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/bug.h> 39 #include <linux/sched/signal.h> 40 #include <linux/module.h> 41 #include <linux/splice.h> 42 #include <crypto/aead.h> 43 44 #include <net/strparser.h> 45 #include <net/tls.h> 46 47 struct tls_decrypt_arg { 48 bool zc; 49 bool async; 50 }; 51 52 noinline void tls_err_abort(struct sock *sk, int err) 53 { 54 WARN_ON_ONCE(err >= 0); 55 /* sk->sk_err should contain a positive error code. */ 56 sk->sk_err = -err; 57 sk_error_report(sk); 58 } 59 60 static int __skb_nsg(struct sk_buff *skb, int offset, int len, 61 unsigned int recursion_level) 62 { 63 int start = skb_headlen(skb); 64 int i, chunk = start - offset; 65 struct sk_buff *frag_iter; 66 int elt = 0; 67 68 if (unlikely(recursion_level >= 24)) 69 return -EMSGSIZE; 70 71 if (chunk > 0) { 72 if (chunk > len) 73 chunk = len; 74 elt++; 75 len -= chunk; 76 if (len == 0) 77 return elt; 78 offset += chunk; 79 } 80 81 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 82 int end; 83 84 WARN_ON(start > offset + len); 85 86 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 87 chunk = end - offset; 88 if (chunk > 0) { 89 if (chunk > len) 90 chunk = len; 91 elt++; 92 len -= chunk; 93 if (len == 0) 94 return elt; 95 offset += chunk; 96 } 97 start = end; 98 } 99 100 if (unlikely(skb_has_frag_list(skb))) { 101 skb_walk_frags(skb, frag_iter) { 102 int end, ret; 103 104 WARN_ON(start > offset + len); 105 106 end = start + frag_iter->len; 107 chunk = end - offset; 108 if (chunk > 0) { 109 if (chunk > len) 110 chunk = len; 111 ret = __skb_nsg(frag_iter, offset - start, chunk, 112 recursion_level + 1); 113 if (unlikely(ret < 0)) 114 return ret; 115 elt += ret; 116 len -= chunk; 117 if (len == 0) 118 return elt; 119 offset += chunk; 120 } 121 start = end; 122 } 123 } 124 BUG_ON(len); 125 return elt; 126 } 127 128 /* Return the number of scatterlist elements required to completely map the 129 * skb, or -EMSGSIZE if the recursion depth is exceeded. 130 */ 131 static int skb_nsg(struct sk_buff *skb, int offset, int len) 132 { 133 return __skb_nsg(skb, offset, len, 0); 134 } 135 136 static int padding_length(struct tls_prot_info *prot, struct sk_buff *skb) 137 { 138 struct strp_msg *rxm = strp_msg(skb); 139 struct tls_msg *tlm = tls_msg(skb); 140 int sub = 0; 141 142 /* Determine zero-padding length */ 143 if (prot->version == TLS_1_3_VERSION) { 144 int offset = rxm->full_len - TLS_TAG_SIZE - 1; 145 char content_type = 0; 146 int err; 147 148 while (content_type == 0) { 149 if (offset < prot->prepend_size) 150 return -EBADMSG; 151 err = skb_copy_bits(skb, rxm->offset + offset, 152 &content_type, 1); 153 if (err) 154 return err; 155 if (content_type) 156 break; 157 sub++; 158 offset--; 159 } 160 tlm->control = content_type; 161 } 162 return sub; 163 } 164 165 static void tls_decrypt_done(struct crypto_async_request *req, int err) 166 { 167 struct aead_request *aead_req = (struct aead_request *)req; 168 struct scatterlist *sgout = aead_req->dst; 169 struct scatterlist *sgin = aead_req->src; 170 struct tls_sw_context_rx *ctx; 171 struct tls_context *tls_ctx; 172 struct tls_prot_info *prot; 173 struct scatterlist *sg; 174 struct sk_buff *skb; 175 unsigned int pages; 176 177 skb = (struct sk_buff *)req->data; 178 tls_ctx = tls_get_ctx(skb->sk); 179 ctx = tls_sw_ctx_rx(tls_ctx); 180 prot = &tls_ctx->prot_info; 181 182 /* Propagate if there was an err */ 183 if (err) { 184 if (err == -EBADMSG) 185 TLS_INC_STATS(sock_net(skb->sk), 186 LINUX_MIB_TLSDECRYPTERROR); 187 ctx->async_wait.err = err; 188 tls_err_abort(skb->sk, err); 189 } else { 190 struct strp_msg *rxm = strp_msg(skb); 191 192 /* No TLS 1.3 support with async crypto */ 193 WARN_ON(prot->tail_size); 194 195 rxm->offset += prot->prepend_size; 196 rxm->full_len -= prot->overhead_size; 197 } 198 199 /* After using skb->sk to propagate sk through crypto async callback 200 * we need to NULL it again. 201 */ 202 skb->sk = NULL; 203 204 205 /* Free the destination pages if skb was not decrypted inplace */ 206 if (sgout != sgin) { 207 /* Skip the first S/G entry as it points to AAD */ 208 for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) { 209 if (!sg) 210 break; 211 put_page(sg_page(sg)); 212 } 213 } 214 215 kfree(aead_req); 216 217 spin_lock_bh(&ctx->decrypt_compl_lock); 218 if (!atomic_dec_return(&ctx->decrypt_pending)) 219 complete(&ctx->async_wait.completion); 220 spin_unlock_bh(&ctx->decrypt_compl_lock); 221 } 222 223 static int tls_do_decryption(struct sock *sk, 224 struct sk_buff *skb, 225 struct scatterlist *sgin, 226 struct scatterlist *sgout, 227 char *iv_recv, 228 size_t data_len, 229 struct aead_request *aead_req, 230 struct tls_decrypt_arg *darg) 231 { 232 struct tls_context *tls_ctx = tls_get_ctx(sk); 233 struct tls_prot_info *prot = &tls_ctx->prot_info; 234 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 235 int ret; 236 237 aead_request_set_tfm(aead_req, ctx->aead_recv); 238 aead_request_set_ad(aead_req, prot->aad_size); 239 aead_request_set_crypt(aead_req, sgin, sgout, 240 data_len + prot->tag_size, 241 (u8 *)iv_recv); 242 243 if (darg->async) { 244 /* Using skb->sk to push sk through to crypto async callback 245 * handler. This allows propagating errors up to the socket 246 * if needed. It _must_ be cleared in the async handler 247 * before consume_skb is called. We _know_ skb->sk is NULL 248 * because it is a clone from strparser. 249 */ 250 skb->sk = sk; 251 aead_request_set_callback(aead_req, 252 CRYPTO_TFM_REQ_MAY_BACKLOG, 253 tls_decrypt_done, skb); 254 atomic_inc(&ctx->decrypt_pending); 255 } else { 256 aead_request_set_callback(aead_req, 257 CRYPTO_TFM_REQ_MAY_BACKLOG, 258 crypto_req_done, &ctx->async_wait); 259 } 260 261 ret = crypto_aead_decrypt(aead_req); 262 if (ret == -EINPROGRESS) { 263 if (darg->async) 264 return 0; 265 266 ret = crypto_wait_req(ret, &ctx->async_wait); 267 } 268 darg->async = false; 269 270 if (ret == -EBADMSG) 271 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTERROR); 272 273 return ret; 274 } 275 276 static void tls_trim_both_msgs(struct sock *sk, int target_size) 277 { 278 struct tls_context *tls_ctx = tls_get_ctx(sk); 279 struct tls_prot_info *prot = &tls_ctx->prot_info; 280 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 281 struct tls_rec *rec = ctx->open_rec; 282 283 sk_msg_trim(sk, &rec->msg_plaintext, target_size); 284 if (target_size > 0) 285 target_size += prot->overhead_size; 286 sk_msg_trim(sk, &rec->msg_encrypted, target_size); 287 } 288 289 static int tls_alloc_encrypted_msg(struct sock *sk, int len) 290 { 291 struct tls_context *tls_ctx = tls_get_ctx(sk); 292 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 293 struct tls_rec *rec = ctx->open_rec; 294 struct sk_msg *msg_en = &rec->msg_encrypted; 295 296 return sk_msg_alloc(sk, msg_en, len, 0); 297 } 298 299 static int tls_clone_plaintext_msg(struct sock *sk, int required) 300 { 301 struct tls_context *tls_ctx = tls_get_ctx(sk); 302 struct tls_prot_info *prot = &tls_ctx->prot_info; 303 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 304 struct tls_rec *rec = ctx->open_rec; 305 struct sk_msg *msg_pl = &rec->msg_plaintext; 306 struct sk_msg *msg_en = &rec->msg_encrypted; 307 int skip, len; 308 309 /* We add page references worth len bytes from encrypted sg 310 * at the end of plaintext sg. It is guaranteed that msg_en 311 * has enough required room (ensured by caller). 312 */ 313 len = required - msg_pl->sg.size; 314 315 /* Skip initial bytes in msg_en's data to be able to use 316 * same offset of both plain and encrypted data. 317 */ 318 skip = prot->prepend_size + msg_pl->sg.size; 319 320 return sk_msg_clone(sk, msg_pl, msg_en, skip, len); 321 } 322 323 static struct tls_rec *tls_get_rec(struct sock *sk) 324 { 325 struct tls_context *tls_ctx = tls_get_ctx(sk); 326 struct tls_prot_info *prot = &tls_ctx->prot_info; 327 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 328 struct sk_msg *msg_pl, *msg_en; 329 struct tls_rec *rec; 330 int mem_size; 331 332 mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send); 333 334 rec = kzalloc(mem_size, sk->sk_allocation); 335 if (!rec) 336 return NULL; 337 338 msg_pl = &rec->msg_plaintext; 339 msg_en = &rec->msg_encrypted; 340 341 sk_msg_init(msg_pl); 342 sk_msg_init(msg_en); 343 344 sg_init_table(rec->sg_aead_in, 2); 345 sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size); 346 sg_unmark_end(&rec->sg_aead_in[1]); 347 348 sg_init_table(rec->sg_aead_out, 2); 349 sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size); 350 sg_unmark_end(&rec->sg_aead_out[1]); 351 352 return rec; 353 } 354 355 static void tls_free_rec(struct sock *sk, struct tls_rec *rec) 356 { 357 sk_msg_free(sk, &rec->msg_encrypted); 358 sk_msg_free(sk, &rec->msg_plaintext); 359 kfree(rec); 360 } 361 362 static void tls_free_open_rec(struct sock *sk) 363 { 364 struct tls_context *tls_ctx = tls_get_ctx(sk); 365 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 366 struct tls_rec *rec = ctx->open_rec; 367 368 if (rec) { 369 tls_free_rec(sk, rec); 370 ctx->open_rec = NULL; 371 } 372 } 373 374 int tls_tx_records(struct sock *sk, int flags) 375 { 376 struct tls_context *tls_ctx = tls_get_ctx(sk); 377 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 378 struct tls_rec *rec, *tmp; 379 struct sk_msg *msg_en; 380 int tx_flags, rc = 0; 381 382 if (tls_is_partially_sent_record(tls_ctx)) { 383 rec = list_first_entry(&ctx->tx_list, 384 struct tls_rec, list); 385 386 if (flags == -1) 387 tx_flags = rec->tx_flags; 388 else 389 tx_flags = flags; 390 391 rc = tls_push_partial_record(sk, tls_ctx, tx_flags); 392 if (rc) 393 goto tx_err; 394 395 /* Full record has been transmitted. 396 * Remove the head of tx_list 397 */ 398 list_del(&rec->list); 399 sk_msg_free(sk, &rec->msg_plaintext); 400 kfree(rec); 401 } 402 403 /* Tx all ready records */ 404 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) { 405 if (READ_ONCE(rec->tx_ready)) { 406 if (flags == -1) 407 tx_flags = rec->tx_flags; 408 else 409 tx_flags = flags; 410 411 msg_en = &rec->msg_encrypted; 412 rc = tls_push_sg(sk, tls_ctx, 413 &msg_en->sg.data[msg_en->sg.curr], 414 0, tx_flags); 415 if (rc) 416 goto tx_err; 417 418 list_del(&rec->list); 419 sk_msg_free(sk, &rec->msg_plaintext); 420 kfree(rec); 421 } else { 422 break; 423 } 424 } 425 426 tx_err: 427 if (rc < 0 && rc != -EAGAIN) 428 tls_err_abort(sk, -EBADMSG); 429 430 return rc; 431 } 432 433 static void tls_encrypt_done(struct crypto_async_request *req, int err) 434 { 435 struct aead_request *aead_req = (struct aead_request *)req; 436 struct sock *sk = req->data; 437 struct tls_context *tls_ctx = tls_get_ctx(sk); 438 struct tls_prot_info *prot = &tls_ctx->prot_info; 439 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 440 struct scatterlist *sge; 441 struct sk_msg *msg_en; 442 struct tls_rec *rec; 443 bool ready = false; 444 int pending; 445 446 rec = container_of(aead_req, struct tls_rec, aead_req); 447 msg_en = &rec->msg_encrypted; 448 449 sge = sk_msg_elem(msg_en, msg_en->sg.curr); 450 sge->offset -= prot->prepend_size; 451 sge->length += prot->prepend_size; 452 453 /* Check if error is previously set on socket */ 454 if (err || sk->sk_err) { 455 rec = NULL; 456 457 /* If err is already set on socket, return the same code */ 458 if (sk->sk_err) { 459 ctx->async_wait.err = -sk->sk_err; 460 } else { 461 ctx->async_wait.err = err; 462 tls_err_abort(sk, err); 463 } 464 } 465 466 if (rec) { 467 struct tls_rec *first_rec; 468 469 /* Mark the record as ready for transmission */ 470 smp_store_mb(rec->tx_ready, true); 471 472 /* If received record is at head of tx_list, schedule tx */ 473 first_rec = list_first_entry(&ctx->tx_list, 474 struct tls_rec, list); 475 if (rec == first_rec) 476 ready = true; 477 } 478 479 spin_lock_bh(&ctx->encrypt_compl_lock); 480 pending = atomic_dec_return(&ctx->encrypt_pending); 481 482 if (!pending && ctx->async_notify) 483 complete(&ctx->async_wait.completion); 484 spin_unlock_bh(&ctx->encrypt_compl_lock); 485 486 if (!ready) 487 return; 488 489 /* Schedule the transmission */ 490 if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) 491 schedule_delayed_work(&ctx->tx_work.work, 1); 492 } 493 494 static int tls_do_encryption(struct sock *sk, 495 struct tls_context *tls_ctx, 496 struct tls_sw_context_tx *ctx, 497 struct aead_request *aead_req, 498 size_t data_len, u32 start) 499 { 500 struct tls_prot_info *prot = &tls_ctx->prot_info; 501 struct tls_rec *rec = ctx->open_rec; 502 struct sk_msg *msg_en = &rec->msg_encrypted; 503 struct scatterlist *sge = sk_msg_elem(msg_en, start); 504 int rc, iv_offset = 0; 505 506 /* For CCM based ciphers, first byte of IV is a constant */ 507 switch (prot->cipher_type) { 508 case TLS_CIPHER_AES_CCM_128: 509 rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE; 510 iv_offset = 1; 511 break; 512 case TLS_CIPHER_SM4_CCM: 513 rec->iv_data[0] = TLS_SM4_CCM_IV_B0_BYTE; 514 iv_offset = 1; 515 break; 516 } 517 518 memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv, 519 prot->iv_size + prot->salt_size); 520 521 xor_iv_with_seq(prot, rec->iv_data + iv_offset, tls_ctx->tx.rec_seq); 522 523 sge->offset += prot->prepend_size; 524 sge->length -= prot->prepend_size; 525 526 msg_en->sg.curr = start; 527 528 aead_request_set_tfm(aead_req, ctx->aead_send); 529 aead_request_set_ad(aead_req, prot->aad_size); 530 aead_request_set_crypt(aead_req, rec->sg_aead_in, 531 rec->sg_aead_out, 532 data_len, rec->iv_data); 533 534 aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG, 535 tls_encrypt_done, sk); 536 537 /* Add the record in tx_list */ 538 list_add_tail((struct list_head *)&rec->list, &ctx->tx_list); 539 atomic_inc(&ctx->encrypt_pending); 540 541 rc = crypto_aead_encrypt(aead_req); 542 if (!rc || rc != -EINPROGRESS) { 543 atomic_dec(&ctx->encrypt_pending); 544 sge->offset -= prot->prepend_size; 545 sge->length += prot->prepend_size; 546 } 547 548 if (!rc) { 549 WRITE_ONCE(rec->tx_ready, true); 550 } else if (rc != -EINPROGRESS) { 551 list_del(&rec->list); 552 return rc; 553 } 554 555 /* Unhook the record from context if encryption is not failure */ 556 ctx->open_rec = NULL; 557 tls_advance_record_sn(sk, prot, &tls_ctx->tx); 558 return rc; 559 } 560 561 static int tls_split_open_record(struct sock *sk, struct tls_rec *from, 562 struct tls_rec **to, struct sk_msg *msg_opl, 563 struct sk_msg *msg_oen, u32 split_point, 564 u32 tx_overhead_size, u32 *orig_end) 565 { 566 u32 i, j, bytes = 0, apply = msg_opl->apply_bytes; 567 struct scatterlist *sge, *osge, *nsge; 568 u32 orig_size = msg_opl->sg.size; 569 struct scatterlist tmp = { }; 570 struct sk_msg *msg_npl; 571 struct tls_rec *new; 572 int ret; 573 574 new = tls_get_rec(sk); 575 if (!new) 576 return -ENOMEM; 577 ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size + 578 tx_overhead_size, 0); 579 if (ret < 0) { 580 tls_free_rec(sk, new); 581 return ret; 582 } 583 584 *orig_end = msg_opl->sg.end; 585 i = msg_opl->sg.start; 586 sge = sk_msg_elem(msg_opl, i); 587 while (apply && sge->length) { 588 if (sge->length > apply) { 589 u32 len = sge->length - apply; 590 591 get_page(sg_page(sge)); 592 sg_set_page(&tmp, sg_page(sge), len, 593 sge->offset + apply); 594 sge->length = apply; 595 bytes += apply; 596 apply = 0; 597 } else { 598 apply -= sge->length; 599 bytes += sge->length; 600 } 601 602 sk_msg_iter_var_next(i); 603 if (i == msg_opl->sg.end) 604 break; 605 sge = sk_msg_elem(msg_opl, i); 606 } 607 608 msg_opl->sg.end = i; 609 msg_opl->sg.curr = i; 610 msg_opl->sg.copybreak = 0; 611 msg_opl->apply_bytes = 0; 612 msg_opl->sg.size = bytes; 613 614 msg_npl = &new->msg_plaintext; 615 msg_npl->apply_bytes = apply; 616 msg_npl->sg.size = orig_size - bytes; 617 618 j = msg_npl->sg.start; 619 nsge = sk_msg_elem(msg_npl, j); 620 if (tmp.length) { 621 memcpy(nsge, &tmp, sizeof(*nsge)); 622 sk_msg_iter_var_next(j); 623 nsge = sk_msg_elem(msg_npl, j); 624 } 625 626 osge = sk_msg_elem(msg_opl, i); 627 while (osge->length) { 628 memcpy(nsge, osge, sizeof(*nsge)); 629 sg_unmark_end(nsge); 630 sk_msg_iter_var_next(i); 631 sk_msg_iter_var_next(j); 632 if (i == *orig_end) 633 break; 634 osge = sk_msg_elem(msg_opl, i); 635 nsge = sk_msg_elem(msg_npl, j); 636 } 637 638 msg_npl->sg.end = j; 639 msg_npl->sg.curr = j; 640 msg_npl->sg.copybreak = 0; 641 642 *to = new; 643 return 0; 644 } 645 646 static void tls_merge_open_record(struct sock *sk, struct tls_rec *to, 647 struct tls_rec *from, u32 orig_end) 648 { 649 struct sk_msg *msg_npl = &from->msg_plaintext; 650 struct sk_msg *msg_opl = &to->msg_plaintext; 651 struct scatterlist *osge, *nsge; 652 u32 i, j; 653 654 i = msg_opl->sg.end; 655 sk_msg_iter_var_prev(i); 656 j = msg_npl->sg.start; 657 658 osge = sk_msg_elem(msg_opl, i); 659 nsge = sk_msg_elem(msg_npl, j); 660 661 if (sg_page(osge) == sg_page(nsge) && 662 osge->offset + osge->length == nsge->offset) { 663 osge->length += nsge->length; 664 put_page(sg_page(nsge)); 665 } 666 667 msg_opl->sg.end = orig_end; 668 msg_opl->sg.curr = orig_end; 669 msg_opl->sg.copybreak = 0; 670 msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size; 671 msg_opl->sg.size += msg_npl->sg.size; 672 673 sk_msg_free(sk, &to->msg_encrypted); 674 sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted); 675 676 kfree(from); 677 } 678 679 static int tls_push_record(struct sock *sk, int flags, 680 unsigned char record_type) 681 { 682 struct tls_context *tls_ctx = tls_get_ctx(sk); 683 struct tls_prot_info *prot = &tls_ctx->prot_info; 684 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 685 struct tls_rec *rec = ctx->open_rec, *tmp = NULL; 686 u32 i, split_point, orig_end; 687 struct sk_msg *msg_pl, *msg_en; 688 struct aead_request *req; 689 bool split; 690 int rc; 691 692 if (!rec) 693 return 0; 694 695 msg_pl = &rec->msg_plaintext; 696 msg_en = &rec->msg_encrypted; 697 698 split_point = msg_pl->apply_bytes; 699 split = split_point && split_point < msg_pl->sg.size; 700 if (unlikely((!split && 701 msg_pl->sg.size + 702 prot->overhead_size > msg_en->sg.size) || 703 (split && 704 split_point + 705 prot->overhead_size > msg_en->sg.size))) { 706 split = true; 707 split_point = msg_en->sg.size; 708 } 709 if (split) { 710 rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en, 711 split_point, prot->overhead_size, 712 &orig_end); 713 if (rc < 0) 714 return rc; 715 /* This can happen if above tls_split_open_record allocates 716 * a single large encryption buffer instead of two smaller 717 * ones. In this case adjust pointers and continue without 718 * split. 719 */ 720 if (!msg_pl->sg.size) { 721 tls_merge_open_record(sk, rec, tmp, orig_end); 722 msg_pl = &rec->msg_plaintext; 723 msg_en = &rec->msg_encrypted; 724 split = false; 725 } 726 sk_msg_trim(sk, msg_en, msg_pl->sg.size + 727 prot->overhead_size); 728 } 729 730 rec->tx_flags = flags; 731 req = &rec->aead_req; 732 733 i = msg_pl->sg.end; 734 sk_msg_iter_var_prev(i); 735 736 rec->content_type = record_type; 737 if (prot->version == TLS_1_3_VERSION) { 738 /* Add content type to end of message. No padding added */ 739 sg_set_buf(&rec->sg_content_type, &rec->content_type, 1); 740 sg_mark_end(&rec->sg_content_type); 741 sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1, 742 &rec->sg_content_type); 743 } else { 744 sg_mark_end(sk_msg_elem(msg_pl, i)); 745 } 746 747 if (msg_pl->sg.end < msg_pl->sg.start) { 748 sg_chain(&msg_pl->sg.data[msg_pl->sg.start], 749 MAX_SKB_FRAGS - msg_pl->sg.start + 1, 750 msg_pl->sg.data); 751 } 752 753 i = msg_pl->sg.start; 754 sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]); 755 756 i = msg_en->sg.end; 757 sk_msg_iter_var_prev(i); 758 sg_mark_end(sk_msg_elem(msg_en, i)); 759 760 i = msg_en->sg.start; 761 sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]); 762 763 tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size, 764 tls_ctx->tx.rec_seq, record_type, prot); 765 766 tls_fill_prepend(tls_ctx, 767 page_address(sg_page(&msg_en->sg.data[i])) + 768 msg_en->sg.data[i].offset, 769 msg_pl->sg.size + prot->tail_size, 770 record_type); 771 772 tls_ctx->pending_open_record_frags = false; 773 774 rc = tls_do_encryption(sk, tls_ctx, ctx, req, 775 msg_pl->sg.size + prot->tail_size, i); 776 if (rc < 0) { 777 if (rc != -EINPROGRESS) { 778 tls_err_abort(sk, -EBADMSG); 779 if (split) { 780 tls_ctx->pending_open_record_frags = true; 781 tls_merge_open_record(sk, rec, tmp, orig_end); 782 } 783 } 784 ctx->async_capable = 1; 785 return rc; 786 } else if (split) { 787 msg_pl = &tmp->msg_plaintext; 788 msg_en = &tmp->msg_encrypted; 789 sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size); 790 tls_ctx->pending_open_record_frags = true; 791 ctx->open_rec = tmp; 792 } 793 794 return tls_tx_records(sk, flags); 795 } 796 797 static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk, 798 bool full_record, u8 record_type, 799 ssize_t *copied, int flags) 800 { 801 struct tls_context *tls_ctx = tls_get_ctx(sk); 802 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 803 struct sk_msg msg_redir = { }; 804 struct sk_psock *psock; 805 struct sock *sk_redir; 806 struct tls_rec *rec; 807 bool enospc, policy; 808 int err = 0, send; 809 u32 delta = 0; 810 811 policy = !(flags & MSG_SENDPAGE_NOPOLICY); 812 psock = sk_psock_get(sk); 813 if (!psock || !policy) { 814 err = tls_push_record(sk, flags, record_type); 815 if (err && sk->sk_err == EBADMSG) { 816 *copied -= sk_msg_free(sk, msg); 817 tls_free_open_rec(sk); 818 err = -sk->sk_err; 819 } 820 if (psock) 821 sk_psock_put(sk, psock); 822 return err; 823 } 824 more_data: 825 enospc = sk_msg_full(msg); 826 if (psock->eval == __SK_NONE) { 827 delta = msg->sg.size; 828 psock->eval = sk_psock_msg_verdict(sk, psock, msg); 829 delta -= msg->sg.size; 830 } 831 if (msg->cork_bytes && msg->cork_bytes > msg->sg.size && 832 !enospc && !full_record) { 833 err = -ENOSPC; 834 goto out_err; 835 } 836 msg->cork_bytes = 0; 837 send = msg->sg.size; 838 if (msg->apply_bytes && msg->apply_bytes < send) 839 send = msg->apply_bytes; 840 841 switch (psock->eval) { 842 case __SK_PASS: 843 err = tls_push_record(sk, flags, record_type); 844 if (err && sk->sk_err == EBADMSG) { 845 *copied -= sk_msg_free(sk, msg); 846 tls_free_open_rec(sk); 847 err = -sk->sk_err; 848 goto out_err; 849 } 850 break; 851 case __SK_REDIRECT: 852 sk_redir = psock->sk_redir; 853 memcpy(&msg_redir, msg, sizeof(*msg)); 854 if (msg->apply_bytes < send) 855 msg->apply_bytes = 0; 856 else 857 msg->apply_bytes -= send; 858 sk_msg_return_zero(sk, msg, send); 859 msg->sg.size -= send; 860 release_sock(sk); 861 err = tcp_bpf_sendmsg_redir(sk_redir, &msg_redir, send, flags); 862 lock_sock(sk); 863 if (err < 0) { 864 *copied -= sk_msg_free_nocharge(sk, &msg_redir); 865 msg->sg.size = 0; 866 } 867 if (msg->sg.size == 0) 868 tls_free_open_rec(sk); 869 break; 870 case __SK_DROP: 871 default: 872 sk_msg_free_partial(sk, msg, send); 873 if (msg->apply_bytes < send) 874 msg->apply_bytes = 0; 875 else 876 msg->apply_bytes -= send; 877 if (msg->sg.size == 0) 878 tls_free_open_rec(sk); 879 *copied -= (send + delta); 880 err = -EACCES; 881 } 882 883 if (likely(!err)) { 884 bool reset_eval = !ctx->open_rec; 885 886 rec = ctx->open_rec; 887 if (rec) { 888 msg = &rec->msg_plaintext; 889 if (!msg->apply_bytes) 890 reset_eval = true; 891 } 892 if (reset_eval) { 893 psock->eval = __SK_NONE; 894 if (psock->sk_redir) { 895 sock_put(psock->sk_redir); 896 psock->sk_redir = NULL; 897 } 898 } 899 if (rec) 900 goto more_data; 901 } 902 out_err: 903 sk_psock_put(sk, psock); 904 return err; 905 } 906 907 static int tls_sw_push_pending_record(struct sock *sk, int flags) 908 { 909 struct tls_context *tls_ctx = tls_get_ctx(sk); 910 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 911 struct tls_rec *rec = ctx->open_rec; 912 struct sk_msg *msg_pl; 913 size_t copied; 914 915 if (!rec) 916 return 0; 917 918 msg_pl = &rec->msg_plaintext; 919 copied = msg_pl->sg.size; 920 if (!copied) 921 return 0; 922 923 return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA, 924 &copied, flags); 925 } 926 927 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size) 928 { 929 long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT); 930 struct tls_context *tls_ctx = tls_get_ctx(sk); 931 struct tls_prot_info *prot = &tls_ctx->prot_info; 932 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 933 bool async_capable = ctx->async_capable; 934 unsigned char record_type = TLS_RECORD_TYPE_DATA; 935 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter); 936 bool eor = !(msg->msg_flags & MSG_MORE); 937 size_t try_to_copy; 938 ssize_t copied = 0; 939 struct sk_msg *msg_pl, *msg_en; 940 struct tls_rec *rec; 941 int required_size; 942 int num_async = 0; 943 bool full_record; 944 int record_room; 945 int num_zc = 0; 946 int orig_size; 947 int ret = 0; 948 int pending; 949 950 if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | 951 MSG_CMSG_COMPAT)) 952 return -EOPNOTSUPP; 953 954 mutex_lock(&tls_ctx->tx_lock); 955 lock_sock(sk); 956 957 if (unlikely(msg->msg_controllen)) { 958 ret = tls_proccess_cmsg(sk, msg, &record_type); 959 if (ret) { 960 if (ret == -EINPROGRESS) 961 num_async++; 962 else if (ret != -EAGAIN) 963 goto send_end; 964 } 965 } 966 967 while (msg_data_left(msg)) { 968 if (sk->sk_err) { 969 ret = -sk->sk_err; 970 goto send_end; 971 } 972 973 if (ctx->open_rec) 974 rec = ctx->open_rec; 975 else 976 rec = ctx->open_rec = tls_get_rec(sk); 977 if (!rec) { 978 ret = -ENOMEM; 979 goto send_end; 980 } 981 982 msg_pl = &rec->msg_plaintext; 983 msg_en = &rec->msg_encrypted; 984 985 orig_size = msg_pl->sg.size; 986 full_record = false; 987 try_to_copy = msg_data_left(msg); 988 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size; 989 if (try_to_copy >= record_room) { 990 try_to_copy = record_room; 991 full_record = true; 992 } 993 994 required_size = msg_pl->sg.size + try_to_copy + 995 prot->overhead_size; 996 997 if (!sk_stream_memory_free(sk)) 998 goto wait_for_sndbuf; 999 1000 alloc_encrypted: 1001 ret = tls_alloc_encrypted_msg(sk, required_size); 1002 if (ret) { 1003 if (ret != -ENOSPC) 1004 goto wait_for_memory; 1005 1006 /* Adjust try_to_copy according to the amount that was 1007 * actually allocated. The difference is due 1008 * to max sg elements limit 1009 */ 1010 try_to_copy -= required_size - msg_en->sg.size; 1011 full_record = true; 1012 } 1013 1014 if (!is_kvec && (full_record || eor) && !async_capable) { 1015 u32 first = msg_pl->sg.end; 1016 1017 ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter, 1018 msg_pl, try_to_copy); 1019 if (ret) 1020 goto fallback_to_reg_send; 1021 1022 num_zc++; 1023 copied += try_to_copy; 1024 1025 sk_msg_sg_copy_set(msg_pl, first); 1026 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record, 1027 record_type, &copied, 1028 msg->msg_flags); 1029 if (ret) { 1030 if (ret == -EINPROGRESS) 1031 num_async++; 1032 else if (ret == -ENOMEM) 1033 goto wait_for_memory; 1034 else if (ctx->open_rec && ret == -ENOSPC) 1035 goto rollback_iter; 1036 else if (ret != -EAGAIN) 1037 goto send_end; 1038 } 1039 continue; 1040 rollback_iter: 1041 copied -= try_to_copy; 1042 sk_msg_sg_copy_clear(msg_pl, first); 1043 iov_iter_revert(&msg->msg_iter, 1044 msg_pl->sg.size - orig_size); 1045 fallback_to_reg_send: 1046 sk_msg_trim(sk, msg_pl, orig_size); 1047 } 1048 1049 required_size = msg_pl->sg.size + try_to_copy; 1050 1051 ret = tls_clone_plaintext_msg(sk, required_size); 1052 if (ret) { 1053 if (ret != -ENOSPC) 1054 goto send_end; 1055 1056 /* Adjust try_to_copy according to the amount that was 1057 * actually allocated. The difference is due 1058 * to max sg elements limit 1059 */ 1060 try_to_copy -= required_size - msg_pl->sg.size; 1061 full_record = true; 1062 sk_msg_trim(sk, msg_en, 1063 msg_pl->sg.size + prot->overhead_size); 1064 } 1065 1066 if (try_to_copy) { 1067 ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter, 1068 msg_pl, try_to_copy); 1069 if (ret < 0) 1070 goto trim_sgl; 1071 } 1072 1073 /* Open records defined only if successfully copied, otherwise 1074 * we would trim the sg but not reset the open record frags. 1075 */ 1076 tls_ctx->pending_open_record_frags = true; 1077 copied += try_to_copy; 1078 if (full_record || eor) { 1079 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record, 1080 record_type, &copied, 1081 msg->msg_flags); 1082 if (ret) { 1083 if (ret == -EINPROGRESS) 1084 num_async++; 1085 else if (ret == -ENOMEM) 1086 goto wait_for_memory; 1087 else if (ret != -EAGAIN) { 1088 if (ret == -ENOSPC) 1089 ret = 0; 1090 goto send_end; 1091 } 1092 } 1093 } 1094 1095 continue; 1096 1097 wait_for_sndbuf: 1098 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1099 wait_for_memory: 1100 ret = sk_stream_wait_memory(sk, &timeo); 1101 if (ret) { 1102 trim_sgl: 1103 if (ctx->open_rec) 1104 tls_trim_both_msgs(sk, orig_size); 1105 goto send_end; 1106 } 1107 1108 if (ctx->open_rec && msg_en->sg.size < required_size) 1109 goto alloc_encrypted; 1110 } 1111 1112 if (!num_async) { 1113 goto send_end; 1114 } else if (num_zc) { 1115 /* Wait for pending encryptions to get completed */ 1116 spin_lock_bh(&ctx->encrypt_compl_lock); 1117 ctx->async_notify = true; 1118 1119 pending = atomic_read(&ctx->encrypt_pending); 1120 spin_unlock_bh(&ctx->encrypt_compl_lock); 1121 if (pending) 1122 crypto_wait_req(-EINPROGRESS, &ctx->async_wait); 1123 else 1124 reinit_completion(&ctx->async_wait.completion); 1125 1126 /* There can be no concurrent accesses, since we have no 1127 * pending encrypt operations 1128 */ 1129 WRITE_ONCE(ctx->async_notify, false); 1130 1131 if (ctx->async_wait.err) { 1132 ret = ctx->async_wait.err; 1133 copied = 0; 1134 } 1135 } 1136 1137 /* Transmit if any encryptions have completed */ 1138 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) { 1139 cancel_delayed_work(&ctx->tx_work.work); 1140 tls_tx_records(sk, msg->msg_flags); 1141 } 1142 1143 send_end: 1144 ret = sk_stream_error(sk, msg->msg_flags, ret); 1145 1146 release_sock(sk); 1147 mutex_unlock(&tls_ctx->tx_lock); 1148 return copied > 0 ? copied : ret; 1149 } 1150 1151 static int tls_sw_do_sendpage(struct sock *sk, struct page *page, 1152 int offset, size_t size, int flags) 1153 { 1154 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); 1155 struct tls_context *tls_ctx = tls_get_ctx(sk); 1156 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 1157 struct tls_prot_info *prot = &tls_ctx->prot_info; 1158 unsigned char record_type = TLS_RECORD_TYPE_DATA; 1159 struct sk_msg *msg_pl; 1160 struct tls_rec *rec; 1161 int num_async = 0; 1162 ssize_t copied = 0; 1163 bool full_record; 1164 int record_room; 1165 int ret = 0; 1166 bool eor; 1167 1168 eor = !(flags & MSG_SENDPAGE_NOTLAST); 1169 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 1170 1171 /* Call the sk_stream functions to manage the sndbuf mem. */ 1172 while (size > 0) { 1173 size_t copy, required_size; 1174 1175 if (sk->sk_err) { 1176 ret = -sk->sk_err; 1177 goto sendpage_end; 1178 } 1179 1180 if (ctx->open_rec) 1181 rec = ctx->open_rec; 1182 else 1183 rec = ctx->open_rec = tls_get_rec(sk); 1184 if (!rec) { 1185 ret = -ENOMEM; 1186 goto sendpage_end; 1187 } 1188 1189 msg_pl = &rec->msg_plaintext; 1190 1191 full_record = false; 1192 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size; 1193 copy = size; 1194 if (copy >= record_room) { 1195 copy = record_room; 1196 full_record = true; 1197 } 1198 1199 required_size = msg_pl->sg.size + copy + prot->overhead_size; 1200 1201 if (!sk_stream_memory_free(sk)) 1202 goto wait_for_sndbuf; 1203 alloc_payload: 1204 ret = tls_alloc_encrypted_msg(sk, required_size); 1205 if (ret) { 1206 if (ret != -ENOSPC) 1207 goto wait_for_memory; 1208 1209 /* Adjust copy according to the amount that was 1210 * actually allocated. The difference is due 1211 * to max sg elements limit 1212 */ 1213 copy -= required_size - msg_pl->sg.size; 1214 full_record = true; 1215 } 1216 1217 sk_msg_page_add(msg_pl, page, copy, offset); 1218 sk_mem_charge(sk, copy); 1219 1220 offset += copy; 1221 size -= copy; 1222 copied += copy; 1223 1224 tls_ctx->pending_open_record_frags = true; 1225 if (full_record || eor || sk_msg_full(msg_pl)) { 1226 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record, 1227 record_type, &copied, flags); 1228 if (ret) { 1229 if (ret == -EINPROGRESS) 1230 num_async++; 1231 else if (ret == -ENOMEM) 1232 goto wait_for_memory; 1233 else if (ret != -EAGAIN) { 1234 if (ret == -ENOSPC) 1235 ret = 0; 1236 goto sendpage_end; 1237 } 1238 } 1239 } 1240 continue; 1241 wait_for_sndbuf: 1242 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1243 wait_for_memory: 1244 ret = sk_stream_wait_memory(sk, &timeo); 1245 if (ret) { 1246 if (ctx->open_rec) 1247 tls_trim_both_msgs(sk, msg_pl->sg.size); 1248 goto sendpage_end; 1249 } 1250 1251 if (ctx->open_rec) 1252 goto alloc_payload; 1253 } 1254 1255 if (num_async) { 1256 /* Transmit if any encryptions have completed */ 1257 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) { 1258 cancel_delayed_work(&ctx->tx_work.work); 1259 tls_tx_records(sk, flags); 1260 } 1261 } 1262 sendpage_end: 1263 ret = sk_stream_error(sk, flags, ret); 1264 return copied > 0 ? copied : ret; 1265 } 1266 1267 int tls_sw_sendpage_locked(struct sock *sk, struct page *page, 1268 int offset, size_t size, int flags) 1269 { 1270 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | 1271 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY | 1272 MSG_NO_SHARED_FRAGS)) 1273 return -EOPNOTSUPP; 1274 1275 return tls_sw_do_sendpage(sk, page, offset, size, flags); 1276 } 1277 1278 int tls_sw_sendpage(struct sock *sk, struct page *page, 1279 int offset, size_t size, int flags) 1280 { 1281 struct tls_context *tls_ctx = tls_get_ctx(sk); 1282 int ret; 1283 1284 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | 1285 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY)) 1286 return -EOPNOTSUPP; 1287 1288 mutex_lock(&tls_ctx->tx_lock); 1289 lock_sock(sk); 1290 ret = tls_sw_do_sendpage(sk, page, offset, size, flags); 1291 release_sock(sk); 1292 mutex_unlock(&tls_ctx->tx_lock); 1293 return ret; 1294 } 1295 1296 static struct sk_buff *tls_wait_data(struct sock *sk, struct sk_psock *psock, 1297 bool nonblock, long timeo, int *err) 1298 { 1299 struct tls_context *tls_ctx = tls_get_ctx(sk); 1300 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 1301 struct sk_buff *skb; 1302 DEFINE_WAIT_FUNC(wait, woken_wake_function); 1303 1304 while (!(skb = ctx->recv_pkt) && sk_psock_queue_empty(psock)) { 1305 if (sk->sk_err) { 1306 *err = sock_error(sk); 1307 return NULL; 1308 } 1309 1310 if (!skb_queue_empty(&sk->sk_receive_queue)) { 1311 __strp_unpause(&ctx->strp); 1312 if (ctx->recv_pkt) 1313 return ctx->recv_pkt; 1314 } 1315 1316 if (sk->sk_shutdown & RCV_SHUTDOWN) 1317 return NULL; 1318 1319 if (sock_flag(sk, SOCK_DONE)) 1320 return NULL; 1321 1322 if (nonblock || !timeo) { 1323 *err = -EAGAIN; 1324 return NULL; 1325 } 1326 1327 add_wait_queue(sk_sleep(sk), &wait); 1328 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 1329 sk_wait_event(sk, &timeo, 1330 ctx->recv_pkt != skb || 1331 !sk_psock_queue_empty(psock), 1332 &wait); 1333 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 1334 remove_wait_queue(sk_sleep(sk), &wait); 1335 1336 /* Handle signals */ 1337 if (signal_pending(current)) { 1338 *err = sock_intr_errno(timeo); 1339 return NULL; 1340 } 1341 } 1342 1343 return skb; 1344 } 1345 1346 static int tls_setup_from_iter(struct iov_iter *from, 1347 int length, int *pages_used, 1348 struct scatterlist *to, 1349 int to_max_pages) 1350 { 1351 int rc = 0, i = 0, num_elem = *pages_used, maxpages; 1352 struct page *pages[MAX_SKB_FRAGS]; 1353 unsigned int size = 0; 1354 ssize_t copied, use; 1355 size_t offset; 1356 1357 while (length > 0) { 1358 i = 0; 1359 maxpages = to_max_pages - num_elem; 1360 if (maxpages == 0) { 1361 rc = -EFAULT; 1362 goto out; 1363 } 1364 copied = iov_iter_get_pages(from, pages, 1365 length, 1366 maxpages, &offset); 1367 if (copied <= 0) { 1368 rc = -EFAULT; 1369 goto out; 1370 } 1371 1372 iov_iter_advance(from, copied); 1373 1374 length -= copied; 1375 size += copied; 1376 while (copied) { 1377 use = min_t(int, copied, PAGE_SIZE - offset); 1378 1379 sg_set_page(&to[num_elem], 1380 pages[i], use, offset); 1381 sg_unmark_end(&to[num_elem]); 1382 /* We do not uncharge memory from this API */ 1383 1384 offset = 0; 1385 copied -= use; 1386 1387 i++; 1388 num_elem++; 1389 } 1390 } 1391 /* Mark the end in the last sg entry if newly added */ 1392 if (num_elem > *pages_used) 1393 sg_mark_end(&to[num_elem - 1]); 1394 out: 1395 if (rc) 1396 iov_iter_revert(from, size); 1397 *pages_used = num_elem; 1398 1399 return rc; 1400 } 1401 1402 /* This function decrypts the input skb into either out_iov or in out_sg 1403 * or in skb buffers itself. The input parameter 'zc' indicates if 1404 * zero-copy mode needs to be tried or not. With zero-copy mode, either 1405 * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are 1406 * NULL, then the decryption happens inside skb buffers itself, i.e. 1407 * zero-copy gets disabled and 'zc' is updated. 1408 */ 1409 1410 static int decrypt_internal(struct sock *sk, struct sk_buff *skb, 1411 struct iov_iter *out_iov, 1412 struct scatterlist *out_sg, 1413 struct tls_decrypt_arg *darg) 1414 { 1415 struct tls_context *tls_ctx = tls_get_ctx(sk); 1416 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 1417 struct tls_prot_info *prot = &tls_ctx->prot_info; 1418 struct strp_msg *rxm = strp_msg(skb); 1419 struct tls_msg *tlm = tls_msg(skb); 1420 int n_sgin, n_sgout, nsg, mem_size, aead_size, err, pages = 0; 1421 struct aead_request *aead_req; 1422 struct sk_buff *unused; 1423 u8 *aad, *iv, *mem = NULL; 1424 struct scatterlist *sgin = NULL; 1425 struct scatterlist *sgout = NULL; 1426 const int data_len = rxm->full_len - prot->overhead_size + 1427 prot->tail_size; 1428 int iv_offset = 0; 1429 1430 if (darg->zc && (out_iov || out_sg)) { 1431 if (out_iov) 1432 n_sgout = 1 + 1433 iov_iter_npages_cap(out_iov, INT_MAX, data_len); 1434 else 1435 n_sgout = sg_nents(out_sg); 1436 n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size, 1437 rxm->full_len - prot->prepend_size); 1438 } else { 1439 n_sgout = 0; 1440 darg->zc = false; 1441 n_sgin = skb_cow_data(skb, 0, &unused); 1442 } 1443 1444 if (n_sgin < 1) 1445 return -EBADMSG; 1446 1447 /* Increment to accommodate AAD */ 1448 n_sgin = n_sgin + 1; 1449 1450 nsg = n_sgin + n_sgout; 1451 1452 aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv); 1453 mem_size = aead_size + (nsg * sizeof(struct scatterlist)); 1454 mem_size = mem_size + prot->aad_size; 1455 mem_size = mem_size + MAX_IV_SIZE; 1456 1457 /* Allocate a single block of memory which contains 1458 * aead_req || sgin[] || sgout[] || aad || iv. 1459 * This order achieves correct alignment for aead_req, sgin, sgout. 1460 */ 1461 mem = kmalloc(mem_size, sk->sk_allocation); 1462 if (!mem) 1463 return -ENOMEM; 1464 1465 /* Segment the allocated memory */ 1466 aead_req = (struct aead_request *)mem; 1467 sgin = (struct scatterlist *)(mem + aead_size); 1468 sgout = sgin + n_sgin; 1469 aad = (u8 *)(sgout + n_sgout); 1470 iv = aad + prot->aad_size; 1471 1472 /* For CCM based ciphers, first byte of nonce+iv is a constant */ 1473 switch (prot->cipher_type) { 1474 case TLS_CIPHER_AES_CCM_128: 1475 iv[0] = TLS_AES_CCM_IV_B0_BYTE; 1476 iv_offset = 1; 1477 break; 1478 case TLS_CIPHER_SM4_CCM: 1479 iv[0] = TLS_SM4_CCM_IV_B0_BYTE; 1480 iv_offset = 1; 1481 break; 1482 } 1483 1484 /* Prepare IV */ 1485 if (prot->version == TLS_1_3_VERSION || 1486 prot->cipher_type == TLS_CIPHER_CHACHA20_POLY1305) { 1487 memcpy(iv + iv_offset, tls_ctx->rx.iv, 1488 prot->iv_size + prot->salt_size); 1489 } else { 1490 err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE, 1491 iv + iv_offset + prot->salt_size, 1492 prot->iv_size); 1493 if (err < 0) { 1494 kfree(mem); 1495 return err; 1496 } 1497 memcpy(iv + iv_offset, tls_ctx->rx.iv, prot->salt_size); 1498 } 1499 xor_iv_with_seq(prot, iv + iv_offset, tls_ctx->rx.rec_seq); 1500 1501 /* Prepare AAD */ 1502 tls_make_aad(aad, rxm->full_len - prot->overhead_size + 1503 prot->tail_size, 1504 tls_ctx->rx.rec_seq, tlm->control, prot); 1505 1506 /* Prepare sgin */ 1507 sg_init_table(sgin, n_sgin); 1508 sg_set_buf(&sgin[0], aad, prot->aad_size); 1509 err = skb_to_sgvec(skb, &sgin[1], 1510 rxm->offset + prot->prepend_size, 1511 rxm->full_len - prot->prepend_size); 1512 if (err < 0) { 1513 kfree(mem); 1514 return err; 1515 } 1516 1517 if (n_sgout) { 1518 if (out_iov) { 1519 sg_init_table(sgout, n_sgout); 1520 sg_set_buf(&sgout[0], aad, prot->aad_size); 1521 1522 err = tls_setup_from_iter(out_iov, data_len, 1523 &pages, &sgout[1], 1524 (n_sgout - 1)); 1525 if (err < 0) 1526 goto fallback_to_reg_recv; 1527 } else if (out_sg) { 1528 memcpy(sgout, out_sg, n_sgout * sizeof(*sgout)); 1529 } else { 1530 goto fallback_to_reg_recv; 1531 } 1532 } else { 1533 fallback_to_reg_recv: 1534 sgout = sgin; 1535 pages = 0; 1536 darg->zc = false; 1537 } 1538 1539 /* Prepare and submit AEAD request */ 1540 err = tls_do_decryption(sk, skb, sgin, sgout, iv, 1541 data_len, aead_req, darg); 1542 if (darg->async) 1543 return 0; 1544 1545 /* Release the pages in case iov was mapped to pages */ 1546 for (; pages > 0; pages--) 1547 put_page(sg_page(&sgout[pages])); 1548 1549 kfree(mem); 1550 return err; 1551 } 1552 1553 static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb, 1554 struct iov_iter *dest, 1555 struct tls_decrypt_arg *darg) 1556 { 1557 struct tls_context *tls_ctx = tls_get_ctx(sk); 1558 struct tls_prot_info *prot = &tls_ctx->prot_info; 1559 struct strp_msg *rxm = strp_msg(skb); 1560 struct tls_msg *tlm = tls_msg(skb); 1561 int pad, err; 1562 1563 if (tlm->decrypted) { 1564 darg->zc = false; 1565 darg->async = false; 1566 return 0; 1567 } 1568 1569 if (tls_ctx->rx_conf == TLS_HW) { 1570 err = tls_device_decrypted(sk, tls_ctx, skb, rxm); 1571 if (err < 0) 1572 return err; 1573 if (err > 0) { 1574 tlm->decrypted = 1; 1575 darg->zc = false; 1576 darg->async = false; 1577 goto decrypt_done; 1578 } 1579 } 1580 1581 err = decrypt_internal(sk, skb, dest, NULL, darg); 1582 if (err < 0) 1583 return err; 1584 if (darg->async) 1585 goto decrypt_next; 1586 1587 decrypt_done: 1588 pad = padding_length(prot, skb); 1589 if (pad < 0) 1590 return pad; 1591 1592 rxm->full_len -= pad; 1593 rxm->offset += prot->prepend_size; 1594 rxm->full_len -= prot->overhead_size; 1595 tlm->decrypted = 1; 1596 decrypt_next: 1597 tls_advance_record_sn(sk, prot, &tls_ctx->rx); 1598 1599 return 0; 1600 } 1601 1602 int decrypt_skb(struct sock *sk, struct sk_buff *skb, 1603 struct scatterlist *sgout) 1604 { 1605 struct tls_decrypt_arg darg = { .zc = true, }; 1606 1607 return decrypt_internal(sk, skb, NULL, sgout, &darg); 1608 } 1609 1610 static int tls_record_content_type(struct msghdr *msg, struct tls_msg *tlm, 1611 u8 *control) 1612 { 1613 int err; 1614 1615 if (!*control) { 1616 *control = tlm->control; 1617 if (!*control) 1618 return -EBADMSG; 1619 1620 err = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE, 1621 sizeof(*control), control); 1622 if (*control != TLS_RECORD_TYPE_DATA) { 1623 if (err || msg->msg_flags & MSG_CTRUNC) 1624 return -EIO; 1625 } 1626 } else if (*control != tlm->control) { 1627 return 0; 1628 } 1629 1630 return 1; 1631 } 1632 1633 /* This function traverses the rx_list in tls receive context to copies the 1634 * decrypted records into the buffer provided by caller zero copy is not 1635 * true. Further, the records are removed from the rx_list if it is not a peek 1636 * case and the record has been consumed completely. 1637 */ 1638 static int process_rx_list(struct tls_sw_context_rx *ctx, 1639 struct msghdr *msg, 1640 u8 *control, 1641 size_t skip, 1642 size_t len, 1643 bool zc, 1644 bool is_peek) 1645 { 1646 struct sk_buff *skb = skb_peek(&ctx->rx_list); 1647 struct tls_msg *tlm; 1648 ssize_t copied = 0; 1649 int err; 1650 1651 while (skip && skb) { 1652 struct strp_msg *rxm = strp_msg(skb); 1653 tlm = tls_msg(skb); 1654 1655 err = tls_record_content_type(msg, tlm, control); 1656 if (err <= 0) 1657 goto out; 1658 1659 if (skip < rxm->full_len) 1660 break; 1661 1662 skip = skip - rxm->full_len; 1663 skb = skb_peek_next(skb, &ctx->rx_list); 1664 } 1665 1666 while (len && skb) { 1667 struct sk_buff *next_skb; 1668 struct strp_msg *rxm = strp_msg(skb); 1669 int chunk = min_t(unsigned int, rxm->full_len - skip, len); 1670 1671 tlm = tls_msg(skb); 1672 1673 err = tls_record_content_type(msg, tlm, control); 1674 if (err <= 0) 1675 goto out; 1676 1677 if (!zc || (rxm->full_len - skip) > len) { 1678 err = skb_copy_datagram_msg(skb, rxm->offset + skip, 1679 msg, chunk); 1680 if (err < 0) 1681 goto out; 1682 } 1683 1684 len = len - chunk; 1685 copied = copied + chunk; 1686 1687 /* Consume the data from record if it is non-peek case*/ 1688 if (!is_peek) { 1689 rxm->offset = rxm->offset + chunk; 1690 rxm->full_len = rxm->full_len - chunk; 1691 1692 /* Return if there is unconsumed data in the record */ 1693 if (rxm->full_len - skip) 1694 break; 1695 } 1696 1697 /* The remaining skip-bytes must lie in 1st record in rx_list. 1698 * So from the 2nd record, 'skip' should be 0. 1699 */ 1700 skip = 0; 1701 1702 if (msg) 1703 msg->msg_flags |= MSG_EOR; 1704 1705 next_skb = skb_peek_next(skb, &ctx->rx_list); 1706 1707 if (!is_peek) { 1708 __skb_unlink(skb, &ctx->rx_list); 1709 consume_skb(skb); 1710 } 1711 1712 skb = next_skb; 1713 } 1714 err = 0; 1715 1716 out: 1717 return copied ? : err; 1718 } 1719 1720 int tls_sw_recvmsg(struct sock *sk, 1721 struct msghdr *msg, 1722 size_t len, 1723 int flags, 1724 int *addr_len) 1725 { 1726 struct tls_context *tls_ctx = tls_get_ctx(sk); 1727 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 1728 struct tls_prot_info *prot = &tls_ctx->prot_info; 1729 struct sk_psock *psock; 1730 unsigned char control = 0; 1731 ssize_t decrypted = 0; 1732 struct strp_msg *rxm; 1733 struct tls_msg *tlm; 1734 struct sk_buff *skb; 1735 ssize_t copied = 0; 1736 bool async = false; 1737 int target, err = 0; 1738 long timeo; 1739 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter); 1740 bool is_peek = flags & MSG_PEEK; 1741 bool bpf_strp_enabled; 1742 bool zc_capable; 1743 1744 if (unlikely(flags & MSG_ERRQUEUE)) 1745 return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR); 1746 1747 psock = sk_psock_get(sk); 1748 lock_sock(sk); 1749 bpf_strp_enabled = sk_psock_strp_enabled(psock); 1750 1751 /* If crypto failed the connection is broken */ 1752 err = ctx->async_wait.err; 1753 if (err) 1754 goto end; 1755 1756 /* Process pending decrypted records. It must be non-zero-copy */ 1757 err = process_rx_list(ctx, msg, &control, 0, len, false, is_peek); 1758 if (err < 0) 1759 goto end; 1760 1761 copied = err; 1762 if (len <= copied) 1763 goto end; 1764 1765 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); 1766 len = len - copied; 1767 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1768 1769 zc_capable = !bpf_strp_enabled && !is_kvec && !is_peek && 1770 prot->version != TLS_1_3_VERSION; 1771 decrypted = 0; 1772 while (len && (decrypted + copied < target || ctx->recv_pkt)) { 1773 struct tls_decrypt_arg darg = {}; 1774 int to_decrypt, chunk; 1775 1776 skb = tls_wait_data(sk, psock, flags & MSG_DONTWAIT, timeo, &err); 1777 if (!skb) { 1778 if (psock) { 1779 chunk = sk_msg_recvmsg(sk, psock, msg, len, 1780 flags); 1781 if (chunk > 0) 1782 goto leave_on_list; 1783 } 1784 goto recv_end; 1785 } 1786 1787 rxm = strp_msg(skb); 1788 tlm = tls_msg(skb); 1789 1790 to_decrypt = rxm->full_len - prot->overhead_size; 1791 1792 if (zc_capable && to_decrypt <= len && 1793 tlm->control == TLS_RECORD_TYPE_DATA) 1794 darg.zc = true; 1795 1796 /* Do not use async mode if record is non-data */ 1797 if (tlm->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled) 1798 darg.async = ctx->async_capable; 1799 else 1800 darg.async = false; 1801 1802 err = decrypt_skb_update(sk, skb, &msg->msg_iter, &darg); 1803 if (err < 0) { 1804 tls_err_abort(sk, -EBADMSG); 1805 goto recv_end; 1806 } 1807 1808 async |= darg.async; 1809 1810 /* If the type of records being processed is not known yet, 1811 * set it to record type just dequeued. If it is already known, 1812 * but does not match the record type just dequeued, go to end. 1813 * We always get record type here since for tls1.2, record type 1814 * is known just after record is dequeued from stream parser. 1815 * For tls1.3, we disable async. 1816 */ 1817 err = tls_record_content_type(msg, tlm, &control); 1818 if (err <= 0) 1819 goto recv_end; 1820 1821 ctx->recv_pkt = NULL; 1822 __strp_unpause(&ctx->strp); 1823 __skb_queue_tail(&ctx->rx_list, skb); 1824 1825 if (async) { 1826 /* TLS 1.2-only, to_decrypt must be text length */ 1827 chunk = min_t(int, to_decrypt, len); 1828 leave_on_list: 1829 decrypted += chunk; 1830 len -= chunk; 1831 continue; 1832 } 1833 /* TLS 1.3 may have updated the length by more than overhead */ 1834 chunk = rxm->full_len; 1835 1836 if (!darg.zc) { 1837 bool partially_consumed = chunk > len; 1838 1839 if (bpf_strp_enabled) { 1840 /* BPF may try to queue the skb */ 1841 __skb_unlink(skb, &ctx->rx_list); 1842 err = sk_psock_tls_strp_read(psock, skb); 1843 if (err != __SK_PASS) { 1844 rxm->offset = rxm->offset + rxm->full_len; 1845 rxm->full_len = 0; 1846 if (err == __SK_DROP) 1847 consume_skb(skb); 1848 continue; 1849 } 1850 __skb_queue_tail(&ctx->rx_list, skb); 1851 } 1852 1853 if (partially_consumed) 1854 chunk = len; 1855 1856 err = skb_copy_datagram_msg(skb, rxm->offset, 1857 msg, chunk); 1858 if (err < 0) 1859 goto recv_end; 1860 1861 if (is_peek) 1862 goto leave_on_list; 1863 1864 if (partially_consumed) { 1865 rxm->offset += chunk; 1866 rxm->full_len -= chunk; 1867 goto leave_on_list; 1868 } 1869 } 1870 1871 decrypted += chunk; 1872 len -= chunk; 1873 1874 __skb_unlink(skb, &ctx->rx_list); 1875 consume_skb(skb); 1876 1877 /* Return full control message to userspace before trying 1878 * to parse another message type 1879 */ 1880 msg->msg_flags |= MSG_EOR; 1881 if (control != TLS_RECORD_TYPE_DATA) 1882 break; 1883 } 1884 1885 recv_end: 1886 if (async) { 1887 int ret, pending; 1888 1889 /* Wait for all previously submitted records to be decrypted */ 1890 spin_lock_bh(&ctx->decrypt_compl_lock); 1891 reinit_completion(&ctx->async_wait.completion); 1892 pending = atomic_read(&ctx->decrypt_pending); 1893 spin_unlock_bh(&ctx->decrypt_compl_lock); 1894 if (pending) { 1895 ret = crypto_wait_req(-EINPROGRESS, &ctx->async_wait); 1896 if (ret) { 1897 if (err >= 0 || err == -EINPROGRESS) 1898 err = ret; 1899 decrypted = 0; 1900 goto end; 1901 } 1902 } 1903 1904 /* Drain records from the rx_list & copy if required */ 1905 if (is_peek || is_kvec) 1906 err = process_rx_list(ctx, msg, &control, copied, 1907 decrypted, false, is_peek); 1908 else 1909 err = process_rx_list(ctx, msg, &control, 0, 1910 decrypted, true, is_peek); 1911 decrypted = max(err, 0); 1912 } 1913 1914 copied += decrypted; 1915 1916 end: 1917 release_sock(sk); 1918 if (psock) 1919 sk_psock_put(sk, psock); 1920 return copied ? : err; 1921 } 1922 1923 ssize_t tls_sw_splice_read(struct socket *sock, loff_t *ppos, 1924 struct pipe_inode_info *pipe, 1925 size_t len, unsigned int flags) 1926 { 1927 struct tls_context *tls_ctx = tls_get_ctx(sock->sk); 1928 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 1929 struct strp_msg *rxm = NULL; 1930 struct sock *sk = sock->sk; 1931 struct tls_msg *tlm; 1932 struct sk_buff *skb; 1933 ssize_t copied = 0; 1934 bool from_queue; 1935 int err = 0; 1936 long timeo; 1937 int chunk; 1938 1939 lock_sock(sk); 1940 1941 timeo = sock_rcvtimeo(sk, flags & SPLICE_F_NONBLOCK); 1942 1943 from_queue = !skb_queue_empty(&ctx->rx_list); 1944 if (from_queue) { 1945 skb = __skb_dequeue(&ctx->rx_list); 1946 } else { 1947 struct tls_decrypt_arg darg = {}; 1948 1949 skb = tls_wait_data(sk, NULL, flags & SPLICE_F_NONBLOCK, timeo, 1950 &err); 1951 if (!skb) 1952 goto splice_read_end; 1953 1954 err = decrypt_skb_update(sk, skb, NULL, &darg); 1955 if (err < 0) { 1956 tls_err_abort(sk, -EBADMSG); 1957 goto splice_read_end; 1958 } 1959 } 1960 1961 rxm = strp_msg(skb); 1962 tlm = tls_msg(skb); 1963 1964 /* splice does not support reading control messages */ 1965 if (tlm->control != TLS_RECORD_TYPE_DATA) { 1966 err = -EINVAL; 1967 goto splice_read_end; 1968 } 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 (!from_queue) { 1976 ctx->recv_pkt = NULL; 1977 __strp_unpause(&ctx->strp); 1978 } 1979 if (chunk < rxm->full_len) { 1980 __skb_queue_head(&ctx->rx_list, skb); 1981 rxm->offset += len; 1982 rxm->full_len -= len; 1983 } else { 1984 consume_skb(skb); 1985 } 1986 1987 splice_read_end: 1988 release_sock(sk); 1989 return copied ? : err; 1990 } 1991 1992 bool tls_sw_sock_is_readable(struct sock *sk) 1993 { 1994 struct tls_context *tls_ctx = tls_get_ctx(sk); 1995 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 1996 bool ingress_empty = true; 1997 struct sk_psock *psock; 1998 1999 rcu_read_lock(); 2000 psock = sk_psock(sk); 2001 if (psock) 2002 ingress_empty = list_empty(&psock->ingress_msg); 2003 rcu_read_unlock(); 2004 2005 return !ingress_empty || ctx->recv_pkt || 2006 !skb_queue_empty(&ctx->rx_list); 2007 } 2008 2009 static int tls_read_size(struct strparser *strp, struct sk_buff *skb) 2010 { 2011 struct tls_context *tls_ctx = tls_get_ctx(strp->sk); 2012 struct tls_prot_info *prot = &tls_ctx->prot_info; 2013 char header[TLS_HEADER_SIZE + MAX_IV_SIZE]; 2014 struct strp_msg *rxm = strp_msg(skb); 2015 struct tls_msg *tlm = tls_msg(skb); 2016 size_t cipher_overhead; 2017 size_t data_len = 0; 2018 int ret; 2019 2020 /* Verify that we have a full TLS header, or wait for more data */ 2021 if (rxm->offset + prot->prepend_size > skb->len) 2022 return 0; 2023 2024 /* Sanity-check size of on-stack buffer. */ 2025 if (WARN_ON(prot->prepend_size > sizeof(header))) { 2026 ret = -EINVAL; 2027 goto read_failure; 2028 } 2029 2030 /* Linearize header to local buffer */ 2031 ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size); 2032 if (ret < 0) 2033 goto read_failure; 2034 2035 tlm->decrypted = 0; 2036 tlm->control = header[0]; 2037 2038 data_len = ((header[4] & 0xFF) | (header[3] << 8)); 2039 2040 cipher_overhead = prot->tag_size; 2041 if (prot->version != TLS_1_3_VERSION && 2042 prot->cipher_type != TLS_CIPHER_CHACHA20_POLY1305) 2043 cipher_overhead += prot->iv_size; 2044 2045 if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead + 2046 prot->tail_size) { 2047 ret = -EMSGSIZE; 2048 goto read_failure; 2049 } 2050 if (data_len < cipher_overhead) { 2051 ret = -EBADMSG; 2052 goto read_failure; 2053 } 2054 2055 /* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */ 2056 if (header[1] != TLS_1_2_VERSION_MINOR || 2057 header[2] != TLS_1_2_VERSION_MAJOR) { 2058 ret = -EINVAL; 2059 goto read_failure; 2060 } 2061 2062 tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE, 2063 TCP_SKB_CB(skb)->seq + rxm->offset); 2064 return data_len + TLS_HEADER_SIZE; 2065 2066 read_failure: 2067 tls_err_abort(strp->sk, ret); 2068 2069 return ret; 2070 } 2071 2072 static void tls_queue(struct strparser *strp, struct sk_buff *skb) 2073 { 2074 struct tls_context *tls_ctx = tls_get_ctx(strp->sk); 2075 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2076 2077 ctx->recv_pkt = skb; 2078 strp_pause(strp); 2079 2080 ctx->saved_data_ready(strp->sk); 2081 } 2082 2083 static void tls_data_ready(struct sock *sk) 2084 { 2085 struct tls_context *tls_ctx = tls_get_ctx(sk); 2086 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2087 struct sk_psock *psock; 2088 2089 strp_data_ready(&ctx->strp); 2090 2091 psock = sk_psock_get(sk); 2092 if (psock) { 2093 if (!list_empty(&psock->ingress_msg)) 2094 ctx->saved_data_ready(sk); 2095 sk_psock_put(sk, psock); 2096 } 2097 } 2098 2099 void tls_sw_cancel_work_tx(struct tls_context *tls_ctx) 2100 { 2101 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 2102 2103 set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask); 2104 set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask); 2105 cancel_delayed_work_sync(&ctx->tx_work.work); 2106 } 2107 2108 void tls_sw_release_resources_tx(struct sock *sk) 2109 { 2110 struct tls_context *tls_ctx = tls_get_ctx(sk); 2111 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 2112 struct tls_rec *rec, *tmp; 2113 int pending; 2114 2115 /* Wait for any pending async encryptions to complete */ 2116 spin_lock_bh(&ctx->encrypt_compl_lock); 2117 ctx->async_notify = true; 2118 pending = atomic_read(&ctx->encrypt_pending); 2119 spin_unlock_bh(&ctx->encrypt_compl_lock); 2120 2121 if (pending) 2122 crypto_wait_req(-EINPROGRESS, &ctx->async_wait); 2123 2124 tls_tx_records(sk, -1); 2125 2126 /* Free up un-sent records in tx_list. First, free 2127 * the partially sent record if any at head of tx_list. 2128 */ 2129 if (tls_ctx->partially_sent_record) { 2130 tls_free_partial_record(sk, tls_ctx); 2131 rec = list_first_entry(&ctx->tx_list, 2132 struct tls_rec, list); 2133 list_del(&rec->list); 2134 sk_msg_free(sk, &rec->msg_plaintext); 2135 kfree(rec); 2136 } 2137 2138 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) { 2139 list_del(&rec->list); 2140 sk_msg_free(sk, &rec->msg_encrypted); 2141 sk_msg_free(sk, &rec->msg_plaintext); 2142 kfree(rec); 2143 } 2144 2145 crypto_free_aead(ctx->aead_send); 2146 tls_free_open_rec(sk); 2147 } 2148 2149 void tls_sw_free_ctx_tx(struct tls_context *tls_ctx) 2150 { 2151 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 2152 2153 kfree(ctx); 2154 } 2155 2156 void tls_sw_release_resources_rx(struct sock *sk) 2157 { 2158 struct tls_context *tls_ctx = tls_get_ctx(sk); 2159 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2160 2161 kfree(tls_ctx->rx.rec_seq); 2162 kfree(tls_ctx->rx.iv); 2163 2164 if (ctx->aead_recv) { 2165 kfree_skb(ctx->recv_pkt); 2166 ctx->recv_pkt = NULL; 2167 __skb_queue_purge(&ctx->rx_list); 2168 crypto_free_aead(ctx->aead_recv); 2169 strp_stop(&ctx->strp); 2170 /* If tls_sw_strparser_arm() was not called (cleanup paths) 2171 * we still want to strp_stop(), but sk->sk_data_ready was 2172 * never swapped. 2173 */ 2174 if (ctx->saved_data_ready) { 2175 write_lock_bh(&sk->sk_callback_lock); 2176 sk->sk_data_ready = ctx->saved_data_ready; 2177 write_unlock_bh(&sk->sk_callback_lock); 2178 } 2179 } 2180 } 2181 2182 void tls_sw_strparser_done(struct tls_context *tls_ctx) 2183 { 2184 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2185 2186 strp_done(&ctx->strp); 2187 } 2188 2189 void tls_sw_free_ctx_rx(struct tls_context *tls_ctx) 2190 { 2191 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2192 2193 kfree(ctx); 2194 } 2195 2196 void tls_sw_free_resources_rx(struct sock *sk) 2197 { 2198 struct tls_context *tls_ctx = tls_get_ctx(sk); 2199 2200 tls_sw_release_resources_rx(sk); 2201 tls_sw_free_ctx_rx(tls_ctx); 2202 } 2203 2204 /* The work handler to transmitt the encrypted records in tx_list */ 2205 static void tx_work_handler(struct work_struct *work) 2206 { 2207 struct delayed_work *delayed_work = to_delayed_work(work); 2208 struct tx_work *tx_work = container_of(delayed_work, 2209 struct tx_work, work); 2210 struct sock *sk = tx_work->sk; 2211 struct tls_context *tls_ctx = tls_get_ctx(sk); 2212 struct tls_sw_context_tx *ctx; 2213 2214 if (unlikely(!tls_ctx)) 2215 return; 2216 2217 ctx = tls_sw_ctx_tx(tls_ctx); 2218 if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask)) 2219 return; 2220 2221 if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) 2222 return; 2223 mutex_lock(&tls_ctx->tx_lock); 2224 lock_sock(sk); 2225 tls_tx_records(sk, -1); 2226 release_sock(sk); 2227 mutex_unlock(&tls_ctx->tx_lock); 2228 } 2229 2230 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx) 2231 { 2232 struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx); 2233 2234 /* Schedule the transmission if tx list is ready */ 2235 if (is_tx_ready(tx_ctx) && 2236 !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask)) 2237 schedule_delayed_work(&tx_ctx->tx_work.work, 0); 2238 } 2239 2240 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx) 2241 { 2242 struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx); 2243 2244 write_lock_bh(&sk->sk_callback_lock); 2245 rx_ctx->saved_data_ready = sk->sk_data_ready; 2246 sk->sk_data_ready = tls_data_ready; 2247 write_unlock_bh(&sk->sk_callback_lock); 2248 2249 strp_check_rcv(&rx_ctx->strp); 2250 } 2251 2252 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx) 2253 { 2254 struct tls_context *tls_ctx = tls_get_ctx(sk); 2255 struct tls_prot_info *prot = &tls_ctx->prot_info; 2256 struct tls_crypto_info *crypto_info; 2257 struct tls_sw_context_tx *sw_ctx_tx = NULL; 2258 struct tls_sw_context_rx *sw_ctx_rx = NULL; 2259 struct cipher_context *cctx; 2260 struct crypto_aead **aead; 2261 struct strp_callbacks cb; 2262 u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size; 2263 struct crypto_tfm *tfm; 2264 char *iv, *rec_seq, *key, *salt, *cipher_name; 2265 size_t keysize; 2266 int rc = 0; 2267 2268 if (!ctx) { 2269 rc = -EINVAL; 2270 goto out; 2271 } 2272 2273 if (tx) { 2274 if (!ctx->priv_ctx_tx) { 2275 sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL); 2276 if (!sw_ctx_tx) { 2277 rc = -ENOMEM; 2278 goto out; 2279 } 2280 ctx->priv_ctx_tx = sw_ctx_tx; 2281 } else { 2282 sw_ctx_tx = 2283 (struct tls_sw_context_tx *)ctx->priv_ctx_tx; 2284 } 2285 } else { 2286 if (!ctx->priv_ctx_rx) { 2287 sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL); 2288 if (!sw_ctx_rx) { 2289 rc = -ENOMEM; 2290 goto out; 2291 } 2292 ctx->priv_ctx_rx = sw_ctx_rx; 2293 } else { 2294 sw_ctx_rx = 2295 (struct tls_sw_context_rx *)ctx->priv_ctx_rx; 2296 } 2297 } 2298 2299 if (tx) { 2300 crypto_init_wait(&sw_ctx_tx->async_wait); 2301 spin_lock_init(&sw_ctx_tx->encrypt_compl_lock); 2302 crypto_info = &ctx->crypto_send.info; 2303 cctx = &ctx->tx; 2304 aead = &sw_ctx_tx->aead_send; 2305 INIT_LIST_HEAD(&sw_ctx_tx->tx_list); 2306 INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler); 2307 sw_ctx_tx->tx_work.sk = sk; 2308 } else { 2309 crypto_init_wait(&sw_ctx_rx->async_wait); 2310 spin_lock_init(&sw_ctx_rx->decrypt_compl_lock); 2311 crypto_info = &ctx->crypto_recv.info; 2312 cctx = &ctx->rx; 2313 skb_queue_head_init(&sw_ctx_rx->rx_list); 2314 aead = &sw_ctx_rx->aead_recv; 2315 } 2316 2317 switch (crypto_info->cipher_type) { 2318 case TLS_CIPHER_AES_GCM_128: { 2319 struct tls12_crypto_info_aes_gcm_128 *gcm_128_info; 2320 2321 gcm_128_info = (void *)crypto_info; 2322 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE; 2323 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE; 2324 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE; 2325 iv = gcm_128_info->iv; 2326 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE; 2327 rec_seq = gcm_128_info->rec_seq; 2328 keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE; 2329 key = gcm_128_info->key; 2330 salt = gcm_128_info->salt; 2331 salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE; 2332 cipher_name = "gcm(aes)"; 2333 break; 2334 } 2335 case TLS_CIPHER_AES_GCM_256: { 2336 struct tls12_crypto_info_aes_gcm_256 *gcm_256_info; 2337 2338 gcm_256_info = (void *)crypto_info; 2339 nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE; 2340 tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE; 2341 iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE; 2342 iv = gcm_256_info->iv; 2343 rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE; 2344 rec_seq = gcm_256_info->rec_seq; 2345 keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE; 2346 key = gcm_256_info->key; 2347 salt = gcm_256_info->salt; 2348 salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE; 2349 cipher_name = "gcm(aes)"; 2350 break; 2351 } 2352 case TLS_CIPHER_AES_CCM_128: { 2353 struct tls12_crypto_info_aes_ccm_128 *ccm_128_info; 2354 2355 ccm_128_info = (void *)crypto_info; 2356 nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE; 2357 tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE; 2358 iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE; 2359 iv = ccm_128_info->iv; 2360 rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE; 2361 rec_seq = ccm_128_info->rec_seq; 2362 keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE; 2363 key = ccm_128_info->key; 2364 salt = ccm_128_info->salt; 2365 salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE; 2366 cipher_name = "ccm(aes)"; 2367 break; 2368 } 2369 case TLS_CIPHER_CHACHA20_POLY1305: { 2370 struct tls12_crypto_info_chacha20_poly1305 *chacha20_poly1305_info; 2371 2372 chacha20_poly1305_info = (void *)crypto_info; 2373 nonce_size = 0; 2374 tag_size = TLS_CIPHER_CHACHA20_POLY1305_TAG_SIZE; 2375 iv_size = TLS_CIPHER_CHACHA20_POLY1305_IV_SIZE; 2376 iv = chacha20_poly1305_info->iv; 2377 rec_seq_size = TLS_CIPHER_CHACHA20_POLY1305_REC_SEQ_SIZE; 2378 rec_seq = chacha20_poly1305_info->rec_seq; 2379 keysize = TLS_CIPHER_CHACHA20_POLY1305_KEY_SIZE; 2380 key = chacha20_poly1305_info->key; 2381 salt = chacha20_poly1305_info->salt; 2382 salt_size = TLS_CIPHER_CHACHA20_POLY1305_SALT_SIZE; 2383 cipher_name = "rfc7539(chacha20,poly1305)"; 2384 break; 2385 } 2386 case TLS_CIPHER_SM4_GCM: { 2387 struct tls12_crypto_info_sm4_gcm *sm4_gcm_info; 2388 2389 sm4_gcm_info = (void *)crypto_info; 2390 nonce_size = TLS_CIPHER_SM4_GCM_IV_SIZE; 2391 tag_size = TLS_CIPHER_SM4_GCM_TAG_SIZE; 2392 iv_size = TLS_CIPHER_SM4_GCM_IV_SIZE; 2393 iv = sm4_gcm_info->iv; 2394 rec_seq_size = TLS_CIPHER_SM4_GCM_REC_SEQ_SIZE; 2395 rec_seq = sm4_gcm_info->rec_seq; 2396 keysize = TLS_CIPHER_SM4_GCM_KEY_SIZE; 2397 key = sm4_gcm_info->key; 2398 salt = sm4_gcm_info->salt; 2399 salt_size = TLS_CIPHER_SM4_GCM_SALT_SIZE; 2400 cipher_name = "gcm(sm4)"; 2401 break; 2402 } 2403 case TLS_CIPHER_SM4_CCM: { 2404 struct tls12_crypto_info_sm4_ccm *sm4_ccm_info; 2405 2406 sm4_ccm_info = (void *)crypto_info; 2407 nonce_size = TLS_CIPHER_SM4_CCM_IV_SIZE; 2408 tag_size = TLS_CIPHER_SM4_CCM_TAG_SIZE; 2409 iv_size = TLS_CIPHER_SM4_CCM_IV_SIZE; 2410 iv = sm4_ccm_info->iv; 2411 rec_seq_size = TLS_CIPHER_SM4_CCM_REC_SEQ_SIZE; 2412 rec_seq = sm4_ccm_info->rec_seq; 2413 keysize = TLS_CIPHER_SM4_CCM_KEY_SIZE; 2414 key = sm4_ccm_info->key; 2415 salt = sm4_ccm_info->salt; 2416 salt_size = TLS_CIPHER_SM4_CCM_SALT_SIZE; 2417 cipher_name = "ccm(sm4)"; 2418 break; 2419 } 2420 default: 2421 rc = -EINVAL; 2422 goto free_priv; 2423 } 2424 2425 /* Sanity-check the sizes for stack allocations. */ 2426 if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE || 2427 rec_seq_size > TLS_MAX_REC_SEQ_SIZE || tag_size != TLS_TAG_SIZE) { 2428 rc = -EINVAL; 2429 goto free_priv; 2430 } 2431 2432 if (crypto_info->version == TLS_1_3_VERSION) { 2433 nonce_size = 0; 2434 prot->aad_size = TLS_HEADER_SIZE; 2435 prot->tail_size = 1; 2436 } else { 2437 prot->aad_size = TLS_AAD_SPACE_SIZE; 2438 prot->tail_size = 0; 2439 } 2440 2441 prot->version = crypto_info->version; 2442 prot->cipher_type = crypto_info->cipher_type; 2443 prot->prepend_size = TLS_HEADER_SIZE + nonce_size; 2444 prot->tag_size = tag_size; 2445 prot->overhead_size = prot->prepend_size + 2446 prot->tag_size + prot->tail_size; 2447 prot->iv_size = iv_size; 2448 prot->salt_size = salt_size; 2449 cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL); 2450 if (!cctx->iv) { 2451 rc = -ENOMEM; 2452 goto free_priv; 2453 } 2454 /* Note: 128 & 256 bit salt are the same size */ 2455 prot->rec_seq_size = rec_seq_size; 2456 memcpy(cctx->iv, salt, salt_size); 2457 memcpy(cctx->iv + salt_size, iv, iv_size); 2458 cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL); 2459 if (!cctx->rec_seq) { 2460 rc = -ENOMEM; 2461 goto free_iv; 2462 } 2463 2464 if (!*aead) { 2465 *aead = crypto_alloc_aead(cipher_name, 0, 0); 2466 if (IS_ERR(*aead)) { 2467 rc = PTR_ERR(*aead); 2468 *aead = NULL; 2469 goto free_rec_seq; 2470 } 2471 } 2472 2473 ctx->push_pending_record = tls_sw_push_pending_record; 2474 2475 rc = crypto_aead_setkey(*aead, key, keysize); 2476 2477 if (rc) 2478 goto free_aead; 2479 2480 rc = crypto_aead_setauthsize(*aead, prot->tag_size); 2481 if (rc) 2482 goto free_aead; 2483 2484 if (sw_ctx_rx) { 2485 tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv); 2486 2487 if (crypto_info->version == TLS_1_3_VERSION) 2488 sw_ctx_rx->async_capable = 0; 2489 else 2490 sw_ctx_rx->async_capable = 2491 !!(tfm->__crt_alg->cra_flags & 2492 CRYPTO_ALG_ASYNC); 2493 2494 /* Set up strparser */ 2495 memset(&cb, 0, sizeof(cb)); 2496 cb.rcv_msg = tls_queue; 2497 cb.parse_msg = tls_read_size; 2498 2499 strp_init(&sw_ctx_rx->strp, sk, &cb); 2500 } 2501 2502 goto out; 2503 2504 free_aead: 2505 crypto_free_aead(*aead); 2506 *aead = NULL; 2507 free_rec_seq: 2508 kfree(cctx->rec_seq); 2509 cctx->rec_seq = NULL; 2510 free_iv: 2511 kfree(cctx->iv); 2512 cctx->iv = NULL; 2513 free_priv: 2514 if (tx) { 2515 kfree(ctx->priv_ctx_tx); 2516 ctx->priv_ctx_tx = NULL; 2517 } else { 2518 kfree(ctx->priv_ctx_rx); 2519 ctx->priv_ctx_rx = NULL; 2520 } 2521 out: 2522 return rc; 2523 } 2524