net/tls/tls_sw.c

   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 #include "tls.h"
  48
  49 struct tls_decrypt_arg {
  50         struct_group(inargs,
  51         bool zc;
  52         bool async;
  53         u8 tail;
  54         );
  55
  56         struct sk_buff *skb;
  57 };
  58
  59 struct tls_decrypt_ctx {
  60         u8 iv[MAX_IV_SIZE];
  61         u8 aad[TLS_MAX_AAD_SIZE];
  62         u8 tail;
  63         struct scatterlist sg[];
  64 };
  65
  66 noinline void tls_err_abort(struct sock *sk, int err)
  67 {
  68         WARN_ON_ONCE(err >= 0);
  69         /* sk->sk_err should contain a positive error code. */
  70         sk->sk_err = -err;
  71         sk_error_report(sk);
  72 }
  73
  74 static int __skb_nsg(struct sk_buff *skb, int offset, int len,
  75                      unsigned int recursion_level)
  76 {
  77         int start = skb_headlen(skb);
  78         int i, chunk = start - offset;
  79         struct sk_buff *frag_iter;
  80         int elt = 0;
  81
  82         if (unlikely(recursion_level >= 24))
  83                 return -EMSGSIZE;
  84
  85         if (chunk > 0) {
  86                 if (chunk > len)
  87                         chunk = len;
  88                 elt++;
  89                 len -= chunk;
  90                 if (len == 0)
  91                         return elt;
  92                 offset += chunk;
  93         }
  94
  95         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
  96                 int end;
  97
  98                 WARN_ON(start > offset + len);
  99
 100                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
 101                 chunk = end - offset;
 102                 if (chunk > 0) {
 103                         if (chunk > len)
 104                                 chunk = len;
 105                         elt++;
 106                         len -= chunk;
 107                         if (len == 0)
 108                                 return elt;
 109                         offset += chunk;
 110                 }
 111                 start = end;
 112         }
 113
 114         if (unlikely(skb_has_frag_list(skb))) {
 115                 skb_walk_frags(skb, frag_iter) {
 116                         int end, ret;
 117
 118                         WARN_ON(start > offset + len);
 119
 120                         end = start + frag_iter->len;
 121                         chunk = end - offset;
 122                         if (chunk > 0) {
 123                                 if (chunk > len)
 124                                         chunk = len;
 125                                 ret = __skb_nsg(frag_iter, offset - start, chunk,
 126                                                 recursion_level + 1);
 127                                 if (unlikely(ret < 0))
 128                                         return ret;
 129                                 elt += ret;
 130                                 len -= chunk;
 131                                 if (len == 0)
 132                                         return elt;
 133                                 offset += chunk;
 134                         }
 135                         start = end;
 136                 }
 137         }
 138         BUG_ON(len);
 139         return elt;
 140 }
 141
 142 /* Return the number of scatterlist elements required to completely map the
 143  * skb, or -EMSGSIZE if the recursion depth is exceeded.
 144  */
 145 static int skb_nsg(struct sk_buff *skb, int offset, int len)
 146 {
 147         return __skb_nsg(skb, offset, len, 0);
 148 }
 149
 150 static int tls_padding_length(struct tls_prot_info *prot, struct sk_buff *skb,
 151                               struct tls_decrypt_arg *darg)
 152 {
 153         struct strp_msg *rxm = strp_msg(skb);
 154         struct tls_msg *tlm = tls_msg(skb);
 155         int sub = 0;
 156
 157         /* Determine zero-padding length */
 158         if (prot->version == TLS_1_3_VERSION) {
 159                 int offset = rxm->full_len - TLS_TAG_SIZE - 1;
 160                 char content_type = darg->zc ? darg->tail : 0;
 161                 int err;
 162
 163                 while (content_type == 0) {
 164                         if (offset < prot->prepend_size)
 165                                 return -EBADMSG;
 166                         err = skb_copy_bits(skb, rxm->offset + offset,
 167                                             &content_type, 1);
 168                         if (err)
 169                                 return err;
 170                         if (content_type)
 171                                 break;
 172                         sub++;
 173                         offset--;
 174                 }
 175                 tlm->control = content_type;
 176         }
 177         return sub;
 178 }
 179
 180 static void tls_decrypt_done(struct crypto_async_request *req, int err)
 181 {
 182         struct aead_request *aead_req = (struct aead_request *)req;
 183         struct scatterlist *sgout = aead_req->dst;
 184         struct scatterlist *sgin = aead_req->src;
 185         struct tls_sw_context_rx *ctx;
 186         struct tls_context *tls_ctx;
 187         struct scatterlist *sg;
 188         unsigned int pages;
 189         struct sock *sk;
 190
 191         sk = (struct sock *)req->data;
 192         tls_ctx = tls_get_ctx(sk);
 193         ctx = tls_sw_ctx_rx(tls_ctx);
 194
 195         /* Propagate if there was an err */
 196         if (err) {
 197                 if (err == -EBADMSG)
 198                         TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTERROR);
 199                 ctx->async_wait.err = err;
 200                 tls_err_abort(sk, err);
 201         }
 202
 203         /* Free the destination pages if skb was not decrypted inplace */
 204         if (sgout != sgin) {
 205                 /* Skip the first S/G entry as it points to AAD */
 206                 for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) {
 207                         if (!sg)
 208                                 break;
 209                         put_page(sg_page(sg));
 210                 }
 211         }
 212
 213         kfree(aead_req);
 214
 215         spin_lock_bh(&ctx->decrypt_compl_lock);
 216         if (!atomic_dec_return(&ctx->decrypt_pending))
 217                 complete(&ctx->async_wait.completion);
 218         spin_unlock_bh(&ctx->decrypt_compl_lock);
 219 }
 220
 221 static int tls_do_decryption(struct sock *sk,
 222                              struct scatterlist *sgin,
 223                              struct scatterlist *sgout,
 224                              char *iv_recv,
 225                              size_t data_len,
 226                              struct aead_request *aead_req,
 227                              struct tls_decrypt_arg *darg)
 228 {
 229         struct tls_context *tls_ctx = tls_get_ctx(sk);
 230         struct tls_prot_info *prot = &tls_ctx->prot_info;
 231         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
 232         int ret;
 233
 234         aead_request_set_tfm(aead_req, ctx->aead_recv);
 235         aead_request_set_ad(aead_req, prot->aad_size);
 236         aead_request_set_crypt(aead_req, sgin, sgout,
 237                                data_len + prot->tag_size,
 238                                (u8 *)iv_recv);
 239
 240         if (darg->async) {
 241                 aead_request_set_callback(aead_req,
 242                                           CRYPTO_TFM_REQ_MAY_BACKLOG,
 243                                           tls_decrypt_done, sk);
 244                 atomic_inc(&ctx->decrypt_pending);
 245         } else {
 246                 aead_request_set_callback(aead_req,
 247                                           CRYPTO_TFM_REQ_MAY_BACKLOG,
 248                                           crypto_req_done, &ctx->async_wait);
 249         }
 250
 251         ret = crypto_aead_decrypt(aead_req);
 252         if (ret == -EINPROGRESS) {
 253                 if (darg->async)
 254                         return 0;
 255
 256                 ret = crypto_wait_req(ret, &ctx->async_wait);
 257         }
 258         darg->async = false;
 259
 260         return ret;
 261 }
 262
 263 static void tls_trim_both_msgs(struct sock *sk, int target_size)
 264 {
 265         struct tls_context *tls_ctx = tls_get_ctx(sk);
 266         struct tls_prot_info *prot = &tls_ctx->prot_info;
 267         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 268         struct tls_rec *rec = ctx->open_rec;
 269
 270         sk_msg_trim(sk, &rec->msg_plaintext, target_size);
 271         if (target_size > 0)
 272                 target_size += prot->overhead_size;
 273         sk_msg_trim(sk, &rec->msg_encrypted, target_size);
 274 }
 275
 276 static int tls_alloc_encrypted_msg(struct sock *sk, int len)
 277 {
 278         struct tls_context *tls_ctx = tls_get_ctx(sk);
 279         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 280         struct tls_rec *rec = ctx->open_rec;
 281         struct sk_msg *msg_en = &rec->msg_encrypted;
 282
 283         return sk_msg_alloc(sk, msg_en, len, 0);
 284 }
 285
 286 static int tls_clone_plaintext_msg(struct sock *sk, int required)
 287 {
 288         struct tls_context *tls_ctx = tls_get_ctx(sk);
 289         struct tls_prot_info *prot = &tls_ctx->prot_info;
 290         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 291         struct tls_rec *rec = ctx->open_rec;
 292         struct sk_msg *msg_pl = &rec->msg_plaintext;
 293         struct sk_msg *msg_en = &rec->msg_encrypted;
 294         int skip, len;
 295
 296         /* We add page references worth len bytes from encrypted sg
 297          * at the end of plaintext sg. It is guaranteed that msg_en
 298          * has enough required room (ensured by caller).
 299          */
 300         len = required - msg_pl->sg.size;
 301
 302         /* Skip initial bytes in msg_en's data to be able to use
 303          * same offset of both plain and encrypted data.
 304          */
 305         skip = prot->prepend_size + msg_pl->sg.size;
 306
 307         return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
 308 }
 309
 310 static struct tls_rec *tls_get_rec(struct sock *sk)
 311 {
 312         struct tls_context *tls_ctx = tls_get_ctx(sk);
 313         struct tls_prot_info *prot = &tls_ctx->prot_info;
 314         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 315         struct sk_msg *msg_pl, *msg_en;
 316         struct tls_rec *rec;
 317         int mem_size;
 318
 319         mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
 320
 321         rec = kzalloc(mem_size, sk->sk_allocation);
 322         if (!rec)
 323                 return NULL;
 324
 325         msg_pl = &rec->msg_plaintext;
 326         msg_en = &rec->msg_encrypted;
 327
 328         sk_msg_init(msg_pl);
 329         sk_msg_init(msg_en);
 330
 331         sg_init_table(rec->sg_aead_in, 2);
 332         sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size);
 333         sg_unmark_end(&rec->sg_aead_in[1]);
 334
 335         sg_init_table(rec->sg_aead_out, 2);
 336         sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size);
 337         sg_unmark_end(&rec->sg_aead_out[1]);
 338
 339         return rec;
 340 }
 341
 342 static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
 343 {
 344         sk_msg_free(sk, &rec->msg_encrypted);
 345         sk_msg_free(sk, &rec->msg_plaintext);
 346         kfree(rec);
 347 }
 348
 349 static void tls_free_open_rec(struct sock *sk)
 350 {
 351         struct tls_context *tls_ctx = tls_get_ctx(sk);
 352         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 353         struct tls_rec *rec = ctx->open_rec;
 354
 355         if (rec) {
 356                 tls_free_rec(sk, rec);
 357                 ctx->open_rec = NULL;
 358         }
 359 }
 360
 361 int tls_tx_records(struct sock *sk, int flags)
 362 {
 363         struct tls_context *tls_ctx = tls_get_ctx(sk);
 364         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 365         struct tls_rec *rec, *tmp;
 366         struct sk_msg *msg_en;
 367         int tx_flags, rc = 0;
 368
 369         if (tls_is_partially_sent_record(tls_ctx)) {
 370                 rec = list_first_entry(&ctx->tx_list,
 371                                        struct tls_rec, list);
 372
 373                 if (flags == -1)
 374                         tx_flags = rec->tx_flags;
 375                 else
 376                         tx_flags = flags;
 377
 378                 rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
 379                 if (rc)
 380                         goto tx_err;
 381
 382                 /* Full record has been transmitted.
 383                  * Remove the head of tx_list
 384                  */
 385                 list_del(&rec->list);
 386                 sk_msg_free(sk, &rec->msg_plaintext);
 387                 kfree(rec);
 388         }
 389
 390         /* Tx all ready records */
 391         list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
 392                 if (READ_ONCE(rec->tx_ready)) {
 393                         if (flags == -1)
 394                                 tx_flags = rec->tx_flags;
 395                         else
 396                                 tx_flags = flags;
 397
 398                         msg_en = &rec->msg_encrypted;
 399                         rc = tls_push_sg(sk, tls_ctx,
 400                                          &msg_en->sg.data[msg_en->sg.curr],
 401                                          0, tx_flags);
 402                         if (rc)
 403                                 goto tx_err;
 404
 405                         list_del(&rec->list);
 406                         sk_msg_free(sk, &rec->msg_plaintext);
 407                         kfree(rec);
 408                 } else {
 409                         break;
 410                 }
 411         }
 412
 413 tx_err:
 414         if (rc < 0 && rc != -EAGAIN)
 415                 tls_err_abort(sk, -EBADMSG);
 416
 417         return rc;
 418 }
 419
 420 static void tls_encrypt_done(struct crypto_async_request *req, int err)
 421 {
 422         struct aead_request *aead_req = (struct aead_request *)req;
 423         struct sock *sk = req->data;
 424         struct tls_context *tls_ctx = tls_get_ctx(sk);
 425         struct tls_prot_info *prot = &tls_ctx->prot_info;
 426         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 427         struct scatterlist *sge;
 428         struct sk_msg *msg_en;
 429         struct tls_rec *rec;
 430         bool ready = false;
 431         int pending;
 432
 433         rec = container_of(aead_req, struct tls_rec, aead_req);
 434         msg_en = &rec->msg_encrypted;
 435
 436         sge = sk_msg_elem(msg_en, msg_en->sg.curr);
 437         sge->offset -= prot->prepend_size;
 438         sge->length += prot->prepend_size;
 439
 440         /* Check if error is previously set on socket */
 441         if (err || sk->sk_err) {
 442                 rec = NULL;
 443
 444                 /* If err is already set on socket, return the same code */
 445                 if (sk->sk_err) {
 446                         ctx->async_wait.err = -sk->sk_err;
 447                 } else {
 448                         ctx->async_wait.err = err;
 449                         tls_err_abort(sk, err);
 450                 }
 451         }
 452
 453         if (rec) {
 454                 struct tls_rec *first_rec;
 455
 456                 /* Mark the record as ready for transmission */
 457                 smp_store_mb(rec->tx_ready, true);
 458
 459                 /* If received record is at head of tx_list, schedule tx */
 460                 first_rec = list_first_entry(&ctx->tx_list,
 461                                              struct tls_rec, list);
 462                 if (rec == first_rec)
 463                         ready = true;
 464         }
 465
 466         spin_lock_bh(&ctx->encrypt_compl_lock);
 467         pending = atomic_dec_return(&ctx->encrypt_pending);
 468
 469         if (!pending && ctx->async_notify)
 470                 complete(&ctx->async_wait.completion);
 471         spin_unlock_bh(&ctx->encrypt_compl_lock);
 472
 473         if (!ready)
 474                 return;
 475
 476         /* Schedule the transmission */
 477         if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
 478                 schedule_delayed_work(&ctx->tx_work.work, 1);
 479 }
 480
 481 static int tls_do_encryption(struct sock *sk,
 482                              struct tls_context *tls_ctx,
 483                              struct tls_sw_context_tx *ctx,
 484                              struct aead_request *aead_req,
 485                              size_t data_len, u32 start)
 486 {
 487         struct tls_prot_info *prot = &tls_ctx->prot_info;
 488         struct tls_rec *rec = ctx->open_rec;
 489         struct sk_msg *msg_en = &rec->msg_encrypted;
 490         struct scatterlist *sge = sk_msg_elem(msg_en, start);
 491         int rc, iv_offset = 0;
 492
 493         /* For CCM based ciphers, first byte of IV is a constant */
 494         switch (prot->cipher_type) {
 495         case TLS_CIPHER_AES_CCM_128:
 496                 rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
 497                 iv_offset = 1;
 498                 break;
 499         case TLS_CIPHER_SM4_CCM:
 500                 rec->iv_data[0] = TLS_SM4_CCM_IV_B0_BYTE;
 501                 iv_offset = 1;
 502                 break;
 503         }
 504
 505         memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
 506                prot->iv_size + prot->salt_size);
 507
 508         tls_xor_iv_with_seq(prot, rec->iv_data + iv_offset,
 509                             tls_ctx->tx.rec_seq);
 510
 511         sge->offset += prot->prepend_size;
 512         sge->length -= prot->prepend_size;
 513
 514         msg_en->sg.curr = start;
 515
 516         aead_request_set_tfm(aead_req, ctx->aead_send);
 517         aead_request_set_ad(aead_req, prot->aad_size);
 518         aead_request_set_crypt(aead_req, rec->sg_aead_in,
 519                                rec->sg_aead_out,
 520                                data_len, rec->iv_data);
 521
 522         aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 523                                   tls_encrypt_done, sk);
 524
 525         /* Add the record in tx_list */
 526         list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
 527         atomic_inc(&ctx->encrypt_pending);
 528
 529         rc = crypto_aead_encrypt(aead_req);
 530         if (!rc || rc != -EINPROGRESS) {
 531                 atomic_dec(&ctx->encrypt_pending);
 532                 sge->offset -= prot->prepend_size;
 533                 sge->length += prot->prepend_size;
 534         }
 535
 536         if (!rc) {
 537                 WRITE_ONCE(rec->tx_ready, true);
 538         } else if (rc != -EINPROGRESS) {
 539                 list_del(&rec->list);
 540                 return rc;
 541         }
 542
 543         /* Unhook the record from context if encryption is not failure */
 544         ctx->open_rec = NULL;
 545         tls_advance_record_sn(sk, prot, &tls_ctx->tx);
 546         return rc;
 547 }
 548
 549 static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
 550                                  struct tls_rec **to, struct sk_msg *msg_opl,
 551                                  struct sk_msg *msg_oen, u32 split_point,
 552                                  u32 tx_overhead_size, u32 *orig_end)
 553 {
 554         u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
 555         struct scatterlist *sge, *osge, *nsge;
 556         u32 orig_size = msg_opl->sg.size;
 557         struct scatterlist tmp = { };
 558         struct sk_msg *msg_npl;
 559         struct tls_rec *new;
 560         int ret;
 561
 562         new = tls_get_rec(sk);
 563         if (!new)
 564                 return -ENOMEM;
 565         ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size +
 566                            tx_overhead_size, 0);
 567         if (ret < 0) {
 568                 tls_free_rec(sk, new);
 569                 return ret;
 570         }
 571
 572         *orig_end = msg_opl->sg.end;
 573         i = msg_opl->sg.start;
 574         sge = sk_msg_elem(msg_opl, i);
 575         while (apply && sge->length) {
 576                 if (sge->length > apply) {
 577                         u32 len = sge->length - apply;
 578
 579                         get_page(sg_page(sge));
 580                         sg_set_page(&tmp, sg_page(sge), len,
 581                                     sge->offset + apply);
 582                         sge->length = apply;
 583                         bytes += apply;
 584                         apply = 0;
 585                 } else {
 586                         apply -= sge->length;
 587                         bytes += sge->length;
 588                 }
 589
 590                 sk_msg_iter_var_next(i);
 591                 if (i == msg_opl->sg.end)
 592                         break;
 593                 sge = sk_msg_elem(msg_opl, i);
 594         }
 595
 596         msg_opl->sg.end = i;
 597         msg_opl->sg.curr = i;
 598         msg_opl->sg.copybreak = 0;
 599         msg_opl->apply_bytes = 0;
 600         msg_opl->sg.size = bytes;
 601
 602         msg_npl = &new->msg_plaintext;
 603         msg_npl->apply_bytes = apply;
 604         msg_npl->sg.size = orig_size - bytes;
 605
 606         j = msg_npl->sg.start;
 607         nsge = sk_msg_elem(msg_npl, j);
 608         if (tmp.length) {
 609                 memcpy(nsge, &tmp, sizeof(*nsge));
 610                 sk_msg_iter_var_next(j);
 611                 nsge = sk_msg_elem(msg_npl, j);
 612         }
 613
 614         osge = sk_msg_elem(msg_opl, i);
 615         while (osge->length) {
 616                 memcpy(nsge, osge, sizeof(*nsge));
 617                 sg_unmark_end(nsge);
 618                 sk_msg_iter_var_next(i);
 619                 sk_msg_iter_var_next(j);
 620                 if (i == *orig_end)
 621                         break;
 622                 osge = sk_msg_elem(msg_opl, i);
 623                 nsge = sk_msg_elem(msg_npl, j);
 624         }
 625
 626         msg_npl->sg.end = j;
 627         msg_npl->sg.curr = j;
 628         msg_npl->sg.copybreak = 0;
 629
 630         *to = new;
 631         return 0;
 632 }
 633
 634 static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
 635                                   struct tls_rec *from, u32 orig_end)
 636 {
 637         struct sk_msg *msg_npl = &from->msg_plaintext;
 638         struct sk_msg *msg_opl = &to->msg_plaintext;
 639         struct scatterlist *osge, *nsge;
 640         u32 i, j;
 641
 642         i = msg_opl->sg.end;
 643         sk_msg_iter_var_prev(i);
 644         j = msg_npl->sg.start;
 645
 646         osge = sk_msg_elem(msg_opl, i);
 647         nsge = sk_msg_elem(msg_npl, j);
 648
 649         if (sg_page(osge) == sg_page(nsge) &&
 650             osge->offset + osge->length == nsge->offset) {
 651                 osge->length += nsge->length;
 652                 put_page(sg_page(nsge));
 653         }
 654
 655         msg_opl->sg.end = orig_end;
 656         msg_opl->sg.curr = orig_end;
 657         msg_opl->sg.copybreak = 0;
 658         msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
 659         msg_opl->sg.size += msg_npl->sg.size;
 660
 661         sk_msg_free(sk, &to->msg_encrypted);
 662         sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted);
 663
 664         kfree(from);
 665 }
 666
 667 static int tls_push_record(struct sock *sk, int flags,
 668                            unsigned char record_type)
 669 {
 670         struct tls_context *tls_ctx = tls_get_ctx(sk);
 671         struct tls_prot_info *prot = &tls_ctx->prot_info;
 672         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 673         struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
 674         u32 i, split_point, orig_end;
 675         struct sk_msg *msg_pl, *msg_en;
 676         struct aead_request *req;
 677         bool split;
 678         int rc;
 679
 680         if (!rec)
 681                 return 0;
 682
 683         msg_pl = &rec->msg_plaintext;
 684         msg_en = &rec->msg_encrypted;
 685
 686         split_point = msg_pl->apply_bytes;
 687         split = split_point && split_point < msg_pl->sg.size;
 688         if (unlikely((!split &&
 689                       msg_pl->sg.size +
 690                       prot->overhead_size > msg_en->sg.size) ||
 691                      (split &&
 692                       split_point +
 693                       prot->overhead_size > msg_en->sg.size))) {
 694                 split = true;
 695                 split_point = msg_en->sg.size;
 696         }
 697         if (split) {
 698                 rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en,
 699                                            split_point, prot->overhead_size,
 700                                            &orig_end);
 701                 if (rc < 0)
 702                         return rc;
 703                 /* This can happen if above tls_split_open_record allocates
 704                  * a single large encryption buffer instead of two smaller
 705                  * ones. In this case adjust pointers and continue without
 706                  * split.
 707                  */
 708                 if (!msg_pl->sg.size) {
 709                         tls_merge_open_record(sk, rec, tmp, orig_end);
 710                         msg_pl = &rec->msg_plaintext;
 711                         msg_en = &rec->msg_encrypted;
 712                         split = false;
 713                 }
 714                 sk_msg_trim(sk, msg_en, msg_pl->sg.size +
 715                             prot->overhead_size);
 716         }
 717
 718         rec->tx_flags = flags;
 719         req = &rec->aead_req;
 720
 721         i = msg_pl->sg.end;
 722         sk_msg_iter_var_prev(i);
 723
 724         rec->content_type = record_type;
 725         if (prot->version == TLS_1_3_VERSION) {
 726                 /* Add content type to end of message.  No padding added */
 727                 sg_set_buf(&rec->sg_content_type, &rec->content_type, 1);
 728                 sg_mark_end(&rec->sg_content_type);
 729                 sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1,
 730                          &rec->sg_content_type);
 731         } else {
 732                 sg_mark_end(sk_msg_elem(msg_pl, i));
 733         }
 734
 735         if (msg_pl->sg.end < msg_pl->sg.start) {
 736                 sg_chain(&msg_pl->sg.data[msg_pl->sg.start],
 737                          MAX_SKB_FRAGS - msg_pl->sg.start + 1,
 738                          msg_pl->sg.data);
 739         }
 740
 741         i = msg_pl->sg.start;
 742         sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]);
 743
 744         i = msg_en->sg.end;
 745         sk_msg_iter_var_prev(i);
 746         sg_mark_end(sk_msg_elem(msg_en, i));
 747
 748         i = msg_en->sg.start;
 749         sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
 750
 751         tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size,
 752                      tls_ctx->tx.rec_seq, record_type, prot);
 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);
 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                                MSG_CMSG_COMPAT))
 940                 return -EOPNOTSUPP;
 941
 942         mutex_lock(&tls_ctx->tx_lock);
 943         lock_sock(sk);
 944
 945         if (unlikely(msg->msg_controllen)) {
 946                 ret = tls_process_cmsg(sk, msg, &record_type);
 947                 if (ret) {
 948                         if (ret == -EINPROGRESS)
 949                                 num_async++;
 950                         else if (ret != -EAGAIN)
 951                                 goto send_end;
 952                 }
 953         }
 954
 955         while (msg_data_left(msg)) {
 956                 if (sk->sk_err) {
 957                         ret = -sk->sk_err;
 958                         goto send_end;
 959                 }
 960
 961                 if (ctx->open_rec)
 962                         rec = ctx->open_rec;
 963                 else
 964                         rec = ctx->open_rec = tls_get_rec(sk);
 965                 if (!rec) {
 966                         ret = -ENOMEM;
 967                         goto send_end;
 968                 }
 969
 970                 msg_pl = &rec->msg_plaintext;
 971                 msg_en = &rec->msg_encrypted;
 972
 973                 orig_size = msg_pl->sg.size;
 974                 full_record = false;
 975                 try_to_copy = msg_data_left(msg);
 976                 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
 977                 if (try_to_copy >= record_room) {
 978                         try_to_copy = record_room;
 979                         full_record = true;
 980                 }
 981
 982                 required_size = msg_pl->sg.size + try_to_copy +
 983                                 prot->overhead_size;
 984
 985                 if (!sk_stream_memory_free(sk))
 986                         goto wait_for_sndbuf;
 987
 988 alloc_encrypted:
 989                 ret = tls_alloc_encrypted_msg(sk, required_size);
 990                 if (ret) {
 991                         if (ret != -ENOSPC)
 992                                 goto wait_for_memory;
 993
 994                         /* Adjust try_to_copy according to the amount that was
 995                          * actually allocated. The difference is due
 996                          * to max sg elements limit
 997                          */
 998                         try_to_copy -= required_size - msg_en->sg.size;
 999                         full_record = true;
1000                 }
1001
1002                 if (!is_kvec && (full_record || eor) && !async_capable) {
1003                         u32 first = msg_pl->sg.end;
1004
1005                         ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter,
1006                                                         msg_pl, try_to_copy);
1007                         if (ret)
1008                                 goto fallback_to_reg_send;
1009
1010                         num_zc++;
1011                         copied += try_to_copy;
1012
1013                         sk_msg_sg_copy_set(msg_pl, first);
1014                         ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1015                                                   record_type, &copied,
1016                                                   msg->msg_flags);
1017                         if (ret) {
1018                                 if (ret == -EINPROGRESS)
1019                                         num_async++;
1020                                 else if (ret == -ENOMEM)
1021                                         goto wait_for_memory;
1022                                 else if (ctx->open_rec && ret == -ENOSPC)
1023                                         goto rollback_iter;
1024                                 else if (ret != -EAGAIN)
1025                                         goto send_end;
1026                         }
1027                         continue;
1028 rollback_iter:
1029                         copied -= try_to_copy;
1030                         sk_msg_sg_copy_clear(msg_pl, first);
1031                         iov_iter_revert(&msg->msg_iter,
1032                                         msg_pl->sg.size - orig_size);
1033 fallback_to_reg_send:
1034                         sk_msg_trim(sk, msg_pl, orig_size);
1035                 }
1036
1037                 required_size = msg_pl->sg.size + try_to_copy;
1038
1039                 ret = tls_clone_plaintext_msg(sk, required_size);
1040                 if (ret) {
1041                         if (ret != -ENOSPC)
1042                                 goto send_end;
1043
1044                         /* Adjust try_to_copy according to the amount that was
1045                          * actually allocated. The difference is due
1046                          * to max sg elements limit
1047                          */
1048                         try_to_copy -= required_size - msg_pl->sg.size;
1049                         full_record = true;
1050                         sk_msg_trim(sk, msg_en,
1051                                     msg_pl->sg.size + prot->overhead_size);
1052                 }
1053
1054                 if (try_to_copy) {
1055                         ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter,
1056                                                        msg_pl, try_to_copy);
1057                         if (ret < 0)
1058                                 goto trim_sgl;
1059                 }
1060
1061                 /* Open records defined only if successfully copied, otherwise
1062                  * we would trim the sg but not reset the open record frags.
1063                  */
1064                 tls_ctx->pending_open_record_frags = true;
1065                 copied += try_to_copy;
1066                 if (full_record || eor) {
1067                         ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1068                                                   record_type, &copied,
1069                                                   msg->msg_flags);
1070                         if (ret) {
1071                                 if (ret == -EINPROGRESS)
1072                                         num_async++;
1073                                 else if (ret == -ENOMEM)
1074                                         goto wait_for_memory;
1075                                 else if (ret != -EAGAIN) {
1076                                         if (ret == -ENOSPC)
1077                                                 ret = 0;
1078                                         goto send_end;
1079                                 }
1080                         }
1081                 }
1082
1083                 continue;
1084
1085 wait_for_sndbuf:
1086                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1087 wait_for_memory:
1088                 ret = sk_stream_wait_memory(sk, &timeo);
1089                 if (ret) {
1090 trim_sgl:
1091                         if (ctx->open_rec)
1092                                 tls_trim_both_msgs(sk, orig_size);
1093                         goto send_end;
1094                 }
1095
1096                 if (ctx->open_rec && msg_en->sg.size < required_size)
1097                         goto alloc_encrypted;
1098         }
1099
1100         if (!num_async) {
1101                 goto send_end;
1102         } else if (num_zc) {
1103                 /* Wait for pending encryptions to get completed */
1104                 spin_lock_bh(&ctx->encrypt_compl_lock);
1105                 ctx->async_notify = true;
1106
1107                 pending = atomic_read(&ctx->encrypt_pending);
1108                 spin_unlock_bh(&ctx->encrypt_compl_lock);
1109                 if (pending)
1110                         crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1111                 else
1112                         reinit_completion(&ctx->async_wait.completion);
1113
1114                 /* There can be no concurrent accesses, since we have no
1115                  * pending encrypt operations
1116                  */
1117                 WRITE_ONCE(ctx->async_notify, false);
1118
1119                 if (ctx->async_wait.err) {
1120                         ret = ctx->async_wait.err;
1121                         copied = 0;
1122                 }
1123         }
1124
1125         /* Transmit if any encryptions have completed */
1126         if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1127                 cancel_delayed_work(&ctx->tx_work.work);
1128                 tls_tx_records(sk, msg->msg_flags);
1129         }
1130
1131 send_end:
1132         ret = sk_stream_error(sk, msg->msg_flags, ret);
1133
1134         release_sock(sk);
1135         mutex_unlock(&tls_ctx->tx_lock);
1136         return copied > 0 ? copied : ret;
1137 }
1138
1139 static int tls_sw_do_sendpage(struct sock *sk, struct page *page,
1140                               int offset, size_t size, int flags)
1141 {
1142         long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1143         struct tls_context *tls_ctx = tls_get_ctx(sk);
1144         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1145         struct tls_prot_info *prot = &tls_ctx->prot_info;
1146         unsigned char record_type = TLS_RECORD_TYPE_DATA;
1147         struct sk_msg *msg_pl;
1148         struct tls_rec *rec;
1149         int num_async = 0;
1150         ssize_t copied = 0;
1151         bool full_record;
1152         int record_room;
1153         int ret = 0;
1154         bool eor;
1155
1156         eor = !(flags & MSG_SENDPAGE_NOTLAST);
1157         sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1158
1159         /* Call the sk_stream functions to manage the sndbuf mem. */
1160         while (size > 0) {
1161                 size_t copy, required_size;
1162
1163                 if (sk->sk_err) {
1164                         ret = -sk->sk_err;
1165                         goto sendpage_end;
1166                 }
1167
1168                 if (ctx->open_rec)
1169                         rec = ctx->open_rec;
1170                 else
1171                         rec = ctx->open_rec = tls_get_rec(sk);
1172                 if (!rec) {
1173                         ret = -ENOMEM;
1174                         goto sendpage_end;
1175                 }
1176
1177                 msg_pl = &rec->msg_plaintext;
1178
1179                 full_record = false;
1180                 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
1181                 copy = size;
1182                 if (copy >= record_room) {
1183                         copy = record_room;
1184                         full_record = true;
1185                 }
1186
1187                 required_size = msg_pl->sg.size + copy + prot->overhead_size;
1188
1189                 if (!sk_stream_memory_free(sk))
1190                         goto wait_for_sndbuf;
1191 alloc_payload:
1192                 ret = tls_alloc_encrypted_msg(sk, required_size);
1193                 if (ret) {
1194                         if (ret != -ENOSPC)
1195                                 goto wait_for_memory;
1196
1197                         /* Adjust copy according to the amount that was
1198                          * actually allocated. The difference is due
1199                          * to max sg elements limit
1200                          */
1201                         copy -= required_size - msg_pl->sg.size;
1202                         full_record = true;
1203                 }
1204
1205                 sk_msg_page_add(msg_pl, page, copy, offset);
1206                 sk_mem_charge(sk, copy);
1207
1208                 offset += copy;
1209                 size -= copy;
1210                 copied += copy;
1211
1212                 tls_ctx->pending_open_record_frags = true;
1213                 if (full_record || eor || sk_msg_full(msg_pl)) {
1214                         ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1215                                                   record_type, &copied, flags);
1216                         if (ret) {
1217                                 if (ret == -EINPROGRESS)
1218                                         num_async++;
1219                                 else if (ret == -ENOMEM)
1220                                         goto wait_for_memory;
1221                                 else if (ret != -EAGAIN) {
1222                                         if (ret == -ENOSPC)
1223                                                 ret = 0;
1224                                         goto sendpage_end;
1225                                 }
1226                         }
1227                 }
1228                 continue;
1229 wait_for_sndbuf:
1230                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1231 wait_for_memory:
1232                 ret = sk_stream_wait_memory(sk, &timeo);
1233                 if (ret) {
1234                         if (ctx->open_rec)
1235                                 tls_trim_both_msgs(sk, msg_pl->sg.size);
1236                         goto sendpage_end;
1237                 }
1238
1239                 if (ctx->open_rec)
1240                         goto alloc_payload;
1241         }
1242
1243         if (num_async) {
1244                 /* Transmit if any encryptions have completed */
1245                 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1246                         cancel_delayed_work(&ctx->tx_work.work);
1247                         tls_tx_records(sk, flags);
1248                 }
1249         }
1250 sendpage_end:
1251         ret = sk_stream_error(sk, flags, ret);
1252         return copied > 0 ? copied : ret;
1253 }
1254
1255 int tls_sw_sendpage_locked(struct sock *sk, struct page *page,
1256                            int offset, size_t size, int flags)
1257 {
1258         if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1259                       MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY |
1260                       MSG_NO_SHARED_FRAGS))
1261                 return -EOPNOTSUPP;
1262
1263         return tls_sw_do_sendpage(sk, page, offset, size, flags);
1264 }
1265
1266 int tls_sw_sendpage(struct sock *sk, struct page *page,
1267                     int offset, size_t size, int flags)
1268 {
1269         struct tls_context *tls_ctx = tls_get_ctx(sk);
1270         int ret;
1271
1272         if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1273                       MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY))
1274                 return -EOPNOTSUPP;
1275
1276         mutex_lock(&tls_ctx->tx_lock);
1277         lock_sock(sk);
1278         ret = tls_sw_do_sendpage(sk, page, offset, size, flags);
1279         release_sock(sk);
1280         mutex_unlock(&tls_ctx->tx_lock);
1281         return ret;
1282 }
1283
1284 static int
1285 tls_rx_rec_wait(struct sock *sk, struct sk_psock *psock, bool nonblock,
1286                 long timeo)
1287 {
1288         struct tls_context *tls_ctx = tls_get_ctx(sk);
1289         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1290         DEFINE_WAIT_FUNC(wait, woken_wake_function);
1291
1292         while (!tls_strp_msg_ready(ctx)) {
1293                 if (!sk_psock_queue_empty(psock))
1294                         return 0;
1295
1296                 if (sk->sk_err)
1297                         return sock_error(sk);
1298
1299                 if (!skb_queue_empty(&sk->sk_receive_queue)) {
1300                         __strp_unpause(&ctx->strp);
1301                         if (tls_strp_msg_ready(ctx))
1302                                 break;
1303                 }
1304
1305                 if (sk->sk_shutdown & RCV_SHUTDOWN)
1306                         return 0;
1307
1308                 if (sock_flag(sk, SOCK_DONE))
1309                         return 0;
1310
1311                 if (nonblock || !timeo)
1312                         return -EAGAIN;
1313
1314                 add_wait_queue(sk_sleep(sk), &wait);
1315                 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1316                 sk_wait_event(sk, &timeo,
1317                               tls_strp_msg_ready(ctx) ||
1318                               !sk_psock_queue_empty(psock),
1319                               &wait);
1320                 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1321                 remove_wait_queue(sk_sleep(sk), &wait);
1322
1323                 /* Handle signals */
1324                 if (signal_pending(current))
1325                         return sock_intr_errno(timeo);
1326         }
1327
1328         return 1;
1329 }
1330
1331 static int tls_setup_from_iter(struct iov_iter *from,
1332                                int length, int *pages_used,
1333                                struct scatterlist *to,
1334                                int to_max_pages)
1335 {
1336         int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1337         struct page *pages[MAX_SKB_FRAGS];
1338         unsigned int size = 0;
1339         ssize_t copied, use;
1340         size_t offset;
1341
1342         while (length > 0) {
1343                 i = 0;
1344                 maxpages = to_max_pages - num_elem;
1345                 if (maxpages == 0) {
1346                         rc = -EFAULT;
1347                         goto out;
1348                 }
1349                 copied = iov_iter_get_pages(from, pages,
1350                                             length,
1351                                             maxpages, &offset);
1352                 if (copied <= 0) {
1353                         rc = -EFAULT;
1354                         goto out;
1355                 }
1356
1357                 iov_iter_advance(from, copied);
1358
1359                 length -= copied;
1360                 size += copied;
1361                 while (copied) {
1362                         use = min_t(int, copied, PAGE_SIZE - offset);
1363
1364                         sg_set_page(&to[num_elem],
1365                                     pages[i], use, offset);
1366                         sg_unmark_end(&to[num_elem]);
1367                         /* We do not uncharge memory from this API */
1368
1369                         offset = 0;
1370                         copied -= use;
1371
1372                         i++;
1373                         num_elem++;
1374                 }
1375         }
1376         /* Mark the end in the last sg entry if newly added */
1377         if (num_elem > *pages_used)
1378                 sg_mark_end(&to[num_elem - 1]);
1379 out:
1380         if (rc)
1381                 iov_iter_revert(from, size);
1382         *pages_used = num_elem;
1383
1384         return rc;
1385 }
1386
1387 static struct sk_buff *
1388 tls_alloc_clrtxt_skb(struct sock *sk, struct sk_buff *skb,
1389                      unsigned int full_len)
1390 {
1391         struct strp_msg *clr_rxm;
1392         struct sk_buff *clr_skb;
1393         int err;
1394
1395         clr_skb = alloc_skb_with_frags(0, full_len, TLS_PAGE_ORDER,
1396                                        &err, sk->sk_allocation);
1397         if (!clr_skb)
1398                 return NULL;
1399
1400         skb_copy_header(clr_skb, skb);
1401         clr_skb->len = full_len;
1402         clr_skb->data_len = full_len;
1403
1404         clr_rxm = strp_msg(clr_skb);
1405         clr_rxm->offset = 0;
1406
1407         return clr_skb;
1408 }
1409
1410 /* Decrypt handlers
1411  *
1412  * tls_decrypt_sw() and tls_decrypt_device() are decrypt handlers.
1413  * They must transform the darg in/out argument are as follows:
1414  *       |          Input            |         Output
1415  * -------------------------------------------------------------------
1416  *    zc | Zero-copy decrypt allowed | Zero-copy performed
1417  * async | Async decrypt allowed     | Async crypto used / in progress
1418  *   skb |            *              | Output skb
1419  *
1420  * If ZC decryption was performed darg.skb will point to the input skb.
1421  */
1422
1423 /* This function decrypts the input skb into either out_iov or in out_sg
1424  * or in skb buffers itself. The input parameter 'darg->zc' indicates if
1425  * zero-copy mode needs to be tried or not. With zero-copy mode, either
1426  * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1427  * NULL, then the decryption happens inside skb buffers itself, i.e.
1428  * zero-copy gets disabled and 'darg->zc' is updated.
1429  */
1430 static int tls_decrypt_sg(struct sock *sk, struct iov_iter *out_iov,
1431                           struct scatterlist *out_sg,
1432                           struct tls_decrypt_arg *darg)
1433 {
1434         struct tls_context *tls_ctx = tls_get_ctx(sk);
1435         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1436         struct tls_prot_info *prot = &tls_ctx->prot_info;
1437         int n_sgin, n_sgout, aead_size, err, pages = 0;
1438         struct sk_buff *skb = tls_strp_msg(ctx);
1439         const struct strp_msg *rxm = strp_msg(skb);
1440         const struct tls_msg *tlm = tls_msg(skb);
1441         struct aead_request *aead_req;
1442         struct scatterlist *sgin = NULL;
1443         struct scatterlist *sgout = NULL;
1444         const int data_len = rxm->full_len - prot->overhead_size;
1445         int tail_pages = !!prot->tail_size;
1446         struct tls_decrypt_ctx *dctx;
1447         struct sk_buff *clear_skb;
1448         int iv_offset = 0;
1449         u8 *mem;
1450
1451         n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size,
1452                          rxm->full_len - prot->prepend_size);
1453         if (n_sgin < 1)
1454                 return n_sgin ?: -EBADMSG;
1455
1456         if (darg->zc && (out_iov || out_sg)) {
1457                 clear_skb = NULL;
1458
1459                 if (out_iov)
1460                         n_sgout = 1 + tail_pages +
1461                                 iov_iter_npages_cap(out_iov, INT_MAX, data_len);
1462                 else
1463                         n_sgout = sg_nents(out_sg);
1464         } else {
1465                 darg->zc = false;
1466
1467                 clear_skb = tls_alloc_clrtxt_skb(sk, skb, rxm->full_len);
1468                 if (!clear_skb)
1469                         return -ENOMEM;
1470
1471                 n_sgout = 1 + skb_shinfo(clear_skb)->nr_frags;
1472         }
1473
1474         /* Increment to accommodate AAD */
1475         n_sgin = n_sgin + 1;
1476
1477         /* Allocate a single block of memory which contains
1478          *   aead_req || tls_decrypt_ctx.
1479          * Both structs are variable length.
1480          */
1481         aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
1482         mem = kmalloc(aead_size + struct_size(dctx, sg, n_sgin + n_sgout),
1483                       sk->sk_allocation);
1484         if (!mem) {
1485                 err = -ENOMEM;
1486                 goto exit_free_skb;
1487         }
1488
1489         /* Segment the allocated memory */
1490         aead_req = (struct aead_request *)mem;
1491         dctx = (struct tls_decrypt_ctx *)(mem + aead_size);
1492         sgin = &dctx->sg[0];
1493         sgout = &dctx->sg[n_sgin];
1494
1495         /* For CCM based ciphers, first byte of nonce+iv is a constant */
1496         switch (prot->cipher_type) {
1497         case TLS_CIPHER_AES_CCM_128:
1498                 dctx->iv[0] = TLS_AES_CCM_IV_B0_BYTE;
1499                 iv_offset = 1;
1500                 break;
1501         case TLS_CIPHER_SM4_CCM:
1502                 dctx->iv[0] = TLS_SM4_CCM_IV_B0_BYTE;
1503                 iv_offset = 1;
1504                 break;
1505         }
1506
1507         /* Prepare IV */
1508         if (prot->version == TLS_1_3_VERSION ||
1509             prot->cipher_type == TLS_CIPHER_CHACHA20_POLY1305) {
1510                 memcpy(&dctx->iv[iv_offset], tls_ctx->rx.iv,
1511                        prot->iv_size + prot->salt_size);
1512         } else {
1513                 err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
1514                                     &dctx->iv[iv_offset] + prot->salt_size,
1515                                     prot->iv_size);
1516                 if (err < 0)
1517                         goto exit_free;
1518                 memcpy(&dctx->iv[iv_offset], tls_ctx->rx.iv, prot->salt_size);
1519         }
1520         tls_xor_iv_with_seq(prot, &dctx->iv[iv_offset], tls_ctx->rx.rec_seq);
1521
1522         /* Prepare AAD */
1523         tls_make_aad(dctx->aad, rxm->full_len - prot->overhead_size +
1524                      prot->tail_size,
1525                      tls_ctx->rx.rec_seq, tlm->control, prot);
1526
1527         /* Prepare sgin */
1528         sg_init_table(sgin, n_sgin);
1529         sg_set_buf(&sgin[0], dctx->aad, prot->aad_size);
1530         err = skb_to_sgvec(skb, &sgin[1],
1531                            rxm->offset + prot->prepend_size,
1532                            rxm->full_len - prot->prepend_size);
1533         if (err < 0)
1534                 goto exit_free;
1535
1536         if (clear_skb) {
1537                 sg_init_table(sgout, n_sgout);
1538                 sg_set_buf(&sgout[0], dctx->aad, prot->aad_size);
1539
1540                 err = skb_to_sgvec(clear_skb, &sgout[1], prot->prepend_size,
1541                                    data_len + prot->tail_size);
1542                 if (err < 0)
1543                         goto exit_free;
1544         } else if (out_iov) {
1545                 sg_init_table(sgout, n_sgout);
1546                 sg_set_buf(&sgout[0], dctx->aad, prot->aad_size);
1547
1548                 err = tls_setup_from_iter(out_iov, data_len, &pages, &sgout[1],
1549                                           (n_sgout - 1 - tail_pages));
1550                 if (err < 0)
1551                         goto exit_free_pages;
1552
1553                 if (prot->tail_size) {
1554                         sg_unmark_end(&sgout[pages]);
1555                         sg_set_buf(&sgout[pages + 1], &dctx->tail,
1556                                    prot->tail_size);
1557                         sg_mark_end(&sgout[pages + 1]);
1558                 }
1559         } else if (out_sg) {
1560                 memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
1561         }
1562
1563         /* Prepare and submit AEAD request */
1564         err = tls_do_decryption(sk, sgin, sgout, dctx->iv,
1565                                 data_len + prot->tail_size, aead_req, darg);
1566         if (err)
1567                 goto exit_free_pages;
1568
1569         darg->skb = clear_skb ?: tls_strp_msg(ctx);
1570         clear_skb = NULL;
1571
1572         if (unlikely(darg->async)) {
1573                 err = tls_strp_msg_hold(sk, skb, &ctx->async_hold);
1574                 if (err)
1575                         __skb_queue_tail(&ctx->async_hold, darg->skb);
1576                 return err;
1577         }
1578
1579         if (prot->tail_size)
1580                 darg->tail = dctx->tail;
1581
1582 exit_free_pages:
1583         /* Release the pages in case iov was mapped to pages */
1584         for (; pages > 0; pages--)
1585                 put_page(sg_page(&sgout[pages]));
1586 exit_free:
1587         kfree(mem);
1588 exit_free_skb:
1589         consume_skb(clear_skb);
1590         return err;
1591 }
1592
1593 static int
1594 tls_decrypt_sw(struct sock *sk, struct tls_context *tls_ctx,
1595                struct msghdr *msg, struct tls_decrypt_arg *darg)
1596 {
1597         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1598         struct tls_prot_info *prot = &tls_ctx->prot_info;
1599         struct strp_msg *rxm;
1600         int pad, err;
1601
1602         err = tls_decrypt_sg(sk, &msg->msg_iter, NULL, darg);
1603         if (err < 0) {
1604                 if (err == -EBADMSG)
1605                         TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTERROR);
1606                 return err;
1607         }
1608         /* keep going even for ->async, the code below is TLS 1.3 */
1609
1610         /* If opportunistic TLS 1.3 ZC failed retry without ZC */
1611         if (unlikely(darg->zc && prot->version == TLS_1_3_VERSION &&
1612                      darg->tail != TLS_RECORD_TYPE_DATA)) {
1613                 darg->zc = false;
1614                 if (!darg->tail)
1615                         TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXNOPADVIOL);
1616                 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTRETRY);
1617                 return tls_decrypt_sw(sk, tls_ctx, msg, darg);
1618         }
1619
1620         pad = tls_padding_length(prot, darg->skb, darg);
1621         if (pad < 0) {
1622                 if (darg->skb != tls_strp_msg(ctx))
1623                         consume_skb(darg->skb);
1624                 return pad;
1625         }
1626
1627         rxm = strp_msg(darg->skb);
1628         rxm->full_len -= pad;
1629
1630         return 0;
1631 }
1632
1633 static int
1634 tls_decrypt_device(struct sock *sk, struct msghdr *msg,
1635                    struct tls_context *tls_ctx, struct tls_decrypt_arg *darg)
1636 {
1637         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1638         struct tls_prot_info *prot = &tls_ctx->prot_info;
1639         struct strp_msg *rxm;
1640         int pad, err;
1641
1642         if (tls_ctx->rx_conf != TLS_HW)
1643                 return 0;
1644
1645         err = tls_device_decrypted(sk, tls_ctx);
1646         if (err <= 0)
1647                 return err;
1648
1649         pad = tls_padding_length(prot, tls_strp_msg(ctx), darg);
1650         if (pad < 0)
1651                 return pad;
1652
1653         darg->async = false;
1654         darg->skb = tls_strp_msg(ctx);
1655         /* ->zc downgrade check, in case TLS 1.3 gets here */
1656         darg->zc &= !(prot->version == TLS_1_3_VERSION &&
1657                       tls_msg(darg->skb)->control != TLS_RECORD_TYPE_DATA);
1658
1659         rxm = strp_msg(darg->skb);
1660         rxm->full_len -= pad;
1661
1662         if (!darg->zc) {
1663                 /* Non-ZC case needs a real skb */
1664                 darg->skb = tls_strp_msg_detach(ctx);
1665                 if (!darg->skb)
1666                         return -ENOMEM;
1667         } else {
1668                 unsigned int off, len;
1669
1670                 /* In ZC case nobody cares about the output skb.
1671                  * Just copy the data here. Note the skb is not fully trimmed.
1672                  */
1673                 off = rxm->offset + prot->prepend_size;
1674                 len = rxm->full_len - prot->overhead_size;
1675
1676                 err = skb_copy_datagram_msg(darg->skb, off, msg, len);
1677                 if (err)
1678                         return err;
1679         }
1680         return 1;
1681 }
1682
1683 static int tls_rx_one_record(struct sock *sk, struct msghdr *msg,
1684                              struct tls_decrypt_arg *darg)
1685 {
1686         struct tls_context *tls_ctx = tls_get_ctx(sk);
1687         struct tls_prot_info *prot = &tls_ctx->prot_info;
1688         struct strp_msg *rxm;
1689         int err;
1690
1691         err = tls_decrypt_device(sk, msg, tls_ctx, darg);
1692         if (!err)
1693                 err = tls_decrypt_sw(sk, tls_ctx, msg, darg);
1694         if (err < 0)
1695                 return err;
1696
1697         rxm = strp_msg(darg->skb);
1698         rxm->offset += prot->prepend_size;
1699         rxm->full_len -= prot->overhead_size;
1700         tls_advance_record_sn(sk, prot, &tls_ctx->rx);
1701
1702         return 0;
1703 }
1704
1705 int decrypt_skb(struct sock *sk, struct scatterlist *sgout)
1706 {
1707         struct tls_decrypt_arg darg = { .zc = true, };
1708
1709         return tls_decrypt_sg(sk, NULL, sgout, &darg);
1710 }
1711
1712 static int tls_record_content_type(struct msghdr *msg, struct tls_msg *tlm,
1713                                    u8 *control)
1714 {
1715         int err;
1716
1717         if (!*control) {
1718                 *control = tlm->control;
1719                 if (!*control)
1720                         return -EBADMSG;
1721
1722                 err = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1723                                sizeof(*control), control);
1724                 if (*control != TLS_RECORD_TYPE_DATA) {
1725                         if (err || msg->msg_flags & MSG_CTRUNC)
1726                                 return -EIO;
1727                 }
1728         } else if (*control != tlm->control) {
1729                 return 0;
1730         }
1731
1732         return 1;
1733 }
1734
1735 static void tls_rx_rec_done(struct tls_sw_context_rx *ctx)
1736 {
1737         consume_skb(ctx->recv_pkt);
1738         ctx->recv_pkt = NULL;
1739         __strp_unpause(&ctx->strp);
1740 }
1741
1742 /* This function traverses the rx_list in tls receive context to copies the
1743  * decrypted records into the buffer provided by caller zero copy is not
1744  * true. Further, the records are removed from the rx_list if it is not a peek
1745  * case and the record has been consumed completely.
1746  */
1747 static int process_rx_list(struct tls_sw_context_rx *ctx,
1748                            struct msghdr *msg,
1749                            u8 *control,
1750                            size_t skip,
1751                            size_t len,
1752                            bool is_peek)
1753 {
1754         struct sk_buff *skb = skb_peek(&ctx->rx_list);
1755         struct tls_msg *tlm;
1756         ssize_t copied = 0;
1757         int err;
1758
1759         while (skip && skb) {
1760                 struct strp_msg *rxm = strp_msg(skb);
1761                 tlm = tls_msg(skb);
1762
1763                 err = tls_record_content_type(msg, tlm, control);
1764                 if (err <= 0)
1765                         goto out;
1766
1767                 if (skip < rxm->full_len)
1768                         break;
1769
1770                 skip = skip - rxm->full_len;
1771                 skb = skb_peek_next(skb, &ctx->rx_list);
1772         }
1773
1774         while (len && skb) {
1775                 struct sk_buff *next_skb;
1776                 struct strp_msg *rxm = strp_msg(skb);
1777                 int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1778
1779                 tlm = tls_msg(skb);
1780
1781                 err = tls_record_content_type(msg, tlm, control);
1782                 if (err <= 0)
1783                         goto out;
1784
1785                 err = skb_copy_datagram_msg(skb, rxm->offset + skip,
1786                                             msg, chunk);
1787                 if (err < 0)
1788                         goto out;
1789
1790                 len = len - chunk;
1791                 copied = copied + chunk;
1792
1793                 /* Consume the data from record if it is non-peek case*/
1794                 if (!is_peek) {
1795                         rxm->offset = rxm->offset + chunk;
1796                         rxm->full_len = rxm->full_len - chunk;
1797
1798                         /* Return if there is unconsumed data in the record */
1799                         if (rxm->full_len - skip)
1800                                 break;
1801                 }
1802
1803                 /* The remaining skip-bytes must lie in 1st record in rx_list.
1804                  * So from the 2nd record, 'skip' should be 0.
1805                  */
1806                 skip = 0;
1807
1808                 if (msg)
1809                         msg->msg_flags |= MSG_EOR;
1810
1811                 next_skb = skb_peek_next(skb, &ctx->rx_list);
1812
1813                 if (!is_peek) {
1814                         __skb_unlink(skb, &ctx->rx_list);
1815                         consume_skb(skb);
1816                 }
1817
1818                 skb = next_skb;
1819         }
1820         err = 0;
1821
1822 out:
1823         return copied ? : err;
1824 }
1825
1826 static void
1827 tls_read_flush_backlog(struct sock *sk, struct tls_prot_info *prot,
1828                        size_t len_left, size_t decrypted, ssize_t done,
1829                        size_t *flushed_at)
1830 {
1831         size_t max_rec;
1832
1833         if (len_left <= decrypted)
1834                 return;
1835
1836         max_rec = prot->overhead_size - prot->tail_size + TLS_MAX_PAYLOAD_SIZE;
1837         if (done - *flushed_at < SZ_128K && tcp_inq(sk) > max_rec)
1838                 return;
1839
1840         *flushed_at = done;
1841         sk_flush_backlog(sk);
1842 }
1843
1844 static long tls_rx_reader_lock(struct sock *sk, struct tls_sw_context_rx *ctx,
1845                                bool nonblock)
1846 {
1847         long timeo;
1848         int err;
1849
1850         lock_sock(sk);
1851
1852         timeo = sock_rcvtimeo(sk, nonblock);
1853
1854         while (unlikely(ctx->reader_present)) {
1855                 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1856
1857                 ctx->reader_contended = 1;
1858
1859                 add_wait_queue(&ctx->wq, &wait);
1860                 sk_wait_event(sk, &timeo,
1861                               !READ_ONCE(ctx->reader_present), &wait);
1862                 remove_wait_queue(&ctx->wq, &wait);
1863
1864                 if (timeo <= 0) {
1865                         err = -EAGAIN;
1866                         goto err_unlock;
1867                 }
1868                 if (signal_pending(current)) {
1869                         err = sock_intr_errno(timeo);
1870                         goto err_unlock;
1871                 }
1872         }
1873
1874         WRITE_ONCE(ctx->reader_present, 1);
1875
1876         return timeo;
1877
1878 err_unlock:
1879         release_sock(sk);
1880         return err;
1881 }
1882
1883 static void tls_rx_reader_unlock(struct sock *sk, struct tls_sw_context_rx *ctx)
1884 {
1885         if (unlikely(ctx->reader_contended)) {
1886                 if (wq_has_sleeper(&ctx->wq))
1887                         wake_up(&ctx->wq);
1888                 else
1889                         ctx->reader_contended = 0;
1890
1891                 WARN_ON_ONCE(!ctx->reader_present);
1892         }
1893
1894         WRITE_ONCE(ctx->reader_present, 0);
1895         release_sock(sk);
1896 }
1897
1898 int tls_sw_recvmsg(struct sock *sk,
1899                    struct msghdr *msg,
1900                    size_t len,
1901                    int flags,
1902                    int *addr_len)
1903 {
1904         struct tls_context *tls_ctx = tls_get_ctx(sk);
1905         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1906         struct tls_prot_info *prot = &tls_ctx->prot_info;
1907         ssize_t decrypted = 0, async_copy_bytes = 0;
1908         struct sk_psock *psock;
1909         unsigned char control = 0;
1910         size_t flushed_at = 0;
1911         struct strp_msg *rxm;
1912         struct tls_msg *tlm;
1913         ssize_t copied = 0;
1914         bool async = false;
1915         int target, err = 0;
1916         long timeo;
1917         bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1918         bool is_peek = flags & MSG_PEEK;
1919         bool bpf_strp_enabled;
1920         bool zc_capable;
1921
1922         if (unlikely(flags & MSG_ERRQUEUE))
1923                 return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
1924
1925         psock = sk_psock_get(sk);
1926         timeo = tls_rx_reader_lock(sk, ctx, flags & MSG_DONTWAIT);
1927         if (timeo < 0)
1928                 return timeo;
1929         bpf_strp_enabled = sk_psock_strp_enabled(psock);
1930
1931         /* If crypto failed the connection is broken */
1932         err = ctx->async_wait.err;
1933         if (err)
1934                 goto end;
1935
1936         /* Process pending decrypted records. It must be non-zero-copy */
1937         err = process_rx_list(ctx, msg, &control, 0, len, is_peek);
1938         if (err < 0)
1939                 goto end;
1940
1941         copied = err;
1942         if (len <= copied)
1943                 goto end;
1944
1945         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1946         len = len - copied;
1947
1948         zc_capable = !bpf_strp_enabled && !is_kvec && !is_peek &&
1949                 ctx->zc_capable;
1950         decrypted = 0;
1951         while (len && (decrypted + copied < target || tls_strp_msg_ready(ctx))) {
1952                 struct tls_decrypt_arg darg;
1953                 int to_decrypt, chunk;
1954
1955                 err = tls_rx_rec_wait(sk, psock, flags & MSG_DONTWAIT, timeo);
1956                 if (err <= 0) {
1957                         if (psock) {
1958                                 chunk = sk_msg_recvmsg(sk, psock, msg, len,
1959                                                        flags);
1960                                 if (chunk > 0) {
1961                                         decrypted += chunk;
1962                                         len -= chunk;
1963                                         continue;
1964                                 }
1965                         }
1966                         goto recv_end;
1967                 }
1968
1969                 memset(&darg.inargs, 0, sizeof(darg.inargs));
1970
1971                 rxm = strp_msg(ctx->recv_pkt);
1972                 tlm = tls_msg(ctx->recv_pkt);
1973
1974                 to_decrypt = rxm->full_len - prot->overhead_size;
1975
1976                 if (zc_capable && to_decrypt <= len &&
1977                     tlm->control == TLS_RECORD_TYPE_DATA)
1978                         darg.zc = true;
1979
1980                 /* Do not use async mode if record is non-data */
1981                 if (tlm->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled)
1982                         darg.async = ctx->async_capable;
1983                 else
1984                         darg.async = false;
1985
1986                 err = tls_rx_one_record(sk, msg, &darg);
1987                 if (err < 0) {
1988                         tls_err_abort(sk, -EBADMSG);
1989                         goto recv_end;
1990                 }
1991
1992                 async |= darg.async;
1993
1994                 /* If the type of records being processed is not known yet,
1995                  * set it to record type just dequeued. If it is already known,
1996                  * but does not match the record type just dequeued, go to end.
1997                  * We always get record type here since for tls1.2, record type
1998                  * is known just after record is dequeued from stream parser.
1999                  * For tls1.3, we disable async.
2000                  */
2001                 err = tls_record_content_type(msg, tls_msg(darg.skb), &control);
2002                 if (err <= 0) {
2003                         DEBUG_NET_WARN_ON_ONCE(darg.zc);
2004                         tls_rx_rec_done(ctx);
2005 put_on_rx_list_err:
2006                         __skb_queue_tail(&ctx->rx_list, darg.skb);
2007                         goto recv_end;
2008                 }
2009
2010                 /* periodically flush backlog, and feed strparser */
2011                 tls_read_flush_backlog(sk, prot, len, to_decrypt,
2012                                        decrypted + copied, &flushed_at);
2013
2014                 /* TLS 1.3 may have updated the length by more than overhead */
2015                 rxm = strp_msg(darg.skb);
2016                 chunk = rxm->full_len;
2017                 tls_rx_rec_done(ctx);
2018
2019                 if (!darg.zc) {
2020                         bool partially_consumed = chunk > len;
2021                         struct sk_buff *skb = darg.skb;
2022
2023                         DEBUG_NET_WARN_ON_ONCE(darg.skb == ctx->recv_pkt);
2024
2025                         if (async) {
2026                                 /* TLS 1.2-only, to_decrypt must be text len */
2027                                 chunk = min_t(int, to_decrypt, len);
2028                                 async_copy_bytes += chunk;
2029 put_on_rx_list:
2030                                 decrypted += chunk;
2031                                 len -= chunk;
2032                                 __skb_queue_tail(&ctx->rx_list, skb);
2033                                 continue;
2034                         }
2035
2036                         if (bpf_strp_enabled) {
2037                                 err = sk_psock_tls_strp_read(psock, skb);
2038                                 if (err != __SK_PASS) {
2039                                         rxm->offset = rxm->offset + rxm->full_len;
2040                                         rxm->full_len = 0;
2041                                         if (err == __SK_DROP)
2042                                                 consume_skb(skb);
2043                                         continue;
2044                                 }
2045                         }
2046
2047                         if (partially_consumed)
2048                                 chunk = len;
2049
2050                         err = skb_copy_datagram_msg(skb, rxm->offset,
2051                                                     msg, chunk);
2052                         if (err < 0)
2053                                 goto put_on_rx_list_err;
2054
2055                         if (is_peek)
2056                                 goto put_on_rx_list;
2057
2058                         if (partially_consumed) {
2059                                 rxm->offset += chunk;
2060                                 rxm->full_len -= chunk;
2061                                 goto put_on_rx_list;
2062                         }
2063
2064                         consume_skb(skb);
2065                 }
2066
2067                 decrypted += chunk;
2068                 len -= chunk;
2069
2070                 /* Return full control message to userspace before trying
2071                  * to parse another message type
2072                  */
2073                 msg->msg_flags |= MSG_EOR;
2074                 if (control != TLS_RECORD_TYPE_DATA)
2075                         break;
2076         }
2077
2078 recv_end:
2079         if (async) {
2080                 int ret, pending;
2081
2082                 /* Wait for all previously submitted records to be decrypted */
2083                 spin_lock_bh(&ctx->decrypt_compl_lock);
2084                 reinit_completion(&ctx->async_wait.completion);
2085                 pending = atomic_read(&ctx->decrypt_pending);
2086                 spin_unlock_bh(&ctx->decrypt_compl_lock);
2087                 ret = 0;
2088                 if (pending)
2089                         ret = crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
2090                 __skb_queue_purge(&ctx->async_hold);
2091
2092                 if (ret) {
2093                         if (err >= 0 || err == -EINPROGRESS)
2094                                 err = ret;
2095                         decrypted = 0;
2096                         goto end;
2097                 }
2098
2099                 /* Drain records from the rx_list & copy if required */
2100                 if (is_peek || is_kvec)
2101                         err = process_rx_list(ctx, msg, &control, copied,
2102                                               decrypted, is_peek);
2103                 else
2104                         err = process_rx_list(ctx, msg, &control, 0,
2105                                               async_copy_bytes, is_peek);
2106                 decrypted = max(err, 0);
2107         }
2108
2109         copied += decrypted;
2110
2111 end:
2112         tls_rx_reader_unlock(sk, ctx);
2113         if (psock)
2114                 sk_psock_put(sk, psock);
2115         return copied ? : err;
2116 }
2117
2118 ssize_t tls_sw_splice_read(struct socket *sock,  loff_t *ppos,
2119                            struct pipe_inode_info *pipe,
2120                            size_t len, unsigned int flags)
2121 {
2122         struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
2123         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2124         struct strp_msg *rxm = NULL;
2125         struct sock *sk = sock->sk;
2126         struct tls_msg *tlm;
2127         struct sk_buff *skb;
2128         ssize_t copied = 0;
2129         int err = 0;
2130         long timeo;
2131         int chunk;
2132
2133         timeo = tls_rx_reader_lock(sk, ctx, flags & SPLICE_F_NONBLOCK);
2134         if (timeo < 0)
2135                 return timeo;
2136
2137         if (!skb_queue_empty(&ctx->rx_list)) {
2138                 skb = __skb_dequeue(&ctx->rx_list);
2139         } else {
2140                 struct tls_decrypt_arg darg;
2141
2142                 err = tls_rx_rec_wait(sk, NULL, flags & SPLICE_F_NONBLOCK,
2143                                       timeo);
2144                 if (err <= 0)
2145                         goto splice_read_end;
2146
2147                 memset(&darg.inargs, 0, sizeof(darg.inargs));
2148
2149                 err = tls_rx_one_record(sk, NULL, &darg);
2150                 if (err < 0) {
2151                         tls_err_abort(sk, -EBADMSG);
2152                         goto splice_read_end;
2153                 }
2154
2155                 tls_rx_rec_done(ctx);
2156                 skb = darg.skb;
2157         }
2158
2159         rxm = strp_msg(skb);
2160         tlm = tls_msg(skb);
2161
2162         /* splice does not support reading control messages */
2163         if (tlm->control != TLS_RECORD_TYPE_DATA) {
2164                 err = -EINVAL;
2165                 goto splice_requeue;
2166         }
2167
2168         chunk = min_t(unsigned int, rxm->full_len, len);
2169         copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
2170         if (copied < 0)
2171                 goto splice_requeue;
2172
2173         if (chunk < rxm->full_len) {
2174                 rxm->offset += len;
2175                 rxm->full_len -= len;
2176                 goto splice_requeue;
2177         }
2178
2179         consume_skb(skb);
2180
2181 splice_read_end:
2182         tls_rx_reader_unlock(sk, ctx);
2183         return copied ? : err;
2184
2185 splice_requeue:
2186         __skb_queue_head(&ctx->rx_list, skb);
2187         goto splice_read_end;
2188 }
2189
2190 bool tls_sw_sock_is_readable(struct sock *sk)
2191 {
2192         struct tls_context *tls_ctx = tls_get_ctx(sk);
2193         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2194         bool ingress_empty = true;
2195         struct sk_psock *psock;
2196
2197         rcu_read_lock();
2198         psock = sk_psock(sk);
2199         if (psock)
2200                 ingress_empty = list_empty(&psock->ingress_msg);
2201         rcu_read_unlock();
2202
2203         return !ingress_empty || tls_strp_msg_ready(ctx) ||
2204                 !skb_queue_empty(&ctx->rx_list);
2205 }
2206
2207 static int tls_read_size(struct strparser *strp, struct sk_buff *skb)
2208 {
2209         struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2210         struct tls_prot_info *prot = &tls_ctx->prot_info;
2211         char header[TLS_HEADER_SIZE + MAX_IV_SIZE];
2212         struct strp_msg *rxm = strp_msg(skb);
2213         struct tls_msg *tlm = tls_msg(skb);
2214         size_t cipher_overhead;
2215         size_t data_len = 0;
2216         int ret;
2217
2218         /* Verify that we have a full TLS header, or wait for more data */
2219         if (rxm->offset + prot->prepend_size > skb->len)
2220                 return 0;
2221
2222         /* Sanity-check size of on-stack buffer. */
2223         if (WARN_ON(prot->prepend_size > sizeof(header))) {
2224                 ret = -EINVAL;
2225                 goto read_failure;
2226         }
2227
2228         /* Linearize header to local buffer */
2229         ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size);
2230         if (ret < 0)
2231                 goto read_failure;
2232
2233         tlm->control = header[0];
2234
2235         data_len = ((header[4] & 0xFF) | (header[3] << 8));
2236
2237         cipher_overhead = prot->tag_size;
2238         if (prot->version != TLS_1_3_VERSION &&
2239             prot->cipher_type != TLS_CIPHER_CHACHA20_POLY1305)
2240                 cipher_overhead += prot->iv_size;
2241
2242         if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
2243             prot->tail_size) {
2244                 ret = -EMSGSIZE;
2245                 goto read_failure;
2246         }
2247         if (data_len < cipher_overhead) {
2248                 ret = -EBADMSG;
2249                 goto read_failure;
2250         }
2251
2252         /* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2253         if (header[1] != TLS_1_2_VERSION_MINOR ||
2254             header[2] != TLS_1_2_VERSION_MAJOR) {
2255                 ret = -EINVAL;
2256                 goto read_failure;
2257         }
2258
2259         tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE,
2260                                      TCP_SKB_CB(skb)->seq + rxm->offset);
2261         return data_len + TLS_HEADER_SIZE;
2262
2263 read_failure:
2264         tls_err_abort(strp->sk, ret);
2265
2266         return ret;
2267 }
2268
2269 static void tls_queue(struct strparser *strp, struct sk_buff *skb)
2270 {
2271         struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2272         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2273
2274         ctx->recv_pkt = skb;
2275         strp_pause(strp);
2276
2277         ctx->saved_data_ready(strp->sk);
2278 }
2279
2280 static void tls_data_ready(struct sock *sk)
2281 {
2282         struct tls_context *tls_ctx = tls_get_ctx(sk);
2283         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2284         struct sk_psock *psock;
2285
2286         strp_data_ready(&ctx->strp);
2287
2288         psock = sk_psock_get(sk);
2289         if (psock) {
2290                 if (!list_empty(&psock->ingress_msg))
2291                         ctx->saved_data_ready(sk);
2292                 sk_psock_put(sk, psock);
2293         }
2294 }
2295
2296 void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
2297 {
2298         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2299
2300         set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask);
2301         set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask);
2302         cancel_delayed_work_sync(&ctx->tx_work.work);
2303 }
2304
2305 void tls_sw_release_resources_tx(struct sock *sk)
2306 {
2307         struct tls_context *tls_ctx = tls_get_ctx(sk);
2308         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2309         struct tls_rec *rec, *tmp;
2310         int pending;
2311
2312         /* Wait for any pending async encryptions to complete */
2313         spin_lock_bh(&ctx->encrypt_compl_lock);
2314         ctx->async_notify = true;
2315         pending = atomic_read(&ctx->encrypt_pending);
2316         spin_unlock_bh(&ctx->encrypt_compl_lock);
2317
2318         if (pending)
2319                 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
2320
2321         tls_tx_records(sk, -1);
2322
2323         /* Free up un-sent records in tx_list. First, free
2324          * the partially sent record if any at head of tx_list.
2325          */
2326         if (tls_ctx->partially_sent_record) {
2327                 tls_free_partial_record(sk, tls_ctx);
2328                 rec = list_first_entry(&ctx->tx_list,
2329                                        struct tls_rec, list);
2330                 list_del(&rec->list);
2331                 sk_msg_free(sk, &rec->msg_plaintext);
2332                 kfree(rec);
2333         }
2334
2335         list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
2336                 list_del(&rec->list);
2337                 sk_msg_free(sk, &rec->msg_encrypted);
2338                 sk_msg_free(sk, &rec->msg_plaintext);
2339                 kfree(rec);
2340         }
2341
2342         crypto_free_aead(ctx->aead_send);
2343         tls_free_open_rec(sk);
2344 }
2345
2346 void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
2347 {
2348         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2349
2350         kfree(ctx);
2351 }
2352
2353 void tls_sw_release_resources_rx(struct sock *sk)
2354 {
2355         struct tls_context *tls_ctx = tls_get_ctx(sk);
2356         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2357
2358         kfree(tls_ctx->rx.rec_seq);
2359         kfree(tls_ctx->rx.iv);
2360
2361         if (ctx->aead_recv) {
2362                 kfree_skb(ctx->recv_pkt);
2363                 ctx->recv_pkt = NULL;
2364                 __skb_queue_purge(&ctx->rx_list);
2365                 crypto_free_aead(ctx->aead_recv);
2366                 strp_stop(&ctx->strp);
2367                 /* If tls_sw_strparser_arm() was not called (cleanup paths)
2368                  * we still want to strp_stop(), but sk->sk_data_ready was
2369                  * never swapped.
2370                  */
2371                 if (ctx->saved_data_ready) {
2372                         write_lock_bh(&sk->sk_callback_lock);
2373                         sk->sk_data_ready = ctx->saved_data_ready;
2374                         write_unlock_bh(&sk->sk_callback_lock);
2375                 }
2376         }
2377 }
2378
2379 void tls_sw_strparser_done(struct tls_context *tls_ctx)
2380 {
2381         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2382
2383         strp_done(&ctx->strp);
2384 }
2385
2386 void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
2387 {
2388         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2389
2390         kfree(ctx);
2391 }
2392
2393 void tls_sw_free_resources_rx(struct sock *sk)
2394 {
2395         struct tls_context *tls_ctx = tls_get_ctx(sk);
2396
2397         tls_sw_release_resources_rx(sk);
2398         tls_sw_free_ctx_rx(tls_ctx);
2399 }
2400
2401 /* The work handler to transmitt the encrypted records in tx_list */
2402 static void tx_work_handler(struct work_struct *work)
2403 {
2404         struct delayed_work *delayed_work = to_delayed_work(work);
2405         struct tx_work *tx_work = container_of(delayed_work,
2406                                                struct tx_work, work);
2407         struct sock *sk = tx_work->sk;
2408         struct tls_context *tls_ctx = tls_get_ctx(sk);
2409         struct tls_sw_context_tx *ctx;
2410
2411         if (unlikely(!tls_ctx))
2412                 return;
2413
2414         ctx = tls_sw_ctx_tx(tls_ctx);
2415         if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
2416                 return;
2417
2418         if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
2419                 return;
2420         mutex_lock(&tls_ctx->tx_lock);
2421         lock_sock(sk);
2422         tls_tx_records(sk, -1);
2423         release_sock(sk);
2424         mutex_unlock(&tls_ctx->tx_lock);
2425 }
2426
2427 static bool tls_is_tx_ready(struct tls_sw_context_tx *ctx)
2428 {
2429         struct tls_rec *rec;
2430
2431         rec = list_first_entry(&ctx->tx_list, struct tls_rec, list);
2432         if (!rec)
2433                 return false;
2434
2435         return READ_ONCE(rec->tx_ready);
2436 }
2437
2438 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
2439 {
2440         struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
2441
2442         /* Schedule the transmission if tx list is ready */
2443         if (tls_is_tx_ready(tx_ctx) &&
2444             !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
2445                 schedule_delayed_work(&tx_ctx->tx_work.work, 0);
2446 }
2447
2448 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
2449 {
2450         struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2451
2452         write_lock_bh(&sk->sk_callback_lock);
2453         rx_ctx->saved_data_ready = sk->sk_data_ready;
2454         sk->sk_data_ready = tls_data_ready;
2455         write_unlock_bh(&sk->sk_callback_lock);
2456
2457         strp_check_rcv(&rx_ctx->strp);
2458 }
2459
2460 void tls_update_rx_zc_capable(struct tls_context *tls_ctx)
2461 {
2462         struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2463
2464         rx_ctx->zc_capable = tls_ctx->rx_no_pad ||
2465                 tls_ctx->prot_info.version != TLS_1_3_VERSION;
2466 }
2467
2468 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
2469 {
2470         struct tls_context *tls_ctx = tls_get_ctx(sk);
2471         struct tls_prot_info *prot = &tls_ctx->prot_info;
2472         struct tls_crypto_info *crypto_info;
2473         struct tls_sw_context_tx *sw_ctx_tx = NULL;
2474         struct tls_sw_context_rx *sw_ctx_rx = NULL;
2475         struct cipher_context *cctx;
2476         struct crypto_aead **aead;
2477         struct strp_callbacks cb;
2478         u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
2479         struct crypto_tfm *tfm;
2480         char *iv, *rec_seq, *key, *salt, *cipher_name;
2481         size_t keysize;
2482         int rc = 0;
2483
2484         if (!ctx) {
2485                 rc = -EINVAL;
2486                 goto out;
2487         }
2488
2489         if (tx) {
2490                 if (!ctx->priv_ctx_tx) {
2491                         sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
2492                         if (!sw_ctx_tx) {
2493                                 rc = -ENOMEM;
2494                                 goto out;
2495                         }
2496                         ctx->priv_ctx_tx = sw_ctx_tx;
2497                 } else {
2498                         sw_ctx_tx =
2499                                 (struct tls_sw_context_tx *)ctx->priv_ctx_tx;
2500                 }
2501         } else {
2502                 if (!ctx->priv_ctx_rx) {
2503                         sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
2504                         if (!sw_ctx_rx) {
2505                                 rc = -ENOMEM;
2506                                 goto out;
2507                         }
2508                         ctx->priv_ctx_rx = sw_ctx_rx;
2509                 } else {
2510                         sw_ctx_rx =
2511                                 (struct tls_sw_context_rx *)ctx->priv_ctx_rx;
2512                 }
2513         }
2514
2515         if (tx) {
2516                 crypto_init_wait(&sw_ctx_tx->async_wait);
2517                 spin_lock_init(&sw_ctx_tx->encrypt_compl_lock);
2518                 crypto_info = &ctx->crypto_send.info;
2519                 cctx = &ctx->tx;
2520                 aead = &sw_ctx_tx->aead_send;
2521                 INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
2522                 INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2523                 sw_ctx_tx->tx_work.sk = sk;
2524         } else {
2525                 crypto_init_wait(&sw_ctx_rx->async_wait);
2526                 spin_lock_init(&sw_ctx_rx->decrypt_compl_lock);
2527                 init_waitqueue_head(&sw_ctx_rx->wq);
2528                 crypto_info = &ctx->crypto_recv.info;
2529                 cctx = &ctx->rx;
2530                 skb_queue_head_init(&sw_ctx_rx->rx_list);
2531                 skb_queue_head_init(&sw_ctx_rx->async_hold);
2532                 aead = &sw_ctx_rx->aead_recv;
2533         }
2534
2535         switch (crypto_info->cipher_type) {
2536         case TLS_CIPHER_AES_GCM_128: {
2537                 struct tls12_crypto_info_aes_gcm_128 *gcm_128_info;
2538
2539                 gcm_128_info = (void *)crypto_info;
2540                 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2541                 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
2542                 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2543                 iv = gcm_128_info->iv;
2544                 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
2545                 rec_seq = gcm_128_info->rec_seq;
2546                 keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE;
2547                 key = gcm_128_info->key;
2548                 salt = gcm_128_info->salt;
2549                 salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
2550                 cipher_name = "gcm(aes)";
2551                 break;
2552         }
2553         case TLS_CIPHER_AES_GCM_256: {
2554                 struct tls12_crypto_info_aes_gcm_256 *gcm_256_info;
2555
2556                 gcm_256_info = (void *)crypto_info;
2557                 nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2558                 tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE;
2559                 iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2560                 iv = gcm_256_info->iv;
2561                 rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE;
2562                 rec_seq = gcm_256_info->rec_seq;
2563                 keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE;
2564                 key = gcm_256_info->key;
2565                 salt = gcm_256_info->salt;
2566                 salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE;
2567                 cipher_name = "gcm(aes)";
2568                 break;
2569         }
2570         case TLS_CIPHER_AES_CCM_128: {
2571                 struct tls12_crypto_info_aes_ccm_128 *ccm_128_info;
2572
2573                 ccm_128_info = (void *)crypto_info;
2574                 nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2575                 tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE;
2576                 iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2577                 iv = ccm_128_info->iv;
2578                 rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE;
2579                 rec_seq = ccm_128_info->rec_seq;
2580                 keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE;
2581                 key = ccm_128_info->key;
2582                 salt = ccm_128_info->salt;
2583                 salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE;
2584                 cipher_name = "ccm(aes)";
2585                 break;
2586         }
2587         case TLS_CIPHER_CHACHA20_POLY1305: {
2588                 struct tls12_crypto_info_chacha20_poly1305 *chacha20_poly1305_info;
2589
2590                 chacha20_poly1305_info = (void *)crypto_info;
2591                 nonce_size = 0;
2592                 tag_size = TLS_CIPHER_CHACHA20_POLY1305_TAG_SIZE;
2593                 iv_size = TLS_CIPHER_CHACHA20_POLY1305_IV_SIZE;
2594                 iv = chacha20_poly1305_info->iv;
2595                 rec_seq_size = TLS_CIPHER_CHACHA20_POLY1305_REC_SEQ_SIZE;
2596                 rec_seq = chacha20_poly1305_info->rec_seq;
2597                 keysize = TLS_CIPHER_CHACHA20_POLY1305_KEY_SIZE;
2598                 key = chacha20_poly1305_info->key;
2599                 salt = chacha20_poly1305_info->salt;
2600                 salt_size = TLS_CIPHER_CHACHA20_POLY1305_SALT_SIZE;
2601                 cipher_name = "rfc7539(chacha20,poly1305)";
2602                 break;
2603         }
2604         case TLS_CIPHER_SM4_GCM: {
2605                 struct tls12_crypto_info_sm4_gcm *sm4_gcm_info;
2606
2607                 sm4_gcm_info = (void *)crypto_info;
2608                 nonce_size = TLS_CIPHER_SM4_GCM_IV_SIZE;
2609                 tag_size = TLS_CIPHER_SM4_GCM_TAG_SIZE;
2610                 iv_size = TLS_CIPHER_SM4_GCM_IV_SIZE;
2611                 iv = sm4_gcm_info->iv;
2612                 rec_seq_size = TLS_CIPHER_SM4_GCM_REC_SEQ_SIZE;
2613                 rec_seq = sm4_gcm_info->rec_seq;
2614                 keysize = TLS_CIPHER_SM4_GCM_KEY_SIZE;
2615                 key = sm4_gcm_info->key;
2616                 salt = sm4_gcm_info->salt;
2617                 salt_size = TLS_CIPHER_SM4_GCM_SALT_SIZE;
2618                 cipher_name = "gcm(sm4)";
2619                 break;
2620         }
2621         case TLS_CIPHER_SM4_CCM: {
2622                 struct tls12_crypto_info_sm4_ccm *sm4_ccm_info;
2623
2624                 sm4_ccm_info = (void *)crypto_info;
2625                 nonce_size = TLS_CIPHER_SM4_CCM_IV_SIZE;
2626                 tag_size = TLS_CIPHER_SM4_CCM_TAG_SIZE;
2627                 iv_size = TLS_CIPHER_SM4_CCM_IV_SIZE;
2628                 iv = sm4_ccm_info->iv;
2629                 rec_seq_size = TLS_CIPHER_SM4_CCM_REC_SEQ_SIZE;
2630                 rec_seq = sm4_ccm_info->rec_seq;
2631                 keysize = TLS_CIPHER_SM4_CCM_KEY_SIZE;
2632                 key = sm4_ccm_info->key;
2633                 salt = sm4_ccm_info->salt;
2634                 salt_size = TLS_CIPHER_SM4_CCM_SALT_SIZE;
2635                 cipher_name = "ccm(sm4)";
2636                 break;
2637         }
2638         default:
2639                 rc = -EINVAL;
2640                 goto free_priv;
2641         }
2642
2643         if (crypto_info->version == TLS_1_3_VERSION) {
2644                 nonce_size = 0;
2645                 prot->aad_size = TLS_HEADER_SIZE;
2646                 prot->tail_size = 1;
2647         } else {
2648                 prot->aad_size = TLS_AAD_SPACE_SIZE;
2649                 prot->tail_size = 0;
2650         }
2651
2652         /* Sanity-check the sizes for stack allocations. */
2653         if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE ||
2654             rec_seq_size > TLS_MAX_REC_SEQ_SIZE || tag_size != TLS_TAG_SIZE ||
2655             prot->aad_size > TLS_MAX_AAD_SIZE) {
2656                 rc = -EINVAL;
2657                 goto free_priv;
2658         }
2659
2660         prot->version = crypto_info->version;
2661         prot->cipher_type = crypto_info->cipher_type;
2662         prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
2663         prot->tag_size = tag_size;
2664         prot->overhead_size = prot->prepend_size +
2665                               prot->tag_size + prot->tail_size;
2666         prot->iv_size = iv_size;
2667         prot->salt_size = salt_size;
2668         cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL);
2669         if (!cctx->iv) {
2670                 rc = -ENOMEM;
2671                 goto free_priv;
2672         }
2673         /* Note: 128 & 256 bit salt are the same size */
2674         prot->rec_seq_size = rec_seq_size;
2675         memcpy(cctx->iv, salt, salt_size);
2676         memcpy(cctx->iv + salt_size, iv, iv_size);
2677         cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
2678         if (!cctx->rec_seq) {
2679                 rc = -ENOMEM;
2680                 goto free_iv;
2681         }
2682
2683         if (!*aead) {
2684                 *aead = crypto_alloc_aead(cipher_name, 0, 0);
2685                 if (IS_ERR(*aead)) {
2686                         rc = PTR_ERR(*aead);
2687                         *aead = NULL;
2688                         goto free_rec_seq;
2689                 }
2690         }
2691
2692         ctx->push_pending_record = tls_sw_push_pending_record;
2693
2694         rc = crypto_aead_setkey(*aead, key, keysize);
2695
2696         if (rc)
2697                 goto free_aead;
2698
2699         rc = crypto_aead_setauthsize(*aead, prot->tag_size);
2700         if (rc)
2701                 goto free_aead;
2702
2703         if (sw_ctx_rx) {
2704                 tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
2705
2706                 tls_update_rx_zc_capable(ctx);
2707                 sw_ctx_rx->async_capable =
2708                         crypto_info->version != TLS_1_3_VERSION &&
2709                         !!(tfm->__crt_alg->cra_flags & CRYPTO_ALG_ASYNC);
2710
2711                 /* Set up strparser */
2712                 memset(&cb, 0, sizeof(cb));
2713                 cb.rcv_msg = tls_queue;
2714                 cb.parse_msg = tls_read_size;
2715
2716                 strp_init(&sw_ctx_rx->strp, sk, &cb);
2717         }
2718
2719         goto out;
2720
2721 free_aead:
2722         crypto_free_aead(*aead);
2723         *aead = NULL;
2724 free_rec_seq:
2725         kfree(cctx->rec_seq);
2726         cctx->rec_seq = NULL;
2727 free_iv:
2728         kfree(cctx->iv);
2729         cctx->iv = NULL;
2730 free_priv:
2731         if (tx) {
2732                 kfree(ctx->priv_ctx_tx);
2733                 ctx->priv_ctx_tx = NULL;
2734         } else {
2735                 kfree(ctx->priv_ctx_rx);
2736                 ctx->priv_ctx_rx = NULL;
2737         }
2738 out:
2739         return rc;
2740 }