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
  41 #include "trace.h"
  42
  43 /* device_offload_lock is used to synchronize tls_dev_add
  44  * against NETDEV_DOWN notifications.
  45  */
  46 static DECLARE_RWSEM(device_offload_lock);
  47
  48 static struct workqueue_struct *destruct_wq __read_mostly;
  49
  50 static LIST_HEAD(tls_device_list);
  51 static LIST_HEAD(tls_device_down_list);
  52 static DEFINE_SPINLOCK(tls_device_lock);
  53
  54 static struct page *dummy_page;
  55
  56 static void tls_device_free_ctx(struct tls_context *ctx)
  57 {
  58         if (ctx->tx_conf == TLS_HW) {
  59                 kfree(tls_offload_ctx_tx(ctx));
  60                 kfree(ctx->tx.rec_seq);
  61                 kfree(ctx->tx.iv);
  62         }
  63
  64         if (ctx->rx_conf == TLS_HW)
  65                 kfree(tls_offload_ctx_rx(ctx));
  66
  67         tls_ctx_free(NULL, ctx);
  68 }
  69
  70 static void tls_device_tx_del_task(struct work_struct *work)
  71 {
  72         struct tls_offload_context_tx *offload_ctx =
  73                 container_of(work, struct tls_offload_context_tx, destruct_work);
  74         struct tls_context *ctx = offload_ctx->ctx;
  75         struct net_device *netdev = ctx->netdev;
  76
  77         netdev->tlsdev_ops->tls_dev_del(netdev, ctx, TLS_OFFLOAD_CTX_DIR_TX);
  78         dev_put(netdev);
  79         ctx->netdev = NULL;
  80         tls_device_free_ctx(ctx);
  81 }
  82
  83 static void tls_device_queue_ctx_destruction(struct tls_context *ctx)
  84 {
  85         unsigned long flags;
  86         bool async_cleanup;
  87
  88         spin_lock_irqsave(&tls_device_lock, flags);
  89         if (unlikely(!refcount_dec_and_test(&ctx->refcount))) {
  90                 spin_unlock_irqrestore(&tls_device_lock, flags);
  91                 return;
  92         }
  93
  94         list_del(&ctx->list); /* Remove from tls_device_list / tls_device_down_list */
  95         async_cleanup = ctx->netdev && ctx->tx_conf == TLS_HW;
  96         if (async_cleanup) {
  97                 struct tls_offload_context_tx *offload_ctx = tls_offload_ctx_tx(ctx);
  98
  99                 /* queue_work inside the spinlock
 100                  * to make sure tls_device_down waits for that work.
 101                  */
 102                 queue_work(destruct_wq, &offload_ctx->destruct_work);
 103         }
 104         spin_unlock_irqrestore(&tls_device_lock, flags);
 105
 106         if (!async_cleanup)
 107                 tls_device_free_ctx(ctx);
 108 }
 109
 110 /* We assume that the socket is already connected */
 111 static struct net_device *get_netdev_for_sock(struct sock *sk)
 112 {
 113         struct dst_entry *dst = sk_dst_get(sk);
 114         struct net_device *netdev = NULL;
 115
 116         if (likely(dst)) {
 117                 netdev = netdev_sk_get_lowest_dev(dst->dev, sk);
 118                 dev_hold(netdev);
 119         }
 120
 121         dst_release(dst);
 122
 123         return netdev;
 124 }
 125
 126 static void destroy_record(struct tls_record_info *record)
 127 {
 128         int i;
 129
 130         for (i = 0; i < record->num_frags; i++)
 131                 __skb_frag_unref(&record->frags[i], false);
 132         kfree(record);
 133 }
 134
 135 static void delete_all_records(struct tls_offload_context_tx *offload_ctx)
 136 {
 137         struct tls_record_info *info, *temp;
 138
 139         list_for_each_entry_safe(info, temp, &offload_ctx->records_list, list) {
 140                 list_del(&info->list);
 141                 destroy_record(info);
 142         }
 143
 144         offload_ctx->retransmit_hint = NULL;
 145 }
 146
 147 static void tls_icsk_clean_acked(struct sock *sk, u32 acked_seq)
 148 {
 149         struct tls_context *tls_ctx = tls_get_ctx(sk);
 150         struct tls_record_info *info, *temp;
 151         struct tls_offload_context_tx *ctx;
 152         u64 deleted_records = 0;
 153         unsigned long flags;
 154
 155         if (!tls_ctx)
 156                 return;
 157
 158         ctx = tls_offload_ctx_tx(tls_ctx);
 159
 160         spin_lock_irqsave(&ctx->lock, flags);
 161         info = ctx->retransmit_hint;
 162         if (info && !before(acked_seq, info->end_seq))
 163                 ctx->retransmit_hint = NULL;
 164
 165         list_for_each_entry_safe(info, temp, &ctx->records_list, list) {
 166                 if (before(acked_seq, info->end_seq))
 167                         break;
 168                 list_del(&info->list);
 169
 170                 destroy_record(info);
 171                 deleted_records++;
 172         }
 173
 174         ctx->unacked_record_sn += deleted_records;
 175         spin_unlock_irqrestore(&ctx->lock, flags);
 176 }
 177
 178 /* At this point, there should be no references on this
 179  * socket and no in-flight SKBs associated with this
 180  * socket, so it is safe to free all the resources.
 181  */
 182 void tls_device_sk_destruct(struct sock *sk)
 183 {
 184         struct tls_context *tls_ctx = tls_get_ctx(sk);
 185         struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
 186
 187         tls_ctx->sk_destruct(sk);
 188
 189         if (tls_ctx->tx_conf == TLS_HW) {
 190                 if (ctx->open_record)
 191                         destroy_record(ctx->open_record);
 192                 delete_all_records(ctx);
 193                 crypto_free_aead(ctx->aead_send);
 194                 clean_acked_data_disable(inet_csk(sk));
 195         }
 196
 197         tls_device_queue_ctx_destruction(tls_ctx);
 198 }
 199 EXPORT_SYMBOL_GPL(tls_device_sk_destruct);
 200
 201 void tls_device_free_resources_tx(struct sock *sk)
 202 {
 203         struct tls_context *tls_ctx = tls_get_ctx(sk);
 204
 205         tls_free_partial_record(sk, tls_ctx);
 206 }
 207
 208 void tls_offload_tx_resync_request(struct sock *sk, u32 got_seq, u32 exp_seq)
 209 {
 210         struct tls_context *tls_ctx = tls_get_ctx(sk);
 211
 212         trace_tls_device_tx_resync_req(sk, got_seq, exp_seq);
 213         WARN_ON(test_and_set_bit(TLS_TX_SYNC_SCHED, &tls_ctx->flags));
 214 }
 215 EXPORT_SYMBOL_GPL(tls_offload_tx_resync_request);
 216
 217 static void tls_device_resync_tx(struct sock *sk, struct tls_context *tls_ctx,
 218                                  u32 seq)
 219 {
 220         struct net_device *netdev;
 221         struct sk_buff *skb;
 222         int err = 0;
 223         u8 *rcd_sn;
 224
 225         skb = tcp_write_queue_tail(sk);
 226         if (skb)
 227                 TCP_SKB_CB(skb)->eor = 1;
 228
 229         rcd_sn = tls_ctx->tx.rec_seq;
 230
 231         trace_tls_device_tx_resync_send(sk, seq, rcd_sn);
 232         down_read(&device_offload_lock);
 233         netdev = tls_ctx->netdev;
 234         if (netdev)
 235                 err = netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq,
 236                                                          rcd_sn,
 237                                                          TLS_OFFLOAD_CTX_DIR_TX);
 238         up_read(&device_offload_lock);
 239         if (err)
 240                 return;
 241
 242         clear_bit_unlock(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
 243 }
 244
 245 static void tls_append_frag(struct tls_record_info *record,
 246                             struct page_frag *pfrag,
 247                             int size)
 248 {
 249         skb_frag_t *frag;
 250
 251         frag = &record->frags[record->num_frags - 1];
 252         if (skb_frag_page(frag) == pfrag->page &&
 253             skb_frag_off(frag) + skb_frag_size(frag) == pfrag->offset) {
 254                 skb_frag_size_add(frag, size);
 255         } else {
 256                 ++frag;
 257                 __skb_frag_set_page(frag, pfrag->page);
 258                 skb_frag_off_set(frag, pfrag->offset);
 259                 skb_frag_size_set(frag, size);
 260                 ++record->num_frags;
 261                 get_page(pfrag->page);
 262         }
 263
 264         pfrag->offset += size;
 265         record->len += size;
 266 }
 267
 268 static int tls_push_record(struct sock *sk,
 269                            struct tls_context *ctx,
 270                            struct tls_offload_context_tx *offload_ctx,
 271                            struct tls_record_info *record,
 272                            int flags)
 273 {
 274         struct tls_prot_info *prot = &ctx->prot_info;
 275         struct tcp_sock *tp = tcp_sk(sk);
 276         skb_frag_t *frag;
 277         int i;
 278
 279         record->end_seq = tp->write_seq + record->len;
 280         list_add_tail_rcu(&record->list, &offload_ctx->records_list);
 281         offload_ctx->open_record = NULL;
 282
 283         if (test_bit(TLS_TX_SYNC_SCHED, &ctx->flags))
 284                 tls_device_resync_tx(sk, ctx, tp->write_seq);
 285
 286         tls_advance_record_sn(sk, prot, &ctx->tx);
 287
 288         for (i = 0; i < record->num_frags; i++) {
 289                 frag = &record->frags[i];
 290                 sg_unmark_end(&offload_ctx->sg_tx_data[i]);
 291                 sg_set_page(&offload_ctx->sg_tx_data[i], skb_frag_page(frag),
 292                             skb_frag_size(frag), skb_frag_off(frag));
 293                 sk_mem_charge(sk, skb_frag_size(frag));
 294                 get_page(skb_frag_page(frag));
 295         }
 296         sg_mark_end(&offload_ctx->sg_tx_data[record->num_frags - 1]);
 297
 298         /* all ready, send */
 299         return tls_push_sg(sk, ctx, offload_ctx->sg_tx_data, 0, flags);
 300 }
 301
 302 static void tls_device_record_close(struct sock *sk,
 303                                     struct tls_context *ctx,
 304                                     struct tls_record_info *record,
 305                                     struct page_frag *pfrag,
 306                                     unsigned char record_type)
 307 {
 308         struct tls_prot_info *prot = &ctx->prot_info;
 309         struct page_frag dummy_tag_frag;
 310
 311         /* append tag
 312          * device will fill in the tag, we just need to append a placeholder
 313          * use socket memory to improve coalescing (re-using a single buffer
 314          * increases frag count)
 315          * if we can't allocate memory now use the dummy page
 316          */
 317         if (unlikely(pfrag->size - pfrag->offset < prot->tag_size) &&
 318             !skb_page_frag_refill(prot->tag_size, pfrag, sk->sk_allocation)) {
 319                 dummy_tag_frag.page = dummy_page;
 320                 dummy_tag_frag.offset = 0;
 321                 pfrag = &dummy_tag_frag;
 322         }
 323         tls_append_frag(record, pfrag, prot->tag_size);
 324
 325         /* fill prepend */
 326         tls_fill_prepend(ctx, skb_frag_address(&record->frags[0]),
 327                          record->len - prot->overhead_size,
 328                          record_type);
 329 }
 330
 331 static int tls_create_new_record(struct tls_offload_context_tx *offload_ctx,
 332                                  struct page_frag *pfrag,
 333                                  size_t prepend_size)
 334 {
 335         struct tls_record_info *record;
 336         skb_frag_t *frag;
 337
 338         record = kmalloc(sizeof(*record), GFP_KERNEL);
 339         if (!record)
 340                 return -ENOMEM;
 341
 342         frag = &record->frags[0];
 343         __skb_frag_set_page(frag, pfrag->page);
 344         skb_frag_off_set(frag, pfrag->offset);
 345         skb_frag_size_set(frag, prepend_size);
 346
 347         get_page(pfrag->page);
 348         pfrag->offset += prepend_size;
 349
 350         record->num_frags = 1;
 351         record->len = prepend_size;
 352         offload_ctx->open_record = record;
 353         return 0;
 354 }
 355
 356 static int tls_do_allocation(struct sock *sk,
 357                              struct tls_offload_context_tx *offload_ctx,
 358                              struct page_frag *pfrag,
 359                              size_t prepend_size)
 360 {
 361         int ret;
 362
 363         if (!offload_ctx->open_record) {
 364                 if (unlikely(!skb_page_frag_refill(prepend_size, pfrag,
 365                                                    sk->sk_allocation))) {
 366                         READ_ONCE(sk->sk_prot)->enter_memory_pressure(sk);
 367                         sk_stream_moderate_sndbuf(sk);
 368                         return -ENOMEM;
 369                 }
 370
 371                 ret = tls_create_new_record(offload_ctx, pfrag, prepend_size);
 372                 if (ret)
 373                         return ret;
 374
 375                 if (pfrag->size > pfrag->offset)
 376                         return 0;
 377         }
 378
 379         if (!sk_page_frag_refill(sk, pfrag))
 380                 return -ENOMEM;
 381
 382         return 0;
 383 }
 384
 385 static int tls_device_copy_data(void *addr, size_t bytes, struct iov_iter *i)
 386 {
 387         size_t pre_copy, nocache;
 388
 389         pre_copy = ~((unsigned long)addr - 1) & (SMP_CACHE_BYTES - 1);
 390         if (pre_copy) {
 391                 pre_copy = min(pre_copy, bytes);
 392                 if (copy_from_iter(addr, pre_copy, i) != pre_copy)
 393                         return -EFAULT;
 394                 bytes -= pre_copy;
 395                 addr += pre_copy;
 396         }
 397
 398         nocache = round_down(bytes, SMP_CACHE_BYTES);
 399         if (copy_from_iter_nocache(addr, nocache, i) != nocache)
 400                 return -EFAULT;
 401         bytes -= nocache;
 402         addr += nocache;
 403
 404         if (bytes && copy_from_iter(addr, bytes, i) != bytes)
 405                 return -EFAULT;
 406
 407         return 0;
 408 }
 409
 410 static int tls_push_data(struct sock *sk,
 411                          struct iov_iter *msg_iter,
 412                          size_t size, int flags,
 413                          unsigned char record_type)
 414 {
 415         struct tls_context *tls_ctx = tls_get_ctx(sk);
 416         struct tls_prot_info *prot = &tls_ctx->prot_info;
 417         struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
 418         struct tls_record_info *record;
 419         int tls_push_record_flags;
 420         struct page_frag *pfrag;
 421         size_t orig_size = size;
 422         u32 max_open_record_len;
 423         bool more = false;
 424         bool done = false;
 425         int copy, rc = 0;
 426         long timeo;
 427
 428         if (flags &
 429             ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_SENDPAGE_NOTLAST))
 430                 return -EOPNOTSUPP;
 431
 432         if (unlikely(sk->sk_err))
 433                 return -sk->sk_err;
 434
 435         flags |= MSG_SENDPAGE_DECRYPTED;
 436         tls_push_record_flags = flags | MSG_SENDPAGE_NOTLAST;
 437
 438         timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
 439         if (tls_is_partially_sent_record(tls_ctx)) {
 440                 rc = tls_push_partial_record(sk, tls_ctx, flags);
 441                 if (rc < 0)
 442                         return rc;
 443         }
 444
 445         pfrag = sk_page_frag(sk);
 446
 447         /* TLS_HEADER_SIZE is not counted as part of the TLS record, and
 448          * we need to leave room for an authentication tag.
 449          */
 450         max_open_record_len = TLS_MAX_PAYLOAD_SIZE +
 451                               prot->prepend_size;
 452         do {
 453                 rc = tls_do_allocation(sk, ctx, pfrag, prot->prepend_size);
 454                 if (unlikely(rc)) {
 455                         rc = sk_stream_wait_memory(sk, &timeo);
 456                         if (!rc)
 457                                 continue;
 458
 459                         record = ctx->open_record;
 460                         if (!record)
 461                                 break;
 462 handle_error:
 463                         if (record_type != TLS_RECORD_TYPE_DATA) {
 464                                 /* avoid sending partial
 465                                  * record with type !=
 466                                  * application_data
 467                                  */
 468                                 size = orig_size;
 469                                 destroy_record(record);
 470                                 ctx->open_record = NULL;
 471                         } else if (record->len > prot->prepend_size) {
 472                                 goto last_record;
 473                         }
 474
 475                         break;
 476                 }
 477
 478                 record = ctx->open_record;
 479                 copy = min_t(size_t, size, (pfrag->size - pfrag->offset));
 480                 copy = min_t(size_t, copy, (max_open_record_len - record->len));
 481
 482                 if (copy) {
 483                         rc = tls_device_copy_data(page_address(pfrag->page) +
 484                                                   pfrag->offset, copy, msg_iter);
 485                         if (rc)
 486                                 goto handle_error;
 487                         tls_append_frag(record, pfrag, copy);
 488                 }
 489
 490                 size -= copy;
 491                 if (!size) {
 492 last_record:
 493                         tls_push_record_flags = flags;
 494                         if (flags & (MSG_SENDPAGE_NOTLAST | MSG_MORE)) {
 495                                 more = true;
 496                                 break;
 497                         }
 498
 499                         done = true;
 500                 }
 501
 502                 if (done || record->len >= max_open_record_len ||
 503                     (record->num_frags >= MAX_SKB_FRAGS - 1)) {
 504                         tls_device_record_close(sk, tls_ctx, record,
 505                                                 pfrag, record_type);
 506
 507                         rc = tls_push_record(sk,
 508                                              tls_ctx,
 509                                              ctx,
 510                                              record,
 511                                              tls_push_record_flags);
 512                         if (rc < 0)
 513                                 break;
 514                 }
 515         } while (!done);
 516
 517         tls_ctx->pending_open_record_frags = more;
 518
 519         if (orig_size - size > 0)
 520                 rc = orig_size - size;
 521
 522         return rc;
 523 }
 524
 525 int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
 526 {
 527         unsigned char record_type = TLS_RECORD_TYPE_DATA;
 528         struct tls_context *tls_ctx = tls_get_ctx(sk);
 529         int rc;
 530
 531         mutex_lock(&tls_ctx->tx_lock);
 532         lock_sock(sk);
 533
 534         if (unlikely(msg->msg_controllen)) {
 535                 rc = tls_proccess_cmsg(sk, msg, &record_type);
 536                 if (rc)
 537                         goto out;
 538         }
 539
 540         rc = tls_push_data(sk, &msg->msg_iter, size,
 541                            msg->msg_flags, record_type);
 542
 543 out:
 544         release_sock(sk);
 545         mutex_unlock(&tls_ctx->tx_lock);
 546         return rc;
 547 }
 548
 549 int tls_device_sendpage(struct sock *sk, struct page *page,
 550                         int offset, size_t size, int flags)
 551 {
 552         struct tls_context *tls_ctx = tls_get_ctx(sk);
 553         struct iov_iter msg_iter;
 554         char *kaddr;
 555         struct kvec iov;
 556         int rc;
 557
 558         if (flags & MSG_SENDPAGE_NOTLAST)
 559                 flags |= MSG_MORE;
 560
 561         mutex_lock(&tls_ctx->tx_lock);
 562         lock_sock(sk);
 563
 564         if (flags & MSG_OOB) {
 565                 rc = -EOPNOTSUPP;
 566                 goto out;
 567         }
 568
 569         kaddr = kmap(page);
 570         iov.iov_base = kaddr + offset;
 571         iov.iov_len = size;
 572         iov_iter_kvec(&msg_iter, WRITE, &iov, 1, size);
 573         rc = tls_push_data(sk, &msg_iter, size,
 574                            flags, TLS_RECORD_TYPE_DATA);
 575         kunmap(page);
 576
 577 out:
 578         release_sock(sk);
 579         mutex_unlock(&tls_ctx->tx_lock);
 580         return rc;
 581 }
 582
 583 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
 584                                        u32 seq, u64 *p_record_sn)
 585 {
 586         u64 record_sn = context->hint_record_sn;
 587         struct tls_record_info *info, *last;
 588
 589         info = context->retransmit_hint;
 590         if (!info ||
 591             before(seq, info->end_seq - info->len)) {
 592                 /* if retransmit_hint is irrelevant start
 593                  * from the beginning of the list
 594                  */
 595                 info = list_first_entry_or_null(&context->records_list,
 596                                                 struct tls_record_info, list);
 597                 if (!info)
 598                         return NULL;
 599                 /* send the start_marker record if seq number is before the
 600                  * tls offload start marker sequence number. This record is
 601                  * required to handle TCP packets which are before TLS offload
 602                  * started.
 603                  *  And if it's not start marker, look if this seq number
 604                  * belongs to the list.
 605                  */
 606                 if (likely(!tls_record_is_start_marker(info))) {
 607                         /* we have the first record, get the last record to see
 608                          * if this seq number belongs to the list.
 609                          */
 610                         last = list_last_entry(&context->records_list,
 611                                                struct tls_record_info, list);
 612
 613                         if (!between(seq, tls_record_start_seq(info),
 614                                      last->end_seq))
 615                                 return NULL;
 616                 }
 617                 record_sn = context->unacked_record_sn;
 618         }
 619
 620         /* We just need the _rcu for the READ_ONCE() */
 621         rcu_read_lock();
 622         list_for_each_entry_from_rcu(info, &context->records_list, list) {
 623                 if (before(seq, info->end_seq)) {
 624                         if (!context->retransmit_hint ||
 625                             after(info->end_seq,
 626                                   context->retransmit_hint->end_seq)) {
 627                                 context->hint_record_sn = record_sn;
 628                                 context->retransmit_hint = info;
 629                         }
 630                         *p_record_sn = record_sn;
 631                         goto exit_rcu_unlock;
 632                 }
 633                 record_sn++;
 634         }
 635         info = NULL;
 636
 637 exit_rcu_unlock:
 638         rcu_read_unlock();
 639         return info;
 640 }
 641 EXPORT_SYMBOL(tls_get_record);
 642
 643 static int tls_device_push_pending_record(struct sock *sk, int flags)
 644 {
 645         struct iov_iter msg_iter;
 646
 647         iov_iter_kvec(&msg_iter, WRITE, NULL, 0, 0);
 648         return tls_push_data(sk, &msg_iter, 0, flags, TLS_RECORD_TYPE_DATA);
 649 }
 650
 651 void tls_device_write_space(struct sock *sk, struct tls_context *ctx)
 652 {
 653         if (tls_is_partially_sent_record(ctx)) {
 654                 gfp_t sk_allocation = sk->sk_allocation;
 655
 656                 WARN_ON_ONCE(sk->sk_write_pending);
 657
 658                 sk->sk_allocation = GFP_ATOMIC;
 659                 tls_push_partial_record(sk, ctx,
 660                                         MSG_DONTWAIT | MSG_NOSIGNAL |
 661                                         MSG_SENDPAGE_DECRYPTED);
 662                 sk->sk_allocation = sk_allocation;
 663         }
 664 }
 665
 666 static void tls_device_resync_rx(struct tls_context *tls_ctx,
 667                                  struct sock *sk, u32 seq, u8 *rcd_sn)
 668 {
 669         struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
 670         struct net_device *netdev;
 671
 672         trace_tls_device_rx_resync_send(sk, seq, rcd_sn, rx_ctx->resync_type);
 673         rcu_read_lock();
 674         netdev = READ_ONCE(tls_ctx->netdev);
 675         if (netdev)
 676                 netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq, rcd_sn,
 677                                                    TLS_OFFLOAD_CTX_DIR_RX);
 678         rcu_read_unlock();
 679         TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXDEVICERESYNC);
 680 }
 681
 682 static bool
 683 tls_device_rx_resync_async(struct tls_offload_resync_async *resync_async,
 684                            s64 resync_req, u32 *seq, u16 *rcd_delta)
 685 {
 686         u32 is_async = resync_req & RESYNC_REQ_ASYNC;
 687         u32 req_seq = resync_req >> 32;
 688         u32 req_end = req_seq + ((resync_req >> 16) & 0xffff);
 689         u16 i;
 690
 691         *rcd_delta = 0;
 692
 693         if (is_async) {
 694                 /* shouldn't get to wraparound:
 695                  * too long in async stage, something bad happened
 696                  */
 697                 if (WARN_ON_ONCE(resync_async->rcd_delta == USHRT_MAX))
 698                         return false;
 699
 700                 /* asynchronous stage: log all headers seq such that
 701                  * req_seq <= seq <= end_seq, and wait for real resync request
 702                  */
 703                 if (before(*seq, req_seq))
 704                         return false;
 705                 if (!after(*seq, req_end) &&
 706                     resync_async->loglen < TLS_DEVICE_RESYNC_ASYNC_LOGMAX)
 707                         resync_async->log[resync_async->loglen++] = *seq;
 708
 709                 resync_async->rcd_delta++;
 710
 711                 return false;
 712         }
 713
 714         /* synchronous stage: check against the logged entries and
 715          * proceed to check the next entries if no match was found
 716          */
 717         for (i = 0; i < resync_async->loglen; i++)
 718                 if (req_seq == resync_async->log[i] &&
 719                     atomic64_try_cmpxchg(&resync_async->req, &resync_req, 0)) {
 720                         *rcd_delta = resync_async->rcd_delta - i;
 721                         *seq = req_seq;
 722                         resync_async->loglen = 0;
 723                         resync_async->rcd_delta = 0;
 724                         return true;
 725                 }
 726
 727         resync_async->loglen = 0;
 728         resync_async->rcd_delta = 0;
 729
 730         if (req_seq == *seq &&
 731             atomic64_try_cmpxchg(&resync_async->req,
 732                                  &resync_req, 0))
 733                 return true;
 734
 735         return false;
 736 }
 737
 738 void tls_device_rx_resync_new_rec(struct sock *sk, u32 rcd_len, u32 seq)
 739 {
 740         struct tls_context *tls_ctx = tls_get_ctx(sk);
 741         struct tls_offload_context_rx *rx_ctx;
 742         u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
 743         u32 sock_data, is_req_pending;
 744         struct tls_prot_info *prot;
 745         s64 resync_req;
 746         u16 rcd_delta;
 747         u32 req_seq;
 748
 749         if (tls_ctx->rx_conf != TLS_HW)
 750                 return;
 751         if (unlikely(test_bit(TLS_RX_DEV_DEGRADED, &tls_ctx->flags)))
 752                 return;
 753
 754         prot = &tls_ctx->prot_info;
 755         rx_ctx = tls_offload_ctx_rx(tls_ctx);
 756         memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
 757
 758         switch (rx_ctx->resync_type) {
 759         case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ:
 760                 resync_req = atomic64_read(&rx_ctx->resync_req);
 761                 req_seq = resync_req >> 32;
 762                 seq += TLS_HEADER_SIZE - 1;
 763                 is_req_pending = resync_req;
 764
 765                 if (likely(!is_req_pending) || req_seq != seq ||
 766                     !atomic64_try_cmpxchg(&rx_ctx->resync_req, &resync_req, 0))
 767                         return;
 768                 break;
 769         case TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT:
 770                 if (likely(!rx_ctx->resync_nh_do_now))
 771                         return;
 772
 773                 /* head of next rec is already in, note that the sock_inq will
 774                  * include the currently parsed message when called from parser
 775                  */
 776                 sock_data = tcp_inq(sk);
 777                 if (sock_data > rcd_len) {
 778                         trace_tls_device_rx_resync_nh_delay(sk, sock_data,
 779                                                             rcd_len);
 780                         return;
 781                 }
 782
 783                 rx_ctx->resync_nh_do_now = 0;
 784                 seq += rcd_len;
 785                 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
 786                 break;
 787         case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC:
 788                 resync_req = atomic64_read(&rx_ctx->resync_async->req);
 789                 is_req_pending = resync_req;
 790                 if (likely(!is_req_pending))
 791                         return;
 792
 793                 if (!tls_device_rx_resync_async(rx_ctx->resync_async,
 794                                                 resync_req, &seq, &rcd_delta))
 795                         return;
 796                 tls_bigint_subtract(rcd_sn, rcd_delta);
 797                 break;
 798         }
 799
 800         tls_device_resync_rx(tls_ctx, sk, seq, rcd_sn);
 801 }
 802
 803 static void tls_device_core_ctrl_rx_resync(struct tls_context *tls_ctx,
 804                                            struct tls_offload_context_rx *ctx,
 805                                            struct sock *sk, struct sk_buff *skb)
 806 {
 807         struct strp_msg *rxm;
 808
 809         /* device will request resyncs by itself based on stream scan */
 810         if (ctx->resync_type != TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT)
 811                 return;
 812         /* already scheduled */
 813         if (ctx->resync_nh_do_now)
 814                 return;
 815         /* seen decrypted fragments since last fully-failed record */
 816         if (ctx->resync_nh_reset) {
 817                 ctx->resync_nh_reset = 0;
 818                 ctx->resync_nh.decrypted_failed = 1;
 819                 ctx->resync_nh.decrypted_tgt = TLS_DEVICE_RESYNC_NH_START_IVAL;
 820                 return;
 821         }
 822
 823         if (++ctx->resync_nh.decrypted_failed <= ctx->resync_nh.decrypted_tgt)
 824                 return;
 825
 826         /* doing resync, bump the next target in case it fails */
 827         if (ctx->resync_nh.decrypted_tgt < TLS_DEVICE_RESYNC_NH_MAX_IVAL)
 828                 ctx->resync_nh.decrypted_tgt *= 2;
 829         else
 830                 ctx->resync_nh.decrypted_tgt += TLS_DEVICE_RESYNC_NH_MAX_IVAL;
 831
 832         rxm = strp_msg(skb);
 833
 834         /* head of next rec is already in, parser will sync for us */
 835         if (tcp_inq(sk) > rxm->full_len) {
 836                 trace_tls_device_rx_resync_nh_schedule(sk);
 837                 ctx->resync_nh_do_now = 1;
 838         } else {
 839                 struct tls_prot_info *prot = &tls_ctx->prot_info;
 840                 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
 841
 842                 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
 843                 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
 844
 845                 tls_device_resync_rx(tls_ctx, sk, tcp_sk(sk)->copied_seq,
 846                                      rcd_sn);
 847         }
 848 }
 849
 850 static int tls_device_reencrypt(struct sock *sk, struct sk_buff *skb)
 851 {
 852         struct strp_msg *rxm = strp_msg(skb);
 853         int err = 0, offset = rxm->offset, copy, nsg, data_len, pos;
 854         struct sk_buff *skb_iter, *unused;
 855         struct scatterlist sg[1];
 856         char *orig_buf, *buf;
 857
 858         orig_buf = kmalloc(rxm->full_len + TLS_HEADER_SIZE +
 859                            TLS_CIPHER_AES_GCM_128_IV_SIZE, sk->sk_allocation);
 860         if (!orig_buf)
 861                 return -ENOMEM;
 862         buf = orig_buf;
 863
 864         nsg = skb_cow_data(skb, 0, &unused);
 865         if (unlikely(nsg < 0)) {
 866                 err = nsg;
 867                 goto free_buf;
 868         }
 869
 870         sg_init_table(sg, 1);
 871         sg_set_buf(&sg[0], buf,
 872                    rxm->full_len + TLS_HEADER_SIZE +
 873                    TLS_CIPHER_AES_GCM_128_IV_SIZE);
 874         err = skb_copy_bits(skb, offset, buf,
 875                             TLS_HEADER_SIZE + TLS_CIPHER_AES_GCM_128_IV_SIZE);
 876         if (err)
 877                 goto free_buf;
 878
 879         /* We are interested only in the decrypted data not the auth */
 880         err = decrypt_skb(sk, skb, sg);
 881         if (err != -EBADMSG)
 882                 goto free_buf;
 883         else
 884                 err = 0;
 885
 886         data_len = rxm->full_len - TLS_CIPHER_AES_GCM_128_TAG_SIZE;
 887
 888         if (skb_pagelen(skb) > offset) {
 889                 copy = min_t(int, skb_pagelen(skb) - offset, data_len);
 890
 891                 if (skb->decrypted) {
 892                         err = skb_store_bits(skb, offset, buf, copy);
 893                         if (err)
 894                                 goto free_buf;
 895                 }
 896
 897                 offset += copy;
 898                 buf += copy;
 899         }
 900
 901         pos = skb_pagelen(skb);
 902         skb_walk_frags(skb, skb_iter) {
 903                 int frag_pos;
 904
 905                 /* Practically all frags must belong to msg if reencrypt
 906                  * is needed with current strparser and coalescing logic,
 907                  * but strparser may "get optimized", so let's be safe.
 908                  */
 909                 if (pos + skb_iter->len <= offset)
 910                         goto done_with_frag;
 911                 if (pos >= data_len + rxm->offset)
 912                         break;
 913
 914                 frag_pos = offset - pos;
 915                 copy = min_t(int, skb_iter->len - frag_pos,
 916                              data_len + rxm->offset - offset);
 917
 918                 if (skb_iter->decrypted) {
 919                         err = skb_store_bits(skb_iter, frag_pos, buf, copy);
 920                         if (err)
 921                                 goto free_buf;
 922                 }
 923
 924                 offset += copy;
 925                 buf += copy;
 926 done_with_frag:
 927                 pos += skb_iter->len;
 928         }
 929
 930 free_buf:
 931         kfree(orig_buf);
 932         return err;
 933 }
 934
 935 int tls_device_decrypted(struct sock *sk, struct tls_context *tls_ctx,
 936                          struct sk_buff *skb, struct strp_msg *rxm)
 937 {
 938         struct tls_offload_context_rx *ctx = tls_offload_ctx_rx(tls_ctx);
 939         int is_decrypted = skb->decrypted;
 940         int is_encrypted = !is_decrypted;
 941         struct sk_buff *skb_iter;
 942
 943         /* Check if all the data is decrypted already */
 944         skb_walk_frags(skb, skb_iter) {
 945                 is_decrypted &= skb_iter->decrypted;
 946                 is_encrypted &= !skb_iter->decrypted;
 947         }
 948
 949         trace_tls_device_decrypted(sk, tcp_sk(sk)->copied_seq - rxm->full_len,
 950                                    tls_ctx->rx.rec_seq, rxm->full_len,
 951                                    is_encrypted, is_decrypted);
 952
 953         ctx->sw.decrypted |= is_decrypted;
 954
 955         if (unlikely(test_bit(TLS_RX_DEV_DEGRADED, &tls_ctx->flags))) {
 956                 if (likely(is_encrypted || is_decrypted))
 957                         return 0;
 958
 959                 /* After tls_device_down disables the offload, the next SKB will
 960                  * likely have initial fragments decrypted, and final ones not
 961                  * decrypted. We need to reencrypt that single SKB.
 962                  */
 963                 return tls_device_reencrypt(sk, skb);
 964         }
 965
 966         /* Return immediately if the record is either entirely plaintext or
 967          * entirely ciphertext. Otherwise handle reencrypt partially decrypted
 968          * record.
 969          */
 970         if (is_decrypted) {
 971                 ctx->resync_nh_reset = 1;
 972                 return 0;
 973         }
 974         if (is_encrypted) {
 975                 tls_device_core_ctrl_rx_resync(tls_ctx, ctx, sk, skb);
 976                 return 0;
 977         }
 978
 979         ctx->resync_nh_reset = 1;
 980         return tls_device_reencrypt(sk, skb);
 981 }
 982
 983 static void tls_device_attach(struct tls_context *ctx, struct sock *sk,
 984                               struct net_device *netdev)
 985 {
 986         if (sk->sk_destruct != tls_device_sk_destruct) {
 987                 refcount_set(&ctx->refcount, 1);
 988                 dev_hold(netdev);
 989                 ctx->netdev = netdev;
 990                 spin_lock_irq(&tls_device_lock);
 991                 list_add_tail(&ctx->list, &tls_device_list);
 992                 spin_unlock_irq(&tls_device_lock);
 993
 994                 ctx->sk_destruct = sk->sk_destruct;
 995                 smp_store_release(&sk->sk_destruct, tls_device_sk_destruct);
 996         }
 997 }
 998
 999 int tls_set_device_offload(struct sock *sk, struct tls_context *ctx)
1000 {
1001         u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
1002         struct tls_context *tls_ctx = tls_get_ctx(sk);
1003         struct tls_prot_info *prot = &tls_ctx->prot_info;
1004         struct tls_record_info *start_marker_record;
1005         struct tls_offload_context_tx *offload_ctx;
1006         struct tls_crypto_info *crypto_info;
1007         struct net_device *netdev;
1008         char *iv, *rec_seq;
1009         struct sk_buff *skb;
1010         __be64 rcd_sn;
1011         int rc;
1012
1013         if (!ctx)
1014                 return -EINVAL;
1015
1016         if (ctx->priv_ctx_tx)
1017                 return -EEXIST;
1018
1019         start_marker_record = kmalloc(sizeof(*start_marker_record), GFP_KERNEL);
1020         if (!start_marker_record)
1021                 return -ENOMEM;
1022
1023         offload_ctx = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_TX, GFP_KERNEL);
1024         if (!offload_ctx) {
1025                 rc = -ENOMEM;
1026                 goto free_marker_record;
1027         }
1028
1029         crypto_info = &ctx->crypto_send.info;
1030         if (crypto_info->version != TLS_1_2_VERSION) {
1031                 rc = -EOPNOTSUPP;
1032                 goto free_offload_ctx;
1033         }
1034
1035         switch (crypto_info->cipher_type) {
1036         case TLS_CIPHER_AES_GCM_128:
1037                 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
1038                 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
1039                 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
1040                 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
1041                 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
1042                 salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
1043                 rec_seq =
1044                  ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
1045                 break;
1046         default:
1047                 rc = -EINVAL;
1048                 goto free_offload_ctx;
1049         }
1050
1051         /* Sanity-check the rec_seq_size for stack allocations */
1052         if (rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
1053                 rc = -EINVAL;
1054                 goto free_offload_ctx;
1055         }
1056
1057         prot->version = crypto_info->version;
1058         prot->cipher_type = crypto_info->cipher_type;
1059         prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
1060         prot->tag_size = tag_size;
1061         prot->overhead_size = prot->prepend_size + prot->tag_size;
1062         prot->iv_size = iv_size;
1063         prot->salt_size = salt_size;
1064         ctx->tx.iv = kmalloc(iv_size + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
1065                              GFP_KERNEL);
1066         if (!ctx->tx.iv) {
1067                 rc = -ENOMEM;
1068                 goto free_offload_ctx;
1069         }
1070
1071         memcpy(ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv, iv_size);
1072
1073         prot->rec_seq_size = rec_seq_size;
1074         ctx->tx.rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
1075         if (!ctx->tx.rec_seq) {
1076                 rc = -ENOMEM;
1077                 goto free_iv;
1078         }
1079
1080         rc = tls_sw_fallback_init(sk, offload_ctx, crypto_info);
1081         if (rc)
1082                 goto free_rec_seq;
1083
1084         /* start at rec_seq - 1 to account for the start marker record */
1085         memcpy(&rcd_sn, ctx->tx.rec_seq, sizeof(rcd_sn));
1086         offload_ctx->unacked_record_sn = be64_to_cpu(rcd_sn) - 1;
1087
1088         start_marker_record->end_seq = tcp_sk(sk)->write_seq;
1089         start_marker_record->len = 0;
1090         start_marker_record->num_frags = 0;
1091
1092         INIT_WORK(&offload_ctx->destruct_work, tls_device_tx_del_task);
1093         offload_ctx->ctx = ctx;
1094
1095         INIT_LIST_HEAD(&offload_ctx->records_list);
1096         list_add_tail(&start_marker_record->list, &offload_ctx->records_list);
1097         spin_lock_init(&offload_ctx->lock);
1098         sg_init_table(offload_ctx->sg_tx_data,
1099                       ARRAY_SIZE(offload_ctx->sg_tx_data));
1100
1101         clean_acked_data_enable(inet_csk(sk), &tls_icsk_clean_acked);
1102         ctx->push_pending_record = tls_device_push_pending_record;
1103
1104         /* TLS offload is greatly simplified if we don't send
1105          * SKBs where only part of the payload needs to be encrypted.
1106          * So mark the last skb in the write queue as end of record.
1107          */
1108         skb = tcp_write_queue_tail(sk);
1109         if (skb)
1110                 TCP_SKB_CB(skb)->eor = 1;
1111
1112         netdev = get_netdev_for_sock(sk);
1113         if (!netdev) {
1114                 pr_err_ratelimited("%s: netdev not found\n", __func__);
1115                 rc = -EINVAL;
1116                 goto disable_cad;
1117         }
1118
1119         if (!(netdev->features & NETIF_F_HW_TLS_TX)) {
1120                 rc = -EOPNOTSUPP;
1121                 goto release_netdev;
1122         }
1123
1124         /* Avoid offloading if the device is down
1125          * We don't want to offload new flows after
1126          * the NETDEV_DOWN event
1127          *
1128          * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1129          * handler thus protecting from the device going down before
1130          * ctx was added to tls_device_list.
1131          */
1132         down_read(&device_offload_lock);
1133         if (!(netdev->flags & IFF_UP)) {
1134                 rc = -EINVAL;
1135                 goto release_lock;
1136         }
1137
1138         ctx->priv_ctx_tx = offload_ctx;
1139         rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_TX,
1140                                              &ctx->crypto_send.info,
1141                                              tcp_sk(sk)->write_seq);
1142         trace_tls_device_offload_set(sk, TLS_OFFLOAD_CTX_DIR_TX,
1143                                      tcp_sk(sk)->write_seq, rec_seq, rc);
1144         if (rc)
1145                 goto release_lock;
1146
1147         tls_device_attach(ctx, sk, netdev);
1148         up_read(&device_offload_lock);
1149
1150         /* following this assignment tls_is_sk_tx_device_offloaded
1151          * will return true and the context might be accessed
1152          * by the netdev's xmit function.
1153          */
1154         smp_store_release(&sk->sk_validate_xmit_skb, tls_validate_xmit_skb);
1155         dev_put(netdev);
1156
1157         return 0;
1158
1159 release_lock:
1160         up_read(&device_offload_lock);
1161 release_netdev:
1162         dev_put(netdev);
1163 disable_cad:
1164         clean_acked_data_disable(inet_csk(sk));
1165         crypto_free_aead(offload_ctx->aead_send);
1166 free_rec_seq:
1167         kfree(ctx->tx.rec_seq);
1168 free_iv:
1169         kfree(ctx->tx.iv);
1170 free_offload_ctx:
1171         kfree(offload_ctx);
1172         ctx->priv_ctx_tx = NULL;
1173 free_marker_record:
1174         kfree(start_marker_record);
1175         return rc;
1176 }
1177
1178 int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx)
1179 {
1180         struct tls12_crypto_info_aes_gcm_128 *info;
1181         struct tls_offload_context_rx *context;
1182         struct net_device *netdev;
1183         int rc = 0;
1184
1185         if (ctx->crypto_recv.info.version != TLS_1_2_VERSION)
1186                 return -EOPNOTSUPP;
1187
1188         netdev = get_netdev_for_sock(sk);
1189         if (!netdev) {
1190                 pr_err_ratelimited("%s: netdev not found\n", __func__);
1191                 return -EINVAL;
1192         }
1193
1194         if (!(netdev->features & NETIF_F_HW_TLS_RX)) {
1195                 rc = -EOPNOTSUPP;
1196                 goto release_netdev;
1197         }
1198
1199         /* Avoid offloading if the device is down
1200          * We don't want to offload new flows after
1201          * the NETDEV_DOWN event
1202          *
1203          * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1204          * handler thus protecting from the device going down before
1205          * ctx was added to tls_device_list.
1206          */
1207         down_read(&device_offload_lock);
1208         if (!(netdev->flags & IFF_UP)) {
1209                 rc = -EINVAL;
1210                 goto release_lock;
1211         }
1212
1213         context = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_RX, GFP_KERNEL);
1214         if (!context) {
1215                 rc = -ENOMEM;
1216                 goto release_lock;
1217         }
1218         context->resync_nh_reset = 1;
1219
1220         ctx->priv_ctx_rx = context;
1221         rc = tls_set_sw_offload(sk, ctx, 0);
1222         if (rc)
1223                 goto release_ctx;
1224
1225         rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_RX,
1226                                              &ctx->crypto_recv.info,
1227                                              tcp_sk(sk)->copied_seq);
1228         info = (void *)&ctx->crypto_recv.info;
1229         trace_tls_device_offload_set(sk, TLS_OFFLOAD_CTX_DIR_RX,
1230                                      tcp_sk(sk)->copied_seq, info->rec_seq, rc);
1231         if (rc)
1232                 goto free_sw_resources;
1233
1234         tls_device_attach(ctx, sk, netdev);
1235         up_read(&device_offload_lock);
1236
1237         dev_put(netdev);
1238
1239         return 0;
1240
1241 free_sw_resources:
1242         up_read(&device_offload_lock);
1243         tls_sw_free_resources_rx(sk);
1244         down_read(&device_offload_lock);
1245 release_ctx:
1246         ctx->priv_ctx_rx = NULL;
1247 release_lock:
1248         up_read(&device_offload_lock);
1249 release_netdev:
1250         dev_put(netdev);
1251         return rc;
1252 }
1253
1254 void tls_device_offload_cleanup_rx(struct sock *sk)
1255 {
1256         struct tls_context *tls_ctx = tls_get_ctx(sk);
1257         struct net_device *netdev;
1258
1259         down_read(&device_offload_lock);
1260         netdev = tls_ctx->netdev;
1261         if (!netdev)
1262                 goto out;
1263
1264         netdev->tlsdev_ops->tls_dev_del(netdev, tls_ctx,
1265                                         TLS_OFFLOAD_CTX_DIR_RX);
1266
1267         if (tls_ctx->tx_conf != TLS_HW) {
1268                 dev_put(netdev);
1269                 tls_ctx->netdev = NULL;
1270         } else {
1271                 set_bit(TLS_RX_DEV_CLOSED, &tls_ctx->flags);
1272         }
1273 out:
1274         up_read(&device_offload_lock);
1275         tls_sw_release_resources_rx(sk);
1276 }
1277
1278 static int tls_device_down(struct net_device *netdev)
1279 {
1280         struct tls_context *ctx, *tmp;
1281         unsigned long flags;
1282         LIST_HEAD(list);
1283
1284         /* Request a write lock to block new offload attempts */
1285         down_write(&device_offload_lock);
1286
1287         spin_lock_irqsave(&tls_device_lock, flags);
1288         list_for_each_entry_safe(ctx, tmp, &tls_device_list, list) {
1289                 if (ctx->netdev != netdev ||
1290                     !refcount_inc_not_zero(&ctx->refcount))
1291                         continue;
1292
1293                 list_move(&ctx->list, &list);
1294         }
1295         spin_unlock_irqrestore(&tls_device_lock, flags);
1296
1297         list_for_each_entry_safe(ctx, tmp, &list, list) {
1298                 /* Stop offloaded TX and switch to the fallback.
1299                  * tls_is_sk_tx_device_offloaded will return false.
1300                  */
1301                 WRITE_ONCE(ctx->sk->sk_validate_xmit_skb, tls_validate_xmit_skb_sw);
1302
1303                 /* Stop the RX and TX resync.
1304                  * tls_dev_resync must not be called after tls_dev_del.
1305                  */
1306                 WRITE_ONCE(ctx->netdev, NULL);
1307
1308                 /* Start skipping the RX resync logic completely. */
1309                 set_bit(TLS_RX_DEV_DEGRADED, &ctx->flags);
1310
1311                 /* Sync with inflight packets. After this point:
1312                  * TX: no non-encrypted packets will be passed to the driver.
1313                  * RX: resync requests from the driver will be ignored.
1314                  */
1315                 synchronize_net();
1316
1317                 /* Release the offload context on the driver side. */
1318                 if (ctx->tx_conf == TLS_HW)
1319                         netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1320                                                         TLS_OFFLOAD_CTX_DIR_TX);
1321                 if (ctx->rx_conf == TLS_HW &&
1322                     !test_bit(TLS_RX_DEV_CLOSED, &ctx->flags))
1323                         netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1324                                                         TLS_OFFLOAD_CTX_DIR_RX);
1325
1326                 dev_put(netdev);
1327
1328                 /* Move the context to a separate list for two reasons:
1329                  * 1. When the context is deallocated, list_del is called.
1330                  * 2. It's no longer an offloaded context, so we don't want to
1331                  *    run offload-specific code on this context.
1332                  */
1333                 spin_lock_irqsave(&tls_device_lock, flags);
1334                 list_move_tail(&ctx->list, &tls_device_down_list);
1335                 spin_unlock_irqrestore(&tls_device_lock, flags);
1336
1337                 /* Device contexts for RX and TX will be freed in on sk_destruct
1338                  * by tls_device_free_ctx. rx_conf and tx_conf stay in TLS_HW.
1339                  * Now release the ref taken above.
1340                  */
1341                 if (refcount_dec_and_test(&ctx->refcount)) {
1342                         /* sk_destruct ran after tls_device_down took a ref, and
1343                          * it returned early. Complete the destruction here.
1344                          */
1345                         list_del(&ctx->list);
1346                         tls_device_free_ctx(ctx);
1347                 }
1348         }
1349
1350         up_write(&device_offload_lock);
1351
1352         flush_workqueue(destruct_wq);
1353
1354         return NOTIFY_DONE;
1355 }
1356
1357 static int tls_dev_event(struct notifier_block *this, unsigned long event,
1358                          void *ptr)
1359 {
1360         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1361
1362         if (!dev->tlsdev_ops &&
1363             !(dev->features & (NETIF_F_HW_TLS_RX | NETIF_F_HW_TLS_TX)))
1364                 return NOTIFY_DONE;
1365
1366         switch (event) {
1367         case NETDEV_REGISTER:
1368         case NETDEV_FEAT_CHANGE:
1369                 if (netif_is_bond_master(dev))
1370                         return NOTIFY_DONE;
1371                 if ((dev->features & NETIF_F_HW_TLS_RX) &&
1372                     !dev->tlsdev_ops->tls_dev_resync)
1373                         return NOTIFY_BAD;
1374
1375                 if  (dev->tlsdev_ops &&
1376                      dev->tlsdev_ops->tls_dev_add &&
1377                      dev->tlsdev_ops->tls_dev_del)
1378                         return NOTIFY_DONE;
1379                 else
1380                         return NOTIFY_BAD;
1381         case NETDEV_DOWN:
1382                 return tls_device_down(dev);
1383         }
1384         return NOTIFY_DONE;
1385 }
1386
1387 static struct notifier_block tls_dev_notifier = {
1388         .notifier_call  = tls_dev_event,
1389 };
1390
1391 int __init tls_device_init(void)
1392 {
1393         int err;
1394
1395         dummy_page = alloc_page(GFP_KERNEL);
1396         if (!dummy_page)
1397                 return -ENOMEM;
1398
1399         destruct_wq = alloc_workqueue("ktls_device_destruct", 0, 0);
1400         if (!destruct_wq) {
1401                 err = -ENOMEM;
1402                 goto err_free_dummy;
1403         }
1404
1405         err = register_netdevice_notifier(&tls_dev_notifier);
1406         if (err)
1407                 goto err_destroy_wq;
1408
1409         return 0;
1410
1411 err_destroy_wq:
1412         destroy_workqueue(destruct_wq);
1413 err_free_dummy:
1414         put_page(dummy_page);
1415         return err;
1416 }
1417
1418 void __exit tls_device_cleanup(void)
1419 {
1420         unregister_netdevice_notifier(&tls_dev_notifier);
1421         flush_workqueue(destruct_wq);
1422         destroy_workqueue(destruct_wq);
1423         clean_acked_data_flush();
1424         put_page(dummy_page);
1425 }