drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use random_ether_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/major.h>
  48 #include <linux/slab.h>
  49 #include <linux/poll.h>
  50 #include <linux/fcntl.h>
  51 #include <linux/init.h>
  52 #include <linux/skbuff.h>
  53 #include <linux/netdevice.h>
  54 #include <linux/etherdevice.h>
  55 #include <linux/miscdevice.h>
  56 #include <linux/ethtool.h>
  57 #include <linux/rtnetlink.h>
  58 #include <linux/compat.h>
  59 #include <linux/if.h>
  60 #include <linux/if_arp.h>
  61 #include <linux/if_ether.h>
  62 #include <linux/if_tun.h>
  63 #include <linux/crc32.h>
  64 #include <linux/nsproxy.h>
  65 #include <linux/virtio_net.h>
  66 #include <linux/rcupdate.h>
  67 #include <net/net_namespace.h>
  68 #include <net/netns/generic.h>
  69 #include <net/rtnetlink.h>
  70 #include <net/sock.h>
  71
  72 #include <asm/system.h>
  73 #include <asm/uaccess.h>
  74
  75 /* Uncomment to enable debugging */
  76 /* #define TUN_DEBUG 1 */
  77
  78 #ifdef TUN_DEBUG
  79 static int debug;
  80
  81 #define tun_debug(level, tun, fmt, args...)                     \
  82 do {                                                            \
  83         if (tun->debug)                                         \
  84                 netdev_printk(level, tun->dev, fmt, ##args);    \
  85 } while (0)
  86 #define DBG1(level, fmt, args...)                               \
  87 do {                                                            \
  88         if (debug == 2)                                         \
  89                 printk(level fmt, ##args);                      \
  90 } while (0)
  91 #else
  92 #define tun_debug(level, tun, fmt, args...)                     \
  93 do {                                                            \
  94         if (0)                                                  \
  95                 netdev_printk(level, tun->dev, fmt, ##args);    \
  96 } while (0)
  97 #define DBG1(level, fmt, args...)                               \
  98 do {                                                            \
  99         if (0)                                                  \
 100                 printk(level fmt, ##args);                      \
 101 } while (0)
 102 #endif
 103
 104 #define FLT_EXACT_COUNT 8
 105 struct tap_filter {
 106         unsigned int    count;    /* Number of addrs. Zero means disabled */
 107         u32             mask[2];  /* Mask of the hashed addrs */
 108         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 109 };
 110
 111 struct tun_file {
 112         atomic_t count;
 113         struct tun_struct *tun;
 114         struct net *net;
 115 };
 116
 117 struct tun_sock;
 118
 119 struct tun_struct {
 120         struct tun_file         *tfile;
 121         unsigned int            flags;
 122         uid_t                   owner;
 123         gid_t                   group;
 124
 125         struct net_device       *dev;
 126         netdev_features_t       set_features;
 127 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 128                           NETIF_F_TSO6|NETIF_F_UFO)
 129         struct fasync_struct    *fasync;
 130
 131         struct tap_filter       txflt;
 132         struct socket           socket;
 133         struct socket_wq        wq;
 134
 135         int                     vnet_hdr_sz;
 136
 137 #ifdef TUN_DEBUG
 138         int debug;
 139 #endif
 140 };
 141
 142 struct tun_sock {
 143         struct sock             sk;
 144         struct tun_struct       *tun;
 145 };
 146
 147 static inline struct tun_sock *tun_sk(struct sock *sk)
 148 {
 149         return container_of(sk, struct tun_sock, sk);
 150 }
 151
 152 static int tun_attach(struct tun_struct *tun, struct file *file)
 153 {
 154         struct tun_file *tfile = file->private_data;
 155         int err;
 156
 157         ASSERT_RTNL();
 158
 159         netif_tx_lock_bh(tun->dev);
 160
 161         err = -EINVAL;
 162         if (tfile->tun)
 163                 goto out;
 164
 165         err = -EBUSY;
 166         if (tun->tfile)
 167                 goto out;
 168
 169         err = 0;
 170         tfile->tun = tun;
 171         tun->tfile = tfile;
 172         tun->socket.file = file;
 173         netif_carrier_on(tun->dev);
 174         dev_hold(tun->dev);
 175         sock_hold(tun->socket.sk);
 176         atomic_inc(&tfile->count);
 177
 178 out:
 179         netif_tx_unlock_bh(tun->dev);
 180         return err;
 181 }
 182
 183 static void __tun_detach(struct tun_struct *tun)
 184 {
 185         /* Detach from net device */
 186         netif_tx_lock_bh(tun->dev);
 187         netif_carrier_off(tun->dev);
 188         tun->tfile = NULL;
 189         tun->socket.file = NULL;
 190         netif_tx_unlock_bh(tun->dev);
 191
 192         /* Drop read queue */
 193         skb_queue_purge(&tun->socket.sk->sk_receive_queue);
 194
 195         /* Drop the extra count on the net device */
 196         dev_put(tun->dev);
 197 }
 198
 199 static void tun_detach(struct tun_struct *tun)
 200 {
 201         rtnl_lock();
 202         __tun_detach(tun);
 203         rtnl_unlock();
 204 }
 205
 206 static struct tun_struct *__tun_get(struct tun_file *tfile)
 207 {
 208         struct tun_struct *tun = NULL;
 209
 210         if (atomic_inc_not_zero(&tfile->count))
 211                 tun = tfile->tun;
 212
 213         return tun;
 214 }
 215
 216 static struct tun_struct *tun_get(struct file *file)
 217 {
 218         return __tun_get(file->private_data);
 219 }
 220
 221 static void tun_put(struct tun_struct *tun)
 222 {
 223         struct tun_file *tfile = tun->tfile;
 224
 225         if (atomic_dec_and_test(&tfile->count))
 226                 tun_detach(tfile->tun);
 227 }
 228
 229 /* TAP filtering */
 230 static void addr_hash_set(u32 *mask, const u8 *addr)
 231 {
 232         int n = ether_crc(ETH_ALEN, addr) >> 26;
 233         mask[n >> 5] |= (1 << (n & 31));
 234 }
 235
 236 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 237 {
 238         int n = ether_crc(ETH_ALEN, addr) >> 26;
 239         return mask[n >> 5] & (1 << (n & 31));
 240 }
 241
 242 static int update_filter(struct tap_filter *filter, void __user *arg)
 243 {
 244         struct { u8 u[ETH_ALEN]; } *addr;
 245         struct tun_filter uf;
 246         int err, alen, n, nexact;
 247
 248         if (copy_from_user(&uf, arg, sizeof(uf)))
 249                 return -EFAULT;
 250
 251         if (!uf.count) {
 252                 /* Disabled */
 253                 filter->count = 0;
 254                 return 0;
 255         }
 256
 257         alen = ETH_ALEN * uf.count;
 258         addr = kmalloc(alen, GFP_KERNEL);
 259         if (!addr)
 260                 return -ENOMEM;
 261
 262         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 263                 err = -EFAULT;
 264                 goto done;
 265         }
 266
 267         /* The filter is updated without holding any locks. Which is
 268          * perfectly safe. We disable it first and in the worst
 269          * case we'll accept a few undesired packets. */
 270         filter->count = 0;
 271         wmb();
 272
 273         /* Use first set of addresses as an exact filter */
 274         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 275                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 276
 277         nexact = n;
 278
 279         /* Remaining multicast addresses are hashed,
 280          * unicast will leave the filter disabled. */
 281         memset(filter->mask, 0, sizeof(filter->mask));
 282         for (; n < uf.count; n++) {
 283                 if (!is_multicast_ether_addr(addr[n].u)) {
 284                         err = 0; /* no filter */
 285                         goto done;
 286                 }
 287                 addr_hash_set(filter->mask, addr[n].u);
 288         }
 289
 290         /* For ALLMULTI just set the mask to all ones.
 291          * This overrides the mask populated above. */
 292         if ((uf.flags & TUN_FLT_ALLMULTI))
 293                 memset(filter->mask, ~0, sizeof(filter->mask));
 294
 295         /* Now enable the filter */
 296         wmb();
 297         filter->count = nexact;
 298
 299         /* Return the number of exact filters */
 300         err = nexact;
 301
 302 done:
 303         kfree(addr);
 304         return err;
 305 }
 306
 307 /* Returns: 0 - drop, !=0 - accept */
 308 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 309 {
 310         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 311          * at this point. */
 312         struct ethhdr *eh = (struct ethhdr *) skb->data;
 313         int i;
 314
 315         /* Exact match */
 316         for (i = 0; i < filter->count; i++)
 317                 if (!compare_ether_addr(eh->h_dest, filter->addr[i]))
 318                         return 1;
 319
 320         /* Inexact match (multicast only) */
 321         if (is_multicast_ether_addr(eh->h_dest))
 322                 return addr_hash_test(filter->mask, eh->h_dest);
 323
 324         return 0;
 325 }
 326
 327 /*
 328  * Checks whether the packet is accepted or not.
 329  * Returns: 0 - drop, !=0 - accept
 330  */
 331 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 332 {
 333         if (!filter->count)
 334                 return 1;
 335
 336         return run_filter(filter, skb);
 337 }
 338
 339 /* Network device part of the driver */
 340
 341 static const struct ethtool_ops tun_ethtool_ops;
 342
 343 /* Net device detach from fd. */
 344 static void tun_net_uninit(struct net_device *dev)
 345 {
 346         struct tun_struct *tun = netdev_priv(dev);
 347         struct tun_file *tfile = tun->tfile;
 348
 349         /* Inform the methods they need to stop using the dev.
 350          */
 351         if (tfile) {
 352                 wake_up_all(&tun->wq.wait);
 353                 if (atomic_dec_and_test(&tfile->count))
 354                         __tun_detach(tun);
 355         }
 356 }
 357
 358 static void tun_free_netdev(struct net_device *dev)
 359 {
 360         struct tun_struct *tun = netdev_priv(dev);
 361
 362         sock_put(tun->socket.sk);
 363 }
 364
 365 /* Net device open. */
 366 static int tun_net_open(struct net_device *dev)
 367 {
 368         netif_start_queue(dev);
 369         return 0;
 370 }
 371
 372 /* Net device close. */
 373 static int tun_net_close(struct net_device *dev)
 374 {
 375         netif_stop_queue(dev);
 376         return 0;
 377 }
 378
 379 /* Net device start xmit */
 380 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 381 {
 382         struct tun_struct *tun = netdev_priv(dev);
 383
 384         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 385
 386         /* Drop packet if interface is not attached */
 387         if (!tun->tfile)
 388                 goto drop;
 389
 390         /* Drop if the filter does not like it.
 391          * This is a noop if the filter is disabled.
 392          * Filter can be enabled only for the TAP devices. */
 393         if (!check_filter(&tun->txflt, skb))
 394                 goto drop;
 395
 396         if (tun->socket.sk->sk_filter &&
 397             sk_filter(tun->socket.sk, skb))
 398                 goto drop;
 399
 400         if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) {
 401                 if (!(tun->flags & TUN_ONE_QUEUE)) {
 402                         /* Normal queueing mode. */
 403                         /* Packet scheduler handles dropping of further packets. */
 404                         netif_stop_queue(dev);
 405
 406                         /* We won't see all dropped packets individually, so overrun
 407                          * error is more appropriate. */
 408                         dev->stats.tx_fifo_errors++;
 409                 } else {
 410                         /* Single queue mode.
 411                          * Driver handles dropping of all packets itself. */
 412                         goto drop;
 413                 }
 414         }
 415
 416         /* Orphan the skb - required as we might hang on to it
 417          * for indefinite time. */
 418         skb_orphan(skb);
 419
 420         /* Enqueue packet */
 421         skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb);
 422
 423         /* Notify and wake up reader process */
 424         if (tun->flags & TUN_FASYNC)
 425                 kill_fasync(&tun->fasync, SIGIO, POLL_IN);
 426         wake_up_interruptible_poll(&tun->wq.wait, POLLIN |
 427                                    POLLRDNORM | POLLRDBAND);
 428         return NETDEV_TX_OK;
 429
 430 drop:
 431         dev->stats.tx_dropped++;
 432         kfree_skb(skb);
 433         return NETDEV_TX_OK;
 434 }
 435
 436 static void tun_net_mclist(struct net_device *dev)
 437 {
 438         /*
 439          * This callback is supposed to deal with mc filter in
 440          * _rx_ path and has nothing to do with the _tx_ path.
 441          * In rx path we always accept everything userspace gives us.
 442          */
 443 }
 444
 445 #define MIN_MTU 68
 446 #define MAX_MTU 65535
 447
 448 static int
 449 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 450 {
 451         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 452                 return -EINVAL;
 453         dev->mtu = new_mtu;
 454         return 0;
 455 }
 456
 457 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 458         netdev_features_t features)
 459 {
 460         struct tun_struct *tun = netdev_priv(dev);
 461
 462         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 463 }
 464 #ifdef CONFIG_NET_POLL_CONTROLLER
 465 static void tun_poll_controller(struct net_device *dev)
 466 {
 467         /*
 468          * Tun only receives frames when:
 469          * 1) the char device endpoint gets data from user space
 470          * 2) the tun socket gets a sendmsg call from user space
 471          * Since both of those are syncronous operations, we are guaranteed
 472          * never to have pending data when we poll for it
 473          * so theres nothing to do here but return.
 474          * We need this though so netpoll recognizes us as an interface that
 475          * supports polling, which enables bridge devices in virt setups to
 476          * still use netconsole
 477          */
 478         return;
 479 }
 480 #endif
 481 static const struct net_device_ops tun_netdev_ops = {
 482         .ndo_uninit             = tun_net_uninit,
 483         .ndo_open               = tun_net_open,
 484         .ndo_stop               = tun_net_close,
 485         .ndo_start_xmit         = tun_net_xmit,
 486         .ndo_change_mtu         = tun_net_change_mtu,
 487         .ndo_fix_features       = tun_net_fix_features,
 488 #ifdef CONFIG_NET_POLL_CONTROLLER
 489         .ndo_poll_controller    = tun_poll_controller,
 490 #endif
 491 };
 492
 493 static const struct net_device_ops tap_netdev_ops = {
 494         .ndo_uninit             = tun_net_uninit,
 495         .ndo_open               = tun_net_open,
 496         .ndo_stop               = tun_net_close,
 497         .ndo_start_xmit         = tun_net_xmit,
 498         .ndo_change_mtu         = tun_net_change_mtu,
 499         .ndo_fix_features       = tun_net_fix_features,
 500         .ndo_set_rx_mode        = tun_net_mclist,
 501         .ndo_set_mac_address    = eth_mac_addr,
 502         .ndo_validate_addr      = eth_validate_addr,
 503 #ifdef CONFIG_NET_POLL_CONTROLLER
 504         .ndo_poll_controller    = tun_poll_controller,
 505 #endif
 506 };
 507
 508 /* Initialize net device. */
 509 static void tun_net_init(struct net_device *dev)
 510 {
 511         struct tun_struct *tun = netdev_priv(dev);
 512
 513         switch (tun->flags & TUN_TYPE_MASK) {
 514         case TUN_TUN_DEV:
 515                 dev->netdev_ops = &tun_netdev_ops;
 516
 517                 /* Point-to-Point TUN Device */
 518                 dev->hard_header_len = 0;
 519                 dev->addr_len = 0;
 520                 dev->mtu = 1500;
 521
 522                 /* Zero header length */
 523                 dev->type = ARPHRD_NONE;
 524                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 525                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 526                 break;
 527
 528         case TUN_TAP_DEV:
 529                 dev->netdev_ops = &tap_netdev_ops;
 530                 /* Ethernet TAP Device */
 531                 ether_setup(dev);
 532                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 533
 534                 random_ether_addr(dev->dev_addr);
 535
 536                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 537                 break;
 538         }
 539 }
 540
 541 /* Character device part */
 542
 543 /* Poll */
 544 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
 545 {
 546         struct tun_file *tfile = file->private_data;
 547         struct tun_struct *tun = __tun_get(tfile);
 548         struct sock *sk;
 549         unsigned int mask = 0;
 550
 551         if (!tun)
 552                 return POLLERR;
 553
 554         sk = tun->socket.sk;
 555
 556         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 557
 558         poll_wait(file, &tun->wq.wait, wait);
 559
 560         if (!skb_queue_empty(&sk->sk_receive_queue))
 561                 mask |= POLLIN | POLLRDNORM;
 562
 563         if (sock_writeable(sk) ||
 564             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 565              sock_writeable(sk)))
 566                 mask |= POLLOUT | POLLWRNORM;
 567
 568         if (tun->dev->reg_state != NETREG_REGISTERED)
 569                 mask = POLLERR;
 570
 571         tun_put(tun);
 572         return mask;
 573 }
 574
 575 /* prepad is the amount to reserve at front.  len is length after that.
 576  * linear is a hint as to how much to copy (usually headers). */
 577 static struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
 578                                      size_t prepad, size_t len,
 579                                      size_t linear, int noblock)
 580 {
 581         struct sock *sk = tun->socket.sk;
 582         struct sk_buff *skb;
 583         int err;
 584
 585         sock_update_classid(sk);
 586
 587         /* Under a page?  Don't bother with paged skb. */
 588         if (prepad + len < PAGE_SIZE || !linear)
 589                 linear = len;
 590
 591         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 592                                    &err);
 593         if (!skb)
 594                 return ERR_PTR(err);
 595
 596         skb_reserve(skb, prepad);
 597         skb_put(skb, linear);
 598         skb->data_len = len - linear;
 599         skb->len += len - linear;
 600
 601         return skb;
 602 }
 603
 604 /* Get packet from user space buffer */
 605 static ssize_t tun_get_user(struct tun_struct *tun,
 606                             const struct iovec *iv, size_t count,
 607                             int noblock)
 608 {
 609         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
 610         struct sk_buff *skb;
 611         size_t len = count, align = NET_SKB_PAD;
 612         struct virtio_net_hdr gso = { 0 };
 613         int offset = 0;
 614
 615         if (!(tun->flags & TUN_NO_PI)) {
 616                 if ((len -= sizeof(pi)) > count)
 617                         return -EINVAL;
 618
 619                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
 620                         return -EFAULT;
 621                 offset += sizeof(pi);
 622         }
 623
 624         if (tun->flags & TUN_VNET_HDR) {
 625                 if ((len -= tun->vnet_hdr_sz) > count)
 626                         return -EINVAL;
 627
 628                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
 629                         return -EFAULT;
 630
 631                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
 632                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
 633                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
 634
 635                 if (gso.hdr_len > len)
 636                         return -EINVAL;
 637                 offset += tun->vnet_hdr_sz;
 638         }
 639
 640         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
 641                 align += NET_IP_ALIGN;
 642                 if (unlikely(len < ETH_HLEN ||
 643                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
 644                         return -EINVAL;
 645         }
 646
 647         skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock);
 648         if (IS_ERR(skb)) {
 649                 if (PTR_ERR(skb) != -EAGAIN)
 650                         tun->dev->stats.rx_dropped++;
 651                 return PTR_ERR(skb);
 652         }
 653
 654         if (skb_copy_datagram_from_iovec(skb, 0, iv, offset, len)) {
 655                 tun->dev->stats.rx_dropped++;
 656                 kfree_skb(skb);
 657                 return -EFAULT;
 658         }
 659
 660         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 661                 if (!skb_partial_csum_set(skb, gso.csum_start,
 662                                           gso.csum_offset)) {
 663                         tun->dev->stats.rx_frame_errors++;
 664                         kfree_skb(skb);
 665                         return -EINVAL;
 666                 }
 667         }
 668
 669         switch (tun->flags & TUN_TYPE_MASK) {
 670         case TUN_TUN_DEV:
 671                 if (tun->flags & TUN_NO_PI) {
 672                         switch (skb->data[0] & 0xf0) {
 673                         case 0x40:
 674                                 pi.proto = htons(ETH_P_IP);
 675                                 break;
 676                         case 0x60:
 677                                 pi.proto = htons(ETH_P_IPV6);
 678                                 break;
 679                         default:
 680                                 tun->dev->stats.rx_dropped++;
 681                                 kfree_skb(skb);
 682                                 return -EINVAL;
 683                         }
 684                 }
 685
 686                 skb_reset_mac_header(skb);
 687                 skb->protocol = pi.proto;
 688                 skb->dev = tun->dev;
 689                 break;
 690         case TUN_TAP_DEV:
 691                 skb->protocol = eth_type_trans(skb, tun->dev);
 692                 break;
 693         }
 694
 695         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 696                 pr_debug("GSO!\n");
 697                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 698                 case VIRTIO_NET_HDR_GSO_TCPV4:
 699                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 700                         break;
 701                 case VIRTIO_NET_HDR_GSO_TCPV6:
 702                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 703                         break;
 704                 case VIRTIO_NET_HDR_GSO_UDP:
 705                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 706                         break;
 707                 default:
 708                         tun->dev->stats.rx_frame_errors++;
 709                         kfree_skb(skb);
 710                         return -EINVAL;
 711                 }
 712
 713                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
 714                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 715
 716                 skb_shinfo(skb)->gso_size = gso.gso_size;
 717                 if (skb_shinfo(skb)->gso_size == 0) {
 718                         tun->dev->stats.rx_frame_errors++;
 719                         kfree_skb(skb);
 720                         return -EINVAL;
 721                 }
 722
 723                 /* Header must be checked, and gso_segs computed. */
 724                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 725                 skb_shinfo(skb)->gso_segs = 0;
 726         }
 727
 728         netif_rx_ni(skb);
 729
 730         tun->dev->stats.rx_packets++;
 731         tun->dev->stats.rx_bytes += len;
 732
 733         return count;
 734 }
 735
 736 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
 737                               unsigned long count, loff_t pos)
 738 {
 739         struct file *file = iocb->ki_filp;
 740         struct tun_struct *tun = tun_get(file);
 741         ssize_t result;
 742
 743         if (!tun)
 744                 return -EBADFD;
 745
 746         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
 747
 748         result = tun_get_user(tun, iv, iov_length(iv, count),
 749                               file->f_flags & O_NONBLOCK);
 750
 751         tun_put(tun);
 752         return result;
 753 }
 754
 755 /* Put packet to the user space buffer */
 756 static ssize_t tun_put_user(struct tun_struct *tun,
 757                             struct sk_buff *skb,
 758                             const struct iovec *iv, int len)
 759 {
 760         struct tun_pi pi = { 0, skb->protocol };
 761         ssize_t total = 0;
 762
 763         if (!(tun->flags & TUN_NO_PI)) {
 764                 if ((len -= sizeof(pi)) < 0)
 765                         return -EINVAL;
 766
 767                 if (len < skb->len) {
 768                         /* Packet will be striped */
 769                         pi.flags |= TUN_PKT_STRIP;
 770                 }
 771
 772                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
 773                         return -EFAULT;
 774                 total += sizeof(pi);
 775         }
 776
 777         if (tun->flags & TUN_VNET_HDR) {
 778                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
 779                 if ((len -= tun->vnet_hdr_sz) < 0)
 780                         return -EINVAL;
 781
 782                 if (skb_is_gso(skb)) {
 783                         struct skb_shared_info *sinfo = skb_shinfo(skb);
 784
 785                         /* This is a hint as to how much should be linear. */
 786                         gso.hdr_len = skb_headlen(skb);
 787                         gso.gso_size = sinfo->gso_size;
 788                         if (sinfo->gso_type & SKB_GSO_TCPV4)
 789                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 790                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
 791                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 792                         else if (sinfo->gso_type & SKB_GSO_UDP)
 793                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
 794                         else {
 795                                 pr_err("unexpected GSO type: "
 796                                        "0x%x, gso_size %d, hdr_len %d\n",
 797                                        sinfo->gso_type, gso.gso_size,
 798                                        gso.hdr_len);
 799                                 print_hex_dump(KERN_ERR, "tun: ",
 800                                                DUMP_PREFIX_NONE,
 801                                                16, 1, skb->head,
 802                                                min((int)gso.hdr_len, 64), true);
 803                                 WARN_ON_ONCE(1);
 804                                 return -EINVAL;
 805                         }
 806                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
 807                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 808                 } else
 809                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
 810
 811                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 812                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 813                         gso.csum_start = skb_checksum_start_offset(skb);
 814                         gso.csum_offset = skb->csum_offset;
 815                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
 816                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
 817                 } /* else everything is zero */
 818
 819                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
 820                                                sizeof(gso))))
 821                         return -EFAULT;
 822                 total += tun->vnet_hdr_sz;
 823         }
 824
 825         len = min_t(int, skb->len, len);
 826
 827         skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
 828         total += skb->len;
 829
 830         tun->dev->stats.tx_packets++;
 831         tun->dev->stats.tx_bytes += len;
 832
 833         return total;
 834 }
 835
 836 static ssize_t tun_do_read(struct tun_struct *tun,
 837                            struct kiocb *iocb, const struct iovec *iv,
 838                            ssize_t len, int noblock)
 839 {
 840         DECLARE_WAITQUEUE(wait, current);
 841         struct sk_buff *skb;
 842         ssize_t ret = 0;
 843
 844         tun_debug(KERN_INFO, tun, "tun_chr_read\n");
 845
 846         if (unlikely(!noblock))
 847                 add_wait_queue(&tun->wq.wait, &wait);
 848         while (len) {
 849                 current->state = TASK_INTERRUPTIBLE;
 850
 851                 /* Read frames from the queue */
 852                 if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
 853                         if (noblock) {
 854                                 ret = -EAGAIN;
 855                                 break;
 856                         }
 857                         if (signal_pending(current)) {
 858                                 ret = -ERESTARTSYS;
 859                                 break;
 860                         }
 861                         if (tun->dev->reg_state != NETREG_REGISTERED) {
 862                                 ret = -EIO;
 863                                 break;
 864                         }
 865
 866                         /* Nothing to read, let's sleep */
 867                         schedule();
 868                         continue;
 869                 }
 870                 netif_wake_queue(tun->dev);
 871
 872                 ret = tun_put_user(tun, skb, iv, len);
 873                 kfree_skb(skb);
 874                 break;
 875         }
 876
 877         current->state = TASK_RUNNING;
 878         if (unlikely(!noblock))
 879                 remove_wait_queue(&tun->wq.wait, &wait);
 880
 881         return ret;
 882 }
 883
 884 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
 885                             unsigned long count, loff_t pos)
 886 {
 887         struct file *file = iocb->ki_filp;
 888         struct tun_file *tfile = file->private_data;
 889         struct tun_struct *tun = __tun_get(tfile);
 890         ssize_t len, ret;
 891
 892         if (!tun)
 893                 return -EBADFD;
 894         len = iov_length(iv, count);
 895         if (len < 0) {
 896                 ret = -EINVAL;
 897                 goto out;
 898         }
 899
 900         ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK);
 901         ret = min_t(ssize_t, ret, len);
 902 out:
 903         tun_put(tun);
 904         return ret;
 905 }
 906
 907 static void tun_setup(struct net_device *dev)
 908 {
 909         struct tun_struct *tun = netdev_priv(dev);
 910
 911         tun->owner = -1;
 912         tun->group = -1;
 913
 914         dev->ethtool_ops = &tun_ethtool_ops;
 915         dev->destructor = tun_free_netdev;
 916 }
 917
 918 /* Trivial set of netlink ops to allow deleting tun or tap
 919  * device with netlink.
 920  */
 921 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
 922 {
 923         return -EINVAL;
 924 }
 925
 926 static struct rtnl_link_ops tun_link_ops __read_mostly = {
 927         .kind           = DRV_NAME,
 928         .priv_size      = sizeof(struct tun_struct),
 929         .setup          = tun_setup,
 930         .validate       = tun_validate,
 931 };
 932
 933 static void tun_sock_write_space(struct sock *sk)
 934 {
 935         struct tun_struct *tun;
 936         wait_queue_head_t *wqueue;
 937
 938         if (!sock_writeable(sk))
 939                 return;
 940
 941         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
 942                 return;
 943
 944         wqueue = sk_sleep(sk);
 945         if (wqueue && waitqueue_active(wqueue))
 946                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
 947                                                 POLLWRNORM | POLLWRBAND);
 948
 949         tun = tun_sk(sk)->tun;
 950         kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
 951 }
 952
 953 static void tun_sock_destruct(struct sock *sk)
 954 {
 955         free_netdev(tun_sk(sk)->tun->dev);
 956 }
 957
 958 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
 959                        struct msghdr *m, size_t total_len)
 960 {
 961         struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
 962         return tun_get_user(tun, m->msg_iov, total_len,
 963                             m->msg_flags & MSG_DONTWAIT);
 964 }
 965
 966 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
 967                        struct msghdr *m, size_t total_len,
 968                        int flags)
 969 {
 970         struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
 971         int ret;
 972         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
 973                 return -EINVAL;
 974         ret = tun_do_read(tun, iocb, m->msg_iov, total_len,
 975                           flags & MSG_DONTWAIT);
 976         if (ret > total_len) {
 977                 m->msg_flags |= MSG_TRUNC;
 978                 ret = flags & MSG_TRUNC ? ret : total_len;
 979         }
 980         return ret;
 981 }
 982
 983 /* Ops structure to mimic raw sockets with tun */
 984 static const struct proto_ops tun_socket_ops = {
 985         .sendmsg = tun_sendmsg,
 986         .recvmsg = tun_recvmsg,
 987 };
 988
 989 static struct proto tun_proto = {
 990         .name           = "tun",
 991         .owner          = THIS_MODULE,
 992         .obj_size       = sizeof(struct tun_sock),
 993 };
 994
 995 static int tun_flags(struct tun_struct *tun)
 996 {
 997         int flags = 0;
 998
 999         if (tun->flags & TUN_TUN_DEV)
1000                 flags |= IFF_TUN;
1001         else
1002                 flags |= IFF_TAP;
1003
1004         if (tun->flags & TUN_NO_PI)
1005                 flags |= IFF_NO_PI;
1006
1007         if (tun->flags & TUN_ONE_QUEUE)
1008                 flags |= IFF_ONE_QUEUE;
1009
1010         if (tun->flags & TUN_VNET_HDR)
1011                 flags |= IFF_VNET_HDR;
1012
1013         return flags;
1014 }
1015
1016 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1017                               char *buf)
1018 {
1019         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1020         return sprintf(buf, "0x%x\n", tun_flags(tun));
1021 }
1022
1023 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1024                               char *buf)
1025 {
1026         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1027         return sprintf(buf, "%d\n", tun->owner);
1028 }
1029
1030 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1031                               char *buf)
1032 {
1033         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1034         return sprintf(buf, "%d\n", tun->group);
1035 }
1036
1037 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1038 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1039 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1040
1041 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1042 {
1043         struct sock *sk;
1044         struct tun_struct *tun;
1045         struct net_device *dev;
1046         int err;
1047
1048         dev = __dev_get_by_name(net, ifr->ifr_name);
1049         if (dev) {
1050                 const struct cred *cred = current_cred();
1051
1052                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1053                         return -EBUSY;
1054                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1055                         tun = netdev_priv(dev);
1056                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1057                         tun = netdev_priv(dev);
1058                 else
1059                         return -EINVAL;
1060
1061                 if (((tun->owner != -1 && cred->euid != tun->owner) ||
1062                      (tun->group != -1 && !in_egroup_p(tun->group))) &&
1063                     !capable(CAP_NET_ADMIN))
1064                         return -EPERM;
1065                 err = security_tun_dev_attach(tun->socket.sk);
1066                 if (err < 0)
1067                         return err;
1068
1069                 err = tun_attach(tun, file);
1070                 if (err < 0)
1071                         return err;
1072         }
1073         else {
1074                 char *name;
1075                 unsigned long flags = 0;
1076
1077                 if (!capable(CAP_NET_ADMIN))
1078                         return -EPERM;
1079                 err = security_tun_dev_create();
1080                 if (err < 0)
1081                         return err;
1082
1083                 /* Set dev type */
1084                 if (ifr->ifr_flags & IFF_TUN) {
1085                         /* TUN device */
1086                         flags |= TUN_TUN_DEV;
1087                         name = "tun%d";
1088                 } else if (ifr->ifr_flags & IFF_TAP) {
1089                         /* TAP device */
1090                         flags |= TUN_TAP_DEV;
1091                         name = "tap%d";
1092                 } else
1093                         return -EINVAL;
1094
1095                 if (*ifr->ifr_name)
1096                         name = ifr->ifr_name;
1097
1098                 dev = alloc_netdev(sizeof(struct tun_struct), name,
1099                                    tun_setup);
1100                 if (!dev)
1101                         return -ENOMEM;
1102
1103                 dev_net_set(dev, net);
1104                 dev->rtnl_link_ops = &tun_link_ops;
1105
1106                 tun = netdev_priv(dev);
1107                 tun->dev = dev;
1108                 tun->flags = flags;
1109                 tun->txflt.count = 0;
1110                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1111
1112                 err = -ENOMEM;
1113                 sk = sk_alloc(net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1114                 if (!sk)
1115                         goto err_free_dev;
1116
1117                 tun->socket.wq = &tun->wq;
1118                 init_waitqueue_head(&tun->wq.wait);
1119                 tun->socket.ops = &tun_socket_ops;
1120                 sock_init_data(&tun->socket, sk);
1121                 sk->sk_write_space = tun_sock_write_space;
1122                 sk->sk_sndbuf = INT_MAX;
1123
1124                 tun_sk(sk)->tun = tun;
1125
1126                 security_tun_dev_post_create(sk);
1127
1128                 tun_net_init(dev);
1129
1130                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1131                         TUN_USER_FEATURES;
1132                 dev->features = dev->hw_features;
1133
1134                 err = register_netdevice(tun->dev);
1135                 if (err < 0)
1136                         goto err_free_sk;
1137
1138                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1139                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1140                     device_create_file(&tun->dev->dev, &dev_attr_group))
1141                         pr_err("Failed to create tun sysfs files\n");
1142
1143                 sk->sk_destruct = tun_sock_destruct;
1144
1145                 err = tun_attach(tun, file);
1146                 if (err < 0)
1147                         goto failed;
1148         }
1149
1150         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1151
1152         if (ifr->ifr_flags & IFF_NO_PI)
1153                 tun->flags |= TUN_NO_PI;
1154         else
1155                 tun->flags &= ~TUN_NO_PI;
1156
1157         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1158                 tun->flags |= TUN_ONE_QUEUE;
1159         else
1160                 tun->flags &= ~TUN_ONE_QUEUE;
1161
1162         if (ifr->ifr_flags & IFF_VNET_HDR)
1163                 tun->flags |= TUN_VNET_HDR;
1164         else
1165                 tun->flags &= ~TUN_VNET_HDR;
1166
1167         /* Make sure persistent devices do not get stuck in
1168          * xoff state.
1169          */
1170         if (netif_running(tun->dev))
1171                 netif_wake_queue(tun->dev);
1172
1173         strcpy(ifr->ifr_name, tun->dev->name);
1174         return 0;
1175
1176  err_free_sk:
1177         sock_put(sk);
1178  err_free_dev:
1179         free_netdev(dev);
1180  failed:
1181         return err;
1182 }
1183
1184 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1185                        struct ifreq *ifr)
1186 {
1187         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1188
1189         strcpy(ifr->ifr_name, tun->dev->name);
1190
1191         ifr->ifr_flags = tun_flags(tun);
1192
1193         return 0;
1194 }
1195
1196 /* This is like a cut-down ethtool ops, except done via tun fd so no
1197  * privs required. */
1198 static int set_offload(struct tun_struct *tun, unsigned long arg)
1199 {
1200         netdev_features_t features = 0;
1201
1202         if (arg & TUN_F_CSUM) {
1203                 features |= NETIF_F_HW_CSUM;
1204                 arg &= ~TUN_F_CSUM;
1205
1206                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1207                         if (arg & TUN_F_TSO_ECN) {
1208                                 features |= NETIF_F_TSO_ECN;
1209                                 arg &= ~TUN_F_TSO_ECN;
1210                         }
1211                         if (arg & TUN_F_TSO4)
1212                                 features |= NETIF_F_TSO;
1213                         if (arg & TUN_F_TSO6)
1214                                 features |= NETIF_F_TSO6;
1215                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1216                 }
1217
1218                 if (arg & TUN_F_UFO) {
1219                         features |= NETIF_F_UFO;
1220                         arg &= ~TUN_F_UFO;
1221                 }
1222         }
1223
1224         /* This gives the user a way to test for new features in future by
1225          * trying to set them. */
1226         if (arg)
1227                 return -EINVAL;
1228
1229         tun->set_features = features;
1230         netdev_update_features(tun->dev);
1231
1232         return 0;
1233 }
1234
1235 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1236                             unsigned long arg, int ifreq_len)
1237 {
1238         struct tun_file *tfile = file->private_data;
1239         struct tun_struct *tun;
1240         void __user* argp = (void __user*)arg;
1241         struct sock_fprog fprog;
1242         struct ifreq ifr;
1243         int sndbuf;
1244         int vnet_hdr_sz;
1245         int ret;
1246
1247         if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
1248                 if (copy_from_user(&ifr, argp, ifreq_len))
1249                         return -EFAULT;
1250
1251         if (cmd == TUNGETFEATURES) {
1252                 /* Currently this just means: "what IFF flags are valid?".
1253                  * This is needed because we never checked for invalid flags on
1254                  * TUNSETIFF. */
1255                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1256                                 IFF_VNET_HDR,
1257                                 (unsigned int __user*)argp);
1258         }
1259
1260         rtnl_lock();
1261
1262         tun = __tun_get(tfile);
1263         if (cmd == TUNSETIFF && !tun) {
1264                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1265
1266                 ret = tun_set_iff(tfile->net, file, &ifr);
1267
1268                 if (ret)
1269                         goto unlock;
1270
1271                 if (copy_to_user(argp, &ifr, ifreq_len))
1272                         ret = -EFAULT;
1273                 goto unlock;
1274         }
1275
1276         ret = -EBADFD;
1277         if (!tun)
1278                 goto unlock;
1279
1280         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %d\n", cmd);
1281
1282         ret = 0;
1283         switch (cmd) {
1284         case TUNGETIFF:
1285                 ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1286                 if (ret)
1287                         break;
1288
1289                 if (copy_to_user(argp, &ifr, ifreq_len))
1290                         ret = -EFAULT;
1291                 break;
1292
1293         case TUNSETNOCSUM:
1294                 /* Disable/Enable checksum */
1295
1296                 /* [unimplemented] */
1297                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1298                           arg ? "disabled" : "enabled");
1299                 break;
1300
1301         case TUNSETPERSIST:
1302                 /* Disable/Enable persist mode */
1303                 if (arg)
1304                         tun->flags |= TUN_PERSIST;
1305                 else
1306                         tun->flags &= ~TUN_PERSIST;
1307
1308                 tun_debug(KERN_INFO, tun, "persist %s\n",
1309                           arg ? "enabled" : "disabled");
1310                 break;
1311
1312         case TUNSETOWNER:
1313                 /* Set owner of the device */
1314                 tun->owner = (uid_t) arg;
1315
1316                 tun_debug(KERN_INFO, tun, "owner set to %d\n", tun->owner);
1317                 break;
1318
1319         case TUNSETGROUP:
1320                 /* Set group of the device */
1321                 tun->group= (gid_t) arg;
1322
1323                 tun_debug(KERN_INFO, tun, "group set to %d\n", tun->group);
1324                 break;
1325
1326         case TUNSETLINK:
1327                 /* Only allow setting the type when the interface is down */
1328                 if (tun->dev->flags & IFF_UP) {
1329                         tun_debug(KERN_INFO, tun,
1330                                   "Linktype set failed because interface is up\n");
1331                         ret = -EBUSY;
1332                 } else {
1333                         tun->dev->type = (int) arg;
1334                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1335                                   tun->dev->type);
1336                         ret = 0;
1337                 }
1338                 break;
1339
1340 #ifdef TUN_DEBUG
1341         case TUNSETDEBUG:
1342                 tun->debug = arg;
1343                 break;
1344 #endif
1345         case TUNSETOFFLOAD:
1346                 ret = set_offload(tun, arg);
1347                 break;
1348
1349         case TUNSETTXFILTER:
1350                 /* Can be set only for TAPs */
1351                 ret = -EINVAL;
1352                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1353                         break;
1354                 ret = update_filter(&tun->txflt, (void __user *)arg);
1355                 break;
1356
1357         case SIOCGIFHWADDR:
1358                 /* Get hw address */
1359                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1360                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1361                 if (copy_to_user(argp, &ifr, ifreq_len))
1362                         ret = -EFAULT;
1363                 break;
1364
1365         case SIOCSIFHWADDR:
1366                 /* Set hw address */
1367                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1368                           ifr.ifr_hwaddr.sa_data);
1369
1370                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1371                 break;
1372
1373         case TUNGETSNDBUF:
1374                 sndbuf = tun->socket.sk->sk_sndbuf;
1375                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1376                         ret = -EFAULT;
1377                 break;
1378
1379         case TUNSETSNDBUF:
1380                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1381                         ret = -EFAULT;
1382                         break;
1383                 }
1384
1385                 tun->socket.sk->sk_sndbuf = sndbuf;
1386                 break;
1387
1388         case TUNGETVNETHDRSZ:
1389                 vnet_hdr_sz = tun->vnet_hdr_sz;
1390                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1391                         ret = -EFAULT;
1392                 break;
1393
1394         case TUNSETVNETHDRSZ:
1395                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1396                         ret = -EFAULT;
1397                         break;
1398                 }
1399                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1400                         ret = -EINVAL;
1401                         break;
1402                 }
1403
1404                 tun->vnet_hdr_sz = vnet_hdr_sz;
1405                 break;
1406
1407         case TUNATTACHFILTER:
1408                 /* Can be set only for TAPs */
1409                 ret = -EINVAL;
1410                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1411                         break;
1412                 ret = -EFAULT;
1413                 if (copy_from_user(&fprog, argp, sizeof(fprog)))
1414                         break;
1415
1416                 ret = sk_attach_filter(&fprog, tun->socket.sk);
1417                 break;
1418
1419         case TUNDETACHFILTER:
1420                 /* Can be set only for TAPs */
1421                 ret = -EINVAL;
1422                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1423                         break;
1424                 ret = sk_detach_filter(tun->socket.sk);
1425                 break;
1426
1427         default:
1428                 ret = -EINVAL;
1429                 break;
1430         }
1431
1432 unlock:
1433         rtnl_unlock();
1434         if (tun)
1435                 tun_put(tun);
1436         return ret;
1437 }
1438
1439 static long tun_chr_ioctl(struct file *file,
1440                           unsigned int cmd, unsigned long arg)
1441 {
1442         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1443 }
1444
1445 #ifdef CONFIG_COMPAT
1446 static long tun_chr_compat_ioctl(struct file *file,
1447                          unsigned int cmd, unsigned long arg)
1448 {
1449         switch (cmd) {
1450         case TUNSETIFF:
1451         case TUNGETIFF:
1452         case TUNSETTXFILTER:
1453         case TUNGETSNDBUF:
1454         case TUNSETSNDBUF:
1455         case SIOCGIFHWADDR:
1456         case SIOCSIFHWADDR:
1457                 arg = (unsigned long)compat_ptr(arg);
1458                 break;
1459         default:
1460                 arg = (compat_ulong_t)arg;
1461                 break;
1462         }
1463
1464         /*
1465          * compat_ifreq is shorter than ifreq, so we must not access beyond
1466          * the end of that structure. All fields that are used in this
1467          * driver are compatible though, we don't need to convert the
1468          * contents.
1469          */
1470         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1471 }
1472 #endif /* CONFIG_COMPAT */
1473
1474 static int tun_chr_fasync(int fd, struct file *file, int on)
1475 {
1476         struct tun_struct *tun = tun_get(file);
1477         int ret;
1478
1479         if (!tun)
1480                 return -EBADFD;
1481
1482         tun_debug(KERN_INFO, tun, "tun_chr_fasync %d\n", on);
1483
1484         if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1485                 goto out;
1486
1487         if (on) {
1488                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1489                 if (ret)
1490                         goto out;
1491                 tun->flags |= TUN_FASYNC;
1492         } else
1493                 tun->flags &= ~TUN_FASYNC;
1494         ret = 0;
1495 out:
1496         tun_put(tun);
1497         return ret;
1498 }
1499
1500 static int tun_chr_open(struct inode *inode, struct file * file)
1501 {
1502         struct tun_file *tfile;
1503
1504         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
1505
1506         tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1507         if (!tfile)
1508                 return -ENOMEM;
1509         atomic_set(&tfile->count, 0);
1510         tfile->tun = NULL;
1511         tfile->net = get_net(current->nsproxy->net_ns);
1512         file->private_data = tfile;
1513         return 0;
1514 }
1515
1516 static int tun_chr_close(struct inode *inode, struct file *file)
1517 {
1518         struct tun_file *tfile = file->private_data;
1519         struct tun_struct *tun;
1520
1521         tun = __tun_get(tfile);
1522         if (tun) {
1523                 struct net_device *dev = tun->dev;
1524
1525                 tun_debug(KERN_INFO, tun, "tun_chr_close\n");
1526
1527                 __tun_detach(tun);
1528
1529                 /* If desirable, unregister the netdevice. */
1530                 if (!(tun->flags & TUN_PERSIST)) {
1531                         rtnl_lock();
1532                         if (dev->reg_state == NETREG_REGISTERED)
1533                                 unregister_netdevice(dev);
1534                         rtnl_unlock();
1535                 }
1536         }
1537
1538         tun = tfile->tun;
1539         if (tun)
1540                 sock_put(tun->socket.sk);
1541
1542         put_net(tfile->net);
1543         kfree(tfile);
1544
1545         return 0;
1546 }
1547
1548 static const struct file_operations tun_fops = {
1549         .owner  = THIS_MODULE,
1550         .llseek = no_llseek,
1551         .read  = do_sync_read,
1552         .aio_read  = tun_chr_aio_read,
1553         .write = do_sync_write,
1554         .aio_write = tun_chr_aio_write,
1555         .poll   = tun_chr_poll,
1556         .unlocked_ioctl = tun_chr_ioctl,
1557 #ifdef CONFIG_COMPAT
1558         .compat_ioctl = tun_chr_compat_ioctl,
1559 #endif
1560         .open   = tun_chr_open,
1561         .release = tun_chr_close,
1562         .fasync = tun_chr_fasync
1563 };
1564
1565 static struct miscdevice tun_miscdev = {
1566         .minor = TUN_MINOR,
1567         .name = "tun",
1568         .nodename = "net/tun",
1569         .fops = &tun_fops,
1570 };
1571
1572 /* ethtool interface */
1573
1574 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1575 {
1576         cmd->supported          = 0;
1577         cmd->advertising        = 0;
1578         ethtool_cmd_speed_set(cmd, SPEED_10);
1579         cmd->duplex             = DUPLEX_FULL;
1580         cmd->port               = PORT_TP;
1581         cmd->phy_address        = 0;
1582         cmd->transceiver        = XCVR_INTERNAL;
1583         cmd->autoneg            = AUTONEG_DISABLE;
1584         cmd->maxtxpkt           = 0;
1585         cmd->maxrxpkt           = 0;
1586         return 0;
1587 }
1588
1589 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1590 {
1591         struct tun_struct *tun = netdev_priv(dev);
1592
1593         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1594         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1595
1596         switch (tun->flags & TUN_TYPE_MASK) {
1597         case TUN_TUN_DEV:
1598                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
1599                 break;
1600         case TUN_TAP_DEV:
1601                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
1602                 break;
1603         }
1604 }
1605
1606 static u32 tun_get_msglevel(struct net_device *dev)
1607 {
1608 #ifdef TUN_DEBUG
1609         struct tun_struct *tun = netdev_priv(dev);
1610         return tun->debug;
1611 #else
1612         return -EOPNOTSUPP;
1613 #endif
1614 }
1615
1616 static void tun_set_msglevel(struct net_device *dev, u32 value)
1617 {
1618 #ifdef TUN_DEBUG
1619         struct tun_struct *tun = netdev_priv(dev);
1620         tun->debug = value;
1621 #endif
1622 }
1623
1624 static const struct ethtool_ops tun_ethtool_ops = {
1625         .get_settings   = tun_get_settings,
1626         .get_drvinfo    = tun_get_drvinfo,
1627         .get_msglevel   = tun_get_msglevel,
1628         .set_msglevel   = tun_set_msglevel,
1629         .get_link       = ethtool_op_get_link,
1630 };
1631
1632
1633 static int __init tun_init(void)
1634 {
1635         int ret = 0;
1636
1637         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1638         pr_info("%s\n", DRV_COPYRIGHT);
1639
1640         ret = rtnl_link_register(&tun_link_ops);
1641         if (ret) {
1642                 pr_err("Can't register link_ops\n");
1643                 goto err_linkops;
1644         }
1645
1646         ret = misc_register(&tun_miscdev);
1647         if (ret) {
1648                 pr_err("Can't register misc device %d\n", TUN_MINOR);
1649                 goto err_misc;
1650         }
1651         return  0;
1652 err_misc:
1653         rtnl_link_unregister(&tun_link_ops);
1654 err_linkops:
1655         return ret;
1656 }
1657
1658 static void tun_cleanup(void)
1659 {
1660         misc_deregister(&tun_miscdev);
1661         rtnl_link_unregister(&tun_link_ops);
1662 }
1663
1664 /* Get an underlying socket object from tun file.  Returns error unless file is
1665  * attached to a device.  The returned object works like a packet socket, it
1666  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1667  * holding a reference to the file for as long as the socket is in use. */
1668 struct socket *tun_get_socket(struct file *file)
1669 {
1670         struct tun_struct *tun;
1671         if (file->f_op != &tun_fops)
1672                 return ERR_PTR(-EINVAL);
1673         tun = tun_get(file);
1674         if (!tun)
1675                 return ERR_PTR(-EBADFD);
1676         tun_put(tun);
1677         return &tun->socket;
1678 }
1679 EXPORT_SYMBOL_GPL(tun_get_socket);
1680
1681 module_init(tun_init);
1682 module_exit(tun_cleanup);
1683 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1684 MODULE_AUTHOR(DRV_COPYRIGHT);
1685 MODULE_LICENSE("GPL");
1686 MODULE_ALIAS_MISCDEV(TUN_MINOR);
1687 MODULE_ALIAS("devname:net/tun");