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