Pileus Git - ~andy/linux/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 eth_random_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 pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/major.h>
  48 #include <linux/slab.h>
  49 #include <linux/poll.h>
  50 #include <linux/fcntl.h>
  51 #include <linux/init.h>
  52 #include <linux/skbuff.h>
  53 #include <linux/netdevice.h>
  54 #include <linux/etherdevice.h>
  55 #include <linux/miscdevice.h>
  56 #include <linux/ethtool.h>
  57 #include <linux/rtnetlink.h>
  58 #include <linux/compat.h>
  59 #include <linux/if.h>
  60 #include <linux/if_arp.h>
  61 #include <linux/if_ether.h>
  62 #include <linux/if_tun.h>
  63 #include <linux/crc32.h>
  64 #include <linux/nsproxy.h>
  65 #include <linux/virtio_net.h>
  66 #include <linux/rcupdate.h>
  67 #include <net/net_namespace.h>
  68 #include <net/netns/generic.h>
  69 #include <net/rtnetlink.h>
  70 #include <net/sock.h>
  71
  72 #include <asm/uaccess.h>
  73
  74 /* Uncomment to enable debugging */
  75 /* #define TUN_DEBUG 1 */
  76
  77 #ifdef TUN_DEBUG
  78 static int debug;
  79
  80 #define tun_debug(level, tun, fmt, args...)                     \
  81 do {                                                            \
  82         if (tun->debug)                                         \
  83                 netdev_printk(level, tun->dev, fmt, ##args);    \
  84 } while (0)
  85 #define DBG1(level, fmt, args...)                               \
  86 do {                                                            \
  87         if (debug == 2)                                         \
  88                 printk(level fmt, ##args);                      \
  89 } while (0)
  90 #else
  91 #define tun_debug(level, tun, fmt, args...)                     \
  92 do {                                                            \
  93         if (0)                                                  \
  94                 netdev_printk(level, tun->dev, fmt, ##args);    \
  95 } while (0)
  96 #define DBG1(level, fmt, args...)                               \
  97 do {                                                            \
  98         if (0)                                                  \
  99                 printk(level fmt, ##args);                      \
 100 } while (0)
 101 #endif
 102
 103 #define GOODCOPY_LEN 128
 104
 105 #define FLT_EXACT_COUNT 8
 106 struct tap_filter {
 107         unsigned int    count;    /* Number of addrs. Zero means disabled */
 108         u32             mask[2];  /* Mask of the hashed addrs */
 109         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 110 };
 111
 112 struct tun_file {
 113         atomic_t count;
 114         struct tun_struct *tun;
 115         struct net *net;
 116 };
 117
 118 struct tun_sock;
 119
 120 struct tun_struct {
 121         struct tun_file         *tfile;
 122         unsigned int            flags;
 123         kuid_t                  owner;
 124         kgid_t                  group;
 125
 126         struct net_device       *dev;
 127         netdev_features_t       set_features;
 128 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 129                           NETIF_F_TSO6|NETIF_F_UFO)
 130         struct fasync_struct    *fasync;
 131
 132         struct tap_filter       txflt;
 133         struct socket           socket;
 134         struct socket_wq        wq;
 135
 136         int                     vnet_hdr_sz;
 137
 138 #ifdef TUN_DEBUG
 139         int debug;
 140 #endif
 141 };
 142
 143 struct tun_sock {
 144         struct sock             sk;
 145         struct tun_struct       *tun;
 146 };
 147
 148 static inline struct tun_sock *tun_sk(struct sock *sk)
 149 {
 150         return container_of(sk, struct tun_sock, sk);
 151 }
 152
 153 static int tun_attach(struct tun_struct *tun, struct file *file)
 154 {
 155         struct tun_file *tfile = file->private_data;
 156         int err;
 157
 158         ASSERT_RTNL();
 159
 160         netif_tx_lock_bh(tun->dev);
 161
 162         err = -EINVAL;
 163         if (tfile->tun)
 164                 goto out;
 165
 166         err = -EBUSY;
 167         if (tun->tfile)
 168                 goto out;
 169
 170         err = 0;
 171         tfile->tun = tun;
 172         tun->tfile = tfile;
 173         tun->socket.file = file;
 174         netif_carrier_on(tun->dev);
 175         dev_hold(tun->dev);
 176         sock_hold(tun->socket.sk);
 177         atomic_inc(&tfile->count);
 178
 179 out:
 180         netif_tx_unlock_bh(tun->dev);
 181         return err;
 182 }
 183
 184 static void __tun_detach(struct tun_struct *tun)
 185 {
 186         /* Detach from net device */
 187         netif_tx_lock_bh(tun->dev);
 188         netif_carrier_off(tun->dev);
 189         tun->tfile = NULL;
 190         netif_tx_unlock_bh(tun->dev);
 191
 192         /* Drop read queue */
 193         skb_queue_purge(&tun->socket.sk->sk_receive_queue);
 194
 195         /* Drop the extra count on the net device */
 196         dev_put(tun->dev);
 197 }
 198
 199 static void tun_detach(struct tun_struct *tun)
 200 {
 201         rtnl_lock();
 202         __tun_detach(tun);
 203         rtnl_unlock();
 204 }
 205
 206 static struct tun_struct *__tun_get(struct tun_file *tfile)
 207 {
 208         struct tun_struct *tun = NULL;
 209
 210         if (atomic_inc_not_zero(&tfile->count))
 211                 tun = tfile->tun;
 212
 213         return tun;
 214 }
 215
 216 static struct tun_struct *tun_get(struct file *file)
 217 {
 218         return __tun_get(file->private_data);
 219 }
 220
 221 static void tun_put(struct tun_struct *tun)
 222 {
 223         struct tun_file *tfile = tun->tfile;
 224
 225         if (atomic_dec_and_test(&tfile->count))
 226                 tun_detach(tfile->tun);
 227 }
 228
 229 /* TAP filtering */
 230 static void addr_hash_set(u32 *mask, const u8 *addr)
 231 {
 232         int n = ether_crc(ETH_ALEN, addr) >> 26;
 233         mask[n >> 5] |= (1 << (n & 31));
 234 }
 235
 236 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 237 {
 238         int n = ether_crc(ETH_ALEN, addr) >> 26;
 239         return mask[n >> 5] & (1 << (n & 31));
 240 }
 241
 242 static int update_filter(struct tap_filter *filter, void __user *arg)
 243 {
 244         struct { u8 u[ETH_ALEN]; } *addr;
 245         struct tun_filter uf;
 246         int err, alen, n, nexact;
 247
 248         if (copy_from_user(&uf, arg, sizeof(uf)))
 249                 return -EFAULT;
 250
 251         if (!uf.count) {
 252                 /* Disabled */
 253                 filter->count = 0;
 254                 return 0;
 255         }
 256
 257         alen = ETH_ALEN * uf.count;
 258         addr = kmalloc(alen, GFP_KERNEL);
 259         if (!addr)
 260                 return -ENOMEM;
 261
 262         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 263                 err = -EFAULT;
 264                 goto done;
 265         }
 266
 267         /* The filter is updated without holding any locks. Which is
 268          * perfectly safe. We disable it first and in the worst
 269          * case we'll accept a few undesired packets. */
 270         filter->count = 0;
 271         wmb();
 272
 273         /* Use first set of addresses as an exact filter */
 274         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 275                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 276
 277         nexact = n;
 278
 279         /* Remaining multicast addresses are hashed,
 280          * unicast will leave the filter disabled. */
 281         memset(filter->mask, 0, sizeof(filter->mask));
 282         for (; n < uf.count; n++) {
 283                 if (!is_multicast_ether_addr(addr[n].u)) {
 284                         err = 0; /* no filter */
 285                         goto done;
 286                 }
 287                 addr_hash_set(filter->mask, addr[n].u);
 288         }
 289
 290         /* For ALLMULTI just set the mask to all ones.
 291          * This overrides the mask populated above. */
 292         if ((uf.flags & TUN_FLT_ALLMULTI))
 293                 memset(filter->mask, ~0, sizeof(filter->mask));
 294
 295         /* Now enable the filter */
 296         wmb();
 297         filter->count = nexact;
 298
 299         /* Return the number of exact filters */
 300         err = nexact;
 301
 302 done:
 303         kfree(addr);
 304         return err;
 305 }
 306
 307 /* Returns: 0 - drop, !=0 - accept */
 308 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 309 {
 310         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 311          * at this point. */
 312         struct ethhdr *eh = (struct ethhdr *) skb->data;
 313         int i;
 314
 315         /* Exact match */
 316         for (i = 0; i < filter->count; i++)
 317                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 318                         return 1;
 319
 320         /* Inexact match (multicast only) */
 321         if (is_multicast_ether_addr(eh->h_dest))
 322                 return addr_hash_test(filter->mask, eh->h_dest);
 323
 324         return 0;
 325 }
 326
 327 /*
 328  * Checks whether the packet is accepted or not.
 329  * Returns: 0 - drop, !=0 - accept
 330  */
 331 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 332 {
 333         if (!filter->count)
 334                 return 1;
 335
 336         return run_filter(filter, skb);
 337 }
 338
 339 /* Network device part of the driver */
 340
 341 static const struct ethtool_ops tun_ethtool_ops;
 342
 343 /* Net device detach from fd. */
 344 static void tun_net_uninit(struct net_device *dev)
 345 {
 346         struct tun_struct *tun = netdev_priv(dev);
 347         struct tun_file *tfile = tun->tfile;
 348
 349         /* Inform the methods they need to stop using the dev.
 350          */
 351         if (tfile) {
 352                 wake_up_all(&tun->wq.wait);
 353                 if (atomic_dec_and_test(&tfile->count))
 354                         __tun_detach(tun);
 355         }
 356 }
 357
 358 static void tun_free_netdev(struct net_device *dev)
 359 {
 360         struct tun_struct *tun = netdev_priv(dev);
 361
 362         BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED, &tun->socket.flags));
 363
 364         sk_release_kernel(tun->socket.sk);
 365 }
 366
 367 /* Net device open. */
 368 static int tun_net_open(struct net_device *dev)
 369 {
 370         netif_start_queue(dev);
 371         return 0;
 372 }
 373
 374 /* Net device close. */
 375 static int tun_net_close(struct net_device *dev)
 376 {
 377         netif_stop_queue(dev);
 378         return 0;
 379 }
 380
 381 /* Net device start xmit */
 382 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 383 {
 384         struct tun_struct *tun = netdev_priv(dev);
 385
 386         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 387
 388         /* Drop packet if interface is not attached */
 389         if (!tun->tfile)
 390                 goto drop;
 391
 392         /* Drop if the filter does not like it.
 393          * This is a noop if the filter is disabled.
 394          * Filter can be enabled only for the TAP devices. */
 395         if (!check_filter(&tun->txflt, skb))
 396                 goto drop;
 397
 398         if (tun->socket.sk->sk_filter &&
 399             sk_filter(tun->socket.sk, skb))
 400                 goto drop;
 401
 402         if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) {
 403                 if (!(tun->flags & TUN_ONE_QUEUE)) {
 404                         /* Normal queueing mode. */
 405                         /* Packet scheduler handles dropping of further packets. */
 406                         netif_stop_queue(dev);
 407
 408                         /* We won't see all dropped packets individually, so overrun
 409                          * error is more appropriate. */
 410                         dev->stats.tx_fifo_errors++;
 411                 } else {
 412                         /* Single queue mode.
 413                          * Driver handles dropping of all packets itself. */
 414                         goto drop;
 415                 }
 416         }
 417
 418         /* Orphan the skb - required as we might hang on to it
 419          * for indefinite time. */
 420         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 421                 goto drop;
 422         skb_orphan(skb);
 423
 424         /* Enqueue packet */
 425         skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb);
 426
 427         /* Notify and wake up reader process */
 428         if (tun->flags & TUN_FASYNC)
 429                 kill_fasync(&tun->fasync, SIGIO, POLL_IN);
 430         wake_up_interruptible_poll(&tun->wq.wait, POLLIN |
 431                                    POLLRDNORM | POLLRDBAND);
 432         return NETDEV_TX_OK;
 433
 434 drop:
 435         dev->stats.tx_dropped++;
 436         kfree_skb(skb);
 437         return NETDEV_TX_OK;
 438 }
 439
 440 static void tun_net_mclist(struct net_device *dev)
 441 {
 442         /*
 443          * This callback is supposed to deal with mc filter in
 444          * _rx_ path and has nothing to do with the _tx_ path.
 445          * In rx path we always accept everything userspace gives us.
 446          */
 447 }
 448
 449 #define MIN_MTU 68
 450 #define MAX_MTU 65535
 451
 452 static int
 453 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 454 {
 455         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 456                 return -EINVAL;
 457         dev->mtu = new_mtu;
 458         return 0;
 459 }
 460
 461 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 462         netdev_features_t features)
 463 {
 464         struct tun_struct *tun = netdev_priv(dev);
 465
 466         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 467 }
 468 #ifdef CONFIG_NET_POLL_CONTROLLER
 469 static void tun_poll_controller(struct net_device *dev)
 470 {
 471         /*
 472          * Tun only receives frames when:
 473          * 1) the char device endpoint gets data from user space
 474          * 2) the tun socket gets a sendmsg call from user space
 475          * Since both of those are syncronous operations, we are guaranteed
 476          * never to have pending data when we poll for it
 477          * so theres nothing to do here but return.
 478          * We need this though so netpoll recognizes us as an interface that
 479          * supports polling, which enables bridge devices in virt setups to
 480          * still use netconsole
 481          */
 482         return;
 483 }
 484 #endif
 485 static const struct net_device_ops tun_netdev_ops = {
 486         .ndo_uninit             = tun_net_uninit,
 487         .ndo_open               = tun_net_open,
 488         .ndo_stop               = tun_net_close,
 489         .ndo_start_xmit         = tun_net_xmit,
 490         .ndo_change_mtu         = tun_net_change_mtu,
 491         .ndo_fix_features       = tun_net_fix_features,
 492 #ifdef CONFIG_NET_POLL_CONTROLLER
 493         .ndo_poll_controller    = tun_poll_controller,
 494 #endif
 495 };
 496
 497 static const struct net_device_ops tap_netdev_ops = {
 498         .ndo_uninit             = tun_net_uninit,
 499         .ndo_open               = tun_net_open,
 500         .ndo_stop               = tun_net_close,
 501         .ndo_start_xmit         = tun_net_xmit,
 502         .ndo_change_mtu         = tun_net_change_mtu,
 503         .ndo_fix_features       = tun_net_fix_features,
 504         .ndo_set_rx_mode        = tun_net_mclist,
 505         .ndo_set_mac_address    = eth_mac_addr,
 506         .ndo_validate_addr      = eth_validate_addr,
 507 #ifdef CONFIG_NET_POLL_CONTROLLER
 508         .ndo_poll_controller    = tun_poll_controller,
 509 #endif
 510 };
 511
 512 /* Initialize net device. */
 513 static void tun_net_init(struct net_device *dev)
 514 {
 515         struct tun_struct *tun = netdev_priv(dev);
 516
 517         switch (tun->flags & TUN_TYPE_MASK) {
 518         case TUN_TUN_DEV:
 519                 dev->netdev_ops = &tun_netdev_ops;
 520
 521                 /* Point-to-Point TUN Device */
 522                 dev->hard_header_len = 0;
 523                 dev->addr_len = 0;
 524                 dev->mtu = 1500;
 525
 526                 /* Zero header length */
 527                 dev->type = ARPHRD_NONE;
 528                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 529                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 530                 break;
 531
 532         case TUN_TAP_DEV:
 533                 dev->netdev_ops = &tap_netdev_ops;
 534                 /* Ethernet TAP Device */
 535                 ether_setup(dev);
 536                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 537
 538                 eth_hw_addr_random(dev);
 539
 540                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 541                 break;
 542         }
 543 }
 544
 545 /* Character device part */
 546
 547 /* Poll */
 548 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
 549 {
 550         struct tun_file *tfile = file->private_data;
 551         struct tun_struct *tun = __tun_get(tfile);
 552         struct sock *sk;
 553         unsigned int mask = 0;
 554
 555         if (!tun)
 556                 return POLLERR;
 557
 558         sk = tun->socket.sk;
 559
 560         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 561
 562         poll_wait(file, &tun->wq.wait, wait);
 563
 564         if (!skb_queue_empty(&sk->sk_receive_queue))
 565                 mask |= POLLIN | POLLRDNORM;
 566
 567         if (sock_writeable(sk) ||
 568             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 569              sock_writeable(sk)))
 570                 mask |= POLLOUT | POLLWRNORM;
 571
 572         if (tun->dev->reg_state != NETREG_REGISTERED)
 573                 mask = POLLERR;
 574
 575         tun_put(tun);
 576         return mask;
 577 }
 578
 579 /* prepad is the amount to reserve at front.  len is length after that.
 580  * linear is a hint as to how much to copy (usually headers). */
 581 static struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
 582                                      size_t prepad, size_t len,
 583                                      size_t linear, int noblock)
 584 {
 585         struct sock *sk = tun->socket.sk;
 586         struct sk_buff *skb;
 587         int err;
 588
 589         /* Under a page?  Don't bother with paged skb. */
 590         if (prepad + len < PAGE_SIZE || !linear)
 591                 linear = len;
 592
 593         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 594                                    &err);
 595         if (!skb)
 596                 return ERR_PTR(err);
 597
 598         skb_reserve(skb, prepad);
 599         skb_put(skb, linear);
 600         skb->data_len = len - linear;
 601         skb->len += len - linear;
 602
 603         return skb;
 604 }
 605
 606 /* set skb frags from iovec, this can move to core network code for reuse */
 607 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 608                                   int offset, size_t count)
 609 {
 610         int len = iov_length(from, count) - offset;
 611         int copy = skb_headlen(skb);
 612         int size, offset1 = 0;
 613         int i = 0;
 614
 615         /* Skip over from offset */
 616         while (count && (offset >= from->iov_len)) {
 617                 offset -= from->iov_len;
 618                 ++from;
 619                 --count;
 620         }
 621
 622         /* copy up to skb headlen */
 623         while (count && (copy > 0)) {
 624                 size = min_t(unsigned int, copy, from->iov_len - offset);
 625                 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
 626                                    size))
 627                         return -EFAULT;
 628                 if (copy > size) {
 629                         ++from;
 630                         --count;
 631                         offset = 0;
 632                 } else
 633                         offset += size;
 634                 copy -= size;
 635                 offset1 += size;
 636         }
 637
 638         if (len == offset1)
 639                 return 0;
 640
 641         while (count--) {
 642                 struct page *page[MAX_SKB_FRAGS];
 643                 int num_pages;
 644                 unsigned long base;
 645                 unsigned long truesize;
 646
 647                 len = from->iov_len - offset;
 648                 if (!len) {
 649                         offset = 0;
 650                         ++from;
 651                         continue;
 652                 }
 653                 base = (unsigned long)from->iov_base + offset;
 654                 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
 655                 if (i + size > MAX_SKB_FRAGS)
 656                         return -EMSGSIZE;
 657                 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
 658                 if (num_pages != size) {
 659                         for (i = 0; i < num_pages; i++)
 660                                 put_page(page[i]);
 661                         return -EFAULT;
 662                 }
 663                 truesize = size * PAGE_SIZE;
 664                 skb->data_len += len;
 665                 skb->len += len;
 666                 skb->truesize += truesize;
 667                 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
 668                 while (len) {
 669                         int off = base & ~PAGE_MASK;
 670                         int size = min_t(int, len, PAGE_SIZE - off);
 671                         __skb_fill_page_desc(skb, i, page[i], off, size);
 672                         skb_shinfo(skb)->nr_frags++;
 673                         /* increase sk_wmem_alloc */
 674                         base += size;
 675                         len -= size;
 676                         i++;
 677                 }
 678                 offset = 0;
 679                 ++from;
 680         }
 681         return 0;
 682 }
 683
 684 /* Get packet from user space buffer */
 685 static ssize_t tun_get_user(struct tun_struct *tun, void *msg_control,
 686                             const struct iovec *iv, size_t total_len,
 687                             size_t count, int noblock)
 688 {
 689         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
 690         struct sk_buff *skb;
 691         size_t len = total_len, align = NET_SKB_PAD;
 692         struct virtio_net_hdr gso = { 0 };
 693         int offset = 0;
 694         int copylen;
 695         bool zerocopy = false;
 696         int err;
 697
 698         if (!(tun->flags & TUN_NO_PI)) {
 699                 if ((len -= sizeof(pi)) > total_len)
 700                         return -EINVAL;
 701
 702                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
 703                         return -EFAULT;
 704                 offset += sizeof(pi);
 705         }
 706
 707         if (tun->flags & TUN_VNET_HDR) {
 708                 if ((len -= tun->vnet_hdr_sz) > total_len)
 709                         return -EINVAL;
 710
 711                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
 712                         return -EFAULT;
 713
 714                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
 715                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
 716                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
 717
 718                 if (gso.hdr_len > len)
 719                         return -EINVAL;
 720                 offset += tun->vnet_hdr_sz;
 721         }
 722
 723         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
 724                 align += NET_IP_ALIGN;
 725                 if (unlikely(len < ETH_HLEN ||
 726                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
 727                         return -EINVAL;
 728         }
 729
 730         if (msg_control)
 731                 zerocopy = true;
 732
 733         if (zerocopy) {
 734                 /* Userspace may produce vectors with count greater than
 735                  * MAX_SKB_FRAGS, so we need to linearize parts of the skb
 736                  * to let the rest of data to be fit in the frags.
 737                  */
 738                 if (count > MAX_SKB_FRAGS) {
 739                         copylen = iov_length(iv, count - MAX_SKB_FRAGS);
 740                         if (copylen < offset)
 741                                 copylen = 0;
 742                         else
 743                                 copylen -= offset;
 744                 } else
 745                                 copylen = 0;
 746                 /* There are 256 bytes to be copied in skb, so there is enough
 747                  * room for skb expand head in case it is used.
 748                  * The rest of the buffer is mapped from userspace.
 749                  */
 750                 if (copylen < gso.hdr_len)
 751                         copylen = gso.hdr_len;
 752                 if (!copylen)
 753                         copylen = GOODCOPY_LEN;
 754         } else
 755                 copylen = len;
 756
 757         skb = tun_alloc_skb(tun, align, copylen, gso.hdr_len, noblock);
 758         if (IS_ERR(skb)) {
 759                 if (PTR_ERR(skb) != -EAGAIN)
 760                         tun->dev->stats.rx_dropped++;
 761                 return PTR_ERR(skb);
 762         }
 763
 764         if (zerocopy)
 765                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
 766         else
 767                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
 768
 769         if (err) {
 770                 tun->dev->stats.rx_dropped++;
 771                 kfree_skb(skb);
 772                 return -EFAULT;
 773         }
 774
 775         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 776                 if (!skb_partial_csum_set(skb, gso.csum_start,
 777                                           gso.csum_offset)) {
 778                         tun->dev->stats.rx_frame_errors++;
 779                         kfree_skb(skb);
 780                         return -EINVAL;
 781                 }
 782         }
 783
 784         switch (tun->flags & TUN_TYPE_MASK) {
 785         case TUN_TUN_DEV:
 786                 if (tun->flags & TUN_NO_PI) {
 787                         switch (skb->data[0] & 0xf0) {
 788                         case 0x40:
 789                                 pi.proto = htons(ETH_P_IP);
 790                                 break;
 791                         case 0x60:
 792                                 pi.proto = htons(ETH_P_IPV6);
 793                                 break;
 794                         default:
 795                                 tun->dev->stats.rx_dropped++;
 796                                 kfree_skb(skb);
 797                                 return -EINVAL;
 798                         }
 799                 }
 800
 801                 skb_reset_mac_header(skb);
 802                 skb->protocol = pi.proto;
 803                 skb->dev = tun->dev;
 804                 break;
 805         case TUN_TAP_DEV:
 806                 skb->protocol = eth_type_trans(skb, tun->dev);
 807                 break;
 808         }
 809
 810         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 811                 pr_debug("GSO!\n");
 812                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 813                 case VIRTIO_NET_HDR_GSO_TCPV4:
 814                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 815                         break;
 816                 case VIRTIO_NET_HDR_GSO_TCPV6:
 817                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 818                         break;
 819                 case VIRTIO_NET_HDR_GSO_UDP:
 820                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 821                         break;
 822                 default:
 823                         tun->dev->stats.rx_frame_errors++;
 824                         kfree_skb(skb);
 825                         return -EINVAL;
 826                 }
 827
 828                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
 829                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 830
 831                 skb_shinfo(skb)->gso_size = gso.gso_size;
 832                 if (skb_shinfo(skb)->gso_size == 0) {
 833                         tun->dev->stats.rx_frame_errors++;
 834                         kfree_skb(skb);
 835                         return -EINVAL;
 836                 }
 837
 838                 /* Header must be checked, and gso_segs computed. */
 839                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 840                 skb_shinfo(skb)->gso_segs = 0;
 841         }
 842
 843         /* copy skb_ubuf_info for callback when skb has no error */
 844         if (zerocopy) {
 845                 skb_shinfo(skb)->destructor_arg = msg_control;
 846                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
 847         }
 848
 849         netif_rx_ni(skb);
 850
 851         tun->dev->stats.rx_packets++;
 852         tun->dev->stats.rx_bytes += len;
 853
 854         return total_len;
 855 }
 856
 857 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
 858                               unsigned long count, loff_t pos)
 859 {
 860         struct file *file = iocb->ki_filp;
 861         struct tun_struct *tun = tun_get(file);
 862         ssize_t result;
 863
 864         if (!tun)
 865                 return -EBADFD;
 866
 867         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
 868
 869         result = tun_get_user(tun, NULL, iv, iov_length(iv, count), count,
 870                               file->f_flags & O_NONBLOCK);
 871
 872         tun_put(tun);
 873         return result;
 874 }
 875
 876 /* Put packet to the user space buffer */
 877 static ssize_t tun_put_user(struct tun_struct *tun,
 878                             struct sk_buff *skb,
 879                             const struct iovec *iv, int len)
 880 {
 881         struct tun_pi pi = { 0, skb->protocol };
 882         ssize_t total = 0;
 883
 884         if (!(tun->flags & TUN_NO_PI)) {
 885                 if ((len -= sizeof(pi)) < 0)
 886                         return -EINVAL;
 887
 888                 if (len < skb->len) {
 889                         /* Packet will be striped */
 890                         pi.flags |= TUN_PKT_STRIP;
 891                 }
 892
 893                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
 894                         return -EFAULT;
 895                 total += sizeof(pi);
 896         }
 897
 898         if (tun->flags & TUN_VNET_HDR) {
 899                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
 900                 if ((len -= tun->vnet_hdr_sz) < 0)
 901                         return -EINVAL;
 902
 903                 if (skb_is_gso(skb)) {
 904                         struct skb_shared_info *sinfo = skb_shinfo(skb);
 905
 906                         /* This is a hint as to how much should be linear. */
 907                         gso.hdr_len = skb_headlen(skb);
 908                         gso.gso_size = sinfo->gso_size;
 909                         if (sinfo->gso_type & SKB_GSO_TCPV4)
 910                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 911                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
 912                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 913                         else if (sinfo->gso_type & SKB_GSO_UDP)
 914                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
 915                         else {
 916                                 pr_err("unexpected GSO type: "
 917                                        "0x%x, gso_size %d, hdr_len %d\n",
 918                                        sinfo->gso_type, gso.gso_size,
 919                                        gso.hdr_len);
 920                                 print_hex_dump(KERN_ERR, "tun: ",
 921                                                DUMP_PREFIX_NONE,
 922                                                16, 1, skb->head,
 923                                                min((int)gso.hdr_len, 64), true);
 924                                 WARN_ON_ONCE(1);
 925                                 return -EINVAL;
 926                         }
 927                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
 928                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 929                 } else
 930                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
 931
 932                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 933                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 934                         gso.csum_start = skb_checksum_start_offset(skb);
 935                         gso.csum_offset = skb->csum_offset;
 936                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
 937                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
 938                 } /* else everything is zero */
 939
 940                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
 941                                                sizeof(gso))))
 942                         return -EFAULT;
 943                 total += tun->vnet_hdr_sz;
 944         }
 945
 946         len = min_t(int, skb->len, len);
 947
 948         skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
 949         total += skb->len;
 950
 951         tun->dev->stats.tx_packets++;
 952         tun->dev->stats.tx_bytes += len;
 953
 954         return total;
 955 }
 956
 957 static ssize_t tun_do_read(struct tun_struct *tun,
 958                            struct kiocb *iocb, const struct iovec *iv,
 959                            ssize_t len, int noblock)
 960 {
 961         DECLARE_WAITQUEUE(wait, current);
 962         struct sk_buff *skb;
 963         ssize_t ret = 0;
 964
 965         tun_debug(KERN_INFO, tun, "tun_chr_read\n");
 966
 967         if (unlikely(!noblock))
 968                 add_wait_queue(&tun->wq.wait, &wait);
 969         while (len) {
 970                 current->state = TASK_INTERRUPTIBLE;
 971
 972                 /* Read frames from the queue */
 973                 if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
 974                         if (noblock) {
 975                                 ret = -EAGAIN;
 976                                 break;
 977                         }
 978                         if (signal_pending(current)) {
 979                                 ret = -ERESTARTSYS;
 980                                 break;
 981                         }
 982                         if (tun->dev->reg_state != NETREG_REGISTERED) {
 983                                 ret = -EIO;
 984                                 break;
 985                         }
 986
 987                         /* Nothing to read, let's sleep */
 988                         schedule();
 989                         continue;
 990                 }
 991                 netif_wake_queue(tun->dev);
 992
 993                 ret = tun_put_user(tun, skb, iv, len);
 994                 kfree_skb(skb);
 995                 break;
 996         }
 997
 998         current->state = TASK_RUNNING;
 999         if (unlikely(!noblock))
1000                 remove_wait_queue(&tun->wq.wait, &wait);
1001
1002         return ret;
1003 }
1004
1005 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1006                             unsigned long count, loff_t pos)
1007 {
1008         struct file *file = iocb->ki_filp;
1009         struct tun_file *tfile = file->private_data;
1010         struct tun_struct *tun = __tun_get(tfile);
1011         ssize_t len, ret;
1012
1013         if (!tun)
1014                 return -EBADFD;
1015         len = iov_length(iv, count);
1016         if (len < 0) {
1017                 ret = -EINVAL;
1018                 goto out;
1019         }
1020
1021         ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK);
1022         ret = min_t(ssize_t, ret, len);
1023 out:
1024         tun_put(tun);
1025         return ret;
1026 }
1027
1028 static void tun_setup(struct net_device *dev)
1029 {
1030         struct tun_struct *tun = netdev_priv(dev);
1031
1032         tun->owner = INVALID_UID;
1033         tun->group = INVALID_GID;
1034
1035         dev->ethtool_ops = &tun_ethtool_ops;
1036         dev->destructor = tun_free_netdev;
1037 }
1038
1039 /* Trivial set of netlink ops to allow deleting tun or tap
1040  * device with netlink.
1041  */
1042 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1043 {
1044         return -EINVAL;
1045 }
1046
1047 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1048         .kind           = DRV_NAME,
1049         .priv_size      = sizeof(struct tun_struct),
1050         .setup          = tun_setup,
1051         .validate       = tun_validate,
1052 };
1053
1054 static void tun_sock_write_space(struct sock *sk)
1055 {
1056         struct tun_struct *tun;
1057         wait_queue_head_t *wqueue;
1058
1059         if (!sock_writeable(sk))
1060                 return;
1061
1062         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1063                 return;
1064
1065         wqueue = sk_sleep(sk);
1066         if (wqueue && waitqueue_active(wqueue))
1067                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1068                                                 POLLWRNORM | POLLWRBAND);
1069
1070         tun = tun_sk(sk)->tun;
1071         kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
1072 }
1073
1074 static void tun_sock_destruct(struct sock *sk)
1075 {
1076         free_netdev(tun_sk(sk)->tun->dev);
1077 }
1078
1079 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1080                        struct msghdr *m, size_t total_len)
1081 {
1082         struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
1083         return tun_get_user(tun, m->msg_control, m->msg_iov, total_len,
1084                             m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1085 }
1086
1087 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1088                        struct msghdr *m, size_t total_len,
1089                        int flags)
1090 {
1091         struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
1092         int ret;
1093         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1094                 return -EINVAL;
1095         ret = tun_do_read(tun, iocb, m->msg_iov, total_len,
1096                           flags & MSG_DONTWAIT);
1097         if (ret > total_len) {
1098                 m->msg_flags |= MSG_TRUNC;
1099                 ret = flags & MSG_TRUNC ? ret : total_len;
1100         }
1101         return ret;
1102 }
1103
1104 static int tun_release(struct socket *sock)
1105 {
1106         if (sock->sk)
1107                 sock_put(sock->sk);
1108         return 0;
1109 }
1110
1111 /* Ops structure to mimic raw sockets with tun */
1112 static const struct proto_ops tun_socket_ops = {
1113         .sendmsg = tun_sendmsg,
1114         .recvmsg = tun_recvmsg,
1115         .release = tun_release,
1116 };
1117
1118 static struct proto tun_proto = {
1119         .name           = "tun",
1120         .owner          = THIS_MODULE,
1121         .obj_size       = sizeof(struct tun_sock),
1122 };
1123
1124 static int tun_flags(struct tun_struct *tun)
1125 {
1126         int flags = 0;
1127
1128         if (tun->flags & TUN_TUN_DEV)
1129                 flags |= IFF_TUN;
1130         else
1131                 flags |= IFF_TAP;
1132
1133         if (tun->flags & TUN_NO_PI)
1134                 flags |= IFF_NO_PI;
1135
1136         if (tun->flags & TUN_ONE_QUEUE)
1137                 flags |= IFF_ONE_QUEUE;
1138
1139         if (tun->flags & TUN_VNET_HDR)
1140                 flags |= IFF_VNET_HDR;
1141
1142         return flags;
1143 }
1144
1145 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1146                               char *buf)
1147 {
1148         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1149         return sprintf(buf, "0x%x\n", tun_flags(tun));
1150 }
1151
1152 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1153                               char *buf)
1154 {
1155         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1156         return uid_valid(tun->owner)?
1157                 sprintf(buf, "%u\n",
1158                         from_kuid_munged(current_user_ns(), tun->owner)):
1159                 sprintf(buf, "-1\n");
1160 }
1161
1162 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1163                               char *buf)
1164 {
1165         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1166         return gid_valid(tun->group) ?
1167                 sprintf(buf, "%u\n",
1168                         from_kgid_munged(current_user_ns(), tun->group)):
1169                 sprintf(buf, "-1\n");
1170 }
1171
1172 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1173 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1174 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1175
1176 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1177 {
1178         struct sock *sk;
1179         struct tun_struct *tun;
1180         struct net_device *dev;
1181         int err;
1182
1183         dev = __dev_get_by_name(net, ifr->ifr_name);
1184         if (dev) {
1185                 const struct cred *cred = current_cred();
1186
1187                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1188                         return -EBUSY;
1189                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1190                         tun = netdev_priv(dev);
1191                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1192                         tun = netdev_priv(dev);
1193                 else
1194                         return -EINVAL;
1195
1196                 if (((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
1197                      (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
1198                     !capable(CAP_NET_ADMIN))
1199                         return -EPERM;
1200                 err = security_tun_dev_attach(tun->socket.sk);
1201                 if (err < 0)
1202                         return err;
1203
1204                 err = tun_attach(tun, file);
1205                 if (err < 0)
1206                         return err;
1207         }
1208         else {
1209                 char *name;
1210                 unsigned long flags = 0;
1211
1212                 if (!capable(CAP_NET_ADMIN))
1213                         return -EPERM;
1214                 err = security_tun_dev_create();
1215                 if (err < 0)
1216                         return err;
1217
1218                 /* Set dev type */
1219                 if (ifr->ifr_flags & IFF_TUN) {
1220                         /* TUN device */
1221                         flags |= TUN_TUN_DEV;
1222                         name = "tun%d";
1223                 } else if (ifr->ifr_flags & IFF_TAP) {
1224                         /* TAP device */
1225                         flags |= TUN_TAP_DEV;
1226                         name = "tap%d";
1227                 } else
1228                         return -EINVAL;
1229
1230                 if (*ifr->ifr_name)
1231                         name = ifr->ifr_name;
1232
1233                 dev = alloc_netdev(sizeof(struct tun_struct), name,
1234                                    tun_setup);
1235                 if (!dev)
1236                         return -ENOMEM;
1237
1238                 dev_net_set(dev, net);
1239                 dev->rtnl_link_ops = &tun_link_ops;
1240
1241                 tun = netdev_priv(dev);
1242                 tun->dev = dev;
1243                 tun->flags = flags;
1244                 tun->txflt.count = 0;
1245                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1246                 set_bit(SOCK_EXTERNALLY_ALLOCATED, &tun->socket.flags);
1247
1248                 err = -ENOMEM;
1249                 sk = sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1250                 if (!sk)
1251                         goto err_free_dev;
1252
1253                 sk_change_net(sk, net);
1254                 tun->socket.wq = &tun->wq;
1255                 init_waitqueue_head(&tun->wq.wait);
1256                 tun->socket.ops = &tun_socket_ops;
1257                 sock_init_data(&tun->socket, sk);
1258                 sk->sk_write_space = tun_sock_write_space;
1259                 sk->sk_sndbuf = INT_MAX;
1260                 sock_set_flag(sk, SOCK_ZEROCOPY);
1261
1262                 tun_sk(sk)->tun = tun;
1263
1264                 security_tun_dev_post_create(sk);
1265
1266                 tun_net_init(dev);
1267
1268                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1269                         TUN_USER_FEATURES;
1270                 dev->features = dev->hw_features;
1271
1272                 err = register_netdevice(tun->dev);
1273                 if (err < 0)
1274                         goto err_free_sk;
1275
1276                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1277                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1278                     device_create_file(&tun->dev->dev, &dev_attr_group))
1279                         pr_err("Failed to create tun sysfs files\n");
1280
1281                 sk->sk_destruct = tun_sock_destruct;
1282
1283                 err = tun_attach(tun, file);
1284                 if (err < 0)
1285                         goto failed;
1286         }
1287
1288         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1289
1290         if (ifr->ifr_flags & IFF_NO_PI)
1291                 tun->flags |= TUN_NO_PI;
1292         else
1293                 tun->flags &= ~TUN_NO_PI;
1294
1295         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1296                 tun->flags |= TUN_ONE_QUEUE;
1297         else
1298                 tun->flags &= ~TUN_ONE_QUEUE;
1299
1300         if (ifr->ifr_flags & IFF_VNET_HDR)
1301                 tun->flags |= TUN_VNET_HDR;
1302         else
1303                 tun->flags &= ~TUN_VNET_HDR;
1304
1305         /* Make sure persistent devices do not get stuck in
1306          * xoff state.
1307          */
1308         if (netif_running(tun->dev))
1309                 netif_wake_queue(tun->dev);
1310
1311         strcpy(ifr->ifr_name, tun->dev->name);
1312         return 0;
1313
1314  err_free_sk:
1315         tun_free_netdev(dev);
1316  err_free_dev:
1317         free_netdev(dev);
1318  failed:
1319         return err;
1320 }
1321
1322 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1323                        struct ifreq *ifr)
1324 {
1325         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1326
1327         strcpy(ifr->ifr_name, tun->dev->name);
1328
1329         ifr->ifr_flags = tun_flags(tun);
1330
1331         return 0;
1332 }
1333
1334 /* This is like a cut-down ethtool ops, except done via tun fd so no
1335  * privs required. */
1336 static int set_offload(struct tun_struct *tun, unsigned long arg)
1337 {
1338         netdev_features_t features = 0;
1339
1340         if (arg & TUN_F_CSUM) {
1341                 features |= NETIF_F_HW_CSUM;
1342                 arg &= ~TUN_F_CSUM;
1343
1344                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1345                         if (arg & TUN_F_TSO_ECN) {
1346                                 features |= NETIF_F_TSO_ECN;
1347                                 arg &= ~TUN_F_TSO_ECN;
1348                         }
1349                         if (arg & TUN_F_TSO4)
1350                                 features |= NETIF_F_TSO;
1351                         if (arg & TUN_F_TSO6)
1352                                 features |= NETIF_F_TSO6;
1353                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1354                 }
1355
1356                 if (arg & TUN_F_UFO) {
1357                         features |= NETIF_F_UFO;
1358                         arg &= ~TUN_F_UFO;
1359                 }
1360         }
1361
1362         /* This gives the user a way to test for new features in future by
1363          * trying to set them. */
1364         if (arg)
1365                 return -EINVAL;
1366
1367         tun->set_features = features;
1368         netdev_update_features(tun->dev);
1369
1370         return 0;
1371 }
1372
1373 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1374                             unsigned long arg, int ifreq_len)
1375 {
1376         struct tun_file *tfile = file->private_data;
1377         struct tun_struct *tun;
1378         void __user* argp = (void __user*)arg;
1379         struct sock_fprog fprog;
1380         struct ifreq ifr;
1381         kuid_t owner;
1382         kgid_t group;
1383         int sndbuf;
1384         int vnet_hdr_sz;
1385         int ret;
1386
1387         if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89) {
1388                 if (copy_from_user(&ifr, argp, ifreq_len))
1389                         return -EFAULT;
1390         } else {
1391                 memset(&ifr, 0, sizeof(ifr));
1392         }
1393         if (cmd == TUNGETFEATURES) {
1394                 /* Currently this just means: "what IFF flags are valid?".
1395                  * This is needed because we never checked for invalid flags on
1396                  * TUNSETIFF. */
1397                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1398                                 IFF_VNET_HDR,
1399                                 (unsigned int __user*)argp);
1400         }
1401
1402         rtnl_lock();
1403
1404         tun = __tun_get(tfile);
1405         if (cmd == TUNSETIFF && !tun) {
1406                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1407
1408                 ret = tun_set_iff(tfile->net, file, &ifr);
1409
1410                 if (ret)
1411                         goto unlock;
1412
1413                 if (copy_to_user(argp, &ifr, ifreq_len))
1414                         ret = -EFAULT;
1415                 goto unlock;
1416         }
1417
1418         ret = -EBADFD;
1419         if (!tun)
1420                 goto unlock;
1421
1422         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %d\n", cmd);
1423
1424         ret = 0;
1425         switch (cmd) {
1426         case TUNGETIFF:
1427                 ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1428                 if (ret)
1429                         break;
1430
1431                 if (copy_to_user(argp, &ifr, ifreq_len))
1432                         ret = -EFAULT;
1433                 break;
1434
1435         case TUNSETNOCSUM:
1436                 /* Disable/Enable checksum */
1437
1438                 /* [unimplemented] */
1439                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1440                           arg ? "disabled" : "enabled");
1441                 break;
1442
1443         case TUNSETPERSIST:
1444                 /* Disable/Enable persist mode */
1445                 if (arg)
1446                         tun->flags |= TUN_PERSIST;
1447                 else
1448                         tun->flags &= ~TUN_PERSIST;
1449
1450                 tun_debug(KERN_INFO, tun, "persist %s\n",
1451                           arg ? "enabled" : "disabled");
1452                 break;
1453
1454         case TUNSETOWNER:
1455                 /* Set owner of the device */
1456                 owner = make_kuid(current_user_ns(), arg);
1457                 if (!uid_valid(owner)) {
1458                         ret = -EINVAL;
1459                         break;
1460                 }
1461                 tun->owner = owner;
1462                 tun_debug(KERN_INFO, tun, "owner set to %d\n",
1463                           from_kuid(&init_user_ns, tun->owner));
1464                 break;
1465
1466         case TUNSETGROUP:
1467                 /* Set group of the device */
1468                 group = make_kgid(current_user_ns(), arg);
1469                 if (!gid_valid(group)) {
1470                         ret = -EINVAL;
1471                         break;
1472                 }
1473                 tun->group = group;
1474                 tun_debug(KERN_INFO, tun, "group set to %d\n",
1475                           from_kgid(&init_user_ns, tun->group));
1476                 break;
1477
1478         case TUNSETLINK:
1479                 /* Only allow setting the type when the interface is down */
1480                 if (tun->dev->flags & IFF_UP) {
1481                         tun_debug(KERN_INFO, tun,
1482                                   "Linktype set failed because interface is up\n");
1483                         ret = -EBUSY;
1484                 } else {
1485                         tun->dev->type = (int) arg;
1486                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1487                                   tun->dev->type);
1488                         ret = 0;
1489                 }
1490                 break;
1491
1492 #ifdef TUN_DEBUG
1493         case TUNSETDEBUG:
1494                 tun->debug = arg;
1495                 break;
1496 #endif
1497         case TUNSETOFFLOAD:
1498                 ret = set_offload(tun, arg);
1499                 break;
1500
1501         case TUNSETTXFILTER:
1502                 /* Can be set only for TAPs */
1503                 ret = -EINVAL;
1504                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1505                         break;
1506                 ret = update_filter(&tun->txflt, (void __user *)arg);
1507                 break;
1508
1509         case SIOCGIFHWADDR:
1510                 /* Get hw address */
1511                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1512                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1513                 if (copy_to_user(argp, &ifr, ifreq_len))
1514                         ret = -EFAULT;
1515                 break;
1516
1517         case SIOCSIFHWADDR:
1518                 /* Set hw address */
1519                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1520                           ifr.ifr_hwaddr.sa_data);
1521
1522                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1523                 break;
1524
1525         case TUNGETSNDBUF:
1526                 sndbuf = tun->socket.sk->sk_sndbuf;
1527                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1528                         ret = -EFAULT;
1529                 break;
1530
1531         case TUNSETSNDBUF:
1532                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1533                         ret = -EFAULT;
1534                         break;
1535                 }
1536
1537                 tun->socket.sk->sk_sndbuf = sndbuf;
1538                 break;
1539
1540         case TUNGETVNETHDRSZ:
1541                 vnet_hdr_sz = tun->vnet_hdr_sz;
1542                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1543                         ret = -EFAULT;
1544                 break;
1545
1546         case TUNSETVNETHDRSZ:
1547                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1548                         ret = -EFAULT;
1549                         break;
1550                 }
1551                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1552                         ret = -EINVAL;
1553                         break;
1554                 }
1555
1556                 tun->vnet_hdr_sz = vnet_hdr_sz;
1557                 break;
1558
1559         case TUNATTACHFILTER:
1560                 /* Can be set only for TAPs */
1561                 ret = -EINVAL;
1562                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1563                         break;
1564                 ret = -EFAULT;
1565                 if (copy_from_user(&fprog, argp, sizeof(fprog)))
1566                         break;
1567
1568                 ret = sk_attach_filter(&fprog, tun->socket.sk);
1569                 break;
1570
1571         case TUNDETACHFILTER:
1572                 /* Can be set only for TAPs */
1573                 ret = -EINVAL;
1574                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1575                         break;
1576                 ret = sk_detach_filter(tun->socket.sk);
1577                 break;
1578
1579         default:
1580                 ret = -EINVAL;
1581                 break;
1582         }
1583
1584 unlock:
1585         rtnl_unlock();
1586         if (tun)
1587                 tun_put(tun);
1588         return ret;
1589 }
1590
1591 static long tun_chr_ioctl(struct file *file,
1592                           unsigned int cmd, unsigned long arg)
1593 {
1594         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1595 }
1596
1597 #ifdef CONFIG_COMPAT
1598 static long tun_chr_compat_ioctl(struct file *file,
1599                          unsigned int cmd, unsigned long arg)
1600 {
1601         switch (cmd) {
1602         case TUNSETIFF:
1603         case TUNGETIFF:
1604         case TUNSETTXFILTER:
1605         case TUNGETSNDBUF:
1606         case TUNSETSNDBUF:
1607         case SIOCGIFHWADDR:
1608         case SIOCSIFHWADDR:
1609                 arg = (unsigned long)compat_ptr(arg);
1610                 break;
1611         default:
1612                 arg = (compat_ulong_t)arg;
1613                 break;
1614         }
1615
1616         /*
1617          * compat_ifreq is shorter than ifreq, so we must not access beyond
1618          * the end of that structure. All fields that are used in this
1619          * driver are compatible though, we don't need to convert the
1620          * contents.
1621          */
1622         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1623 }
1624 #endif /* CONFIG_COMPAT */
1625
1626 static int tun_chr_fasync(int fd, struct file *file, int on)
1627 {
1628         struct tun_struct *tun = tun_get(file);
1629         int ret;
1630
1631         if (!tun)
1632                 return -EBADFD;
1633
1634         tun_debug(KERN_INFO, tun, "tun_chr_fasync %d\n", on);
1635
1636         if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1637                 goto out;
1638
1639         if (on) {
1640                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1641                 if (ret)
1642                         goto out;
1643                 tun->flags |= TUN_FASYNC;
1644         } else
1645                 tun->flags &= ~TUN_FASYNC;
1646         ret = 0;
1647 out:
1648         tun_put(tun);
1649         return ret;
1650 }
1651
1652 static int tun_chr_open(struct inode *inode, struct file * file)
1653 {
1654         struct tun_file *tfile;
1655
1656         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
1657
1658         tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1659         if (!tfile)
1660                 return -ENOMEM;
1661         atomic_set(&tfile->count, 0);
1662         tfile->tun = NULL;
1663         tfile->net = get_net(current->nsproxy->net_ns);
1664         file->private_data = tfile;
1665         return 0;
1666 }
1667
1668 static int tun_chr_close(struct inode *inode, struct file *file)
1669 {
1670         struct tun_file *tfile = file->private_data;
1671         struct tun_struct *tun;
1672
1673         tun = __tun_get(tfile);
1674         if (tun) {
1675                 struct net_device *dev = tun->dev;
1676
1677                 tun_debug(KERN_INFO, tun, "tun_chr_close\n");
1678
1679                 __tun_detach(tun);
1680
1681                 /* If desirable, unregister the netdevice. */
1682                 if (!(tun->flags & TUN_PERSIST)) {
1683                         rtnl_lock();
1684                         if (dev->reg_state == NETREG_REGISTERED)
1685                                 unregister_netdevice(dev);
1686                         rtnl_unlock();
1687                 }
1688         }
1689
1690         tun = tfile->tun;
1691         if (tun)
1692                 sock_put(tun->socket.sk);
1693
1694         put_net(tfile->net);
1695         kfree(tfile);
1696
1697         return 0;
1698 }
1699
1700 static const struct file_operations tun_fops = {
1701         .owner  = THIS_MODULE,
1702         .llseek = no_llseek,
1703         .read  = do_sync_read,
1704         .aio_read  = tun_chr_aio_read,
1705         .write = do_sync_write,
1706         .aio_write = tun_chr_aio_write,
1707         .poll   = tun_chr_poll,
1708         .unlocked_ioctl = tun_chr_ioctl,
1709 #ifdef CONFIG_COMPAT
1710         .compat_ioctl = tun_chr_compat_ioctl,
1711 #endif
1712         .open   = tun_chr_open,
1713         .release = tun_chr_close,
1714         .fasync = tun_chr_fasync
1715 };
1716
1717 static struct miscdevice tun_miscdev = {
1718         .minor = TUN_MINOR,
1719         .name = "tun",
1720         .nodename = "net/tun",
1721         .fops = &tun_fops,
1722 };
1723
1724 /* ethtool interface */
1725
1726 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1727 {
1728         cmd->supported          = 0;
1729         cmd->advertising        = 0;
1730         ethtool_cmd_speed_set(cmd, SPEED_10);
1731         cmd->duplex             = DUPLEX_FULL;
1732         cmd->port               = PORT_TP;
1733         cmd->phy_address        = 0;
1734         cmd->transceiver        = XCVR_INTERNAL;
1735         cmd->autoneg            = AUTONEG_DISABLE;
1736         cmd->maxtxpkt           = 0;
1737         cmd->maxrxpkt           = 0;
1738         return 0;
1739 }
1740
1741 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1742 {
1743         struct tun_struct *tun = netdev_priv(dev);
1744
1745         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1746         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1747
1748         switch (tun->flags & TUN_TYPE_MASK) {
1749         case TUN_TUN_DEV:
1750                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
1751                 break;
1752         case TUN_TAP_DEV:
1753                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
1754                 break;
1755         }
1756 }
1757
1758 static u32 tun_get_msglevel(struct net_device *dev)
1759 {
1760 #ifdef TUN_DEBUG
1761         struct tun_struct *tun = netdev_priv(dev);
1762         return tun->debug;
1763 #else
1764         return -EOPNOTSUPP;
1765 #endif
1766 }
1767
1768 static void tun_set_msglevel(struct net_device *dev, u32 value)
1769 {
1770 #ifdef TUN_DEBUG
1771         struct tun_struct *tun = netdev_priv(dev);
1772         tun->debug = value;
1773 #endif
1774 }
1775
1776 static const struct ethtool_ops tun_ethtool_ops = {
1777         .get_settings   = tun_get_settings,
1778         .get_drvinfo    = tun_get_drvinfo,
1779         .get_msglevel   = tun_get_msglevel,
1780         .set_msglevel   = tun_set_msglevel,
1781         .get_link       = ethtool_op_get_link,
1782 };
1783
1784
1785 static int __init tun_init(void)
1786 {
1787         int ret = 0;
1788
1789         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1790         pr_info("%s\n", DRV_COPYRIGHT);
1791
1792         ret = rtnl_link_register(&tun_link_ops);
1793         if (ret) {
1794                 pr_err("Can't register link_ops\n");
1795                 goto err_linkops;
1796         }
1797
1798         ret = misc_register(&tun_miscdev);
1799         if (ret) {
1800                 pr_err("Can't register misc device %d\n", TUN_MINOR);
1801                 goto err_misc;
1802         }
1803         return  0;
1804 err_misc:
1805         rtnl_link_unregister(&tun_link_ops);
1806 err_linkops:
1807         return ret;
1808 }
1809
1810 static void tun_cleanup(void)
1811 {
1812         misc_deregister(&tun_miscdev);
1813         rtnl_link_unregister(&tun_link_ops);
1814 }
1815
1816 /* Get an underlying socket object from tun file.  Returns error unless file is
1817  * attached to a device.  The returned object works like a packet socket, it
1818  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1819  * holding a reference to the file for as long as the socket is in use. */
1820 struct socket *tun_get_socket(struct file *file)
1821 {
1822         struct tun_struct *tun;
1823         if (file->f_op != &tun_fops)
1824                 return ERR_PTR(-EINVAL);
1825         tun = tun_get(file);
1826         if (!tun)
1827                 return ERR_PTR(-EBADFD);
1828         tun_put(tun);
1829         return &tun->socket;
1830 }
1831 EXPORT_SYMBOL_GPL(tun_get_socket);
1832
1833 module_init(tun_init);
1834 module_exit(tun_cleanup);
1835 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1836 MODULE_AUTHOR(DRV_COPYRIGHT);
1837 MODULE_LICENSE("GPL");
1838 MODULE_ALIAS_MISCDEV(TUN_MINOR);
1839 MODULE_ALIAS("devname:net/tun");