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