| 1 | /* |
| 2 | * Copyright (c) 2016-2017, Mellanox Technologies. All rights reserved. |
| 3 | * Copyright (c) 2016-2017, Dave Watson <davejwatson@fb.com>. All rights reserved. |
| 4 | * Copyright (c) 2016-2017, Lance Chao <lancerchao@fb.com>. All rights reserved. |
| 5 | * Copyright (c) 2016, Fridolin Pokorny <fridolin.pokorny@gmail.com>. All rights reserved. |
| 6 | * Copyright (c) 2016, Nikos Mavrogiannopoulos <nmav@gnutls.org>. All rights reserved. |
| 7 | * Copyright (c) 2018, Covalent IO, Inc. http://covalent.io |
| 8 | * |
| 9 | * This software is available to you under a choice of one of two |
| 10 | * licenses. You may choose to be licensed under the terms of the GNU |
| 11 | * General Public License (GPL) Version 2, available from the file |
| 12 | * COPYING in the main directory of this source tree, or the |
| 13 | * OpenIB.org BSD license below: |
| 14 | * |
| 15 | * Redistribution and use in source and binary forms, with or |
| 16 | * without modification, are permitted provided that the following |
| 17 | * conditions are met: |
| 18 | * |
| 19 | * - Redistributions of source code must retain the above |
| 20 | * copyright notice, this list of conditions and the following |
| 21 | * disclaimer. |
| 22 | * |
| 23 | * - Redistributions in binary form must reproduce the above |
| 24 | * copyright notice, this list of conditions and the following |
| 25 | * disclaimer in the documentation and/or other materials |
| 26 | * provided with the distribution. |
| 27 | * |
| 28 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, |
| 29 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF |
| 30 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND |
| 31 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS |
| 32 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN |
| 33 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN |
| 34 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
| 35 | * SOFTWARE. |
| 36 | */ |
| 37 | |
| 38 | #include <linux/sched/signal.h> |
| 39 | #include <linux/module.h> |
| 40 | #include <crypto/aead.h> |
| 41 | |
| 42 | #include <net/strparser.h> |
| 43 | #include <net/tls.h> |
| 44 | |
| 45 | static int __skb_nsg(struct sk_buff *skb, int offset, int len, |
| 46 | unsigned int recursion_level) |
| 47 | { |
| 48 | int start = skb_headlen(skb); |
| 49 | int i, chunk = start - offset; |
| 50 | struct sk_buff *frag_iter; |
| 51 | int elt = 0; |
| 52 | |
| 53 | if (unlikely(recursion_level >= 24)) |
| 54 | return -EMSGSIZE; |
| 55 | |
| 56 | if (chunk > 0) { |
| 57 | if (chunk > len) |
| 58 | chunk = len; |
| 59 | elt++; |
| 60 | len -= chunk; |
| 61 | if (len == 0) |
| 62 | return elt; |
| 63 | offset += chunk; |
| 64 | } |
| 65 | |
| 66 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
| 67 | int end; |
| 68 | |
| 69 | WARN_ON(start > offset + len); |
| 70 | |
| 71 | end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); |
| 72 | chunk = end - offset; |
| 73 | if (chunk > 0) { |
| 74 | if (chunk > len) |
| 75 | chunk = len; |
| 76 | elt++; |
| 77 | len -= chunk; |
| 78 | if (len == 0) |
| 79 | return elt; |
| 80 | offset += chunk; |
| 81 | } |
| 82 | start = end; |
| 83 | } |
| 84 | |
| 85 | if (unlikely(skb_has_frag_list(skb))) { |
| 86 | skb_walk_frags(skb, frag_iter) { |
| 87 | int end, ret; |
| 88 | |
| 89 | WARN_ON(start > offset + len); |
| 90 | |
| 91 | end = start + frag_iter->len; |
| 92 | chunk = end - offset; |
| 93 | if (chunk > 0) { |
| 94 | if (chunk > len) |
| 95 | chunk = len; |
| 96 | ret = __skb_nsg(frag_iter, offset - start, chunk, |
| 97 | recursion_level + 1); |
| 98 | if (unlikely(ret < 0)) |
| 99 | return ret; |
| 100 | elt += ret; |
| 101 | len -= chunk; |
| 102 | if (len == 0) |
| 103 | return elt; |
| 104 | offset += chunk; |
| 105 | } |
| 106 | start = end; |
| 107 | } |
| 108 | } |
| 109 | BUG_ON(len); |
| 110 | return elt; |
| 111 | } |
| 112 | |
| 113 | /* Return the number of scatterlist elements required to completely map the |
| 114 | * skb, or -EMSGSIZE if the recursion depth is exceeded. |
| 115 | */ |
| 116 | static int skb_nsg(struct sk_buff *skb, int offset, int len) |
| 117 | { |
| 118 | return __skb_nsg(skb, offset, len, 0); |
| 119 | } |
| 120 | |
| 121 | static int padding_length(struct tls_sw_context_rx *ctx, |
| 122 | struct tls_prot_info *prot, struct sk_buff *skb) |
| 123 | { |
| 124 | struct strp_msg *rxm = strp_msg(skb); |
| 125 | int sub = 0; |
| 126 | |
| 127 | /* Determine zero-padding length */ |
| 128 | if (prot->version == TLS_1_3_VERSION) { |
| 129 | char content_type = 0; |
| 130 | int err; |
| 131 | int back = 17; |
| 132 | |
| 133 | while (content_type == 0) { |
| 134 | if (back > rxm->full_len - prot->prepend_size) |
| 135 | return -EBADMSG; |
| 136 | err = skb_copy_bits(skb, |
| 137 | rxm->offset + rxm->full_len - back, |
| 138 | &content_type, 1); |
| 139 | if (err) |
| 140 | return err; |
| 141 | if (content_type) |
| 142 | break; |
| 143 | sub++; |
| 144 | back++; |
| 145 | } |
| 146 | ctx->control = content_type; |
| 147 | } |
| 148 | return sub; |
| 149 | } |
| 150 | |
| 151 | static void tls_decrypt_done(struct crypto_async_request *req, int err) |
| 152 | { |
| 153 | struct aead_request *aead_req = (struct aead_request *)req; |
| 154 | struct scatterlist *sgout = aead_req->dst; |
| 155 | struct scatterlist *sgin = aead_req->src; |
| 156 | struct tls_sw_context_rx *ctx; |
| 157 | struct tls_context *tls_ctx; |
| 158 | struct tls_prot_info *prot; |
| 159 | struct scatterlist *sg; |
| 160 | struct sk_buff *skb; |
| 161 | unsigned int pages; |
| 162 | int pending; |
| 163 | |
| 164 | skb = (struct sk_buff *)req->data; |
| 165 | tls_ctx = tls_get_ctx(skb->sk); |
| 166 | ctx = tls_sw_ctx_rx(tls_ctx); |
| 167 | prot = &tls_ctx->prot_info; |
| 168 | |
| 169 | /* Propagate if there was an err */ |
| 170 | if (err) { |
| 171 | if (err == -EBADMSG) |
| 172 | TLS_INC_STATS(sock_net(skb->sk), |
| 173 | LINUX_MIB_TLSDECRYPTERROR); |
| 174 | ctx->async_wait.err = err; |
| 175 | tls_err_abort(skb->sk, err); |
| 176 | } else { |
| 177 | struct strp_msg *rxm = strp_msg(skb); |
| 178 | int pad; |
| 179 | |
| 180 | pad = padding_length(ctx, prot, skb); |
| 181 | if (pad < 0) { |
| 182 | ctx->async_wait.err = pad; |
| 183 | tls_err_abort(skb->sk, pad); |
| 184 | } else { |
| 185 | rxm->full_len -= pad; |
| 186 | rxm->offset += prot->prepend_size; |
| 187 | rxm->full_len -= prot->overhead_size; |
| 188 | } |
| 189 | } |
| 190 | |
| 191 | /* After using skb->sk to propagate sk through crypto async callback |
| 192 | * we need to NULL it again. |
| 193 | */ |
| 194 | skb->sk = NULL; |
| 195 | |
| 196 | |
| 197 | /* Free the destination pages if skb was not decrypted inplace */ |
| 198 | if (sgout != sgin) { |
| 199 | /* Skip the first S/G entry as it points to AAD */ |
| 200 | for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) { |
| 201 | if (!sg) |
| 202 | break; |
| 203 | put_page(sg_page(sg)); |
| 204 | } |
| 205 | } |
| 206 | |
| 207 | kfree(aead_req); |
| 208 | |
| 209 | spin_lock_bh(&ctx->decrypt_compl_lock); |
| 210 | pending = atomic_dec_return(&ctx->decrypt_pending); |
| 211 | |
| 212 | if (!pending && ctx->async_notify) |
| 213 | complete(&ctx->async_wait.completion); |
| 214 | spin_unlock_bh(&ctx->decrypt_compl_lock); |
| 215 | } |
| 216 | |
| 217 | static int tls_do_decryption(struct sock *sk, |
| 218 | struct sk_buff *skb, |
| 219 | struct scatterlist *sgin, |
| 220 | struct scatterlist *sgout, |
| 221 | char *iv_recv, |
| 222 | size_t data_len, |
| 223 | struct aead_request *aead_req, |
| 224 | bool async) |
| 225 | { |
| 226 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 227 | struct tls_prot_info *prot = &tls_ctx->prot_info; |
| 228 | struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); |
| 229 | int ret; |
| 230 | |
| 231 | aead_request_set_tfm(aead_req, ctx->aead_recv); |
| 232 | aead_request_set_ad(aead_req, prot->aad_size); |
| 233 | aead_request_set_crypt(aead_req, sgin, sgout, |
| 234 | data_len + prot->tag_size, |
| 235 | (u8 *)iv_recv); |
| 236 | |
| 237 | if (async) { |
| 238 | /* Using skb->sk to push sk through to crypto async callback |
| 239 | * handler. This allows propagating errors up to the socket |
| 240 | * if needed. It _must_ be cleared in the async handler |
| 241 | * before consume_skb is called. We _know_ skb->sk is NULL |
| 242 | * because it is a clone from strparser. |
| 243 | */ |
| 244 | skb->sk = sk; |
| 245 | aead_request_set_callback(aead_req, |
| 246 | CRYPTO_TFM_REQ_MAY_BACKLOG, |
| 247 | tls_decrypt_done, skb); |
| 248 | atomic_inc(&ctx->decrypt_pending); |
| 249 | } else { |
| 250 | aead_request_set_callback(aead_req, |
| 251 | CRYPTO_TFM_REQ_MAY_BACKLOG, |
| 252 | crypto_req_done, &ctx->async_wait); |
| 253 | } |
| 254 | |
| 255 | ret = crypto_aead_decrypt(aead_req); |
| 256 | if (ret == -EINPROGRESS) { |
| 257 | if (async) |
| 258 | return ret; |
| 259 | |
| 260 | ret = crypto_wait_req(ret, &ctx->async_wait); |
| 261 | } |
| 262 | |
| 263 | if (async) |
| 264 | atomic_dec(&ctx->decrypt_pending); |
| 265 | |
| 266 | return ret; |
| 267 | } |
| 268 | |
| 269 | static void tls_trim_both_msgs(struct sock *sk, int target_size) |
| 270 | { |
| 271 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 272 | struct tls_prot_info *prot = &tls_ctx->prot_info; |
| 273 | struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); |
| 274 | struct tls_rec *rec = ctx->open_rec; |
| 275 | |
| 276 | sk_msg_trim(sk, &rec->msg_plaintext, target_size); |
| 277 | if (target_size > 0) |
| 278 | target_size += prot->overhead_size; |
| 279 | sk_msg_trim(sk, &rec->msg_encrypted, target_size); |
| 280 | } |
| 281 | |
| 282 | static int tls_alloc_encrypted_msg(struct sock *sk, int len) |
| 283 | { |
| 284 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 285 | struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); |
| 286 | struct tls_rec *rec = ctx->open_rec; |
| 287 | struct sk_msg *msg_en = &rec->msg_encrypted; |
| 288 | |
| 289 | return sk_msg_alloc(sk, msg_en, len, 0); |
| 290 | } |
| 291 | |
| 292 | static int tls_clone_plaintext_msg(struct sock *sk, int required) |
| 293 | { |
| 294 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 295 | struct tls_prot_info *prot = &tls_ctx->prot_info; |
| 296 | struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); |
| 297 | struct tls_rec *rec = ctx->open_rec; |
| 298 | struct sk_msg *msg_pl = &rec->msg_plaintext; |
| 299 | struct sk_msg *msg_en = &rec->msg_encrypted; |
| 300 | int skip, len; |
| 301 | |
| 302 | /* We add page references worth len bytes from encrypted sg |
| 303 | * at the end of plaintext sg. It is guaranteed that msg_en |
| 304 | * has enough required room (ensured by caller). |
| 305 | */ |
| 306 | len = required - msg_pl->sg.size; |
| 307 | |
| 308 | /* Skip initial bytes in msg_en's data to be able to use |
| 309 | * same offset of both plain and encrypted data. |
| 310 | */ |
| 311 | skip = prot->prepend_size + msg_pl->sg.size; |
| 312 | |
| 313 | return sk_msg_clone(sk, msg_pl, msg_en, skip, len); |
| 314 | } |
| 315 | |
| 316 | static struct tls_rec *tls_get_rec(struct sock *sk) |
| 317 | { |
| 318 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 319 | struct tls_prot_info *prot = &tls_ctx->prot_info; |
| 320 | struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); |
| 321 | struct sk_msg *msg_pl, *msg_en; |
| 322 | struct tls_rec *rec; |
| 323 | int mem_size; |
| 324 | |
| 325 | mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send); |
| 326 | |
| 327 | rec = kzalloc(mem_size, sk->sk_allocation); |
| 328 | if (!rec) |
| 329 | return NULL; |
| 330 | |
| 331 | msg_pl = &rec->msg_plaintext; |
| 332 | msg_en = &rec->msg_encrypted; |
| 333 | |
| 334 | sk_msg_init(msg_pl); |
| 335 | sk_msg_init(msg_en); |
| 336 | |
| 337 | sg_init_table(rec->sg_aead_in, 2); |
| 338 | sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size); |
| 339 | sg_unmark_end(&rec->sg_aead_in[1]); |
| 340 | |
| 341 | sg_init_table(rec->sg_aead_out, 2); |
| 342 | sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size); |
| 343 | sg_unmark_end(&rec->sg_aead_out[1]); |
| 344 | |
| 345 | return rec; |
| 346 | } |
| 347 | |
| 348 | static void tls_free_rec(struct sock *sk, struct tls_rec *rec) |
| 349 | { |
| 350 | sk_msg_free(sk, &rec->msg_encrypted); |
| 351 | sk_msg_free(sk, &rec->msg_plaintext); |
| 352 | kfree(rec); |
| 353 | } |
| 354 | |
| 355 | static void tls_free_open_rec(struct sock *sk) |
| 356 | { |
| 357 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 358 | struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); |
| 359 | struct tls_rec *rec = ctx->open_rec; |
| 360 | |
| 361 | if (rec) { |
| 362 | tls_free_rec(sk, rec); |
| 363 | ctx->open_rec = NULL; |
| 364 | } |
| 365 | } |
| 366 | |
| 367 | int tls_tx_records(struct sock *sk, int flags) |
| 368 | { |
| 369 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 370 | struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); |
| 371 | struct tls_rec *rec, *tmp; |
| 372 | struct sk_msg *msg_en; |
| 373 | int tx_flags, rc = 0; |
| 374 | |
| 375 | if (tls_is_partially_sent_record(tls_ctx)) { |
| 376 | rec = list_first_entry(&ctx->tx_list, |
| 377 | struct tls_rec, list); |
| 378 | |
| 379 | if (flags == -1) |
| 380 | tx_flags = rec->tx_flags; |
| 381 | else |
| 382 | tx_flags = flags; |
| 383 | |
| 384 | rc = tls_push_partial_record(sk, tls_ctx, tx_flags); |
| 385 | if (rc) |
| 386 | goto tx_err; |
| 387 | |
| 388 | /* Full record has been transmitted. |
| 389 | * Remove the head of tx_list |
| 390 | */ |
| 391 | list_del(&rec->list); |
| 392 | sk_msg_free(sk, &rec->msg_plaintext); |
| 393 | kfree(rec); |
| 394 | } |
| 395 | |
| 396 | /* Tx all ready records */ |
| 397 | list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) { |
| 398 | if (READ_ONCE(rec->tx_ready)) { |
| 399 | if (flags == -1) |
| 400 | tx_flags = rec->tx_flags; |
| 401 | else |
| 402 | tx_flags = flags; |
| 403 | |
| 404 | msg_en = &rec->msg_encrypted; |
| 405 | rc = tls_push_sg(sk, tls_ctx, |
| 406 | &msg_en->sg.data[msg_en->sg.curr], |
| 407 | 0, tx_flags); |
| 408 | if (rc) |
| 409 | goto tx_err; |
| 410 | |
| 411 | list_del(&rec->list); |
| 412 | sk_msg_free(sk, &rec->msg_plaintext); |
| 413 | kfree(rec); |
| 414 | } else { |
| 415 | break; |
| 416 | } |
| 417 | } |
| 418 | |
| 419 | tx_err: |
| 420 | if (rc < 0 && rc != -EAGAIN) |
| 421 | tls_err_abort(sk, EBADMSG); |
| 422 | |
| 423 | return rc; |
| 424 | } |
| 425 | |
| 426 | static void tls_encrypt_done(struct crypto_async_request *req, int err) |
| 427 | { |
| 428 | struct aead_request *aead_req = (struct aead_request *)req; |
| 429 | struct sock *sk = req->data; |
| 430 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 431 | struct tls_prot_info *prot = &tls_ctx->prot_info; |
| 432 | struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); |
| 433 | struct scatterlist *sge; |
| 434 | struct sk_msg *msg_en; |
| 435 | struct tls_rec *rec; |
| 436 | bool ready = false; |
| 437 | int pending; |
| 438 | |
| 439 | rec = container_of(aead_req, struct tls_rec, aead_req); |
| 440 | msg_en = &rec->msg_encrypted; |
| 441 | |
| 442 | sge = sk_msg_elem(msg_en, msg_en->sg.curr); |
| 443 | sge->offset -= prot->prepend_size; |
| 444 | sge->length += prot->prepend_size; |
| 445 | |
| 446 | /* Check if error is previously set on socket */ |
| 447 | if (err || sk->sk_err) { |
| 448 | rec = NULL; |
| 449 | |
| 450 | /* If err is already set on socket, return the same code */ |
| 451 | if (sk->sk_err) { |
| 452 | ctx->async_wait.err = sk->sk_err; |
| 453 | } else { |
| 454 | ctx->async_wait.err = err; |
| 455 | tls_err_abort(sk, err); |
| 456 | } |
| 457 | } |
| 458 | |
| 459 | if (rec) { |
| 460 | struct tls_rec *first_rec; |
| 461 | |
| 462 | /* Mark the record as ready for transmission */ |
| 463 | smp_store_mb(rec->tx_ready, true); |
| 464 | |
| 465 | /* If received record is at head of tx_list, schedule tx */ |
| 466 | first_rec = list_first_entry(&ctx->tx_list, |
| 467 | struct tls_rec, list); |
| 468 | if (rec == first_rec) |
| 469 | ready = true; |
| 470 | } |
| 471 | |
| 472 | spin_lock_bh(&ctx->encrypt_compl_lock); |
| 473 | pending = atomic_dec_return(&ctx->encrypt_pending); |
| 474 | |
| 475 | if (!pending && ctx->async_notify) |
| 476 | complete(&ctx->async_wait.completion); |
| 477 | spin_unlock_bh(&ctx->encrypt_compl_lock); |
| 478 | |
| 479 | if (!ready) |
| 480 | return; |
| 481 | |
| 482 | /* Schedule the transmission */ |
| 483 | if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) |
| 484 | schedule_delayed_work(&ctx->tx_work.work, 1); |
| 485 | } |
| 486 | |
| 487 | static int tls_do_encryption(struct sock *sk, |
| 488 | struct tls_context *tls_ctx, |
| 489 | struct tls_sw_context_tx *ctx, |
| 490 | struct aead_request *aead_req, |
| 491 | size_t data_len, u32 start) |
| 492 | { |
| 493 | struct tls_prot_info *prot = &tls_ctx->prot_info; |
| 494 | struct tls_rec *rec = ctx->open_rec; |
| 495 | struct sk_msg *msg_en = &rec->msg_encrypted; |
| 496 | struct scatterlist *sge = sk_msg_elem(msg_en, start); |
| 497 | int rc, iv_offset = 0; |
| 498 | |
| 499 | /* For CCM based ciphers, first byte of IV is a constant */ |
| 500 | if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) { |
| 501 | rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE; |
| 502 | iv_offset = 1; |
| 503 | } |
| 504 | |
| 505 | memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv, |
| 506 | prot->iv_size + prot->salt_size); |
| 507 | |
| 508 | xor_iv_with_seq(prot->version, rec->iv_data, tls_ctx->tx.rec_seq); |
| 509 | |
| 510 | sge->offset += prot->prepend_size; |
| 511 | sge->length -= prot->prepend_size; |
| 512 | |
| 513 | msg_en->sg.curr = start; |
| 514 | |
| 515 | aead_request_set_tfm(aead_req, ctx->aead_send); |
| 516 | aead_request_set_ad(aead_req, prot->aad_size); |
| 517 | aead_request_set_crypt(aead_req, rec->sg_aead_in, |
| 518 | rec->sg_aead_out, |
| 519 | data_len, rec->iv_data); |
| 520 | |
| 521 | aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG, |
| 522 | tls_encrypt_done, sk); |
| 523 | |
| 524 | /* Add the record in tx_list */ |
| 525 | list_add_tail((struct list_head *)&rec->list, &ctx->tx_list); |
| 526 | atomic_inc(&ctx->encrypt_pending); |
| 527 | |
| 528 | rc = crypto_aead_encrypt(aead_req); |
| 529 | if (!rc || rc != -EINPROGRESS) { |
| 530 | atomic_dec(&ctx->encrypt_pending); |
| 531 | sge->offset -= prot->prepend_size; |
| 532 | sge->length += prot->prepend_size; |
| 533 | } |
| 534 | |
| 535 | if (!rc) { |
| 536 | WRITE_ONCE(rec->tx_ready, true); |
| 537 | } else if (rc != -EINPROGRESS) { |
| 538 | list_del(&rec->list); |
| 539 | return rc; |
| 540 | } |
| 541 | |
| 542 | /* Unhook the record from context if encryption is not failure */ |
| 543 | ctx->open_rec = NULL; |
| 544 | tls_advance_record_sn(sk, prot, &tls_ctx->tx); |
| 545 | return rc; |
| 546 | } |
| 547 | |
| 548 | static int tls_split_open_record(struct sock *sk, struct tls_rec *from, |
| 549 | struct tls_rec **to, struct sk_msg *msg_opl, |
| 550 | struct sk_msg *msg_oen, u32 split_point, |
| 551 | u32 tx_overhead_size, u32 *orig_end) |
| 552 | { |
| 553 | u32 i, j, bytes = 0, apply = msg_opl->apply_bytes; |
| 554 | struct scatterlist *sge, *osge, *nsge; |
| 555 | u32 orig_size = msg_opl->sg.size; |
| 556 | struct scatterlist tmp = { }; |
| 557 | struct sk_msg *msg_npl; |
| 558 | struct tls_rec *new; |
| 559 | int ret; |
| 560 | |
| 561 | new = tls_get_rec(sk); |
| 562 | if (!new) |
| 563 | return -ENOMEM; |
| 564 | ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size + |
| 565 | tx_overhead_size, 0); |
| 566 | if (ret < 0) { |
| 567 | tls_free_rec(sk, new); |
| 568 | return ret; |
| 569 | } |
| 570 | |
| 571 | *orig_end = msg_opl->sg.end; |
| 572 | i = msg_opl->sg.start; |
| 573 | sge = sk_msg_elem(msg_opl, i); |
| 574 | while (apply && sge->length) { |
| 575 | if (sge->length > apply) { |
| 576 | u32 len = sge->length - apply; |
| 577 | |
| 578 | get_page(sg_page(sge)); |
| 579 | sg_set_page(&tmp, sg_page(sge), len, |
| 580 | sge->offset + apply); |
| 581 | sge->length = apply; |
| 582 | bytes += apply; |
| 583 | apply = 0; |
| 584 | } else { |
| 585 | apply -= sge->length; |
| 586 | bytes += sge->length; |
| 587 | } |
| 588 | |
| 589 | sk_msg_iter_var_next(i); |
| 590 | if (i == msg_opl->sg.end) |
| 591 | break; |
| 592 | sge = sk_msg_elem(msg_opl, i); |
| 593 | } |
| 594 | |
| 595 | msg_opl->sg.end = i; |
| 596 | msg_opl->sg.curr = i; |
| 597 | msg_opl->sg.copybreak = 0; |
| 598 | msg_opl->apply_bytes = 0; |
| 599 | msg_opl->sg.size = bytes; |
| 600 | |
| 601 | msg_npl = &new->msg_plaintext; |
| 602 | msg_npl->apply_bytes = apply; |
| 603 | msg_npl->sg.size = orig_size - bytes; |
| 604 | |
| 605 | j = msg_npl->sg.start; |
| 606 | nsge = sk_msg_elem(msg_npl, j); |
| 607 | if (tmp.length) { |
| 608 | memcpy(nsge, &tmp, sizeof(*nsge)); |
| 609 | sk_msg_iter_var_next(j); |
| 610 | nsge = sk_msg_elem(msg_npl, j); |
| 611 | } |
| 612 | |
| 613 | osge = sk_msg_elem(msg_opl, i); |
| 614 | while (osge->length) { |
| 615 | memcpy(nsge, osge, sizeof(*nsge)); |
| 616 | sg_unmark_end(nsge); |
| 617 | sk_msg_iter_var_next(i); |
| 618 | sk_msg_iter_var_next(j); |
| 619 | if (i == *orig_end) |
| 620 | break; |
| 621 | osge = sk_msg_elem(msg_opl, i); |
| 622 | nsge = sk_msg_elem(msg_npl, j); |
| 623 | } |
| 624 | |
| 625 | msg_npl->sg.end = j; |
| 626 | msg_npl->sg.curr = j; |
| 627 | msg_npl->sg.copybreak = 0; |
| 628 | |
| 629 | *to = new; |
| 630 | return 0; |
| 631 | } |
| 632 | |
| 633 | static void tls_merge_open_record(struct sock *sk, struct tls_rec *to, |
| 634 | struct tls_rec *from, u32 orig_end) |
| 635 | { |
| 636 | struct sk_msg *msg_npl = &from->msg_plaintext; |
| 637 | struct sk_msg *msg_opl = &to->msg_plaintext; |
| 638 | struct scatterlist *osge, *nsge; |
| 639 | u32 i, j; |
| 640 | |
| 641 | i = msg_opl->sg.end; |
| 642 | sk_msg_iter_var_prev(i); |
| 643 | j = msg_npl->sg.start; |
| 644 | |
| 645 | osge = sk_msg_elem(msg_opl, i); |
| 646 | nsge = sk_msg_elem(msg_npl, j); |
| 647 | |
| 648 | if (sg_page(osge) == sg_page(nsge) && |
| 649 | osge->offset + osge->length == nsge->offset) { |
| 650 | osge->length += nsge->length; |
| 651 | put_page(sg_page(nsge)); |
| 652 | } |
| 653 | |
| 654 | msg_opl->sg.end = orig_end; |
| 655 | msg_opl->sg.curr = orig_end; |
| 656 | msg_opl->sg.copybreak = 0; |
| 657 | msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size; |
| 658 | msg_opl->sg.size += msg_npl->sg.size; |
| 659 | |
| 660 | sk_msg_free(sk, &to->msg_encrypted); |
| 661 | sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted); |
| 662 | |
| 663 | kfree(from); |
| 664 | } |
| 665 | |
| 666 | static int tls_push_record(struct sock *sk, int flags, |
| 667 | unsigned char record_type) |
| 668 | { |
| 669 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 670 | struct tls_prot_info *prot = &tls_ctx->prot_info; |
| 671 | struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); |
| 672 | struct tls_rec *rec = ctx->open_rec, *tmp = NULL; |
| 673 | u32 i, split_point, uninitialized_var(orig_end); |
| 674 | struct sk_msg *msg_pl, *msg_en; |
| 675 | struct aead_request *req; |
| 676 | bool split; |
| 677 | int rc; |
| 678 | |
| 679 | if (!rec) |
| 680 | return 0; |
| 681 | |
| 682 | msg_pl = &rec->msg_plaintext; |
| 683 | msg_en = &rec->msg_encrypted; |
| 684 | |
| 685 | split_point = msg_pl->apply_bytes; |
| 686 | split = split_point && split_point < msg_pl->sg.size; |
| 687 | if (unlikely((!split && |
| 688 | msg_pl->sg.size + |
| 689 | prot->overhead_size > msg_en->sg.size) || |
| 690 | (split && |
| 691 | split_point + |
| 692 | prot->overhead_size > msg_en->sg.size))) { |
| 693 | split = true; |
| 694 | split_point = msg_en->sg.size; |
| 695 | } |
| 696 | if (split) { |
| 697 | rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en, |
| 698 | split_point, prot->overhead_size, |
| 699 | &orig_end); |
| 700 | if (rc < 0) |
| 701 | return rc; |
| 702 | /* This can happen if above tls_split_open_record allocates |
| 703 | * a single large encryption buffer instead of two smaller |
| 704 | * ones. In this case adjust pointers and continue without |
| 705 | * split. |
| 706 | */ |
| 707 | if (!msg_pl->sg.size) { |
| 708 | tls_merge_open_record(sk, rec, tmp, orig_end); |
| 709 | msg_pl = &rec->msg_plaintext; |
| 710 | msg_en = &rec->msg_encrypted; |
| 711 | split = false; |
| 712 | } |
| 713 | sk_msg_trim(sk, msg_en, msg_pl->sg.size + |
| 714 | prot->overhead_size); |
| 715 | } |
| 716 | |
| 717 | rec->tx_flags = flags; |
| 718 | req = &rec->aead_req; |
| 719 | |
| 720 | i = msg_pl->sg.end; |
| 721 | sk_msg_iter_var_prev(i); |
| 722 | |
| 723 | rec->content_type = record_type; |
| 724 | if (prot->version == TLS_1_3_VERSION) { |
| 725 | /* Add content type to end of message. No padding added */ |
| 726 | sg_set_buf(&rec->sg_content_type, &rec->content_type, 1); |
| 727 | sg_mark_end(&rec->sg_content_type); |
| 728 | sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1, |
| 729 | &rec->sg_content_type); |
| 730 | } else { |
| 731 | sg_mark_end(sk_msg_elem(msg_pl, i)); |
| 732 | } |
| 733 | |
| 734 | if (msg_pl->sg.end < msg_pl->sg.start) { |
| 735 | sg_chain(&msg_pl->sg.data[msg_pl->sg.start], |
| 736 | MAX_SKB_FRAGS - msg_pl->sg.start + 1, |
| 737 | msg_pl->sg.data); |
| 738 | } |
| 739 | |
| 740 | i = msg_pl->sg.start; |
| 741 | sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]); |
| 742 | |
| 743 | i = msg_en->sg.end; |
| 744 | sk_msg_iter_var_prev(i); |
| 745 | sg_mark_end(sk_msg_elem(msg_en, i)); |
| 746 | |
| 747 | i = msg_en->sg.start; |
| 748 | sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]); |
| 749 | |
| 750 | tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size, |
| 751 | tls_ctx->tx.rec_seq, prot->rec_seq_size, |
| 752 | record_type, prot->version); |
| 753 | |
| 754 | tls_fill_prepend(tls_ctx, |
| 755 | page_address(sg_page(&msg_en->sg.data[i])) + |
| 756 | msg_en->sg.data[i].offset, |
| 757 | msg_pl->sg.size + prot->tail_size, |
| 758 | record_type, prot->version); |
| 759 | |
| 760 | tls_ctx->pending_open_record_frags = false; |
| 761 | |
| 762 | rc = tls_do_encryption(sk, tls_ctx, ctx, req, |
| 763 | msg_pl->sg.size + prot->tail_size, i); |
| 764 | if (rc < 0) { |
| 765 | if (rc != -EINPROGRESS) { |
| 766 | tls_err_abort(sk, EBADMSG); |
| 767 | if (split) { |
| 768 | tls_ctx->pending_open_record_frags = true; |
| 769 | tls_merge_open_record(sk, rec, tmp, orig_end); |
| 770 | } |
| 771 | } |
| 772 | ctx->async_capable = 1; |
| 773 | return rc; |
| 774 | } else if (split) { |
| 775 | msg_pl = &tmp->msg_plaintext; |
| 776 | msg_en = &tmp->msg_encrypted; |
| 777 | sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size); |
| 778 | tls_ctx->pending_open_record_frags = true; |
| 779 | ctx->open_rec = tmp; |
| 780 | } |
| 781 | |
| 782 | return tls_tx_records(sk, flags); |
| 783 | } |
| 784 | |
| 785 | static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk, |
| 786 | bool full_record, u8 record_type, |
| 787 | ssize_t *copied, int flags) |
| 788 | { |
| 789 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 790 | struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); |
| 791 | struct sk_msg msg_redir = { }; |
| 792 | struct sk_psock *psock; |
| 793 | struct sock *sk_redir; |
| 794 | struct tls_rec *rec; |
| 795 | bool enospc, policy; |
| 796 | int err = 0, send; |
| 797 | u32 delta = 0; |
| 798 | |
| 799 | policy = !(flags & MSG_SENDPAGE_NOPOLICY); |
| 800 | psock = sk_psock_get(sk); |
| 801 | if (!psock || !policy) { |
| 802 | err = tls_push_record(sk, flags, record_type); |
| 803 | if (err && sk->sk_err == EBADMSG) { |
| 804 | *copied -= sk_msg_free(sk, msg); |
| 805 | tls_free_open_rec(sk); |
| 806 | err = -sk->sk_err; |
| 807 | } |
| 808 | if (psock) |
| 809 | sk_psock_put(sk, psock); |
| 810 | return err; |
| 811 | } |
| 812 | more_data: |
| 813 | enospc = sk_msg_full(msg); |
| 814 | if (psock->eval == __SK_NONE) { |
| 815 | delta = msg->sg.size; |
| 816 | psock->eval = sk_psock_msg_verdict(sk, psock, msg); |
| 817 | delta -= msg->sg.size; |
| 818 | } |
| 819 | if (msg->cork_bytes && msg->cork_bytes > msg->sg.size && |
| 820 | !enospc && !full_record) { |
| 821 | err = -ENOSPC; |
| 822 | goto out_err; |
| 823 | } |
| 824 | msg->cork_bytes = 0; |
| 825 | send = msg->sg.size; |
| 826 | if (msg->apply_bytes && msg->apply_bytes < send) |
| 827 | send = msg->apply_bytes; |
| 828 | |
| 829 | switch (psock->eval) { |
| 830 | case __SK_PASS: |
| 831 | err = tls_push_record(sk, flags, record_type); |
| 832 | if (err && sk->sk_err == EBADMSG) { |
| 833 | *copied -= sk_msg_free(sk, msg); |
| 834 | tls_free_open_rec(sk); |
| 835 | err = -sk->sk_err; |
| 836 | goto out_err; |
| 837 | } |
| 838 | break; |
| 839 | case __SK_REDIRECT: |
| 840 | sk_redir = psock->sk_redir; |
| 841 | memcpy(&msg_redir, msg, sizeof(*msg)); |
| 842 | if (msg->apply_bytes < send) |
| 843 | msg->apply_bytes = 0; |
| 844 | else |
| 845 | msg->apply_bytes -= send; |
| 846 | sk_msg_return_zero(sk, msg, send); |
| 847 | msg->sg.size -= send; |
| 848 | release_sock(sk); |
| 849 | err = tcp_bpf_sendmsg_redir(sk_redir, &msg_redir, send, flags); |
| 850 | lock_sock(sk); |
| 851 | if (err < 0) { |
| 852 | *copied -= sk_msg_free_nocharge(sk, &msg_redir); |
| 853 | msg->sg.size = 0; |
| 854 | } |
| 855 | if (msg->sg.size == 0) |
| 856 | tls_free_open_rec(sk); |
| 857 | break; |
| 858 | case __SK_DROP: |
| 859 | default: |
| 860 | sk_msg_free_partial(sk, msg, send); |
| 861 | if (msg->apply_bytes < send) |
| 862 | msg->apply_bytes = 0; |
| 863 | else |
| 864 | msg->apply_bytes -= send; |
| 865 | if (msg->sg.size == 0) |
| 866 | tls_free_open_rec(sk); |
| 867 | *copied -= (send + delta); |
| 868 | err = -EACCES; |
| 869 | } |
| 870 | |
| 871 | if (likely(!err)) { |
| 872 | bool reset_eval = !ctx->open_rec; |
| 873 | |
| 874 | rec = ctx->open_rec; |
| 875 | if (rec) { |
| 876 | msg = &rec->msg_plaintext; |
| 877 | if (!msg->apply_bytes) |
| 878 | reset_eval = true; |
| 879 | } |
| 880 | if (reset_eval) { |
| 881 | psock->eval = __SK_NONE; |
| 882 | if (psock->sk_redir) { |
| 883 | sock_put(psock->sk_redir); |
| 884 | psock->sk_redir = NULL; |
| 885 | } |
| 886 | } |
| 887 | if (rec) |
| 888 | goto more_data; |
| 889 | } |
| 890 | out_err: |
| 891 | sk_psock_put(sk, psock); |
| 892 | return err; |
| 893 | } |
| 894 | |
| 895 | static int tls_sw_push_pending_record(struct sock *sk, int flags) |
| 896 | { |
| 897 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 898 | struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); |
| 899 | struct tls_rec *rec = ctx->open_rec; |
| 900 | struct sk_msg *msg_pl; |
| 901 | size_t copied; |
| 902 | |
| 903 | if (!rec) |
| 904 | return 0; |
| 905 | |
| 906 | msg_pl = &rec->msg_plaintext; |
| 907 | copied = msg_pl->sg.size; |
| 908 | if (!copied) |
| 909 | return 0; |
| 910 | |
| 911 | return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA, |
| 912 | &copied, flags); |
| 913 | } |
| 914 | |
| 915 | int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size) |
| 916 | { |
| 917 | long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT); |
| 918 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 919 | struct tls_prot_info *prot = &tls_ctx->prot_info; |
| 920 | struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); |
| 921 | bool async_capable = ctx->async_capable; |
| 922 | unsigned char record_type = TLS_RECORD_TYPE_DATA; |
| 923 | bool is_kvec = iov_iter_is_kvec(&msg->msg_iter); |
| 924 | bool eor = !(msg->msg_flags & MSG_MORE); |
| 925 | size_t try_to_copy; |
| 926 | ssize_t copied = 0; |
| 927 | struct sk_msg *msg_pl, *msg_en; |
| 928 | struct tls_rec *rec; |
| 929 | int required_size; |
| 930 | int num_async = 0; |
| 931 | bool full_record; |
| 932 | int record_room; |
| 933 | int num_zc = 0; |
| 934 | int orig_size; |
| 935 | int ret = 0; |
| 936 | int pending; |
| 937 | |
| 938 | if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL)) |
| 939 | return -EOPNOTSUPP; |
| 940 | |
| 941 | mutex_lock(&tls_ctx->tx_lock); |
| 942 | lock_sock(sk); |
| 943 | |
| 944 | if (unlikely(msg->msg_controllen)) { |
| 945 | ret = tls_proccess_cmsg(sk, msg, &record_type); |
| 946 | if (ret) { |
| 947 | if (ret == -EINPROGRESS) |
| 948 | num_async++; |
| 949 | else if (ret != -EAGAIN) |
| 950 | goto send_end; |
| 951 | } |
| 952 | } |
| 953 | |
| 954 | while (msg_data_left(msg)) { |
| 955 | if (sk->sk_err) { |
| 956 | ret = -sk->sk_err; |
| 957 | goto send_end; |
| 958 | } |
| 959 | |
| 960 | if (ctx->open_rec) |
| 961 | rec = ctx->open_rec; |
| 962 | else |
| 963 | rec = ctx->open_rec = tls_get_rec(sk); |
| 964 | if (!rec) { |
| 965 | ret = -ENOMEM; |
| 966 | goto send_end; |
| 967 | } |
| 968 | |
| 969 | msg_pl = &rec->msg_plaintext; |
| 970 | msg_en = &rec->msg_encrypted; |
| 971 | |
| 972 | orig_size = msg_pl->sg.size; |
| 973 | full_record = false; |
| 974 | try_to_copy = msg_data_left(msg); |
| 975 | record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size; |
| 976 | if (try_to_copy >= record_room) { |
| 977 | try_to_copy = record_room; |
| 978 | full_record = true; |
| 979 | } |
| 980 | |
| 981 | required_size = msg_pl->sg.size + try_to_copy + |
| 982 | prot->overhead_size; |
| 983 | |
| 984 | if (!sk_stream_memory_free(sk)) |
| 985 | goto wait_for_sndbuf; |
| 986 | |
| 987 | alloc_encrypted: |
| 988 | ret = tls_alloc_encrypted_msg(sk, required_size); |
| 989 | if (ret) { |
| 990 | if (ret != -ENOSPC) |
| 991 | goto wait_for_memory; |
| 992 | |
| 993 | /* Adjust try_to_copy according to the amount that was |
| 994 | * actually allocated. The difference is due |
| 995 | * to max sg elements limit |
| 996 | */ |
| 997 | try_to_copy -= required_size - msg_en->sg.size; |
| 998 | full_record = true; |
| 999 | } |
| 1000 | |
| 1001 | if (!is_kvec && (full_record || eor) && !async_capable) { |
| 1002 | u32 first = msg_pl->sg.end; |
| 1003 | |
| 1004 | ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter, |
| 1005 | msg_pl, try_to_copy); |
| 1006 | if (ret) |
| 1007 | goto fallback_to_reg_send; |
| 1008 | |
| 1009 | num_zc++; |
| 1010 | copied += try_to_copy; |
| 1011 | |
| 1012 | sk_msg_sg_copy_set(msg_pl, first); |
| 1013 | ret = bpf_exec_tx_verdict(msg_pl, sk, full_record, |
| 1014 | record_type, &copied, |
| 1015 | msg->msg_flags); |
| 1016 | if (ret) { |
| 1017 | if (ret == -EINPROGRESS) |
| 1018 | num_async++; |
| 1019 | else if (ret == -ENOMEM) |
| 1020 | goto wait_for_memory; |
| 1021 | else if (ctx->open_rec && ret == -ENOSPC) |
| 1022 | goto rollback_iter; |
| 1023 | else if (ret != -EAGAIN) |
| 1024 | goto send_end; |
| 1025 | } |
| 1026 | continue; |
| 1027 | rollback_iter: |
| 1028 | copied -= try_to_copy; |
| 1029 | sk_msg_sg_copy_clear(msg_pl, first); |
| 1030 | iov_iter_revert(&msg->msg_iter, |
| 1031 | msg_pl->sg.size - orig_size); |
| 1032 | fallback_to_reg_send: |
| 1033 | sk_msg_trim(sk, msg_pl, orig_size); |
| 1034 | } |
| 1035 | |
| 1036 | required_size = msg_pl->sg.size + try_to_copy; |
| 1037 | |
| 1038 | ret = tls_clone_plaintext_msg(sk, required_size); |
| 1039 | if (ret) { |
| 1040 | if (ret != -ENOSPC) |
| 1041 | goto send_end; |
| 1042 | |
| 1043 | /* Adjust try_to_copy according to the amount that was |
| 1044 | * actually allocated. The difference is due |
| 1045 | * to max sg elements limit |
| 1046 | */ |
| 1047 | try_to_copy -= required_size - msg_pl->sg.size; |
| 1048 | full_record = true; |
| 1049 | sk_msg_trim(sk, msg_en, |
| 1050 | msg_pl->sg.size + prot->overhead_size); |
| 1051 | } |
| 1052 | |
| 1053 | if (try_to_copy) { |
| 1054 | ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter, |
| 1055 | msg_pl, try_to_copy); |
| 1056 | if (ret < 0) |
| 1057 | goto trim_sgl; |
| 1058 | } |
| 1059 | |
| 1060 | /* Open records defined only if successfully copied, otherwise |
| 1061 | * we would trim the sg but not reset the open record frags. |
| 1062 | */ |
| 1063 | tls_ctx->pending_open_record_frags = true; |
| 1064 | copied += try_to_copy; |
| 1065 | if (full_record || eor) { |
| 1066 | ret = bpf_exec_tx_verdict(msg_pl, sk, full_record, |
| 1067 | record_type, &copied, |
| 1068 | msg->msg_flags); |
| 1069 | if (ret) { |
| 1070 | if (ret == -EINPROGRESS) |
| 1071 | num_async++; |
| 1072 | else if (ret == -ENOMEM) |
| 1073 | goto wait_for_memory; |
| 1074 | else if (ret != -EAGAIN) { |
| 1075 | if (ret == -ENOSPC) |
| 1076 | ret = 0; |
| 1077 | goto send_end; |
| 1078 | } |
| 1079 | } |
| 1080 | } |
| 1081 | |
| 1082 | continue; |
| 1083 | |
| 1084 | wait_for_sndbuf: |
| 1085 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
| 1086 | wait_for_memory: |
| 1087 | ret = sk_stream_wait_memory(sk, &timeo); |
| 1088 | if (ret) { |
| 1089 | trim_sgl: |
| 1090 | if (ctx->open_rec) |
| 1091 | tls_trim_both_msgs(sk, orig_size); |
| 1092 | goto send_end; |
| 1093 | } |
| 1094 | |
| 1095 | if (ctx->open_rec && msg_en->sg.size < required_size) |
| 1096 | goto alloc_encrypted; |
| 1097 | } |
| 1098 | |
| 1099 | if (!num_async) { |
| 1100 | goto send_end; |
| 1101 | } else if (num_zc) { |
| 1102 | /* Wait for pending encryptions to get completed */ |
| 1103 | spin_lock_bh(&ctx->encrypt_compl_lock); |
| 1104 | ctx->async_notify = true; |
| 1105 | |
| 1106 | pending = atomic_read(&ctx->encrypt_pending); |
| 1107 | spin_unlock_bh(&ctx->encrypt_compl_lock); |
| 1108 | if (pending) |
| 1109 | crypto_wait_req(-EINPROGRESS, &ctx->async_wait); |
| 1110 | else |
| 1111 | reinit_completion(&ctx->async_wait.completion); |
| 1112 | |
| 1113 | /* There can be no concurrent accesses, since we have no |
| 1114 | * pending encrypt operations |
| 1115 | */ |
| 1116 | WRITE_ONCE(ctx->async_notify, false); |
| 1117 | |
| 1118 | if (ctx->async_wait.err) { |
| 1119 | ret = ctx->async_wait.err; |
| 1120 | copied = 0; |
| 1121 | } |
| 1122 | } |
| 1123 | |
| 1124 | /* Transmit if any encryptions have completed */ |
| 1125 | if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) { |
| 1126 | cancel_delayed_work(&ctx->tx_work.work); |
| 1127 | tls_tx_records(sk, msg->msg_flags); |
| 1128 | } |
| 1129 | |
| 1130 | send_end: |
| 1131 | ret = sk_stream_error(sk, msg->msg_flags, ret); |
| 1132 | |
| 1133 | release_sock(sk); |
| 1134 | mutex_unlock(&tls_ctx->tx_lock); |
| 1135 | return copied > 0 ? copied : ret; |
| 1136 | } |
| 1137 | |
| 1138 | static int tls_sw_do_sendpage(struct sock *sk, struct page *page, |
| 1139 | int offset, size_t size, int flags) |
| 1140 | { |
| 1141 | long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); |
| 1142 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 1143 | struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); |
| 1144 | struct tls_prot_info *prot = &tls_ctx->prot_info; |
| 1145 | unsigned char record_type = TLS_RECORD_TYPE_DATA; |
| 1146 | struct sk_msg *msg_pl; |
| 1147 | struct tls_rec *rec; |
| 1148 | int num_async = 0; |
| 1149 | ssize_t copied = 0; |
| 1150 | bool full_record; |
| 1151 | int record_room; |
| 1152 | int ret = 0; |
| 1153 | bool eor; |
| 1154 | |
| 1155 | eor = !(flags & (MSG_MORE | MSG_SENDPAGE_NOTLAST)); |
| 1156 | sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
| 1157 | |
| 1158 | /* Call the sk_stream functions to manage the sndbuf mem. */ |
| 1159 | while (size > 0) { |
| 1160 | size_t copy, required_size; |
| 1161 | |
| 1162 | if (sk->sk_err) { |
| 1163 | ret = -sk->sk_err; |
| 1164 | goto sendpage_end; |
| 1165 | } |
| 1166 | |
| 1167 | if (ctx->open_rec) |
| 1168 | rec = ctx->open_rec; |
| 1169 | else |
| 1170 | rec = ctx->open_rec = tls_get_rec(sk); |
| 1171 | if (!rec) { |
| 1172 | ret = -ENOMEM; |
| 1173 | goto sendpage_end; |
| 1174 | } |
| 1175 | |
| 1176 | msg_pl = &rec->msg_plaintext; |
| 1177 | |
| 1178 | full_record = false; |
| 1179 | record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size; |
| 1180 | copy = size; |
| 1181 | if (copy >= record_room) { |
| 1182 | copy = record_room; |
| 1183 | full_record = true; |
| 1184 | } |
| 1185 | |
| 1186 | required_size = msg_pl->sg.size + copy + prot->overhead_size; |
| 1187 | |
| 1188 | if (!sk_stream_memory_free(sk)) |
| 1189 | goto wait_for_sndbuf; |
| 1190 | alloc_payload: |
| 1191 | ret = tls_alloc_encrypted_msg(sk, required_size); |
| 1192 | if (ret) { |
| 1193 | if (ret != -ENOSPC) |
| 1194 | goto wait_for_memory; |
| 1195 | |
| 1196 | /* Adjust copy according to the amount that was |
| 1197 | * actually allocated. The difference is due |
| 1198 | * to max sg elements limit |
| 1199 | */ |
| 1200 | copy -= required_size - msg_pl->sg.size; |
| 1201 | full_record = true; |
| 1202 | } |
| 1203 | |
| 1204 | sk_msg_page_add(msg_pl, page, copy, offset); |
| 1205 | sk_mem_charge(sk, copy); |
| 1206 | |
| 1207 | offset += copy; |
| 1208 | size -= copy; |
| 1209 | copied += copy; |
| 1210 | |
| 1211 | tls_ctx->pending_open_record_frags = true; |
| 1212 | if (full_record || eor || sk_msg_full(msg_pl)) { |
| 1213 | ret = bpf_exec_tx_verdict(msg_pl, sk, full_record, |
| 1214 | record_type, &copied, flags); |
| 1215 | if (ret) { |
| 1216 | if (ret == -EINPROGRESS) |
| 1217 | num_async++; |
| 1218 | else if (ret == -ENOMEM) |
| 1219 | goto wait_for_memory; |
| 1220 | else if (ret != -EAGAIN) { |
| 1221 | if (ret == -ENOSPC) |
| 1222 | ret = 0; |
| 1223 | goto sendpage_end; |
| 1224 | } |
| 1225 | } |
| 1226 | } |
| 1227 | continue; |
| 1228 | wait_for_sndbuf: |
| 1229 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
| 1230 | wait_for_memory: |
| 1231 | ret = sk_stream_wait_memory(sk, &timeo); |
| 1232 | if (ret) { |
| 1233 | if (ctx->open_rec) |
| 1234 | tls_trim_both_msgs(sk, msg_pl->sg.size); |
| 1235 | goto sendpage_end; |
| 1236 | } |
| 1237 | |
| 1238 | if (ctx->open_rec) |
| 1239 | goto alloc_payload; |
| 1240 | } |
| 1241 | |
| 1242 | if (num_async) { |
| 1243 | /* Transmit if any encryptions have completed */ |
| 1244 | if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) { |
| 1245 | cancel_delayed_work(&ctx->tx_work.work); |
| 1246 | tls_tx_records(sk, flags); |
| 1247 | } |
| 1248 | } |
| 1249 | sendpage_end: |
| 1250 | ret = sk_stream_error(sk, flags, ret); |
| 1251 | return copied > 0 ? copied : ret; |
| 1252 | } |
| 1253 | |
| 1254 | int tls_sw_sendpage_locked(struct sock *sk, struct page *page, |
| 1255 | int offset, size_t size, int flags) |
| 1256 | { |
| 1257 | if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | |
| 1258 | MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY | |
| 1259 | MSG_NO_SHARED_FRAGS)) |
| 1260 | return -EOPNOTSUPP; |
| 1261 | |
| 1262 | return tls_sw_do_sendpage(sk, page, offset, size, flags); |
| 1263 | } |
| 1264 | |
| 1265 | int tls_sw_sendpage(struct sock *sk, struct page *page, |
| 1266 | int offset, size_t size, int flags) |
| 1267 | { |
| 1268 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 1269 | int ret; |
| 1270 | |
| 1271 | if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | |
| 1272 | MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY)) |
| 1273 | return -EOPNOTSUPP; |
| 1274 | |
| 1275 | mutex_lock(&tls_ctx->tx_lock); |
| 1276 | lock_sock(sk); |
| 1277 | ret = tls_sw_do_sendpage(sk, page, offset, size, flags); |
| 1278 | release_sock(sk); |
| 1279 | mutex_unlock(&tls_ctx->tx_lock); |
| 1280 | return ret; |
| 1281 | } |
| 1282 | |
| 1283 | static struct sk_buff *tls_wait_data(struct sock *sk, struct sk_psock *psock, |
| 1284 | int flags, long timeo, int *err) |
| 1285 | { |
| 1286 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 1287 | struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); |
| 1288 | struct sk_buff *skb; |
| 1289 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
| 1290 | |
| 1291 | while (!(skb = ctx->recv_pkt) && sk_psock_queue_empty(psock)) { |
| 1292 | if (sk->sk_err) { |
| 1293 | *err = sock_error(sk); |
| 1294 | return NULL; |
| 1295 | } |
| 1296 | |
| 1297 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
| 1298 | return NULL; |
| 1299 | |
| 1300 | if (sock_flag(sk, SOCK_DONE)) |
| 1301 | return NULL; |
| 1302 | |
| 1303 | if ((flags & MSG_DONTWAIT) || !timeo) { |
| 1304 | *err = -EAGAIN; |
| 1305 | return NULL; |
| 1306 | } |
| 1307 | |
| 1308 | add_wait_queue(sk_sleep(sk), &wait); |
| 1309 | sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); |
| 1310 | sk_wait_event(sk, &timeo, |
| 1311 | ctx->recv_pkt != skb || |
| 1312 | !sk_psock_queue_empty(psock), |
| 1313 | &wait); |
| 1314 | sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); |
| 1315 | remove_wait_queue(sk_sleep(sk), &wait); |
| 1316 | |
| 1317 | /* Handle signals */ |
| 1318 | if (signal_pending(current)) { |
| 1319 | *err = sock_intr_errno(timeo); |
| 1320 | return NULL; |
| 1321 | } |
| 1322 | } |
| 1323 | |
| 1324 | return skb; |
| 1325 | } |
| 1326 | |
| 1327 | static int tls_setup_from_iter(struct sock *sk, struct iov_iter *from, |
| 1328 | int length, int *pages_used, |
| 1329 | unsigned int *size_used, |
| 1330 | struct scatterlist *to, |
| 1331 | int to_max_pages) |
| 1332 | { |
| 1333 | int rc = 0, i = 0, num_elem = *pages_used, maxpages; |
| 1334 | struct page *pages[MAX_SKB_FRAGS]; |
| 1335 | unsigned int size = *size_used; |
| 1336 | ssize_t copied, use; |
| 1337 | size_t offset; |
| 1338 | |
| 1339 | while (length > 0) { |
| 1340 | i = 0; |
| 1341 | maxpages = to_max_pages - num_elem; |
| 1342 | if (maxpages == 0) { |
| 1343 | rc = -EFAULT; |
| 1344 | goto out; |
| 1345 | } |
| 1346 | copied = iov_iter_get_pages(from, pages, |
| 1347 | length, |
| 1348 | maxpages, &offset); |
| 1349 | if (copied <= 0) { |
| 1350 | rc = -EFAULT; |
| 1351 | goto out; |
| 1352 | } |
| 1353 | |
| 1354 | iov_iter_advance(from, copied); |
| 1355 | |
| 1356 | length -= copied; |
| 1357 | size += copied; |
| 1358 | while (copied) { |
| 1359 | use = min_t(int, copied, PAGE_SIZE - offset); |
| 1360 | |
| 1361 | sg_set_page(&to[num_elem], |
| 1362 | pages[i], use, offset); |
| 1363 | sg_unmark_end(&to[num_elem]); |
| 1364 | /* We do not uncharge memory from this API */ |
| 1365 | |
| 1366 | offset = 0; |
| 1367 | copied -= use; |
| 1368 | |
| 1369 | i++; |
| 1370 | num_elem++; |
| 1371 | } |
| 1372 | } |
| 1373 | /* Mark the end in the last sg entry if newly added */ |
| 1374 | if (num_elem > *pages_used) |
| 1375 | sg_mark_end(&to[num_elem - 1]); |
| 1376 | out: |
| 1377 | if (rc) |
| 1378 | iov_iter_revert(from, size - *size_used); |
| 1379 | *size_used = size; |
| 1380 | *pages_used = num_elem; |
| 1381 | |
| 1382 | return rc; |
| 1383 | } |
| 1384 | |
| 1385 | /* This function decrypts the input skb into either out_iov or in out_sg |
| 1386 | * or in skb buffers itself. The input parameter 'zc' indicates if |
| 1387 | * zero-copy mode needs to be tried or not. With zero-copy mode, either |
| 1388 | * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are |
| 1389 | * NULL, then the decryption happens inside skb buffers itself, i.e. |
| 1390 | * zero-copy gets disabled and 'zc' is updated. |
| 1391 | */ |
| 1392 | |
| 1393 | static int decrypt_internal(struct sock *sk, struct sk_buff *skb, |
| 1394 | struct iov_iter *out_iov, |
| 1395 | struct scatterlist *out_sg, |
| 1396 | int *chunk, bool *zc, bool async) |
| 1397 | { |
| 1398 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 1399 | struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); |
| 1400 | struct tls_prot_info *prot = &tls_ctx->prot_info; |
| 1401 | struct strp_msg *rxm = strp_msg(skb); |
| 1402 | int n_sgin, n_sgout, nsg, mem_size, aead_size, err, pages = 0; |
| 1403 | struct aead_request *aead_req; |
| 1404 | struct sk_buff *unused; |
| 1405 | u8 *aad, *iv, *mem = NULL; |
| 1406 | struct scatterlist *sgin = NULL; |
| 1407 | struct scatterlist *sgout = NULL; |
| 1408 | const int data_len = rxm->full_len - prot->overhead_size + |
| 1409 | prot->tail_size; |
| 1410 | int iv_offset = 0; |
| 1411 | |
| 1412 | if (*zc && (out_iov || out_sg)) { |
| 1413 | if (out_iov) |
| 1414 | n_sgout = iov_iter_npages(out_iov, INT_MAX) + 1; |
| 1415 | else |
| 1416 | n_sgout = sg_nents(out_sg); |
| 1417 | n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size, |
| 1418 | rxm->full_len - prot->prepend_size); |
| 1419 | } else { |
| 1420 | n_sgout = 0; |
| 1421 | *zc = false; |
| 1422 | n_sgin = skb_cow_data(skb, 0, &unused); |
| 1423 | } |
| 1424 | |
| 1425 | if (n_sgin < 1) |
| 1426 | return -EBADMSG; |
| 1427 | |
| 1428 | /* Increment to accommodate AAD */ |
| 1429 | n_sgin = n_sgin + 1; |
| 1430 | |
| 1431 | nsg = n_sgin + n_sgout; |
| 1432 | |
| 1433 | aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv); |
| 1434 | mem_size = aead_size + (nsg * sizeof(struct scatterlist)); |
| 1435 | mem_size = mem_size + prot->aad_size; |
| 1436 | mem_size = mem_size + crypto_aead_ivsize(ctx->aead_recv); |
| 1437 | |
| 1438 | /* Allocate a single block of memory which contains |
| 1439 | * aead_req || sgin[] || sgout[] || aad || iv. |
| 1440 | * This order achieves correct alignment for aead_req, sgin, sgout. |
| 1441 | */ |
| 1442 | mem = kmalloc(mem_size, sk->sk_allocation); |
| 1443 | if (!mem) |
| 1444 | return -ENOMEM; |
| 1445 | |
| 1446 | /* Segment the allocated memory */ |
| 1447 | aead_req = (struct aead_request *)mem; |
| 1448 | sgin = (struct scatterlist *)(mem + aead_size); |
| 1449 | sgout = sgin + n_sgin; |
| 1450 | aad = (u8 *)(sgout + n_sgout); |
| 1451 | iv = aad + prot->aad_size; |
| 1452 | |
| 1453 | /* For CCM based ciphers, first byte of nonce+iv is always '2' */ |
| 1454 | if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) { |
| 1455 | iv[0] = 2; |
| 1456 | iv_offset = 1; |
| 1457 | } |
| 1458 | |
| 1459 | /* Prepare IV */ |
| 1460 | err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE, |
| 1461 | iv + iv_offset + prot->salt_size, |
| 1462 | prot->iv_size); |
| 1463 | if (err < 0) { |
| 1464 | kfree(mem); |
| 1465 | return err; |
| 1466 | } |
| 1467 | if (prot->version == TLS_1_3_VERSION) |
| 1468 | memcpy(iv + iv_offset, tls_ctx->rx.iv, |
| 1469 | crypto_aead_ivsize(ctx->aead_recv)); |
| 1470 | else |
| 1471 | memcpy(iv + iv_offset, tls_ctx->rx.iv, prot->salt_size); |
| 1472 | |
| 1473 | xor_iv_with_seq(prot->version, iv, tls_ctx->rx.rec_seq); |
| 1474 | |
| 1475 | /* Prepare AAD */ |
| 1476 | tls_make_aad(aad, rxm->full_len - prot->overhead_size + |
| 1477 | prot->tail_size, |
| 1478 | tls_ctx->rx.rec_seq, prot->rec_seq_size, |
| 1479 | ctx->control, prot->version); |
| 1480 | |
| 1481 | /* Prepare sgin */ |
| 1482 | sg_init_table(sgin, n_sgin); |
| 1483 | sg_set_buf(&sgin[0], aad, prot->aad_size); |
| 1484 | err = skb_to_sgvec(skb, &sgin[1], |
| 1485 | rxm->offset + prot->prepend_size, |
| 1486 | rxm->full_len - prot->prepend_size); |
| 1487 | if (err < 0) { |
| 1488 | kfree(mem); |
| 1489 | return err; |
| 1490 | } |
| 1491 | |
| 1492 | if (n_sgout) { |
| 1493 | if (out_iov) { |
| 1494 | sg_init_table(sgout, n_sgout); |
| 1495 | sg_set_buf(&sgout[0], aad, prot->aad_size); |
| 1496 | |
| 1497 | *chunk = 0; |
| 1498 | err = tls_setup_from_iter(sk, out_iov, data_len, |
| 1499 | &pages, chunk, &sgout[1], |
| 1500 | (n_sgout - 1)); |
| 1501 | if (err < 0) |
| 1502 | goto fallback_to_reg_recv; |
| 1503 | } else if (out_sg) { |
| 1504 | memcpy(sgout, out_sg, n_sgout * sizeof(*sgout)); |
| 1505 | } else { |
| 1506 | goto fallback_to_reg_recv; |
| 1507 | } |
| 1508 | } else { |
| 1509 | fallback_to_reg_recv: |
| 1510 | sgout = sgin; |
| 1511 | pages = 0; |
| 1512 | *chunk = data_len; |
| 1513 | *zc = false; |
| 1514 | } |
| 1515 | |
| 1516 | /* Prepare and submit AEAD request */ |
| 1517 | err = tls_do_decryption(sk, skb, sgin, sgout, iv, |
| 1518 | data_len, aead_req, async); |
| 1519 | if (err == -EINPROGRESS) |
| 1520 | return err; |
| 1521 | |
| 1522 | /* Release the pages in case iov was mapped to pages */ |
| 1523 | for (; pages > 0; pages--) |
| 1524 | put_page(sg_page(&sgout[pages])); |
| 1525 | |
| 1526 | kfree(mem); |
| 1527 | return err; |
| 1528 | } |
| 1529 | |
| 1530 | static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb, |
| 1531 | struct iov_iter *dest, int *chunk, bool *zc, |
| 1532 | bool async) |
| 1533 | { |
| 1534 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 1535 | struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); |
| 1536 | struct tls_prot_info *prot = &tls_ctx->prot_info; |
| 1537 | struct strp_msg *rxm = strp_msg(skb); |
| 1538 | int pad, err = 0; |
| 1539 | |
| 1540 | if (!ctx->decrypted) { |
| 1541 | if (tls_ctx->rx_conf == TLS_HW) { |
| 1542 | err = tls_device_decrypted(sk, tls_ctx, skb, rxm); |
| 1543 | if (err < 0) |
| 1544 | return err; |
| 1545 | } |
| 1546 | |
| 1547 | /* Still not decrypted after tls_device */ |
| 1548 | if (!ctx->decrypted) { |
| 1549 | err = decrypt_internal(sk, skb, dest, NULL, chunk, zc, |
| 1550 | async); |
| 1551 | if (err < 0) { |
| 1552 | if (err == -EINPROGRESS) |
| 1553 | tls_advance_record_sn(sk, prot, |
| 1554 | &tls_ctx->rx); |
| 1555 | else if (err == -EBADMSG) |
| 1556 | TLS_INC_STATS(sock_net(sk), |
| 1557 | LINUX_MIB_TLSDECRYPTERROR); |
| 1558 | return err; |
| 1559 | } |
| 1560 | } else { |
| 1561 | *zc = false; |
| 1562 | } |
| 1563 | |
| 1564 | pad = padding_length(ctx, prot, skb); |
| 1565 | if (pad < 0) |
| 1566 | return pad; |
| 1567 | |
| 1568 | rxm->full_len -= pad; |
| 1569 | rxm->offset += prot->prepend_size; |
| 1570 | rxm->full_len -= prot->overhead_size; |
| 1571 | tls_advance_record_sn(sk, prot, &tls_ctx->rx); |
| 1572 | ctx->decrypted = 1; |
| 1573 | ctx->saved_data_ready(sk); |
| 1574 | } else { |
| 1575 | *zc = false; |
| 1576 | } |
| 1577 | |
| 1578 | return err; |
| 1579 | } |
| 1580 | |
| 1581 | int decrypt_skb(struct sock *sk, struct sk_buff *skb, |
| 1582 | struct scatterlist *sgout) |
| 1583 | { |
| 1584 | bool zc = true; |
| 1585 | int chunk; |
| 1586 | |
| 1587 | return decrypt_internal(sk, skb, NULL, sgout, &chunk, &zc, false); |
| 1588 | } |
| 1589 | |
| 1590 | static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb, |
| 1591 | unsigned int len) |
| 1592 | { |
| 1593 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 1594 | struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); |
| 1595 | |
| 1596 | if (skb) { |
| 1597 | struct strp_msg *rxm = strp_msg(skb); |
| 1598 | |
| 1599 | if (len < rxm->full_len) { |
| 1600 | rxm->offset += len; |
| 1601 | rxm->full_len -= len; |
| 1602 | return false; |
| 1603 | } |
| 1604 | consume_skb(skb); |
| 1605 | } |
| 1606 | |
| 1607 | /* Finished with message */ |
| 1608 | ctx->recv_pkt = NULL; |
| 1609 | __strp_unpause(&ctx->strp); |
| 1610 | |
| 1611 | return true; |
| 1612 | } |
| 1613 | |
| 1614 | /* This function traverses the rx_list in tls receive context to copies the |
| 1615 | * decrypted records into the buffer provided by caller zero copy is not |
| 1616 | * true. Further, the records are removed from the rx_list if it is not a peek |
| 1617 | * case and the record has been consumed completely. |
| 1618 | */ |
| 1619 | static int process_rx_list(struct tls_sw_context_rx *ctx, |
| 1620 | struct msghdr *msg, |
| 1621 | u8 *control, |
| 1622 | bool *cmsg, |
| 1623 | size_t skip, |
| 1624 | size_t len, |
| 1625 | bool zc, |
| 1626 | bool is_peek) |
| 1627 | { |
| 1628 | struct sk_buff *skb = skb_peek(&ctx->rx_list); |
| 1629 | u8 ctrl = *control; |
| 1630 | u8 msgc = *cmsg; |
| 1631 | struct tls_msg *tlm; |
| 1632 | ssize_t copied = 0; |
| 1633 | |
| 1634 | /* Set the record type in 'control' if caller didn't pass it */ |
| 1635 | if (!ctrl && skb) { |
| 1636 | tlm = tls_msg(skb); |
| 1637 | ctrl = tlm->control; |
| 1638 | } |
| 1639 | |
| 1640 | while (skip && skb) { |
| 1641 | struct strp_msg *rxm = strp_msg(skb); |
| 1642 | tlm = tls_msg(skb); |
| 1643 | |
| 1644 | /* Cannot process a record of different type */ |
| 1645 | if (ctrl != tlm->control) |
| 1646 | return 0; |
| 1647 | |
| 1648 | if (skip < rxm->full_len) |
| 1649 | break; |
| 1650 | |
| 1651 | skip = skip - rxm->full_len; |
| 1652 | skb = skb_peek_next(skb, &ctx->rx_list); |
| 1653 | } |
| 1654 | |
| 1655 | while (len && skb) { |
| 1656 | struct sk_buff *next_skb; |
| 1657 | struct strp_msg *rxm = strp_msg(skb); |
| 1658 | int chunk = min_t(unsigned int, rxm->full_len - skip, len); |
| 1659 | |
| 1660 | tlm = tls_msg(skb); |
| 1661 | |
| 1662 | /* Cannot process a record of different type */ |
| 1663 | if (ctrl != tlm->control) |
| 1664 | return 0; |
| 1665 | |
| 1666 | /* Set record type if not already done. For a non-data record, |
| 1667 | * do not proceed if record type could not be copied. |
| 1668 | */ |
| 1669 | if (!msgc) { |
| 1670 | int cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE, |
| 1671 | sizeof(ctrl), &ctrl); |
| 1672 | msgc = true; |
| 1673 | if (ctrl != TLS_RECORD_TYPE_DATA) { |
| 1674 | if (cerr || msg->msg_flags & MSG_CTRUNC) |
| 1675 | return -EIO; |
| 1676 | |
| 1677 | *cmsg = msgc; |
| 1678 | } |
| 1679 | } |
| 1680 | |
| 1681 | if (!zc || (rxm->full_len - skip) > len) { |
| 1682 | int err = skb_copy_datagram_msg(skb, rxm->offset + skip, |
| 1683 | msg, chunk); |
| 1684 | if (err < 0) |
| 1685 | return err; |
| 1686 | } |
| 1687 | |
| 1688 | len = len - chunk; |
| 1689 | copied = copied + chunk; |
| 1690 | |
| 1691 | /* Consume the data from record if it is non-peek case*/ |
| 1692 | if (!is_peek) { |
| 1693 | rxm->offset = rxm->offset + chunk; |
| 1694 | rxm->full_len = rxm->full_len - chunk; |
| 1695 | |
| 1696 | /* Return if there is unconsumed data in the record */ |
| 1697 | if (rxm->full_len - skip) |
| 1698 | break; |
| 1699 | } |
| 1700 | |
| 1701 | /* The remaining skip-bytes must lie in 1st record in rx_list. |
| 1702 | * So from the 2nd record, 'skip' should be 0. |
| 1703 | */ |
| 1704 | skip = 0; |
| 1705 | |
| 1706 | if (msg) |
| 1707 | msg->msg_flags |= MSG_EOR; |
| 1708 | |
| 1709 | next_skb = skb_peek_next(skb, &ctx->rx_list); |
| 1710 | |
| 1711 | if (!is_peek) { |
| 1712 | skb_unlink(skb, &ctx->rx_list); |
| 1713 | consume_skb(skb); |
| 1714 | } |
| 1715 | |
| 1716 | skb = next_skb; |
| 1717 | } |
| 1718 | |
| 1719 | *control = ctrl; |
| 1720 | return copied; |
| 1721 | } |
| 1722 | |
| 1723 | int tls_sw_recvmsg(struct sock *sk, |
| 1724 | struct msghdr *msg, |
| 1725 | size_t len, |
| 1726 | int nonblock, |
| 1727 | int flags, |
| 1728 | int *addr_len) |
| 1729 | { |
| 1730 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 1731 | struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); |
| 1732 | struct tls_prot_info *prot = &tls_ctx->prot_info; |
| 1733 | struct sk_psock *psock; |
| 1734 | unsigned char control = 0; |
| 1735 | ssize_t decrypted = 0; |
| 1736 | struct strp_msg *rxm; |
| 1737 | struct tls_msg *tlm; |
| 1738 | struct sk_buff *skb; |
| 1739 | ssize_t copied = 0; |
| 1740 | bool cmsg = false; |
| 1741 | int target, err = 0; |
| 1742 | long timeo; |
| 1743 | bool is_kvec = iov_iter_is_kvec(&msg->msg_iter); |
| 1744 | bool is_peek = flags & MSG_PEEK; |
| 1745 | bool bpf_strp_enabled; |
| 1746 | int num_async = 0; |
| 1747 | int pending; |
| 1748 | |
| 1749 | flags |= nonblock; |
| 1750 | |
| 1751 | if (unlikely(flags & MSG_ERRQUEUE)) |
| 1752 | return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR); |
| 1753 | |
| 1754 | psock = sk_psock_get(sk); |
| 1755 | lock_sock(sk); |
| 1756 | bpf_strp_enabled = sk_psock_strp_enabled(psock); |
| 1757 | |
| 1758 | /* Process pending decrypted records. It must be non-zero-copy */ |
| 1759 | err = process_rx_list(ctx, msg, &control, &cmsg, 0, len, false, |
| 1760 | is_peek); |
| 1761 | if (err < 0) { |
| 1762 | tls_err_abort(sk, err); |
| 1763 | goto end; |
| 1764 | } else { |
| 1765 | copied = err; |
| 1766 | } |
| 1767 | |
| 1768 | if (len <= copied) |
| 1769 | goto recv_end; |
| 1770 | |
| 1771 | target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); |
| 1772 | len = len - copied; |
| 1773 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); |
| 1774 | |
| 1775 | while (len && (decrypted + copied < target || ctx->recv_pkt)) { |
| 1776 | bool retain_skb = false; |
| 1777 | bool zc = false; |
| 1778 | int to_decrypt; |
| 1779 | int chunk = 0; |
| 1780 | bool async_capable; |
| 1781 | bool async = false; |
| 1782 | |
| 1783 | skb = tls_wait_data(sk, psock, flags, timeo, &err); |
| 1784 | if (!skb) { |
| 1785 | if (psock) { |
| 1786 | int ret = __tcp_bpf_recvmsg(sk, psock, |
| 1787 | msg, len, flags); |
| 1788 | |
| 1789 | if (ret > 0) { |
| 1790 | decrypted += ret; |
| 1791 | len -= ret; |
| 1792 | continue; |
| 1793 | } |
| 1794 | } |
| 1795 | goto recv_end; |
| 1796 | } else { |
| 1797 | tlm = tls_msg(skb); |
| 1798 | if (prot->version == TLS_1_3_VERSION) |
| 1799 | tlm->control = 0; |
| 1800 | else |
| 1801 | tlm->control = ctx->control; |
| 1802 | } |
| 1803 | |
| 1804 | rxm = strp_msg(skb); |
| 1805 | |
| 1806 | to_decrypt = rxm->full_len - prot->overhead_size; |
| 1807 | |
| 1808 | if (to_decrypt <= len && !is_kvec && !is_peek && |
| 1809 | ctx->control == TLS_RECORD_TYPE_DATA && |
| 1810 | prot->version != TLS_1_3_VERSION && |
| 1811 | !bpf_strp_enabled) |
| 1812 | zc = true; |
| 1813 | |
| 1814 | /* Do not use async mode if record is non-data */ |
| 1815 | if (ctx->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled) |
| 1816 | async_capable = ctx->async_capable; |
| 1817 | else |
| 1818 | async_capable = false; |
| 1819 | |
| 1820 | err = decrypt_skb_update(sk, skb, &msg->msg_iter, |
| 1821 | &chunk, &zc, async_capable); |
| 1822 | if (err < 0 && err != -EINPROGRESS) { |
| 1823 | tls_err_abort(sk, EBADMSG); |
| 1824 | goto recv_end; |
| 1825 | } |
| 1826 | |
| 1827 | if (err == -EINPROGRESS) { |
| 1828 | async = true; |
| 1829 | num_async++; |
| 1830 | } else if (prot->version == TLS_1_3_VERSION) { |
| 1831 | tlm->control = ctx->control; |
| 1832 | } |
| 1833 | |
| 1834 | /* If the type of records being processed is not known yet, |
| 1835 | * set it to record type just dequeued. If it is already known, |
| 1836 | * but does not match the record type just dequeued, go to end. |
| 1837 | * We always get record type here since for tls1.2, record type |
| 1838 | * is known just after record is dequeued from stream parser. |
| 1839 | * For tls1.3, we disable async. |
| 1840 | */ |
| 1841 | |
| 1842 | if (!control) |
| 1843 | control = tlm->control; |
| 1844 | else if (control != tlm->control) |
| 1845 | goto recv_end; |
| 1846 | |
| 1847 | if (!cmsg) { |
| 1848 | int cerr; |
| 1849 | |
| 1850 | cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE, |
| 1851 | sizeof(control), &control); |
| 1852 | cmsg = true; |
| 1853 | if (control != TLS_RECORD_TYPE_DATA) { |
| 1854 | if (cerr || msg->msg_flags & MSG_CTRUNC) { |
| 1855 | err = -EIO; |
| 1856 | goto recv_end; |
| 1857 | } |
| 1858 | } |
| 1859 | } |
| 1860 | |
| 1861 | if (async) |
| 1862 | goto pick_next_record; |
| 1863 | |
| 1864 | if (!zc) { |
| 1865 | if (bpf_strp_enabled) { |
| 1866 | err = sk_psock_tls_strp_read(psock, skb); |
| 1867 | if (err != __SK_PASS) { |
| 1868 | rxm->offset = rxm->offset + rxm->full_len; |
| 1869 | rxm->full_len = 0; |
| 1870 | if (err == __SK_DROP) |
| 1871 | consume_skb(skb); |
| 1872 | ctx->recv_pkt = NULL; |
| 1873 | __strp_unpause(&ctx->strp); |
| 1874 | continue; |
| 1875 | } |
| 1876 | } |
| 1877 | |
| 1878 | if (rxm->full_len > len) { |
| 1879 | retain_skb = true; |
| 1880 | chunk = len; |
| 1881 | } else { |
| 1882 | chunk = rxm->full_len; |
| 1883 | } |
| 1884 | |
| 1885 | err = skb_copy_datagram_msg(skb, rxm->offset, |
| 1886 | msg, chunk); |
| 1887 | if (err < 0) |
| 1888 | goto recv_end; |
| 1889 | |
| 1890 | if (!is_peek) { |
| 1891 | rxm->offset = rxm->offset + chunk; |
| 1892 | rxm->full_len = rxm->full_len - chunk; |
| 1893 | } |
| 1894 | } |
| 1895 | |
| 1896 | pick_next_record: |
| 1897 | if (chunk > len) |
| 1898 | chunk = len; |
| 1899 | |
| 1900 | decrypted += chunk; |
| 1901 | len -= chunk; |
| 1902 | |
| 1903 | /* For async or peek case, queue the current skb */ |
| 1904 | if (async || is_peek || retain_skb) { |
| 1905 | skb_queue_tail(&ctx->rx_list, skb); |
| 1906 | skb = NULL; |
| 1907 | } |
| 1908 | |
| 1909 | if (tls_sw_advance_skb(sk, skb, chunk)) { |
| 1910 | /* Return full control message to |
| 1911 | * userspace before trying to parse |
| 1912 | * another message type |
| 1913 | */ |
| 1914 | msg->msg_flags |= MSG_EOR; |
| 1915 | if (ctx->control != TLS_RECORD_TYPE_DATA) |
| 1916 | goto recv_end; |
| 1917 | } else { |
| 1918 | break; |
| 1919 | } |
| 1920 | } |
| 1921 | |
| 1922 | recv_end: |
| 1923 | if (num_async) { |
| 1924 | /* Wait for all previously submitted records to be decrypted */ |
| 1925 | spin_lock_bh(&ctx->decrypt_compl_lock); |
| 1926 | ctx->async_notify = true; |
| 1927 | pending = atomic_read(&ctx->decrypt_pending); |
| 1928 | spin_unlock_bh(&ctx->decrypt_compl_lock); |
| 1929 | if (pending) { |
| 1930 | err = crypto_wait_req(-EINPROGRESS, &ctx->async_wait); |
| 1931 | if (err) { |
| 1932 | /* one of async decrypt failed */ |
| 1933 | tls_err_abort(sk, err); |
| 1934 | copied = 0; |
| 1935 | decrypted = 0; |
| 1936 | goto end; |
| 1937 | } |
| 1938 | } else { |
| 1939 | reinit_completion(&ctx->async_wait.completion); |
| 1940 | } |
| 1941 | |
| 1942 | /* There can be no concurrent accesses, since we have no |
| 1943 | * pending decrypt operations |
| 1944 | */ |
| 1945 | WRITE_ONCE(ctx->async_notify, false); |
| 1946 | |
| 1947 | /* Drain records from the rx_list & copy if required */ |
| 1948 | if (is_peek || is_kvec) |
| 1949 | err = process_rx_list(ctx, msg, &control, &cmsg, copied, |
| 1950 | decrypted, false, is_peek); |
| 1951 | else |
| 1952 | err = process_rx_list(ctx, msg, &control, &cmsg, 0, |
| 1953 | decrypted, true, is_peek); |
| 1954 | if (err < 0) { |
| 1955 | tls_err_abort(sk, err); |
| 1956 | copied = 0; |
| 1957 | goto end; |
| 1958 | } |
| 1959 | } |
| 1960 | |
| 1961 | copied += decrypted; |
| 1962 | |
| 1963 | end: |
| 1964 | release_sock(sk); |
| 1965 | if (psock) |
| 1966 | sk_psock_put(sk, psock); |
| 1967 | return copied ? : err; |
| 1968 | } |
| 1969 | |
| 1970 | ssize_t tls_sw_splice_read(struct socket *sock, loff_t *ppos, |
| 1971 | struct pipe_inode_info *pipe, |
| 1972 | size_t len, unsigned int flags) |
| 1973 | { |
| 1974 | struct tls_context *tls_ctx = tls_get_ctx(sock->sk); |
| 1975 | struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); |
| 1976 | struct strp_msg *rxm = NULL; |
| 1977 | struct sock *sk = sock->sk; |
| 1978 | struct sk_buff *skb; |
| 1979 | ssize_t copied = 0; |
| 1980 | int err = 0; |
| 1981 | long timeo; |
| 1982 | int chunk; |
| 1983 | bool zc = false; |
| 1984 | |
| 1985 | lock_sock(sk); |
| 1986 | |
| 1987 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); |
| 1988 | |
| 1989 | skb = tls_wait_data(sk, NULL, flags, timeo, &err); |
| 1990 | if (!skb) |
| 1991 | goto splice_read_end; |
| 1992 | |
| 1993 | if (!ctx->decrypted) { |
| 1994 | err = decrypt_skb_update(sk, skb, NULL, &chunk, &zc, false); |
| 1995 | |
| 1996 | /* splice does not support reading control messages */ |
| 1997 | if (ctx->control != TLS_RECORD_TYPE_DATA) { |
| 1998 | err = -EINVAL; |
| 1999 | goto splice_read_end; |
| 2000 | } |
| 2001 | |
| 2002 | if (err < 0) { |
| 2003 | tls_err_abort(sk, EBADMSG); |
| 2004 | goto splice_read_end; |
| 2005 | } |
| 2006 | ctx->decrypted = 1; |
| 2007 | } |
| 2008 | rxm = strp_msg(skb); |
| 2009 | |
| 2010 | chunk = min_t(unsigned int, rxm->full_len, len); |
| 2011 | copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags); |
| 2012 | if (copied < 0) |
| 2013 | goto splice_read_end; |
| 2014 | |
| 2015 | if (likely(!(flags & MSG_PEEK))) |
| 2016 | tls_sw_advance_skb(sk, skb, copied); |
| 2017 | |
| 2018 | splice_read_end: |
| 2019 | release_sock(sk); |
| 2020 | return copied ? : err; |
| 2021 | } |
| 2022 | |
| 2023 | bool tls_sw_stream_read(const struct sock *sk) |
| 2024 | { |
| 2025 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 2026 | struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); |
| 2027 | bool ingress_empty = true; |
| 2028 | struct sk_psock *psock; |
| 2029 | |
| 2030 | rcu_read_lock(); |
| 2031 | psock = sk_psock(sk); |
| 2032 | if (psock) |
| 2033 | ingress_empty = list_empty(&psock->ingress_msg); |
| 2034 | rcu_read_unlock(); |
| 2035 | |
| 2036 | return !ingress_empty || ctx->recv_pkt || |
| 2037 | !skb_queue_empty(&ctx->rx_list); |
| 2038 | } |
| 2039 | |
| 2040 | static int tls_read_size(struct strparser *strp, struct sk_buff *skb) |
| 2041 | { |
| 2042 | struct tls_context *tls_ctx = tls_get_ctx(strp->sk); |
| 2043 | struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); |
| 2044 | struct tls_prot_info *prot = &tls_ctx->prot_info; |
| 2045 | char header[TLS_HEADER_SIZE + MAX_IV_SIZE]; |
| 2046 | struct strp_msg *rxm = strp_msg(skb); |
| 2047 | size_t cipher_overhead; |
| 2048 | size_t data_len = 0; |
| 2049 | int ret; |
| 2050 | |
| 2051 | /* Verify that we have a full TLS header, or wait for more data */ |
| 2052 | if (rxm->offset + prot->prepend_size > skb->len) |
| 2053 | return 0; |
| 2054 | |
| 2055 | /* Sanity-check size of on-stack buffer. */ |
| 2056 | if (WARN_ON(prot->prepend_size > sizeof(header))) { |
| 2057 | ret = -EINVAL; |
| 2058 | goto read_failure; |
| 2059 | } |
| 2060 | |
| 2061 | /* Linearize header to local buffer */ |
| 2062 | ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size); |
| 2063 | |
| 2064 | if (ret < 0) |
| 2065 | goto read_failure; |
| 2066 | |
| 2067 | ctx->control = header[0]; |
| 2068 | |
| 2069 | data_len = ((header[4] & 0xFF) | (header[3] << 8)); |
| 2070 | |
| 2071 | cipher_overhead = prot->tag_size; |
| 2072 | if (prot->version != TLS_1_3_VERSION) |
| 2073 | cipher_overhead += prot->iv_size; |
| 2074 | |
| 2075 | if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead + |
| 2076 | prot->tail_size) { |
| 2077 | ret = -EMSGSIZE; |
| 2078 | goto read_failure; |
| 2079 | } |
| 2080 | if (data_len < cipher_overhead) { |
| 2081 | ret = -EBADMSG; |
| 2082 | goto read_failure; |
| 2083 | } |
| 2084 | |
| 2085 | /* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */ |
| 2086 | if (header[1] != TLS_1_2_VERSION_MINOR || |
| 2087 | header[2] != TLS_1_2_VERSION_MAJOR) { |
| 2088 | ret = -EINVAL; |
| 2089 | goto read_failure; |
| 2090 | } |
| 2091 | |
| 2092 | tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE, |
| 2093 | TCP_SKB_CB(skb)->seq + rxm->offset); |
| 2094 | return data_len + TLS_HEADER_SIZE; |
| 2095 | |
| 2096 | read_failure: |
| 2097 | tls_err_abort(strp->sk, ret); |
| 2098 | |
| 2099 | return ret; |
| 2100 | } |
| 2101 | |
| 2102 | static void tls_queue(struct strparser *strp, struct sk_buff *skb) |
| 2103 | { |
| 2104 | struct tls_context *tls_ctx = tls_get_ctx(strp->sk); |
| 2105 | struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); |
| 2106 | |
| 2107 | ctx->decrypted = 0; |
| 2108 | |
| 2109 | ctx->recv_pkt = skb; |
| 2110 | strp_pause(strp); |
| 2111 | |
| 2112 | ctx->saved_data_ready(strp->sk); |
| 2113 | } |
| 2114 | |
| 2115 | static void tls_data_ready(struct sock *sk) |
| 2116 | { |
| 2117 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 2118 | struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); |
| 2119 | struct sk_psock *psock; |
| 2120 | |
| 2121 | strp_data_ready(&ctx->strp); |
| 2122 | |
| 2123 | psock = sk_psock_get(sk); |
| 2124 | if (psock) { |
| 2125 | if (!list_empty(&psock->ingress_msg)) |
| 2126 | ctx->saved_data_ready(sk); |
| 2127 | sk_psock_put(sk, psock); |
| 2128 | } |
| 2129 | } |
| 2130 | |
| 2131 | void tls_sw_cancel_work_tx(struct tls_context *tls_ctx) |
| 2132 | { |
| 2133 | struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); |
| 2134 | |
| 2135 | set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask); |
| 2136 | set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask); |
| 2137 | cancel_delayed_work_sync(&ctx->tx_work.work); |
| 2138 | } |
| 2139 | |
| 2140 | void tls_sw_release_resources_tx(struct sock *sk) |
| 2141 | { |
| 2142 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 2143 | struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); |
| 2144 | struct tls_rec *rec, *tmp; |
| 2145 | |
| 2146 | /* Wait for any pending async encryptions to complete */ |
| 2147 | smp_store_mb(ctx->async_notify, true); |
| 2148 | if (atomic_read(&ctx->encrypt_pending)) |
| 2149 | crypto_wait_req(-EINPROGRESS, &ctx->async_wait); |
| 2150 | |
| 2151 | tls_tx_records(sk, -1); |
| 2152 | |
| 2153 | /* Free up un-sent records in tx_list. First, free |
| 2154 | * the partially sent record if any at head of tx_list. |
| 2155 | */ |
| 2156 | if (tls_ctx->partially_sent_record) { |
| 2157 | tls_free_partial_record(sk, tls_ctx); |
| 2158 | rec = list_first_entry(&ctx->tx_list, |
| 2159 | struct tls_rec, list); |
| 2160 | list_del(&rec->list); |
| 2161 | sk_msg_free(sk, &rec->msg_plaintext); |
| 2162 | kfree(rec); |
| 2163 | } |
| 2164 | |
| 2165 | list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) { |
| 2166 | list_del(&rec->list); |
| 2167 | sk_msg_free(sk, &rec->msg_encrypted); |
| 2168 | sk_msg_free(sk, &rec->msg_plaintext); |
| 2169 | kfree(rec); |
| 2170 | } |
| 2171 | |
| 2172 | crypto_free_aead(ctx->aead_send); |
| 2173 | tls_free_open_rec(sk); |
| 2174 | } |
| 2175 | |
| 2176 | void tls_sw_free_ctx_tx(struct tls_context *tls_ctx) |
| 2177 | { |
| 2178 | struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx); |
| 2179 | |
| 2180 | kfree(ctx); |
| 2181 | } |
| 2182 | |
| 2183 | void tls_sw_release_resources_rx(struct sock *sk) |
| 2184 | { |
| 2185 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 2186 | struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); |
| 2187 | |
| 2188 | kfree(tls_ctx->rx.rec_seq); |
| 2189 | kfree(tls_ctx->rx.iv); |
| 2190 | |
| 2191 | if (ctx->aead_recv) { |
| 2192 | kfree_skb(ctx->recv_pkt); |
| 2193 | ctx->recv_pkt = NULL; |
| 2194 | skb_queue_purge(&ctx->rx_list); |
| 2195 | crypto_free_aead(ctx->aead_recv); |
| 2196 | strp_stop(&ctx->strp); |
| 2197 | /* If tls_sw_strparser_arm() was not called (cleanup paths) |
| 2198 | * we still want to strp_stop(), but sk->sk_data_ready was |
| 2199 | * never swapped. |
| 2200 | */ |
| 2201 | if (ctx->saved_data_ready) { |
| 2202 | write_lock_bh(&sk->sk_callback_lock); |
| 2203 | sk->sk_data_ready = ctx->saved_data_ready; |
| 2204 | write_unlock_bh(&sk->sk_callback_lock); |
| 2205 | } |
| 2206 | } |
| 2207 | } |
| 2208 | |
| 2209 | void tls_sw_strparser_done(struct tls_context *tls_ctx) |
| 2210 | { |
| 2211 | struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); |
| 2212 | |
| 2213 | strp_done(&ctx->strp); |
| 2214 | } |
| 2215 | |
| 2216 | void tls_sw_free_ctx_rx(struct tls_context *tls_ctx) |
| 2217 | { |
| 2218 | struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx); |
| 2219 | |
| 2220 | kfree(ctx); |
| 2221 | } |
| 2222 | |
| 2223 | void tls_sw_free_resources_rx(struct sock *sk) |
| 2224 | { |
| 2225 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 2226 | |
| 2227 | tls_sw_release_resources_rx(sk); |
| 2228 | tls_sw_free_ctx_rx(tls_ctx); |
| 2229 | } |
| 2230 | |
| 2231 | /* The work handler to transmitt the encrypted records in tx_list */ |
| 2232 | static void tx_work_handler(struct work_struct *work) |
| 2233 | { |
| 2234 | struct delayed_work *delayed_work = to_delayed_work(work); |
| 2235 | struct tx_work *tx_work = container_of(delayed_work, |
| 2236 | struct tx_work, work); |
| 2237 | struct sock *sk = tx_work->sk; |
| 2238 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 2239 | struct tls_sw_context_tx *ctx; |
| 2240 | |
| 2241 | if (unlikely(!tls_ctx)) |
| 2242 | return; |
| 2243 | |
| 2244 | ctx = tls_sw_ctx_tx(tls_ctx); |
| 2245 | if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask)) |
| 2246 | return; |
| 2247 | |
| 2248 | if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) |
| 2249 | return; |
| 2250 | mutex_lock(&tls_ctx->tx_lock); |
| 2251 | lock_sock(sk); |
| 2252 | tls_tx_records(sk, -1); |
| 2253 | release_sock(sk); |
| 2254 | mutex_unlock(&tls_ctx->tx_lock); |
| 2255 | } |
| 2256 | |
| 2257 | void tls_sw_write_space(struct sock *sk, struct tls_context *ctx) |
| 2258 | { |
| 2259 | struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx); |
| 2260 | |
| 2261 | /* Schedule the transmission if tx list is ready */ |
| 2262 | if (is_tx_ready(tx_ctx) && |
| 2263 | !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask)) |
| 2264 | schedule_delayed_work(&tx_ctx->tx_work.work, 0); |
| 2265 | } |
| 2266 | |
| 2267 | void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx) |
| 2268 | { |
| 2269 | struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx); |
| 2270 | |
| 2271 | write_lock_bh(&sk->sk_callback_lock); |
| 2272 | rx_ctx->saved_data_ready = sk->sk_data_ready; |
| 2273 | sk->sk_data_ready = tls_data_ready; |
| 2274 | write_unlock_bh(&sk->sk_callback_lock); |
| 2275 | |
| 2276 | strp_check_rcv(&rx_ctx->strp); |
| 2277 | } |
| 2278 | |
| 2279 | int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx) |
| 2280 | { |
| 2281 | struct tls_context *tls_ctx = tls_get_ctx(sk); |
| 2282 | struct tls_prot_info *prot = &tls_ctx->prot_info; |
| 2283 | struct tls_crypto_info *crypto_info; |
| 2284 | struct tls12_crypto_info_aes_gcm_128 *gcm_128_info; |
| 2285 | struct tls12_crypto_info_aes_gcm_256 *gcm_256_info; |
| 2286 | struct tls12_crypto_info_aes_ccm_128 *ccm_128_info; |
| 2287 | struct tls_sw_context_tx *sw_ctx_tx = NULL; |
| 2288 | struct tls_sw_context_rx *sw_ctx_rx = NULL; |
| 2289 | struct cipher_context *cctx; |
| 2290 | struct crypto_aead **aead; |
| 2291 | struct strp_callbacks cb; |
| 2292 | u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size; |
| 2293 | struct crypto_tfm *tfm; |
| 2294 | char *iv, *rec_seq, *key, *salt, *cipher_name; |
| 2295 | size_t keysize; |
| 2296 | int rc = 0; |
| 2297 | |
| 2298 | if (!ctx) { |
| 2299 | rc = -EINVAL; |
| 2300 | goto out; |
| 2301 | } |
| 2302 | |
| 2303 | if (tx) { |
| 2304 | if (!ctx->priv_ctx_tx) { |
| 2305 | sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL); |
| 2306 | if (!sw_ctx_tx) { |
| 2307 | rc = -ENOMEM; |
| 2308 | goto out; |
| 2309 | } |
| 2310 | ctx->priv_ctx_tx = sw_ctx_tx; |
| 2311 | } else { |
| 2312 | sw_ctx_tx = |
| 2313 | (struct tls_sw_context_tx *)ctx->priv_ctx_tx; |
| 2314 | } |
| 2315 | } else { |
| 2316 | if (!ctx->priv_ctx_rx) { |
| 2317 | sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL); |
| 2318 | if (!sw_ctx_rx) { |
| 2319 | rc = -ENOMEM; |
| 2320 | goto out; |
| 2321 | } |
| 2322 | ctx->priv_ctx_rx = sw_ctx_rx; |
| 2323 | } else { |
| 2324 | sw_ctx_rx = |
| 2325 | (struct tls_sw_context_rx *)ctx->priv_ctx_rx; |
| 2326 | } |
| 2327 | } |
| 2328 | |
| 2329 | if (tx) { |
| 2330 | crypto_init_wait(&sw_ctx_tx->async_wait); |
| 2331 | spin_lock_init(&sw_ctx_tx->encrypt_compl_lock); |
| 2332 | crypto_info = &ctx->crypto_send.info; |
| 2333 | cctx = &ctx->tx; |
| 2334 | aead = &sw_ctx_tx->aead_send; |
| 2335 | INIT_LIST_HEAD(&sw_ctx_tx->tx_list); |
| 2336 | INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler); |
| 2337 | sw_ctx_tx->tx_work.sk = sk; |
| 2338 | } else { |
| 2339 | crypto_init_wait(&sw_ctx_rx->async_wait); |
| 2340 | spin_lock_init(&sw_ctx_rx->decrypt_compl_lock); |
| 2341 | crypto_info = &ctx->crypto_recv.info; |
| 2342 | cctx = &ctx->rx; |
| 2343 | skb_queue_head_init(&sw_ctx_rx->rx_list); |
| 2344 | aead = &sw_ctx_rx->aead_recv; |
| 2345 | } |
| 2346 | |
| 2347 | switch (crypto_info->cipher_type) { |
| 2348 | case TLS_CIPHER_AES_GCM_128: { |
| 2349 | nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE; |
| 2350 | tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE; |
| 2351 | iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE; |
| 2352 | iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv; |
| 2353 | rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE; |
| 2354 | rec_seq = |
| 2355 | ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq; |
| 2356 | gcm_128_info = |
| 2357 | (struct tls12_crypto_info_aes_gcm_128 *)crypto_info; |
| 2358 | keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE; |
| 2359 | key = gcm_128_info->key; |
| 2360 | salt = gcm_128_info->salt; |
| 2361 | salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE; |
| 2362 | cipher_name = "gcm(aes)"; |
| 2363 | break; |
| 2364 | } |
| 2365 | case TLS_CIPHER_AES_GCM_256: { |
| 2366 | nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE; |
| 2367 | tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE; |
| 2368 | iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE; |
| 2369 | iv = ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->iv; |
| 2370 | rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE; |
| 2371 | rec_seq = |
| 2372 | ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->rec_seq; |
| 2373 | gcm_256_info = |
| 2374 | (struct tls12_crypto_info_aes_gcm_256 *)crypto_info; |
| 2375 | keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE; |
| 2376 | key = gcm_256_info->key; |
| 2377 | salt = gcm_256_info->salt; |
| 2378 | salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE; |
| 2379 | cipher_name = "gcm(aes)"; |
| 2380 | break; |
| 2381 | } |
| 2382 | case TLS_CIPHER_AES_CCM_128: { |
| 2383 | nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE; |
| 2384 | tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE; |
| 2385 | iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE; |
| 2386 | iv = ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->iv; |
| 2387 | rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE; |
| 2388 | rec_seq = |
| 2389 | ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->rec_seq; |
| 2390 | ccm_128_info = |
| 2391 | (struct tls12_crypto_info_aes_ccm_128 *)crypto_info; |
| 2392 | keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE; |
| 2393 | key = ccm_128_info->key; |
| 2394 | salt = ccm_128_info->salt; |
| 2395 | salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE; |
| 2396 | cipher_name = "ccm(aes)"; |
| 2397 | break; |
| 2398 | } |
| 2399 | default: |
| 2400 | rc = -EINVAL; |
| 2401 | goto free_priv; |
| 2402 | } |
| 2403 | |
| 2404 | /* Sanity-check the sizes for stack allocations. */ |
| 2405 | if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE || |
| 2406 | rec_seq_size > TLS_MAX_REC_SEQ_SIZE) { |
| 2407 | rc = -EINVAL; |
| 2408 | goto free_priv; |
| 2409 | } |
| 2410 | |
| 2411 | if (crypto_info->version == TLS_1_3_VERSION) { |
| 2412 | nonce_size = 0; |
| 2413 | prot->aad_size = TLS_HEADER_SIZE; |
| 2414 | prot->tail_size = 1; |
| 2415 | } else { |
| 2416 | prot->aad_size = TLS_AAD_SPACE_SIZE; |
| 2417 | prot->tail_size = 0; |
| 2418 | } |
| 2419 | |
| 2420 | prot->version = crypto_info->version; |
| 2421 | prot->cipher_type = crypto_info->cipher_type; |
| 2422 | prot->prepend_size = TLS_HEADER_SIZE + nonce_size; |
| 2423 | prot->tag_size = tag_size; |
| 2424 | prot->overhead_size = prot->prepend_size + |
| 2425 | prot->tag_size + prot->tail_size; |
| 2426 | prot->iv_size = iv_size; |
| 2427 | prot->salt_size = salt_size; |
| 2428 | cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL); |
| 2429 | if (!cctx->iv) { |
| 2430 | rc = -ENOMEM; |
| 2431 | goto free_priv; |
| 2432 | } |
| 2433 | /* Note: 128 & 256 bit salt are the same size */ |
| 2434 | prot->rec_seq_size = rec_seq_size; |
| 2435 | memcpy(cctx->iv, salt, salt_size); |
| 2436 | memcpy(cctx->iv + salt_size, iv, iv_size); |
| 2437 | cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL); |
| 2438 | if (!cctx->rec_seq) { |
| 2439 | rc = -ENOMEM; |
| 2440 | goto free_iv; |
| 2441 | } |
| 2442 | |
| 2443 | if (!*aead) { |
| 2444 | *aead = crypto_alloc_aead(cipher_name, 0, 0); |
| 2445 | if (IS_ERR(*aead)) { |
| 2446 | rc = PTR_ERR(*aead); |
| 2447 | *aead = NULL; |
| 2448 | goto free_rec_seq; |
| 2449 | } |
| 2450 | } |
| 2451 | |
| 2452 | ctx->push_pending_record = tls_sw_push_pending_record; |
| 2453 | |
| 2454 | rc = crypto_aead_setkey(*aead, key, keysize); |
| 2455 | |
| 2456 | if (rc) |
| 2457 | goto free_aead; |
| 2458 | |
| 2459 | rc = crypto_aead_setauthsize(*aead, prot->tag_size); |
| 2460 | if (rc) |
| 2461 | goto free_aead; |
| 2462 | |
| 2463 | if (sw_ctx_rx) { |
| 2464 | tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv); |
| 2465 | |
| 2466 | if (crypto_info->version == TLS_1_3_VERSION) |
| 2467 | sw_ctx_rx->async_capable = 0; |
| 2468 | else |
| 2469 | sw_ctx_rx->async_capable = |
| 2470 | !!(tfm->__crt_alg->cra_flags & |
| 2471 | CRYPTO_ALG_ASYNC); |
| 2472 | |
| 2473 | /* Set up strparser */ |
| 2474 | memset(&cb, 0, sizeof(cb)); |
| 2475 | cb.rcv_msg = tls_queue; |
| 2476 | cb.parse_msg = tls_read_size; |
| 2477 | |
| 2478 | strp_init(&sw_ctx_rx->strp, sk, &cb); |
| 2479 | } |
| 2480 | |
| 2481 | goto out; |
| 2482 | |
| 2483 | free_aead: |
| 2484 | crypto_free_aead(*aead); |
| 2485 | *aead = NULL; |
| 2486 | free_rec_seq: |
| 2487 | kfree(cctx->rec_seq); |
| 2488 | cctx->rec_seq = NULL; |
| 2489 | free_iv: |
| 2490 | kfree(cctx->iv); |
| 2491 | cctx->iv = NULL; |
| 2492 | free_priv: |
| 2493 | if (tx) { |
| 2494 | kfree(ctx->priv_ctx_tx); |
| 2495 | ctx->priv_ctx_tx = NULL; |
| 2496 | } else { |
| 2497 | kfree(ctx->priv_ctx_rx); |
| 2498 | ctx->priv_ctx_rx = NULL; |
| 2499 | } |
| 2500 | out: |
| 2501 | return rc; |
| 2502 | } |