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