ctx_tx->tx_list); 2294 INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler); 2295 sw_ctx_tx->tx_work.sk = sk; 2296 } else { 2297 crypto_init_wait(&sw_ctx_rx->async_wait); 2298 crypto_info = &ctx->crypto_recv.info; 2299 cctx = &ctx->rx; 2300 skb_queue_head_init(&sw_ctx_rx->rx_list); 2301 aead = &sw_ctx_rx->aead_recv; 2302 } 2303 2304 switch (crypto_info->cipher_type) { 2305 case TLS_CIPHER_AES_GCM_128: { 2306 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE; 2307 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE; 2308 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE; 2309 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv; 2310 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE; 2311 rec_seq = 2312 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq; 2313 gcm_128_info = 2314 (struct tls12_crypto_info_aes_gcm_128 *)crypto_info; 2315 keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE; 2316 key = gcm_128_info->key; 2317 salt = gcm_128_info->salt; 2318 salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE; 2319 cipher_name = "gcm(aes)"; 2320 break; 2321 } 2322 case TLS_CIPHER_AES_GCM_256: { 2323 nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE; 2324 tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE; 2325 iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE; 2326 iv = ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->iv; 2327 rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE; 2328 rec_seq = 2329 ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->rec_seq; 2330 gcm_256_info = 2331 (struct tls12_crypto_info_aes_gcm_256 *)crypto_info; 2332 keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE; 2333 key = gcm_256_info->key; 2334 salt = gcm_256_info->salt; 2335 salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE; 2336 cipher_name = "gcm(aes)"; 2337 break; 2338 } 2339 case TLS_CIPHER_AES_CCM_128: { 2340 nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE; 2341 tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE; 2342 iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE; 2343 iv = ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->iv; 2344 rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE; 2345 rec_seq = 2346 ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->rec_seq; 2347 ccm_128_info = 2348 (struct tls12_crypto_info_aes_ccm_128 *)crypto_info; 2349 keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE; 2350 key = ccm_128_info->key; 2351 salt = ccm_128_info->salt; 2352 salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE; 2353 cipher_name = "ccm(aes)"; 2354 break; 2355 } 2356 default: 2357 rc = -EINVAL; 2358 goto free_priv; 2359 } 2360 2361 /* Sanity-check the sizes for stack allocations. */ 2362 if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE || 2363 rec_seq_size > TLS_MAX_REC_SEQ_SIZE) { 2364 rc = -EINVAL; 2365 goto free_priv; 2366 } 2367 2368 if (crypto_info->version == TLS_1_3_VERSION) { 2369 nonce_size = 0; 2370 prot->aad_size = TLS_HEADER_SIZE; 2371 prot->tail_size = 1; 2372 } else { 2373 prot->aad_size = TLS_AAD_SPACE_SIZE; 2374 prot->tail_size = 0; 2375 } 2376 2377 prot->version = crypto_info->version; 2378 prot->cipher_type = crypto_info->cipher_type; 2379 prot->prepend_size = TLS_HEADER_SIZE + nonce_size; 2380 prot->tag_size = tag_size; 2381 prot->overhead_size = prot->prepend_size + 2382 prot->tag_size + prot->tail_size; 2383 prot->iv_size = iv_size; 2384 prot->salt_size = salt_size; 2385 cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL); 2386 if (!cctx->iv) { 2387 rc = -ENOMEM; 2388 goto free_priv; 2389 } 2390 /* Note: 128 & 256 bit salt are the same size */ 2391 prot->rec_seq_size = rec_seq_size; 2392 memcpy(cctx->iv, salt, salt_size); 2393 memcpy(cctx->iv + salt_size, iv, iv_size); 2394 cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL); 2395 if (!cctx->rec_seq) { 2396 rc = -ENOMEM; 2397 goto free_iv; 2398 } 2399 2400 if (!*aead) { 2401 *aead = crypto_alloc_aead(cipher_name, 0, 0); 2402 if (IS_ERR(*aead)) { 2403 rc = PTR_ERR(*aead); 2404 *aead = NULL; 2405 goto free_rec_seq; 2406 } 2407 } 2408 2409 ctx->push_pending_record = tls_sw_push_pending_record; 2410 2411 rc = crypto_aead_setkey(*aead, key, keysize); 2412 2413 if (rc) 2414 goto free_aead; 2415 2416 rc = crypto_aead_setauthsize(*aead, prot->tag_size); 2417 if (rc) 2418 goto free_aead; 2419 2420 if (sw_ctx_rx) { 2421 tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv); 2422 2423 if (crypto_info->version == TLS_1_3_VERSION) 2424 sw_ctx_rx->async_capable = 0; 2425 else 2426 sw_ctx_rx->async_capable = 2427 !!(tfm->__crt_alg->cra_flags & 2428 CRYPTO_ALG_ASYNC); 2429 2430 /* Set up strparser */ 2431 memset(&cb, 0, sizeof(cb)); 2432 cb.rcv_msg = tls_queue; 2433 cb.parse_msg = tls_read_size; 2434 2435 strp_init(&sw_ctx_rx->strp, sk, &cb); 2436 } 2437 2438 goto out; 2439 2440 free_aead: 2441 crypto_free_aead(*aead); 2442 *aead = NULL; 2443 free_rec_seq: 2444 kfree(cctx->rec_seq); 2445 cctx->rec_seq = NULL; 2446 free_iv: 2447 kfree(cctx->iv); 2448 cctx->iv = NULL; 2449 free_priv: 2450 if (tx) { 2451 kfree(ctx->priv_ctx_tx); 2452 ctx->priv_ctx_tx = NULL; 2453 } else { 2454 kfree(ctx->priv_ctx_rx); 2455 ctx->priv_ctx_rx = NULL; 2456 } 2457 out: 2458 return rc; 2459 } 2460