Pileus Git - ~andy/linux/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 eth_random_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/uaccess.h>
  73
  74 /* Uncomment to enable debugging */
  75 /* #define TUN_DEBUG 1 */
  76
  77 #ifdef TUN_DEBUG
  78 static int debug;
  79
  80 #define tun_debug(level, tun, fmt, args...)                     \
  81 do {                                                            \
  82         if (tun->debug)                                         \
  83                 netdev_printk(level, tun->dev, fmt, ##args);    \
  84 } while (0)
  85 #define DBG1(level, fmt, args...)                               \
  86 do {                                                            \
  87         if (debug == 2)                                         \
  88                 printk(level fmt, ##args);                      \
  89 } while (0)
  90 #else
  91 #define tun_debug(level, tun, fmt, args...)                     \
  92 do {                                                            \
  93         if (0)                                                  \
  94                 netdev_printk(level, tun->dev, fmt, ##args);    \
  95 } while (0)
  96 #define DBG1(level, fmt, args...)                               \
  97 do {                                                            \
  98         if (0)                                                  \
  99                 printk(level fmt, ##args);                      \
 100 } while (0)
 101 #endif
 102
 103 #define FLT_EXACT_COUNT 8
 104 struct tap_filter {
 105         unsigned int    count;    /* Number of addrs. Zero means disabled */
 106         u32             mask[2];  /* Mask of the hashed addrs */
 107         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 108 };
 109
 110 struct tun_file {
 111         atomic_t count;
 112         struct tun_struct *tun;
 113         struct net *net;
 114 };
 115
 116 struct tun_sock;
 117
 118 struct tun_struct {
 119         struct tun_file         *tfile;
 120         unsigned int            flags;
 121         uid_t                   owner;
 122         gid_t                   group;
 123
 124         struct net_device       *dev;
 125         netdev_features_t       set_features;
 126 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 127                           NETIF_F_TSO6|NETIF_F_UFO)
 128         struct fasync_struct    *fasync;
 129
 130         struct tap_filter       txflt;
 131         struct socket           socket;
 132         struct socket_wq        wq;
 133
 134         int                     vnet_hdr_sz;
 135
 136 #ifdef TUN_DEBUG
 137         int debug;
 138 #endif
 139 };
 140
 141 struct tun_sock {
 142         struct sock             sk;
 143         struct tun_struct       *tun;
 144 };
 145
 146 static inline struct tun_sock *tun_sk(struct sock *sk)
 147 {
 148         return container_of(sk, struct tun_sock, sk);
 149 }
 150
 151 static int tun_attach(struct tun_struct *tun, struct file *file)
 152 {
 153         struct tun_file *tfile = file->private_data;
 154         int err;
 155
 156         ASSERT_RTNL();
 157
 158         netif_tx_lock_bh(tun->dev);
 159
 160         err = -EINVAL;
 161         if (tfile->tun)
 162                 goto out;
 163
 164         err = -EBUSY;
 165         if (tun->tfile)
 166                 goto out;
 167
 168         err = 0;
 169         tfile->tun = tun;
 170         tun->tfile = tfile;
 171         tun->socket.file = file;
 172         netif_carrier_on(tun->dev);
 173         dev_hold(tun->dev);
 174         sock_hold(tun->socket.sk);
 175         atomic_inc(&tfile->count);
 176
 177 out:
 178         netif_tx_unlock_bh(tun->dev);
 179         return err;
 180 }
 181
 182 static void __tun_detach(struct tun_struct *tun)
 183 {
 184         /* Detach from net device */
 185         netif_tx_lock_bh(tun->dev);
 186         netif_carrier_off(tun->dev);
 187         tun->tfile = NULL;
 188         tun->socket.file = NULL;
 189         netif_tx_unlock_bh(tun->dev);
 190
 191         /* Drop read queue */
 192         skb_queue_purge(&tun->socket.sk->sk_receive_queue);
 193
 194         /* Drop the extra count on the net device */
 195         dev_put(tun->dev);
 196 }
 197
 198 static void tun_detach(struct tun_struct *tun)
 199 {
 200         rtnl_lock();
 201         __tun_detach(tun);
 202         rtnl_unlock();
 203 }
 204
 205 static struct tun_struct *__tun_get(struct tun_file *tfile)
 206 {
 207         struct tun_struct *tun = NULL;
 208
 209         if (atomic_inc_not_zero(&tfile->count))
 210                 tun = tfile->tun;
 211
 212         return tun;
 213 }
 214
 215 static struct tun_struct *tun_get(struct file *file)
 216 {
 217         return __tun_get(file->private_data);
 218 }
 219
 220 static void tun_put(struct tun_struct *tun)
 221 {
 222         struct tun_file *tfile = tun->tfile;
 223
 224         if (atomic_dec_and_test(&tfile->count))
 225                 tun_detach(tfile->tun);
 226 }
 227
 228 /* TAP filtering */
 229 static void addr_hash_set(u32 *mask, const u8 *addr)
 230 {
 231         int n = ether_crc(ETH_ALEN, addr) >> 26;
 232         mask[n >> 5] |= (1 << (n & 31));
 233 }
 234
 235 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 236 {
 237         int n = ether_crc(ETH_ALEN, addr) >> 26;
 238         return mask[n >> 5] & (1 << (n & 31));
 239 }
 240
 241 static int update_filter(struct tap_filter *filter, void __user *arg)
 242 {
 243         struct { u8 u[ETH_ALEN]; } *addr;
 244         struct tun_filter uf;
 245         int err, alen, n, nexact;
 246
 247         if (copy_from_user(&uf, arg, sizeof(uf)))
 248                 return -EFAULT;
 249
 250         if (!uf.count) {
 251                 /* Disabled */
 252                 filter->count = 0;
 253                 return 0;
 254         }
 255
 256         alen = ETH_ALEN * uf.count;
 257         addr = kmalloc(alen, GFP_KERNEL);
 258         if (!addr)
 259                 return -ENOMEM;
 260
 261         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 262                 err = -EFAULT;
 263                 goto done;
 264         }
 265
 266         /* The filter is updated without holding any locks. Which is
 267          * perfectly safe. We disable it first and in the worst
 268          * case we'll accept a few undesired packets. */
 269         filter->count = 0;
 270         wmb();
 271
 272         /* Use first set of addresses as an exact filter */
 273         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 274                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 275
 276         nexact = n;
 277
 278         /* Remaining multicast addresses are hashed,
 279          * unicast will leave the filter disabled. */
 280         memset(filter->mask, 0, sizeof(filter->mask));
 281         for (; n < uf.count; n++) {
 282                 if (!is_multicast_ether_addr(addr[n].u)) {
 283                         err = 0; /* no filter */
 284                         goto done;
 285                 }
 286                 addr_hash_set(filter->mask, addr[n].u);
 287         }
 288
 289         /* For ALLMULTI just set the mask to all ones.
 290          * This overrides the mask populated above. */
 291         if ((uf.flags & TUN_FLT_ALLMULTI))
 292                 memset(filter->mask, ~0, sizeof(filter->mask));
 293
 294         /* Now enable the filter */
 295         wmb();
 296         filter->count = nexact;
 297
 298         /* Return the number of exact filters */
 299         err = nexact;
 300
 301 done:
 302         kfree(addr);
 303         return err;
 304 }
 305
 306 /* Returns: 0 - drop, !=0 - accept */
 307 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 308 {
 309         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 310          * at this point. */
 311         struct ethhdr *eh = (struct ethhdr *) skb->data;
 312         int i;
 313
 314         /* Exact match */
 315         for (i = 0; i < filter->count; i++)
 316                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 317                         return 1;
 318
 319         /* Inexact match (multicast only) */
 320         if (is_multicast_ether_addr(eh->h_dest))
 321                 return addr_hash_test(filter->mask, eh->h_dest);
 322
 323         return 0;
 324 }
 325
 326 /*
 327  * Checks whether the packet is accepted or not.
 328  * Returns: 0 - drop, !=0 - accept
 329  */
 330 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 331 {
 332         if (!filter->count)
 333                 return 1;
 334
 335         return run_filter(filter, skb);
 336 }
 337
 338 /* Network device part of the driver */
 339
 340 static const struct ethtool_ops tun_ethtool_ops;
 341
 342 /* Net device detach from fd. */
 343 static void tun_net_uninit(struct net_device *dev)
 344 {
 345         struct tun_struct *tun = netdev_priv(dev);
 346         struct tun_file *tfile = tun->tfile;
 347
 348         /* Inform the methods they need to stop using the dev.
 349          */
 350         if (tfile) {
 351                 wake_up_all(&tun->wq.wait);
 352                 if (atomic_dec_and_test(&tfile->count))
 353                         __tun_detach(tun);
 354         }
 355 }
 356
 357 static void tun_free_netdev(struct net_device *dev)
 358 {
 359         struct tun_struct *tun = netdev_priv(dev);
 360
 361         BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED, &tun->socket.flags));
 362
 363         sk_release_kernel(tun->socket.sk);
 364 }
 365
 366 /* Net device open. */
 367 static int tun_net_open(struct net_device *dev)
 368 {
 369         netif_start_queue(dev);
 370         return 0;
 371 }
 372
 373 /* Net device close. */
 374 static int tun_net_close(struct net_device *dev)
 375 {
 376         netif_stop_queue(dev);
 377         return 0;
 378 }
 379
 380 /* Net device start xmit */
 381 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 382 {
 383         struct tun_struct *tun = netdev_priv(dev);
 384
 385         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 386
 387         /* Drop packet if interface is not attached */
 388         if (!tun->tfile)
 389                 goto drop;
 390
 391         /* Drop if the filter does not like it.
 392          * This is a noop if the filter is disabled.
 393          * Filter can be enabled only for the TAP devices. */
 394         if (!check_filter(&tun->txflt, skb))
 395                 goto drop;
 396
 397         if (tun->socket.sk->sk_filter &&
 398             sk_filter(tun->socket.sk, skb))
 399                 goto drop;
 400
 401         if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) {
 402                 if (!(tun->flags & TUN_ONE_QUEUE)) {
 403                         /* Normal queueing mode. */
 404                         /* Packet scheduler handles dropping of further packets. */
 405                         netif_stop_queue(dev);
 406
 407                         /* We won't see all dropped packets individually, so overrun
 408                          * error is more appropriate. */
 409                         dev->stats.tx_fifo_errors++;
 410                 } else {
 411                         /* Single queue mode.
 412                          * Driver handles dropping of all packets itself. */
 413                         goto drop;
 414                 }
 415         }
 416
 417         /* Orphan the skb - required as we might hang on to it
 418          * for indefinite time. */
 419         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 420                 goto drop;
 421         skb_orphan(skb);
 422
 423         /* Enqueue packet */
 424         skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb);
 425
 426         /* Notify and wake up reader process */
 427         if (tun->flags & TUN_FASYNC)
 428                 kill_fasync(&tun->fasync, SIGIO, POLL_IN);
 429         wake_up_interruptible_poll(&tun->wq.wait, POLLIN |
 430                                    POLLRDNORM | POLLRDBAND);
 431         return NETDEV_TX_OK;
 432
 433 drop:
 434         dev->stats.tx_dropped++;
 435         kfree_skb(skb);
 436         return NETDEV_TX_OK;
 437 }
 438
 439 static void tun_net_mclist(struct net_device *dev)
 440 {
 441         /*
 442          * This callback is supposed to deal with mc filter in
 443          * _rx_ path and has nothing to do with the _tx_ path.
 444          * In rx path we always accept everything userspace gives us.
 445          */
 446 }
 447
 448 #define MIN_MTU 68
 449 #define MAX_MTU 65535
 450
 451 static int
 452 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 453 {
 454         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 455                 return -EINVAL;
 456         dev->mtu = new_mtu;
 457         return 0;
 458 }
 459
 460 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 461         netdev_features_t features)
 462 {
 463         struct tun_struct *tun = netdev_priv(dev);
 464
 465         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 466 }
 467 #ifdef CONFIG_NET_POLL_CONTROLLER
 468 static void tun_poll_controller(struct net_device *dev)
 469 {
 470         /*
 471          * Tun only receives frames when:
 472          * 1) the char device endpoint gets data from user space
 473          * 2) the tun socket gets a sendmsg call from user space
 474          * Since both of those are syncronous operations, we are guaranteed
 475          * never to have pending data when we poll for it
 476          * so theres nothing to do here but return.
 477          * We need this though so netpoll recognizes us as an interface that
 478          * supports polling, which enables bridge devices in virt setups to
 479          * still use netconsole
 480          */
 481         return;
 482 }
 483 #endif
 484 static const struct net_device_ops tun_netdev_ops = {
 485         .ndo_uninit             = tun_net_uninit,
 486         .ndo_open               = tun_net_open,
 487         .ndo_stop               = tun_net_close,
 488         .ndo_start_xmit         = tun_net_xmit,
 489         .ndo_change_mtu         = tun_net_change_mtu,
 490         .ndo_fix_features       = tun_net_fix_features,
 491 #ifdef CONFIG_NET_POLL_CONTROLLER
 492         .ndo_poll_controller    = tun_poll_controller,
 493 #endif
 494 };
 495
 496 static const struct net_device_ops tap_netdev_ops = {
 497         .ndo_uninit             = tun_net_uninit,
 498         .ndo_open               = tun_net_open,
 499         .ndo_stop               = tun_net_close,
 500         .ndo_start_xmit         = tun_net_xmit,
 501         .ndo_change_mtu         = tun_net_change_mtu,
 502         .ndo_fix_features       = tun_net_fix_features,
 503         .ndo_set_rx_mode        = tun_net_mclist,
 504         .ndo_set_mac_address    = eth_mac_addr,
 505         .ndo_validate_addr      = eth_validate_addr,
 506 #ifdef CONFIG_NET_POLL_CONTROLLER
 507         .ndo_poll_controller    = tun_poll_controller,
 508 #endif
 509 };
 510
 511 /* Initialize net device. */
 512 static void tun_net_init(struct net_device *dev)
 513 {
 514         struct tun_struct *tun = netdev_priv(dev);
 515
 516         switch (tun->flags & TUN_TYPE_MASK) {
 517         case TUN_TUN_DEV:
 518                 dev->netdev_ops = &tun_netdev_ops;
 519
 520                 /* Point-to-Point TUN Device */
 521                 dev->hard_header_len = 0;
 522                 dev->addr_len = 0;
 523                 dev->mtu = 1500;
 524
 525                 /* Zero header length */
 526                 dev->type = ARPHRD_NONE;
 527                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 528                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 529                 break;
 530
 531         case TUN_TAP_DEV:
 532                 dev->netdev_ops = &tap_netdev_ops;
 533                 /* Ethernet TAP Device */
 534                 ether_setup(dev);
 535                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 536
 537                 eth_hw_addr_random(dev);
 538
 539                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 540                 break;
 541         }
 542 }
 543
 544 /* Character device part */
 545
 546 /* Poll */
 547 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
 548 {
 549         struct tun_file *tfile = file->private_data;
 550         struct tun_struct *tun = __tun_get(tfile);
 551         struct sock *sk;
 552         unsigned int mask = 0;
 553
 554         if (!tun)
 555                 return POLLERR;
 556
 557         sk = tun->socket.sk;
 558
 559         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 560
 561         poll_wait(file, &tun->wq.wait, wait);
 562
 563         if (!skb_queue_empty(&sk->sk_receive_queue))
 564                 mask |= POLLIN | POLLRDNORM;
 565
 566         if (sock_writeable(sk) ||
 567             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 568              sock_writeable(sk)))
 569                 mask |= POLLOUT | POLLWRNORM;
 570
 571         if (tun->dev->reg_state != NETREG_REGISTERED)
 572                 mask = POLLERR;
 573
 574         tun_put(tun);
 575         return mask;
 576 }
 577
 578 /* prepad is the amount to reserve at front.  len is length after that.
 579  * linear is a hint as to how much to copy (usually headers). */
 580 static struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
 581                                      size_t prepad, size_t len,
 582                                      size_t linear, int noblock)
 583 {
 584         struct sock *sk = tun->socket.sk;
 585         struct sk_buff *skb;
 586         int err;
 587
 588         sock_update_classid(sk);
 589
 590         /* Under a page?  Don't bother with paged skb. */
 591         if (prepad + len < PAGE_SIZE || !linear)
 592                 linear = len;
 593
 594         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 595                                    &err);
 596         if (!skb)
 597                 return ERR_PTR(err);
 598
 599         skb_reserve(skb, prepad);
 600         skb_put(skb, linear);
 601         skb->data_len = len - linear;
 602         skb->len += len - linear;
 603
 604         return skb;
 605 }
 606
 607 /* Get packet from user space buffer */
 608 static ssize_t tun_get_user(struct tun_struct *tun,
 609                             const struct iovec *iv, size_t count,
 610                             int noblock)
 611 {
 612         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
 613         struct sk_buff *skb;
 614         size_t len = count, align = NET_SKB_PAD;
 615         struct virtio_net_hdr gso = { 0 };
 616         int offset = 0;
 617
 618         if (!(tun->flags & TUN_NO_PI)) {
 619                 if ((len -= sizeof(pi)) > count)
 620                         return -EINVAL;
 621
 622                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
 623                         return -EFAULT;
 624                 offset += sizeof(pi);
 625         }
 626
 627         if (tun->flags & TUN_VNET_HDR) {
 628                 if ((len -= tun->vnet_hdr_sz) > count)
 629                         return -EINVAL;
 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 static int tun_release(struct socket *sock)
 987 {
 988         if (sock->sk)
 989                 sock_put(sock->sk);
 990         return 0;
 991 }
 992
 993 /* Ops structure to mimic raw sockets with tun */
 994 static const struct proto_ops tun_socket_ops = {
 995         .sendmsg = tun_sendmsg,
 996         .recvmsg = tun_recvmsg,
 997         .release = tun_release,
 998 };
 999
1000 static struct proto tun_proto = {
1001         .name           = "tun",
1002         .owner          = THIS_MODULE,
1003         .obj_size       = sizeof(struct tun_sock),
1004 };
1005
1006 static int tun_flags(struct tun_struct *tun)
1007 {
1008         int flags = 0;
1009
1010         if (tun->flags & TUN_TUN_DEV)
1011                 flags |= IFF_TUN;
1012         else
1013                 flags |= IFF_TAP;
1014
1015         if (tun->flags & TUN_NO_PI)
1016                 flags |= IFF_NO_PI;
1017
1018         if (tun->flags & TUN_ONE_QUEUE)
1019                 flags |= IFF_ONE_QUEUE;
1020
1021         if (tun->flags & TUN_VNET_HDR)
1022                 flags |= IFF_VNET_HDR;
1023
1024         return flags;
1025 }
1026
1027 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1028                               char *buf)
1029 {
1030         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1031         return sprintf(buf, "0x%x\n", tun_flags(tun));
1032 }
1033
1034 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1035                               char *buf)
1036 {
1037         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1038         return sprintf(buf, "%d\n", tun->owner);
1039 }
1040
1041 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1042                               char *buf)
1043 {
1044         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1045         return sprintf(buf, "%d\n", tun->group);
1046 }
1047
1048 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1049 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1050 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1051
1052 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1053 {
1054         struct sock *sk;
1055         struct tun_struct *tun;
1056         struct net_device *dev;
1057         int err;
1058
1059         dev = __dev_get_by_name(net, ifr->ifr_name);
1060         if (dev) {
1061                 const struct cred *cred = current_cred();
1062
1063                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1064                         return -EBUSY;
1065                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1066                         tun = netdev_priv(dev);
1067                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1068                         tun = netdev_priv(dev);
1069                 else
1070                         return -EINVAL;
1071
1072                 if (((tun->owner != -1 && cred->euid != tun->owner) ||
1073                      (tun->group != -1 && !in_egroup_p(tun->group))) &&
1074                     !capable(CAP_NET_ADMIN))
1075                         return -EPERM;
1076                 err = security_tun_dev_attach(tun->socket.sk);
1077                 if (err < 0)
1078                         return err;
1079
1080                 err = tun_attach(tun, file);
1081                 if (err < 0)
1082                         return err;
1083         }
1084         else {
1085                 char *name;
1086                 unsigned long flags = 0;
1087
1088                 if (!capable(CAP_NET_ADMIN))
1089                         return -EPERM;
1090                 err = security_tun_dev_create();
1091                 if (err < 0)
1092                         return err;
1093
1094                 /* Set dev type */
1095                 if (ifr->ifr_flags & IFF_TUN) {
1096                         /* TUN device */
1097                         flags |= TUN_TUN_DEV;
1098                         name = "tun%d";
1099                 } else if (ifr->ifr_flags & IFF_TAP) {
1100                         /* TAP device */
1101                         flags |= TUN_TAP_DEV;
1102                         name = "tap%d";
1103                 } else
1104                         return -EINVAL;
1105
1106                 if (*ifr->ifr_name)
1107                         name = ifr->ifr_name;
1108
1109                 dev = alloc_netdev(sizeof(struct tun_struct), name,
1110                                    tun_setup);
1111                 if (!dev)
1112                         return -ENOMEM;
1113
1114                 dev_net_set(dev, net);
1115                 dev->rtnl_link_ops = &tun_link_ops;
1116
1117                 tun = netdev_priv(dev);
1118                 tun->dev = dev;
1119                 tun->flags = flags;
1120                 tun->txflt.count = 0;
1121                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1122                 set_bit(SOCK_EXTERNALLY_ALLOCATED, &tun->socket.flags);
1123
1124                 err = -ENOMEM;
1125                 sk = sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1126                 if (!sk)
1127                         goto err_free_dev;
1128
1129                 sk_change_net(sk, net);
1130                 tun->socket.wq = &tun->wq;
1131                 init_waitqueue_head(&tun->wq.wait);
1132                 tun->socket.ops = &tun_socket_ops;
1133                 sock_init_data(&tun->socket, sk);
1134                 sk->sk_write_space = tun_sock_write_space;
1135                 sk->sk_sndbuf = INT_MAX;
1136
1137                 tun_sk(sk)->tun = tun;
1138
1139                 security_tun_dev_post_create(sk);
1140
1141                 tun_net_init(dev);
1142
1143                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1144                         TUN_USER_FEATURES;
1145                 dev->features = dev->hw_features;
1146
1147                 err = register_netdevice(tun->dev);
1148                 if (err < 0)
1149                         goto err_free_sk;
1150
1151                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1152                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1153                     device_create_file(&tun->dev->dev, &dev_attr_group))
1154                         pr_err("Failed to create tun sysfs files\n");
1155
1156                 sk->sk_destruct = tun_sock_destruct;
1157
1158                 err = tun_attach(tun, file);
1159                 if (err < 0)
1160                         goto failed;
1161         }
1162
1163         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1164
1165         if (ifr->ifr_flags & IFF_NO_PI)
1166                 tun->flags |= TUN_NO_PI;
1167         else
1168                 tun->flags &= ~TUN_NO_PI;
1169
1170         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1171                 tun->flags |= TUN_ONE_QUEUE;
1172         else
1173                 tun->flags &= ~TUN_ONE_QUEUE;
1174
1175         if (ifr->ifr_flags & IFF_VNET_HDR)
1176                 tun->flags |= TUN_VNET_HDR;
1177         else
1178                 tun->flags &= ~TUN_VNET_HDR;
1179
1180         /* Make sure persistent devices do not get stuck in
1181          * xoff state.
1182          */
1183         if (netif_running(tun->dev))
1184                 netif_wake_queue(tun->dev);
1185
1186         strcpy(ifr->ifr_name, tun->dev->name);
1187         return 0;
1188
1189  err_free_sk:
1190         tun_free_netdev(dev);
1191  err_free_dev:
1192         free_netdev(dev);
1193  failed:
1194         return err;
1195 }
1196
1197 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1198                        struct ifreq *ifr)
1199 {
1200         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1201
1202         strcpy(ifr->ifr_name, tun->dev->name);
1203
1204         ifr->ifr_flags = tun_flags(tun);
1205
1206         return 0;
1207 }
1208
1209 /* This is like a cut-down ethtool ops, except done via tun fd so no
1210  * privs required. */
1211 static int set_offload(struct tun_struct *tun, unsigned long arg)
1212 {
1213         netdev_features_t features = 0;
1214
1215         if (arg & TUN_F_CSUM) {
1216                 features |= NETIF_F_HW_CSUM;
1217                 arg &= ~TUN_F_CSUM;
1218
1219                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1220                         if (arg & TUN_F_TSO_ECN) {
1221                                 features |= NETIF_F_TSO_ECN;
1222                                 arg &= ~TUN_F_TSO_ECN;
1223                         }
1224                         if (arg & TUN_F_TSO4)
1225                                 features |= NETIF_F_TSO;
1226                         if (arg & TUN_F_TSO6)
1227                                 features |= NETIF_F_TSO6;
1228                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1229                 }
1230
1231                 if (arg & TUN_F_UFO) {
1232                         features |= NETIF_F_UFO;
1233                         arg &= ~TUN_F_UFO;
1234                 }
1235         }
1236
1237         /* This gives the user a way to test for new features in future by
1238          * trying to set them. */
1239         if (arg)
1240                 return -EINVAL;
1241
1242         tun->set_features = features;
1243         netdev_update_features(tun->dev);
1244
1245         return 0;
1246 }
1247
1248 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1249                             unsigned long arg, int ifreq_len)
1250 {
1251         struct tun_file *tfile = file->private_data;
1252         struct tun_struct *tun;
1253         void __user* argp = (void __user*)arg;
1254         struct sock_fprog fprog;
1255         struct ifreq ifr;
1256         int sndbuf;
1257         int vnet_hdr_sz;
1258         int ret;
1259
1260         if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
1261                 if (copy_from_user(&ifr, argp, ifreq_len))
1262                         return -EFAULT;
1263
1264         if (cmd == TUNGETFEATURES) {
1265                 /* Currently this just means: "what IFF flags are valid?".
1266                  * This is needed because we never checked for invalid flags on
1267                  * TUNSETIFF. */
1268                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1269                                 IFF_VNET_HDR,
1270                                 (unsigned int __user*)argp);
1271         }
1272
1273         rtnl_lock();
1274
1275         tun = __tun_get(tfile);
1276         if (cmd == TUNSETIFF && !tun) {
1277                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1278
1279                 ret = tun_set_iff(tfile->net, file, &ifr);
1280
1281                 if (ret)
1282                         goto unlock;
1283
1284                 if (copy_to_user(argp, &ifr, ifreq_len))
1285                         ret = -EFAULT;
1286                 goto unlock;
1287         }
1288
1289         ret = -EBADFD;
1290         if (!tun)
1291                 goto unlock;
1292
1293         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %d\n", cmd);
1294
1295         ret = 0;
1296         switch (cmd) {
1297         case TUNGETIFF:
1298                 ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1299                 if (ret)
1300                         break;
1301
1302                 if (copy_to_user(argp, &ifr, ifreq_len))
1303                         ret = -EFAULT;
1304                 break;
1305
1306         case TUNSETNOCSUM:
1307                 /* Disable/Enable checksum */
1308
1309                 /* [unimplemented] */
1310                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1311                           arg ? "disabled" : "enabled");
1312                 break;
1313
1314         case TUNSETPERSIST:
1315                 /* Disable/Enable persist mode */
1316                 if (arg)
1317                         tun->flags |= TUN_PERSIST;
1318                 else
1319                         tun->flags &= ~TUN_PERSIST;
1320
1321                 tun_debug(KERN_INFO, tun, "persist %s\n",
1322                           arg ? "enabled" : "disabled");
1323                 break;
1324
1325         case TUNSETOWNER:
1326                 /* Set owner of the device */
1327                 tun->owner = (uid_t) arg;
1328
1329                 tun_debug(KERN_INFO, tun, "owner set to %d\n", tun->owner);
1330                 break;
1331
1332         case TUNSETGROUP:
1333                 /* Set group of the device */
1334                 tun->group= (gid_t) arg;
1335
1336                 tun_debug(KERN_INFO, tun, "group set to %d\n", tun->group);
1337                 break;
1338
1339         case TUNSETLINK:
1340                 /* Only allow setting the type when the interface is down */
1341                 if (tun->dev->flags & IFF_UP) {
1342                         tun_debug(KERN_INFO, tun,
1343                                   "Linktype set failed because interface is up\n");
1344                         ret = -EBUSY;
1345                 } else {
1346                         tun->dev->type = (int) arg;
1347                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1348                                   tun->dev->type);
1349                         ret = 0;
1350                 }
1351                 break;
1352
1353 #ifdef TUN_DEBUG
1354         case TUNSETDEBUG:
1355                 tun->debug = arg;
1356                 break;
1357 #endif
1358         case TUNSETOFFLOAD:
1359                 ret = set_offload(tun, arg);
1360                 break;
1361
1362         case TUNSETTXFILTER:
1363                 /* Can be set only for TAPs */
1364                 ret = -EINVAL;
1365                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1366                         break;
1367                 ret = update_filter(&tun->txflt, (void __user *)arg);
1368                 break;
1369
1370         case SIOCGIFHWADDR:
1371                 /* Get hw address */
1372                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1373                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1374                 if (copy_to_user(argp, &ifr, ifreq_len))
1375                         ret = -EFAULT;
1376                 break;
1377
1378         case SIOCSIFHWADDR:
1379                 /* Set hw address */
1380                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1381                           ifr.ifr_hwaddr.sa_data);
1382
1383                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1384                 break;
1385
1386         case TUNGETSNDBUF:
1387                 sndbuf = tun->socket.sk->sk_sndbuf;
1388                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1389                         ret = -EFAULT;
1390                 break;
1391
1392         case TUNSETSNDBUF:
1393                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1394                         ret = -EFAULT;
1395                         break;
1396                 }
1397
1398                 tun->socket.sk->sk_sndbuf = sndbuf;
1399                 break;
1400
1401         case TUNGETVNETHDRSZ:
1402                 vnet_hdr_sz = tun->vnet_hdr_sz;
1403                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1404                         ret = -EFAULT;
1405                 break;
1406
1407         case TUNSETVNETHDRSZ:
1408                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1409                         ret = -EFAULT;
1410                         break;
1411                 }
1412                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1413                         ret = -EINVAL;
1414                         break;
1415                 }
1416
1417                 tun->vnet_hdr_sz = vnet_hdr_sz;
1418                 break;
1419
1420         case TUNATTACHFILTER:
1421                 /* Can be set only for TAPs */
1422                 ret = -EINVAL;
1423                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1424                         break;
1425                 ret = -EFAULT;
1426                 if (copy_from_user(&fprog, argp, sizeof(fprog)))
1427                         break;
1428
1429                 ret = sk_attach_filter(&fprog, tun->socket.sk);
1430                 break;
1431
1432         case TUNDETACHFILTER:
1433                 /* Can be set only for TAPs */
1434                 ret = -EINVAL;
1435                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1436                         break;
1437                 ret = sk_detach_filter(tun->socket.sk);
1438                 break;
1439
1440         default:
1441                 ret = -EINVAL;
1442                 break;
1443         }
1444
1445 unlock:
1446         rtnl_unlock();
1447         if (tun)
1448                 tun_put(tun);
1449         return ret;
1450 }
1451
1452 static long tun_chr_ioctl(struct file *file,
1453                           unsigned int cmd, unsigned long arg)
1454 {
1455         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1456 }
1457
1458 #ifdef CONFIG_COMPAT
1459 static long tun_chr_compat_ioctl(struct file *file,
1460                          unsigned int cmd, unsigned long arg)
1461 {
1462         switch (cmd) {
1463         case TUNSETIFF:
1464         case TUNGETIFF:
1465         case TUNSETTXFILTER:
1466         case TUNGETSNDBUF:
1467         case TUNSETSNDBUF:
1468         case SIOCGIFHWADDR:
1469         case SIOCSIFHWADDR:
1470                 arg = (unsigned long)compat_ptr(arg);
1471                 break;
1472         default:
1473                 arg = (compat_ulong_t)arg;
1474                 break;
1475         }
1476
1477         /*
1478          * compat_ifreq is shorter than ifreq, so we must not access beyond
1479          * the end of that structure. All fields that are used in this
1480          * driver are compatible though, we don't need to convert the
1481          * contents.
1482          */
1483         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1484 }
1485 #endif /* CONFIG_COMPAT */
1486
1487 static int tun_chr_fasync(int fd, struct file *file, int on)
1488 {
1489         struct tun_struct *tun = tun_get(file);
1490         int ret;
1491
1492         if (!tun)
1493                 return -EBADFD;
1494
1495         tun_debug(KERN_INFO, tun, "tun_chr_fasync %d\n", on);
1496
1497         if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1498                 goto out;
1499
1500         if (on) {
1501                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1502                 if (ret)
1503                         goto out;
1504                 tun->flags |= TUN_FASYNC;
1505         } else
1506                 tun->flags &= ~TUN_FASYNC;
1507         ret = 0;
1508 out:
1509         tun_put(tun);
1510         return ret;
1511 }
1512
1513 static int tun_chr_open(struct inode *inode, struct file * file)
1514 {
1515         struct tun_file *tfile;
1516
1517         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
1518
1519         tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1520         if (!tfile)
1521                 return -ENOMEM;
1522         atomic_set(&tfile->count, 0);
1523         tfile->tun = NULL;
1524         tfile->net = get_net(current->nsproxy->net_ns);
1525         file->private_data = tfile;
1526         return 0;
1527 }
1528
1529 static int tun_chr_close(struct inode *inode, struct file *file)
1530 {
1531         struct tun_file *tfile = file->private_data;
1532         struct tun_struct *tun;
1533
1534         tun = __tun_get(tfile);
1535         if (tun) {
1536                 struct net_device *dev = tun->dev;
1537
1538                 tun_debug(KERN_INFO, tun, "tun_chr_close\n");
1539
1540                 __tun_detach(tun);
1541
1542                 /* If desirable, unregister the netdevice. */
1543                 if (!(tun->flags & TUN_PERSIST)) {
1544                         rtnl_lock();
1545                         if (dev->reg_state == NETREG_REGISTERED)
1546                                 unregister_netdevice(dev);
1547                         rtnl_unlock();
1548                 }
1549         }
1550
1551         tun = tfile->tun;
1552         if (tun)
1553                 sock_put(tun->socket.sk);
1554
1555         put_net(tfile->net);
1556         kfree(tfile);
1557
1558         return 0;
1559 }
1560
1561 static const struct file_operations tun_fops = {
1562         .owner  = THIS_MODULE,
1563         .llseek = no_llseek,
1564         .read  = do_sync_read,
1565         .aio_read  = tun_chr_aio_read,
1566         .write = do_sync_write,
1567         .aio_write = tun_chr_aio_write,
1568         .poll   = tun_chr_poll,
1569         .unlocked_ioctl = tun_chr_ioctl,
1570 #ifdef CONFIG_COMPAT
1571         .compat_ioctl = tun_chr_compat_ioctl,
1572 #endif
1573         .open   = tun_chr_open,
1574         .release = tun_chr_close,
1575         .fasync = tun_chr_fasync
1576 };
1577
1578 static struct miscdevice tun_miscdev = {
1579         .minor = TUN_MINOR,
1580         .name = "tun",
1581         .nodename = "net/tun",
1582         .fops = &tun_fops,
1583 };
1584
1585 /* ethtool interface */
1586
1587 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1588 {
1589         cmd->supported          = 0;
1590         cmd->advertising        = 0;
1591         ethtool_cmd_speed_set(cmd, SPEED_10);
1592         cmd->duplex             = DUPLEX_FULL;
1593         cmd->port               = PORT_TP;
1594         cmd->phy_address        = 0;
1595         cmd->transceiver        = XCVR_INTERNAL;
1596         cmd->autoneg            = AUTONEG_DISABLE;
1597         cmd->maxtxpkt           = 0;
1598         cmd->maxrxpkt           = 0;
1599         return 0;
1600 }
1601
1602 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1603 {
1604         struct tun_struct *tun = netdev_priv(dev);
1605
1606         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1607         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1608
1609         switch (tun->flags & TUN_TYPE_MASK) {
1610         case TUN_TUN_DEV:
1611                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
1612                 break;
1613         case TUN_TAP_DEV:
1614                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
1615                 break;
1616         }
1617 }
1618
1619 static u32 tun_get_msglevel(struct net_device *dev)
1620 {
1621 #ifdef TUN_DEBUG
1622         struct tun_struct *tun = netdev_priv(dev);
1623         return tun->debug;
1624 #else
1625         return -EOPNOTSUPP;
1626 #endif
1627 }
1628
1629 static void tun_set_msglevel(struct net_device *dev, u32 value)
1630 {
1631 #ifdef TUN_DEBUG
1632         struct tun_struct *tun = netdev_priv(dev);
1633         tun->debug = value;
1634 #endif
1635 }
1636
1637 static const struct ethtool_ops tun_ethtool_ops = {
1638         .get_settings   = tun_get_settings,
1639         .get_drvinfo    = tun_get_drvinfo,
1640         .get_msglevel   = tun_get_msglevel,
1641         .set_msglevel   = tun_set_msglevel,
1642         .get_link       = ethtool_op_get_link,
1643 };
1644
1645
1646 static int __init tun_init(void)
1647 {
1648         int ret = 0;
1649
1650         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1651         pr_info("%s\n", DRV_COPYRIGHT);
1652
1653         ret = rtnl_link_register(&tun_link_ops);
1654         if (ret) {
1655                 pr_err("Can't register link_ops\n");
1656                 goto err_linkops;
1657         }
1658
1659         ret = misc_register(&tun_miscdev);
1660         if (ret) {
1661                 pr_err("Can't register misc device %d\n", TUN_MINOR);
1662                 goto err_misc;
1663         }
1664         return  0;
1665 err_misc:
1666         rtnl_link_unregister(&tun_link_ops);
1667 err_linkops:
1668         return ret;
1669 }
1670
1671 static void tun_cleanup(void)
1672 {
1673         misc_deregister(&tun_miscdev);
1674         rtnl_link_unregister(&tun_link_ops);
1675 }
1676
1677 /* Get an underlying socket object from tun file.  Returns error unless file is
1678  * attached to a device.  The returned object works like a packet socket, it
1679  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1680  * holding a reference to the file for as long as the socket is in use. */
1681 struct socket *tun_get_socket(struct file *file)
1682 {
1683         struct tun_struct *tun;
1684         if (file->f_op != &tun_fops)
1685                 return ERR_PTR(-EINVAL);
1686         tun = tun_get(file);
1687         if (!tun)
1688                 return ERR_PTR(-EBADFD);
1689         tun_put(tun);
1690         return &tun->socket;
1691 }
1692 EXPORT_SYMBOL_GPL(tun_get_socket);
1693
1694 module_init(tun_init);
1695 module_exit(tun_cleanup);
1696 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1697 MODULE_AUTHOR(DRV_COPYRIGHT);
1698 MODULE_LICENSE("GPL");
1699 MODULE_ALIAS_MISCDEV(TUN_MINOR);
1700 MODULE_ALIAS("devname:net/tun");