net/tls/tls_sw.c

   1 /*
   2  * Copyright (c) 2016-2017, Mellanox Technologies. All rights reserved.
   3  * Copyright (c) 2016-2017, Dave Watson <davejwatson@fb.com>. All rights reserved.
   4  * Copyright (c) 2016-2017, Lance Chao <lancerchao@fb.com>. All rights reserved.
   5  * Copyright (c) 2016, Fridolin Pokorny <fridolin.pokorny@gmail.com>. All rights reserved.
   6  * Copyright (c) 2016, Nikos Mavrogiannopoulos <nmav@gnutls.org>. All rights reserved.
   7  * Copyright (c) 2018, Covalent IO, Inc. http://covalent.io
   8  *
   9  * This software is available to you under a choice of one of two
  10  * licenses.  You may choose to be licensed under the terms of the GNU
  11  * General Public License (GPL) Version 2, available from the file
  12  * COPYING in the main directory of this source tree, or the
  13  * OpenIB.org BSD license below:
  14  *
  15  *     Redistribution and use in source and binary forms, with or
  16  *     without modification, are permitted provided that the following
  17  *     conditions are met:
  18  *
  19  *      - Redistributions of source code must retain the above
  20  *        copyright notice, this list of conditions and the following
  21  *        disclaimer.
  22  *
  23  *      - Redistributions in binary form must reproduce the above
  24  *        copyright notice, this list of conditions and the following
  25  *        disclaimer in the documentation and/or other materials
  26  *        provided with the distribution.
  27  *
  28  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  29  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  30  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  31  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  32  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  33  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  34  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  35  * SOFTWARE.
  36  */
  37
  38 #include <linux/bug.h>
  39 #include <linux/sched/signal.h>
  40 #include <linux/module.h>
  41 #include <linux/splice.h>
  42 #include <crypto/aead.h>
  43
  44 #include <net/strparser.h>
  45 #include <net/tls.h>
  46
  47 #include "tls.h"
  48
  49 struct tls_decrypt_arg {
  50         struct_group(inargs,
  51         bool zc;
  52         bool async;
  53         u8 tail;
  54         );
  55
  56         struct sk_buff *skb;
  57 };
  58
  59 struct tls_decrypt_ctx {
  60         u8 iv[MAX_IV_SIZE];
  61         u8 aad[TLS_MAX_AAD_SIZE];
  62         u8 tail;
  63         struct scatterlist sg[];
  64 };
  65
  66 noinline void tls_err_abort(struct sock *sk, int err)
  67 {
  68         WARN_ON_ONCE(err >= 0);
  69         /* sk->sk_err should contain a positive error code. */
  70         sk->sk_err = -err;
  71         sk_error_report(sk);
  72 }
  73
  74 static int __skb_nsg(struct sk_buff *skb, int offset, int len,
  75                      unsigned int recursion_level)
  76 {
  77         int start = skb_headlen(skb);
  78         int i, chunk = start - offset;
  79         struct sk_buff *frag_iter;
  80         int elt = 0;
  81
  82         if (unlikely(recursion_level >= 24))
  83                 return -EMSGSIZE;
  84
  85         if (chunk > 0) {
  86                 if (chunk > len)
  87                         chunk = len;
  88                 elt++;
  89                 len -= chunk;
  90                 if (len == 0)
  91                         return elt;
  92                 offset += chunk;
  93         }
  94
  95         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
  96                 int end;
  97
  98                 WARN_ON(start > offset + len);
  99
 100                 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
 101                 chunk = end - offset;
 102                 if (chunk > 0) {
 103                         if (chunk > len)
 104                                 chunk = len;
 105                         elt++;
 106                         len -= chunk;
 107                         if (len == 0)
 108                                 return elt;
 109                         offset += chunk;
 110                 }
 111                 start = end;
 112         }
 113
 114         if (unlikely(skb_has_frag_list(skb))) {
 115                 skb_walk_frags(skb, frag_iter) {
 116                         int end, ret;
 117
 118                         WARN_ON(start > offset + len);
 119
 120                         end = start + frag_iter->len;
 121                         chunk = end - offset;
 122                         if (chunk > 0) {
 123                                 if (chunk > len)
 124                                         chunk = len;
 125                                 ret = __skb_nsg(frag_iter, offset - start, chunk,
 126                                                 recursion_level + 1);
 127                                 if (unlikely(ret < 0))
 128                                         return ret;
 129                                 elt += ret;
 130                                 len -= chunk;
 131                                 if (len == 0)
 132                                         return elt;
 133                                 offset += chunk;
 134                         }
 135                         start = end;
 136                 }
 137         }
 138         BUG_ON(len);
 139         return elt;
 140 }
 141
 142 /* Return the number of scatterlist elements required to completely map the
 143  * skb, or -EMSGSIZE if the recursion depth is exceeded.
 144  */
 145 static int skb_nsg(struct sk_buff *skb, int offset, int len)
 146 {
 147         return __skb_nsg(skb, offset, len, 0);
 148 }
 149
 150 static int tls_padding_length(struct tls_prot_info *prot, struct sk_buff *skb,
 151                               struct tls_decrypt_arg *darg)
 152 {
 153         struct strp_msg *rxm = strp_msg(skb);
 154         struct tls_msg *tlm = tls_msg(skb);
 155         int sub = 0;
 156
 157         /* Determine zero-padding length */
 158         if (prot->version == TLS_1_3_VERSION) {
 159                 int offset = rxm->full_len - TLS_TAG_SIZE - 1;
 160                 char content_type = darg->zc ? darg->tail : 0;
 161                 int err;
 162
 163                 while (content_type == 0) {
 164                         if (offset < prot->prepend_size)
 165                                 return -EBADMSG;
 166                         err = skb_copy_bits(skb, rxm->offset + offset,
 167                                             &content_type, 1);
 168                         if (err)
 169                                 return err;
 170                         if (content_type)
 171                                 break;
 172                         sub++;
 173                         offset--;
 174                 }
 175                 tlm->control = content_type;
 176         }
 177         return sub;
 178 }
 179
 180 static void tls_decrypt_done(struct crypto_async_request *req, int err)
 181 {
 182         struct aead_request *aead_req = (struct aead_request *)req;
 183         struct scatterlist *sgout = aead_req->dst;
 184         struct scatterlist *sgin = aead_req->src;
 185         struct tls_sw_context_rx *ctx;
 186         struct tls_context *tls_ctx;
 187         struct scatterlist *sg;
 188         unsigned int pages;
 189         struct sock *sk;
 190
 191         sk = (struct sock *)req->data;
 192         tls_ctx = tls_get_ctx(sk);
 193         ctx = tls_sw_ctx_rx(tls_ctx);
 194
 195         /* Propagate if there was an err */
 196         if (err) {
 197                 if (err == -EBADMSG)
 198                         TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTERROR);
 199                 ctx->async_wait.err = err;
 200                 tls_err_abort(sk, err);
 201         }
 202
 203         /* Free the destination pages if skb was not decrypted inplace */
 204         if (sgout != sgin) {
 205                 /* Skip the first S/G entry as it points to AAD */
 206                 for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) {
 207                         if (!sg)
 208                                 break;
 209                         put_page(sg_page(sg));
 210                 }
 211         }
 212
 213         kfree(aead_req);
 214
 215         spin_lock_bh(&ctx->decrypt_compl_lock);
 216         if (!atomic_dec_return(&ctx->decrypt_pending))
 217                 complete(&ctx->async_wait.completion);
 218         spin_unlock_bh(&ctx->decrypt_compl_lock);
 219 }
 220
 221 static int tls_do_decryption(struct sock *sk,
 222                              struct scatterlist *sgin,
 223                              struct scatterlist *sgout,
 224                              char *iv_recv,
 225                              size_t data_len,
 226                              struct aead_request *aead_req,
 227                              struct tls_decrypt_arg *darg)
 228 {
 229         struct tls_context *tls_ctx = tls_get_ctx(sk);
 230         struct tls_prot_info *prot = &tls_ctx->prot_info;
 231         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
 232         int ret;
 233
 234         aead_request_set_tfm(aead_req, ctx->aead_recv);
 235         aead_request_set_ad(aead_req, prot->aad_size);
 236         aead_request_set_crypt(aead_req, sgin, sgout,
 237                                data_len + prot->tag_size,
 238                                (u8 *)iv_recv);
 239
 240         if (darg->async) {
 241                 aead_request_set_callback(aead_req,
 242                                           CRYPTO_TFM_REQ_MAY_BACKLOG,
 243                                           tls_decrypt_done, sk);
 244                 atomic_inc(&ctx->decrypt_pending);
 245         } else {
 246                 aead_request_set_callback(aead_req,
 247                                           CRYPTO_TFM_REQ_MAY_BACKLOG,
 248                                           crypto_req_done, &ctx->async_wait);
 249         }
 250
 251         ret = crypto_aead_decrypt(aead_req);
 252         if (ret == -EINPROGRESS) {
 253                 if (darg->async)
 254                         return 0;
 255
 256                 ret = crypto_wait_req(ret, &ctx->async_wait);
 257         }
 258         darg->async = false;
 259
 260         return ret;
 261 }
 262
 263 static void tls_trim_both_msgs(struct sock *sk, int target_size)
 264 {
 265         struct tls_context *tls_ctx = tls_get_ctx(sk);
 266         struct tls_prot_info *prot = &tls_ctx->prot_info;
 267         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 268         struct tls_rec *rec = ctx->open_rec;
 269
 270         sk_msg_trim(sk, &rec->msg_plaintext, target_size);
 271         if (target_size > 0)
 272                 target_size += prot->overhead_size;
 273         sk_msg_trim(sk, &rec->msg_encrypted, target_size);
 274 }
 275
 276 static int tls_alloc_encrypted_msg(struct sock *sk, int len)
 277 {
 278         struct tls_context *tls_ctx = tls_get_ctx(sk);
 279         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 280         struct tls_rec *rec = ctx->open_rec;
 281         struct sk_msg *msg_en = &rec->msg_encrypted;
 282
 283         return sk_msg_alloc(sk, msg_en, len, 0);
 284 }
 285
 286 static int tls_clone_plaintext_msg(struct sock *sk, int required)
 287 {
 288         struct tls_context *tls_ctx = tls_get_ctx(sk);
 289         struct tls_prot_info *prot = &tls_ctx->prot_info;
 290         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 291         struct tls_rec *rec = ctx->open_rec;
 292         struct sk_msg *msg_pl = &rec->msg_plaintext;
 293         struct sk_msg *msg_en = &rec->msg_encrypted;
 294         int skip, len;
 295
 296         /* We add page references worth len bytes from encrypted sg
 297          * at the end of plaintext sg. It is guaranteed that msg_en
 298          * has enough required room (ensured by caller).
 299          */
 300         len = required - msg_pl->sg.size;
 301
 302         /* Skip initial bytes in msg_en's data to be able to use
 303          * same offset of both plain and encrypted data.
 304          */
 305         skip = prot->prepend_size + msg_pl->sg.size;
 306
 307         return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
 308 }
 309
 310 static struct tls_rec *tls_get_rec(struct sock *sk)
 311 {
 312         struct tls_context *tls_ctx = tls_get_ctx(sk);
 313         struct tls_prot_info *prot = &tls_ctx->prot_info;
 314         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 315         struct sk_msg *msg_pl, *msg_en;
 316         struct tls_rec *rec;
 317         int mem_size;
 318
 319         mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
 320
 321         rec = kzalloc(mem_size, sk->sk_allocation);
 322         if (!rec)
 323                 return NULL;
 324
 325         msg_pl = &rec->msg_plaintext;
 326         msg_en = &rec->msg_encrypted;
 327
 328         sk_msg_init(msg_pl);
 329         sk_msg_init(msg_en);
 330
 331         sg_init_table(rec->sg_aead_in, 2);
 332         sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size);
 333         sg_unmark_end(&rec->sg_aead_in[1]);
 334
 335         sg_init_table(rec->sg_aead_out, 2);
 336         sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size);
 337         sg_unmark_end(&rec->sg_aead_out[1]);
 338
 339         return rec;
 340 }
 341
 342 static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
 343 {
 344         sk_msg_free(sk, &rec->msg_encrypted);
 345         sk_msg_free(sk, &rec->msg_plaintext);
 346         kfree(rec);
 347 }
 348
 349 static void tls_free_open_rec(struct sock *sk)
 350 {
 351         struct tls_context *tls_ctx = tls_get_ctx(sk);
 352         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 353         struct tls_rec *rec = ctx->open_rec;
 354
 355         if (rec) {
 356                 tls_free_rec(sk, rec);
 357                 ctx->open_rec = NULL;
 358         }
 359 }
 360
 361 int tls_tx_records(struct sock *sk, int flags)
 362 {
 363         struct tls_context *tls_ctx = tls_get_ctx(sk);
 364         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 365         struct tls_rec *rec, *tmp;
 366         struct sk_msg *msg_en;
 367         int tx_flags, rc = 0;
 368
 369         if (tls_is_partially_sent_record(tls_ctx)) {
 370                 rec = list_first_entry(&ctx->tx_list,
 371                                        struct tls_rec, list);
 372
 373                 if (flags == -1)
 374                         tx_flags = rec->tx_flags;
 375                 else
 376                         tx_flags = flags;
 377
 378                 rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
 379                 if (rc)
 380                         goto tx_err;
 381
 382                 /* Full record has been transmitted.
 383                  * Remove the head of tx_list
 384                  */
 385                 list_del(&rec->list);
 386                 sk_msg_free(sk, &rec->msg_plaintext);
 387                 kfree(rec);
 388         }
 389
 390         /* Tx all ready records */
 391         list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
 392                 if (READ_ONCE(rec->tx_ready)) {
 393                         if (flags == -1)
 394                                 tx_flags = rec->tx_flags;
 395                         else
 396                                 tx_flags = flags;
 397
 398                         msg_en = &rec->msg_encrypted;
 399                         rc = tls_push_sg(sk, tls_ctx,
 400                                          &msg_en->sg.data[msg_en->sg.curr],
 401                                          0, tx_flags);
 402                         if (rc)
 403                                 goto tx_err;
 404
 405                         list_del(&rec->list);
 406                         sk_msg_free(sk, &rec->msg_plaintext);
 407                         kfree(rec);
 408                 } else {
 409                         break;
 410                 }
 411         }
 412
 413 tx_err:
 414         if (rc < 0 && rc != -EAGAIN)
 415                 tls_err_abort(sk, -EBADMSG);
 416
 417         return rc;
 418 }
 419
 420 static void tls_encrypt_done(struct crypto_async_request *req, int err)
 421 {
 422         struct aead_request *aead_req = (struct aead_request *)req;
 423         struct sock *sk = req->data;
 424         struct tls_context *tls_ctx = tls_get_ctx(sk);
 425         struct tls_prot_info *prot = &tls_ctx->prot_info;
 426         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 427         struct scatterlist *sge;
 428         struct sk_msg *msg_en;
 429         struct tls_rec *rec;
 430         bool ready = false;
 431         int pending;
 432
 433         rec = container_of(aead_req, struct tls_rec, aead_req);
 434         msg_en = &rec->msg_encrypted;
 435
 436         sge = sk_msg_elem(msg_en, msg_en->sg.curr);
 437         sge->offset -= prot->prepend_size;
 438         sge->length += prot->prepend_size;
 439
 440         /* Check if error is previously set on socket */
 441         if (err || sk->sk_err) {
 442                 rec = NULL;
 443
 444                 /* If err is already set on socket, return the same code */
 445                 if (sk->sk_err) {
 446                         ctx->async_wait.err = -sk->sk_err;
 447                 } else {
 448                         ctx->async_wait.err = err;
 449                         tls_err_abort(sk, err);
 450                 }
 451         }
 452
 453         if (rec) {
 454                 struct tls_rec *first_rec;
 455
 456                 /* Mark the record as ready for transmission */
 457                 smp_store_mb(rec->tx_ready, true);
 458
 459                 /* If received record is at head of tx_list, schedule tx */
 460                 first_rec = list_first_entry(&ctx->tx_list,
 461                                              struct tls_rec, list);
 462                 if (rec == first_rec)
 463                         ready = true;
 464         }
 465
 466         spin_lock_bh(&ctx->encrypt_compl_lock);
 467         pending = atomic_dec_return(&ctx->encrypt_pending);
 468
 469         if (!pending && ctx->async_notify)
 470                 complete(&ctx->async_wait.completion);
 471         spin_unlock_bh(&ctx->encrypt_compl_lock);
 472
 473         if (!ready)
 474                 return;
 475
 476         /* Schedule the transmission */
 477         if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
 478                 schedule_delayed_work(&ctx->tx_work.work, 1);
 479 }
 480
 481 static int tls_do_encryption(struct sock *sk,
 482                              struct tls_context *tls_ctx,
 483                              struct tls_sw_context_tx *ctx,
 484                              struct aead_request *aead_req,
 485                              size_t data_len, u32 start)
 486 {
 487         struct tls_prot_info *prot = &tls_ctx->prot_info;
 488         struct tls_rec *rec = ctx->open_rec;
 489         struct sk_msg *msg_en = &rec->msg_encrypted;
 490         struct scatterlist *sge = sk_msg_elem(msg_en, start);
 491         int rc, iv_offset = 0;
 492
 493         /* For CCM based ciphers, first byte of IV is a constant */
 494         switch (prot->cipher_type) {
 495         case TLS_CIPHER_AES_CCM_128:
 496                 rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
 497                 iv_offset = 1;
 498                 break;
 499         case TLS_CIPHER_SM4_CCM:
 500                 rec->iv_data[0] = TLS_SM4_CCM_IV_B0_BYTE;
 501                 iv_offset = 1;
 502                 break;
 503         }
 504
 505         memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
 506                prot->iv_size + prot->salt_size);
 507
 508         tls_xor_iv_with_seq(prot, rec->iv_data + iv_offset,
 509                             tls_ctx->tx.rec_seq);
 510
 511         sge->offset += prot->prepend_size;
 512         sge->length -= prot->prepend_size;
 513
 514         msg_en->sg.curr = start;
 515
 516         aead_request_set_tfm(aead_req, ctx->aead_send);
 517         aead_request_set_ad(aead_req, prot->aad_size);
 518         aead_request_set_crypt(aead_req, rec->sg_aead_in,
 519                                rec->sg_aead_out,
 520                                data_len, rec->iv_data);
 521
 522         aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 523                                   tls_encrypt_done, sk);
 524
 525         /* Add the record in tx_list */
 526         list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
 527         atomic_inc(&ctx->encrypt_pending);
 528
 529         rc = crypto_aead_encrypt(aead_req);
 530         if (!rc || rc != -EINPROGRESS) {
 531                 atomic_dec(&ctx->encrypt_pending);
 532                 sge->offset -= prot->prepend_size;
 533                 sge->length += prot->prepend_size;
 534         }
 535
 536         if (!rc) {
 537                 WRITE_ONCE(rec->tx_ready, true);
 538         } else if (rc != -EINPROGRESS) {
 539                 list_del(&rec->list);
 540                 return rc;
 541         }
 542
 543         /* Unhook the record from context if encryption is not failure */
 544         ctx->open_rec = NULL;
 545         tls_advance_record_sn(sk, prot, &tls_ctx->tx);
 546         return rc;
 547 }
 548
 549 static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
 550                                  struct tls_rec **to, struct sk_msg *msg_opl,
 551                                  struct sk_msg *msg_oen, u32 split_point,
 552                                  u32 tx_overhead_size, u32 *orig_end)
 553 {
 554         u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
 555         struct scatterlist *sge, *osge, *nsge;
 556         u32 orig_size = msg_opl->sg.size;
 557         struct scatterlist tmp = { };
 558         struct sk_msg *msg_npl;
 559         struct tls_rec *new;
 560         int ret;
 561
 562         new = tls_get_rec(sk);
 563         if (!new)
 564                 return -ENOMEM;
 565         ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size +
 566                            tx_overhead_size, 0);
 567         if (ret < 0) {
 568                 tls_free_rec(sk, new);
 569                 return ret;
 570         }
 571
 572         *orig_end = msg_opl->sg.end;
 573         i = msg_opl->sg.start;
 574         sge = sk_msg_elem(msg_opl, i);
 575         while (apply && sge->length) {
 576                 if (sge->length > apply) {
 577                         u32 len = sge->length - apply;
 578
 579                         get_page(sg_page(sge));
 580                         sg_set_page(&tmp, sg_page(sge), len,
 581                                     sge->offset + apply);
 582                         sge->length = apply;
 583                         bytes += apply;
 584                         apply = 0;
 585                 } else {
 586                         apply -= sge->length;
 587                         bytes += sge->length;
 588                 }
 589
 590                 sk_msg_iter_var_next(i);
 591                 if (i == msg_opl->sg.end)
 592                         break;
 593                 sge = sk_msg_elem(msg_opl, i);
 594         }
 595
 596         msg_opl->sg.end = i;
 597         msg_opl->sg.curr = i;
 598         msg_opl->sg.copybreak = 0;
 599         msg_opl->apply_bytes = 0;
 600         msg_opl->sg.size = bytes;
 601
 602         msg_npl = &new->msg_plaintext;
 603         msg_npl->apply_bytes = apply;
 604         msg_npl->sg.size = orig_size - bytes;
 605
 606         j = msg_npl->sg.start;
 607         nsge = sk_msg_elem(msg_npl, j);
 608         if (tmp.length) {
 609                 memcpy(nsge, &tmp, sizeof(*nsge));
 610                 sk_msg_iter_var_next(j);
 611                 nsge = sk_msg_elem(msg_npl, j);
 612         }
 613
 614         osge = sk_msg_elem(msg_opl, i);
 615         while (osge->length) {
 616                 memcpy(nsge, osge, sizeof(*nsge));
 617                 sg_unmark_end(nsge);
 618                 sk_msg_iter_var_next(i);
 619                 sk_msg_iter_var_next(j);
 620                 if (i == *orig_end)
 621                         break;
 622                 osge = sk_msg_elem(msg_opl, i);
 623                 nsge = sk_msg_elem(msg_npl, j);
 624         }
 625
 626         msg_npl->sg.end = j;
 627         msg_npl->sg.curr = j;
 628         msg_npl->sg.copybreak = 0;
 629
 630         *to = new;
 631         return 0;
 632 }
 633
 634 static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
 635                                   struct tls_rec *from, u32 orig_end)
 636 {
 637         struct sk_msg *msg_npl = &from->msg_plaintext;
 638         struct sk_msg *msg_opl = &to->msg_plaintext;
 639         struct scatterlist *osge, *nsge;
 640         u32 i, j;
 641
 642         i = msg_opl->sg.end;
 643         sk_msg_iter_var_prev(i);
 644         j = msg_npl->sg.start;
 645
 646         osge = sk_msg_elem(msg_opl, i);
 647         nsge = sk_msg_elem(msg_npl, j);
 648
 649         if (sg_page(osge) == sg_page(nsge) &&
 650             osge->offset + osge->length == nsge->offset) {
 651                 osge->length += nsge->length;
 652                 put_page(sg_page(nsge));
 653         }
 654
 655         msg_opl->sg.end = orig_end;
 656         msg_opl->sg.curr = orig_end;
 657         msg_opl->sg.copybreak = 0;
 658         msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
 659         msg_opl->sg.size += msg_npl->sg.size;
 660
 661         sk_msg_free(sk, &to->msg_encrypted);
 662         sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted);
 663
 664         kfree(from);
 665 }
 666
 667 static int tls_push_record(struct sock *sk, int flags,
 668                            unsigned char record_type)
 669 {
 670         struct tls_context *tls_ctx = tls_get_ctx(sk);
 671         struct tls_prot_info *prot = &tls_ctx->prot_info;
 672         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 673         struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
 674         u32 i, split_point, orig_end;
 675         struct sk_msg *msg_pl, *msg_en;
 676         struct aead_request *req;
 677         bool split;
 678         int rc;
 679
 680         if (!rec)
 681                 return 0;
 682
 683         msg_pl = &rec->msg_plaintext;
 684         msg_en = &rec->msg_encrypted;
 685
 686         split_point = msg_pl->apply_bytes;
 687         split = split_point && split_point < msg_pl->sg.size;
 688         if (unlikely((!split &&
 689                       msg_pl->sg.size +
 690                       prot->overhead_size > msg_en->sg.size) ||
 691                      (split &&
 692                       split_point +
 693                       prot->overhead_size > msg_en->sg.size))) {
 694                 split = true;
 695                 split_point = msg_en->sg.size;
 696         }
 697         if (split) {
 698                 rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en,
 699                                            split_point, prot->overhead_size,
 700                                            &orig_end);
 701                 if (rc < 0)
 702                         return rc;
 703                 /* This can happen if above tls_split_open_record allocates
 704                  * a single large encryption buffer instead of two smaller
 705                  * ones. In this case adjust pointers and continue without
 706                  * split.
 707                  */
 708                 if (!msg_pl->sg.size) {
 709                         tls_merge_open_record(sk, rec, tmp, orig_end);
 710                         msg_pl = &rec->msg_plaintext;
 711                         msg_en = &rec->msg_encrypted;
 712                         split = false;
 713                 }
 714                 sk_msg_trim(sk, msg_en, msg_pl->sg.size +
 715                             prot->overhead_size);
 716         }
 717
 718         rec->tx_flags = flags;
 719         req = &rec->aead_req;
 720
 721         i = msg_pl->sg.end;
 722         sk_msg_iter_var_prev(i);
 723
 724         rec->content_type = record_type;
 725         if (prot->version == TLS_1_3_VERSION) {
 726                 /* Add content type to end of message.  No padding added */
 727                 sg_set_buf(&rec->sg_content_type, &rec->content_type, 1);
 728                 sg_mark_end(&rec->sg_content_type);
 729                 sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1,
 730                          &rec->sg_content_type);
 731         } else {
 732                 sg_mark_end(sk_msg_elem(msg_pl, i));
 733         }
 734
 735         if (msg_pl->sg.end < msg_pl->sg.start) {
 736                 sg_chain(&msg_pl->sg.data[msg_pl->sg.start],
 737                          MAX_SKB_FRAGS - msg_pl->sg.start + 1,
 738                          msg_pl->sg.data);
 739         }
 740
 741         i = msg_pl->sg.start;
 742         sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]);
 743
 744         i = msg_en->sg.end;
 745         sk_msg_iter_var_prev(i);
 746         sg_mark_end(sk_msg_elem(msg_en, i));
 747
 748         i = msg_en->sg.start;
 749         sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
 750
 751         tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size,
 752                      tls_ctx->tx.rec_seq, record_type, prot);
 753
 754         tls_fill_prepend(tls_ctx,
 755                          page_address(sg_page(&msg_en->sg.data[i])) +
 756                          msg_en->sg.data[i].offset,
 757                          msg_pl->sg.size + prot->tail_size,
 758                          record_type);
 759
 760         tls_ctx->pending_open_record_frags = false;
 761
 762         rc = tls_do_encryption(sk, tls_ctx, ctx, req,
 763                                msg_pl->sg.size + prot->tail_size, i);
 764         if (rc < 0) {
 765                 if (rc != -EINPROGRESS) {
 766                         tls_err_abort(sk, -EBADMSG);
 767                         if (split) {
 768                                 tls_ctx->pending_open_record_frags = true;
 769                                 tls_merge_open_record(sk, rec, tmp, orig_end);
 770                         }
 771                 }
 772                 ctx->async_capable = 1;
 773                 return rc;
 774         } else if (split) {
 775                 msg_pl = &tmp->msg_plaintext;
 776                 msg_en = &tmp->msg_encrypted;
 777                 sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size);
 778                 tls_ctx->pending_open_record_frags = true;
 779                 ctx->open_rec = tmp;
 780         }
 781
 782         return tls_tx_records(sk, flags);
 783 }
 784
 785 static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk,
 786                                bool full_record, u8 record_type,
 787                                ssize_t *copied, int flags)
 788 {
 789         struct tls_context *tls_ctx = tls_get_ctx(sk);
 790         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 791         struct sk_msg msg_redir = { };
 792         struct sk_psock *psock;
 793         struct sock *sk_redir;
 794         struct tls_rec *rec;
 795         bool enospc, policy, redir_ingress;
 796         int err = 0, send;
 797         u32 delta = 0;
 798
 799         policy = !(flags & MSG_SENDPAGE_NOPOLICY);
 800         psock = sk_psock_get(sk);
 801         if (!psock || !policy) {
 802                 err = tls_push_record(sk, flags, record_type);
 803                 if (err && sk->sk_err == EBADMSG) {
 804                         *copied -= sk_msg_free(sk, msg);
 805                         tls_free_open_rec(sk);
 806                         err = -sk->sk_err;
 807                 }
 808                 if (psock)
 809                         sk_psock_put(sk, psock);
 810                 return err;
 811         }
 812 more_data:
 813         enospc = sk_msg_full(msg);
 814         if (psock->eval == __SK_NONE) {
 815                 delta = msg->sg.size;
 816                 psock->eval = sk_psock_msg_verdict(sk, psock, msg);
 817                 delta -= msg->sg.size;
 818         }
 819         if (msg->cork_bytes && msg->cork_bytes > msg->sg.size &&
 820             !enospc && !full_record) {
 821                 err = -ENOSPC;
 822                 goto out_err;
 823         }
 824         msg->cork_bytes = 0;
 825         send = msg->sg.size;
 826         if (msg->apply_bytes && msg->apply_bytes < send)
 827                 send = msg->apply_bytes;
 828
 829         switch (psock->eval) {
 830         case __SK_PASS:
 831                 err = tls_push_record(sk, flags, record_type);
 832                 if (err && sk->sk_err == EBADMSG) {
 833                         *copied -= sk_msg_free(sk, msg);
 834                         tls_free_open_rec(sk);
 835                         err = -sk->sk_err;
 836                         goto out_err;
 837                 }
 838                 break;
 839         case __SK_REDIRECT:
 840                 redir_ingress = psock->redir_ingress;
 841                 sk_redir = psock->sk_redir;
 842                 memcpy(&msg_redir, msg, sizeof(*msg));
 843                 if (msg->apply_bytes < send)
 844                         msg->apply_bytes = 0;
 845                 else
 846                         msg->apply_bytes -= send;
 847                 sk_msg_return_zero(sk, msg, send);
 848                 msg->sg.size -= send;
 849                 release_sock(sk);
 850                 err = tcp_bpf_sendmsg_redir(sk_redir, redir_ingress,
 851                                             &msg_redir, send, flags);
 852                 lock_sock(sk);
 853                 if (err < 0) {
 854                         *copied -= sk_msg_free_nocharge(sk, &msg_redir);
 855                         msg->sg.size = 0;
 856                 }
 857                 if (msg->sg.size == 0)
 858                         tls_free_open_rec(sk);
 859                 break;
 860         case __SK_DROP:
 861         default:
 862                 sk_msg_free_partial(sk, msg, send);
 863                 if (msg->apply_bytes < send)
 864                         msg->apply_bytes = 0;
 865                 else
 866                         msg->apply_bytes -= send;
 867                 if (msg->sg.size == 0)
 868                         tls_free_open_rec(sk);
 869                 *copied -= (send + delta);
 870                 err = -EACCES;
 871         }
 872
 873         if (likely(!err)) {
 874                 bool reset_eval = !ctx->open_rec;
 875
 876                 rec = ctx->open_rec;
 877                 if (rec) {
 878                         msg = &rec->msg_plaintext;
 879                         if (!msg->apply_bytes)
 880                                 reset_eval = true;
 881                 }
 882                 if (reset_eval) {
 883                         psock->eval = __SK_NONE;
 884                         if (psock->sk_redir) {
 885                                 sock_put(psock->sk_redir);
 886                                 psock->sk_redir = NULL;
 887                         }
 888                 }
 889                 if (rec)
 890                         goto more_data;
 891         }
 892  out_err:
 893         sk_psock_put(sk, psock);
 894         return err;
 895 }
 896
 897 static int tls_sw_push_pending_record(struct sock *sk, int flags)
 898 {
 899         struct tls_context *tls_ctx = tls_get_ctx(sk);
 900         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 901         struct tls_rec *rec = ctx->open_rec;
 902         struct sk_msg *msg_pl;
 903         size_t copied;
 904
 905         if (!rec)
 906                 return 0;
 907
 908         msg_pl = &rec->msg_plaintext;
 909         copied = msg_pl->sg.size;
 910         if (!copied)
 911                 return 0;
 912
 913         return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA,
 914                                    &copied, flags);
 915 }
 916
 917 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
 918 {
 919         long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
 920         struct tls_context *tls_ctx = tls_get_ctx(sk);
 921         struct tls_prot_info *prot = &tls_ctx->prot_info;
 922         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 923         bool async_capable = ctx->async_capable;
 924         unsigned char record_type = TLS_RECORD_TYPE_DATA;
 925         bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
 926         bool eor = !(msg->msg_flags & MSG_MORE);
 927         size_t try_to_copy;
 928         ssize_t copied = 0;
 929         struct sk_msg *msg_pl, *msg_en;
 930         struct tls_rec *rec;
 931         int required_size;
 932         int num_async = 0;
 933         bool full_record;
 934         int record_room;
 935         int num_zc = 0;
 936         int orig_size;
 937         int ret = 0;
 938         int pending;
 939
 940         if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
 941                                MSG_CMSG_COMPAT))
 942                 return -EOPNOTSUPP;
 943
 944         ret = mutex_lock_interruptible(&tls_ctx->tx_lock);
 945         if (ret)
 946                 return ret;
 947         lock_sock(sk);
 948
 949         if (unlikely(msg->msg_controllen)) {
 950                 ret = tls_process_cmsg(sk, msg, &record_type);
 951                 if (ret) {
 952                         if (ret == -EINPROGRESS)
 953                                 num_async++;
 954                         else if (ret != -EAGAIN)
 955                                 goto send_end;
 956                 }
 957         }
 958
 959         while (msg_data_left(msg)) {
 960                 if (sk->sk_err) {
 961                         ret = -sk->sk_err;
 962                         goto send_end;
 963                 }
 964
 965                 if (ctx->open_rec)
 966                         rec = ctx->open_rec;
 967                 else
 968                         rec = ctx->open_rec = tls_get_rec(sk);
 969                 if (!rec) {
 970                         ret = -ENOMEM;
 971                         goto send_end;
 972                 }
 973
 974                 msg_pl = &rec->msg_plaintext;
 975                 msg_en = &rec->msg_encrypted;
 976
 977                 orig_size = msg_pl->sg.size;
 978                 full_record = false;
 979                 try_to_copy = msg_data_left(msg);
 980                 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
 981                 if (try_to_copy >= record_room) {
 982                         try_to_copy = record_room;
 983                         full_record = true;
 984                 }
 985
 986                 required_size = msg_pl->sg.size + try_to_copy +
 987                                 prot->overhead_size;
 988
 989                 if (!sk_stream_memory_free(sk))
 990                         goto wait_for_sndbuf;
 991
 992 alloc_encrypted:
 993                 ret = tls_alloc_encrypted_msg(sk, required_size);
 994                 if (ret) {
 995                         if (ret != -ENOSPC)
 996                                 goto wait_for_memory;
 997
 998                         /* Adjust try_to_copy according to the amount that was
 999                          * actually allocated. The difference is due
1000                          * to max sg elements limit
1001                          */
1002                         try_to_copy -= required_size - msg_en->sg.size;
1003                         full_record = true;
1004                 }
1005
1006                 if (!is_kvec && (full_record || eor) && !async_capable) {
1007                         u32 first = msg_pl->sg.end;
1008
1009                         ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter,
1010                                                         msg_pl, try_to_copy);
1011                         if (ret)
1012                                 goto fallback_to_reg_send;
1013
1014                         num_zc++;
1015                         copied += try_to_copy;
1016
1017                         sk_msg_sg_copy_set(msg_pl, first);
1018                         ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1019                                                   record_type, &copied,
1020                                                   msg->msg_flags);
1021                         if (ret) {
1022                                 if (ret == -EINPROGRESS)
1023                                         num_async++;
1024                                 else if (ret == -ENOMEM)
1025                                         goto wait_for_memory;
1026                                 else if (ctx->open_rec && ret == -ENOSPC)
1027                                         goto rollback_iter;
1028                                 else if (ret != -EAGAIN)
1029                                         goto send_end;
1030                         }
1031                         continue;
1032 rollback_iter:
1033                         copied -= try_to_copy;
1034                         sk_msg_sg_copy_clear(msg_pl, first);
1035                         iov_iter_revert(&msg->msg_iter,
1036                                         msg_pl->sg.size - orig_size);
1037 fallback_to_reg_send:
1038                         sk_msg_trim(sk, msg_pl, orig_size);
1039                 }
1040
1041                 required_size = msg_pl->sg.size + try_to_copy;
1042
1043                 ret = tls_clone_plaintext_msg(sk, required_size);
1044                 if (ret) {
1045                         if (ret != -ENOSPC)
1046                                 goto send_end;
1047
1048                         /* Adjust try_to_copy according to the amount that was
1049                          * actually allocated. The difference is due
1050                          * to max sg elements limit
1051                          */
1052                         try_to_copy -= required_size - msg_pl->sg.size;
1053                         full_record = true;
1054                         sk_msg_trim(sk, msg_en,
1055                                     msg_pl->sg.size + prot->overhead_size);
1056                 }
1057
1058                 if (try_to_copy) {
1059                         ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter,
1060                                                        msg_pl, try_to_copy);
1061                         if (ret < 0)
1062                                 goto trim_sgl;
1063                 }
1064
1065                 /* Open records defined only if successfully copied, otherwise
1066                  * we would trim the sg but not reset the open record frags.
1067                  */
1068                 tls_ctx->pending_open_record_frags = true;
1069                 copied += try_to_copy;
1070                 if (full_record || eor) {
1071                         ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1072                                                   record_type, &copied,
1073                                                   msg->msg_flags);
1074                         if (ret) {
1075                                 if (ret == -EINPROGRESS)
1076                                         num_async++;
1077                                 else if (ret == -ENOMEM)
1078                                         goto wait_for_memory;
1079                                 else if (ret != -EAGAIN) {
1080                                         if (ret == -ENOSPC)
1081                                                 ret = 0;
1082                                         goto send_end;
1083                                 }
1084                         }
1085                 }
1086
1087                 continue;
1088
1089 wait_for_sndbuf:
1090                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1091 wait_for_memory:
1092                 ret = sk_stream_wait_memory(sk, &timeo);
1093                 if (ret) {
1094 trim_sgl:
1095                         if (ctx->open_rec)
1096                                 tls_trim_both_msgs(sk, orig_size);
1097                         goto send_end;
1098                 }
1099
1100                 if (ctx->open_rec && msg_en->sg.size < required_size)
1101                         goto alloc_encrypted;
1102         }
1103
1104         if (!num_async) {
1105                 goto send_end;
1106         } else if (num_zc) {
1107                 /* Wait for pending encryptions to get completed */
1108                 spin_lock_bh(&ctx->encrypt_compl_lock);
1109                 ctx->async_notify = true;
1110
1111                 pending = atomic_read(&ctx->encrypt_pending);
1112                 spin_unlock_bh(&ctx->encrypt_compl_lock);
1113                 if (pending)
1114                         crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1115                 else
1116                         reinit_completion(&ctx->async_wait.completion);
1117
1118                 /* There can be no concurrent accesses, since we have no
1119                  * pending encrypt operations
1120                  */
1121                 WRITE_ONCE(ctx->async_notify, false);
1122
1123                 if (ctx->async_wait.err) {
1124                         ret = ctx->async_wait.err;
1125                         copied = 0;
1126                 }
1127         }
1128
1129         /* Transmit if any encryptions have completed */
1130         if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1131                 cancel_delayed_work(&ctx->tx_work.work);
1132                 tls_tx_records(sk, msg->msg_flags);
1133         }
1134
1135 send_end:
1136         ret = sk_stream_error(sk, msg->msg_flags, ret);
1137
1138         release_sock(sk);
1139         mutex_unlock(&tls_ctx->tx_lock);
1140         return copied > 0 ? copied : ret;
1141 }
1142
1143 static int tls_sw_do_sendpage(struct sock *sk, struct page *page,
1144                               int offset, size_t size, int flags)
1145 {
1146         long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1147         struct tls_context *tls_ctx = tls_get_ctx(sk);
1148         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1149         struct tls_prot_info *prot = &tls_ctx->prot_info;
1150         unsigned char record_type = TLS_RECORD_TYPE_DATA;
1151         struct sk_msg *msg_pl;
1152         struct tls_rec *rec;
1153         int num_async = 0;
1154         ssize_t copied = 0;
1155         bool full_record;
1156         int record_room;
1157         int ret = 0;
1158         bool eor;
1159
1160         eor = !(flags & MSG_SENDPAGE_NOTLAST);
1161         sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1162
1163         /* Call the sk_stream functions to manage the sndbuf mem. */
1164         while (size > 0) {
1165                 size_t copy, required_size;
1166
1167                 if (sk->sk_err) {
1168                         ret = -sk->sk_err;
1169                         goto sendpage_end;
1170                 }
1171
1172                 if (ctx->open_rec)
1173                         rec = ctx->open_rec;
1174                 else
1175                         rec = ctx->open_rec = tls_get_rec(sk);
1176                 if (!rec) {
1177                         ret = -ENOMEM;
1178                         goto sendpage_end;
1179                 }
1180
1181                 msg_pl = &rec->msg_plaintext;
1182
1183                 full_record = false;
1184                 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
1185                 copy = size;
1186                 if (copy >= record_room) {
1187                         copy = record_room;
1188                         full_record = true;
1189                 }
1190
1191                 required_size = msg_pl->sg.size + copy + prot->overhead_size;
1192
1193                 if (!sk_stream_memory_free(sk))
1194                         goto wait_for_sndbuf;
1195 alloc_payload:
1196                 ret = tls_alloc_encrypted_msg(sk, required_size);
1197                 if (ret) {
1198                         if (ret != -ENOSPC)
1199                                 goto wait_for_memory;
1200
1201                         /* Adjust copy according to the amount that was
1202                          * actually allocated. The difference is due
1203                          * to max sg elements limit
1204                          */
1205                         copy -= required_size - msg_pl->sg.size;
1206                         full_record = true;
1207                 }
1208
1209                 sk_msg_page_add(msg_pl, page, copy, offset);
1210                 sk_mem_charge(sk, copy);
1211
1212                 offset += copy;
1213                 size -= copy;
1214                 copied += copy;
1215
1216                 tls_ctx->pending_open_record_frags = true;
1217                 if (full_record || eor || sk_msg_full(msg_pl)) {
1218                         ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1219                                                   record_type, &copied, flags);
1220                         if (ret) {
1221                                 if (ret == -EINPROGRESS)
1222                                         num_async++;
1223                                 else if (ret == -ENOMEM)
1224                                         goto wait_for_memory;
1225                                 else if (ret != -EAGAIN) {
1226                                         if (ret == -ENOSPC)
1227                                                 ret = 0;
1228                                         goto sendpage_end;
1229                                 }
1230                         }
1231                 }
1232                 continue;
1233 wait_for_sndbuf:
1234                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1235 wait_for_memory:
1236                 ret = sk_stream_wait_memory(sk, &timeo);
1237                 if (ret) {
1238                         if (ctx->open_rec)
1239                                 tls_trim_both_msgs(sk, msg_pl->sg.size);
1240                         goto sendpage_end;
1241                 }
1242
1243                 if (ctx->open_rec)
1244                         goto alloc_payload;
1245         }
1246
1247         if (num_async) {
1248                 /* Transmit if any encryptions have completed */
1249                 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1250                         cancel_delayed_work(&ctx->tx_work.work);
1251                         tls_tx_records(sk, flags);
1252                 }
1253         }
1254 sendpage_end:
1255         ret = sk_stream_error(sk, flags, ret);
1256         return copied > 0 ? copied : ret;
1257 }
1258
1259 int tls_sw_sendpage_locked(struct sock *sk, struct page *page,
1260                            int offset, size_t size, int flags)
1261 {
1262         if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1263                       MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY |
1264                       MSG_NO_SHARED_FRAGS))
1265                 return -EOPNOTSUPP;
1266
1267         return tls_sw_do_sendpage(sk, page, offset, size, flags);
1268 }
1269
1270 int tls_sw_sendpage(struct sock *sk, struct page *page,
1271                     int offset, size_t size, int flags)
1272 {
1273         struct tls_context *tls_ctx = tls_get_ctx(sk);
1274         int ret;
1275
1276         if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1277                       MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY))
1278                 return -EOPNOTSUPP;
1279
1280         ret = mutex_lock_interruptible(&tls_ctx->tx_lock);
1281         if (ret)
1282                 return ret;
1283         lock_sock(sk);
1284         ret = tls_sw_do_sendpage(sk, page, offset, size, flags);
1285         release_sock(sk);
1286         mutex_unlock(&tls_ctx->tx_lock);
1287         return ret;
1288 }
1289
1290 static int
1291 tls_rx_rec_wait(struct sock *sk, struct sk_psock *psock, bool nonblock,
1292                 bool released)
1293 {
1294         struct tls_context *tls_ctx = tls_get_ctx(sk);
1295         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1296         DEFINE_WAIT_FUNC(wait, woken_wake_function);
1297         long timeo;
1298
1299         timeo = sock_rcvtimeo(sk, nonblock);
1300
1301         while (!tls_strp_msg_ready(ctx)) {
1302                 if (!sk_psock_queue_empty(psock))
1303                         return 0;
1304
1305                 if (sk->sk_err)
1306                         return sock_error(sk);
1307
1308                 if (!skb_queue_empty(&sk->sk_receive_queue)) {
1309                         tls_strp_check_rcv(&ctx->strp);
1310                         if (tls_strp_msg_ready(ctx))
1311                                 break;
1312                 }
1313
1314                 if (sk->sk_shutdown & RCV_SHUTDOWN)
1315                         return 0;
1316
1317                 if (sock_flag(sk, SOCK_DONE))
1318                         return 0;
1319
1320                 if (!timeo)
1321                         return -EAGAIN;
1322
1323                 released = true;
1324                 add_wait_queue(sk_sleep(sk), &wait);
1325                 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1326                 sk_wait_event(sk, &timeo,
1327                               tls_strp_msg_ready(ctx) ||
1328                               !sk_psock_queue_empty(psock),
1329                               &wait);
1330                 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1331                 remove_wait_queue(sk_sleep(sk), &wait);
1332
1333                 /* Handle signals */
1334                 if (signal_pending(current))
1335                         return sock_intr_errno(timeo);
1336         }
1337
1338         tls_strp_msg_load(&ctx->strp, released);
1339
1340         return 1;
1341 }
1342
1343 static int tls_setup_from_iter(struct iov_iter *from,
1344                                int length, int *pages_used,
1345                                struct scatterlist *to,
1346                                int to_max_pages)
1347 {
1348         int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1349         struct page *pages[MAX_SKB_FRAGS];
1350         unsigned int size = 0;
1351         ssize_t copied, use;
1352         size_t offset;
1353
1354         while (length > 0) {
1355                 i = 0;
1356                 maxpages = to_max_pages - num_elem;
1357                 if (maxpages == 0) {
1358                         rc = -EFAULT;
1359                         goto out;
1360                 }
1361                 copied = iov_iter_get_pages2(from, pages,
1362                                             length,
1363                                             maxpages, &offset);
1364                 if (copied <= 0) {
1365                         rc = -EFAULT;
1366                         goto out;
1367                 }
1368
1369                 length -= copied;
1370                 size += copied;
1371                 while (copied) {
1372                         use = min_t(int, copied, PAGE_SIZE - offset);
1373
1374                         sg_set_page(&to[num_elem],
1375                                     pages[i], use, offset);
1376                         sg_unmark_end(&to[num_elem]);
1377                         /* We do not uncharge memory from this API */
1378
1379                         offset = 0;
1380                         copied -= use;
1381
1382                         i++;
1383                         num_elem++;
1384                 }
1385         }
1386         /* Mark the end in the last sg entry if newly added */
1387         if (num_elem > *pages_used)
1388                 sg_mark_end(&to[num_elem - 1]);
1389 out:
1390         if (rc)
1391                 iov_iter_revert(from, size);
1392         *pages_used = num_elem;
1393
1394         return rc;
1395 }
1396
1397 static struct sk_buff *
1398 tls_alloc_clrtxt_skb(struct sock *sk, struct sk_buff *skb,
1399                      unsigned int full_len)
1400 {
1401         struct strp_msg *clr_rxm;
1402         struct sk_buff *clr_skb;
1403         int err;
1404
1405         clr_skb = alloc_skb_with_frags(0, full_len, TLS_PAGE_ORDER,
1406                                        &err, sk->sk_allocation);
1407         if (!clr_skb)
1408                 return NULL;
1409
1410         skb_copy_header(clr_skb, skb);
1411         clr_skb->len = full_len;
1412         clr_skb->data_len = full_len;
1413
1414         clr_rxm = strp_msg(clr_skb);
1415         clr_rxm->offset = 0;
1416
1417         return clr_skb;
1418 }
1419
1420 /* Decrypt handlers
1421  *
1422  * tls_decrypt_sw() and tls_decrypt_device() are decrypt handlers.
1423  * They must transform the darg in/out argument are as follows:
1424  *       |          Input            |         Output
1425  * -------------------------------------------------------------------
1426  *    zc | Zero-copy decrypt allowed | Zero-copy performed
1427  * async | Async decrypt allowed     | Async crypto used / in progress
1428  *   skb |            *              | Output skb
1429  *
1430  * If ZC decryption was performed darg.skb will point to the input skb.
1431  */
1432
1433 /* This function decrypts the input skb into either out_iov or in out_sg
1434  * or in skb buffers itself. The input parameter 'darg->zc' indicates if
1435  * zero-copy mode needs to be tried or not. With zero-copy mode, either
1436  * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1437  * NULL, then the decryption happens inside skb buffers itself, i.e.
1438  * zero-copy gets disabled and 'darg->zc' is updated.
1439  */
1440 static int tls_decrypt_sg(struct sock *sk, struct iov_iter *out_iov,
1441                           struct scatterlist *out_sg,
1442                           struct tls_decrypt_arg *darg)
1443 {
1444         struct tls_context *tls_ctx = tls_get_ctx(sk);
1445         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1446         struct tls_prot_info *prot = &tls_ctx->prot_info;
1447         int n_sgin, n_sgout, aead_size, err, pages = 0;
1448         struct sk_buff *skb = tls_strp_msg(ctx);
1449         const struct strp_msg *rxm = strp_msg(skb);
1450         const struct tls_msg *tlm = tls_msg(skb);
1451         struct aead_request *aead_req;
1452         struct scatterlist *sgin = NULL;
1453         struct scatterlist *sgout = NULL;
1454         const int data_len = rxm->full_len - prot->overhead_size;
1455         int tail_pages = !!prot->tail_size;
1456         struct tls_decrypt_ctx *dctx;
1457         struct sk_buff *clear_skb;
1458         int iv_offset = 0;
1459         u8 *mem;
1460
1461         n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size,
1462                          rxm->full_len - prot->prepend_size);
1463         if (n_sgin < 1)
1464                 return n_sgin ?: -EBADMSG;
1465
1466         if (darg->zc && (out_iov || out_sg)) {
1467                 clear_skb = NULL;
1468
1469                 if (out_iov)
1470                         n_sgout = 1 + tail_pages +
1471                                 iov_iter_npages_cap(out_iov, INT_MAX, data_len);
1472                 else
1473                         n_sgout = sg_nents(out_sg);
1474         } else {
1475                 darg->zc = false;
1476
1477                 clear_skb = tls_alloc_clrtxt_skb(sk, skb, rxm->full_len);
1478                 if (!clear_skb)
1479                         return -ENOMEM;
1480
1481                 n_sgout = 1 + skb_shinfo(clear_skb)->nr_frags;
1482         }
1483
1484         /* Increment to accommodate AAD */
1485         n_sgin = n_sgin + 1;
1486
1487         /* Allocate a single block of memory which contains
1488          *   aead_req || tls_decrypt_ctx.
1489          * Both structs are variable length.
1490          */
1491         aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
1492         mem = kmalloc(aead_size + struct_size(dctx, sg, n_sgin + n_sgout),
1493                       sk->sk_allocation);
1494         if (!mem) {
1495                 err = -ENOMEM;
1496                 goto exit_free_skb;
1497         }
1498
1499         /* Segment the allocated memory */
1500         aead_req = (struct aead_request *)mem;
1501         dctx = (struct tls_decrypt_ctx *)(mem + aead_size);
1502         sgin = &dctx->sg[0];
1503         sgout = &dctx->sg[n_sgin];
1504
1505         /* For CCM based ciphers, first byte of nonce+iv is a constant */
1506         switch (prot->cipher_type) {
1507         case TLS_CIPHER_AES_CCM_128:
1508                 dctx->iv[0] = TLS_AES_CCM_IV_B0_BYTE;
1509                 iv_offset = 1;
1510                 break;
1511         case TLS_CIPHER_SM4_CCM:
1512                 dctx->iv[0] = TLS_SM4_CCM_IV_B0_BYTE;
1513                 iv_offset = 1;
1514                 break;
1515         }
1516
1517         /* Prepare IV */
1518         if (prot->version == TLS_1_3_VERSION ||
1519             prot->cipher_type == TLS_CIPHER_CHACHA20_POLY1305) {
1520                 memcpy(&dctx->iv[iv_offset], tls_ctx->rx.iv,
1521                        prot->iv_size + prot->salt_size);
1522         } else {
1523                 err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
1524                                     &dctx->iv[iv_offset] + prot->salt_size,
1525                                     prot->iv_size);
1526                 if (err < 0)
1527                         goto exit_free;
1528                 memcpy(&dctx->iv[iv_offset], tls_ctx->rx.iv, prot->salt_size);
1529         }
1530         tls_xor_iv_with_seq(prot, &dctx->iv[iv_offset], tls_ctx->rx.rec_seq);
1531
1532         /* Prepare AAD */
1533         tls_make_aad(dctx->aad, rxm->full_len - prot->overhead_size +
1534                      prot->tail_size,
1535                      tls_ctx->rx.rec_seq, tlm->control, prot);
1536
1537         /* Prepare sgin */
1538         sg_init_table(sgin, n_sgin);
1539         sg_set_buf(&sgin[0], dctx->aad, prot->aad_size);
1540         err = skb_to_sgvec(skb, &sgin[1],
1541                            rxm->offset + prot->prepend_size,
1542                            rxm->full_len - prot->prepend_size);
1543         if (err < 0)
1544                 goto exit_free;
1545
1546         if (clear_skb) {
1547                 sg_init_table(sgout, n_sgout);
1548                 sg_set_buf(&sgout[0], dctx->aad, prot->aad_size);
1549
1550                 err = skb_to_sgvec(clear_skb, &sgout[1], prot->prepend_size,
1551                                    data_len + prot->tail_size);
1552                 if (err < 0)
1553                         goto exit_free;
1554         } else if (out_iov) {
1555                 sg_init_table(sgout, n_sgout);
1556                 sg_set_buf(&sgout[0], dctx->aad, prot->aad_size);
1557
1558                 err = tls_setup_from_iter(out_iov, data_len, &pages, &sgout[1],
1559                                           (n_sgout - 1 - tail_pages));
1560                 if (err < 0)
1561                         goto exit_free_pages;
1562
1563                 if (prot->tail_size) {
1564                         sg_unmark_end(&sgout[pages]);
1565                         sg_set_buf(&sgout[pages + 1], &dctx->tail,
1566                                    prot->tail_size);
1567                         sg_mark_end(&sgout[pages + 1]);
1568                 }
1569         } else if (out_sg) {
1570                 memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
1571         }
1572
1573         /* Prepare and submit AEAD request */
1574         err = tls_do_decryption(sk, sgin, sgout, dctx->iv,
1575                                 data_len + prot->tail_size, aead_req, darg);
1576         if (err)
1577                 goto exit_free_pages;
1578
1579         darg->skb = clear_skb ?: tls_strp_msg(ctx);
1580         clear_skb = NULL;
1581
1582         if (unlikely(darg->async)) {
1583                 err = tls_strp_msg_hold(&ctx->strp, &ctx->async_hold);
1584                 if (err)
1585                         __skb_queue_tail(&ctx->async_hold, darg->skb);
1586                 return err;
1587         }
1588
1589         if (prot->tail_size)
1590                 darg->tail = dctx->tail;
1591
1592 exit_free_pages:
1593         /* Release the pages in case iov was mapped to pages */
1594         for (; pages > 0; pages--)
1595                 put_page(sg_page(&sgout[pages]));
1596 exit_free:
1597         kfree(mem);
1598 exit_free_skb:
1599         consume_skb(clear_skb);
1600         return err;
1601 }
1602
1603 static int
1604 tls_decrypt_sw(struct sock *sk, struct tls_context *tls_ctx,
1605                struct msghdr *msg, struct tls_decrypt_arg *darg)
1606 {
1607         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1608         struct tls_prot_info *prot = &tls_ctx->prot_info;
1609         struct strp_msg *rxm;
1610         int pad, err;
1611
1612         err = tls_decrypt_sg(sk, &msg->msg_iter, NULL, darg);
1613         if (err < 0) {
1614                 if (err == -EBADMSG)
1615                         TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTERROR);
1616                 return err;
1617         }
1618         /* keep going even for ->async, the code below is TLS 1.3 */
1619
1620         /* If opportunistic TLS 1.3 ZC failed retry without ZC */
1621         if (unlikely(darg->zc && prot->version == TLS_1_3_VERSION &&
1622                      darg->tail != TLS_RECORD_TYPE_DATA)) {
1623                 darg->zc = false;
1624                 if (!darg->tail)
1625                         TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXNOPADVIOL);
1626                 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTRETRY);
1627                 return tls_decrypt_sw(sk, tls_ctx, msg, darg);
1628         }
1629
1630         pad = tls_padding_length(prot, darg->skb, darg);
1631         if (pad < 0) {
1632                 if (darg->skb != tls_strp_msg(ctx))
1633                         consume_skb(darg->skb);
1634                 return pad;
1635         }
1636
1637         rxm = strp_msg(darg->skb);
1638         rxm->full_len -= pad;
1639
1640         return 0;
1641 }
1642
1643 static int
1644 tls_decrypt_device(struct sock *sk, struct msghdr *msg,
1645                    struct tls_context *tls_ctx, struct tls_decrypt_arg *darg)
1646 {
1647         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1648         struct tls_prot_info *prot = &tls_ctx->prot_info;
1649         struct strp_msg *rxm;
1650         int pad, err;
1651
1652         if (tls_ctx->rx_conf != TLS_HW)
1653                 return 0;
1654
1655         err = tls_device_decrypted(sk, tls_ctx);
1656         if (err <= 0)
1657                 return err;
1658
1659         pad = tls_padding_length(prot, tls_strp_msg(ctx), darg);
1660         if (pad < 0)
1661                 return pad;
1662
1663         darg->async = false;
1664         darg->skb = tls_strp_msg(ctx);
1665         /* ->zc downgrade check, in case TLS 1.3 gets here */
1666         darg->zc &= !(prot->version == TLS_1_3_VERSION &&
1667                       tls_msg(darg->skb)->control != TLS_RECORD_TYPE_DATA);
1668
1669         rxm = strp_msg(darg->skb);
1670         rxm->full_len -= pad;
1671
1672         if (!darg->zc) {
1673                 /* Non-ZC case needs a real skb */
1674                 darg->skb = tls_strp_msg_detach(ctx);
1675                 if (!darg->skb)
1676                         return -ENOMEM;
1677         } else {
1678                 unsigned int off, len;
1679
1680                 /* In ZC case nobody cares about the output skb.
1681                  * Just copy the data here. Note the skb is not fully trimmed.
1682                  */
1683                 off = rxm->offset + prot->prepend_size;
1684                 len = rxm->full_len - prot->overhead_size;
1685
1686                 err = skb_copy_datagram_msg(darg->skb, off, msg, len);
1687                 if (err)
1688                         return err;
1689         }
1690         return 1;
1691 }
1692
1693 static int tls_rx_one_record(struct sock *sk, struct msghdr *msg,
1694                              struct tls_decrypt_arg *darg)
1695 {
1696         struct tls_context *tls_ctx = tls_get_ctx(sk);
1697         struct tls_prot_info *prot = &tls_ctx->prot_info;
1698         struct strp_msg *rxm;
1699         int err;
1700
1701         err = tls_decrypt_device(sk, msg, tls_ctx, darg);
1702         if (!err)
1703                 err = tls_decrypt_sw(sk, tls_ctx, msg, darg);
1704         if (err < 0)
1705                 return err;
1706
1707         rxm = strp_msg(darg->skb);
1708         rxm->offset += prot->prepend_size;
1709         rxm->full_len -= prot->overhead_size;
1710         tls_advance_record_sn(sk, prot, &tls_ctx->rx);
1711
1712         return 0;
1713 }
1714
1715 int decrypt_skb(struct sock *sk, struct scatterlist *sgout)
1716 {
1717         struct tls_decrypt_arg darg = { .zc = true, };
1718
1719         return tls_decrypt_sg(sk, NULL, sgout, &darg);
1720 }
1721
1722 static int tls_record_content_type(struct msghdr *msg, struct tls_msg *tlm,
1723                                    u8 *control)
1724 {
1725         int err;
1726
1727         if (!*control) {
1728                 *control = tlm->control;
1729                 if (!*control)
1730                         return -EBADMSG;
1731
1732                 err = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1733                                sizeof(*control), control);
1734                 if (*control != TLS_RECORD_TYPE_DATA) {
1735                         if (err || msg->msg_flags & MSG_CTRUNC)
1736                                 return -EIO;
1737                 }
1738         } else if (*control != tlm->control) {
1739                 return 0;
1740         }
1741
1742         return 1;
1743 }
1744
1745 static void tls_rx_rec_done(struct tls_sw_context_rx *ctx)
1746 {
1747         tls_strp_msg_done(&ctx->strp);
1748 }
1749
1750 /* This function traverses the rx_list in tls receive context to copies the
1751  * decrypted records into the buffer provided by caller zero copy is not
1752  * true. Further, the records are removed from the rx_list if it is not a peek
1753  * case and the record has been consumed completely.
1754  */
1755 static int process_rx_list(struct tls_sw_context_rx *ctx,
1756                            struct msghdr *msg,
1757                            u8 *control,
1758                            size_t skip,
1759                            size_t len,
1760                            bool is_peek)
1761 {
1762         struct sk_buff *skb = skb_peek(&ctx->rx_list);
1763         struct tls_msg *tlm;
1764         ssize_t copied = 0;
1765         int err;
1766
1767         while (skip && skb) {
1768                 struct strp_msg *rxm = strp_msg(skb);
1769                 tlm = tls_msg(skb);
1770
1771                 err = tls_record_content_type(msg, tlm, control);
1772                 if (err <= 0)
1773                         goto out;
1774
1775                 if (skip < rxm->full_len)
1776                         break;
1777
1778                 skip = skip - rxm->full_len;
1779                 skb = skb_peek_next(skb, &ctx->rx_list);
1780         }
1781
1782         while (len && skb) {
1783                 struct sk_buff *next_skb;
1784                 struct strp_msg *rxm = strp_msg(skb);
1785                 int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1786
1787                 tlm = tls_msg(skb);
1788
1789                 err = tls_record_content_type(msg, tlm, control);
1790                 if (err <= 0)
1791                         goto out;
1792
1793                 err = skb_copy_datagram_msg(skb, rxm->offset + skip,
1794                                             msg, chunk);
1795                 if (err < 0)
1796                         goto out;
1797
1798                 len = len - chunk;
1799                 copied = copied + chunk;
1800
1801                 /* Consume the data from record if it is non-peek case*/
1802                 if (!is_peek) {
1803                         rxm->offset = rxm->offset + chunk;
1804                         rxm->full_len = rxm->full_len - chunk;
1805
1806                         /* Return if there is unconsumed data in the record */
1807                         if (rxm->full_len - skip)
1808                                 break;
1809                 }
1810
1811                 /* The remaining skip-bytes must lie in 1st record in rx_list.
1812                  * So from the 2nd record, 'skip' should be 0.
1813                  */
1814                 skip = 0;
1815
1816                 if (msg)
1817                         msg->msg_flags |= MSG_EOR;
1818
1819                 next_skb = skb_peek_next(skb, &ctx->rx_list);
1820
1821                 if (!is_peek) {
1822                         __skb_unlink(skb, &ctx->rx_list);
1823                         consume_skb(skb);
1824                 }
1825
1826                 skb = next_skb;
1827         }
1828         err = 0;
1829
1830 out:
1831         return copied ? : err;
1832 }
1833
1834 static bool
1835 tls_read_flush_backlog(struct sock *sk, struct tls_prot_info *prot,
1836                        size_t len_left, size_t decrypted, ssize_t done,
1837                        size_t *flushed_at)
1838 {
1839         size_t max_rec;
1840
1841         if (len_left <= decrypted)
1842                 return false;
1843
1844         max_rec = prot->overhead_size - prot->tail_size + TLS_MAX_PAYLOAD_SIZE;
1845         if (done - *flushed_at < SZ_128K && tcp_inq(sk) > max_rec)
1846                 return false;
1847
1848         *flushed_at = done;
1849         return sk_flush_backlog(sk);
1850 }
1851
1852 static int tls_rx_reader_lock(struct sock *sk, struct tls_sw_context_rx *ctx,
1853                               bool nonblock)
1854 {
1855         long timeo;
1856         int err;
1857
1858         lock_sock(sk);
1859
1860         timeo = sock_rcvtimeo(sk, nonblock);
1861
1862         while (unlikely(ctx->reader_present)) {
1863                 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1864
1865                 ctx->reader_contended = 1;
1866
1867                 add_wait_queue(&ctx->wq, &wait);
1868                 sk_wait_event(sk, &timeo,
1869                               !READ_ONCE(ctx->reader_present), &wait);
1870                 remove_wait_queue(&ctx->wq, &wait);
1871
1872                 if (timeo <= 0) {
1873                         err = -EAGAIN;
1874                         goto err_unlock;
1875                 }
1876                 if (signal_pending(current)) {
1877                         err = sock_intr_errno(timeo);
1878                         goto err_unlock;
1879                 }
1880         }
1881
1882         WRITE_ONCE(ctx->reader_present, 1);
1883
1884         return 0;
1885
1886 err_unlock:
1887         release_sock(sk);
1888         return err;
1889 }
1890
1891 static void tls_rx_reader_unlock(struct sock *sk, struct tls_sw_context_rx *ctx)
1892 {
1893         if (unlikely(ctx->reader_contended)) {
1894                 if (wq_has_sleeper(&ctx->wq))
1895                         wake_up(&ctx->wq);
1896                 else
1897                         ctx->reader_contended = 0;
1898
1899                 WARN_ON_ONCE(!ctx->reader_present);
1900         }
1901
1902         WRITE_ONCE(ctx->reader_present, 0);
1903         release_sock(sk);
1904 }
1905
1906 int tls_sw_recvmsg(struct sock *sk,
1907                    struct msghdr *msg,
1908                    size_t len,
1909                    int flags,
1910                    int *addr_len)
1911 {
1912         struct tls_context *tls_ctx = tls_get_ctx(sk);
1913         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1914         struct tls_prot_info *prot = &tls_ctx->prot_info;
1915         ssize_t decrypted = 0, async_copy_bytes = 0;
1916         struct sk_psock *psock;
1917         unsigned char control = 0;
1918         size_t flushed_at = 0;
1919         struct strp_msg *rxm;
1920         struct tls_msg *tlm;
1921         ssize_t copied = 0;
1922         bool async = false;
1923         int target, err;
1924         bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1925         bool is_peek = flags & MSG_PEEK;
1926         bool released = true;
1927         bool bpf_strp_enabled;
1928         bool zc_capable;
1929
1930         if (unlikely(flags & MSG_ERRQUEUE))
1931                 return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
1932
1933         psock = sk_psock_get(sk);
1934         err = tls_rx_reader_lock(sk, ctx, flags & MSG_DONTWAIT);
1935         if (err < 0)
1936                 return err;
1937         bpf_strp_enabled = sk_psock_strp_enabled(psock);
1938
1939         /* If crypto failed the connection is broken */
1940         err = ctx->async_wait.err;
1941         if (err)
1942                 goto end;
1943
1944         /* Process pending decrypted records. It must be non-zero-copy */
1945         err = process_rx_list(ctx, msg, &control, 0, len, is_peek);
1946         if (err < 0)
1947                 goto end;
1948
1949         copied = err;
1950         if (len <= copied)
1951                 goto end;
1952
1953         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1954         len = len - copied;
1955
1956         zc_capable = !bpf_strp_enabled && !is_kvec && !is_peek &&
1957                 ctx->zc_capable;
1958         decrypted = 0;
1959         while (len && (decrypted + copied < target || tls_strp_msg_ready(ctx))) {
1960                 struct tls_decrypt_arg darg;
1961                 int to_decrypt, chunk;
1962
1963                 err = tls_rx_rec_wait(sk, psock, flags & MSG_DONTWAIT,
1964                                       released);
1965                 if (err <= 0) {
1966                         if (psock) {
1967                                 chunk = sk_msg_recvmsg(sk, psock, msg, len,
1968                                                        flags);
1969                                 if (chunk > 0) {
1970                                         decrypted += chunk;
1971                                         len -= chunk;
1972                                         continue;
1973                                 }
1974                         }
1975                         goto recv_end;
1976                 }
1977
1978                 memset(&darg.inargs, 0, sizeof(darg.inargs));
1979
1980                 rxm = strp_msg(tls_strp_msg(ctx));
1981                 tlm = tls_msg(tls_strp_msg(ctx));
1982
1983                 to_decrypt = rxm->full_len - prot->overhead_size;
1984
1985                 if (zc_capable && to_decrypt <= len &&
1986                     tlm->control == TLS_RECORD_TYPE_DATA)
1987                         darg.zc = true;
1988
1989                 /* Do not use async mode if record is non-data */
1990                 if (tlm->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled)
1991                         darg.async = ctx->async_capable;
1992                 else
1993                         darg.async = false;
1994
1995                 err = tls_rx_one_record(sk, msg, &darg);
1996                 if (err < 0) {
1997                         tls_err_abort(sk, -EBADMSG);
1998                         goto recv_end;
1999                 }
2000
2001                 async |= darg.async;
2002
2003                 /* If the type of records being processed is not known yet,
2004                  * set it to record type just dequeued. If it is already known,
2005                  * but does not match the record type just dequeued, go to end.
2006                  * We always get record type here since for tls1.2, record type
2007                  * is known just after record is dequeued from stream parser.
2008                  * For tls1.3, we disable async.
2009                  */
2010                 err = tls_record_content_type(msg, tls_msg(darg.skb), &control);
2011                 if (err <= 0) {
2012                         DEBUG_NET_WARN_ON_ONCE(darg.zc);
2013                         tls_rx_rec_done(ctx);
2014 put_on_rx_list_err:
2015                         __skb_queue_tail(&ctx->rx_list, darg.skb);
2016                         goto recv_end;
2017                 }
2018
2019                 /* periodically flush backlog, and feed strparser */
2020                 released = tls_read_flush_backlog(sk, prot, len, to_decrypt,
2021                                                   decrypted + copied,
2022                                                   &flushed_at);
2023
2024                 /* TLS 1.3 may have updated the length by more than overhead */
2025                 rxm = strp_msg(darg.skb);
2026                 chunk = rxm->full_len;
2027                 tls_rx_rec_done(ctx);
2028
2029                 if (!darg.zc) {
2030                         bool partially_consumed = chunk > len;
2031                         struct sk_buff *skb = darg.skb;
2032
2033                         DEBUG_NET_WARN_ON_ONCE(darg.skb == ctx->strp.anchor);
2034
2035                         if (async) {
2036                                 /* TLS 1.2-only, to_decrypt must be text len */
2037                                 chunk = min_t(int, to_decrypt, len);
2038                                 async_copy_bytes += chunk;
2039 put_on_rx_list:
2040                                 decrypted += chunk;
2041                                 len -= chunk;
2042                                 __skb_queue_tail(&ctx->rx_list, skb);
2043                                 continue;
2044                         }
2045
2046                         if (bpf_strp_enabled) {
2047                                 released = true;
2048                                 err = sk_psock_tls_strp_read(psock, skb);
2049                                 if (err != __SK_PASS) {
2050                                         rxm->offset = rxm->offset + rxm->full_len;
2051                                         rxm->full_len = 0;
2052                                         if (err == __SK_DROP)
2053                                                 consume_skb(skb);
2054                                         continue;
2055                                 }
2056                         }
2057
2058                         if (partially_consumed)
2059                                 chunk = len;
2060
2061                         err = skb_copy_datagram_msg(skb, rxm->offset,
2062                                                     msg, chunk);
2063                         if (err < 0)
2064                                 goto put_on_rx_list_err;
2065
2066                         if (is_peek)
2067                                 goto put_on_rx_list;
2068
2069                         if (partially_consumed) {
2070                                 rxm->offset += chunk;
2071                                 rxm->full_len -= chunk;
2072                                 goto put_on_rx_list;
2073                         }
2074
2075                         consume_skb(skb);
2076                 }
2077
2078                 decrypted += chunk;
2079                 len -= chunk;
2080
2081                 /* Return full control message to userspace before trying
2082                  * to parse another message type
2083                  */
2084                 msg->msg_flags |= MSG_EOR;
2085                 if (control != TLS_RECORD_TYPE_DATA)
2086                         break;
2087         }
2088
2089 recv_end:
2090         if (async) {
2091                 int ret, pending;
2092
2093                 /* Wait for all previously submitted records to be decrypted */
2094                 spin_lock_bh(&ctx->decrypt_compl_lock);
2095                 reinit_completion(&ctx->async_wait.completion);
2096                 pending = atomic_read(&ctx->decrypt_pending);
2097                 spin_unlock_bh(&ctx->decrypt_compl_lock);
2098                 ret = 0;
2099                 if (pending)
2100                         ret = crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
2101                 __skb_queue_purge(&ctx->async_hold);
2102
2103                 if (ret) {
2104                         if (err >= 0 || err == -EINPROGRESS)
2105                                 err = ret;
2106                         decrypted = 0;
2107                         goto end;
2108                 }
2109
2110                 /* Drain records from the rx_list & copy if required */
2111                 if (is_peek || is_kvec)
2112                         err = process_rx_list(ctx, msg, &control, copied,
2113                                               decrypted, is_peek);
2114                 else
2115                         err = process_rx_list(ctx, msg, &control, 0,
2116                                               async_copy_bytes, is_peek);
2117                 decrypted += max(err, 0);
2118         }
2119
2120         copied += decrypted;
2121
2122 end:
2123         tls_rx_reader_unlock(sk, ctx);
2124         if (psock)
2125                 sk_psock_put(sk, psock);
2126         return copied ? : err;
2127 }
2128
2129 ssize_t tls_sw_splice_read(struct socket *sock,  loff_t *ppos,
2130                            struct pipe_inode_info *pipe,
2131                            size_t len, unsigned int flags)
2132 {
2133         struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
2134         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2135         struct strp_msg *rxm = NULL;
2136         struct sock *sk = sock->sk;
2137         struct tls_msg *tlm;
2138         struct sk_buff *skb;
2139         ssize_t copied = 0;
2140         int chunk;
2141         int err;
2142
2143         err = tls_rx_reader_lock(sk, ctx, flags & SPLICE_F_NONBLOCK);
2144         if (err < 0)
2145                 return err;
2146
2147         if (!skb_queue_empty(&ctx->rx_list)) {
2148                 skb = __skb_dequeue(&ctx->rx_list);
2149         } else {
2150                 struct tls_decrypt_arg darg;
2151
2152                 err = tls_rx_rec_wait(sk, NULL, flags & SPLICE_F_NONBLOCK,
2153                                       true);
2154                 if (err <= 0)
2155                         goto splice_read_end;
2156
2157                 memset(&darg.inargs, 0, sizeof(darg.inargs));
2158
2159                 err = tls_rx_one_record(sk, NULL, &darg);
2160                 if (err < 0) {
2161                         tls_err_abort(sk, -EBADMSG);
2162                         goto splice_read_end;
2163                 }
2164
2165                 tls_rx_rec_done(ctx);
2166                 skb = darg.skb;
2167         }
2168
2169         rxm = strp_msg(skb);
2170         tlm = tls_msg(skb);
2171
2172         /* splice does not support reading control messages */
2173         if (tlm->control != TLS_RECORD_TYPE_DATA) {
2174                 err = -EINVAL;
2175                 goto splice_requeue;
2176         }
2177
2178         chunk = min_t(unsigned int, rxm->full_len, len);
2179         copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
2180         if (copied < 0)
2181                 goto splice_requeue;
2182
2183         if (chunk < rxm->full_len) {
2184                 rxm->offset += len;
2185                 rxm->full_len -= len;
2186                 goto splice_requeue;
2187         }
2188
2189         consume_skb(skb);
2190
2191 splice_read_end:
2192         tls_rx_reader_unlock(sk, ctx);
2193         return copied ? : err;
2194
2195 splice_requeue:
2196         __skb_queue_head(&ctx->rx_list, skb);
2197         goto splice_read_end;
2198 }
2199
2200 bool tls_sw_sock_is_readable(struct sock *sk)
2201 {
2202         struct tls_context *tls_ctx = tls_get_ctx(sk);
2203         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2204         bool ingress_empty = true;
2205         struct sk_psock *psock;
2206
2207         rcu_read_lock();
2208         psock = sk_psock(sk);
2209         if (psock)
2210                 ingress_empty = list_empty(&psock->ingress_msg);
2211         rcu_read_unlock();
2212
2213         return !ingress_empty || tls_strp_msg_ready(ctx) ||
2214                 !skb_queue_empty(&ctx->rx_list);
2215 }
2216
2217 int tls_rx_msg_size(struct tls_strparser *strp, struct sk_buff *skb)
2218 {
2219         struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2220         struct tls_prot_info *prot = &tls_ctx->prot_info;
2221         char header[TLS_HEADER_SIZE + MAX_IV_SIZE];
2222         size_t cipher_overhead;
2223         size_t data_len = 0;
2224         int ret;
2225
2226         /* Verify that we have a full TLS header, or wait for more data */
2227         if (strp->stm.offset + prot->prepend_size > skb->len)
2228                 return 0;
2229
2230         /* Sanity-check size of on-stack buffer. */
2231         if (WARN_ON(prot->prepend_size > sizeof(header))) {
2232                 ret = -EINVAL;
2233                 goto read_failure;
2234         }
2235
2236         /* Linearize header to local buffer */
2237         ret = skb_copy_bits(skb, strp->stm.offset, header, prot->prepend_size);
2238         if (ret < 0)
2239                 goto read_failure;
2240
2241         strp->mark = header[0];
2242
2243         data_len = ((header[4] & 0xFF) | (header[3] << 8));
2244
2245         cipher_overhead = prot->tag_size;
2246         if (prot->version != TLS_1_3_VERSION &&
2247             prot->cipher_type != TLS_CIPHER_CHACHA20_POLY1305)
2248                 cipher_overhead += prot->iv_size;
2249
2250         if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
2251             prot->tail_size) {
2252                 ret = -EMSGSIZE;
2253                 goto read_failure;
2254         }
2255         if (data_len < cipher_overhead) {
2256                 ret = -EBADMSG;
2257                 goto read_failure;
2258         }
2259
2260         /* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2261         if (header[1] != TLS_1_2_VERSION_MINOR ||
2262             header[2] != TLS_1_2_VERSION_MAJOR) {
2263                 ret = -EINVAL;
2264                 goto read_failure;
2265         }
2266
2267         tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE,
2268                                      TCP_SKB_CB(skb)->seq + strp->stm.offset);
2269         return data_len + TLS_HEADER_SIZE;
2270
2271 read_failure:
2272         tls_err_abort(strp->sk, ret);
2273
2274         return ret;
2275 }
2276
2277 void tls_rx_msg_ready(struct tls_strparser *strp)
2278 {
2279         struct tls_sw_context_rx *ctx;
2280
2281         ctx = container_of(strp, struct tls_sw_context_rx, strp);
2282         ctx->saved_data_ready(strp->sk);
2283 }
2284
2285 static void tls_data_ready(struct sock *sk)
2286 {
2287         struct tls_context *tls_ctx = tls_get_ctx(sk);
2288         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2289         struct sk_psock *psock;
2290
2291         tls_strp_data_ready(&ctx->strp);
2292
2293         psock = sk_psock_get(sk);
2294         if (psock) {
2295                 if (!list_empty(&psock->ingress_msg))
2296                         ctx->saved_data_ready(sk);
2297                 sk_psock_put(sk, psock);
2298         }
2299 }
2300
2301 void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
2302 {
2303         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2304
2305         set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask);
2306         set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask);
2307         cancel_delayed_work_sync(&ctx->tx_work.work);
2308 }
2309
2310 void tls_sw_release_resources_tx(struct sock *sk)
2311 {
2312         struct tls_context *tls_ctx = tls_get_ctx(sk);
2313         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2314         struct tls_rec *rec, *tmp;
2315         int pending;
2316
2317         /* Wait for any pending async encryptions to complete */
2318         spin_lock_bh(&ctx->encrypt_compl_lock);
2319         ctx->async_notify = true;
2320         pending = atomic_read(&ctx->encrypt_pending);
2321         spin_unlock_bh(&ctx->encrypt_compl_lock);
2322
2323         if (pending)
2324                 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
2325
2326         tls_tx_records(sk, -1);
2327
2328         /* Free up un-sent records in tx_list. First, free
2329          * the partially sent record if any at head of tx_list.
2330          */
2331         if (tls_ctx->partially_sent_record) {
2332                 tls_free_partial_record(sk, tls_ctx);
2333                 rec = list_first_entry(&ctx->tx_list,
2334                                        struct tls_rec, list);
2335                 list_del(&rec->list);
2336                 sk_msg_free(sk, &rec->msg_plaintext);
2337                 kfree(rec);
2338         }
2339
2340         list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
2341                 list_del(&rec->list);
2342                 sk_msg_free(sk, &rec->msg_encrypted);
2343                 sk_msg_free(sk, &rec->msg_plaintext);
2344                 kfree(rec);
2345         }
2346
2347         crypto_free_aead(ctx->aead_send);
2348         tls_free_open_rec(sk);
2349 }
2350
2351 void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
2352 {
2353         struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2354
2355         kfree(ctx);
2356 }
2357
2358 void tls_sw_release_resources_rx(struct sock *sk)
2359 {
2360         struct tls_context *tls_ctx = tls_get_ctx(sk);
2361         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2362
2363         kfree(tls_ctx->rx.rec_seq);
2364         kfree(tls_ctx->rx.iv);
2365
2366         if (ctx->aead_recv) {
2367                 __skb_queue_purge(&ctx->rx_list);
2368                 crypto_free_aead(ctx->aead_recv);
2369                 tls_strp_stop(&ctx->strp);
2370                 /* If tls_sw_strparser_arm() was not called (cleanup paths)
2371                  * we still want to tls_strp_stop(), but sk->sk_data_ready was
2372                  * never swapped.
2373                  */
2374                 if (ctx->saved_data_ready) {
2375                         write_lock_bh(&sk->sk_callback_lock);
2376                         sk->sk_data_ready = ctx->saved_data_ready;
2377                         write_unlock_bh(&sk->sk_callback_lock);
2378                 }
2379         }
2380 }
2381
2382 void tls_sw_strparser_done(struct tls_context *tls_ctx)
2383 {
2384         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2385
2386         tls_strp_done(&ctx->strp);
2387 }
2388
2389 void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
2390 {
2391         struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2392
2393         kfree(ctx);
2394 }
2395
2396 void tls_sw_free_resources_rx(struct sock *sk)
2397 {
2398         struct tls_context *tls_ctx = tls_get_ctx(sk);
2399
2400         tls_sw_release_resources_rx(sk);
2401         tls_sw_free_ctx_rx(tls_ctx);
2402 }
2403
2404 /* The work handler to transmitt the encrypted records in tx_list */
2405 static void tx_work_handler(struct work_struct *work)
2406 {
2407         struct delayed_work *delayed_work = to_delayed_work(work);
2408         struct tx_work *tx_work = container_of(delayed_work,
2409                                                struct tx_work, work);
2410         struct sock *sk = tx_work->sk;
2411         struct tls_context *tls_ctx = tls_get_ctx(sk);
2412         struct tls_sw_context_tx *ctx;
2413
2414         if (unlikely(!tls_ctx))
2415                 return;
2416
2417         ctx = tls_sw_ctx_tx(tls_ctx);
2418         if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
2419                 return;
2420
2421         if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
2422                 return;
2423
2424         if (mutex_trylock(&tls_ctx->tx_lock)) {
2425                 lock_sock(sk);
2426                 tls_tx_records(sk, -1);
2427                 release_sock(sk);
2428                 mutex_unlock(&tls_ctx->tx_lock);
2429         } else if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
2430                 /* Someone is holding the tx_lock, they will likely run Tx
2431                  * and cancel the work on their way out of the lock section.
2432                  * Schedule a long delay just in case.
2433                  */
2434                 schedule_delayed_work(&ctx->tx_work.work, msecs_to_jiffies(10));
2435         }
2436 }
2437
2438 static bool tls_is_tx_ready(struct tls_sw_context_tx *ctx)
2439 {
2440         struct tls_rec *rec;
2441
2442         rec = list_first_entry_or_null(&ctx->tx_list, struct tls_rec, list);
2443         if (!rec)
2444                 return false;
2445
2446         return READ_ONCE(rec->tx_ready);
2447 }
2448
2449 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
2450 {
2451         struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
2452
2453         /* Schedule the transmission if tx list is ready */
2454         if (tls_is_tx_ready(tx_ctx) &&
2455             !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
2456                 schedule_delayed_work(&tx_ctx->tx_work.work, 0);
2457 }
2458
2459 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
2460 {
2461         struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2462
2463         write_lock_bh(&sk->sk_callback_lock);
2464         rx_ctx->saved_data_ready = sk->sk_data_ready;
2465         sk->sk_data_ready = tls_data_ready;
2466         write_unlock_bh(&sk->sk_callback_lock);
2467 }
2468
2469 void tls_update_rx_zc_capable(struct tls_context *tls_ctx)
2470 {
2471         struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2472
2473         rx_ctx->zc_capable = tls_ctx->rx_no_pad ||
2474                 tls_ctx->prot_info.version != TLS_1_3_VERSION;
2475 }
2476
2477 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
2478 {
2479         struct tls_context *tls_ctx = tls_get_ctx(sk);
2480         struct tls_prot_info *prot = &tls_ctx->prot_info;
2481         struct tls_crypto_info *crypto_info;
2482         struct tls_sw_context_tx *sw_ctx_tx = NULL;
2483         struct tls_sw_context_rx *sw_ctx_rx = NULL;
2484         struct cipher_context *cctx;
2485         struct crypto_aead **aead;
2486         u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
2487         struct crypto_tfm *tfm;
2488         char *iv, *rec_seq, *key, *salt, *cipher_name;
2489         size_t keysize;
2490         int rc = 0;
2491
2492         if (!ctx) {
2493                 rc = -EINVAL;
2494                 goto out;
2495         }
2496
2497         if (tx) {
2498                 if (!ctx->priv_ctx_tx) {
2499                         sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
2500                         if (!sw_ctx_tx) {
2501                                 rc = -ENOMEM;
2502                                 goto out;
2503                         }
2504                         ctx->priv_ctx_tx = sw_ctx_tx;
2505                 } else {
2506                         sw_ctx_tx =
2507                                 (struct tls_sw_context_tx *)ctx->priv_ctx_tx;
2508                 }
2509         } else {
2510                 if (!ctx->priv_ctx_rx) {
2511                         sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
2512                         if (!sw_ctx_rx) {
2513                                 rc = -ENOMEM;
2514                                 goto out;
2515                         }
2516                         ctx->priv_ctx_rx = sw_ctx_rx;
2517                 } else {
2518                         sw_ctx_rx =
2519                                 (struct tls_sw_context_rx *)ctx->priv_ctx_rx;
2520                 }
2521         }
2522
2523         if (tx) {
2524                 crypto_init_wait(&sw_ctx_tx->async_wait);
2525                 spin_lock_init(&sw_ctx_tx->encrypt_compl_lock);
2526                 crypto_info = &ctx->crypto_send.info;
2527                 cctx = &ctx->tx;
2528                 aead = &sw_ctx_tx->aead_send;
2529                 INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
2530                 INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2531                 sw_ctx_tx->tx_work.sk = sk;
2532         } else {
2533                 crypto_init_wait(&sw_ctx_rx->async_wait);
2534                 spin_lock_init(&sw_ctx_rx->decrypt_compl_lock);
2535                 init_waitqueue_head(&sw_ctx_rx->wq);
2536                 crypto_info = &ctx->crypto_recv.info;
2537                 cctx = &ctx->rx;
2538                 skb_queue_head_init(&sw_ctx_rx->rx_list);
2539                 skb_queue_head_init(&sw_ctx_rx->async_hold);
2540                 aead = &sw_ctx_rx->aead_recv;
2541         }
2542
2543         switch (crypto_info->cipher_type) {
2544         case TLS_CIPHER_AES_GCM_128: {
2545                 struct tls12_crypto_info_aes_gcm_128 *gcm_128_info;
2546
2547                 gcm_128_info = (void *)crypto_info;
2548                 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2549                 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
2550                 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2551                 iv = gcm_128_info->iv;
2552                 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
2553                 rec_seq = gcm_128_info->rec_seq;
2554                 keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE;
2555                 key = gcm_128_info->key;
2556                 salt = gcm_128_info->salt;
2557                 salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
2558                 cipher_name = "gcm(aes)";
2559                 break;
2560         }
2561         case TLS_CIPHER_AES_GCM_256: {
2562                 struct tls12_crypto_info_aes_gcm_256 *gcm_256_info;
2563
2564                 gcm_256_info = (void *)crypto_info;
2565                 nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2566                 tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE;
2567                 iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2568                 iv = gcm_256_info->iv;
2569                 rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE;
2570                 rec_seq = gcm_256_info->rec_seq;
2571                 keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE;
2572                 key = gcm_256_info->key;
2573                 salt = gcm_256_info->salt;
2574                 salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE;
2575                 cipher_name = "gcm(aes)";
2576                 break;
2577         }
2578         case TLS_CIPHER_AES_CCM_128: {
2579                 struct tls12_crypto_info_aes_ccm_128 *ccm_128_info;
2580
2581                 ccm_128_info = (void *)crypto_info;
2582                 nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2583                 tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE;
2584                 iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2585                 iv = ccm_128_info->iv;
2586                 rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE;
2587                 rec_seq = ccm_128_info->rec_seq;
2588                 keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE;
2589                 key = ccm_128_info->key;
2590                 salt = ccm_128_info->salt;
2591                 salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE;
2592                 cipher_name = "ccm(aes)";
2593                 break;
2594         }
2595         case TLS_CIPHER_CHACHA20_POLY1305: {
2596                 struct tls12_crypto_info_chacha20_poly1305 *chacha20_poly1305_info;
2597
2598                 chacha20_poly1305_info = (void *)crypto_info;
2599                 nonce_size = 0;
2600                 tag_size = TLS_CIPHER_CHACHA20_POLY1305_TAG_SIZE;
2601                 iv_size = TLS_CIPHER_CHACHA20_POLY1305_IV_SIZE;
2602                 iv = chacha20_poly1305_info->iv;
2603                 rec_seq_size = TLS_CIPHER_CHACHA20_POLY1305_REC_SEQ_SIZE;
2604                 rec_seq = chacha20_poly1305_info->rec_seq;
2605                 keysize = TLS_CIPHER_CHACHA20_POLY1305_KEY_SIZE;
2606                 key = chacha20_poly1305_info->key;
2607                 salt = chacha20_poly1305_info->salt;
2608                 salt_size = TLS_CIPHER_CHACHA20_POLY1305_SALT_SIZE;
2609                 cipher_name = "rfc7539(chacha20,poly1305)";
2610                 break;
2611         }
2612         case TLS_CIPHER_SM4_GCM: {
2613                 struct tls12_crypto_info_sm4_gcm *sm4_gcm_info;
2614
2615                 sm4_gcm_info = (void *)crypto_info;
2616                 nonce_size = TLS_CIPHER_SM4_GCM_IV_SIZE;
2617                 tag_size = TLS_CIPHER_SM4_GCM_TAG_SIZE;
2618                 iv_size = TLS_CIPHER_SM4_GCM_IV_SIZE;
2619                 iv = sm4_gcm_info->iv;
2620                 rec_seq_size = TLS_CIPHER_SM4_GCM_REC_SEQ_SIZE;
2621                 rec_seq = sm4_gcm_info->rec_seq;
2622                 keysize = TLS_CIPHER_SM4_GCM_KEY_SIZE;
2623                 key = sm4_gcm_info->key;
2624                 salt = sm4_gcm_info->salt;
2625                 salt_size = TLS_CIPHER_SM4_GCM_SALT_SIZE;
2626                 cipher_name = "gcm(sm4)";
2627                 break;
2628         }
2629         case TLS_CIPHER_SM4_CCM: {
2630                 struct tls12_crypto_info_sm4_ccm *sm4_ccm_info;
2631
2632                 sm4_ccm_info = (void *)crypto_info;
2633                 nonce_size = TLS_CIPHER_SM4_CCM_IV_SIZE;
2634                 tag_size = TLS_CIPHER_SM4_CCM_TAG_SIZE;
2635                 iv_size = TLS_CIPHER_SM4_CCM_IV_SIZE;
2636                 iv = sm4_ccm_info->iv;
2637                 rec_seq_size = TLS_CIPHER_SM4_CCM_REC_SEQ_SIZE;
2638                 rec_seq = sm4_ccm_info->rec_seq;
2639                 keysize = TLS_CIPHER_SM4_CCM_KEY_SIZE;
2640                 key = sm4_ccm_info->key;
2641                 salt = sm4_ccm_info->salt;
2642                 salt_size = TLS_CIPHER_SM4_CCM_SALT_SIZE;
2643                 cipher_name = "ccm(sm4)";
2644                 break;
2645         }
2646         case TLS_CIPHER_ARIA_GCM_128: {
2647                 struct tls12_crypto_info_aria_gcm_128 *aria_gcm_128_info;
2648
2649                 aria_gcm_128_info = (void *)crypto_info;
2650                 nonce_size = TLS_CIPHER_ARIA_GCM_128_IV_SIZE;
2651                 tag_size = TLS_CIPHER_ARIA_GCM_128_TAG_SIZE;
2652                 iv_size = TLS_CIPHER_ARIA_GCM_128_IV_SIZE;
2653                 iv = aria_gcm_128_info->iv;
2654                 rec_seq_size = TLS_CIPHER_ARIA_GCM_128_REC_SEQ_SIZE;
2655                 rec_seq = aria_gcm_128_info->rec_seq;
2656                 keysize = TLS_CIPHER_ARIA_GCM_128_KEY_SIZE;
2657                 key = aria_gcm_128_info->key;
2658                 salt = aria_gcm_128_info->salt;
2659                 salt_size = TLS_CIPHER_ARIA_GCM_128_SALT_SIZE;
2660                 cipher_name = "gcm(aria)";
2661                 break;
2662         }
2663         case TLS_CIPHER_ARIA_GCM_256: {
2664                 struct tls12_crypto_info_aria_gcm_256 *gcm_256_info;
2665
2666                 gcm_256_info = (void *)crypto_info;
2667                 nonce_size = TLS_CIPHER_ARIA_GCM_256_IV_SIZE;
2668                 tag_size = TLS_CIPHER_ARIA_GCM_256_TAG_SIZE;
2669                 iv_size = TLS_CIPHER_ARIA_GCM_256_IV_SIZE;
2670                 iv = gcm_256_info->iv;
2671                 rec_seq_size = TLS_CIPHER_ARIA_GCM_256_REC_SEQ_SIZE;
2672                 rec_seq = gcm_256_info->rec_seq;
2673                 keysize = TLS_CIPHER_ARIA_GCM_256_KEY_SIZE;
2674                 key = gcm_256_info->key;
2675                 salt = gcm_256_info->salt;
2676                 salt_size = TLS_CIPHER_ARIA_GCM_256_SALT_SIZE;
2677                 cipher_name = "gcm(aria)";
2678                 break;
2679         }
2680         default:
2681                 rc = -EINVAL;
2682                 goto free_priv;
2683         }
2684
2685         if (crypto_info->version == TLS_1_3_VERSION) {
2686                 nonce_size = 0;
2687                 prot->aad_size = TLS_HEADER_SIZE;
2688                 prot->tail_size = 1;
2689         } else {
2690                 prot->aad_size = TLS_AAD_SPACE_SIZE;
2691                 prot->tail_size = 0;
2692         }
2693
2694         /* Sanity-check the sizes for stack allocations. */
2695         if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE ||
2696             rec_seq_size > TLS_MAX_REC_SEQ_SIZE || tag_size != TLS_TAG_SIZE ||
2697             prot->aad_size > TLS_MAX_AAD_SIZE) {
2698                 rc = -EINVAL;
2699                 goto free_priv;
2700         }
2701
2702         prot->version = crypto_info->version;
2703         prot->cipher_type = crypto_info->cipher_type;
2704         prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
2705         prot->tag_size = tag_size;
2706         prot->overhead_size = prot->prepend_size +
2707                               prot->tag_size + prot->tail_size;
2708         prot->iv_size = iv_size;
2709         prot->salt_size = salt_size;
2710         cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL);
2711         if (!cctx->iv) {
2712                 rc = -ENOMEM;
2713                 goto free_priv;
2714         }
2715         /* Note: 128 & 256 bit salt are the same size */
2716         prot->rec_seq_size = rec_seq_size;
2717         memcpy(cctx->iv, salt, salt_size);
2718         memcpy(cctx->iv + salt_size, iv, iv_size);
2719         cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
2720         if (!cctx->rec_seq) {
2721                 rc = -ENOMEM;
2722                 goto free_iv;
2723         }
2724
2725         if (!*aead) {
2726                 *aead = crypto_alloc_aead(cipher_name, 0, 0);
2727                 if (IS_ERR(*aead)) {
2728                         rc = PTR_ERR(*aead);
2729                         *aead = NULL;
2730                         goto free_rec_seq;
2731                 }
2732         }
2733
2734         ctx->push_pending_record = tls_sw_push_pending_record;
2735
2736         rc = crypto_aead_setkey(*aead, key, keysize);
2737
2738         if (rc)
2739                 goto free_aead;
2740
2741         rc = crypto_aead_setauthsize(*aead, prot->tag_size);
2742         if (rc)
2743                 goto free_aead;
2744
2745         if (sw_ctx_rx) {
2746                 tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
2747
2748                 tls_update_rx_zc_capable(ctx);
2749                 sw_ctx_rx->async_capable =
2750                         crypto_info->version != TLS_1_3_VERSION &&
2751                         !!(tfm->__crt_alg->cra_flags & CRYPTO_ALG_ASYNC);
2752
2753                 rc = tls_strp_init(&sw_ctx_rx->strp, sk);
2754                 if (rc)
2755                         goto free_aead;
2756         }
2757
2758         goto out;
2759
2760 free_aead:
2761         crypto_free_aead(*aead);
2762         *aead = NULL;
2763 free_rec_seq:
2764         kfree(cctx->rec_seq);
2765         cctx->rec_seq = NULL;
2766 free_iv:
2767         kfree(cctx->iv);
2768         cctx->iv = NULL;
2769 free_priv:
2770         if (tx) {
2771                 kfree(ctx->priv_ctx_tx);
2772                 ctx->priv_ctx_tx = NULL;
2773         } else {
2774                 kfree(ctx->priv_ctx_rx);
2775                 ctx->priv_ctx_rx = NULL;
2776         }
2777 out:
2778         return rc;
2779 }