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