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