git.oblomov.eu Git - linux-2.6/blob - 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/smp_lock.h>
  48 #include <linux/poll.h>
  49 #include <linux/fcntl.h>
  50 #include <linux/init.h>
  51 #include <linux/skbuff.h>
  52 #include <linux/netdevice.h>
  53 #include <linux/etherdevice.h>
  54 #include <linux/miscdevice.h>
  55 #include <linux/ethtool.h>
  56 #include <linux/rtnetlink.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 <net/net_namespace.h>
  65 #include <net/netns/generic.h>
  66
  67 #include <asm/system.h>
  68 #include <asm/uaccess.h>
  69
  70 /* Uncomment to enable debugging */
  71 /* #define TUN_DEBUG 1 */
  72
  73 #ifdef TUN_DEBUG
  74 static int debug;
  75
  76 #define DBG  if(tun->debug)printk
  77 #define DBG1 if(debug==2)printk
  78 #else
  79 #define DBG( a... )
  80 #define DBG1( a... )
  81 #endif
  82
  83 #define FLT_EXACT_COUNT 8
  84 struct tap_filter {
  85         unsigned int    count;    /* Number of addrs. Zero means disabled */
  86         u32             mask[2];  /* Mask of the hashed addrs */
  87         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
  88 };
  89
  90 struct tun_file {
  91         struct tun_struct *tun;
  92 };
  93
  94 struct tun_struct {
  95         struct tun_file         *tfile;
  96         unsigned int            flags;
  97         uid_t                   owner;
  98         gid_t                   group;
  99
 100         wait_queue_head_t       read_wait;
 101         struct sk_buff_head     readq;
 102
 103         struct net_device       *dev;
 104         struct fasync_struct    *fasync;
 105
 106         struct tap_filter       txflt;
 107
 108 #ifdef TUN_DEBUG
 109         int debug;
 110 #endif
 111 };
 112
 113 static int tun_attach(struct tun_struct *tun, struct file *file)
 114 {
 115         struct tun_file *tfile = file->private_data;
 116         const struct cred *cred = current_cred();
 117
 118         ASSERT_RTNL();
 119
 120         if (tfile->tun)
 121                 return -EINVAL;
 122
 123         if (tun->tfile)
 124                 return -EBUSY;
 125
 126         /* Check permissions */
 127         if (((tun->owner != -1 && cred->euid != tun->owner) ||
 128              (tun->group != -1 && cred->egid != tun->group)) &&
 129                 !capable(CAP_NET_ADMIN))
 130                 return -EPERM;
 131
 132         tfile->tun = tun;
 133         tun->tfile = tfile;
 134         get_net(dev_net(tun->dev));
 135
 136         return 0;
 137 }
 138
 139 static void __tun_detach(struct tun_struct *tun)
 140 {
 141         struct tun_file *tfile = tun->tfile;
 142
 143         /* Detach from net device */
 144         tfile->tun = NULL;
 145         tun->tfile = NULL;
 146         put_net(dev_net(tun->dev));
 147
 148         /* Drop read queue */
 149         skb_queue_purge(&tun->readq);
 150 }
 151
 152 static struct tun_struct *__tun_get(struct tun_file *tfile)
 153 {
 154         return tfile->tun;
 155 }
 156
 157 static struct tun_struct *tun_get(struct file *file)
 158 {
 159         return __tun_get(file->private_data);
 160 }
 161
 162 static void tun_put(struct tun_struct *tun)
 163 {
 164         /* Noop for now */
 165 }
 166
 167 /* TAP filterting */
 168 static void addr_hash_set(u32 *mask, const u8 *addr)
 169 {
 170         int n = ether_crc(ETH_ALEN, addr) >> 26;
 171         mask[n >> 5] |= (1 << (n & 31));
 172 }
 173
 174 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 175 {
 176         int n = ether_crc(ETH_ALEN, addr) >> 26;
 177         return mask[n >> 5] & (1 << (n & 31));
 178 }
 179
 180 static int update_filter(struct tap_filter *filter, void __user *arg)
 181 {
 182         struct { u8 u[ETH_ALEN]; } *addr;
 183         struct tun_filter uf;
 184         int err, alen, n, nexact;
 185
 186         if (copy_from_user(&uf, arg, sizeof(uf)))
 187                 return -EFAULT;
 188
 189         if (!uf.count) {
 190                 /* Disabled */
 191                 filter->count = 0;
 192                 return 0;
 193         }
 194
 195         alen = ETH_ALEN * uf.count;
 196         addr = kmalloc(alen, GFP_KERNEL);
 197         if (!addr)
 198                 return -ENOMEM;
 199
 200         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 201                 err = -EFAULT;
 202                 goto done;
 203         }
 204
 205         /* The filter is updated without holding any locks. Which is
 206          * perfectly safe. We disable it first and in the worst
 207          * case we'll accept a few undesired packets. */
 208         filter->count = 0;
 209         wmb();
 210
 211         /* Use first set of addresses as an exact filter */
 212         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 213                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 214
 215         nexact = n;
 216
 217         /* The rest is hashed */
 218         memset(filter->mask, 0, sizeof(filter->mask));
 219         for (; n < uf.count; n++)
 220                 addr_hash_set(filter->mask, addr[n].u);
 221
 222         /* For ALLMULTI just set the mask to all ones.
 223          * This overrides the mask populated above. */
 224         if ((uf.flags & TUN_FLT_ALLMULTI))
 225                 memset(filter->mask, ~0, sizeof(filter->mask));
 226
 227         /* Now enable the filter */
 228         wmb();
 229         filter->count = nexact;
 230
 231         /* Return the number of exact filters */
 232         err = nexact;
 233
 234 done:
 235         kfree(addr);
 236         return err;
 237 }
 238
 239 /* Returns: 0 - drop, !=0 - accept */
 240 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 241 {
 242         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 243          * at this point. */
 244         struct ethhdr *eh = (struct ethhdr *) skb->data;
 245         int i;
 246
 247         /* Exact match */
 248         for (i = 0; i < filter->count; i++)
 249                 if (!compare_ether_addr(eh->h_dest, filter->addr[i]))
 250                         return 1;
 251
 252         /* Inexact match (multicast only) */
 253         if (is_multicast_ether_addr(eh->h_dest))
 254                 return addr_hash_test(filter->mask, eh->h_dest);
 255
 256         return 0;
 257 }
 258
 259 /*
 260  * Checks whether the packet is accepted or not.
 261  * Returns: 0 - drop, !=0 - accept
 262  */
 263 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 264 {
 265         if (!filter->count)
 266                 return 1;
 267
 268         return run_filter(filter, skb);
 269 }
 270
 271 /* Network device part of the driver */
 272
 273 static const struct ethtool_ops tun_ethtool_ops;
 274
 275 /* Net device open. */
 276 static int tun_net_open(struct net_device *dev)
 277 {
 278         netif_start_queue(dev);
 279         return 0;
 280 }
 281
 282 /* Net device close. */
 283 static int tun_net_close(struct net_device *dev)
 284 {
 285         netif_stop_queue(dev);
 286         return 0;
 287 }
 288
 289 /* Net device start xmit */
 290 static int tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 291 {
 292         struct tun_struct *tun = netdev_priv(dev);
 293
 294         DBG(KERN_INFO "%s: tun_net_xmit %d\n", tun->dev->name, skb->len);
 295
 296         /* Drop packet if interface is not attached */
 297         if (!tun->tfile)
 298                 goto drop;
 299
 300         /* Drop if the filter does not like it.
 301          * This is a noop if the filter is disabled.
 302          * Filter can be enabled only for the TAP devices. */
 303         if (!check_filter(&tun->txflt, skb))
 304                 goto drop;
 305
 306         if (skb_queue_len(&tun->readq) >= dev->tx_queue_len) {
 307                 if (!(tun->flags & TUN_ONE_QUEUE)) {
 308                         /* Normal queueing mode. */
 309                         /* Packet scheduler handles dropping of further packets. */
 310                         netif_stop_queue(dev);
 311
 312                         /* We won't see all dropped packets individually, so overrun
 313                          * error is more appropriate. */
 314                         dev->stats.tx_fifo_errors++;
 315                 } else {
 316                         /* Single queue mode.
 317                          * Driver handles dropping of all packets itself. */
 318                         goto drop;
 319                 }
 320         }
 321
 322         /* Enqueue packet */
 323         skb_queue_tail(&tun->readq, skb);
 324         dev->trans_start = jiffies;
 325
 326         /* Notify and wake up reader process */
 327         if (tun->flags & TUN_FASYNC)
 328                 kill_fasync(&tun->fasync, SIGIO, POLL_IN);
 329         wake_up_interruptible(&tun->read_wait);
 330         return 0;
 331
 332 drop:
 333         dev->stats.tx_dropped++;
 334         kfree_skb(skb);
 335         return 0;
 336 }
 337
 338 static void tun_net_mclist(struct net_device *dev)
 339 {
 340         /*
 341          * This callback is supposed to deal with mc filter in
 342          * _rx_ path and has nothing to do with the _tx_ path.
 343          * In rx path we always accept everything userspace gives us.
 344          */
 345         return;
 346 }
 347
 348 #define MIN_MTU 68
 349 #define MAX_MTU 65535
 350
 351 static int
 352 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 353 {
 354         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 355                 return -EINVAL;
 356         dev->mtu = new_mtu;
 357         return 0;
 358 }
 359
 360 static const struct net_device_ops tun_netdev_ops = {
 361         .ndo_open               = tun_net_open,
 362         .ndo_stop               = tun_net_close,
 363         .ndo_start_xmit         = tun_net_xmit,
 364         .ndo_change_mtu         = tun_net_change_mtu,
 365 };
 366
 367 static const struct net_device_ops tap_netdev_ops = {
 368         .ndo_open               = tun_net_open,
 369         .ndo_stop               = tun_net_close,
 370         .ndo_start_xmit         = tun_net_xmit,
 371         .ndo_change_mtu         = tun_net_change_mtu,
 372         .ndo_set_multicast_list = tun_net_mclist,
 373         .ndo_set_mac_address    = eth_mac_addr,
 374         .ndo_validate_addr      = eth_validate_addr,
 375 };
 376
 377 /* Initialize net device. */
 378 static void tun_net_init(struct net_device *dev)
 379 {
 380         struct tun_struct *tun = netdev_priv(dev);
 381
 382         switch (tun->flags & TUN_TYPE_MASK) {
 383         case TUN_TUN_DEV:
 384                 dev->netdev_ops = &tun_netdev_ops;
 385
 386                 /* Point-to-Point TUN Device */
 387                 dev->hard_header_len = 0;
 388                 dev->addr_len = 0;
 389                 dev->mtu = 1500;
 390
 391                 /* Zero header length */
 392                 dev->type = ARPHRD_NONE;
 393                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 394                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 395                 break;
 396
 397         case TUN_TAP_DEV:
 398                 dev->netdev_ops = &tap_netdev_ops;
 399                 /* Ethernet TAP Device */
 400                 ether_setup(dev);
 401
 402                 random_ether_addr(dev->dev_addr);
 403
 404                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 405                 break;
 406         }
 407 }
 408
 409 /* Character device part */
 410
 411 /* Poll */
 412 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
 413 {
 414         struct tun_struct *tun = tun_get(file);
 415         unsigned int mask = POLLOUT | POLLWRNORM;
 416
 417         if (!tun)
 418                 return POLLERR;
 419
 420         DBG(KERN_INFO "%s: tun_chr_poll\n", tun->dev->name);
 421
 422         poll_wait(file, &tun->read_wait, wait);
 423
 424         if (!skb_queue_empty(&tun->readq))
 425                 mask |= POLLIN | POLLRDNORM;
 426
 427         tun_put(tun);
 428         return mask;
 429 }
 430
 431 /* prepad is the amount to reserve at front.  len is length after that.
 432  * linear is a hint as to how much to copy (usually headers). */
 433 static struct sk_buff *tun_alloc_skb(size_t prepad, size_t len, size_t linear,
 434                                      gfp_t gfp)
 435 {
 436         struct sk_buff *skb;
 437         unsigned int i;
 438
 439         skb = alloc_skb(prepad + len, gfp|__GFP_NOWARN);
 440         if (skb) {
 441                 skb_reserve(skb, prepad);
 442                 skb_put(skb, len);
 443                 return skb;
 444         }
 445
 446         /* Under a page?  Don't bother with paged skb. */
 447         if (prepad + len < PAGE_SIZE)
 448                 return NULL;
 449
 450         /* Start with a normal skb, and add pages. */
 451         skb = alloc_skb(prepad + linear, gfp);
 452         if (!skb)
 453                 return NULL;
 454
 455         skb_reserve(skb, prepad);
 456         skb_put(skb, linear);
 457
 458         len -= linear;
 459
 460         for (i = 0; i < MAX_SKB_FRAGS; i++) {
 461                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 462
 463                 f->page = alloc_page(gfp|__GFP_ZERO);
 464                 if (!f->page)
 465                         break;
 466
 467                 f->page_offset = 0;
 468                 f->size = PAGE_SIZE;
 469
 470                 skb->data_len += PAGE_SIZE;
 471                 skb->len += PAGE_SIZE;
 472                 skb->truesize += PAGE_SIZE;
 473                 skb_shinfo(skb)->nr_frags++;
 474
 475                 if (len < PAGE_SIZE) {
 476                         len = 0;
 477                         break;
 478                 }
 479                 len -= PAGE_SIZE;
 480         }
 481
 482         /* Too large, or alloc fail? */
 483         if (unlikely(len)) {
 484                 kfree_skb(skb);
 485                 skb = NULL;
 486         }
 487
 488         return skb;
 489 }
 490
 491 /* Get packet from user space buffer */
 492 static __inline__ ssize_t tun_get_user(struct tun_struct *tun, struct iovec *iv, size_t count)
 493 {
 494         struct tun_pi pi = { 0, __constant_htons(ETH_P_IP) };
 495         struct sk_buff *skb;
 496         size_t len = count, align = 0;
 497         struct virtio_net_hdr gso = { 0 };
 498
 499         if (!(tun->flags & TUN_NO_PI)) {
 500                 if ((len -= sizeof(pi)) > count)
 501                         return -EINVAL;
 502
 503                 if(memcpy_fromiovec((void *)&pi, iv, sizeof(pi)))
 504                         return -EFAULT;
 505         }
 506
 507         if (tun->flags & TUN_VNET_HDR) {
 508                 if ((len -= sizeof(gso)) > count)
 509                         return -EINVAL;
 510
 511                 if (memcpy_fromiovec((void *)&gso, iv, sizeof(gso)))
 512                         return -EFAULT;
 513
 514                 if (gso.hdr_len > len)
 515                         return -EINVAL;
 516         }
 517
 518         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
 519                 align = NET_IP_ALIGN;
 520                 if (unlikely(len < ETH_HLEN))
 521                         return -EINVAL;
 522         }
 523
 524         if (!(skb = tun_alloc_skb(align, len, gso.hdr_len, GFP_KERNEL))) {
 525                 tun->dev->stats.rx_dropped++;
 526                 return -ENOMEM;
 527         }
 528
 529         if (skb_copy_datagram_from_iovec(skb, 0, iv, len)) {
 530                 tun->dev->stats.rx_dropped++;
 531                 kfree_skb(skb);
 532                 return -EFAULT;
 533         }
 534
 535         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 536                 if (!skb_partial_csum_set(skb, gso.csum_start,
 537                                           gso.csum_offset)) {
 538                         tun->dev->stats.rx_frame_errors++;
 539                         kfree_skb(skb);
 540                         return -EINVAL;
 541                 }
 542         } else if (tun->flags & TUN_NOCHECKSUM)
 543                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 544
 545         switch (tun->flags & TUN_TYPE_MASK) {
 546         case TUN_TUN_DEV:
 547                 if (tun->flags & TUN_NO_PI) {
 548                         switch (skb->data[0] & 0xf0) {
 549                         case 0x40:
 550                                 pi.proto = htons(ETH_P_IP);
 551                                 break;
 552                         case 0x60:
 553                                 pi.proto = htons(ETH_P_IPV6);
 554                                 break;
 555                         default:
 556                                 tun->dev->stats.rx_dropped++;
 557                                 kfree_skb(skb);
 558                                 return -EINVAL;
 559                         }
 560                 }
 561
 562                 skb_reset_mac_header(skb);
 563                 skb->protocol = pi.proto;
 564                 skb->dev = tun->dev;
 565                 break;
 566         case TUN_TAP_DEV:
 567                 skb->protocol = eth_type_trans(skb, tun->dev);
 568                 break;
 569         };
 570
 571         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 572                 pr_debug("GSO!\n");
 573                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 574                 case VIRTIO_NET_HDR_GSO_TCPV4:
 575                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 576                         break;
 577                 case VIRTIO_NET_HDR_GSO_TCPV6:
 578                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 579                         break;
 580                 default:
 581                         tun->dev->stats.rx_frame_errors++;
 582                         kfree_skb(skb);
 583                         return -EINVAL;
 584                 }
 585
 586                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
 587                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 588
 589                 skb_shinfo(skb)->gso_size = gso.gso_size;
 590                 if (skb_shinfo(skb)->gso_size == 0) {
 591                         tun->dev->stats.rx_frame_errors++;
 592                         kfree_skb(skb);
 593                         return -EINVAL;
 594                 }
 595
 596                 /* Header must be checked, and gso_segs computed. */
 597                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 598                 skb_shinfo(skb)->gso_segs = 0;
 599         }
 600
 601         netif_rx_ni(skb);
 602
 603         tun->dev->stats.rx_packets++;
 604         tun->dev->stats.rx_bytes += len;
 605
 606         return count;
 607 }
 608
 609 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
 610                               unsigned long count, loff_t pos)
 611 {
 612         struct tun_struct *tun = tun_get(iocb->ki_filp);
 613         ssize_t result;
 614
 615         if (!tun)
 616                 return -EBADFD;
 617
 618         DBG(KERN_INFO "%s: tun_chr_write %ld\n", tun->dev->name, count);
 619
 620         result = tun_get_user(tun, (struct iovec *) iv, iov_length(iv, count));
 621
 622         tun_put(tun);
 623         return result;
 624 }
 625
 626 /* Put packet to the user space buffer */
 627 static __inline__ ssize_t tun_put_user(struct tun_struct *tun,
 628                                        struct sk_buff *skb,
 629                                        struct iovec *iv, int len)
 630 {
 631         struct tun_pi pi = { 0, skb->protocol };
 632         ssize_t total = 0;
 633
 634         if (!(tun->flags & TUN_NO_PI)) {
 635                 if ((len -= sizeof(pi)) < 0)
 636                         return -EINVAL;
 637
 638                 if (len < skb->len) {
 639                         /* Packet will be striped */
 640                         pi.flags |= TUN_PKT_STRIP;
 641                 }
 642
 643                 if (memcpy_toiovec(iv, (void *) &pi, sizeof(pi)))
 644                         return -EFAULT;
 645                 total += sizeof(pi);
 646         }
 647
 648         if (tun->flags & TUN_VNET_HDR) {
 649                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
 650                 if ((len -= sizeof(gso)) < 0)
 651                         return -EINVAL;
 652
 653                 if (skb_is_gso(skb)) {
 654                         struct skb_shared_info *sinfo = skb_shinfo(skb);
 655
 656                         /* This is a hint as to how much should be linear. */
 657                         gso.hdr_len = skb_headlen(skb);
 658                         gso.gso_size = sinfo->gso_size;
 659                         if (sinfo->gso_type & SKB_GSO_TCPV4)
 660                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 661                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
 662                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 663                         else
 664                                 BUG();
 665                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
 666                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 667                 } else
 668                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
 669
 670                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 671                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 672                         gso.csum_start = skb->csum_start - skb_headroom(skb);
 673                         gso.csum_offset = skb->csum_offset;
 674                 } /* else everything is zero */
 675
 676                 if (unlikely(memcpy_toiovec(iv, (void *)&gso, sizeof(gso))))
 677                         return -EFAULT;
 678                 total += sizeof(gso);
 679         }
 680
 681         len = min_t(int, skb->len, len);
 682
 683         skb_copy_datagram_iovec(skb, 0, iv, len);
 684         total += len;
 685
 686         tun->dev->stats.tx_packets++;
 687         tun->dev->stats.tx_bytes += len;
 688
 689         return total;
 690 }
 691
 692 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
 693                             unsigned long count, loff_t pos)
 694 {
 695         struct file *file = iocb->ki_filp;
 696         struct tun_struct *tun = tun_get(file);
 697         DECLARE_WAITQUEUE(wait, current);
 698         struct sk_buff *skb;
 699         ssize_t len, ret = 0;
 700
 701         if (!tun)
 702                 return -EBADFD;
 703
 704         DBG(KERN_INFO "%s: tun_chr_read\n", tun->dev->name);
 705
 706         len = iov_length(iv, count);
 707         if (len < 0) {
 708                 ret = -EINVAL;
 709                 goto out;
 710         }
 711
 712         add_wait_queue(&tun->read_wait, &wait);
 713         while (len) {
 714                 current->state = TASK_INTERRUPTIBLE;
 715
 716                 /* Read frames from the queue */
 717                 if (!(skb=skb_dequeue(&tun->readq))) {
 718                         if (file->f_flags & O_NONBLOCK) {
 719                                 ret = -EAGAIN;
 720                                 break;
 721                         }
 722                         if (signal_pending(current)) {
 723                                 ret = -ERESTARTSYS;
 724                                 break;
 725                         }
 726
 727                         /* Nothing to read, let's sleep */
 728                         schedule();
 729                         continue;
 730                 }
 731                 netif_wake_queue(tun->dev);
 732
 733                 ret = tun_put_user(tun, skb, (struct iovec *) iv, len);
 734                 kfree_skb(skb);
 735                 break;
 736         }
 737
 738         current->state = TASK_RUNNING;
 739         remove_wait_queue(&tun->read_wait, &wait);
 740
 741 out:
 742         tun_put(tun);
 743         return ret;
 744 }
 745
 746 static void tun_setup(struct net_device *dev)
 747 {
 748         struct tun_struct *tun = netdev_priv(dev);
 749
 750         skb_queue_head_init(&tun->readq);
 751         init_waitqueue_head(&tun->read_wait);
 752
 753         tun->owner = -1;
 754         tun->group = -1;
 755
 756         dev->ethtool_ops = &tun_ethtool_ops;
 757         dev->destructor = free_netdev;
 758         dev->features |= NETIF_F_NETNS_LOCAL;
 759 }
 760
 761 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
 762 {
 763         struct tun_struct *tun;
 764         struct net_device *dev;
 765         int err;
 766
 767         dev = __dev_get_by_name(net, ifr->ifr_name);
 768         if (dev) {
 769                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
 770                         tun = netdev_priv(dev);
 771                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
 772                         tun = netdev_priv(dev);
 773                 else
 774                         return -EINVAL;
 775
 776                 err = tun_attach(tun, file);
 777                 if (err < 0)
 778                         return err;
 779         }
 780         else {
 781                 char *name;
 782                 unsigned long flags = 0;
 783
 784                 err = -EINVAL;
 785
 786                 if (!capable(CAP_NET_ADMIN))
 787                         return -EPERM;
 788
 789                 /* Set dev type */
 790                 if (ifr->ifr_flags & IFF_TUN) {
 791                         /* TUN device */
 792                         flags |= TUN_TUN_DEV;
 793                         name = "tun%d";
 794                 } else if (ifr->ifr_flags & IFF_TAP) {
 795                         /* TAP device */
 796                         flags |= TUN_TAP_DEV;
 797                         name = "tap%d";
 798                 } else
 799                         goto failed;
 800
 801                 if (*ifr->ifr_name)
 802                         name = ifr->ifr_name;
 803
 804                 dev = alloc_netdev(sizeof(struct tun_struct), name,
 805                                    tun_setup);
 806                 if (!dev)
 807                         return -ENOMEM;
 808
 809                 dev_net_set(dev, net);
 810
 811                 tun = netdev_priv(dev);
 812                 tun->dev = dev;
 813                 tun->flags = flags;
 814                 tun->txflt.count = 0;
 815
 816                 tun_net_init(dev);
 817
 818                 if (strchr(dev->name, '%')) {
 819                         err = dev_alloc_name(dev, dev->name);
 820                         if (err < 0)
 821                                 goto err_free_dev;
 822                 }
 823
 824                 err = register_netdevice(tun->dev);
 825                 if (err < 0)
 826                         goto err_free_dev;
 827
 828                 err = tun_attach(tun, file);
 829                 if (err < 0)
 830                         goto err_free_dev;
 831         }
 832
 833         DBG(KERN_INFO "%s: tun_set_iff\n", tun->dev->name);
 834
 835         if (ifr->ifr_flags & IFF_NO_PI)
 836                 tun->flags |= TUN_NO_PI;
 837         else
 838                 tun->flags &= ~TUN_NO_PI;
 839
 840         if (ifr->ifr_flags & IFF_ONE_QUEUE)
 841                 tun->flags |= TUN_ONE_QUEUE;
 842         else
 843                 tun->flags &= ~TUN_ONE_QUEUE;
 844
 845         if (ifr->ifr_flags & IFF_VNET_HDR)
 846                 tun->flags |= TUN_VNET_HDR;
 847         else
 848                 tun->flags &= ~TUN_VNET_HDR;
 849
 850         /* Make sure persistent devices do not get stuck in
 851          * xoff state.
 852          */
 853         if (netif_running(tun->dev))
 854                 netif_wake_queue(tun->dev);
 855
 856         strcpy(ifr->ifr_name, tun->dev->name);
 857         return 0;
 858
 859  err_free_dev:
 860         free_netdev(dev);
 861  failed:
 862         return err;
 863 }
 864
 865 static int tun_get_iff(struct net *net, struct file *file, struct ifreq *ifr)
 866 {
 867         struct tun_struct *tun = tun_get(file);
 868
 869         if (!tun)
 870                 return -EBADFD;
 871
 872         DBG(KERN_INFO "%s: tun_get_iff\n", tun->dev->name);
 873
 874         strcpy(ifr->ifr_name, tun->dev->name);
 875
 876         ifr->ifr_flags = 0;
 877
 878         if (ifr->ifr_flags & TUN_TUN_DEV)
 879                 ifr->ifr_flags |= IFF_TUN;
 880         else
 881                 ifr->ifr_flags |= IFF_TAP;
 882
 883         if (tun->flags & TUN_NO_PI)
 884                 ifr->ifr_flags |= IFF_NO_PI;
 885
 886         if (tun->flags & TUN_ONE_QUEUE)
 887                 ifr->ifr_flags |= IFF_ONE_QUEUE;
 888
 889         if (tun->flags & TUN_VNET_HDR)
 890                 ifr->ifr_flags |= IFF_VNET_HDR;
 891
 892         tun_put(tun);
 893         return 0;
 894 }
 895
 896 /* This is like a cut-down ethtool ops, except done via tun fd so no
 897  * privs required. */
 898 static int set_offload(struct net_device *dev, unsigned long arg)
 899 {
 900         unsigned int old_features, features;
 901
 902         old_features = dev->features;
 903         /* Unset features, set them as we chew on the arg. */
 904         features = (old_features & ~(NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST
 905                                     |NETIF_F_TSO_ECN|NETIF_F_TSO|NETIF_F_TSO6));
 906
 907         if (arg & TUN_F_CSUM) {
 908                 features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
 909                 arg &= ~TUN_F_CSUM;
 910
 911                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
 912                         if (arg & TUN_F_TSO_ECN) {
 913                                 features |= NETIF_F_TSO_ECN;
 914                                 arg &= ~TUN_F_TSO_ECN;
 915                         }
 916                         if (arg & TUN_F_TSO4)
 917                                 features |= NETIF_F_TSO;
 918                         if (arg & TUN_F_TSO6)
 919                                 features |= NETIF_F_TSO6;
 920                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
 921                 }
 922         }
 923
 924         /* This gives the user a way to test for new features in future by
 925          * trying to set them. */
 926         if (arg)
 927                 return -EINVAL;
 928
 929         dev->features = features;
 930         if (old_features != dev->features)
 931                 netdev_features_change(dev);
 932
 933         return 0;
 934 }
 935
 936 static int tun_chr_ioctl(struct inode *inode, struct file *file,
 937                          unsigned int cmd, unsigned long arg)
 938 {
 939         struct tun_struct *tun;
 940         void __user* argp = (void __user*)arg;
 941         struct ifreq ifr;
 942         int ret;
 943
 944         if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
 945                 if (copy_from_user(&ifr, argp, sizeof ifr))
 946                         return -EFAULT;
 947
 948         if (cmd == TUNGETFEATURES) {
 949                 /* Currently this just means: "what IFF flags are valid?".
 950                  * This is needed because we never checked for invalid flags on
 951                  * TUNSETIFF. */
 952                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
 953                                 IFF_VNET_HDR,
 954                                 (unsigned int __user*)argp);
 955         }
 956
 957         tun = tun_get(file);
 958         if (cmd == TUNSETIFF && !tun) {
 959                 int err;
 960
 961                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
 962
 963                 rtnl_lock();
 964                 err = tun_set_iff(current->nsproxy->net_ns, file, &ifr);
 965                 rtnl_unlock();
 966
 967                 if (err)
 968                         return err;
 969
 970                 if (copy_to_user(argp, &ifr, sizeof(ifr)))
 971                         return -EFAULT;
 972                 return 0;
 973         }
 974
 975
 976         if (!tun)
 977                 return -EBADFD;
 978
 979         DBG(KERN_INFO "%s: tun_chr_ioctl cmd %d\n", tun->dev->name, cmd);
 980
 981         ret = 0;
 982         switch (cmd) {
 983         case TUNGETIFF:
 984                 ret = tun_get_iff(current->nsproxy->net_ns, file, &ifr);
 985                 if (ret)
 986                         break;
 987
 988                 if (copy_to_user(argp, &ifr, sizeof(ifr)))
 989                         ret = -EFAULT;
 990                 break;
 991
 992         case TUNSETNOCSUM:
 993                 /* Disable/Enable checksum */
 994                 if (arg)
 995                         tun->flags |= TUN_NOCHECKSUM;
 996                 else
 997                         tun->flags &= ~TUN_NOCHECKSUM;
 998
 999                 DBG(KERN_INFO "%s: checksum %s\n",
1000                     tun->dev->name, arg ? "disabled" : "enabled");
1001                 break;
1002
1003         case TUNSETPERSIST:
1004                 /* Disable/Enable persist mode */
1005                 if (arg)
1006                         tun->flags |= TUN_PERSIST;
1007                 else
1008                         tun->flags &= ~TUN_PERSIST;
1009
1010                 DBG(KERN_INFO "%s: persist %s\n",
1011                     tun->dev->name, arg ? "enabled" : "disabled");
1012                 break;
1013
1014         case TUNSETOWNER:
1015                 /* Set owner of the device */
1016                 tun->owner = (uid_t) arg;
1017
1018                 DBG(KERN_INFO "%s: owner set to %d\n", tun->dev->name, tun->owner);
1019                 break;
1020
1021         case TUNSETGROUP:
1022                 /* Set group of the device */
1023                 tun->group= (gid_t) arg;
1024
1025                 DBG(KERN_INFO "%s: group set to %d\n", tun->dev->name, tun->group);
1026                 break;
1027
1028         case TUNSETLINK:
1029                 /* Only allow setting the type when the interface is down */
1030                 rtnl_lock();
1031                 if (tun->dev->flags & IFF_UP) {
1032                         DBG(KERN_INFO "%s: Linktype set failed because interface is up\n",
1033                                 tun->dev->name);
1034                         ret = -EBUSY;
1035                 } else {
1036                         tun->dev->type = (int) arg;
1037                         DBG(KERN_INFO "%s: linktype set to %d\n", tun->dev->name, tun->dev->type);
1038                         ret = 0;
1039                 }
1040                 rtnl_unlock();
1041                 break;
1042
1043 #ifdef TUN_DEBUG
1044         case TUNSETDEBUG:
1045                 tun->debug = arg;
1046                 break;
1047 #endif
1048         case TUNSETOFFLOAD:
1049                 rtnl_lock();
1050                 ret = set_offload(tun->dev, arg);
1051                 rtnl_unlock();
1052                 break;
1053
1054         case TUNSETTXFILTER:
1055                 /* Can be set only for TAPs */
1056                 ret = -EINVAL;
1057                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1058                         break;
1059                 rtnl_lock();
1060                 ret = update_filter(&tun->txflt, (void __user *)arg);
1061                 rtnl_unlock();
1062                 break;
1063
1064         case SIOCGIFHWADDR:
1065                 /* Get hw addres */
1066                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1067                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1068                 if (copy_to_user(argp, &ifr, sizeof ifr))
1069                         ret = -EFAULT;
1070                 break;
1071
1072         case SIOCSIFHWADDR:
1073                 /* Set hw address */
1074                 DBG(KERN_DEBUG "%s: set hw address: %pM\n",
1075                         tun->dev->name, ifr.ifr_hwaddr.sa_data);
1076
1077                 rtnl_lock();
1078                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1079                 rtnl_unlock();
1080                 break;
1081         default:
1082                 ret = -EINVAL;
1083                 break;
1084         };
1085
1086         tun_put(tun);
1087         return ret;
1088 }
1089
1090 static int tun_chr_fasync(int fd, struct file *file, int on)
1091 {
1092         struct tun_struct *tun = tun_get(file);
1093         int ret;
1094
1095         if (!tun)
1096                 return -EBADFD;
1097
1098         DBG(KERN_INFO "%s: tun_chr_fasync %d\n", tun->dev->name, on);
1099
1100         lock_kernel();
1101         if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1102                 goto out;
1103
1104         if (on) {
1105                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1106                 if (ret)
1107                         goto out;
1108                 tun->flags |= TUN_FASYNC;
1109         } else
1110                 tun->flags &= ~TUN_FASYNC;
1111         ret = 0;
1112 out:
1113         unlock_kernel();
1114         tun_put(tun);
1115         return ret;
1116 }
1117
1118 static int tun_chr_open(struct inode *inode, struct file * file)
1119 {
1120         struct tun_file *tfile;
1121         cycle_kernel_lock();
1122         DBG1(KERN_INFO "tunX: tun_chr_open\n");
1123
1124         tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1125         if (!tfile)
1126                 return -ENOMEM;
1127         tfile->tun = NULL;
1128         file->private_data = tfile;
1129         return 0;
1130 }
1131
1132 static int tun_chr_close(struct inode *inode, struct file *file)
1133 {
1134         struct tun_file *tfile = file->private_data;
1135         struct tun_struct *tun = __tun_get(tfile);
1136
1137
1138         if (tun) {
1139                 DBG(KERN_INFO "%s: tun_chr_close\n", tun->dev->name);
1140
1141                 rtnl_lock();
1142                 __tun_detach(tun);
1143
1144                 /* If desireable, unregister the netdevice. */
1145                 if (!(tun->flags & TUN_PERSIST))
1146                         unregister_netdevice(tun->dev);
1147
1148                 rtnl_unlock();
1149         }
1150
1151         kfree(tfile);
1152
1153         return 0;
1154 }
1155
1156 static const struct file_operations tun_fops = {
1157         .owner  = THIS_MODULE,
1158         .llseek = no_llseek,
1159         .read  = do_sync_read,
1160         .aio_read  = tun_chr_aio_read,
1161         .write = do_sync_write,
1162         .aio_write = tun_chr_aio_write,
1163         .poll   = tun_chr_poll,
1164         .ioctl  = tun_chr_ioctl,
1165         .open   = tun_chr_open,
1166         .release = tun_chr_close,
1167         .fasync = tun_chr_fasync
1168 };
1169
1170 static struct miscdevice tun_miscdev = {
1171         .minor = TUN_MINOR,
1172         .name = "tun",
1173         .fops = &tun_fops,
1174 };
1175
1176 /* ethtool interface */
1177
1178 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1179 {
1180         cmd->supported          = 0;
1181         cmd->advertising        = 0;
1182         cmd->speed              = SPEED_10;
1183         cmd->duplex             = DUPLEX_FULL;
1184         cmd->port               = PORT_TP;
1185         cmd->phy_address        = 0;
1186         cmd->transceiver        = XCVR_INTERNAL;
1187         cmd->autoneg            = AUTONEG_DISABLE;
1188         cmd->maxtxpkt           = 0;
1189         cmd->maxrxpkt           = 0;
1190         return 0;
1191 }
1192
1193 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1194 {
1195         struct tun_struct *tun = netdev_priv(dev);
1196
1197         strcpy(info->driver, DRV_NAME);
1198         strcpy(info->version, DRV_VERSION);
1199         strcpy(info->fw_version, "N/A");
1200
1201         switch (tun->flags & TUN_TYPE_MASK) {
1202         case TUN_TUN_DEV:
1203                 strcpy(info->bus_info, "tun");
1204                 break;
1205         case TUN_TAP_DEV:
1206                 strcpy(info->bus_info, "tap");
1207                 break;
1208         }
1209 }
1210
1211 static u32 tun_get_msglevel(struct net_device *dev)
1212 {
1213 #ifdef TUN_DEBUG
1214         struct tun_struct *tun = netdev_priv(dev);
1215         return tun->debug;
1216 #else
1217         return -EOPNOTSUPP;
1218 #endif
1219 }
1220
1221 static void tun_set_msglevel(struct net_device *dev, u32 value)
1222 {
1223 #ifdef TUN_DEBUG
1224         struct tun_struct *tun = netdev_priv(dev);
1225         tun->debug = value;
1226 #endif
1227 }
1228
1229 static u32 tun_get_link(struct net_device *dev)
1230 {
1231         struct tun_struct *tun = netdev_priv(dev);
1232         return !!tun->tfile;
1233 }
1234
1235 static u32 tun_get_rx_csum(struct net_device *dev)
1236 {
1237         struct tun_struct *tun = netdev_priv(dev);
1238         return (tun->flags & TUN_NOCHECKSUM) == 0;
1239 }
1240
1241 static int tun_set_rx_csum(struct net_device *dev, u32 data)
1242 {
1243         struct tun_struct *tun = netdev_priv(dev);
1244         if (data)
1245                 tun->flags &= ~TUN_NOCHECKSUM;
1246         else
1247                 tun->flags |= TUN_NOCHECKSUM;
1248         return 0;
1249 }
1250
1251 static const struct ethtool_ops tun_ethtool_ops = {
1252         .get_settings   = tun_get_settings,
1253         .get_drvinfo    = tun_get_drvinfo,
1254         .get_msglevel   = tun_get_msglevel,
1255         .set_msglevel   = tun_set_msglevel,
1256         .get_link       = tun_get_link,
1257         .get_rx_csum    = tun_get_rx_csum,
1258         .set_rx_csum    = tun_set_rx_csum
1259 };
1260
1261 static int tun_init_net(struct net *net)
1262 {
1263         return 0;
1264 }
1265
1266 static void tun_exit_net(struct net *net)
1267 {
1268         struct net_device *dev, *next;
1269
1270         rtnl_lock();
1271         for_each_netdev_safe(net, dev, next) {
1272                 if (dev->ethtool_ops != &tun_ethtool_ops)
1273                         continue;
1274                 DBG(KERN_INFO "%s cleaned up\n", dev->name);
1275                 unregister_netdevice(dev);
1276         }
1277         rtnl_unlock();
1278 }
1279
1280 static struct pernet_operations tun_net_ops = {
1281         .init = tun_init_net,
1282         .exit = tun_exit_net,
1283 };
1284
1285 static int __init tun_init(void)
1286 {
1287         int ret = 0;
1288
1289         printk(KERN_INFO "tun: %s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1290         printk(KERN_INFO "tun: %s\n", DRV_COPYRIGHT);
1291
1292         ret = register_pernet_device(&tun_net_ops);
1293         if (ret) {
1294                 printk(KERN_ERR "tun: Can't register pernet ops\n");
1295                 goto err_pernet;
1296         }
1297
1298         ret = misc_register(&tun_miscdev);
1299         if (ret) {
1300                 printk(KERN_ERR "tun: Can't register misc device %d\n", TUN_MINOR);
1301                 goto err_misc;
1302         }
1303         return 0;
1304
1305 err_misc:
1306         unregister_pernet_device(&tun_net_ops);
1307 err_pernet:
1308         return ret;
1309 }
1310
1311 static void tun_cleanup(void)
1312 {
1313         misc_deregister(&tun_miscdev);
1314         unregister_pernet_device(&tun_net_ops);
1315 }
1316
1317 module_init(tun_init);
1318 module_exit(tun_cleanup);
1319 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1320 MODULE_AUTHOR(DRV_COPYRIGHT);
1321 MODULE_LICENSE("GPL");
1322 MODULE_ALIAS_MISCDEV(TUN_MINOR);