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