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/ipv6.h>
  68 #include <net/net_namespace.h>
  69 #include <net/netns/generic.h>
  70 #include <net/rtnetlink.h>
  71 #include <net/sock.h>
  72
  73 #include <asm/system.h>
  74 #include <asm/uaccess.h>
  75
  76 /* Uncomment to enable debugging */
  77 /* #define TUN_DEBUG 1 */
  78
  79 #ifdef TUN_DEBUG
  80 static int debug;
  81
  82 #define tun_debug(level, tun, fmt, args...)                     \
  83 do {                                                            \
  84         if (tun->debug)                                         \
  85                 netdev_printk(level, tun->dev, fmt, ##args);    \
  86 } while (0)
  87 #define DBG1(level, fmt, args...)                               \
  88 do {                                                            \
  89         if (debug == 2)                                         \
  90                 printk(level fmt, ##args);                      \
  91 } while (0)
  92 #else
  93 #define tun_debug(level, tun, fmt, args...)                     \
  94 do {                                                            \
  95         if (0)                                                  \
  96                 netdev_printk(level, tun->dev, fmt, ##args);    \
  97 } while (0)
  98 #define DBG1(level, fmt, args...)                               \
  99 do {                                                            \
 100         if (0)                                                  \
 101                 printk(level fmt, ##args);                      \
 102 } while (0)
 103 #endif
 104
 105 #define FLT_EXACT_COUNT 8
 106 struct tap_filter {
 107         unsigned int    count;    /* Number of addrs. Zero means disabled */
 108         u32             mask[2];  /* Mask of the hashed addrs */
 109         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 110 };
 111
 112 struct tun_file {
 113         atomic_t count;
 114         struct tun_struct *tun;
 115         struct net *net;
 116 };
 117
 118 struct tun_sock;
 119
 120 struct tun_struct {
 121         struct tun_file         *tfile;
 122         unsigned int            flags;
 123         uid_t                   owner;
 124         gid_t                   group;
 125
 126         struct net_device       *dev;
 127         u32                     set_features;
 128 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 129                           NETIF_F_TSO6|NETIF_F_UFO)
 130         struct fasync_struct    *fasync;
 131
 132         struct tap_filter       txflt;
 133         struct socket           socket;
 134         struct socket_wq        wq;
 135
 136         int                     vnet_hdr_sz;
 137
 138 #ifdef TUN_DEBUG
 139         int debug;
 140 #endif
 141 };
 142
 143 struct tun_sock {
 144         struct sock             sk;
 145         struct tun_struct       *tun;
 146 };
 147
 148 static inline struct tun_sock *tun_sk(struct sock *sk)
 149 {
 150         return container_of(sk, struct tun_sock, sk);
 151 }
 152
 153 static int tun_attach(struct tun_struct *tun, struct file *file)
 154 {
 155         struct tun_file *tfile = file->private_data;
 156         int err;
 157
 158         ASSERT_RTNL();
 159
 160         netif_tx_lock_bh(tun->dev);
 161
 162         err = -EINVAL;
 163         if (tfile->tun)
 164                 goto out;
 165
 166         err = -EBUSY;
 167         if (tun->tfile)
 168                 goto out;
 169
 170         err = 0;
 171         tfile->tun = tun;
 172         tun->tfile = tfile;
 173         tun->socket.file = file;
 174         netif_carrier_on(tun->dev);
 175         dev_hold(tun->dev);
 176         sock_hold(tun->socket.sk);
 177         atomic_inc(&tfile->count);
 178
 179 out:
 180         netif_tx_unlock_bh(tun->dev);
 181         return err;
 182 }
 183
 184 static void __tun_detach(struct tun_struct *tun)
 185 {
 186         /* Detach from net device */
 187         netif_tx_lock_bh(tun->dev);
 188         netif_carrier_off(tun->dev);
 189         tun->tfile = NULL;
 190         tun->socket.file = NULL;
 191         netif_tx_unlock_bh(tun->dev);
 192
 193         /* Drop read queue */
 194         skb_queue_purge(&tun->socket.sk->sk_receive_queue);
 195
 196         /* Drop the extra count on the net device */
 197         dev_put(tun->dev);
 198 }
 199
 200 static void tun_detach(struct tun_struct *tun)
 201 {
 202         rtnl_lock();
 203         __tun_detach(tun);
 204         rtnl_unlock();
 205 }
 206
 207 static struct tun_struct *__tun_get(struct tun_file *tfile)
 208 {
 209         struct tun_struct *tun = NULL;
 210
 211         if (atomic_inc_not_zero(&tfile->count))
 212                 tun = tfile->tun;
 213
 214         return tun;
 215 }
 216
 217 static struct tun_struct *tun_get(struct file *file)
 218 {
 219         return __tun_get(file->private_data);
 220 }
 221
 222 static void tun_put(struct tun_struct *tun)
 223 {
 224         struct tun_file *tfile = tun->tfile;
 225
 226         if (atomic_dec_and_test(&tfile->count))
 227                 tun_detach(tfile->tun);
 228 }
 229
 230 /* TAP filtering */
 231 static void addr_hash_set(u32 *mask, const u8 *addr)
 232 {
 233         int n = ether_crc(ETH_ALEN, addr) >> 26;
 234         mask[n >> 5] |= (1 << (n & 31));
 235 }
 236
 237 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 238 {
 239         int n = ether_crc(ETH_ALEN, addr) >> 26;
 240         return mask[n >> 5] & (1 << (n & 31));
 241 }
 242
 243 static int update_filter(struct tap_filter *filter, void __user *arg)
 244 {
 245         struct { u8 u[ETH_ALEN]; } *addr;
 246         struct tun_filter uf;
 247         int err, alen, n, nexact;
 248
 249         if (copy_from_user(&uf, arg, sizeof(uf)))
 250                 return -EFAULT;
 251
 252         if (!uf.count) {
 253                 /* Disabled */
 254                 filter->count = 0;
 255                 return 0;
 256         }
 257
 258         alen = ETH_ALEN * uf.count;
 259         addr = kmalloc(alen, GFP_KERNEL);
 260         if (!addr)
 261                 return -ENOMEM;
 262
 263         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 264                 err = -EFAULT;
 265                 goto done;
 266         }
 267
 268         /* The filter is updated without holding any locks. Which is
 269          * perfectly safe. We disable it first and in the worst
 270          * case we'll accept a few undesired packets. */
 271         filter->count = 0;
 272         wmb();
 273
 274         /* Use first set of addresses as an exact filter */
 275         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 276                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 277
 278         nexact = n;
 279
 280         /* Remaining multicast addresses are hashed,
 281          * unicast will leave the filter disabled. */
 282         memset(filter->mask, 0, sizeof(filter->mask));
 283         for (; n < uf.count; n++) {
 284                 if (!is_multicast_ether_addr(addr[n].u)) {
 285                         err = 0; /* no filter */
 286                         goto done;
 287                 }
 288                 addr_hash_set(filter->mask, addr[n].u);
 289         }
 290
 291         /* For ALLMULTI just set the mask to all ones.
 292          * This overrides the mask populated above. */
 293         if ((uf.flags & TUN_FLT_ALLMULTI))
 294                 memset(filter->mask, ~0, sizeof(filter->mask));
 295
 296         /* Now enable the filter */
 297         wmb();
 298         filter->count = nexact;
 299
 300         /* Return the number of exact filters */
 301         err = nexact;
 302
 303 done:
 304         kfree(addr);
 305         return err;
 306 }
 307
 308 /* Returns: 0 - drop, !=0 - accept */
 309 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 310 {
 311         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 312          * at this point. */
 313         struct ethhdr *eh = (struct ethhdr *) skb->data;
 314         int i;
 315
 316         /* Exact match */
 317         for (i = 0; i < filter->count; i++)
 318                 if (!compare_ether_addr(eh->h_dest, filter->addr[i]))
 319                         return 1;
 320
 321         /* Inexact match (multicast only) */
 322         if (is_multicast_ether_addr(eh->h_dest))
 323                 return addr_hash_test(filter->mask, eh->h_dest);
 324
 325         return 0;
 326 }
 327
 328 /*
 329  * Checks whether the packet is accepted or not.
 330  * Returns: 0 - drop, !=0 - accept
 331  */
 332 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 333 {
 334         if (!filter->count)
 335                 return 1;
 336
 337         return run_filter(filter, skb);
 338 }
 339
 340 /* Network device part of the driver */
 341
 342 static const struct ethtool_ops tun_ethtool_ops;
 343
 344 /* Net device detach from fd. */
 345 static void tun_net_uninit(struct net_device *dev)
 346 {
 347         struct tun_struct *tun = netdev_priv(dev);
 348         struct tun_file *tfile = tun->tfile;
 349
 350         /* Inform the methods they need to stop using the dev.
 351          */
 352         if (tfile) {
 353                 wake_up_all(&tun->wq.wait);
 354                 if (atomic_dec_and_test(&tfile->count))
 355                         __tun_detach(tun);
 356         }
 357 }
 358
 359 static void tun_free_netdev(struct net_device *dev)
 360 {
 361         struct tun_struct *tun = netdev_priv(dev);
 362
 363         sock_put(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         skb_orphan(skb);
 420
 421         nf_reset(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 u32 tun_net_fix_features(struct net_device *dev, u32 features)
 461 {
 462         struct tun_struct *tun = netdev_priv(dev);
 463
 464         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 465 }
 466 #ifdef CONFIG_NET_POLL_CONTROLLER
 467 static void tun_poll_controller(struct net_device *dev)
 468 {
 469         /*
 470          * Tun only receives frames when:
 471          * 1) the char device endpoint gets data from user space
 472          * 2) the tun socket gets a sendmsg call from user space
 473          * Since both of those are syncronous operations, we are guaranteed
 474          * never to have pending data when we poll for it
 475          * so theres nothing to do here but return.
 476          * We need this though so netpoll recognizes us as an interface that
 477          * supports polling, which enables bridge devices in virt setups to
 478          * still use netconsole
 479          */
 480         return;
 481 }
 482 #endif
 483 static const struct net_device_ops tun_netdev_ops = {
 484         .ndo_uninit             = tun_net_uninit,
 485         .ndo_open               = tun_net_open,
 486         .ndo_stop               = tun_net_close,
 487         .ndo_start_xmit         = tun_net_xmit,
 488         .ndo_change_mtu         = tun_net_change_mtu,
 489         .ndo_fix_features       = tun_net_fix_features,
 490 #ifdef CONFIG_NET_POLL_CONTROLLER
 491         .ndo_poll_controller    = tun_poll_controller,
 492 #endif
 493 };
 494
 495 static const struct net_device_ops tap_netdev_ops = {
 496         .ndo_uninit             = tun_net_uninit,
 497         .ndo_open               = tun_net_open,
 498         .ndo_stop               = tun_net_close,
 499         .ndo_start_xmit         = tun_net_xmit,
 500         .ndo_change_mtu         = tun_net_change_mtu,
 501         .ndo_fix_features       = tun_net_fix_features,
 502         .ndo_set_rx_mode        = tun_net_mclist,
 503         .ndo_set_mac_address    = eth_mac_addr,
 504         .ndo_validate_addr      = eth_validate_addr,
 505 #ifdef CONFIG_NET_POLL_CONTROLLER
 506         .ndo_poll_controller    = tun_poll_controller,
 507 #endif
 508 };
 509
 510 /* Initialize net device. */
 511 static void tun_net_init(struct net_device *dev)
 512 {
 513         struct tun_struct *tun = netdev_priv(dev);
 514
 515         switch (tun->flags & TUN_TYPE_MASK) {
 516         case TUN_TUN_DEV:
 517                 dev->netdev_ops = &tun_netdev_ops;
 518
 519                 /* Point-to-Point TUN Device */
 520                 dev->hard_header_len = 0;
 521                 dev->addr_len = 0;
 522                 dev->mtu = 1500;
 523
 524                 /* Zero header length */
 525                 dev->type = ARPHRD_NONE;
 526                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 527                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 528                 break;
 529
 530         case TUN_TAP_DEV:
 531                 dev->netdev_ops = &tap_netdev_ops;
 532                 /* Ethernet TAP Device */
 533                 ether_setup(dev);
 534                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 535
 536                 random_ether_addr(dev->dev_addr);
 537
 538                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 539                 break;
 540         }
 541 }
 542
 543 /* Character device part */
 544
 545 /* Poll */
 546 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
 547 {
 548         struct tun_file *tfile = file->private_data;
 549         struct tun_struct *tun = __tun_get(tfile);
 550         struct sock *sk;
 551         unsigned int mask = 0;
 552
 553         if (!tun)
 554                 return POLLERR;
 555
 556         sk = tun->socket.sk;
 557
 558         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 559
 560         poll_wait(file, &tun->wq.wait, wait);
 561
 562         if (!skb_queue_empty(&sk->sk_receive_queue))
 563                 mask |= POLLIN | POLLRDNORM;
 564
 565         if (sock_writeable(sk) ||
 566             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 567              sock_writeable(sk)))
 568                 mask |= POLLOUT | POLLWRNORM;
 569
 570         if (tun->dev->reg_state != NETREG_REGISTERED)
 571                 mask = POLLERR;
 572
 573         tun_put(tun);
 574         return mask;
 575 }
 576
 577 /* prepad is the amount to reserve at front.  len is length after that.
 578  * linear is a hint as to how much to copy (usually headers). */
 579 static struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
 580                                      size_t prepad, size_t len,
 581                                      size_t linear, int noblock)
 582 {
 583         struct sock *sk = tun->socket.sk;
 584         struct sk_buff *skb;
 585         int err;
 586
 587         sock_update_classid(sk);
 588
 589         /* Under a page?  Don't bother with paged skb. */
 590         if (prepad + len < PAGE_SIZE || !linear)
 591                 linear = len;
 592
 593         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 594                                    &err);
 595         if (!skb)
 596                 return ERR_PTR(err);
 597
 598         skb_reserve(skb, prepad);
 599         skb_put(skb, linear);
 600         skb->data_len = len - linear;
 601         skb->len += len - linear;
 602
 603         return skb;
 604 }
 605
 606 /* Get packet from user space buffer */
 607 static ssize_t tun_get_user(struct tun_struct *tun,
 608                             const struct iovec *iv, size_t count,
 609                             int noblock)
 610 {
 611         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
 612         struct sk_buff *skb;
 613         size_t len = count, align = NET_SKB_PAD;
 614         struct virtio_net_hdr gso = { 0 };
 615         int offset = 0;
 616
 617         if (!(tun->flags & TUN_NO_PI)) {
 618                 if (len < sizeof(pi))
 619                         return -EINVAL;
 620                 len -= sizeof(pi);
 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)
 629                         return -EINVAL;
 630                 len -= tun->vnet_hdr_sz;
 631
 632                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
 633                         return -EFAULT;
 634
 635                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
 636                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
 637                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
 638
 639                 if (gso.hdr_len > len)
 640                         return -EINVAL;
 641                 offset += tun->vnet_hdr_sz;
 642         }
 643
 644         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
 645                 align += NET_IP_ALIGN;
 646                 if (unlikely(len < ETH_HLEN ||
 647                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
 648                         return -EINVAL;
 649         }
 650
 651         skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock);
 652         if (IS_ERR(skb)) {
 653                 if (PTR_ERR(skb) != -EAGAIN)
 654                         tun->dev->stats.rx_dropped++;
 655                 return PTR_ERR(skb);
 656         }
 657
 658         if (skb_copy_datagram_from_iovec(skb, 0, iv, offset, len)) {
 659                 tun->dev->stats.rx_dropped++;
 660                 kfree_skb(skb);
 661                 return -EFAULT;
 662         }
 663
 664         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 665                 if (!skb_partial_csum_set(skb, gso.csum_start,
 666                                           gso.csum_offset)) {
 667                         tun->dev->stats.rx_frame_errors++;
 668                         kfree_skb(skb);
 669                         return -EINVAL;
 670                 }
 671         }
 672
 673         switch (tun->flags & TUN_TYPE_MASK) {
 674         case TUN_TUN_DEV:
 675                 if (tun->flags & TUN_NO_PI) {
 676                         switch (skb->data[0] & 0xf0) {
 677                         case 0x40:
 678                                 pi.proto = htons(ETH_P_IP);
 679                                 break;
 680                         case 0x60:
 681                                 pi.proto = htons(ETH_P_IPV6);
 682                                 break;
 683                         default:
 684                                 tun->dev->stats.rx_dropped++;
 685                                 kfree_skb(skb);
 686                                 return -EINVAL;
 687                         }
 688                 }
 689
 690                 skb_reset_mac_header(skb);
 691                 skb->protocol = pi.proto;
 692                 skb->dev = tun->dev;
 693                 break;
 694         case TUN_TAP_DEV:
 695                 skb->protocol = eth_type_trans(skb, tun->dev);
 696                 break;
 697         }
 698
 699         skb_reset_network_header(skb);
 700
 701         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 702                 pr_debug("GSO!\n");
 703                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 704                 case VIRTIO_NET_HDR_GSO_TCPV4:
 705                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 706                         break;
 707                 case VIRTIO_NET_HDR_GSO_TCPV6:
 708                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 709                         break;
 710                 case VIRTIO_NET_HDR_GSO_UDP:
 711                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 712                         if (skb->protocol == htons(ETH_P_IPV6))
 713                                 ipv6_proxy_select_ident(skb);
 714                         break;
 715                 default:
 716                         tun->dev->stats.rx_frame_errors++;
 717                         kfree_skb(skb);
 718                         return -EINVAL;
 719                 }
 720
 721                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
 722                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 723
 724                 skb_shinfo(skb)->gso_size = gso.gso_size;
 725                 if (skb_shinfo(skb)->gso_size == 0) {
 726                         tun->dev->stats.rx_frame_errors++;
 727                         kfree_skb(skb);
 728                         return -EINVAL;
 729                 }
 730
 731                 /* Header must be checked, and gso_segs computed. */
 732                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 733                 skb_shinfo(skb)->gso_segs = 0;
 734         }
 735
 736         netif_rx_ni(skb);
 737
 738         tun->dev->stats.rx_packets++;
 739         tun->dev->stats.rx_bytes += len;
 740
 741         return count;
 742 }
 743
 744 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
 745                               unsigned long count, loff_t pos)
 746 {
 747         struct file *file = iocb->ki_filp;
 748         struct tun_struct *tun = tun_get(file);
 749         ssize_t result;
 750
 751         if (!tun)
 752                 return -EBADFD;
 753
 754         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
 755
 756         result = tun_get_user(tun, iv, iov_length(iv, count),
 757                               file->f_flags & O_NONBLOCK);
 758
 759         tun_put(tun);
 760         return result;
 761 }
 762
 763 /* Put packet to the user space buffer */
 764 static ssize_t tun_put_user(struct tun_struct *tun,
 765                             struct sk_buff *skb,
 766                             const struct iovec *iv, int len)
 767 {
 768         struct tun_pi pi = { 0, skb->protocol };
 769         ssize_t total = 0;
 770         int vnet_hdr_sz = 0;
 771
 772         if (tun->flags & TUN_VNET_HDR)
 773                 vnet_hdr_sz = tun->vnet_hdr_sz;
 774
 775         if (!(tun->flags & TUN_NO_PI)) {
 776                 if ((len -= sizeof(pi)) < 0)
 777                         return -EINVAL;
 778
 779                 if (len < skb->len + vnet_hdr_sz) {
 780                         /* Packet will be striped */
 781                         pi.flags |= TUN_PKT_STRIP;
 782                 }
 783
 784                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
 785                         return -EFAULT;
 786                 total += sizeof(pi);
 787         }
 788
 789         if (vnet_hdr_sz) {
 790                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
 791                 if ((len -= vnet_hdr_sz) < 0)
 792                         return -EINVAL;
 793
 794                 if (skb_is_gso(skb)) {
 795                         struct skb_shared_info *sinfo = skb_shinfo(skb);
 796
 797                         /* This is a hint as to how much should be linear. */
 798                         gso.hdr_len = skb_headlen(skb);
 799                         gso.gso_size = sinfo->gso_size;
 800                         if (sinfo->gso_type & SKB_GSO_TCPV4)
 801                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 802                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
 803                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 804                         else if (sinfo->gso_type & SKB_GSO_UDP)
 805                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
 806                         else {
 807                                 pr_err("unexpected GSO type: "
 808                                        "0x%x, gso_size %d, hdr_len %d\n",
 809                                        sinfo->gso_type, gso.gso_size,
 810                                        gso.hdr_len);
 811                                 print_hex_dump(KERN_ERR, "tun: ",
 812                                                DUMP_PREFIX_NONE,
 813                                                16, 1, skb->head,
 814                                                min((int)gso.hdr_len, 64), true);
 815                                 WARN_ON_ONCE(1);
 816                                 return -EINVAL;
 817                         }
 818                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
 819                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 820                 } else
 821                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
 822
 823                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 824                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 825                         gso.csum_start = skb_checksum_start_offset(skb);
 826                         gso.csum_offset = skb->csum_offset;
 827                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
 828                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
 829                 } /* else everything is zero */
 830
 831                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
 832                                                sizeof(gso))))
 833                         return -EFAULT;
 834                 total += vnet_hdr_sz;
 835         }
 836
 837         len = min_t(int, skb->len, len);
 838
 839         skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
 840         total += skb->len;
 841
 842         tun->dev->stats.tx_packets++;
 843         tun->dev->stats.tx_bytes += len;
 844
 845         return total;
 846 }
 847
 848 static ssize_t tun_do_read(struct tun_struct *tun,
 849                            struct kiocb *iocb, const struct iovec *iv,
 850                            ssize_t len, int noblock)
 851 {
 852         DECLARE_WAITQUEUE(wait, current);
 853         struct sk_buff *skb;
 854         ssize_t ret = 0;
 855
 856         tun_debug(KERN_INFO, tun, "tun_chr_read\n");
 857
 858         if (unlikely(!noblock))
 859                 add_wait_queue(&tun->wq.wait, &wait);
 860         while (len) {
 861                 current->state = TASK_INTERRUPTIBLE;
 862
 863                 /* Read frames from the queue */
 864                 if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
 865                         if (noblock) {
 866                                 ret = -EAGAIN;
 867                                 break;
 868                         }
 869                         if (signal_pending(current)) {
 870                                 ret = -ERESTARTSYS;
 871                                 break;
 872                         }
 873                         if (tun->dev->reg_state != NETREG_REGISTERED) {
 874                                 ret = -EIO;
 875                                 break;
 876                         }
 877
 878                         /* Nothing to read, let's sleep */
 879                         schedule();
 880                         continue;
 881                 }
 882                 netif_wake_queue(tun->dev);
 883
 884                 ret = tun_put_user(tun, skb, iv, len);
 885                 kfree_skb(skb);
 886                 break;
 887         }
 888
 889         current->state = TASK_RUNNING;
 890         if (unlikely(!noblock))
 891                 remove_wait_queue(&tun->wq.wait, &wait);
 892
 893         return ret;
 894 }
 895
 896 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
 897                             unsigned long count, loff_t pos)
 898 {
 899         struct file *file = iocb->ki_filp;
 900         struct tun_file *tfile = file->private_data;
 901         struct tun_struct *tun = __tun_get(tfile);
 902         ssize_t len, ret;
 903
 904         if (!tun)
 905                 return -EBADFD;
 906         len = iov_length(iv, count);
 907         if (len < 0) {
 908                 ret = -EINVAL;
 909                 goto out;
 910         }
 911
 912         ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK);
 913         ret = min_t(ssize_t, ret, len);
 914 out:
 915         tun_put(tun);
 916         return ret;
 917 }
 918
 919 static void tun_setup(struct net_device *dev)
 920 {
 921         struct tun_struct *tun = netdev_priv(dev);
 922
 923         tun->owner = -1;
 924         tun->group = -1;
 925
 926         dev->ethtool_ops = &tun_ethtool_ops;
 927         dev->destructor = tun_free_netdev;
 928 }
 929
 930 /* Trivial set of netlink ops to allow deleting tun or tap
 931  * device with netlink.
 932  */
 933 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
 934 {
 935         return -EINVAL;
 936 }
 937
 938 static struct rtnl_link_ops tun_link_ops __read_mostly = {
 939         .kind           = DRV_NAME,
 940         .priv_size      = sizeof(struct tun_struct),
 941         .setup          = tun_setup,
 942         .validate       = tun_validate,
 943 };
 944
 945 static void tun_sock_write_space(struct sock *sk)
 946 {
 947         struct tun_struct *tun;
 948         wait_queue_head_t *wqueue;
 949
 950         if (!sock_writeable(sk))
 951                 return;
 952
 953         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
 954                 return;
 955
 956         wqueue = sk_sleep(sk);
 957         if (wqueue && waitqueue_active(wqueue))
 958                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
 959                                                 POLLWRNORM | POLLWRBAND);
 960
 961         tun = tun_sk(sk)->tun;
 962         kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
 963 }
 964
 965 static void tun_sock_destruct(struct sock *sk)
 966 {
 967         free_netdev(tun_sk(sk)->tun->dev);
 968 }
 969
 970 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
 971                        struct msghdr *m, size_t total_len)
 972 {
 973         struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
 974         return tun_get_user(tun, m->msg_iov, total_len,
 975                             m->msg_flags & MSG_DONTWAIT);
 976 }
 977
 978 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
 979                        struct msghdr *m, size_t total_len,
 980                        int flags)
 981 {
 982         struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
 983         int ret;
 984         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
 985                 return -EINVAL;
 986         ret = tun_do_read(tun, iocb, m->msg_iov, total_len,
 987                           flags & MSG_DONTWAIT);
 988         if (ret > total_len) {
 989                 m->msg_flags |= MSG_TRUNC;
 990                 ret = flags & MSG_TRUNC ? ret : total_len;
 991         }
 992         return ret;
 993 }
 994
 995 /* Ops structure to mimic raw sockets with tun */
 996 static const struct proto_ops tun_socket_ops = {
 997         .sendmsg = tun_sendmsg,
 998         .recvmsg = tun_recvmsg,
 999 };
1000
1001 static struct proto tun_proto = {
1002         .name           = "tun",
1003         .owner          = THIS_MODULE,
1004         .obj_size       = sizeof(struct tun_sock),
1005 };
1006
1007 static int tun_flags(struct tun_struct *tun)
1008 {
1009         int flags = 0;
1010
1011         if (tun->flags & TUN_TUN_DEV)
1012                 flags |= IFF_TUN;
1013         else
1014                 flags |= IFF_TAP;
1015
1016         if (tun->flags & TUN_NO_PI)
1017                 flags |= IFF_NO_PI;
1018
1019         if (tun->flags & TUN_ONE_QUEUE)
1020                 flags |= IFF_ONE_QUEUE;
1021
1022         if (tun->flags & TUN_VNET_HDR)
1023                 flags |= IFF_VNET_HDR;
1024
1025         return flags;
1026 }
1027
1028 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1029                               char *buf)
1030 {
1031         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1032         return sprintf(buf, "0x%x\n", tun_flags(tun));
1033 }
1034
1035 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1036                               char *buf)
1037 {
1038         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1039         return sprintf(buf, "%d\n", tun->owner);
1040 }
1041
1042 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1043                               char *buf)
1044 {
1045         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1046         return sprintf(buf, "%d\n", tun->group);
1047 }
1048
1049 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1050 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1051 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1052
1053 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1054 {
1055         struct sock *sk;
1056         struct tun_struct *tun;
1057         struct net_device *dev;
1058         int err;
1059
1060         dev = __dev_get_by_name(net, ifr->ifr_name);
1061         if (dev) {
1062                 const struct cred *cred = current_cred();
1063
1064                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1065                         return -EBUSY;
1066                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1067                         tun = netdev_priv(dev);
1068                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1069                         tun = netdev_priv(dev);
1070                 else
1071                         return -EINVAL;
1072
1073                 if (((tun->owner != -1 && cred->euid != tun->owner) ||
1074                      (tun->group != -1 && !in_egroup_p(tun->group))) &&
1075                     !capable(CAP_NET_ADMIN))
1076                         return -EPERM;
1077                 err = security_tun_dev_attach(tun->socket.sk);
1078                 if (err < 0)
1079                         return err;
1080
1081                 err = tun_attach(tun, file);
1082                 if (err < 0)
1083                         return err;
1084         }
1085         else {
1086                 char *name;
1087                 unsigned long flags = 0;
1088
1089                 if (!capable(CAP_NET_ADMIN))
1090                         return -EPERM;
1091                 err = security_tun_dev_create();
1092                 if (err < 0)
1093                         return err;
1094
1095                 /* Set dev type */
1096                 if (ifr->ifr_flags & IFF_TUN) {
1097                         /* TUN device */
1098                         flags |= TUN_TUN_DEV;
1099                         name = "tun%d";
1100                 } else if (ifr->ifr_flags & IFF_TAP) {
1101                         /* TAP device */
1102                         flags |= TUN_TAP_DEV;
1103                         name = "tap%d";
1104                 } else
1105                         return -EINVAL;
1106
1107                 if (*ifr->ifr_name)
1108                         name = ifr->ifr_name;
1109
1110                 dev = alloc_netdev(sizeof(struct tun_struct), name,
1111                                    tun_setup);
1112                 if (!dev)
1113                         return -ENOMEM;
1114
1115                 dev_net_set(dev, net);
1116                 dev->rtnl_link_ops = &tun_link_ops;
1117
1118                 tun = netdev_priv(dev);
1119                 tun->dev = dev;
1120                 tun->flags = flags;
1121                 tun->txflt.count = 0;
1122                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1123
1124                 err = -ENOMEM;
1125                 sk = sk_alloc(net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1126                 if (!sk)
1127                         goto err_free_dev;
1128
1129                 tun->socket.wq = &tun->wq;
1130                 init_waitqueue_head(&tun->wq.wait);
1131                 tun->socket.ops = &tun_socket_ops;
1132                 sock_init_data(&tun->socket, sk);
1133                 sk->sk_write_space = tun_sock_write_space;
1134                 sk->sk_sndbuf = INT_MAX;
1135
1136                 tun_sk(sk)->tun = tun;
1137
1138                 security_tun_dev_post_create(sk);
1139
1140                 tun_net_init(dev);
1141
1142                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1143                         TUN_USER_FEATURES;
1144                 dev->features = dev->hw_features;
1145
1146                 err = register_netdevice(tun->dev);
1147                 if (err < 0)
1148                         goto err_free_sk;
1149
1150                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1151                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1152                     device_create_file(&tun->dev->dev, &dev_attr_group))
1153                         pr_err("Failed to create tun sysfs files\n");
1154
1155                 sk->sk_destruct = tun_sock_destruct;
1156
1157                 err = tun_attach(tun, file);
1158                 if (err < 0)
1159                         goto failed;
1160         }
1161
1162         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1163
1164         if (ifr->ifr_flags & IFF_NO_PI)
1165                 tun->flags |= TUN_NO_PI;
1166         else
1167                 tun->flags &= ~TUN_NO_PI;
1168
1169         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1170                 tun->flags |= TUN_ONE_QUEUE;
1171         else
1172                 tun->flags &= ~TUN_ONE_QUEUE;
1173
1174         if (ifr->ifr_flags & IFF_VNET_HDR)
1175                 tun->flags |= TUN_VNET_HDR;
1176         else
1177                 tun->flags &= ~TUN_VNET_HDR;
1178
1179         /* Make sure persistent devices do not get stuck in
1180          * xoff state.
1181          */
1182         if (netif_running(tun->dev))
1183                 netif_wake_queue(tun->dev);
1184
1185         strcpy(ifr->ifr_name, tun->dev->name);
1186         return 0;
1187
1188  err_free_sk:
1189         sock_put(sk);
1190  err_free_dev:
1191         free_netdev(dev);
1192  failed:
1193         return err;
1194 }
1195
1196 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1197                        struct ifreq *ifr)
1198 {
1199         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1200
1201         strcpy(ifr->ifr_name, tun->dev->name);
1202
1203         ifr->ifr_flags = tun_flags(tun);
1204
1205         return 0;
1206 }
1207
1208 /* This is like a cut-down ethtool ops, except done via tun fd so no
1209  * privs required. */
1210 static int set_offload(struct tun_struct *tun, unsigned long arg)
1211 {
1212         u32 features = 0;
1213
1214         if (arg & TUN_F_CSUM) {
1215                 features |= NETIF_F_HW_CSUM;
1216                 arg &= ~TUN_F_CSUM;
1217
1218                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1219                         if (arg & TUN_F_TSO_ECN) {
1220                                 features |= NETIF_F_TSO_ECN;
1221                                 arg &= ~TUN_F_TSO_ECN;
1222                         }
1223                         if (arg & TUN_F_TSO4)
1224                                 features |= NETIF_F_TSO;
1225                         if (arg & TUN_F_TSO6)
1226                                 features |= NETIF_F_TSO6;
1227                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1228                 }
1229
1230                 if (arg & TUN_F_UFO) {
1231                         features |= NETIF_F_UFO;
1232                         arg &= ~TUN_F_UFO;
1233                 }
1234         }
1235
1236         /* This gives the user a way to test for new features in future by
1237          * trying to set them. */
1238         if (arg)
1239                 return -EINVAL;
1240
1241         tun->set_features = features;
1242         netdev_update_features(tun->dev);
1243
1244         return 0;
1245 }
1246
1247 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1248                             unsigned long arg, int ifreq_len)
1249 {
1250         struct tun_file *tfile = file->private_data;
1251         struct tun_struct *tun;
1252         void __user* argp = (void __user*)arg;
1253         struct sock_fprog fprog;
1254         struct ifreq ifr;
1255         int sndbuf;
1256         int vnet_hdr_sz;
1257         int ret;
1258
1259         if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89) {
1260                 if (copy_from_user(&ifr, argp, ifreq_len))
1261                         return -EFAULT;
1262         } else {
1263                 memset(&ifr, 0, sizeof(ifr));
1264         }
1265         if (cmd == TUNGETFEATURES) {
1266                 /* Currently this just means: "what IFF flags are valid?".
1267                  * This is needed because we never checked for invalid flags on
1268                  * TUNSETIFF. */
1269                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1270                                 IFF_VNET_HDR,
1271                                 (unsigned int __user*)argp);
1272         }
1273
1274         rtnl_lock();
1275
1276         tun = __tun_get(tfile);
1277         if (cmd == TUNSETIFF && !tun) {
1278                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1279
1280                 ret = tun_set_iff(tfile->net, file, &ifr);
1281
1282                 if (ret)
1283                         goto unlock;
1284
1285                 if (copy_to_user(argp, &ifr, ifreq_len))
1286                         ret = -EFAULT;
1287                 goto unlock;
1288         }
1289
1290         ret = -EBADFD;
1291         if (!tun)
1292                 goto unlock;
1293
1294         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %d\n", cmd);
1295
1296         ret = 0;
1297         switch (cmd) {
1298         case TUNGETIFF:
1299                 ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1300                 if (ret)
1301                         break;
1302
1303                 if (copy_to_user(argp, &ifr, ifreq_len))
1304                         ret = -EFAULT;
1305                 break;
1306
1307         case TUNSETNOCSUM:
1308                 /* Disable/Enable checksum */
1309
1310                 /* [unimplemented] */
1311                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1312                           arg ? "disabled" : "enabled");
1313                 break;
1314
1315         case TUNSETPERSIST:
1316                 /* Disable/Enable persist mode */
1317                 if (arg)
1318                         tun->flags |= TUN_PERSIST;
1319                 else
1320                         tun->flags &= ~TUN_PERSIST;
1321
1322                 tun_debug(KERN_INFO, tun, "persist %s\n",
1323                           arg ? "enabled" : "disabled");
1324                 break;
1325
1326         case TUNSETOWNER:
1327                 /* Set owner of the device */
1328                 tun->owner = (uid_t) arg;
1329
1330                 tun_debug(KERN_INFO, tun, "owner set to %d\n", tun->owner);
1331                 break;
1332
1333         case TUNSETGROUP:
1334                 /* Set group of the device */
1335                 tun->group= (gid_t) arg;
1336
1337                 tun_debug(KERN_INFO, tun, "group set to %d\n", tun->group);
1338                 break;
1339
1340         case TUNSETLINK:
1341                 /* Only allow setting the type when the interface is down */
1342                 if (tun->dev->flags & IFF_UP) {
1343                         tun_debug(KERN_INFO, tun,
1344                                   "Linktype set failed because interface is up\n");
1345                         ret = -EBUSY;
1346                 } else {
1347                         tun->dev->type = (int) arg;
1348                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1349                                   tun->dev->type);
1350                         ret = 0;
1351                 }
1352                 break;
1353
1354 #ifdef TUN_DEBUG
1355         case TUNSETDEBUG:
1356                 tun->debug = arg;
1357                 break;
1358 #endif
1359         case TUNSETOFFLOAD:
1360                 ret = set_offload(tun, arg);
1361                 break;
1362
1363         case TUNSETTXFILTER:
1364                 /* Can be set only for TAPs */
1365                 ret = -EINVAL;
1366                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1367                         break;
1368                 ret = update_filter(&tun->txflt, (void __user *)arg);
1369                 break;
1370
1371         case SIOCGIFHWADDR:
1372                 /* Get hw address */
1373                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1374                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1375                 if (copy_to_user(argp, &ifr, ifreq_len))
1376                         ret = -EFAULT;
1377                 break;
1378
1379         case SIOCSIFHWADDR:
1380                 /* Set hw address */
1381                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1382                           ifr.ifr_hwaddr.sa_data);
1383
1384                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1385                 break;
1386
1387         case TUNGETSNDBUF:
1388                 sndbuf = tun->socket.sk->sk_sndbuf;
1389                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1390                         ret = -EFAULT;
1391                 break;
1392
1393         case TUNSETSNDBUF:
1394                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1395                         ret = -EFAULT;
1396                         break;
1397                 }
1398
1399                 tun->socket.sk->sk_sndbuf = sndbuf;
1400                 break;
1401
1402         case TUNGETVNETHDRSZ:
1403                 vnet_hdr_sz = tun->vnet_hdr_sz;
1404                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1405                         ret = -EFAULT;
1406                 break;
1407
1408         case TUNSETVNETHDRSZ:
1409                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1410                         ret = -EFAULT;
1411                         break;
1412                 }
1413                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1414                         ret = -EINVAL;
1415                         break;
1416                 }
1417
1418                 tun->vnet_hdr_sz = vnet_hdr_sz;
1419                 break;
1420
1421         case TUNATTACHFILTER:
1422                 /* Can be set only for TAPs */
1423                 ret = -EINVAL;
1424                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1425                         break;
1426                 ret = -EFAULT;
1427                 if (copy_from_user(&fprog, argp, sizeof(fprog)))
1428                         break;
1429
1430                 ret = sk_attach_filter(&fprog, tun->socket.sk);
1431                 break;
1432
1433         case TUNDETACHFILTER:
1434                 /* Can be set only for TAPs */
1435                 ret = -EINVAL;
1436                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1437                         break;
1438                 ret = sk_detach_filter(tun->socket.sk);
1439                 break;
1440
1441         default:
1442                 ret = -EINVAL;
1443                 break;
1444         }
1445
1446 unlock:
1447         rtnl_unlock();
1448         if (tun)
1449                 tun_put(tun);
1450         return ret;
1451 }
1452
1453 static long tun_chr_ioctl(struct file *file,
1454                           unsigned int cmd, unsigned long arg)
1455 {
1456         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1457 }
1458
1459 #ifdef CONFIG_COMPAT
1460 static long tun_chr_compat_ioctl(struct file *file,
1461                          unsigned int cmd, unsigned long arg)
1462 {
1463         switch (cmd) {
1464         case TUNSETIFF:
1465         case TUNGETIFF:
1466         case TUNSETTXFILTER:
1467         case TUNGETSNDBUF:
1468         case TUNSETSNDBUF:
1469         case SIOCGIFHWADDR:
1470         case SIOCSIFHWADDR:
1471                 arg = (unsigned long)compat_ptr(arg);
1472                 break;
1473         default:
1474                 arg = (compat_ulong_t)arg;
1475                 break;
1476         }
1477
1478         /*
1479          * compat_ifreq is shorter than ifreq, so we must not access beyond
1480          * the end of that structure. All fields that are used in this
1481          * driver are compatible though, we don't need to convert the
1482          * contents.
1483          */
1484         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1485 }
1486 #endif /* CONFIG_COMPAT */
1487
1488 static int tun_chr_fasync(int fd, struct file *file, int on)
1489 {
1490         struct tun_struct *tun = tun_get(file);
1491         int ret;
1492
1493         if (!tun)
1494                 return -EBADFD;
1495
1496         tun_debug(KERN_INFO, tun, "tun_chr_fasync %d\n", on);
1497
1498         if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1499                 goto out;
1500
1501         if (on) {
1502                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1503                 if (ret)
1504                         goto out;
1505                 tun->flags |= TUN_FASYNC;
1506         } else
1507                 tun->flags &= ~TUN_FASYNC;
1508         ret = 0;
1509 out:
1510         tun_put(tun);
1511         return ret;
1512 }
1513
1514 static int tun_chr_open(struct inode *inode, struct file * file)
1515 {
1516         struct tun_file *tfile;
1517
1518         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
1519
1520         tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1521         if (!tfile)
1522                 return -ENOMEM;
1523         atomic_set(&tfile->count, 0);
1524         tfile->tun = NULL;
1525         tfile->net = get_net(current->nsproxy->net_ns);
1526         file->private_data = tfile;
1527         return 0;
1528 }
1529
1530 static int tun_chr_close(struct inode *inode, struct file *file)
1531 {
1532         struct tun_file *tfile = file->private_data;
1533         struct tun_struct *tun;
1534
1535         tun = __tun_get(tfile);
1536         if (tun) {
1537                 struct net_device *dev = tun->dev;
1538
1539                 tun_debug(KERN_INFO, tun, "tun_chr_close\n");
1540
1541                 __tun_detach(tun);
1542
1543                 /* If desirable, unregister the netdevice. */
1544                 if (!(tun->flags & TUN_PERSIST)) {
1545                         rtnl_lock();
1546                         if (dev->reg_state == NETREG_REGISTERED)
1547                                 unregister_netdevice(dev);
1548                         rtnl_unlock();
1549                 }
1550         }
1551
1552         tun = tfile->tun;
1553         if (tun)
1554                 sock_put(tun->socket.sk);
1555
1556         put_net(tfile->net);
1557         kfree(tfile);
1558
1559         return 0;
1560 }
1561
1562 static const struct file_operations tun_fops = {
1563         .owner  = THIS_MODULE,
1564         .llseek = no_llseek,
1565         .read  = do_sync_read,
1566         .aio_read  = tun_chr_aio_read,
1567         .write = do_sync_write,
1568         .aio_write = tun_chr_aio_write,
1569         .poll   = tun_chr_poll,
1570         .unlocked_ioctl = tun_chr_ioctl,
1571 #ifdef CONFIG_COMPAT
1572         .compat_ioctl = tun_chr_compat_ioctl,
1573 #endif
1574         .open   = tun_chr_open,
1575         .release = tun_chr_close,
1576         .fasync = tun_chr_fasync
1577 };
1578
1579 static struct miscdevice tun_miscdev = {
1580         .minor = TUN_MINOR,
1581         .name = "tun",
1582         .nodename = "net/tun",
1583         .fops = &tun_fops,
1584 };
1585
1586 /* ethtool interface */
1587
1588 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1589 {
1590         cmd->supported          = 0;
1591         cmd->advertising        = 0;
1592         ethtool_cmd_speed_set(cmd, SPEED_10);
1593         cmd->duplex             = DUPLEX_FULL;
1594         cmd->port               = PORT_TP;
1595         cmd->phy_address        = 0;
1596         cmd->transceiver        = XCVR_INTERNAL;
1597         cmd->autoneg            = AUTONEG_DISABLE;
1598         cmd->maxtxpkt           = 0;
1599         cmd->maxrxpkt           = 0;
1600         return 0;
1601 }
1602
1603 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1604 {
1605         struct tun_struct *tun = netdev_priv(dev);
1606
1607         strcpy(info->driver, DRV_NAME);
1608         strcpy(info->version, DRV_VERSION);
1609         strcpy(info->fw_version, "N/A");
1610
1611         switch (tun->flags & TUN_TYPE_MASK) {
1612         case TUN_TUN_DEV:
1613                 strcpy(info->bus_info, "tun");
1614                 break;
1615         case TUN_TAP_DEV:
1616                 strcpy(info->bus_info, "tap");
1617                 break;
1618         }
1619 }
1620
1621 static u32 tun_get_msglevel(struct net_device *dev)
1622 {
1623 #ifdef TUN_DEBUG
1624         struct tun_struct *tun = netdev_priv(dev);
1625         return tun->debug;
1626 #else
1627         return -EOPNOTSUPP;
1628 #endif
1629 }
1630
1631 static void tun_set_msglevel(struct net_device *dev, u32 value)
1632 {
1633 #ifdef TUN_DEBUG
1634         struct tun_struct *tun = netdev_priv(dev);
1635         tun->debug = value;
1636 #endif
1637 }
1638
1639 static const struct ethtool_ops tun_ethtool_ops = {
1640         .get_settings   = tun_get_settings,
1641         .get_drvinfo    = tun_get_drvinfo,
1642         .get_msglevel   = tun_get_msglevel,
1643         .set_msglevel   = tun_set_msglevel,
1644         .get_link       = ethtool_op_get_link,
1645 };
1646
1647
1648 static int __init tun_init(void)
1649 {
1650         int ret = 0;
1651
1652         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1653         pr_info("%s\n", DRV_COPYRIGHT);
1654
1655         ret = rtnl_link_register(&tun_link_ops);
1656         if (ret) {
1657                 pr_err("Can't register link_ops\n");
1658                 goto err_linkops;
1659         }
1660
1661         ret = misc_register(&tun_miscdev);
1662         if (ret) {
1663                 pr_err("Can't register misc device %d\n", TUN_MINOR);
1664                 goto err_misc;
1665         }
1666         return  0;
1667 err_misc:
1668         rtnl_link_unregister(&tun_link_ops);
1669 err_linkops:
1670         return ret;
1671 }
1672
1673 static void tun_cleanup(void)
1674 {
1675         misc_deregister(&tun_miscdev);
1676         rtnl_link_unregister(&tun_link_ops);
1677 }
1678
1679 /* Get an underlying socket object from tun file.  Returns error unless file is
1680  * attached to a device.  The returned object works like a packet socket, it
1681  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1682  * holding a reference to the file for as long as the socket is in use. */
1683 struct socket *tun_get_socket(struct file *file)
1684 {
1685         struct tun_struct *tun;
1686         if (file->f_op != &tun_fops)
1687                 return ERR_PTR(-EINVAL);
1688         tun = tun_get(file);
1689         if (!tun)
1690                 return ERR_PTR(-EBADFD);
1691         tun_put(tun);
1692         return &tun->socket;
1693 }
1694 EXPORT_SYMBOL_GPL(tun_get_socket);
1695
1696 module_init(tun_init);
1697 module_exit(tun_cleanup);
1698 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1699 MODULE_AUTHOR(DRV_COPYRIGHT);
1700 MODULE_LICENSE("GPL");
1701 MODULE_ALIAS_MISCDEV(TUN_MINOR);
1702 MODULE_ALIAS("devname:net/tun");