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  *  Brian Braunstein <linuxkernel@bristyle.com> 2007/03/23
  22  *    Fixed hw address handling.  Now net_device.dev_addr is kept consistent
  23  *    with tun.dev_addr when the address is set by this module.
  24  *
  25  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  26  *    Add TUNSETLINK ioctl to set the link encapsulation
  27  *
  28  *  Mark Smith <markzzzsmith@yahoo.com.au>
  29  *   Use random_ether_addr() for tap MAC address.
  30  *
  31  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  32  *    Fixes in packet dropping, queue length setting and queue wakeup.
  33  *    Increased default tx queue length.
  34  *    Added ethtool API.
  35  *    Minor cleanups
  36  *
  37  *  Daniel Podlejski <underley@underley.eu.org>
  38  *    Modifications for 2.3.99-pre5 kernel.
  39  */
  40
  41 #define DRV_NAME        "tun"
  42 #define DRV_VERSION     "1.6"
  43 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  44 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  45
  46 #include <linux/module.h>
  47 #include <linux/errno.h>
  48 #include <linux/kernel.h>
  49 #include <linux/major.h>
  50 #include <linux/slab.h>
  51 #include <linux/poll.h>
  52 #include <linux/fcntl.h>
  53 #include <linux/init.h>
  54 #include <linux/skbuff.h>
  55 #include <linux/netdevice.h>
  56 #include <linux/etherdevice.h>
  57 #include <linux/miscdevice.h>
  58 #include <linux/ethtool.h>
  59 #include <linux/rtnetlink.h>
  60 #include <linux/if.h>
  61 #include <linux/if_arp.h>
  62 #include <linux/if_ether.h>
  63 #include <linux/if_tun.h>
  64 #include <linux/crc32.h>
  65 #include <linux/nsproxy.h>
  66 #include <linux/virtio_net.h>
  67 #include <net/net_namespace.h>
  68 #include <net/netns/generic.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 struct tun_struct {
  87         struct list_head        list;
  88         unsigned long           flags;
  89         int                     attached;
  90         uid_t                   owner;
  91         gid_t                   group;
  92
  93         wait_queue_head_t       read_wait;
  94         struct sk_buff_head     readq;
  95
  96         struct net_device       *dev;
  97
  98         struct fasync_struct    *fasync;
  99
 100         unsigned long if_flags;
 101         u8 dev_addr[ETH_ALEN];
 102         u32 chr_filter[2];
 103         u32 net_filter[2];
 104
 105 #ifdef TUN_DEBUG
 106         int debug;
 107 #endif
 108 };
 109
 110 /* Network device part of the driver */
 111
 112 static unsigned int tun_net_id;
 113 struct tun_net {
 114         struct list_head dev_list;
 115 };
 116
 117 static const struct ethtool_ops tun_ethtool_ops;
 118
 119 /* Net device open. */
 120 static int tun_net_open(struct net_device *dev)
 121 {
 122         netif_start_queue(dev);
 123         return 0;
 124 }
 125
 126 /* Net device close. */
 127 static int tun_net_close(struct net_device *dev)
 128 {
 129         netif_stop_queue(dev);
 130         return 0;
 131 }
 132
 133 /* Net device start xmit */
 134 static int tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 135 {
 136         struct tun_struct *tun = netdev_priv(dev);
 137
 138         DBG(KERN_INFO "%s: tun_net_xmit %d\n", tun->dev->name, skb->len);
 139
 140         /* Drop packet if interface is not attached */
 141         if (!tun->attached)
 142                 goto drop;
 143
 144         /* Packet dropping */
 145         if (skb_queue_len(&tun->readq) >= dev->tx_queue_len) {
 146                 if (!(tun->flags & TUN_ONE_QUEUE)) {
 147                         /* Normal queueing mode. */
 148                         /* Packet scheduler handles dropping of further packets. */
 149                         netif_stop_queue(dev);
 150
 151                         /* We won't see all dropped packets individually, so overrun
 152                          * error is more appropriate. */
 153                         dev->stats.tx_fifo_errors++;
 154                 } else {
 155                         /* Single queue mode.
 156                          * Driver handles dropping of all packets itself. */
 157                         goto drop;
 158                 }
 159         }
 160
 161         /* Queue packet */
 162         skb_queue_tail(&tun->readq, skb);
 163         dev->trans_start = jiffies;
 164
 165         /* Notify and wake up reader process */
 166         if (tun->flags & TUN_FASYNC)
 167                 kill_fasync(&tun->fasync, SIGIO, POLL_IN);
 168         wake_up_interruptible(&tun->read_wait);
 169         return 0;
 170
 171 drop:
 172         dev->stats.tx_dropped++;
 173         kfree_skb(skb);
 174         return 0;
 175 }
 176
 177 /** Add the specified Ethernet address to this multicast filter. */
 178 static void
 179 add_multi(u32* filter, const u8* addr)
 180 {
 181         int bit_nr = ether_crc(ETH_ALEN, addr) >> 26;
 182         filter[bit_nr >> 5] |= 1 << (bit_nr & 31);
 183 }
 184
 185 /** Remove the specified Ethernet addres from this multicast filter. */
 186 static void
 187 del_multi(u32* filter, const u8* addr)
 188 {
 189         int bit_nr = ether_crc(ETH_ALEN, addr) >> 26;
 190         filter[bit_nr >> 5] &= ~(1 << (bit_nr & 31));
 191 }
 192
 193 /** Update the list of multicast groups to which the network device belongs.
 194  * This list is used to filter packets being sent from the character device to
 195  * the network device. */
 196 static void
 197 tun_net_mclist(struct net_device *dev)
 198 {
 199         struct tun_struct *tun = netdev_priv(dev);
 200         const struct dev_mc_list *mclist;
 201         int i;
 202         DECLARE_MAC_BUF(mac);
 203         DBG(KERN_DEBUG "%s: tun_net_mclist: mc_count %d\n",
 204                         dev->name, dev->mc_count);
 205         memset(tun->chr_filter, 0, sizeof tun->chr_filter);
 206         for (i = 0, mclist = dev->mc_list; i < dev->mc_count && mclist != NULL;
 207                         i++, mclist = mclist->next) {
 208                 add_multi(tun->net_filter, mclist->dmi_addr);
 209                 DBG(KERN_DEBUG "%s: tun_net_mclist: %s\n",
 210                     dev->name, print_mac(mac, mclist->dmi_addr));
 211         }
 212 }
 213
 214 #define MIN_MTU 68
 215 #define MAX_MTU 65535
 216
 217 static int
 218 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 219 {
 220         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 221                 return -EINVAL;
 222         dev->mtu = new_mtu;
 223         return 0;
 224 }
 225
 226 /* Initialize net device. */
 227 static void tun_net_init(struct net_device *dev)
 228 {
 229         struct tun_struct *tun = netdev_priv(dev);
 230
 231         switch (tun->flags & TUN_TYPE_MASK) {
 232         case TUN_TUN_DEV:
 233                 /* Point-to-Point TUN Device */
 234                 dev->hard_header_len = 0;
 235                 dev->addr_len = 0;
 236                 dev->mtu = 1500;
 237                 dev->change_mtu = tun_net_change_mtu;
 238
 239                 /* Zero header length */
 240                 dev->type = ARPHRD_NONE;
 241                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 242                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 243                 break;
 244
 245         case TUN_TAP_DEV:
 246                 /* Ethernet TAP Device */
 247                 dev->set_multicast_list = tun_net_mclist;
 248
 249                 ether_setup(dev);
 250                 dev->change_mtu = tun_net_change_mtu;
 251
 252                 /* random address already created for us by tun_set_iff, use it */
 253                 memcpy(dev->dev_addr, tun->dev_addr, min(sizeof(tun->dev_addr), sizeof(dev->dev_addr)) );
 254
 255                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 256                 break;
 257         }
 258 }
 259
 260 /* Character device part */
 261
 262 /* Poll */
 263 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
 264 {
 265         struct tun_struct *tun = file->private_data;
 266         unsigned int mask = POLLOUT | POLLWRNORM;
 267
 268         if (!tun)
 269                 return -EBADFD;
 270
 271         DBG(KERN_INFO "%s: tun_chr_poll\n", tun->dev->name);
 272
 273         poll_wait(file, &tun->read_wait, wait);
 274
 275         if (!skb_queue_empty(&tun->readq))
 276                 mask |= POLLIN | POLLRDNORM;
 277
 278         return mask;
 279 }
 280
 281 /* Get packet from user space buffer */
 282 static __inline__ ssize_t tun_get_user(struct tun_struct *tun, struct iovec *iv, size_t count)
 283 {
 284         struct tun_pi pi = { 0, __constant_htons(ETH_P_IP) };
 285         struct sk_buff *skb;
 286         size_t len = count, align = 0;
 287         struct virtio_net_hdr gso = { 0 };
 288
 289         if (!(tun->flags & TUN_NO_PI)) {
 290                 if ((len -= sizeof(pi)) > count)
 291                         return -EINVAL;
 292
 293                 if(memcpy_fromiovec((void *)&pi, iv, sizeof(pi)))
 294                         return -EFAULT;
 295         }
 296
 297         if (tun->flags & TUN_VNET_HDR) {
 298                 if ((len -= sizeof(gso)) > count)
 299                         return -EINVAL;
 300
 301                 if (memcpy_fromiovec((void *)&gso, iv, sizeof(gso)))
 302                         return -EFAULT;
 303
 304                 if (gso.hdr_len > len)
 305                         return -EINVAL;
 306         }
 307
 308         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
 309                 align = NET_IP_ALIGN;
 310                 if (unlikely(len < ETH_HLEN))
 311                         return -EINVAL;
 312         }
 313
 314         if (!(skb = alloc_skb(len + align, GFP_KERNEL))) {
 315                 tun->dev->stats.rx_dropped++;
 316                 return -ENOMEM;
 317         }
 318
 319         if (align)
 320                 skb_reserve(skb, align);
 321         if (memcpy_fromiovec(skb_put(skb, len), iv, len)) {
 322                 tun->dev->stats.rx_dropped++;
 323                 kfree_skb(skb);
 324                 return -EFAULT;
 325         }
 326
 327         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 328                 if (!skb_partial_csum_set(skb, gso.csum_start,
 329                                           gso.csum_offset)) {
 330                         tun->dev->stats.rx_frame_errors++;
 331                         kfree_skb(skb);
 332                         return -EINVAL;
 333                 }
 334         } else if (tun->flags & TUN_NOCHECKSUM)
 335                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 336
 337         switch (tun->flags & TUN_TYPE_MASK) {
 338         case TUN_TUN_DEV:
 339                 if (tun->flags & TUN_NO_PI) {
 340                         switch (skb->data[0] & 0xf0) {
 341                         case 0x40:
 342                                 pi.proto = htons(ETH_P_IP);
 343                                 break;
 344                         case 0x60:
 345                                 pi.proto = htons(ETH_P_IPV6);
 346                                 break;
 347                         default:
 348                                 tun->dev->stats.rx_dropped++;
 349                                 kfree_skb(skb);
 350                                 return -EINVAL;
 351                         }
 352                 }
 353
 354                 skb_reset_mac_header(skb);
 355                 skb->protocol = pi.proto;
 356                 skb->dev = tun->dev;
 357                 break;
 358         case TUN_TAP_DEV:
 359                 skb->protocol = eth_type_trans(skb, tun->dev);
 360                 break;
 361         };
 362
 363         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 364                 pr_debug("GSO!\n");
 365                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 366                 case VIRTIO_NET_HDR_GSO_TCPV4:
 367                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 368                         break;
 369                 case VIRTIO_NET_HDR_GSO_TCPV6:
 370                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 371                         break;
 372                 default:
 373                         tun->dev->stats.rx_frame_errors++;
 374                         kfree_skb(skb);
 375                         return -EINVAL;
 376                 }
 377
 378                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
 379                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 380
 381                 skb_shinfo(skb)->gso_size = gso.gso_size;
 382                 if (skb_shinfo(skb)->gso_size == 0) {
 383                         tun->dev->stats.rx_frame_errors++;
 384                         kfree_skb(skb);
 385                         return -EINVAL;
 386                 }
 387
 388                 /* Header must be checked, and gso_segs computed. */
 389                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 390                 skb_shinfo(skb)->gso_segs = 0;
 391         }
 392
 393         netif_rx_ni(skb);
 394         tun->dev->last_rx = jiffies;
 395
 396         tun->dev->stats.rx_packets++;
 397         tun->dev->stats.rx_bytes += len;
 398
 399         return count;
 400 }
 401
 402 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
 403                               unsigned long count, loff_t pos)
 404 {
 405         struct tun_struct *tun = iocb->ki_filp->private_data;
 406
 407         if (!tun)
 408                 return -EBADFD;
 409
 410         DBG(KERN_INFO "%s: tun_chr_write %ld\n", tun->dev->name, count);
 411
 412         return tun_get_user(tun, (struct iovec *) iv, iov_length(iv, count));
 413 }
 414
 415 /* Put packet to the user space buffer */
 416 static __inline__ ssize_t tun_put_user(struct tun_struct *tun,
 417                                        struct sk_buff *skb,
 418                                        struct iovec *iv, int len)
 419 {
 420         struct tun_pi pi = { 0, skb->protocol };
 421         ssize_t total = 0;
 422
 423         if (!(tun->flags & TUN_NO_PI)) {
 424                 if ((len -= sizeof(pi)) < 0)
 425                         return -EINVAL;
 426
 427                 if (len < skb->len) {
 428                         /* Packet will be striped */
 429                         pi.flags |= TUN_PKT_STRIP;
 430                 }
 431
 432                 if (memcpy_toiovec(iv, (void *) &pi, sizeof(pi)))
 433                         return -EFAULT;
 434                 total += sizeof(pi);
 435         }
 436
 437         if (tun->flags & TUN_VNET_HDR) {
 438                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
 439                 if ((len -= sizeof(gso)) < 0)
 440                         return -EINVAL;
 441
 442                 if (skb_is_gso(skb)) {
 443                         struct skb_shared_info *sinfo = skb_shinfo(skb);
 444
 445                         /* This is a hint as to how much should be linear. */
 446                         gso.hdr_len = skb_headlen(skb);
 447                         gso.gso_size = sinfo->gso_size;
 448                         if (sinfo->gso_type & SKB_GSO_TCPV4)
 449                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 450                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
 451                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 452                         else
 453                                 BUG();
 454                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
 455                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 456                 } else
 457                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
 458
 459                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 460                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 461                         gso.csum_start = skb->csum_start - skb_headroom(skb);
 462                         gso.csum_offset = skb->csum_offset;
 463                 } /* else everything is zero */
 464
 465                 if (unlikely(memcpy_toiovec(iv, (void *)&gso, sizeof(gso))))
 466                         return -EFAULT;
 467                 total += sizeof(gso);
 468         }
 469
 470         len = min_t(int, skb->len, len);
 471
 472         skb_copy_datagram_iovec(skb, 0, iv, len);
 473         total += len;
 474
 475         tun->dev->stats.tx_packets++;
 476         tun->dev->stats.tx_bytes += len;
 477
 478         return total;
 479 }
 480
 481 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
 482                             unsigned long count, loff_t pos)
 483 {
 484         struct file *file = iocb->ki_filp;
 485         struct tun_struct *tun = file->private_data;
 486         DECLARE_WAITQUEUE(wait, current);
 487         struct sk_buff *skb;
 488         ssize_t len, ret = 0;
 489         DECLARE_MAC_BUF(mac);
 490
 491         if (!tun)
 492                 return -EBADFD;
 493
 494         DBG(KERN_INFO "%s: tun_chr_read\n", tun->dev->name);
 495
 496         len = iov_length(iv, count);
 497         if (len < 0)
 498                 return -EINVAL;
 499
 500         add_wait_queue(&tun->read_wait, &wait);
 501         while (len) {
 502                 const u8 ones[ ETH_ALEN] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
 503                 u8 addr[ ETH_ALEN];
 504                 int bit_nr;
 505
 506                 current->state = TASK_INTERRUPTIBLE;
 507
 508                 /* Read frames from the queue */
 509                 if (!(skb=skb_dequeue(&tun->readq))) {
 510                         if (file->f_flags & O_NONBLOCK) {
 511                                 ret = -EAGAIN;
 512                                 break;
 513                         }
 514                         if (signal_pending(current)) {
 515                                 ret = -ERESTARTSYS;
 516                                 break;
 517                         }
 518
 519                         /* Nothing to read, let's sleep */
 520                         schedule();
 521                         continue;
 522                 }
 523                 netif_wake_queue(tun->dev);
 524
 525                 /** Decide whether to accept this packet. This code is designed to
 526                  * behave identically to an Ethernet interface. Accept the packet if
 527                  * - we are promiscuous.
 528                  * - the packet is addressed to us.
 529                  * - the packet is broadcast.
 530                  * - the packet is multicast and
 531                  *   - we are multicast promiscous.
 532                  *   - we belong to the multicast group.
 533                  */
 534                 skb_copy_from_linear_data(skb, addr, min_t(size_t, sizeof addr,
 535                                                                    skb->len));
 536                 bit_nr = ether_crc(sizeof addr, addr) >> 26;
 537                 if ((tun->if_flags & IFF_PROMISC) ||
 538                                 memcmp(addr, tun->dev_addr, sizeof addr) == 0 ||
 539                                 memcmp(addr, ones, sizeof addr) == 0 ||
 540                                 (((addr[0] == 1 && addr[1] == 0 && addr[2] == 0x5e) ||
 541                                   (addr[0] == 0x33 && addr[1] == 0x33)) &&
 542                                  ((tun->if_flags & IFF_ALLMULTI) ||
 543                                   (tun->chr_filter[bit_nr >> 5] & (1 << (bit_nr & 31)))))) {
 544                         DBG(KERN_DEBUG "%s: tun_chr_readv: accepted: %s\n",
 545                                         tun->dev->name, print_mac(mac, addr));
 546                         ret = tun_put_user(tun, skb, (struct iovec *) iv, len);
 547                         kfree_skb(skb);
 548                         break;
 549                 } else {
 550                         DBG(KERN_DEBUG "%s: tun_chr_readv: rejected: %s\n",
 551                                         tun->dev->name, print_mac(mac, addr));
 552                         kfree_skb(skb);
 553                         continue;
 554                 }
 555         }
 556
 557         current->state = TASK_RUNNING;
 558         remove_wait_queue(&tun->read_wait, &wait);
 559
 560         return ret;
 561 }
 562
 563 static void tun_setup(struct net_device *dev)
 564 {
 565         struct tun_struct *tun = netdev_priv(dev);
 566
 567         skb_queue_head_init(&tun->readq);
 568         init_waitqueue_head(&tun->read_wait);
 569
 570         tun->owner = -1;
 571         tun->group = -1;
 572
 573         dev->open = tun_net_open;
 574         dev->hard_start_xmit = tun_net_xmit;
 575         dev->stop = tun_net_close;
 576         dev->ethtool_ops = &tun_ethtool_ops;
 577         dev->destructor = free_netdev;
 578         dev->features |= NETIF_F_NETNS_LOCAL;
 579 }
 580
 581 static struct tun_struct *tun_get_by_name(struct tun_net *tn, const char *name)
 582 {
 583         struct tun_struct *tun;
 584
 585         ASSERT_RTNL();
 586         list_for_each_entry(tun, &tn->dev_list, list) {
 587                 if (!strncmp(tun->dev->name, name, IFNAMSIZ))
 588                     return tun;
 589         }
 590
 591         return NULL;
 592 }
 593
 594 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
 595 {
 596         struct tun_net *tn;
 597         struct tun_struct *tun;
 598         struct net_device *dev;
 599         int err;
 600
 601         tn = net_generic(net, tun_net_id);
 602         tun = tun_get_by_name(tn, ifr->ifr_name);
 603         if (tun) {
 604                 if (tun->attached)
 605                         return -EBUSY;
 606
 607                 /* Check permissions */
 608                 if (((tun->owner != -1 &&
 609                       current->euid != tun->owner) ||
 610                      (tun->group != -1 &&
 611                       current->egid != tun->group)) &&
 612                      !capable(CAP_NET_ADMIN))
 613                         return -EPERM;
 614         }
 615         else if (__dev_get_by_name(net, ifr->ifr_name))
 616                 return -EINVAL;
 617         else {
 618                 char *name;
 619                 unsigned long flags = 0;
 620
 621                 err = -EINVAL;
 622
 623                 if (!capable(CAP_NET_ADMIN))
 624                         return -EPERM;
 625
 626                 /* Set dev type */
 627                 if (ifr->ifr_flags & IFF_TUN) {
 628                         /* TUN device */
 629                         flags |= TUN_TUN_DEV;
 630                         name = "tun%d";
 631                 } else if (ifr->ifr_flags & IFF_TAP) {
 632                         /* TAP device */
 633                         flags |= TUN_TAP_DEV;
 634                         name = "tap%d";
 635                 } else
 636                         goto failed;
 637
 638                 if (*ifr->ifr_name)
 639                         name = ifr->ifr_name;
 640
 641                 dev = alloc_netdev(sizeof(struct tun_struct), name,
 642                                    tun_setup);
 643                 if (!dev)
 644                         return -ENOMEM;
 645
 646                 dev_net_set(dev, net);
 647                 tun = netdev_priv(dev);
 648                 tun->dev = dev;
 649                 tun->flags = flags;
 650                 /* Be promiscuous by default to maintain previous behaviour. */
 651                 tun->if_flags = IFF_PROMISC;
 652                 /* Generate random Ethernet address. */
 653                 *(__be16 *)tun->dev_addr = htons(0x00FF);
 654                 get_random_bytes(tun->dev_addr + sizeof(u16), 4);
 655                 memset(tun->chr_filter, 0, sizeof tun->chr_filter);
 656
 657                 tun_net_init(dev);
 658
 659                 if (strchr(dev->name, '%')) {
 660                         err = dev_alloc_name(dev, dev->name);
 661                         if (err < 0)
 662                                 goto err_free_dev;
 663                 }
 664
 665                 err = register_netdevice(tun->dev);
 666                 if (err < 0)
 667                         goto err_free_dev;
 668
 669                 list_add(&tun->list, &tn->dev_list);
 670         }
 671
 672         DBG(KERN_INFO "%s: tun_set_iff\n", tun->dev->name);
 673
 674         if (ifr->ifr_flags & IFF_NO_PI)
 675                 tun->flags |= TUN_NO_PI;
 676         else
 677                 tun->flags &= ~TUN_NO_PI;
 678
 679         if (ifr->ifr_flags & IFF_ONE_QUEUE)
 680                 tun->flags |= TUN_ONE_QUEUE;
 681         else
 682                 tun->flags &= ~TUN_ONE_QUEUE;
 683
 684         if (ifr->ifr_flags & IFF_VNET_HDR)
 685                 tun->flags |= TUN_VNET_HDR;
 686         else
 687                 tun->flags &= ~TUN_VNET_HDR;
 688
 689         file->private_data = tun;
 690         tun->attached = 1;
 691         get_net(dev_net(tun->dev));
 692
 693         strcpy(ifr->ifr_name, tun->dev->name);
 694         return 0;
 695
 696  err_free_dev:
 697         free_netdev(dev);
 698  failed:
 699         return err;
 700 }
 701
 702 /* This is like a cut-down ethtool ops, except done via tun fd so no
 703  * privs required. */
 704 static int set_offload(struct net_device *dev, unsigned long arg)
 705 {
 706         unsigned int old_features, features;
 707
 708         old_features = dev->features;
 709         /* Unset features, set them as we chew on the arg. */
 710         features = (old_features & ~(NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST
 711                                     |NETIF_F_TSO_ECN|NETIF_F_TSO|NETIF_F_TSO6));
 712
 713         if (arg & TUN_F_CSUM) {
 714                 features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
 715                 arg &= ~TUN_F_CSUM;
 716
 717                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
 718                         if (arg & TUN_F_TSO_ECN) {
 719                                 features |= NETIF_F_TSO_ECN;
 720                                 arg &= ~TUN_F_TSO_ECN;
 721                         }
 722                         if (arg & TUN_F_TSO4)
 723                                 features |= NETIF_F_TSO;
 724                         if (arg & TUN_F_TSO6)
 725                                 features |= NETIF_F_TSO6;
 726                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
 727                 }
 728         }
 729
 730         /* This gives the user a way to test for new features in future by
 731          * trying to set them. */
 732         if (arg)
 733                 return -EINVAL;
 734
 735         dev->features = features;
 736         if (old_features != dev->features)
 737                 netdev_features_change(dev);
 738
 739         return 0;
 740 }
 741
 742 static int tun_chr_ioctl(struct inode *inode, struct file *file,
 743                          unsigned int cmd, unsigned long arg)
 744 {
 745         struct tun_struct *tun = file->private_data;
 746         void __user* argp = (void __user*)arg;
 747         struct ifreq ifr;
 748         DECLARE_MAC_BUF(mac);
 749
 750         if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
 751                 if (copy_from_user(&ifr, argp, sizeof ifr))
 752                         return -EFAULT;
 753
 754         if (cmd == TUNSETIFF && !tun) {
 755                 int err;
 756
 757                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
 758
 759                 rtnl_lock();
 760                 err = tun_set_iff(current->nsproxy->net_ns, file, &ifr);
 761                 rtnl_unlock();
 762
 763                 if (err)
 764                         return err;
 765
 766                 if (copy_to_user(argp, &ifr, sizeof(ifr)))
 767                         return -EFAULT;
 768                 return 0;
 769         }
 770
 771         if (cmd == TUNGETFEATURES) {
 772                 /* Currently this just means: "what IFF flags are valid?".
 773                  * This is needed because we never checked for invalid flags on
 774                  * TUNSETIFF. */
 775                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
 776                                 IFF_VNET_HDR,
 777                                 (unsigned int __user*)argp);
 778         }
 779
 780         if (!tun)
 781                 return -EBADFD;
 782
 783         DBG(KERN_INFO "%s: tun_chr_ioctl cmd %d\n", tun->dev->name, cmd);
 784
 785         switch (cmd) {
 786         case TUNSETNOCSUM:
 787                 /* Disable/Enable checksum */
 788                 if (arg)
 789                         tun->flags |= TUN_NOCHECKSUM;
 790                 else
 791                         tun->flags &= ~TUN_NOCHECKSUM;
 792
 793                 DBG(KERN_INFO "%s: checksum %s\n",
 794                     tun->dev->name, arg ? "disabled" : "enabled");
 795                 break;
 796
 797         case TUNSETPERSIST:
 798                 /* Disable/Enable persist mode */
 799                 if (arg)
 800                         tun->flags |= TUN_PERSIST;
 801                 else
 802                         tun->flags &= ~TUN_PERSIST;
 803
 804                 DBG(KERN_INFO "%s: persist %s\n",
 805                     tun->dev->name, arg ? "enabled" : "disabled");
 806                 break;
 807
 808         case TUNSETOWNER:
 809                 /* Set owner of the device */
 810                 tun->owner = (uid_t) arg;
 811
 812                 DBG(KERN_INFO "%s: owner set to %d\n", tun->dev->name, tun->owner);
 813                 break;
 814
 815         case TUNSETGROUP:
 816                 /* Set group of the device */
 817                 tun->group= (gid_t) arg;
 818
 819                 DBG(KERN_INFO "%s: group set to %d\n", tun->dev->name, tun->group);
 820                 break;
 821
 822         case TUNSETLINK:
 823         {
 824                 int ret;
 825
 826                 /* Only allow setting the type when the interface is down */
 827                 rtnl_lock();
 828                 if (tun->dev->flags & IFF_UP) {
 829                         DBG(KERN_INFO "%s: Linktype set failed because interface is up\n",
 830                                 tun->dev->name);
 831                         ret = -EBUSY;
 832                 } else {
 833                         tun->dev->type = (int) arg;
 834                         DBG(KERN_INFO "%s: linktype set to %d\n", tun->dev->name, tun->dev->type);
 835                         ret = 0;
 836                 }
 837                 rtnl_unlock();
 838                 return ret;
 839         }
 840
 841 #ifdef TUN_DEBUG
 842         case TUNSETDEBUG:
 843                 tun->debug = arg;
 844                 break;
 845 #endif
 846
 847         case TUNSETOFFLOAD:
 848         {
 849                 int ret;
 850                 rtnl_lock();
 851                 ret = set_offload(tun->dev, arg);
 852                 rtnl_unlock();
 853                 return ret;
 854         }
 855
 856         case SIOCGIFFLAGS:
 857                 ifr.ifr_flags = tun->if_flags;
 858                 if (copy_to_user( argp, &ifr, sizeof ifr))
 859                         return -EFAULT;
 860                 return 0;
 861
 862         case SIOCSIFFLAGS:
 863                 /** Set the character device's interface flags. Currently only
 864                  * IFF_PROMISC and IFF_ALLMULTI are used. */
 865                 tun->if_flags = ifr.ifr_flags;
 866                 DBG(KERN_INFO "%s: interface flags 0x%lx\n",
 867                                 tun->dev->name, tun->if_flags);
 868                 return 0;
 869
 870         case SIOCGIFHWADDR:
 871                 /* Note: the actual net device's address may be different */
 872                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev_addr,
 873                                 min(sizeof ifr.ifr_hwaddr.sa_data, sizeof tun->dev_addr));
 874                 if (copy_to_user( argp, &ifr, sizeof ifr))
 875                         return -EFAULT;
 876                 return 0;
 877
 878         case SIOCSIFHWADDR:
 879         {
 880                 /* try to set the actual net device's hw address */
 881                 int ret;
 882
 883                 rtnl_lock();
 884                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
 885                 rtnl_unlock();
 886
 887                 if (ret == 0) {
 888                         /** Set the character device's hardware address. This is used when
 889                          * filtering packets being sent from the network device to the character
 890                          * device. */
 891                         memcpy(tun->dev_addr, ifr.ifr_hwaddr.sa_data,
 892                                         min(sizeof ifr.ifr_hwaddr.sa_data, sizeof tun->dev_addr));
 893                         DBG(KERN_DEBUG "%s: set hardware address: %x:%x:%x:%x:%x:%x\n",
 894                                         tun->dev->name,
 895                                         tun->dev_addr[0], tun->dev_addr[1], tun->dev_addr[2],
 896                                         tun->dev_addr[3], tun->dev_addr[4], tun->dev_addr[5]);
 897                 }
 898
 899                 return  ret;
 900         }
 901
 902         case SIOCADDMULTI:
 903                 /** Add the specified group to the character device's multicast filter
 904                  * list. */
 905                 rtnl_lock();
 906                 netif_tx_lock_bh(tun->dev);
 907                 add_multi(tun->chr_filter, ifr.ifr_hwaddr.sa_data);
 908                 netif_tx_unlock_bh(tun->dev);
 909                 rtnl_unlock();
 910
 911                 DBG(KERN_DEBUG "%s: add multi: %s\n",
 912                     tun->dev->name, print_mac(mac, ifr.ifr_hwaddr.sa_data));
 913                 return 0;
 914
 915         case SIOCDELMULTI:
 916                 /** Remove the specified group from the character device's multicast
 917                  * filter list. */
 918                 rtnl_lock();
 919                 netif_tx_lock_bh(tun->dev);
 920                 del_multi(tun->chr_filter, ifr.ifr_hwaddr.sa_data);
 921                 netif_tx_unlock_bh(tun->dev);
 922                 rtnl_unlock();
 923
 924                 DBG(KERN_DEBUG "%s: del multi: %s\n",
 925                     tun->dev->name, print_mac(mac, ifr.ifr_hwaddr.sa_data));
 926                 return 0;
 927
 928         default:
 929                 return -EINVAL;
 930         };
 931
 932         return 0;
 933 }
 934
 935 static int tun_chr_fasync(int fd, struct file *file, int on)
 936 {
 937         struct tun_struct *tun = file->private_data;
 938         int ret;
 939
 940         if (!tun)
 941                 return -EBADFD;
 942
 943         DBG(KERN_INFO "%s: tun_chr_fasync %d\n", tun->dev->name, on);
 944
 945         if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
 946                 return ret;
 947
 948         if (on) {
 949                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
 950                 if (ret)
 951                         return ret;
 952                 tun->flags |= TUN_FASYNC;
 953         } else
 954                 tun->flags &= ~TUN_FASYNC;
 955
 956         return 0;
 957 }
 958
 959 static int tun_chr_open(struct inode *inode, struct file * file)
 960 {
 961         DBG1(KERN_INFO "tunX: tun_chr_open\n");
 962         file->private_data = NULL;
 963         return 0;
 964 }
 965
 966 static int tun_chr_close(struct inode *inode, struct file *file)
 967 {
 968         struct tun_struct *tun = file->private_data;
 969
 970         if (!tun)
 971                 return 0;
 972
 973         DBG(KERN_INFO "%s: tun_chr_close\n", tun->dev->name);
 974
 975         tun_chr_fasync(-1, file, 0);
 976
 977         rtnl_lock();
 978
 979         /* Detach from net device */
 980         file->private_data = NULL;
 981         tun->attached = 0;
 982         put_net(dev_net(tun->dev));
 983
 984         /* Drop read queue */
 985         skb_queue_purge(&tun->readq);
 986
 987         if (!(tun->flags & TUN_PERSIST)) {
 988                 list_del(&tun->list);
 989                 unregister_netdevice(tun->dev);
 990         }
 991
 992         rtnl_unlock();
 993
 994         return 0;
 995 }
 996
 997 static const struct file_operations tun_fops = {
 998         .owner  = THIS_MODULE,
 999         .llseek = no_llseek,
1000         .read  = do_sync_read,
1001         .aio_read  = tun_chr_aio_read,
1002         .write = do_sync_write,
1003         .aio_write = tun_chr_aio_write,
1004         .poll   = tun_chr_poll,
1005         .ioctl  = tun_chr_ioctl,
1006         .open   = tun_chr_open,
1007         .release = tun_chr_close,
1008         .fasync = tun_chr_fasync
1009 };
1010
1011 static struct miscdevice tun_miscdev = {
1012         .minor = TUN_MINOR,
1013         .name = "tun",
1014         .fops = &tun_fops,
1015 };
1016
1017 /* ethtool interface */
1018
1019 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1020 {
1021         cmd->supported          = 0;
1022         cmd->advertising        = 0;
1023         cmd->speed              = SPEED_10;
1024         cmd->duplex             = DUPLEX_FULL;
1025         cmd->port               = PORT_TP;
1026         cmd->phy_address        = 0;
1027         cmd->transceiver        = XCVR_INTERNAL;
1028         cmd->autoneg            = AUTONEG_DISABLE;
1029         cmd->maxtxpkt           = 0;
1030         cmd->maxrxpkt           = 0;
1031         return 0;
1032 }
1033
1034 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1035 {
1036         struct tun_struct *tun = netdev_priv(dev);
1037
1038         strcpy(info->driver, DRV_NAME);
1039         strcpy(info->version, DRV_VERSION);
1040         strcpy(info->fw_version, "N/A");
1041
1042         switch (tun->flags & TUN_TYPE_MASK) {
1043         case TUN_TUN_DEV:
1044                 strcpy(info->bus_info, "tun");
1045                 break;
1046         case TUN_TAP_DEV:
1047                 strcpy(info->bus_info, "tap");
1048                 break;
1049         }
1050 }
1051
1052 static u32 tun_get_msglevel(struct net_device *dev)
1053 {
1054 #ifdef TUN_DEBUG
1055         struct tun_struct *tun = netdev_priv(dev);
1056         return tun->debug;
1057 #else
1058         return -EOPNOTSUPP;
1059 #endif
1060 }
1061
1062 static void tun_set_msglevel(struct net_device *dev, u32 value)
1063 {
1064 #ifdef TUN_DEBUG
1065         struct tun_struct *tun = netdev_priv(dev);
1066         tun->debug = value;
1067 #endif
1068 }
1069
1070 static u32 tun_get_link(struct net_device *dev)
1071 {
1072         struct tun_struct *tun = netdev_priv(dev);
1073         return tun->attached;
1074 }
1075
1076 static u32 tun_get_rx_csum(struct net_device *dev)
1077 {
1078         struct tun_struct *tun = netdev_priv(dev);
1079         return (tun->flags & TUN_NOCHECKSUM) == 0;
1080 }
1081
1082 static int tun_set_rx_csum(struct net_device *dev, u32 data)
1083 {
1084         struct tun_struct *tun = netdev_priv(dev);
1085         if (data)
1086                 tun->flags &= ~TUN_NOCHECKSUM;
1087         else
1088                 tun->flags |= TUN_NOCHECKSUM;
1089         return 0;
1090 }
1091
1092 static const struct ethtool_ops tun_ethtool_ops = {
1093         .get_settings   = tun_get_settings,
1094         .get_drvinfo    = tun_get_drvinfo,
1095         .get_msglevel   = tun_get_msglevel,
1096         .set_msglevel   = tun_set_msglevel,
1097         .get_link       = tun_get_link,
1098         .get_rx_csum    = tun_get_rx_csum,
1099         .set_rx_csum    = tun_set_rx_csum
1100 };
1101
1102 static int tun_init_net(struct net *net)
1103 {
1104         struct tun_net *tn;
1105
1106         tn = kmalloc(sizeof(*tn), GFP_KERNEL);
1107         if (tn == NULL)
1108                 return -ENOMEM;
1109
1110         INIT_LIST_HEAD(&tn->dev_list);
1111
1112         if (net_assign_generic(net, tun_net_id, tn)) {
1113                 kfree(tn);
1114                 return -ENOMEM;
1115         }
1116
1117         return 0;
1118 }
1119
1120 static void tun_exit_net(struct net *net)
1121 {
1122         struct tun_net *tn;
1123         struct tun_struct *tun, *nxt;
1124
1125         tn = net_generic(net, tun_net_id);
1126
1127         rtnl_lock();
1128         list_for_each_entry_safe(tun, nxt, &tn->dev_list, list) {
1129                 DBG(KERN_INFO "%s cleaned up\n", tun->dev->name);
1130                 unregister_netdevice(tun->dev);
1131         }
1132         rtnl_unlock();
1133
1134         kfree(tn);
1135 }
1136
1137 static struct pernet_operations tun_net_ops = {
1138         .init = tun_init_net,
1139         .exit = tun_exit_net,
1140 };
1141
1142 static int __init tun_init(void)
1143 {
1144         int ret = 0;
1145
1146         printk(KERN_INFO "tun: %s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1147         printk(KERN_INFO "tun: %s\n", DRV_COPYRIGHT);
1148
1149         ret = register_pernet_gen_device(&tun_net_id, &tun_net_ops);
1150         if (ret) {
1151                 printk(KERN_ERR "tun: Can't register pernet ops\n");
1152                 goto err_pernet;
1153         }
1154
1155         ret = misc_register(&tun_miscdev);
1156         if (ret) {
1157                 printk(KERN_ERR "tun: Can't register misc device %d\n", TUN_MINOR);
1158                 goto err_misc;
1159         }
1160         return 0;
1161
1162 err_misc:
1163         unregister_pernet_gen_device(tun_net_id, &tun_net_ops);
1164 err_pernet:
1165         return ret;
1166 }
1167
1168 static void tun_cleanup(void)
1169 {
1170         misc_deregister(&tun_miscdev);
1171         unregister_pernet_gen_device(tun_net_id, &tun_net_ops);
1172 }
1173
1174 module_init(tun_init);
1175 module_exit(tun_cleanup);
1176 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1177 MODULE_AUTHOR(DRV_COPYRIGHT);
1178 MODULE_LICENSE("GPL");
1179 MODULE_ALIAS_MISCDEV(TUN_MINOR);