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