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 /* device_offload_lock is used to synchronize tls_dev_add
  42  * against NETDEV_DOWN notifications.
  43  */
  44 static DECLARE_RWSEM(device_offload_lock);
  45
  46 static void tls_device_gc_task(struct work_struct *work);
  47
  48 static DECLARE_WORK(tls_device_gc_work, tls_device_gc_task);
  49 static LIST_HEAD(tls_device_gc_list);
  50 static LIST_HEAD(tls_device_list);
  51 static DEFINE_SPINLOCK(tls_device_lock);
  52
  53 static void tls_device_free_ctx(struct tls_context *ctx)
  54 {
  55         if (ctx->tx_conf == TLS_HW) {
  56                 kfree(tls_offload_ctx_tx(ctx));
  57                 kfree(ctx->tx.rec_seq);
  58                 kfree(ctx->tx.iv);
  59         }
  60
  61         if (ctx->rx_conf == TLS_HW)
  62                 kfree(tls_offload_ctx_rx(ctx));
  63
  64         tls_ctx_free(ctx);
  65 }
  66
  67 static void tls_device_gc_task(struct work_struct *work)
  68 {
  69         struct tls_context *ctx, *tmp;
  70         unsigned long flags;
  71         LIST_HEAD(gc_list);
  72
  73         spin_lock_irqsave(&tls_device_lock, flags);
  74         list_splice_init(&tls_device_gc_list, &gc_list);
  75         spin_unlock_irqrestore(&tls_device_lock, flags);
  76
  77         list_for_each_entry_safe(ctx, tmp, &gc_list, list) {
  78                 struct net_device *netdev = ctx->netdev;
  79
  80                 if (netdev && ctx->tx_conf == TLS_HW) {
  81                         netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
  82                                                         TLS_OFFLOAD_CTX_DIR_TX);
  83                         dev_put(netdev);
  84                         ctx->netdev = NULL;
  85                 }
  86
  87                 list_del(&ctx->list);
  88                 tls_device_free_ctx(ctx);
  89         }
  90 }
  91
  92 static void tls_device_queue_ctx_destruction(struct tls_context *ctx)
  93 {
  94         unsigned long flags;
  95
  96         spin_lock_irqsave(&tls_device_lock, flags);
  97         list_move_tail(&ctx->list, &tls_device_gc_list);
  98
  99         /* schedule_work inside the spinlock
 100          * to make sure tls_device_down waits for that work.
 101          */
 102         schedule_work(&tls_device_gc_work);
 103
 104         spin_unlock_irqrestore(&tls_device_lock, flags);
 105 }
 106
 107 /* We assume that the socket is already connected */
 108 static struct net_device *get_netdev_for_sock(struct sock *sk)
 109 {
 110         struct dst_entry *dst = sk_dst_get(sk);
 111         struct net_device *netdev = NULL;
 112
 113         if (likely(dst)) {
 114                 netdev = dst->dev;
 115                 dev_hold(netdev);
 116         }
 117
 118         dst_release(dst);
 119
 120         return netdev;
 121 }
 122
 123 static void destroy_record(struct tls_record_info *record)
 124 {
 125         int nr_frags = record->num_frags;
 126         skb_frag_t *frag;
 127
 128         while (nr_frags-- > 0) {
 129                 frag = &record->frags[nr_frags];
 130                 __skb_frag_unref(frag);
 131         }
 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                 list_del(&info->list);
 165                 destroy_record(info);
 166                 deleted_records++;
 167         }
 168
 169         list_for_each_entry_safe(info, temp, &ctx->records_list, list) {
 170                 if (before(acked_seq, info->end_seq))
 171                         break;
 172                 list_del(&info->list);
 173
 174                 destroy_record(info);
 175                 deleted_records++;
 176         }
 177
 178         ctx->unacked_record_sn += deleted_records;
 179         spin_unlock_irqrestore(&ctx->lock, flags);
 180 }
 181
 182 /* At this point, there should be no references on this
 183  * socket and no in-flight SKBs associated with this
 184  * socket, so it is safe to free all the resources.
 185  */
 186 static void tls_device_sk_destruct(struct sock *sk)
 187 {
 188         struct tls_context *tls_ctx = tls_get_ctx(sk);
 189         struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
 190
 191         tls_ctx->sk_destruct(sk);
 192
 193         if (tls_ctx->tx_conf == TLS_HW) {
 194                 if (ctx->open_record)
 195                         destroy_record(ctx->open_record);
 196                 delete_all_records(ctx);
 197                 crypto_free_aead(ctx->aead_send);
 198                 clean_acked_data_disable(inet_csk(sk));
 199         }
 200
 201         if (refcount_dec_and_test(&tls_ctx->refcount))
 202                 tls_device_queue_ctx_destruction(tls_ctx);
 203 }
 204
 205 void tls_device_free_resources_tx(struct sock *sk)
 206 {
 207         struct tls_context *tls_ctx = tls_get_ctx(sk);
 208
 209         tls_free_partial_record(sk, tls_ctx);
 210 }
 211
 212 static void tls_device_resync_tx(struct sock *sk, struct tls_context *tls_ctx,
 213                                  u32 seq)
 214 {
 215         struct net_device *netdev;
 216         struct sk_buff *skb;
 217         u8 *rcd_sn;
 218
 219         skb = tcp_write_queue_tail(sk);
 220         if (skb)
 221                 TCP_SKB_CB(skb)->eor = 1;
 222
 223         rcd_sn = tls_ctx->tx.rec_seq;
 224
 225         down_read(&device_offload_lock);
 226         netdev = tls_ctx->netdev;
 227         if (netdev)
 228                 netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq, rcd_sn,
 229                                                    TLS_OFFLOAD_CTX_DIR_TX);
 230         up_read(&device_offload_lock);
 231
 232         clear_bit_unlock(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
 233 }
 234
 235 static void tls_append_frag(struct tls_record_info *record,
 236                             struct page_frag *pfrag,
 237                             int size)
 238 {
 239         skb_frag_t *frag;
 240
 241         frag = &record->frags[record->num_frags - 1];
 242         if (frag->page.p == pfrag->page &&
 243             frag->page_offset + frag->size == pfrag->offset) {
 244                 frag->size += size;
 245         } else {
 246                 ++frag;
 247                 frag->page.p = pfrag->page;
 248                 frag->page_offset = pfrag->offset;
 249                 frag->size = size;
 250                 ++record->num_frags;
 251                 get_page(pfrag->page);
 252         }
 253
 254         pfrag->offset += size;
 255         record->len += size;
 256 }
 257
 258 static int tls_push_record(struct sock *sk,
 259                            struct tls_context *ctx,
 260                            struct tls_offload_context_tx *offload_ctx,
 261                            struct tls_record_info *record,
 262                            struct page_frag *pfrag,
 263                            int flags,
 264                            unsigned char record_type)
 265 {
 266         struct tls_prot_info *prot = &ctx->prot_info;
 267         struct tcp_sock *tp = tcp_sk(sk);
 268         struct page_frag dummy_tag_frag;
 269         skb_frag_t *frag;
 270         int i;
 271
 272         /* fill prepend */
 273         frag = &record->frags[0];
 274         tls_fill_prepend(ctx,
 275                          skb_frag_address(frag),
 276                          record->len - prot->prepend_size,
 277                          record_type,
 278                          prot->version);
 279
 280         /* HW doesn't care about the data in the tag, because it fills it. */
 281         dummy_tag_frag.page = skb_frag_page(frag);
 282         dummy_tag_frag.offset = 0;
 283
 284         tls_append_frag(record, &dummy_tag_frag, prot->tag_size);
 285         record->end_seq = tp->write_seq + record->len;
 286         spin_lock_irq(&offload_ctx->lock);
 287         list_add_tail(&record->list, &offload_ctx->records_list);
 288         spin_unlock_irq(&offload_ctx->lock);
 289         offload_ctx->open_record = NULL;
 290
 291         if (test_bit(TLS_TX_SYNC_SCHED, &ctx->flags))
 292                 tls_device_resync_tx(sk, ctx, tp->write_seq);
 293
 294         tls_advance_record_sn(sk, prot, &ctx->tx);
 295
 296         for (i = 0; i < record->num_frags; i++) {
 297                 frag = &record->frags[i];
 298                 sg_unmark_end(&offload_ctx->sg_tx_data[i]);
 299                 sg_set_page(&offload_ctx->sg_tx_data[i], skb_frag_page(frag),
 300                             frag->size, frag->page_offset);
 301                 sk_mem_charge(sk, frag->size);
 302                 get_page(skb_frag_page(frag));
 303         }
 304         sg_mark_end(&offload_ctx->sg_tx_data[record->num_frags - 1]);
 305
 306         /* all ready, send */
 307         return tls_push_sg(sk, ctx, offload_ctx->sg_tx_data, 0, flags);
 308 }
 309
 310 static int tls_create_new_record(struct tls_offload_context_tx *offload_ctx,
 311                                  struct page_frag *pfrag,
 312                                  size_t prepend_size)
 313 {
 314         struct tls_record_info *record;
 315         skb_frag_t *frag;
 316
 317         record = kmalloc(sizeof(*record), GFP_KERNEL);
 318         if (!record)
 319                 return -ENOMEM;
 320
 321         frag = &record->frags[0];
 322         __skb_frag_set_page(frag, pfrag->page);
 323         frag->page_offset = pfrag->offset;
 324         skb_frag_size_set(frag, prepend_size);
 325
 326         get_page(pfrag->page);
 327         pfrag->offset += prepend_size;
 328
 329         record->num_frags = 1;
 330         record->len = prepend_size;
 331         offload_ctx->open_record = record;
 332         return 0;
 333 }
 334
 335 static int tls_do_allocation(struct sock *sk,
 336                              struct tls_offload_context_tx *offload_ctx,
 337                              struct page_frag *pfrag,
 338                              size_t prepend_size)
 339 {
 340         int ret;
 341
 342         if (!offload_ctx->open_record) {
 343                 if (unlikely(!skb_page_frag_refill(prepend_size, pfrag,
 344                                                    sk->sk_allocation))) {
 345                         sk->sk_prot->enter_memory_pressure(sk);
 346                         sk_stream_moderate_sndbuf(sk);
 347                         return -ENOMEM;
 348                 }
 349
 350                 ret = tls_create_new_record(offload_ctx, pfrag, prepend_size);
 351                 if (ret)
 352                         return ret;
 353
 354                 if (pfrag->size > pfrag->offset)
 355                         return 0;
 356         }
 357
 358         if (!sk_page_frag_refill(sk, pfrag))
 359                 return -ENOMEM;
 360
 361         return 0;
 362 }
 363
 364 static int tls_push_data(struct sock *sk,
 365                          struct iov_iter *msg_iter,
 366                          size_t size, int flags,
 367                          unsigned char record_type)
 368 {
 369         struct tls_context *tls_ctx = tls_get_ctx(sk);
 370         struct tls_prot_info *prot = &tls_ctx->prot_info;
 371         struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
 372         int tls_push_record_flags = flags | MSG_SENDPAGE_NOTLAST;
 373         int more = flags & (MSG_SENDPAGE_NOTLAST | MSG_MORE);
 374         struct tls_record_info *record = ctx->open_record;
 375         struct page_frag *pfrag;
 376         size_t orig_size = size;
 377         u32 max_open_record_len;
 378         int copy, rc = 0;
 379         bool done = false;
 380         long timeo;
 381
 382         if (flags &
 383             ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_SENDPAGE_NOTLAST))
 384                 return -ENOTSUPP;
 385
 386         if (sk->sk_err)
 387                 return -sk->sk_err;
 388
 389         timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
 390         if (tls_is_partially_sent_record(tls_ctx)) {
 391                 rc = tls_push_partial_record(sk, tls_ctx, flags);
 392                 if (rc < 0)
 393                         return rc;
 394         }
 395
 396         pfrag = sk_page_frag(sk);
 397
 398         /* TLS_HEADER_SIZE is not counted as part of the TLS record, and
 399          * we need to leave room for an authentication tag.
 400          */
 401         max_open_record_len = TLS_MAX_PAYLOAD_SIZE +
 402                               prot->prepend_size;
 403         do {
 404                 rc = tls_do_allocation(sk, ctx, pfrag,
 405                                        prot->prepend_size);
 406                 if (rc) {
 407                         rc = sk_stream_wait_memory(sk, &timeo);
 408                         if (!rc)
 409                                 continue;
 410
 411                         record = ctx->open_record;
 412                         if (!record)
 413                                 break;
 414 handle_error:
 415                         if (record_type != TLS_RECORD_TYPE_DATA) {
 416                                 /* avoid sending partial
 417                                  * record with type !=
 418                                  * application_data
 419                                  */
 420                                 size = orig_size;
 421                                 destroy_record(record);
 422                                 ctx->open_record = NULL;
 423                         } else if (record->len > prot->prepend_size) {
 424                                 goto last_record;
 425                         }
 426
 427                         break;
 428                 }
 429
 430                 record = ctx->open_record;
 431                 copy = min_t(size_t, size, (pfrag->size - pfrag->offset));
 432                 copy = min_t(size_t, copy, (max_open_record_len - record->len));
 433
 434                 if (copy_from_iter_nocache(page_address(pfrag->page) +
 435                                                pfrag->offset,
 436                                            copy, msg_iter) != copy) {
 437                         rc = -EFAULT;
 438                         goto handle_error;
 439                 }
 440                 tls_append_frag(record, pfrag, copy);
 441
 442                 size -= copy;
 443                 if (!size) {
 444 last_record:
 445                         tls_push_record_flags = flags;
 446                         if (more) {
 447                                 tls_ctx->pending_open_record_frags =
 448                                                 !!record->num_frags;
 449                                 break;
 450                         }
 451
 452                         done = true;
 453                 }
 454
 455                 if (done || record->len >= max_open_record_len ||
 456                     (record->num_frags >= MAX_SKB_FRAGS - 1)) {
 457                         rc = tls_push_record(sk,
 458                                              tls_ctx,
 459                                              ctx,
 460                                              record,
 461                                              pfrag,
 462                                              tls_push_record_flags,
 463                                              record_type);
 464                         if (rc < 0)
 465                                 break;
 466                 }
 467         } while (!done);
 468
 469         if (orig_size - size > 0)
 470                 rc = orig_size - size;
 471
 472         return rc;
 473 }
 474
 475 int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
 476 {
 477         unsigned char record_type = TLS_RECORD_TYPE_DATA;
 478         int rc;
 479
 480         lock_sock(sk);
 481
 482         if (unlikely(msg->msg_controllen)) {
 483                 rc = tls_proccess_cmsg(sk, msg, &record_type);
 484                 if (rc)
 485                         goto out;
 486         }
 487
 488         rc = tls_push_data(sk, &msg->msg_iter, size,
 489                            msg->msg_flags, record_type);
 490
 491 out:
 492         release_sock(sk);
 493         return rc;
 494 }
 495
 496 int tls_device_sendpage(struct sock *sk, struct page *page,
 497                         int offset, size_t size, int flags)
 498 {
 499         struct iov_iter msg_iter;
 500         char *kaddr = kmap(page);
 501         struct kvec iov;
 502         int rc;
 503
 504         if (flags & MSG_SENDPAGE_NOTLAST)
 505                 flags |= MSG_MORE;
 506
 507         lock_sock(sk);
 508
 509         if (flags & MSG_OOB) {
 510                 rc = -ENOTSUPP;
 511                 goto out;
 512         }
 513
 514         iov.iov_base = kaddr + offset;
 515         iov.iov_len = size;
 516         iov_iter_kvec(&msg_iter, WRITE, &iov, 1, size);
 517         rc = tls_push_data(sk, &msg_iter, size,
 518                            flags, TLS_RECORD_TYPE_DATA);
 519         kunmap(page);
 520
 521 out:
 522         release_sock(sk);
 523         return rc;
 524 }
 525
 526 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
 527                                        u32 seq, u64 *p_record_sn)
 528 {
 529         u64 record_sn = context->hint_record_sn;
 530         struct tls_record_info *info;
 531
 532         info = context->retransmit_hint;
 533         if (!info ||
 534             before(seq, info->end_seq - info->len)) {
 535                 /* if retransmit_hint is irrelevant start
 536                  * from the beggining of the list
 537                  */
 538                 info = list_first_entry(&context->records_list,
 539                                         struct tls_record_info, list);
 540                 record_sn = context->unacked_record_sn;
 541         }
 542
 543         list_for_each_entry_from(info, &context->records_list, list) {
 544                 if (before(seq, info->end_seq)) {
 545                         if (!context->retransmit_hint ||
 546                             after(info->end_seq,
 547                                   context->retransmit_hint->end_seq)) {
 548                                 context->hint_record_sn = record_sn;
 549                                 context->retransmit_hint = info;
 550                         }
 551                         *p_record_sn = record_sn;
 552                         return info;
 553                 }
 554                 record_sn++;
 555         }
 556
 557         return NULL;
 558 }
 559 EXPORT_SYMBOL(tls_get_record);
 560
 561 static int tls_device_push_pending_record(struct sock *sk, int flags)
 562 {
 563         struct iov_iter msg_iter;
 564
 565         iov_iter_kvec(&msg_iter, WRITE, NULL, 0, 0);
 566         return tls_push_data(sk, &msg_iter, 0, flags, TLS_RECORD_TYPE_DATA);
 567 }
 568
 569 void tls_device_write_space(struct sock *sk, struct tls_context *ctx)
 570 {
 571         if (!sk->sk_write_pending && tls_is_partially_sent_record(ctx)) {
 572                 gfp_t sk_allocation = sk->sk_allocation;
 573
 574                 sk->sk_allocation = GFP_ATOMIC;
 575                 tls_push_partial_record(sk, ctx, MSG_DONTWAIT | MSG_NOSIGNAL);
 576                 sk->sk_allocation = sk_allocation;
 577         }
 578 }
 579
 580 static void tls_device_resync_rx(struct tls_context *tls_ctx,
 581                                  struct sock *sk, u32 seq, u8 *rcd_sn)
 582 {
 583         struct net_device *netdev;
 584
 585         if (WARN_ON(test_and_set_bit(TLS_RX_SYNC_RUNNING, &tls_ctx->flags)))
 586                 return;
 587         netdev = READ_ONCE(tls_ctx->netdev);
 588         if (netdev)
 589                 netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq, rcd_sn,
 590                                                    TLS_OFFLOAD_CTX_DIR_RX);
 591         clear_bit_unlock(TLS_RX_SYNC_RUNNING, &tls_ctx->flags);
 592 }
 593
 594 void tls_device_rx_resync_new_rec(struct sock *sk, u32 rcd_len, u32 seq)
 595 {
 596         struct tls_context *tls_ctx = tls_get_ctx(sk);
 597         struct tls_offload_context_rx *rx_ctx;
 598         u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
 599         struct tls_prot_info *prot;
 600         u32 is_req_pending;
 601         s64 resync_req;
 602         u32 req_seq;
 603
 604         if (tls_ctx->rx_conf != TLS_HW)
 605                 return;
 606
 607         prot = &tls_ctx->prot_info;
 608         rx_ctx = tls_offload_ctx_rx(tls_ctx);
 609         memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
 610
 611         switch (rx_ctx->resync_type) {
 612         case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ:
 613                 resync_req = atomic64_read(&rx_ctx->resync_req);
 614                 req_seq = resync_req >> 32;
 615                 seq += TLS_HEADER_SIZE - 1;
 616                 is_req_pending = resync_req;
 617
 618                 if (likely(!is_req_pending) || req_seq != seq ||
 619                     !atomic64_try_cmpxchg(&rx_ctx->resync_req, &resync_req, 0))
 620                         return;
 621                 break;
 622         case TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT:
 623                 if (likely(!rx_ctx->resync_nh_do_now))
 624                         return;
 625
 626                 /* head of next rec is already in, note that the sock_inq will
 627                  * include the currently parsed message when called from parser
 628                  */
 629                 if (tcp_inq(sk) > rcd_len)
 630                         return;
 631
 632                 rx_ctx->resync_nh_do_now = 0;
 633                 seq += rcd_len;
 634                 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
 635                 break;
 636         }
 637
 638         tls_device_resync_rx(tls_ctx, sk, seq, rcd_sn);
 639 }
 640
 641 static void tls_device_core_ctrl_rx_resync(struct tls_context *tls_ctx,
 642                                            struct tls_offload_context_rx *ctx,
 643                                            struct sock *sk, struct sk_buff *skb)
 644 {
 645         struct strp_msg *rxm;
 646
 647         /* device will request resyncs by itself based on stream scan */
 648         if (ctx->resync_type != TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT)
 649                 return;
 650         /* already scheduled */
 651         if (ctx->resync_nh_do_now)
 652                 return;
 653         /* seen decrypted fragments since last fully-failed record */
 654         if (ctx->resync_nh_reset) {
 655                 ctx->resync_nh_reset = 0;
 656                 ctx->resync_nh.decrypted_failed = 1;
 657                 ctx->resync_nh.decrypted_tgt = TLS_DEVICE_RESYNC_NH_START_IVAL;
 658                 return;
 659         }
 660
 661         if (++ctx->resync_nh.decrypted_failed <= ctx->resync_nh.decrypted_tgt)
 662                 return;
 663
 664         /* doing resync, bump the next target in case it fails */
 665         if (ctx->resync_nh.decrypted_tgt < TLS_DEVICE_RESYNC_NH_MAX_IVAL)
 666                 ctx->resync_nh.decrypted_tgt *= 2;
 667         else
 668                 ctx->resync_nh.decrypted_tgt += TLS_DEVICE_RESYNC_NH_MAX_IVAL;
 669
 670         rxm = strp_msg(skb);
 671
 672         /* head of next rec is already in, parser will sync for us */
 673         if (tcp_inq(sk) > rxm->full_len) {
 674                 ctx->resync_nh_do_now = 1;
 675         } else {
 676                 struct tls_prot_info *prot = &tls_ctx->prot_info;
 677                 u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
 678
 679                 memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
 680                 tls_bigint_increment(rcd_sn, prot->rec_seq_size);
 681
 682                 tls_device_resync_rx(tls_ctx, sk, tcp_sk(sk)->copied_seq,
 683                                      rcd_sn);
 684         }
 685 }
 686
 687 static int tls_device_reencrypt(struct sock *sk, struct sk_buff *skb)
 688 {
 689         struct strp_msg *rxm = strp_msg(skb);
 690         int err = 0, offset = rxm->offset, copy, nsg, data_len, pos;
 691         struct sk_buff *skb_iter, *unused;
 692         struct scatterlist sg[1];
 693         char *orig_buf, *buf;
 694
 695         orig_buf = kmalloc(rxm->full_len + TLS_HEADER_SIZE +
 696                            TLS_CIPHER_AES_GCM_128_IV_SIZE, sk->sk_allocation);
 697         if (!orig_buf)
 698                 return -ENOMEM;
 699         buf = orig_buf;
 700
 701         nsg = skb_cow_data(skb, 0, &unused);
 702         if (unlikely(nsg < 0)) {
 703                 err = nsg;
 704                 goto free_buf;
 705         }
 706
 707         sg_init_table(sg, 1);
 708         sg_set_buf(&sg[0], buf,
 709                    rxm->full_len + TLS_HEADER_SIZE +
 710                    TLS_CIPHER_AES_GCM_128_IV_SIZE);
 711         err = skb_copy_bits(skb, offset, buf,
 712                             TLS_HEADER_SIZE + TLS_CIPHER_AES_GCM_128_IV_SIZE);
 713         if (err)
 714                 goto free_buf;
 715
 716         /* We are interested only in the decrypted data not the auth */
 717         err = decrypt_skb(sk, skb, sg);
 718         if (err != -EBADMSG)
 719                 goto free_buf;
 720         else
 721                 err = 0;
 722
 723         data_len = rxm->full_len - TLS_CIPHER_AES_GCM_128_TAG_SIZE;
 724
 725         if (skb_pagelen(skb) > offset) {
 726                 copy = min_t(int, skb_pagelen(skb) - offset, data_len);
 727
 728                 if (skb->decrypted) {
 729                         err = skb_store_bits(skb, offset, buf, copy);
 730                         if (err)
 731                                 goto free_buf;
 732                 }
 733
 734                 offset += copy;
 735                 buf += copy;
 736         }
 737
 738         pos = skb_pagelen(skb);
 739         skb_walk_frags(skb, skb_iter) {
 740                 int frag_pos;
 741
 742                 /* Practically all frags must belong to msg if reencrypt
 743                  * is needed with current strparser and coalescing logic,
 744                  * but strparser may "get optimized", so let's be safe.
 745                  */
 746                 if (pos + skb_iter->len <= offset)
 747                         goto done_with_frag;
 748                 if (pos >= data_len + rxm->offset)
 749                         break;
 750
 751                 frag_pos = offset - pos;
 752                 copy = min_t(int, skb_iter->len - frag_pos,
 753                              data_len + rxm->offset - offset);
 754
 755                 if (skb_iter->decrypted) {
 756                         err = skb_store_bits(skb_iter, frag_pos, buf, copy);
 757                         if (err)
 758                                 goto free_buf;
 759                 }
 760
 761                 offset += copy;
 762                 buf += copy;
 763 done_with_frag:
 764                 pos += skb_iter->len;
 765         }
 766
 767 free_buf:
 768         kfree(orig_buf);
 769         return err;
 770 }
 771
 772 int tls_device_decrypted(struct sock *sk, struct sk_buff *skb)
 773 {
 774         struct tls_context *tls_ctx = tls_get_ctx(sk);
 775         struct tls_offload_context_rx *ctx = tls_offload_ctx_rx(tls_ctx);
 776         int is_decrypted = skb->decrypted;
 777         int is_encrypted = !is_decrypted;
 778         struct sk_buff *skb_iter;
 779
 780         /* Check if all the data is decrypted already */
 781         skb_walk_frags(skb, skb_iter) {
 782                 is_decrypted &= skb_iter->decrypted;
 783                 is_encrypted &= !skb_iter->decrypted;
 784         }
 785
 786         ctx->sw.decrypted |= is_decrypted;
 787
 788         /* Return immediately if the record is either entirely plaintext or
 789          * entirely ciphertext. Otherwise handle reencrypt partially decrypted
 790          * record.
 791          */
 792         if (is_decrypted) {
 793                 ctx->resync_nh_reset = 1;
 794                 return 0;
 795         }
 796         if (is_encrypted) {
 797                 tls_device_core_ctrl_rx_resync(tls_ctx, ctx, sk, skb);
 798                 return 0;
 799         }
 800
 801         ctx->resync_nh_reset = 1;
 802         return tls_device_reencrypt(sk, skb);
 803 }
 804
 805 static void tls_device_attach(struct tls_context *ctx, struct sock *sk,
 806                               struct net_device *netdev)
 807 {
 808         if (sk->sk_destruct != tls_device_sk_destruct) {
 809                 refcount_set(&ctx->refcount, 1);
 810                 dev_hold(netdev);
 811                 ctx->netdev = netdev;
 812                 spin_lock_irq(&tls_device_lock);
 813                 list_add_tail(&ctx->list, &tls_device_list);
 814                 spin_unlock_irq(&tls_device_lock);
 815
 816                 ctx->sk_destruct = sk->sk_destruct;
 817                 sk->sk_destruct = tls_device_sk_destruct;
 818         }
 819 }
 820
 821 int tls_set_device_offload(struct sock *sk, struct tls_context *ctx)
 822 {
 823         u16 nonce_size, tag_size, iv_size, rec_seq_size;
 824         struct tls_context *tls_ctx = tls_get_ctx(sk);
 825         struct tls_prot_info *prot = &tls_ctx->prot_info;
 826         struct tls_record_info *start_marker_record;
 827         struct tls_offload_context_tx *offload_ctx;
 828         struct tls_crypto_info *crypto_info;
 829         struct net_device *netdev;
 830         char *iv, *rec_seq;
 831         struct sk_buff *skb;
 832         int rc = -EINVAL;
 833         __be64 rcd_sn;
 834
 835         if (!ctx)
 836                 goto out;
 837
 838         if (ctx->priv_ctx_tx) {
 839                 rc = -EEXIST;
 840                 goto out;
 841         }
 842
 843         start_marker_record = kmalloc(sizeof(*start_marker_record), GFP_KERNEL);
 844         if (!start_marker_record) {
 845                 rc = -ENOMEM;
 846                 goto out;
 847         }
 848
 849         offload_ctx = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_TX, GFP_KERNEL);
 850         if (!offload_ctx) {
 851                 rc = -ENOMEM;
 852                 goto free_marker_record;
 853         }
 854
 855         crypto_info = &ctx->crypto_send.info;
 856         if (crypto_info->version != TLS_1_2_VERSION) {
 857                 rc = -EOPNOTSUPP;
 858                 goto free_offload_ctx;
 859         }
 860
 861         switch (crypto_info->cipher_type) {
 862         case TLS_CIPHER_AES_GCM_128:
 863                 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
 864                 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
 865                 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
 866                 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
 867                 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
 868                 rec_seq =
 869                  ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
 870                 break;
 871         default:
 872                 rc = -EINVAL;
 873                 goto free_offload_ctx;
 874         }
 875
 876         /* Sanity-check the rec_seq_size for stack allocations */
 877         if (rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
 878                 rc = -EINVAL;
 879                 goto free_offload_ctx;
 880         }
 881
 882         prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
 883         prot->tag_size = tag_size;
 884         prot->overhead_size = prot->prepend_size + prot->tag_size;
 885         prot->iv_size = iv_size;
 886         ctx->tx.iv = kmalloc(iv_size + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
 887                              GFP_KERNEL);
 888         if (!ctx->tx.iv) {
 889                 rc = -ENOMEM;
 890                 goto free_offload_ctx;
 891         }
 892
 893         memcpy(ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv, iv_size);
 894
 895         prot->rec_seq_size = rec_seq_size;
 896         ctx->tx.rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
 897         if (!ctx->tx.rec_seq) {
 898                 rc = -ENOMEM;
 899                 goto free_iv;
 900         }
 901
 902         rc = tls_sw_fallback_init(sk, offload_ctx, crypto_info);
 903         if (rc)
 904                 goto free_rec_seq;
 905
 906         /* start at rec_seq - 1 to account for the start marker record */
 907         memcpy(&rcd_sn, ctx->tx.rec_seq, sizeof(rcd_sn));
 908         offload_ctx->unacked_record_sn = be64_to_cpu(rcd_sn) - 1;
 909
 910         start_marker_record->end_seq = tcp_sk(sk)->write_seq;
 911         start_marker_record->len = 0;
 912         start_marker_record->num_frags = 0;
 913
 914         INIT_LIST_HEAD(&offload_ctx->records_list);
 915         list_add_tail(&start_marker_record->list, &offload_ctx->records_list);
 916         spin_lock_init(&offload_ctx->lock);
 917         sg_init_table(offload_ctx->sg_tx_data,
 918                       ARRAY_SIZE(offload_ctx->sg_tx_data));
 919
 920         clean_acked_data_enable(inet_csk(sk), &tls_icsk_clean_acked);
 921         ctx->push_pending_record = tls_device_push_pending_record;
 922
 923         /* TLS offload is greatly simplified if we don't send
 924          * SKBs where only part of the payload needs to be encrypted.
 925          * So mark the last skb in the write queue as end of record.
 926          */
 927         skb = tcp_write_queue_tail(sk);
 928         if (skb)
 929                 TCP_SKB_CB(skb)->eor = 1;
 930
 931         /* We support starting offload on multiple sockets
 932          * concurrently, so we only need a read lock here.
 933          * This lock must precede get_netdev_for_sock to prevent races between
 934          * NETDEV_DOWN and setsockopt.
 935          */
 936         down_read(&device_offload_lock);
 937         netdev = get_netdev_for_sock(sk);
 938         if (!netdev) {
 939                 pr_err_ratelimited("%s: netdev not found\n", __func__);
 940                 rc = -EINVAL;
 941                 goto release_lock;
 942         }
 943
 944         if (!(netdev->features & NETIF_F_HW_TLS_TX)) {
 945                 rc = -ENOTSUPP;
 946                 goto release_netdev;
 947         }
 948
 949         /* Avoid offloading if the device is down
 950          * We don't want to offload new flows after
 951          * the NETDEV_DOWN event
 952          */
 953         if (!(netdev->flags & IFF_UP)) {
 954                 rc = -EINVAL;
 955                 goto release_netdev;
 956         }
 957
 958         ctx->priv_ctx_tx = offload_ctx;
 959         rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_TX,
 960                                              &ctx->crypto_send.info,
 961                                              tcp_sk(sk)->write_seq);
 962         if (rc)
 963                 goto release_netdev;
 964
 965         tls_device_attach(ctx, sk, netdev);
 966
 967         /* following this assignment tls_is_sk_tx_device_offloaded
 968          * will return true and the context might be accessed
 969          * by the netdev's xmit function.
 970          */
 971         smp_store_release(&sk->sk_validate_xmit_skb, tls_validate_xmit_skb);
 972         dev_put(netdev);
 973         up_read(&device_offload_lock);
 974         goto out;
 975
 976 release_netdev:
 977         dev_put(netdev);
 978 release_lock:
 979         up_read(&device_offload_lock);
 980         clean_acked_data_disable(inet_csk(sk));
 981         crypto_free_aead(offload_ctx->aead_send);
 982 free_rec_seq:
 983         kfree(ctx->tx.rec_seq);
 984 free_iv:
 985         kfree(ctx->tx.iv);
 986 free_offload_ctx:
 987         kfree(offload_ctx);
 988         ctx->priv_ctx_tx = NULL;
 989 free_marker_record:
 990         kfree(start_marker_record);
 991 out:
 992         return rc;
 993 }
 994
 995 int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx)
 996 {
 997         struct tls_offload_context_rx *context;
 998         struct net_device *netdev;
 999         int rc = 0;
1000
1001         if (ctx->crypto_recv.info.version != TLS_1_2_VERSION)
1002                 return -EOPNOTSUPP;
1003
1004         /* We support starting offload on multiple sockets
1005          * concurrently, so we only need a read lock here.
1006          * This lock must precede get_netdev_for_sock to prevent races between
1007          * NETDEV_DOWN and setsockopt.
1008          */
1009         down_read(&device_offload_lock);
1010         netdev = get_netdev_for_sock(sk);
1011         if (!netdev) {
1012                 pr_err_ratelimited("%s: netdev not found\n", __func__);
1013                 rc = -EINVAL;
1014                 goto release_lock;
1015         }
1016
1017         if (!(netdev->features & NETIF_F_HW_TLS_RX)) {
1018                 rc = -ENOTSUPP;
1019                 goto release_netdev;
1020         }
1021
1022         /* Avoid offloading if the device is down
1023          * We don't want to offload new flows after
1024          * the NETDEV_DOWN event
1025          */
1026         if (!(netdev->flags & IFF_UP)) {
1027                 rc = -EINVAL;
1028                 goto release_netdev;
1029         }
1030
1031         context = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_RX, GFP_KERNEL);
1032         if (!context) {
1033                 rc = -ENOMEM;
1034                 goto release_netdev;
1035         }
1036         context->resync_nh_reset = 1;
1037
1038         ctx->priv_ctx_rx = context;
1039         rc = tls_set_sw_offload(sk, ctx, 0);
1040         if (rc)
1041                 goto release_ctx;
1042
1043         rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_RX,
1044                                              &ctx->crypto_recv.info,
1045                                              tcp_sk(sk)->copied_seq);
1046         if (rc)
1047                 goto free_sw_resources;
1048
1049         tls_device_attach(ctx, sk, netdev);
1050         goto release_netdev;
1051
1052 free_sw_resources:
1053         up_read(&device_offload_lock);
1054         tls_sw_free_resources_rx(sk);
1055         down_read(&device_offload_lock);
1056 release_ctx:
1057         ctx->priv_ctx_rx = NULL;
1058 release_netdev:
1059         dev_put(netdev);
1060 release_lock:
1061         up_read(&device_offload_lock);
1062         return rc;
1063 }
1064
1065 void tls_device_offload_cleanup_rx(struct sock *sk)
1066 {
1067         struct tls_context *tls_ctx = tls_get_ctx(sk);
1068         struct net_device *netdev;
1069
1070         down_read(&device_offload_lock);
1071         netdev = tls_ctx->netdev;
1072         if (!netdev)
1073                 goto out;
1074
1075         netdev->tlsdev_ops->tls_dev_del(netdev, tls_ctx,
1076                                         TLS_OFFLOAD_CTX_DIR_RX);
1077
1078         if (tls_ctx->tx_conf != TLS_HW) {
1079                 dev_put(netdev);
1080                 tls_ctx->netdev = NULL;
1081         }
1082 out:
1083         up_read(&device_offload_lock);
1084         tls_sw_release_resources_rx(sk);
1085 }
1086
1087 static int tls_device_down(struct net_device *netdev)
1088 {
1089         struct tls_context *ctx, *tmp;
1090         unsigned long flags;
1091         LIST_HEAD(list);
1092
1093         /* Request a write lock to block new offload attempts */
1094         down_write(&device_offload_lock);
1095
1096         spin_lock_irqsave(&tls_device_lock, flags);
1097         list_for_each_entry_safe(ctx, tmp, &tls_device_list, list) {
1098                 if (ctx->netdev != netdev ||
1099                     !refcount_inc_not_zero(&ctx->refcount))
1100                         continue;
1101
1102                 list_move(&ctx->list, &list);
1103         }
1104         spin_unlock_irqrestore(&tls_device_lock, flags);
1105
1106         list_for_each_entry_safe(ctx, tmp, &list, list) {
1107                 if (ctx->tx_conf == TLS_HW)
1108                         netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1109                                                         TLS_OFFLOAD_CTX_DIR_TX);
1110                 if (ctx->rx_conf == TLS_HW)
1111                         netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1112                                                         TLS_OFFLOAD_CTX_DIR_RX);
1113                 WRITE_ONCE(ctx->netdev, NULL);
1114                 smp_mb__before_atomic(); /* pairs with test_and_set_bit() */
1115                 while (test_bit(TLS_RX_SYNC_RUNNING, &ctx->flags))
1116                         usleep_range(10, 200);
1117                 dev_put(netdev);
1118                 list_del_init(&ctx->list);
1119
1120                 if (refcount_dec_and_test(&ctx->refcount))
1121                         tls_device_free_ctx(ctx);
1122         }
1123
1124         up_write(&device_offload_lock);
1125
1126         flush_work(&tls_device_gc_work);
1127
1128         return NOTIFY_DONE;
1129 }
1130
1131 static int tls_dev_event(struct notifier_block *this, unsigned long event,
1132                          void *ptr)
1133 {
1134         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1135
1136         if (!dev->tlsdev_ops &&
1137             !(dev->features & (NETIF_F_HW_TLS_RX | NETIF_F_HW_TLS_TX)))
1138                 return NOTIFY_DONE;
1139
1140         switch (event) {
1141         case NETDEV_REGISTER:
1142         case NETDEV_FEAT_CHANGE:
1143                 if ((dev->features & NETIF_F_HW_TLS_RX) &&
1144                     !dev->tlsdev_ops->tls_dev_resync)
1145                         return NOTIFY_BAD;
1146
1147                 if  (dev->tlsdev_ops &&
1148                      dev->tlsdev_ops->tls_dev_add &&
1149                      dev->tlsdev_ops->tls_dev_del)
1150                         return NOTIFY_DONE;
1151                 else
1152                         return NOTIFY_BAD;
1153         case NETDEV_DOWN:
1154                 return tls_device_down(dev);
1155         }
1156         return NOTIFY_DONE;
1157 }
1158
1159 static struct notifier_block tls_dev_notifier = {
1160         .notifier_call  = tls_dev_event,
1161 };
1162
1163 void __init tls_device_init(void)
1164 {
1165         register_netdevice_notifier(&tls_dev_notifier);
1166 }
1167
1168 void __exit tls_device_cleanup(void)
1169 {
1170         unregister_netdevice_notifier(&tls_dev_notifier);
1171         flush_work(&tls_device_gc_work);
1172         clean_acked_data_flush();
1173 }