drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <[email protected]>
   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 <[email protected]> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <[email protected]>
  25  *    Use random_ether_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <[email protected]>  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 <[email protected]>
  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 <[email protected]>"
  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
 113 #ifdef TUN_DEBUG
 114         int debug;
 115 #endif
 116 };
 117
 118 struct tun_sock {
 119         struct sock             sk;
 120         struct tun_struct       *tun;
 121 };
 122
 123 static inline struct tun_sock *tun_sk(struct sock *sk)
 124 {
 125         return container_of(sk, struct tun_sock, sk);
 126 }
 127
 128 static int tun_attach(struct tun_struct *tun, struct file *file)
 129 {
 130         struct tun_file *tfile = file->private_data;
 131         int err;
 132
 133         ASSERT_RTNL();
 134
 135         netif_tx_lock_bh(tun->dev);
 136
 137         err = -EINVAL;
 138         if (tfile->tun)
 139                 goto out;
 140
 141         err = -EBUSY;
 142         if (tun->tfile)
 143                 goto out;
 144
 145         err = 0;
 146         tfile->tun = tun;
 147         tun->tfile = tfile;
 148         tun->socket.file = file;
 149         dev_hold(tun->dev);
 150         sock_hold(tun->socket.sk);
 151         atomic_inc(&tfile->count);
 152
 153 out:
 154         netif_tx_unlock_bh(tun->dev);
 155         return err;
 156 }
 157
 158 static void __tun_detach(struct tun_struct *tun)
 159 {
 160         /* Detach from net device */
 161         netif_tx_lock_bh(tun->dev);
 162         tun->tfile = NULL;
 163         tun->socket.file = NULL;
 164         netif_tx_unlock_bh(tun->dev);
 165
 166         /* Drop read queue */
 167         skb_queue_purge(&tun->socket.sk->sk_receive_queue);
 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->socket.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 netdev_tx_t 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 (tun->socket.sk->sk_filter &&
 371             sk_filter(tun->socket.sk, skb))
 372                 goto drop;
 373
 374         if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) {
 375                 if (!(tun->flags & TUN_ONE_QUEUE)) {
 376                         /* Normal queueing mode. */
 377                         /* Packet scheduler handles dropping of further packets. */
 378                         netif_stop_queue(dev);
 379
 380                         /* We won't see all dropped packets individually, so overrun
 381                          * error is more appropriate. */
 382                         dev->stats.tx_fifo_errors++;
 383                 } else {
 384                         /* Single queue mode.
 385                          * Driver handles dropping of all packets itself. */
 386                         goto drop;
 387                 }
 388         }
 389
 390         /* Enqueue packet */
 391         skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb);
 392         dev->trans_start = jiffies;
 393
 394         /* Notify and wake up reader process */
 395         if (tun->flags & TUN_FASYNC)
 396                 kill_fasync(&tun->fasync, SIGIO, POLL_IN);
 397         wake_up_interruptible_poll(&tun->socket.wait, POLLIN |
 398                                    POLLRDNORM | POLLRDBAND);
 399         return NETDEV_TX_OK;
 400
 401 drop:
 402         dev->stats.tx_dropped++;
 403         kfree_skb(skb);
 404         return NETDEV_TX_OK;
 405 }
 406
 407 static void tun_net_mclist(struct net_device *dev)
 408 {
 409         /*
 410          * This callback is supposed to deal with mc filter in
 411          * _rx_ path and has nothing to do with the _tx_ path.
 412          * In rx path we always accept everything userspace gives us.
 413          */
 414         return;
 415 }
 416
 417 #define MIN_MTU 68
 418 #define MAX_MTU 65535
 419
 420 static int
 421 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 422 {
 423         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 424                 return -EINVAL;
 425         dev->mtu = new_mtu;
 426         return 0;
 427 }
 428
 429 static const struct net_device_ops tun_netdev_ops = {
 430         .ndo_uninit             = tun_net_uninit,
 431         .ndo_open               = tun_net_open,
 432         .ndo_stop               = tun_net_close,
 433         .ndo_start_xmit         = tun_net_xmit,
 434         .ndo_change_mtu         = tun_net_change_mtu,
 435 };
 436
 437 static const struct net_device_ops tap_netdev_ops = {
 438         .ndo_uninit             = tun_net_uninit,
 439         .ndo_open               = tun_net_open,
 440         .ndo_stop               = tun_net_close,
 441         .ndo_start_xmit         = tun_net_xmit,
 442         .ndo_change_mtu         = tun_net_change_mtu,
 443         .ndo_set_multicast_list = tun_net_mclist,
 444         .ndo_set_mac_address    = eth_mac_addr,
 445         .ndo_validate_addr      = eth_validate_addr,
 446 };
 447
 448 /* Initialize net device. */
 449 static void tun_net_init(struct net_device *dev)
 450 {
 451         struct tun_struct *tun = netdev_priv(dev);
 452
 453         switch (tun->flags & TUN_TYPE_MASK) {
 454         case TUN_TUN_DEV:
 455                 dev->netdev_ops = &tun_netdev_ops;
 456
 457                 /* Point-to-Point TUN Device */
 458                 dev->hard_header_len = 0;
 459                 dev->addr_len = 0;
 460                 dev->mtu = 1500;
 461
 462                 /* Zero header length */
 463                 dev->type = ARPHRD_NONE;
 464                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 465                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 466                 break;
 467
 468         case TUN_TAP_DEV:
 469                 dev->netdev_ops = &tap_netdev_ops;
 470                 /* Ethernet TAP Device */
 471                 ether_setup(dev);
 472
 473                 random_ether_addr(dev->dev_addr);
 474
 475                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 476                 break;
 477         }
 478 }
 479
 480 /* Character device part */
 481
 482 /* Poll */
 483 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
 484 {
 485         struct tun_file *tfile = file->private_data;
 486         struct tun_struct *tun = __tun_get(tfile);
 487         struct sock *sk;
 488         unsigned int mask = 0;
 489
 490         if (!tun)
 491                 return POLLERR;
 492
 493         sk = tun->socket.sk;
 494
 495         DBG(KERN_INFO "%s: tun_chr_poll\n", tun->dev->name);
 496
 497         poll_wait(file, &tun->socket.wait, wait);
 498
 499         if (!skb_queue_empty(&sk->sk_receive_queue))
 500                 mask |= POLLIN | POLLRDNORM;
 501
 502         if (sock_writeable(sk) ||
 503             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 504              sock_writeable(sk)))
 505                 mask |= POLLOUT | POLLWRNORM;
 506
 507         if (tun->dev->reg_state != NETREG_REGISTERED)
 508                 mask = POLLERR;
 509
 510         tun_put(tun);
 511         return mask;
 512 }
 513
 514 /* prepad is the amount to reserve at front.  len is length after that.
 515  * linear is a hint as to how much to copy (usually headers). */
 516 static inline struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
 517                                             size_t prepad, size_t len,
 518                                             size_t linear, int noblock)
 519 {
 520         struct sock *sk = tun->socket.sk;
 521         struct sk_buff *skb;
 522         int err;
 523
 524         /* Under a page?  Don't bother with paged skb. */
 525         if (prepad + len < PAGE_SIZE || !linear)
 526                 linear = len;
 527
 528         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 529                                    &err);
 530         if (!skb)
 531                 return ERR_PTR(err);
 532
 533         skb_reserve(skb, prepad);
 534         skb_put(skb, linear);
 535         skb->data_len = len - linear;
 536         skb->len += len - linear;
 537
 538         return skb;
 539 }
 540
 541 /* Get packet from user space buffer */
 542 static __inline__ ssize_t tun_get_user(struct tun_struct *tun,
 543                                        const struct iovec *iv, size_t count,
 544                                        int noblock)
 545 {
 546         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
 547         struct sk_buff *skb;
 548         size_t len = count, align = 0;
 549         struct virtio_net_hdr gso = { 0 };
 550         int offset = 0;
 551
 552         if (!(tun->flags & TUN_NO_PI)) {
 553                 if ((len -= sizeof(pi)) > count)
 554                         return -EINVAL;
 555
 556                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
 557                         return -EFAULT;
 558                 offset += sizeof(pi);
 559         }
 560
 561         if (tun->flags & TUN_VNET_HDR) {
 562                 if ((len -= sizeof(gso)) > count)
 563                         return -EINVAL;
 564
 565                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
 566                         return -EFAULT;
 567
 568                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
 569                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
 570                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
 571
 572                 if (gso.hdr_len > len)
 573                         return -EINVAL;
 574                 offset += sizeof(gso);
 575         }
 576
 577         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
 578                 align = NET_IP_ALIGN;
 579                 if (unlikely(len < ETH_HLEN ||
 580                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
 581                         return -EINVAL;
 582         }
 583
 584         skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock);
 585         if (IS_ERR(skb)) {
 586                 if (PTR_ERR(skb) != -EAGAIN)
 587                         tun->dev->stats.rx_dropped++;
 588                 return PTR_ERR(skb);
 589         }
 590
 591         if (skb_copy_datagram_from_iovec(skb, 0, iv, offset, len)) {
 592                 tun->dev->stats.rx_dropped++;
 593                 kfree_skb(skb);
 594                 return -EFAULT;
 595         }
 596
 597         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 598                 if (!skb_partial_csum_set(skb, gso.csum_start,
 599                                           gso.csum_offset)) {
 600                         tun->dev->stats.rx_frame_errors++;
 601                         kfree_skb(skb);
 602                         return -EINVAL;
 603                 }
 604         } else if (tun->flags & TUN_NOCHECKSUM)
 605                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 606
 607         switch (tun->flags & TUN_TYPE_MASK) {
 608         case TUN_TUN_DEV:
 609                 if (tun->flags & TUN_NO_PI) {
 610                         switch (skb->data[0] & 0xf0) {
 611                         case 0x40:
 612                                 pi.proto = htons(ETH_P_IP);
 613                                 break;
 614                         case 0x60:
 615                                 pi.proto = htons(ETH_P_IPV6);
 616                                 break;
 617                         default:
 618                                 tun->dev->stats.rx_dropped++;
 619                                 kfree_skb(skb);
 620                                 return -EINVAL;
 621                         }
 622                 }
 623
 624                 skb_reset_mac_header(skb);
 625                 skb->protocol = pi.proto;
 626                 skb->dev = tun->dev;
 627                 break;
 628         case TUN_TAP_DEV:
 629                 skb->protocol = eth_type_trans(skb, tun->dev);
 630                 break;
 631         };
 632
 633         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 634                 pr_debug("GSO!\n");
 635                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 636                 case VIRTIO_NET_HDR_GSO_TCPV4:
 637                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 638                         break;
 639                 case VIRTIO_NET_HDR_GSO_TCPV6:
 640                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 641                         break;
 642                 case VIRTIO_NET_HDR_GSO_UDP:
 643                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 644                         break;
 645                 default:
 646                         tun->dev->stats.rx_frame_errors++;
 647                         kfree_skb(skb);
 648                         return -EINVAL;
 649                 }
 650
 651                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
 652                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 653
 654                 skb_shinfo(skb)->gso_size = gso.gso_size;
 655                 if (skb_shinfo(skb)->gso_size == 0) {
 656                         tun->dev->stats.rx_frame_errors++;
 657                         kfree_skb(skb);
 658                         return -EINVAL;
 659                 }
 660
 661                 /* Header must be checked, and gso_segs computed. */
 662                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 663                 skb_shinfo(skb)->gso_segs = 0;
 664         }
 665
 666         netif_rx_ni(skb);
 667
 668         tun->dev->stats.rx_packets++;
 669         tun->dev->stats.rx_bytes += len;
 670
 671         return count;
 672 }
 673
 674 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
 675                               unsigned long count, loff_t pos)
 676 {
 677         struct file *file = iocb->ki_filp;
 678         struct tun_struct *tun = tun_get(file);
 679         ssize_t result;
 680
 681         if (!tun)
 682                 return -EBADFD;
 683
 684         DBG(KERN_INFO "%s: tun_chr_write %ld\n", tun->dev->name, count);
 685
 686         result = tun_get_user(tun, iv, iov_length(iv, count),
 687                               file->f_flags & O_NONBLOCK);
 688
 689         tun_put(tun);
 690         return result;
 691 }
 692
 693 /* Put packet to the user space buffer */
 694 static __inline__ ssize_t tun_put_user(struct tun_struct *tun,
 695                                        struct sk_buff *skb,
 696                                        const struct iovec *iv, int len)
 697 {
 698         struct tun_pi pi = { 0, skb->protocol };
 699         ssize_t total = 0;
 700
 701         if (!(tun->flags & TUN_NO_PI)) {
 702                 if ((len -= sizeof(pi)) < 0)
 703                         return -EINVAL;
 704
 705                 if (len < skb->len) {
 706                         /* Packet will be striped */
 707                         pi.flags |= TUN_PKT_STRIP;
 708                 }
 709
 710                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
 711                         return -EFAULT;
 712                 total += sizeof(pi);
 713         }
 714
 715         if (tun->flags & TUN_VNET_HDR) {
 716                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
 717                 if ((len -= sizeof(gso)) < 0)
 718                         return -EINVAL;
 719
 720                 if (skb_is_gso(skb)) {
 721                         struct skb_shared_info *sinfo = skb_shinfo(skb);
 722
 723                         /* This is a hint as to how much should be linear. */
 724                         gso.hdr_len = skb_headlen(skb);
 725                         gso.gso_size = sinfo->gso_size;
 726                         if (sinfo->gso_type & SKB_GSO_TCPV4)
 727                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 728                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
 729                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 730                         else if (sinfo->gso_type & SKB_GSO_UDP)
 731                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
 732                         else
 733                                 BUG();
 734                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
 735                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 736                 } else
 737                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
 738
 739                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 740                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 741                         gso.csum_start = skb->csum_start - skb_headroom(skb);
 742                         gso.csum_offset = skb->csum_offset;
 743                 } /* else everything is zero */
 744
 745                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
 746                                                sizeof(gso))))
 747                         return -EFAULT;
 748                 total += sizeof(gso);
 749         }
 750
 751         len = min_t(int, skb->len, len);
 752
 753         skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
 754         total += skb->len;
 755
 756         tun->dev->stats.tx_packets++;
 757         tun->dev->stats.tx_bytes += len;
 758
 759         return total;
 760 }
 761
 762 static ssize_t tun_do_read(struct tun_struct *tun,
 763                            struct kiocb *iocb, const struct iovec *iv,
 764                            ssize_t len, int noblock)
 765 {
 766         DECLARE_WAITQUEUE(wait, current);
 767         struct sk_buff *skb;
 768         ssize_t ret = 0;
 769
 770         DBG(KERN_INFO "%s: tun_chr_read\n", tun->dev->name);
 771
 772         add_wait_queue(&tun->socket.wait, &wait);
 773         while (len) {
 774                 current->state = TASK_INTERRUPTIBLE;
 775
 776                 /* Read frames from the queue */
 777                 if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
 778                         if (noblock) {
 779                                 ret = -EAGAIN;
 780                                 break;
 781                         }
 782                         if (signal_pending(current)) {
 783                                 ret = -ERESTARTSYS;
 784                                 break;
 785                         }
 786                         if (tun->dev->reg_state != NETREG_REGISTERED) {
 787                                 ret = -EIO;
 788                                 break;
 789                         }
 790
 791                         /* Nothing to read, let's sleep */
 792                         schedule();
 793                         continue;
 794                 }
 795                 netif_wake_queue(tun->dev);
 796
 797                 ret = tun_put_user(tun, skb, iv, len);
 798                 kfree_skb(skb);
 799                 break;
 800         }
 801
 802         current->state = TASK_RUNNING;
 803         remove_wait_queue(&tun->socket.wait, &wait);
 804
 805         return ret;
 806 }
 807
 808 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
 809                             unsigned long count, loff_t pos)
 810 {
 811         struct file *file = iocb->ki_filp;
 812         struct tun_file *tfile = file->private_data;
 813         struct tun_struct *tun = __tun_get(tfile);
 814         ssize_t len, ret;
 815
 816         if (!tun)
 817                 return -EBADFD;
 818         len = iov_length(iv, count);
 819         if (len < 0) {
 820                 ret = -EINVAL;
 821                 goto out;
 822         }
 823
 824         ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK);
 825         ret = min_t(ssize_t, ret, len);
 826 out:
 827         tun_put(tun);
 828         return ret;
 829 }
 830
 831 static void tun_setup(struct net_device *dev)
 832 {
 833         struct tun_struct *tun = netdev_priv(dev);
 834
 835         tun->owner = -1;
 836         tun->group = -1;
 837
 838         dev->ethtool_ops = &tun_ethtool_ops;
 839         dev->destructor = tun_free_netdev;
 840 }
 841
 842 /* Trivial set of netlink ops to allow deleting tun or tap
 843  * device with netlink.
 844  */
 845 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
 846 {
 847         return -EINVAL;
 848 }
 849
 850 static struct rtnl_link_ops tun_link_ops __read_mostly = {
 851         .kind           = DRV_NAME,
 852         .priv_size      = sizeof(struct tun_struct),
 853         .setup          = tun_setup,
 854         .validate       = tun_validate,
 855 };
 856
 857 static void tun_sock_write_space(struct sock *sk)
 858 {
 859         struct tun_struct *tun;
 860
 861         if (!sock_writeable(sk))
 862                 return;
 863
 864         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
 865                 return;
 866
 867         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
 868                 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
 869                                                 POLLWRNORM | POLLWRBAND);
 870
 871         tun = tun_sk(sk)->tun;
 872         kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
 873 }
 874
 875 static void tun_sock_destruct(struct sock *sk)
 876 {
 877         free_netdev(tun_sk(sk)->tun->dev);
 878 }
 879
 880 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
 881                        struct msghdr *m, size_t total_len)
 882 {
 883         struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
 884         return tun_get_user(tun, m->msg_iov, total_len,
 885                             m->msg_flags & MSG_DONTWAIT);
 886 }
 887
 888 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
 889                        struct msghdr *m, size_t total_len,
 890                        int flags)
 891 {
 892         struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
 893         int ret;
 894         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
 895                 return -EINVAL;
 896         ret = tun_do_read(tun, iocb, m->msg_iov, total_len,
 897                           flags & MSG_DONTWAIT);
 898         if (ret > total_len) {
 899                 m->msg_flags |= MSG_TRUNC;
 900                 ret = flags & MSG_TRUNC ? ret : total_len;
 901         }
 902         return ret;
 903 }
 904
 905 /* Ops structure to mimic raw sockets with tun */
 906 static const struct proto_ops tun_socket_ops = {
 907         .sendmsg = tun_sendmsg,
 908         .recvmsg = tun_recvmsg,
 909 };
 910
 911 static struct proto tun_proto = {
 912         .name           = "tun",
 913         .owner          = THIS_MODULE,
 914         .obj_size       = sizeof(struct tun_sock),
 915 };
 916
 917 static int tun_flags(struct tun_struct *tun)
 918 {
 919         int flags = 0;
 920
 921         if (tun->flags & TUN_TUN_DEV)
 922                 flags |= IFF_TUN;
 923         else
 924                 flags |= IFF_TAP;
 925
 926         if (tun->flags & TUN_NO_PI)
 927                 flags |= IFF_NO_PI;
 928
 929         if (tun->flags & TUN_ONE_QUEUE)
 930                 flags |= IFF_ONE_QUEUE;
 931
 932         if (tun->flags & TUN_VNET_HDR)
 933                 flags |= IFF_VNET_HDR;
 934
 935         return flags;
 936 }
 937
 938 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
 939                               char *buf)
 940 {
 941         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
 942         return sprintf(buf, "0x%x\n", tun_flags(tun));
 943 }
 944
 945 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
 946                               char *buf)
 947 {
 948         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
 949         return sprintf(buf, "%d\n", tun->owner);
 950 }
 951
 952 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
 953                               char *buf)
 954 {
 955         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
 956         return sprintf(buf, "%d\n", tun->group);
 957 }
 958
 959 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
 960 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
 961 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
 962
 963 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
 964 {
 965         struct sock *sk;
 966         struct tun_struct *tun;
 967         struct net_device *dev;
 968         int err;
 969
 970         dev = __dev_get_by_name(net, ifr->ifr_name);
 971         if (dev) {
 972                 const struct cred *cred = current_cred();
 973
 974                 if (ifr->ifr_flags & IFF_TUN_EXCL)
 975                         return -EBUSY;
 976                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
 977                         tun = netdev_priv(dev);
 978                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
 979                         tun = netdev_priv(dev);
 980                 else
 981                         return -EINVAL;
 982
 983                 if (((tun->owner != -1 && cred->euid != tun->owner) ||
 984                      (tun->group != -1 && !in_egroup_p(tun->group))) &&
 985                     !capable(CAP_NET_ADMIN))
 986                         return -EPERM;
 987                 err = security_tun_dev_attach(tun->socket.sk);
 988                 if (err < 0)
 989                         return err;
 990
 991                 err = tun_attach(tun, file);
 992                 if (err < 0)
 993                         return err;
 994         }
 995         else {
 996                 char *name;
 997                 unsigned long flags = 0;
 998
 999                 if (!capable(CAP_NET_ADMIN))
1000                         return -EPERM;
1001                 err = security_tun_dev_create();
1002                 if (err < 0)
1003                         return err;
1004
1005                 /* Set dev type */
1006                 if (ifr->ifr_flags & IFF_TUN) {
1007                         /* TUN device */
1008                         flags |= TUN_TUN_DEV;
1009                         name = "tun%d";
1010                 } else if (ifr->ifr_flags & IFF_TAP) {
1011                         /* TAP device */
1012                         flags |= TUN_TAP_DEV;
1013                         name = "tap%d";
1014                 } else
1015                         return -EINVAL;
1016
1017                 if (*ifr->ifr_name)
1018                         name = ifr->ifr_name;
1019
1020                 dev = alloc_netdev(sizeof(struct tun_struct), name,
1021                                    tun_setup);
1022                 if (!dev)
1023                         return -ENOMEM;
1024
1025                 dev_net_set(dev, net);
1026                 dev->rtnl_link_ops = &tun_link_ops;
1027
1028                 tun = netdev_priv(dev);
1029                 tun->dev = dev;
1030                 tun->flags = flags;
1031                 tun->txflt.count = 0;
1032
1033                 err = -ENOMEM;
1034                 sk = sk_alloc(net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1035                 if (!sk)
1036                         goto err_free_dev;
1037
1038                 init_waitqueue_head(&tun->socket.wait);
1039                 tun->socket.ops = &tun_socket_ops;
1040                 sock_init_data(&tun->socket, sk);
1041                 sk->sk_write_space = tun_sock_write_space;
1042                 sk->sk_sndbuf = INT_MAX;
1043
1044                 tun_sk(sk)->tun = tun;
1045
1046                 security_tun_dev_post_create(sk);
1047
1048                 tun_net_init(dev);
1049
1050                 if (strchr(dev->name, '%')) {
1051                         err = dev_alloc_name(dev, dev->name);
1052                         if (err < 0)
1053                                 goto err_free_sk;
1054                 }
1055
1056                 err = register_netdevice(tun->dev);
1057                 if (err < 0)
1058                         goto err_free_sk;
1059
1060                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1061                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1062                     device_create_file(&tun->dev->dev, &dev_attr_group))
1063                         printk(KERN_ERR "Failed to create tun sysfs files\n");
1064
1065                 sk->sk_destruct = tun_sock_destruct;
1066
1067                 err = tun_attach(tun, file);
1068                 if (err < 0)
1069                         goto failed;
1070         }
1071
1072         DBG(KERN_INFO "%s: tun_set_iff\n", tun->dev->name);
1073
1074         if (ifr->ifr_flags & IFF_NO_PI)
1075                 tun->flags |= TUN_NO_PI;
1076         else
1077                 tun->flags &= ~TUN_NO_PI;
1078
1079         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1080                 tun->flags |= TUN_ONE_QUEUE;
1081         else
1082                 tun->flags &= ~TUN_ONE_QUEUE;
1083
1084         if (ifr->ifr_flags & IFF_VNET_HDR)
1085                 tun->flags |= TUN_VNET_HDR;
1086         else
1087                 tun->flags &= ~TUN_VNET_HDR;
1088
1089         /* Make sure persistent devices do not get stuck in
1090          * xoff state.
1091          */
1092         if (netif_running(tun->dev))
1093                 netif_wake_queue(tun->dev);
1094
1095         strcpy(ifr->ifr_name, tun->dev->name);
1096         return 0;
1097
1098  err_free_sk:
1099         sock_put(sk);
1100  err_free_dev:
1101         free_netdev(dev);
1102  failed:
1103         return err;
1104 }
1105
1106 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1107                        struct ifreq *ifr)
1108 {
1109         DBG(KERN_INFO "%s: tun_get_iff\n", tun->dev->name);
1110
1111         strcpy(ifr->ifr_name, tun->dev->name);
1112
1113         ifr->ifr_flags = tun_flags(tun);
1114
1115         return 0;
1116 }
1117
1118 /* This is like a cut-down ethtool ops, except done via tun fd so no
1119  * privs required. */
1120 static int set_offload(struct net_device *dev, unsigned long arg)
1121 {
1122         unsigned int old_features, features;
1123
1124         old_features = dev->features;
1125         /* Unset features, set them as we chew on the arg. */
1126         features = (old_features & ~(NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST
1127                                     |NETIF_F_TSO_ECN|NETIF_F_TSO|NETIF_F_TSO6
1128                                     |NETIF_F_UFO));
1129
1130         if (arg & TUN_F_CSUM) {
1131                 features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1132                 arg &= ~TUN_F_CSUM;
1133
1134                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1135                         if (arg & TUN_F_TSO_ECN) {
1136                                 features |= NETIF_F_TSO_ECN;
1137                                 arg &= ~TUN_F_TSO_ECN;
1138                         }
1139                         if (arg & TUN_F_TSO4)
1140                                 features |= NETIF_F_TSO;
1141                         if (arg & TUN_F_TSO6)
1142                                 features |= NETIF_F_TSO6;
1143                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1144                 }
1145
1146                 if (arg & TUN_F_UFO) {
1147                         features |= NETIF_F_UFO;
1148                         arg &= ~TUN_F_UFO;
1149                 }
1150         }
1151
1152         /* This gives the user a way to test for new features in future by
1153          * trying to set them. */
1154         if (arg)
1155                 return -EINVAL;
1156
1157         dev->features = features;
1158         if (old_features != dev->features)
1159                 netdev_features_change(dev);
1160
1161         return 0;
1162 }
1163
1164 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1165                             unsigned long arg, int ifreq_len)
1166 {
1167         struct tun_file *tfile = file->private_data;
1168         struct tun_struct *tun;
1169         void __user* argp = (void __user*)arg;
1170         struct sock_fprog fprog;
1171         struct ifreq ifr;
1172         int sndbuf;
1173         int ret;
1174
1175         if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
1176                 if (copy_from_user(&ifr, argp, ifreq_len))
1177                         return -EFAULT;
1178
1179         if (cmd == TUNGETFEATURES) {
1180                 /* Currently this just means: "what IFF flags are valid?".
1181                  * This is needed because we never checked for invalid flags on
1182                  * TUNSETIFF. */
1183                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1184                                 IFF_VNET_HDR,
1185                                 (unsigned int __user*)argp);
1186         }
1187
1188         rtnl_lock();
1189
1190         tun = __tun_get(tfile);
1191         if (cmd == TUNSETIFF && !tun) {
1192                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1193
1194                 ret = tun_set_iff(tfile->net, file, &ifr);
1195
1196                 if (ret)
1197                         goto unlock;
1198
1199                 if (copy_to_user(argp, &ifr, ifreq_len))
1200                         ret = -EFAULT;
1201                 goto unlock;
1202         }
1203
1204         ret = -EBADFD;
1205         if (!tun)
1206                 goto unlock;
1207
1208         DBG(KERN_INFO "%s: tun_chr_ioctl cmd %d\n", tun->dev->name, cmd);
1209
1210         ret = 0;
1211         switch (cmd) {
1212         case TUNGETIFF:
1213                 ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1214                 if (ret)
1215                         break;
1216
1217                 if (copy_to_user(argp, &ifr, ifreq_len))
1218                         ret = -EFAULT;
1219                 break;
1220
1221         case TUNSETNOCSUM:
1222                 /* Disable/Enable checksum */
1223                 if (arg)
1224                         tun->flags |= TUN_NOCHECKSUM;
1225                 else
1226                         tun->flags &= ~TUN_NOCHECKSUM;
1227
1228                 DBG(KERN_INFO "%s: checksum %s\n",
1229                     tun->dev->name, arg ? "disabled" : "enabled");
1230                 break;
1231
1232         case TUNSETPERSIST:
1233                 /* Disable/Enable persist mode */
1234                 if (arg)
1235                         tun->flags |= TUN_PERSIST;
1236                 else
1237                         tun->flags &= ~TUN_PERSIST;
1238
1239                 DBG(KERN_INFO "%s: persist %s\n",
1240                     tun->dev->name, arg ? "enabled" : "disabled");
1241                 break;
1242
1243         case TUNSETOWNER:
1244                 /* Set owner of the device */
1245                 tun->owner = (uid_t) arg;
1246
1247                 DBG(KERN_INFO "%s: owner set to %d\n", tun->dev->name, tun->owner);
1248                 break;
1249
1250         case TUNSETGROUP:
1251                 /* Set group of the device */
1252                 tun->group= (gid_t) arg;
1253
1254                 DBG(KERN_INFO "%s: group set to %d\n", tun->dev->name, tun->group);
1255                 break;
1256
1257         case TUNSETLINK:
1258                 /* Only allow setting the type when the interface is down */
1259                 if (tun->dev->flags & IFF_UP) {
1260                         DBG(KERN_INFO "%s: Linktype set failed because interface is up\n",
1261                                 tun->dev->name);
1262                         ret = -EBUSY;
1263                 } else {
1264                         tun->dev->type = (int) arg;
1265                         DBG(KERN_INFO "%s: linktype set to %d\n", tun->dev->name, tun->dev->type);
1266                         ret = 0;
1267                 }
1268                 break;
1269
1270 #ifdef TUN_DEBUG
1271         case TUNSETDEBUG:
1272                 tun->debug = arg;
1273                 break;
1274 #endif
1275         case TUNSETOFFLOAD:
1276                 ret = set_offload(tun->dev, arg);
1277                 break;
1278
1279         case TUNSETTXFILTER:
1280                 /* Can be set only for TAPs */
1281                 ret = -EINVAL;
1282                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1283                         break;
1284                 ret = update_filter(&tun->txflt, (void __user *)arg);
1285                 break;
1286
1287         case SIOCGIFHWADDR:
1288                 /* Get hw addres */
1289                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1290                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1291                 if (copy_to_user(argp, &ifr, ifreq_len))
1292                         ret = -EFAULT;
1293                 break;
1294
1295         case SIOCSIFHWADDR:
1296                 /* Set hw address */
1297                 DBG(KERN_DEBUG "%s: set hw address: %pM\n",
1298                         tun->dev->name, ifr.ifr_hwaddr.sa_data);
1299
1300                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1301                 break;
1302
1303         case TUNGETSNDBUF:
1304                 sndbuf = tun->socket.sk->sk_sndbuf;
1305                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1306                         ret = -EFAULT;
1307                 break;
1308
1309         case TUNSETSNDBUF:
1310                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1311                         ret = -EFAULT;
1312                         break;
1313                 }
1314
1315                 tun->socket.sk->sk_sndbuf = sndbuf;
1316                 break;
1317
1318         case TUNATTACHFILTER:
1319                 /* Can be set only for TAPs */
1320                 ret = -EINVAL;
1321                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1322                         break;
1323                 ret = -EFAULT;
1324                 if (copy_from_user(&fprog, argp, sizeof(fprog)))
1325                         break;
1326
1327                 ret = sk_attach_filter(&fprog, tun->socket.sk);
1328                 break;
1329
1330         case TUNDETACHFILTER:
1331                 /* Can be set only for TAPs */
1332                 ret = -EINVAL;
1333                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1334                         break;
1335                 ret = sk_detach_filter(tun->socket.sk);
1336                 break;
1337
1338         default:
1339                 ret = -EINVAL;
1340                 break;
1341         };
1342
1343 unlock:
1344         rtnl_unlock();
1345         if (tun)
1346                 tun_put(tun);
1347         return ret;
1348 }
1349
1350 static long tun_chr_ioctl(struct file *file,
1351                           unsigned int cmd, unsigned long arg)
1352 {
1353         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1354 }
1355
1356 #ifdef CONFIG_COMPAT
1357 static long tun_chr_compat_ioctl(struct file *file,
1358                          unsigned int cmd, unsigned long arg)
1359 {
1360         switch (cmd) {
1361         case TUNSETIFF:
1362         case TUNGETIFF:
1363         case TUNSETTXFILTER:
1364         case TUNGETSNDBUF:
1365         case TUNSETSNDBUF:
1366         case SIOCGIFHWADDR:
1367         case SIOCSIFHWADDR:
1368                 arg = (unsigned long)compat_ptr(arg);
1369                 break;
1370         default:
1371                 arg = (compat_ulong_t)arg;
1372                 break;
1373         }
1374
1375         /*
1376          * compat_ifreq is shorter than ifreq, so we must not access beyond
1377          * the end of that structure. All fields that are used in this
1378          * driver are compatible though, we don't need to convert the
1379          * contents.
1380          */
1381         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1382 }
1383 #endif /* CONFIG_COMPAT */
1384
1385 static int tun_chr_fasync(int fd, struct file *file, int on)
1386 {
1387         struct tun_struct *tun = tun_get(file);
1388         int ret;
1389
1390         if (!tun)
1391                 return -EBADFD;
1392
1393         DBG(KERN_INFO "%s: tun_chr_fasync %d\n", tun->dev->name, on);
1394
1395         if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1396                 goto out;
1397
1398         if (on) {
1399                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1400                 if (ret)
1401                         goto out;
1402                 tun->flags |= TUN_FASYNC;
1403         } else
1404                 tun->flags &= ~TUN_FASYNC;
1405         ret = 0;
1406 out:
1407         tun_put(tun);
1408         return ret;
1409 }
1410
1411 static int tun_chr_open(struct inode *inode, struct file * file)
1412 {
1413         struct tun_file *tfile;
1414
1415         DBG1(KERN_INFO "tunX: tun_chr_open\n");
1416
1417         tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1418         if (!tfile)
1419                 return -ENOMEM;
1420         atomic_set(&tfile->count, 0);
1421         tfile->tun = NULL;
1422         tfile->net = get_net(current->nsproxy->net_ns);
1423         file->private_data = tfile;
1424         return 0;
1425 }
1426
1427 static int tun_chr_close(struct inode *inode, struct file *file)
1428 {
1429         struct tun_file *tfile = file->private_data;
1430         struct tun_struct *tun;
1431
1432         tun = __tun_get(tfile);
1433         if (tun) {
1434                 struct net_device *dev = tun->dev;
1435
1436                 DBG(KERN_INFO "%s: tun_chr_close\n", dev->name);
1437
1438                 __tun_detach(tun);
1439
1440                 /* If desireable, unregister the netdevice. */
1441                 if (!(tun->flags & TUN_PERSIST)) {
1442                         rtnl_lock();
1443                         if (dev->reg_state == NETREG_REGISTERED)
1444                                 unregister_netdevice(dev);
1445                         rtnl_unlock();
1446                 }
1447         }
1448
1449         tun = tfile->tun;
1450         if (tun)
1451                 sock_put(tun->socket.sk);
1452
1453         put_net(tfile->net);
1454         kfree(tfile);
1455
1456         return 0;
1457 }
1458
1459 static const struct file_operations tun_fops = {
1460         .owner  = THIS_MODULE,
1461         .llseek = no_llseek,
1462         .read  = do_sync_read,
1463         .aio_read  = tun_chr_aio_read,
1464         .write = do_sync_write,
1465         .aio_write = tun_chr_aio_write,
1466         .poll   = tun_chr_poll,
1467         .unlocked_ioctl = tun_chr_ioctl,
1468 #ifdef CONFIG_COMPAT
1469         .compat_ioctl = tun_chr_compat_ioctl,
1470 #endif
1471         .open   = tun_chr_open,
1472         .release = tun_chr_close,
1473         .fasync = tun_chr_fasync
1474 };
1475
1476 static struct miscdevice tun_miscdev = {
1477         .minor = TUN_MINOR,
1478         .name = "tun",
1479         .nodename = "net/tun",
1480         .fops = &tun_fops,
1481 };
1482
1483 /* ethtool interface */
1484
1485 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1486 {
1487         cmd->supported          = 0;
1488         cmd->advertising        = 0;
1489         cmd->speed              = SPEED_10;
1490         cmd->duplex             = DUPLEX_FULL;
1491         cmd->port               = PORT_TP;
1492         cmd->phy_address        = 0;
1493         cmd->transceiver        = XCVR_INTERNAL;
1494         cmd->autoneg            = AUTONEG_DISABLE;
1495         cmd->maxtxpkt           = 0;
1496         cmd->maxrxpkt           = 0;
1497         return 0;
1498 }
1499
1500 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1501 {
1502         struct tun_struct *tun = netdev_priv(dev);
1503
1504         strcpy(info->driver, DRV_NAME);
1505         strcpy(info->version, DRV_VERSION);
1506         strcpy(info->fw_version, "N/A");
1507
1508         switch (tun->flags & TUN_TYPE_MASK) {
1509         case TUN_TUN_DEV:
1510                 strcpy(info->bus_info, "tun");
1511                 break;
1512         case TUN_TAP_DEV:
1513                 strcpy(info->bus_info, "tap");
1514                 break;
1515         }
1516 }
1517
1518 static u32 tun_get_msglevel(struct net_device *dev)
1519 {
1520 #ifdef TUN_DEBUG
1521         struct tun_struct *tun = netdev_priv(dev);
1522         return tun->debug;
1523 #else
1524         return -EOPNOTSUPP;
1525 #endif
1526 }
1527
1528 static void tun_set_msglevel(struct net_device *dev, u32 value)
1529 {
1530 #ifdef TUN_DEBUG
1531         struct tun_struct *tun = netdev_priv(dev);
1532         tun->debug = value;
1533 #endif
1534 }
1535
1536 static u32 tun_get_link(struct net_device *dev)
1537 {
1538         struct tun_struct *tun = netdev_priv(dev);
1539         return !!tun->tfile;
1540 }
1541
1542 static u32 tun_get_rx_csum(struct net_device *dev)
1543 {
1544         struct tun_struct *tun = netdev_priv(dev);
1545         return (tun->flags & TUN_NOCHECKSUM) == 0;
1546 }
1547
1548 static int tun_set_rx_csum(struct net_device *dev, u32 data)
1549 {
1550         struct tun_struct *tun = netdev_priv(dev);
1551         if (data)
1552                 tun->flags &= ~TUN_NOCHECKSUM;
1553         else
1554                 tun->flags |= TUN_NOCHECKSUM;
1555         return 0;
1556 }
1557
1558 static const struct ethtool_ops tun_ethtool_ops = {
1559         .get_settings   = tun_get_settings,
1560         .get_drvinfo    = tun_get_drvinfo,
1561         .get_msglevel   = tun_get_msglevel,
1562         .set_msglevel   = tun_set_msglevel,
1563         .get_link       = tun_get_link,
1564         .get_rx_csum    = tun_get_rx_csum,
1565         .set_rx_csum    = tun_set_rx_csum
1566 };
1567
1568
1569 static int __init tun_init(void)
1570 {
1571         int ret = 0;
1572
1573         printk(KERN_INFO "tun: %s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1574         printk(KERN_INFO "tun: %s\n", DRV_COPYRIGHT);
1575
1576         ret = rtnl_link_register(&tun_link_ops);
1577         if (ret) {
1578                 printk(KERN_ERR "tun: Can't register link_ops\n");
1579                 goto err_linkops;
1580         }
1581
1582         ret = misc_register(&tun_miscdev);
1583         if (ret) {
1584                 printk(KERN_ERR "tun: Can't register misc device %d\n", TUN_MINOR);
1585                 goto err_misc;
1586         }
1587         return  0;
1588 err_misc:
1589         rtnl_link_unregister(&tun_link_ops);
1590 err_linkops:
1591         return ret;
1592 }
1593
1594 static void tun_cleanup(void)
1595 {
1596         misc_deregister(&tun_miscdev);
1597         rtnl_link_unregister(&tun_link_ops);
1598 }
1599
1600 /* Get an underlying socket object from tun file.  Returns error unless file is
1601  * attached to a device.  The returned object works like a packet socket, it
1602  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1603  * holding a reference to the file for as long as the socket is in use. */
1604 struct socket *tun_get_socket(struct file *file)
1605 {
1606         struct tun_struct *tun;
1607         if (file->f_op != &tun_fops)
1608                 return ERR_PTR(-EINVAL);
1609         tun = tun_get(file);
1610         if (!tun)
1611                 return ERR_PTR(-EBADFD);
1612         tun_put(tun);
1613         return &tun->socket;
1614 }
1615 EXPORT_SYMBOL_GPL(tun_get_socket);
1616
1617 module_init(tun_init);
1618 module_exit(tun_cleanup);
1619 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1620 MODULE_AUTHOR(DRV_COPYRIGHT);
1621 MODULE_LICENSE("GPL");
1622 MODULE_ALIAS_MISCDEV(TUN_MINOR);