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