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_attach(struct tls_context *ctx, struct sock *sk,
  93                               struct net_device *netdev)
  94 {
  95         if (sk->sk_destruct != tls_device_sk_destruct) {
  96                 refcount_set(&ctx->refcount, 1);
  97                 dev_hold(netdev);
  98                 ctx->netdev = netdev;
  99                 spin_lock_irq(&tls_device_lock);
 100                 list_add_tail(&ctx->list, &tls_device_list);
 101                 spin_unlock_irq(&tls_device_lock);
 102
 103                 ctx->sk_destruct = sk->sk_destruct;
 104                 sk->sk_destruct = tls_device_sk_destruct;
 105         }
 106 }
 107
 108 static void tls_device_queue_ctx_destruction(struct tls_context *ctx)
 109 {
 110         unsigned long flags;
 111
 112         spin_lock_irqsave(&tls_device_lock, flags);
 113         list_move_tail(&ctx->list, &tls_device_gc_list);
 114
 115         /* schedule_work inside the spinlock
 116          * to make sure tls_device_down waits for that work.
 117          */
 118         schedule_work(&tls_device_gc_work);
 119
 120         spin_unlock_irqrestore(&tls_device_lock, flags);
 121 }
 122
 123 /* We assume that the socket is already connected */
 124 static struct net_device *get_netdev_for_sock(struct sock *sk)
 125 {
 126         struct dst_entry *dst = sk_dst_get(sk);
 127         struct net_device *netdev = NULL;
 128
 129         if (likely(dst)) {
 130                 netdev = dst->dev;
 131                 dev_hold(netdev);
 132         }
 133
 134         dst_release(dst);
 135
 136         return netdev;
 137 }
 138
 139 static void destroy_record(struct tls_record_info *record)
 140 {
 141         int nr_frags = record->num_frags;
 142         skb_frag_t *frag;
 143
 144         while (nr_frags-- > 0) {
 145                 frag = &record->frags[nr_frags];
 146                 __skb_frag_unref(frag);
 147         }
 148         kfree(record);
 149 }
 150
 151 static void delete_all_records(struct tls_offload_context_tx *offload_ctx)
 152 {
 153         struct tls_record_info *info, *temp;
 154
 155         list_for_each_entry_safe(info, temp, &offload_ctx->records_list, list) {
 156                 list_del(&info->list);
 157                 destroy_record(info);
 158         }
 159
 160         offload_ctx->retransmit_hint = NULL;
 161 }
 162
 163 static void tls_icsk_clean_acked(struct sock *sk, u32 acked_seq)
 164 {
 165         struct tls_context *tls_ctx = tls_get_ctx(sk);
 166         struct tls_record_info *info, *temp;
 167         struct tls_offload_context_tx *ctx;
 168         u64 deleted_records = 0;
 169         unsigned long flags;
 170
 171         if (!tls_ctx)
 172                 return;
 173
 174         ctx = tls_offload_ctx_tx(tls_ctx);
 175
 176         spin_lock_irqsave(&ctx->lock, flags);
 177         info = ctx->retransmit_hint;
 178         if (info && !before(acked_seq, info->end_seq)) {
 179                 ctx->retransmit_hint = NULL;
 180                 list_del(&info->list);
 181                 destroy_record(info);
 182                 deleted_records++;
 183         }
 184
 185         list_for_each_entry_safe(info, temp, &ctx->records_list, list) {
 186                 if (before(acked_seq, info->end_seq))
 187                         break;
 188                 list_del(&info->list);
 189
 190                 destroy_record(info);
 191                 deleted_records++;
 192         }
 193
 194         ctx->unacked_record_sn += deleted_records;
 195         spin_unlock_irqrestore(&ctx->lock, flags);
 196 }
 197
 198 /* At this point, there should be no references on this
 199  * socket and no in-flight SKBs associated with this
 200  * socket, so it is safe to free all the resources.
 201  */
 202 void tls_device_sk_destruct(struct sock *sk)
 203 {
 204         struct tls_context *tls_ctx = tls_get_ctx(sk);
 205         struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
 206
 207         tls_ctx->sk_destruct(sk);
 208
 209         if (tls_ctx->tx_conf == TLS_HW) {
 210                 if (ctx->open_record)
 211                         destroy_record(ctx->open_record);
 212                 delete_all_records(ctx);
 213                 crypto_free_aead(ctx->aead_send);
 214                 clean_acked_data_disable(inet_csk(sk));
 215         }
 216
 217         if (refcount_dec_and_test(&tls_ctx->refcount))
 218                 tls_device_queue_ctx_destruction(tls_ctx);
 219 }
 220 EXPORT_SYMBOL(tls_device_sk_destruct);
 221
 222 static void tls_append_frag(struct tls_record_info *record,
 223                             struct page_frag *pfrag,
 224                             int size)
 225 {
 226         skb_frag_t *frag;
 227
 228         frag = &record->frags[record->num_frags - 1];
 229         if (frag->page.p == pfrag->page &&
 230             frag->page_offset + frag->size == pfrag->offset) {
 231                 frag->size += size;
 232         } else {
 233                 ++frag;
 234                 frag->page.p = pfrag->page;
 235                 frag->page_offset = pfrag->offset;
 236                 frag->size = size;
 237                 ++record->num_frags;
 238                 get_page(pfrag->page);
 239         }
 240
 241         pfrag->offset += size;
 242         record->len += size;
 243 }
 244
 245 static int tls_push_record(struct sock *sk,
 246                            struct tls_context *ctx,
 247                            struct tls_offload_context_tx *offload_ctx,
 248                            struct tls_record_info *record,
 249                            struct page_frag *pfrag,
 250                            int flags,
 251                            unsigned char record_type)
 252 {
 253         struct tcp_sock *tp = tcp_sk(sk);
 254         struct page_frag dummy_tag_frag;
 255         skb_frag_t *frag;
 256         int i;
 257
 258         /* fill prepend */
 259         frag = &record->frags[0];
 260         tls_fill_prepend(ctx,
 261                          skb_frag_address(frag),
 262                          record->len - ctx->tx.prepend_size,
 263                          record_type);
 264
 265         /* HW doesn't care about the data in the tag, because it fills it. */
 266         dummy_tag_frag.page = skb_frag_page(frag);
 267         dummy_tag_frag.offset = 0;
 268
 269         tls_append_frag(record, &dummy_tag_frag, ctx->tx.tag_size);
 270         record->end_seq = tp->write_seq + record->len;
 271         spin_lock_irq(&offload_ctx->lock);
 272         list_add_tail(&record->list, &offload_ctx->records_list);
 273         spin_unlock_irq(&offload_ctx->lock);
 274         offload_ctx->open_record = NULL;
 275         set_bit(TLS_PENDING_CLOSED_RECORD, &ctx->flags);
 276         tls_advance_record_sn(sk, &ctx->tx);
 277
 278         for (i = 0; i < record->num_frags; i++) {
 279                 frag = &record->frags[i];
 280                 sg_unmark_end(&offload_ctx->sg_tx_data[i]);
 281                 sg_set_page(&offload_ctx->sg_tx_data[i], skb_frag_page(frag),
 282                             frag->size, frag->page_offset);
 283                 sk_mem_charge(sk, frag->size);
 284                 get_page(skb_frag_page(frag));
 285         }
 286         sg_mark_end(&offload_ctx->sg_tx_data[record->num_frags - 1]);
 287
 288         /* all ready, send */
 289         return tls_push_sg(sk, ctx, offload_ctx->sg_tx_data, 0, flags);
 290 }
 291
 292 static int tls_create_new_record(struct tls_offload_context_tx *offload_ctx,
 293                                  struct page_frag *pfrag,
 294                                  size_t prepend_size)
 295 {
 296         struct tls_record_info *record;
 297         skb_frag_t *frag;
 298
 299         record = kmalloc(sizeof(*record), GFP_KERNEL);
 300         if (!record)
 301                 return -ENOMEM;
 302
 303         frag = &record->frags[0];
 304         __skb_frag_set_page(frag, pfrag->page);
 305         frag->page_offset = pfrag->offset;
 306         skb_frag_size_set(frag, prepend_size);
 307
 308         get_page(pfrag->page);
 309         pfrag->offset += prepend_size;
 310
 311         record->num_frags = 1;
 312         record->len = prepend_size;
 313         offload_ctx->open_record = record;
 314         return 0;
 315 }
 316
 317 static int tls_do_allocation(struct sock *sk,
 318                              struct tls_offload_context_tx *offload_ctx,
 319                              struct page_frag *pfrag,
 320                              size_t prepend_size)
 321 {
 322         int ret;
 323
 324         if (!offload_ctx->open_record) {
 325                 if (unlikely(!skb_page_frag_refill(prepend_size, pfrag,
 326                                                    sk->sk_allocation))) {
 327                         sk->sk_prot->enter_memory_pressure(sk);
 328                         sk_stream_moderate_sndbuf(sk);
 329                         return -ENOMEM;
 330                 }
 331
 332                 ret = tls_create_new_record(offload_ctx, pfrag, prepend_size);
 333                 if (ret)
 334                         return ret;
 335
 336                 if (pfrag->size > pfrag->offset)
 337                         return 0;
 338         }
 339
 340         if (!sk_page_frag_refill(sk, pfrag))
 341                 return -ENOMEM;
 342
 343         return 0;
 344 }
 345
 346 static int tls_push_data(struct sock *sk,
 347                          struct iov_iter *msg_iter,
 348                          size_t size, int flags,
 349                          unsigned char record_type)
 350 {
 351         struct tls_context *tls_ctx = tls_get_ctx(sk);
 352         struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
 353         int tls_push_record_flags = flags | MSG_SENDPAGE_NOTLAST;
 354         struct tls_record_info *record = ctx->open_record;
 355         struct page_frag *pfrag;
 356         size_t orig_size = size;
 357         u32 max_open_record_len;
 358         bool more = false;
 359         bool done = false;
 360         int copy, rc = 0;
 361         long timeo;
 362
 363         if (flags &
 364             ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_SENDPAGE_NOTLAST))
 365                 return -ENOTSUPP;
 366
 367         if (sk->sk_err)
 368                 return -sk->sk_err;
 369
 370         timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
 371         rc = tls_complete_pending_work(sk, tls_ctx, flags, &timeo);
 372         if (rc < 0)
 373                 return rc;
 374
 375         pfrag = sk_page_frag(sk);
 376
 377         /* TLS_HEADER_SIZE is not counted as part of the TLS record, and
 378          * we need to leave room for an authentication tag.
 379          */
 380         max_open_record_len = TLS_MAX_PAYLOAD_SIZE +
 381                               tls_ctx->tx.prepend_size;
 382         do {
 383                 rc = tls_do_allocation(sk, ctx, pfrag,
 384                                        tls_ctx->tx.prepend_size);
 385                 if (rc) {
 386                         rc = sk_stream_wait_memory(sk, &timeo);
 387                         if (!rc)
 388                                 continue;
 389
 390                         record = ctx->open_record;
 391                         if (!record)
 392                                 break;
 393 handle_error:
 394                         if (record_type != TLS_RECORD_TYPE_DATA) {
 395                                 /* avoid sending partial
 396                                  * record with type !=
 397                                  * application_data
 398                                  */
 399                                 size = orig_size;
 400                                 destroy_record(record);
 401                                 ctx->open_record = NULL;
 402                         } else if (record->len > tls_ctx->tx.prepend_size) {
 403                                 goto last_record;
 404                         }
 405
 406                         break;
 407                 }
 408
 409                 record = ctx->open_record;
 410                 copy = min_t(size_t, size, (pfrag->size - pfrag->offset));
 411                 copy = min_t(size_t, copy, (max_open_record_len - record->len));
 412
 413                 if (copy_from_iter_nocache(page_address(pfrag->page) +
 414                                                pfrag->offset,
 415                                            copy, msg_iter) != copy) {
 416                         rc = -EFAULT;
 417                         goto handle_error;
 418                 }
 419                 tls_append_frag(record, pfrag, copy);
 420
 421                 size -= copy;
 422                 if (!size) {
 423 last_record:
 424                         tls_push_record_flags = flags;
 425                         if (flags & (MSG_SENDPAGE_NOTLAST | MSG_MORE)) {
 426                                 more = true;
 427                                 break;
 428                         }
 429
 430                         done = true;
 431                 }
 432
 433                 if (done || record->len >= max_open_record_len ||
 434                     (record->num_frags >= MAX_SKB_FRAGS - 1)) {
 435                         rc = tls_push_record(sk,
 436                                              tls_ctx,
 437                                              ctx,
 438                                              record,
 439                                              pfrag,
 440                                              tls_push_record_flags,
 441                                              record_type);
 442                         if (rc < 0)
 443                                 break;
 444                 }
 445         } while (!done);
 446
 447         tls_ctx->pending_open_record_frags = more;
 448
 449         if (orig_size - size > 0)
 450                 rc = orig_size - size;
 451
 452         return rc;
 453 }
 454
 455 int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
 456 {
 457         unsigned char record_type = TLS_RECORD_TYPE_DATA;
 458         int rc;
 459
 460         lock_sock(sk);
 461
 462         if (unlikely(msg->msg_controllen)) {
 463                 rc = tls_proccess_cmsg(sk, msg, &record_type);
 464                 if (rc)
 465                         goto out;
 466         }
 467
 468         rc = tls_push_data(sk, &msg->msg_iter, size,
 469                            msg->msg_flags, record_type);
 470
 471 out:
 472         release_sock(sk);
 473         return rc;
 474 }
 475
 476 int tls_device_sendpage(struct sock *sk, struct page *page,
 477                         int offset, size_t size, int flags)
 478 {
 479         struct iov_iter msg_iter;
 480         char *kaddr;
 481         struct kvec iov;
 482         int rc;
 483
 484         if (flags & MSG_SENDPAGE_NOTLAST)
 485                 flags |= MSG_MORE;
 486
 487         lock_sock(sk);
 488
 489         if (flags & MSG_OOB) {
 490                 rc = -ENOTSUPP;
 491                 goto out;
 492         }
 493
 494         kaddr = kmap(page);
 495         iov.iov_base = kaddr + offset;
 496         iov.iov_len = size;
 497         iov_iter_kvec(&msg_iter, WRITE | ITER_KVEC, &iov, 1, size);
 498         rc = tls_push_data(sk, &msg_iter, size,
 499                            flags, TLS_RECORD_TYPE_DATA);
 500         kunmap(page);
 501
 502 out:
 503         release_sock(sk);
 504         return rc;
 505 }
 506
 507 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
 508                                        u32 seq, u64 *p_record_sn)
 509 {
 510         u64 record_sn = context->hint_record_sn;
 511         struct tls_record_info *info, *last;
 512
 513         info = context->retransmit_hint;
 514         if (!info ||
 515             before(seq, info->end_seq - info->len)) {
 516                 /* if retransmit_hint is irrelevant start
 517                  * from the beggining of the list
 518                  */
 519                 info = list_first_entry(&context->records_list,
 520                                         struct tls_record_info, list);
 521
 522                 /* send the start_marker record if seq number is before the
 523                  * tls offload start marker sequence number. This record is
 524                  * required to handle TCP packets which are before TLS offload
 525                  * started.
 526                  *  And if it's not start marker, look if this seq number
 527                  * belongs to the list.
 528                  */
 529                 if (likely(!tls_record_is_start_marker(info))) {
 530                         /* we have the first record, get the last record to see
 531                          * if this seq number belongs to the list.
 532                          */
 533                         last = list_last_entry(&context->records_list,
 534                                                struct tls_record_info, list);
 535
 536                         if (!between(seq, tls_record_start_seq(info),
 537                                      last->end_seq))
 538                                 return NULL;
 539                 }
 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 | ITER_KVEC, NULL, 0, 0);
 566         return tls_push_data(sk, &msg_iter, 0, flags, TLS_RECORD_TYPE_DATA);
 567 }
 568
 569 static void tls_device_resync_rx(struct tls_context *tls_ctx,
 570                                  struct sock *sk, u32 seq, u64 rcd_sn)
 571 {
 572         struct net_device *netdev;
 573
 574         if (WARN_ON(test_and_set_bit(TLS_RX_SYNC_RUNNING, &tls_ctx->flags)))
 575                 return;
 576         netdev = READ_ONCE(tls_ctx->netdev);
 577         if (netdev)
 578                 netdev->tlsdev_ops->tls_dev_resync_rx(netdev, sk, seq, rcd_sn);
 579         clear_bit_unlock(TLS_RX_SYNC_RUNNING, &tls_ctx->flags);
 580 }
 581
 582 void handle_device_resync(struct sock *sk, u32 seq, u64 rcd_sn)
 583 {
 584         struct tls_context *tls_ctx = tls_get_ctx(sk);
 585         struct tls_offload_context_rx *rx_ctx;
 586         u32 is_req_pending;
 587         s64 resync_req;
 588         u32 req_seq;
 589
 590         if (tls_ctx->rx_conf != TLS_HW)
 591                 return;
 592
 593         rx_ctx = tls_offload_ctx_rx(tls_ctx);
 594         resync_req = atomic64_read(&rx_ctx->resync_req);
 595         req_seq = ntohl(resync_req >> 32) - ((u32)TLS_HEADER_SIZE - 1);
 596         is_req_pending = resync_req;
 597
 598         if (unlikely(is_req_pending) && req_seq == seq &&
 599             atomic64_try_cmpxchg(&rx_ctx->resync_req, &resync_req, 0)) {
 600                 seq += TLS_HEADER_SIZE - 1;
 601                 tls_device_resync_rx(tls_ctx, sk, seq, rcd_sn);
 602         }
 603 }
 604
 605 static int tls_device_reencrypt(struct sock *sk, struct sk_buff *skb)
 606 {
 607         struct strp_msg *rxm = strp_msg(skb);
 608         int err = 0, offset = rxm->offset, copy, nsg, data_len, pos;
 609         struct sk_buff *skb_iter, *unused;
 610         struct scatterlist sg[1];
 611         char *orig_buf, *buf;
 612
 613         orig_buf = kmalloc(rxm->full_len + TLS_HEADER_SIZE +
 614                            TLS_CIPHER_AES_GCM_128_IV_SIZE, sk->sk_allocation);
 615         if (!orig_buf)
 616                 return -ENOMEM;
 617         buf = orig_buf;
 618
 619         nsg = skb_cow_data(skb, 0, &unused);
 620         if (unlikely(nsg < 0)) {
 621                 err = nsg;
 622                 goto free_buf;
 623         }
 624
 625         sg_init_table(sg, 1);
 626         sg_set_buf(&sg[0], buf,
 627                    rxm->full_len + TLS_HEADER_SIZE +
 628                    TLS_CIPHER_AES_GCM_128_IV_SIZE);
 629         skb_copy_bits(skb, offset, buf,
 630                       TLS_HEADER_SIZE + TLS_CIPHER_AES_GCM_128_IV_SIZE);
 631
 632         /* We are interested only in the decrypted data not the auth */
 633         err = decrypt_skb(sk, skb, sg);
 634         if (err != -EBADMSG)
 635                 goto free_buf;
 636         else
 637                 err = 0;
 638
 639         data_len = rxm->full_len - TLS_CIPHER_AES_GCM_128_TAG_SIZE;
 640
 641         if (skb_pagelen(skb) > offset) {
 642                 copy = min_t(int, skb_pagelen(skb) - offset, data_len);
 643
 644                 if (skb->decrypted)
 645                         skb_store_bits(skb, offset, buf, copy);
 646
 647                 offset += copy;
 648                 buf += copy;
 649         }
 650
 651         pos = skb_pagelen(skb);
 652         skb_walk_frags(skb, skb_iter) {
 653                 int frag_pos;
 654
 655                 /* Practically all frags must belong to msg if reencrypt
 656                  * is needed with current strparser and coalescing logic,
 657                  * but strparser may "get optimized", so let's be safe.
 658                  */
 659                 if (pos + skb_iter->len <= offset)
 660                         goto done_with_frag;
 661                 if (pos >= data_len + rxm->offset)
 662                         break;
 663
 664                 frag_pos = offset - pos;
 665                 copy = min_t(int, skb_iter->len - frag_pos,
 666                              data_len + rxm->offset - offset);
 667
 668                 if (skb_iter->decrypted)
 669                         skb_store_bits(skb_iter, frag_pos, buf, copy);
 670
 671                 offset += copy;
 672                 buf += copy;
 673 done_with_frag:
 674                 pos += skb_iter->len;
 675         }
 676
 677 free_buf:
 678         kfree(orig_buf);
 679         return err;
 680 }
 681
 682 int tls_device_decrypted(struct sock *sk, struct sk_buff *skb)
 683 {
 684         struct tls_context *tls_ctx = tls_get_ctx(sk);
 685         struct tls_offload_context_rx *ctx = tls_offload_ctx_rx(tls_ctx);
 686         int is_decrypted = skb->decrypted;
 687         int is_encrypted = !is_decrypted;
 688         struct sk_buff *skb_iter;
 689
 690         /* Skip if it is already decrypted */
 691         if (ctx->sw.decrypted)
 692                 return 0;
 693
 694         /* Check if all the data is decrypted already */
 695         skb_walk_frags(skb, skb_iter) {
 696                 is_decrypted &= skb_iter->decrypted;
 697                 is_encrypted &= !skb_iter->decrypted;
 698         }
 699
 700         ctx->sw.decrypted |= is_decrypted;
 701
 702         /* Return immedeatly if the record is either entirely plaintext or
 703          * entirely ciphertext. Otherwise handle reencrypt partially decrypted
 704          * record.
 705          */
 706         return (is_encrypted || is_decrypted) ? 0 :
 707                 tls_device_reencrypt(sk, skb);
 708 }
 709
 710 int tls_set_device_offload(struct sock *sk, struct tls_context *ctx)
 711 {
 712         u16 nonce_size, tag_size, iv_size, rec_seq_size;
 713         struct tls_record_info *start_marker_record;
 714         struct tls_offload_context_tx *offload_ctx;
 715         struct tls_crypto_info *crypto_info;
 716         struct net_device *netdev;
 717         char *iv, *rec_seq;
 718         struct sk_buff *skb;
 719         int rc = -EINVAL;
 720         __be64 rcd_sn;
 721
 722         if (!ctx)
 723                 goto out;
 724
 725         if (ctx->priv_ctx_tx) {
 726                 rc = -EEXIST;
 727                 goto out;
 728         }
 729
 730         start_marker_record = kmalloc(sizeof(*start_marker_record), GFP_KERNEL);
 731         if (!start_marker_record) {
 732                 rc = -ENOMEM;
 733                 goto out;
 734         }
 735
 736         offload_ctx = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_TX, GFP_KERNEL);
 737         if (!offload_ctx) {
 738                 rc = -ENOMEM;
 739                 goto free_marker_record;
 740         }
 741
 742         crypto_info = &ctx->crypto_send.info;
 743         switch (crypto_info->cipher_type) {
 744         case TLS_CIPHER_AES_GCM_128:
 745                 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
 746                 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
 747                 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
 748                 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
 749                 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
 750                 rec_seq =
 751                  ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
 752                 break;
 753         default:
 754                 rc = -EINVAL;
 755                 goto free_offload_ctx;
 756         }
 757
 758         ctx->tx.prepend_size = TLS_HEADER_SIZE + nonce_size;
 759         ctx->tx.tag_size = tag_size;
 760         ctx->tx.overhead_size = ctx->tx.prepend_size + ctx->tx.tag_size;
 761         ctx->tx.iv_size = iv_size;
 762         ctx->tx.iv = kmalloc(iv_size + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
 763                              GFP_KERNEL);
 764         if (!ctx->tx.iv) {
 765                 rc = -ENOMEM;
 766                 goto free_offload_ctx;
 767         }
 768
 769         memcpy(ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv, iv_size);
 770
 771         ctx->tx.rec_seq_size = rec_seq_size;
 772         ctx->tx.rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
 773         if (!ctx->tx.rec_seq) {
 774                 rc = -ENOMEM;
 775                 goto free_iv;
 776         }
 777
 778         rc = tls_sw_fallback_init(sk, offload_ctx, crypto_info);
 779         if (rc)
 780                 goto free_rec_seq;
 781
 782         /* start at rec_seq - 1 to account for the start marker record */
 783         memcpy(&rcd_sn, ctx->tx.rec_seq, sizeof(rcd_sn));
 784         offload_ctx->unacked_record_sn = be64_to_cpu(rcd_sn) - 1;
 785
 786         start_marker_record->end_seq = tcp_sk(sk)->write_seq;
 787         start_marker_record->len = 0;
 788         start_marker_record->num_frags = 0;
 789
 790         INIT_LIST_HEAD(&offload_ctx->records_list);
 791         list_add_tail(&start_marker_record->list, &offload_ctx->records_list);
 792         spin_lock_init(&offload_ctx->lock);
 793         sg_init_table(offload_ctx->sg_tx_data,
 794                       ARRAY_SIZE(offload_ctx->sg_tx_data));
 795
 796         clean_acked_data_enable(inet_csk(sk), &tls_icsk_clean_acked);
 797         ctx->push_pending_record = tls_device_push_pending_record;
 798
 799         /* TLS offload is greatly simplified if we don't send
 800          * SKBs where only part of the payload needs to be encrypted.
 801          * So mark the last skb in the write queue as end of record.
 802          */
 803         skb = tcp_write_queue_tail(sk);
 804         if (skb)
 805                 TCP_SKB_CB(skb)->eor = 1;
 806
 807         /* We support starting offload on multiple sockets
 808          * concurrently, so we only need a read lock here.
 809          * This lock must precede get_netdev_for_sock to prevent races between
 810          * NETDEV_DOWN and setsockopt.
 811          */
 812         down_read(&device_offload_lock);
 813         netdev = get_netdev_for_sock(sk);
 814         if (!netdev) {
 815                 pr_err_ratelimited("%s: netdev not found\n", __func__);
 816                 rc = -EINVAL;
 817                 goto release_lock;
 818         }
 819
 820         if (!(netdev->features & NETIF_F_HW_TLS_TX)) {
 821                 rc = -ENOTSUPP;
 822                 goto release_netdev;
 823         }
 824
 825         /* Avoid offloading if the device is down
 826          * We don't want to offload new flows after
 827          * the NETDEV_DOWN event
 828          */
 829         if (!(netdev->flags & IFF_UP)) {
 830                 rc = -EINVAL;
 831                 goto release_netdev;
 832         }
 833
 834         ctx->priv_ctx_tx = offload_ctx;
 835         rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_TX,
 836                                              &ctx->crypto_send.info,
 837                                              tcp_sk(sk)->write_seq);
 838         if (rc)
 839                 goto release_netdev;
 840
 841         tls_device_attach(ctx, sk, netdev);
 842
 843         /* following this assignment tls_is_sk_tx_device_offloaded
 844          * will return true and the context might be accessed
 845          * by the netdev's xmit function.
 846          */
 847         smp_store_release(&sk->sk_validate_xmit_skb, tls_validate_xmit_skb);
 848         dev_put(netdev);
 849         up_read(&device_offload_lock);
 850         goto out;
 851
 852 release_netdev:
 853         dev_put(netdev);
 854 release_lock:
 855         up_read(&device_offload_lock);
 856         clean_acked_data_disable(inet_csk(sk));
 857         crypto_free_aead(offload_ctx->aead_send);
 858 free_rec_seq:
 859         kfree(ctx->tx.rec_seq);
 860 free_iv:
 861         kfree(ctx->tx.iv);
 862 free_offload_ctx:
 863         kfree(offload_ctx);
 864         ctx->priv_ctx_tx = NULL;
 865 free_marker_record:
 866         kfree(start_marker_record);
 867 out:
 868         return rc;
 869 }
 870
 871 int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx)
 872 {
 873         struct tls_offload_context_rx *context;
 874         struct net_device *netdev;
 875         int rc = 0;
 876
 877         /* We support starting offload on multiple sockets
 878          * concurrently, so we only need a read lock here.
 879          * This lock must precede get_netdev_for_sock to prevent races between
 880          * NETDEV_DOWN and setsockopt.
 881          */
 882         down_read(&device_offload_lock);
 883         netdev = get_netdev_for_sock(sk);
 884         if (!netdev) {
 885                 pr_err_ratelimited("%s: netdev not found\n", __func__);
 886                 rc = -EINVAL;
 887                 goto release_lock;
 888         }
 889
 890         if (!(netdev->features & NETIF_F_HW_TLS_RX)) {
 891                 pr_err_ratelimited("%s: netdev %s with no TLS offload\n",
 892                                    __func__, netdev->name);
 893                 rc = -ENOTSUPP;
 894                 goto release_netdev;
 895         }
 896
 897         /* Avoid offloading if the device is down
 898          * We don't want to offload new flows after
 899          * the NETDEV_DOWN event
 900          */
 901         if (!(netdev->flags & IFF_UP)) {
 902                 rc = -EINVAL;
 903                 goto release_netdev;
 904         }
 905
 906         context = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_RX, GFP_KERNEL);
 907         if (!context) {
 908                 rc = -ENOMEM;
 909                 goto release_netdev;
 910         }
 911
 912         ctx->priv_ctx_rx = context;
 913         rc = tls_set_sw_offload(sk, ctx, 0);
 914         if (rc)
 915                 goto release_ctx;
 916
 917         rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_RX,
 918                                              &ctx->crypto_recv.info,
 919                                              tcp_sk(sk)->copied_seq);
 920         if (rc) {
 921                 pr_err_ratelimited("%s: The netdev has refused to offload this socket\n",
 922                                    __func__);
 923                 goto free_sw_resources;
 924         }
 925
 926         tls_device_attach(ctx, sk, netdev);
 927         goto release_netdev;
 928
 929 free_sw_resources:
 930         up_read(&device_offload_lock);
 931         tls_sw_free_resources_rx(sk);
 932         down_read(&device_offload_lock);
 933 release_ctx:
 934         ctx->priv_ctx_rx = NULL;
 935 release_netdev:
 936         dev_put(netdev);
 937 release_lock:
 938         up_read(&device_offload_lock);
 939         return rc;
 940 }
 941
 942 void tls_device_offload_cleanup_rx(struct sock *sk)
 943 {
 944         struct tls_context *tls_ctx = tls_get_ctx(sk);
 945         struct net_device *netdev;
 946
 947         down_read(&device_offload_lock);
 948         netdev = tls_ctx->netdev;
 949         if (!netdev)
 950                 goto out;
 951
 952         netdev->tlsdev_ops->tls_dev_del(netdev, tls_ctx,
 953                                         TLS_OFFLOAD_CTX_DIR_RX);
 954
 955         if (tls_ctx->tx_conf != TLS_HW) {
 956                 dev_put(netdev);
 957                 tls_ctx->netdev = NULL;
 958         } else {
 959                 set_bit(TLS_RX_DEV_CLOSED, &tls_ctx->flags);
 960         }
 961 out:
 962         up_read(&device_offload_lock);
 963         tls_sw_release_resources_rx(sk);
 964 }
 965
 966 static int tls_device_down(struct net_device *netdev)
 967 {
 968         struct tls_context *ctx, *tmp;
 969         unsigned long flags;
 970         LIST_HEAD(list);
 971
 972         /* Request a write lock to block new offload attempts */
 973         down_write(&device_offload_lock);
 974
 975         spin_lock_irqsave(&tls_device_lock, flags);
 976         list_for_each_entry_safe(ctx, tmp, &tls_device_list, list) {
 977                 if (ctx->netdev != netdev ||
 978                     !refcount_inc_not_zero(&ctx->refcount))
 979                         continue;
 980
 981                 list_move(&ctx->list, &list);
 982         }
 983         spin_unlock_irqrestore(&tls_device_lock, flags);
 984
 985         list_for_each_entry_safe(ctx, tmp, &list, list) {
 986                 if (ctx->tx_conf == TLS_HW)
 987                         netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
 988                                                         TLS_OFFLOAD_CTX_DIR_TX);
 989                 if (ctx->rx_conf == TLS_HW &&
 990                     !test_bit(TLS_RX_DEV_CLOSED, &ctx->flags))
 991                         netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
 992                                                         TLS_OFFLOAD_CTX_DIR_RX);
 993                 WRITE_ONCE(ctx->netdev, NULL);
 994                 smp_mb__before_atomic(); /* pairs with test_and_set_bit() */
 995                 while (test_bit(TLS_RX_SYNC_RUNNING, &ctx->flags))
 996                         usleep_range(10, 200);
 997                 dev_put(netdev);
 998                 list_del_init(&ctx->list);
 999
1000                 if (refcount_dec_and_test(&ctx->refcount))
1001                         tls_device_free_ctx(ctx);
1002         }
1003
1004         up_write(&device_offload_lock);
1005
1006         flush_work(&tls_device_gc_work);
1007
1008         return NOTIFY_DONE;
1009 }
1010
1011 static int tls_dev_event(struct notifier_block *this, unsigned long event,
1012                          void *ptr)
1013 {
1014         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1015
1016         if (!dev->tlsdev_ops &&
1017             !(dev->features & (NETIF_F_HW_TLS_RX | NETIF_F_HW_TLS_TX)))
1018                 return NOTIFY_DONE;
1019
1020         switch (event) {
1021         case NETDEV_REGISTER:
1022         case NETDEV_FEAT_CHANGE:
1023                 if ((dev->features & NETIF_F_HW_TLS_RX) &&
1024                     !dev->tlsdev_ops->tls_dev_resync_rx)
1025                         return NOTIFY_BAD;
1026
1027                 if  (dev->tlsdev_ops &&
1028                      dev->tlsdev_ops->tls_dev_add &&
1029                      dev->tlsdev_ops->tls_dev_del)
1030                         return NOTIFY_DONE;
1031                 else
1032                         return NOTIFY_BAD;
1033         case NETDEV_DOWN:
1034                 return tls_device_down(dev);
1035         }
1036         return NOTIFY_DONE;
1037 }
1038
1039 static struct notifier_block tls_dev_notifier = {
1040         .notifier_call  = tls_dev_event,
1041 };
1042
1043 void __init tls_device_init(void)
1044 {
1045         register_netdevice_notifier(&tls_dev_notifier);
1046 }
1047
1048 void __exit tls_device_cleanup(void)
1049 {
1050         unregister_netdevice_notifier(&tls_dev_notifier);
1051         flush_work(&tls_device_gc_work);
1052 }