drivers/net/tun.c

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