drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use random_ether_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define DRV_NAME        "tun"
  38 #define DRV_VERSION     "1.6"
  39 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  40 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  41
  42 #include <linux/module.h>
  43 #include <linux/errno.h>
  44 #include <linux/kernel.h>
  45 #include <linux/major.h>
  46 #include <linux/slab.h>
  47 #include <linux/poll.h>
  48 #include <linux/fcntl.h>
  49 #include <linux/init.h>
  50 #include <linux/skbuff.h>
  51 #include <linux/netdevice.h>
  52 #include <linux/etherdevice.h>
  53 #include <linux/miscdevice.h>
  54 #include <linux/ethtool.h>
  55 #include <linux/rtnetlink.h>
  56 #include <linux/compat.h>
  57 #include <linux/if.h>
  58 #include <linux/if_arp.h>
  59 #include <linux/if_ether.h>
  60 #include <linux/if_tun.h>
  61 #include <linux/crc32.h>
  62 #include <linux/nsproxy.h>
  63 #include <linux/virtio_net.h>
  64 #include <linux/rcupdate.h>
  65 #include <net/net_namespace.h>
  66 #include <net/netns/generic.h>
  67 #include <net/rtnetlink.h>
  68 #include <net/sock.h>
  69
  70 #include <asm/system.h>
  71 #include <asm/uaccess.h>
  72
  73 /* Uncomment to enable debugging */
  74 /* #define TUN_DEBUG 1 */
  75
  76 #ifdef TUN_DEBUG
  77 static int debug;
  78
  79 #define DBG  if(tun->debug)printk
  80 #define DBG1 if(debug==2)printk
  81 #else
  82 #define DBG( a... )
  83 #define DBG1( a... )
  84 #endif
  85
  86 #define FLT_EXACT_COUNT 8
  87 struct tap_filter {
  88         unsigned int    count;    /* Number of addrs. Zero means disabled */
  89         u32             mask[2];  /* Mask of the hashed addrs */
  90         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
  91 };
  92
  93 struct tun_file {
  94         atomic_t count;
  95         struct tun_struct *tun;
  96         struct net *net;
  97 };
  98
  99 struct tun_sock;
 100
 101 struct tun_struct {
 102         struct tun_file         *tfile;
 103         unsigned int            flags;
 104         uid_t                   owner;
 105         gid_t                   group;
 106
 107         struct net_device       *dev;
 108         struct fasync_struct    *fasync;
 109
 110         struct tap_filter       txflt;
 111         struct socket           socket;
 112         struct socket_wq        wq;
 113
 114         int                     vnet_hdr_sz;
 115
 116 #ifdef TUN_DEBUG
 117         int debug;
 118 #endif
 119 };
 120
 121 struct tun_sock {
 122         struct sock             sk;
 123         struct tun_struct       *tun;
 124 };
 125
 126 static inline struct tun_sock *tun_sk(struct sock *sk)
 127 {
 128         return container_of(sk, struct tun_sock, sk);
 129 }
 130
 131 static int tun_attach(struct tun_struct *tun, struct file *file)
 132 {
 133         struct tun_file *tfile = file->private_data;
 134         int err;
 135
 136         ASSERT_RTNL();
 137
 138         netif_tx_lock_bh(tun->dev);
 139
 140         err = -EINVAL;
 141         if (tfile->tun)
 142                 goto out;
 143
 144         err = -EBUSY;
 145         if (tun->tfile)
 146                 goto out;
 147
 148         err = 0;
 149         tfile->tun = tun;
 150         tun->tfile = tfile;
 151         tun->socket.file = file;
 152         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         /* Under a page?  Don't bother with paged skb. */
 530         if (prepad + len < PAGE_SIZE || !linear)
 531                 linear = len;
 532
 533         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 534                                    &err);
 535         if (!skb)
 536                 return ERR_PTR(err);
 537
 538         skb_reserve(skb, prepad);
 539         skb_put(skb, linear);
 540         skb->data_len = len - linear;
 541         skb->len += len - linear;
 542
 543         return skb;
 544 }
 545
 546 /* Get packet from user space buffer */
 547 static __inline__ ssize_t tun_get_user(struct tun_struct *tun,
 548                                        const struct iovec *iv, size_t count,
 549                                        int noblock)
 550 {
 551         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
 552         struct sk_buff *skb;
 553         size_t len = count, align = 0;
 554         struct virtio_net_hdr gso = { 0 };
 555         int offset = 0;
 556
 557         if (!(tun->flags & TUN_NO_PI)) {
 558                 if ((len -= sizeof(pi)) > count)
 559                         return -EINVAL;
 560
 561                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
 562                         return -EFAULT;
 563                 offset += sizeof(pi);
 564         }
 565
 566         if (tun->flags & TUN_VNET_HDR) {
 567                 if ((len -= tun->vnet_hdr_sz) > count)
 568                         return -EINVAL;
 569
 570                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
 571                         return -EFAULT;
 572
 573                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
 574                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
 575                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
 576
 577                 if (gso.hdr_len > len)
 578                         return -EINVAL;
 579                 offset += tun->vnet_hdr_sz;
 580         }
 581
 582         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
 583                 align = NET_IP_ALIGN;
 584                 if (unlikely(len < ETH_HLEN ||
 585                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
 586                         return -EINVAL;
 587         }
 588
 589         skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock);
 590         if (IS_ERR(skb)) {
 591                 if (PTR_ERR(skb) != -EAGAIN)
 592                         tun->dev->stats.rx_dropped++;
 593                 return PTR_ERR(skb);
 594         }
 595
 596         if (skb_copy_datagram_from_iovec(skb, 0, iv, offset, len)) {
 597                 tun->dev->stats.rx_dropped++;
 598                 kfree_skb(skb);
 599                 return -EFAULT;
 600         }
 601
 602         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 603                 if (!skb_partial_csum_set(skb, gso.csum_start,
 604                                           gso.csum_offset)) {
 605                         tun->dev->stats.rx_frame_errors++;
 606                         kfree_skb(skb);
 607                         return -EINVAL;
 608                 }
 609         } else if (tun->flags & TUN_NOCHECKSUM)
 610                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 611
 612         switch (tun->flags & TUN_TYPE_MASK) {
 613         case TUN_TUN_DEV:
 614                 if (tun->flags & TUN_NO_PI) {
 615                         switch (skb->data[0] & 0xf0) {
 616                         case 0x40:
 617                                 pi.proto = htons(ETH_P_IP);
 618                                 break;
 619                         case 0x60:
 620                                 pi.proto = htons(ETH_P_IPV6);
 621                                 break;
 622                         default:
 623                                 tun->dev->stats.rx_dropped++;
 624                                 kfree_skb(skb);
 625                                 return -EINVAL;
 626                         }
 627                 }
 628
 629                 skb_reset_mac_header(skb);
 630                 skb->protocol = pi.proto;
 631                 skb->dev = tun->dev;
 632                 break;
 633         case TUN_TAP_DEV:
 634                 skb->protocol = eth_type_trans(skb, tun->dev);
 635                 break;
 636         };
 637
 638         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 639                 pr_debug("GSO!\n");
 640                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 641                 case VIRTIO_NET_HDR_GSO_TCPV4:
 642                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 643                         break;
 644                 case VIRTIO_NET_HDR_GSO_TCPV6:
 645                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 646                         break;
 647                 case VIRTIO_NET_HDR_GSO_UDP:
 648                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 649                         break;
 650                 default:
 651                         tun->dev->stats.rx_frame_errors++;
 652                         kfree_skb(skb);
 653                         return -EINVAL;
 654                 }
 655
 656                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
 657                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 658
 659                 skb_shinfo(skb)->gso_size = gso.gso_size;
 660                 if (skb_shinfo(skb)->gso_size == 0) {
 661                         tun->dev->stats.rx_frame_errors++;
 662                         kfree_skb(skb);
 663                         return -EINVAL;
 664                 }
 665
 666                 /* Header must be checked, and gso_segs computed. */
 667                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 668                 skb_shinfo(skb)->gso_segs = 0;
 669         }
 670
 671         netif_rx_ni(skb);
 672
 673         tun->dev->stats.rx_packets++;
 674         tun->dev->stats.rx_bytes += len;
 675
 676         return count;
 677 }
 678
 679 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
 680                               unsigned long count, loff_t pos)
 681 {
 682         struct file *file = iocb->ki_filp;
 683         struct tun_struct *tun = tun_get(file);
 684         ssize_t result;
 685
 686         if (!tun)
 687                 return -EBADFD;
 688
 689         DBG(KERN_INFO "%s: tun_chr_write %ld\n", tun->dev->name, count);
 690
 691         result = tun_get_user(tun, iv, iov_length(iv, count),
 692                               file->f_flags & O_NONBLOCK);
 693
 694         tun_put(tun);
 695         return result;
 696 }
 697
 698 /* Put packet to the user space buffer */
 699 static __inline__ ssize_t tun_put_user(struct tun_struct *tun,
 700                                        struct sk_buff *skb,
 701                                        const struct iovec *iv, int len)
 702 {
 703         struct tun_pi pi = { 0, skb->protocol };
 704         ssize_t total = 0;
 705
 706         if (!(tun->flags & TUN_NO_PI)) {
 707                 if ((len -= sizeof(pi)) < 0)
 708                         return -EINVAL;
 709
 710                 if (len < skb->len) {
 711                         /* Packet will be striped */
 712                         pi.flags |= TUN_PKT_STRIP;
 713                 }
 714
 715                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
 716                         return -EFAULT;
 717                 total += sizeof(pi);
 718         }
 719
 720         if (tun->flags & TUN_VNET_HDR) {
 721                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
 722                 if ((len -= tun->vnet_hdr_sz) < 0)
 723                         return -EINVAL;
 724
 725                 if (skb_is_gso(skb)) {
 726                         struct skb_shared_info *sinfo = skb_shinfo(skb);
 727
 728                         /* This is a hint as to how much should be linear. */
 729                         gso.hdr_len = skb_headlen(skb);
 730                         gso.gso_size = sinfo->gso_size;
 731                         if (sinfo->gso_type & SKB_GSO_TCPV4)
 732                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 733                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
 734                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 735                         else if (sinfo->gso_type & SKB_GSO_UDP)
 736                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
 737                         else
 738                                 BUG();
 739                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
 740                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 741                 } else
 742                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
 743
 744                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 745                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 746                         gso.csum_start = skb->csum_start - skb_headroom(skb);
 747                         gso.csum_offset = skb->csum_offset;
 748                 } /* else everything is zero */
 749
 750                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
 751                                                sizeof(gso))))
 752                         return -EFAULT;
 753                 total += tun->vnet_hdr_sz;
 754         }
 755
 756         len = min_t(int, skb->len, len);
 757
 758         skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
 759         total += skb->len;
 760
 761         tun->dev->stats.tx_packets++;
 762         tun->dev->stats.tx_bytes += len;
 763
 764         return total;
 765 }
 766
 767 static ssize_t tun_do_read(struct tun_struct *tun,
 768                            struct kiocb *iocb, const struct iovec *iv,
 769                            ssize_t len, int noblock)
 770 {
 771         DECLARE_WAITQUEUE(wait, current);
 772         struct sk_buff *skb;
 773         ssize_t ret = 0;
 774
 775         DBG(KERN_INFO "%s: tun_chr_read\n", tun->dev->name);
 776
 777         add_wait_queue(&tun->wq.wait, &wait);
 778         while (len) {
 779                 current->state = TASK_INTERRUPTIBLE;
 780
 781                 /* Read frames from the queue */
 782                 if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
 783                         if (noblock) {
 784                                 ret = -EAGAIN;
 785                                 break;
 786                         }
 787                         if (signal_pending(current)) {
 788                                 ret = -ERESTARTSYS;
 789                                 break;
 790                         }
 791                         if (tun->dev->reg_state != NETREG_REGISTERED) {
 792                                 ret = -EIO;
 793                                 break;
 794                         }
 795
 796                         /* Nothing to read, let's sleep */
 797                         schedule();
 798                         continue;
 799                 }
 800                 netif_wake_queue(tun->dev);
 801
 802                 ret = tun_put_user(tun, skb, iv, len);
 803                 kfree_skb(skb);
 804                 break;
 805         }
 806
 807         current->state = TASK_RUNNING;
 808         remove_wait_queue(&tun->wq.wait, &wait);
 809
 810         return ret;
 811 }
 812
 813 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
 814                             unsigned long count, loff_t pos)
 815 {
 816         struct file *file = iocb->ki_filp;
 817         struct tun_file *tfile = file->private_data;
 818         struct tun_struct *tun = __tun_get(tfile);
 819         ssize_t len, ret;
 820
 821         if (!tun)
 822                 return -EBADFD;
 823         len = iov_length(iv, count);
 824         if (len < 0) {
 825                 ret = -EINVAL;
 826                 goto out;
 827         }
 828
 829         ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK);
 830         ret = min_t(ssize_t, ret, len);
 831 out:
 832         tun_put(tun);
 833         return ret;
 834 }
 835
 836 static void tun_setup(struct net_device *dev)
 837 {
 838         struct tun_struct *tun = netdev_priv(dev);
 839
 840         tun->owner = -1;
 841         tun->group = -1;
 842
 843         dev->ethtool_ops = &tun_ethtool_ops;
 844         dev->destructor = tun_free_netdev;
 845 }
 846
 847 /* Trivial set of netlink ops to allow deleting tun or tap
 848  * device with netlink.
 849  */
 850 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
 851 {
 852         return -EINVAL;
 853 }
 854
 855 static struct rtnl_link_ops tun_link_ops __read_mostly = {
 856         .kind           = DRV_NAME,
 857         .priv_size      = sizeof(struct tun_struct),
 858         .setup          = tun_setup,
 859         .validate       = tun_validate,
 860 };
 861
 862 static void tun_sock_write_space(struct sock *sk)
 863 {
 864         struct tun_struct *tun;
 865         wait_queue_head_t *wqueue;
 866
 867         if (!sock_writeable(sk))
 868                 return;
 869
 870         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
 871                 return;
 872
 873         wqueue = sk_sleep(sk);
 874         if (wqueue && waitqueue_active(wqueue))
 875                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
 876                                                 POLLWRNORM | POLLWRBAND);
 877
 878         tun = tun_sk(sk)->tun;
 879         kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
 880 }
 881
 882 static void tun_sock_destruct(struct sock *sk)
 883 {
 884         free_netdev(tun_sk(sk)->tun->dev);
 885 }
 886
 887 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
 888                        struct msghdr *m, size_t total_len)
 889 {
 890         struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
 891         return tun_get_user(tun, m->msg_iov, total_len,
 892                             m->msg_flags & MSG_DONTWAIT);
 893 }
 894
 895 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
 896                        struct msghdr *m, size_t total_len,
 897                        int flags)
 898 {
 899         struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
 900         int ret;
 901         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
 902                 return -EINVAL;
 903         ret = tun_do_read(tun, iocb, m->msg_iov, total_len,
 904                           flags & MSG_DONTWAIT);
 905         if (ret > total_len) {
 906                 m->msg_flags |= MSG_TRUNC;
 907                 ret = flags & MSG_TRUNC ? ret : total_len;
 908         }
 909         return ret;
 910 }
 911
 912 /* Ops structure to mimic raw sockets with tun */
 913 static const struct proto_ops tun_socket_ops = {
 914         .sendmsg = tun_sendmsg,
 915         .recvmsg = tun_recvmsg,
 916 };
 917
 918 static struct proto tun_proto = {
 919         .name           = "tun",
 920         .owner          = THIS_MODULE,
 921         .obj_size       = sizeof(struct tun_sock),
 922 };
 923
 924 static int tun_flags(struct tun_struct *tun)
 925 {
 926         int flags = 0;
 927
 928         if (tun->flags & TUN_TUN_DEV)
 929                 flags |= IFF_TUN;
 930         else
 931                 flags |= IFF_TAP;
 932
 933         if (tun->flags & TUN_NO_PI)
 934                 flags |= IFF_NO_PI;
 935
 936         if (tun->flags & TUN_ONE_QUEUE)
 937                 flags |= IFF_ONE_QUEUE;
 938
 939         if (tun->flags & TUN_VNET_HDR)
 940                 flags |= IFF_VNET_HDR;
 941
 942         return flags;
 943 }
 944
 945 static ssize_t tun_show_flags(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, "0x%x\n", tun_flags(tun));
 950 }
 951
 952 static ssize_t tun_show_owner(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->owner);
 957 }
 958
 959 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
 960                               char *buf)
 961 {
 962         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
 963         return sprintf(buf, "%d\n", tun->group);
 964 }
 965
 966 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
 967 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
 968 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
 969
 970 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
 971 {
 972         struct sock *sk;
 973         struct tun_struct *tun;
 974         struct net_device *dev;
 975         int err;
 976
 977         dev = __dev_get_by_name(net, ifr->ifr_name);
 978         if (dev) {
 979                 const struct cred *cred = current_cred();
 980
 981                 if (ifr->ifr_flags & IFF_TUN_EXCL)
 982                         return -EBUSY;
 983                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
 984                         tun = netdev_priv(dev);
 985                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
 986                         tun = netdev_priv(dev);
 987                 else
 988                         return -EINVAL;
 989
 990                 if (((tun->owner != -1 && cred->euid != tun->owner) ||
 991                      (tun->group != -1 && !in_egroup_p(tun->group))) &&
 992                     !capable(CAP_NET_ADMIN))
 993                         return -EPERM;
 994                 err = security_tun_dev_attach(tun->socket.sk);
 995                 if (err < 0)
 996                         return err;
 997
 998                 err = tun_attach(tun, file);
 999                 if (err < 0)
1000                         return err;
1001         }
1002         else {
1003                 char *name;
1004                 unsigned long flags = 0;
1005
1006                 if (!capable(CAP_NET_ADMIN))
1007                         return -EPERM;
1008                 err = security_tun_dev_create();
1009                 if (err < 0)
1010                         return err;
1011
1012                 /* Set dev type */
1013                 if (ifr->ifr_flags & IFF_TUN) {
1014                         /* TUN device */
1015                         flags |= TUN_TUN_DEV;
1016                         name = "tun%d";
1017                 } else if (ifr->ifr_flags & IFF_TAP) {
1018                         /* TAP device */
1019                         flags |= TUN_TAP_DEV;
1020                         name = "tap%d";
1021                 } else
1022                         return -EINVAL;
1023
1024                 if (*ifr->ifr_name)
1025                         name = ifr->ifr_name;
1026
1027                 dev = alloc_netdev(sizeof(struct tun_struct), name,
1028                                    tun_setup);
1029                 if (!dev)
1030                         return -ENOMEM;
1031
1032                 dev_net_set(dev, net);
1033                 dev->rtnl_link_ops = &tun_link_ops;
1034
1035                 tun = netdev_priv(dev);
1036                 tun->dev = dev;
1037                 tun->flags = flags;
1038                 tun->txflt.count = 0;
1039                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1040
1041                 err = -ENOMEM;
1042                 sk = sk_alloc(net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1043                 if (!sk)
1044                         goto err_free_dev;
1045
1046                 tun->socket.wq = &tun->wq;
1047                 init_waitqueue_head(&tun->wq.wait);
1048                 tun->socket.ops = &tun_socket_ops;
1049                 sock_init_data(&tun->socket, sk);
1050                 sk->sk_write_space = tun_sock_write_space;
1051                 sk->sk_sndbuf = INT_MAX;
1052
1053                 tun_sk(sk)->tun = tun;
1054
1055                 security_tun_dev_post_create(sk);
1056
1057                 tun_net_init(dev);
1058
1059                 if (strchr(dev->name, '%')) {
1060                         err = dev_alloc_name(dev, dev->name);
1061                         if (err < 0)
1062                                 goto err_free_sk;
1063                 }
1064
1065                 err = register_netdevice(tun->dev);
1066                 if (err < 0)
1067                         goto err_free_sk;
1068
1069                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1070                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1071                     device_create_file(&tun->dev->dev, &dev_attr_group))
1072                         printk(KERN_ERR "Failed to create tun sysfs files\n");
1073
1074                 sk->sk_destruct = tun_sock_destruct;
1075
1076                 err = tun_attach(tun, file);
1077                 if (err < 0)
1078                         goto failed;
1079         }
1080
1081         DBG(KERN_INFO "%s: tun_set_iff\n", tun->dev->name);
1082
1083         if (ifr->ifr_flags & IFF_NO_PI)
1084                 tun->flags |= TUN_NO_PI;
1085         else
1086                 tun->flags &= ~TUN_NO_PI;
1087
1088         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1089                 tun->flags |= TUN_ONE_QUEUE;
1090         else
1091                 tun->flags &= ~TUN_ONE_QUEUE;
1092
1093         if (ifr->ifr_flags & IFF_VNET_HDR)
1094                 tun->flags |= TUN_VNET_HDR;
1095         else
1096                 tun->flags &= ~TUN_VNET_HDR;
1097
1098         /* Make sure persistent devices do not get stuck in
1099          * xoff state.
1100          */
1101         if (netif_running(tun->dev))
1102                 netif_wake_queue(tun->dev);
1103
1104         strcpy(ifr->ifr_name, tun->dev->name);
1105         return 0;
1106
1107  err_free_sk:
1108         sock_put(sk);
1109  err_free_dev:
1110         free_netdev(dev);
1111  failed:
1112         return err;
1113 }
1114
1115 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1116                        struct ifreq *ifr)
1117 {
1118         DBG(KERN_INFO "%s: tun_get_iff\n", tun->dev->name);
1119
1120         strcpy(ifr->ifr_name, tun->dev->name);
1121
1122         ifr->ifr_flags = tun_flags(tun);
1123
1124         return 0;
1125 }
1126
1127 /* This is like a cut-down ethtool ops, except done via tun fd so no
1128  * privs required. */
1129 static int set_offload(struct net_device *dev, unsigned long arg)
1130 {
1131         unsigned int old_features, features;
1132
1133         old_features = dev->features;
1134         /* Unset features, set them as we chew on the arg. */
1135         features = (old_features & ~(NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST
1136                                     |NETIF_F_TSO_ECN|NETIF_F_TSO|NETIF_F_TSO6
1137                                     |NETIF_F_UFO));
1138
1139         if (arg & TUN_F_CSUM) {
1140                 features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1141                 arg &= ~TUN_F_CSUM;
1142
1143                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1144                         if (arg & TUN_F_TSO_ECN) {
1145                                 features |= NETIF_F_TSO_ECN;
1146                                 arg &= ~TUN_F_TSO_ECN;
1147                         }
1148                         if (arg & TUN_F_TSO4)
1149                                 features |= NETIF_F_TSO;
1150                         if (arg & TUN_F_TSO6)
1151                                 features |= NETIF_F_TSO6;
1152                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1153                 }
1154
1155                 if (arg & TUN_F_UFO) {
1156                         features |= NETIF_F_UFO;
1157                         arg &= ~TUN_F_UFO;
1158                 }
1159         }
1160
1161         /* This gives the user a way to test for new features in future by
1162          * trying to set them. */
1163         if (arg)
1164                 return -EINVAL;
1165
1166         dev->features = features;
1167         if (old_features != dev->features)
1168                 netdev_features_change(dev);
1169
1170         return 0;
1171 }
1172
1173 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1174                             unsigned long arg, int ifreq_len)
1175 {
1176         struct tun_file *tfile = file->private_data;
1177         struct tun_struct *tun;
1178         void __user* argp = (void __user*)arg;
1179         struct sock_fprog fprog;
1180         struct ifreq ifr;
1181         int sndbuf;
1182         int vnet_hdr_sz;
1183         int ret;
1184
1185         if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
1186                 if (copy_from_user(&ifr, argp, ifreq_len))
1187                         return -EFAULT;
1188
1189         if (cmd == TUNGETFEATURES) {
1190                 /* Currently this just means: "what IFF flags are valid?".
1191                  * This is needed because we never checked for invalid flags on
1192                  * TUNSETIFF. */
1193                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1194                                 IFF_VNET_HDR,
1195                                 (unsigned int __user*)argp);
1196         }
1197
1198         rtnl_lock();
1199
1200         tun = __tun_get(tfile);
1201         if (cmd == TUNSETIFF && !tun) {
1202                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1203
1204                 ret = tun_set_iff(tfile->net, file, &ifr);
1205
1206                 if (ret)
1207                         goto unlock;
1208
1209                 if (copy_to_user(argp, &ifr, ifreq_len))
1210                         ret = -EFAULT;
1211                 goto unlock;
1212         }
1213
1214         ret = -EBADFD;
1215         if (!tun)
1216                 goto unlock;
1217
1218         DBG(KERN_INFO "%s: tun_chr_ioctl cmd %d\n", tun->dev->name, cmd);
1219
1220         ret = 0;
1221         switch (cmd) {
1222         case TUNGETIFF:
1223                 ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1224                 if (ret)
1225                         break;
1226
1227                 if (copy_to_user(argp, &ifr, ifreq_len))
1228                         ret = -EFAULT;
1229                 break;
1230
1231         case TUNSETNOCSUM:
1232                 /* Disable/Enable checksum */
1233                 if (arg)
1234                         tun->flags |= TUN_NOCHECKSUM;
1235                 else
1236                         tun->flags &= ~TUN_NOCHECKSUM;
1237
1238                 DBG(KERN_INFO "%s: checksum %s\n",
1239                     tun->dev->name, arg ? "disabled" : "enabled");
1240                 break;
1241
1242         case TUNSETPERSIST:
1243                 /* Disable/Enable persist mode */
1244                 if (arg)
1245                         tun->flags |= TUN_PERSIST;
1246                 else
1247                         tun->flags &= ~TUN_PERSIST;
1248
1249                 DBG(KERN_INFO "%s: persist %s\n",
1250                     tun->dev->name, arg ? "enabled" : "disabled");
1251                 break;
1252
1253         case TUNSETOWNER:
1254                 /* Set owner of the device */
1255                 tun->owner = (uid_t) arg;
1256
1257                 DBG(KERN_INFO "%s: owner set to %d\n", tun->dev->name, tun->owner);
1258                 break;
1259
1260         case TUNSETGROUP:
1261                 /* Set group of the device */
1262                 tun->group= (gid_t) arg;
1263
1264                 DBG(KERN_INFO "%s: group set to %d\n", tun->dev->name, tun->group);
1265                 break;
1266
1267         case TUNSETLINK:
1268                 /* Only allow setting the type when the interface is down */
1269                 if (tun->dev->flags & IFF_UP) {
1270                         DBG(KERN_INFO "%s: Linktype set failed because interface is up\n",
1271                                 tun->dev->name);
1272                         ret = -EBUSY;
1273                 } else {
1274                         tun->dev->type = (int) arg;
1275                         DBG(KERN_INFO "%s: linktype set to %d\n", tun->dev->name, tun->dev->type);
1276                         ret = 0;
1277                 }
1278                 break;
1279
1280 #ifdef TUN_DEBUG
1281         case TUNSETDEBUG:
1282                 tun->debug = arg;
1283                 break;
1284 #endif
1285         case TUNSETOFFLOAD:
1286                 ret = set_offload(tun->dev, arg);
1287                 break;
1288
1289         case TUNSETTXFILTER:
1290                 /* Can be set only for TAPs */
1291                 ret = -EINVAL;
1292                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1293                         break;
1294                 ret = update_filter(&tun->txflt, (void __user *)arg);
1295                 break;
1296
1297         case SIOCGIFHWADDR:
1298                 /* Get hw addres */
1299                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1300                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1301                 if (copy_to_user(argp, &ifr, ifreq_len))
1302                         ret = -EFAULT;
1303                 break;
1304
1305         case SIOCSIFHWADDR:
1306                 /* Set hw address */
1307                 DBG(KERN_DEBUG "%s: set hw address: %pM\n",
1308                         tun->dev->name, ifr.ifr_hwaddr.sa_data);
1309
1310                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1311                 break;
1312
1313         case TUNGETSNDBUF:
1314                 sndbuf = tun->socket.sk->sk_sndbuf;
1315                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1316                         ret = -EFAULT;
1317                 break;
1318
1319         case TUNSETSNDBUF:
1320                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1321                         ret = -EFAULT;
1322                         break;
1323                 }
1324
1325                 tun->socket.sk->sk_sndbuf = sndbuf;
1326                 break;
1327
1328         case TUNGETVNETHDRSZ:
1329                 vnet_hdr_sz = tun->vnet_hdr_sz;
1330                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1331                         ret = -EFAULT;
1332                 break;
1333
1334         case TUNSETVNETHDRSZ:
1335                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1336                         ret = -EFAULT;
1337                         break;
1338                 }
1339                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1340                         ret = -EINVAL;
1341                         break;
1342                 }
1343
1344                 tun->vnet_hdr_sz = vnet_hdr_sz;
1345                 break;
1346
1347         case TUNATTACHFILTER:
1348                 /* Can be set only for TAPs */
1349                 ret = -EINVAL;
1350                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1351                         break;
1352                 ret = -EFAULT;
1353                 if (copy_from_user(&fprog, argp, sizeof(fprog)))
1354                         break;
1355
1356                 ret = sk_attach_filter(&fprog, tun->socket.sk);
1357                 break;
1358
1359         case TUNDETACHFILTER:
1360                 /* Can be set only for TAPs */
1361                 ret = -EINVAL;
1362                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1363                         break;
1364                 ret = sk_detach_filter(tun->socket.sk);
1365                 break;
1366
1367         default:
1368                 ret = -EINVAL;
1369                 break;
1370         }
1371
1372 unlock:
1373         rtnl_unlock();
1374         if (tun)
1375                 tun_put(tun);
1376         return ret;
1377 }
1378
1379 static long tun_chr_ioctl(struct file *file,
1380                           unsigned int cmd, unsigned long arg)
1381 {
1382         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1383 }
1384
1385 #ifdef CONFIG_COMPAT
1386 static long tun_chr_compat_ioctl(struct file *file,
1387                          unsigned int cmd, unsigned long arg)
1388 {
1389         switch (cmd) {
1390         case TUNSETIFF:
1391         case TUNGETIFF:
1392         case TUNSETTXFILTER:
1393         case TUNGETSNDBUF:
1394         case TUNSETSNDBUF:
1395         case SIOCGIFHWADDR:
1396         case SIOCSIFHWADDR:
1397                 arg = (unsigned long)compat_ptr(arg);
1398                 break;
1399         default:
1400                 arg = (compat_ulong_t)arg;
1401                 break;
1402         }
1403
1404         /*
1405          * compat_ifreq is shorter than ifreq, so we must not access beyond
1406          * the end of that structure. All fields that are used in this
1407          * driver are compatible though, we don't need to convert the
1408          * contents.
1409          */
1410         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1411 }
1412 #endif /* CONFIG_COMPAT */
1413
1414 static int tun_chr_fasync(int fd, struct file *file, int on)
1415 {
1416         struct tun_struct *tun = tun_get(file);
1417         int ret;
1418
1419         if (!tun)
1420                 return -EBADFD;
1421
1422         DBG(KERN_INFO "%s: tun_chr_fasync %d\n", tun->dev->name, on);
1423
1424         if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1425                 goto out;
1426
1427         if (on) {
1428                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1429                 if (ret)
1430                         goto out;
1431                 tun->flags |= TUN_FASYNC;
1432         } else
1433                 tun->flags &= ~TUN_FASYNC;
1434         ret = 0;
1435 out:
1436         tun_put(tun);
1437         return ret;
1438 }
1439
1440 static int tun_chr_open(struct inode *inode, struct file * file)
1441 {
1442         struct tun_file *tfile;
1443
1444         DBG1(KERN_INFO "tunX: tun_chr_open\n");
1445
1446         tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1447         if (!tfile)
1448                 return -ENOMEM;
1449         atomic_set(&tfile->count, 0);
1450         tfile->tun = NULL;
1451         tfile->net = get_net(current->nsproxy->net_ns);
1452         file->private_data = tfile;
1453         return 0;
1454 }
1455
1456 static int tun_chr_close(struct inode *inode, struct file *file)
1457 {
1458         struct tun_file *tfile = file->private_data;
1459         struct tun_struct *tun;
1460
1461         tun = __tun_get(tfile);
1462         if (tun) {
1463                 struct net_device *dev = tun->dev;
1464
1465                 DBG(KERN_INFO "%s: tun_chr_close\n", dev->name);
1466
1467                 __tun_detach(tun);
1468
1469                 /* If desirable, unregister the netdevice. */
1470                 if (!(tun->flags & TUN_PERSIST)) {
1471                         rtnl_lock();
1472                         if (dev->reg_state == NETREG_REGISTERED)
1473                                 unregister_netdevice(dev);
1474                         rtnl_unlock();
1475                 }
1476         }
1477
1478         tun = tfile->tun;
1479         if (tun)
1480                 sock_put(tun->socket.sk);
1481
1482         put_net(tfile->net);
1483         kfree(tfile);
1484
1485         return 0;
1486 }
1487
1488 static const struct file_operations tun_fops = {
1489         .owner  = THIS_MODULE,
1490         .llseek = no_llseek,
1491         .read  = do_sync_read,
1492         .aio_read  = tun_chr_aio_read,
1493         .write = do_sync_write,
1494         .aio_write = tun_chr_aio_write,
1495         .poll   = tun_chr_poll,
1496         .unlocked_ioctl = tun_chr_ioctl,
1497 #ifdef CONFIG_COMPAT
1498         .compat_ioctl = tun_chr_compat_ioctl,
1499 #endif
1500         .open   = tun_chr_open,
1501         .release = tun_chr_close,
1502         .fasync = tun_chr_fasync
1503 };
1504
1505 static struct miscdevice tun_miscdev = {
1506         .minor = TUN_MINOR,
1507         .name = "tun",
1508         .nodename = "net/tun",
1509         .fops = &tun_fops,
1510 };
1511
1512 /* ethtool interface */
1513
1514 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1515 {
1516         cmd->supported          = 0;
1517         cmd->advertising        = 0;
1518         cmd->speed              = SPEED_10;
1519         cmd->duplex             = DUPLEX_FULL;
1520         cmd->port               = PORT_TP;
1521         cmd->phy_address        = 0;
1522         cmd->transceiver        = XCVR_INTERNAL;
1523         cmd->autoneg            = AUTONEG_DISABLE;
1524         cmd->maxtxpkt           = 0;
1525         cmd->maxrxpkt           = 0;
1526         return 0;
1527 }
1528
1529 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1530 {
1531         struct tun_struct *tun = netdev_priv(dev);
1532
1533         strcpy(info->driver, DRV_NAME);
1534         strcpy(info->version, DRV_VERSION);
1535         strcpy(info->fw_version, "N/A");
1536
1537         switch (tun->flags & TUN_TYPE_MASK) {
1538         case TUN_TUN_DEV:
1539                 strcpy(info->bus_info, "tun");
1540                 break;
1541         case TUN_TAP_DEV:
1542                 strcpy(info->bus_info, "tap");
1543                 break;
1544         }
1545 }
1546
1547 static u32 tun_get_msglevel(struct net_device *dev)
1548 {
1549 #ifdef TUN_DEBUG
1550         struct tun_struct *tun = netdev_priv(dev);
1551         return tun->debug;
1552 #else
1553         return -EOPNOTSUPP;
1554 #endif
1555 }
1556
1557 static void tun_set_msglevel(struct net_device *dev, u32 value)
1558 {
1559 #ifdef TUN_DEBUG
1560         struct tun_struct *tun = netdev_priv(dev);
1561         tun->debug = value;
1562 #endif
1563 }
1564
1565 static u32 tun_get_link(struct net_device *dev)
1566 {
1567         struct tun_struct *tun = netdev_priv(dev);
1568         return !!tun->tfile;
1569 }
1570
1571 static u32 tun_get_rx_csum(struct net_device *dev)
1572 {
1573         struct tun_struct *tun = netdev_priv(dev);
1574         return (tun->flags & TUN_NOCHECKSUM) == 0;
1575 }
1576
1577 static int tun_set_rx_csum(struct net_device *dev, u32 data)
1578 {
1579         struct tun_struct *tun = netdev_priv(dev);
1580         if (data)
1581                 tun->flags &= ~TUN_NOCHECKSUM;
1582         else
1583                 tun->flags |= TUN_NOCHECKSUM;
1584         return 0;
1585 }
1586
1587 static const struct ethtool_ops tun_ethtool_ops = {
1588         .get_settings   = tun_get_settings,
1589         .get_drvinfo    = tun_get_drvinfo,
1590         .get_msglevel   = tun_get_msglevel,
1591         .set_msglevel   = tun_set_msglevel,
1592         .get_link       = tun_get_link,
1593         .get_rx_csum    = tun_get_rx_csum,
1594         .set_rx_csum    = tun_set_rx_csum
1595 };
1596
1597
1598 static int __init tun_init(void)
1599 {
1600         int ret = 0;
1601
1602         printk(KERN_INFO "tun: %s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1603         printk(KERN_INFO "tun: %s\n", DRV_COPYRIGHT);
1604
1605         ret = rtnl_link_register(&tun_link_ops);
1606         if (ret) {
1607                 printk(KERN_ERR "tun: Can't register link_ops\n");
1608                 goto err_linkops;
1609         }
1610
1611         ret = misc_register(&tun_miscdev);
1612         if (ret) {
1613                 printk(KERN_ERR "tun: Can't register misc device %d\n", TUN_MINOR);
1614                 goto err_misc;
1615         }
1616         return  0;
1617 err_misc:
1618         rtnl_link_unregister(&tun_link_ops);
1619 err_linkops:
1620         return ret;
1621 }
1622
1623 static void tun_cleanup(void)
1624 {
1625         misc_deregister(&tun_miscdev);
1626         rtnl_link_unregister(&tun_link_ops);
1627 }
1628
1629 /* Get an underlying socket object from tun file.  Returns error unless file is
1630  * attached to a device.  The returned object works like a packet socket, it
1631  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1632  * holding a reference to the file for as long as the socket is in use. */
1633 struct socket *tun_get_socket(struct file *file)
1634 {
1635         struct tun_struct *tun;
1636         if (file->f_op != &tun_fops)
1637                 return ERR_PTR(-EINVAL);
1638         tun = tun_get(file);
1639         if (!tun)
1640                 return ERR_PTR(-EBADFD);
1641         tun_put(tun);
1642         return &tun->socket;
1643 }
1644 EXPORT_SYMBOL_GPL(tun_get_socket);
1645
1646 module_init(tun_init);
1647 module_exit(tun_cleanup);
1648 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1649 MODULE_AUTHOR(DRV_COPYRIGHT);
1650 MODULE_LICENSE("GPL");
1651 MODULE_ALIAS_MISCDEV(TUN_MINOR);