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 noinline void tls_err_abort(struct sock *sk, int err)
  48 {
  49         WARN_ON_ONCE(err >= 0);
  50         /* sk->sk_err should contain a positive error code. */
  51         sk->sk_err = -err;
  52         sk_error_report(sk);
  53 }
  54
  55 static int __skb_nsg(struct sk_buff *skb, int offset, int len,
  56                      unsigned int recursion_level)
  57 {
  58         int start = skb_headlen(skb);
  59         int i, chunk = start - offset;
  60         struct sk_buff *frag_iter;
  61         int elt = 0;
  62
  63         if (unlikely(recursion_level >= 24))
  64                 return -EMSGSIZE;
  65
  66         if (chunk > 0) {
  67                 if (chunk > len)
  68                         chunk = len;
  69                 elt++;
  70                 len -= chunk;
  71                 if (len == 0)
  72                         return elt;
  73                 offset += chunk;
  74         }
  75
  76         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
  77                 int end;
  78
  79                 WARN_ON(start > offset + len);
  80
  81                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
  82                 chunk = end - offset;
  83                 if (chunk > 0) {
  84                         if (chunk > len)
  85                                 chunk = len;
  86                         elt++;
  87                         len -= chunk;
  88                         if (len == 0)
  89                                 return elt;
  90                         offset += chunk;
  91                 }
  92                 start = end;
  93         }
  94
  95         if (unlikely(skb_has_frag_list(skb))) {
  96                 skb_walk_frags(skb, frag_iter) {
  97                         int end, ret;
  98
  99                         WARN_ON(start > offset + len);
 100
 101                         end = start + frag_iter->len;
 102                         chunk = end - offset;
 103                         if (chunk > 0) {
 104                                 if (chunk > len)
 105                                         chunk = len;
 106                                 ret = __skb_nsg(frag_iter, offset - start, chunk,
 107                                                 recursion_level + 1);
 108                                 if (unlikely(ret < 0))
 109                                         return ret;
 110                                 elt += ret;
 111                                 len -= chunk;
 112                                 if (len == 0)
 113                                         return elt;
 114                                 offset += chunk;
 115                         }
 116                         start = end;
 117                 }
 118         }
 119         BUG_ON(len);
 120         return elt;
 121 }
 122
 123 /* Return the number of scatterlist elements required to completely map the
 124  * skb, or -EMSGSIZE if the recursion depth is exceeded.
 125  */
 126 static int skb_nsg(struct sk_buff *skb, int offset, int len)
 127 {
 128         return __skb_nsg(skb, offset, len, 0);
 129 }
 130
 131 static int padding_length(struct tls_sw_context_rx *ctx,
 132                           struct tls_prot_info *prot, struct sk_buff *skb)
 133 {
 134         struct strp_msg *rxm = strp_msg(skb);
 135         int sub = 0;
 136
 137         /* Determine zero-padding length */
 138         if (prot->version == TLS_1_3_VERSION) {
 139                 char content_type = 0;
 140                 int err;
 141                 int back = 17;
 142
 143                 while (content_type == 0) {
 144                         if (back > rxm->full_len - prot->prepend_size)
 145                                 return -EBADMSG;
 146                         err = skb_copy_bits(skb,
 147                                             rxm->offset + rxm->full_len - back,
 148                                             &content_type, 1);
 149                         if (err)
 150                                 return err;
 151                         if (content_type)
 152                                 break;
 153                         sub++;
 154                         back++;
 155                 }
 156                 ctx->control = content_type;
 157         }
 158         return sub;
 159 }
 160
 161 static void tls_decrypt_done(struct crypto_async_request *req, int err)
 162 {
 163         struct aead_request *aead_req = (struct aead_request *)req;
 164         struct scatterlist *sgout = aead_req->dst;
 165         struct scatterlist *sgin = aead_req->src;
 166         struct tls_sw_context_rx *ctx;
 167         struct tls_context *tls_ctx;
 168         struct tls_prot_info *prot;
 169         struct scatterlist *sg;
 170         struct sk_buff *skb;
 171         unsigned int pages;
 172         int pending;
 173
 174         skb = (struct sk_buff *)req->data;
 175         tls_ctx = tls_get_ctx(skb->sk);
 176         ctx = tls_sw_ctx_rx(tls_ctx);
 177         prot = &tls_ctx->prot_info;
 178
 179         /* Propagate if there was an err */
 180         if (err) {
 181                 if (err == -EBADMSG)
 182                         TLS_INC_STATS(sock_net(skb->sk),
 183                                       LINUX_MIB_TLSDECRYPTERROR);
 184                 ctx->async_wait.err = err;
 185                 tls_err_abort(skb->sk, err);
 186         } else {
 187                 struct strp_msg *rxm = strp_msg(skb);
 188                 int pad;
 189
 190                 pad = padding_length(ctx, prot, skb);
 191                 if (pad < 0) {
 192                         ctx->async_wait.err = pad;
 193                         tls_err_abort(skb->sk, pad);
 194                 } else {
 195                         rxm->full_len -= pad;
 196                         rxm->offset += prot->prepend_size;
 197                         rxm->full_len -= prot->overhead_size;
 198                 }
 199         }
 200
 201         /* After using skb->sk to propagate sk through crypto async callback
 202          * we need to NULL it again.
 203          */
 204         skb->sk = NULL;
 205
 206
 207         /* Free the destination pages if skb was not decrypted inplace */
 208         if (sgout != sgin) {
 209                 /* Skip the first S/G entry as it points to AAD */
 210                 for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) {
 211                         if (!sg)
 212                                 break;
 213                         put_page(sg_page(sg));
 214                 }
 215         }
 216
 217         kfree(aead_req);
 218
 219         spin_lock_bh(&ctx->decrypt_compl_lock);
 220         pending = atomic_dec_return(&ctx->decrypt_pending);
 221
 222         if (!pending && ctx->async_notify)
 223                 complete(&ctx->async_wait.completion);
 224         spin_unlock_bh(&ctx->decrypt_compl_lock);
 225 }
 226
 227 static int tls_do_decryption(struct sock *sk,
 228                              struct sk_buff *skb,
 229                              struct scatterlist *sgin,
 230                              struct scatterlist *sgout,
 231                              char *iv_recv,
 232                              size_t data_len,
 233                              struct aead_request *aead_req,
 234                              bool async)
 235 {
 236         struct tls_context *tls_ctx = tls_get_ctx(sk);
 237         struct tls_prot_info *prot = &tls_ctx->prot_info;
 238         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
 239         int ret;
 240
 241         aead_request_set_tfm(aead_req, ctx->aead_recv);
 242         aead_request_set_ad(aead_req, prot->aad_size);
 243         aead_request_set_crypt(aead_req, sgin, sgout,
 244                                data_len + prot->tag_size,
 245                                (u8 *)iv_recv);
 246
 247         if (async) {
 248                 /* Using skb->sk to push sk through to crypto async callback
 249                  * handler. This allows propagating errors up to the socket
 250                  * if needed. It _must_ be cleared in the async handler
 251                  * before consume_skb is called. We _know_ skb->sk is NULL
 252                  * because it is a clone from strparser.
 253                  */
 254                 skb->sk = sk;
 255                 aead_request_set_callback(aead_req,
 256                                           CRYPTO_TFM_REQ_MAY_BACKLOG,
 257                                           tls_decrypt_done, skb);
 258                 atomic_inc(&ctx->decrypt_pending);
 259         } else {
 260                 aead_request_set_callback(aead_req,
 261                                           CRYPTO_TFM_REQ_MAY_BACKLOG,
 262                                           crypto_req_done, &ctx->async_wait);
 263         }
 264
 265         ret = crypto_aead_decrypt(aead_req);
 266         if (ret == -EINPROGRESS) {
 267                 if (async)
 268                         return ret;
 269
 270                 ret = crypto_wait_req(ret, &ctx->async_wait);
 271         }
 272
 273         if (async)
 274                 atomic_dec(&ctx->decrypt_pending);
 275
 276         return ret;
 277 }
 278
 279 static void tls_trim_both_msgs(struct sock *sk, int target_size)
 280 {
 281         struct tls_context *tls_ctx = tls_get_ctx(sk);
 282         struct tls_prot_info *prot = &tls_ctx->prot_info;
 283         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 284         struct tls_rec *rec = ctx->open_rec;
 285
 286         sk_msg_trim(sk, &rec->msg_plaintext, target_size);
 287         if (target_size > 0)
 288                 target_size += prot->overhead_size;
 289         sk_msg_trim(sk, &rec->msg_encrypted, target_size);
 290 }
 291
 292 static int tls_alloc_encrypted_msg(struct sock *sk, int len)
 293 {
 294         struct tls_context *tls_ctx = tls_get_ctx(sk);
 295         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 296         struct tls_rec *rec = ctx->open_rec;
 297         struct sk_msg *msg_en = &rec->msg_encrypted;
 298
 299         return sk_msg_alloc(sk, msg_en, len, 0);
 300 }
 301
 302 static int tls_clone_plaintext_msg(struct sock *sk, int required)
 303 {
 304         struct tls_context *tls_ctx = tls_get_ctx(sk);
 305         struct tls_prot_info *prot = &tls_ctx->prot_info;
 306         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 307         struct tls_rec *rec = ctx->open_rec;
 308         struct sk_msg *msg_pl = &rec->msg_plaintext;
 309         struct sk_msg *msg_en = &rec->msg_encrypted;
 310         int skip, len;
 311
 312         /* We add page references worth len bytes from encrypted sg
 313          * at the end of plaintext sg. It is guaranteed that msg_en
 314          * has enough required room (ensured by caller).
 315          */
 316         len = required - msg_pl->sg.size;
 317
 318         /* Skip initial bytes in msg_en's data to be able to use
 319          * same offset of both plain and encrypted data.
 320          */
 321         skip = prot->prepend_size + msg_pl->sg.size;
 322
 323         return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
 324 }
 325
 326 static struct tls_rec *tls_get_rec(struct sock *sk)
 327 {
 328         struct tls_context *tls_ctx = tls_get_ctx(sk);
 329         struct tls_prot_info *prot = &tls_ctx->prot_info;
 330         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 331         struct sk_msg *msg_pl, *msg_en;
 332         struct tls_rec *rec;
 333         int mem_size;
 334
 335         mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
 336
 337         rec = kzalloc(mem_size, sk->sk_allocation);
 338         if (!rec)
 339                 return NULL;
 340
 341         msg_pl = &rec->msg_plaintext;
 342         msg_en = &rec->msg_encrypted;
 343
 344         sk_msg_init(msg_pl);
 345         sk_msg_init(msg_en);
 346
 347         sg_init_table(rec->sg_aead_in, 2);
 348         sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size);
 349         sg_unmark_end(&rec->sg_aead_in[1]);
 350
 351         sg_init_table(rec->sg_aead_out, 2);
 352         sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size);
 353         sg_unmark_end(&rec->sg_aead_out[1]);
 354
 355         return rec;
 356 }
 357
 358 static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
 359 {
 360         sk_msg_free(sk, &rec->msg_encrypted);
 361         sk_msg_free(sk, &rec->msg_plaintext);
 362         kfree(rec);
 363 }
 364
 365 static void tls_free_open_rec(struct sock *sk)
 366 {
 367         struct tls_context *tls_ctx = tls_get_ctx(sk);
 368         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 369         struct tls_rec *rec = ctx->open_rec;
 370
 371         if (rec) {
 372                 tls_free_rec(sk, rec);
 373                 ctx->open_rec = NULL;
 374         }
 375 }
 376
 377 int tls_tx_records(struct sock *sk, int flags)
 378 {
 379         struct tls_context *tls_ctx = tls_get_ctx(sk);
 380         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 381         struct tls_rec *rec, *tmp;
 382         struct sk_msg *msg_en;
 383         int tx_flags, rc = 0;
 384
 385         if (tls_is_partially_sent_record(tls_ctx)) {
 386                 rec = list_first_entry(&ctx->tx_list,
 387                                        struct tls_rec, list);
 388
 389                 if (flags == -1)
 390                         tx_flags = rec->tx_flags;
 391                 else
 392                         tx_flags = flags;
 393
 394                 rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
 395                 if (rc)
 396                         goto tx_err;
 397
 398                 /* Full record has been transmitted.
 399                  * Remove the head of tx_list
 400                  */
 401                 list_del(&rec->list);
 402                 sk_msg_free(sk, &rec->msg_plaintext);
 403                 kfree(rec);
 404         }
 405
 406         /* Tx all ready records */
 407         list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
 408                 if (READ_ONCE(rec->tx_ready)) {
 409                         if (flags == -1)
 410                                 tx_flags = rec->tx_flags;
 411                         else
 412                                 tx_flags = flags;
 413
 414                         msg_en = &rec->msg_encrypted;
 415                         rc = tls_push_sg(sk, tls_ctx,
 416                                          &msg_en->sg.data[msg_en->sg.curr],
 417                                          0, tx_flags);
 418                         if (rc)
 419                                 goto tx_err;
 420
 421                         list_del(&rec->list);
 422                         sk_msg_free(sk, &rec->msg_plaintext);
 423                         kfree(rec);
 424                 } else {
 425                         break;
 426                 }
 427         }
 428
 429 tx_err:
 430         if (rc < 0 && rc != -EAGAIN)
 431                 tls_err_abort(sk, -EBADMSG);
 432
 433         return rc;
 434 }
 435
 436 static void tls_encrypt_done(struct crypto_async_request *req, int err)
 437 {
 438         struct aead_request *aead_req = (struct aead_request *)req;
 439         struct sock *sk = req->data;
 440         struct tls_context *tls_ctx = tls_get_ctx(sk);
 441         struct tls_prot_info *prot = &tls_ctx->prot_info;
 442         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 443         struct scatterlist *sge;
 444         struct sk_msg *msg_en;
 445         struct tls_rec *rec;
 446         bool ready = false;
 447         int pending;
 448
 449         rec = container_of(aead_req, struct tls_rec, aead_req);
 450         msg_en = &rec->msg_encrypted;
 451
 452         sge = sk_msg_elem(msg_en, msg_en->sg.curr);
 453         sge->offset -= prot->prepend_size;
 454         sge->length += prot->prepend_size;
 455
 456         /* Check if error is previously set on socket */
 457         if (err || sk->sk_err) {
 458                 rec = NULL;
 459
 460                 /* If err is already set on socket, return the same code */
 461                 if (sk->sk_err) {
 462                         ctx->async_wait.err = -sk->sk_err;
 463                 } else {
 464                         ctx->async_wait.err = err;
 465                         tls_err_abort(sk, err);
 466                 }
 467         }
 468
 469         if (rec) {
 470                 struct tls_rec *first_rec;
 471
 472                 /* Mark the record as ready for transmission */
 473                 smp_store_mb(rec->tx_ready, true);
 474
 475                 /* If received record is at head of tx_list, schedule tx */
 476                 first_rec = list_first_entry(&ctx->tx_list,
 477                                              struct tls_rec, list);
 478                 if (rec == first_rec)
 479                         ready = true;
 480         }
 481
 482         spin_lock_bh(&ctx->encrypt_compl_lock);
 483         pending = atomic_dec_return(&ctx->encrypt_pending);
 484
 485         if (!pending && ctx->async_notify)
 486                 complete(&ctx->async_wait.completion);
 487         spin_unlock_bh(&ctx->encrypt_compl_lock);
 488
 489         if (!ready)
 490                 return;
 491
 492         /* Schedule the transmission */
 493         if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
 494                 schedule_delayed_work(&ctx->tx_work.work, 1);
 495 }
 496
 497 static int tls_do_encryption(struct sock *sk,
 498                              struct tls_context *tls_ctx,
 499                              struct tls_sw_context_tx *ctx,
 500                              struct aead_request *aead_req,
 501                              size_t data_len, u32 start)
 502 {
 503         struct tls_prot_info *prot = &tls_ctx->prot_info;
 504         struct tls_rec *rec = ctx->open_rec;
 505         struct sk_msg *msg_en = &rec->msg_encrypted;
 506         struct scatterlist *sge = sk_msg_elem(msg_en, start);
 507         int rc, iv_offset = 0;
 508
 509         /* For CCM based ciphers, first byte of IV is a constant */
 510         switch (prot->cipher_type) {
 511         case TLS_CIPHER_AES_CCM_128:
 512                 rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
 513                 iv_offset = 1;
 514                 break;
 515         case TLS_CIPHER_SM4_CCM:
 516                 rec->iv_data[0] = TLS_SM4_CCM_IV_B0_BYTE;
 517                 iv_offset = 1;
 518                 break;
 519         }
 520
 521         memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
 522                prot->iv_size + prot->salt_size);
 523
 524         xor_iv_with_seq(prot, rec->iv_data + iv_offset, tls_ctx->tx.rec_seq);
 525
 526         sge->offset += prot->prepend_size;
 527         sge->length -= prot->prepend_size;
 528
 529         msg_en->sg.curr = start;
 530
 531         aead_request_set_tfm(aead_req, ctx->aead_send);
 532         aead_request_set_ad(aead_req, prot->aad_size);
 533         aead_request_set_crypt(aead_req, rec->sg_aead_in,
 534                                rec->sg_aead_out,
 535                                data_len, rec->iv_data);
 536
 537         aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 538                                   tls_encrypt_done, sk);
 539
 540         /* Add the record in tx_list */
 541         list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
 542         atomic_inc(&ctx->encrypt_pending);
 543
 544         rc = crypto_aead_encrypt(aead_req);
 545         if (!rc || rc != -EINPROGRESS) {
 546                 atomic_dec(&ctx->encrypt_pending);
 547                 sge->offset -= prot->prepend_size;
 548                 sge->length += prot->prepend_size;
 549         }
 550
 551         if (!rc) {
 552                 WRITE_ONCE(rec->tx_ready, true);
 553         } else if (rc != -EINPROGRESS) {
 554                 list_del(&rec->list);
 555                 return rc;
 556         }
 557
 558         /* Unhook the record from context if encryption is not failure */
 559         ctx->open_rec = NULL;
 560         tls_advance_record_sn(sk, prot, &tls_ctx->tx);
 561         return rc;
 562 }
 563
 564 static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
 565                                  struct tls_rec **to, struct sk_msg *msg_opl,
 566                                  struct sk_msg *msg_oen, u32 split_point,
 567                                  u32 tx_overhead_size, u32 *orig_end)
 568 {
 569         u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
 570         struct scatterlist *sge, *osge, *nsge;
 571         u32 orig_size = msg_opl->sg.size;
 572         struct scatterlist tmp = { };
 573         struct sk_msg *msg_npl;
 574         struct tls_rec *new;
 575         int ret;
 576
 577         new = tls_get_rec(sk);
 578         if (!new)
 579                 return -ENOMEM;
 580         ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size +
 581                            tx_overhead_size, 0);
 582         if (ret < 0) {
 583                 tls_free_rec(sk, new);
 584                 return ret;
 585         }
 586
 587         *orig_end = msg_opl->sg.end;
 588         i = msg_opl->sg.start;
 589         sge = sk_msg_elem(msg_opl, i);
 590         while (apply && sge->length) {
 591                 if (sge->length > apply) {
 592                         u32 len = sge->length - apply;
 593
 594                         get_page(sg_page(sge));
 595                         sg_set_page(&tmp, sg_page(sge), len,
 596                                     sge->offset + apply);
 597                         sge->length = apply;
 598                         bytes += apply;
 599                         apply = 0;
 600                 } else {
 601                         apply -= sge->length;
 602                         bytes += sge->length;
 603                 }
 604
 605                 sk_msg_iter_var_next(i);
 606                 if (i == msg_opl->sg.end)
 607                         break;
 608                 sge = sk_msg_elem(msg_opl, i);
 609         }
 610
 611         msg_opl->sg.end = i;
 612         msg_opl->sg.curr = i;
 613         msg_opl->sg.copybreak = 0;
 614         msg_opl->apply_bytes = 0;
 615         msg_opl->sg.size = bytes;
 616
 617         msg_npl = &new->msg_plaintext;
 618         msg_npl->apply_bytes = apply;
 619         msg_npl->sg.size = orig_size - bytes;
 620
 621         j = msg_npl->sg.start;
 622         nsge = sk_msg_elem(msg_npl, j);
 623         if (tmp.length) {
 624                 memcpy(nsge, &tmp, sizeof(*nsge));
 625                 sk_msg_iter_var_next(j);
 626                 nsge = sk_msg_elem(msg_npl, j);
 627         }
 628
 629         osge = sk_msg_elem(msg_opl, i);
 630         while (osge->length) {
 631                 memcpy(nsge, osge, sizeof(*nsge));
 632                 sg_unmark_end(nsge);
 633                 sk_msg_iter_var_next(i);
 634                 sk_msg_iter_var_next(j);
 635                 if (i == *orig_end)
 636                         break;
 637                 osge = sk_msg_elem(msg_opl, i);
 638                 nsge = sk_msg_elem(msg_npl, j);
 639         }
 640
 641         msg_npl->sg.end = j;
 642         msg_npl->sg.curr = j;
 643         msg_npl->sg.copybreak = 0;
 644
 645         *to = new;
 646         return 0;
 647 }
 648
 649 static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
 650                                   struct tls_rec *from, u32 orig_end)
 651 {
 652         struct sk_msg *msg_npl = &from->msg_plaintext;
 653         struct sk_msg *msg_opl = &to->msg_plaintext;
 654         struct scatterlist *osge, *nsge;
 655         u32 i, j;
 656
 657         i = msg_opl->sg.end;
 658         sk_msg_iter_var_prev(i);
 659         j = msg_npl->sg.start;
 660
 661         osge = sk_msg_elem(msg_opl, i);
 662         nsge = sk_msg_elem(msg_npl, j);
 663
 664         if (sg_page(osge) == sg_page(nsge) &&
 665             osge->offset + osge->length == nsge->offset) {
 666                 osge->length += nsge->length;
 667                 put_page(sg_page(nsge));
 668         }
 669
 670         msg_opl->sg.end = orig_end;
 671         msg_opl->sg.curr = orig_end;
 672         msg_opl->sg.copybreak = 0;
 673         msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
 674         msg_opl->sg.size += msg_npl->sg.size;
 675
 676         sk_msg_free(sk, &to->msg_encrypted);
 677         sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted);
 678
 679         kfree(from);
 680 }
 681
 682 static int tls_push_record(struct sock *sk, int flags,
 683                            unsigned char record_type)
 684 {
 685         struct tls_context *tls_ctx = tls_get_ctx(sk);
 686         struct tls_prot_info *prot = &tls_ctx->prot_info;
 687         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 688         struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
 689         u32 i, split_point, orig_end;
 690         struct sk_msg *msg_pl, *msg_en;
 691         struct aead_request *req;
 692         bool split;
 693         int rc;
 694
 695         if (!rec)
 696                 return 0;
 697
 698         msg_pl = &rec->msg_plaintext;
 699         msg_en = &rec->msg_encrypted;
 700
 701         split_point = msg_pl->apply_bytes;
 702         split = split_point && split_point < msg_pl->sg.size;
 703         if (unlikely((!split &&
 704                       msg_pl->sg.size +
 705                       prot->overhead_size > msg_en->sg.size) ||
 706                      (split &&
 707                       split_point +
 708                       prot->overhead_size > msg_en->sg.size))) {
 709                 split = true;
 710                 split_point = msg_en->sg.size;
 711         }
 712         if (split) {
 713                 rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en,
 714                                            split_point, prot->overhead_size,
 715                                            &orig_end);
 716                 if (rc < 0)
 717                         return rc;
 718                 /* This can happen if above tls_split_open_record allocates
 719                  * a single large encryption buffer instead of two smaller
 720                  * ones. In this case adjust pointers and continue without
 721                  * split.
 722                  */
 723                 if (!msg_pl->sg.size) {
 724                         tls_merge_open_record(sk, rec, tmp, orig_end);
 725                         msg_pl = &rec->msg_plaintext;
 726                         msg_en = &rec->msg_encrypted;
 727                         split = false;
 728                 }
 729                 sk_msg_trim(sk, msg_en, msg_pl->sg.size +
 730                             prot->overhead_size);
 731         }
 732
 733         rec->tx_flags = flags;
 734         req = &rec->aead_req;
 735
 736         i = msg_pl->sg.end;
 737         sk_msg_iter_var_prev(i);
 738
 739         rec->content_type = record_type;
 740         if (prot->version == TLS_1_3_VERSION) {
 741                 /* Add content type to end of message.  No padding added */
 742                 sg_set_buf(&rec->sg_content_type, &rec->content_type, 1);
 743                 sg_mark_end(&rec->sg_content_type);
 744                 sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1,
 745                          &rec->sg_content_type);
 746         } else {
 747                 sg_mark_end(sk_msg_elem(msg_pl, i));
 748         }
 749
 750         if (msg_pl->sg.end < msg_pl->sg.start) {
 751                 sg_chain(&msg_pl->sg.data[msg_pl->sg.start],
 752                          MAX_SKB_FRAGS - msg_pl->sg.start + 1,
 753                          msg_pl->sg.data);
 754         }
 755
 756         i = msg_pl->sg.start;
 757         sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]);
 758
 759         i = msg_en->sg.end;
 760         sk_msg_iter_var_prev(i);
 761         sg_mark_end(sk_msg_elem(msg_en, i));
 762
 763         i = msg_en->sg.start;
 764         sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
 765
 766         tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size,
 767                      tls_ctx->tx.rec_seq, record_type, prot);
 768
 769         tls_fill_prepend(tls_ctx,
 770                          page_address(sg_page(&msg_en->sg.data[i])) +
 771                          msg_en->sg.data[i].offset,
 772                          msg_pl->sg.size + prot->tail_size,
 773                          record_type);
 774
 775         tls_ctx->pending_open_record_frags = false;
 776
 777         rc = tls_do_encryption(sk, tls_ctx, ctx, req,
 778                                msg_pl->sg.size + prot->tail_size, i);
 779         if (rc < 0) {
 780                 if (rc != -EINPROGRESS) {
 781                         tls_err_abort(sk, -EBADMSG);
 782                         if (split) {
 783                                 tls_ctx->pending_open_record_frags = true;
 784                                 tls_merge_open_record(sk, rec, tmp, orig_end);
 785                         }
 786                 }
 787                 ctx->async_capable = 1;
 788                 return rc;
 789         } else if (split) {
 790                 msg_pl = &tmp->msg_plaintext;
 791                 msg_en = &tmp->msg_encrypted;
 792                 sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size);
 793                 tls_ctx->pending_open_record_frags = true;
 794                 ctx->open_rec = tmp;
 795         }
 796
 797         return tls_tx_records(sk, flags);
 798 }
 799
 800 static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk,
 801                                bool full_record, u8 record_type,
 802                                ssize_t *copied, int flags)
 803 {
 804         struct tls_context *tls_ctx = tls_get_ctx(sk);
 805         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 806         struct sk_msg msg_redir = { };
 807         struct sk_psock *psock;
 808         struct sock *sk_redir;
 809         struct tls_rec *rec;
 810         bool enospc, policy;
 811         int err = 0, send;
 812         u32 delta = 0;
 813
 814         policy = !(flags & MSG_SENDPAGE_NOPOLICY);
 815         psock = sk_psock_get(sk);
 816         if (!psock || !policy) {
 817                 err = tls_push_record(sk, flags, record_type);
 818                 if (err && sk->sk_err == EBADMSG) {
 819                         *copied -= sk_msg_free(sk, msg);
 820                         tls_free_open_rec(sk);
 821                         err = -sk->sk_err;
 822                 }
 823                 if (psock)
 824                         sk_psock_put(sk, psock);
 825                 return err;
 826         }
 827 more_data:
 828         enospc = sk_msg_full(msg);
 829         if (psock->eval == __SK_NONE) {
 830                 delta = msg->sg.size;
 831                 psock->eval = sk_psock_msg_verdict(sk, psock, msg);
 832                 delta -= msg->sg.size;
 833         }
 834         if (msg->cork_bytes && msg->cork_bytes > msg->sg.size &&
 835             !enospc && !full_record) {
 836                 err = -ENOSPC;
 837                 goto out_err;
 838         }
 839         msg->cork_bytes = 0;
 840         send = msg->sg.size;
 841         if (msg->apply_bytes && msg->apply_bytes < send)
 842                 send = msg->apply_bytes;
 843
 844         switch (psock->eval) {
 845         case __SK_PASS:
 846                 err = tls_push_record(sk, flags, record_type);
 847                 if (err && sk->sk_err == EBADMSG) {
 848                         *copied -= sk_msg_free(sk, msg);
 849                         tls_free_open_rec(sk);
 850                         err = -sk->sk_err;
 851                         goto out_err;
 852                 }
 853                 break;
 854         case __SK_REDIRECT:
 855                 sk_redir = psock->sk_redir;
 856                 memcpy(&msg_redir, msg, sizeof(*msg));
 857                 if (msg->apply_bytes < send)
 858                         msg->apply_bytes = 0;
 859                 else
 860                         msg->apply_bytes -= send;
 861                 sk_msg_return_zero(sk, msg, send);
 862                 msg->sg.size -= send;
 863                 release_sock(sk);
 864                 err = tcp_bpf_sendmsg_redir(sk_redir, &msg_redir, send, flags);
 865                 lock_sock(sk);
 866                 if (err < 0) {
 867                         *copied -= sk_msg_free_nocharge(sk, &msg_redir);
 868                         msg->sg.size = 0;
 869                 }
 870                 if (msg->sg.size == 0)
 871                         tls_free_open_rec(sk);
 872                 break;
 873         case __SK_DROP:
 874         default:
 875                 sk_msg_free_partial(sk, msg, send);
 876                 if (msg->apply_bytes < send)
 877                         msg->apply_bytes = 0;
 878                 else
 879                         msg->apply_bytes -= send;
 880                 if (msg->sg.size == 0)
 881                         tls_free_open_rec(sk);
 882                 *copied -= (send + delta);
 883                 err = -EACCES;
 884         }
 885
 886         if (likely(!err)) {
 887                 bool reset_eval = !ctx->open_rec;
 888
 889                 rec = ctx->open_rec;
 890                 if (rec) {
 891                         msg = &rec->msg_plaintext;
 892                         if (!msg->apply_bytes)
 893                                 reset_eval = true;
 894                 }
 895                 if (reset_eval) {
 896                         psock->eval = __SK_NONE;
 897                         if (psock->sk_redir) {
 898                                 sock_put(psock->sk_redir);
 899                                 psock->sk_redir = NULL;
 900                         }
 901                 }
 902                 if (rec)
 903                         goto more_data;
 904         }
 905  out_err:
 906         sk_psock_put(sk, psock);
 907         return err;
 908 }
 909
 910 static int tls_sw_push_pending_record(struct sock *sk, int flags)
 911 {
 912         struct tls_context *tls_ctx = tls_get_ctx(sk);
 913         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 914         struct tls_rec *rec = ctx->open_rec;
 915         struct sk_msg *msg_pl;
 916         size_t copied;
 917
 918         if (!rec)
 919                 return 0;
 920
 921         msg_pl = &rec->msg_plaintext;
 922         copied = msg_pl->sg.size;
 923         if (!copied)
 924                 return 0;
 925
 926         return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA,
 927                                    &copied, flags);
 928 }
 929
 930 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
 931 {
 932         long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
 933         struct tls_context *tls_ctx = tls_get_ctx(sk);
 934         struct tls_prot_info *prot = &tls_ctx->prot_info;
 935         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 936         bool async_capable = ctx->async_capable;
 937         unsigned char record_type = TLS_RECORD_TYPE_DATA;
 938         bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
 939         bool eor = !(msg->msg_flags & MSG_MORE);
 940         size_t try_to_copy;
 941         ssize_t copied = 0;
 942         struct sk_msg *msg_pl, *msg_en;
 943         struct tls_rec *rec;
 944         int required_size;
 945         int num_async = 0;
 946         bool full_record;
 947         int record_room;
 948         int num_zc = 0;
 949         int orig_size;
 950         int ret = 0;
 951         int pending;
 952
 953         if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
 954                                MSG_CMSG_COMPAT))
 955                 return -EOPNOTSUPP;
 956
 957         mutex_lock(&tls_ctx->tx_lock);
 958         lock_sock(sk);
 959
 960         if (unlikely(msg->msg_controllen)) {
 961                 ret = tls_proccess_cmsg(sk, msg, &record_type);
 962                 if (ret) {
 963                         if (ret == -EINPROGRESS)
 964                                 num_async++;
 965                         else if (ret != -EAGAIN)
 966                                 goto send_end;
 967                 }
 968         }
 969
 970         while (msg_data_left(msg)) {
 971                 if (sk->sk_err) {
 972                         ret = -sk->sk_err;
 973                         goto send_end;
 974                 }
 975
 976                 if (ctx->open_rec)
 977                         rec = ctx->open_rec;
 978                 else
 979                         rec = ctx->open_rec = tls_get_rec(sk);
 980                 if (!rec) {
 981                         ret = -ENOMEM;
 982                         goto send_end;
 983                 }
 984
 985                 msg_pl = &rec->msg_plaintext;
 986                 msg_en = &rec->msg_encrypted;
 987
 988                 orig_size = msg_pl->sg.size;
 989                 full_record = false;
 990                 try_to_copy = msg_data_left(msg);
 991                 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
 992                 if (try_to_copy >= record_room) {
 993                         try_to_copy = record_room;
 994                         full_record = true;
 995                 }
 996
 997                 required_size = msg_pl->sg.size + try_to_copy +
 998                                 prot->overhead_size;
 999
1000                 if (!sk_stream_memory_free(sk))
1001                         goto wait_for_sndbuf;
1002
1003 alloc_encrypted:
1004                 ret = tls_alloc_encrypted_msg(sk, required_size);
1005                 if (ret) {
1006                         if (ret != -ENOSPC)
1007                                 goto wait_for_memory;
1008
1009                         /* Adjust try_to_copy according to the amount that was
1010                          * actually allocated. The difference is due
1011                          * to max sg elements limit
1012                          */
1013                         try_to_copy -= required_size - msg_en->sg.size;
1014                         full_record = true;
1015                 }
1016
1017                 if (!is_kvec && (full_record || eor) && !async_capable) {
1018                         u32 first = msg_pl->sg.end;
1019
1020                         ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter,
1021                                                         msg_pl, try_to_copy);
1022                         if (ret)
1023                                 goto fallback_to_reg_send;
1024
1025                         num_zc++;
1026                         copied += try_to_copy;
1027
1028                         sk_msg_sg_copy_set(msg_pl, first);
1029                         ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1030                                                   record_type, &copied,
1031                                                   msg->msg_flags);
1032                         if (ret) {
1033                                 if (ret == -EINPROGRESS)
1034                                         num_async++;
1035                                 else if (ret == -ENOMEM)
1036                                         goto wait_for_memory;
1037                                 else if (ctx->open_rec && ret == -ENOSPC)
1038                                         goto rollback_iter;
1039                                 else if (ret != -EAGAIN)
1040                                         goto send_end;
1041                         }
1042                         continue;
1043 rollback_iter:
1044                         copied -= try_to_copy;
1045                         sk_msg_sg_copy_clear(msg_pl, first);
1046                         iov_iter_revert(&msg->msg_iter,
1047                                         msg_pl->sg.size - orig_size);
1048 fallback_to_reg_send:
1049                         sk_msg_trim(sk, msg_pl, orig_size);
1050                 }
1051
1052                 required_size = msg_pl->sg.size + try_to_copy;
1053
1054                 ret = tls_clone_plaintext_msg(sk, required_size);
1055                 if (ret) {
1056                         if (ret != -ENOSPC)
1057                                 goto send_end;
1058
1059                         /* Adjust try_to_copy according to the amount that was
1060                          * actually allocated. The difference is due
1061                          * to max sg elements limit
1062                          */
1063                         try_to_copy -= required_size - msg_pl->sg.size;
1064                         full_record = true;
1065                         sk_msg_trim(sk, msg_en,
1066                                     msg_pl->sg.size + prot->overhead_size);
1067                 }
1068
1069                 if (try_to_copy) {
1070                         ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter,
1071                                                        msg_pl, try_to_copy);
1072                         if (ret < 0)
1073                                 goto trim_sgl;
1074                 }
1075
1076                 /* Open records defined only if successfully copied, otherwise
1077                  * we would trim the sg but not reset the open record frags.
1078                  */
1079                 tls_ctx->pending_open_record_frags = true;
1080                 copied += try_to_copy;
1081                 if (full_record || eor) {
1082                         ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1083                                                   record_type, &copied,
1084                                                   msg->msg_flags);
1085                         if (ret) {
1086                                 if (ret == -EINPROGRESS)
1087                                         num_async++;
1088                                 else if (ret == -ENOMEM)
1089                                         goto wait_for_memory;
1090                                 else if (ret != -EAGAIN) {
1091                                         if (ret == -ENOSPC)
1092                                                 ret = 0;
1093                                         goto send_end;
1094                                 }
1095                         }
1096                 }
1097
1098                 continue;
1099
1100 wait_for_sndbuf:
1101                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1102 wait_for_memory:
1103                 ret = sk_stream_wait_memory(sk, &timeo);
1104                 if (ret) {
1105 trim_sgl:
1106                         if (ctx->open_rec)
1107                                 tls_trim_both_msgs(sk, orig_size);
1108                         goto send_end;
1109                 }
1110
1111                 if (ctx->open_rec && msg_en->sg.size < required_size)
1112                         goto alloc_encrypted;
1113         }
1114
1115         if (!num_async) {
1116                 goto send_end;
1117         } else if (num_zc) {
1118                 /* Wait for pending encryptions to get completed */
1119                 spin_lock_bh(&ctx->encrypt_compl_lock);
1120                 ctx->async_notify = true;
1121
1122                 pending = atomic_read(&ctx->encrypt_pending);
1123                 spin_unlock_bh(&ctx->encrypt_compl_lock);
1124                 if (pending)
1125                         crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1126                 else
1127                         reinit_completion(&ctx->async_wait.completion);
1128
1129                 /* There can be no concurrent accesses, since we have no
1130                  * pending encrypt operations
1131                  */
1132                 WRITE_ONCE(ctx->async_notify, false);
1133
1134                 if (ctx->async_wait.err) {
1135                         ret = ctx->async_wait.err;
1136                         copied = 0;
1137                 }
1138         }
1139
1140         /* Transmit if any encryptions have completed */
1141         if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1142                 cancel_delayed_work(&ctx->tx_work.work);
1143                 tls_tx_records(sk, msg->msg_flags);
1144         }
1145
1146 send_end:
1147         ret = sk_stream_error(sk, msg->msg_flags, ret);
1148
1149         release_sock(sk);
1150         mutex_unlock(&tls_ctx->tx_lock);
1151         return copied > 0 ? copied : ret;
1152 }
1153
1154 static int tls_sw_do_sendpage(struct sock *sk, struct page *page,
1155                               int offset, size_t size, int flags)
1156 {
1157         long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1158         struct tls_context *tls_ctx = tls_get_ctx(sk);
1159         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1160         struct tls_prot_info *prot = &tls_ctx->prot_info;
1161         unsigned char record_type = TLS_RECORD_TYPE_DATA;
1162         struct sk_msg *msg_pl;
1163         struct tls_rec *rec;
1164         int num_async = 0;
1165         ssize_t copied = 0;
1166         bool full_record;
1167         int record_room;
1168         int ret = 0;
1169         bool eor;
1170
1171         eor = !(flags & MSG_SENDPAGE_NOTLAST);
1172         sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1173
1174         /* Call the sk_stream functions to manage the sndbuf mem. */
1175         while (size > 0) {
1176                 size_t copy, required_size;
1177
1178                 if (sk->sk_err) {
1179                         ret = -sk->sk_err;
1180                         goto sendpage_end;
1181                 }
1182
1183                 if (ctx->open_rec)
1184                         rec = ctx->open_rec;
1185                 else
1186                         rec = ctx->open_rec = tls_get_rec(sk);
1187                 if (!rec) {
1188                         ret = -ENOMEM;
1189                         goto sendpage_end;
1190                 }
1191
1192                 msg_pl = &rec->msg_plaintext;
1193
1194                 full_record = false;
1195                 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
1196                 copy = size;
1197                 if (copy >= record_room) {
1198                         copy = record_room;
1199                         full_record = true;
1200                 }
1201
1202                 required_size = msg_pl->sg.size + copy + prot->overhead_size;
1203
1204                 if (!sk_stream_memory_free(sk))
1205                         goto wait_for_sndbuf;
1206 alloc_payload:
1207                 ret = tls_alloc_encrypted_msg(sk, required_size);
1208                 if (ret) {
1209                         if (ret != -ENOSPC)
1210                                 goto wait_for_memory;
1211
1212                         /* Adjust copy according to the amount that was
1213                          * actually allocated. The difference is due
1214                          * to max sg elements limit
1215                          */
1216                         copy -= required_size - msg_pl->sg.size;
1217                         full_record = true;
1218                 }
1219
1220                 sk_msg_page_add(msg_pl, page, copy, offset);
1221                 sk_mem_charge(sk, copy);
1222
1223                 offset += copy;
1224                 size -= copy;
1225                 copied += copy;
1226
1227                 tls_ctx->pending_open_record_frags = true;
1228                 if (full_record || eor || sk_msg_full(msg_pl)) {
1229                         ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1230                                                   record_type, &copied, flags);
1231                         if (ret) {
1232                                 if (ret == -EINPROGRESS)
1233                                         num_async++;
1234                                 else if (ret == -ENOMEM)
1235                                         goto wait_for_memory;
1236                                 else if (ret != -EAGAIN) {
1237                                         if (ret == -ENOSPC)
1238                                                 ret = 0;
1239                                         goto sendpage_end;
1240                                 }
1241                         }
1242                 }
1243                 continue;
1244 wait_for_sndbuf:
1245                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1246 wait_for_memory:
1247                 ret = sk_stream_wait_memory(sk, &timeo);
1248                 if (ret) {
1249                         if (ctx->open_rec)
1250                                 tls_trim_both_msgs(sk, msg_pl->sg.size);
1251                         goto sendpage_end;
1252                 }
1253
1254                 if (ctx->open_rec)
1255                         goto alloc_payload;
1256         }
1257
1258         if (num_async) {
1259                 /* Transmit if any encryptions have completed */
1260                 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1261                         cancel_delayed_work(&ctx->tx_work.work);
1262                         tls_tx_records(sk, flags);
1263                 }
1264         }
1265 sendpage_end:
1266         ret = sk_stream_error(sk, flags, ret);
1267         return copied > 0 ? copied : ret;
1268 }
1269
1270 int tls_sw_sendpage_locked(struct sock *sk, struct page *page,
1271                            int offset, size_t size, int flags)
1272 {
1273         if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1274                       MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY |
1275                       MSG_NO_SHARED_FRAGS))
1276                 return -EOPNOTSUPP;
1277
1278         return tls_sw_do_sendpage(sk, page, offset, size, flags);
1279 }
1280
1281 int tls_sw_sendpage(struct sock *sk, struct page *page,
1282                     int offset, size_t size, int flags)
1283 {
1284         struct tls_context *tls_ctx = tls_get_ctx(sk);
1285         int ret;
1286
1287         if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1288                       MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY))
1289                 return -EOPNOTSUPP;
1290
1291         mutex_lock(&tls_ctx->tx_lock);
1292         lock_sock(sk);
1293         ret = tls_sw_do_sendpage(sk, page, offset, size, flags);
1294         release_sock(sk);
1295         mutex_unlock(&tls_ctx->tx_lock);
1296         return ret;
1297 }
1298
1299 static struct sk_buff *tls_wait_data(struct sock *sk, struct sk_psock *psock,
1300                                      bool nonblock, long timeo, int *err)
1301 {
1302         struct tls_context *tls_ctx = tls_get_ctx(sk);
1303         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1304         struct sk_buff *skb;
1305         DEFINE_WAIT_FUNC(wait, woken_wake_function);
1306
1307         while (!(skb = ctx->recv_pkt) && sk_psock_queue_empty(psock)) {
1308                 if (sk->sk_err) {
1309                         *err = sock_error(sk);
1310                         return NULL;
1311                 }
1312
1313                 if (!skb_queue_empty(&sk->sk_receive_queue)) {
1314                         __strp_unpause(&ctx->strp);
1315                         if (ctx->recv_pkt)
1316                                 return ctx->recv_pkt;
1317                 }
1318
1319                 if (sk->sk_shutdown & RCV_SHUTDOWN)
1320                         return NULL;
1321
1322                 if (sock_flag(sk, SOCK_DONE))
1323                         return NULL;
1324
1325                 if (nonblock || !timeo) {
1326                         *err = -EAGAIN;
1327                         return NULL;
1328                 }
1329
1330                 add_wait_queue(sk_sleep(sk), &wait);
1331                 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1332                 sk_wait_event(sk, &timeo,
1333                               ctx->recv_pkt != skb ||
1334                               !sk_psock_queue_empty(psock),
1335                               &wait);
1336                 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1337                 remove_wait_queue(sk_sleep(sk), &wait);
1338
1339                 /* Handle signals */
1340                 if (signal_pending(current)) {
1341                         *err = sock_intr_errno(timeo);
1342                         return NULL;
1343                 }
1344         }
1345
1346         return skb;
1347 }
1348
1349 static int tls_setup_from_iter(struct sock *sk, struct iov_iter *from,
1350                                int length, int *pages_used,
1351                                unsigned int *size_used,
1352                                struct scatterlist *to,
1353                                int to_max_pages)
1354 {
1355         int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1356         struct page *pages[MAX_SKB_FRAGS];
1357         unsigned int size = *size_used;
1358         ssize_t copied, use;
1359         size_t offset;
1360
1361         while (length > 0) {
1362                 i = 0;
1363                 maxpages = to_max_pages - num_elem;
1364                 if (maxpages == 0) {
1365                         rc = -EFAULT;
1366                         goto out;
1367                 }
1368                 copied = iov_iter_get_pages(from, pages,
1369                                             length,
1370                                             maxpages, &offset);
1371                 if (copied <= 0) {
1372                         rc = -EFAULT;
1373                         goto out;
1374                 }
1375
1376                 iov_iter_advance(from, copied);
1377
1378                 length -= copied;
1379                 size += copied;
1380                 while (copied) {
1381                         use = min_t(int, copied, PAGE_SIZE - offset);
1382
1383                         sg_set_page(&to[num_elem],
1384                                     pages[i], use, offset);
1385                         sg_unmark_end(&to[num_elem]);
1386                         /* We do not uncharge memory from this API */
1387
1388                         offset = 0;
1389                         copied -= use;
1390
1391                         i++;
1392                         num_elem++;
1393                 }
1394         }
1395         /* Mark the end in the last sg entry if newly added */
1396         if (num_elem > *pages_used)
1397                 sg_mark_end(&to[num_elem - 1]);
1398 out:
1399         if (rc)
1400                 iov_iter_revert(from, size - *size_used);
1401         *size_used = size;
1402         *pages_used = num_elem;
1403
1404         return rc;
1405 }
1406
1407 /* This function decrypts the input skb into either out_iov or in out_sg
1408  * or in skb buffers itself. The input parameter 'zc' indicates if
1409  * zero-copy mode needs to be tried or not. With zero-copy mode, either
1410  * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1411  * NULL, then the decryption happens inside skb buffers itself, i.e.
1412  * zero-copy gets disabled and 'zc' is updated.
1413  */
1414
1415 static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
1416                             struct iov_iter *out_iov,
1417                             struct scatterlist *out_sg,
1418                             int *chunk, bool *zc, bool async)
1419 {
1420         struct tls_context *tls_ctx = tls_get_ctx(sk);
1421         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1422         struct tls_prot_info *prot = &tls_ctx->prot_info;
1423         struct strp_msg *rxm = strp_msg(skb);
1424         int n_sgin, n_sgout, nsg, mem_size, aead_size, err, pages = 0;
1425         struct aead_request *aead_req;
1426         struct sk_buff *unused;
1427         u8 *aad, *iv, *mem = NULL;
1428         struct scatterlist *sgin = NULL;
1429         struct scatterlist *sgout = NULL;
1430         const int data_len = rxm->full_len - prot->overhead_size +
1431                              prot->tail_size;
1432         int iv_offset = 0;
1433
1434         if (*zc && (out_iov || out_sg)) {
1435                 if (out_iov)
1436                         n_sgout = 1 +
1437                                 iov_iter_npages_cap(out_iov, INT_MAX, data_len);
1438                 else
1439                         n_sgout = sg_nents(out_sg);
1440                 n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size,
1441                                  rxm->full_len - prot->prepend_size);
1442         } else {
1443                 n_sgout = 0;
1444                 *zc = false;
1445                 n_sgin = skb_cow_data(skb, 0, &unused);
1446         }
1447
1448         if (n_sgin < 1)
1449                 return -EBADMSG;
1450
1451         /* Increment to accommodate AAD */
1452         n_sgin = n_sgin + 1;
1453
1454         nsg = n_sgin + n_sgout;
1455
1456         aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
1457         mem_size = aead_size + (nsg * sizeof(struct scatterlist));
1458         mem_size = mem_size + prot->aad_size;
1459         mem_size = mem_size + crypto_aead_ivsize(ctx->aead_recv);
1460
1461         /* Allocate a single block of memory which contains
1462          * aead_req || sgin[] || sgout[] || aad || iv.
1463          * This order achieves correct alignment for aead_req, sgin, sgout.
1464          */
1465         mem = kmalloc(mem_size, sk->sk_allocation);
1466         if (!mem)
1467                 return -ENOMEM;
1468
1469         /* Segment the allocated memory */
1470         aead_req = (struct aead_request *)mem;
1471         sgin = (struct scatterlist *)(mem + aead_size);
1472         sgout = sgin + n_sgin;
1473         aad = (u8 *)(sgout + n_sgout);
1474         iv = aad + prot->aad_size;
1475
1476         /* For CCM based ciphers, first byte of nonce+iv is a constant */
1477         switch (prot->cipher_type) {
1478         case TLS_CIPHER_AES_CCM_128:
1479                 iv[0] = TLS_AES_CCM_IV_B0_BYTE;
1480                 iv_offset = 1;
1481                 break;
1482         case TLS_CIPHER_SM4_CCM:
1483                 iv[0] = TLS_SM4_CCM_IV_B0_BYTE;
1484                 iv_offset = 1;
1485                 break;
1486         }
1487
1488         /* Prepare IV */
1489         err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
1490                             iv + iv_offset + prot->salt_size,
1491                             prot->iv_size);
1492         if (err < 0) {
1493                 kfree(mem);
1494                 return err;
1495         }
1496         if (prot->version == TLS_1_3_VERSION ||
1497             prot->cipher_type == TLS_CIPHER_CHACHA20_POLY1305)
1498                 memcpy(iv + iv_offset, tls_ctx->rx.iv,
1499                        prot->iv_size + prot->salt_size);
1500         else
1501                 memcpy(iv + iv_offset, tls_ctx->rx.iv, prot->salt_size);
1502
1503         xor_iv_with_seq(prot, iv + iv_offset, tls_ctx->rx.rec_seq);
1504
1505         /* Prepare AAD */
1506         tls_make_aad(aad, rxm->full_len - prot->overhead_size +
1507                      prot->tail_size,
1508                      tls_ctx->rx.rec_seq, ctx->control, prot);
1509
1510         /* Prepare sgin */
1511         sg_init_table(sgin, n_sgin);
1512         sg_set_buf(&sgin[0], aad, prot->aad_size);
1513         err = skb_to_sgvec(skb, &sgin[1],
1514                            rxm->offset + prot->prepend_size,
1515                            rxm->full_len - prot->prepend_size);
1516         if (err < 0) {
1517                 kfree(mem);
1518                 return err;
1519         }
1520
1521         if (n_sgout) {
1522                 if (out_iov) {
1523                         sg_init_table(sgout, n_sgout);
1524                         sg_set_buf(&sgout[0], aad, prot->aad_size);
1525
1526                         *chunk = 0;
1527                         err = tls_setup_from_iter(sk, out_iov, data_len,
1528                                                   &pages, chunk, &sgout[1],
1529                                                   (n_sgout - 1));
1530                         if (err < 0)
1531                                 goto fallback_to_reg_recv;
1532                 } else if (out_sg) {
1533                         memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
1534                 } else {
1535                         goto fallback_to_reg_recv;
1536                 }
1537         } else {
1538 fallback_to_reg_recv:
1539                 sgout = sgin;
1540                 pages = 0;
1541                 *chunk = data_len;
1542                 *zc = false;
1543         }
1544
1545         /* Prepare and submit AEAD request */
1546         err = tls_do_decryption(sk, skb, sgin, sgout, iv,
1547                                 data_len, aead_req, async);
1548         if (err == -EINPROGRESS)
1549                 return err;
1550
1551         /* Release the pages in case iov was mapped to pages */
1552         for (; pages > 0; pages--)
1553                 put_page(sg_page(&sgout[pages]));
1554
1555         kfree(mem);
1556         return err;
1557 }
1558
1559 static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
1560                               struct iov_iter *dest, int *chunk, bool *zc,
1561                               bool async)
1562 {
1563         struct tls_context *tls_ctx = tls_get_ctx(sk);
1564         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1565         struct tls_prot_info *prot = &tls_ctx->prot_info;
1566         struct strp_msg *rxm = strp_msg(skb);
1567         int pad, err = 0;
1568
1569         if (!ctx->decrypted) {
1570                 if (tls_ctx->rx_conf == TLS_HW) {
1571                         err = tls_device_decrypted(sk, tls_ctx, skb, rxm);
1572                         if (err < 0)
1573                                 return err;
1574                 }
1575
1576                 /* Still not decrypted after tls_device */
1577                 if (!ctx->decrypted) {
1578                         err = decrypt_internal(sk, skb, dest, NULL, chunk, zc,
1579                                                async);
1580                         if (err < 0) {
1581                                 if (err == -EINPROGRESS)
1582                                         tls_advance_record_sn(sk, prot,
1583                                                               &tls_ctx->rx);
1584                                 else if (err == -EBADMSG)
1585                                         TLS_INC_STATS(sock_net(sk),
1586                                                       LINUX_MIB_TLSDECRYPTERROR);
1587                                 return err;
1588                         }
1589                 } else {
1590                         *zc = false;
1591                 }
1592
1593                 pad = padding_length(ctx, prot, skb);
1594                 if (pad < 0)
1595                         return pad;
1596
1597                 rxm->full_len -= pad;
1598                 rxm->offset += prot->prepend_size;
1599                 rxm->full_len -= prot->overhead_size;
1600                 tls_advance_record_sn(sk, prot, &tls_ctx->rx);
1601                 ctx->decrypted = 1;
1602                 ctx->saved_data_ready(sk);
1603         } else {
1604                 *zc = false;
1605         }
1606
1607         return err;
1608 }
1609
1610 int decrypt_skb(struct sock *sk, struct sk_buff *skb,
1611                 struct scatterlist *sgout)
1612 {
1613         bool zc = true;
1614         int chunk;
1615
1616         return decrypt_internal(sk, skb, NULL, sgout, &chunk, &zc, false);
1617 }
1618
1619 static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
1620                                unsigned int len)
1621 {
1622         struct tls_context *tls_ctx = tls_get_ctx(sk);
1623         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1624
1625         if (skb) {
1626                 struct strp_msg *rxm = strp_msg(skb);
1627
1628                 if (len < rxm->full_len) {
1629                         rxm->offset += len;
1630                         rxm->full_len -= len;
1631                         return false;
1632                 }
1633                 consume_skb(skb);
1634         }
1635
1636         /* Finished with message */
1637         ctx->recv_pkt = NULL;
1638         __strp_unpause(&ctx->strp);
1639
1640         return true;
1641 }
1642
1643 /* This function traverses the rx_list in tls receive context to copies the
1644  * decrypted records into the buffer provided by caller zero copy is not
1645  * true. Further, the records are removed from the rx_list if it is not a peek
1646  * case and the record has been consumed completely.
1647  */
1648 static int process_rx_list(struct tls_sw_context_rx *ctx,
1649                            struct msghdr *msg,
1650                            u8 *control,
1651                            bool *cmsg,
1652                            size_t skip,
1653                            size_t len,
1654                            bool zc,
1655                            bool is_peek)
1656 {
1657         struct sk_buff *skb = skb_peek(&ctx->rx_list);
1658         u8 ctrl = *control;
1659         u8 msgc = *cmsg;
1660         struct tls_msg *tlm;
1661         ssize_t copied = 0;
1662
1663         /* Set the record type in 'control' if caller didn't pass it */
1664         if (!ctrl && skb) {
1665                 tlm = tls_msg(skb);
1666                 ctrl = tlm->control;
1667         }
1668
1669         while (skip && skb) {
1670                 struct strp_msg *rxm = strp_msg(skb);
1671                 tlm = tls_msg(skb);
1672
1673                 /* Cannot process a record of different type */
1674                 if (ctrl != tlm->control)
1675                         return 0;
1676
1677                 if (skip < rxm->full_len)
1678                         break;
1679
1680                 skip = skip - rxm->full_len;
1681                 skb = skb_peek_next(skb, &ctx->rx_list);
1682         }
1683
1684         while (len && skb) {
1685                 struct sk_buff *next_skb;
1686                 struct strp_msg *rxm = strp_msg(skb);
1687                 int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1688
1689                 tlm = tls_msg(skb);
1690
1691                 /* Cannot process a record of different type */
1692                 if (ctrl != tlm->control)
1693                         return 0;
1694
1695                 /* Set record type if not already done. For a non-data record,
1696                  * do not proceed if record type could not be copied.
1697                  */
1698                 if (!msgc) {
1699                         int cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1700                                             sizeof(ctrl), &ctrl);
1701                         msgc = true;
1702                         if (ctrl != TLS_RECORD_TYPE_DATA) {
1703                                 if (cerr || msg->msg_flags & MSG_CTRUNC)
1704                                         return -EIO;
1705
1706                                 *cmsg = msgc;
1707                         }
1708                 }
1709
1710                 if (!zc || (rxm->full_len - skip) > len) {
1711                         int err = skb_copy_datagram_msg(skb, rxm->offset + skip,
1712                                                     msg, chunk);
1713                         if (err < 0)
1714                                 return err;
1715                 }
1716
1717                 len = len - chunk;
1718                 copied = copied + chunk;
1719
1720                 /* Consume the data from record if it is non-peek case*/
1721                 if (!is_peek) {
1722                         rxm->offset = rxm->offset + chunk;
1723                         rxm->full_len = rxm->full_len - chunk;
1724
1725                         /* Return if there is unconsumed data in the record */
1726                         if (rxm->full_len - skip)
1727                                 break;
1728                 }
1729
1730                 /* The remaining skip-bytes must lie in 1st record in rx_list.
1731                  * So from the 2nd record, 'skip' should be 0.
1732                  */
1733                 skip = 0;
1734
1735                 if (msg)
1736                         msg->msg_flags |= MSG_EOR;
1737
1738                 next_skb = skb_peek_next(skb, &ctx->rx_list);
1739
1740                 if (!is_peek) {
1741                         skb_unlink(skb, &ctx->rx_list);
1742                         consume_skb(skb);
1743                 }
1744
1745                 skb = next_skb;
1746         }
1747
1748         *control = ctrl;
1749         return copied;
1750 }
1751
1752 int tls_sw_recvmsg(struct sock *sk,
1753                    struct msghdr *msg,
1754                    size_t len,
1755                    int nonblock,
1756                    int flags,
1757                    int *addr_len)
1758 {
1759         struct tls_context *tls_ctx = tls_get_ctx(sk);
1760         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1761         struct tls_prot_info *prot = &tls_ctx->prot_info;
1762         struct sk_psock *psock;
1763         unsigned char control = 0;
1764         ssize_t decrypted = 0;
1765         struct strp_msg *rxm;
1766         struct tls_msg *tlm;
1767         struct sk_buff *skb;
1768         ssize_t copied = 0;
1769         bool cmsg = false;
1770         int target, err = 0;
1771         long timeo;
1772         bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1773         bool is_peek = flags & MSG_PEEK;
1774         bool bpf_strp_enabled;
1775         int num_async = 0;
1776         int pending;
1777
1778         flags |= nonblock;
1779
1780         if (unlikely(flags & MSG_ERRQUEUE))
1781                 return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
1782
1783         psock = sk_psock_get(sk);
1784         lock_sock(sk);
1785         bpf_strp_enabled = sk_psock_strp_enabled(psock);
1786
1787         /* Process pending decrypted records. It must be non-zero-copy */
1788         err = process_rx_list(ctx, msg, &control, &cmsg, 0, len, false,
1789                               is_peek);
1790         if (err < 0) {
1791                 tls_err_abort(sk, err);
1792                 goto end;
1793         } else {
1794                 copied = err;
1795         }
1796
1797         if (len <= copied)
1798                 goto recv_end;
1799
1800         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1801         len = len - copied;
1802         timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1803
1804         while (len && (decrypted + copied < target || ctx->recv_pkt)) {
1805                 bool retain_skb = false;
1806                 bool zc = false;
1807                 int to_decrypt;
1808                 int chunk = 0;
1809                 bool async_capable;
1810                 bool async = false;
1811
1812                 skb = tls_wait_data(sk, psock, flags & MSG_DONTWAIT, timeo, &err);
1813                 if (!skb) {
1814                         if (psock) {
1815                                 int ret = sk_msg_recvmsg(sk, psock, msg, len,
1816                                                          flags);
1817
1818                                 if (ret > 0) {
1819                                         decrypted += ret;
1820                                         len -= ret;
1821                                         continue;
1822                                 }
1823                         }
1824                         goto recv_end;
1825                 } else {
1826                         tlm = tls_msg(skb);
1827                         if (prot->version == TLS_1_3_VERSION)
1828                                 tlm->control = 0;
1829                         else
1830                                 tlm->control = ctx->control;
1831                 }
1832
1833                 rxm = strp_msg(skb);
1834
1835                 to_decrypt = rxm->full_len - prot->overhead_size;
1836
1837                 if (to_decrypt <= len && !is_kvec && !is_peek &&
1838                     ctx->control == TLS_RECORD_TYPE_DATA &&
1839                     prot->version != TLS_1_3_VERSION &&
1840                     !bpf_strp_enabled)
1841                         zc = true;
1842
1843                 /* Do not use async mode if record is non-data */
1844                 if (ctx->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled)
1845                         async_capable = ctx->async_capable;
1846                 else
1847                         async_capable = false;
1848
1849                 err = decrypt_skb_update(sk, skb, &msg->msg_iter,
1850                                          &chunk, &zc, async_capable);
1851                 if (err < 0 && err != -EINPROGRESS) {
1852                         tls_err_abort(sk, -EBADMSG);
1853                         goto recv_end;
1854                 }
1855
1856                 if (err == -EINPROGRESS) {
1857                         async = true;
1858                         num_async++;
1859                 } else if (prot->version == TLS_1_3_VERSION) {
1860                         tlm->control = ctx->control;
1861                 }
1862
1863                 /* If the type of records being processed is not known yet,
1864                  * set it to record type just dequeued. If it is already known,
1865                  * but does not match the record type just dequeued, go to end.
1866                  * We always get record type here since for tls1.2, record type
1867                  * is known just after record is dequeued from stream parser.
1868                  * For tls1.3, we disable async.
1869                  */
1870
1871                 if (!control)
1872                         control = tlm->control;
1873                 else if (control != tlm->control)
1874                         goto recv_end;
1875
1876                 if (!cmsg) {
1877                         int cerr;
1878
1879                         cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1880                                         sizeof(control), &control);
1881                         cmsg = true;
1882                         if (control != TLS_RECORD_TYPE_DATA) {
1883                                 if (cerr || msg->msg_flags & MSG_CTRUNC) {
1884                                         err = -EIO;
1885                                         goto recv_end;
1886                                 }
1887                         }
1888                 }
1889
1890                 if (async)
1891                         goto pick_next_record;
1892
1893                 if (!zc) {
1894                         if (bpf_strp_enabled) {
1895                                 err = sk_psock_tls_strp_read(psock, skb);
1896                                 if (err != __SK_PASS) {
1897                                         rxm->offset = rxm->offset + rxm->full_len;
1898                                         rxm->full_len = 0;
1899                                         if (err == __SK_DROP)
1900                                                 consume_skb(skb);
1901                                         ctx->recv_pkt = NULL;
1902                                         __strp_unpause(&ctx->strp);
1903                                         continue;
1904                                 }
1905                         }
1906
1907                         if (rxm->full_len > len) {
1908                                 retain_skb = true;
1909                                 chunk = len;
1910                         } else {
1911                                 chunk = rxm->full_len;
1912                         }
1913
1914                         err = skb_copy_datagram_msg(skb, rxm->offset,
1915                                                     msg, chunk);
1916                         if (err < 0)
1917                                 goto recv_end;
1918
1919                         if (!is_peek) {
1920                                 rxm->offset = rxm->offset + chunk;
1921                                 rxm->full_len = rxm->full_len - chunk;
1922                         }
1923                 }
1924
1925 pick_next_record:
1926                 if (chunk > len)
1927                         chunk = len;
1928
1929                 decrypted += chunk;
1930                 len -= chunk;
1931
1932                 /* For async or peek case, queue the current skb */
1933                 if (async || is_peek || retain_skb) {
1934                         skb_queue_tail(&ctx->rx_list, skb);
1935                         skb = NULL;
1936                 }
1937
1938                 if (tls_sw_advance_skb(sk, skb, chunk)) {
1939                         /* Return full control message to
1940                          * userspace before trying to parse
1941                          * another message type
1942                          */
1943                         msg->msg_flags |= MSG_EOR;
1944                         if (control != TLS_RECORD_TYPE_DATA)
1945                                 goto recv_end;
1946                 } else {
1947                         break;
1948                 }
1949         }
1950
1951 recv_end:
1952         if (num_async) {
1953                 /* Wait for all previously submitted records to be decrypted */
1954                 spin_lock_bh(&ctx->decrypt_compl_lock);
1955                 ctx->async_notify = true;
1956                 pending = atomic_read(&ctx->decrypt_pending);
1957                 spin_unlock_bh(&ctx->decrypt_compl_lock);
1958                 if (pending) {
1959                         err = crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1960                         if (err) {
1961                                 /* one of async decrypt failed */
1962                                 tls_err_abort(sk, err);
1963                                 copied = 0;
1964                                 decrypted = 0;
1965                                 goto end;
1966                         }
1967                 } else {
1968                         reinit_completion(&ctx->async_wait.completion);
1969                 }
1970
1971                 /* There can be no concurrent accesses, since we have no
1972                  * pending decrypt operations
1973                  */
1974                 WRITE_ONCE(ctx->async_notify, false);
1975
1976                 /* Drain records from the rx_list & copy if required */
1977                 if (is_peek || is_kvec)
1978                         err = process_rx_list(ctx, msg, &control, &cmsg, copied,
1979                                               decrypted, false, is_peek);
1980                 else
1981                         err = process_rx_list(ctx, msg, &control, &cmsg, 0,
1982                                               decrypted, true, is_peek);
1983                 if (err < 0) {
1984                         tls_err_abort(sk, err);
1985                         copied = 0;
1986                         goto end;
1987                 }
1988         }
1989
1990         copied += decrypted;
1991
1992 end:
1993         release_sock(sk);
1994         sk_defer_free_flush(sk);
1995         if (psock)
1996                 sk_psock_put(sk, psock);
1997         return copied ? : err;
1998 }
1999
2000 ssize_t tls_sw_splice_read(struct socket *sock,  loff_t *ppos,
2001                            struct pipe_inode_info *pipe,
2002                            size_t len, unsigned int flags)
2003 {
2004         struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
2005         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2006         struct strp_msg *rxm = NULL;
2007         struct sock *sk = sock->sk;
2008         struct sk_buff *skb;
2009         ssize_t copied = 0;
2010         bool from_queue;
2011         int err = 0;
2012         long timeo;
2013         int chunk;
2014         bool zc = false;
2015
2016         lock_sock(sk);
2017
2018         timeo = sock_rcvtimeo(sk, flags & SPLICE_F_NONBLOCK);
2019
2020         from_queue = !skb_queue_empty(&ctx->rx_list);
2021         if (from_queue) {
2022                 skb = __skb_dequeue(&ctx->rx_list);
2023         } else {
2024                 skb = tls_wait_data(sk, NULL, flags & SPLICE_F_NONBLOCK, timeo,
2025                                     &err);
2026                 if (!skb)
2027                         goto splice_read_end;
2028
2029                 err = decrypt_skb_update(sk, skb, NULL, &chunk, &zc, false);
2030                 if (err < 0) {
2031                         tls_err_abort(sk, -EBADMSG);
2032                         goto splice_read_end;
2033                 }
2034         }
2035
2036         /* splice does not support reading control messages */
2037         if (ctx->control != TLS_RECORD_TYPE_DATA) {
2038                 err = -EINVAL;
2039                 goto splice_read_end;
2040         }
2041
2042         rxm = strp_msg(skb);
2043
2044         chunk = min_t(unsigned int, rxm->full_len, len);
2045         copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
2046         if (copied < 0)
2047                 goto splice_read_end;
2048
2049         if (!from_queue) {
2050                 ctx->recv_pkt = NULL;
2051                 __strp_unpause(&ctx->strp);
2052         }
2053         if (chunk < rxm->full_len) {
2054                 __skb_queue_head(&ctx->rx_list, skb);
2055                 rxm->offset += len;
2056                 rxm->full_len -= len;
2057         } else {
2058                 consume_skb(skb);
2059         }
2060
2061 splice_read_end:
2062         release_sock(sk);
2063         sk_defer_free_flush(sk);
2064         return copied ? : err;
2065 }
2066
2067 bool tls_sw_sock_is_readable(struct sock *sk)
2068 {
2069         struct tls_context *tls_ctx = tls_get_ctx(sk);
2070         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2071         bool ingress_empty = true;
2072         struct sk_psock *psock;
2073
2074         rcu_read_lock();
2075         psock = sk_psock(sk);
2076         if (psock)
2077                 ingress_empty = list_empty(&psock->ingress_msg);
2078         rcu_read_unlock();
2079
2080         return !ingress_empty || ctx->recv_pkt ||
2081                 !skb_queue_empty(&ctx->rx_list);
2082 }
2083
2084 static int tls_read_size(struct strparser *strp, struct sk_buff *skb)
2085 {
2086         struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2087         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2088         struct tls_prot_info *prot = &tls_ctx->prot_info;
2089         char header[TLS_HEADER_SIZE + MAX_IV_SIZE];
2090         struct strp_msg *rxm = strp_msg(skb);
2091         size_t cipher_overhead;
2092         size_t data_len = 0;
2093         int ret;
2094
2095         /* Verify that we have a full TLS header, or wait for more data */
2096         if (rxm->offset + prot->prepend_size > skb->len)
2097                 return 0;
2098
2099         /* Sanity-check size of on-stack buffer. */
2100         if (WARN_ON(prot->prepend_size > sizeof(header))) {
2101                 ret = -EINVAL;
2102                 goto read_failure;
2103         }
2104
2105         /* Linearize header to local buffer */
2106         ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size);
2107
2108         if (ret < 0)
2109                 goto read_failure;
2110
2111         ctx->control = header[0];
2112
2113         data_len = ((header[4] & 0xFF) | (header[3] << 8));
2114
2115         cipher_overhead = prot->tag_size;
2116         if (prot->version != TLS_1_3_VERSION &&
2117             prot->cipher_type != TLS_CIPHER_CHACHA20_POLY1305)
2118                 cipher_overhead += prot->iv_size;
2119
2120         if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
2121             prot->tail_size) {
2122                 ret = -EMSGSIZE;
2123                 goto read_failure;
2124         }
2125         if (data_len < cipher_overhead) {
2126                 ret = -EBADMSG;
2127                 goto read_failure;
2128         }
2129
2130         /* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2131         if (header[1] != TLS_1_2_VERSION_MINOR ||
2132             header[2] != TLS_1_2_VERSION_MAJOR) {
2133                 ret = -EINVAL;
2134                 goto read_failure;
2135         }
2136
2137         tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE,
2138                                      TCP_SKB_CB(skb)->seq + rxm->offset);
2139         return data_len + TLS_HEADER_SIZE;
2140
2141 read_failure:
2142         tls_err_abort(strp->sk, ret);
2143
2144         return ret;
2145 }
2146
2147 static void tls_queue(struct strparser *strp, struct sk_buff *skb)
2148 {
2149         struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2150         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2151
2152         ctx->decrypted = 0;
2153
2154         ctx->recv_pkt = skb;
2155         strp_pause(strp);
2156
2157         ctx->saved_data_ready(strp->sk);
2158 }
2159
2160 static void tls_data_ready(struct sock *sk)
2161 {
2162         struct tls_context *tls_ctx = tls_get_ctx(sk);
2163         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2164         struct sk_psock *psock;
2165
2166         strp_data_ready(&ctx->strp);
2167
2168         psock = sk_psock_get(sk);
2169         if (psock) {
2170                 if (!list_empty(&psock->ingress_msg))
2171                         ctx->saved_data_ready(sk);
2172                 sk_psock_put(sk, psock);
2173         }
2174 }
2175
2176 void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
2177 {
2178         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2179
2180         set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask);
2181         set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask);
2182         cancel_delayed_work_sync(&ctx->tx_work.work);
2183 }
2184
2185 void tls_sw_release_resources_tx(struct sock *sk)
2186 {
2187         struct tls_context *tls_ctx = tls_get_ctx(sk);
2188         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2189         struct tls_rec *rec, *tmp;
2190         int pending;
2191
2192         /* Wait for any pending async encryptions to complete */
2193         spin_lock_bh(&ctx->encrypt_compl_lock);
2194         ctx->async_notify = true;
2195         pending = atomic_read(&ctx->encrypt_pending);
2196         spin_unlock_bh(&ctx->encrypt_compl_lock);
2197
2198         if (pending)
2199                 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
2200
2201         tls_tx_records(sk, -1);
2202
2203         /* Free up un-sent records in tx_list. First, free
2204          * the partially sent record if any at head of tx_list.
2205          */
2206         if (tls_ctx->partially_sent_record) {
2207                 tls_free_partial_record(sk, tls_ctx);
2208                 rec = list_first_entry(&ctx->tx_list,
2209                                        struct tls_rec, list);
2210                 list_del(&rec->list);
2211                 sk_msg_free(sk, &rec->msg_plaintext);
2212                 kfree(rec);
2213         }
2214
2215         list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
2216                 list_del(&rec->list);
2217                 sk_msg_free(sk, &rec->msg_encrypted);
2218                 sk_msg_free(sk, &rec->msg_plaintext);
2219                 kfree(rec);
2220         }
2221
2222         crypto_free_aead(ctx->aead_send);
2223         tls_free_open_rec(sk);
2224 }
2225
2226 void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
2227 {
2228         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2229
2230         kfree(ctx);
2231 }
2232
2233 void tls_sw_release_resources_rx(struct sock *sk)
2234 {
2235         struct tls_context *tls_ctx = tls_get_ctx(sk);
2236         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2237
2238         kfree(tls_ctx->rx.rec_seq);
2239         kfree(tls_ctx->rx.iv);
2240
2241         if (ctx->aead_recv) {
2242                 kfree_skb(ctx->recv_pkt);
2243                 ctx->recv_pkt = NULL;
2244                 skb_queue_purge(&ctx->rx_list);
2245                 crypto_free_aead(ctx->aead_recv);
2246                 strp_stop(&ctx->strp);
2247                 /* If tls_sw_strparser_arm() was not called (cleanup paths)
2248                  * we still want to strp_stop(), but sk->sk_data_ready was
2249                  * never swapped.
2250                  */
2251                 if (ctx->saved_data_ready) {
2252                         write_lock_bh(&sk->sk_callback_lock);
2253                         sk->sk_data_ready = ctx->saved_data_ready;
2254                         write_unlock_bh(&sk->sk_callback_lock);
2255                 }
2256         }
2257 }
2258
2259 void tls_sw_strparser_done(struct tls_context *tls_ctx)
2260 {
2261         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2262
2263         strp_done(&ctx->strp);
2264 }
2265
2266 void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
2267 {
2268         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2269
2270         kfree(ctx);
2271 }
2272
2273 void tls_sw_free_resources_rx(struct sock *sk)
2274 {
2275         struct tls_context *tls_ctx = tls_get_ctx(sk);
2276
2277         tls_sw_release_resources_rx(sk);
2278         tls_sw_free_ctx_rx(tls_ctx);
2279 }
2280
2281 /* The work handler to transmitt the encrypted records in tx_list */
2282 static void tx_work_handler(struct work_struct *work)
2283 {
2284         struct delayed_work *delayed_work = to_delayed_work(work);
2285         struct tx_work *tx_work = container_of(delayed_work,
2286                                                struct tx_work, work);
2287         struct sock *sk = tx_work->sk;
2288         struct tls_context *tls_ctx = tls_get_ctx(sk);
2289         struct tls_sw_context_tx *ctx;
2290
2291         if (unlikely(!tls_ctx))
2292                 return;
2293
2294         ctx = tls_sw_ctx_tx(tls_ctx);
2295         if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
2296                 return;
2297
2298         if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
2299                 return;
2300         mutex_lock(&tls_ctx->tx_lock);
2301         lock_sock(sk);
2302         tls_tx_records(sk, -1);
2303         release_sock(sk);
2304         mutex_unlock(&tls_ctx->tx_lock);
2305 }
2306
2307 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
2308 {
2309         struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
2310
2311         /* Schedule the transmission if tx list is ready */
2312         if (is_tx_ready(tx_ctx) &&
2313             !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
2314                 schedule_delayed_work(&tx_ctx->tx_work.work, 0);
2315 }
2316
2317 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
2318 {
2319         struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2320
2321         write_lock_bh(&sk->sk_callback_lock);
2322         rx_ctx->saved_data_ready = sk->sk_data_ready;
2323         sk->sk_data_ready = tls_data_ready;
2324         write_unlock_bh(&sk->sk_callback_lock);
2325
2326         strp_check_rcv(&rx_ctx->strp);
2327 }
2328
2329 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
2330 {
2331         struct tls_context *tls_ctx = tls_get_ctx(sk);
2332         struct tls_prot_info *prot = &tls_ctx->prot_info;
2333         struct tls_crypto_info *crypto_info;
2334         struct tls_sw_context_tx *sw_ctx_tx = NULL;
2335         struct tls_sw_context_rx *sw_ctx_rx = NULL;
2336         struct cipher_context *cctx;
2337         struct crypto_aead **aead;
2338         struct strp_callbacks cb;
2339         u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
2340         struct crypto_tfm *tfm;
2341         char *iv, *rec_seq, *key, *salt, *cipher_name;
2342         size_t keysize;
2343         int rc = 0;
2344
2345         if (!ctx) {
2346                 rc = -EINVAL;
2347                 goto out;
2348         }
2349
2350         if (tx) {
2351                 if (!ctx->priv_ctx_tx) {
2352                         sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
2353                         if (!sw_ctx_tx) {
2354                                 rc = -ENOMEM;
2355                                 goto out;
2356                         }
2357                         ctx->priv_ctx_tx = sw_ctx_tx;
2358                 } else {
2359                         sw_ctx_tx =
2360                                 (struct tls_sw_context_tx *)ctx->priv_ctx_tx;
2361                 }
2362         } else {
2363                 if (!ctx->priv_ctx_rx) {
2364                         sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
2365                         if (!sw_ctx_rx) {
2366                                 rc = -ENOMEM;
2367                                 goto out;
2368                         }
2369                         ctx->priv_ctx_rx = sw_ctx_rx;
2370                 } else {
2371                         sw_ctx_rx =
2372                                 (struct tls_sw_context_rx *)ctx->priv_ctx_rx;
2373                 }
2374         }
2375
2376         if (tx) {
2377                 crypto_init_wait(&sw_ctx_tx->async_wait);
2378                 spin_lock_init(&sw_ctx_tx->encrypt_compl_lock);
2379                 crypto_info = &ctx->crypto_send.info;
2380                 cctx = &ctx->tx;
2381                 aead = &sw_ctx_tx->aead_send;
2382                 INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
2383                 INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2384                 sw_ctx_tx->tx_work.sk = sk;
2385         } else {
2386                 crypto_init_wait(&sw_ctx_rx->async_wait);
2387                 spin_lock_init(&sw_ctx_rx->decrypt_compl_lock);
2388                 crypto_info = &ctx->crypto_recv.info;
2389                 cctx = &ctx->rx;
2390                 skb_queue_head_init(&sw_ctx_rx->rx_list);
2391                 aead = &sw_ctx_rx->aead_recv;
2392         }
2393
2394         switch (crypto_info->cipher_type) {
2395         case TLS_CIPHER_AES_GCM_128: {
2396                 struct tls12_crypto_info_aes_gcm_128 *gcm_128_info;
2397
2398                 gcm_128_info = (void *)crypto_info;
2399                 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2400                 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
2401                 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2402                 iv = gcm_128_info->iv;
2403                 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
2404                 rec_seq = gcm_128_info->rec_seq;
2405                 keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE;
2406                 key = gcm_128_info->key;
2407                 salt = gcm_128_info->salt;
2408                 salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
2409                 cipher_name = "gcm(aes)";
2410                 break;
2411         }
2412         case TLS_CIPHER_AES_GCM_256: {
2413                 struct tls12_crypto_info_aes_gcm_256 *gcm_256_info;
2414
2415                 gcm_256_info = (void *)crypto_info;
2416                 nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2417                 tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE;
2418                 iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2419                 iv = gcm_256_info->iv;
2420                 rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE;
2421                 rec_seq = gcm_256_info->rec_seq;
2422                 keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE;
2423                 key = gcm_256_info->key;
2424                 salt = gcm_256_info->salt;
2425                 salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE;
2426                 cipher_name = "gcm(aes)";
2427                 break;
2428         }
2429         case TLS_CIPHER_AES_CCM_128: {
2430                 struct tls12_crypto_info_aes_ccm_128 *ccm_128_info;
2431
2432                 ccm_128_info = (void *)crypto_info;
2433                 nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2434                 tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE;
2435                 iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2436                 iv = ccm_128_info->iv;
2437                 rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE;
2438                 rec_seq = ccm_128_info->rec_seq;
2439                 keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE;
2440                 key = ccm_128_info->key;
2441                 salt = ccm_128_info->salt;
2442                 salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE;
2443                 cipher_name = "ccm(aes)";
2444                 break;
2445         }
2446         case TLS_CIPHER_CHACHA20_POLY1305: {
2447                 struct tls12_crypto_info_chacha20_poly1305 *chacha20_poly1305_info;
2448
2449                 chacha20_poly1305_info = (void *)crypto_info;
2450                 nonce_size = 0;
2451                 tag_size = TLS_CIPHER_CHACHA20_POLY1305_TAG_SIZE;
2452                 iv_size = TLS_CIPHER_CHACHA20_POLY1305_IV_SIZE;
2453                 iv = chacha20_poly1305_info->iv;
2454                 rec_seq_size = TLS_CIPHER_CHACHA20_POLY1305_REC_SEQ_SIZE;
2455                 rec_seq = chacha20_poly1305_info->rec_seq;
2456                 keysize = TLS_CIPHER_CHACHA20_POLY1305_KEY_SIZE;
2457                 key = chacha20_poly1305_info->key;
2458                 salt = chacha20_poly1305_info->salt;
2459                 salt_size = TLS_CIPHER_CHACHA20_POLY1305_SALT_SIZE;
2460                 cipher_name = "rfc7539(chacha20,poly1305)";
2461                 break;
2462         }
2463         case TLS_CIPHER_SM4_GCM: {
2464                 struct tls12_crypto_info_sm4_gcm *sm4_gcm_info;
2465
2466                 sm4_gcm_info = (void *)crypto_info;
2467                 nonce_size = TLS_CIPHER_SM4_GCM_IV_SIZE;
2468                 tag_size = TLS_CIPHER_SM4_GCM_TAG_SIZE;
2469                 iv_size = TLS_CIPHER_SM4_GCM_IV_SIZE;
2470                 iv = sm4_gcm_info->iv;
2471                 rec_seq_size = TLS_CIPHER_SM4_GCM_REC_SEQ_SIZE;
2472                 rec_seq = sm4_gcm_info->rec_seq;
2473                 keysize = TLS_CIPHER_SM4_GCM_KEY_SIZE;
2474                 key = sm4_gcm_info->key;
2475                 salt = sm4_gcm_info->salt;
2476                 salt_size = TLS_CIPHER_SM4_GCM_SALT_SIZE;
2477                 cipher_name = "gcm(sm4)";
2478                 break;
2479         }
2480         case TLS_CIPHER_SM4_CCM: {
2481                 struct tls12_crypto_info_sm4_ccm *sm4_ccm_info;
2482
2483                 sm4_ccm_info = (void *)crypto_info;
2484                 nonce_size = TLS_CIPHER_SM4_CCM_IV_SIZE;
2485                 tag_size = TLS_CIPHER_SM4_CCM_TAG_SIZE;
2486                 iv_size = TLS_CIPHER_SM4_CCM_IV_SIZE;
2487                 iv = sm4_ccm_info->iv;
2488                 rec_seq_size = TLS_CIPHER_SM4_CCM_REC_SEQ_SIZE;
2489                 rec_seq = sm4_ccm_info->rec_seq;
2490                 keysize = TLS_CIPHER_SM4_CCM_KEY_SIZE;
2491                 key = sm4_ccm_info->key;
2492                 salt = sm4_ccm_info->salt;
2493                 salt_size = TLS_CIPHER_SM4_CCM_SALT_SIZE;
2494                 cipher_name = "ccm(sm4)";
2495                 break;
2496         }
2497         default:
2498                 rc = -EINVAL;
2499                 goto free_priv;
2500         }
2501
2502         /* Sanity-check the sizes for stack allocations. */
2503         if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE ||
2504             rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
2505                 rc = -EINVAL;
2506                 goto free_priv;
2507         }
2508
2509         if (crypto_info->version == TLS_1_3_VERSION) {
2510                 nonce_size = 0;
2511                 prot->aad_size = TLS_HEADER_SIZE;
2512                 prot->tail_size = 1;
2513         } else {
2514                 prot->aad_size = TLS_AAD_SPACE_SIZE;
2515                 prot->tail_size = 0;
2516         }
2517
2518         prot->version = crypto_info->version;
2519         prot->cipher_type = crypto_info->cipher_type;
2520         prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
2521         prot->tag_size = tag_size;
2522         prot->overhead_size = prot->prepend_size +
2523                               prot->tag_size + prot->tail_size;
2524         prot->iv_size = iv_size;
2525         prot->salt_size = salt_size;
2526         cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL);
2527         if (!cctx->iv) {
2528                 rc = -ENOMEM;
2529                 goto free_priv;
2530         }
2531         /* Note: 128 & 256 bit salt are the same size */
2532         prot->rec_seq_size = rec_seq_size;
2533         memcpy(cctx->iv, salt, salt_size);
2534         memcpy(cctx->iv + salt_size, iv, iv_size);
2535         cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
2536         if (!cctx->rec_seq) {
2537                 rc = -ENOMEM;
2538                 goto free_iv;
2539         }
2540
2541         if (!*aead) {
2542                 *aead = crypto_alloc_aead(cipher_name, 0, 0);
2543                 if (IS_ERR(*aead)) {
2544                         rc = PTR_ERR(*aead);
2545                         *aead = NULL;
2546                         goto free_rec_seq;
2547                 }
2548         }
2549
2550         ctx->push_pending_record = tls_sw_push_pending_record;
2551
2552         rc = crypto_aead_setkey(*aead, key, keysize);
2553
2554         if (rc)
2555                 goto free_aead;
2556
2557         rc = crypto_aead_setauthsize(*aead, prot->tag_size);
2558         if (rc)
2559                 goto free_aead;
2560
2561         if (sw_ctx_rx) {
2562                 tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
2563
2564                 if (crypto_info->version == TLS_1_3_VERSION)
2565                         sw_ctx_rx->async_capable = 0;
2566                 else
2567                         sw_ctx_rx->async_capable =
2568                                 !!(tfm->__crt_alg->cra_flags &
2569                                    CRYPTO_ALG_ASYNC);
2570
2571                 /* Set up strparser */
2572                 memset(&cb, 0, sizeof(cb));
2573                 cb.rcv_msg = tls_queue;
2574                 cb.parse_msg = tls_read_size;
2575
2576                 strp_init(&sw_ctx_rx->strp, sk, &cb);
2577         }
2578
2579         goto out;
2580
2581 free_aead:
2582         crypto_free_aead(*aead);
2583         *aead = NULL;
2584 free_rec_seq:
2585         kfree(cctx->rec_seq);
2586         cctx->rec_seq = NULL;
2587 free_iv:
2588         kfree(cctx->iv);
2589         cctx->iv = NULL;
2590 free_priv:
2591         if (tx) {
2592                 kfree(ctx->priv_ctx_tx);
2593                 ctx->priv_ctx_tx = NULL;
2594         } else {
2595                 kfree(ctx->priv_ctx_rx);
2596                 ctx->priv_ctx_rx = NULL;
2597         }
2598 out:
2599         return rc;
2600 }