SAFE public projects git trees. - safe/jmp/linux-2.6/blob - drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use random_ether_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define DRV_NAME        "tun"
  38 #define DRV_VERSION     "1.6"
  39 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  40 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  41
  42 #include <linux/module.h>
  43 #include <linux/errno.h>
  44 #include <linux/kernel.h>
  45 #include <linux/major.h>
  46 #include <linux/slab.h>
  47 #include <linux/poll.h>
  48 #include <linux/fcntl.h>
  49 #include <linux/init.h>
  50 #include <linux/skbuff.h>
  51 #include <linux/netdevice.h>
  52 #include <linux/etherdevice.h>
  53 #include <linux/miscdevice.h>
  54 #include <linux/ethtool.h>
  55 #include <linux/rtnetlink.h>
  56 #include <linux/compat.h>
  57 #include <linux/if.h>
  58 #include <linux/if_arp.h>
  59 #include <linux/if_ether.h>
  60 #include <linux/if_tun.h>
  61 #include <linux/crc32.h>
  62 #include <linux/nsproxy.h>
  63 #include <linux/virtio_net.h>
  64 #include <net/net_namespace.h>
  65 #include <net/netns/generic.h>
  66 #include <net/rtnetlink.h>
  67 #include <net/sock.h>
  68
  69 #include <asm/system.h>
  70 #include <asm/uaccess.h>
  71
  72 /* Uncomment to enable debugging */
  73 /* #define TUN_DEBUG 1 */
  74
  75 #ifdef TUN_DEBUG
  76 static int debug;
  77
  78 #define DBG  if(tun->debug)printk
  79 #define DBG1 if(debug==2)printk
  80 #else
  81 #define DBG( a... )
  82 #define DBG1( a... )
  83 #endif
  84
  85 #define FLT_EXACT_COUNT 8
  86 struct tap_filter {
  87         unsigned int    count;    /* Number of addrs. Zero means disabled */
  88         u32             mask[2];  /* Mask of the hashed addrs */
  89         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
  90 };
  91
  92 struct tun_file {
  93         atomic_t count;
  94         struct tun_struct *tun;
  95         struct net *net;
  96 };
  97
  98 struct tun_sock;
  99
 100 struct tun_struct {
 101         struct tun_file         *tfile;
 102         unsigned int            flags;
 103         uid_t                   owner;
 104         gid_t                   group;
 105
 106         struct net_device       *dev;
 107         struct fasync_struct    *fasync;
 108
 109         struct tap_filter       txflt;
 110         struct socket           socket;
 111
 112 #ifdef TUN_DEBUG
 113         int debug;
 114 #endif
 115 };
 116
 117 struct tun_sock {
 118         struct sock             sk;
 119         struct tun_struct       *tun;
 120 };
 121
 122 static inline struct tun_sock *tun_sk(struct sock *sk)
 123 {
 124         return container_of(sk, struct tun_sock, sk);
 125 }
 126
 127 static int tun_attach(struct tun_struct *tun, struct file *file)
 128 {
 129         struct tun_file *tfile = file->private_data;
 130         int err;
 131
 132         ASSERT_RTNL();
 133
 134         netif_tx_lock_bh(tun->dev);
 135
 136         err = -EINVAL;
 137         if (tfile->tun)
 138                 goto out;
 139
 140         err = -EBUSY;
 141         if (tun->tfile)
 142                 goto out;
 143
 144         err = 0;
 145         tfile->tun = tun;
 146         tun->tfile = tfile;
 147         tun->socket.file = file;
 148         dev_hold(tun->dev);
 149         sock_hold(tun->socket.sk);
 150         atomic_inc(&tfile->count);
 151
 152 out:
 153         netif_tx_unlock_bh(tun->dev);
 154         return err;
 155 }
 156
 157 static void __tun_detach(struct tun_struct *tun)
 158 {
 159         /* Detach from net device */
 160         netif_tx_lock_bh(tun->dev);
 161         tun->tfile = NULL;
 162         tun->socket.file = NULL;
 163         netif_tx_unlock_bh(tun->dev);
 164
 165         /* Drop read queue */
 166         skb_queue_purge(&tun->socket.sk->sk_receive_queue);
 167
 168         /* Drop the extra count on the net device */
 169         dev_put(tun->dev);
 170 }
 171
 172 static void tun_detach(struct tun_struct *tun)
 173 {
 174         rtnl_lock();
 175         __tun_detach(tun);
 176         rtnl_unlock();
 177 }
 178
 179 static struct tun_struct *__tun_get(struct tun_file *tfile)
 180 {
 181         struct tun_struct *tun = NULL;
 182
 183         if (atomic_inc_not_zero(&tfile->count))
 184                 tun = tfile->tun;
 185
 186         return tun;
 187 }
 188
 189 static struct tun_struct *tun_get(struct file *file)
 190 {
 191         return __tun_get(file->private_data);
 192 }
 193
 194 static void tun_put(struct tun_struct *tun)
 195 {
 196         struct tun_file *tfile = tun->tfile;
 197
 198         if (atomic_dec_and_test(&tfile->count))
 199                 tun_detach(tfile->tun);
 200 }
 201
 202 /* TAP filterting */
 203 static void addr_hash_set(u32 *mask, const u8 *addr)
 204 {
 205         int n = ether_crc(ETH_ALEN, addr) >> 26;
 206         mask[n >> 5] |= (1 << (n & 31));
 207 }
 208
 209 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 210 {
 211         int n = ether_crc(ETH_ALEN, addr) >> 26;
 212         return mask[n >> 5] & (1 << (n & 31));
 213 }
 214
 215 static int update_filter(struct tap_filter *filter, void __user *arg)
 216 {
 217         struct { u8 u[ETH_ALEN]; } *addr;
 218         struct tun_filter uf;
 219         int err, alen, n, nexact;
 220
 221         if (copy_from_user(&uf, arg, sizeof(uf)))
 222                 return -EFAULT;
 223
 224         if (!uf.count) {
 225                 /* Disabled */
 226                 filter->count = 0;
 227                 return 0;
 228         }
 229
 230         alen = ETH_ALEN * uf.count;
 231         addr = kmalloc(alen, GFP_KERNEL);
 232         if (!addr)
 233                 return -ENOMEM;
 234
 235         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 236                 err = -EFAULT;
 237                 goto done;
 238         }
 239
 240         /* The filter is updated without holding any locks. Which is
 241          * perfectly safe. We disable it first and in the worst
 242          * case we'll accept a few undesired packets. */
 243         filter->count = 0;
 244         wmb();
 245
 246         /* Use first set of addresses as an exact filter */
 247         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 248                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 249
 250         nexact = n;
 251
 252         /* Remaining multicast addresses are hashed,
 253          * unicast will leave the filter disabled. */
 254         memset(filter->mask, 0, sizeof(filter->mask));
 255         for (; n < uf.count; n++) {
 256                 if (!is_multicast_ether_addr(addr[n].u)) {
 257                         err = 0; /* no filter */
 258                         goto done;
 259                 }
 260                 addr_hash_set(filter->mask, addr[n].u);
 261         }
 262
 263         /* For ALLMULTI just set the mask to all ones.
 264          * This overrides the mask populated above. */
 265         if ((uf.flags & TUN_FLT_ALLMULTI))
 266                 memset(filter->mask, ~0, sizeof(filter->mask));
 267
 268         /* Now enable the filter */
 269         wmb();
 270         filter->count = nexact;
 271
 272         /* Return the number of exact filters */
 273         err = nexact;
 274
 275 done:
 276         kfree(addr);
 277         return err;
 278 }
 279
 280 /* Returns: 0 - drop, !=0 - accept */
 281 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 282 {
 283         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 284          * at this point. */
 285         struct ethhdr *eh = (struct ethhdr *) skb->data;
 286         int i;
 287
 288         /* Exact match */
 289         for (i = 0; i < filter->count; i++)
 290                 if (!compare_ether_addr(eh->h_dest, filter->addr[i]))
 291                         return 1;
 292
 293         /* Inexact match (multicast only) */
 294         if (is_multicast_ether_addr(eh->h_dest))
 295                 return addr_hash_test(filter->mask, eh->h_dest);
 296
 297         return 0;
 298 }
 299
 300 /*
 301  * Checks whether the packet is accepted or not.
 302  * Returns: 0 - drop, !=0 - accept
 303  */
 304 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 305 {
 306         if (!filter->count)
 307                 return 1;
 308
 309         return run_filter(filter, skb);
 310 }
 311
 312 /* Network device part of the driver */
 313
 314 static const struct ethtool_ops tun_ethtool_ops;
 315
 316 /* Net device detach from fd. */
 317 static void tun_net_uninit(struct net_device *dev)
 318 {
 319         struct tun_struct *tun = netdev_priv(dev);
 320         struct tun_file *tfile = tun->tfile;
 321
 322         /* Inform the methods they need to stop using the dev.
 323          */
 324         if (tfile) {
 325                 wake_up_all(&tun->socket.wait);
 326                 if (atomic_dec_and_test(&tfile->count))
 327                         __tun_detach(tun);
 328         }
 329 }
 330
 331 static void tun_free_netdev(struct net_device *dev)
 332 {
 333         struct tun_struct *tun = netdev_priv(dev);
 334
 335         sock_put(tun->socket.sk);
 336 }
 337
 338 /* Net device open. */
 339 static int tun_net_open(struct net_device *dev)
 340 {
 341         netif_start_queue(dev);
 342         return 0;
 343 }
 344
 345 /* Net device close. */
 346 static int tun_net_close(struct net_device *dev)
 347 {
 348         netif_stop_queue(dev);
 349         return 0;
 350 }
 351
 352 /* Net device start xmit */
 353 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 354 {
 355         struct tun_struct *tun = netdev_priv(dev);
 356
 357         DBG(KERN_INFO "%s: tun_net_xmit %d\n", tun->dev->name, skb->len);
 358
 359         /* Drop packet if interface is not attached */
 360         if (!tun->tfile)
 361                 goto drop;
 362
 363         /* Drop if the filter does not like it.
 364          * This is a noop if the filter is disabled.
 365          * Filter can be enabled only for the TAP devices. */
 366         if (!check_filter(&tun->txflt, skb))
 367                 goto drop;
 368
 369         if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) {
 370                 if (!(tun->flags & TUN_ONE_QUEUE)) {
 371                         /* Normal queueing mode. */
 372                         /* Packet scheduler handles dropping of further packets. */
 373                         netif_stop_queue(dev);
 374
 375                         /* We won't see all dropped packets individually, so overrun
 376                          * error is more appropriate. */
 377                         dev->stats.tx_fifo_errors++;
 378                 } else {
 379                         /* Single queue mode.
 380                          * Driver handles dropping of all packets itself. */
 381                         goto drop;
 382                 }
 383         }
 384
 385         /* Enqueue packet */
 386         skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb);
 387         dev->trans_start = jiffies;
 388
 389         /* Notify and wake up reader process */
 390         if (tun->flags & TUN_FASYNC)
 391                 kill_fasync(&tun->fasync, SIGIO, POLL_IN);
 392         wake_up_interruptible_poll(&tun->socket.wait, POLLIN |
 393                                    POLLRDNORM | POLLRDBAND);
 394         return NETDEV_TX_OK;
 395
 396 drop:
 397         dev->stats.tx_dropped++;
 398         kfree_skb(skb);
 399         return NETDEV_TX_OK;
 400 }
 401
 402 static void tun_net_mclist(struct net_device *dev)
 403 {
 404         /*
 405          * This callback is supposed to deal with mc filter in
 406          * _rx_ path and has nothing to do with the _tx_ path.
 407          * In rx path we always accept everything userspace gives us.
 408          */
 409         return;
 410 }
 411
 412 #define MIN_MTU 68
 413 #define MAX_MTU 65535
 414
 415 static int
 416 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 417 {
 418         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 419                 return -EINVAL;
 420         dev->mtu = new_mtu;
 421         return 0;
 422 }
 423
 424 static const struct net_device_ops tun_netdev_ops = {
 425         .ndo_uninit             = tun_net_uninit,
 426         .ndo_open               = tun_net_open,
 427         .ndo_stop               = tun_net_close,
 428         .ndo_start_xmit         = tun_net_xmit,
 429         .ndo_change_mtu         = tun_net_change_mtu,
 430 };
 431
 432 static const struct net_device_ops tap_netdev_ops = {
 433         .ndo_uninit             = tun_net_uninit,
 434         .ndo_open               = tun_net_open,
 435         .ndo_stop               = tun_net_close,
 436         .ndo_start_xmit         = tun_net_xmit,
 437         .ndo_change_mtu         = tun_net_change_mtu,
 438         .ndo_set_multicast_list = tun_net_mclist,
 439         .ndo_set_mac_address    = eth_mac_addr,
 440         .ndo_validate_addr      = eth_validate_addr,
 441 };
 442
 443 /* Initialize net device. */
 444 static void tun_net_init(struct net_device *dev)
 445 {
 446         struct tun_struct *tun = netdev_priv(dev);
 447
 448         switch (tun->flags & TUN_TYPE_MASK) {
 449         case TUN_TUN_DEV:
 450                 dev->netdev_ops = &tun_netdev_ops;
 451
 452                 /* Point-to-Point TUN Device */
 453                 dev->hard_header_len = 0;
 454                 dev->addr_len = 0;
 455                 dev->mtu = 1500;
 456
 457                 /* Zero header length */
 458                 dev->type = ARPHRD_NONE;
 459                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 460                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 461                 break;
 462
 463         case TUN_TAP_DEV:
 464                 dev->netdev_ops = &tap_netdev_ops;
 465                 /* Ethernet TAP Device */
 466                 ether_setup(dev);
 467
 468                 random_ether_addr(dev->dev_addr);
 469
 470                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 471                 break;
 472         }
 473 }
 474
 475 /* Character device part */
 476
 477 /* Poll */
 478 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
 479 {
 480         struct tun_file *tfile = file->private_data;
 481         struct tun_struct *tun = __tun_get(tfile);
 482         struct sock *sk;
 483         unsigned int mask = 0;
 484
 485         if (!tun)
 486                 return POLLERR;
 487
 488         sk = tun->socket.sk;
 489
 490         DBG(KERN_INFO "%s: tun_chr_poll\n", tun->dev->name);
 491
 492         poll_wait(file, &tun->socket.wait, wait);
 493
 494         if (!skb_queue_empty(&sk->sk_receive_queue))
 495                 mask |= POLLIN | POLLRDNORM;
 496
 497         if (sock_writeable(sk) ||
 498             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 499              sock_writeable(sk)))
 500                 mask |= POLLOUT | POLLWRNORM;
 501
 502         if (tun->dev->reg_state != NETREG_REGISTERED)
 503                 mask = POLLERR;
 504
 505         tun_put(tun);
 506         return mask;
 507 }
 508
 509 /* prepad is the amount to reserve at front.  len is length after that.
 510  * linear is a hint as to how much to copy (usually headers). */
 511 static inline struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
 512                                             size_t prepad, size_t len,
 513                                             size_t linear, int noblock)
 514 {
 515         struct sock *sk = tun->socket.sk;
 516         struct sk_buff *skb;
 517         int err;
 518
 519         /* Under a page?  Don't bother with paged skb. */
 520         if (prepad + len < PAGE_SIZE || !linear)
 521                 linear = len;
 522
 523         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 524                                    &err);
 525         if (!skb)
 526                 return ERR_PTR(err);
 527
 528         skb_reserve(skb, prepad);
 529         skb_put(skb, linear);
 530         skb->data_len = len - linear;
 531         skb->len += len - linear;
 532
 533         return skb;
 534 }
 535
 536 /* Get packet from user space buffer */
 537 static __inline__ ssize_t tun_get_user(struct tun_struct *tun,
 538                                        const struct iovec *iv, size_t count,
 539                                        int noblock)
 540 {
 541         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
 542         struct sk_buff *skb;
 543         size_t len = count, align = 0;
 544         struct virtio_net_hdr gso = { 0 };
 545         int offset = 0;
 546
 547         if (!(tun->flags & TUN_NO_PI)) {
 548                 if ((len -= sizeof(pi)) > count)
 549                         return -EINVAL;
 550
 551                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
 552                         return -EFAULT;
 553                 offset += sizeof(pi);
 554         }
 555
 556         if (tun->flags & TUN_VNET_HDR) {
 557                 if ((len -= sizeof(gso)) > count)
 558                         return -EINVAL;
 559
 560                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
 561                         return -EFAULT;
 562
 563                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
 564                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
 565                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
 566
 567                 if (gso.hdr_len > len)
 568                         return -EINVAL;
 569                 offset += sizeof(gso);
 570         }
 571
 572         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
 573                 align = NET_IP_ALIGN;
 574                 if (unlikely(len < ETH_HLEN ||
 575                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
 576                         return -EINVAL;
 577         }
 578
 579         skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock);
 580         if (IS_ERR(skb)) {
 581                 if (PTR_ERR(skb) != -EAGAIN)
 582                         tun->dev->stats.rx_dropped++;
 583                 return PTR_ERR(skb);
 584         }
 585
 586         if (skb_copy_datagram_from_iovec(skb, 0, iv, offset, len)) {
 587                 tun->dev->stats.rx_dropped++;
 588                 kfree_skb(skb);
 589                 return -EFAULT;
 590         }
 591
 592         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 593                 if (!skb_partial_csum_set(skb, gso.csum_start,
 594                                           gso.csum_offset)) {
 595                         tun->dev->stats.rx_frame_errors++;
 596                         kfree_skb(skb);
 597                         return -EINVAL;
 598                 }
 599         } else if (tun->flags & TUN_NOCHECKSUM)
 600                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 601
 602         switch (tun->flags & TUN_TYPE_MASK) {
 603         case TUN_TUN_DEV:
 604                 if (tun->flags & TUN_NO_PI) {
 605                         switch (skb->data[0] & 0xf0) {
 606                         case 0x40:
 607                                 pi.proto = htons(ETH_P_IP);
 608                                 break;
 609                         case 0x60:
 610                                 pi.proto = htons(ETH_P_IPV6);
 611                                 break;
 612                         default:
 613                                 tun->dev->stats.rx_dropped++;
 614                                 kfree_skb(skb);
 615                                 return -EINVAL;
 616                         }
 617                 }
 618
 619                 skb_reset_mac_header(skb);
 620                 skb->protocol = pi.proto;
 621                 skb->dev = tun->dev;
 622                 break;
 623         case TUN_TAP_DEV:
 624                 skb->protocol = eth_type_trans(skb, tun->dev);
 625                 break;
 626         };
 627
 628         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 629                 pr_debug("GSO!\n");
 630                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 631                 case VIRTIO_NET_HDR_GSO_TCPV4:
 632                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 633                         break;
 634                 case VIRTIO_NET_HDR_GSO_TCPV6:
 635                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 636                         break;
 637                 case VIRTIO_NET_HDR_GSO_UDP:
 638                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 639                         break;
 640                 default:
 641                         tun->dev->stats.rx_frame_errors++;
 642                         kfree_skb(skb);
 643                         return -EINVAL;
 644                 }
 645
 646                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
 647                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 648
 649                 skb_shinfo(skb)->gso_size = gso.gso_size;
 650                 if (skb_shinfo(skb)->gso_size == 0) {
 651                         tun->dev->stats.rx_frame_errors++;
 652                         kfree_skb(skb);
 653                         return -EINVAL;
 654                 }
 655
 656                 /* Header must be checked, and gso_segs computed. */
 657                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 658                 skb_shinfo(skb)->gso_segs = 0;
 659         }
 660
 661         netif_rx_ni(skb);
 662
 663         tun->dev->stats.rx_packets++;
 664         tun->dev->stats.rx_bytes += len;
 665
 666         return count;
 667 }
 668
 669 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
 670                               unsigned long count, loff_t pos)
 671 {
 672         struct file *file = iocb->ki_filp;
 673         struct tun_struct *tun = tun_get(file);
 674         ssize_t result;
 675
 676         if (!tun)
 677                 return -EBADFD;
 678
 679         DBG(KERN_INFO "%s: tun_chr_write %ld\n", tun->dev->name, count);
 680
 681         result = tun_get_user(tun, iv, iov_length(iv, count),
 682                               file->f_flags & O_NONBLOCK);
 683
 684         tun_put(tun);
 685         return result;
 686 }
 687
 688 /* Put packet to the user space buffer */
 689 static __inline__ ssize_t tun_put_user(struct tun_struct *tun,
 690                                        struct sk_buff *skb,
 691                                        const struct iovec *iv, int len)
 692 {
 693         struct tun_pi pi = { 0, skb->protocol };
 694         ssize_t total = 0;
 695
 696         if (!(tun->flags & TUN_NO_PI)) {
 697                 if ((len -= sizeof(pi)) < 0)
 698                         return -EINVAL;
 699
 700                 if (len < skb->len) {
 701                         /* Packet will be striped */
 702                         pi.flags |= TUN_PKT_STRIP;
 703                 }
 704
 705                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
 706                         return -EFAULT;
 707                 total += sizeof(pi);
 708         }
 709
 710         if (tun->flags & TUN_VNET_HDR) {
 711                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
 712                 if ((len -= sizeof(gso)) < 0)
 713                         return -EINVAL;
 714
 715                 if (skb_is_gso(skb)) {
 716                         struct skb_shared_info *sinfo = skb_shinfo(skb);
 717
 718                         /* This is a hint as to how much should be linear. */
 719                         gso.hdr_len = skb_headlen(skb);
 720                         gso.gso_size = sinfo->gso_size;
 721                         if (sinfo->gso_type & SKB_GSO_TCPV4)
 722                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 723                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
 724                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 725                         else if (sinfo->gso_type & SKB_GSO_UDP)
 726                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
 727                         else
 728                                 BUG();
 729                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
 730                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 731                 } else
 732                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
 733
 734                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 735                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 736                         gso.csum_start = skb->csum_start - skb_headroom(skb);
 737                         gso.csum_offset = skb->csum_offset;
 738                 } /* else everything is zero */
 739
 740                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
 741                                                sizeof(gso))))
 742                         return -EFAULT;
 743                 total += sizeof(gso);
 744         }
 745
 746         len = min_t(int, skb->len, len);
 747
 748         skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
 749         total += skb->len;
 750
 751         tun->dev->stats.tx_packets++;
 752         tun->dev->stats.tx_bytes += len;
 753
 754         return total;
 755 }
 756
 757 static ssize_t tun_do_read(struct tun_struct *tun,
 758                            struct kiocb *iocb, const struct iovec *iv,
 759                            ssize_t len, int noblock)
 760 {
 761         DECLARE_WAITQUEUE(wait, current);
 762         struct sk_buff *skb;
 763         ssize_t ret = 0;
 764
 765         DBG(KERN_INFO "%s: tun_chr_read\n", tun->dev->name);
 766
 767         add_wait_queue(&tun->socket.wait, &wait);
 768         while (len) {
 769                 current->state = TASK_INTERRUPTIBLE;
 770
 771                 /* Read frames from the queue */
 772                 if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
 773                         if (noblock) {
 774                                 ret = -EAGAIN;
 775                                 break;
 776                         }
 777                         if (signal_pending(current)) {
 778                                 ret = -ERESTARTSYS;
 779                                 break;
 780                         }
 781                         if (tun->dev->reg_state != NETREG_REGISTERED) {
 782                                 ret = -EIO;
 783                                 break;
 784                         }
 785
 786                         /* Nothing to read, let's sleep */
 787                         schedule();
 788                         continue;
 789                 }
 790                 netif_wake_queue(tun->dev);
 791
 792                 ret = tun_put_user(tun, skb, iv, len);
 793                 kfree_skb(skb);
 794                 break;
 795         }
 796
 797         current->state = TASK_RUNNING;
 798         remove_wait_queue(&tun->socket.wait, &wait);
 799
 800         return ret;
 801 }
 802
 803 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
 804                             unsigned long count, loff_t pos)
 805 {
 806         struct file *file = iocb->ki_filp;
 807         struct tun_file *tfile = file->private_data;
 808         struct tun_struct *tun = __tun_get(tfile);
 809         ssize_t len, ret;
 810
 811         if (!tun)
 812                 return -EBADFD;
 813         len = iov_length(iv, count);
 814         if (len < 0) {
 815                 ret = -EINVAL;
 816                 goto out;
 817         }
 818
 819         ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK);
 820         ret = min_t(ssize_t, ret, len);
 821 out:
 822         tun_put(tun);
 823         return ret;
 824 }
 825
 826 static void tun_setup(struct net_device *dev)
 827 {
 828         struct tun_struct *tun = netdev_priv(dev);
 829
 830         tun->owner = -1;
 831         tun->group = -1;
 832
 833         dev->ethtool_ops = &tun_ethtool_ops;
 834         dev->destructor = tun_free_netdev;
 835 }
 836
 837 /* Trivial set of netlink ops to allow deleting tun or tap
 838  * device with netlink.
 839  */
 840 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
 841 {
 842         return -EINVAL;
 843 }
 844
 845 static struct rtnl_link_ops tun_link_ops __read_mostly = {
 846         .kind           = DRV_NAME,
 847         .priv_size      = sizeof(struct tun_struct),
 848         .setup          = tun_setup,
 849         .validate       = tun_validate,
 850 };
 851
 852 static void tun_sock_write_space(struct sock *sk)
 853 {
 854         struct tun_struct *tun;
 855
 856         if (!sock_writeable(sk))
 857                 return;
 858
 859         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
 860                 return;
 861
 862         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
 863                 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
 864                                                 POLLWRNORM | POLLWRBAND);
 865
 866         tun = tun_sk(sk)->tun;
 867         kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
 868 }
 869
 870 static void tun_sock_destruct(struct sock *sk)
 871 {
 872         free_netdev(tun_sk(sk)->tun->dev);
 873 }
 874
 875 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
 876                        struct msghdr *m, size_t total_len)
 877 {
 878         struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
 879         return tun_get_user(tun, m->msg_iov, total_len,
 880                             m->msg_flags & MSG_DONTWAIT);
 881 }
 882
 883 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
 884                        struct msghdr *m, size_t total_len,
 885                        int flags)
 886 {
 887         struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
 888         int ret;
 889         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
 890                 return -EINVAL;
 891         ret = tun_do_read(tun, iocb, m->msg_iov, total_len,
 892                           flags & MSG_DONTWAIT);
 893         if (ret > total_len) {
 894                 m->msg_flags |= MSG_TRUNC;
 895                 ret = flags & MSG_TRUNC ? ret : total_len;
 896         }
 897         return ret;
 898 }
 899
 900 /* Ops structure to mimic raw sockets with tun */
 901 static const struct proto_ops tun_socket_ops = {
 902         .sendmsg = tun_sendmsg,
 903         .recvmsg = tun_recvmsg,
 904 };
 905
 906 static struct proto tun_proto = {
 907         .name           = "tun",
 908         .owner          = THIS_MODULE,
 909         .obj_size       = sizeof(struct tun_sock),
 910 };
 911
 912 static int tun_flags(struct tun_struct *tun)
 913 {
 914         int flags = 0;
 915
 916         if (tun->flags & TUN_TUN_DEV)
 917                 flags |= IFF_TUN;
 918         else
 919                 flags |= IFF_TAP;
 920
 921         if (tun->flags & TUN_NO_PI)
 922                 flags |= IFF_NO_PI;
 923
 924         if (tun->flags & TUN_ONE_QUEUE)
 925                 flags |= IFF_ONE_QUEUE;
 926
 927         if (tun->flags & TUN_VNET_HDR)
 928                 flags |= IFF_VNET_HDR;
 929
 930         return flags;
 931 }
 932
 933 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
 934                               char *buf)
 935 {
 936         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
 937         return sprintf(buf, "0x%x\n", tun_flags(tun));
 938 }
 939
 940 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
 941                               char *buf)
 942 {
 943         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
 944         return sprintf(buf, "%d\n", tun->owner);
 945 }
 946
 947 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
 948                               char *buf)
 949 {
 950         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
 951         return sprintf(buf, "%d\n", tun->group);
 952 }
 953
 954 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
 955 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
 956 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
 957
 958 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
 959 {
 960         struct sock *sk;
 961         struct tun_struct *tun;
 962         struct net_device *dev;
 963         int err;
 964
 965         dev = __dev_get_by_name(net, ifr->ifr_name);
 966         if (dev) {
 967                 const struct cred *cred = current_cred();
 968
 969                 if (ifr->ifr_flags & IFF_TUN_EXCL)
 970                         return -EBUSY;
 971                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
 972                         tun = netdev_priv(dev);
 973                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
 974                         tun = netdev_priv(dev);
 975                 else
 976                         return -EINVAL;
 977
 978                 if (((tun->owner != -1 && cred->euid != tun->owner) ||
 979                      (tun->group != -1 && !in_egroup_p(tun->group))) &&
 980                     !capable(CAP_NET_ADMIN))
 981                         return -EPERM;
 982                 err = security_tun_dev_attach(tun->socket.sk);
 983                 if (err < 0)
 984                         return err;
 985
 986                 err = tun_attach(tun, file);
 987                 if (err < 0)
 988                         return err;
 989         }
 990         else {
 991                 char *name;
 992                 unsigned long flags = 0;
 993
 994                 if (!capable(CAP_NET_ADMIN))
 995                         return -EPERM;
 996                 err = security_tun_dev_create();
 997                 if (err < 0)
 998                         return err;
 999
1000                 /* Set dev type */
1001                 if (ifr->ifr_flags & IFF_TUN) {
1002                         /* TUN device */
1003                         flags |= TUN_TUN_DEV;
1004                         name = "tun%d";
1005                 } else if (ifr->ifr_flags & IFF_TAP) {
1006                         /* TAP device */
1007                         flags |= TUN_TAP_DEV;
1008                         name = "tap%d";
1009                 } else
1010                         return -EINVAL;
1011
1012                 if (*ifr->ifr_name)
1013                         name = ifr->ifr_name;
1014
1015                 dev = alloc_netdev(sizeof(struct tun_struct), name,
1016                                    tun_setup);
1017                 if (!dev)
1018                         return -ENOMEM;
1019
1020                 dev_net_set(dev, net);
1021                 dev->rtnl_link_ops = &tun_link_ops;
1022
1023                 tun = netdev_priv(dev);
1024                 tun->dev = dev;
1025                 tun->flags = flags;
1026                 tun->txflt.count = 0;
1027
1028                 err = -ENOMEM;
1029                 sk = sk_alloc(net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1030                 if (!sk)
1031                         goto err_free_dev;
1032
1033                 init_waitqueue_head(&tun->socket.wait);
1034                 tun->socket.ops = &tun_socket_ops;
1035                 sock_init_data(&tun->socket, sk);
1036                 sk->sk_write_space = tun_sock_write_space;
1037                 sk->sk_sndbuf = INT_MAX;
1038
1039                 tun_sk(sk)->tun = tun;
1040
1041                 security_tun_dev_post_create(sk);
1042
1043                 tun_net_init(dev);
1044
1045                 if (strchr(dev->name, '%')) {
1046                         err = dev_alloc_name(dev, dev->name);
1047                         if (err < 0)
1048                                 goto err_free_sk;
1049                 }
1050
1051                 err = register_netdevice(tun->dev);
1052                 if (err < 0)
1053                         goto err_free_sk;
1054
1055                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1056                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1057                     device_create_file(&tun->dev->dev, &dev_attr_group))
1058                         printk(KERN_ERR "Failed to create tun sysfs files\n");
1059
1060                 sk->sk_destruct = tun_sock_destruct;
1061
1062                 err = tun_attach(tun, file);
1063                 if (err < 0)
1064                         goto failed;
1065         }
1066
1067         DBG(KERN_INFO "%s: tun_set_iff\n", tun->dev->name);
1068
1069         if (ifr->ifr_flags & IFF_NO_PI)
1070                 tun->flags |= TUN_NO_PI;
1071         else
1072                 tun->flags &= ~TUN_NO_PI;
1073
1074         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1075                 tun->flags |= TUN_ONE_QUEUE;
1076         else
1077                 tun->flags &= ~TUN_ONE_QUEUE;
1078
1079         if (ifr->ifr_flags & IFF_VNET_HDR)
1080                 tun->flags |= TUN_VNET_HDR;
1081         else
1082                 tun->flags &= ~TUN_VNET_HDR;
1083
1084         /* Make sure persistent devices do not get stuck in
1085          * xoff state.
1086          */
1087         if (netif_running(tun->dev))
1088                 netif_wake_queue(tun->dev);
1089
1090         strcpy(ifr->ifr_name, tun->dev->name);
1091         return 0;
1092
1093  err_free_sk:
1094         sock_put(sk);
1095  err_free_dev:
1096         free_netdev(dev);
1097  failed:
1098         return err;
1099 }
1100
1101 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1102                        struct ifreq *ifr)
1103 {
1104         DBG(KERN_INFO "%s: tun_get_iff\n", tun->dev->name);
1105
1106         strcpy(ifr->ifr_name, tun->dev->name);
1107
1108         ifr->ifr_flags = tun_flags(tun);
1109
1110         return 0;
1111 }
1112
1113 /* This is like a cut-down ethtool ops, except done via tun fd so no
1114  * privs required. */
1115 static int set_offload(struct net_device *dev, unsigned long arg)
1116 {
1117         unsigned int old_features, features;
1118
1119         old_features = dev->features;
1120         /* Unset features, set them as we chew on the arg. */
1121         features = (old_features & ~(NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST
1122                                     |NETIF_F_TSO_ECN|NETIF_F_TSO|NETIF_F_TSO6
1123                                     |NETIF_F_UFO));
1124
1125         if (arg & TUN_F_CSUM) {
1126                 features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1127                 arg &= ~TUN_F_CSUM;
1128
1129                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1130                         if (arg & TUN_F_TSO_ECN) {
1131                                 features |= NETIF_F_TSO_ECN;
1132                                 arg &= ~TUN_F_TSO_ECN;
1133                         }
1134                         if (arg & TUN_F_TSO4)
1135                                 features |= NETIF_F_TSO;
1136                         if (arg & TUN_F_TSO6)
1137                                 features |= NETIF_F_TSO6;
1138                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1139                 }
1140
1141                 if (arg & TUN_F_UFO) {
1142                         features |= NETIF_F_UFO;
1143                         arg &= ~TUN_F_UFO;
1144                 }
1145         }
1146
1147         /* This gives the user a way to test for new features in future by
1148          * trying to set them. */
1149         if (arg)
1150                 return -EINVAL;
1151
1152         dev->features = features;
1153         if (old_features != dev->features)
1154                 netdev_features_change(dev);
1155
1156         return 0;
1157 }
1158
1159 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1160                             unsigned long arg, int ifreq_len)
1161 {
1162         struct tun_file *tfile = file->private_data;
1163         struct tun_struct *tun;
1164         void __user* argp = (void __user*)arg;
1165         struct ifreq ifr;
1166         int sndbuf;
1167         int ret;
1168
1169         if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
1170                 if (copy_from_user(&ifr, argp, ifreq_len))
1171                         return -EFAULT;
1172
1173         if (cmd == TUNGETFEATURES) {
1174                 /* Currently this just means: "what IFF flags are valid?".
1175                  * This is needed because we never checked for invalid flags on
1176                  * TUNSETIFF. */
1177                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1178                                 IFF_VNET_HDR,
1179                                 (unsigned int __user*)argp);
1180         }
1181
1182         rtnl_lock();
1183
1184         tun = __tun_get(tfile);
1185         if (cmd == TUNSETIFF && !tun) {
1186                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1187
1188                 ret = tun_set_iff(tfile->net, file, &ifr);
1189
1190                 if (ret)
1191                         goto unlock;
1192
1193                 if (copy_to_user(argp, &ifr, ifreq_len))
1194                         ret = -EFAULT;
1195                 goto unlock;
1196         }
1197
1198         ret = -EBADFD;
1199         if (!tun)
1200                 goto unlock;
1201
1202         DBG(KERN_INFO "%s: tun_chr_ioctl cmd %d\n", tun->dev->name, cmd);
1203
1204         ret = 0;
1205         switch (cmd) {
1206         case TUNGETIFF:
1207                 ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1208                 if (ret)
1209                         break;
1210
1211                 if (copy_to_user(argp, &ifr, ifreq_len))
1212                         ret = -EFAULT;
1213                 break;
1214
1215         case TUNSETNOCSUM:
1216                 /* Disable/Enable checksum */
1217                 if (arg)
1218                         tun->flags |= TUN_NOCHECKSUM;
1219                 else
1220                         tun->flags &= ~TUN_NOCHECKSUM;
1221
1222                 DBG(KERN_INFO "%s: checksum %s\n",
1223                     tun->dev->name, arg ? "disabled" : "enabled");
1224                 break;
1225
1226         case TUNSETPERSIST:
1227                 /* Disable/Enable persist mode */
1228                 if (arg)
1229                         tun->flags |= TUN_PERSIST;
1230                 else
1231                         tun->flags &= ~TUN_PERSIST;
1232
1233                 DBG(KERN_INFO "%s: persist %s\n",
1234                     tun->dev->name, arg ? "enabled" : "disabled");
1235                 break;
1236
1237         case TUNSETOWNER:
1238                 /* Set owner of the device */
1239                 tun->owner = (uid_t) arg;
1240
1241                 DBG(KERN_INFO "%s: owner set to %d\n", tun->dev->name, tun->owner);
1242                 break;
1243
1244         case TUNSETGROUP:
1245                 /* Set group of the device */
1246                 tun->group= (gid_t) arg;
1247
1248                 DBG(KERN_INFO "%s: group set to %d\n", tun->dev->name, tun->group);
1249                 break;
1250
1251         case TUNSETLINK:
1252                 /* Only allow setting the type when the interface is down */
1253                 if (tun->dev->flags & IFF_UP) {
1254                         DBG(KERN_INFO "%s: Linktype set failed because interface is up\n",
1255                                 tun->dev->name);
1256                         ret = -EBUSY;
1257                 } else {
1258                         tun->dev->type = (int) arg;
1259                         DBG(KERN_INFO "%s: linktype set to %d\n", tun->dev->name, tun->dev->type);
1260                         ret = 0;
1261                 }
1262                 break;
1263
1264 #ifdef TUN_DEBUG
1265         case TUNSETDEBUG:
1266                 tun->debug = arg;
1267                 break;
1268 #endif
1269         case TUNSETOFFLOAD:
1270                 ret = set_offload(tun->dev, arg);
1271                 break;
1272
1273         case TUNSETTXFILTER:
1274                 /* Can be set only for TAPs */
1275                 ret = -EINVAL;
1276                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1277                         break;
1278                 ret = update_filter(&tun->txflt, (void __user *)arg);
1279                 break;
1280
1281         case SIOCGIFHWADDR:
1282                 /* Get hw addres */
1283                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1284                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1285                 if (copy_to_user(argp, &ifr, ifreq_len))
1286                         ret = -EFAULT;
1287                 break;
1288
1289         case SIOCSIFHWADDR:
1290                 /* Set hw address */
1291                 DBG(KERN_DEBUG "%s: set hw address: %pM\n",
1292                         tun->dev->name, ifr.ifr_hwaddr.sa_data);
1293
1294                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1295                 break;
1296
1297         case TUNGETSNDBUF:
1298                 sndbuf = tun->socket.sk->sk_sndbuf;
1299                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1300                         ret = -EFAULT;
1301                 break;
1302
1303         case TUNSETSNDBUF:
1304                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1305                         ret = -EFAULT;
1306                         break;
1307                 }
1308
1309                 tun->socket.sk->sk_sndbuf = sndbuf;
1310                 break;
1311
1312         default:
1313                 ret = -EINVAL;
1314                 break;
1315         };
1316
1317 unlock:
1318         rtnl_unlock();
1319         if (tun)
1320                 tun_put(tun);
1321         return ret;
1322 }
1323
1324 static long tun_chr_ioctl(struct file *file,
1325                           unsigned int cmd, unsigned long arg)
1326 {
1327         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1328 }
1329
1330 #ifdef CONFIG_COMPAT
1331 static long tun_chr_compat_ioctl(struct file *file,
1332                          unsigned int cmd, unsigned long arg)
1333 {
1334         switch (cmd) {
1335         case TUNSETIFF:
1336         case TUNGETIFF:
1337         case TUNSETTXFILTER:
1338         case TUNGETSNDBUF:
1339         case TUNSETSNDBUF:
1340         case SIOCGIFHWADDR:
1341         case SIOCSIFHWADDR:
1342                 arg = (unsigned long)compat_ptr(arg);
1343                 break;
1344         default:
1345                 arg = (compat_ulong_t)arg;
1346                 break;
1347         }
1348
1349         /*
1350          * compat_ifreq is shorter than ifreq, so we must not access beyond
1351          * the end of that structure. All fields that are used in this
1352          * driver are compatible though, we don't need to convert the
1353          * contents.
1354          */
1355         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1356 }
1357 #endif /* CONFIG_COMPAT */
1358
1359 static int tun_chr_fasync(int fd, struct file *file, int on)
1360 {
1361         struct tun_struct *tun = tun_get(file);
1362         int ret;
1363
1364         if (!tun)
1365                 return -EBADFD;
1366
1367         DBG(KERN_INFO "%s: tun_chr_fasync %d\n", tun->dev->name, on);
1368
1369         if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1370                 goto out;
1371
1372         if (on) {
1373                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1374                 if (ret)
1375                         goto out;
1376                 tun->flags |= TUN_FASYNC;
1377         } else
1378                 tun->flags &= ~TUN_FASYNC;
1379         ret = 0;
1380 out:
1381         tun_put(tun);
1382         return ret;
1383 }
1384
1385 static int tun_chr_open(struct inode *inode, struct file * file)
1386 {
1387         struct tun_file *tfile;
1388
1389         DBG1(KERN_INFO "tunX: tun_chr_open\n");
1390
1391         tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1392         if (!tfile)
1393                 return -ENOMEM;
1394         atomic_set(&tfile->count, 0);
1395         tfile->tun = NULL;
1396         tfile->net = get_net(current->nsproxy->net_ns);
1397         file->private_data = tfile;
1398         return 0;
1399 }
1400
1401 static int tun_chr_close(struct inode *inode, struct file *file)
1402 {
1403         struct tun_file *tfile = file->private_data;
1404         struct tun_struct *tun;
1405
1406         tun = __tun_get(tfile);
1407         if (tun) {
1408                 struct net_device *dev = tun->dev;
1409
1410                 DBG(KERN_INFO "%s: tun_chr_close\n", dev->name);
1411
1412                 __tun_detach(tun);
1413
1414                 /* If desireable, unregister the netdevice. */
1415                 if (!(tun->flags & TUN_PERSIST)) {
1416                         rtnl_lock();
1417                         if (dev->reg_state == NETREG_REGISTERED)
1418                                 unregister_netdevice(dev);
1419                         rtnl_unlock();
1420                 }
1421         }
1422
1423         tun = tfile->tun;
1424         if (tun)
1425                 sock_put(tun->socket.sk);
1426
1427         put_net(tfile->net);
1428         kfree(tfile);
1429
1430         return 0;
1431 }
1432
1433 static const struct file_operations tun_fops = {
1434         .owner  = THIS_MODULE,
1435         .llseek = no_llseek,
1436         .read  = do_sync_read,
1437         .aio_read  = tun_chr_aio_read,
1438         .write = do_sync_write,
1439         .aio_write = tun_chr_aio_write,
1440         .poll   = tun_chr_poll,
1441         .unlocked_ioctl = tun_chr_ioctl,
1442 #ifdef CONFIG_COMPAT
1443         .compat_ioctl = tun_chr_compat_ioctl,
1444 #endif
1445         .open   = tun_chr_open,
1446         .release = tun_chr_close,
1447         .fasync = tun_chr_fasync
1448 };
1449
1450 static struct miscdevice tun_miscdev = {
1451         .minor = TUN_MINOR,
1452         .name = "tun",
1453         .nodename = "net/tun",
1454         .fops = &tun_fops,
1455 };
1456
1457 /* ethtool interface */
1458
1459 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1460 {
1461         cmd->supported          = 0;
1462         cmd->advertising        = 0;
1463         cmd->speed              = SPEED_10;
1464         cmd->duplex             = DUPLEX_FULL;
1465         cmd->port               = PORT_TP;
1466         cmd->phy_address        = 0;
1467         cmd->transceiver        = XCVR_INTERNAL;
1468         cmd->autoneg            = AUTONEG_DISABLE;
1469         cmd->maxtxpkt           = 0;
1470         cmd->maxrxpkt           = 0;
1471         return 0;
1472 }
1473
1474 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1475 {
1476         struct tun_struct *tun = netdev_priv(dev);
1477
1478         strcpy(info->driver, DRV_NAME);
1479         strcpy(info->version, DRV_VERSION);
1480         strcpy(info->fw_version, "N/A");
1481
1482         switch (tun->flags & TUN_TYPE_MASK) {
1483         case TUN_TUN_DEV:
1484                 strcpy(info->bus_info, "tun");
1485                 break;
1486         case TUN_TAP_DEV:
1487                 strcpy(info->bus_info, "tap");
1488                 break;
1489         }
1490 }
1491
1492 static u32 tun_get_msglevel(struct net_device *dev)
1493 {
1494 #ifdef TUN_DEBUG
1495         struct tun_struct *tun = netdev_priv(dev);
1496         return tun->debug;
1497 #else
1498         return -EOPNOTSUPP;
1499 #endif
1500 }
1501
1502 static void tun_set_msglevel(struct net_device *dev, u32 value)
1503 {
1504 #ifdef TUN_DEBUG
1505         struct tun_struct *tun = netdev_priv(dev);
1506         tun->debug = value;
1507 #endif
1508 }
1509
1510 static u32 tun_get_link(struct net_device *dev)
1511 {
1512         struct tun_struct *tun = netdev_priv(dev);
1513         return !!tun->tfile;
1514 }
1515
1516 static u32 tun_get_rx_csum(struct net_device *dev)
1517 {
1518         struct tun_struct *tun = netdev_priv(dev);
1519         return (tun->flags & TUN_NOCHECKSUM) == 0;
1520 }
1521
1522 static int tun_set_rx_csum(struct net_device *dev, u32 data)
1523 {
1524         struct tun_struct *tun = netdev_priv(dev);
1525         if (data)
1526                 tun->flags &= ~TUN_NOCHECKSUM;
1527         else
1528                 tun->flags |= TUN_NOCHECKSUM;
1529         return 0;
1530 }
1531
1532 static const struct ethtool_ops tun_ethtool_ops = {
1533         .get_settings   = tun_get_settings,
1534         .get_drvinfo    = tun_get_drvinfo,
1535         .get_msglevel   = tun_get_msglevel,
1536         .set_msglevel   = tun_set_msglevel,
1537         .get_link       = tun_get_link,
1538         .get_rx_csum    = tun_get_rx_csum,
1539         .set_rx_csum    = tun_set_rx_csum
1540 };
1541
1542
1543 static int __init tun_init(void)
1544 {
1545         int ret = 0;
1546
1547         printk(KERN_INFO "tun: %s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1548         printk(KERN_INFO "tun: %s\n", DRV_COPYRIGHT);
1549
1550         ret = rtnl_link_register(&tun_link_ops);
1551         if (ret) {
1552                 printk(KERN_ERR "tun: Can't register link_ops\n");
1553                 goto err_linkops;
1554         }
1555
1556         ret = misc_register(&tun_miscdev);
1557         if (ret) {
1558                 printk(KERN_ERR "tun: Can't register misc device %d\n", TUN_MINOR);
1559                 goto err_misc;
1560         }
1561         return  0;
1562 err_misc:
1563         rtnl_link_unregister(&tun_link_ops);
1564 err_linkops:
1565         return ret;
1566 }
1567
1568 static void tun_cleanup(void)
1569 {
1570         misc_deregister(&tun_miscdev);
1571         rtnl_link_unregister(&tun_link_ops);
1572 }
1573
1574 /* Get an underlying socket object from tun file.  Returns error unless file is
1575  * attached to a device.  The returned object works like a packet socket, it
1576  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1577  * holding a reference to the file for as long as the socket is in use. */
1578 struct socket *tun_get_socket(struct file *file)
1579 {
1580         struct tun_struct *tun;
1581         if (file->f_op != &tun_fops)
1582                 return ERR_PTR(-EINVAL);
1583         tun = tun_get(file);
1584         if (!tun)
1585                 return ERR_PTR(-EBADFD);
1586         tun_put(tun);
1587         return &tun->socket;
1588 }
1589 EXPORT_SYMBOL_GPL(tun_get_socket);
1590
1591 module_init(tun_init);
1592 module_exit(tun_cleanup);
1593 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1594 MODULE_AUTHOR(DRV_COPYRIGHT);
1595 MODULE_LICENSE("GPL");
1596 MODULE_ALIAS_MISCDEV(TUN_MINOR);