drivers/net/tap.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 #include <linux/etherdevice.h>
   3 #include <linux/if_tap.h>
   4 #include <linux/if_vlan.h>
   5 #include <linux/interrupt.h>
   6 #include <linux/nsproxy.h>
   7 #include <linux/compat.h>
   8 #include <linux/if_tun.h>
   9 #include <linux/module.h>
  10 #include <linux/skbuff.h>
  11 #include <linux/cache.h>
  12 #include <linux/sched/signal.h>
  13 #include <linux/types.h>
  14 #include <linux/slab.h>
  15 #include <linux/wait.h>
  16 #include <linux/cdev.h>
  17 #include <linux/idr.h>
  18 #include <linux/fs.h>
  19 #include <linux/uio.h>
  20
  21 #include <net/gso.h>
  22 #include <net/net_namespace.h>
  23 #include <net/rtnetlink.h>
  24 #include <net/sock.h>
  25 #include <linux/virtio_net.h>
  26 #include <linux/skb_array.h>
  27
  28 #define TAP_IFFEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE)
  29
  30 #define TAP_VNET_LE 0x80000000
  31 #define TAP_VNET_BE 0x40000000
  32
  33 #ifdef CONFIG_TUN_VNET_CROSS_LE
  34 static inline bool tap_legacy_is_little_endian(struct tap_queue *q)
  35 {
  36         return q->flags & TAP_VNET_BE ? false :
  37                 virtio_legacy_is_little_endian();
  38 }
  39
  40 static long tap_get_vnet_be(struct tap_queue *q, int __user *sp)
  41 {
  42         int s = !!(q->flags & TAP_VNET_BE);
  43
  44         if (put_user(s, sp))
  45                 return -EFAULT;
  46
  47         return 0;
  48 }
  49
  50 static long tap_set_vnet_be(struct tap_queue *q, int __user *sp)
  51 {
  52         int s;
  53
  54         if (get_user(s, sp))
  55                 return -EFAULT;
  56
  57         if (s)
  58                 q->flags |= TAP_VNET_BE;
  59         else
  60                 q->flags &= ~TAP_VNET_BE;
  61
  62         return 0;
  63 }
  64 #else
  65 static inline bool tap_legacy_is_little_endian(struct tap_queue *q)
  66 {
  67         return virtio_legacy_is_little_endian();
  68 }
  69
  70 static long tap_get_vnet_be(struct tap_queue *q, int __user *argp)
  71 {
  72         return -EINVAL;
  73 }
  74
  75 static long tap_set_vnet_be(struct tap_queue *q, int __user *argp)
  76 {
  77         return -EINVAL;
  78 }
  79 #endif /* CONFIG_TUN_VNET_CROSS_LE */
  80
  81 static inline bool tap_is_little_endian(struct tap_queue *q)
  82 {
  83         return q->flags & TAP_VNET_LE ||
  84                 tap_legacy_is_little_endian(q);
  85 }
  86
  87 static inline u16 tap16_to_cpu(struct tap_queue *q, __virtio16 val)
  88 {
  89         return __virtio16_to_cpu(tap_is_little_endian(q), val);
  90 }
  91
  92 static inline __virtio16 cpu_to_tap16(struct tap_queue *q, u16 val)
  93 {
  94         return __cpu_to_virtio16(tap_is_little_endian(q), val);
  95 }
  96
  97 static struct proto tap_proto = {
  98         .name = "tap",
  99         .owner = THIS_MODULE,
 100         .obj_size = sizeof(struct tap_queue),
 101 };
 102
 103 #define TAP_NUM_DEVS (1U << MINORBITS)
 104
 105 static LIST_HEAD(major_list);
 106
 107 struct major_info {
 108         struct rcu_head rcu;
 109         dev_t major;
 110         struct idr minor_idr;
 111         spinlock_t minor_lock;
 112         const char *device_name;
 113         struct list_head next;
 114 };
 115
 116 #define GOODCOPY_LEN 128
 117
 118 static const struct proto_ops tap_socket_ops;
 119
 120 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
 121 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST)
 122
 123 static struct tap_dev *tap_dev_get_rcu(const struct net_device *dev)
 124 {
 125         return rcu_dereference(dev->rx_handler_data);
 126 }
 127
 128 /*
 129  * RCU usage:
 130  * The tap_queue and the macvlan_dev are loosely coupled, the
 131  * pointers from one to the other can only be read while rcu_read_lock
 132  * or rtnl is held.
 133  *
 134  * Both the file and the macvlan_dev hold a reference on the tap_queue
 135  * through sock_hold(&q->sk). When the macvlan_dev goes away first,
 136  * q->vlan becomes inaccessible. When the files gets closed,
 137  * tap_get_queue() fails.
 138  *
 139  * There may still be references to the struct sock inside of the
 140  * queue from outbound SKBs, but these never reference back to the
 141  * file or the dev. The data structure is freed through __sk_free
 142  * when both our references and any pending SKBs are gone.
 143  */
 144
 145 static int tap_enable_queue(struct tap_dev *tap, struct file *file,
 146                             struct tap_queue *q)
 147 {
 148         int err = -EINVAL;
 149
 150         ASSERT_RTNL();
 151
 152         if (q->enabled)
 153                 goto out;
 154
 155         err = 0;
 156         rcu_assign_pointer(tap->taps[tap->numvtaps], q);
 157         q->queue_index = tap->numvtaps;
 158         q->enabled = true;
 159
 160         tap->numvtaps++;
 161 out:
 162         return err;
 163 }
 164
 165 /* Requires RTNL */
 166 static int tap_set_queue(struct tap_dev *tap, struct file *file,
 167                          struct tap_queue *q)
 168 {
 169         if (tap->numqueues == MAX_TAP_QUEUES)
 170                 return -EBUSY;
 171
 172         rcu_assign_pointer(q->tap, tap);
 173         rcu_assign_pointer(tap->taps[tap->numvtaps], q);
 174         sock_hold(&q->sk);
 175
 176         q->file = file;
 177         q->queue_index = tap->numvtaps;
 178         q->enabled = true;
 179         file->private_data = q;
 180         list_add_tail(&q->next, &tap->queue_list);
 181
 182         tap->numvtaps++;
 183         tap->numqueues++;
 184
 185         return 0;
 186 }
 187
 188 static int tap_disable_queue(struct tap_queue *q)
 189 {
 190         struct tap_dev *tap;
 191         struct tap_queue *nq;
 192
 193         ASSERT_RTNL();
 194         if (!q->enabled)
 195                 return -EINVAL;
 196
 197         tap = rtnl_dereference(q->tap);
 198
 199         if (tap) {
 200                 int index = q->queue_index;
 201                 BUG_ON(index >= tap->numvtaps);
 202                 nq = rtnl_dereference(tap->taps[tap->numvtaps - 1]);
 203                 nq->queue_index = index;
 204
 205                 rcu_assign_pointer(tap->taps[index], nq);
 206                 RCU_INIT_POINTER(tap->taps[tap->numvtaps - 1], NULL);
 207                 q->enabled = false;
 208
 209                 tap->numvtaps--;
 210         }
 211
 212         return 0;
 213 }
 214
 215 /*
 216  * The file owning the queue got closed, give up both
 217  * the reference that the files holds as well as the
 218  * one from the macvlan_dev if that still exists.
 219  *
 220  * Using the spinlock makes sure that we don't get
 221  * to the queue again after destroying it.
 222  */
 223 static void tap_put_queue(struct tap_queue *q)
 224 {
 225         struct tap_dev *tap;
 226
 227         rtnl_lock();
 228         tap = rtnl_dereference(q->tap);
 229
 230         if (tap) {
 231                 if (q->enabled)
 232                         BUG_ON(tap_disable_queue(q));
 233
 234                 tap->numqueues--;
 235                 RCU_INIT_POINTER(q->tap, NULL);
 236                 sock_put(&q->sk);
 237                 list_del_init(&q->next);
 238         }
 239
 240         rtnl_unlock();
 241
 242         synchronize_rcu();
 243         sock_put(&q->sk);
 244 }
 245
 246 /*
 247  * Select a queue based on the rxq of the device on which this packet
 248  * arrived. If the incoming device is not mq, calculate a flow hash
 249  * to select a queue. If all fails, find the first available queue.
 250  * Cache vlan->numvtaps since it can become zero during the execution
 251  * of this function.
 252  */
 253 static struct tap_queue *tap_get_queue(struct tap_dev *tap,
 254                                        struct sk_buff *skb)
 255 {
 256         struct tap_queue *queue = NULL;
 257         /* Access to taps array is protected by rcu, but access to numvtaps
 258          * isn't. Below we use it to lookup a queue, but treat it as a hint
 259          * and validate that the result isn't NULL - in case we are
 260          * racing against queue removal.
 261          */
 262         int numvtaps = READ_ONCE(tap->numvtaps);
 263         __u32 rxq;
 264
 265         if (!numvtaps)
 266                 goto out;
 267
 268         if (numvtaps == 1)
 269                 goto single;
 270
 271         /* Check if we can use flow to select a queue */
 272         rxq = skb_get_hash(skb);
 273         if (rxq) {
 274                 queue = rcu_dereference(tap->taps[rxq % numvtaps]);
 275                 goto out;
 276         }
 277
 278         if (likely(skb_rx_queue_recorded(skb))) {
 279                 rxq = skb_get_rx_queue(skb);
 280
 281                 while (unlikely(rxq >= numvtaps))
 282                         rxq -= numvtaps;
 283
 284                 queue = rcu_dereference(tap->taps[rxq]);
 285                 goto out;
 286         }
 287
 288 single:
 289         queue = rcu_dereference(tap->taps[0]);
 290 out:
 291         return queue;
 292 }
 293
 294 /*
 295  * The net_device is going away, give up the reference
 296  * that it holds on all queues and safely set the pointer
 297  * from the queues to NULL.
 298  */
 299 void tap_del_queues(struct tap_dev *tap)
 300 {
 301         struct tap_queue *q, *tmp;
 302
 303         ASSERT_RTNL();
 304         list_for_each_entry_safe(q, tmp, &tap->queue_list, next) {
 305                 list_del_init(&q->next);
 306                 RCU_INIT_POINTER(q->tap, NULL);
 307                 if (q->enabled)
 308                         tap->numvtaps--;
 309                 tap->numqueues--;
 310                 sock_put(&q->sk);
 311         }
 312         BUG_ON(tap->numvtaps);
 313         BUG_ON(tap->numqueues);
 314         /* guarantee that any future tap_set_queue will fail */
 315         tap->numvtaps = MAX_TAP_QUEUES;
 316 }
 317 EXPORT_SYMBOL_GPL(tap_del_queues);
 318
 319 rx_handler_result_t tap_handle_frame(struct sk_buff **pskb)
 320 {
 321         struct sk_buff *skb = *pskb;
 322         struct net_device *dev = skb->dev;
 323         struct tap_dev *tap;
 324         struct tap_queue *q;
 325         netdev_features_t features = TAP_FEATURES;
 326         enum skb_drop_reason drop_reason;
 327
 328         tap = tap_dev_get_rcu(dev);
 329         if (!tap)
 330                 return RX_HANDLER_PASS;
 331
 332         q = tap_get_queue(tap, skb);
 333         if (!q)
 334                 return RX_HANDLER_PASS;
 335
 336         skb_push(skb, ETH_HLEN);
 337
 338         /* Apply the forward feature mask so that we perform segmentation
 339          * according to users wishes.  This only works if VNET_HDR is
 340          * enabled.
 341          */
 342         if (q->flags & IFF_VNET_HDR)
 343                 features |= tap->tap_features;
 344         if (netif_needs_gso(skb, features)) {
 345                 struct sk_buff *segs = __skb_gso_segment(skb, features, false);
 346                 struct sk_buff *next;
 347
 348                 if (IS_ERR(segs)) {
 349                         drop_reason = SKB_DROP_REASON_SKB_GSO_SEG;
 350                         goto drop;
 351                 }
 352
 353                 if (!segs) {
 354                         if (ptr_ring_produce(&q->ring, skb)) {
 355                                 drop_reason = SKB_DROP_REASON_FULL_RING;
 356                                 goto drop;
 357                         }
 358                         goto wake_up;
 359                 }
 360
 361                 consume_skb(skb);
 362                 skb_list_walk_safe(segs, skb, next) {
 363                         skb_mark_not_on_list(skb);
 364                         if (ptr_ring_produce(&q->ring, skb)) {
 365                                 drop_reason = SKB_DROP_REASON_FULL_RING;
 366                                 kfree_skb_reason(skb, drop_reason);
 367                                 kfree_skb_list_reason(next, drop_reason);
 368                                 break;
 369                         }
 370                 }
 371         } else {
 372                 /* If we receive a partial checksum and the tap side
 373                  * doesn't support checksum offload, compute the checksum.
 374                  * Note: it doesn't matter which checksum feature to
 375                  *        check, we either support them all or none.
 376                  */
 377                 if (skb->ip_summed == CHECKSUM_PARTIAL &&
 378                     !(features & NETIF_F_CSUM_MASK) &&
 379                     skb_checksum_help(skb)) {
 380                         drop_reason = SKB_DROP_REASON_SKB_CSUM;
 381                         goto drop;
 382                 }
 383                 if (ptr_ring_produce(&q->ring, skb)) {
 384                         drop_reason = SKB_DROP_REASON_FULL_RING;
 385                         goto drop;
 386                 }
 387         }
 388
 389 wake_up:
 390         wake_up_interruptible_poll(sk_sleep(&q->sk), EPOLLIN | EPOLLRDNORM | EPOLLRDBAND);
 391         return RX_HANDLER_CONSUMED;
 392
 393 drop:
 394         /* Count errors/drops only here, thus don't care about args. */
 395         if (tap->count_rx_dropped)
 396                 tap->count_rx_dropped(tap);
 397         kfree_skb_reason(skb, drop_reason);
 398         return RX_HANDLER_CONSUMED;
 399 }
 400 EXPORT_SYMBOL_GPL(tap_handle_frame);
 401
 402 static struct major_info *tap_get_major(int major)
 403 {
 404         struct major_info *tap_major;
 405
 406         list_for_each_entry_rcu(tap_major, &major_list, next) {
 407                 if (tap_major->major == major)
 408                         return tap_major;
 409         }
 410
 411         return NULL;
 412 }
 413
 414 int tap_get_minor(dev_t major, struct tap_dev *tap)
 415 {
 416         int retval = -ENOMEM;
 417         struct major_info *tap_major;
 418
 419         rcu_read_lock();
 420         tap_major = tap_get_major(MAJOR(major));
 421         if (!tap_major) {
 422                 retval = -EINVAL;
 423                 goto unlock;
 424         }
 425
 426         spin_lock(&tap_major->minor_lock);
 427         retval = idr_alloc(&tap_major->minor_idr, tap, 1, TAP_NUM_DEVS, GFP_ATOMIC);
 428         if (retval >= 0) {
 429                 tap->minor = retval;
 430         } else if (retval == -ENOSPC) {
 431                 netdev_err(tap->dev, "Too many tap devices\n");
 432                 retval = -EINVAL;
 433         }
 434         spin_unlock(&tap_major->minor_lock);
 435
 436 unlock:
 437         rcu_read_unlock();
 438         return retval < 0 ? retval : 0;
 439 }
 440 EXPORT_SYMBOL_GPL(tap_get_minor);
 441
 442 void tap_free_minor(dev_t major, struct tap_dev *tap)
 443 {
 444         struct major_info *tap_major;
 445
 446         rcu_read_lock();
 447         tap_major = tap_get_major(MAJOR(major));
 448         if (!tap_major) {
 449                 goto unlock;
 450         }
 451
 452         spin_lock(&tap_major->minor_lock);
 453         if (tap->minor) {
 454                 idr_remove(&tap_major->minor_idr, tap->minor);
 455                 tap->minor = 0;
 456         }
 457         spin_unlock(&tap_major->minor_lock);
 458
 459 unlock:
 460         rcu_read_unlock();
 461 }
 462 EXPORT_SYMBOL_GPL(tap_free_minor);
 463
 464 static struct tap_dev *dev_get_by_tap_file(int major, int minor)
 465 {
 466         struct net_device *dev = NULL;
 467         struct tap_dev *tap;
 468         struct major_info *tap_major;
 469
 470         rcu_read_lock();
 471         tap_major = tap_get_major(major);
 472         if (!tap_major) {
 473                 tap = NULL;
 474                 goto unlock;
 475         }
 476
 477         spin_lock(&tap_major->minor_lock);
 478         tap = idr_find(&tap_major->minor_idr, minor);
 479         if (tap) {
 480                 dev = tap->dev;
 481                 dev_hold(dev);
 482         }
 483         spin_unlock(&tap_major->minor_lock);
 484
 485 unlock:
 486         rcu_read_unlock();
 487         return tap;
 488 }
 489
 490 static void tap_sock_write_space(struct sock *sk)
 491 {
 492         wait_queue_head_t *wqueue;
 493
 494         if (!sock_writeable(sk) ||
 495             !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
 496                 return;
 497
 498         wqueue = sk_sleep(sk);
 499         if (wqueue && waitqueue_active(wqueue))
 500                 wake_up_interruptible_poll(wqueue, EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
 501 }
 502
 503 static void tap_sock_destruct(struct sock *sk)
 504 {
 505         struct tap_queue *q = container_of(sk, struct tap_queue, sk);
 506
 507         ptr_ring_cleanup(&q->ring, __skb_array_destroy_skb);
 508 }
 509
 510 static int tap_open(struct inode *inode, struct file *file)
 511 {
 512         struct net *net = current->nsproxy->net_ns;
 513         struct tap_dev *tap;
 514         struct tap_queue *q;
 515         int err = -ENODEV;
 516
 517         rtnl_lock();
 518         tap = dev_get_by_tap_file(imajor(inode), iminor(inode));
 519         if (!tap)
 520                 goto err;
 521
 522         err = -ENOMEM;
 523         q = (struct tap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
 524                                              &tap_proto, 0);
 525         if (!q)
 526                 goto err;
 527         if (ptr_ring_init(&q->ring, tap->dev->tx_queue_len, GFP_KERNEL)) {
 528                 sk_free(&q->sk);
 529                 goto err;
 530         }
 531
 532         init_waitqueue_head(&q->sock.wq.wait);
 533         q->sock.type = SOCK_RAW;
 534         q->sock.state = SS_CONNECTED;
 535         q->sock.file = file;
 536         q->sock.ops = &tap_socket_ops;
 537         sock_init_data_uid(&q->sock, &q->sk, inode->i_uid);
 538         q->sk.sk_write_space = tap_sock_write_space;
 539         q->sk.sk_destruct = tap_sock_destruct;
 540         q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
 541         q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
 542
 543         /*
 544          * so far only KVM virtio_net uses tap, enable zero copy between
 545          * guest kernel and host kernel when lower device supports zerocopy
 546          *
 547          * The macvlan supports zerocopy iff the lower device supports zero
 548          * copy so we don't have to look at the lower device directly.
 549          */
 550         if ((tap->dev->features & NETIF_F_HIGHDMA) && (tap->dev->features & NETIF_F_SG))
 551                 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
 552
 553         err = tap_set_queue(tap, file, q);
 554         if (err) {
 555                 /* tap_sock_destruct() will take care of freeing ptr_ring */
 556                 goto err_put;
 557         }
 558
 559         /* tap groks IOCB_NOWAIT just fine, mark it as such */
 560         file->f_mode |= FMODE_NOWAIT;
 561
 562         dev_put(tap->dev);
 563
 564         rtnl_unlock();
 565         return err;
 566
 567 err_put:
 568         sock_put(&q->sk);
 569 err:
 570         if (tap)
 571                 dev_put(tap->dev);
 572
 573         rtnl_unlock();
 574         return err;
 575 }
 576
 577 static int tap_release(struct inode *inode, struct file *file)
 578 {
 579         struct tap_queue *q = file->private_data;
 580         tap_put_queue(q);
 581         return 0;
 582 }
 583
 584 static __poll_t tap_poll(struct file *file, poll_table *wait)
 585 {
 586         struct tap_queue *q = file->private_data;
 587         __poll_t mask = EPOLLERR;
 588
 589         if (!q)
 590                 goto out;
 591
 592         mask = 0;
 593         poll_wait(file, &q->sock.wq.wait, wait);
 594
 595         if (!ptr_ring_empty(&q->ring))
 596                 mask |= EPOLLIN | EPOLLRDNORM;
 597
 598         if (sock_writeable(&q->sk) ||
 599             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) &&
 600              sock_writeable(&q->sk)))
 601                 mask |= EPOLLOUT | EPOLLWRNORM;
 602
 603 out:
 604         return mask;
 605 }
 606
 607 static inline struct sk_buff *tap_alloc_skb(struct sock *sk, size_t prepad,
 608                                             size_t len, size_t linear,
 609                                                 int noblock, int *err)
 610 {
 611         struct sk_buff *skb;
 612
 613         /* Under a page?  Don't bother with paged skb. */
 614         if (prepad + len < PAGE_SIZE || !linear)
 615                 linear = len;
 616
 617         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 618                                    err, 0);
 619         if (!skb)
 620                 return NULL;
 621
 622         skb_reserve(skb, prepad);
 623         skb_put(skb, linear);
 624         skb->data_len = len - linear;
 625         skb->len += len - linear;
 626
 627         return skb;
 628 }
 629
 630 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */
 631 #define TAP_RESERVE HH_DATA_OFF(ETH_HLEN)
 632
 633 /* Get packet from user space buffer */
 634 static ssize_t tap_get_user(struct tap_queue *q, void *msg_control,
 635                             struct iov_iter *from, int noblock)
 636 {
 637         int good_linear = SKB_MAX_HEAD(TAP_RESERVE);
 638         struct sk_buff *skb;
 639         struct tap_dev *tap;
 640         unsigned long total_len = iov_iter_count(from);
 641         unsigned long len = total_len;
 642         int err;
 643         struct virtio_net_hdr vnet_hdr = { 0 };
 644         int vnet_hdr_len = 0;
 645         int copylen = 0;
 646         int depth;
 647         bool zerocopy = false;
 648         size_t linear;
 649         enum skb_drop_reason drop_reason;
 650
 651         if (q->flags & IFF_VNET_HDR) {
 652                 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
 653
 654                 err = -EINVAL;
 655                 if (len < vnet_hdr_len)
 656                         goto err;
 657                 len -= vnet_hdr_len;
 658
 659                 err = -EFAULT;
 660                 if (!copy_from_iter_full(&vnet_hdr, sizeof(vnet_hdr), from))
 661                         goto err;
 662                 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr));
 663                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
 664                      tap16_to_cpu(q, vnet_hdr.csum_start) +
 665                      tap16_to_cpu(q, vnet_hdr.csum_offset) + 2 >
 666                              tap16_to_cpu(q, vnet_hdr.hdr_len))
 667                         vnet_hdr.hdr_len = cpu_to_tap16(q,
 668                                  tap16_to_cpu(q, vnet_hdr.csum_start) +
 669                                  tap16_to_cpu(q, vnet_hdr.csum_offset) + 2);
 670                 err = -EINVAL;
 671                 if (tap16_to_cpu(q, vnet_hdr.hdr_len) > len)
 672                         goto err;
 673         }
 674
 675         err = -EINVAL;
 676         if (unlikely(len < ETH_HLEN))
 677                 goto err;
 678
 679         if (msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
 680                 struct iov_iter i;
 681
 682                 copylen = vnet_hdr.hdr_len ?
 683                         tap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN;
 684                 if (copylen > good_linear)
 685                         copylen = good_linear;
 686                 else if (copylen < ETH_HLEN)
 687                         copylen = ETH_HLEN;
 688                 linear = copylen;
 689                 i = *from;
 690                 iov_iter_advance(&i, copylen);
 691                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
 692                         zerocopy = true;
 693         }
 694
 695         if (!zerocopy) {
 696                 copylen = len;
 697                 linear = tap16_to_cpu(q, vnet_hdr.hdr_len);
 698                 if (linear > good_linear)
 699                         linear = good_linear;
 700                 else if (linear < ETH_HLEN)
 701                         linear = ETH_HLEN;
 702         }
 703
 704         skb = tap_alloc_skb(&q->sk, TAP_RESERVE, copylen,
 705                             linear, noblock, &err);
 706         if (!skb)
 707                 goto err;
 708
 709         if (zerocopy)
 710                 err = zerocopy_sg_from_iter(skb, from);
 711         else
 712                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
 713
 714         if (err) {
 715                 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
 716                 goto err_kfree;
 717         }
 718
 719         skb_set_network_header(skb, ETH_HLEN);
 720         skb_reset_mac_header(skb);
 721         skb->protocol = eth_hdr(skb)->h_proto;
 722
 723         rcu_read_lock();
 724         tap = rcu_dereference(q->tap);
 725         if (!tap) {
 726                 kfree_skb(skb);
 727                 rcu_read_unlock();
 728                 return total_len;
 729         }
 730         skb->dev = tap->dev;
 731
 732         if (vnet_hdr_len) {
 733                 err = virtio_net_hdr_to_skb(skb, &vnet_hdr,
 734                                             tap_is_little_endian(q));
 735                 if (err) {
 736                         rcu_read_unlock();
 737                         drop_reason = SKB_DROP_REASON_DEV_HDR;
 738                         goto err_kfree;
 739                 }
 740         }
 741
 742         skb_probe_transport_header(skb);
 743
 744         /* Move network header to the right position for VLAN tagged packets */
 745         if (eth_type_vlan(skb->protocol) &&
 746             vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0)
 747                 skb_set_network_header(skb, depth);
 748
 749         /* copy skb_ubuf_info for callback when skb has no error */
 750         if (zerocopy) {
 751                 skb_zcopy_init(skb, msg_control);
 752         } else if (msg_control) {
 753                 struct ubuf_info *uarg = msg_control;
 754                 uarg->callback(NULL, uarg, false);
 755         }
 756
 757         dev_queue_xmit(skb);
 758         rcu_read_unlock();
 759         return total_len;
 760
 761 err_kfree:
 762         kfree_skb_reason(skb, drop_reason);
 763
 764 err:
 765         rcu_read_lock();
 766         tap = rcu_dereference(q->tap);
 767         if (tap && tap->count_tx_dropped)
 768                 tap->count_tx_dropped(tap);
 769         rcu_read_unlock();
 770
 771         return err;
 772 }
 773
 774 static ssize_t tap_write_iter(struct kiocb *iocb, struct iov_iter *from)
 775 {
 776         struct file *file = iocb->ki_filp;
 777         struct tap_queue *q = file->private_data;
 778         int noblock = 0;
 779
 780         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
 781                 noblock = 1;
 782
 783         return tap_get_user(q, NULL, from, noblock);
 784 }
 785
 786 /* Put packet to the user space buffer */
 787 static ssize_t tap_put_user(struct tap_queue *q,
 788                             const struct sk_buff *skb,
 789                             struct iov_iter *iter)
 790 {
 791         int ret;
 792         int vnet_hdr_len = 0;
 793         int vlan_offset = 0;
 794         int total;
 795
 796         if (q->flags & IFF_VNET_HDR) {
 797                 int vlan_hlen = skb_vlan_tag_present(skb) ? VLAN_HLEN : 0;
 798                 struct virtio_net_hdr vnet_hdr;
 799
 800                 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
 801                 if (iov_iter_count(iter) < vnet_hdr_len)
 802                         return -EINVAL;
 803
 804                 if (virtio_net_hdr_from_skb(skb, &vnet_hdr,
 805                                             tap_is_little_endian(q), true,
 806                                             vlan_hlen))
 807                         BUG();
 808
 809                 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) !=
 810                     sizeof(vnet_hdr))
 811                         return -EFAULT;
 812
 813                 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr));
 814         }
 815         total = vnet_hdr_len;
 816         total += skb->len;
 817
 818         if (skb_vlan_tag_present(skb)) {
 819                 struct {
 820                         __be16 h_vlan_proto;
 821                         __be16 h_vlan_TCI;
 822                 } veth;
 823                 veth.h_vlan_proto = skb->vlan_proto;
 824                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
 825
 826                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
 827                 total += VLAN_HLEN;
 828
 829                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
 830                 if (ret || !iov_iter_count(iter))
 831                         goto done;
 832
 833                 ret = copy_to_iter(&veth, sizeof(veth), iter);
 834                 if (ret != sizeof(veth) || !iov_iter_count(iter))
 835                         goto done;
 836         }
 837
 838         ret = skb_copy_datagram_iter(skb, vlan_offset, iter,
 839                                      skb->len - vlan_offset);
 840
 841 done:
 842         return ret ? ret : total;
 843 }
 844
 845 static ssize_t tap_do_read(struct tap_queue *q,
 846                            struct iov_iter *to,
 847                            int noblock, struct sk_buff *skb)
 848 {
 849         DEFINE_WAIT(wait);
 850         ssize_t ret = 0;
 851
 852         if (!iov_iter_count(to)) {
 853                 kfree_skb(skb);
 854                 return 0;
 855         }
 856
 857         if (skb)
 858                 goto put;
 859
 860         while (1) {
 861                 if (!noblock)
 862                         prepare_to_wait(sk_sleep(&q->sk), &wait,
 863                                         TASK_INTERRUPTIBLE);
 864
 865                 /* Read frames from the queue */
 866                 skb = ptr_ring_consume(&q->ring);
 867                 if (skb)
 868                         break;
 869                 if (noblock) {
 870                         ret = -EAGAIN;
 871                         break;
 872                 }
 873                 if (signal_pending(current)) {
 874                         ret = -ERESTARTSYS;
 875                         break;
 876                 }
 877                 /* Nothing to read, let's sleep */
 878                 schedule();
 879         }
 880         if (!noblock)
 881                 finish_wait(sk_sleep(&q->sk), &wait);
 882
 883 put:
 884         if (skb) {
 885                 ret = tap_put_user(q, skb, to);
 886                 if (unlikely(ret < 0))
 887                         kfree_skb(skb);
 888                 else
 889                         consume_skb(skb);
 890         }
 891         return ret;
 892 }
 893
 894 static ssize_t tap_read_iter(struct kiocb *iocb, struct iov_iter *to)
 895 {
 896         struct file *file = iocb->ki_filp;
 897         struct tap_queue *q = file->private_data;
 898         ssize_t len = iov_iter_count(to), ret;
 899         int noblock = 0;
 900
 901         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
 902                 noblock = 1;
 903
 904         ret = tap_do_read(q, to, noblock, NULL);
 905         ret = min_t(ssize_t, ret, len);
 906         if (ret > 0)
 907                 iocb->ki_pos = ret;
 908         return ret;
 909 }
 910
 911 static struct tap_dev *tap_get_tap_dev(struct tap_queue *q)
 912 {
 913         struct tap_dev *tap;
 914
 915         ASSERT_RTNL();
 916         tap = rtnl_dereference(q->tap);
 917         if (tap)
 918                 dev_hold(tap->dev);
 919
 920         return tap;
 921 }
 922
 923 static void tap_put_tap_dev(struct tap_dev *tap)
 924 {
 925         dev_put(tap->dev);
 926 }
 927
 928 static int tap_ioctl_set_queue(struct file *file, unsigned int flags)
 929 {
 930         struct tap_queue *q = file->private_data;
 931         struct tap_dev *tap;
 932         int ret;
 933
 934         tap = tap_get_tap_dev(q);
 935         if (!tap)
 936                 return -EINVAL;
 937
 938         if (flags & IFF_ATTACH_QUEUE)
 939                 ret = tap_enable_queue(tap, file, q);
 940         else if (flags & IFF_DETACH_QUEUE)
 941                 ret = tap_disable_queue(q);
 942         else
 943                 ret = -EINVAL;
 944
 945         tap_put_tap_dev(tap);
 946         return ret;
 947 }
 948
 949 static int set_offload(struct tap_queue *q, unsigned long arg)
 950 {
 951         struct tap_dev *tap;
 952         netdev_features_t features;
 953         netdev_features_t feature_mask = 0;
 954
 955         tap = rtnl_dereference(q->tap);
 956         if (!tap)
 957                 return -ENOLINK;
 958
 959         features = tap->dev->features;
 960
 961         if (arg & TUN_F_CSUM) {
 962                 feature_mask = NETIF_F_HW_CSUM;
 963
 964                 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
 965                         if (arg & TUN_F_TSO_ECN)
 966                                 feature_mask |= NETIF_F_TSO_ECN;
 967                         if (arg & TUN_F_TSO4)
 968                                 feature_mask |= NETIF_F_TSO;
 969                         if (arg & TUN_F_TSO6)
 970                                 feature_mask |= NETIF_F_TSO6;
 971                 }
 972
 973                 /* TODO: for now USO4 and USO6 should work simultaneously */
 974                 if ((arg & (TUN_F_USO4 | TUN_F_USO6)) == (TUN_F_USO4 | TUN_F_USO6))
 975                         features |= NETIF_F_GSO_UDP_L4;
 976         }
 977
 978         /* tun/tap driver inverts the usage for TSO offloads, where
 979          * setting the TSO bit means that the userspace wants to
 980          * accept TSO frames and turning it off means that user space
 981          * does not support TSO.
 982          * For tap, we have to invert it to mean the same thing.
 983          * When user space turns off TSO, we turn off GSO/LRO so that
 984          * user-space will not receive TSO frames.
 985          */
 986         if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6) ||
 987             (feature_mask & (TUN_F_USO4 | TUN_F_USO6)) == (TUN_F_USO4 | TUN_F_USO6))
 988                 features |= RX_OFFLOADS;
 989         else
 990                 features &= ~RX_OFFLOADS;
 991
 992         /* tap_features are the same as features on tun/tap and
 993          * reflect user expectations.
 994          */
 995         tap->tap_features = feature_mask;
 996         if (tap->update_features)
 997                 tap->update_features(tap, features);
 998
 999         return 0;
1000 }
1001
1002 /*
1003  * provide compatibility with generic tun/tap interface
1004  */
1005 static long tap_ioctl(struct file *file, unsigned int cmd,
1006                       unsigned long arg)
1007 {
1008         struct tap_queue *q = file->private_data;
1009         struct tap_dev *tap;
1010         void __user *argp = (void __user *)arg;
1011         struct ifreq __user *ifr = argp;
1012         unsigned int __user *up = argp;
1013         unsigned short u;
1014         int __user *sp = argp;
1015         struct sockaddr sa;
1016         int s;
1017         int ret;
1018
1019         switch (cmd) {
1020         case TUNSETIFF:
1021                 /* ignore the name, just look at flags */
1022                 if (get_user(u, &ifr->ifr_flags))
1023                         return -EFAULT;
1024
1025                 ret = 0;
1026                 if ((u & ~TAP_IFFEATURES) != (IFF_NO_PI | IFF_TAP))
1027                         ret = -EINVAL;
1028                 else
1029                         q->flags = (q->flags & ~TAP_IFFEATURES) | u;
1030
1031                 return ret;
1032
1033         case TUNGETIFF:
1034                 rtnl_lock();
1035                 tap = tap_get_tap_dev(q);
1036                 if (!tap) {
1037                         rtnl_unlock();
1038                         return -ENOLINK;
1039                 }
1040
1041                 ret = 0;
1042                 u = q->flags;
1043                 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) ||
1044                     put_user(u, &ifr->ifr_flags))
1045                         ret = -EFAULT;
1046                 tap_put_tap_dev(tap);
1047                 rtnl_unlock();
1048                 return ret;
1049
1050         case TUNSETQUEUE:
1051                 if (get_user(u, &ifr->ifr_flags))
1052                         return -EFAULT;
1053                 rtnl_lock();
1054                 ret = tap_ioctl_set_queue(file, u);
1055                 rtnl_unlock();
1056                 return ret;
1057
1058         case TUNGETFEATURES:
1059                 if (put_user(IFF_TAP | IFF_NO_PI | TAP_IFFEATURES, up))
1060                         return -EFAULT;
1061                 return 0;
1062
1063         case TUNSETSNDBUF:
1064                 if (get_user(s, sp))
1065                         return -EFAULT;
1066                 if (s <= 0)
1067                         return -EINVAL;
1068
1069                 q->sk.sk_sndbuf = s;
1070                 return 0;
1071
1072         case TUNGETVNETHDRSZ:
1073                 s = q->vnet_hdr_sz;
1074                 if (put_user(s, sp))
1075                         return -EFAULT;
1076                 return 0;
1077
1078         case TUNSETVNETHDRSZ:
1079                 if (get_user(s, sp))
1080                         return -EFAULT;
1081                 if (s < (int)sizeof(struct virtio_net_hdr))
1082                         return -EINVAL;
1083
1084                 q->vnet_hdr_sz = s;
1085                 return 0;
1086
1087         case TUNGETVNETLE:
1088                 s = !!(q->flags & TAP_VNET_LE);
1089                 if (put_user(s, sp))
1090                         return -EFAULT;
1091                 return 0;
1092
1093         case TUNSETVNETLE:
1094                 if (get_user(s, sp))
1095                         return -EFAULT;
1096                 if (s)
1097                         q->flags |= TAP_VNET_LE;
1098                 else
1099                         q->flags &= ~TAP_VNET_LE;
1100                 return 0;
1101
1102         case TUNGETVNETBE:
1103                 return tap_get_vnet_be(q, sp);
1104
1105         case TUNSETVNETBE:
1106                 return tap_set_vnet_be(q, sp);
1107
1108         case TUNSETOFFLOAD:
1109                 /* let the user check for future flags */
1110                 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1111                             TUN_F_TSO_ECN | TUN_F_UFO |
1112                             TUN_F_USO4 | TUN_F_USO6))
1113                         return -EINVAL;
1114
1115                 rtnl_lock();
1116                 ret = set_offload(q, arg);
1117                 rtnl_unlock();
1118                 return ret;
1119
1120         case SIOCGIFHWADDR:
1121                 rtnl_lock();
1122                 tap = tap_get_tap_dev(q);
1123                 if (!tap) {
1124                         rtnl_unlock();
1125                         return -ENOLINK;
1126                 }
1127                 ret = 0;
1128                 dev_get_mac_address(&sa, dev_net(tap->dev), tap->dev->name);
1129                 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) ||
1130                     copy_to_user(&ifr->ifr_hwaddr, &sa, sizeof(sa)))
1131                         ret = -EFAULT;
1132                 tap_put_tap_dev(tap);
1133                 rtnl_unlock();
1134                 return ret;
1135
1136         case SIOCSIFHWADDR:
1137                 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa)))
1138                         return -EFAULT;
1139                 rtnl_lock();
1140                 tap = tap_get_tap_dev(q);
1141                 if (!tap) {
1142                         rtnl_unlock();
1143                         return -ENOLINK;
1144                 }
1145                 ret = dev_set_mac_address_user(tap->dev, &sa, NULL);
1146                 tap_put_tap_dev(tap);
1147                 rtnl_unlock();
1148                 return ret;
1149
1150         default:
1151                 return -EINVAL;
1152         }
1153 }
1154
1155 static const struct file_operations tap_fops = {
1156         .owner          = THIS_MODULE,
1157         .open           = tap_open,
1158         .release        = tap_release,
1159         .read_iter      = tap_read_iter,
1160         .write_iter     = tap_write_iter,
1161         .poll           = tap_poll,
1162         .llseek         = no_llseek,
1163         .unlocked_ioctl = tap_ioctl,
1164         .compat_ioctl   = compat_ptr_ioctl,
1165 };
1166
1167 static int tap_get_user_xdp(struct tap_queue *q, struct xdp_buff *xdp)
1168 {
1169         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
1170         struct virtio_net_hdr *gso = &hdr->gso;
1171         int buflen = hdr->buflen;
1172         int vnet_hdr_len = 0;
1173         struct tap_dev *tap;
1174         struct sk_buff *skb;
1175         int err, depth;
1176
1177         if (q->flags & IFF_VNET_HDR)
1178                 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
1179
1180         skb = build_skb(xdp->data_hard_start, buflen);
1181         if (!skb) {
1182                 err = -ENOMEM;
1183                 goto err;
1184         }
1185
1186         skb_reserve(skb, xdp->data - xdp->data_hard_start);
1187         skb_put(skb, xdp->data_end - xdp->data);
1188
1189         skb_set_network_header(skb, ETH_HLEN);
1190         skb_reset_mac_header(skb);
1191         skb->protocol = eth_hdr(skb)->h_proto;
1192
1193         if (vnet_hdr_len) {
1194                 err = virtio_net_hdr_to_skb(skb, gso, tap_is_little_endian(q));
1195                 if (err)
1196                         goto err_kfree;
1197         }
1198
1199         /* Move network header to the right position for VLAN tagged packets */
1200         if (eth_type_vlan(skb->protocol) &&
1201             vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0)
1202                 skb_set_network_header(skb, depth);
1203
1204         rcu_read_lock();
1205         tap = rcu_dereference(q->tap);
1206         if (tap) {
1207                 skb->dev = tap->dev;
1208                 skb_probe_transport_header(skb);
1209                 dev_queue_xmit(skb);
1210         } else {
1211                 kfree_skb(skb);
1212         }
1213         rcu_read_unlock();
1214
1215         return 0;
1216
1217 err_kfree:
1218         kfree_skb(skb);
1219 err:
1220         rcu_read_lock();
1221         tap = rcu_dereference(q->tap);
1222         if (tap && tap->count_tx_dropped)
1223                 tap->count_tx_dropped(tap);
1224         rcu_read_unlock();
1225         return err;
1226 }
1227
1228 static int tap_sendmsg(struct socket *sock, struct msghdr *m,
1229                        size_t total_len)
1230 {
1231         struct tap_queue *q = container_of(sock, struct tap_queue, sock);
1232         struct tun_msg_ctl *ctl = m->msg_control;
1233         struct xdp_buff *xdp;
1234         int i;
1235
1236         if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
1237             ctl && ctl->type == TUN_MSG_PTR) {
1238                 for (i = 0; i < ctl->num; i++) {
1239                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
1240                         tap_get_user_xdp(q, xdp);
1241                 }
1242                 return 0;
1243         }
1244
1245         return tap_get_user(q, ctl ? ctl->ptr : NULL, &m->msg_iter,
1246                             m->msg_flags & MSG_DONTWAIT);
1247 }
1248
1249 static int tap_recvmsg(struct socket *sock, struct msghdr *m,
1250                        size_t total_len, int flags)
1251 {
1252         struct tap_queue *q = container_of(sock, struct tap_queue, sock);
1253         struct sk_buff *skb = m->msg_control;
1254         int ret;
1255         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) {
1256                 kfree_skb(skb);
1257                 return -EINVAL;
1258         }
1259         ret = tap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT, skb);
1260         if (ret > total_len) {
1261                 m->msg_flags |= MSG_TRUNC;
1262                 ret = flags & MSG_TRUNC ? ret : total_len;
1263         }
1264         return ret;
1265 }
1266
1267 static int tap_peek_len(struct socket *sock)
1268 {
1269         struct tap_queue *q = container_of(sock, struct tap_queue,
1270                                                sock);
1271         return PTR_RING_PEEK_CALL(&q->ring, __skb_array_len_with_tag);
1272 }
1273
1274 /* Ops structure to mimic raw sockets with tun */
1275 static const struct proto_ops tap_socket_ops = {
1276         .sendmsg = tap_sendmsg,
1277         .recvmsg = tap_recvmsg,
1278         .peek_len = tap_peek_len,
1279 };
1280
1281 /* Get an underlying socket object from tun file.  Returns error unless file is
1282  * attached to a device.  The returned object works like a packet socket, it
1283  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1284  * holding a reference to the file for as long as the socket is in use. */
1285 struct socket *tap_get_socket(struct file *file)
1286 {
1287         struct tap_queue *q;
1288         if (file->f_op != &tap_fops)
1289                 return ERR_PTR(-EINVAL);
1290         q = file->private_data;
1291         if (!q)
1292                 return ERR_PTR(-EBADFD);
1293         return &q->sock;
1294 }
1295 EXPORT_SYMBOL_GPL(tap_get_socket);
1296
1297 struct ptr_ring *tap_get_ptr_ring(struct file *file)
1298 {
1299         struct tap_queue *q;
1300
1301         if (file->f_op != &tap_fops)
1302                 return ERR_PTR(-EINVAL);
1303         q = file->private_data;
1304         if (!q)
1305                 return ERR_PTR(-EBADFD);
1306         return &q->ring;
1307 }
1308 EXPORT_SYMBOL_GPL(tap_get_ptr_ring);
1309
1310 int tap_queue_resize(struct tap_dev *tap)
1311 {
1312         struct net_device *dev = tap->dev;
1313         struct tap_queue *q;
1314         struct ptr_ring **rings;
1315         int n = tap->numqueues;
1316         int ret, i = 0;
1317
1318         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
1319         if (!rings)
1320                 return -ENOMEM;
1321
1322         list_for_each_entry(q, &tap->queue_list, next)
1323                 rings[i++] = &q->ring;
1324
1325         ret = ptr_ring_resize_multiple(rings, n,
1326                                        dev->tx_queue_len, GFP_KERNEL,
1327                                        __skb_array_destroy_skb);
1328
1329         kfree(rings);
1330         return ret;
1331 }
1332 EXPORT_SYMBOL_GPL(tap_queue_resize);
1333
1334 static int tap_list_add(dev_t major, const char *device_name)
1335 {
1336         struct major_info *tap_major;
1337
1338         tap_major = kzalloc(sizeof(*tap_major), GFP_ATOMIC);
1339         if (!tap_major)
1340                 return -ENOMEM;
1341
1342         tap_major->major = MAJOR(major);
1343
1344         idr_init(&tap_major->minor_idr);
1345         spin_lock_init(&tap_major->minor_lock);
1346
1347         tap_major->device_name = device_name;
1348
1349         list_add_tail_rcu(&tap_major->next, &major_list);
1350         return 0;
1351 }
1352
1353 int tap_create_cdev(struct cdev *tap_cdev, dev_t *tap_major,
1354                     const char *device_name, struct module *module)
1355 {
1356         int err;
1357
1358         err = alloc_chrdev_region(tap_major, 0, TAP_NUM_DEVS, device_name);
1359         if (err)
1360                 goto out1;
1361
1362         cdev_init(tap_cdev, &tap_fops);
1363         tap_cdev->owner = module;
1364         err = cdev_add(tap_cdev, *tap_major, TAP_NUM_DEVS);
1365         if (err)
1366                 goto out2;
1367
1368         err =  tap_list_add(*tap_major, device_name);
1369         if (err)
1370                 goto out3;
1371
1372         return 0;
1373
1374 out3:
1375         cdev_del(tap_cdev);
1376 out2:
1377         unregister_chrdev_region(*tap_major, TAP_NUM_DEVS);
1378 out1:
1379         return err;
1380 }
1381 EXPORT_SYMBOL_GPL(tap_create_cdev);
1382
1383 void tap_destroy_cdev(dev_t major, struct cdev *tap_cdev)
1384 {
1385         struct major_info *tap_major, *tmp;
1386
1387         cdev_del(tap_cdev);
1388         unregister_chrdev_region(major, TAP_NUM_DEVS);
1389         list_for_each_entry_safe(tap_major, tmp, &major_list, next) {
1390                 if (tap_major->major == MAJOR(major)) {
1391                         idr_destroy(&tap_major->minor_idr);
1392                         list_del_rcu(&tap_major->next);
1393                         kfree_rcu(tap_major, rcu);
1394                 }
1395         }
1396 }
1397 EXPORT_SYMBOL_GPL(tap_destroy_cdev);
1398
1399 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1400 MODULE_AUTHOR("Sainath Grandhi <sainath.grandhi@intel.com>");
1401 MODULE_LICENSE("GPL");