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