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 void tls_device_gc_task(struct work_struct *work);
  49
  50 static DECLARE_WORK(tls_device_gc_work, tls_device_gc_task);
  51 static LIST_HEAD(tls_device_gc_list);
  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 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_gc_task(struct work_struct *work)
  71 {
  72         struct tls_context *ctx, *tmp;
  73         unsigned long flags;
  74         LIST_HEAD(gc_list);
  75
  76         spin_lock_irqsave(&tls_device_lock, flags);
  77         list_splice_init(&tls_device_gc_list, &gc_list);
  78         spin_unlock_irqrestore(&tls_device_lock, flags);
  79
  80         list_for_each_entry_safe(ctx, tmp, &gc_list, list) {
  81                 struct net_device *netdev = ctx->netdev;
  82
  83                 if (netdev && ctx->tx_conf == TLS_HW) {
  84                         netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
  85                                                         TLS_OFFLOAD_CTX_DIR_TX);
  86                         dev_put(netdev);
  87                         ctx->netdev = NULL;
  88                 }
  89
  90                 list_del(&ctx->list);
  91                 tls_device_free_ctx(ctx);
  92         }
  93 }
  94
  95 static void tls_device_queue_ctx_destruction(struct tls_context *ctx)
  96 {
  97         unsigned long flags;
  98
  99         spin_lock_irqsave(&tls_device_lock, flags);
 100         list_move_tail(&ctx->list, &tls_device_gc_list);
 101
 102         /* schedule_work inside the spinlock
 103          * to make sure tls_device_down waits for that work.
 104          */
 105         schedule_work(&tls_device_gc_work);
 106
 107         spin_unlock_irqrestore(&tls_device_lock, flags);
 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         if (refcount_dec_and_test(&tls_ctx->refcount))
 198                 tls_device_queue_ctx_destruction(tls_ctx);
 199 }
 200 EXPORT_SYMBOL_GPL(tls_device_sk_destruct);
 201
 202 void tls_device_free_resources_tx(struct sock *sk)
 203 {
 204         struct tls_context *tls_ctx = tls_get_ctx(sk);
 205
 206         tls_free_partial_record(sk, tls_ctx);
 207 }
 208
 209 void tls_offload_tx_resync_request(struct sock *sk, u32 got_seq, u32 exp_seq)
 210 {
 211         struct tls_context *tls_ctx = tls_get_ctx(sk);
 212
 213         trace_tls_device_tx_resync_req(sk, got_seq, exp_seq);
 214         WARN_ON(test_and_set_bit(TLS_TX_SYNC_SCHED, &tls_ctx->flags));
 215 }
 216 EXPORT_SYMBOL_GPL(tls_offload_tx_resync_request);
 217
 218 static void tls_device_resync_tx(struct sock *sk, struct tls_context *tls_ctx,
 219                                  u32 seq)
 220 {
 221         struct net_device *netdev;
 222         struct sk_buff *skb;
 223         int err = 0;
 224         u8 *rcd_sn;
 225
 226         skb = tcp_write_queue_tail(sk);
 227         if (skb)
 228                 TCP_SKB_CB(skb)->eor = 1;
 229
 230         rcd_sn = tls_ctx->tx.rec_seq;
 231
 232         trace_tls_device_tx_resync_send(sk, seq, rcd_sn);
 233         down_read(&device_offload_lock);
 234         netdev = tls_ctx->netdev;
 235         if (netdev)
 236                 err = netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq,
 237                                                          rcd_sn,
 238                                                          TLS_OFFLOAD_CTX_DIR_TX);
 239         up_read(&device_offload_lock);
 240         if (err)
 241                 return;
 242
 243         clear_bit_unlock(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
 244 }
 245
 246 static void tls_append_frag(struct tls_record_info *record,
 247                             struct page_frag *pfrag,
 248                             int size)
 249 {
 250         skb_frag_t *frag;
 251
 252         frag = &record->frags[record->num_frags - 1];
 253         if (skb_frag_page(frag) == pfrag->page &&
 254             skb_frag_off(frag) + skb_frag_size(frag) == pfrag->offset) {
 255                 skb_frag_size_add(frag, size);
 256         } else {
 257                 ++frag;
 258                 __skb_frag_set_page(frag, pfrag->page);
 259                 skb_frag_off_set(frag, pfrag->offset);
 260                 skb_frag_size_set(frag, size);
 261                 ++record->num_frags;
 262                 get_page(pfrag->page);
 263         }
 264
 265         pfrag->offset += size;
 266         record->len += size;
 267 }
 268
 269 static int tls_push_record(struct sock *sk,
 270                            struct tls_context *ctx,
 271                            struct tls_offload_context_tx *offload_ctx,
 272                            struct tls_record_info *record,
 273                            int flags)
 274 {
 275         struct tls_prot_info *prot = &ctx->prot_info;
 276         struct tcp_sock *tp = tcp_sk(sk);
 277         skb_frag_t *frag;
 278         int i;
 279
 280         record->end_seq = tp->write_seq + record->len;
 281         list_add_tail_rcu(&record->list, &offload_ctx->records_list);
 282         offload_ctx->open_record = NULL;
 283
 284         if (test_bit(TLS_TX_SYNC_SCHED, &ctx->flags))
 285                 tls_device_resync_tx(sk, ctx, tp->write_seq);
 286
 287         tls_advance_record_sn(sk, prot, &ctx->tx);
 288
 289         for (i = 0; i < record->num_frags; i++) {
 290                 frag = &record->frags[i];
 291                 sg_unmark_end(&offload_ctx->sg_tx_data[i]);
 292                 sg_set_page(&offload_ctx->sg_tx_data[i], skb_frag_page(frag),
 293                             skb_frag_size(frag), skb_frag_off(frag));
 294                 sk_mem_charge(sk, skb_frag_size(frag));
 295                 get_page(skb_frag_page(frag));
 296         }
 297         sg_mark_end(&offload_ctx->sg_tx_data[record->num_frags - 1]);
 298
 299         /* all ready, send */
 300         return tls_push_sg(sk, ctx, offload_ctx->sg_tx_data, 0, flags);
 301 }
 302
 303 static int tls_device_record_close(struct sock *sk,
 304                                    struct tls_context *ctx,
 305                                    struct tls_record_info *record,
 306                                    struct page_frag *pfrag,
 307                                    unsigned char record_type)
 308 {
 309         struct tls_prot_info *prot = &ctx->prot_info;
 310         int ret;
 311
 312         /* append tag
 313          * device will fill in the tag, we just need to append a placeholder
 314          * use socket memory to improve coalescing (re-using a single buffer
 315          * increases frag count)
 316          * if we can't allocate memory now, steal some back from data
 317          */
 318         if (likely(skb_page_frag_refill(prot->tag_size, pfrag,
 319                                         sk->sk_allocation))) {
 320                 ret = 0;
 321                 tls_append_frag(record, pfrag, prot->tag_size);
 322         } else {
 323                 ret = prot->tag_size;
 324                 if (record->len <= prot->overhead_size)
 325                         return -ENOMEM;
 326         }
 327
 328         /* fill prepend */
 329         tls_fill_prepend(ctx, skb_frag_address(&record->frags[0]),
 330                          record->len - prot->overhead_size,
 331                          record_type);
 332         return ret;
 333 }
 334
 335 static int tls_create_new_record(struct tls_offload_context_tx *offload_ctx,
 336                                  struct page_frag *pfrag,
 337                                  size_t prepend_size)
 338 {
 339         struct tls_record_info *record;
 340         skb_frag_t *frag;
 341
 342         record = kmalloc(sizeof(*record), GFP_KERNEL);
 343         if (!record)
 344                 return -ENOMEM;
 345
 346         frag = &record->frags[0];
 347         __skb_frag_set_page(frag, pfrag->page);
 348         skb_frag_off_set(frag, pfrag->offset);
 349         skb_frag_size_set(frag, prepend_size);
 350
 351         get_page(pfrag->page);
 352         pfrag->offset += prepend_size;
 353
 354         record->num_frags = 1;
 355         record->len = prepend_size;
 356         offload_ctx->open_record = record;
 357         return 0;
 358 }
 359
 360 static int tls_do_allocation(struct sock *sk,
 361                              struct tls_offload_context_tx *offload_ctx,
 362                              struct page_frag *pfrag,
 363                              size_t prepend_size)
 364 {
 365         int ret;
 366
 367         if (!offload_ctx->open_record) {
 368                 if (unlikely(!skb_page_frag_refill(prepend_size, pfrag,
 369                                                    sk->sk_allocation))) {
 370                         READ_ONCE(sk->sk_prot)->enter_memory_pressure(sk);
 371                         sk_stream_moderate_sndbuf(sk);
 372                         return -ENOMEM;
 373                 }
 374
 375                 ret = tls_create_new_record(offload_ctx, pfrag, prepend_size);
 376                 if (ret)
 377                         return ret;
 378
 379                 if (pfrag->size > pfrag->offset)
 380                         return 0;
 381         }
 382
 383         if (!sk_page_frag_refill(sk, pfrag))
 384                 return -ENOMEM;
 385
 386         return 0;
 387 }
 388
 389 static int tls_device_copy_data(void *addr, size_t bytes, struct iov_iter *i)
 390 {
 391         size_t pre_copy, nocache;
 392
 393         pre_copy = ~((unsigned long)addr - 1) & (SMP_CACHE_BYTES - 1);
 394         if (pre_copy) {
 395                 pre_copy = min(pre_copy, bytes);
 396                 if (copy_from_iter(addr, pre_copy, i) != pre_copy)
 397                         return -EFAULT;
 398                 bytes -= pre_copy;
 399                 addr += pre_copy;
 400         }
 401
 402         nocache = round_down(bytes, SMP_CACHE_BYTES);
 403         if (copy_from_iter_nocache(addr, nocache, i) != nocache)
 404                 return -EFAULT;
 405         bytes -= nocache;
 406         addr += nocache;
 407
 408         if (bytes && copy_from_iter(addr, bytes, i) != bytes)
 409                 return -EFAULT;
 410
 411         return 0;
 412 }
 413
 414 union tls_iter_offset {
 415         struct iov_iter *msg_iter;
 416         int offset;
 417 };
 418
 419 static int tls_push_data(struct sock *sk,
 420                          union tls_iter_offset iter_offset,
 421                          size_t size, int flags,
 422                          unsigned char record_type,
 423                          struct page *zc_page)
 424 {
 425         struct tls_context *tls_ctx = tls_get_ctx(sk);
 426         struct tls_prot_info *prot = &tls_ctx->prot_info;
 427         struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
 428         struct tls_record_info *record;
 429         int tls_push_record_flags;
 430         struct page_frag *pfrag;
 431         size_t orig_size = size;
 432         u32 max_open_record_len;
 433         bool more = false;
 434         bool done = false;
 435         int copy, rc = 0;
 436         long timeo;
 437
 438         if (flags &
 439             ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_SENDPAGE_NOTLAST))
 440                 return -EOPNOTSUPP;
 441
 442         if (unlikely(sk->sk_err))
 443                 return -sk->sk_err;
 444
 445         flags |= MSG_SENDPAGE_DECRYPTED;
 446         tls_push_record_flags = flags | MSG_SENDPAGE_NOTLAST;
 447
 448         timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
 449         if (tls_is_partially_sent_record(tls_ctx)) {
 450                 rc = tls_push_partial_record(sk, tls_ctx, flags);
 451                 if (rc < 0)
 452                         return rc;
 453         }
 454
 455         pfrag = sk_page_frag(sk);
 456
 457         /* TLS_HEADER_SIZE is not counted as part of the TLS record, and
 458          * we need to leave room for an authentication tag.
 459          */
 460         max_open_record_len = TLS_MAX_PAYLOAD_SIZE +
 461                               prot->prepend_size;
 462         do {
 463                 rc = tls_do_allocation(sk, ctx, pfrag, prot->prepend_size);
 464                 if (unlikely(rc)) {
 465                         rc = sk_stream_wait_memory(sk, &timeo);
 466                         if (!rc)
 467                                 continue;
 468
 469                         record = ctx->open_record;
 470                         if (!record)
 471                                 break;
 472 handle_error:
 473                         if (record_type != TLS_RECORD_TYPE_DATA) {
 474                                 /* avoid sending partial
 475                                  * record with type !=
 476                                  * application_data
 477                                  */
 478                                 size = orig_size;
 479                                 destroy_record(record);
 480                                 ctx->open_record = NULL;
 481                         } else if (record->len > prot->prepend_size) {
 482                                 goto last_record;
 483                         }
 484
 485                         break;
 486                 }
 487
 488                 record = ctx->open_record;
 489
 490                 copy = min_t(size_t, size, max_open_record_len - record->len);
 491                 if (copy && zc_page) {
 492                         struct page_frag zc_pfrag;
 493
 494                         zc_pfrag.page = zc_page;
 495                         zc_pfrag.offset = iter_offset.offset;
 496                         zc_pfrag.size = copy;
 497                         tls_append_frag(record, &zc_pfrag, copy);
 498                 } else if (copy) {
 499                         copy = min_t(size_t, copy, pfrag->size - pfrag->offset);
 500
 501                         rc = tls_device_copy_data(page_address(pfrag->page) +
 502                                                   pfrag->offset, copy,
 503                                                   iter_offset.msg_iter);
 504                         if (rc)
 505                                 goto handle_error;
 506                         tls_append_frag(record, pfrag, copy);
 507                 }
 508
 509                 size -= copy;
 510                 if (!size) {
 511 last_record:
 512                         tls_push_record_flags = flags;
 513                         if (flags & (MSG_SENDPAGE_NOTLAST | MSG_MORE)) {
 514                                 more = true;
 515                                 break;
 516                         }
 517
 518                         done = true;
 519                 }
 520
 521                 if (done || record->len >= max_open_record_len ||
 522                     (record->num_frags >= MAX_SKB_FRAGS - 1)) {
 523                         rc = tls_device_record_close(sk, tls_ctx, record,
 524                                                      pfrag, record_type);
 525                         if (rc) {
 526                                 if (rc > 0) {
 527                                         size += rc;
 528                                 } else {
 529                                         size = orig_size;
 530                                         destroy_record(record);
 531                                         ctx->open_record = NULL;
 532                                         break;
 533                                 }
 534                         }
 535
 536                         rc = tls_push_record(sk,
 537                                              tls_ctx,
 538                                              ctx,
 539                                              record,
 540                                              tls_push_record_flags);
 541                         if (rc < 0)
 542                                 break;
 543                 }
 544         } while (!done);
 545
 546         tls_ctx->pending_open_record_frags = more;
 547
 548         if (orig_size - size > 0)
 549                 rc = orig_size - size;
 550
 551         return rc;
 552 }
 553
 554 int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
 555 {
 556         unsigned char record_type = TLS_RECORD_TYPE_DATA;
 557         struct tls_context *tls_ctx = tls_get_ctx(sk);
 558         union tls_iter_offset iter;
 559         int rc;
 560
 561         mutex_lock(&tls_ctx->tx_lock);
 562         lock_sock(sk);
 563
 564         if (unlikely(msg->msg_controllen)) {
 565                 rc = tls_proccess_cmsg(sk, msg, &record_type);
 566                 if (rc)
 567                         goto out;
 568         }
 569
 570         iter.msg_iter = &msg->msg_iter;
 571         rc = tls_push_data(sk, iter, size, msg->msg_flags, record_type, NULL);
 572
 573 out:
 574         release_sock(sk);
 575         mutex_unlock(&tls_ctx->tx_lock);
 576         return rc;
 577 }
 578
 579 int tls_device_sendpage(struct sock *sk, struct page *page,
 580                         int offset, size_t size, int flags)
 581 {
 582         struct tls_context *tls_ctx = tls_get_ctx(sk);
 583         union tls_iter_offset iter_offset;
 584         struct iov_iter msg_iter;
 585         char *kaddr;
 586         struct kvec iov;
 587         int rc;
 588
 589         if (flags & MSG_SENDPAGE_NOTLAST)
 590                 flags |= MSG_MORE;
 591
 592         mutex_lock(&tls_ctx->tx_lock);
 593         lock_sock(sk);
 594
 595         if (flags & MSG_OOB) {
 596                 rc = -EOPNOTSUPP;
 597                 goto out;
 598         }
 599
 600         if (tls_ctx->zerocopy_sendfile) {
 601                 iter_offset.offset = offset;
 602                 rc = tls_push_data(sk, iter_offset, size,
 603                                    flags, TLS_RECORD_TYPE_DATA, page);
 604                 goto out;
 605         }
 606
 607         kaddr = kmap(page);
 608         iov.iov_base = kaddr + offset;
 609         iov.iov_len = size;
 610         iov_iter_kvec(&msg_iter, WRITE, &iov, 1, size);
 611         iter_offset.msg_iter = &msg_iter;
 612         rc = tls_push_data(sk, iter_offset, size, flags, TLS_RECORD_TYPE_DATA,
 613                            NULL);
 614         kunmap(page);
 615
 616 out:
 617         release_sock(sk);
 618         mutex_unlock(&tls_ctx->tx_lock);
 619         return rc;
 620 }
 621
 622 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
 623                                        u32 seq, u64 *p_record_sn)
 624 {
 625         u64 record_sn = context->hint_record_sn;
 626         struct tls_record_info *info, *last;
 627
 628         info = context->retransmit_hint;
 629         if (!info ||
 630             before(seq, info->end_seq - info->len)) {
 631                 /* if retransmit_hint is irrelevant start
 632                  * from the beginning of the list
 633                  */
 634                 info = list_first_entry_or_null(&context->records_list,
 635                                                 struct tls_record_info, list);
 636                 if (!info)
 637                         return NULL;
 638                 /* send the start_marker record if seq number is before the
 639                  * tls offload start marker sequence number. This record is
 640                  * required to handle TCP packets which are before TLS offload
 641                  * started.
 642                  *  And if it's not start marker, look if this seq number
 643                  * belongs to the list.
 644                  */
 645                 if (likely(!tls_record_is_start_marker(info))) {
 646                         /* we have the first record, get the last record to see
 647                          * if this seq number belongs to the list.
 648                          */
 649                         last = list_last_entry(&context->records_list,
 650                                                struct tls_record_info, list);
 651
 652                         if (!between(seq, tls_record_start_seq(info),
 653                                      last->end_seq))
 654                                 return NULL;
 655                 }
 656                 record_sn = context->unacked_record_sn;
 657         }
 658
 659         /* We just need the _rcu for the READ_ONCE() */
 660         rcu_read_lock();
 661         list_for_each_entry_from_rcu(info, &context->records_list, list) {
 662                 if (before(seq, info->end_seq)) {
 663                         if (!context->retransmit_hint ||
 664                             after(info->end_seq,
 665                                   context->retransmit_hint->end_seq)) {
 666                                 context->hint_record_sn = record_sn;
 667                                 context->retransmit_hint = info;
 668                         }
 669                         *p_record_sn = record_sn;
 670                         goto exit_rcu_unlock;
 671                 }
 672                 record_sn++;
 673         }
 674         info = NULL;
 675
 676 exit_rcu_unlock:
 677         rcu_read_unlock();
 678         return info;
 679 }
 680 EXPORT_SYMBOL(tls_get_record);
 681
 682 static int tls_device_push_pending_record(struct sock *sk, int flags)
 683 {
 684         union tls_iter_offset iter;
 685         struct iov_iter msg_iter;
 686
 687         iov_iter_kvec(&msg_iter, WRITE, NULL, 0, 0);
 688         iter.msg_iter = &msg_iter;
 689         return tls_push_data(sk, iter, 0, flags, TLS_RECORD_TYPE_DATA, NULL);
 690 }
 691
 692 void tls_device_write_space(struct sock *sk, struct tls_context *ctx)
 693 {
 694         if (tls_is_partially_sent_record(ctx)) {
 695                 gfp_t sk_allocation = sk->sk_allocation;
 696
 697                 WARN_ON_ONCE(sk->sk_write_pending);
 698
 699                 sk->sk_allocation = GFP_ATOMIC;
 700                 tls_push_partial_record(sk, ctx,
 701                                         MSG_DONTWAIT | MSG_NOSIGNAL |
 702                                         MSG_SENDPAGE_DECRYPTED);
 703                 sk->sk_allocation = sk_allocation;
 704         }
 705 }
 706
 707 static void tls_device_resync_rx(struct tls_context *tls_ctx,
 708                                  struct sock *sk, u32 seq, u8 *rcd_sn)
 709 {
 710         struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
 711         struct net_device *netdev;
 712
 713         trace_tls_device_rx_resync_send(sk, seq, rcd_sn, rx_ctx->resync_type);
 714         rcu_read_lock();
 715         netdev = READ_ONCE(tls_ctx->netdev);
 716         if (netdev)
 717                 netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq, rcd_sn,
 718                                                    TLS_OFFLOAD_CTX_DIR_RX);
 719         rcu_read_unlock();
 720         TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXDEVICERESYNC);
 721 }
 722
 723 static bool
 724 tls_device_rx_resync_async(struct tls_offload_resync_async *resync_async,
 725                            s64 resync_req, u32 *seq, u16 *rcd_delta)
 726 {
 727         u32 is_async = resync_req & RESYNC_REQ_ASYNC;
 728         u32 req_seq = resync_req >> 32;
 729         u32 req_end = req_seq + ((resync_req >> 16) & 0xffff);
 730         u16 i;
 731
 732         *rcd_delta = 0;
 733
 734         if (is_async) {
 735                 /* shouldn't get to wraparound:
 736                  * too long in async stage, something bad happened
 737                  */
 738                 if (WARN_ON_ONCE(resync_async->rcd_delta == USHRT_MAX))
 739                         return false;
 740
 741                 /* asynchronous stage: log all headers seq such that
 742                  * req_seq <= seq <= end_seq, and wait for real resync request
 743                  */
 744                 if (before(*seq, req_seq))
 745                         return false;
 746                 if (!after(*seq, req_end) &&
 747                     resync_async->loglen < TLS_DEVICE_RESYNC_ASYNC_LOGMAX)
 748                         resync_async->log[resync_async->loglen++] = *seq;
 749
 750                 resync_async->rcd_delta++;
 751
 752                 return false;
 753         }
 754
 755         /* synchronous stage: check against the logged entries and
 756          * proceed to check the next entries if no match was found
 757          */
 758         for (i = 0; i < resync_async->loglen; i++)
 759                 if (req_seq == resync_async->log[i] &&
 760                     atomic64_try_cmpxchg(&resync_async->req, &resync_req, 0)) {
 761                         *rcd_delta = resync_async->rcd_delta - i;
 762                         *seq = req_seq;
 763                         resync_async->loglen = 0;
 764                         resync_async->rcd_delta = 0;
 765                         return true;
 766                 }
 767
 768         resync_async->loglen = 0;
 769         resync_async->rcd_delta = 0;
 770
 771         if (req_seq == *seq &&
 772             atomic64_try_cmpxchg(&resync_async->req,
 773                                  &resync_req, 0))
 774                 return true;
 775
 776         return false;
 777 }
 778
 779 void tls_device_rx_resync_new_rec(struct sock *sk, u32 rcd_len, u32 seq)
 780 {
 781         struct tls_context *tls_ctx = tls_get_ctx(sk);
 782         struct tls_offload_context_rx *rx_ctx;
 783         u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
 784         u32 sock_data, is_req_pending;
 785         struct tls_prot_info *prot;
 786         s64 resync_req;
 787         u16 rcd_delta;
 788         u32 req_seq;
 789
 790         if (tls_ctx->rx_conf != TLS_HW)
 791                 return;
 792         if (unlikely(test_bit(TLS_RX_DEV_DEGRADED, &tls_ctx->flags)))
 793                 return;
 794
 795         prot = &tls_ctx->prot_info;
 796         rx_ctx = tls_offload_ctx_rx(tls_ctx);
 797         memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
 798
 799         switch (rx_ctx->resync_type) {
 800         case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ:
 801                 resync_req = atomic64_read(&rx_ctx->resync_req);
 802                 req_seq = resync_req >> 32;
 803                 seq += TLS_HEADER_SIZE - 1;
 804                 is_req_pending = resync_req;
 805
 806                 if (likely(!is_req_pending) || req_seq != seq ||
 807                     !atomic64_try_cmpxchg(&rx_ctx->resync_req, &resync_req, 0))
 808                         return;
 809                 break;
 810         case TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT:
 811                 if (likely(!rx_ctx->resync_nh_do_now))
 812                         return;
 813
 814                 /* head of next rec is already in, note that the sock_inq will
 815                  * include the currently parsed message when called from parser
 816                  */
 817                 sock_data = tcp_inq(sk);
 818                 if (sock_data > rcd_len) {
 819                         trace_tls_device_rx_resync_nh_delay(sk, sock_data,
 820                                                             rcd_len);
 821                         return;
 822                 }
 823
 824                 rx_ctx->resync_nh_do_now = 0;
 825                 seq += rcd_len;
 826                 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
 827                 break;
 828         case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC:
 829                 resync_req = atomic64_read(&rx_ctx->resync_async->req);
 830                 is_req_pending = resync_req;
 831                 if (likely(!is_req_pending))
 832                         return;
 833
 834                 if (!tls_device_rx_resync_async(rx_ctx->resync_async,
 835                                                 resync_req, &seq, &rcd_delta))
 836                         return;
 837                 tls_bigint_subtract(rcd_sn, rcd_delta);
 838                 break;
 839         }
 840
 841         tls_device_resync_rx(tls_ctx, sk, seq, rcd_sn);
 842 }
 843
 844 static void tls_device_core_ctrl_rx_resync(struct tls_context *tls_ctx,
 845                                            struct tls_offload_context_rx *ctx,
 846                                            struct sock *sk, struct sk_buff *skb)
 847 {
 848         struct strp_msg *rxm;
 849
 850         /* device will request resyncs by itself based on stream scan */
 851         if (ctx->resync_type != TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT)
 852                 return;
 853         /* already scheduled */
 854         if (ctx->resync_nh_do_now)
 855                 return;
 856         /* seen decrypted fragments since last fully-failed record */
 857         if (ctx->resync_nh_reset) {
 858                 ctx->resync_nh_reset = 0;
 859                 ctx->resync_nh.decrypted_failed = 1;
 860                 ctx->resync_nh.decrypted_tgt = TLS_DEVICE_RESYNC_NH_START_IVAL;
 861                 return;
 862         }
 863
 864         if (++ctx->resync_nh.decrypted_failed <= ctx->resync_nh.decrypted_tgt)
 865                 return;
 866
 867         /* doing resync, bump the next target in case it fails */
 868         if (ctx->resync_nh.decrypted_tgt < TLS_DEVICE_RESYNC_NH_MAX_IVAL)
 869                 ctx->resync_nh.decrypted_tgt *= 2;
 870         else
 871                 ctx->resync_nh.decrypted_tgt += TLS_DEVICE_RESYNC_NH_MAX_IVAL;
 872
 873         rxm = strp_msg(skb);
 874
 875         /* head of next rec is already in, parser will sync for us */
 876         if (tcp_inq(sk) > rxm->full_len) {
 877                 trace_tls_device_rx_resync_nh_schedule(sk);
 878                 ctx->resync_nh_do_now = 1;
 879         } else {
 880                 struct tls_prot_info *prot = &tls_ctx->prot_info;
 881                 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
 882
 883                 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
 884                 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
 885
 886                 tls_device_resync_rx(tls_ctx, sk, tcp_sk(sk)->copied_seq,
 887                                      rcd_sn);
 888         }
 889 }
 890
 891 static int tls_device_reencrypt(struct sock *sk, struct sk_buff *skb)
 892 {
 893         struct strp_msg *rxm = strp_msg(skb);
 894         int err = 0, offset = rxm->offset, copy, nsg, data_len, pos;
 895         struct sk_buff *skb_iter, *unused;
 896         struct scatterlist sg[1];
 897         char *orig_buf, *buf;
 898
 899         orig_buf = kmalloc(rxm->full_len + TLS_HEADER_SIZE +
 900                            TLS_CIPHER_AES_GCM_128_IV_SIZE, sk->sk_allocation);
 901         if (!orig_buf)
 902                 return -ENOMEM;
 903         buf = orig_buf;
 904
 905         nsg = skb_cow_data(skb, 0, &unused);
 906         if (unlikely(nsg < 0)) {
 907                 err = nsg;
 908                 goto free_buf;
 909         }
 910
 911         sg_init_table(sg, 1);
 912         sg_set_buf(&sg[0], buf,
 913                    rxm->full_len + TLS_HEADER_SIZE +
 914                    TLS_CIPHER_AES_GCM_128_IV_SIZE);
 915         err = skb_copy_bits(skb, offset, buf,
 916                             TLS_HEADER_SIZE + TLS_CIPHER_AES_GCM_128_IV_SIZE);
 917         if (err)
 918                 goto free_buf;
 919
 920         /* We are interested only in the decrypted data not the auth */
 921         err = decrypt_skb(sk, skb, sg);
 922         if (err != -EBADMSG)
 923                 goto free_buf;
 924         else
 925                 err = 0;
 926
 927         data_len = rxm->full_len - TLS_CIPHER_AES_GCM_128_TAG_SIZE;
 928
 929         if (skb_pagelen(skb) > offset) {
 930                 copy = min_t(int, skb_pagelen(skb) - offset, data_len);
 931
 932                 if (skb->decrypted) {
 933                         err = skb_store_bits(skb, offset, buf, copy);
 934                         if (err)
 935                                 goto free_buf;
 936                 }
 937
 938                 offset += copy;
 939                 buf += copy;
 940         }
 941
 942         pos = skb_pagelen(skb);
 943         skb_walk_frags(skb, skb_iter) {
 944                 int frag_pos;
 945
 946                 /* Practically all frags must belong to msg if reencrypt
 947                  * is needed with current strparser and coalescing logic,
 948                  * but strparser may "get optimized", so let's be safe.
 949                  */
 950                 if (pos + skb_iter->len <= offset)
 951                         goto done_with_frag;
 952                 if (pos >= data_len + rxm->offset)
 953                         break;
 954
 955                 frag_pos = offset - pos;
 956                 copy = min_t(int, skb_iter->len - frag_pos,
 957                              data_len + rxm->offset - offset);
 958
 959                 if (skb_iter->decrypted) {
 960                         err = skb_store_bits(skb_iter, frag_pos, buf, copy);
 961                         if (err)
 962                                 goto free_buf;
 963                 }
 964
 965                 offset += copy;
 966                 buf += copy;
 967 done_with_frag:
 968                 pos += skb_iter->len;
 969         }
 970
 971 free_buf:
 972         kfree(orig_buf);
 973         return err;
 974 }
 975
 976 int tls_device_decrypted(struct sock *sk, struct tls_context *tls_ctx,
 977                          struct sk_buff *skb, struct strp_msg *rxm)
 978 {
 979         struct tls_offload_context_rx *ctx = tls_offload_ctx_rx(tls_ctx);
 980         int is_decrypted = skb->decrypted;
 981         int is_encrypted = !is_decrypted;
 982         struct sk_buff *skb_iter;
 983
 984         /* Check if all the data is decrypted already */
 985         skb_walk_frags(skb, skb_iter) {
 986                 is_decrypted &= skb_iter->decrypted;
 987                 is_encrypted &= !skb_iter->decrypted;
 988         }
 989
 990         trace_tls_device_decrypted(sk, tcp_sk(sk)->copied_seq - rxm->full_len,
 991                                    tls_ctx->rx.rec_seq, rxm->full_len,
 992                                    is_encrypted, is_decrypted);
 993
 994         if (unlikely(test_bit(TLS_RX_DEV_DEGRADED, &tls_ctx->flags))) {
 995                 if (likely(is_encrypted || is_decrypted))
 996                         return is_decrypted;
 997
 998                 /* After tls_device_down disables the offload, the next SKB will
 999                  * likely have initial fragments decrypted, and final ones not
1000                  * decrypted. We need to reencrypt that single SKB.
1001                  */
1002                 return tls_device_reencrypt(sk, skb);
1003         }
1004
1005         /* Return immediately if the record is either entirely plaintext or
1006          * entirely ciphertext. Otherwise handle reencrypt partially decrypted
1007          * record.
1008          */
1009         if (is_decrypted) {
1010                 ctx->resync_nh_reset = 1;
1011                 return is_decrypted;
1012         }
1013         if (is_encrypted) {
1014                 tls_device_core_ctrl_rx_resync(tls_ctx, ctx, sk, skb);
1015                 return 0;
1016         }
1017
1018         ctx->resync_nh_reset = 1;
1019         return tls_device_reencrypt(sk, skb);
1020 }
1021
1022 static void tls_device_attach(struct tls_context *ctx, struct sock *sk,
1023                               struct net_device *netdev)
1024 {
1025         if (sk->sk_destruct != tls_device_sk_destruct) {
1026                 refcount_set(&ctx->refcount, 1);
1027                 dev_hold(netdev);
1028                 ctx->netdev = netdev;
1029                 spin_lock_irq(&tls_device_lock);
1030                 list_add_tail(&ctx->list, &tls_device_list);
1031                 spin_unlock_irq(&tls_device_lock);
1032
1033                 ctx->sk_destruct = sk->sk_destruct;
1034                 smp_store_release(&sk->sk_destruct, tls_device_sk_destruct);
1035         }
1036 }
1037
1038 int tls_set_device_offload(struct sock *sk, struct tls_context *ctx)
1039 {
1040         u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
1041         struct tls_context *tls_ctx = tls_get_ctx(sk);
1042         struct tls_prot_info *prot = &tls_ctx->prot_info;
1043         struct tls_record_info *start_marker_record;
1044         struct tls_offload_context_tx *offload_ctx;
1045         struct tls_crypto_info *crypto_info;
1046         struct net_device *netdev;
1047         char *iv, *rec_seq;
1048         struct sk_buff *skb;
1049         __be64 rcd_sn;
1050         int rc;
1051
1052         if (!ctx)
1053                 return -EINVAL;
1054
1055         if (ctx->priv_ctx_tx)
1056                 return -EEXIST;
1057
1058         netdev = get_netdev_for_sock(sk);
1059         if (!netdev) {
1060                 pr_err_ratelimited("%s: netdev not found\n", __func__);
1061                 return -EINVAL;
1062         }
1063
1064         if (!(netdev->features & NETIF_F_HW_TLS_TX)) {
1065                 rc = -EOPNOTSUPP;
1066                 goto release_netdev;
1067         }
1068
1069         crypto_info = &ctx->crypto_send.info;
1070         if (crypto_info->version != TLS_1_2_VERSION) {
1071                 rc = -EOPNOTSUPP;
1072                 goto release_netdev;
1073         }
1074
1075         switch (crypto_info->cipher_type) {
1076         case TLS_CIPHER_AES_GCM_128:
1077                 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
1078                 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
1079                 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
1080                 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
1081                 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
1082                 salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
1083                 rec_seq =
1084                  ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
1085                 break;
1086         default:
1087                 rc = -EINVAL;
1088                 goto release_netdev;
1089         }
1090
1091         /* Sanity-check the rec_seq_size for stack allocations */
1092         if (rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
1093                 rc = -EINVAL;
1094                 goto release_netdev;
1095         }
1096
1097         prot->version = crypto_info->version;
1098         prot->cipher_type = crypto_info->cipher_type;
1099         prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
1100         prot->tag_size = tag_size;
1101         prot->overhead_size = prot->prepend_size + prot->tag_size;
1102         prot->iv_size = iv_size;
1103         prot->salt_size = salt_size;
1104         ctx->tx.iv = kmalloc(iv_size + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
1105                              GFP_KERNEL);
1106         if (!ctx->tx.iv) {
1107                 rc = -ENOMEM;
1108                 goto release_netdev;
1109         }
1110
1111         memcpy(ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv, iv_size);
1112
1113         prot->rec_seq_size = rec_seq_size;
1114         ctx->tx.rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
1115         if (!ctx->tx.rec_seq) {
1116                 rc = -ENOMEM;
1117                 goto free_iv;
1118         }
1119
1120         start_marker_record = kmalloc(sizeof(*start_marker_record), GFP_KERNEL);
1121         if (!start_marker_record) {
1122                 rc = -ENOMEM;
1123                 goto free_rec_seq;
1124         }
1125
1126         offload_ctx = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_TX, GFP_KERNEL);
1127         if (!offload_ctx) {
1128                 rc = -ENOMEM;
1129                 goto free_marker_record;
1130         }
1131
1132         rc = tls_sw_fallback_init(sk, offload_ctx, crypto_info);
1133         if (rc)
1134                 goto free_offload_ctx;
1135
1136         /* start at rec_seq - 1 to account for the start marker record */
1137         memcpy(&rcd_sn, ctx->tx.rec_seq, sizeof(rcd_sn));
1138         offload_ctx->unacked_record_sn = be64_to_cpu(rcd_sn) - 1;
1139
1140         start_marker_record->end_seq = tcp_sk(sk)->write_seq;
1141         start_marker_record->len = 0;
1142         start_marker_record->num_frags = 0;
1143
1144         INIT_LIST_HEAD(&offload_ctx->records_list);
1145         list_add_tail(&start_marker_record->list, &offload_ctx->records_list);
1146         spin_lock_init(&offload_ctx->lock);
1147         sg_init_table(offload_ctx->sg_tx_data,
1148                       ARRAY_SIZE(offload_ctx->sg_tx_data));
1149
1150         clean_acked_data_enable(inet_csk(sk), &tls_icsk_clean_acked);
1151         ctx->push_pending_record = tls_device_push_pending_record;
1152
1153         /* TLS offload is greatly simplified if we don't send
1154          * SKBs where only part of the payload needs to be encrypted.
1155          * So mark the last skb in the write queue as end of record.
1156          */
1157         skb = tcp_write_queue_tail(sk);
1158         if (skb)
1159                 TCP_SKB_CB(skb)->eor = 1;
1160
1161         /* Avoid offloading if the device is down
1162          * We don't want to offload new flows after
1163          * the NETDEV_DOWN event
1164          *
1165          * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1166          * handler thus protecting from the device going down before
1167          * ctx was added to tls_device_list.
1168          */
1169         down_read(&device_offload_lock);
1170         if (!(netdev->flags & IFF_UP)) {
1171                 rc = -EINVAL;
1172                 goto release_lock;
1173         }
1174
1175         ctx->priv_ctx_tx = offload_ctx;
1176         rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_TX,
1177                                              &ctx->crypto_send.info,
1178                                              tcp_sk(sk)->write_seq);
1179         trace_tls_device_offload_set(sk, TLS_OFFLOAD_CTX_DIR_TX,
1180                                      tcp_sk(sk)->write_seq, rec_seq, rc);
1181         if (rc)
1182                 goto release_lock;
1183
1184         tls_device_attach(ctx, sk, netdev);
1185         up_read(&device_offload_lock);
1186
1187         /* following this assignment tls_is_sk_tx_device_offloaded
1188          * will return true and the context might be accessed
1189          * by the netdev's xmit function.
1190          */
1191         smp_store_release(&sk->sk_validate_xmit_skb, tls_validate_xmit_skb);
1192         dev_put(netdev);
1193
1194         return 0;
1195
1196 release_lock:
1197         up_read(&device_offload_lock);
1198         clean_acked_data_disable(inet_csk(sk));
1199         crypto_free_aead(offload_ctx->aead_send);
1200 free_offload_ctx:
1201         kfree(offload_ctx);
1202         ctx->priv_ctx_tx = NULL;
1203 free_marker_record:
1204         kfree(start_marker_record);
1205 free_rec_seq:
1206         kfree(ctx->tx.rec_seq);
1207 free_iv:
1208         kfree(ctx->tx.iv);
1209 release_netdev:
1210         dev_put(netdev);
1211         return rc;
1212 }
1213
1214 int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx)
1215 {
1216         struct tls12_crypto_info_aes_gcm_128 *info;
1217         struct tls_offload_context_rx *context;
1218         struct net_device *netdev;
1219         int rc = 0;
1220
1221         if (ctx->crypto_recv.info.version != TLS_1_2_VERSION)
1222                 return -EOPNOTSUPP;
1223
1224         netdev = get_netdev_for_sock(sk);
1225         if (!netdev) {
1226                 pr_err_ratelimited("%s: netdev not found\n", __func__);
1227                 return -EINVAL;
1228         }
1229
1230         if (!(netdev->features & NETIF_F_HW_TLS_RX)) {
1231                 rc = -EOPNOTSUPP;
1232                 goto release_netdev;
1233         }
1234
1235         /* Avoid offloading if the device is down
1236          * We don't want to offload new flows after
1237          * the NETDEV_DOWN event
1238          *
1239          * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1240          * handler thus protecting from the device going down before
1241          * ctx was added to tls_device_list.
1242          */
1243         down_read(&device_offload_lock);
1244         if (!(netdev->flags & IFF_UP)) {
1245                 rc = -EINVAL;
1246                 goto release_lock;
1247         }
1248
1249         context = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_RX, GFP_KERNEL);
1250         if (!context) {
1251                 rc = -ENOMEM;
1252                 goto release_lock;
1253         }
1254         context->resync_nh_reset = 1;
1255
1256         ctx->priv_ctx_rx = context;
1257         rc = tls_set_sw_offload(sk, ctx, 0);
1258         if (rc)
1259                 goto release_ctx;
1260
1261         rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_RX,
1262                                              &ctx->crypto_recv.info,
1263                                              tcp_sk(sk)->copied_seq);
1264         info = (void *)&ctx->crypto_recv.info;
1265         trace_tls_device_offload_set(sk, TLS_OFFLOAD_CTX_DIR_RX,
1266                                      tcp_sk(sk)->copied_seq, info->rec_seq, rc);
1267         if (rc)
1268                 goto free_sw_resources;
1269
1270         tls_device_attach(ctx, sk, netdev);
1271         up_read(&device_offload_lock);
1272
1273         dev_put(netdev);
1274
1275         return 0;
1276
1277 free_sw_resources:
1278         up_read(&device_offload_lock);
1279         tls_sw_free_resources_rx(sk);
1280         down_read(&device_offload_lock);
1281 release_ctx:
1282         ctx->priv_ctx_rx = NULL;
1283 release_lock:
1284         up_read(&device_offload_lock);
1285 release_netdev:
1286         dev_put(netdev);
1287         return rc;
1288 }
1289
1290 void tls_device_offload_cleanup_rx(struct sock *sk)
1291 {
1292         struct tls_context *tls_ctx = tls_get_ctx(sk);
1293         struct net_device *netdev;
1294
1295         down_read(&device_offload_lock);
1296         netdev = tls_ctx->netdev;
1297         if (!netdev)
1298                 goto out;
1299
1300         netdev->tlsdev_ops->tls_dev_del(netdev, tls_ctx,
1301                                         TLS_OFFLOAD_CTX_DIR_RX);
1302
1303         if (tls_ctx->tx_conf != TLS_HW) {
1304                 dev_put(netdev);
1305                 tls_ctx->netdev = NULL;
1306         } else {
1307                 set_bit(TLS_RX_DEV_CLOSED, &tls_ctx->flags);
1308         }
1309 out:
1310         up_read(&device_offload_lock);
1311         tls_sw_release_resources_rx(sk);
1312 }
1313
1314 static int tls_device_down(struct net_device *netdev)
1315 {
1316         struct tls_context *ctx, *tmp;
1317         unsigned long flags;
1318         LIST_HEAD(list);
1319
1320         /* Request a write lock to block new offload attempts */
1321         down_write(&device_offload_lock);
1322
1323         spin_lock_irqsave(&tls_device_lock, flags);
1324         list_for_each_entry_safe(ctx, tmp, &tls_device_list, list) {
1325                 if (ctx->netdev != netdev ||
1326                     !refcount_inc_not_zero(&ctx->refcount))
1327                         continue;
1328
1329                 list_move(&ctx->list, &list);
1330         }
1331         spin_unlock_irqrestore(&tls_device_lock, flags);
1332
1333         list_for_each_entry_safe(ctx, tmp, &list, list) {
1334                 /* Stop offloaded TX and switch to the fallback.
1335                  * tls_is_sk_tx_device_offloaded will return false.
1336                  */
1337                 WRITE_ONCE(ctx->sk->sk_validate_xmit_skb, tls_validate_xmit_skb_sw);
1338
1339                 /* Stop the RX and TX resync.
1340                  * tls_dev_resync must not be called after tls_dev_del.
1341                  */
1342                 WRITE_ONCE(ctx->netdev, NULL);
1343
1344                 /* Start skipping the RX resync logic completely. */
1345                 set_bit(TLS_RX_DEV_DEGRADED, &ctx->flags);
1346
1347                 /* Sync with inflight packets. After this point:
1348                  * TX: no non-encrypted packets will be passed to the driver.
1349                  * RX: resync requests from the driver will be ignored.
1350                  */
1351                 synchronize_net();
1352
1353                 /* Release the offload context on the driver side. */
1354                 if (ctx->tx_conf == TLS_HW)
1355                         netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1356                                                         TLS_OFFLOAD_CTX_DIR_TX);
1357                 if (ctx->rx_conf == TLS_HW &&
1358                     !test_bit(TLS_RX_DEV_CLOSED, &ctx->flags))
1359                         netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1360                                                         TLS_OFFLOAD_CTX_DIR_RX);
1361
1362                 dev_put(netdev);
1363
1364                 /* Move the context to a separate list for two reasons:
1365                  * 1. When the context is deallocated, list_del is called.
1366                  * 2. It's no longer an offloaded context, so we don't want to
1367                  *    run offload-specific code on this context.
1368                  */
1369                 spin_lock_irqsave(&tls_device_lock, flags);
1370                 list_move_tail(&ctx->list, &tls_device_down_list);
1371                 spin_unlock_irqrestore(&tls_device_lock, flags);
1372
1373                 /* Device contexts for RX and TX will be freed in on sk_destruct
1374                  * by tls_device_free_ctx. rx_conf and tx_conf stay in TLS_HW.
1375                  * Now release the ref taken above.
1376                  */
1377                 if (refcount_dec_and_test(&ctx->refcount))
1378                         tls_device_free_ctx(ctx);
1379         }
1380
1381         up_write(&device_offload_lock);
1382
1383         flush_work(&tls_device_gc_work);
1384
1385         return NOTIFY_DONE;
1386 }
1387
1388 static int tls_dev_event(struct notifier_block *this, unsigned long event,
1389                          void *ptr)
1390 {
1391         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1392
1393         if (!dev->tlsdev_ops &&
1394             !(dev->features & (NETIF_F_HW_TLS_RX | NETIF_F_HW_TLS_TX)))
1395                 return NOTIFY_DONE;
1396
1397         switch (event) {
1398         case NETDEV_REGISTER:
1399         case NETDEV_FEAT_CHANGE:
1400                 if (netif_is_bond_master(dev))
1401                         return NOTIFY_DONE;
1402                 if ((dev->features & NETIF_F_HW_TLS_RX) &&
1403                     !dev->tlsdev_ops->tls_dev_resync)
1404                         return NOTIFY_BAD;
1405
1406                 if  (dev->tlsdev_ops &&
1407                      dev->tlsdev_ops->tls_dev_add &&
1408                      dev->tlsdev_ops->tls_dev_del)
1409                         return NOTIFY_DONE;
1410                 else
1411                         return NOTIFY_BAD;
1412         case NETDEV_DOWN:
1413                 return tls_device_down(dev);
1414         }
1415         return NOTIFY_DONE;
1416 }
1417
1418 static struct notifier_block tls_dev_notifier = {
1419         .notifier_call  = tls_dev_event,
1420 };
1421
1422 void __init tls_device_init(void)
1423 {
1424         register_netdevice_notifier(&tls_dev_notifier);
1425 }
1426
1427 void __exit tls_device_cleanup(void)
1428 {
1429         unregister_netdevice_notifier(&tls_dev_notifier);
1430         flush_work(&tls_device_gc_work);
1431         clean_acked_data_flush();
1432 }