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