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 pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/major.h>
  48 #include <linux/slab.h>
  49 #include <linux/poll.h>
  50 #include <linux/fcntl.h>
  51 #include <linux/init.h>
  52 #include <linux/skbuff.h>
  53 #include <linux/netdevice.h>
  54 #include <linux/etherdevice.h>
  55 #include <linux/miscdevice.h>
  56 #include <linux/ethtool.h>
  57 #include <linux/rtnetlink.h>
  58 #include <linux/compat.h>
  59 #include <linux/if.h>
  60 #include <linux/if_arp.h>
  61 #include <linux/if_ether.h>
  62 #include <linux/if_tun.h>
  63 #include <linux/crc32.h>
  64 #include <linux/nsproxy.h>
  65 #include <linux/virtio_net.h>
  66 #include <linux/rcupdate.h>
  67 #include <net/net_namespace.h>
  68 #include <net/netns/generic.h>
  69 #include <net/rtnetlink.h>
  70 #include <net/sock.h>
  71
  72 #include <asm/system.h>
  73 #include <asm/uaccess.h>
  74
  75 /* Uncomment to enable debugging */
  76 /* #define TUN_DEBUG 1 */
  77
  78 #ifdef TUN_DEBUG
  79 static int debug;
  80
  81 #define tun_debug(level, tun, fmt, args...)                     \
  82 do {                                                            \
  83         if (tun->debug)                                         \
  84                 netdev_printk(level, tun->dev, fmt, ##args);    \
  85 } while (0)
  86 #define DBG1(level, fmt, args...)                               \
  87 do {                                                            \
  88         if (debug == 2)                                         \
  89                 printk(level fmt, ##args);                      \
  90 } while (0)
  91 #else
  92 #define tun_debug(level, tun, fmt, args...)                     \
  93 do {                                                            \
  94         if (0)                                                  \
  95                 netdev_printk(level, tun->dev, fmt, ##args);    \
  96 } while (0)
  97 #define DBG1(level, fmt, args...)                               \
  98 do {                                                            \
  99         if (0)                                                  \
 100                 printk(level fmt, ##args);                      \
 101 } while (0)
 102 #endif
 103
 104 #define FLT_EXACT_COUNT 8
 105 struct tap_filter {
 106         unsigned int    count;    /* Number of addrs. Zero means disabled */
 107         u32             mask[2];  /* Mask of the hashed addrs */
 108         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 109 };
 110
 111 struct tun_file {
 112         atomic_t count;
 113         struct tun_struct *tun;
 114         struct net *net;
 115 };
 116
 117 struct tun_sock;
 118
 119 struct tun_struct {
 120         struct tun_file         *tfile;
 121         unsigned int            flags;
 122         uid_t                   owner;
 123         gid_t                   group;
 124
 125         struct net_device       *dev;
 126         u32                     set_features;
 127 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 128                           NETIF_F_TSO6|NETIF_F_UFO)
 129         struct fasync_struct    *fasync;
 130
 131         struct tap_filter       txflt;
 132         struct socket           socket;
 133         struct socket_wq        wq;
 134
 135         int                     vnet_hdr_sz;
 136
 137 #ifdef TUN_DEBUG
 138         int debug;
 139 #endif
 140 };
 141
 142 struct tun_sock {
 143         struct sock             sk;
 144         struct tun_struct       *tun;
 145 };
 146
 147 static inline struct tun_sock *tun_sk(struct sock *sk)
 148 {
 149         return container_of(sk, struct tun_sock, sk);
 150 }
 151
 152 static int tun_attach(struct tun_struct *tun, struct file *file)
 153 {
 154         struct tun_file *tfile = file->private_data;
 155         int err;
 156
 157         ASSERT_RTNL();
 158
 159         netif_tx_lock_bh(tun->dev);
 160
 161         err = -EINVAL;
 162         if (tfile->tun)
 163                 goto out;
 164
 165         err = -EBUSY;
 166         if (tun->tfile)
 167                 goto out;
 168
 169         err = 0;
 170         tfile->tun = tun;
 171         tun->tfile = tfile;
 172         tun->socket.file = file;
 173         netif_carrier_on(tun->dev);
 174         dev_hold(tun->dev);
 175         sock_hold(tun->socket.sk);
 176         atomic_inc(&tfile->count);
 177
 178 out:
 179         netif_tx_unlock_bh(tun->dev);
 180         return err;
 181 }
 182
 183 static void __tun_detach(struct tun_struct *tun)
 184 {
 185         /* Detach from net device */
 186         netif_tx_lock_bh(tun->dev);
 187         netif_carrier_off(tun->dev);
 188         tun->tfile = NULL;
 189         tun->socket.file = NULL;
 190         netif_tx_unlock_bh(tun->dev);
 191
 192         /* Drop read queue */
 193         skb_queue_purge(&tun->socket.sk->sk_receive_queue);
 194
 195         /* Drop the extra count on the net device */
 196         dev_put(tun->dev);
 197 }
 198
 199 static void tun_detach(struct tun_struct *tun)
 200 {
 201         rtnl_lock();
 202         __tun_detach(tun);
 203         rtnl_unlock();
 204 }
 205
 206 static struct tun_struct *__tun_get(struct tun_file *tfile)
 207 {
 208         struct tun_struct *tun = NULL;
 209
 210         if (atomic_inc_not_zero(&tfile->count))
 211                 tun = tfile->tun;
 212
 213         return tun;
 214 }
 215
 216 static struct tun_struct *tun_get(struct file *file)
 217 {
 218         return __tun_get(file->private_data);
 219 }
 220
 221 static void tun_put(struct tun_struct *tun)
 222 {
 223         struct tun_file *tfile = tun->tfile;
 224
 225         if (atomic_dec_and_test(&tfile->count))
 226                 tun_detach(tfile->tun);
 227 }
 228
 229 /* TAP filtering */
 230 static void addr_hash_set(u32 *mask, const u8 *addr)
 231 {
 232         int n = ether_crc(ETH_ALEN, addr) >> 26;
 233         mask[n >> 5] |= (1 << (n & 31));
 234 }
 235
 236 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 237 {
 238         int n = ether_crc(ETH_ALEN, addr) >> 26;
 239         return mask[n >> 5] & (1 << (n & 31));
 240 }
 241
 242 static int update_filter(struct tap_filter *filter, void __user *arg)
 243 {
 244         struct { u8 u[ETH_ALEN]; } *addr;
 245         struct tun_filter uf;
 246         int err, alen, n, nexact;
 247
 248         if (copy_from_user(&uf, arg, sizeof(uf)))
 249                 return -EFAULT;
 250
 251         if (!uf.count) {
 252                 /* Disabled */
 253                 filter->count = 0;
 254                 return 0;
 255         }
 256
 257         alen = ETH_ALEN * uf.count;
 258         addr = kmalloc(alen, GFP_KERNEL);
 259         if (!addr)
 260                 return -ENOMEM;
 261
 262         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 263                 err = -EFAULT;
 264                 goto done;
 265         }
 266
 267         /* The filter is updated without holding any locks. Which is
 268          * perfectly safe. We disable it first and in the worst
 269          * case we'll accept a few undesired packets. */
 270         filter->count = 0;
 271         wmb();
 272
 273         /* Use first set of addresses as an exact filter */
 274         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 275                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 276
 277         nexact = n;
 278
 279         /* Remaining multicast addresses are hashed,
 280          * unicast will leave the filter disabled. */
 281         memset(filter->mask, 0, sizeof(filter->mask));
 282         for (; n < uf.count; n++) {
 283                 if (!is_multicast_ether_addr(addr[n].u)) {
 284                         err = 0; /* no filter */
 285                         goto done;
 286                 }
 287                 addr_hash_set(filter->mask, addr[n].u);
 288         }
 289
 290         /* For ALLMULTI just set the mask to all ones.
 291          * This overrides the mask populated above. */
 292         if ((uf.flags & TUN_FLT_ALLMULTI))
 293                 memset(filter->mask, ~0, sizeof(filter->mask));
 294
 295         /* Now enable the filter */
 296         wmb();
 297         filter->count = nexact;
 298
 299         /* Return the number of exact filters */
 300         err = nexact;
 301
 302 done:
 303         kfree(addr);
 304         return err;
 305 }
 306
 307 /* Returns: 0 - drop, !=0 - accept */
 308 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 309 {
 310         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 311          * at this point. */
 312         struct ethhdr *eh = (struct ethhdr *) skb->data;
 313         int i;
 314
 315         /* Exact match */
 316         for (i = 0; i < filter->count; i++)
 317                 if (!compare_ether_addr(eh->h_dest, filter->addr[i]))
 318                         return 1;
 319
 320         /* Inexact match (multicast only) */
 321         if (is_multicast_ether_addr(eh->h_dest))
 322                 return addr_hash_test(filter->mask, eh->h_dest);
 323
 324         return 0;
 325 }
 326
 327 /*
 328  * Checks whether the packet is accepted or not.
 329  * Returns: 0 - drop, !=0 - accept
 330  */
 331 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 332 {
 333         if (!filter->count)
 334                 return 1;
 335
 336         return run_filter(filter, skb);
 337 }
 338
 339 /* Network device part of the driver */
 340
 341 static const struct ethtool_ops tun_ethtool_ops;
 342
 343 /* Net device detach from fd. */
 344 static void tun_net_uninit(struct net_device *dev)
 345 {
 346         struct tun_struct *tun = netdev_priv(dev);
 347         struct tun_file *tfile = tun->tfile;
 348
 349         /* Inform the methods they need to stop using the dev.
 350          */
 351         if (tfile) {
 352                 wake_up_all(&tun->wq.wait);
 353                 if (atomic_dec_and_test(&tfile->count))
 354                         __tun_detach(tun);
 355         }
 356 }
 357
 358 static void tun_free_netdev(struct net_device *dev)
 359 {
 360         struct tun_struct *tun = netdev_priv(dev);
 361
 362         sock_put(tun->socket.sk);
 363 }
 364
 365 /* Net device open. */
 366 static int tun_net_open(struct net_device *dev)
 367 {
 368         netif_start_queue(dev);
 369         return 0;
 370 }
 371
 372 /* Net device close. */
 373 static int tun_net_close(struct net_device *dev)
 374 {
 375         netif_stop_queue(dev);
 376         return 0;
 377 }
 378
 379 /* Net device start xmit */
 380 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 381 {
 382         struct tun_struct *tun = netdev_priv(dev);
 383
 384         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 385
 386         /* Drop packet if interface is not attached */
 387         if (!tun->tfile)
 388                 goto drop;
 389
 390         /* Drop if the filter does not like it.
 391          * This is a noop if the filter is disabled.
 392          * Filter can be enabled only for the TAP devices. */
 393         if (!check_filter(&tun->txflt, skb))
 394                 goto drop;
 395
 396         if (tun->socket.sk->sk_filter &&
 397             sk_filter(tun->socket.sk, skb))
 398                 goto drop;
 399
 400         if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) {
 401                 if (!(tun->flags & TUN_ONE_QUEUE)) {
 402                         /* Normal queueing mode. */
 403                         /* Packet scheduler handles dropping of further packets. */
 404                         netif_stop_queue(dev);
 405
 406                         /* We won't see all dropped packets individually, so overrun
 407                          * error is more appropriate. */
 408                         dev->stats.tx_fifo_errors++;
 409                 } else {
 410                         /* Single queue mode.
 411                          * Driver handles dropping of all packets itself. */
 412                         goto drop;
 413                 }
 414         }
 415
 416         /* Orphan the skb - required as we might hang on to it
 417          * for indefinite time. */
 418         skb_orphan(skb);
 419
 420         nf_reset(skb);
 421
 422         /* Enqueue packet */
 423         skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb);
 424
 425         /* Notify and wake up reader process */
 426         if (tun->flags & TUN_FASYNC)
 427                 kill_fasync(&tun->fasync, SIGIO, POLL_IN);
 428         wake_up_interruptible_poll(&tun->wq.wait, POLLIN |
 429                                    POLLRDNORM | POLLRDBAND);
 430         return NETDEV_TX_OK;
 431
 432 drop:
 433         dev->stats.tx_dropped++;
 434         kfree_skb(skb);
 435         return NETDEV_TX_OK;
 436 }
 437
 438 static void tun_net_mclist(struct net_device *dev)
 439 {
 440         /*
 441          * This callback is supposed to deal with mc filter in
 442          * _rx_ path and has nothing to do with the _tx_ path.
 443          * In rx path we always accept everything userspace gives us.
 444          */
 445 }
 446
 447 #define MIN_MTU 68
 448 #define MAX_MTU 65535
 449
 450 static int
 451 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 452 {
 453         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 454                 return -EINVAL;
 455         dev->mtu = new_mtu;
 456         return 0;
 457 }
 458
 459 static u32 tun_net_fix_features(struct net_device *dev, u32 features)
 460 {
 461         struct tun_struct *tun = netdev_priv(dev);
 462
 463         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 464 }
 465 #ifdef CONFIG_NET_POLL_CONTROLLER
 466 static void tun_poll_controller(struct net_device *dev)
 467 {
 468         /*
 469          * Tun only receives frames when:
 470          * 1) the char device endpoint gets data from user space
 471          * 2) the tun socket gets a sendmsg call from user space
 472          * Since both of those are syncronous operations, we are guaranteed
 473          * never to have pending data when we poll for it
 474          * so theres nothing to do here but return.
 475          * We need this though so netpoll recognizes us as an interface that
 476          * supports polling, which enables bridge devices in virt setups to
 477          * still use netconsole
 478          */
 479         return;
 480 }
 481 #endif
 482 static const struct net_device_ops tun_netdev_ops = {
 483         .ndo_uninit             = tun_net_uninit,
 484         .ndo_open               = tun_net_open,
 485         .ndo_stop               = tun_net_close,
 486         .ndo_start_xmit         = tun_net_xmit,
 487         .ndo_change_mtu         = tun_net_change_mtu,
 488         .ndo_fix_features       = tun_net_fix_features,
 489 #ifdef CONFIG_NET_POLL_CONTROLLER
 490         .ndo_poll_controller    = tun_poll_controller,
 491 #endif
 492 };
 493
 494 static const struct net_device_ops tap_netdev_ops = {
 495         .ndo_uninit             = tun_net_uninit,
 496         .ndo_open               = tun_net_open,
 497         .ndo_stop               = tun_net_close,
 498         .ndo_start_xmit         = tun_net_xmit,
 499         .ndo_change_mtu         = tun_net_change_mtu,
 500         .ndo_fix_features       = tun_net_fix_features,
 501         .ndo_set_rx_mode        = tun_net_mclist,
 502         .ndo_set_mac_address    = eth_mac_addr,
 503         .ndo_validate_addr      = eth_validate_addr,
 504 #ifdef CONFIG_NET_POLL_CONTROLLER
 505         .ndo_poll_controller    = tun_poll_controller,
 506 #endif
 507 };
 508
 509 /* Initialize net device. */
 510 static void tun_net_init(struct net_device *dev)
 511 {
 512         struct tun_struct *tun = netdev_priv(dev);
 513
 514         switch (tun->flags & TUN_TYPE_MASK) {
 515         case TUN_TUN_DEV:
 516                 dev->netdev_ops = &tun_netdev_ops;
 517
 518                 /* Point-to-Point TUN Device */
 519                 dev->hard_header_len = 0;
 520                 dev->addr_len = 0;
 521                 dev->mtu = 1500;
 522
 523                 /* Zero header length */
 524                 dev->type = ARPHRD_NONE;
 525                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 526                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 527                 break;
 528
 529         case TUN_TAP_DEV:
 530                 dev->netdev_ops = &tap_netdev_ops;
 531                 /* Ethernet TAP Device */
 532                 ether_setup(dev);
 533                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 534
 535                 random_ether_addr(dev->dev_addr);
 536
 537                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 538                 break;
 539         }
 540 }
 541
 542 /* Character device part */
 543
 544 /* Poll */
 545 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
 546 {
 547         struct tun_file *tfile = file->private_data;
 548         struct tun_struct *tun = __tun_get(tfile);
 549         struct sock *sk;
 550         unsigned int mask = 0;
 551
 552         if (!tun)
 553                 return POLLERR;
 554
 555         sk = tun->socket.sk;
 556
 557         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 558
 559         poll_wait(file, &tun->wq.wait, wait);
 560
 561         if (!skb_queue_empty(&sk->sk_receive_queue))
 562                 mask |= POLLIN | POLLRDNORM;
 563
 564         if (sock_writeable(sk) ||
 565             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 566              sock_writeable(sk)))
 567                 mask |= POLLOUT | POLLWRNORM;
 568
 569         if (tun->dev->reg_state != NETREG_REGISTERED)
 570                 mask = POLLERR;
 571
 572         tun_put(tun);
 573         return mask;
 574 }
 575
 576 /* prepad is the amount to reserve at front.  len is length after that.
 577  * linear is a hint as to how much to copy (usually headers). */
 578 static struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
 579                                      size_t prepad, size_t len,
 580                                      size_t linear, int noblock)
 581 {
 582         struct sock *sk = tun->socket.sk;
 583         struct sk_buff *skb;
 584         int err;
 585
 586         sock_update_classid(sk);
 587
 588         /* Under a page?  Don't bother with paged skb. */
 589         if (prepad + len < PAGE_SIZE || !linear)
 590                 linear = len;
 591
 592         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 593                                    &err);
 594         if (!skb)
 595                 return ERR_PTR(err);
 596
 597         skb_reserve(skb, prepad);
 598         skb_put(skb, linear);
 599         skb->data_len = len - linear;
 600         skb->len += len - linear;
 601
 602         return skb;
 603 }
 604
 605 /* Get packet from user space buffer */
 606 static ssize_t tun_get_user(struct tun_struct *tun,
 607                             const struct iovec *iv, size_t count,
 608                             int noblock)
 609 {
 610         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
 611         struct sk_buff *skb;
 612         size_t len = count, align = NET_SKB_PAD;
 613         struct virtio_net_hdr gso = { 0 };
 614         int offset = 0;
 615
 616         if (!(tun->flags & TUN_NO_PI)) {
 617                 if (len < sizeof(pi))
 618                         return -EINVAL;
 619                 len -= sizeof(pi);
 620
 621                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
 622                         return -EFAULT;
 623                 offset += sizeof(pi);
 624         }
 625
 626         if (tun->flags & TUN_VNET_HDR) {
 627                 if (len < tun->vnet_hdr_sz)
 628                         return -EINVAL;
 629                 len -= tun->vnet_hdr_sz;
 630
 631                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
 632                         return -EFAULT;
 633
 634                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
 635                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
 636                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
 637
 638                 if (gso.hdr_len > len)
 639                         return -EINVAL;
 640                 offset += tun->vnet_hdr_sz;
 641         }
 642
 643         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
 644                 align += NET_IP_ALIGN;
 645                 if (unlikely(len < ETH_HLEN ||
 646                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
 647                         return -EINVAL;
 648         }
 649
 650         skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock);
 651         if (IS_ERR(skb)) {
 652                 if (PTR_ERR(skb) != -EAGAIN)
 653                         tun->dev->stats.rx_dropped++;
 654                 return PTR_ERR(skb);
 655         }
 656
 657         if (skb_copy_datagram_from_iovec(skb, 0, iv, offset, len)) {
 658                 tun->dev->stats.rx_dropped++;
 659                 kfree_skb(skb);
 660                 return -EFAULT;
 661         }
 662
 663         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 664                 if (!skb_partial_csum_set(skb, gso.csum_start,
 665                                           gso.csum_offset)) {
 666                         tun->dev->stats.rx_frame_errors++;
 667                         kfree_skb(skb);
 668                         return -EINVAL;
 669                 }
 670         }
 671
 672         switch (tun->flags & TUN_TYPE_MASK) {
 673         case TUN_TUN_DEV:
 674                 if (tun->flags & TUN_NO_PI) {
 675                         switch (skb->data[0] & 0xf0) {
 676                         case 0x40:
 677                                 pi.proto = htons(ETH_P_IP);
 678                                 break;
 679                         case 0x60:
 680                                 pi.proto = htons(ETH_P_IPV6);
 681                                 break;
 682                         default:
 683                                 tun->dev->stats.rx_dropped++;
 684                                 kfree_skb(skb);
 685                                 return -EINVAL;
 686                         }
 687                 }
 688
 689                 skb_reset_mac_header(skb);
 690                 skb->protocol = pi.proto;
 691                 skb->dev = tun->dev;
 692                 break;
 693         case TUN_TAP_DEV:
 694                 skb->protocol = eth_type_trans(skb, tun->dev);
 695                 break;
 696         }
 697
 698         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 699                 pr_debug("GSO!\n");
 700                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 701                 case VIRTIO_NET_HDR_GSO_TCPV4:
 702                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 703                         break;
 704                 case VIRTIO_NET_HDR_GSO_TCPV6:
 705                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 706                         break;
 707                 case VIRTIO_NET_HDR_GSO_UDP:
 708                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 709                         break;
 710                 default:
 711                         tun->dev->stats.rx_frame_errors++;
 712                         kfree_skb(skb);
 713                         return -EINVAL;
 714                 }
 715
 716                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
 717                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 718
 719                 skb_shinfo(skb)->gso_size = gso.gso_size;
 720                 if (skb_shinfo(skb)->gso_size == 0) {
 721                         tun->dev->stats.rx_frame_errors++;
 722                         kfree_skb(skb);
 723                         return -EINVAL;
 724                 }
 725
 726                 /* Header must be checked, and gso_segs computed. */
 727                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 728                 skb_shinfo(skb)->gso_segs = 0;
 729         }
 730
 731         netif_rx_ni(skb);
 732
 733         tun->dev->stats.rx_packets++;
 734         tun->dev->stats.rx_bytes += len;
 735
 736         return count;
 737 }
 738
 739 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
 740                               unsigned long count, loff_t pos)
 741 {
 742         struct file *file = iocb->ki_filp;
 743         struct tun_struct *tun = tun_get(file);
 744         ssize_t result;
 745
 746         if (!tun)
 747                 return -EBADFD;
 748
 749         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
 750
 751         result = tun_get_user(tun, iv, iov_length(iv, count),
 752                               file->f_flags & O_NONBLOCK);
 753
 754         tun_put(tun);
 755         return result;
 756 }
 757
 758 /* Put packet to the user space buffer */
 759 static ssize_t tun_put_user(struct tun_struct *tun,
 760                             struct sk_buff *skb,
 761                             const struct iovec *iv, int len)
 762 {
 763         struct tun_pi pi = { 0, skb->protocol };
 764         ssize_t total = 0;
 765
 766         if (!(tun->flags & TUN_NO_PI)) {
 767                 if ((len -= sizeof(pi)) < 0)
 768                         return -EINVAL;
 769
 770                 if (len < skb->len) {
 771                         /* Packet will be striped */
 772                         pi.flags |= TUN_PKT_STRIP;
 773                 }
 774
 775                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
 776                         return -EFAULT;
 777                 total += sizeof(pi);
 778         }
 779
 780         if (tun->flags & TUN_VNET_HDR) {
 781                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
 782                 if ((len -= tun->vnet_hdr_sz) < 0)
 783                         return -EINVAL;
 784
 785                 if (skb_is_gso(skb)) {
 786                         struct skb_shared_info *sinfo = skb_shinfo(skb);
 787
 788                         /* This is a hint as to how much should be linear. */
 789                         gso.hdr_len = skb_headlen(skb);
 790                         gso.gso_size = sinfo->gso_size;
 791                         if (sinfo->gso_type & SKB_GSO_TCPV4)
 792                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 793                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
 794                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 795                         else if (sinfo->gso_type & SKB_GSO_UDP)
 796                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
 797                         else {
 798                                 pr_err("unexpected GSO type: "
 799                                        "0x%x, gso_size %d, hdr_len %d\n",
 800                                        sinfo->gso_type, gso.gso_size,
 801                                        gso.hdr_len);
 802                                 print_hex_dump(KERN_ERR, "tun: ",
 803                                                DUMP_PREFIX_NONE,
 804                                                16, 1, skb->head,
 805                                                min((int)gso.hdr_len, 64), true);
 806                                 WARN_ON_ONCE(1);
 807                                 return -EINVAL;
 808                         }
 809                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
 810                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 811                 } else
 812                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
 813
 814                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 815                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 816                         gso.csum_start = skb_checksum_start_offset(skb);
 817                         gso.csum_offset = skb->csum_offset;
 818                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
 819                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
 820                 } /* else everything is zero */
 821
 822                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
 823                                                sizeof(gso))))
 824                         return -EFAULT;
 825                 total += tun->vnet_hdr_sz;
 826         }
 827
 828         len = min_t(int, skb->len, len);
 829
 830         skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
 831         total += skb->len;
 832
 833         tun->dev->stats.tx_packets++;
 834         tun->dev->stats.tx_bytes += len;
 835
 836         return total;
 837 }
 838
 839 static ssize_t tun_do_read(struct tun_struct *tun,
 840                            struct kiocb *iocb, const struct iovec *iv,
 841                            ssize_t len, int noblock)
 842 {
 843         DECLARE_WAITQUEUE(wait, current);
 844         struct sk_buff *skb;
 845         ssize_t ret = 0;
 846
 847         tun_debug(KERN_INFO, tun, "tun_chr_read\n");
 848
 849         if (unlikely(!noblock))
 850                 add_wait_queue(&tun->wq.wait, &wait);
 851         while (len) {
 852                 current->state = TASK_INTERRUPTIBLE;
 853
 854                 /* Read frames from the queue */
 855                 if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
 856                         if (noblock) {
 857                                 ret = -EAGAIN;
 858                                 break;
 859                         }
 860                         if (signal_pending(current)) {
 861                                 ret = -ERESTARTSYS;
 862                                 break;
 863                         }
 864                         if (tun->dev->reg_state != NETREG_REGISTERED) {
 865                                 ret = -EIO;
 866                                 break;
 867                         }
 868
 869                         /* Nothing to read, let's sleep */
 870                         schedule();
 871                         continue;
 872                 }
 873                 netif_wake_queue(tun->dev);
 874
 875                 ret = tun_put_user(tun, skb, iv, len);
 876                 kfree_skb(skb);
 877                 break;
 878         }
 879
 880         current->state = TASK_RUNNING;
 881         if (unlikely(!noblock))
 882                 remove_wait_queue(&tun->wq.wait, &wait);
 883
 884         return ret;
 885 }
 886
 887 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
 888                             unsigned long count, loff_t pos)
 889 {
 890         struct file *file = iocb->ki_filp;
 891         struct tun_file *tfile = file->private_data;
 892         struct tun_struct *tun = __tun_get(tfile);
 893         ssize_t len, ret;
 894
 895         if (!tun)
 896                 return -EBADFD;
 897         len = iov_length(iv, count);
 898         if (len < 0) {
 899                 ret = -EINVAL;
 900                 goto out;
 901         }
 902
 903         ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK);
 904         ret = min_t(ssize_t, ret, len);
 905 out:
 906         tun_put(tun);
 907         return ret;
 908 }
 909
 910 static void tun_setup(struct net_device *dev)
 911 {
 912         struct tun_struct *tun = netdev_priv(dev);
 913
 914         tun->owner = -1;
 915         tun->group = -1;
 916
 917         dev->ethtool_ops = &tun_ethtool_ops;
 918         dev->destructor = tun_free_netdev;
 919 }
 920
 921 /* Trivial set of netlink ops to allow deleting tun or tap
 922  * device with netlink.
 923  */
 924 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
 925 {
 926         return -EINVAL;
 927 }
 928
 929 static struct rtnl_link_ops tun_link_ops __read_mostly = {
 930         .kind           = DRV_NAME,
 931         .priv_size      = sizeof(struct tun_struct),
 932         .setup          = tun_setup,
 933         .validate       = tun_validate,
 934 };
 935
 936 static void tun_sock_write_space(struct sock *sk)
 937 {
 938         struct tun_struct *tun;
 939         wait_queue_head_t *wqueue;
 940
 941         if (!sock_writeable(sk))
 942                 return;
 943
 944         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
 945                 return;
 946
 947         wqueue = sk_sleep(sk);
 948         if (wqueue && waitqueue_active(wqueue))
 949                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
 950                                                 POLLWRNORM | POLLWRBAND);
 951
 952         tun = tun_sk(sk)->tun;
 953         kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
 954 }
 955
 956 static void tun_sock_destruct(struct sock *sk)
 957 {
 958         free_netdev(tun_sk(sk)->tun->dev);
 959 }
 960
 961 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
 962                        struct msghdr *m, size_t total_len)
 963 {
 964         struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
 965         return tun_get_user(tun, m->msg_iov, total_len,
 966                             m->msg_flags & MSG_DONTWAIT);
 967 }
 968
 969 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
 970                        struct msghdr *m, size_t total_len,
 971                        int flags)
 972 {
 973         struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
 974         int ret;
 975         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
 976                 return -EINVAL;
 977         ret = tun_do_read(tun, iocb, m->msg_iov, total_len,
 978                           flags & MSG_DONTWAIT);
 979         if (ret > total_len) {
 980                 m->msg_flags |= MSG_TRUNC;
 981                 ret = flags & MSG_TRUNC ? ret : total_len;
 982         }
 983         return ret;
 984 }
 985
 986 /* Ops structure to mimic raw sockets with tun */
 987 static const struct proto_ops tun_socket_ops = {
 988         .sendmsg = tun_sendmsg,
 989         .recvmsg = tun_recvmsg,
 990 };
 991
 992 static struct proto tun_proto = {
 993         .name           = "tun",
 994         .owner          = THIS_MODULE,
 995         .obj_size       = sizeof(struct tun_sock),
 996 };
 997
 998 static int tun_flags(struct tun_struct *tun)
 999 {
1000         int flags = 0;
1001
1002         if (tun->flags & TUN_TUN_DEV)
1003                 flags |= IFF_TUN;
1004         else
1005                 flags |= IFF_TAP;
1006
1007         if (tun->flags & TUN_NO_PI)
1008                 flags |= IFF_NO_PI;
1009
1010         if (tun->flags & TUN_ONE_QUEUE)
1011                 flags |= IFF_ONE_QUEUE;
1012
1013         if (tun->flags & TUN_VNET_HDR)
1014                 flags |= IFF_VNET_HDR;
1015
1016         return flags;
1017 }
1018
1019 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1020                               char *buf)
1021 {
1022         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1023         return sprintf(buf, "0x%x\n", tun_flags(tun));
1024 }
1025
1026 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1027                               char *buf)
1028 {
1029         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1030         return sprintf(buf, "%d\n", tun->owner);
1031 }
1032
1033 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1034                               char *buf)
1035 {
1036         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1037         return sprintf(buf, "%d\n", tun->group);
1038 }
1039
1040 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1041 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1042 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1043
1044 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1045 {
1046         struct sock *sk;
1047         struct tun_struct *tun;
1048         struct net_device *dev;
1049         int err;
1050
1051         dev = __dev_get_by_name(net, ifr->ifr_name);
1052         if (dev) {
1053                 const struct cred *cred = current_cred();
1054
1055                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1056                         return -EBUSY;
1057                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1058                         tun = netdev_priv(dev);
1059                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1060                         tun = netdev_priv(dev);
1061                 else
1062                         return -EINVAL;
1063
1064                 if (((tun->owner != -1 && cred->euid != tun->owner) ||
1065                      (tun->group != -1 && !in_egroup_p(tun->group))) &&
1066                     !capable(CAP_NET_ADMIN))
1067                         return -EPERM;
1068                 err = security_tun_dev_attach(tun->socket.sk);
1069                 if (err < 0)
1070                         return err;
1071
1072                 err = tun_attach(tun, file);
1073                 if (err < 0)
1074                         return err;
1075         }
1076         else {
1077                 char *name;
1078                 unsigned long flags = 0;
1079
1080                 if (!capable(CAP_NET_ADMIN))
1081                         return -EPERM;
1082                 err = security_tun_dev_create();
1083                 if (err < 0)
1084                         return err;
1085
1086                 /* Set dev type */
1087                 if (ifr->ifr_flags & IFF_TUN) {
1088                         /* TUN device */
1089                         flags |= TUN_TUN_DEV;
1090                         name = "tun%d";
1091                 } else if (ifr->ifr_flags & IFF_TAP) {
1092                         /* TAP device */
1093                         flags |= TUN_TAP_DEV;
1094                         name = "tap%d";
1095                 } else
1096                         return -EINVAL;
1097
1098                 if (*ifr->ifr_name)
1099                         name = ifr->ifr_name;
1100
1101                 dev = alloc_netdev(sizeof(struct tun_struct), name,
1102                                    tun_setup);
1103                 if (!dev)
1104                         return -ENOMEM;
1105
1106                 dev_net_set(dev, net);
1107                 dev->rtnl_link_ops = &tun_link_ops;
1108
1109                 tun = netdev_priv(dev);
1110                 tun->dev = dev;
1111                 tun->flags = flags;
1112                 tun->txflt.count = 0;
1113                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1114
1115                 err = -ENOMEM;
1116                 sk = sk_alloc(net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1117                 if (!sk)
1118                         goto err_free_dev;
1119
1120                 tun->socket.wq = &tun->wq;
1121                 init_waitqueue_head(&tun->wq.wait);
1122                 tun->socket.ops = &tun_socket_ops;
1123                 sock_init_data(&tun->socket, sk);
1124                 sk->sk_write_space = tun_sock_write_space;
1125                 sk->sk_sndbuf = INT_MAX;
1126
1127                 tun_sk(sk)->tun = tun;
1128
1129                 security_tun_dev_post_create(sk);
1130
1131                 tun_net_init(dev);
1132
1133                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1134                         TUN_USER_FEATURES;
1135                 dev->features = dev->hw_features;
1136
1137                 err = register_netdevice(tun->dev);
1138                 if (err < 0)
1139                         goto err_free_sk;
1140
1141                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1142                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1143                     device_create_file(&tun->dev->dev, &dev_attr_group))
1144                         pr_err("Failed to create tun sysfs files\n");
1145
1146                 sk->sk_destruct = tun_sock_destruct;
1147
1148                 err = tun_attach(tun, file);
1149                 if (err < 0)
1150                         goto failed;
1151         }
1152
1153         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1154
1155         if (ifr->ifr_flags & IFF_NO_PI)
1156                 tun->flags |= TUN_NO_PI;
1157         else
1158                 tun->flags &= ~TUN_NO_PI;
1159
1160         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1161                 tun->flags |= TUN_ONE_QUEUE;
1162         else
1163                 tun->flags &= ~TUN_ONE_QUEUE;
1164
1165         if (ifr->ifr_flags & IFF_VNET_HDR)
1166                 tun->flags |= TUN_VNET_HDR;
1167         else
1168                 tun->flags &= ~TUN_VNET_HDR;
1169
1170         /* Make sure persistent devices do not get stuck in
1171          * xoff state.
1172          */
1173         if (netif_running(tun->dev))
1174                 netif_wake_queue(tun->dev);
1175
1176         strcpy(ifr->ifr_name, tun->dev->name);
1177         return 0;
1178
1179  err_free_sk:
1180         sock_put(sk);
1181  err_free_dev:
1182         free_netdev(dev);
1183  failed:
1184         return err;
1185 }
1186
1187 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1188                        struct ifreq *ifr)
1189 {
1190         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1191
1192         strcpy(ifr->ifr_name, tun->dev->name);
1193
1194         ifr->ifr_flags = tun_flags(tun);
1195
1196         return 0;
1197 }
1198
1199 /* This is like a cut-down ethtool ops, except done via tun fd so no
1200  * privs required. */
1201 static int set_offload(struct tun_struct *tun, unsigned long arg)
1202 {
1203         u32 features = 0;
1204
1205         if (arg & TUN_F_CSUM) {
1206                 features |= NETIF_F_HW_CSUM;
1207                 arg &= ~TUN_F_CSUM;
1208
1209                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1210                         if (arg & TUN_F_TSO_ECN) {
1211                                 features |= NETIF_F_TSO_ECN;
1212                                 arg &= ~TUN_F_TSO_ECN;
1213                         }
1214                         if (arg & TUN_F_TSO4)
1215                                 features |= NETIF_F_TSO;
1216                         if (arg & TUN_F_TSO6)
1217                                 features |= NETIF_F_TSO6;
1218                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1219                 }
1220
1221                 if (arg & TUN_F_UFO) {
1222                         features |= NETIF_F_UFO;
1223                         arg &= ~TUN_F_UFO;
1224                 }
1225         }
1226
1227         /* This gives the user a way to test for new features in future by
1228          * trying to set them. */
1229         if (arg)
1230                 return -EINVAL;
1231
1232         tun->set_features = features;
1233         netdev_update_features(tun->dev);
1234
1235         return 0;
1236 }
1237
1238 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1239                             unsigned long arg, int ifreq_len)
1240 {
1241         struct tun_file *tfile = file->private_data;
1242         struct tun_struct *tun;
1243         void __user* argp = (void __user*)arg;
1244         struct sock_fprog fprog;
1245         struct ifreq ifr;
1246         int sndbuf;
1247         int vnet_hdr_sz;
1248         int ret;
1249
1250         if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89) {
1251                 if (copy_from_user(&ifr, argp, ifreq_len))
1252                         return -EFAULT;
1253         } else {
1254                 memset(&ifr, 0, sizeof(ifr));
1255         }
1256         if (cmd == TUNGETFEATURES) {
1257                 /* Currently this just means: "what IFF flags are valid?".
1258                  * This is needed because we never checked for invalid flags on
1259                  * TUNSETIFF. */
1260                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1261                                 IFF_VNET_HDR,
1262                                 (unsigned int __user*)argp);
1263         }
1264
1265         rtnl_lock();
1266
1267         tun = __tun_get(tfile);
1268         if (cmd == TUNSETIFF && !tun) {
1269                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1270
1271                 ret = tun_set_iff(tfile->net, file, &ifr);
1272
1273                 if (ret)
1274                         goto unlock;
1275
1276                 if (copy_to_user(argp, &ifr, ifreq_len))
1277                         ret = -EFAULT;
1278                 goto unlock;
1279         }
1280
1281         ret = -EBADFD;
1282         if (!tun)
1283                 goto unlock;
1284
1285         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %d\n", cmd);
1286
1287         ret = 0;
1288         switch (cmd) {
1289         case TUNGETIFF:
1290                 ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1291                 if (ret)
1292                         break;
1293
1294                 if (copy_to_user(argp, &ifr, ifreq_len))
1295                         ret = -EFAULT;
1296                 break;
1297
1298         case TUNSETNOCSUM:
1299                 /* Disable/Enable checksum */
1300
1301                 /* [unimplemented] */
1302                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1303                           arg ? "disabled" : "enabled");
1304                 break;
1305
1306         case TUNSETPERSIST:
1307                 /* Disable/Enable persist mode */
1308                 if (arg)
1309                         tun->flags |= TUN_PERSIST;
1310                 else
1311                         tun->flags &= ~TUN_PERSIST;
1312
1313                 tun_debug(KERN_INFO, tun, "persist %s\n",
1314                           arg ? "enabled" : "disabled");
1315                 break;
1316
1317         case TUNSETOWNER:
1318                 /* Set owner of the device */
1319                 tun->owner = (uid_t) arg;
1320
1321                 tun_debug(KERN_INFO, tun, "owner set to %d\n", tun->owner);
1322                 break;
1323
1324         case TUNSETGROUP:
1325                 /* Set group of the device */
1326                 tun->group= (gid_t) arg;
1327
1328                 tun_debug(KERN_INFO, tun, "group set to %d\n", tun->group);
1329                 break;
1330
1331         case TUNSETLINK:
1332                 /* Only allow setting the type when the interface is down */
1333                 if (tun->dev->flags & IFF_UP) {
1334                         tun_debug(KERN_INFO, tun,
1335                                   "Linktype set failed because interface is up\n");
1336                         ret = -EBUSY;
1337                 } else {
1338                         tun->dev->type = (int) arg;
1339                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1340                                   tun->dev->type);
1341                         ret = 0;
1342                 }
1343                 break;
1344
1345 #ifdef TUN_DEBUG
1346         case TUNSETDEBUG:
1347                 tun->debug = arg;
1348                 break;
1349 #endif
1350         case TUNSETOFFLOAD:
1351                 ret = set_offload(tun, arg);
1352                 break;
1353
1354         case TUNSETTXFILTER:
1355                 /* Can be set only for TAPs */
1356                 ret = -EINVAL;
1357                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1358                         break;
1359                 ret = update_filter(&tun->txflt, (void __user *)arg);
1360                 break;
1361
1362         case SIOCGIFHWADDR:
1363                 /* Get hw address */
1364                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1365                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1366                 if (copy_to_user(argp, &ifr, ifreq_len))
1367                         ret = -EFAULT;
1368                 break;
1369
1370         case SIOCSIFHWADDR:
1371                 /* Set hw address */
1372                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1373                           ifr.ifr_hwaddr.sa_data);
1374
1375                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1376                 break;
1377
1378         case TUNGETSNDBUF:
1379                 sndbuf = tun->socket.sk->sk_sndbuf;
1380                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1381                         ret = -EFAULT;
1382                 break;
1383
1384         case TUNSETSNDBUF:
1385                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1386                         ret = -EFAULT;
1387                         break;
1388                 }
1389
1390                 tun->socket.sk->sk_sndbuf = sndbuf;
1391                 break;
1392
1393         case TUNGETVNETHDRSZ:
1394                 vnet_hdr_sz = tun->vnet_hdr_sz;
1395                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1396                         ret = -EFAULT;
1397                 break;
1398
1399         case TUNSETVNETHDRSZ:
1400                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1401                         ret = -EFAULT;
1402                         break;
1403                 }
1404                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1405                         ret = -EINVAL;
1406                         break;
1407                 }
1408
1409                 tun->vnet_hdr_sz = vnet_hdr_sz;
1410                 break;
1411
1412         case TUNATTACHFILTER:
1413                 /* Can be set only for TAPs */
1414                 ret = -EINVAL;
1415                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1416                         break;
1417                 ret = -EFAULT;
1418                 if (copy_from_user(&fprog, argp, sizeof(fprog)))
1419                         break;
1420
1421                 ret = sk_attach_filter(&fprog, tun->socket.sk);
1422                 break;
1423
1424         case TUNDETACHFILTER:
1425                 /* Can be set only for TAPs */
1426                 ret = -EINVAL;
1427                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1428                         break;
1429                 ret = sk_detach_filter(tun->socket.sk);
1430                 break;
1431
1432         default:
1433                 ret = -EINVAL;
1434                 break;
1435         }
1436
1437 unlock:
1438         rtnl_unlock();
1439         if (tun)
1440                 tun_put(tun);
1441         return ret;
1442 }
1443
1444 static long tun_chr_ioctl(struct file *file,
1445                           unsigned int cmd, unsigned long arg)
1446 {
1447         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1448 }
1449
1450 #ifdef CONFIG_COMPAT
1451 static long tun_chr_compat_ioctl(struct file *file,
1452                          unsigned int cmd, unsigned long arg)
1453 {
1454         switch (cmd) {
1455         case TUNSETIFF:
1456         case TUNGETIFF:
1457         case TUNSETTXFILTER:
1458         case TUNGETSNDBUF:
1459         case TUNSETSNDBUF:
1460         case SIOCGIFHWADDR:
1461         case SIOCSIFHWADDR:
1462                 arg = (unsigned long)compat_ptr(arg);
1463                 break;
1464         default:
1465                 arg = (compat_ulong_t)arg;
1466                 break;
1467         }
1468
1469         /*
1470          * compat_ifreq is shorter than ifreq, so we must not access beyond
1471          * the end of that structure. All fields that are used in this
1472          * driver are compatible though, we don't need to convert the
1473          * contents.
1474          */
1475         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1476 }
1477 #endif /* CONFIG_COMPAT */
1478
1479 static int tun_chr_fasync(int fd, struct file *file, int on)
1480 {
1481         struct tun_struct *tun = tun_get(file);
1482         int ret;
1483
1484         if (!tun)
1485                 return -EBADFD;
1486
1487         tun_debug(KERN_INFO, tun, "tun_chr_fasync %d\n", on);
1488
1489         if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1490                 goto out;
1491
1492         if (on) {
1493                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1494                 if (ret)
1495                         goto out;
1496                 tun->flags |= TUN_FASYNC;
1497         } else
1498                 tun->flags &= ~TUN_FASYNC;
1499         ret = 0;
1500 out:
1501         tun_put(tun);
1502         return ret;
1503 }
1504
1505 static int tun_chr_open(struct inode *inode, struct file * file)
1506 {
1507         struct tun_file *tfile;
1508
1509         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
1510
1511         tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1512         if (!tfile)
1513                 return -ENOMEM;
1514         atomic_set(&tfile->count, 0);
1515         tfile->tun = NULL;
1516         tfile->net = get_net(current->nsproxy->net_ns);
1517         file->private_data = tfile;
1518         return 0;
1519 }
1520
1521 static int tun_chr_close(struct inode *inode, struct file *file)
1522 {
1523         struct tun_file *tfile = file->private_data;
1524         struct tun_struct *tun;
1525
1526         tun = __tun_get(tfile);
1527         if (tun) {
1528                 struct net_device *dev = tun->dev;
1529
1530                 tun_debug(KERN_INFO, tun, "tun_chr_close\n");
1531
1532                 __tun_detach(tun);
1533
1534                 /* If desirable, unregister the netdevice. */
1535                 if (!(tun->flags & TUN_PERSIST)) {
1536                         rtnl_lock();
1537                         if (dev->reg_state == NETREG_REGISTERED)
1538                                 unregister_netdevice(dev);
1539                         rtnl_unlock();
1540                 }
1541         }
1542
1543         tun = tfile->tun;
1544         if (tun)
1545                 sock_put(tun->socket.sk);
1546
1547         put_net(tfile->net);
1548         kfree(tfile);
1549
1550         return 0;
1551 }
1552
1553 static const struct file_operations tun_fops = {
1554         .owner  = THIS_MODULE,
1555         .llseek = no_llseek,
1556         .read  = do_sync_read,
1557         .aio_read  = tun_chr_aio_read,
1558         .write = do_sync_write,
1559         .aio_write = tun_chr_aio_write,
1560         .poll   = tun_chr_poll,
1561         .unlocked_ioctl = tun_chr_ioctl,
1562 #ifdef CONFIG_COMPAT
1563         .compat_ioctl = tun_chr_compat_ioctl,
1564 #endif
1565         .open   = tun_chr_open,
1566         .release = tun_chr_close,
1567         .fasync = tun_chr_fasync
1568 };
1569
1570 static struct miscdevice tun_miscdev = {
1571         .minor = TUN_MINOR,
1572         .name = "tun",
1573         .nodename = "net/tun",
1574         .fops = &tun_fops,
1575 };
1576
1577 /* ethtool interface */
1578
1579 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1580 {
1581         cmd->supported          = 0;
1582         cmd->advertising        = 0;
1583         ethtool_cmd_speed_set(cmd, SPEED_10);
1584         cmd->duplex             = DUPLEX_FULL;
1585         cmd->port               = PORT_TP;
1586         cmd->phy_address        = 0;
1587         cmd->transceiver        = XCVR_INTERNAL;
1588         cmd->autoneg            = AUTONEG_DISABLE;
1589         cmd->maxtxpkt           = 0;
1590         cmd->maxrxpkt           = 0;
1591         return 0;
1592 }
1593
1594 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1595 {
1596         struct tun_struct *tun = netdev_priv(dev);
1597
1598         strcpy(info->driver, DRV_NAME);
1599         strcpy(info->version, DRV_VERSION);
1600         strcpy(info->fw_version, "N/A");
1601
1602         switch (tun->flags & TUN_TYPE_MASK) {
1603         case TUN_TUN_DEV:
1604                 strcpy(info->bus_info, "tun");
1605                 break;
1606         case TUN_TAP_DEV:
1607                 strcpy(info->bus_info, "tap");
1608                 break;
1609         }
1610 }
1611
1612 static u32 tun_get_msglevel(struct net_device *dev)
1613 {
1614 #ifdef TUN_DEBUG
1615         struct tun_struct *tun = netdev_priv(dev);
1616         return tun->debug;
1617 #else
1618         return -EOPNOTSUPP;
1619 #endif
1620 }
1621
1622 static void tun_set_msglevel(struct net_device *dev, u32 value)
1623 {
1624 #ifdef TUN_DEBUG
1625         struct tun_struct *tun = netdev_priv(dev);
1626         tun->debug = value;
1627 #endif
1628 }
1629
1630 static const struct ethtool_ops tun_ethtool_ops = {
1631         .get_settings   = tun_get_settings,
1632         .get_drvinfo    = tun_get_drvinfo,
1633         .get_msglevel   = tun_get_msglevel,
1634         .set_msglevel   = tun_set_msglevel,
1635         .get_link       = ethtool_op_get_link,
1636 };
1637
1638
1639 static int __init tun_init(void)
1640 {
1641         int ret = 0;
1642
1643         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1644         pr_info("%s\n", DRV_COPYRIGHT);
1645
1646         ret = rtnl_link_register(&tun_link_ops);
1647         if (ret) {
1648                 pr_err("Can't register link_ops\n");
1649                 goto err_linkops;
1650         }
1651
1652         ret = misc_register(&tun_miscdev);
1653         if (ret) {
1654                 pr_err("Can't register misc device %d\n", TUN_MINOR);
1655                 goto err_misc;
1656         }
1657         return  0;
1658 err_misc:
1659         rtnl_link_unregister(&tun_link_ops);
1660 err_linkops:
1661         return ret;
1662 }
1663
1664 static void tun_cleanup(void)
1665 {
1666         misc_deregister(&tun_miscdev);
1667         rtnl_link_unregister(&tun_link_ops);
1668 }
1669
1670 /* Get an underlying socket object from tun file.  Returns error unless file is
1671  * attached to a device.  The returned object works like a packet socket, it
1672  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1673  * holding a reference to the file for as long as the socket is in use. */
1674 struct socket *tun_get_socket(struct file *file)
1675 {
1676         struct tun_struct *tun;
1677         if (file->f_op != &tun_fops)
1678                 return ERR_PTR(-EINVAL);
1679         tun = tun_get(file);
1680         if (!tun)
1681                 return ERR_PTR(-EBADFD);
1682         tun_put(tun);
1683         return &tun->socket;
1684 }
1685 EXPORT_SYMBOL_GPL(tun_get_socket);
1686
1687 module_init(tun_init);
1688 module_exit(tun_cleanup);
1689 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1690 MODULE_AUTHOR(DRV_COPYRIGHT);
1691 MODULE_LICENSE("GPL");
1692 MODULE_ALIAS_MISCDEV(TUN_MINOR);
1693 MODULE_ALIAS("devname:net/tun");