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
  58         if (ctx->rx_conf == TLS_HW)
  59                 kfree(tls_offload_ctx_rx(ctx));
  60
  61         kfree(ctx);
  62 }
  63
  64 static void tls_device_gc_task(struct work_struct *work)
  65 {
  66         struct tls_context *ctx, *tmp;
  67         unsigned long flags;
  68         LIST_HEAD(gc_list);
  69
  70         spin_lock_irqsave(&tls_device_lock, flags);
  71         list_splice_init(&tls_device_gc_list, &gc_list);
  72         spin_unlock_irqrestore(&tls_device_lock, flags);
  73
  74         list_for_each_entry_safe(ctx, tmp, &gc_list, list) {
  75                 struct net_device *netdev = ctx->netdev;
  76
  77                 if (netdev && ctx->tx_conf == TLS_HW) {
  78                         netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
  79                                                         TLS_OFFLOAD_CTX_DIR_TX);
  80                         dev_put(netdev);
  81                         ctx->netdev = NULL;
  82                 }
  83
  84                 list_del(&ctx->list);
  85                 tls_device_free_ctx(ctx);
  86         }
  87 }
  88
  89 static void tls_device_attach(struct tls_context *ctx, struct sock *sk,
  90                               struct net_device *netdev)
  91 {
  92         if (sk->sk_destruct != tls_device_sk_destruct) {
  93                 refcount_set(&ctx->refcount, 1);
  94                 dev_hold(netdev);
  95                 ctx->netdev = netdev;
  96                 spin_lock_irq(&tls_device_lock);
  97                 list_add_tail(&ctx->list, &tls_device_list);
  98                 spin_unlock_irq(&tls_device_lock);
  99
 100                 ctx->sk_destruct = sk->sk_destruct;
 101                 sk->sk_destruct = tls_device_sk_destruct;
 102         }
 103 }
 104
 105 static void tls_device_queue_ctx_destruction(struct tls_context *ctx)
 106 {
 107         unsigned long flags;
 108
 109         spin_lock_irqsave(&tls_device_lock, flags);
 110         list_move_tail(&ctx->list, &tls_device_gc_list);
 111
 112         /* schedule_work inside the spinlock
 113          * to make sure tls_device_down waits for that work.
 114          */
 115         schedule_work(&tls_device_gc_work);
 116
 117         spin_unlock_irqrestore(&tls_device_lock, flags);
 118 }
 119
 120 /* We assume that the socket is already connected */
 121 static struct net_device *get_netdev_for_sock(struct sock *sk)
 122 {
 123         struct dst_entry *dst = sk_dst_get(sk);
 124         struct net_device *netdev = NULL;
 125
 126         if (likely(dst)) {
 127                 netdev = dst->dev;
 128                 dev_hold(netdev);
 129         }
 130
 131         dst_release(dst);
 132
 133         return netdev;
 134 }
 135
 136 static void destroy_record(struct tls_record_info *record)
 137 {
 138         int nr_frags = record->num_frags;
 139         skb_frag_t *frag;
 140
 141         while (nr_frags-- > 0) {
 142                 frag = &record->frags[nr_frags];
 143                 __skb_frag_unref(frag);
 144         }
 145         kfree(record);
 146 }
 147
 148 static void delete_all_records(struct tls_offload_context_tx *offload_ctx)
 149 {
 150         struct tls_record_info *info, *temp;
 151
 152         list_for_each_entry_safe(info, temp, &offload_ctx->records_list, list) {
 153                 list_del(&info->list);
 154                 destroy_record(info);
 155         }
 156
 157         offload_ctx->retransmit_hint = NULL;
 158 }
 159
 160 static void tls_icsk_clean_acked(struct sock *sk, u32 acked_seq)
 161 {
 162         struct tls_context *tls_ctx = tls_get_ctx(sk);
 163         struct tls_record_info *info, *temp;
 164         struct tls_offload_context_tx *ctx;
 165         u64 deleted_records = 0;
 166         unsigned long flags;
 167
 168         if (!tls_ctx)
 169                 return;
 170
 171         ctx = tls_offload_ctx_tx(tls_ctx);
 172
 173         spin_lock_irqsave(&ctx->lock, flags);
 174         info = ctx->retransmit_hint;
 175         if (info && !before(acked_seq, info->end_seq)) {
 176                 ctx->retransmit_hint = NULL;
 177                 list_del(&info->list);
 178                 destroy_record(info);
 179                 deleted_records++;
 180         }
 181
 182         list_for_each_entry_safe(info, temp, &ctx->records_list, list) {
 183                 if (before(acked_seq, info->end_seq))
 184                         break;
 185                 list_del(&info->list);
 186
 187                 destroy_record(info);
 188                 deleted_records++;
 189         }
 190
 191         ctx->unacked_record_sn += deleted_records;
 192         spin_unlock_irqrestore(&ctx->lock, flags);
 193 }
 194
 195 /* At this point, there should be no references on this
 196  * socket and no in-flight SKBs associated with this
 197  * socket, so it is safe to free all the resources.
 198  */
 199 void tls_device_sk_destruct(struct sock *sk)
 200 {
 201         struct tls_context *tls_ctx = tls_get_ctx(sk);
 202         struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
 203
 204         tls_ctx->sk_destruct(sk);
 205
 206         if (tls_ctx->tx_conf == TLS_HW) {
 207                 if (ctx->open_record)
 208                         destroy_record(ctx->open_record);
 209                 delete_all_records(ctx);
 210                 crypto_free_aead(ctx->aead_send);
 211                 clean_acked_data_disable(inet_csk(sk));
 212         }
 213
 214         if (refcount_dec_and_test(&tls_ctx->refcount))
 215                 tls_device_queue_ctx_destruction(tls_ctx);
 216 }
 217 EXPORT_SYMBOL(tls_device_sk_destruct);
 218
 219 static void tls_append_frag(struct tls_record_info *record,
 220                             struct page_frag *pfrag,
 221                             int size)
 222 {
 223         skb_frag_t *frag;
 224
 225         frag = &record->frags[record->num_frags - 1];
 226         if (frag->page.p == pfrag->page &&
 227             frag->page_offset + frag->size == pfrag->offset) {
 228                 frag->size += size;
 229         } else {
 230                 ++frag;
 231                 frag->page.p = pfrag->page;
 232                 frag->page_offset = pfrag->offset;
 233                 frag->size = size;
 234                 ++record->num_frags;
 235                 get_page(pfrag->page);
 236         }
 237
 238         pfrag->offset += size;
 239         record->len += size;
 240 }
 241
 242 static int tls_push_record(struct sock *sk,
 243                            struct tls_context *ctx,
 244                            struct tls_offload_context_tx *offload_ctx,
 245                            struct tls_record_info *record,
 246                            struct page_frag *pfrag,
 247                            int flags,
 248                            unsigned char record_type)
 249 {
 250         struct tcp_sock *tp = tcp_sk(sk);
 251         struct page_frag dummy_tag_frag;
 252         skb_frag_t *frag;
 253         int i;
 254
 255         /* fill prepend */
 256         frag = &record->frags[0];
 257         tls_fill_prepend(ctx,
 258                          skb_frag_address(frag),
 259                          record->len - ctx->tx.prepend_size,
 260                          record_type);
 261
 262         /* HW doesn't care about the data in the tag, because it fills it. */
 263         dummy_tag_frag.page = skb_frag_page(frag);
 264         dummy_tag_frag.offset = 0;
 265
 266         tls_append_frag(record, &dummy_tag_frag, ctx->tx.tag_size);
 267         record->end_seq = tp->write_seq + record->len;
 268         spin_lock_irq(&offload_ctx->lock);
 269         list_add_tail(&record->list, &offload_ctx->records_list);
 270         spin_unlock_irq(&offload_ctx->lock);
 271         offload_ctx->open_record = NULL;
 272         set_bit(TLS_PENDING_CLOSED_RECORD, &ctx->flags);
 273         tls_advance_record_sn(sk, &ctx->tx);
 274
 275         for (i = 0; i < record->num_frags; i++) {
 276                 frag = &record->frags[i];
 277                 sg_unmark_end(&offload_ctx->sg_tx_data[i]);
 278                 sg_set_page(&offload_ctx->sg_tx_data[i], skb_frag_page(frag),
 279                             frag->size, frag->page_offset);
 280                 sk_mem_charge(sk, frag->size);
 281                 get_page(skb_frag_page(frag));
 282         }
 283         sg_mark_end(&offload_ctx->sg_tx_data[record->num_frags - 1]);
 284
 285         /* all ready, send */
 286         return tls_push_sg(sk, ctx, offload_ctx->sg_tx_data, 0, flags);
 287 }
 288
 289 static int tls_create_new_record(struct tls_offload_context_tx *offload_ctx,
 290                                  struct page_frag *pfrag,
 291                                  size_t prepend_size)
 292 {
 293         struct tls_record_info *record;
 294         skb_frag_t *frag;
 295
 296         record = kmalloc(sizeof(*record), GFP_KERNEL);
 297         if (!record)
 298                 return -ENOMEM;
 299
 300         frag = &record->frags[0];
 301         __skb_frag_set_page(frag, pfrag->page);
 302         frag->page_offset = pfrag->offset;
 303         skb_frag_size_set(frag, prepend_size);
 304
 305         get_page(pfrag->page);
 306         pfrag->offset += prepend_size;
 307
 308         record->num_frags = 1;
 309         record->len = prepend_size;
 310         offload_ctx->open_record = record;
 311         return 0;
 312 }
 313
 314 static int tls_do_allocation(struct sock *sk,
 315                              struct tls_offload_context_tx *offload_ctx,
 316                              struct page_frag *pfrag,
 317                              size_t prepend_size)
 318 {
 319         int ret;
 320
 321         if (!offload_ctx->open_record) {
 322                 if (unlikely(!skb_page_frag_refill(prepend_size, pfrag,
 323                                                    sk->sk_allocation))) {
 324                         sk->sk_prot->enter_memory_pressure(sk);
 325                         sk_stream_moderate_sndbuf(sk);
 326                         return -ENOMEM;
 327                 }
 328
 329                 ret = tls_create_new_record(offload_ctx, pfrag, prepend_size);
 330                 if (ret)
 331                         return ret;
 332
 333                 if (pfrag->size > pfrag->offset)
 334                         return 0;
 335         }
 336
 337         if (!sk_page_frag_refill(sk, pfrag))
 338                 return -ENOMEM;
 339
 340         return 0;
 341 }
 342
 343 static int tls_push_data(struct sock *sk,
 344                          struct iov_iter *msg_iter,
 345                          size_t size, int flags,
 346                          unsigned char record_type)
 347 {
 348         struct tls_context *tls_ctx = tls_get_ctx(sk);
 349         struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
 350         int tls_push_record_flags = flags | MSG_SENDPAGE_NOTLAST;
 351         int more = flags & (MSG_SENDPAGE_NOTLAST | MSG_MORE);
 352         struct tls_record_info *record = ctx->open_record;
 353         struct page_frag *pfrag;
 354         size_t orig_size = size;
 355         u32 max_open_record_len;
 356         int copy, rc = 0;
 357         bool done = false;
 358         long timeo;
 359
 360         if (flags &
 361             ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_SENDPAGE_NOTLAST))
 362                 return -ENOTSUPP;
 363
 364         if (sk->sk_err)
 365                 return -sk->sk_err;
 366
 367         timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
 368         rc = tls_complete_pending_work(sk, tls_ctx, flags, &timeo);
 369         if (rc < 0)
 370                 return rc;
 371
 372         pfrag = sk_page_frag(sk);
 373
 374         /* TLS_HEADER_SIZE is not counted as part of the TLS record, and
 375          * we need to leave room for an authentication tag.
 376          */
 377         max_open_record_len = TLS_MAX_PAYLOAD_SIZE +
 378                               tls_ctx->tx.prepend_size;
 379         do {
 380                 rc = tls_do_allocation(sk, ctx, pfrag,
 381                                        tls_ctx->tx.prepend_size);
 382                 if (rc) {
 383                         rc = sk_stream_wait_memory(sk, &timeo);
 384                         if (!rc)
 385                                 continue;
 386
 387                         record = ctx->open_record;
 388                         if (!record)
 389                                 break;
 390 handle_error:
 391                         if (record_type != TLS_RECORD_TYPE_DATA) {
 392                                 /* avoid sending partial
 393                                  * record with type !=
 394                                  * application_data
 395                                  */
 396                                 size = orig_size;
 397                                 destroy_record(record);
 398                                 ctx->open_record = NULL;
 399                         } else if (record->len > tls_ctx->tx.prepend_size) {
 400                                 goto last_record;
 401                         }
 402
 403                         break;
 404                 }
 405
 406                 record = ctx->open_record;
 407                 copy = min_t(size_t, size, (pfrag->size - pfrag->offset));
 408                 copy = min_t(size_t, copy, (max_open_record_len - record->len));
 409
 410                 if (copy_from_iter_nocache(page_address(pfrag->page) +
 411                                                pfrag->offset,
 412                                            copy, msg_iter) != copy) {
 413                         rc = -EFAULT;
 414                         goto handle_error;
 415                 }
 416                 tls_append_frag(record, pfrag, copy);
 417
 418                 size -= copy;
 419                 if (!size) {
 420 last_record:
 421                         tls_push_record_flags = flags;
 422                         if (more) {
 423                                 tls_ctx->pending_open_record_frags =
 424                                                 !!record->num_frags;
 425                                 break;
 426                         }
 427
 428                         done = true;
 429                 }
 430
 431                 if (done || record->len >= max_open_record_len ||
 432                     (record->num_frags >= MAX_SKB_FRAGS - 1)) {
 433                         rc = tls_push_record(sk,
 434                                              tls_ctx,
 435                                              ctx,
 436                                              record,
 437                                              pfrag,
 438                                              tls_push_record_flags,
 439                                              record_type);
 440                         if (rc < 0)
 441                                 break;
 442                 }
 443         } while (!done);
 444
 445         if (orig_size - size > 0)
 446                 rc = orig_size - size;
 447
 448         return rc;
 449 }
 450
 451 int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
 452 {
 453         unsigned char record_type = TLS_RECORD_TYPE_DATA;
 454         int rc;
 455
 456         lock_sock(sk);
 457
 458         if (unlikely(msg->msg_controllen)) {
 459                 rc = tls_proccess_cmsg(sk, msg, &record_type);
 460                 if (rc)
 461                         goto out;
 462         }
 463
 464         rc = tls_push_data(sk, &msg->msg_iter, size,
 465                            msg->msg_flags, record_type);
 466
 467 out:
 468         release_sock(sk);
 469         return rc;
 470 }
 471
 472 int tls_device_sendpage(struct sock *sk, struct page *page,
 473                         int offset, size_t size, int flags)
 474 {
 475         struct iov_iter msg_iter;
 476         char *kaddr = kmap(page);
 477         struct kvec iov;
 478         int rc;
 479
 480         if (flags & MSG_SENDPAGE_NOTLAST)
 481                 flags |= MSG_MORE;
 482
 483         lock_sock(sk);
 484
 485         if (flags & MSG_OOB) {
 486                 rc = -ENOTSUPP;
 487                 goto out;
 488         }
 489
 490         iov.iov_base = kaddr + offset;
 491         iov.iov_len = size;
 492         iov_iter_kvec(&msg_iter, WRITE | ITER_KVEC, &iov, 1, size);
 493         rc = tls_push_data(sk, &msg_iter, size,
 494                            flags, TLS_RECORD_TYPE_DATA);
 495         kunmap(page);
 496
 497 out:
 498         release_sock(sk);
 499         return rc;
 500 }
 501
 502 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
 503                                        u32 seq, u64 *p_record_sn)
 504 {
 505         u64 record_sn = context->hint_record_sn;
 506         struct tls_record_info *info;
 507
 508         info = context->retransmit_hint;
 509         if (!info ||
 510             before(seq, info->end_seq - info->len)) {
 511                 /* if retransmit_hint is irrelevant start
 512                  * from the beggining of the list
 513                  */
 514                 info = list_first_entry(&context->records_list,
 515                                         struct tls_record_info, list);
 516                 record_sn = context->unacked_record_sn;
 517         }
 518
 519         list_for_each_entry_from(info, &context->records_list, list) {
 520                 if (before(seq, info->end_seq)) {
 521                         if (!context->retransmit_hint ||
 522                             after(info->end_seq,
 523                                   context->retransmit_hint->end_seq)) {
 524                                 context->hint_record_sn = record_sn;
 525                                 context->retransmit_hint = info;
 526                         }
 527                         *p_record_sn = record_sn;
 528                         return info;
 529                 }
 530                 record_sn++;
 531         }
 532
 533         return NULL;
 534 }
 535 EXPORT_SYMBOL(tls_get_record);
 536
 537 static int tls_device_push_pending_record(struct sock *sk, int flags)
 538 {
 539         struct iov_iter msg_iter;
 540
 541         iov_iter_kvec(&msg_iter, WRITE | ITER_KVEC, NULL, 0, 0);
 542         return tls_push_data(sk, &msg_iter, 0, flags, TLS_RECORD_TYPE_DATA);
 543 }
 544
 545 void handle_device_resync(struct sock *sk, u32 seq, u64 rcd_sn)
 546 {
 547         struct tls_context *tls_ctx = tls_get_ctx(sk);
 548         struct net_device *netdev = tls_ctx->netdev;
 549         struct tls_offload_context_rx *rx_ctx;
 550         u32 is_req_pending;
 551         s64 resync_req;
 552         u32 req_seq;
 553
 554         if (tls_ctx->rx_conf != TLS_HW)
 555                 return;
 556
 557         rx_ctx = tls_offload_ctx_rx(tls_ctx);
 558         resync_req = atomic64_read(&rx_ctx->resync_req);
 559         req_seq = ntohl(resync_req >> 32) - ((u32)TLS_HEADER_SIZE - 1);
 560         is_req_pending = resync_req;
 561
 562         if (unlikely(is_req_pending) && req_seq == seq &&
 563             atomic64_try_cmpxchg(&rx_ctx->resync_req, &resync_req, 0))
 564                 netdev->tlsdev_ops->tls_dev_resync_rx(netdev, sk,
 565                                                       seq + TLS_HEADER_SIZE - 1,
 566                                                       rcd_sn);
 567 }
 568
 569 static int tls_device_reencrypt(struct sock *sk, struct sk_buff *skb)
 570 {
 571         struct strp_msg *rxm = strp_msg(skb);
 572         int err = 0, offset = rxm->offset, copy, nsg;
 573         struct sk_buff *skb_iter, *unused;
 574         struct scatterlist sg[1];
 575         char *orig_buf, *buf;
 576
 577         orig_buf = kmalloc(rxm->full_len + TLS_HEADER_SIZE +
 578                            TLS_CIPHER_AES_GCM_128_IV_SIZE, sk->sk_allocation);
 579         if (!orig_buf)
 580                 return -ENOMEM;
 581         buf = orig_buf;
 582
 583         nsg = skb_cow_data(skb, 0, &unused);
 584         if (unlikely(nsg < 0)) {
 585                 err = nsg;
 586                 goto free_buf;
 587         }
 588
 589         sg_init_table(sg, 1);
 590         sg_set_buf(&sg[0], buf,
 591                    rxm->full_len + TLS_HEADER_SIZE +
 592                    TLS_CIPHER_AES_GCM_128_IV_SIZE);
 593         skb_copy_bits(skb, offset, buf,
 594                       TLS_HEADER_SIZE + TLS_CIPHER_AES_GCM_128_IV_SIZE);
 595
 596         /* We are interested only in the decrypted data not the auth */
 597         err = decrypt_skb(sk, skb, sg);
 598         if (err != -EBADMSG)
 599                 goto free_buf;
 600         else
 601                 err = 0;
 602
 603         copy = min_t(int, skb_pagelen(skb) - offset,
 604                      rxm->full_len - TLS_CIPHER_AES_GCM_128_TAG_SIZE);
 605
 606         if (skb->decrypted)
 607                 skb_store_bits(skb, offset, buf, copy);
 608
 609         offset += copy;
 610         buf += copy;
 611
 612         skb_walk_frags(skb, skb_iter) {
 613                 copy = min_t(int, skb_iter->len,
 614                              rxm->full_len - offset + rxm->offset -
 615                              TLS_CIPHER_AES_GCM_128_TAG_SIZE);
 616
 617                 if (skb_iter->decrypted)
 618                         skb_store_bits(skb_iter, offset, buf, copy);
 619
 620                 offset += copy;
 621                 buf += copy;
 622         }
 623
 624 free_buf:
 625         kfree(orig_buf);
 626         return err;
 627 }
 628
 629 int tls_device_decrypted(struct sock *sk, struct sk_buff *skb)
 630 {
 631         struct tls_context *tls_ctx = tls_get_ctx(sk);
 632         struct tls_offload_context_rx *ctx = tls_offload_ctx_rx(tls_ctx);
 633         int is_decrypted = skb->decrypted;
 634         int is_encrypted = !is_decrypted;
 635         struct sk_buff *skb_iter;
 636
 637         /* Skip if it is already decrypted */
 638         if (ctx->sw.decrypted)
 639                 return 0;
 640
 641         /* Check if all the data is decrypted already */
 642         skb_walk_frags(skb, skb_iter) {
 643                 is_decrypted &= skb_iter->decrypted;
 644                 is_encrypted &= !skb_iter->decrypted;
 645         }
 646
 647         ctx->sw.decrypted |= is_decrypted;
 648
 649         /* Return immedeatly if the record is either entirely plaintext or
 650          * entirely ciphertext. Otherwise handle reencrypt partially decrypted
 651          * record.
 652          */
 653         return (is_encrypted || is_decrypted) ? 0 :
 654                 tls_device_reencrypt(sk, skb);
 655 }
 656
 657 int tls_set_device_offload(struct sock *sk, struct tls_context *ctx)
 658 {
 659         u16 nonce_size, tag_size, iv_size, rec_seq_size;
 660         struct tls_record_info *start_marker_record;
 661         struct tls_offload_context_tx *offload_ctx;
 662         struct tls_crypto_info *crypto_info;
 663         struct net_device *netdev;
 664         char *iv, *rec_seq;
 665         struct sk_buff *skb;
 666         int rc = -EINVAL;
 667         __be64 rcd_sn;
 668
 669         if (!ctx)
 670                 goto out;
 671
 672         if (ctx->priv_ctx_tx) {
 673                 rc = -EEXIST;
 674                 goto out;
 675         }
 676
 677         start_marker_record = kmalloc(sizeof(*start_marker_record), GFP_KERNEL);
 678         if (!start_marker_record) {
 679                 rc = -ENOMEM;
 680                 goto out;
 681         }
 682
 683         offload_ctx = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_TX, GFP_KERNEL);
 684         if (!offload_ctx) {
 685                 rc = -ENOMEM;
 686                 goto free_marker_record;
 687         }
 688
 689         crypto_info = &ctx->crypto_send.info;
 690         switch (crypto_info->cipher_type) {
 691         case TLS_CIPHER_AES_GCM_128:
 692                 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
 693                 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
 694                 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
 695                 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
 696                 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
 697                 rec_seq =
 698                  ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
 699                 break;
 700         default:
 701                 rc = -EINVAL;
 702                 goto free_offload_ctx;
 703         }
 704
 705         ctx->tx.prepend_size = TLS_HEADER_SIZE + nonce_size;
 706         ctx->tx.tag_size = tag_size;
 707         ctx->tx.overhead_size = ctx->tx.prepend_size + ctx->tx.tag_size;
 708         ctx->tx.iv_size = iv_size;
 709         ctx->tx.iv = kmalloc(iv_size + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
 710                              GFP_KERNEL);
 711         if (!ctx->tx.iv) {
 712                 rc = -ENOMEM;
 713                 goto free_offload_ctx;
 714         }
 715
 716         memcpy(ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv, iv_size);
 717
 718         ctx->tx.rec_seq_size = rec_seq_size;
 719         ctx->tx.rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
 720         if (!ctx->tx.rec_seq) {
 721                 rc = -ENOMEM;
 722                 goto free_iv;
 723         }
 724
 725         rc = tls_sw_fallback_init(sk, offload_ctx, crypto_info);
 726         if (rc)
 727                 goto free_rec_seq;
 728
 729         /* start at rec_seq - 1 to account for the start marker record */
 730         memcpy(&rcd_sn, ctx->tx.rec_seq, sizeof(rcd_sn));
 731         offload_ctx->unacked_record_sn = be64_to_cpu(rcd_sn) - 1;
 732
 733         start_marker_record->end_seq = tcp_sk(sk)->write_seq;
 734         start_marker_record->len = 0;
 735         start_marker_record->num_frags = 0;
 736
 737         INIT_LIST_HEAD(&offload_ctx->records_list);
 738         list_add_tail(&start_marker_record->list, &offload_ctx->records_list);
 739         spin_lock_init(&offload_ctx->lock);
 740         sg_init_table(offload_ctx->sg_tx_data,
 741                       ARRAY_SIZE(offload_ctx->sg_tx_data));
 742
 743         clean_acked_data_enable(inet_csk(sk), &tls_icsk_clean_acked);
 744         ctx->push_pending_record = tls_device_push_pending_record;
 745
 746         /* TLS offload is greatly simplified if we don't send
 747          * SKBs where only part of the payload needs to be encrypted.
 748          * So mark the last skb in the write queue as end of record.
 749          */
 750         skb = tcp_write_queue_tail(sk);
 751         if (skb)
 752                 TCP_SKB_CB(skb)->eor = 1;
 753
 754         /* We support starting offload on multiple sockets
 755          * concurrently, so we only need a read lock here.
 756          * This lock must precede get_netdev_for_sock to prevent races between
 757          * NETDEV_DOWN and setsockopt.
 758          */
 759         down_read(&device_offload_lock);
 760         netdev = get_netdev_for_sock(sk);
 761         if (!netdev) {
 762                 pr_err_ratelimited("%s: netdev not found\n", __func__);
 763                 rc = -EINVAL;
 764                 goto release_lock;
 765         }
 766
 767         if (!(netdev->features & NETIF_F_HW_TLS_TX)) {
 768                 rc = -ENOTSUPP;
 769                 goto release_netdev;
 770         }
 771
 772         /* Avoid offloading if the device is down
 773          * We don't want to offload new flows after
 774          * the NETDEV_DOWN event
 775          */
 776         if (!(netdev->flags & IFF_UP)) {
 777                 rc = -EINVAL;
 778                 goto release_netdev;
 779         }
 780
 781         ctx->priv_ctx_tx = offload_ctx;
 782         rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_TX,
 783                                              &ctx->crypto_send.info,
 784                                              tcp_sk(sk)->write_seq);
 785         if (rc)
 786                 goto release_netdev;
 787
 788         tls_device_attach(ctx, sk, netdev);
 789
 790         /* following this assignment tls_is_sk_tx_device_offloaded
 791          * will return true and the context might be accessed
 792          * by the netdev's xmit function.
 793          */
 794         smp_store_release(&sk->sk_validate_xmit_skb, tls_validate_xmit_skb);
 795         dev_put(netdev);
 796         up_read(&device_offload_lock);
 797         goto out;
 798
 799 release_netdev:
 800         dev_put(netdev);
 801 release_lock:
 802         up_read(&device_offload_lock);
 803         clean_acked_data_disable(inet_csk(sk));
 804         crypto_free_aead(offload_ctx->aead_send);
 805 free_rec_seq:
 806         kfree(ctx->tx.rec_seq);
 807 free_iv:
 808         kfree(ctx->tx.iv);
 809 free_offload_ctx:
 810         kfree(offload_ctx);
 811         ctx->priv_ctx_tx = NULL;
 812 free_marker_record:
 813         kfree(start_marker_record);
 814 out:
 815         return rc;
 816 }
 817
 818 int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx)
 819 {
 820         struct tls_offload_context_rx *context;
 821         struct net_device *netdev;
 822         int rc = 0;
 823
 824         /* We support starting offload on multiple sockets
 825          * concurrently, so we only need a read lock here.
 826          * This lock must precede get_netdev_for_sock to prevent races between
 827          * NETDEV_DOWN and setsockopt.
 828          */
 829         down_read(&device_offload_lock);
 830         netdev = get_netdev_for_sock(sk);
 831         if (!netdev) {
 832                 pr_err_ratelimited("%s: netdev not found\n", __func__);
 833                 rc = -EINVAL;
 834                 goto release_lock;
 835         }
 836
 837         if (!(netdev->features & NETIF_F_HW_TLS_RX)) {
 838                 pr_err_ratelimited("%s: netdev %s with no TLS offload\n",
 839                                    __func__, netdev->name);
 840                 rc = -ENOTSUPP;
 841                 goto release_netdev;
 842         }
 843
 844         /* Avoid offloading if the device is down
 845          * We don't want to offload new flows after
 846          * the NETDEV_DOWN event
 847          */
 848         if (!(netdev->flags & IFF_UP)) {
 849                 rc = -EINVAL;
 850                 goto release_netdev;
 851         }
 852
 853         context = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_RX, GFP_KERNEL);
 854         if (!context) {
 855                 rc = -ENOMEM;
 856                 goto release_netdev;
 857         }
 858
 859         ctx->priv_ctx_rx = context;
 860         rc = tls_set_sw_offload(sk, ctx, 0);
 861         if (rc)
 862                 goto release_ctx;
 863
 864         rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_RX,
 865                                              &ctx->crypto_recv.info,
 866                                              tcp_sk(sk)->copied_seq);
 867         if (rc) {
 868                 pr_err_ratelimited("%s: The netdev has refused to offload this socket\n",
 869                                    __func__);
 870                 goto free_sw_resources;
 871         }
 872
 873         tls_device_attach(ctx, sk, netdev);
 874         goto release_netdev;
 875
 876 free_sw_resources:
 877         tls_sw_free_resources_rx(sk);
 878 release_ctx:
 879         ctx->priv_ctx_rx = NULL;
 880 release_netdev:
 881         dev_put(netdev);
 882 release_lock:
 883         up_read(&device_offload_lock);
 884         return rc;
 885 }
 886
 887 void tls_device_offload_cleanup_rx(struct sock *sk)
 888 {
 889         struct tls_context *tls_ctx = tls_get_ctx(sk);
 890         struct net_device *netdev;
 891
 892         down_read(&device_offload_lock);
 893         netdev = tls_ctx->netdev;
 894         if (!netdev)
 895                 goto out;
 896
 897         if (!(netdev->features & NETIF_F_HW_TLS_RX)) {
 898                 pr_err_ratelimited("%s: device is missing NETIF_F_HW_TLS_RX cap\n",
 899                                    __func__);
 900                 goto out;
 901         }
 902
 903         netdev->tlsdev_ops->tls_dev_del(netdev, tls_ctx,
 904                                         TLS_OFFLOAD_CTX_DIR_RX);
 905
 906         if (tls_ctx->tx_conf != TLS_HW) {
 907                 dev_put(netdev);
 908                 tls_ctx->netdev = NULL;
 909         }
 910 out:
 911         up_read(&device_offload_lock);
 912         kfree(tls_ctx->rx.rec_seq);
 913         kfree(tls_ctx->rx.iv);
 914         tls_sw_release_resources_rx(sk);
 915 }
 916
 917 static int tls_device_down(struct net_device *netdev)
 918 {
 919         struct tls_context *ctx, *tmp;
 920         unsigned long flags;
 921         LIST_HEAD(list);
 922
 923         /* Request a write lock to block new offload attempts */
 924         down_write(&device_offload_lock);
 925
 926         spin_lock_irqsave(&tls_device_lock, flags);
 927         list_for_each_entry_safe(ctx, tmp, &tls_device_list, list) {
 928                 if (ctx->netdev != netdev ||
 929                     !refcount_inc_not_zero(&ctx->refcount))
 930                         continue;
 931
 932                 list_move(&ctx->list, &list);
 933         }
 934         spin_unlock_irqrestore(&tls_device_lock, flags);
 935
 936         list_for_each_entry_safe(ctx, tmp, &list, list) {
 937                 if (ctx->tx_conf == TLS_HW)
 938                         netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
 939                                                         TLS_OFFLOAD_CTX_DIR_TX);
 940                 if (ctx->rx_conf == TLS_HW)
 941                         netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
 942                                                         TLS_OFFLOAD_CTX_DIR_RX);
 943                 ctx->netdev = NULL;
 944                 dev_put(netdev);
 945                 list_del_init(&ctx->list);
 946
 947                 if (refcount_dec_and_test(&ctx->refcount))
 948                         tls_device_free_ctx(ctx);
 949         }
 950
 951         up_write(&device_offload_lock);
 952
 953         flush_work(&tls_device_gc_work);
 954
 955         return NOTIFY_DONE;
 956 }
 957
 958 static int tls_dev_event(struct notifier_block *this, unsigned long event,
 959                          void *ptr)
 960 {
 961         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 962
 963         if (!(dev->features & (NETIF_F_HW_TLS_RX | NETIF_F_HW_TLS_TX)))
 964                 return NOTIFY_DONE;
 965
 966         switch (event) {
 967         case NETDEV_REGISTER:
 968         case NETDEV_FEAT_CHANGE:
 969                 if ((dev->features & NETIF_F_HW_TLS_RX) &&
 970                     !dev->tlsdev_ops->tls_dev_resync_rx)
 971                         return NOTIFY_BAD;
 972
 973                 if  (dev->tlsdev_ops &&
 974                      dev->tlsdev_ops->tls_dev_add &&
 975                      dev->tlsdev_ops->tls_dev_del)
 976                         return NOTIFY_DONE;
 977                 else
 978                         return NOTIFY_BAD;
 979         case NETDEV_DOWN:
 980                 return tls_device_down(dev);
 981         }
 982         return NOTIFY_DONE;
 983 }
 984
 985 static struct notifier_block tls_dev_notifier = {
 986         .notifier_call  = tls_dev_event,
 987 };
 988
 989 void __init tls_device_init(void)
 990 {
 991         register_netdevice_notifier(&tls_dev_notifier);
 992 }
 993
 994 void __exit tls_device_cleanup(void)
 995 {
 996         unregister_netdevice_notifier(&tls_dev_notifier);
 997         flush_work(&tls_device_gc_work);
 998 }