git.oblomov.eu Git - linux-2.6/blob - drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use random_ether_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define DRV_NAME        "tun"
  38 #define DRV_VERSION     "1.6"
  39 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  40 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  41
  42 #include <linux/module.h>
  43 #include <linux/errno.h>
  44 #include <linux/kernel.h>
  45 #include <linux/major.h>
  46 #include <linux/slab.h>
  47 #include <linux/smp_lock.h>
  48 #include <linux/poll.h>
  49 #include <linux/fcntl.h>
  50 #include <linux/init.h>
  51 #include <linux/skbuff.h>
  52 #include <linux/netdevice.h>
  53 #include <linux/etherdevice.h>
  54 #include <linux/miscdevice.h>
  55 #include <linux/ethtool.h>
  56 #include <linux/rtnetlink.h>
  57 #include <linux/if.h>
  58 #include <linux/if_arp.h>
  59 #include <linux/if_ether.h>
  60 #include <linux/if_tun.h>
  61 #include <linux/crc32.h>
  62 #include <linux/nsproxy.h>
  63 #include <linux/virtio_net.h>
  64 #include <net/net_namespace.h>
  65 #include <net/netns/generic.h>
  66
  67 #include <asm/system.h>
  68 #include <asm/uaccess.h>
  69
  70 /* Uncomment to enable debugging */
  71 /* #define TUN_DEBUG 1 */
  72
  73 #ifdef TUN_DEBUG
  74 static int debug;
  75
  76 #define DBG  if(tun->debug)printk
  77 #define DBG1 if(debug==2)printk
  78 #else
  79 #define DBG( a... )
  80 #define DBG1( a... )
  81 #endif
  82
  83 #define FLT_EXACT_COUNT 8
  84 struct tap_filter {
  85         unsigned int    count;    /* Number of addrs. Zero means disabled */
  86         u32             mask[2];  /* Mask of the hashed addrs */
  87         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
  88 };
  89
  90 struct tun_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         /* The rest is hashed */
 161         memset(filter->mask, 0, sizeof(filter->mask));
 162         for (; n < uf.count; n++)
 163                 addr_hash_set(filter->mask, addr[n].u);
 164
 165         /* For ALLMULTI just set the mask to all ones.
 166          * This overrides the mask populated above. */
 167         if ((uf.flags & TUN_FLT_ALLMULTI))
 168                 memset(filter->mask, ~0, sizeof(filter->mask));
 169
 170         /* Now enable the filter */
 171         wmb();
 172         filter->count = nexact;
 173
 174         /* Return the number of exact filters */
 175         err = nexact;
 176
 177 done:
 178         kfree(addr);
 179         return err;
 180 }
 181
 182 /* Returns: 0 - drop, !=0 - accept */
 183 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 184 {
 185         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 186          * at this point. */
 187         struct ethhdr *eh = (struct ethhdr *) skb->data;
 188         int i;
 189
 190         /* Exact match */
 191         for (i = 0; i < filter->count; i++)
 192                 if (!compare_ether_addr(eh->h_dest, filter->addr[i]))
 193                         return 1;
 194
 195         /* Inexact match (multicast only) */
 196         if (is_multicast_ether_addr(eh->h_dest))
 197                 return addr_hash_test(filter->mask, eh->h_dest);
 198
 199         return 0;
 200 }
 201
 202 /*
 203  * Checks whether the packet is accepted or not.
 204  * Returns: 0 - drop, !=0 - accept
 205  */
 206 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 207 {
 208         if (!filter->count)
 209                 return 1;
 210
 211         return run_filter(filter, skb);
 212 }
 213
 214 /* Network device part of the driver */
 215
 216 static unsigned int tun_net_id;
 217 struct tun_net {
 218         struct list_head dev_list;
 219 };
 220
 221 static const struct ethtool_ops tun_ethtool_ops;
 222
 223 /* Net device open. */
 224 static int tun_net_open(struct net_device *dev)
 225 {
 226         netif_start_queue(dev);
 227         return 0;
 228 }
 229
 230 /* Net device close. */
 231 static int tun_net_close(struct net_device *dev)
 232 {
 233         netif_stop_queue(dev);
 234         return 0;
 235 }
 236
 237 /* Net device start xmit */
 238 static int tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 239 {
 240         struct tun_struct *tun = netdev_priv(dev);
 241
 242         DBG(KERN_INFO "%s: tun_net_xmit %d\n", tun->dev->name, skb->len);
 243
 244         /* Drop packet if interface is not attached */
 245         if (!tun->attached)
 246                 goto drop;
 247
 248         /* Drop if the filter does not like it.
 249          * This is a noop if the filter is disabled.
 250          * Filter can be enabled only for the TAP devices. */
 251         if (!check_filter(&tun->txflt, skb))
 252                 goto drop;
 253
 254         if (skb_queue_len(&tun->readq) >= dev->tx_queue_len) {
 255                 if (!(tun->flags & TUN_ONE_QUEUE)) {
 256                         /* Normal queueing mode. */
 257                         /* Packet scheduler handles dropping of further packets. */
 258                         netif_stop_queue(dev);
 259
 260                         /* We won't see all dropped packets individually, so overrun
 261                          * error is more appropriate. */
 262                         dev->stats.tx_fifo_errors++;
 263                 } else {
 264                         /* Single queue mode.
 265                          * Driver handles dropping of all packets itself. */
 266                         goto drop;
 267                 }
 268         }
 269
 270         /* Enqueue packet */
 271         skb_queue_tail(&tun->readq, skb);
 272         dev->trans_start = jiffies;
 273
 274         /* Notify and wake up reader process */
 275         if (tun->flags & TUN_FASYNC)
 276                 kill_fasync(&tun->fasync, SIGIO, POLL_IN);
 277         wake_up_interruptible(&tun->read_wait);
 278         return 0;
 279
 280 drop:
 281         dev->stats.tx_dropped++;
 282         kfree_skb(skb);
 283         return 0;
 284 }
 285
 286 static void tun_net_mclist(struct net_device *dev)
 287 {
 288         /*
 289          * This callback is supposed to deal with mc filter in
 290          * _rx_ path and has nothing to do with the _tx_ path.
 291          * In rx path we always accept everything userspace gives us.
 292          */
 293         return;
 294 }
 295
 296 #define MIN_MTU 68
 297 #define MAX_MTU 65535
 298
 299 static int
 300 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 301 {
 302         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 303                 return -EINVAL;
 304         dev->mtu = new_mtu;
 305         return 0;
 306 }
 307
 308 static const struct net_device_ops tun_netdev_ops = {
 309         .ndo_open               = tun_net_open,
 310         .ndo_stop               = tun_net_close,
 311         .ndo_start_xmit         = tun_net_xmit,
 312         .ndo_change_mtu         = tun_net_change_mtu,
 313 };
 314
 315 static const struct net_device_ops tap_netdev_ops = {
 316         .ndo_open               = tun_net_open,
 317         .ndo_stop               = tun_net_close,
 318         .ndo_start_xmit         = tun_net_xmit,
 319         .ndo_change_mtu         = tun_net_change_mtu,
 320         .ndo_set_multicast_list = tun_net_mclist,
 321         .ndo_set_mac_address    = eth_mac_addr,
 322         .ndo_validate_addr      = eth_validate_addr,
 323 };
 324
 325 /* Initialize net device. */
 326 static void tun_net_init(struct net_device *dev)
 327 {
 328         struct tun_struct *tun = netdev_priv(dev);
 329
 330         switch (tun->flags & TUN_TYPE_MASK) {
 331         case TUN_TUN_DEV:
 332                 dev->netdev_ops = &tun_netdev_ops;
 333
 334                 /* Point-to-Point TUN Device */
 335                 dev->hard_header_len = 0;
 336                 dev->addr_len = 0;
 337                 dev->mtu = 1500;
 338
 339                 /* Zero header length */
 340                 dev->type = ARPHRD_NONE;
 341                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 342                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 343                 break;
 344
 345         case TUN_TAP_DEV:
 346                 dev->netdev_ops = &tun_netdev_ops;
 347                 /* Ethernet TAP Device */
 348                 ether_setup(dev);
 349
 350                 random_ether_addr(dev->dev_addr);
 351
 352                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 353                 break;
 354         }
 355 }
 356
 357 /* Character device part */
 358
 359 /* Poll */
 360 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
 361 {
 362         struct tun_struct *tun = file->private_data;
 363         unsigned int mask = POLLOUT | POLLWRNORM;
 364
 365         if (!tun)
 366                 return -EBADFD;
 367
 368         DBG(KERN_INFO "%s: tun_chr_poll\n", tun->dev->name);
 369
 370         poll_wait(file, &tun->read_wait, wait);
 371
 372         if (!skb_queue_empty(&tun->readq))
 373                 mask |= POLLIN | POLLRDNORM;
 374
 375         return mask;
 376 }
 377
 378 /* prepad is the amount to reserve at front.  len is length after that.
 379  * linear is a hint as to how much to copy (usually headers). */
 380 static struct sk_buff *tun_alloc_skb(size_t prepad, size_t len, size_t linear,
 381                                      gfp_t gfp)
 382 {
 383         struct sk_buff *skb;
 384         unsigned int i;
 385
 386         skb = alloc_skb(prepad + len, gfp|__GFP_NOWARN);
 387         if (skb) {
 388                 skb_reserve(skb, prepad);
 389                 skb_put(skb, len);
 390                 return skb;
 391         }
 392
 393         /* Under a page?  Don't bother with paged skb. */
 394         if (prepad + len < PAGE_SIZE)
 395                 return NULL;
 396
 397         /* Start with a normal skb, and add pages. */
 398         skb = alloc_skb(prepad + linear, gfp);
 399         if (!skb)
 400                 return NULL;
 401
 402         skb_reserve(skb, prepad);
 403         skb_put(skb, linear);
 404
 405         len -= linear;
 406
 407         for (i = 0; i < MAX_SKB_FRAGS; i++) {
 408                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 409
 410                 f->page = alloc_page(gfp|__GFP_ZERO);
 411                 if (!f->page)
 412                         break;
 413
 414                 f->page_offset = 0;
 415                 f->size = PAGE_SIZE;
 416
 417                 skb->data_len += PAGE_SIZE;
 418                 skb->len += PAGE_SIZE;
 419                 skb->truesize += PAGE_SIZE;
 420                 skb_shinfo(skb)->nr_frags++;
 421
 422                 if (len < PAGE_SIZE) {
 423                         len = 0;
 424                         break;
 425                 }
 426                 len -= PAGE_SIZE;
 427         }
 428
 429         /* Too large, or alloc fail? */
 430         if (unlikely(len)) {
 431                 kfree_skb(skb);
 432                 skb = NULL;
 433         }
 434
 435         return skb;
 436 }
 437
 438 /* Get packet from user space buffer */
 439 static __inline__ ssize_t tun_get_user(struct tun_struct *tun, struct iovec *iv, size_t count)
 440 {
 441         struct tun_pi pi = { 0, __constant_htons(ETH_P_IP) };
 442         struct sk_buff *skb;
 443         size_t len = count, align = 0;
 444         struct virtio_net_hdr gso = { 0 };
 445
 446         if (!(tun->flags & TUN_NO_PI)) {
 447                 if ((len -= sizeof(pi)) > count)
 448                         return -EINVAL;
 449
 450                 if(memcpy_fromiovec((void *)&pi, iv, sizeof(pi)))
 451                         return -EFAULT;
 452         }
 453
 454         if (tun->flags & TUN_VNET_HDR) {
 455                 if ((len -= sizeof(gso)) > count)
 456                         return -EINVAL;
 457
 458                 if (memcpy_fromiovec((void *)&gso, iv, sizeof(gso)))
 459                         return -EFAULT;
 460
 461                 if (gso.hdr_len > len)
 462                         return -EINVAL;
 463         }
 464
 465         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
 466                 align = NET_IP_ALIGN;
 467                 if (unlikely(len < ETH_HLEN))
 468                         return -EINVAL;
 469         }
 470
 471         if (!(skb = tun_alloc_skb(align, len, gso.hdr_len, GFP_KERNEL))) {
 472                 tun->dev->stats.rx_dropped++;
 473                 return -ENOMEM;
 474         }
 475
 476         if (skb_copy_datagram_from_iovec(skb, 0, iv, len)) {
 477                 tun->dev->stats.rx_dropped++;
 478                 kfree_skb(skb);
 479                 return -EFAULT;
 480         }
 481
 482         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 483                 if (!skb_partial_csum_set(skb, gso.csum_start,
 484                                           gso.csum_offset)) {
 485                         tun->dev->stats.rx_frame_errors++;
 486                         kfree_skb(skb);
 487                         return -EINVAL;
 488                 }
 489         } else if (tun->flags & TUN_NOCHECKSUM)
 490                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 491
 492         switch (tun->flags & TUN_TYPE_MASK) {
 493         case TUN_TUN_DEV:
 494                 if (tun->flags & TUN_NO_PI) {
 495                         switch (skb->data[0] & 0xf0) {
 496                         case 0x40:
 497                                 pi.proto = htons(ETH_P_IP);
 498                                 break;
 499                         case 0x60:
 500                                 pi.proto = htons(ETH_P_IPV6);
 501                                 break;
 502                         default:
 503                                 tun->dev->stats.rx_dropped++;
 504                                 kfree_skb(skb);
 505                                 return -EINVAL;
 506                         }
 507                 }
 508
 509                 skb_reset_mac_header(skb);
 510                 skb->protocol = pi.proto;
 511                 skb->dev = tun->dev;
 512                 break;
 513         case TUN_TAP_DEV:
 514                 skb->protocol = eth_type_trans(skb, tun->dev);
 515                 break;
 516         };
 517
 518         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 519                 pr_debug("GSO!\n");
 520                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 521                 case VIRTIO_NET_HDR_GSO_TCPV4:
 522                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 523                         break;
 524                 case VIRTIO_NET_HDR_GSO_TCPV6:
 525                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 526                         break;
 527                 default:
 528                         tun->dev->stats.rx_frame_errors++;
 529                         kfree_skb(skb);
 530                         return -EINVAL;
 531                 }
 532
 533                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
 534                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 535
 536                 skb_shinfo(skb)->gso_size = gso.gso_size;
 537                 if (skb_shinfo(skb)->gso_size == 0) {
 538                         tun->dev->stats.rx_frame_errors++;
 539                         kfree_skb(skb);
 540                         return -EINVAL;
 541                 }
 542
 543                 /* Header must be checked, and gso_segs computed. */
 544                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 545                 skb_shinfo(skb)->gso_segs = 0;
 546         }
 547
 548         netif_rx_ni(skb);
 549
 550         tun->dev->stats.rx_packets++;
 551         tun->dev->stats.rx_bytes += len;
 552
 553         return count;
 554 }
 555
 556 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
 557                               unsigned long count, loff_t pos)
 558 {
 559         struct tun_struct *tun = iocb->ki_filp->private_data;
 560
 561         if (!tun)
 562                 return -EBADFD;
 563
 564         DBG(KERN_INFO "%s: tun_chr_write %ld\n", tun->dev->name, count);
 565
 566         return tun_get_user(tun, (struct iovec *) iv, iov_length(iv, count));
 567 }
 568
 569 /* Put packet to the user space buffer */
 570 static __inline__ ssize_t tun_put_user(struct tun_struct *tun,
 571                                        struct sk_buff *skb,
 572                                        struct iovec *iv, int len)
 573 {
 574         struct tun_pi pi = { 0, skb->protocol };
 575         ssize_t total = 0;
 576
 577         if (!(tun->flags & TUN_NO_PI)) {
 578                 if ((len -= sizeof(pi)) < 0)
 579                         return -EINVAL;
 580
 581                 if (len < skb->len) {
 582                         /* Packet will be striped */
 583                         pi.flags |= TUN_PKT_STRIP;
 584                 }
 585
 586                 if (memcpy_toiovec(iv, (void *) &pi, sizeof(pi)))
 587                         return -EFAULT;
 588                 total += sizeof(pi);
 589         }
 590
 591         if (tun->flags & TUN_VNET_HDR) {
 592                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
 593                 if ((len -= sizeof(gso)) < 0)
 594                         return -EINVAL;
 595
 596                 if (skb_is_gso(skb)) {
 597                         struct skb_shared_info *sinfo = skb_shinfo(skb);
 598
 599                         /* This is a hint as to how much should be linear. */
 600                         gso.hdr_len = skb_headlen(skb);
 601                         gso.gso_size = sinfo->gso_size;
 602                         if (sinfo->gso_type & SKB_GSO_TCPV4)
 603                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 604                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
 605                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 606                         else
 607                                 BUG();
 608                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
 609                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 610                 } else
 611                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
 612
 613                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 614                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 615                         gso.csum_start = skb->csum_start - skb_headroom(skb);
 616                         gso.csum_offset = skb->csum_offset;
 617                 } /* else everything is zero */
 618
 619                 if (unlikely(memcpy_toiovec(iv, (void *)&gso, sizeof(gso))))
 620                         return -EFAULT;
 621                 total += sizeof(gso);
 622         }
 623
 624         len = min_t(int, skb->len, len);
 625
 626         skb_copy_datagram_iovec(skb, 0, iv, len);
 627         total += len;
 628
 629         tun->dev->stats.tx_packets++;
 630         tun->dev->stats.tx_bytes += len;
 631
 632         return total;
 633 }
 634
 635 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
 636                             unsigned long count, loff_t pos)
 637 {
 638         struct file *file = iocb->ki_filp;
 639         struct tun_struct *tun = file->private_data;
 640         DECLARE_WAITQUEUE(wait, current);
 641         struct sk_buff *skb;
 642         ssize_t len, ret = 0;
 643
 644         if (!tun)
 645                 return -EBADFD;
 646
 647         DBG(KERN_INFO "%s: tun_chr_read\n", tun->dev->name);
 648
 649         len = iov_length(iv, count);
 650         if (len < 0)
 651                 return -EINVAL;
 652
 653         add_wait_queue(&tun->read_wait, &wait);
 654         while (len) {
 655                 current->state = TASK_INTERRUPTIBLE;
 656
 657                 /* Read frames from the queue */
 658                 if (!(skb=skb_dequeue(&tun->readq))) {
 659                         if (file->f_flags & O_NONBLOCK) {
 660                                 ret = -EAGAIN;
 661                                 break;
 662                         }
 663                         if (signal_pending(current)) {
 664                                 ret = -ERESTARTSYS;
 665                                 break;
 666                         }
 667
 668                         /* Nothing to read, let's sleep */
 669                         schedule();
 670                         continue;
 671                 }
 672                 netif_wake_queue(tun->dev);
 673
 674                 ret = tun_put_user(tun, skb, (struct iovec *) iv, len);
 675                 kfree_skb(skb);
 676                 break;
 677         }
 678
 679         current->state = TASK_RUNNING;
 680         remove_wait_queue(&tun->read_wait, &wait);
 681
 682         return ret;
 683 }
 684
 685 static void tun_setup(struct net_device *dev)
 686 {
 687         struct tun_struct *tun = netdev_priv(dev);
 688
 689         skb_queue_head_init(&tun->readq);
 690         init_waitqueue_head(&tun->read_wait);
 691
 692         tun->owner = -1;
 693         tun->group = -1;
 694
 695         dev->ethtool_ops = &tun_ethtool_ops;
 696         dev->destructor = free_netdev;
 697         dev->features |= NETIF_F_NETNS_LOCAL;
 698 }
 699
 700 static struct tun_struct *tun_get_by_name(struct tun_net *tn, const char *name)
 701 {
 702         struct tun_struct *tun;
 703
 704         ASSERT_RTNL();
 705         list_for_each_entry(tun, &tn->dev_list, list) {
 706                 if (!strncmp(tun->dev->name, name, IFNAMSIZ))
 707                     return tun;
 708         }
 709
 710         return NULL;
 711 }
 712
 713 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
 714 {
 715         struct tun_net *tn;
 716         struct tun_struct *tun;
 717         struct net_device *dev;
 718         const struct cred *cred = current_cred();
 719         int err;
 720
 721         tn = net_generic(net, tun_net_id);
 722         tun = tun_get_by_name(tn, ifr->ifr_name);
 723         if (tun) {
 724                 if (tun->attached)
 725                         return -EBUSY;
 726
 727                 /* Check permissions */
 728                 if (((tun->owner != -1 &&
 729                       cred->euid != tun->owner) ||
 730                      (tun->group != -1 &&
 731                       cred->egid != tun->group)) &&
 732                     !capable(CAP_NET_ADMIN)) {
 733                         return -EPERM;
 734                 }
 735         }
 736         else if (__dev_get_by_name(net, ifr->ifr_name))
 737                 return -EINVAL;
 738         else {
 739                 char *name;
 740                 unsigned long flags = 0;
 741
 742                 err = -EINVAL;
 743
 744                 if (!capable(CAP_NET_ADMIN))
 745                         return -EPERM;
 746
 747                 /* Set dev type */
 748                 if (ifr->ifr_flags & IFF_TUN) {
 749                         /* TUN device */
 750                         flags |= TUN_TUN_DEV;
 751                         name = "tun%d";
 752                 } else if (ifr->ifr_flags & IFF_TAP) {
 753                         /* TAP device */
 754                         flags |= TUN_TAP_DEV;
 755                         name = "tap%d";
 756                 } else
 757                         goto failed;
 758
 759                 if (*ifr->ifr_name)
 760                         name = ifr->ifr_name;
 761
 762                 dev = alloc_netdev(sizeof(struct tun_struct), name,
 763                                    tun_setup);
 764                 if (!dev)
 765                         return -ENOMEM;
 766
 767                 dev_net_set(dev, net);
 768
 769                 tun = netdev_priv(dev);
 770                 tun->dev = dev;
 771                 tun->flags = flags;
 772                 tun->txflt.count = 0;
 773
 774                 tun_net_init(dev);
 775
 776                 if (strchr(dev->name, '%')) {
 777                         err = dev_alloc_name(dev, dev->name);
 778                         if (err < 0)
 779                                 goto err_free_dev;
 780                 }
 781
 782                 err = register_netdevice(tun->dev);
 783                 if (err < 0)
 784                         goto err_free_dev;
 785
 786                 list_add(&tun->list, &tn->dev_list);
 787         }
 788
 789         DBG(KERN_INFO "%s: tun_set_iff\n", tun->dev->name);
 790
 791         if (ifr->ifr_flags & IFF_NO_PI)
 792                 tun->flags |= TUN_NO_PI;
 793         else
 794                 tun->flags &= ~TUN_NO_PI;
 795
 796         if (ifr->ifr_flags & IFF_ONE_QUEUE)
 797                 tun->flags |= TUN_ONE_QUEUE;
 798         else
 799                 tun->flags &= ~TUN_ONE_QUEUE;
 800
 801         if (ifr->ifr_flags & IFF_VNET_HDR)
 802                 tun->flags |= TUN_VNET_HDR;
 803         else
 804                 tun->flags &= ~TUN_VNET_HDR;
 805
 806         file->private_data = tun;
 807         tun->attached = 1;
 808         get_net(dev_net(tun->dev));
 809
 810         /* Make sure persistent devices do not get stuck in
 811          * xoff state.
 812          */
 813         if (netif_running(tun->dev))
 814                 netif_wake_queue(tun->dev);
 815
 816         strcpy(ifr->ifr_name, tun->dev->name);
 817         return 0;
 818
 819  err_free_dev:
 820         free_netdev(dev);
 821  failed:
 822         return err;
 823 }
 824
 825 static int tun_get_iff(struct net *net, struct file *file, struct ifreq *ifr)
 826 {
 827         struct tun_struct *tun = file->private_data;
 828
 829         if (!tun)
 830                 return -EBADFD;
 831
 832         DBG(KERN_INFO "%s: tun_get_iff\n", tun->dev->name);
 833
 834         strcpy(ifr->ifr_name, tun->dev->name);
 835
 836         ifr->ifr_flags = 0;
 837
 838         if (ifr->ifr_flags & TUN_TUN_DEV)
 839                 ifr->ifr_flags |= IFF_TUN;
 840         else
 841                 ifr->ifr_flags |= IFF_TAP;
 842
 843         if (tun->flags & TUN_NO_PI)
 844                 ifr->ifr_flags |= IFF_NO_PI;
 845
 846         if (tun->flags & TUN_ONE_QUEUE)
 847                 ifr->ifr_flags |= IFF_ONE_QUEUE;
 848
 849         if (tun->flags & TUN_VNET_HDR)
 850                 ifr->ifr_flags |= IFF_VNET_HDR;
 851
 852         return 0;
 853 }
 854
 855 /* This is like a cut-down ethtool ops, except done via tun fd so no
 856  * privs required. */
 857 static int set_offload(struct net_device *dev, unsigned long arg)
 858 {
 859         unsigned int old_features, features;
 860
 861         old_features = dev->features;
 862         /* Unset features, set them as we chew on the arg. */
 863         features = (old_features & ~(NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST
 864                                     |NETIF_F_TSO_ECN|NETIF_F_TSO|NETIF_F_TSO6));
 865
 866         if (arg & TUN_F_CSUM) {
 867                 features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
 868                 arg &= ~TUN_F_CSUM;
 869
 870                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
 871                         if (arg & TUN_F_TSO_ECN) {
 872                                 features |= NETIF_F_TSO_ECN;
 873                                 arg &= ~TUN_F_TSO_ECN;
 874                         }
 875                         if (arg & TUN_F_TSO4)
 876                                 features |= NETIF_F_TSO;
 877                         if (arg & TUN_F_TSO6)
 878                                 features |= NETIF_F_TSO6;
 879                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
 880                 }
 881         }
 882
 883         /* This gives the user a way to test for new features in future by
 884          * trying to set them. */
 885         if (arg)
 886                 return -EINVAL;
 887
 888         dev->features = features;
 889         if (old_features != dev->features)
 890                 netdev_features_change(dev);
 891
 892         return 0;
 893 }
 894
 895 static int tun_chr_ioctl(struct inode *inode, struct file *file,
 896                          unsigned int cmd, unsigned long arg)
 897 {
 898         struct tun_struct *tun = file->private_data;
 899         void __user* argp = (void __user*)arg;
 900         struct ifreq ifr;
 901         int ret;
 902
 903         if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
 904                 if (copy_from_user(&ifr, argp, sizeof ifr))
 905                         return -EFAULT;
 906
 907         if (cmd == TUNSETIFF && !tun) {
 908                 int err;
 909
 910                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
 911
 912                 rtnl_lock();
 913                 err = tun_set_iff(current->nsproxy->net_ns, file, &ifr);
 914                 rtnl_unlock();
 915
 916                 if (err)
 917                         return err;
 918
 919                 if (copy_to_user(argp, &ifr, sizeof(ifr)))
 920                         return -EFAULT;
 921                 return 0;
 922         }
 923
 924         if (cmd == TUNGETFEATURES) {
 925                 /* Currently this just means: "what IFF flags are valid?".
 926                  * This is needed because we never checked for invalid flags on
 927                  * TUNSETIFF. */
 928                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
 929                                 IFF_VNET_HDR,
 930                                 (unsigned int __user*)argp);
 931         }
 932
 933         if (!tun)
 934                 return -EBADFD;
 935
 936         DBG(KERN_INFO "%s: tun_chr_ioctl cmd %d\n", tun->dev->name, cmd);
 937
 938         switch (cmd) {
 939         case TUNGETIFF:
 940                 ret = tun_get_iff(current->nsproxy->net_ns, file, &ifr);
 941                 if (ret)
 942                         return ret;
 943
 944                 if (copy_to_user(argp, &ifr, sizeof(ifr)))
 945                         return -EFAULT;
 946                 break;
 947
 948         case TUNSETNOCSUM:
 949                 /* Disable/Enable checksum */
 950                 if (arg)
 951                         tun->flags |= TUN_NOCHECKSUM;
 952                 else
 953                         tun->flags &= ~TUN_NOCHECKSUM;
 954
 955                 DBG(KERN_INFO "%s: checksum %s\n",
 956                     tun->dev->name, arg ? "disabled" : "enabled");
 957                 break;
 958
 959         case TUNSETPERSIST:
 960                 /* Disable/Enable persist mode */
 961                 if (arg)
 962                         tun->flags |= TUN_PERSIST;
 963                 else
 964                         tun->flags &= ~TUN_PERSIST;
 965
 966                 DBG(KERN_INFO "%s: persist %s\n",
 967                     tun->dev->name, arg ? "enabled" : "disabled");
 968                 break;
 969
 970         case TUNSETOWNER:
 971                 /* Set owner of the device */
 972                 tun->owner = (uid_t) arg;
 973
 974                 DBG(KERN_INFO "%s: owner set to %d\n", tun->dev->name, tun->owner);
 975                 break;
 976
 977         case TUNSETGROUP:
 978                 /* Set group of the device */
 979                 tun->group= (gid_t) arg;
 980
 981                 DBG(KERN_INFO "%s: group set to %d\n", tun->dev->name, tun->group);
 982                 break;
 983
 984         case TUNSETLINK:
 985                 /* Only allow setting the type when the interface is down */
 986                 rtnl_lock();
 987                 if (tun->dev->flags & IFF_UP) {
 988                         DBG(KERN_INFO "%s: Linktype set failed because interface is up\n",
 989                                 tun->dev->name);
 990                         ret = -EBUSY;
 991                 } else {
 992                         tun->dev->type = (int) arg;
 993                         DBG(KERN_INFO "%s: linktype set to %d\n", tun->dev->name, tun->dev->type);
 994                         ret = 0;
 995                 }
 996                 rtnl_unlock();
 997                 return ret;
 998
 999 #ifdef TUN_DEBUG
1000         case TUNSETDEBUG:
1001                 tun->debug = arg;
1002                 break;
1003 #endif
1004         case TUNSETOFFLOAD:
1005                 rtnl_lock();
1006                 ret = set_offload(tun->dev, arg);
1007                 rtnl_unlock();
1008                 return ret;
1009
1010         case TUNSETTXFILTER:
1011                 /* Can be set only for TAPs */
1012                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1013                         return -EINVAL;
1014                 rtnl_lock();
1015                 ret = update_filter(&tun->txflt, (void __user *)arg);
1016                 rtnl_unlock();
1017                 return ret;
1018
1019         case SIOCGIFHWADDR:
1020                 /* Get hw addres */
1021                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1022                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1023                 if (copy_to_user(argp, &ifr, sizeof ifr))
1024                         return -EFAULT;
1025                 return 0;
1026
1027         case SIOCSIFHWADDR:
1028                 /* Set hw address */
1029                 DBG(KERN_DEBUG "%s: set hw address: %pM\n",
1030                         tun->dev->name, ifr.ifr_hwaddr.sa_data);
1031
1032                 rtnl_lock();
1033                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1034                 rtnl_unlock();
1035                 return ret;
1036
1037         default:
1038                 return -EINVAL;
1039         };
1040
1041         return 0;
1042 }
1043
1044 static int tun_chr_fasync(int fd, struct file *file, int on)
1045 {
1046         struct tun_struct *tun = file->private_data;
1047         int ret;
1048
1049         if (!tun)
1050                 return -EBADFD;
1051
1052         DBG(KERN_INFO "%s: tun_chr_fasync %d\n", tun->dev->name, on);
1053
1054         lock_kernel();
1055         if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1056                 goto out;
1057
1058         if (on) {
1059                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1060                 if (ret)
1061                         goto out;
1062                 tun->flags |= TUN_FASYNC;
1063         } else
1064                 tun->flags &= ~TUN_FASYNC;
1065         ret = 0;
1066 out:
1067         unlock_kernel();
1068         return ret;
1069 }
1070
1071 static int tun_chr_open(struct inode *inode, struct file * file)
1072 {
1073         cycle_kernel_lock();
1074         DBG1(KERN_INFO "tunX: tun_chr_open\n");
1075         file->private_data = NULL;
1076         return 0;
1077 }
1078
1079 static int tun_chr_close(struct inode *inode, struct file *file)
1080 {
1081         struct tun_struct *tun = file->private_data;
1082
1083         if (!tun)
1084                 return 0;
1085
1086         DBG(KERN_INFO "%s: tun_chr_close\n", tun->dev->name);
1087
1088         rtnl_lock();
1089
1090         /* Detach from net device */
1091         file->private_data = NULL;
1092         tun->attached = 0;
1093         put_net(dev_net(tun->dev));
1094
1095         /* Drop read queue */
1096         skb_queue_purge(&tun->readq);
1097
1098         if (!(tun->flags & TUN_PERSIST)) {
1099                 list_del(&tun->list);
1100                 unregister_netdevice(tun->dev);
1101         }
1102
1103         rtnl_unlock();
1104
1105         return 0;
1106 }
1107
1108 static const struct file_operations tun_fops = {
1109         .owner  = THIS_MODULE,
1110         .llseek = no_llseek,
1111         .read  = do_sync_read,
1112         .aio_read  = tun_chr_aio_read,
1113         .write = do_sync_write,
1114         .aio_write = tun_chr_aio_write,
1115         .poll   = tun_chr_poll,
1116         .ioctl  = tun_chr_ioctl,
1117         .open   = tun_chr_open,
1118         .release = tun_chr_close,
1119         .fasync = tun_chr_fasync
1120 };
1121
1122 static struct miscdevice tun_miscdev = {
1123         .minor = TUN_MINOR,
1124         .name = "tun",
1125         .fops = &tun_fops,
1126 };
1127
1128 /* ethtool interface */
1129
1130 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1131 {
1132         cmd->supported          = 0;
1133         cmd->advertising        = 0;
1134         cmd->speed              = SPEED_10;
1135         cmd->duplex             = DUPLEX_FULL;
1136         cmd->port               = PORT_TP;
1137         cmd->phy_address        = 0;
1138         cmd->transceiver        = XCVR_INTERNAL;
1139         cmd->autoneg            = AUTONEG_DISABLE;
1140         cmd->maxtxpkt           = 0;
1141         cmd->maxrxpkt           = 0;
1142         return 0;
1143 }
1144
1145 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1146 {
1147         struct tun_struct *tun = netdev_priv(dev);
1148
1149         strcpy(info->driver, DRV_NAME);
1150         strcpy(info->version, DRV_VERSION);
1151         strcpy(info->fw_version, "N/A");
1152
1153         switch (tun->flags & TUN_TYPE_MASK) {
1154         case TUN_TUN_DEV:
1155                 strcpy(info->bus_info, "tun");
1156                 break;
1157         case TUN_TAP_DEV:
1158                 strcpy(info->bus_info, "tap");
1159                 break;
1160         }
1161 }
1162
1163 static u32 tun_get_msglevel(struct net_device *dev)
1164 {
1165 #ifdef TUN_DEBUG
1166         struct tun_struct *tun = netdev_priv(dev);
1167         return tun->debug;
1168 #else
1169         return -EOPNOTSUPP;
1170 #endif
1171 }
1172
1173 static void tun_set_msglevel(struct net_device *dev, u32 value)
1174 {
1175 #ifdef TUN_DEBUG
1176         struct tun_struct *tun = netdev_priv(dev);
1177         tun->debug = value;
1178 #endif
1179 }
1180
1181 static u32 tun_get_link(struct net_device *dev)
1182 {
1183         struct tun_struct *tun = netdev_priv(dev);
1184         return tun->attached;
1185 }
1186
1187 static u32 tun_get_rx_csum(struct net_device *dev)
1188 {
1189         struct tun_struct *tun = netdev_priv(dev);
1190         return (tun->flags & TUN_NOCHECKSUM) == 0;
1191 }
1192
1193 static int tun_set_rx_csum(struct net_device *dev, u32 data)
1194 {
1195         struct tun_struct *tun = netdev_priv(dev);
1196         if (data)
1197                 tun->flags &= ~TUN_NOCHECKSUM;
1198         else
1199                 tun->flags |= TUN_NOCHECKSUM;
1200         return 0;
1201 }
1202
1203 static const struct ethtool_ops tun_ethtool_ops = {
1204         .get_settings   = tun_get_settings,
1205         .get_drvinfo    = tun_get_drvinfo,
1206         .get_msglevel   = tun_get_msglevel,
1207         .set_msglevel   = tun_set_msglevel,
1208         .get_link       = tun_get_link,
1209         .get_rx_csum    = tun_get_rx_csum,
1210         .set_rx_csum    = tun_set_rx_csum
1211 };
1212
1213 static int tun_init_net(struct net *net)
1214 {
1215         struct tun_net *tn;
1216
1217         tn = kmalloc(sizeof(*tn), GFP_KERNEL);
1218         if (tn == NULL)
1219                 return -ENOMEM;
1220
1221         INIT_LIST_HEAD(&tn->dev_list);
1222
1223         if (net_assign_generic(net, tun_net_id, tn)) {
1224                 kfree(tn);
1225                 return -ENOMEM;
1226         }
1227
1228         return 0;
1229 }
1230
1231 static void tun_exit_net(struct net *net)
1232 {
1233         struct tun_net *tn;
1234         struct tun_struct *tun, *nxt;
1235
1236         tn = net_generic(net, tun_net_id);
1237
1238         rtnl_lock();
1239         list_for_each_entry_safe(tun, nxt, &tn->dev_list, list) {
1240                 DBG(KERN_INFO "%s cleaned up\n", tun->dev->name);
1241                 unregister_netdevice(tun->dev);
1242         }
1243         rtnl_unlock();
1244
1245         kfree(tn);
1246 }
1247
1248 static struct pernet_operations tun_net_ops = {
1249         .init = tun_init_net,
1250         .exit = tun_exit_net,
1251 };
1252
1253 static int __init tun_init(void)
1254 {
1255         int ret = 0;
1256
1257         printk(KERN_INFO "tun: %s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1258         printk(KERN_INFO "tun: %s\n", DRV_COPYRIGHT);
1259
1260         ret = register_pernet_gen_device(&tun_net_id, &tun_net_ops);
1261         if (ret) {
1262                 printk(KERN_ERR "tun: Can't register pernet ops\n");
1263                 goto err_pernet;
1264         }
1265
1266         ret = misc_register(&tun_miscdev);
1267         if (ret) {
1268                 printk(KERN_ERR "tun: Can't register misc device %d\n", TUN_MINOR);
1269                 goto err_misc;
1270         }
1271         return 0;
1272
1273 err_misc:
1274         unregister_pernet_gen_device(tun_net_id, &tun_net_ops);
1275 err_pernet:
1276         return ret;
1277 }
1278
1279 static void tun_cleanup(void)
1280 {
1281         misc_deregister(&tun_miscdev);
1282         unregister_pernet_gen_device(tun_net_id, &tun_net_ops);
1283 }
1284
1285 module_init(tun_init);
1286 module_exit(tun_cleanup);
1287 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1288 MODULE_AUTHOR(DRV_COPYRIGHT);
1289 MODULE_LICENSE("GPL");
1290 MODULE_ALIAS_MISCDEV(TUN_MINOR);