net/tls/tls_device.c

   1 /* Copyright (c) 2018, Mellanox Technologies All rights reserved.
   2  *
   3  * This software is available to you under a choice of one of two
   4  * licenses.  You may choose to be licensed under the terms of the GNU
   5  * General Public License (GPL) Version 2, available from the file
   6  * COPYING in the main directory of this source tree, or the
   7  * OpenIB.org BSD license below:
   8  *
   9  *     Redistribution and use in source and binary forms, with or
  10  *     without modification, are permitted provided that the following
  11  *     conditions are met:
  12  *
  13  *      - Redistributions of source code must retain the above
  14  *        copyright notice, this list of conditions and the following
  15  *        disclaimer.
  16  *
  17  *      - Redistributions in binary form must reproduce the above
  18  *        copyright notice, this list of conditions and the following
  19  *        disclaimer in the documentation and/or other materials
  20  *        provided with the distribution.
  21  *
  22  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  23  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  24  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  25  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  26  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  27  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  28  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  29  * SOFTWARE.
  30  */
  31
  32 #include <crypto/aead.h>
  33 #include <linux/highmem.h>
  34 #include <linux/module.h>
  35 #include <linux/netdevice.h>
  36 #include <net/dst.h>
  37 #include <net/inet_connection_sock.h>
  38 #include <net/tcp.h>
  39 #include <net/tls.h>
  40 #include <linux/skbuff_ref.h>
  41
  42 #include "tls.h"
  43 #include "trace.h"
  44
  45 /* device_offload_lock is used to synchronize tls_dev_add
  46  * against NETDEV_DOWN notifications.
  47  */
  48 static DECLARE_RWSEM(device_offload_lock);
  49
  50 static struct workqueue_struct *destruct_wq __read_mostly;
  51
  52 static LIST_HEAD(tls_device_list);
  53 static LIST_HEAD(tls_device_down_list);
  54 static DEFINE_SPINLOCK(tls_device_lock);
  55
  56 static struct page *dummy_page;
  57
  58 static void tls_device_free_ctx(struct tls_context *ctx)
  59 {
  60         if (ctx->tx_conf == TLS_HW)
  61                 kfree(tls_offload_ctx_tx(ctx));
  62
  63         if (ctx->rx_conf == TLS_HW)
  64                 kfree(tls_offload_ctx_rx(ctx));
  65
  66         tls_ctx_free(NULL, ctx);
  67 }
  68
  69 static void tls_device_tx_del_task(struct work_struct *work)
  70 {
  71         struct tls_offload_context_tx *offload_ctx =
  72                 container_of(work, struct tls_offload_context_tx, destruct_work);
  73         struct tls_context *ctx = offload_ctx->ctx;
  74         struct net_device *netdev;
  75
  76         /* Safe, because this is the destroy flow, refcount is 0, so
  77          * tls_device_down can't store this field in parallel.
  78          */
  79         netdev = rcu_dereference_protected(ctx->netdev,
  80                                            !refcount_read(&ctx->refcount));
  81
  82         netdev->tlsdev_ops->tls_dev_del(netdev, ctx, TLS_OFFLOAD_CTX_DIR_TX);
  83         dev_put(netdev);
  84         ctx->netdev = NULL;
  85         tls_device_free_ctx(ctx);
  86 }
  87
  88 static void tls_device_queue_ctx_destruction(struct tls_context *ctx)
  89 {
  90         struct net_device *netdev;
  91         unsigned long flags;
  92         bool async_cleanup;
  93
  94         spin_lock_irqsave(&tls_device_lock, flags);
  95         if (unlikely(!refcount_dec_and_test(&ctx->refcount))) {
  96                 spin_unlock_irqrestore(&tls_device_lock, flags);
  97                 return;
  98         }
  99
 100         list_del(&ctx->list); /* Remove from tls_device_list / tls_device_down_list */
 101
 102         /* Safe, because this is the destroy flow, refcount is 0, so
 103          * tls_device_down can't store this field in parallel.
 104          */
 105         netdev = rcu_dereference_protected(ctx->netdev,
 106                                            !refcount_read(&ctx->refcount));
 107
 108         async_cleanup = netdev && ctx->tx_conf == TLS_HW;
 109         if (async_cleanup) {
 110                 struct tls_offload_context_tx *offload_ctx = tls_offload_ctx_tx(ctx);
 111
 112                 /* queue_work inside the spinlock
 113                  * to make sure tls_device_down waits for that work.
 114                  */
 115                 queue_work(destruct_wq, &offload_ctx->destruct_work);
 116         }
 117         spin_unlock_irqrestore(&tls_device_lock, flags);
 118
 119         if (!async_cleanup)
 120                 tls_device_free_ctx(ctx);
 121 }
 122
 123 /* We assume that the socket is already connected */
 124 static struct net_device *get_netdev_for_sock(struct sock *sk)
 125 {
 126         struct dst_entry *dst = sk_dst_get(sk);
 127         struct net_device *netdev = NULL;
 128
 129         if (likely(dst)) {
 130                 netdev = netdev_sk_get_lowest_dev(dst->dev, sk);
 131                 dev_hold(netdev);
 132         }
 133
 134         dst_release(dst);
 135
 136         return netdev;
 137 }
 138
 139 static void destroy_record(struct tls_record_info *record)
 140 {
 141         int i;
 142
 143         for (i = 0; i < record->num_frags; i++)
 144                 __skb_frag_unref(&record->frags[i], false);
 145         kfree(record);
 146 }
 147
 148 static void delete_all_records(struct tls_offload_context_tx *offload_ctx)
 149 {
 150         struct tls_record_info *info, *temp;
 151
 152         list_for_each_entry_safe(info, temp, &offload_ctx->records_list, list) {
 153                 list_del(&info->list);
 154                 destroy_record(info);
 155         }
 156
 157         offload_ctx->retransmit_hint = NULL;
 158 }
 159
 160 static void tls_icsk_clean_acked(struct sock *sk, u32 acked_seq)
 161 {
 162         struct tls_context *tls_ctx = tls_get_ctx(sk);
 163         struct tls_record_info *info, *temp;
 164         struct tls_offload_context_tx *ctx;
 165         u64 deleted_records = 0;
 166         unsigned long flags;
 167
 168         if (!tls_ctx)
 169                 return;
 170
 171         ctx = tls_offload_ctx_tx(tls_ctx);
 172
 173         spin_lock_irqsave(&ctx->lock, flags);
 174         info = ctx->retransmit_hint;
 175         if (info && !before(acked_seq, info->end_seq))
 176                 ctx->retransmit_hint = NULL;
 177
 178         list_for_each_entry_safe(info, temp, &ctx->records_list, list) {
 179                 if (before(acked_seq, info->end_seq))
 180                         break;
 181                 list_del(&info->list);
 182
 183                 destroy_record(info);
 184                 deleted_records++;
 185         }
 186
 187         ctx->unacked_record_sn += deleted_records;
 188         spin_unlock_irqrestore(&ctx->lock, flags);
 189 }
 190
 191 /* At this point, there should be no references on this
 192  * socket and no in-flight SKBs associated with this
 193  * socket, so it is safe to free all the resources.
 194  */
 195 void tls_device_sk_destruct(struct sock *sk)
 196 {
 197         struct tls_context *tls_ctx = tls_get_ctx(sk);
 198         struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
 199
 200         tls_ctx->sk_destruct(sk);
 201
 202         if (tls_ctx->tx_conf == TLS_HW) {
 203                 if (ctx->open_record)
 204                         destroy_record(ctx->open_record);
 205                 delete_all_records(ctx);
 206                 crypto_free_aead(ctx->aead_send);
 207                 clean_acked_data_disable(inet_csk(sk));
 208         }
 209
 210         tls_device_queue_ctx_destruction(tls_ctx);
 211 }
 212 EXPORT_SYMBOL_GPL(tls_device_sk_destruct);
 213
 214 void tls_device_free_resources_tx(struct sock *sk)
 215 {
 216         struct tls_context *tls_ctx = tls_get_ctx(sk);
 217
 218         tls_free_partial_record(sk, tls_ctx);
 219 }
 220
 221 void tls_offload_tx_resync_request(struct sock *sk, u32 got_seq, u32 exp_seq)
 222 {
 223         struct tls_context *tls_ctx = tls_get_ctx(sk);
 224
 225         trace_tls_device_tx_resync_req(sk, got_seq, exp_seq);
 226         WARN_ON(test_and_set_bit(TLS_TX_SYNC_SCHED, &tls_ctx->flags));
 227 }
 228 EXPORT_SYMBOL_GPL(tls_offload_tx_resync_request);
 229
 230 static void tls_device_resync_tx(struct sock *sk, struct tls_context *tls_ctx,
 231                                  u32 seq)
 232 {
 233         struct net_device *netdev;
 234         int err = 0;
 235         u8 *rcd_sn;
 236
 237         tcp_write_collapse_fence(sk);
 238         rcd_sn = tls_ctx->tx.rec_seq;
 239
 240         trace_tls_device_tx_resync_send(sk, seq, rcd_sn);
 241         down_read(&device_offload_lock);
 242         netdev = rcu_dereference_protected(tls_ctx->netdev,
 243                                            lockdep_is_held(&device_offload_lock));
 244         if (netdev)
 245                 err = netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq,
 246                                                          rcd_sn,
 247                                                          TLS_OFFLOAD_CTX_DIR_TX);
 248         up_read(&device_offload_lock);
 249         if (err)
 250                 return;
 251
 252         clear_bit_unlock(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
 253 }
 254
 255 static void tls_append_frag(struct tls_record_info *record,
 256                             struct page_frag *pfrag,
 257                             int size)
 258 {
 259         skb_frag_t *frag;
 260
 261         frag = &record->frags[record->num_frags - 1];
 262         if (skb_frag_page(frag) == pfrag->page &&
 263             skb_frag_off(frag) + skb_frag_size(frag) == pfrag->offset) {
 264                 skb_frag_size_add(frag, size);
 265         } else {
 266                 ++frag;
 267                 skb_frag_fill_page_desc(frag, pfrag->page, pfrag->offset,
 268                                         size);
 269                 ++record->num_frags;
 270                 get_page(pfrag->page);
 271         }
 272
 273         pfrag->offset += size;
 274         record->len += size;
 275 }
 276
 277 static int tls_push_record(struct sock *sk,
 278                            struct tls_context *ctx,
 279                            struct tls_offload_context_tx *offload_ctx,
 280                            struct tls_record_info *record,
 281                            int flags)
 282 {
 283         struct tls_prot_info *prot = &ctx->prot_info;
 284         struct tcp_sock *tp = tcp_sk(sk);
 285         skb_frag_t *frag;
 286         int i;
 287
 288         record->end_seq = tp->write_seq + record->len;
 289         list_add_tail_rcu(&record->list, &offload_ctx->records_list);
 290         offload_ctx->open_record = NULL;
 291
 292         if (test_bit(TLS_TX_SYNC_SCHED, &ctx->flags))
 293                 tls_device_resync_tx(sk, ctx, tp->write_seq);
 294
 295         tls_advance_record_sn(sk, prot, &ctx->tx);
 296
 297         for (i = 0; i < record->num_frags; i++) {
 298                 frag = &record->frags[i];
 299                 sg_unmark_end(&offload_ctx->sg_tx_data[i]);
 300                 sg_set_page(&offload_ctx->sg_tx_data[i], skb_frag_page(frag),
 301                             skb_frag_size(frag), skb_frag_off(frag));
 302                 sk_mem_charge(sk, skb_frag_size(frag));
 303                 get_page(skb_frag_page(frag));
 304         }
 305         sg_mark_end(&offload_ctx->sg_tx_data[record->num_frags - 1]);
 306
 307         /* all ready, send */
 308         return tls_push_sg(sk, ctx, offload_ctx->sg_tx_data, 0, flags);
 309 }
 310
 311 static void tls_device_record_close(struct sock *sk,
 312                                     struct tls_context *ctx,
 313                                     struct tls_record_info *record,
 314                                     struct page_frag *pfrag,
 315                                     unsigned char record_type)
 316 {
 317         struct tls_prot_info *prot = &ctx->prot_info;
 318         struct page_frag dummy_tag_frag;
 319
 320         /* append tag
 321          * device will fill in the tag, we just need to append a placeholder
 322          * use socket memory to improve coalescing (re-using a single buffer
 323          * increases frag count)
 324          * if we can't allocate memory now use the dummy page
 325          */
 326         if (unlikely(pfrag->size - pfrag->offset < prot->tag_size) &&
 327             !skb_page_frag_refill(prot->tag_size, pfrag, sk->sk_allocation)) {
 328                 dummy_tag_frag.page = dummy_page;
 329                 dummy_tag_frag.offset = 0;
 330                 pfrag = &dummy_tag_frag;
 331         }
 332         tls_append_frag(record, pfrag, prot->tag_size);
 333
 334         /* fill prepend */
 335         tls_fill_prepend(ctx, skb_frag_address(&record->frags[0]),
 336                          record->len - prot->overhead_size,
 337                          record_type);
 338 }
 339
 340 static int tls_create_new_record(struct tls_offload_context_tx *offload_ctx,
 341                                  struct page_frag *pfrag,
 342                                  size_t prepend_size)
 343 {
 344         struct tls_record_info *record;
 345         skb_frag_t *frag;
 346
 347         record = kmalloc(sizeof(*record), GFP_KERNEL);
 348         if (!record)
 349                 return -ENOMEM;
 350
 351         frag = &record->frags[0];
 352         skb_frag_fill_page_desc(frag, pfrag->page, pfrag->offset,
 353                                 prepend_size);
 354
 355         get_page(pfrag->page);
 356         pfrag->offset += prepend_size;
 357
 358         record->num_frags = 1;
 359         record->len = prepend_size;
 360         offload_ctx->open_record = record;
 361         return 0;
 362 }
 363
 364 static int tls_do_allocation(struct sock *sk,
 365                              struct tls_offload_context_tx *offload_ctx,
 366                              struct page_frag *pfrag,
 367                              size_t prepend_size)
 368 {
 369         int ret;
 370
 371         if (!offload_ctx->open_record) {
 372                 if (unlikely(!skb_page_frag_refill(prepend_size, pfrag,
 373                                                    sk->sk_allocation))) {
 374                         READ_ONCE(sk->sk_prot)->enter_memory_pressure(sk);
 375                         sk_stream_moderate_sndbuf(sk);
 376                         return -ENOMEM;
 377                 }
 378
 379                 ret = tls_create_new_record(offload_ctx, pfrag, prepend_size);
 380                 if (ret)
 381                         return ret;
 382
 383                 if (pfrag->size > pfrag->offset)
 384                         return 0;
 385         }
 386
 387         if (!sk_page_frag_refill(sk, pfrag))
 388                 return -ENOMEM;
 389
 390         return 0;
 391 }
 392
 393 static int tls_device_copy_data(void *addr, size_t bytes, struct iov_iter *i)
 394 {
 395         size_t pre_copy, nocache;
 396
 397         pre_copy = ~((unsigned long)addr - 1) & (SMP_CACHE_BYTES - 1);
 398         if (pre_copy) {
 399                 pre_copy = min(pre_copy, bytes);
 400                 if (copy_from_iter(addr, pre_copy, i) != pre_copy)
 401                         return -EFAULT;
 402                 bytes -= pre_copy;
 403                 addr += pre_copy;
 404         }
 405
 406         nocache = round_down(bytes, SMP_CACHE_BYTES);
 407         if (copy_from_iter_nocache(addr, nocache, i) != nocache)
 408                 return -EFAULT;
 409         bytes -= nocache;
 410         addr += nocache;
 411
 412         if (bytes && copy_from_iter(addr, bytes, i) != bytes)
 413                 return -EFAULT;
 414
 415         return 0;
 416 }
 417
 418 static int tls_push_data(struct sock *sk,
 419                          struct iov_iter *iter,
 420                          size_t size, int flags,
 421                          unsigned char record_type)
 422 {
 423         struct tls_context *tls_ctx = tls_get_ctx(sk);
 424         struct tls_prot_info *prot = &tls_ctx->prot_info;
 425         struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
 426         struct tls_record_info *record;
 427         int tls_push_record_flags;
 428         struct page_frag *pfrag;
 429         size_t orig_size = size;
 430         u32 max_open_record_len;
 431         bool more = false;
 432         bool done = false;
 433         int copy, rc = 0;
 434         long timeo;
 435
 436         if (flags &
 437             ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
 438               MSG_SPLICE_PAGES | MSG_EOR))
 439                 return -EOPNOTSUPP;
 440
 441         if ((flags & (MSG_MORE | MSG_EOR)) == (MSG_MORE | MSG_EOR))
 442                 return -EINVAL;
 443
 444         if (unlikely(sk->sk_err))
 445                 return -sk->sk_err;
 446
 447         flags |= MSG_SENDPAGE_DECRYPTED;
 448         tls_push_record_flags = flags | MSG_MORE;
 449
 450         timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
 451         if (tls_is_partially_sent_record(tls_ctx)) {
 452                 rc = tls_push_partial_record(sk, tls_ctx, flags);
 453                 if (rc < 0)
 454                         return rc;
 455         }
 456
 457         pfrag = sk_page_frag(sk);
 458
 459         /* TLS_HEADER_SIZE is not counted as part of the TLS record, and
 460          * we need to leave room for an authentication tag.
 461          */
 462         max_open_record_len = TLS_MAX_PAYLOAD_SIZE +
 463                               prot->prepend_size;
 464         do {
 465                 rc = tls_do_allocation(sk, ctx, pfrag, prot->prepend_size);
 466                 if (unlikely(rc)) {
 467                         rc = sk_stream_wait_memory(sk, &timeo);
 468                         if (!rc)
 469                                 continue;
 470
 471                         record = ctx->open_record;
 472                         if (!record)
 473                                 break;
 474 handle_error:
 475                         if (record_type != TLS_RECORD_TYPE_DATA) {
 476                                 /* avoid sending partial
 477                                  * record with type !=
 478                                  * application_data
 479                                  */
 480                                 size = orig_size;
 481                                 destroy_record(record);
 482                                 ctx->open_record = NULL;
 483                         } else if (record->len > prot->prepend_size) {
 484                                 goto last_record;
 485                         }
 486
 487                         break;
 488                 }
 489
 490                 record = ctx->open_record;
 491
 492                 copy = min_t(size_t, size, max_open_record_len - record->len);
 493                 if (copy && (flags & MSG_SPLICE_PAGES)) {
 494                         struct page_frag zc_pfrag;
 495                         struct page **pages = &zc_pfrag.page;
 496                         size_t off;
 497
 498                         rc = iov_iter_extract_pages(iter, &pages,
 499                                                     copy, 1, 0, &off);
 500                         if (rc <= 0) {
 501                                 if (rc == 0)
 502                                         rc = -EIO;
 503                                 goto handle_error;
 504                         }
 505                         copy = rc;
 506
 507                         if (WARN_ON_ONCE(!sendpage_ok(zc_pfrag.page))) {
 508                                 iov_iter_revert(iter, copy);
 509                                 rc = -EIO;
 510                                 goto handle_error;
 511                         }
 512
 513                         zc_pfrag.offset = off;
 514                         zc_pfrag.size = copy;
 515                         tls_append_frag(record, &zc_pfrag, copy);
 516                 } else if (copy) {
 517                         copy = min_t(size_t, copy, pfrag->size - pfrag->offset);
 518
 519                         rc = tls_device_copy_data(page_address(pfrag->page) +
 520                                                   pfrag->offset, copy,
 521                                                   iter);
 522                         if (rc)
 523                                 goto handle_error;
 524                         tls_append_frag(record, pfrag, copy);
 525                 }
 526
 527                 size -= copy;
 528                 if (!size) {
 529 last_record:
 530                         tls_push_record_flags = flags;
 531                         if (flags & MSG_MORE) {
 532                                 more = true;
 533                                 break;
 534                         }
 535
 536                         done = true;
 537                 }
 538
 539                 if (done || record->len >= max_open_record_len ||
 540                     (record->num_frags >= MAX_SKB_FRAGS - 1)) {
 541                         tls_device_record_close(sk, tls_ctx, record,
 542                                                 pfrag, record_type);
 543
 544                         rc = tls_push_record(sk,
 545                                              tls_ctx,
 546                                              ctx,
 547                                              record,
 548                                              tls_push_record_flags);
 549                         if (rc < 0)
 550                                 break;
 551                 }
 552         } while (!done);
 553
 554         tls_ctx->pending_open_record_frags = more;
 555
 556         if (orig_size - size > 0)
 557                 rc = orig_size - size;
 558
 559         return rc;
 560 }
 561
 562 int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
 563 {
 564         unsigned char record_type = TLS_RECORD_TYPE_DATA;
 565         struct tls_context *tls_ctx = tls_get_ctx(sk);
 566         int rc;
 567
 568         if (!tls_ctx->zerocopy_sendfile)
 569                 msg->msg_flags &= ~MSG_SPLICE_PAGES;
 570
 571         mutex_lock(&tls_ctx->tx_lock);
 572         lock_sock(sk);
 573
 574         if (unlikely(msg->msg_controllen)) {
 575                 rc = tls_process_cmsg(sk, msg, &record_type);
 576                 if (rc)
 577                         goto out;
 578         }
 579
 580         rc = tls_push_data(sk, &msg->msg_iter, size, msg->msg_flags,
 581                            record_type);
 582
 583 out:
 584         release_sock(sk);
 585         mutex_unlock(&tls_ctx->tx_lock);
 586         return rc;
 587 }
 588
 589 void tls_device_splice_eof(struct socket *sock)
 590 {
 591         struct sock *sk = sock->sk;
 592         struct tls_context *tls_ctx = tls_get_ctx(sk);
 593         struct iov_iter iter = {};
 594
 595         if (!tls_is_partially_sent_record(tls_ctx))
 596                 return;
 597
 598         mutex_lock(&tls_ctx->tx_lock);
 599         lock_sock(sk);
 600
 601         if (tls_is_partially_sent_record(tls_ctx)) {
 602                 iov_iter_bvec(&iter, ITER_SOURCE, NULL, 0, 0);
 603                 tls_push_data(sk, &iter, 0, 0, TLS_RECORD_TYPE_DATA);
 604         }
 605
 606         release_sock(sk);
 607         mutex_unlock(&tls_ctx->tx_lock);
 608 }
 609
 610 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
 611                                        u32 seq, u64 *p_record_sn)
 612 {
 613         u64 record_sn = context->hint_record_sn;
 614         struct tls_record_info *info, *last;
 615
 616         info = context->retransmit_hint;
 617         if (!info ||
 618             before(seq, info->end_seq - info->len)) {
 619                 /* if retransmit_hint is irrelevant start
 620                  * from the beginning of the list
 621                  */
 622                 info = list_first_entry_or_null(&context->records_list,
 623                                                 struct tls_record_info, list);
 624                 if (!info)
 625                         return NULL;
 626                 /* send the start_marker record if seq number is before the
 627                  * tls offload start marker sequence number. This record is
 628                  * required to handle TCP packets which are before TLS offload
 629                  * started.
 630                  *  And if it's not start marker, look if this seq number
 631                  * belongs to the list.
 632                  */
 633                 if (likely(!tls_record_is_start_marker(info))) {
 634                         /* we have the first record, get the last record to see
 635                          * if this seq number belongs to the list.
 636                          */
 637                         last = list_last_entry(&context->records_list,
 638                                                struct tls_record_info, list);
 639
 640                         if (!between(seq, tls_record_start_seq(info),
 641                                      last->end_seq))
 642                                 return NULL;
 643                 }
 644                 record_sn = context->unacked_record_sn;
 645         }
 646
 647         /* We just need the _rcu for the READ_ONCE() */
 648         rcu_read_lock();
 649         list_for_each_entry_from_rcu(info, &context->records_list, list) {
 650                 if (before(seq, info->end_seq)) {
 651                         if (!context->retransmit_hint ||
 652                             after(info->end_seq,
 653                                   context->retransmit_hint->end_seq)) {
 654                                 context->hint_record_sn = record_sn;
 655                                 context->retransmit_hint = info;
 656                         }
 657                         *p_record_sn = record_sn;
 658                         goto exit_rcu_unlock;
 659                 }
 660                 record_sn++;
 661         }
 662         info = NULL;
 663
 664 exit_rcu_unlock:
 665         rcu_read_unlock();
 666         return info;
 667 }
 668 EXPORT_SYMBOL(tls_get_record);
 669
 670 static int tls_device_push_pending_record(struct sock *sk, int flags)
 671 {
 672         struct iov_iter iter;
 673
 674         iov_iter_kvec(&iter, ITER_SOURCE, NULL, 0, 0);
 675         return tls_push_data(sk, &iter, 0, flags, TLS_RECORD_TYPE_DATA);
 676 }
 677
 678 void tls_device_write_space(struct sock *sk, struct tls_context *ctx)
 679 {
 680         if (tls_is_partially_sent_record(ctx)) {
 681                 gfp_t sk_allocation = sk->sk_allocation;
 682
 683                 WARN_ON_ONCE(sk->sk_write_pending);
 684
 685                 sk->sk_allocation = GFP_ATOMIC;
 686                 tls_push_partial_record(sk, ctx,
 687                                         MSG_DONTWAIT | MSG_NOSIGNAL |
 688                                         MSG_SENDPAGE_DECRYPTED);
 689                 sk->sk_allocation = sk_allocation;
 690         }
 691 }
 692
 693 static void tls_device_resync_rx(struct tls_context *tls_ctx,
 694                                  struct sock *sk, u32 seq, u8 *rcd_sn)
 695 {
 696         struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
 697         struct net_device *netdev;
 698
 699         trace_tls_device_rx_resync_send(sk, seq, rcd_sn, rx_ctx->resync_type);
 700         rcu_read_lock();
 701         netdev = rcu_dereference(tls_ctx->netdev);
 702         if (netdev)
 703                 netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq, rcd_sn,
 704                                                    TLS_OFFLOAD_CTX_DIR_RX);
 705         rcu_read_unlock();
 706         TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXDEVICERESYNC);
 707 }
 708
 709 static bool
 710 tls_device_rx_resync_async(struct tls_offload_resync_async *resync_async,
 711                            s64 resync_req, u32 *seq, u16 *rcd_delta)
 712 {
 713         u32 is_async = resync_req & RESYNC_REQ_ASYNC;
 714         u32 req_seq = resync_req >> 32;
 715         u32 req_end = req_seq + ((resync_req >> 16) & 0xffff);
 716         u16 i;
 717
 718         *rcd_delta = 0;
 719
 720         if (is_async) {
 721                 /* shouldn't get to wraparound:
 722                  * too long in async stage, something bad happened
 723                  */
 724                 if (WARN_ON_ONCE(resync_async->rcd_delta == USHRT_MAX))
 725                         return false;
 726
 727                 /* asynchronous stage: log all headers seq such that
 728                  * req_seq <= seq <= end_seq, and wait for real resync request
 729                  */
 730                 if (before(*seq, req_seq))
 731                         return false;
 732                 if (!after(*seq, req_end) &&
 733                     resync_async->loglen < TLS_DEVICE_RESYNC_ASYNC_LOGMAX)
 734                         resync_async->log[resync_async->loglen++] = *seq;
 735
 736                 resync_async->rcd_delta++;
 737
 738                 return false;
 739         }
 740
 741         /* synchronous stage: check against the logged entries and
 742          * proceed to check the next entries if no match was found
 743          */
 744         for (i = 0; i < resync_async->loglen; i++)
 745                 if (req_seq == resync_async->log[i] &&
 746                     atomic64_try_cmpxchg(&resync_async->req, &resync_req, 0)) {
 747                         *rcd_delta = resync_async->rcd_delta - i;
 748                         *seq = req_seq;
 749                         resync_async->loglen = 0;
 750                         resync_async->rcd_delta = 0;
 751                         return true;
 752                 }
 753
 754         resync_async->loglen = 0;
 755         resync_async->rcd_delta = 0;
 756
 757         if (req_seq == *seq &&
 758             atomic64_try_cmpxchg(&resync_async->req,
 759                                  &resync_req, 0))
 760                 return true;
 761
 762         return false;
 763 }
 764
 765 void tls_device_rx_resync_new_rec(struct sock *sk, u32 rcd_len, u32 seq)
 766 {
 767         struct tls_context *tls_ctx = tls_get_ctx(sk);
 768         struct tls_offload_context_rx *rx_ctx;
 769         u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
 770         u32 sock_data, is_req_pending;
 771         struct tls_prot_info *prot;
 772         s64 resync_req;
 773         u16 rcd_delta;
 774         u32 req_seq;
 775
 776         if (tls_ctx->rx_conf != TLS_HW)
 777                 return;
 778         if (unlikely(test_bit(TLS_RX_DEV_DEGRADED, &tls_ctx->flags)))
 779                 return;
 780
 781         prot = &tls_ctx->prot_info;
 782         rx_ctx = tls_offload_ctx_rx(tls_ctx);
 783         memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
 784
 785         switch (rx_ctx->resync_type) {
 786         case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ:
 787                 resync_req = atomic64_read(&rx_ctx->resync_req);
 788                 req_seq = resync_req >> 32;
 789                 seq += TLS_HEADER_SIZE - 1;
 790                 is_req_pending = resync_req;
 791
 792                 if (likely(!is_req_pending) || req_seq != seq ||
 793                     !atomic64_try_cmpxchg(&rx_ctx->resync_req, &resync_req, 0))
 794                         return;
 795                 break;
 796         case TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT:
 797                 if (likely(!rx_ctx->resync_nh_do_now))
 798                         return;
 799
 800                 /* head of next rec is already in, note that the sock_inq will
 801                  * include the currently parsed message when called from parser
 802                  */
 803                 sock_data = tcp_inq(sk);
 804                 if (sock_data > rcd_len) {
 805                         trace_tls_device_rx_resync_nh_delay(sk, sock_data,
 806                                                             rcd_len);
 807                         return;
 808                 }
 809
 810                 rx_ctx->resync_nh_do_now = 0;
 811                 seq += rcd_len;
 812                 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
 813                 break;
 814         case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC:
 815                 resync_req = atomic64_read(&rx_ctx->resync_async->req);
 816                 is_req_pending = resync_req;
 817                 if (likely(!is_req_pending))
 818                         return;
 819
 820                 if (!tls_device_rx_resync_async(rx_ctx->resync_async,
 821                                                 resync_req, &seq, &rcd_delta))
 822                         return;
 823                 tls_bigint_subtract(rcd_sn, rcd_delta);
 824                 break;
 825         }
 826
 827         tls_device_resync_rx(tls_ctx, sk, seq, rcd_sn);
 828 }
 829
 830 static void tls_device_core_ctrl_rx_resync(struct tls_context *tls_ctx,
 831                                            struct tls_offload_context_rx *ctx,
 832                                            struct sock *sk, struct sk_buff *skb)
 833 {
 834         struct strp_msg *rxm;
 835
 836         /* device will request resyncs by itself based on stream scan */
 837         if (ctx->resync_type != TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT)
 838                 return;
 839         /* already scheduled */
 840         if (ctx->resync_nh_do_now)
 841                 return;
 842         /* seen decrypted fragments since last fully-failed record */
 843         if (ctx->resync_nh_reset) {
 844                 ctx->resync_nh_reset = 0;
 845                 ctx->resync_nh.decrypted_failed = 1;
 846                 ctx->resync_nh.decrypted_tgt = TLS_DEVICE_RESYNC_NH_START_IVAL;
 847                 return;
 848         }
 849
 850         if (++ctx->resync_nh.decrypted_failed <= ctx->resync_nh.decrypted_tgt)
 851                 return;
 852
 853         /* doing resync, bump the next target in case it fails */
 854         if (ctx->resync_nh.decrypted_tgt < TLS_DEVICE_RESYNC_NH_MAX_IVAL)
 855                 ctx->resync_nh.decrypted_tgt *= 2;
 856         else
 857                 ctx->resync_nh.decrypted_tgt += TLS_DEVICE_RESYNC_NH_MAX_IVAL;
 858
 859         rxm = strp_msg(skb);
 860
 861         /* head of next rec is already in, parser will sync for us */
 862         if (tcp_inq(sk) > rxm->full_len) {
 863                 trace_tls_device_rx_resync_nh_schedule(sk);
 864                 ctx->resync_nh_do_now = 1;
 865         } else {
 866                 struct tls_prot_info *prot = &tls_ctx->prot_info;
 867                 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
 868
 869                 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
 870                 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
 871
 872                 tls_device_resync_rx(tls_ctx, sk, tcp_sk(sk)->copied_seq,
 873                                      rcd_sn);
 874         }
 875 }
 876
 877 static int
 878 tls_device_reencrypt(struct sock *sk, struct tls_context *tls_ctx)
 879 {
 880         struct tls_sw_context_rx *sw_ctx = tls_sw_ctx_rx(tls_ctx);
 881         const struct tls_cipher_desc *cipher_desc;
 882         int err, offset, copy, data_len, pos;
 883         struct sk_buff *skb, *skb_iter;
 884         struct scatterlist sg[1];
 885         struct strp_msg *rxm;
 886         char *orig_buf, *buf;
 887
 888         cipher_desc = get_cipher_desc(tls_ctx->crypto_recv.info.cipher_type);
 889         DEBUG_NET_WARN_ON_ONCE(!cipher_desc || !cipher_desc->offloadable);
 890
 891         rxm = strp_msg(tls_strp_msg(sw_ctx));
 892         orig_buf = kmalloc(rxm->full_len + TLS_HEADER_SIZE + cipher_desc->iv,
 893                            sk->sk_allocation);
 894         if (!orig_buf)
 895                 return -ENOMEM;
 896         buf = orig_buf;
 897
 898         err = tls_strp_msg_cow(sw_ctx);
 899         if (unlikely(err))
 900                 goto free_buf;
 901
 902         skb = tls_strp_msg(sw_ctx);
 903         rxm = strp_msg(skb);
 904         offset = rxm->offset;
 905
 906         sg_init_table(sg, 1);
 907         sg_set_buf(&sg[0], buf,
 908                    rxm->full_len + TLS_HEADER_SIZE + cipher_desc->iv);
 909         err = skb_copy_bits(skb, offset, buf, TLS_HEADER_SIZE + cipher_desc->iv);
 910         if (err)
 911                 goto free_buf;
 912
 913         /* We are interested only in the decrypted data not the auth */
 914         err = decrypt_skb(sk, sg);
 915         if (err != -EBADMSG)
 916                 goto free_buf;
 917         else
 918                 err = 0;
 919
 920         data_len = rxm->full_len - cipher_desc->tag;
 921
 922         if (skb_pagelen(skb) > offset) {
 923                 copy = min_t(int, skb_pagelen(skb) - offset, data_len);
 924
 925                 if (skb->decrypted) {
 926                         err = skb_store_bits(skb, offset, buf, copy);
 927                         if (err)
 928                                 goto free_buf;
 929                 }
 930
 931                 offset += copy;
 932                 buf += copy;
 933         }
 934
 935         pos = skb_pagelen(skb);
 936         skb_walk_frags(skb, skb_iter) {
 937                 int frag_pos;
 938
 939                 /* Practically all frags must belong to msg if reencrypt
 940                  * is needed with current strparser and coalescing logic,
 941                  * but strparser may "get optimized", so let's be safe.
 942                  */
 943                 if (pos + skb_iter->len <= offset)
 944                         goto done_with_frag;
 945                 if (pos >= data_len + rxm->offset)
 946                         break;
 947
 948                 frag_pos = offset - pos;
 949                 copy = min_t(int, skb_iter->len - frag_pos,
 950                              data_len + rxm->offset - offset);
 951
 952                 if (skb_iter->decrypted) {
 953                         err = skb_store_bits(skb_iter, frag_pos, buf, copy);
 954                         if (err)
 955                                 goto free_buf;
 956                 }
 957
 958                 offset += copy;
 959                 buf += copy;
 960 done_with_frag:
 961                 pos += skb_iter->len;
 962         }
 963
 964 free_buf:
 965         kfree(orig_buf);
 966         return err;
 967 }
 968
 969 int tls_device_decrypted(struct sock *sk, struct tls_context *tls_ctx)
 970 {
 971         struct tls_offload_context_rx *ctx = tls_offload_ctx_rx(tls_ctx);
 972         struct tls_sw_context_rx *sw_ctx = tls_sw_ctx_rx(tls_ctx);
 973         struct sk_buff *skb = tls_strp_msg(sw_ctx);
 974         struct strp_msg *rxm = strp_msg(skb);
 975         int is_decrypted, is_encrypted;
 976
 977         if (!tls_strp_msg_mixed_decrypted(sw_ctx)) {
 978                 is_decrypted = skb->decrypted;
 979                 is_encrypted = !is_decrypted;
 980         } else {
 981                 is_decrypted = 0;
 982                 is_encrypted = 0;
 983         }
 984
 985         trace_tls_device_decrypted(sk, tcp_sk(sk)->copied_seq - rxm->full_len,
 986                                    tls_ctx->rx.rec_seq, rxm->full_len,
 987                                    is_encrypted, is_decrypted);
 988
 989         if (unlikely(test_bit(TLS_RX_DEV_DEGRADED, &tls_ctx->flags))) {
 990                 if (likely(is_encrypted || is_decrypted))
 991                         return is_decrypted;
 992
 993                 /* After tls_device_down disables the offload, the next SKB will
 994                  * likely have initial fragments decrypted, and final ones not
 995                  * decrypted. We need to reencrypt that single SKB.
 996                  */
 997                 return tls_device_reencrypt(sk, tls_ctx);
 998         }
 999
1000         /* Return immediately if the record is either entirely plaintext or
1001          * entirely ciphertext. Otherwise handle reencrypt partially decrypted
1002          * record.
1003          */
1004         if (is_decrypted) {
1005                 ctx->resync_nh_reset = 1;
1006                 return is_decrypted;
1007         }
1008         if (is_encrypted) {
1009                 tls_device_core_ctrl_rx_resync(tls_ctx, ctx, sk, skb);
1010                 return 0;
1011         }
1012
1013         ctx->resync_nh_reset = 1;
1014         return tls_device_reencrypt(sk, tls_ctx);
1015 }
1016
1017 static void tls_device_attach(struct tls_context *ctx, struct sock *sk,
1018                               struct net_device *netdev)
1019 {
1020         if (sk->sk_destruct != tls_device_sk_destruct) {
1021                 refcount_set(&ctx->refcount, 1);
1022                 dev_hold(netdev);
1023                 RCU_INIT_POINTER(ctx->netdev, netdev);
1024                 spin_lock_irq(&tls_device_lock);
1025                 list_add_tail(&ctx->list, &tls_device_list);
1026                 spin_unlock_irq(&tls_device_lock);
1027
1028                 ctx->sk_destruct = sk->sk_destruct;
1029                 smp_store_release(&sk->sk_destruct, tls_device_sk_destruct);
1030         }
1031 }
1032
1033 static struct tls_offload_context_tx *alloc_offload_ctx_tx(struct tls_context *ctx)
1034 {
1035         struct tls_offload_context_tx *offload_ctx;
1036         __be64 rcd_sn;
1037
1038         offload_ctx = kzalloc(sizeof(*offload_ctx), GFP_KERNEL);
1039         if (!offload_ctx)
1040                 return NULL;
1041
1042         INIT_WORK(&offload_ctx->destruct_work, tls_device_tx_del_task);
1043         INIT_LIST_HEAD(&offload_ctx->records_list);
1044         spin_lock_init(&offload_ctx->lock);
1045         sg_init_table(offload_ctx->sg_tx_data,
1046                       ARRAY_SIZE(offload_ctx->sg_tx_data));
1047
1048         /* start at rec_seq - 1 to account for the start marker record */
1049         memcpy(&rcd_sn, ctx->tx.rec_seq, sizeof(rcd_sn));
1050         offload_ctx->unacked_record_sn = be64_to_cpu(rcd_sn) - 1;
1051
1052         offload_ctx->ctx = ctx;
1053
1054         return offload_ctx;
1055 }
1056
1057 int tls_set_device_offload(struct sock *sk)
1058 {
1059         struct tls_record_info *start_marker_record;
1060         struct tls_offload_context_tx *offload_ctx;
1061         const struct tls_cipher_desc *cipher_desc;
1062         struct tls_crypto_info *crypto_info;
1063         struct tls_prot_info *prot;
1064         struct net_device *netdev;
1065         struct tls_context *ctx;
1066         char *iv, *rec_seq;
1067         int rc;
1068
1069         ctx = tls_get_ctx(sk);
1070         prot = &ctx->prot_info;
1071
1072         if (ctx->priv_ctx_tx)
1073                 return -EEXIST;
1074
1075         netdev = get_netdev_for_sock(sk);
1076         if (!netdev) {
1077                 pr_err_ratelimited("%s: netdev not found\n", __func__);
1078                 return -EINVAL;
1079         }
1080
1081         if (!(netdev->features & NETIF_F_HW_TLS_TX)) {
1082                 rc = -EOPNOTSUPP;
1083                 goto release_netdev;
1084         }
1085
1086         crypto_info = &ctx->crypto_send.info;
1087         if (crypto_info->version != TLS_1_2_VERSION) {
1088                 rc = -EOPNOTSUPP;
1089                 goto release_netdev;
1090         }
1091
1092         cipher_desc = get_cipher_desc(crypto_info->cipher_type);
1093         if (!cipher_desc || !cipher_desc->offloadable) {
1094                 rc = -EINVAL;
1095                 goto release_netdev;
1096         }
1097
1098         rc = init_prot_info(prot, crypto_info, cipher_desc);
1099         if (rc)
1100                 goto release_netdev;
1101
1102         iv = crypto_info_iv(crypto_info, cipher_desc);
1103         rec_seq = crypto_info_rec_seq(crypto_info, cipher_desc);
1104
1105         memcpy(ctx->tx.iv + cipher_desc->salt, iv, cipher_desc->iv);
1106         memcpy(ctx->tx.rec_seq, rec_seq, cipher_desc->rec_seq);
1107
1108         start_marker_record = kmalloc(sizeof(*start_marker_record), GFP_KERNEL);
1109         if (!start_marker_record) {
1110                 rc = -ENOMEM;
1111                 goto release_netdev;
1112         }
1113
1114         offload_ctx = alloc_offload_ctx_tx(ctx);
1115         if (!offload_ctx) {
1116                 rc = -ENOMEM;
1117                 goto free_marker_record;
1118         }
1119
1120         rc = tls_sw_fallback_init(sk, offload_ctx, crypto_info);
1121         if (rc)
1122                 goto free_offload_ctx;
1123
1124         start_marker_record->end_seq = tcp_sk(sk)->write_seq;
1125         start_marker_record->len = 0;
1126         start_marker_record->num_frags = 0;
1127         list_add_tail(&start_marker_record->list, &offload_ctx->records_list);
1128
1129         clean_acked_data_enable(inet_csk(sk), &tls_icsk_clean_acked);
1130         ctx->push_pending_record = tls_device_push_pending_record;
1131
1132         /* TLS offload is greatly simplified if we don't send
1133          * SKBs where only part of the payload needs to be encrypted.
1134          * So mark the last skb in the write queue as end of record.
1135          */
1136         tcp_write_collapse_fence(sk);
1137
1138         /* Avoid offloading if the device is down
1139          * We don't want to offload new flows after
1140          * the NETDEV_DOWN event
1141          *
1142          * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1143          * handler thus protecting from the device going down before
1144          * ctx was added to tls_device_list.
1145          */
1146         down_read(&device_offload_lock);
1147         if (!(netdev->flags & IFF_UP)) {
1148                 rc = -EINVAL;
1149                 goto release_lock;
1150         }
1151
1152         ctx->priv_ctx_tx = offload_ctx;
1153         rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_TX,
1154                                              &ctx->crypto_send.info,
1155                                              tcp_sk(sk)->write_seq);
1156         trace_tls_device_offload_set(sk, TLS_OFFLOAD_CTX_DIR_TX,
1157                                      tcp_sk(sk)->write_seq, rec_seq, rc);
1158         if (rc)
1159                 goto release_lock;
1160
1161         tls_device_attach(ctx, sk, netdev);
1162         up_read(&device_offload_lock);
1163
1164         /* following this assignment tls_is_skb_tx_device_offloaded
1165          * will return true and the context might be accessed
1166          * by the netdev's xmit function.
1167          */
1168         smp_store_release(&sk->sk_validate_xmit_skb, tls_validate_xmit_skb);
1169         dev_put(netdev);
1170
1171         return 0;
1172
1173 release_lock:
1174         up_read(&device_offload_lock);
1175         clean_acked_data_disable(inet_csk(sk));
1176         crypto_free_aead(offload_ctx->aead_send);
1177 free_offload_ctx:
1178         kfree(offload_ctx);
1179         ctx->priv_ctx_tx = NULL;
1180 free_marker_record:
1181         kfree(start_marker_record);
1182 release_netdev:
1183         dev_put(netdev);
1184         return rc;
1185 }
1186
1187 int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx)
1188 {
1189         struct tls12_crypto_info_aes_gcm_128 *info;
1190         struct tls_offload_context_rx *context;
1191         struct net_device *netdev;
1192         int rc = 0;
1193
1194         if (ctx->crypto_recv.info.version != TLS_1_2_VERSION)
1195                 return -EOPNOTSUPP;
1196
1197         netdev = get_netdev_for_sock(sk);
1198         if (!netdev) {
1199                 pr_err_ratelimited("%s: netdev not found\n", __func__);
1200                 return -EINVAL;
1201         }
1202
1203         if (!(netdev->features & NETIF_F_HW_TLS_RX)) {
1204                 rc = -EOPNOTSUPP;
1205                 goto release_netdev;
1206         }
1207
1208         /* Avoid offloading if the device is down
1209          * We don't want to offload new flows after
1210          * the NETDEV_DOWN event
1211          *
1212          * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1213          * handler thus protecting from the device going down before
1214          * ctx was added to tls_device_list.
1215          */
1216         down_read(&device_offload_lock);
1217         if (!(netdev->flags & IFF_UP)) {
1218                 rc = -EINVAL;
1219                 goto release_lock;
1220         }
1221
1222         context = kzalloc(sizeof(*context), GFP_KERNEL);
1223         if (!context) {
1224                 rc = -ENOMEM;
1225                 goto release_lock;
1226         }
1227         context->resync_nh_reset = 1;
1228
1229         ctx->priv_ctx_rx = context;
1230         rc = tls_set_sw_offload(sk, 0);
1231         if (rc)
1232                 goto release_ctx;
1233
1234         rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_RX,
1235                                              &ctx->crypto_recv.info,
1236                                              tcp_sk(sk)->copied_seq);
1237         info = (void *)&ctx->crypto_recv.info;
1238         trace_tls_device_offload_set(sk, TLS_OFFLOAD_CTX_DIR_RX,
1239                                      tcp_sk(sk)->copied_seq, info->rec_seq, rc);
1240         if (rc)
1241                 goto free_sw_resources;
1242
1243         tls_device_attach(ctx, sk, netdev);
1244         up_read(&device_offload_lock);
1245
1246         dev_put(netdev);
1247
1248         return 0;
1249
1250 free_sw_resources:
1251         up_read(&device_offload_lock);
1252         tls_sw_free_resources_rx(sk);
1253         down_read(&device_offload_lock);
1254 release_ctx:
1255         ctx->priv_ctx_rx = NULL;
1256 release_lock:
1257         up_read(&device_offload_lock);
1258 release_netdev:
1259         dev_put(netdev);
1260         return rc;
1261 }
1262
1263 void tls_device_offload_cleanup_rx(struct sock *sk)
1264 {
1265         struct tls_context *tls_ctx = tls_get_ctx(sk);
1266         struct net_device *netdev;
1267
1268         down_read(&device_offload_lock);
1269         netdev = rcu_dereference_protected(tls_ctx->netdev,
1270                                            lockdep_is_held(&device_offload_lock));
1271         if (!netdev)
1272                 goto out;
1273
1274         netdev->tlsdev_ops->tls_dev_del(netdev, tls_ctx,
1275                                         TLS_OFFLOAD_CTX_DIR_RX);
1276
1277         if (tls_ctx->tx_conf != TLS_HW) {
1278                 dev_put(netdev);
1279                 rcu_assign_pointer(tls_ctx->netdev, NULL);
1280         } else {
1281                 set_bit(TLS_RX_DEV_CLOSED, &tls_ctx->flags);
1282         }
1283 out:
1284         up_read(&device_offload_lock);
1285         tls_sw_release_resources_rx(sk);
1286 }
1287
1288 static int tls_device_down(struct net_device *netdev)
1289 {
1290         struct tls_context *ctx, *tmp;
1291         unsigned long flags;
1292         LIST_HEAD(list);
1293
1294         /* Request a write lock to block new offload attempts */
1295         down_write(&device_offload_lock);
1296
1297         spin_lock_irqsave(&tls_device_lock, flags);
1298         list_for_each_entry_safe(ctx, tmp, &tls_device_list, list) {
1299                 struct net_device *ctx_netdev =
1300                         rcu_dereference_protected(ctx->netdev,
1301                                                   lockdep_is_held(&device_offload_lock));
1302
1303                 if (ctx_netdev != netdev ||
1304                     !refcount_inc_not_zero(&ctx->refcount))
1305                         continue;
1306
1307                 list_move(&ctx->list, &list);
1308         }
1309         spin_unlock_irqrestore(&tls_device_lock, flags);
1310
1311         list_for_each_entry_safe(ctx, tmp, &list, list) {
1312                 /* Stop offloaded TX and switch to the fallback.
1313                  * tls_is_skb_tx_device_offloaded will return false.
1314                  */
1315                 WRITE_ONCE(ctx->sk->sk_validate_xmit_skb, tls_validate_xmit_skb_sw);
1316
1317                 /* Stop the RX and TX resync.
1318                  * tls_dev_resync must not be called after tls_dev_del.
1319                  */
1320                 rcu_assign_pointer(ctx->netdev, NULL);
1321
1322                 /* Start skipping the RX resync logic completely. */
1323                 set_bit(TLS_RX_DEV_DEGRADED, &ctx->flags);
1324
1325                 /* Sync with inflight packets. After this point:
1326                  * TX: no non-encrypted packets will be passed to the driver.
1327                  * RX: resync requests from the driver will be ignored.
1328                  */
1329                 synchronize_net();
1330
1331                 /* Release the offload context on the driver side. */
1332                 if (ctx->tx_conf == TLS_HW)
1333                         netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1334                                                         TLS_OFFLOAD_CTX_DIR_TX);
1335                 if (ctx->rx_conf == TLS_HW &&
1336                     !test_bit(TLS_RX_DEV_CLOSED, &ctx->flags))
1337                         netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1338                                                         TLS_OFFLOAD_CTX_DIR_RX);
1339
1340                 dev_put(netdev);
1341
1342                 /* Move the context to a separate list for two reasons:
1343                  * 1. When the context is deallocated, list_del is called.
1344                  * 2. It's no longer an offloaded context, so we don't want to
1345                  *    run offload-specific code on this context.
1346                  */
1347                 spin_lock_irqsave(&tls_device_lock, flags);
1348                 list_move_tail(&ctx->list, &tls_device_down_list);
1349                 spin_unlock_irqrestore(&tls_device_lock, flags);
1350
1351                 /* Device contexts for RX and TX will be freed in on sk_destruct
1352                  * by tls_device_free_ctx. rx_conf and tx_conf stay in TLS_HW.
1353                  * Now release the ref taken above.
1354                  */
1355                 if (refcount_dec_and_test(&ctx->refcount)) {
1356                         /* sk_destruct ran after tls_device_down took a ref, and
1357                          * it returned early. Complete the destruction here.
1358                          */
1359                         list_del(&ctx->list);
1360                         tls_device_free_ctx(ctx);
1361                 }
1362         }
1363
1364         up_write(&device_offload_lock);
1365
1366         flush_workqueue(destruct_wq);
1367
1368         return NOTIFY_DONE;
1369 }
1370
1371 static int tls_dev_event(struct notifier_block *this, unsigned long event,
1372                          void *ptr)
1373 {
1374         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1375
1376         if (!dev->tlsdev_ops &&
1377             !(dev->features & (NETIF_F_HW_TLS_RX | NETIF_F_HW_TLS_TX)))
1378                 return NOTIFY_DONE;
1379
1380         switch (event) {
1381         case NETDEV_REGISTER:
1382         case NETDEV_FEAT_CHANGE:
1383                 if (netif_is_bond_master(dev))
1384                         return NOTIFY_DONE;
1385                 if ((dev->features & NETIF_F_HW_TLS_RX) &&
1386                     !dev->tlsdev_ops->tls_dev_resync)
1387                         return NOTIFY_BAD;
1388
1389                 if  (dev->tlsdev_ops &&
1390                      dev->tlsdev_ops->tls_dev_add &&
1391                      dev->tlsdev_ops->tls_dev_del)
1392                         return NOTIFY_DONE;
1393                 else
1394                         return NOTIFY_BAD;
1395         case NETDEV_DOWN:
1396                 return tls_device_down(dev);
1397         }
1398         return NOTIFY_DONE;
1399 }
1400
1401 static struct notifier_block tls_dev_notifier = {
1402         .notifier_call  = tls_dev_event,
1403 };
1404
1405 int __init tls_device_init(void)
1406 {
1407         int err;
1408
1409         dummy_page = alloc_page(GFP_KERNEL);
1410         if (!dummy_page)
1411                 return -ENOMEM;
1412
1413         destruct_wq = alloc_workqueue("ktls_device_destruct", 0, 0);
1414         if (!destruct_wq) {
1415                 err = -ENOMEM;
1416                 goto err_free_dummy;
1417         }
1418
1419         err = register_netdevice_notifier(&tls_dev_notifier);
1420         if (err)
1421                 goto err_destroy_wq;
1422
1423         return 0;
1424
1425 err_destroy_wq:
1426         destroy_workqueue(destruct_wq);
1427 err_free_dummy:
1428         put_page(dummy_page);
1429         return err;
1430 }
1431
1432 void __exit tls_device_cleanup(void)
1433 {
1434         unregister_netdevice_notifier(&tls_dev_notifier);
1435         destroy_workqueue(destruct_wq);
1436         clean_acked_data_flush();
1437         put_page(dummy_page);
1438 }