drivers/net/tap.c

   1 #include <linux/etherdevice.h>
   2 #include <linux/if_tap.h>
   3 #include <linux/if_vlan.h>
   4 #include <linux/interrupt.h>
   5 #include <linux/nsproxy.h>
   6 #include <linux/compat.h>
   7 #include <linux/if_tun.h>
   8 #include <linux/module.h>
   9 #include <linux/skbuff.h>
  10 #include <linux/cache.h>
  11 #include <linux/sched/signal.h>
  12 #include <linux/types.h>
  13 #include <linux/slab.h>
  14 #include <linux/wait.h>
  15 #include <linux/cdev.h>
  16 #include <linux/idr.h>
  17 #include <linux/fs.h>
  18 #include <linux/uio.h>
  19
  20 #include <net/net_namespace.h>
  21 #include <net/rtnetlink.h>
  22 #include <net/sock.h>
  23 #include <linux/virtio_net.h>
  24 #include <linux/skb_array.h>
  25
  26 #define TAP_IFFEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE)
  27
  28 #define TAP_VNET_LE 0x80000000
  29 #define TAP_VNET_BE 0x40000000
  30
  31 #ifdef CONFIG_TUN_VNET_CROSS_LE
  32 static inline bool tap_legacy_is_little_endian(struct tap_queue *q)
  33 {
  34         return q->flags & TAP_VNET_BE ? false :
  35                 virtio_legacy_is_little_endian();
  36 }
  37
  38 static long tap_get_vnet_be(struct tap_queue *q, int __user *sp)
  39 {
  40         int s = !!(q->flags & TAP_VNET_BE);
  41
  42         if (put_user(s, sp))
  43                 return -EFAULT;
  44
  45         return 0;
  46 }
  47
  48 static long tap_set_vnet_be(struct tap_queue *q, int __user *sp)
  49 {
  50         int s;
  51
  52         if (get_user(s, sp))
  53                 return -EFAULT;
  54
  55         if (s)
  56                 q->flags |= TAP_VNET_BE;
  57         else
  58                 q->flags &= ~TAP_VNET_BE;
  59
  60         return 0;
  61 }
  62 #else
  63 static inline bool tap_legacy_is_little_endian(struct tap_queue *q)
  64 {
  65         return virtio_legacy_is_little_endian();
  66 }
  67
  68 static long tap_get_vnet_be(struct tap_queue *q, int __user *argp)
  69 {
  70         return -EINVAL;
  71 }
  72
  73 static long tap_set_vnet_be(struct tap_queue *q, int __user *argp)
  74 {
  75         return -EINVAL;
  76 }
  77 #endif /* CONFIG_TUN_VNET_CROSS_LE */
  78
  79 static inline bool tap_is_little_endian(struct tap_queue *q)
  80 {
  81         return q->flags & TAP_VNET_LE ||
  82                 tap_legacy_is_little_endian(q);
  83 }
  84
  85 static inline u16 tap16_to_cpu(struct tap_queue *q, __virtio16 val)
  86 {
  87         return __virtio16_to_cpu(tap_is_little_endian(q), val);
  88 }
  89
  90 static inline __virtio16 cpu_to_tap16(struct tap_queue *q, u16 val)
  91 {
  92         return __cpu_to_virtio16(tap_is_little_endian(q), val);
  93 }
  94
  95 static struct proto tap_proto = {
  96         .name = "tap",
  97         .owner = THIS_MODULE,
  98         .obj_size = sizeof(struct tap_queue),
  99 };
 100
 101 #define TAP_NUM_DEVS (1U << MINORBITS)
 102
 103 static LIST_HEAD(major_list);
 104
 105 struct major_info {
 106         struct rcu_head rcu;
 107         dev_t major;
 108         struct idr minor_idr;
 109         spinlock_t minor_lock;
 110         const char *device_name;
 111         struct list_head next;
 112 };
 113
 114 #define GOODCOPY_LEN 128
 115
 116 static const struct proto_ops tap_socket_ops;
 117
 118 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
 119 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST)
 120
 121 static struct tap_dev *tap_dev_get_rcu(const struct net_device *dev)
 122 {
 123         return rcu_dereference(dev->rx_handler_data);
 124 }
 125
 126 /*
 127  * RCU usage:
 128  * The tap_queue and the macvlan_dev are loosely coupled, the
 129  * pointers from one to the other can only be read while rcu_read_lock
 130  * or rtnl is held.
 131  *
 132  * Both the file and the macvlan_dev hold a reference on the tap_queue
 133  * through sock_hold(&q->sk). When the macvlan_dev goes away first,
 134  * q->vlan becomes inaccessible. When the files gets closed,
 135  * tap_get_queue() fails.
 136  *
 137  * There may still be references to the struct sock inside of the
 138  * queue from outbound SKBs, but these never reference back to the
 139  * file or the dev. The data structure is freed through __sk_free
 140  * when both our references and any pending SKBs are gone.
 141  */
 142
 143 static int tap_enable_queue(struct tap_dev *tap, struct file *file,
 144                             struct tap_queue *q)
 145 {
 146         int err = -EINVAL;
 147
 148         ASSERT_RTNL();
 149
 150         if (q->enabled)
 151                 goto out;
 152
 153         err = 0;
 154         rcu_assign_pointer(tap->taps[tap->numvtaps], q);
 155         q->queue_index = tap->numvtaps;
 156         q->enabled = true;
 157
 158         tap->numvtaps++;
 159 out:
 160         return err;
 161 }
 162
 163 /* Requires RTNL */
 164 static int tap_set_queue(struct tap_dev *tap, struct file *file,
 165                          struct tap_queue *q)
 166 {
 167         if (tap->numqueues == MAX_TAP_QUEUES)
 168                 return -EBUSY;
 169
 170         rcu_assign_pointer(q->tap, tap);
 171         rcu_assign_pointer(tap->taps[tap->numvtaps], q);
 172         sock_hold(&q->sk);
 173
 174         q->file = file;
 175         q->queue_index = tap->numvtaps;
 176         q->enabled = true;
 177         file->private_data = q;
 178         list_add_tail(&q->next, &tap->queue_list);
 179
 180         tap->numvtaps++;
 181         tap->numqueues++;
 182
 183         return 0;
 184 }
 185
 186 static int tap_disable_queue(struct tap_queue *q)
 187 {
 188         struct tap_dev *tap;
 189         struct tap_queue *nq;
 190
 191         ASSERT_RTNL();
 192         if (!q->enabled)
 193                 return -EINVAL;
 194
 195         tap = rtnl_dereference(q->tap);
 196
 197         if (tap) {
 198                 int index = q->queue_index;
 199                 BUG_ON(index >= tap->numvtaps);
 200                 nq = rtnl_dereference(tap->taps[tap->numvtaps - 1]);
 201                 nq->queue_index = index;
 202
 203                 rcu_assign_pointer(tap->taps[index], nq);
 204                 RCU_INIT_POINTER(tap->taps[tap->numvtaps - 1], NULL);
 205                 q->enabled = false;
 206
 207                 tap->numvtaps--;
 208         }
 209
 210         return 0;
 211 }
 212
 213 /*
 214  * The file owning the queue got closed, give up both
 215  * the reference that the files holds as well as the
 216  * one from the macvlan_dev if that still exists.
 217  *
 218  * Using the spinlock makes sure that we don't get
 219  * to the queue again after destroying it.
 220  */
 221 static void tap_put_queue(struct tap_queue *q)
 222 {
 223         struct tap_dev *tap;
 224
 225         rtnl_lock();
 226         tap = rtnl_dereference(q->tap);
 227
 228         if (tap) {
 229                 if (q->enabled)
 230                         BUG_ON(tap_disable_queue(q));
 231
 232                 tap->numqueues--;
 233                 RCU_INIT_POINTER(q->tap, NULL);
 234                 sock_put(&q->sk);
 235                 list_del_init(&q->next);
 236         }
 237
 238         rtnl_unlock();
 239
 240         synchronize_rcu();
 241         sock_put(&q->sk);
 242 }
 243
 244 /*
 245  * Select a queue based on the rxq of the device on which this packet
 246  * arrived. If the incoming device is not mq, calculate a flow hash
 247  * to select a queue. If all fails, find the first available queue.
 248  * Cache vlan->numvtaps since it can become zero during the execution
 249  * of this function.
 250  */
 251 static struct tap_queue *tap_get_queue(struct tap_dev *tap,
 252                                        struct sk_buff *skb)
 253 {
 254         struct tap_queue *queue = NULL;
 255         /* Access to taps array is protected by rcu, but access to numvtaps
 256          * isn't. Below we use it to lookup a queue, but treat it as a hint
 257          * and validate that the result isn't NULL - in case we are
 258          * racing against queue removal.
 259          */
 260         int numvtaps = READ_ONCE(tap->numvtaps);
 261         __u32 rxq;
 262
 263         if (!numvtaps)
 264                 goto out;
 265
 266         if (numvtaps == 1)
 267                 goto single;
 268
 269         /* Check if we can use flow to select a queue */
 270         rxq = skb_get_hash(skb);
 271         if (rxq) {
 272                 queue = rcu_dereference(tap->taps[rxq % numvtaps]);
 273                 goto out;
 274         }
 275
 276         if (likely(skb_rx_queue_recorded(skb))) {
 277                 rxq = skb_get_rx_queue(skb);
 278
 279                 while (unlikely(rxq >= numvtaps))
 280                         rxq -= numvtaps;
 281
 282                 queue = rcu_dereference(tap->taps[rxq]);
 283                 goto out;
 284         }
 285
 286 single:
 287         queue = rcu_dereference(tap->taps[0]);
 288 out:
 289         return queue;
 290 }
 291
 292 /*
 293  * The net_device is going away, give up the reference
 294  * that it holds on all queues and safely set the pointer
 295  * from the queues to NULL.
 296  */
 297 void tap_del_queues(struct tap_dev *tap)
 298 {
 299         struct tap_queue *q, *tmp;
 300
 301         ASSERT_RTNL();
 302         list_for_each_entry_safe(q, tmp, &tap->queue_list, next) {
 303                 list_del_init(&q->next);
 304                 RCU_INIT_POINTER(q->tap, NULL);
 305                 if (q->enabled)
 306                         tap->numvtaps--;
 307                 tap->numqueues--;
 308                 sock_put(&q->sk);
 309         }
 310         BUG_ON(tap->numvtaps);
 311         BUG_ON(tap->numqueues);
 312         /* guarantee that any future tap_set_queue will fail */
 313         tap->numvtaps = MAX_TAP_QUEUES;
 314 }
 315 EXPORT_SYMBOL_GPL(tap_del_queues);
 316
 317 rx_handler_result_t tap_handle_frame(struct sk_buff **pskb)
 318 {
 319         struct sk_buff *skb = *pskb;
 320         struct net_device *dev = skb->dev;
 321         struct tap_dev *tap;
 322         struct tap_queue *q;
 323         netdev_features_t features = TAP_FEATURES;
 324
 325         tap = tap_dev_get_rcu(dev);
 326         if (!tap)
 327                 return RX_HANDLER_PASS;
 328
 329         q = tap_get_queue(tap, skb);
 330         if (!q)
 331                 return RX_HANDLER_PASS;
 332
 333         skb_push(skb, ETH_HLEN);
 334
 335         /* Apply the forward feature mask so that we perform segmentation
 336          * according to users wishes.  This only works if VNET_HDR is
 337          * enabled.
 338          */
 339         if (q->flags & IFF_VNET_HDR)
 340                 features |= tap->tap_features;
 341         if (netif_needs_gso(skb, features)) {
 342                 struct sk_buff *segs = __skb_gso_segment(skb, features, false);
 343
 344                 if (IS_ERR(segs))
 345                         goto drop;
 346
 347                 if (!segs) {
 348                         if (ptr_ring_produce(&q->ring, skb))
 349                                 goto drop;
 350                         goto wake_up;
 351                 }
 352
 353                 consume_skb(skb);
 354                 while (segs) {
 355                         struct sk_buff *nskb = segs->next;
 356
 357                         segs->next = NULL;
 358                         if (ptr_ring_produce(&q->ring, segs)) {
 359                                 kfree_skb(segs);
 360                                 kfree_skb_list(nskb);
 361                                 break;
 362                         }
 363                         segs = nskb;
 364                 }
 365         } else {
 366                 /* If we receive a partial checksum and the tap side
 367                  * doesn't support checksum offload, compute the checksum.
 368                  * Note: it doesn't matter which checksum feature to
 369                  *        check, we either support them all or none.
 370                  */
 371                 if (skb->ip_summed == CHECKSUM_PARTIAL &&
 372                     !(features & NETIF_F_CSUM_MASK) &&
 373                     skb_checksum_help(skb))
 374                         goto drop;
 375                 if (ptr_ring_produce(&q->ring, skb))
 376                         goto drop;
 377         }
 378
 379 wake_up:
 380         wake_up_interruptible_poll(sk_sleep(&q->sk), EPOLLIN | EPOLLRDNORM | EPOLLRDBAND);
 381         return RX_HANDLER_CONSUMED;
 382
 383 drop:
 384         /* Count errors/drops only here, thus don't care about args. */
 385         if (tap->count_rx_dropped)
 386                 tap->count_rx_dropped(tap);
 387         kfree_skb(skb);
 388         return RX_HANDLER_CONSUMED;
 389 }
 390 EXPORT_SYMBOL_GPL(tap_handle_frame);
 391
 392 static struct major_info *tap_get_major(int major)
 393 {
 394         struct major_info *tap_major;
 395
 396         list_for_each_entry_rcu(tap_major, &major_list, next) {
 397                 if (tap_major->major == major)
 398                         return tap_major;
 399         }
 400
 401         return NULL;
 402 }
 403
 404 int tap_get_minor(dev_t major, struct tap_dev *tap)
 405 {
 406         int retval = -ENOMEM;
 407         struct major_info *tap_major;
 408
 409         rcu_read_lock();
 410         tap_major = tap_get_major(MAJOR(major));
 411         if (!tap_major) {
 412                 retval = -EINVAL;
 413                 goto unlock;
 414         }
 415
 416         spin_lock(&tap_major->minor_lock);
 417         retval = idr_alloc(&tap_major->minor_idr, tap, 1, TAP_NUM_DEVS, GFP_ATOMIC);
 418         if (retval >= 0) {
 419                 tap->minor = retval;
 420         } else if (retval == -ENOSPC) {
 421                 netdev_err(tap->dev, "Too many tap devices\n");
 422                 retval = -EINVAL;
 423         }
 424         spin_unlock(&tap_major->minor_lock);
 425
 426 unlock:
 427         rcu_read_unlock();
 428         return retval < 0 ? retval : 0;
 429 }
 430 EXPORT_SYMBOL_GPL(tap_get_minor);
 431
 432 void tap_free_minor(dev_t major, struct tap_dev *tap)
 433 {
 434         struct major_info *tap_major;
 435
 436         rcu_read_lock();
 437         tap_major = tap_get_major(MAJOR(major));
 438         if (!tap_major) {
 439                 goto unlock;
 440         }
 441
 442         spin_lock(&tap_major->minor_lock);
 443         if (tap->minor) {
 444                 idr_remove(&tap_major->minor_idr, tap->minor);
 445                 tap->minor = 0;
 446         }
 447         spin_unlock(&tap_major->minor_lock);
 448
 449 unlock:
 450         rcu_read_unlock();
 451 }
 452 EXPORT_SYMBOL_GPL(tap_free_minor);
 453
 454 static struct tap_dev *dev_get_by_tap_file(int major, int minor)
 455 {
 456         struct net_device *dev = NULL;
 457         struct tap_dev *tap;
 458         struct major_info *tap_major;
 459
 460         rcu_read_lock();
 461         tap_major = tap_get_major(major);
 462         if (!tap_major) {
 463                 tap = NULL;
 464                 goto unlock;
 465         }
 466
 467         spin_lock(&tap_major->minor_lock);
 468         tap = idr_find(&tap_major->minor_idr, minor);
 469         if (tap) {
 470                 dev = tap->dev;
 471                 dev_hold(dev);
 472         }
 473         spin_unlock(&tap_major->minor_lock);
 474
 475 unlock:
 476         rcu_read_unlock();
 477         return tap;
 478 }
 479
 480 static void tap_sock_write_space(struct sock *sk)
 481 {
 482         wait_queue_head_t *wqueue;
 483
 484         if (!sock_writeable(sk) ||
 485             !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
 486                 return;
 487
 488         wqueue = sk_sleep(sk);
 489         if (wqueue && waitqueue_active(wqueue))
 490                 wake_up_interruptible_poll(wqueue, EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
 491 }
 492
 493 static void tap_sock_destruct(struct sock *sk)
 494 {
 495         struct tap_queue *q = container_of(sk, struct tap_queue, sk);
 496
 497         ptr_ring_cleanup(&q->ring, __skb_array_destroy_skb);
 498 }
 499
 500 static int tap_open(struct inode *inode, struct file *file)
 501 {
 502         struct net *net = current->nsproxy->net_ns;
 503         struct tap_dev *tap;
 504         struct tap_queue *q;
 505         int err = -ENODEV;
 506
 507         rtnl_lock();
 508         tap = dev_get_by_tap_file(imajor(inode), iminor(inode));
 509         if (!tap)
 510                 goto err;
 511
 512         err = -ENOMEM;
 513         q = (struct tap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
 514                                              &tap_proto, 0);
 515         if (!q)
 516                 goto err;
 517         if (ptr_ring_init(&q->ring, tap->dev->tx_queue_len, GFP_KERNEL)) {
 518                 sk_free(&q->sk);
 519                 goto err;
 520         }
 521
 522         RCU_INIT_POINTER(q->sock.wq, &q->wq);
 523         init_waitqueue_head(&q->wq.wait);
 524         q->sock.type = SOCK_RAW;
 525         q->sock.state = SS_CONNECTED;
 526         q->sock.file = file;
 527         q->sock.ops = &tap_socket_ops;
 528         sock_init_data(&q->sock, &q->sk);
 529         q->sk.sk_write_space = tap_sock_write_space;
 530         q->sk.sk_destruct = tap_sock_destruct;
 531         q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
 532         q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
 533
 534         /*
 535          * so far only KVM virtio_net uses tap, enable zero copy between
 536          * guest kernel and host kernel when lower device supports zerocopy
 537          *
 538          * The macvlan supports zerocopy iff the lower device supports zero
 539          * copy so we don't have to look at the lower device directly.
 540          */
 541         if ((tap->dev->features & NETIF_F_HIGHDMA) && (tap->dev->features & NETIF_F_SG))
 542                 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
 543
 544         err = tap_set_queue(tap, file, q);
 545         if (err) {
 546                 /* tap_sock_destruct() will take care of freeing ptr_ring */
 547                 goto err_put;
 548         }
 549
 550         dev_put(tap->dev);
 551
 552         rtnl_unlock();
 553         return err;
 554
 555 err_put:
 556         sock_put(&q->sk);
 557 err:
 558         if (tap)
 559                 dev_put(tap->dev);
 560
 561         rtnl_unlock();
 562         return err;
 563 }
 564
 565 static int tap_release(struct inode *inode, struct file *file)
 566 {
 567         struct tap_queue *q = file->private_data;
 568         tap_put_queue(q);
 569         return 0;
 570 }
 571
 572 static __poll_t tap_poll(struct file *file, poll_table *wait)
 573 {
 574         struct tap_queue *q = file->private_data;
 575         __poll_t mask = EPOLLERR;
 576
 577         if (!q)
 578                 goto out;
 579
 580         mask = 0;
 581         poll_wait(file, &q->wq.wait, wait);
 582
 583         if (!ptr_ring_empty(&q->ring))
 584                 mask |= EPOLLIN | EPOLLRDNORM;
 585
 586         if (sock_writeable(&q->sk) ||
 587             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) &&
 588              sock_writeable(&q->sk)))
 589                 mask |= EPOLLOUT | EPOLLWRNORM;
 590
 591 out:
 592         return mask;
 593 }
 594
 595 static inline struct sk_buff *tap_alloc_skb(struct sock *sk, size_t prepad,
 596                                             size_t len, size_t linear,
 597                                                 int noblock, int *err)
 598 {
 599         struct sk_buff *skb;
 600
 601         /* Under a page?  Don't bother with paged skb. */
 602         if (prepad + len < PAGE_SIZE || !linear)
 603                 linear = len;
 604
 605         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 606                                    err, 0);
 607         if (!skb)
 608                 return NULL;
 609
 610         skb_reserve(skb, prepad);
 611         skb_put(skb, linear);
 612         skb->data_len = len - linear;
 613         skb->len += len - linear;
 614
 615         return skb;
 616 }
 617
 618 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */
 619 #define TAP_RESERVE HH_DATA_OFF(ETH_HLEN)
 620
 621 /* Get packet from user space buffer */
 622 static ssize_t tap_get_user(struct tap_queue *q, struct msghdr *m,
 623                             struct iov_iter *from, int noblock)
 624 {
 625         int good_linear = SKB_MAX_HEAD(TAP_RESERVE);
 626         struct sk_buff *skb;
 627         struct tap_dev *tap;
 628         unsigned long total_len = iov_iter_count(from);
 629         unsigned long len = total_len;
 630         int err;
 631         struct virtio_net_hdr vnet_hdr = { 0 };
 632         int vnet_hdr_len = 0;
 633         int copylen = 0;
 634         int depth;
 635         bool zerocopy = false;
 636         size_t linear;
 637
 638         if (q->flags & IFF_VNET_HDR) {
 639                 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
 640
 641                 err = -EINVAL;
 642                 if (len < vnet_hdr_len)
 643                         goto err;
 644                 len -= vnet_hdr_len;
 645
 646                 err = -EFAULT;
 647                 if (!copy_from_iter_full(&vnet_hdr, sizeof(vnet_hdr), from))
 648                         goto err;
 649                 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr));
 650                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
 651                      tap16_to_cpu(q, vnet_hdr.csum_start) +
 652                      tap16_to_cpu(q, vnet_hdr.csum_offset) + 2 >
 653                              tap16_to_cpu(q, vnet_hdr.hdr_len))
 654                         vnet_hdr.hdr_len = cpu_to_tap16(q,
 655                                  tap16_to_cpu(q, vnet_hdr.csum_start) +
 656                                  tap16_to_cpu(q, vnet_hdr.csum_offset) + 2);
 657                 err = -EINVAL;
 658                 if (tap16_to_cpu(q, vnet_hdr.hdr_len) > len)
 659                         goto err;
 660         }
 661
 662         err = -EINVAL;
 663         if (unlikely(len < ETH_HLEN))
 664                 goto err;
 665
 666         if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
 667                 struct iov_iter i;
 668
 669                 copylen = vnet_hdr.hdr_len ?
 670                         tap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN;
 671                 if (copylen > good_linear)
 672                         copylen = good_linear;
 673                 else if (copylen < ETH_HLEN)
 674                         copylen = ETH_HLEN;
 675                 linear = copylen;
 676                 i = *from;
 677                 iov_iter_advance(&i, copylen);
 678                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
 679                         zerocopy = true;
 680         }
 681
 682         if (!zerocopy) {
 683                 copylen = len;
 684                 linear = tap16_to_cpu(q, vnet_hdr.hdr_len);
 685                 if (linear > good_linear)
 686                         linear = good_linear;
 687                 else if (linear < ETH_HLEN)
 688                         linear = ETH_HLEN;
 689         }
 690
 691         skb = tap_alloc_skb(&q->sk, TAP_RESERVE, copylen,
 692                             linear, noblock, &err);
 693         if (!skb)
 694                 goto err;
 695
 696         if (zerocopy)
 697                 err = zerocopy_sg_from_iter(skb, from);
 698         else
 699                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
 700
 701         if (err)
 702                 goto err_kfree;
 703
 704         skb_set_network_header(skb, ETH_HLEN);
 705         skb_reset_mac_header(skb);
 706         skb->protocol = eth_hdr(skb)->h_proto;
 707
 708         if (vnet_hdr_len) {
 709                 err = virtio_net_hdr_to_skb(skb, &vnet_hdr,
 710                                             tap_is_little_endian(q));
 711                 if (err)
 712                         goto err_kfree;
 713         }
 714
 715         skb_probe_transport_header(skb, ETH_HLEN);
 716
 717         /* Move network header to the right position for VLAN tagged packets */
 718         if ((skb->protocol == htons(ETH_P_8021Q) ||
 719              skb->protocol == htons(ETH_P_8021AD)) &&
 720             __vlan_get_protocol(skb, skb->protocol, &depth) != 0)
 721                 skb_set_network_header(skb, depth);
 722
 723         rcu_read_lock();
 724         tap = rcu_dereference(q->tap);
 725         /* copy skb_ubuf_info for callback when skb has no error */
 726         if (zerocopy) {
 727                 skb_shinfo(skb)->destructor_arg = m->msg_control;
 728                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
 729                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
 730         } else if (m && m->msg_control) {
 731                 struct ubuf_info *uarg = m->msg_control;
 732                 uarg->callback(uarg, false);
 733         }
 734
 735         if (tap) {
 736                 skb->dev = tap->dev;
 737                 dev_queue_xmit(skb);
 738         } else {
 739                 kfree_skb(skb);
 740         }
 741         rcu_read_unlock();
 742
 743         return total_len;
 744
 745 err_kfree:
 746         kfree_skb(skb);
 747
 748 err:
 749         rcu_read_lock();
 750         tap = rcu_dereference(q->tap);
 751         if (tap && tap->count_tx_dropped)
 752                 tap->count_tx_dropped(tap);
 753         rcu_read_unlock();
 754
 755         return err;
 756 }
 757
 758 static ssize_t tap_write_iter(struct kiocb *iocb, struct iov_iter *from)
 759 {
 760         struct file *file = iocb->ki_filp;
 761         struct tap_queue *q = file->private_data;
 762
 763         return tap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK);
 764 }
 765
 766 /* Put packet to the user space buffer */
 767 static ssize_t tap_put_user(struct tap_queue *q,
 768                             const struct sk_buff *skb,
 769                             struct iov_iter *iter)
 770 {
 771         int ret;
 772         int vnet_hdr_len = 0;
 773         int vlan_offset = 0;
 774         int total;
 775
 776         if (q->flags & IFF_VNET_HDR) {
 777                 struct virtio_net_hdr vnet_hdr;
 778                 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
 779                 if (iov_iter_count(iter) < vnet_hdr_len)
 780                         return -EINVAL;
 781
 782                 if (virtio_net_hdr_from_skb(skb, &vnet_hdr,
 783                                             tap_is_little_endian(q), true))
 784                         BUG();
 785
 786                 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) !=
 787                     sizeof(vnet_hdr))
 788                         return -EFAULT;
 789
 790                 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr));
 791         }
 792         total = vnet_hdr_len;
 793         total += skb->len;
 794
 795         if (skb_vlan_tag_present(skb)) {
 796                 struct {
 797                         __be16 h_vlan_proto;
 798                         __be16 h_vlan_TCI;
 799                 } veth;
 800                 veth.h_vlan_proto = skb->vlan_proto;
 801                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
 802
 803                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
 804                 total += VLAN_HLEN;
 805
 806                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
 807                 if (ret || !iov_iter_count(iter))
 808                         goto done;
 809
 810                 ret = copy_to_iter(&veth, sizeof(veth), iter);
 811                 if (ret != sizeof(veth) || !iov_iter_count(iter))
 812                         goto done;
 813         }
 814
 815         ret = skb_copy_datagram_iter(skb, vlan_offset, iter,
 816                                      skb->len - vlan_offset);
 817
 818 done:
 819         return ret ? ret : total;
 820 }
 821
 822 static ssize_t tap_do_read(struct tap_queue *q,
 823                            struct iov_iter *to,
 824                            int noblock, struct sk_buff *skb)
 825 {
 826         DEFINE_WAIT(wait);
 827         ssize_t ret = 0;
 828
 829         if (!iov_iter_count(to)) {
 830                 if (skb)
 831                         kfree_skb(skb);
 832                 return 0;
 833         }
 834
 835         if (skb)
 836                 goto put;
 837
 838         while (1) {
 839                 if (!noblock)
 840                         prepare_to_wait(sk_sleep(&q->sk), &wait,
 841                                         TASK_INTERRUPTIBLE);
 842
 843                 /* Read frames from the queue */
 844                 skb = ptr_ring_consume(&q->ring);
 845                 if (skb)
 846                         break;
 847                 if (noblock) {
 848                         ret = -EAGAIN;
 849                         break;
 850                 }
 851                 if (signal_pending(current)) {
 852                         ret = -ERESTARTSYS;
 853                         break;
 854                 }
 855                 /* Nothing to read, let's sleep */
 856                 schedule();
 857         }
 858         if (!noblock)
 859                 finish_wait(sk_sleep(&q->sk), &wait);
 860
 861 put:
 862         if (skb) {
 863                 ret = tap_put_user(q, skb, to);
 864                 if (unlikely(ret < 0))
 865                         kfree_skb(skb);
 866                 else
 867                         consume_skb(skb);
 868         }
 869         return ret;
 870 }
 871
 872 static ssize_t tap_read_iter(struct kiocb *iocb, struct iov_iter *to)
 873 {
 874         struct file *file = iocb->ki_filp;
 875         struct tap_queue *q = file->private_data;
 876         ssize_t len = iov_iter_count(to), ret;
 877
 878         ret = tap_do_read(q, to, file->f_flags & O_NONBLOCK, NULL);
 879         ret = min_t(ssize_t, ret, len);
 880         if (ret > 0)
 881                 iocb->ki_pos = ret;
 882         return ret;
 883 }
 884
 885 static struct tap_dev *tap_get_tap_dev(struct tap_queue *q)
 886 {
 887         struct tap_dev *tap;
 888
 889         ASSERT_RTNL();
 890         tap = rtnl_dereference(q->tap);
 891         if (tap)
 892                 dev_hold(tap->dev);
 893
 894         return tap;
 895 }
 896
 897 static void tap_put_tap_dev(struct tap_dev *tap)
 898 {
 899         dev_put(tap->dev);
 900 }
 901
 902 static int tap_ioctl_set_queue(struct file *file, unsigned int flags)
 903 {
 904         struct tap_queue *q = file->private_data;
 905         struct tap_dev *tap;
 906         int ret;
 907
 908         tap = tap_get_tap_dev(q);
 909         if (!tap)
 910                 return -EINVAL;
 911
 912         if (flags & IFF_ATTACH_QUEUE)
 913                 ret = tap_enable_queue(tap, file, q);
 914         else if (flags & IFF_DETACH_QUEUE)
 915                 ret = tap_disable_queue(q);
 916         else
 917                 ret = -EINVAL;
 918
 919         tap_put_tap_dev(tap);
 920         return ret;
 921 }
 922
 923 static int set_offload(struct tap_queue *q, unsigned long arg)
 924 {
 925         struct tap_dev *tap;
 926         netdev_features_t features;
 927         netdev_features_t feature_mask = 0;
 928
 929         tap = rtnl_dereference(q->tap);
 930         if (!tap)
 931                 return -ENOLINK;
 932
 933         features = tap->dev->features;
 934
 935         if (arg & TUN_F_CSUM) {
 936                 feature_mask = NETIF_F_HW_CSUM;
 937
 938                 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
 939                         if (arg & TUN_F_TSO_ECN)
 940                                 feature_mask |= NETIF_F_TSO_ECN;
 941                         if (arg & TUN_F_TSO4)
 942                                 feature_mask |= NETIF_F_TSO;
 943                         if (arg & TUN_F_TSO6)
 944                                 feature_mask |= NETIF_F_TSO6;
 945                 }
 946         }
 947
 948         /* tun/tap driver inverts the usage for TSO offloads, where
 949          * setting the TSO bit means that the userspace wants to
 950          * accept TSO frames and turning it off means that user space
 951          * does not support TSO.
 952          * For tap, we have to invert it to mean the same thing.
 953          * When user space turns off TSO, we turn off GSO/LRO so that
 954          * user-space will not receive TSO frames.
 955          */
 956         if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6))
 957                 features |= RX_OFFLOADS;
 958         else
 959                 features &= ~RX_OFFLOADS;
 960
 961         /* tap_features are the same as features on tun/tap and
 962          * reflect user expectations.
 963          */
 964         tap->tap_features = feature_mask;
 965         if (tap->update_features)
 966                 tap->update_features(tap, features);
 967
 968         return 0;
 969 }
 970
 971 /*
 972  * provide compatibility with generic tun/tap interface
 973  */
 974 static long tap_ioctl(struct file *file, unsigned int cmd,
 975                       unsigned long arg)
 976 {
 977         struct tap_queue *q = file->private_data;
 978         struct tap_dev *tap;
 979         void __user *argp = (void __user *)arg;
 980         struct ifreq __user *ifr = argp;
 981         unsigned int __user *up = argp;
 982         unsigned short u;
 983         int __user *sp = argp;
 984         struct sockaddr sa;
 985         int s;
 986         int ret;
 987
 988         switch (cmd) {
 989         case TUNSETIFF:
 990                 /* ignore the name, just look at flags */
 991                 if (get_user(u, &ifr->ifr_flags))
 992                         return -EFAULT;
 993
 994                 ret = 0;
 995                 if ((u & ~TAP_IFFEATURES) != (IFF_NO_PI | IFF_TAP))
 996                         ret = -EINVAL;
 997                 else
 998                         q->flags = (q->flags & ~TAP_IFFEATURES) | u;
 999
1000                 return ret;
1001
1002         case TUNGETIFF:
1003                 rtnl_lock();
1004                 tap = tap_get_tap_dev(q);
1005                 if (!tap) {
1006                         rtnl_unlock();
1007                         return -ENOLINK;
1008                 }
1009
1010                 ret = 0;
1011                 u = q->flags;
1012                 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) ||
1013                     put_user(u, &ifr->ifr_flags))
1014                         ret = -EFAULT;
1015                 tap_put_tap_dev(tap);
1016                 rtnl_unlock();
1017                 return ret;
1018
1019         case TUNSETQUEUE:
1020                 if (get_user(u, &ifr->ifr_flags))
1021                         return -EFAULT;
1022                 rtnl_lock();
1023                 ret = tap_ioctl_set_queue(file, u);
1024                 rtnl_unlock();
1025                 return ret;
1026
1027         case TUNGETFEATURES:
1028                 if (put_user(IFF_TAP | IFF_NO_PI | TAP_IFFEATURES, up))
1029                         return -EFAULT;
1030                 return 0;
1031
1032         case TUNSETSNDBUF:
1033                 if (get_user(s, sp))
1034                         return -EFAULT;
1035                 if (s <= 0)
1036                         return -EINVAL;
1037
1038                 q->sk.sk_sndbuf = s;
1039                 return 0;
1040
1041         case TUNGETVNETHDRSZ:
1042                 s = q->vnet_hdr_sz;
1043                 if (put_user(s, sp))
1044                         return -EFAULT;
1045                 return 0;
1046
1047         case TUNSETVNETHDRSZ:
1048                 if (get_user(s, sp))
1049                         return -EFAULT;
1050                 if (s < (int)sizeof(struct virtio_net_hdr))
1051                         return -EINVAL;
1052
1053                 q->vnet_hdr_sz = s;
1054                 return 0;
1055
1056         case TUNGETVNETLE:
1057                 s = !!(q->flags & TAP_VNET_LE);
1058                 if (put_user(s, sp))
1059                         return -EFAULT;
1060                 return 0;
1061
1062         case TUNSETVNETLE:
1063                 if (get_user(s, sp))
1064                         return -EFAULT;
1065                 if (s)
1066                         q->flags |= TAP_VNET_LE;
1067                 else
1068                         q->flags &= ~TAP_VNET_LE;
1069                 return 0;
1070
1071         case TUNGETVNETBE:
1072                 return tap_get_vnet_be(q, sp);
1073
1074         case TUNSETVNETBE:
1075                 return tap_set_vnet_be(q, sp);
1076
1077         case TUNSETOFFLOAD:
1078                 /* let the user check for future flags */
1079                 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1080                             TUN_F_TSO_ECN | TUN_F_UFO))
1081                         return -EINVAL;
1082
1083                 rtnl_lock();
1084                 ret = set_offload(q, arg);
1085                 rtnl_unlock();
1086                 return ret;
1087
1088         case SIOCGIFHWADDR:
1089                 rtnl_lock();
1090                 tap = tap_get_tap_dev(q);
1091                 if (!tap) {
1092                         rtnl_unlock();
1093                         return -ENOLINK;
1094                 }
1095                 ret = 0;
1096                 u = tap->dev->type;
1097                 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) ||
1098                     copy_to_user(&ifr->ifr_hwaddr.sa_data, tap->dev->dev_addr, ETH_ALEN) ||
1099                     put_user(u, &ifr->ifr_hwaddr.sa_family))
1100                         ret = -EFAULT;
1101                 tap_put_tap_dev(tap);
1102                 rtnl_unlock();
1103                 return ret;
1104
1105         case SIOCSIFHWADDR:
1106                 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa)))
1107                         return -EFAULT;
1108                 rtnl_lock();
1109                 tap = tap_get_tap_dev(q);
1110                 if (!tap) {
1111                         rtnl_unlock();
1112                         return -ENOLINK;
1113                 }
1114                 ret = dev_set_mac_address(tap->dev, &sa);
1115                 tap_put_tap_dev(tap);
1116                 rtnl_unlock();
1117                 return ret;
1118
1119         default:
1120                 return -EINVAL;
1121         }
1122 }
1123
1124 #ifdef CONFIG_COMPAT
1125 static long tap_compat_ioctl(struct file *file, unsigned int cmd,
1126                              unsigned long arg)
1127 {
1128         return tap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
1129 }
1130 #endif
1131
1132 static const struct file_operations tap_fops = {
1133         .owner          = THIS_MODULE,
1134         .open           = tap_open,
1135         .release        = tap_release,
1136         .read_iter      = tap_read_iter,
1137         .write_iter     = tap_write_iter,
1138         .poll           = tap_poll,
1139         .llseek         = no_llseek,
1140         .unlocked_ioctl = tap_ioctl,
1141 #ifdef CONFIG_COMPAT
1142         .compat_ioctl   = tap_compat_ioctl,
1143 #endif
1144 };
1145
1146 static int tap_sendmsg(struct socket *sock, struct msghdr *m,
1147                        size_t total_len)
1148 {
1149         struct tap_queue *q = container_of(sock, struct tap_queue, sock);
1150         return tap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT);
1151 }
1152
1153 static int tap_recvmsg(struct socket *sock, struct msghdr *m,
1154                        size_t total_len, int flags)
1155 {
1156         struct tap_queue *q = container_of(sock, struct tap_queue, sock);
1157         struct sk_buff *skb = m->msg_control;
1158         int ret;
1159         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) {
1160                 if (skb)
1161                         kfree_skb(skb);
1162                 return -EINVAL;
1163         }
1164         ret = tap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT, skb);
1165         if (ret > total_len) {
1166                 m->msg_flags |= MSG_TRUNC;
1167                 ret = flags & MSG_TRUNC ? ret : total_len;
1168         }
1169         return ret;
1170 }
1171
1172 static int tap_peek_len(struct socket *sock)
1173 {
1174         struct tap_queue *q = container_of(sock, struct tap_queue,
1175                                                sock);
1176         return PTR_RING_PEEK_CALL(&q->ring, __skb_array_len_with_tag);
1177 }
1178
1179 /* Ops structure to mimic raw sockets with tun */
1180 static const struct proto_ops tap_socket_ops = {
1181         .sendmsg = tap_sendmsg,
1182         .recvmsg = tap_recvmsg,
1183         .peek_len = tap_peek_len,
1184 };
1185
1186 /* Get an underlying socket object from tun file.  Returns error unless file is
1187  * attached to a device.  The returned object works like a packet socket, it
1188  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1189  * holding a reference to the file for as long as the socket is in use. */
1190 struct socket *tap_get_socket(struct file *file)
1191 {
1192         struct tap_queue *q;
1193         if (file->f_op != &tap_fops)
1194                 return ERR_PTR(-EINVAL);
1195         q = file->private_data;
1196         if (!q)
1197                 return ERR_PTR(-EBADFD);
1198         return &q->sock;
1199 }
1200 EXPORT_SYMBOL_GPL(tap_get_socket);
1201
1202 struct ptr_ring *tap_get_ptr_ring(struct file *file)
1203 {
1204         struct tap_queue *q;
1205
1206         if (file->f_op != &tap_fops)
1207                 return ERR_PTR(-EINVAL);
1208         q = file->private_data;
1209         if (!q)
1210                 return ERR_PTR(-EBADFD);
1211         return &q->ring;
1212 }
1213 EXPORT_SYMBOL_GPL(tap_get_ptr_ring);
1214
1215 int tap_queue_resize(struct tap_dev *tap)
1216 {
1217         struct net_device *dev = tap->dev;
1218         struct tap_queue *q;
1219         struct ptr_ring **rings;
1220         int n = tap->numqueues;
1221         int ret, i = 0;
1222
1223         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
1224         if (!rings)
1225                 return -ENOMEM;
1226
1227         list_for_each_entry(q, &tap->queue_list, next)
1228                 rings[i++] = &q->ring;
1229
1230         ret = ptr_ring_resize_multiple(rings, n,
1231                                        dev->tx_queue_len, GFP_KERNEL,
1232                                        __skb_array_destroy_skb);
1233
1234         kfree(rings);
1235         return ret;
1236 }
1237 EXPORT_SYMBOL_GPL(tap_queue_resize);
1238
1239 static int tap_list_add(dev_t major, const char *device_name)
1240 {
1241         struct major_info *tap_major;
1242
1243         tap_major = kzalloc(sizeof(*tap_major), GFP_ATOMIC);
1244         if (!tap_major)
1245                 return -ENOMEM;
1246
1247         tap_major->major = MAJOR(major);
1248
1249         idr_init(&tap_major->minor_idr);
1250         spin_lock_init(&tap_major->minor_lock);
1251
1252         tap_major->device_name = device_name;
1253
1254         list_add_tail_rcu(&tap_major->next, &major_list);
1255         return 0;
1256 }
1257
1258 int tap_create_cdev(struct cdev *tap_cdev, dev_t *tap_major,
1259                     const char *device_name, struct module *module)
1260 {
1261         int err;
1262
1263         err = alloc_chrdev_region(tap_major, 0, TAP_NUM_DEVS, device_name);
1264         if (err)
1265                 goto out1;
1266
1267         cdev_init(tap_cdev, &tap_fops);
1268         tap_cdev->owner = module;
1269         err = cdev_add(tap_cdev, *tap_major, TAP_NUM_DEVS);
1270         if (err)
1271                 goto out2;
1272
1273         err =  tap_list_add(*tap_major, device_name);
1274         if (err)
1275                 goto out3;
1276
1277         return 0;
1278
1279 out3:
1280         cdev_del(tap_cdev);
1281 out2:
1282         unregister_chrdev_region(*tap_major, TAP_NUM_DEVS);
1283 out1:
1284         return err;
1285 }
1286 EXPORT_SYMBOL_GPL(tap_create_cdev);
1287
1288 void tap_destroy_cdev(dev_t major, struct cdev *tap_cdev)
1289 {
1290         struct major_info *tap_major, *tmp;
1291
1292         cdev_del(tap_cdev);
1293         unregister_chrdev_region(major, TAP_NUM_DEVS);
1294         list_for_each_entry_safe(tap_major, tmp, &major_list, next) {
1295                 if (tap_major->major == MAJOR(major)) {
1296                         idr_destroy(&tap_major->minor_idr);
1297                         list_del_rcu(&tap_major->next);
1298                         kfree_rcu(tap_major, rcu);
1299                 }
1300         }
1301 }
1302 EXPORT_SYMBOL_GPL(tap_destroy_cdev);
1303
1304 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1305 MODULE_AUTHOR("Sainath Grandhi <sainath.grandhi@intel.com>");
1306 MODULE_LICENSE("GPL");