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 noinline void tls_err_abort(struct sock *sk, int err) 48 { 49 WARN_ON_ONCE(err >= 0); 50 /* sk->sk_err should contain a positive error code. */ 51 sk->sk_err = -err; 52 sk_error_report(sk); 53 } 54 55 static int __skb_nsg(struct sk_buff *skb, int offset, int len, 56 unsigned int recursion_level) 57 { 58 int start = skb_headlen(skb); 59 int i, chunk = start - offset; 60 struct sk_buff *frag_iter; 61 int elt = 0; 62 63 if (unlikely(recursion_level >= 24)) 64 return -EMSGSIZE; 65 66 if (chunk > 0) { 67 if (chunk > len) 68 chunk = len; 69 elt++; 70 len -= chunk; 71 if (len == 0) 72 return elt; 73 offset += chunk; 74 } 75 76 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 77 int end; 78 79 WARN_ON(start > offset + len); 80 81 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 82 chunk = end - offset; 83 if (chunk > 0) { 84 if (chunk > len) 85 chunk = len; 86 elt++; 87 len -= chunk; 88 if (len == 0) 89 return elt; 90 offset += chunk; 91 } 92 start = end; 93 } 94 95 if (unlikely(skb_has_frag_list(skb))) { 96 skb_walk_frags(skb, frag_iter) { 97 int end, ret; 98 99 WARN_ON(start > offset + len); 100 101 end = start + frag_iter->len; 102 chunk = end - offset; 103 if (chunk > 0) { 104 if (chunk > len) 105 chunk = len; 106 ret = __skb_nsg(frag_iter, offset - start, chunk, 107 recursion_level + 1); 108 if (unlikely(ret < 0)) 109 return ret; 110 elt += ret; 111 len -= chunk; 112 if (len == 0) 113 return elt; 114 offset += chunk; 115 } 116 start = end; 117 } 118 } 119 BUG_ON(len); 120 return elt; 121 } 122 123 /* Return the number of scatterlist elements required to completely map the 124 * skb, or -EMSGSIZE if the recursion depth is exceeded. 125 */ 126 static int skb_nsg(struct sk_buff *skb, int offset, int len) 127 { 128 return __skb_nsg(skb, offset, len, 0); 129 } 130 131 static int padding_length(struct tls_sw_context_rx *ctx, 132 struct tls_prot_info *prot, struct sk_buff *skb) 133 { 134 struct strp_msg *rxm = strp_msg(skb); 135 int sub = 0; 136 137 /* Determine zero-padding length */ 138 if (prot->version == TLS_1_3_VERSION) { 139 char content_type = 0; 140 int err; 141 int back = 17; 142 143 while (content_type == 0) { 144 if (back > rxm->full_len - prot->prepend_size) 145 return -EBADMSG; 146 err = skb_copy_bits(skb, 147 rxm->offset + rxm->full_len - back, 148 &content_type, 1); 149 if (err) 150 return err; 151 if (content_type) 152 break; 153 sub++; 154 back++; 155 } 156 ctx->control = content_type; 157 } 158 return sub; 159 } 160 161 static void tls_decrypt_done(struct crypto_async_request *req, int err) 162 { 163 struct aead_request *aead_req = (struct aead_request *)req; 164 struct scatterlist *sgout = aead_req->dst; 165 struct scatterlist *sgin = aead_req->src; 166 struct tls_sw_context_rx *ctx; 167 struct tls_context *tls_ctx; 168 struct tls_prot_info *prot; 169 struct scatterlist *sg; 170 struct sk_buff *skb; 171 unsigned int pages; 172 int pending; 173 174 skb = (struct sk_buff *)req->data; 175 tls_ctx = tls_get_ctx(skb->sk); 176 ctx = tls_sw_ctx_rx(tls_ctx); 177 prot = &tls_ctx->prot_info; 178 179 /* Propagate if there was an err */ 180 if (err) { 181 if (err == -EBADMSG) 182 TLS_INC_STATS(sock_net(skb->sk), 183 LINUX_MIB_TLSDECRYPTERROR); 184 ctx->async_wait.err = err; 185 tls_err_abort(skb->sk, err); 186 } else { 187 struct strp_msg *rxm = strp_msg(skb); 188 int pad; 189 190 pad = padding_length(ctx, prot, skb); 191 if (pad < 0) { 192 ctx->async_wait.err = pad; 193 tls_err_abort(skb->sk, pad); 194 } else { 195 rxm->full_len -= pad; 196 rxm->offset += prot->prepend_size; 197 rxm->full_len -= prot->overhead_size; 198 } 199 } 200 201 /* After using skb->sk to propagate sk through crypto async callback 202 * we need to NULL it again. 203 */ 204 skb->sk = NULL; 205 206 207 /* Free the destination pages if skb was not decrypted inplace */ 208 if (sgout != sgin) { 209 /* Skip the first S/G entry as it points to AAD */ 210 for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) { 211 if (!sg) 212 break; 213 put_page(sg_page(sg)); 214 } 215 } 216 217 kfree(aead_req); 218 219 spin_lock_bh(&ctx->decrypt_compl_lock); 220 pending = atomic_dec_return(&ctx->decrypt_pending); 221 222 if (!pending && ctx->async_notify) 223 complete(&ctx->async_wait.completion); 224 spin_unlock_bh(&ctx->decrypt_compl_lock); 225 } 226 227 static int tls_do_decryption(struct sock *sk, 228 struct sk_buff *skb, 229 struct scatterlist *sgin, 230 struct scatterlist *sgout, 231 char *iv_recv, 232 size_t data_len, 233 struct aead_request *aead_req, 234 bool async) 235 { 236 struct tls_context *tls_ctx = tls_get_ctx(sk); 237 struct tls_prot_info *prot = &tls_ctx->prot_info; 238 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 239 int ret; 240 241 aead_request_set_tfm(aead_req, ctx->aead_recv); 242 aead_request_set_ad(aead_req, prot->aad_size); 243 aead_request_set_crypt(aead_req, sgin, sgout, 244 data_len + prot->tag_size, 245 (u8 *)iv_recv); 246 247 if (async) { 248 /* Using skb->sk to push sk through to crypto async callback 249 * handler. This allows propagating errors up to the socket 250 * if needed. It _must_ be cleared in the async handler 251 * before consume_skb is called. We _know_ skb->sk is NULL 252 * because it is a clone from strparser. 253 */ 254 skb->sk = sk; 255 aead_request_set_callback(aead_req, 256 CRYPTO_TFM_REQ_MAY_BACKLOG, 257 tls_decrypt_done, skb); 258 atomic_inc(&ctx->decrypt_pending); 259 } else { 260 aead_request_set_callback(aead_req, 261 CRYPTO_TFM_REQ_MAY_BACKLOG, 262 crypto_req_done, &ctx->async_wait); 263 } 264 265 ret = crypto_aead_decrypt(aead_req); 266 if (ret == -EINPROGRESS) { 267 if (async) 268 return ret; 269 270 ret = crypto_wait_req(ret, &ctx->async_wait); 271 } 272 273 if (async) 274 atomic_dec(&ctx->decrypt_pending); 275 276 return ret; 277 } 278 279 static void tls_trim_both_msgs(struct sock *sk, int target_size) 280 { 281 struct tls_context *tls_ctx = tls_get_ctx(sk); 282 struct tls_prot_info *prot = &tls_ctx->prot_info; 283 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 284 struct tls_rec *rec = ctx->open_rec; 285 286 sk_msg_trim(sk, &rec->msg_plaintext, target_size); 287 if (target_size > 0) 288 target_size += prot->overhead_size; 289 sk_msg_trim(sk, &rec->msg_encrypted, target_size); 290 } 291 292 static int tls_alloc_encrypted_msg(struct sock *sk, int len) 293 { 294 struct tls_context *tls_ctx = tls_get_ctx(sk); 295 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 296 struct tls_rec *rec = ctx->open_rec; 297 struct sk_msg *msg_en = &rec->msg_encrypted; 298 299 return sk_msg_alloc(sk, msg_en, len, 0); 300 } 301 302 static int tls_clone_plaintext_msg(struct sock *sk, int required) 303 { 304 struct tls_context *tls_ctx = tls_get_ctx(sk); 305 struct tls_prot_info *prot = &tls_ctx->prot_info; 306 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 307 struct tls_rec *rec = ctx->open_rec; 308 struct sk_msg *msg_pl = &rec->msg_plaintext; 309 struct sk_msg *msg_en = &rec->msg_encrypted; 310 int skip, len; 311 312 /* We add page references worth len bytes from encrypted sg 313 * at the end of plaintext sg. It is guaranteed that msg_en 314 * has enough required room (ensured by caller). 315 */ 316 len = required - msg_pl->sg.size; 317 318 /* Skip initial bytes in msg_en's data to be able to use 319 * same offset of both plain and encrypted data. 320 */ 321 skip = prot->prepend_size + msg_pl->sg.size; 322 323 return sk_msg_clone(sk, msg_pl, msg_en, skip, len); 324 } 325 326 static struct tls_rec *tls_get_rec(struct sock *sk) 327 { 328 struct tls_context *tls_ctx = tls_get_ctx(sk); 329 struct tls_prot_info *prot = &tls_ctx->prot_info; 330 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 331 struct sk_msg *msg_pl, *msg_en; 332 struct tls_rec *rec; 333 int mem_size; 334 335 mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send); 336 337 rec = kzalloc(mem_size, sk->sk_allocation); 338 if (!rec) 339 return NULL; 340 341 msg_pl = &rec->msg_plaintext; 342 msg_en = &rec->msg_encrypted; 343 344 sk_msg_init(msg_pl); 345 sk_msg_init(msg_en); 346 347 sg_init_table(rec->sg_aead_in, 2); 348 sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size); 349 sg_unmark_end(&rec->sg_aead_in[1]); 350 351 sg_init_table(rec->sg_aead_out, 2); 352 sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size); 353 sg_unmark_end(&rec->sg_aead_out[1]); 354 355 return rec; 356 } 357 358 static void tls_free_rec(struct sock *sk, struct tls_rec *rec) 359 { 360 sk_msg_free(sk, &rec->msg_encrypted); 361 sk_msg_free(sk, &rec->msg_plaintext); 362 kfree(rec); 363 } 364 365 static void tls_free_open_rec(struct sock *sk) 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 = ctx->open_rec; 370 371 if (rec) { 372 tls_free_rec(sk, rec); 373 ctx->open_rec = NULL; 374 } 375 } 376 377 int tls_tx_records(struct sock *sk, int flags) 378 { 379 struct tls_context *tls_ctx = tls_get_ctx(sk); 380 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 381 struct tls_rec *rec, *tmp; 382 struct sk_msg *msg_en; 383 int tx_flags, rc = 0; 384 385 if (tls_is_partially_sent_record(tls_ctx)) { 386 rec = list_first_entry(&ctx->tx_list, 387 struct tls_rec, list); 388 389 if (flags == -1) 390 tx_flags = rec->tx_flags; 391 else 392 tx_flags = flags; 393 394 rc = tls_push_partial_record(sk, tls_ctx, tx_flags); 395 if (rc) 396 goto tx_err; 397 398 /* Full record has been transmitted. 399 * Remove the head of tx_list 400 */ 401 list_del(&rec->list); 402 sk_msg_free(sk, &rec->msg_plaintext); 403 kfree(rec); 404 } 405 406 /* Tx all ready records */ 407 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) { 408 if (READ_ONCE(rec->tx_ready)) { 409 if (flags == -1) 410 tx_flags = rec->tx_flags; 411 else 412 tx_flags = flags; 413 414 msg_en = &rec->msg_encrypted; 415 rc = tls_push_sg(sk, tls_ctx, 416 &msg_en->sg.data[msg_en->sg.curr], 417 0, tx_flags); 418 if (rc) 419 goto tx_err; 420 421 list_del(&rec->list); 422 sk_msg_free(sk, &rec->msg_plaintext); 423 kfree(rec); 424 } else { 425 break; 426 } 427 } 428 429 tx_err: 430 if (rc < 0 && rc != -EAGAIN) 431 tls_err_abort(sk, -EBADMSG); 432 433 return rc; 434 } 435 436 static void tls_encrypt_done(struct crypto_async_request *req, int err) 437 { 438 struct aead_request *aead_req = (struct aead_request *)req; 439 struct sock *sk = req->data; 440 struct tls_context *tls_ctx = tls_get_ctx(sk); 441 struct tls_prot_info *prot = &tls_ctx->prot_info; 442 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 443 struct scatterlist *sge; 444 struct sk_msg *msg_en; 445 struct tls_rec *rec; 446 bool ready = false; 447 int pending; 448 449 rec = container_of(aead_req, struct tls_rec, aead_req); 450 msg_en = &rec->msg_encrypted; 451 452 sge = sk_msg_elem(msg_en, msg_en->sg.curr); 453 sge->offset -= prot->prepend_size; 454 sge->length += prot->prepend_size; 455 456 /* Check if error is previously set on socket */ 457 if (err || sk->sk_err) { 458 rec = NULL; 459 460 /* If err is already set on socket, return the same code */ 461 if (sk->sk_err) { 462 ctx->async_wait.err = -sk->sk_err; 463 } else { 464 ctx->async_wait.err = err; 465 tls_err_abort(sk, err); 466 } 467 } 468 469 if (rec) { 470 struct tls_rec *first_rec; 471 472 /* Mark the record as ready for transmission */ 473 smp_store_mb(rec->tx_ready, true); 474 475 /* If received record is at head of tx_list, schedule tx */ 476 first_rec = list_first_entry(&ctx->tx_list, 477 struct tls_rec, list); 478 if (rec == first_rec) 479 ready = true; 480 } 481 482 spin_lock_bh(&ctx->encrypt_compl_lock); 483 pending = atomic_dec_return(&ctx->encrypt_pending); 484 485 if (!pending && ctx->async_notify) 486 complete(&ctx->async_wait.completion); 487 spin_unlock_bh(&ctx->encrypt_compl_lock); 488 489 if (!ready) 490 return; 491 492 /* Schedule the transmission */ 493 if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) 494 schedule_delayed_work(&ctx->tx_work.work, 1); 495 } 496 497 static int tls_do_encryption(struct sock *sk, 498 struct tls_context *tls_ctx, 499 struct tls_sw_context_tx *ctx, 500 struct aead_request *aead_req, 501 size_t data_len, u32 start) 502 { 503 struct tls_prot_info *prot = &tls_ctx->prot_info; 504 struct tls_rec *rec = ctx->open_rec; 505 struct sk_msg *msg_en = &rec->msg_encrypted; 506 struct scatterlist *sge = sk_msg_elem(msg_en, start); 507 int rc, iv_offset = 0; 508 509 /* For CCM based ciphers, first byte of IV is a constant */ 510 switch (prot->cipher_type) { 511 case TLS_CIPHER_AES_CCM_128: 512 rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE; 513 iv_offset = 1; 514 break; 515 case TLS_CIPHER_SM4_CCM: 516 rec->iv_data[0] = TLS_SM4_CCM_IV_B0_BYTE; 517 iv_offset = 1; 518 break; 519 } 520 521 memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv, 522 prot->iv_size + prot->salt_size); 523 524 xor_iv_with_seq(prot, rec->iv_data + iv_offset, tls_ctx->tx.rec_seq); 525 526 sge->offset += prot->prepend_size; 527 sge->length -= prot->prepend_size; 528 529 msg_en->sg.curr = start; 530 531 aead_request_set_tfm(aead_req, ctx->aead_send); 532 aead_request_set_ad(aead_req, prot->aad_size); 533 aead_request_set_crypt(aead_req, rec->sg_aead_in, 534 rec->sg_aead_out, 535 data_len, rec->iv_data); 536 537 aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG, 538 tls_encrypt_done, sk); 539 540 /* Add the record in tx_list */ 541 list_add_tail((struct list_head *)&rec->list, &ctx->tx_list); 542 atomic_inc(&ctx->encrypt_pending); 543 544 rc = crypto_aead_encrypt(aead_req); 545 if (!rc || rc != -EINPROGRESS) { 546 atomic_dec(&ctx->encrypt_pending); 547 sge->offset -= prot->prepend_size; 548 sge->length += prot->prepend_size; 549 } 550 551 if (!rc) { 552 WRITE_ONCE(rec->tx_ready, true); 553 } else if (rc != -EINPROGRESS) { 554 list_del(&rec->list); 555 return rc; 556 } 557 558 /* Unhook the record from context if encryption is not failure */ 559 ctx->open_rec = NULL; 560 tls_advance_record_sn(sk, prot, &tls_ctx->tx); 561 return rc; 562 } 563 564 static int tls_split_open_record(struct sock *sk, struct tls_rec *from, 565 struct tls_rec **to, struct sk_msg *msg_opl, 566 struct sk_msg *msg_oen, u32 split_point, 567 u32 tx_overhead_size, u32 *orig_end) 568 { 569 u32 i, j, bytes = 0, apply = msg_opl->apply_bytes; 570 struct scatterlist *sge, *osge, *nsge; 571 u32 orig_size = msg_opl->sg.size; 572 struct scatterlist tmp = { }; 573 struct sk_msg *msg_npl; 574 struct tls_rec *new; 575 int ret; 576 577 new = tls_get_rec(sk); 578 if (!new) 579 return -ENOMEM; 580 ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size + 581 tx_overhead_size, 0); 582 if (ret < 0) { 583 tls_free_rec(sk, new); 584 return ret; 585 } 586 587 *orig_end = msg_opl->sg.end; 588 i = msg_opl->sg.start; 589 sge = sk_msg_elem(msg_opl, i); 590 while (apply && sge->length) { 591 if (sge->length > apply) { 592 u32 len = sge->length - apply; 593 594 get_page(sg_page(sge)); 595 sg_set_page(&tmp, sg_page(sge), len, 596 sge->offset + apply); 597 sge->length = apply; 598 bytes += apply; 599 apply = 0; 600 } else { 601 apply -= sge->length; 602 bytes += sge->length; 603 } 604 605 sk_msg_iter_var_next(i); 606 if (i == msg_opl->sg.end) 607 break; 608 sge = sk_msg_elem(msg_opl, i); 609 } 610 611 msg_opl->sg.end = i; 612 msg_opl->sg.curr = i; 613 msg_opl->sg.copybreak = 0; 614 msg_opl->apply_bytes = 0; 615 msg_opl->sg.size = bytes; 616 617 msg_npl = &new->msg_plaintext; 618 msg_npl->apply_bytes = apply; 619 msg_npl->sg.size = orig_size - bytes; 620 621 j = msg_npl->sg.start; 622 nsge = sk_msg_elem(msg_npl, j); 623 if (tmp.length) { 624 memcpy(nsge, &tmp, sizeof(*nsge)); 625 sk_msg_iter_var_next(j); 626 nsge = sk_msg_elem(msg_npl, j); 627 } 628 629 osge = sk_msg_elem(msg_opl, i); 630 while (osge->length) { 631 memcpy(nsge, osge, sizeof(*nsge)); 632 sg_unmark_end(nsge); 633 sk_msg_iter_var_next(i); 634 sk_msg_iter_var_next(j); 635 if (i == *orig_end) 636 break; 637 osge = sk_msg_elem(msg_opl, i); 638 nsge = sk_msg_elem(msg_npl, j); 639 } 640 641 msg_npl->sg.end = j; 642 msg_npl->sg.curr = j; 643 msg_npl->sg.copybreak = 0; 644 645 *to = new; 646 return 0; 647 } 648 649 static void tls_merge_open_record(struct sock *sk, struct tls_rec *to, 650 struct tls_rec *from, u32 orig_end) 651 { 652 struct sk_msg *msg_npl = &from->msg_plaintext; 653 struct sk_msg *msg_opl = &to->msg_plaintext; 654 struct scatterlist *osge, *nsge; 655 u32 i, j; 656 657 i = msg_opl->sg.end; 658 sk_msg_iter_var_prev(i); 659 j = msg_npl->sg.start; 660 661 osge = sk_msg_elem(msg_opl, i); 662 nsge = sk_msg_elem(msg_npl, j); 663 664 if (sg_page(osge) == sg_page(nsge) && 665 osge->offset + osge->length == nsge->offset) { 666 osge->length += nsge->length; 667 put_page(sg_page(nsge)); 668 } 669 670 msg_opl->sg.end = orig_end; 671 msg_opl->sg.curr = orig_end; 672 msg_opl->sg.copybreak = 0; 673 msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size; 674 msg_opl->sg.size += msg_npl->sg.size; 675 676 sk_msg_free(sk, &to->msg_encrypted); 677 sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted); 678 679 kfree(from); 680 } 681 682 static int tls_push_record(struct sock *sk, int flags, 683 unsigned char record_type) 684 { 685 struct tls_context *tls_ctx = tls_get_ctx(sk); 686 struct tls_prot_info *prot = &tls_ctx->prot_info; 687 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 688 struct tls_rec *rec = ctx->open_rec, *tmp = NULL; 689 u32 i, split_point, orig_end; 690 struct sk_msg *msg_pl, *msg_en; 691 struct aead_request *req; 692 bool split; 693 int rc; 694 695 if (!rec) 696 return 0; 697 698 msg_pl = &rec->msg_plaintext; 699 msg_en = &rec->msg_encrypted; 700 701 split_point = msg_pl->apply_bytes; 702 split = split_point && split_point < msg_pl->sg.size; 703 if (unlikely((!split && 704 msg_pl->sg.size + 705 prot->overhead_size > msg_en->sg.size) || 706 (split && 707 split_point + 708 prot->overhead_size > msg_en->sg.size))) { 709 split = true; 710 split_point = msg_en->sg.size; 711 } 712 if (split) { 713 rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en, 714 split_point, prot->overhead_size, 715 &orig_end); 716 if (rc < 0) 717 return rc; 718 /* This can happen if above tls_split_open_record allocates 719 * a single large encryption buffer instead of two smaller 720 * ones. In this case adjust pointers and continue without 721 * split. 722 */ 723 if (!msg_pl->sg.size) { 724 tls_merge_open_record(sk, rec, tmp, orig_end); 725 msg_pl = &rec->msg_plaintext; 726 msg_en = &rec->msg_encrypted; 727 split = false; 728 } 729 sk_msg_trim(sk, msg_en, msg_pl->sg.size + 730 prot->overhead_size); 731 } 732 733 rec->tx_flags = flags; 734 req = &rec->aead_req; 735 736 i = msg_pl->sg.end; 737 sk_msg_iter_var_prev(i); 738 739 rec->content_type = record_type; 740 if (prot->version == TLS_1_3_VERSION) { 741 /* Add content type to end of message. No padding added */ 742 sg_set_buf(&rec->sg_content_type, &rec->content_type, 1); 743 sg_mark_end(&rec->sg_content_type); 744 sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1, 745 &rec->sg_content_type); 746 } else { 747 sg_mark_end(sk_msg_elem(msg_pl, i)); 748 } 749 750 if (msg_pl->sg.end < msg_pl->sg.start) { 751 sg_chain(&msg_pl->sg.data[msg_pl->sg.start], 752 MAX_SKB_FRAGS - msg_pl->sg.start + 1, 753 msg_pl->sg.data); 754 } 755 756 i = msg_pl->sg.start; 757 sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]); 758 759 i = msg_en->sg.end; 760 sk_msg_iter_var_prev(i); 761 sg_mark_end(sk_msg_elem(msg_en, i)); 762 763 i = msg_en->sg.start; 764 sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]); 765 766 tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size, 767 tls_ctx->tx.rec_seq, record_type, prot); 768 769 tls_fill_prepend(tls_ctx, 770 page_address(sg_page(&msg_en->sg.data[i])) + 771 msg_en->sg.data[i].offset, 772 msg_pl->sg.size + prot->tail_size, 773 record_type); 774 775 tls_ctx->pending_open_record_frags = false; 776 777 rc = tls_do_encryption(sk, tls_ctx, ctx, req, 778 msg_pl->sg.size + prot->tail_size, i); 779 if (rc < 0) { 780 if (rc != -EINPROGRESS) { 781 tls_err_abort(sk, -EBADMSG); 782 if (split) { 783 tls_ctx->pending_open_record_frags = true; 784 tls_merge_open_record(sk, rec, tmp, orig_end); 785 } 786 } 787 ctx->async_capable = 1; 788 return rc; 789 } else if (split) { 790 msg_pl = &tmp->msg_plaintext; 791 msg_en = &tmp->msg_encrypted; 792 sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size); 793 tls_ctx->pending_open_record_frags = true; 794 ctx->open_rec = tmp; 795 } 796 797 return tls_tx_records(sk, flags); 798 } 799 800 static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk, 801 bool full_record, u8 record_type, 802 ssize_t *copied, int flags) 803 { 804 struct tls_context *tls_ctx = tls_get_ctx(sk); 805 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 806 struct sk_msg msg_redir = { }; 807 struct sk_psock *psock; 808 struct sock *sk_redir; 809 struct tls_rec *rec; 810 bool enospc, policy; 811 int err = 0, send; 812 u32 delta = 0; 813 814 policy = !(flags & MSG_SENDPAGE_NOPOLICY); 815 psock = sk_psock_get(sk); 816 if (!psock || !policy) { 817 err = tls_push_record(sk, flags, record_type); 818 if (err && sk->sk_err == EBADMSG) { 819 *copied -= sk_msg_free(sk, msg); 820 tls_free_open_rec(sk); 821 err = -sk->sk_err; 822 } 823 if (psock) 824 sk_psock_put(sk, psock); 825 return err; 826 } 827 more_data: 828 enospc = sk_msg_full(msg); 829 if (psock->eval == __SK_NONE) { 830 delta = msg->sg.size; 831 psock->eval = sk_psock_msg_verdict(sk, psock, msg); 832 delta -= msg->sg.size; 833 } 834 if (msg->cork_bytes && msg->cork_bytes > msg->sg.size && 835 !enospc && !full_record) { 836 err = -ENOSPC; 837 goto out_err; 838 } 839 msg->cork_bytes = 0; 840 send = msg->sg.size; 841 if (msg->apply_bytes && msg->apply_bytes < send) 842 send = msg->apply_bytes; 843 844 switch (psock->eval) { 845 case __SK_PASS: 846 err = tls_push_record(sk, flags, record_type); 847 if (err && sk->sk_err == EBADMSG) { 848 *copied -= sk_msg_free(sk, msg); 849 tls_free_open_rec(sk); 850 err = -sk->sk_err; 851 goto out_err; 852 } 853 break; 854 case __SK_REDIRECT: 855 sk_redir = psock->sk_redir; 856 memcpy(&msg_redir, msg, sizeof(*msg)); 857 if (msg->apply_bytes < send) 858 msg->apply_bytes = 0; 859 else 860 msg->apply_bytes -= send; 861 sk_msg_return_zero(sk, msg, send); 862 msg->sg.size -= send; 863 release_sock(sk); 864 err = tcp_bpf_sendmsg_redir(sk_redir, &msg_redir, send, flags); 865 lock_sock(sk); 866 if (err < 0) { 867 *copied -= sk_msg_free_nocharge(sk, &msg_redir); 868 msg->sg.size = 0; 869 } 870 if (msg->sg.size == 0) 871 tls_free_open_rec(sk); 872 break; 873 case __SK_DROP: 874 default: 875 sk_msg_free_partial(sk, msg, send); 876 if (msg->apply_bytes < send) 877 msg->apply_bytes = 0; 878 else 879 msg->apply_bytes -= send; 880 if (msg->sg.size == 0) 881 tls_free_open_rec(sk); 882 *copied -= (send + delta); 883 err = -EACCES; 884 } 885 886 if (likely(!err)) { 887 bool reset_eval = !ctx->open_rec; 888 889 rec = ctx->open_rec; 890 if (rec) { 891 msg = &rec->msg_plaintext; 892 if (!msg->apply_bytes) 893 reset_eval = true; 894 } 895 if (reset_eval) { 896 psock->eval = __SK_NONE; 897 if (psock->sk_redir) { 898 sock_put(psock->sk_redir); 899 psock->sk_redir = NULL; 900 } 901 } 902 if (rec) 903 goto more_data; 904 } 905 out_err: 906 sk_psock_put(sk, psock); 907 return err; 908 } 909 910 static int tls_sw_push_pending_record(struct sock *sk, int flags) 911 { 912 struct tls_context *tls_ctx = tls_get_ctx(sk); 913 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 914 struct tls_rec *rec = ctx->open_rec; 915 struct sk_msg *msg_pl; 916 size_t copied; 917 918 if (!rec) 919 return 0; 920 921 msg_pl = &rec->msg_plaintext; 922 copied = msg_pl->sg.size; 923 if (!copied) 924 return 0; 925 926 return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA, 927 &copied, flags); 928 } 929 930 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size) 931 { 932 long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT); 933 struct tls_context *tls_ctx = tls_get_ctx(sk); 934 struct tls_prot_info *prot = &tls_ctx->prot_info; 935 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 936 bool async_capable = ctx->async_capable; 937 unsigned char record_type = TLS_RECORD_TYPE_DATA; 938 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter); 939 bool eor = !(msg->msg_flags & MSG_MORE); 940 size_t try_to_copy; 941 ssize_t copied = 0; 942 struct sk_msg *msg_pl, *msg_en; 943 struct tls_rec *rec; 944 int required_size; 945 int num_async = 0; 946 bool full_record; 947 int record_room; 948 int num_zc = 0; 949 int orig_size; 950 int ret = 0; 951 int pending; 952 953 if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | 954 MSG_CMSG_COMPAT)) 955 return -EOPNOTSUPP; 956 957 mutex_lock(&tls_ctx->tx_lock); 958 lock_sock(sk); 959 960 if (unlikely(msg->msg_controllen)) { 961 ret = tls_proccess_cmsg(sk, msg, &record_type); 962 if (ret) { 963 if (ret == -EINPROGRESS) 964 num_async++; 965 else if (ret != -EAGAIN) 966 goto send_end; 967 } 968 } 969 970 while (msg_data_left(msg)) { 971 if (sk->sk_err) { 972 ret = -sk->sk_err; 973 goto send_end; 974 } 975 976 if (ctx->open_rec) 977 rec = ctx->open_rec; 978 else 979 rec = ctx->open_rec = tls_get_rec(sk); 980 if (!rec) { 981 ret = -ENOMEM; 982 goto send_end; 983 } 984 985 msg_pl = &rec->msg_plaintext; 986 msg_en = &rec->msg_encrypted; 987 988 orig_size = msg_pl->sg.size; 989 full_record = false; 990 try_to_copy = msg_data_left(msg); 991 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size; 992 if (try_to_copy >= record_room) { 993 try_to_copy = record_room; 994 full_record = true; 995 } 996 997 required_size = msg_pl->sg.size + try_to_copy + 998 prot->overhead_size; 999 1000 if (!sk_stream_memory_free(sk)) 1001 goto wait_for_sndbuf; 1002 1003 alloc_encrypted: 1004 ret = tls_alloc_encrypted_msg(sk, required_size); 1005 if (ret) { 1006 if (ret != -ENOSPC) 1007 goto wait_for_memory; 1008 1009 /* Adjust try_to_copy according to the amount that was 1010 * actually allocated. The difference is due 1011 * to max sg elements limit 1012 */ 1013 try_to_copy -= required_size - msg_en->sg.size; 1014 full_record = true; 1015 } 1016 1017 if (!is_kvec && (full_record || eor) && !async_capable) { 1018 u32 first = msg_pl->sg.end; 1019 1020 ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter, 1021 msg_pl, try_to_copy); 1022 if (ret) 1023 goto fallback_to_reg_send; 1024 1025 num_zc++; 1026 copied += try_to_copy; 1027 1028 sk_msg_sg_copy_set(msg_pl, first); 1029 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record, 1030 record_type, &copied, 1031 msg->msg_flags); 1032 if (ret) { 1033 if (ret == -EINPROGRESS) 1034 num_async++; 1035 else if (ret == -ENOMEM) 1036 goto wait_for_memory; 1037 else if (ctx->open_rec && ret == -ENOSPC) 1038 goto rollback_iter; 1039 else if (ret != -EAGAIN) 1040 goto send_end; 1041 } 1042 continue; 1043 rollback_iter: 1044 copied -= try_to_copy; 1045 sk_msg_sg_copy_clear(msg_pl, first); 1046 iov_iter_revert(&msg->msg_iter, 1047 msg_pl->sg.size - orig_size); 1048 fallback_to_reg_send: 1049 sk_msg_trim(sk, msg_pl, orig_size); 1050 } 1051 1052 required_size = msg_pl->sg.size + try_to_copy; 1053 1054 ret = tls_clone_plaintext_msg(sk, required_size); 1055 if (ret) { 1056 if (ret != -ENOSPC) 1057 goto send_end; 1058 1059 /* Adjust try_to_copy according to the amount that was 1060 * actually allocated. The difference is due 1061 * to max sg elements limit 1062 */ 1063 try_to_copy -= required_size - msg_pl->sg.size; 1064 full_record = true; 1065 sk_msg_trim(sk, msg_en, 1066 msg_pl->sg.size + prot->overhead_size); 1067 } 1068 1069 if (try_to_copy) { 1070 ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter, 1071 msg_pl, try_to_copy); 1072 if (ret < 0) 1073 goto trim_sgl; 1074 } 1075 1076 /* Open records defined only if successfully copied, otherwise 1077 * we would trim the sg but not reset the open record frags. 1078 */ 1079 tls_ctx->pending_open_record_frags = true; 1080 copied += try_to_copy; 1081 if (full_record || eor) { 1082 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record, 1083 record_type, &copied, 1084 msg->msg_flags); 1085 if (ret) { 1086 if (ret == -EINPROGRESS) 1087 num_async++; 1088 else if (ret == -ENOMEM) 1089 goto wait_for_memory; 1090 else if (ret != -EAGAIN) { 1091 if (ret == -ENOSPC) 1092 ret = 0; 1093 goto send_end; 1094 } 1095 } 1096 } 1097 1098 continue; 1099 1100 wait_for_sndbuf: 1101 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1102 wait_for_memory: 1103 ret = sk_stream_wait_memory(sk, &timeo); 1104 if (ret) { 1105 trim_sgl: 1106 if (ctx->open_rec) 1107 tls_trim_both_msgs(sk, orig_size); 1108 goto send_end; 1109 } 1110 1111 if (ctx->open_rec && msg_en->sg.size < required_size) 1112 goto alloc_encrypted; 1113 } 1114 1115 if (!num_async) { 1116 goto send_end; 1117 } else if (num_zc) { 1118 /* Wait for pending encryptions to get completed */ 1119 spin_lock_bh(&ctx->encrypt_compl_lock); 1120 ctx->async_notify = true; 1121 1122 pending = atomic_read(&ctx->encrypt_pending); 1123 spin_unlock_bh(&ctx->encrypt_compl_lock); 1124 if (pending) 1125 crypto_wait_req(-EINPROGRESS, &ctx->async_wait); 1126 else 1127 reinit_completion(&ctx->async_wait.completion); 1128 1129 /* There can be no concurrent accesses, since we have no 1130 * pending encrypt operations 1131 */ 1132 WRITE_ONCE(ctx->async_notify, false); 1133 1134 if (ctx->async_wait.err) { 1135 ret = ctx->async_wait.err; 1136 copied = 0; 1137 } 1138 } 1139 1140 /* Transmit if any encryptions have completed */ 1141 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) { 1142 cancel_delayed_work(&ctx->tx_work.work); 1143 tls_tx_records(sk, msg->msg_flags); 1144 } 1145 1146 send_end: 1147 ret = sk_stream_error(sk, msg->msg_flags, ret); 1148 1149 release_sock(sk); 1150 mutex_unlock(&tls_ctx->tx_lock); 1151 return copied > 0 ? copied : ret; 1152 } 1153 1154 static int tls_sw_do_sendpage(struct sock *sk, struct page *page, 1155 int offset, size_t size, int flags) 1156 { 1157 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); 1158 struct tls_context *tls_ctx = tls_get_ctx(sk); 1159 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 1160 struct tls_prot_info *prot = &tls_ctx->prot_info; 1161 unsigned char record_type = TLS_RECORD_TYPE_DATA; 1162 struct sk_msg *msg_pl; 1163 struct tls_rec *rec; 1164 int num_async = 0; 1165 ssize_t copied = 0; 1166 bool full_record; 1167 int record_room; 1168 int ret = 0; 1169 bool eor; 1170 1171 eor = !(flags & MSG_SENDPAGE_NOTLAST); 1172 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 1173 1174 /* Call the sk_stream functions to manage the sndbuf mem. */ 1175 while (size > 0) { 1176 size_t copy, required_size; 1177 1178 if (sk->sk_err) { 1179 ret = -sk->sk_err; 1180 goto sendpage_end; 1181 } 1182 1183 if (ctx->open_rec) 1184 rec = ctx->open_rec; 1185 else 1186 rec = ctx->open_rec = tls_get_rec(sk); 1187 if (!rec) { 1188 ret = -ENOMEM; 1189 goto sendpage_end; 1190 } 1191 1192 msg_pl = &rec->msg_plaintext; 1193 1194 full_record = false; 1195 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size; 1196 copy = size; 1197 if (copy >= record_room) { 1198 copy = record_room; 1199 full_record = true; 1200 } 1201 1202 required_size = msg_pl->sg.size + copy + prot->overhead_size; 1203 1204 if (!sk_stream_memory_free(sk)) 1205 goto wait_for_sndbuf; 1206 alloc_payload: 1207 ret = tls_alloc_encrypted_msg(sk, required_size); 1208 if (ret) { 1209 if (ret != -ENOSPC) 1210 goto wait_for_memory; 1211 1212 /* Adjust copy according to the amount that was 1213 * actually allocated. The difference is due 1214 * to max sg elements limit 1215 */ 1216 copy -= required_size - msg_pl->sg.size; 1217 full_record = true; 1218 } 1219 1220 sk_msg_page_add(msg_pl, page, copy, offset); 1221 sk_mem_charge(sk, copy); 1222 1223 offset += copy; 1224 size -= copy; 1225 copied += copy; 1226 1227 tls_ctx->pending_open_record_frags = true; 1228 if (full_record || eor || sk_msg_full(msg_pl)) { 1229 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record, 1230 record_type, &copied, flags); 1231 if (ret) { 1232 if (ret == -EINPROGRESS) 1233 num_async++; 1234 else if (ret == -ENOMEM) 1235 goto wait_for_memory; 1236 else if (ret != -EAGAIN) { 1237 if (ret == -ENOSPC) 1238 ret = 0; 1239 goto sendpage_end; 1240 } 1241 } 1242 } 1243 continue; 1244 wait_for_sndbuf: 1245 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1246 wait_for_memory: 1247 ret = sk_stream_wait_memory(sk, &timeo); 1248 if (ret) { 1249 if (ctx->open_rec) 1250 tls_trim_both_msgs(sk, msg_pl->sg.size); 1251 goto sendpage_end; 1252 } 1253 1254 if (ctx->open_rec) 1255 goto alloc_payload; 1256 } 1257 1258 if (num_async) { 1259 /* Transmit if any encryptions have completed */ 1260 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) { 1261 cancel_delayed_work(&ctx->tx_work.work); 1262 tls_tx_records(sk, flags); 1263 } 1264 } 1265 sendpage_end: 1266 ret = sk_stream_error(sk, flags, ret); 1267 return copied > 0 ? copied : ret; 1268 } 1269 1270 int tls_sw_sendpage_locked(struct sock *sk, struct page *page, 1271 int offset, size_t size, int flags) 1272 { 1273 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | 1274 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY | 1275 MSG_NO_SHARED_FRAGS)) 1276 return -EOPNOTSUPP; 1277 1278 return tls_sw_do_sendpage(sk, page, offset, size, flags); 1279 } 1280 1281 int tls_sw_sendpage(struct sock *sk, struct page *page, 1282 int offset, size_t size, int flags) 1283 { 1284 struct tls_context *tls_ctx = tls_get_ctx(sk); 1285 int ret; 1286 1287 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | 1288 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY)) 1289 return -EOPNOTSUPP; 1290 1291 mutex_lock(&tls_ctx->tx_lock); 1292 lock_sock(sk); 1293 ret = tls_sw_do_sendpage(sk, page, offset, size, flags); 1294 release_sock(sk); 1295 mutex_unlock(&tls_ctx->tx_lock); 1296 return ret; 1297 } 1298 1299 static struct sk_buff *tls_wait_data(struct sock *sk, struct sk_psock *psock, 1300 bool nonblock, long timeo, int *err) 1301 { 1302 struct tls_context *tls_ctx = tls_get_ctx(sk); 1303 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 1304 struct sk_buff *skb; 1305 DEFINE_WAIT_FUNC(wait, woken_wake_function); 1306 1307 while (!(skb = ctx->recv_pkt) && sk_psock_queue_empty(psock)) { 1308 if (sk->sk_err) { 1309 *err = sock_error(sk); 1310 return NULL; 1311 } 1312 1313 if (!skb_queue_empty(&sk->sk_receive_queue)) { 1314 __strp_unpause(&ctx->strp); 1315 if (ctx->recv_pkt) 1316 return ctx->recv_pkt; 1317 } 1318 1319 if (sk->sk_shutdown & RCV_SHUTDOWN) 1320 return NULL; 1321 1322 if (sock_flag(sk, SOCK_DONE)) 1323 return NULL; 1324 1325 if (nonblock || !timeo) { 1326 *err = -EAGAIN; 1327 return NULL; 1328 } 1329 1330 add_wait_queue(sk_sleep(sk), &wait); 1331 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 1332 sk_wait_event(sk, &timeo, 1333 ctx->recv_pkt != skb || 1334 !sk_psock_queue_empty(psock), 1335 &wait); 1336 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 1337 remove_wait_queue(sk_sleep(sk), &wait); 1338 1339 /* Handle signals */ 1340 if (signal_pending(current)) { 1341 *err = sock_intr_errno(timeo); 1342 return NULL; 1343 } 1344 } 1345 1346 return skb; 1347 } 1348 1349 static int tls_setup_from_iter(struct sock *sk, struct iov_iter *from, 1350 int length, int *pages_used, 1351 unsigned int *size_used, 1352 struct scatterlist *to, 1353 int to_max_pages) 1354 { 1355 int rc = 0, i = 0, num_elem = *pages_used, maxpages; 1356 struct page *pages[MAX_SKB_FRAGS]; 1357 unsigned int size = *size_used; 1358 ssize_t copied, use; 1359 size_t offset; 1360 1361 while (length > 0) { 1362 i = 0; 1363 maxpages = to_max_pages - num_elem; 1364 if (maxpages == 0) { 1365 rc = -EFAULT; 1366 goto out; 1367 } 1368 copied = iov_iter_get_pages(from, pages, 1369 length, 1370 maxpages, &offset); 1371 if (copied <= 0) { 1372 rc = -EFAULT; 1373 goto out; 1374 } 1375 1376 iov_iter_advance(from, copied); 1377 1378 length -= copied; 1379 size += copied; 1380 while (copied) { 1381 use = min_t(int, copied, PAGE_SIZE - offset); 1382 1383 sg_set_page(&to[num_elem], 1384 pages[i], use, offset); 1385 sg_unmark_end(&to[num_elem]); 1386 /* We do not uncharge memory from this API */ 1387 1388 offset = 0; 1389 copied -= use; 1390 1391 i++; 1392 num_elem++; 1393 } 1394 } 1395 /* Mark the end in the last sg entry if newly added */ 1396 if (num_elem > *pages_used) 1397 sg_mark_end(&to[num_elem - 1]); 1398 out: 1399 if (rc) 1400 iov_iter_revert(from, size - *size_used); 1401 *size_used = size; 1402 *pages_used = num_elem; 1403 1404 return rc; 1405 } 1406 1407 /* This function decrypts the input skb into either out_iov or in out_sg 1408 * or in skb buffers itself. The input parameter 'zc' indicates if 1409 * zero-copy mode needs to be tried or not. With zero-copy mode, either 1410 * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are 1411 * NULL, then the decryption happens inside skb buffers itself, i.e. 1412 * zero-copy gets disabled and 'zc' is updated. 1413 */ 1414 1415 static int decrypt_internal(struct sock *sk, struct sk_buff *skb, 1416 struct iov_iter *out_iov, 1417 struct scatterlist *out_sg, 1418 int *chunk, bool *zc, bool async) 1419 { 1420 struct tls_context *tls_ctx = tls_get_ctx(sk); 1421 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 1422 struct tls_prot_info *prot = &tls_ctx->prot_info; 1423 struct strp_msg *rxm = strp_msg(skb); 1424 int n_sgin, n_sgout, nsg, mem_size, aead_size, err, pages = 0; 1425 struct aead_request *aead_req; 1426 struct sk_buff *unused; 1427 u8 *aad, *iv, *mem = NULL; 1428 struct scatterlist *sgin = NULL; 1429 struct scatterlist *sgout = NULL; 1430 const int data_len = rxm->full_len - prot->overhead_size + 1431 prot->tail_size; 1432 int iv_offset = 0; 1433 1434 if (*zc && (out_iov || out_sg)) { 1435 if (out_iov) 1436 n_sgout = iov_iter_npages(out_iov, INT_MAX) + 1; 1437 else 1438 n_sgout = sg_nents(out_sg); 1439 n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size, 1440 rxm->full_len - prot->prepend_size); 1441 } else { 1442 n_sgout = 0; 1443 *zc = false; 1444 n_sgin = skb_cow_data(skb, 0, &unused); 1445 } 1446 1447 if (n_sgin < 1) 1448 return -EBADMSG; 1449 1450 /* Increment to accommodate AAD */ 1451 n_sgin = n_sgin + 1; 1452 1453 nsg = n_sgin + n_sgout; 1454 1455 aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv); 1456 mem_size = aead_size + (nsg * sizeof(struct scatterlist)); 1457 mem_size = mem_size + prot->aad_size; 1458 mem_size = mem_size + crypto_aead_ivsize(ctx->aead_recv); 1459 1460 /* Allocate a single block of memory which contains 1461 * aead_req || sgin[] || sgout[] || aad || iv. 1462 * This order achieves correct alignment for aead_req, sgin, sgout. 1463 */ 1464 mem = kmalloc(mem_size, sk->sk_allocation); 1465 if (!mem) 1466 return -ENOMEM; 1467 1468 /* Segment the allocated memory */ 1469 aead_req = (struct aead_request *)mem; 1470 sgin = (struct scatterlist *)(mem + aead_size); 1471 sgout = sgin + n_sgin; 1472 aad = (u8 *)(sgout + n_sgout); 1473 iv = aad + prot->aad_size; 1474 1475 /* For CCM based ciphers, first byte of nonce+iv is a constant */ 1476 switch (prot->cipher_type) { 1477 case TLS_CIPHER_AES_CCM_128: 1478 iv[0] = TLS_AES_CCM_IV_B0_BYTE; 1479 iv_offset = 1; 1480 break; 1481 case TLS_CIPHER_SM4_CCM: 1482 iv[0] = TLS_SM4_CCM_IV_B0_BYTE; 1483 iv_offset = 1; 1484 break; 1485 } 1486 1487 /* Prepare IV */ 1488 err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE, 1489 iv + iv_offset + prot->salt_size, 1490 prot->iv_size); 1491 if (err < 0) { 1492 kfree(mem); 1493 return err; 1494 } 1495 if (prot->version == TLS_1_3_VERSION || 1496 prot->cipher_type == TLS_CIPHER_CHACHA20_POLY1305) 1497 memcpy(iv + iv_offset, tls_ctx->rx.iv, 1498 crypto_aead_ivsize(ctx->aead_recv)); 1499 else 1500 memcpy(iv + iv_offset, tls_ctx->rx.iv, prot->salt_size); 1501 1502 xor_iv_with_seq(prot, iv + iv_offset, tls_ctx->rx.rec_seq); 1503 1504 /* Prepare AAD */ 1505 tls_make_aad(aad, rxm->full_len - prot->overhead_size + 1506 prot->tail_size, 1507 tls_ctx->rx.rec_seq, ctx->control, prot); 1508 1509 /* Prepare sgin */ 1510 sg_init_table(sgin, n_sgin); 1511 sg_set_buf(&sgin[0], aad, prot->aad_size); 1512 err = skb_to_sgvec(skb, &sgin[1], 1513 rxm->offset + prot->prepend_size, 1514 rxm->full_len - prot->prepend_size); 1515 if (err < 0) { 1516 kfree(mem); 1517 return err; 1518 } 1519 1520 if (n_sgout) { 1521 if (out_iov) { 1522 sg_init_table(sgout, n_sgout); 1523 sg_set_buf(&sgout[0], aad, prot->aad_size); 1524 1525 *chunk = 0; 1526 err = tls_setup_from_iter(sk, out_iov, data_len, 1527 &pages, chunk, &sgout[1], 1528 (n_sgout - 1)); 1529 if (err < 0) 1530 goto fallback_to_reg_recv; 1531 } else if (out_sg) { 1532 memcpy(sgout, out_sg, n_sgout * sizeof(*sgout)); 1533 } else { 1534 goto fallback_to_reg_recv; 1535 } 1536 } else { 1537 fallback_to_reg_recv: 1538 sgout = sgin; 1539 pages = 0; 1540 *chunk = data_len; 1541 *zc = false; 1542 } 1543 1544 /* Prepare and submit AEAD request */ 1545 err = tls_do_decryption(sk, skb, sgin, sgout, iv, 1546 data_len, aead_req, async); 1547 if (err == -EINPROGRESS) 1548 return err; 1549 1550 /* Release the pages in case iov was mapped to pages */ 1551 for (; pages > 0; pages--) 1552 put_page(sg_page(&sgout[pages])); 1553 1554 kfree(mem); 1555 return err; 1556 } 1557 1558 static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb, 1559 struct iov_iter *dest, int *chunk, bool *zc, 1560 bool async) 1561 { 1562 struct tls_context *tls_ctx = tls_get_ctx(sk); 1563 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 1564 struct tls_prot_info *prot = &tls_ctx->prot_info; 1565 struct strp_msg *rxm = strp_msg(skb); 1566 int pad, err = 0; 1567 1568 if (!ctx->decrypted) { 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 } 1574 1575 /* Still not decrypted after tls_device */ 1576 if (!ctx->decrypted) { 1577 err = decrypt_internal(sk, skb, dest, NULL, chunk, zc, 1578 async); 1579 if (err < 0) { 1580 if (err == -EINPROGRESS) 1581 tls_advance_record_sn(sk, prot, 1582 &tls_ctx->rx); 1583 else if (err == -EBADMSG) 1584 TLS_INC_STATS(sock_net(sk), 1585 LINUX_MIB_TLSDECRYPTERROR); 1586 return err; 1587 } 1588 } else { 1589 *zc = false; 1590 } 1591 1592 pad = padding_length(ctx, prot, skb); 1593 if (pad < 0) 1594 return pad; 1595 1596 rxm->full_len -= pad; 1597 rxm->offset += prot->prepend_size; 1598 rxm->full_len -= prot->overhead_size; 1599 tls_advance_record_sn(sk, prot, &tls_ctx->rx); 1600 ctx->decrypted = 1; 1601 ctx->saved_data_ready(sk); 1602 } else { 1603 *zc = false; 1604 } 1605 1606 return err; 1607 } 1608 1609 int decrypt_skb(struct sock *sk, struct sk_buff *skb, 1610 struct scatterlist *sgout) 1611 { 1612 bool zc = true; 1613 int chunk; 1614 1615 return decrypt_internal(sk, skb, NULL, sgout, &chunk, &zc, false); 1616 } 1617 1618 static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb, 1619 unsigned int len) 1620 { 1621 struct tls_context *tls_ctx = tls_get_ctx(sk); 1622 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 1623 1624 if (skb) { 1625 struct strp_msg *rxm = strp_msg(skb); 1626 1627 if (len < rxm->full_len) { 1628 rxm->offset += len; 1629 rxm->full_len -= len; 1630 return false; 1631 } 1632 consume_skb(skb); 1633 } 1634 1635 /* Finished with message */ 1636 ctx->recv_pkt = NULL; 1637 __strp_unpause(&ctx->strp); 1638 1639 return true; 1640 } 1641 1642 /* This function traverses the rx_list in tls receive context to copies the 1643 * decrypted records into the buffer provided by caller zero copy is not 1644 * true. Further, the records are removed from the rx_list if it is not a peek 1645 * case and the record has been consumed completely. 1646 */ 1647 static int process_rx_list(struct tls_sw_context_rx *ctx, 1648 struct msghdr *msg, 1649 u8 *control, 1650 bool *cmsg, 1651 size_t skip, 1652 size_t len, 1653 bool zc, 1654 bool is_peek) 1655 { 1656 struct sk_buff *skb = skb_peek(&ctx->rx_list); 1657 u8 ctrl = *control; 1658 u8 msgc = *cmsg; 1659 struct tls_msg *tlm; 1660 ssize_t copied = 0; 1661 1662 /* Set the record type in 'control' if caller didn't pass it */ 1663 if (!ctrl && skb) { 1664 tlm = tls_msg(skb); 1665 ctrl = tlm->control; 1666 } 1667 1668 while (skip && skb) { 1669 struct strp_msg *rxm = strp_msg(skb); 1670 tlm = tls_msg(skb); 1671 1672 /* Cannot process a record of different type */ 1673 if (ctrl != tlm->control) 1674 return 0; 1675 1676 if (skip < rxm->full_len) 1677 break; 1678 1679 skip = skip - rxm->full_len; 1680 skb = skb_peek_next(skb, &ctx->rx_list); 1681 } 1682 1683 while (len && skb) { 1684 struct sk_buff *next_skb; 1685 struct strp_msg *rxm = strp_msg(skb); 1686 int chunk = min_t(unsigned int, rxm->full_len - skip, len); 1687 1688 tlm = tls_msg(skb); 1689 1690 /* Cannot process a record of different type */ 1691 if (ctrl != tlm->control) 1692 return 0; 1693 1694 /* Set record type if not already done. For a non-data record, 1695 * do not proceed if record type could not be copied. 1696 */ 1697 if (!msgc) { 1698 int cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE, 1699 sizeof(ctrl), &ctrl); 1700 msgc = true; 1701 if (ctrl != TLS_RECORD_TYPE_DATA) { 1702 if (cerr || msg->msg_flags & MSG_CTRUNC) 1703 return -EIO; 1704 1705 *cmsg = msgc; 1706 } 1707 } 1708 1709 if (!zc || (rxm->full_len - skip) > len) { 1710 int err = skb_copy_datagram_msg(skb, rxm->offset + skip, 1711 msg, chunk); 1712 if (err < 0) 1713 return err; 1714 } 1715 1716 len = len - chunk; 1717 copied = copied + chunk; 1718 1719 /* Consume the data from record if it is non-peek case*/ 1720 if (!is_peek) { 1721 rxm->offset = rxm->offset + chunk; 1722 rxm->full_len = rxm->full_len - chunk; 1723 1724 /* Return if there is unconsumed data in the record */ 1725 if (rxm->full_len - skip) 1726 break; 1727 } 1728 1729 /* The remaining skip-bytes must lie in 1st record in rx_list. 1730 * So from the 2nd record, 'skip' should be 0. 1731 */ 1732 skip = 0; 1733 1734 if (msg) 1735 msg->msg_flags |= MSG_EOR; 1736 1737 next_skb = skb_peek_next(skb, &ctx->rx_list); 1738 1739 if (!is_peek) { 1740 skb_unlink(skb, &ctx->rx_list); 1741 consume_skb(skb); 1742 } 1743 1744 skb = next_skb; 1745 } 1746 1747 *control = ctrl; 1748 return copied; 1749 } 1750 1751 int tls_sw_recvmsg(struct sock *sk, 1752 struct msghdr *msg, 1753 size_t len, 1754 int nonblock, 1755 int flags, 1756 int *addr_len) 1757 { 1758 struct tls_context *tls_ctx = tls_get_ctx(sk); 1759 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 1760 struct tls_prot_info *prot = &tls_ctx->prot_info; 1761 struct sk_psock *psock; 1762 unsigned char control = 0; 1763 ssize_t decrypted = 0; 1764 struct strp_msg *rxm; 1765 struct tls_msg *tlm; 1766 struct sk_buff *skb; 1767 ssize_t copied = 0; 1768 bool cmsg = false; 1769 int target, err = 0; 1770 long timeo; 1771 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter); 1772 bool is_peek = flags & MSG_PEEK; 1773 bool bpf_strp_enabled; 1774 int num_async = 0; 1775 int pending; 1776 1777 flags |= nonblock; 1778 1779 if (unlikely(flags & MSG_ERRQUEUE)) 1780 return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR); 1781 1782 psock = sk_psock_get(sk); 1783 lock_sock(sk); 1784 bpf_strp_enabled = sk_psock_strp_enabled(psock); 1785 1786 /* Process pending decrypted records. It must be non-zero-copy */ 1787 err = process_rx_list(ctx, msg, &control, &cmsg, 0, len, false, 1788 is_peek); 1789 if (err < 0) { 1790 tls_err_abort(sk, err); 1791 goto end; 1792 } else { 1793 copied = err; 1794 } 1795 1796 if (len <= copied) 1797 goto recv_end; 1798 1799 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); 1800 len = len - copied; 1801 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1802 1803 while (len && (decrypted + copied < target || ctx->recv_pkt)) { 1804 bool retain_skb = false; 1805 bool zc = false; 1806 int to_decrypt; 1807 int chunk = 0; 1808 bool async_capable; 1809 bool async = false; 1810 1811 skb = tls_wait_data(sk, psock, flags & MSG_DONTWAIT, timeo, &err); 1812 if (!skb) { 1813 if (psock) { 1814 int ret = sk_msg_recvmsg(sk, psock, msg, len, 1815 flags); 1816 1817 if (ret > 0) { 1818 decrypted += ret; 1819 len -= ret; 1820 continue; 1821 } 1822 } 1823 goto recv_end; 1824 } else { 1825 tlm = tls_msg(skb); 1826 if (prot->version == TLS_1_3_VERSION) 1827 tlm->control = 0; 1828 else 1829 tlm->control = ctx->control; 1830 } 1831 1832 rxm = strp_msg(skb); 1833 1834 to_decrypt = rxm->full_len - prot->overhead_size; 1835 1836 if (to_decrypt <= len && !is_kvec && !is_peek && 1837 ctx->control == TLS_RECORD_TYPE_DATA && 1838 prot->version != TLS_1_3_VERSION && 1839 !bpf_strp_enabled) 1840 zc = true; 1841 1842 /* Do not use async mode if record is non-data */ 1843 if (ctx->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled) 1844 async_capable = ctx->async_capable; 1845 else 1846 async_capable = false; 1847 1848 err = decrypt_skb_update(sk, skb, &msg->msg_iter, 1849 &chunk, &zc, async_capable); 1850 if (err < 0 && err != -EINPROGRESS) { 1851 tls_err_abort(sk, -EBADMSG); 1852 goto recv_end; 1853 } 1854 1855 if (err == -EINPROGRESS) { 1856 async = true; 1857 num_async++; 1858 } else if (prot->version == TLS_1_3_VERSION) { 1859 tlm->control = ctx->control; 1860 } 1861 1862 /* If the type of records being processed is not known yet, 1863 * set it to record type just dequeued. If it is already known, 1864 * but does not match the record type just dequeued, go to end. 1865 * We always get record type here since for tls1.2, record type 1866 * is known just after record is dequeued from stream parser. 1867 * For tls1.3, we disable async. 1868 */ 1869 1870 if (!control) 1871 control = tlm->control; 1872 else if (control != tlm->control) 1873 goto recv_end; 1874 1875 if (!cmsg) { 1876 int cerr; 1877 1878 cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE, 1879 sizeof(control), &control); 1880 cmsg = true; 1881 if (control != TLS_RECORD_TYPE_DATA) { 1882 if (cerr || msg->msg_flags & MSG_CTRUNC) { 1883 err = -EIO; 1884 goto recv_end; 1885 } 1886 } 1887 } 1888 1889 if (async) 1890 goto pick_next_record; 1891 1892 if (!zc) { 1893 if (bpf_strp_enabled) { 1894 err = sk_psock_tls_strp_read(psock, skb); 1895 if (err != __SK_PASS) { 1896 rxm->offset = rxm->offset + rxm->full_len; 1897 rxm->full_len = 0; 1898 if (err == __SK_DROP) 1899 consume_skb(skb); 1900 ctx->recv_pkt = NULL; 1901 __strp_unpause(&ctx->strp); 1902 continue; 1903 } 1904 } 1905 1906 if (rxm->full_len > len) { 1907 retain_skb = true; 1908 chunk = len; 1909 } else { 1910 chunk = rxm->full_len; 1911 } 1912 1913 err = skb_copy_datagram_msg(skb, rxm->offset, 1914 msg, chunk); 1915 if (err < 0) 1916 goto recv_end; 1917 1918 if (!is_peek) { 1919 rxm->offset = rxm->offset + chunk; 1920 rxm->full_len = rxm->full_len - chunk; 1921 } 1922 } 1923 1924 pick_next_record: 1925 if (chunk > len) 1926 chunk = len; 1927 1928 decrypted += chunk; 1929 len -= chunk; 1930 1931 /* For async or peek case, queue the current skb */ 1932 if (async || is_peek || retain_skb) { 1933 skb_queue_tail(&ctx->rx_list, skb); 1934 skb = NULL; 1935 } 1936 1937 if (tls_sw_advance_skb(sk, skb, chunk)) { 1938 /* Return full control message to 1939 * userspace before trying to parse 1940 * another message type 1941 */ 1942 msg->msg_flags |= MSG_EOR; 1943 if (control != TLS_RECORD_TYPE_DATA) 1944 goto recv_end; 1945 } else { 1946 break; 1947 } 1948 } 1949 1950 recv_end: 1951 if (num_async) { 1952 /* Wait for all previously submitted records to be decrypted */ 1953 spin_lock_bh(&ctx->decrypt_compl_lock); 1954 ctx->async_notify = true; 1955 pending = atomic_read(&ctx->decrypt_pending); 1956 spin_unlock_bh(&ctx->decrypt_compl_lock); 1957 if (pending) { 1958 err = crypto_wait_req(-EINPROGRESS, &ctx->async_wait); 1959 if (err) { 1960 /* one of async decrypt failed */ 1961 tls_err_abort(sk, err); 1962 copied = 0; 1963 decrypted = 0; 1964 goto end; 1965 } 1966 } else { 1967 reinit_completion(&ctx->async_wait.completion); 1968 } 1969 1970 /* There can be no concurrent accesses, since we have no 1971 * pending decrypt operations 1972 */ 1973 WRITE_ONCE(ctx->async_notify, false); 1974 1975 /* Drain records from the rx_list & copy if required */ 1976 if (is_peek || is_kvec) 1977 err = process_rx_list(ctx, msg, &control, &cmsg, copied, 1978 decrypted, false, is_peek); 1979 else 1980 err = process_rx_list(ctx, msg, &control, &cmsg, 0, 1981 decrypted, true, is_peek); 1982 if (err < 0) { 1983 tls_err_abort(sk, err); 1984 copied = 0; 1985 goto end; 1986 } 1987 } 1988 1989 copied += decrypted; 1990 1991 end: 1992 release_sock(sk); 1993 sk_defer_free_flush(sk); 1994 if (psock) 1995 sk_psock_put(sk, psock); 1996 return copied ? : err; 1997 } 1998 1999 ssize_t tls_sw_splice_read(struct socket *sock, loff_t *ppos, 2000 struct pipe_inode_info *pipe, 2001 size_t len, unsigned int flags) 2002 { 2003 struct tls_context *tls_ctx = tls_get_ctx(sock->sk); 2004 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2005 struct strp_msg *rxm = NULL; 2006 struct sock *sk = sock->sk; 2007 struct sk_buff *skb; 2008 ssize_t copied = 0; 2009 bool from_queue; 2010 int err = 0; 2011 long timeo; 2012 int chunk; 2013 bool zc = false; 2014 2015 lock_sock(sk); 2016 2017 timeo = sock_rcvtimeo(sk, flags & SPLICE_F_NONBLOCK); 2018 2019 from_queue = !skb_queue_empty(&ctx->rx_list); 2020 if (from_queue) { 2021 skb = __skb_dequeue(&ctx->rx_list); 2022 } else { 2023 skb = tls_wait_data(sk, NULL, flags & SPLICE_F_NONBLOCK, timeo, 2024 &err); 2025 if (!skb) 2026 goto splice_read_end; 2027 2028 err = decrypt_skb_update(sk, skb, NULL, &chunk, &zc, false); 2029 if (err < 0) { 2030 tls_err_abort(sk, -EBADMSG); 2031 goto splice_read_end; 2032 } 2033 } 2034 2035 /* splice does not support reading control messages */ 2036 if (ctx->control != TLS_RECORD_TYPE_DATA) { 2037 err = -EINVAL; 2038 goto splice_read_end; 2039 } 2040 2041 rxm = strp_msg(skb); 2042 2043 chunk = min_t(unsigned int, rxm->full_len, len); 2044 copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags); 2045 if (copied < 0) 2046 goto splice_read_end; 2047 2048 if (!from_queue) { 2049 ctx->recv_pkt = NULL; 2050 __strp_unpause(&ctx->strp); 2051 } 2052 if (chunk < rxm->full_len) { 2053 __skb_queue_head(&ctx->rx_list, skb); 2054 rxm->offset += len; 2055 rxm->full_len -= len; 2056 } else { 2057 consume_skb(skb); 2058 } 2059 2060 splice_read_end: 2061 release_sock(sk); 2062 sk_defer_free_flush(sk); 2063 return copied ? : err; 2064 } 2065 2066 bool tls_sw_sock_is_readable(struct sock *sk) 2067 { 2068 struct tls_context *tls_ctx = tls_get_ctx(sk); 2069 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2070 bool ingress_empty = true; 2071 struct sk_psock *psock; 2072 2073 rcu_read_lock(); 2074 psock = sk_psock(sk); 2075 if (psock) 2076 ingress_empty = list_empty(&psock->ingress_msg); 2077 rcu_read_unlock(); 2078 2079 return !ingress_empty || ctx->recv_pkt || 2080 !skb_queue_empty(&ctx->rx_list); 2081 } 2082 2083 static int tls_read_size(struct strparser *strp, struct sk_buff *skb) 2084 { 2085 struct tls_context *tls_ctx = tls_get_ctx(strp->sk); 2086 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2087 struct tls_prot_info *prot = &tls_ctx->prot_info; 2088 char header[TLS_HEADER_SIZE + MAX_IV_SIZE]; 2089 struct strp_msg *rxm = strp_msg(skb); 2090 size_t cipher_overhead; 2091 size_t data_len = 0; 2092 int ret; 2093 2094 /* Verify that we have a full TLS header, or wait for more data */ 2095 if (rxm->offset + prot->prepend_size > skb->len) 2096 return 0; 2097 2098 /* Sanity-check size of on-stack buffer. */ 2099 if (WARN_ON(prot->prepend_size > sizeof(header))) { 2100 ret = -EINVAL; 2101 goto read_failure; 2102 } 2103 2104 /* Linearize header to local buffer */ 2105 ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size); 2106 2107 if (ret < 0) 2108 goto read_failure; 2109 2110 ctx->control = header[0]; 2111 2112 data_len = ((header[4] & 0xFF) | (header[3] << 8)); 2113 2114 cipher_overhead = prot->tag_size; 2115 if (prot->version != TLS_1_3_VERSION && 2116 prot->cipher_type != TLS_CIPHER_CHACHA20_POLY1305) 2117 cipher_overhead += prot->iv_size; 2118 2119 if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead + 2120 prot->tail_size) { 2121 ret = -EMSGSIZE; 2122 goto read_failure; 2123 } 2124 if (data_len < cipher_overhead) { 2125 ret = -EBADMSG; 2126 goto read_failure; 2127 } 2128 2129 /* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */ 2130 if (header[1] != TLS_1_2_VERSION_MINOR || 2131 header[2] != TLS_1_2_VERSION_MAJOR) { 2132 ret = -EINVAL; 2133 goto read_failure; 2134 } 2135 2136 tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE, 2137 TCP_SKB_CB(skb)->seq + rxm->offset); 2138 return data_len + TLS_HEADER_SIZE; 2139 2140 read_failure: 2141 tls_err_abort(strp->sk, ret); 2142 2143 return ret; 2144 } 2145 2146 static void tls_queue(struct strparser *strp, struct sk_buff *skb) 2147 { 2148 struct tls_context *tls_ctx = tls_get_ctx(strp->sk); 2149 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2150 2151 ctx->decrypted = 0; 2152 2153 ctx->recv_pkt = skb; 2154 strp_pause(strp); 2155 2156 ctx->saved_data_ready(strp->sk); 2157 } 2158 2159 static void tls_data_ready(struct sock *sk) 2160 { 2161 struct tls_context *tls_ctx = tls_get_ctx(sk); 2162 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2163 struct sk_psock *psock; 2164 2165 strp_data_ready(&ctx->strp); 2166 2167 psock = sk_psock_get(sk); 2168 if (psock) { 2169 if (!list_empty(&psock->ingress_msg)) 2170 ctx->saved_data_ready(sk); 2171 sk_psock_put(sk, psock); 2172 } 2173 } 2174 2175 void tls_sw_cancel_work_tx(struct tls_context *tls_ctx) 2176 { 2177 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 2178 2179 set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask); 2180 set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask); 2181 cancel_delayed_work_sync(&ctx->tx_work.work); 2182 } 2183 2184 void tls_sw_release_resources_tx(struct sock *sk) 2185 { 2186 struct tls_context *tls_ctx = tls_get_ctx(sk); 2187 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 2188 struct tls_rec *rec, *tmp; 2189 int pending; 2190 2191 /* Wait for any pending async encryptions to complete */ 2192 spin_lock_bh(&ctx->encrypt_compl_lock); 2193 ctx->async_notify = true; 2194 pending = atomic_read(&ctx->encrypt_pending); 2195 spin_unlock_bh(&ctx->encrypt_compl_lock); 2196 2197 if (pending) 2198 crypto_wait_req(-EINPROGRESS, &ctx->async_wait); 2199 2200 tls_tx_records(sk, -1); 2201 2202 /* Free up un-sent records in tx_list. First, free 2203 * the partially sent record if any at head of tx_list. 2204 */ 2205 if (tls_ctx->partially_sent_record) { 2206 tls_free_partial_record(sk, tls_ctx); 2207 rec = list_first_entry(&ctx->tx_list, 2208 struct tls_rec, list); 2209 list_del(&rec->list); 2210 sk_msg_free(sk, &rec->msg_plaintext); 2211 kfree(rec); 2212 } 2213 2214 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) { 2215 list_del(&rec->list); 2216 sk_msg_free(sk, &rec->msg_encrypted); 2217 sk_msg_free(sk, &rec->msg_plaintext); 2218 kfree(rec); 2219 } 2220 2221 crypto_free_aead(ctx->aead_send); 2222 tls_free_open_rec(sk); 2223 } 2224 2225 void tls_sw_free_ctx_tx(struct tls_context *tls_ctx) 2226 { 2227 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); 2228 2229 kfree(ctx); 2230 } 2231 2232 void tls_sw_release_resources_rx(struct sock *sk) 2233 { 2234 struct tls_context *tls_ctx = tls_get_ctx(sk); 2235 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2236 2237 kfree(tls_ctx->rx.rec_seq); 2238 kfree(tls_ctx->rx.iv); 2239 2240 if (ctx->aead_recv) { 2241 kfree_skb(ctx->recv_pkt); 2242 ctx->recv_pkt = NULL; 2243 skb_queue_purge(&ctx->rx_list); 2244 crypto_free_aead(ctx->aead_recv); 2245 strp_stop(&ctx->strp); 2246 /* If tls_sw_strparser_arm() was not called (cleanup paths) 2247 * we still want to strp_stop(), but sk->sk_data_ready was 2248 * never swapped. 2249 */ 2250 if (ctx->saved_data_ready) { 2251 write_lock_bh(&sk->sk_callback_lock); 2252 sk->sk_data_ready = ctx->saved_data_ready; 2253 write_unlock_bh(&sk->sk_callback_lock); 2254 } 2255 } 2256 } 2257 2258 void tls_sw_strparser_done(struct tls_context *tls_ctx) 2259 { 2260 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2261 2262 strp_done(&ctx->strp); 2263 } 2264 2265 void tls_sw_free_ctx_rx(struct tls_context *tls_ctx) 2266 { 2267 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); 2268 2269 kfree(ctx); 2270 } 2271 2272 void tls_sw_free_resources_rx(struct sock *sk) 2273 { 2274 struct tls_context *tls_ctx = tls_get_ctx(sk); 2275 2276 tls_sw_release_resources_rx(sk); 2277 tls_sw_free_ctx_rx(tls_ctx); 2278 } 2279 2280 /* The work handler to transmitt the encrypted records in tx_list */ 2281 static void tx_work_handler(struct work_struct *work) 2282 { 2283 struct delayed_work *delayed_work = to_delayed_work(work); 2284 struct tx_work *tx_work = container_of(delayed_work, 2285 struct tx_work, work); 2286 struct sock *sk = tx_work->sk; 2287 struct tls_context *tls_ctx = tls_get_ctx(sk); 2288 struct tls_sw_context_tx *ctx; 2289 2290 if (unlikely(!tls_ctx)) 2291 return; 2292 2293 ctx = tls_sw_ctx_tx(tls_ctx); 2294 if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask)) 2295 return; 2296 2297 if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) 2298 return; 2299 mutex_lock(&tls_ctx->tx_lock); 2300 lock_sock(sk); 2301 tls_tx_records(sk, -1); 2302 release_sock(sk); 2303 mutex_unlock(&tls_ctx->tx_lock); 2304 } 2305 2306 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx) 2307 { 2308 struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx); 2309 2310 /* Schedule the transmission if tx list is ready */ 2311 if (is_tx_ready(tx_ctx) && 2312 !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask)) 2313 schedule_delayed_work(&tx_ctx->tx_work.work, 0); 2314 } 2315 2316 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx) 2317 { 2318 struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx); 2319 2320 write_lock_bh(&sk->sk_callback_lock); 2321 rx_ctx->saved_data_ready = sk->sk_data_ready; 2322 sk->sk_data_ready = tls_data_ready; 2323 write_unlock_bh(&sk->sk_callback_lock); 2324 2325 strp_check_rcv(&rx_ctx->strp); 2326 } 2327 2328 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx) 2329 { 2330 struct tls_context *tls_ctx = tls_get_ctx(sk); 2331 struct tls_prot_info *prot = &tls_ctx->prot_info; 2332 struct tls_crypto_info *crypto_info; 2333 struct tls_sw_context_tx *sw_ctx_tx = NULL; 2334 struct tls_sw_context_rx *sw_ctx_rx = NULL; 2335 struct cipher_context *cctx; 2336 struct crypto_aead **aead; 2337 struct strp_callbacks cb; 2338 u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size; 2339 struct crypto_tfm *tfm; 2340 char *iv, *rec_seq, *key, *salt, *cipher_name; 2341 size_t keysize; 2342 int rc = 0; 2343 2344 if (!ctx) { 2345 rc = -EINVAL; 2346 goto out; 2347 } 2348 2349 if (tx) { 2350 if (!ctx->priv_ctx_tx) { 2351 sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL); 2352 if (!sw_ctx_tx) { 2353 rc = -ENOMEM; 2354 goto out; 2355 } 2356 ctx->priv_ctx_tx = sw_ctx_tx; 2357 } else { 2358 sw_ctx_tx = 2359 (struct tls_sw_context_tx *)ctx->priv_ctx_tx; 2360 } 2361 } else { 2362 if (!ctx->priv_ctx_rx) { 2363 sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL); 2364 if (!sw_ctx_rx) { 2365 rc = -ENOMEM; 2366 goto out; 2367 } 2368 ctx->priv_ctx_rx = sw_ctx_rx; 2369 } else { 2370 sw_ctx_rx = 2371 (struct tls_sw_context_rx *)ctx->priv_ctx_rx; 2372 } 2373 } 2374 2375 if (tx) { 2376 crypto_init_wait(&sw_ctx_tx->async_wait); 2377 spin_lock_init(&sw_ctx_tx->encrypt_compl_lock); 2378 crypto_info = &ctx->crypto_send.info; 2379 cctx = &ctx->tx; 2380 aead = &sw_ctx_tx->aead_send; 2381 INIT_LIST_HEAD(&sw_ctx_tx->tx_list); 2382 INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler); 2383 sw_ctx_tx->tx_work.sk = sk; 2384 } else { 2385 crypto_init_wait(&sw_ctx_rx->async_wait); 2386 spin_lock_init(&sw_ctx_rx->decrypt_compl_lock); 2387 crypto_info = &ctx->crypto_recv.info; 2388 cctx = &ctx->rx; 2389 skb_queue_head_init(&sw_ctx_rx->rx_list); 2390 aead = &sw_ctx_rx->aead_recv; 2391 } 2392 2393 switch (crypto_info->cipher_type) { 2394 case TLS_CIPHER_AES_GCM_128: { 2395 struct tls12_crypto_info_aes_gcm_128 *gcm_128_info; 2396 2397 gcm_128_info = (void *)crypto_info; 2398 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE; 2399 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE; 2400 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE; 2401 iv = gcm_128_info->iv; 2402 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE; 2403 rec_seq = gcm_128_info->rec_seq; 2404 keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE; 2405 key = gcm_128_info->key; 2406 salt = gcm_128_info->salt; 2407 salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE; 2408 cipher_name = "gcm(aes)"; 2409 break; 2410 } 2411 case TLS_CIPHER_AES_GCM_256: { 2412 struct tls12_crypto_info_aes_gcm_256 *gcm_256_info; 2413 2414 gcm_256_info = (void *)crypto_info; 2415 nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE; 2416 tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE; 2417 iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE; 2418 iv = gcm_256_info->iv; 2419 rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE; 2420 rec_seq = gcm_256_info->rec_seq; 2421 keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE; 2422 key = gcm_256_info->key; 2423 salt = gcm_256_info->salt; 2424 salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE; 2425 cipher_name = "gcm(aes)"; 2426 break; 2427 } 2428 case TLS_CIPHER_AES_CCM_128: { 2429 struct tls12_crypto_info_aes_ccm_128 *ccm_128_info; 2430 2431 ccm_128_info = (void *)crypto_info; 2432 nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE; 2433 tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE; 2434 iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE; 2435 iv = ccm_128_info->iv; 2436 rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE; 2437 rec_seq = ccm_128_info->rec_seq; 2438 keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE; 2439 key = ccm_128_info->key; 2440 salt = ccm_128_info->salt; 2441 salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE; 2442 cipher_name = "ccm(aes)"; 2443 break; 2444 } 2445 case TLS_CIPHER_CHACHA20_POLY1305: { 2446 struct tls12_crypto_info_chacha20_poly1305 *chacha20_poly1305_info; 2447 2448 chacha20_poly1305_info = (void *)crypto_info; 2449 nonce_size = 0; 2450 tag_size = TLS_CIPHER_CHACHA20_POLY1305_TAG_SIZE; 2451 iv_size = TLS_CIPHER_CHACHA20_POLY1305_IV_SIZE; 2452 iv = chacha20_poly1305_info->iv; 2453 rec_seq_size = TLS_CIPHER_CHACHA20_POLY1305_REC_SEQ_SIZE; 2454 rec_seq = chacha20_poly1305_info->rec_seq; 2455 keysize = TLS_CIPHER_CHACHA20_POLY1305_KEY_SIZE; 2456 key = chacha20_poly1305_info->key; 2457 salt = chacha20_poly1305_info->salt; 2458 salt_size = TLS_CIPHER_CHACHA20_POLY1305_SALT_SIZE; 2459 cipher_name = "rfc7539(chacha20,poly1305)"; 2460 break; 2461 } 2462 case TLS_CIPHER_SM4_GCM: { 2463 struct tls12_crypto_info_sm4_gcm *sm4_gcm_info; 2464 2465 sm4_gcm_info = (void *)crypto_info; 2466 nonce_size = TLS_CIPHER_SM4_GCM_IV_SIZE; 2467 tag_size = TLS_CIPHER_SM4_GCM_TAG_SIZE; 2468 iv_size = TLS_CIPHER_SM4_GCM_IV_SIZE; 2469 iv = sm4_gcm_info->iv; 2470 rec_seq_size = TLS_CIPHER_SM4_GCM_REC_SEQ_SIZE; 2471 rec_seq = sm4_gcm_info->rec_seq; 2472 keysize = TLS_CIPHER_SM4_GCM_KEY_SIZE; 2473 key = sm4_gcm_info->key; 2474 salt = sm4_gcm_info->salt; 2475 salt_size = TLS_CIPHER_SM4_GCM_SALT_SIZE; 2476 cipher_name = "gcm(sm4)"; 2477 break; 2478 } 2479 case TLS_CIPHER_SM4_CCM: { 2480 struct tls12_crypto_info_sm4_ccm *sm4_ccm_info; 2481 2482 sm4_ccm_info = (void *)crypto_info; 2483 nonce_size = TLS_CIPHER_SM4_CCM_IV_SIZE; 2484 tag_size = TLS_CIPHER_SM4_CCM_TAG_SIZE; 2485 iv_size = TLS_CIPHER_SM4_CCM_IV_SIZE; 2486 iv = sm4_ccm_info->iv; 2487 rec_seq_size = TLS_CIPHER_SM4_CCM_REC_SEQ_SIZE; 2488 rec_seq = sm4_ccm_info->rec_seq; 2489 keysize = TLS_CIPHER_SM4_CCM_KEY_SIZE; 2490 key = sm4_ccm_info->key; 2491 salt = sm4_ccm_info->salt; 2492 salt_size = TLS_CIPHER_SM4_CCM_SALT_SIZE; 2493 cipher_name = "ccm(sm4)"; 2494 break; 2495 } 2496 default: 2497 rc = -EINVAL; 2498 goto free_priv; 2499 } 2500 2501 /* Sanity-check the sizes for stack allocations. */ 2502 if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE || 2503 rec_seq_size > TLS_MAX_REC_SEQ_SIZE) { 2504 rc = -EINVAL; 2505 goto free_priv; 2506 } 2507 2508 if (crypto_info->version == TLS_1_3_VERSION) { 2509 nonce_size = 0; 2510 prot->aad_size = TLS_HEADER_SIZE; 2511 prot->tail_size = 1; 2512 } else { 2513 prot->aad_size = TLS_AAD_SPACE_SIZE; 2514 prot->tail_size = 0; 2515 } 2516 2517 prot->version = crypto_info->version; 2518 prot->cipher_type = crypto_info->cipher_type; 2519 prot->prepend_size = TLS_HEADER_SIZE + nonce_size; 2520 prot->tag_size = tag_size; 2521 prot->overhead_size = prot->prepend_size + 2522 prot->tag_size + prot->tail_size; 2523 prot->iv_size = iv_size; 2524 prot->salt_size = salt_size; 2525 cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL); 2526 if (!cctx->iv) { 2527 rc = -ENOMEM; 2528 goto free_priv; 2529 } 2530 /* Note: 128 & 256 bit salt are the same size */ 2531 prot->rec_seq_size = rec_seq_size; 2532 memcpy(cctx->iv, salt, salt_size); 2533 memcpy(cctx->iv + salt_size, iv, iv_size); 2534 cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL); 2535 if (!cctx->rec_seq) { 2536 rc = -ENOMEM; 2537 goto free_iv; 2538 } 2539 2540 if (!*aead) { 2541 *aead = crypto_alloc_aead(cipher_name, 0, 0); 2542 if (IS_ERR(*aead)) { 2543 rc = PTR_ERR(*aead); 2544 *aead = NULL; 2545 goto free_rec_seq; 2546 } 2547 } 2548 2549 ctx->push_pending_record = tls_sw_push_pending_record; 2550 2551 rc = crypto_aead_setkey(*aead, key, keysize); 2552 2553 if (rc) 2554 goto free_aead; 2555 2556 rc = crypto_aead_setauthsize(*aead, prot->tag_size); 2557 if (rc) 2558 goto free_aead; 2559 2560 if (sw_ctx_rx) { 2561 tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv); 2562 2563 if (crypto_info->version == TLS_1_3_VERSION) 2564 sw_ctx_rx->async_capable = 0; 2565 else 2566 sw_ctx_rx->async_capable = 2567 !!(tfm->__crt_alg->cra_flags & 2568 CRYPTO_ALG_ASYNC); 2569 2570 /* Set up strparser */ 2571 memset(&cb, 0, sizeof(cb)); 2572 cb.rcv_msg = tls_queue; 2573 cb.parse_msg = tls_read_size; 2574 2575 strp_init(&sw_ctx_rx->strp, sk, &cb); 2576 } 2577 2578 goto out; 2579 2580 free_aead: 2581 crypto_free_aead(*aead); 2582 *aead = NULL; 2583 free_rec_seq: 2584 kfree(cctx->rec_seq); 2585 cctx->rec_seq = NULL; 2586 free_iv: 2587 kfree(cctx->iv); 2588 cctx->iv = NULL; 2589 free_priv: 2590 if (tx) { 2591 kfree(ctx->priv_ctx_tx); 2592 ctx->priv_ctx_tx = NULL; 2593 } else { 2594 kfree(ctx->priv_ctx_rx); 2595 ctx->priv_ctx_rx = NULL; 2596 } 2597 out: 2598 return rc; 2599 } 2600