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 #include <net/rtnetlink.h>
  67
  68 #include <asm/system.h>
  69 #include <asm/uaccess.h>
  70
  71 /* Uncomment to enable debugging */
  72 /* #define TUN_DEBUG 1 */
  73
  74 #ifdef TUN_DEBUG
  75 static int debug;
  76
  77 #define DBG  if(tun->debug)printk
  78 #define DBG1 if(debug==2)printk
  79 #else
  80 #define DBG( a... )
  81 #define DBG1( a... )
  82 #endif
  83
  84 #define FLT_EXACT_COUNT 8
  85 struct tap_filter {
  86         unsigned int    count;    /* Number of addrs. Zero means disabled */
  87         u32             mask[2];  /* Mask of the hashed addrs */
  88         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
  89 };
  90
  91 struct tun_file {
  92         atomic_t count;
  93         struct tun_struct *tun;
  94         struct net *net;
  95         wait_queue_head_t       read_wait;
  96 };
  97
  98 struct tun_struct {
  99         struct tun_file         *tfile;
 100         unsigned int            flags;
 101         uid_t                   owner;
 102         gid_t                   group;
 103
 104         struct sk_buff_head     readq;
 105
 106         struct net_device       *dev;
 107         struct fasync_struct    *fasync;
 108
 109         struct tap_filter       txflt;
 110
 111 #ifdef TUN_DEBUG
 112         int debug;
 113 #endif
 114 };
 115
 116 static int tun_attach(struct tun_struct *tun, struct file *file)
 117 {
 118         struct tun_file *tfile = file->private_data;
 119         const struct cred *cred = current_cred();
 120         int err;
 121
 122         ASSERT_RTNL();
 123
 124         /* Check permissions */
 125         if (((tun->owner != -1 && cred->euid != tun->owner) ||
 126              (tun->group != -1 && cred->egid != tun->group)) &&
 127                 !capable(CAP_NET_ADMIN))
 128                 return -EPERM;
 129
 130         netif_tx_lock_bh(tun->dev);
 131
 132         err = -EINVAL;
 133         if (tfile->tun)
 134                 goto out;
 135
 136         err = -EBUSY;
 137         if (tun->tfile)
 138                 goto out;
 139
 140         err = 0;
 141         tfile->tun = tun;
 142         tun->tfile = tfile;
 143         dev_hold(tun->dev);
 144         atomic_inc(&tfile->count);
 145
 146 out:
 147         netif_tx_unlock_bh(tun->dev);
 148         return err;
 149 }
 150
 151 static void __tun_detach(struct tun_struct *tun)
 152 {
 153         struct tun_file *tfile = tun->tfile;
 154
 155         /* Detach from net device */
 156         netif_tx_lock_bh(tun->dev);
 157         tfile->tun = NULL;
 158         tun->tfile = NULL;
 159         netif_tx_unlock_bh(tun->dev);
 160
 161         /* Drop read queue */
 162         skb_queue_purge(&tun->readq);
 163
 164         /* Drop the extra count on the net device */
 165         dev_put(tun->dev);
 166 }
 167
 168 static void tun_detach(struct tun_struct *tun)
 169 {
 170         rtnl_lock();
 171         __tun_detach(tun);
 172         rtnl_unlock();
 173 }
 174
 175 static struct tun_struct *__tun_get(struct tun_file *tfile)
 176 {
 177         struct tun_struct *tun = NULL;
 178
 179         if (atomic_inc_not_zero(&tfile->count))
 180                 tun = tfile->tun;
 181
 182         return tun;
 183 }
 184
 185 static struct tun_struct *tun_get(struct file *file)
 186 {
 187         return __tun_get(file->private_data);
 188 }
 189
 190 static void tun_put(struct tun_struct *tun)
 191 {
 192         struct tun_file *tfile = tun->tfile;
 193
 194         if (atomic_dec_and_test(&tfile->count))
 195                 tun_detach(tfile->tun);
 196 }
 197
 198 /* TAP filterting */
 199 static void addr_hash_set(u32 *mask, const u8 *addr)
 200 {
 201         int n = ether_crc(ETH_ALEN, addr) >> 26;
 202         mask[n >> 5] |= (1 << (n & 31));
 203 }
 204
 205 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 206 {
 207         int n = ether_crc(ETH_ALEN, addr) >> 26;
 208         return mask[n >> 5] & (1 << (n & 31));
 209 }
 210
 211 static int update_filter(struct tap_filter *filter, void __user *arg)
 212 {
 213         struct { u8 u[ETH_ALEN]; } *addr;
 214         struct tun_filter uf;
 215         int err, alen, n, nexact;
 216
 217         if (copy_from_user(&uf, arg, sizeof(uf)))
 218                 return -EFAULT;
 219
 220         if (!uf.count) {
 221                 /* Disabled */
 222                 filter->count = 0;
 223                 return 0;
 224         }
 225
 226         alen = ETH_ALEN * uf.count;
 227         addr = kmalloc(alen, GFP_KERNEL);
 228         if (!addr)
 229                 return -ENOMEM;
 230
 231         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 232                 err = -EFAULT;
 233                 goto done;
 234         }
 235
 236         /* The filter is updated without holding any locks. Which is
 237          * perfectly safe. We disable it first and in the worst
 238          * case we'll accept a few undesired packets. */
 239         filter->count = 0;
 240         wmb();
 241
 242         /* Use first set of addresses as an exact filter */
 243         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 244                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 245
 246         nexact = n;
 247
 248         /* The rest is hashed */
 249         memset(filter->mask, 0, sizeof(filter->mask));
 250         for (; n < uf.count; n++)
 251                 addr_hash_set(filter->mask, addr[n].u);
 252
 253         /* For ALLMULTI just set the mask to all ones.
 254          * This overrides the mask populated above. */
 255         if ((uf.flags & TUN_FLT_ALLMULTI))
 256                 memset(filter->mask, ~0, sizeof(filter->mask));
 257
 258         /* Now enable the filter */
 259         wmb();
 260         filter->count = nexact;
 261
 262         /* Return the number of exact filters */
 263         err = nexact;
 264
 265 done:
 266         kfree(addr);
 267         return err;
 268 }
 269
 270 /* Returns: 0 - drop, !=0 - accept */
 271 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 272 {
 273         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 274          * at this point. */
 275         struct ethhdr *eh = (struct ethhdr *) skb->data;
 276         int i;
 277
 278         /* Exact match */
 279         for (i = 0; i < filter->count; i++)
 280                 if (!compare_ether_addr(eh->h_dest, filter->addr[i]))
 281                         return 1;
 282
 283         /* Inexact match (multicast only) */
 284         if (is_multicast_ether_addr(eh->h_dest))
 285                 return addr_hash_test(filter->mask, eh->h_dest);
 286
 287         return 0;
 288 }
 289
 290 /*
 291  * Checks whether the packet is accepted or not.
 292  * Returns: 0 - drop, !=0 - accept
 293  */
 294 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 295 {
 296         if (!filter->count)
 297                 return 1;
 298
 299         return run_filter(filter, skb);
 300 }
 301
 302 /* Network device part of the driver */
 303
 304 static const struct ethtool_ops tun_ethtool_ops;
 305
 306 /* Net device detach from fd. */
 307 static void tun_net_uninit(struct net_device *dev)
 308 {
 309         struct tun_struct *tun = netdev_priv(dev);
 310         struct tun_file *tfile = tun->tfile;
 311
 312         /* Inform the methods they need to stop using the dev.
 313          */
 314         if (tfile) {
 315                 wake_up_all(&tfile->read_wait);
 316                 if (atomic_dec_and_test(&tfile->count))
 317                         __tun_detach(tun);
 318         }
 319 }
 320
 321 /* Net device open. */
 322 static int tun_net_open(struct net_device *dev)
 323 {
 324         netif_start_queue(dev);
 325         return 0;
 326 }
 327
 328 /* Net device close. */
 329 static int tun_net_close(struct net_device *dev)
 330 {
 331         netif_stop_queue(dev);
 332         return 0;
 333 }
 334
 335 /* Net device start xmit */
 336 static int tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 337 {
 338         struct tun_struct *tun = netdev_priv(dev);
 339
 340         DBG(KERN_INFO "%s: tun_net_xmit %d\n", tun->dev->name, skb->len);
 341
 342         /* Drop packet if interface is not attached */
 343         if (!tun->tfile)
 344                 goto drop;
 345
 346         /* Drop if the filter does not like it.
 347          * This is a noop if the filter is disabled.
 348          * Filter can be enabled only for the TAP devices. */
 349         if (!check_filter(&tun->txflt, skb))
 350                 goto drop;
 351
 352         if (skb_queue_len(&tun->readq) >= dev->tx_queue_len) {
 353                 if (!(tun->flags & TUN_ONE_QUEUE)) {
 354                         /* Normal queueing mode. */
 355                         /* Packet scheduler handles dropping of further packets. */
 356                         netif_stop_queue(dev);
 357
 358                         /* We won't see all dropped packets individually, so overrun
 359                          * error is more appropriate. */
 360                         dev->stats.tx_fifo_errors++;
 361                 } else {
 362                         /* Single queue mode.
 363                          * Driver handles dropping of all packets itself. */
 364                         goto drop;
 365                 }
 366         }
 367
 368         /* Enqueue packet */
 369         skb_queue_tail(&tun->readq, skb);
 370         dev->trans_start = jiffies;
 371
 372         /* Notify and wake up reader process */
 373         if (tun->flags & TUN_FASYNC)
 374                 kill_fasync(&tun->fasync, SIGIO, POLL_IN);
 375         wake_up_interruptible(&tun->tfile->read_wait);
 376         return 0;
 377
 378 drop:
 379         dev->stats.tx_dropped++;
 380         kfree_skb(skb);
 381         return 0;
 382 }
 383
 384 static void tun_net_mclist(struct net_device *dev)
 385 {
 386         /*
 387          * This callback is supposed to deal with mc filter in
 388          * _rx_ path and has nothing to do with the _tx_ path.
 389          * In rx path we always accept everything userspace gives us.
 390          */
 391         return;
 392 }
 393
 394 #define MIN_MTU 68
 395 #define MAX_MTU 65535
 396
 397 static int
 398 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 399 {
 400         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 401                 return -EINVAL;
 402         dev->mtu = new_mtu;
 403         return 0;
 404 }
 405
 406 static const struct net_device_ops tun_netdev_ops = {
 407         .ndo_uninit             = tun_net_uninit,
 408         .ndo_open               = tun_net_open,
 409         .ndo_stop               = tun_net_close,
 410         .ndo_start_xmit         = tun_net_xmit,
 411         .ndo_change_mtu         = tun_net_change_mtu,
 412 };
 413
 414 static const struct net_device_ops tap_netdev_ops = {
 415         .ndo_uninit             = tun_net_uninit,
 416         .ndo_open               = tun_net_open,
 417         .ndo_stop               = tun_net_close,
 418         .ndo_start_xmit         = tun_net_xmit,
 419         .ndo_change_mtu         = tun_net_change_mtu,
 420         .ndo_set_multicast_list = tun_net_mclist,
 421         .ndo_set_mac_address    = eth_mac_addr,
 422         .ndo_validate_addr      = eth_validate_addr,
 423 };
 424
 425 /* Initialize net device. */
 426 static void tun_net_init(struct net_device *dev)
 427 {
 428         struct tun_struct *tun = netdev_priv(dev);
 429
 430         switch (tun->flags & TUN_TYPE_MASK) {
 431         case TUN_TUN_DEV:
 432                 dev->netdev_ops = &tun_netdev_ops;
 433
 434                 /* Point-to-Point TUN Device */
 435                 dev->hard_header_len = 0;
 436                 dev->addr_len = 0;
 437                 dev->mtu = 1500;
 438
 439                 /* Zero header length */
 440                 dev->type = ARPHRD_NONE;
 441                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 442                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 443                 break;
 444
 445         case TUN_TAP_DEV:
 446                 dev->netdev_ops = &tap_netdev_ops;
 447                 /* Ethernet TAP Device */
 448                 ether_setup(dev);
 449
 450                 random_ether_addr(dev->dev_addr);
 451
 452                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 453                 break;
 454         }
 455 }
 456
 457 /* Character device part */
 458
 459 /* Poll */
 460 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
 461 {
 462         struct tun_file *tfile = file->private_data;
 463         struct tun_struct *tun = __tun_get(tfile);
 464         unsigned int mask = POLLOUT | POLLWRNORM;
 465
 466         if (!tun)
 467                 return POLLERR;
 468
 469         DBG(KERN_INFO "%s: tun_chr_poll\n", tun->dev->name);
 470
 471         poll_wait(file, &tfile->read_wait, wait);
 472
 473         if (!skb_queue_empty(&tun->readq))
 474                 mask |= POLLIN | POLLRDNORM;
 475
 476         if (tun->dev->reg_state != NETREG_REGISTERED)
 477                 mask = POLLERR;
 478
 479         tun_put(tun);
 480         return mask;
 481 }
 482
 483 /* prepad is the amount to reserve at front.  len is length after that.
 484  * linear is a hint as to how much to copy (usually headers). */
 485 static struct sk_buff *tun_alloc_skb(size_t prepad, size_t len, size_t linear,
 486                                      gfp_t gfp)
 487 {
 488         struct sk_buff *skb;
 489         unsigned int i;
 490
 491         skb = alloc_skb(prepad + len, gfp|__GFP_NOWARN);
 492         if (skb) {
 493                 skb_reserve(skb, prepad);
 494                 skb_put(skb, len);
 495                 return skb;
 496         }
 497
 498         /* Under a page?  Don't bother with paged skb. */
 499         if (prepad + len < PAGE_SIZE)
 500                 return NULL;
 501
 502         /* Start with a normal skb, and add pages. */
 503         skb = alloc_skb(prepad + linear, gfp);
 504         if (!skb)
 505                 return NULL;
 506
 507         skb_reserve(skb, prepad);
 508         skb_put(skb, linear);
 509
 510         len -= linear;
 511
 512         for (i = 0; i < MAX_SKB_FRAGS; i++) {
 513                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 514
 515                 f->page = alloc_page(gfp|__GFP_ZERO);
 516                 if (!f->page)
 517                         break;
 518
 519                 f->page_offset = 0;
 520                 f->size = PAGE_SIZE;
 521
 522                 skb->data_len += PAGE_SIZE;
 523                 skb->len += PAGE_SIZE;
 524                 skb->truesize += PAGE_SIZE;
 525                 skb_shinfo(skb)->nr_frags++;
 526
 527                 if (len < PAGE_SIZE) {
 528                         len = 0;
 529                         break;
 530                 }
 531                 len -= PAGE_SIZE;
 532         }
 533
 534         /* Too large, or alloc fail? */
 535         if (unlikely(len)) {
 536                 kfree_skb(skb);
 537                 skb = NULL;
 538         }
 539
 540         return skb;
 541 }
 542
 543 /* Get packet from user space buffer */
 544 static __inline__ ssize_t tun_get_user(struct tun_struct *tun, struct iovec *iv, size_t count)
 545 {
 546         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
 547         struct sk_buff *skb;
 548         size_t len = count, align = 0;
 549         struct virtio_net_hdr gso = { 0 };
 550
 551         if (!(tun->flags & TUN_NO_PI)) {
 552                 if ((len -= sizeof(pi)) > count)
 553                         return -EINVAL;
 554
 555                 if(memcpy_fromiovec((void *)&pi, iv, sizeof(pi)))
 556                         return -EFAULT;
 557         }
 558
 559         if (tun->flags & TUN_VNET_HDR) {
 560                 if ((len -= sizeof(gso)) > count)
 561                         return -EINVAL;
 562
 563                 if (memcpy_fromiovec((void *)&gso, iv, sizeof(gso)))
 564                         return -EFAULT;
 565
 566                 if (gso.hdr_len > len)
 567                         return -EINVAL;
 568         }
 569
 570         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
 571                 align = NET_IP_ALIGN;
 572                 if (unlikely(len < ETH_HLEN))
 573                         return -EINVAL;
 574         }
 575
 576         if (!(skb = tun_alloc_skb(align, len, gso.hdr_len, GFP_KERNEL))) {
 577                 tun->dev->stats.rx_dropped++;
 578                 return -ENOMEM;
 579         }
 580
 581         if (skb_copy_datagram_from_iovec(skb, 0, iv, len)) {
 582                 tun->dev->stats.rx_dropped++;
 583                 kfree_skb(skb);
 584                 return -EFAULT;
 585         }
 586
 587         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 588                 if (!skb_partial_csum_set(skb, gso.csum_start,
 589                                           gso.csum_offset)) {
 590                         tun->dev->stats.rx_frame_errors++;
 591                         kfree_skb(skb);
 592                         return -EINVAL;
 593                 }
 594         } else if (tun->flags & TUN_NOCHECKSUM)
 595                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 596
 597         switch (tun->flags & TUN_TYPE_MASK) {
 598         case TUN_TUN_DEV:
 599                 if (tun->flags & TUN_NO_PI) {
 600                         switch (skb->data[0] & 0xf0) {
 601                         case 0x40:
 602                                 pi.proto = htons(ETH_P_IP);
 603                                 break;
 604                         case 0x60:
 605                                 pi.proto = htons(ETH_P_IPV6);
 606                                 break;
 607                         default:
 608                                 tun->dev->stats.rx_dropped++;
 609                                 kfree_skb(skb);
 610                                 return -EINVAL;
 611                         }
 612                 }
 613
 614                 skb_reset_mac_header(skb);
 615                 skb->protocol = pi.proto;
 616                 skb->dev = tun->dev;
 617                 break;
 618         case TUN_TAP_DEV:
 619                 skb->protocol = eth_type_trans(skb, tun->dev);
 620                 break;
 621         };
 622
 623         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 624                 pr_debug("GSO!\n");
 625                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 626                 case VIRTIO_NET_HDR_GSO_TCPV4:
 627                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 628                         break;
 629                 case VIRTIO_NET_HDR_GSO_TCPV6:
 630                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 631                         break;
 632                 default:
 633                         tun->dev->stats.rx_frame_errors++;
 634                         kfree_skb(skb);
 635                         return -EINVAL;
 636                 }
 637
 638                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
 639                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 640
 641                 skb_shinfo(skb)->gso_size = gso.gso_size;
 642                 if (skb_shinfo(skb)->gso_size == 0) {
 643                         tun->dev->stats.rx_frame_errors++;
 644                         kfree_skb(skb);
 645                         return -EINVAL;
 646                 }
 647
 648                 /* Header must be checked, and gso_segs computed. */
 649                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 650                 skb_shinfo(skb)->gso_segs = 0;
 651         }
 652
 653         netif_rx_ni(skb);
 654
 655         tun->dev->stats.rx_packets++;
 656         tun->dev->stats.rx_bytes += len;
 657
 658         return count;
 659 }
 660
 661 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
 662                               unsigned long count, loff_t pos)
 663 {
 664         struct tun_struct *tun = tun_get(iocb->ki_filp);
 665         ssize_t result;
 666
 667         if (!tun)
 668                 return -EBADFD;
 669
 670         DBG(KERN_INFO "%s: tun_chr_write %ld\n", tun->dev->name, count);
 671
 672         result = tun_get_user(tun, (struct iovec *) iv, iov_length(iv, count));
 673
 674         tun_put(tun);
 675         return result;
 676 }
 677
 678 /* Put packet to the user space buffer */
 679 static __inline__ ssize_t tun_put_user(struct tun_struct *tun,
 680                                        struct sk_buff *skb,
 681                                        struct iovec *iv, int len)
 682 {
 683         struct tun_pi pi = { 0, skb->protocol };
 684         ssize_t total = 0;
 685
 686         if (!(tun->flags & TUN_NO_PI)) {
 687                 if ((len -= sizeof(pi)) < 0)
 688                         return -EINVAL;
 689
 690                 if (len < skb->len) {
 691                         /* Packet will be striped */
 692                         pi.flags |= TUN_PKT_STRIP;
 693                 }
 694
 695                 if (memcpy_toiovec(iv, (void *) &pi, sizeof(pi)))
 696                         return -EFAULT;
 697                 total += sizeof(pi);
 698         }
 699
 700         if (tun->flags & TUN_VNET_HDR) {
 701                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
 702                 if ((len -= sizeof(gso)) < 0)
 703                         return -EINVAL;
 704
 705                 if (skb_is_gso(skb)) {
 706                         struct skb_shared_info *sinfo = skb_shinfo(skb);
 707
 708                         /* This is a hint as to how much should be linear. */
 709                         gso.hdr_len = skb_headlen(skb);
 710                         gso.gso_size = sinfo->gso_size;
 711                         if (sinfo->gso_type & SKB_GSO_TCPV4)
 712                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 713                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
 714                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 715                         else
 716                                 BUG();
 717                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
 718                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 719                 } else
 720                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
 721
 722                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 723                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 724                         gso.csum_start = skb->csum_start - skb_headroom(skb);
 725                         gso.csum_offset = skb->csum_offset;
 726                 } /* else everything is zero */
 727
 728                 if (unlikely(memcpy_toiovec(iv, (void *)&gso, sizeof(gso))))
 729                         return -EFAULT;
 730                 total += sizeof(gso);
 731         }
 732
 733         len = min_t(int, skb->len, len);
 734
 735         skb_copy_datagram_iovec(skb, 0, iv, len);
 736         total += len;
 737
 738         tun->dev->stats.tx_packets++;
 739         tun->dev->stats.tx_bytes += len;
 740
 741         return total;
 742 }
 743
 744 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
 745                             unsigned long count, loff_t pos)
 746 {
 747         struct file *file = iocb->ki_filp;
 748         struct tun_file *tfile = file->private_data;
 749         struct tun_struct *tun = __tun_get(tfile);
 750         DECLARE_WAITQUEUE(wait, current);
 751         struct sk_buff *skb;
 752         ssize_t len, ret = 0;
 753
 754         if (!tun)
 755                 return -EBADFD;
 756
 757         DBG(KERN_INFO "%s: tun_chr_read\n", tun->dev->name);
 758
 759         len = iov_length(iv, count);
 760         if (len < 0) {
 761                 ret = -EINVAL;
 762                 goto out;
 763         }
 764
 765         add_wait_queue(&tfile->read_wait, &wait);
 766         while (len) {
 767                 current->state = TASK_INTERRUPTIBLE;
 768
 769                 /* Read frames from the queue */
 770                 if (!(skb=skb_dequeue(&tun->readq))) {
 771                         if (file->f_flags & O_NONBLOCK) {
 772                                 ret = -EAGAIN;
 773                                 break;
 774                         }
 775                         if (signal_pending(current)) {
 776                                 ret = -ERESTARTSYS;
 777                                 break;
 778                         }
 779                         if (tun->dev->reg_state != NETREG_REGISTERED) {
 780                                 ret = -EIO;
 781                                 break;
 782                         }
 783
 784                         /* Nothing to read, let's sleep */
 785                         schedule();
 786                         continue;
 787                 }
 788                 netif_wake_queue(tun->dev);
 789
 790                 ret = tun_put_user(tun, skb, (struct iovec *) iv, len);
 791                 kfree_skb(skb);
 792                 break;
 793         }
 794
 795         current->state = TASK_RUNNING;
 796         remove_wait_queue(&tfile->read_wait, &wait);
 797
 798 out:
 799         tun_put(tun);
 800         return ret;
 801 }
 802
 803 static void tun_setup(struct net_device *dev)
 804 {
 805         struct tun_struct *tun = netdev_priv(dev);
 806
 807         skb_queue_head_init(&tun->readq);
 808
 809         tun->owner = -1;
 810         tun->group = -1;
 811
 812         dev->ethtool_ops = &tun_ethtool_ops;
 813         dev->destructor = free_netdev;
 814 }
 815
 816 /* Trivial set of netlink ops to allow deleting tun or tap
 817  * device with netlink.
 818  */
 819 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
 820 {
 821         return -EINVAL;
 822 }
 823
 824 static struct rtnl_link_ops tun_link_ops __read_mostly = {
 825         .kind           = DRV_NAME,
 826         .priv_size      = sizeof(struct tun_struct),
 827         .setup          = tun_setup,
 828         .validate       = tun_validate,
 829 };
 830
 831
 832 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
 833 {
 834         struct tun_struct *tun;
 835         struct net_device *dev;
 836         int err;
 837
 838         dev = __dev_get_by_name(net, ifr->ifr_name);
 839         if (dev) {
 840                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
 841                         tun = netdev_priv(dev);
 842                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
 843                         tun = netdev_priv(dev);
 844                 else
 845                         return -EINVAL;
 846
 847                 err = tun_attach(tun, file);
 848                 if (err < 0)
 849                         return err;
 850         }
 851         else {
 852                 char *name;
 853                 unsigned long flags = 0;
 854
 855                 err = -EINVAL;
 856
 857                 if (!capable(CAP_NET_ADMIN))
 858                         return -EPERM;
 859
 860                 /* Set dev type */
 861                 if (ifr->ifr_flags & IFF_TUN) {
 862                         /* TUN device */
 863                         flags |= TUN_TUN_DEV;
 864                         name = "tun%d";
 865                 } else if (ifr->ifr_flags & IFF_TAP) {
 866                         /* TAP device */
 867                         flags |= TUN_TAP_DEV;
 868                         name = "tap%d";
 869                 } else
 870                         goto failed;
 871
 872                 if (*ifr->ifr_name)
 873                         name = ifr->ifr_name;
 874
 875                 dev = alloc_netdev(sizeof(struct tun_struct), name,
 876                                    tun_setup);
 877                 if (!dev)
 878                         return -ENOMEM;
 879
 880                 dev_net_set(dev, net);
 881                 dev->rtnl_link_ops = &tun_link_ops;
 882
 883                 tun = netdev_priv(dev);
 884                 tun->dev = dev;
 885                 tun->flags = flags;
 886                 tun->txflt.count = 0;
 887
 888                 tun_net_init(dev);
 889
 890                 if (strchr(dev->name, '%')) {
 891                         err = dev_alloc_name(dev, dev->name);
 892                         if (err < 0)
 893                                 goto err_free_dev;
 894                 }
 895
 896                 err = register_netdevice(tun->dev);
 897                 if (err < 0)
 898                         goto err_free_dev;
 899
 900                 err = tun_attach(tun, file);
 901                 if (err < 0)
 902                         goto err_free_dev;
 903         }
 904
 905         DBG(KERN_INFO "%s: tun_set_iff\n", tun->dev->name);
 906
 907         if (ifr->ifr_flags & IFF_NO_PI)
 908                 tun->flags |= TUN_NO_PI;
 909         else
 910                 tun->flags &= ~TUN_NO_PI;
 911
 912         if (ifr->ifr_flags & IFF_ONE_QUEUE)
 913                 tun->flags |= TUN_ONE_QUEUE;
 914         else
 915                 tun->flags &= ~TUN_ONE_QUEUE;
 916
 917         if (ifr->ifr_flags & IFF_VNET_HDR)
 918                 tun->flags |= TUN_VNET_HDR;
 919         else
 920                 tun->flags &= ~TUN_VNET_HDR;
 921
 922         /* Make sure persistent devices do not get stuck in
 923          * xoff state.
 924          */
 925         if (netif_running(tun->dev))
 926                 netif_wake_queue(tun->dev);
 927
 928         strcpy(ifr->ifr_name, tun->dev->name);
 929         return 0;
 930
 931  err_free_dev:
 932         free_netdev(dev);
 933  failed:
 934         return err;
 935 }
 936
 937 static int tun_get_iff(struct net *net, struct file *file, struct ifreq *ifr)
 938 {
 939         struct tun_struct *tun = tun_get(file);
 940
 941         if (!tun)
 942                 return -EBADFD;
 943
 944         DBG(KERN_INFO "%s: tun_get_iff\n", tun->dev->name);
 945
 946         strcpy(ifr->ifr_name, tun->dev->name);
 947
 948         ifr->ifr_flags = 0;
 949
 950         if (ifr->ifr_flags & TUN_TUN_DEV)
 951                 ifr->ifr_flags |= IFF_TUN;
 952         else
 953                 ifr->ifr_flags |= IFF_TAP;
 954
 955         if (tun->flags & TUN_NO_PI)
 956                 ifr->ifr_flags |= IFF_NO_PI;
 957
 958         if (tun->flags & TUN_ONE_QUEUE)
 959                 ifr->ifr_flags |= IFF_ONE_QUEUE;
 960
 961         if (tun->flags & TUN_VNET_HDR)
 962                 ifr->ifr_flags |= IFF_VNET_HDR;
 963
 964         tun_put(tun);
 965         return 0;
 966 }
 967
 968 /* This is like a cut-down ethtool ops, except done via tun fd so no
 969  * privs required. */
 970 static int set_offload(struct net_device *dev, unsigned long arg)
 971 {
 972         unsigned int old_features, features;
 973
 974         old_features = dev->features;
 975         /* Unset features, set them as we chew on the arg. */
 976         features = (old_features & ~(NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST
 977                                     |NETIF_F_TSO_ECN|NETIF_F_TSO|NETIF_F_TSO6));
 978
 979         if (arg & TUN_F_CSUM) {
 980                 features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
 981                 arg &= ~TUN_F_CSUM;
 982
 983                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
 984                         if (arg & TUN_F_TSO_ECN) {
 985                                 features |= NETIF_F_TSO_ECN;
 986                                 arg &= ~TUN_F_TSO_ECN;
 987                         }
 988                         if (arg & TUN_F_TSO4)
 989                                 features |= NETIF_F_TSO;
 990                         if (arg & TUN_F_TSO6)
 991                                 features |= NETIF_F_TSO6;
 992                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
 993                 }
 994         }
 995
 996         /* This gives the user a way to test for new features in future by
 997          * trying to set them. */
 998         if (arg)
 999                 return -EINVAL;
1000
1001         dev->features = features;
1002         if (old_features != dev->features)
1003                 netdev_features_change(dev);
1004
1005         return 0;
1006 }
1007
1008 static int tun_chr_ioctl(struct inode *inode, struct file *file,
1009                          unsigned int cmd, unsigned long arg)
1010 {
1011         struct tun_file *tfile = file->private_data;
1012         struct tun_struct *tun;
1013         void __user* argp = (void __user*)arg;
1014         struct ifreq ifr;
1015         int ret;
1016
1017         if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
1018                 if (copy_from_user(&ifr, argp, sizeof ifr))
1019                         return -EFAULT;
1020
1021         if (cmd == TUNGETFEATURES) {
1022                 /* Currently this just means: "what IFF flags are valid?".
1023                  * This is needed because we never checked for invalid flags on
1024                  * TUNSETIFF. */
1025                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1026                                 IFF_VNET_HDR,
1027                                 (unsigned int __user*)argp);
1028         }
1029
1030         tun = __tun_get(tfile);
1031         if (cmd == TUNSETIFF && !tun) {
1032                 int err;
1033
1034                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1035
1036                 rtnl_lock();
1037                 err = tun_set_iff(tfile->net, file, &ifr);
1038                 rtnl_unlock();
1039
1040                 if (err)
1041                         return err;
1042
1043                 if (copy_to_user(argp, &ifr, sizeof(ifr)))
1044                         return -EFAULT;
1045                 return 0;
1046         }
1047
1048
1049         if (!tun)
1050                 return -EBADFD;
1051
1052         DBG(KERN_INFO "%s: tun_chr_ioctl cmd %d\n", tun->dev->name, cmd);
1053
1054         ret = 0;
1055         switch (cmd) {
1056         case TUNGETIFF:
1057                 ret = tun_get_iff(current->nsproxy->net_ns, file, &ifr);
1058                 if (ret)
1059                         break;
1060
1061                 if (copy_to_user(argp, &ifr, sizeof(ifr)))
1062                         ret = -EFAULT;
1063                 break;
1064
1065         case TUNSETNOCSUM:
1066                 /* Disable/Enable checksum */
1067                 if (arg)
1068                         tun->flags |= TUN_NOCHECKSUM;
1069                 else
1070                         tun->flags &= ~TUN_NOCHECKSUM;
1071
1072                 DBG(KERN_INFO "%s: checksum %s\n",
1073                     tun->dev->name, arg ? "disabled" : "enabled");
1074                 break;
1075
1076         case TUNSETPERSIST:
1077                 /* Disable/Enable persist mode */
1078                 if (arg)
1079                         tun->flags |= TUN_PERSIST;
1080                 else
1081                         tun->flags &= ~TUN_PERSIST;
1082
1083                 DBG(KERN_INFO "%s: persist %s\n",
1084                     tun->dev->name, arg ? "enabled" : "disabled");
1085                 break;
1086
1087         case TUNSETOWNER:
1088                 /* Set owner of the device */
1089                 tun->owner = (uid_t) arg;
1090
1091                 DBG(KERN_INFO "%s: owner set to %d\n", tun->dev->name, tun->owner);
1092                 break;
1093
1094         case TUNSETGROUP:
1095                 /* Set group of the device */
1096                 tun->group= (gid_t) arg;
1097
1098                 DBG(KERN_INFO "%s: group set to %d\n", tun->dev->name, tun->group);
1099                 break;
1100
1101         case TUNSETLINK:
1102                 /* Only allow setting the type when the interface is down */
1103                 rtnl_lock();
1104                 if (tun->dev->flags & IFF_UP) {
1105                         DBG(KERN_INFO "%s: Linktype set failed because interface is up\n",
1106                                 tun->dev->name);
1107                         ret = -EBUSY;
1108                 } else {
1109                         tun->dev->type = (int) arg;
1110                         DBG(KERN_INFO "%s: linktype set to %d\n", tun->dev->name, tun->dev->type);
1111                         ret = 0;
1112                 }
1113                 rtnl_unlock();
1114                 break;
1115
1116 #ifdef TUN_DEBUG
1117         case TUNSETDEBUG:
1118                 tun->debug = arg;
1119                 break;
1120 #endif
1121         case TUNSETOFFLOAD:
1122                 rtnl_lock();
1123                 ret = set_offload(tun->dev, arg);
1124                 rtnl_unlock();
1125                 break;
1126
1127         case TUNSETTXFILTER:
1128                 /* Can be set only for TAPs */
1129                 ret = -EINVAL;
1130                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1131                         break;
1132                 rtnl_lock();
1133                 ret = update_filter(&tun->txflt, (void __user *)arg);
1134                 rtnl_unlock();
1135                 break;
1136
1137         case SIOCGIFHWADDR:
1138                 /* Get hw addres */
1139                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1140                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1141                 if (copy_to_user(argp, &ifr, sizeof ifr))
1142                         ret = -EFAULT;
1143                 break;
1144
1145         case SIOCSIFHWADDR:
1146                 /* Set hw address */
1147                 DBG(KERN_DEBUG "%s: set hw address: %pM\n",
1148                         tun->dev->name, ifr.ifr_hwaddr.sa_data);
1149
1150                 rtnl_lock();
1151                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1152                 rtnl_unlock();
1153                 break;
1154         default:
1155                 ret = -EINVAL;
1156                 break;
1157         };
1158
1159         tun_put(tun);
1160         return ret;
1161 }
1162
1163 static int tun_chr_fasync(int fd, struct file *file, int on)
1164 {
1165         struct tun_struct *tun = tun_get(file);
1166         int ret;
1167
1168         if (!tun)
1169                 return -EBADFD;
1170
1171         DBG(KERN_INFO "%s: tun_chr_fasync %d\n", tun->dev->name, on);
1172
1173         lock_kernel();
1174         if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1175                 goto out;
1176
1177         if (on) {
1178                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1179                 if (ret)
1180                         goto out;
1181                 tun->flags |= TUN_FASYNC;
1182         } else
1183                 tun->flags &= ~TUN_FASYNC;
1184         ret = 0;
1185 out:
1186         unlock_kernel();
1187         tun_put(tun);
1188         return ret;
1189 }
1190
1191 static int tun_chr_open(struct inode *inode, struct file * file)
1192 {
1193         struct tun_file *tfile;
1194         cycle_kernel_lock();
1195         DBG1(KERN_INFO "tunX: tun_chr_open\n");
1196
1197         tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1198         if (!tfile)
1199                 return -ENOMEM;
1200         atomic_set(&tfile->count, 0);
1201         tfile->tun = NULL;
1202         tfile->net = get_net(current->nsproxy->net_ns);
1203         init_waitqueue_head(&tfile->read_wait);
1204         file->private_data = tfile;
1205         return 0;
1206 }
1207
1208 static int tun_chr_close(struct inode *inode, struct file *file)
1209 {
1210         struct tun_file *tfile = file->private_data;
1211         struct tun_struct *tun = __tun_get(tfile);
1212
1213
1214         if (tun) {
1215                 DBG(KERN_INFO "%s: tun_chr_close\n", tun->dev->name);
1216
1217                 rtnl_lock();
1218                 __tun_detach(tun);
1219
1220                 /* If desireable, unregister the netdevice. */
1221                 if (!(tun->flags & TUN_PERSIST))
1222                         unregister_netdevice(tun->dev);
1223
1224                 rtnl_unlock();
1225         }
1226
1227         put_net(tfile->net);
1228         kfree(tfile);
1229
1230         return 0;
1231 }
1232
1233 static const struct file_operations tun_fops = {
1234         .owner  = THIS_MODULE,
1235         .llseek = no_llseek,
1236         .read  = do_sync_read,
1237         .aio_read  = tun_chr_aio_read,
1238         .write = do_sync_write,
1239         .aio_write = tun_chr_aio_write,
1240         .poll   = tun_chr_poll,
1241         .ioctl  = tun_chr_ioctl,
1242         .open   = tun_chr_open,
1243         .release = tun_chr_close,
1244         .fasync = tun_chr_fasync
1245 };
1246
1247 static struct miscdevice tun_miscdev = {
1248         .minor = TUN_MINOR,
1249         .name = "tun",
1250         .fops = &tun_fops,
1251 };
1252
1253 /* ethtool interface */
1254
1255 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1256 {
1257         cmd->supported          = 0;
1258         cmd->advertising        = 0;
1259         cmd->speed              = SPEED_10;
1260         cmd->duplex             = DUPLEX_FULL;
1261         cmd->port               = PORT_TP;
1262         cmd->phy_address        = 0;
1263         cmd->transceiver        = XCVR_INTERNAL;
1264         cmd->autoneg            = AUTONEG_DISABLE;
1265         cmd->maxtxpkt           = 0;
1266         cmd->maxrxpkt           = 0;
1267         return 0;
1268 }
1269
1270 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1271 {
1272         struct tun_struct *tun = netdev_priv(dev);
1273
1274         strcpy(info->driver, DRV_NAME);
1275         strcpy(info->version, DRV_VERSION);
1276         strcpy(info->fw_version, "N/A");
1277
1278         switch (tun->flags & TUN_TYPE_MASK) {
1279         case TUN_TUN_DEV:
1280                 strcpy(info->bus_info, "tun");
1281                 break;
1282         case TUN_TAP_DEV:
1283                 strcpy(info->bus_info, "tap");
1284                 break;
1285         }
1286 }
1287
1288 static u32 tun_get_msglevel(struct net_device *dev)
1289 {
1290 #ifdef TUN_DEBUG
1291         struct tun_struct *tun = netdev_priv(dev);
1292         return tun->debug;
1293 #else
1294         return -EOPNOTSUPP;
1295 #endif
1296 }
1297
1298 static void tun_set_msglevel(struct net_device *dev, u32 value)
1299 {
1300 #ifdef TUN_DEBUG
1301         struct tun_struct *tun = netdev_priv(dev);
1302         tun->debug = value;
1303 #endif
1304 }
1305
1306 static u32 tun_get_link(struct net_device *dev)
1307 {
1308         struct tun_struct *tun = netdev_priv(dev);
1309         return !!tun->tfile;
1310 }
1311
1312 static u32 tun_get_rx_csum(struct net_device *dev)
1313 {
1314         struct tun_struct *tun = netdev_priv(dev);
1315         return (tun->flags & TUN_NOCHECKSUM) == 0;
1316 }
1317
1318 static int tun_set_rx_csum(struct net_device *dev, u32 data)
1319 {
1320         struct tun_struct *tun = netdev_priv(dev);
1321         if (data)
1322                 tun->flags &= ~TUN_NOCHECKSUM;
1323         else
1324                 tun->flags |= TUN_NOCHECKSUM;
1325         return 0;
1326 }
1327
1328 static const struct ethtool_ops tun_ethtool_ops = {
1329         .get_settings   = tun_get_settings,
1330         .get_drvinfo    = tun_get_drvinfo,
1331         .get_msglevel   = tun_get_msglevel,
1332         .set_msglevel   = tun_set_msglevel,
1333         .get_link       = tun_get_link,
1334         .get_rx_csum    = tun_get_rx_csum,
1335         .set_rx_csum    = tun_set_rx_csum
1336 };
1337
1338
1339 static int __init tun_init(void)
1340 {
1341         int ret = 0;
1342
1343         printk(KERN_INFO "tun: %s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1344         printk(KERN_INFO "tun: %s\n", DRV_COPYRIGHT);
1345
1346         ret = rtnl_link_register(&tun_link_ops);
1347         if (ret) {
1348                 printk(KERN_ERR "tun: Can't register link_ops\n");
1349                 goto err_linkops;
1350         }
1351
1352         ret = misc_register(&tun_miscdev);
1353         if (ret) {
1354                 printk(KERN_ERR "tun: Can't register misc device %d\n", TUN_MINOR);
1355                 goto err_misc;
1356         }
1357         return  0;
1358 err_misc:
1359         rtnl_link_unregister(&tun_link_ops);
1360 err_linkops:
1361         return ret;
1362 }
1363
1364 static void tun_cleanup(void)
1365 {
1366         misc_deregister(&tun_miscdev);
1367         rtnl_link_unregister(&tun_link_ops);
1368 }
1369
1370 module_init(tun_init);
1371 module_exit(tun_cleanup);
1372 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1373 MODULE_AUTHOR(DRV_COPYRIGHT);
1374 MODULE_LICENSE("GPL");
1375 MODULE_ALIAS_MISCDEV(TUN_MINOR);