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/if_vlan.h>
  64 #include <linux/crc32.h>
  65 #include <linux/nsproxy.h>
  66 #include <linux/virtio_net.h>
  67 #include <linux/rcupdate.h>
  68 #include <net/net_namespace.h>
  69 #include <net/netns/generic.h>
  70 #include <net/rtnetlink.h>
  71 #include <net/sock.h>
  72
  73 #include <asm/uaccess.h>
  74
  75 /* Uncomment to enable debugging */
  76 /* #define TUN_DEBUG 1 */
  77
  78 #ifdef TUN_DEBUG
  79 static int debug;
  80
  81 #define tun_debug(level, tun, fmt, args...)                     \
  82 do {                                                            \
  83         if (tun->debug)                                         \
  84                 netdev_printk(level, tun->dev, fmt, ##args);    \
  85 } while (0)
  86 #define DBG1(level, fmt, args...)                               \
  87 do {                                                            \
  88         if (debug == 2)                                         \
  89                 printk(level fmt, ##args);                      \
  90 } while (0)
  91 #else
  92 #define tun_debug(level, tun, fmt, args...)                     \
  93 do {                                                            \
  94         if (0)                                                  \
  95                 netdev_printk(level, tun->dev, fmt, ##args);    \
  96 } while (0)
  97 #define DBG1(level, fmt, args...)                               \
  98 do {                                                            \
  99         if (0)                                                  \
 100                 printk(level fmt, ##args);                      \
 101 } while (0)
 102 #endif
 103
 104 #define GOODCOPY_LEN 128
 105
 106 #define FLT_EXACT_COUNT 8
 107 struct tap_filter {
 108         unsigned int    count;    /* Number of addrs. Zero means disabled */
 109         u32             mask[2];  /* Mask of the hashed addrs */
 110         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 111 };
 112
 113 /* DEFAULT_MAX_NUM_RSS_QUEUES were choosed to let the rx/tx queues allocated for
 114  * the netdevice to be fit in one page. So we can make sure the success of
 115  * memory allocation. TODO: increase the limit. */
 116 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
 117 #define MAX_TAP_FLOWS  4096
 118
 119 #define TUN_FLOW_EXPIRE (3 * HZ)
 120
 121 /* A tun_file connects an open character device to a tuntap netdevice. It
 122  * also contains all socket related strctures (except sock_fprog and tap_filter)
 123  * to serve as one transmit queue for tuntap device. The sock_fprog and
 124  * tap_filter were kept in tun_struct since they were used for filtering for the
 125  * netdevice not for a specific queue (at least I didn't see the requirement for
 126  * this).
 127  *
 128  * RCU usage:
 129  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 130  * other can only be read while rcu_read_lock or rtnl_lock is held.
 131  */
 132 struct tun_file {
 133         struct sock sk;
 134         struct socket socket;
 135         struct socket_wq wq;
 136         struct tun_struct __rcu *tun;
 137         struct net *net;
 138         struct fasync_struct *fasync;
 139         /* only used for fasnyc */
 140         unsigned int flags;
 141         u16 queue_index;
 142         struct list_head next;
 143         struct tun_struct *detached;
 144 };
 145
 146 struct tun_flow_entry {
 147         struct hlist_node hash_link;
 148         struct rcu_head rcu;
 149         struct tun_struct *tun;
 150
 151         u32 rxhash;
 152         int queue_index;
 153         unsigned long updated;
 154 };
 155
 156 #define TUN_NUM_FLOW_ENTRIES 1024
 157
 158 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 159  * device, socket filter, sndbuf and vnet header size were restore when the
 160  * file were attached to a persist device.
 161  */
 162 struct tun_struct {
 163         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 164         unsigned int            numqueues;
 165         unsigned int            flags;
 166         kuid_t                  owner;
 167         kgid_t                  group;
 168
 169         struct net_device       *dev;
 170         netdev_features_t       set_features;
 171 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 172                           NETIF_F_TSO6|NETIF_F_UFO)
 173
 174         int                     vnet_hdr_sz;
 175         int                     sndbuf;
 176         struct tap_filter       txflt;
 177         struct sock_fprog       fprog;
 178         /* protected by rtnl lock */
 179         bool                    filter_attached;
 180 #ifdef TUN_DEBUG
 181         int debug;
 182 #endif
 183         spinlock_t lock;
 184         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 185         struct timer_list flow_gc_timer;
 186         unsigned long ageing_time;
 187         unsigned int numdisabled;
 188         struct list_head disabled;
 189         void *security;
 190         u32 flow_count;
 191 };
 192
 193 static inline u32 tun_hashfn(u32 rxhash)
 194 {
 195         return rxhash & 0x3ff;
 196 }
 197
 198 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 199 {
 200         struct tun_flow_entry *e;
 201
 202         hlist_for_each_entry_rcu(e, head, hash_link) {
 203                 if (e->rxhash == rxhash)
 204                         return e;
 205         }
 206         return NULL;
 207 }
 208
 209 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 210                                               struct hlist_head *head,
 211                                               u32 rxhash, u16 queue_index)
 212 {
 213         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 214
 215         if (e) {
 216                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 217                           rxhash, queue_index);
 218                 e->updated = jiffies;
 219                 e->rxhash = rxhash;
 220                 e->queue_index = queue_index;
 221                 e->tun = tun;
 222                 hlist_add_head_rcu(&e->hash_link, head);
 223                 ++tun->flow_count;
 224         }
 225         return e;
 226 }
 227
 228 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 229 {
 230         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 231                   e->rxhash, e->queue_index);
 232         hlist_del_rcu(&e->hash_link);
 233         kfree_rcu(e, rcu);
 234         --tun->flow_count;
 235 }
 236
 237 static void tun_flow_flush(struct tun_struct *tun)
 238 {
 239         int i;
 240
 241         spin_lock_bh(&tun->lock);
 242         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 243                 struct tun_flow_entry *e;
 244                 struct hlist_node *n;
 245
 246                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 247                         tun_flow_delete(tun, e);
 248         }
 249         spin_unlock_bh(&tun->lock);
 250 }
 251
 252 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 253 {
 254         int i;
 255
 256         spin_lock_bh(&tun->lock);
 257         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 258                 struct tun_flow_entry *e;
 259                 struct hlist_node *n;
 260
 261                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 262                         if (e->queue_index == queue_index)
 263                                 tun_flow_delete(tun, e);
 264                 }
 265         }
 266         spin_unlock_bh(&tun->lock);
 267 }
 268
 269 static void tun_flow_cleanup(unsigned long data)
 270 {
 271         struct tun_struct *tun = (struct tun_struct *)data;
 272         unsigned long delay = tun->ageing_time;
 273         unsigned long next_timer = jiffies + delay;
 274         unsigned long count = 0;
 275         int i;
 276
 277         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 278
 279         spin_lock_bh(&tun->lock);
 280         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 281                 struct tun_flow_entry *e;
 282                 struct hlist_node *n;
 283
 284                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 285                         unsigned long this_timer;
 286                         count++;
 287                         this_timer = e->updated + delay;
 288                         if (time_before_eq(this_timer, jiffies))
 289                                 tun_flow_delete(tun, e);
 290                         else if (time_before(this_timer, next_timer))
 291                                 next_timer = this_timer;
 292                 }
 293         }
 294
 295         if (count)
 296                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 297         spin_unlock_bh(&tun->lock);
 298 }
 299
 300 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 301                             struct tun_file *tfile)
 302 {
 303         struct hlist_head *head;
 304         struct tun_flow_entry *e;
 305         unsigned long delay = tun->ageing_time;
 306         u16 queue_index = tfile->queue_index;
 307
 308         if (!rxhash)
 309                 return;
 310         else
 311                 head = &tun->flows[tun_hashfn(rxhash)];
 312
 313         rcu_read_lock();
 314
 315         /* We may get a very small possibility of OOO during switching, not
 316          * worth to optimize.*/
 317         if (tun->numqueues == 1 || tfile->detached)
 318                 goto unlock;
 319
 320         e = tun_flow_find(head, rxhash);
 321         if (likely(e)) {
 322                 /* TODO: keep queueing to old queue until it's empty? */
 323                 e->queue_index = queue_index;
 324                 e->updated = jiffies;
 325         } else {
 326                 spin_lock_bh(&tun->lock);
 327                 if (!tun_flow_find(head, rxhash) &&
 328                     tun->flow_count < MAX_TAP_FLOWS)
 329                         tun_flow_create(tun, head, rxhash, queue_index);
 330
 331                 if (!timer_pending(&tun->flow_gc_timer))
 332                         mod_timer(&tun->flow_gc_timer,
 333                                   round_jiffies_up(jiffies + delay));
 334                 spin_unlock_bh(&tun->lock);
 335         }
 336
 337 unlock:
 338         rcu_read_unlock();
 339 }
 340
 341 /* We try to identify a flow through its rxhash first. The reason that
 342  * we do not check rxq no. is becuase some cards(e.g 82599), chooses
 343  * the rxq based on the txq where the last packet of the flow comes. As
 344  * the userspace application move between processors, we may get a
 345  * different rxq no. here. If we could not get rxhash, then we would
 346  * hope the rxq no. may help here.
 347  */
 348 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
 349 {
 350         struct tun_struct *tun = netdev_priv(dev);
 351         struct tun_flow_entry *e;
 352         u32 txq = 0;
 353         u32 numqueues = 0;
 354
 355         rcu_read_lock();
 356         numqueues = ACCESS_ONCE(tun->numqueues);
 357
 358         txq = skb_get_rxhash(skb);
 359         if (txq) {
 360                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 361                 if (e)
 362                         txq = e->queue_index;
 363                 else
 364                         /* use multiply and shift instead of expensive divide */
 365                         txq = ((u64)txq * numqueues) >> 32;
 366         } else if (likely(skb_rx_queue_recorded(skb))) {
 367                 txq = skb_get_rx_queue(skb);
 368                 while (unlikely(txq >= numqueues))
 369                         txq -= numqueues;
 370         }
 371
 372         rcu_read_unlock();
 373         return txq;
 374 }
 375
 376 static inline bool tun_not_capable(struct tun_struct *tun)
 377 {
 378         const struct cred *cred = current_cred();
 379         struct net *net = dev_net(tun->dev);
 380
 381         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 382                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 383                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 384 }
 385
 386 static void tun_set_real_num_queues(struct tun_struct *tun)
 387 {
 388         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 389         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 390 }
 391
 392 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 393 {
 394         tfile->detached = tun;
 395         list_add_tail(&tfile->next, &tun->disabled);
 396         ++tun->numdisabled;
 397 }
 398
 399 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 400 {
 401         struct tun_struct *tun = tfile->detached;
 402
 403         tfile->detached = NULL;
 404         list_del_init(&tfile->next);
 405         --tun->numdisabled;
 406         return tun;
 407 }
 408
 409 static void __tun_detach(struct tun_file *tfile, bool clean)
 410 {
 411         struct tun_file *ntfile;
 412         struct tun_struct *tun;
 413
 414         tun = rtnl_dereference(tfile->tun);
 415
 416         if (tun && !tfile->detached) {
 417                 u16 index = tfile->queue_index;
 418                 BUG_ON(index >= tun->numqueues);
 419
 420                 rcu_assign_pointer(tun->tfiles[index],
 421                                    tun->tfiles[tun->numqueues - 1]);
 422                 ntfile = rtnl_dereference(tun->tfiles[index]);
 423                 ntfile->queue_index = index;
 424
 425                 --tun->numqueues;
 426                 if (clean) {
 427                         rcu_assign_pointer(tfile->tun, NULL);
 428                         sock_put(&tfile->sk);
 429                 } else
 430                         tun_disable_queue(tun, tfile);
 431
 432                 synchronize_net();
 433                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 434                 /* Drop read queue */
 435                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 436                 tun_set_real_num_queues(tun);
 437         } else if (tfile->detached && clean) {
 438                 tun = tun_enable_queue(tfile);
 439                 sock_put(&tfile->sk);
 440         }
 441
 442         if (clean) {
 443                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 444                         netif_carrier_off(tun->dev);
 445
 446                         if (!(tun->flags & TUN_PERSIST) &&
 447                             tun->dev->reg_state == NETREG_REGISTERED)
 448                                 unregister_netdevice(tun->dev);
 449                 }
 450
 451                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 452                                  &tfile->socket.flags));
 453                 sk_release_kernel(&tfile->sk);
 454         }
 455 }
 456
 457 static void tun_detach(struct tun_file *tfile, bool clean)
 458 {
 459         rtnl_lock();
 460         __tun_detach(tfile, clean);
 461         rtnl_unlock();
 462 }
 463
 464 static void tun_detach_all(struct net_device *dev)
 465 {
 466         struct tun_struct *tun = netdev_priv(dev);
 467         struct tun_file *tfile, *tmp;
 468         int i, n = tun->numqueues;
 469
 470         for (i = 0; i < n; i++) {
 471                 tfile = rtnl_dereference(tun->tfiles[i]);
 472                 BUG_ON(!tfile);
 473                 wake_up_all(&tfile->wq.wait);
 474                 rcu_assign_pointer(tfile->tun, NULL);
 475                 --tun->numqueues;
 476         }
 477         list_for_each_entry(tfile, &tun->disabled, next) {
 478                 wake_up_all(&tfile->wq.wait);
 479                 rcu_assign_pointer(tfile->tun, NULL);
 480         }
 481         BUG_ON(tun->numqueues != 0);
 482
 483         synchronize_net();
 484         for (i = 0; i < n; i++) {
 485                 tfile = rtnl_dereference(tun->tfiles[i]);
 486                 /* Drop read queue */
 487                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 488                 sock_put(&tfile->sk);
 489         }
 490         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 491                 tun_enable_queue(tfile);
 492                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 493                 sock_put(&tfile->sk);
 494         }
 495         BUG_ON(tun->numdisabled != 0);
 496
 497         if (tun->flags & TUN_PERSIST)
 498                 module_put(THIS_MODULE);
 499 }
 500
 501 static int tun_attach(struct tun_struct *tun, struct file *file)
 502 {
 503         struct tun_file *tfile = file->private_data;
 504         int err;
 505
 506         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 507         if (err < 0)
 508                 goto out;
 509
 510         err = -EINVAL;
 511         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 512                 goto out;
 513
 514         err = -EBUSY;
 515         if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 516                 goto out;
 517
 518         err = -E2BIG;
 519         if (!tfile->detached &&
 520             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 521                 goto out;
 522
 523         err = 0;
 524
 525         /* Re-attach the filter to presist device */
 526         if (tun->filter_attached == true) {
 527                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 528                 if (!err)
 529                         goto out;
 530         }
 531         tfile->queue_index = tun->numqueues;
 532         rcu_assign_pointer(tfile->tun, tun);
 533         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 534         tun->numqueues++;
 535
 536         if (tfile->detached)
 537                 tun_enable_queue(tfile);
 538         else
 539                 sock_hold(&tfile->sk);
 540
 541         tun_set_real_num_queues(tun);
 542
 543         /* device is allowed to go away first, so no need to hold extra
 544          * refcnt.
 545          */
 546
 547 out:
 548         return err;
 549 }
 550
 551 static struct tun_struct *__tun_get(struct tun_file *tfile)
 552 {
 553         struct tun_struct *tun;
 554
 555         rcu_read_lock();
 556         tun = rcu_dereference(tfile->tun);
 557         if (tun)
 558                 dev_hold(tun->dev);
 559         rcu_read_unlock();
 560
 561         return tun;
 562 }
 563
 564 static struct tun_struct *tun_get(struct file *file)
 565 {
 566         return __tun_get(file->private_data);
 567 }
 568
 569 static void tun_put(struct tun_struct *tun)
 570 {
 571         dev_put(tun->dev);
 572 }
 573
 574 /* TAP filtering */
 575 static void addr_hash_set(u32 *mask, const u8 *addr)
 576 {
 577         int n = ether_crc(ETH_ALEN, addr) >> 26;
 578         mask[n >> 5] |= (1 << (n & 31));
 579 }
 580
 581 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 582 {
 583         int n = ether_crc(ETH_ALEN, addr) >> 26;
 584         return mask[n >> 5] & (1 << (n & 31));
 585 }
 586
 587 static int update_filter(struct tap_filter *filter, void __user *arg)
 588 {
 589         struct { u8 u[ETH_ALEN]; } *addr;
 590         struct tun_filter uf;
 591         int err, alen, n, nexact;
 592
 593         if (copy_from_user(&uf, arg, sizeof(uf)))
 594                 return -EFAULT;
 595
 596         if (!uf.count) {
 597                 /* Disabled */
 598                 filter->count = 0;
 599                 return 0;
 600         }
 601
 602         alen = ETH_ALEN * uf.count;
 603         addr = kmalloc(alen, GFP_KERNEL);
 604         if (!addr)
 605                 return -ENOMEM;
 606
 607         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 608                 err = -EFAULT;
 609                 goto done;
 610         }
 611
 612         /* The filter is updated without holding any locks. Which is
 613          * perfectly safe. We disable it first and in the worst
 614          * case we'll accept a few undesired packets. */
 615         filter->count = 0;
 616         wmb();
 617
 618         /* Use first set of addresses as an exact filter */
 619         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 620                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 621
 622         nexact = n;
 623
 624         /* Remaining multicast addresses are hashed,
 625          * unicast will leave the filter disabled. */
 626         memset(filter->mask, 0, sizeof(filter->mask));
 627         for (; n < uf.count; n++) {
 628                 if (!is_multicast_ether_addr(addr[n].u)) {
 629                         err = 0; /* no filter */
 630                         goto done;
 631                 }
 632                 addr_hash_set(filter->mask, addr[n].u);
 633         }
 634
 635         /* For ALLMULTI just set the mask to all ones.
 636          * This overrides the mask populated above. */
 637         if ((uf.flags & TUN_FLT_ALLMULTI))
 638                 memset(filter->mask, ~0, sizeof(filter->mask));
 639
 640         /* Now enable the filter */
 641         wmb();
 642         filter->count = nexact;
 643
 644         /* Return the number of exact filters */
 645         err = nexact;
 646
 647 done:
 648         kfree(addr);
 649         return err;
 650 }
 651
 652 /* Returns: 0 - drop, !=0 - accept */
 653 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 654 {
 655         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 656          * at this point. */
 657         struct ethhdr *eh = (struct ethhdr *) skb->data;
 658         int i;
 659
 660         /* Exact match */
 661         for (i = 0; i < filter->count; i++)
 662                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 663                         return 1;
 664
 665         /* Inexact match (multicast only) */
 666         if (is_multicast_ether_addr(eh->h_dest))
 667                 return addr_hash_test(filter->mask, eh->h_dest);
 668
 669         return 0;
 670 }
 671
 672 /*
 673  * Checks whether the packet is accepted or not.
 674  * Returns: 0 - drop, !=0 - accept
 675  */
 676 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 677 {
 678         if (!filter->count)
 679                 return 1;
 680
 681         return run_filter(filter, skb);
 682 }
 683
 684 /* Network device part of the driver */
 685
 686 static const struct ethtool_ops tun_ethtool_ops;
 687
 688 /* Net device detach from fd. */
 689 static void tun_net_uninit(struct net_device *dev)
 690 {
 691         tun_detach_all(dev);
 692 }
 693
 694 /* Net device open. */
 695 static int tun_net_open(struct net_device *dev)
 696 {
 697         netif_tx_start_all_queues(dev);
 698         return 0;
 699 }
 700
 701 /* Net device close. */
 702 static int tun_net_close(struct net_device *dev)
 703 {
 704         netif_tx_stop_all_queues(dev);
 705         return 0;
 706 }
 707
 708 /* Net device start xmit */
 709 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 710 {
 711         struct tun_struct *tun = netdev_priv(dev);
 712         int txq = skb->queue_mapping;
 713         struct tun_file *tfile;
 714
 715         rcu_read_lock();
 716         tfile = rcu_dereference(tun->tfiles[txq]);
 717
 718         /* Drop packet if interface is not attached */
 719         if (txq >= tun->numqueues)
 720                 goto drop;
 721
 722         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 723
 724         BUG_ON(!tfile);
 725
 726         /* Drop if the filter does not like it.
 727          * This is a noop if the filter is disabled.
 728          * Filter can be enabled only for the TAP devices. */
 729         if (!check_filter(&tun->txflt, skb))
 730                 goto drop;
 731
 732         if (tfile->socket.sk->sk_filter &&
 733             sk_filter(tfile->socket.sk, skb))
 734                 goto drop;
 735
 736         /* Limit the number of packets queued by dividing txq length with the
 737          * number of queues.
 738          */
 739         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
 740                           >= dev->tx_queue_len / tun->numqueues)
 741                 goto drop;
 742
 743         if (skb->sk) {
 744                 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
 745                 sw_tx_timestamp(skb);
 746         }
 747
 748         /* Orphan the skb - required as we might hang on to it
 749          * for indefinite time. */
 750         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 751                 goto drop;
 752         skb_orphan(skb);
 753
 754         nf_reset(skb);
 755
 756         /* Enqueue packet */
 757         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 758
 759         /* Notify and wake up reader process */
 760         if (tfile->flags & TUN_FASYNC)
 761                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 762         wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
 763                                    POLLRDNORM | POLLRDBAND);
 764
 765         rcu_read_unlock();
 766         return NETDEV_TX_OK;
 767
 768 drop:
 769         dev->stats.tx_dropped++;
 770         skb_tx_error(skb);
 771         kfree_skb(skb);
 772         rcu_read_unlock();
 773         return NETDEV_TX_OK;
 774 }
 775
 776 static void tun_net_mclist(struct net_device *dev)
 777 {
 778         /*
 779          * This callback is supposed to deal with mc filter in
 780          * _rx_ path and has nothing to do with the _tx_ path.
 781          * In rx path we always accept everything userspace gives us.
 782          */
 783 }
 784
 785 #define MIN_MTU 68
 786 #define MAX_MTU 65535
 787
 788 static int
 789 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 790 {
 791         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 792                 return -EINVAL;
 793         dev->mtu = new_mtu;
 794         return 0;
 795 }
 796
 797 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 798         netdev_features_t features)
 799 {
 800         struct tun_struct *tun = netdev_priv(dev);
 801
 802         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 803 }
 804 #ifdef CONFIG_NET_POLL_CONTROLLER
 805 static void tun_poll_controller(struct net_device *dev)
 806 {
 807         /*
 808          * Tun only receives frames when:
 809          * 1) the char device endpoint gets data from user space
 810          * 2) the tun socket gets a sendmsg call from user space
 811          * Since both of those are syncronous operations, we are guaranteed
 812          * never to have pending data when we poll for it
 813          * so theres nothing to do here but return.
 814          * We need this though so netpoll recognizes us as an interface that
 815          * supports polling, which enables bridge devices in virt setups to
 816          * still use netconsole
 817          */
 818         return;
 819 }
 820 #endif
 821 static const struct net_device_ops tun_netdev_ops = {
 822         .ndo_uninit             = tun_net_uninit,
 823         .ndo_open               = tun_net_open,
 824         .ndo_stop               = tun_net_close,
 825         .ndo_start_xmit         = tun_net_xmit,
 826         .ndo_change_mtu         = tun_net_change_mtu,
 827         .ndo_fix_features       = tun_net_fix_features,
 828         .ndo_select_queue       = tun_select_queue,
 829 #ifdef CONFIG_NET_POLL_CONTROLLER
 830         .ndo_poll_controller    = tun_poll_controller,
 831 #endif
 832 };
 833
 834 static const struct net_device_ops tap_netdev_ops = {
 835         .ndo_uninit             = tun_net_uninit,
 836         .ndo_open               = tun_net_open,
 837         .ndo_stop               = tun_net_close,
 838         .ndo_start_xmit         = tun_net_xmit,
 839         .ndo_change_mtu         = tun_net_change_mtu,
 840         .ndo_fix_features       = tun_net_fix_features,
 841         .ndo_set_rx_mode        = tun_net_mclist,
 842         .ndo_set_mac_address    = eth_mac_addr,
 843         .ndo_validate_addr      = eth_validate_addr,
 844         .ndo_select_queue       = tun_select_queue,
 845 #ifdef CONFIG_NET_POLL_CONTROLLER
 846         .ndo_poll_controller    = tun_poll_controller,
 847 #endif
 848 };
 849
 850 static void tun_flow_init(struct tun_struct *tun)
 851 {
 852         int i;
 853
 854         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 855                 INIT_HLIST_HEAD(&tun->flows[i]);
 856
 857         tun->ageing_time = TUN_FLOW_EXPIRE;
 858         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 859         mod_timer(&tun->flow_gc_timer,
 860                   round_jiffies_up(jiffies + tun->ageing_time));
 861 }
 862
 863 static void tun_flow_uninit(struct tun_struct *tun)
 864 {
 865         del_timer_sync(&tun->flow_gc_timer);
 866         tun_flow_flush(tun);
 867 }
 868
 869 /* Initialize net device. */
 870 static void tun_net_init(struct net_device *dev)
 871 {
 872         struct tun_struct *tun = netdev_priv(dev);
 873
 874         switch (tun->flags & TUN_TYPE_MASK) {
 875         case TUN_TUN_DEV:
 876                 dev->netdev_ops = &tun_netdev_ops;
 877
 878                 /* Point-to-Point TUN Device */
 879                 dev->hard_header_len = 0;
 880                 dev->addr_len = 0;
 881                 dev->mtu = 1500;
 882
 883                 /* Zero header length */
 884                 dev->type = ARPHRD_NONE;
 885                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 886                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 887                 break;
 888
 889         case TUN_TAP_DEV:
 890                 dev->netdev_ops = &tap_netdev_ops;
 891                 /* Ethernet TAP Device */
 892                 ether_setup(dev);
 893                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 894                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 895
 896                 eth_hw_addr_random(dev);
 897
 898                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 899                 break;
 900         }
 901 }
 902
 903 /* Character device part */
 904
 905 /* Poll */
 906 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 907 {
 908         struct tun_file *tfile = file->private_data;
 909         struct tun_struct *tun = __tun_get(tfile);
 910         struct sock *sk;
 911         unsigned int mask = 0;
 912
 913         if (!tun)
 914                 return POLLERR;
 915
 916         sk = tfile->socket.sk;
 917
 918         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 919
 920         poll_wait(file, &tfile->wq.wait, wait);
 921
 922         if (!skb_queue_empty(&sk->sk_receive_queue))
 923                 mask |= POLLIN | POLLRDNORM;
 924
 925         if (sock_writeable(sk) ||
 926             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 927              sock_writeable(sk)))
 928                 mask |= POLLOUT | POLLWRNORM;
 929
 930         if (tun->dev->reg_state != NETREG_REGISTERED)
 931                 mask = POLLERR;
 932
 933         tun_put(tun);
 934         return mask;
 935 }
 936
 937 /* prepad is the amount to reserve at front.  len is length after that.
 938  * linear is a hint as to how much to copy (usually headers). */
 939 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 940                                      size_t prepad, size_t len,
 941                                      size_t linear, int noblock)
 942 {
 943         struct sock *sk = tfile->socket.sk;
 944         struct sk_buff *skb;
 945         int err;
 946
 947         /* Under a page?  Don't bother with paged skb. */
 948         if (prepad + len < PAGE_SIZE || !linear)
 949                 linear = len;
 950
 951         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 952                                    &err);
 953         if (!skb)
 954                 return ERR_PTR(err);
 955
 956         skb_reserve(skb, prepad);
 957         skb_put(skb, linear);
 958         skb->data_len = len - linear;
 959         skb->len += len - linear;
 960
 961         return skb;
 962 }
 963
 964 /* Get packet from user space buffer */
 965 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
 966                             void *msg_control, const struct iovec *iv,
 967                             size_t total_len, size_t count, int noblock)
 968 {
 969         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
 970         struct sk_buff *skb;
 971         size_t len = total_len, align = NET_SKB_PAD, linear;
 972         struct virtio_net_hdr gso = { 0 };
 973         int offset = 0;
 974         int copylen;
 975         bool zerocopy = false;
 976         int err;
 977         u32 rxhash;
 978
 979         if (!(tun->flags & TUN_NO_PI)) {
 980                 if ((len -= sizeof(pi)) > total_len)
 981                         return -EINVAL;
 982
 983                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
 984                         return -EFAULT;
 985                 offset += sizeof(pi);
 986         }
 987
 988         if (tun->flags & TUN_VNET_HDR) {
 989                 if ((len -= tun->vnet_hdr_sz) > total_len)
 990                         return -EINVAL;
 991
 992                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
 993                         return -EFAULT;
 994
 995                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
 996                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
 997                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
 998
 999                 if (gso.hdr_len > len)
1000                         return -EINVAL;
1001                 offset += tun->vnet_hdr_sz;
1002         }
1003
1004         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1005                 align += NET_IP_ALIGN;
1006                 if (unlikely(len < ETH_HLEN ||
1007                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1008                         return -EINVAL;
1009         }
1010
1011         if (msg_control) {
1012                 /* There are 256 bytes to be copied in skb, so there is
1013                  * enough room for skb expand head in case it is used.
1014                  * The rest of the buffer is mapped from userspace.
1015                  */
1016                 copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1017                 linear = copylen;
1018                 if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1019                         zerocopy = true;
1020         }
1021
1022         if (!zerocopy) {
1023                 copylen = len;
1024                 linear = gso.hdr_len;
1025         }
1026
1027         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1028         if (IS_ERR(skb)) {
1029                 if (PTR_ERR(skb) != -EAGAIN)
1030                         tun->dev->stats.rx_dropped++;
1031                 return PTR_ERR(skb);
1032         }
1033
1034         if (zerocopy)
1035                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1036         else {
1037                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1038                 if (!err && msg_control) {
1039                         struct ubuf_info *uarg = msg_control;
1040                         uarg->callback(uarg, false);
1041                 }
1042         }
1043
1044         if (err) {
1045                 tun->dev->stats.rx_dropped++;
1046                 kfree_skb(skb);
1047                 return -EFAULT;
1048         }
1049
1050         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1051                 if (!skb_partial_csum_set(skb, gso.csum_start,
1052                                           gso.csum_offset)) {
1053                         tun->dev->stats.rx_frame_errors++;
1054                         kfree_skb(skb);
1055                         return -EINVAL;
1056                 }
1057         }
1058
1059         switch (tun->flags & TUN_TYPE_MASK) {
1060         case TUN_TUN_DEV:
1061                 if (tun->flags & TUN_NO_PI) {
1062                         switch (skb->data[0] & 0xf0) {
1063                         case 0x40:
1064                                 pi.proto = htons(ETH_P_IP);
1065                                 break;
1066                         case 0x60:
1067                                 pi.proto = htons(ETH_P_IPV6);
1068                                 break;
1069                         default:
1070                                 tun->dev->stats.rx_dropped++;
1071                                 kfree_skb(skb);
1072                                 return -EINVAL;
1073                         }
1074                 }
1075
1076                 skb_reset_mac_header(skb);
1077                 skb->protocol = pi.proto;
1078                 skb->dev = tun->dev;
1079                 break;
1080         case TUN_TAP_DEV:
1081                 skb->protocol = eth_type_trans(skb, tun->dev);
1082                 break;
1083         }
1084
1085         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1086                 pr_debug("GSO!\n");
1087                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1088                 case VIRTIO_NET_HDR_GSO_TCPV4:
1089                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1090                         break;
1091                 case VIRTIO_NET_HDR_GSO_TCPV6:
1092                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1093                         break;
1094                 case VIRTIO_NET_HDR_GSO_UDP:
1095                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1096                         break;
1097                 default:
1098                         tun->dev->stats.rx_frame_errors++;
1099                         kfree_skb(skb);
1100                         return -EINVAL;
1101                 }
1102
1103                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1104                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1105
1106                 skb_shinfo(skb)->gso_size = gso.gso_size;
1107                 if (skb_shinfo(skb)->gso_size == 0) {
1108                         tun->dev->stats.rx_frame_errors++;
1109                         kfree_skb(skb);
1110                         return -EINVAL;
1111                 }
1112
1113                 /* Header must be checked, and gso_segs computed. */
1114                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1115                 skb_shinfo(skb)->gso_segs = 0;
1116         }
1117
1118         /* copy skb_ubuf_info for callback when skb has no error */
1119         if (zerocopy) {
1120                 skb_shinfo(skb)->destructor_arg = msg_control;
1121                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1122                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1123         }
1124
1125         skb_reset_network_header(skb);
1126         skb_probe_transport_header(skb, 0);
1127
1128         rxhash = skb_get_rxhash(skb);
1129         netif_rx_ni(skb);
1130
1131         tun->dev->stats.rx_packets++;
1132         tun->dev->stats.rx_bytes += len;
1133
1134         tun_flow_update(tun, rxhash, tfile);
1135         return total_len;
1136 }
1137
1138 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1139                               unsigned long count, loff_t pos)
1140 {
1141         struct file *file = iocb->ki_filp;
1142         struct tun_struct *tun = tun_get(file);
1143         struct tun_file *tfile = file->private_data;
1144         ssize_t result;
1145
1146         if (!tun)
1147                 return -EBADFD;
1148
1149         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1150
1151         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1152                               count, file->f_flags & O_NONBLOCK);
1153
1154         tun_put(tun);
1155         return result;
1156 }
1157
1158 /* Put packet to the user space buffer */
1159 static ssize_t tun_put_user(struct tun_struct *tun,
1160                             struct tun_file *tfile,
1161                             struct sk_buff *skb,
1162                             const struct iovec *iv, int len)
1163 {
1164         struct tun_pi pi = { 0, skb->protocol };
1165         ssize_t total = 0;
1166         int vlan_offset = 0;
1167
1168         if (!(tun->flags & TUN_NO_PI)) {
1169                 if ((len -= sizeof(pi)) < 0)
1170                         return -EINVAL;
1171
1172                 if (len < skb->len) {
1173                         /* Packet will be striped */
1174                         pi.flags |= TUN_PKT_STRIP;
1175                 }
1176
1177                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1178                         return -EFAULT;
1179                 total += sizeof(pi);
1180         }
1181
1182         if (tun->flags & TUN_VNET_HDR) {
1183                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1184                 if ((len -= tun->vnet_hdr_sz) < 0)
1185                         return -EINVAL;
1186
1187                 if (skb_is_gso(skb)) {
1188                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1189
1190                         /* This is a hint as to how much should be linear. */
1191                         gso.hdr_len = skb_headlen(skb);
1192                         gso.gso_size = sinfo->gso_size;
1193                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1194                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1195                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1196                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1197                         else if (sinfo->gso_type & SKB_GSO_UDP)
1198                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1199                         else {
1200                                 pr_err("unexpected GSO type: "
1201                                        "0x%x, gso_size %d, hdr_len %d\n",
1202                                        sinfo->gso_type, gso.gso_size,
1203                                        gso.hdr_len);
1204                                 print_hex_dump(KERN_ERR, "tun: ",
1205                                                DUMP_PREFIX_NONE,
1206                                                16, 1, skb->head,
1207                                                min((int)gso.hdr_len, 64), true);
1208                                 WARN_ON_ONCE(1);
1209                                 return -EINVAL;
1210                         }
1211                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1212                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1213                 } else
1214                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1215
1216                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1217                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1218                         gso.csum_start = skb_checksum_start_offset(skb);
1219                         gso.csum_offset = skb->csum_offset;
1220                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1221                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1222                 } /* else everything is zero */
1223
1224                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1225                                                sizeof(gso))))
1226                         return -EFAULT;
1227                 total += tun->vnet_hdr_sz;
1228         }
1229
1230         if (!vlan_tx_tag_present(skb)) {
1231                 len = min_t(int, skb->len, len);
1232         } else {
1233                 int copy, ret;
1234                 struct {
1235                         __be16 h_vlan_proto;
1236                         __be16 h_vlan_TCI;
1237                 } veth;
1238
1239                 veth.h_vlan_proto = skb->vlan_proto;
1240                 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1241
1242                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1243                 len = min_t(int, skb->len + VLAN_HLEN, len);
1244
1245                 copy = min_t(int, vlan_offset, len);
1246                 ret = skb_copy_datagram_const_iovec(skb, 0, iv, total, copy);
1247                 len -= copy;
1248                 total += copy;
1249                 if (ret || !len)
1250                         goto done;
1251
1252                 copy = min_t(int, sizeof(veth), len);
1253                 ret = memcpy_toiovecend(iv, (void *)&veth, total, copy);
1254                 len -= copy;
1255                 total += copy;
1256                 if (ret || !len)
1257                         goto done;
1258         }
1259
1260         skb_copy_datagram_const_iovec(skb, vlan_offset, iv, total, len);
1261         total += len;
1262
1263 done:
1264         tun->dev->stats.tx_packets++;
1265         tun->dev->stats.tx_bytes += len;
1266
1267         return total;
1268 }
1269
1270 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1271                            struct kiocb *iocb, const struct iovec *iv,
1272                            ssize_t len, int noblock)
1273 {
1274         DECLARE_WAITQUEUE(wait, current);
1275         struct sk_buff *skb;
1276         ssize_t ret = 0;
1277
1278         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1279
1280         if (unlikely(!noblock))
1281                 add_wait_queue(&tfile->wq.wait, &wait);
1282         while (len) {
1283                 current->state = TASK_INTERRUPTIBLE;
1284
1285                 /* Read frames from the queue */
1286                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1287                         if (noblock) {
1288                                 ret = -EAGAIN;
1289                                 break;
1290                         }
1291                         if (signal_pending(current)) {
1292                                 ret = -ERESTARTSYS;
1293                                 break;
1294                         }
1295                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1296                                 ret = -EIO;
1297                                 break;
1298                         }
1299
1300                         /* Nothing to read, let's sleep */
1301                         schedule();
1302                         continue;
1303                 }
1304
1305                 ret = tun_put_user(tun, tfile, skb, iv, len);
1306                 kfree_skb(skb);
1307                 break;
1308         }
1309
1310         current->state = TASK_RUNNING;
1311         if (unlikely(!noblock))
1312                 remove_wait_queue(&tfile->wq.wait, &wait);
1313
1314         return ret;
1315 }
1316
1317 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1318                             unsigned long count, loff_t pos)
1319 {
1320         struct file *file = iocb->ki_filp;
1321         struct tun_file *tfile = file->private_data;
1322         struct tun_struct *tun = __tun_get(tfile);
1323         ssize_t len, ret;
1324
1325         if (!tun)
1326                 return -EBADFD;
1327         len = iov_length(iv, count);
1328         if (len < 0) {
1329                 ret = -EINVAL;
1330                 goto out;
1331         }
1332
1333         ret = tun_do_read(tun, tfile, iocb, iv, len,
1334                           file->f_flags & O_NONBLOCK);
1335         ret = min_t(ssize_t, ret, len);
1336 out:
1337         tun_put(tun);
1338         return ret;
1339 }
1340
1341 static void tun_free_netdev(struct net_device *dev)
1342 {
1343         struct tun_struct *tun = netdev_priv(dev);
1344
1345         BUG_ON(!(list_empty(&tun->disabled)));
1346         tun_flow_uninit(tun);
1347         security_tun_dev_free_security(tun->security);
1348         free_netdev(dev);
1349 }
1350
1351 static void tun_setup(struct net_device *dev)
1352 {
1353         struct tun_struct *tun = netdev_priv(dev);
1354
1355         tun->owner = INVALID_UID;
1356         tun->group = INVALID_GID;
1357
1358         dev->ethtool_ops = &tun_ethtool_ops;
1359         dev->destructor = tun_free_netdev;
1360 }
1361
1362 /* Trivial set of netlink ops to allow deleting tun or tap
1363  * device with netlink.
1364  */
1365 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1366 {
1367         return -EINVAL;
1368 }
1369
1370 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1371         .kind           = DRV_NAME,
1372         .priv_size      = sizeof(struct tun_struct),
1373         .setup          = tun_setup,
1374         .validate       = tun_validate,
1375 };
1376
1377 static void tun_sock_write_space(struct sock *sk)
1378 {
1379         struct tun_file *tfile;
1380         wait_queue_head_t *wqueue;
1381
1382         if (!sock_writeable(sk))
1383                 return;
1384
1385         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1386                 return;
1387
1388         wqueue = sk_sleep(sk);
1389         if (wqueue && waitqueue_active(wqueue))
1390                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1391                                                 POLLWRNORM | POLLWRBAND);
1392
1393         tfile = container_of(sk, struct tun_file, sk);
1394         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1395 }
1396
1397 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1398                        struct msghdr *m, size_t total_len)
1399 {
1400         int ret;
1401         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1402         struct tun_struct *tun = __tun_get(tfile);
1403
1404         if (!tun)
1405                 return -EBADFD;
1406         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1407                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1408         tun_put(tun);
1409         return ret;
1410 }
1411
1412 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1413                        struct msghdr *m, size_t total_len,
1414                        int flags)
1415 {
1416         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1417         struct tun_struct *tun = __tun_get(tfile);
1418         int ret;
1419
1420         if (!tun)
1421                 return -EBADFD;
1422
1423         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1424                 ret = -EINVAL;
1425                 goto out;
1426         }
1427         if (flags & MSG_ERRQUEUE) {
1428                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1429                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1430                 goto out;
1431         }
1432         ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1433                           flags & MSG_DONTWAIT);
1434         if (ret > total_len) {
1435                 m->msg_flags |= MSG_TRUNC;
1436                 ret = flags & MSG_TRUNC ? ret : total_len;
1437         }
1438 out:
1439         tun_put(tun);
1440         return ret;
1441 }
1442
1443 static int tun_release(struct socket *sock)
1444 {
1445         if (sock->sk)
1446                 sock_put(sock->sk);
1447         return 0;
1448 }
1449
1450 /* Ops structure to mimic raw sockets with tun */
1451 static const struct proto_ops tun_socket_ops = {
1452         .sendmsg = tun_sendmsg,
1453         .recvmsg = tun_recvmsg,
1454         .release = tun_release,
1455 };
1456
1457 static struct proto tun_proto = {
1458         .name           = "tun",
1459         .owner          = THIS_MODULE,
1460         .obj_size       = sizeof(struct tun_file),
1461 };
1462
1463 static int tun_flags(struct tun_struct *tun)
1464 {
1465         int flags = 0;
1466
1467         if (tun->flags & TUN_TUN_DEV)
1468                 flags |= IFF_TUN;
1469         else
1470                 flags |= IFF_TAP;
1471
1472         if (tun->flags & TUN_NO_PI)
1473                 flags |= IFF_NO_PI;
1474
1475         /* This flag has no real effect.  We track the value for backwards
1476          * compatibility.
1477          */
1478         if (tun->flags & TUN_ONE_QUEUE)
1479                 flags |= IFF_ONE_QUEUE;
1480
1481         if (tun->flags & TUN_VNET_HDR)
1482                 flags |= IFF_VNET_HDR;
1483
1484         if (tun->flags & TUN_TAP_MQ)
1485                 flags |= IFF_MULTI_QUEUE;
1486
1487         if (tun->flags & TUN_PERSIST)
1488                 flags |= IFF_PERSIST;
1489
1490         return flags;
1491 }
1492
1493 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1494                               char *buf)
1495 {
1496         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1497         return sprintf(buf, "0x%x\n", tun_flags(tun));
1498 }
1499
1500 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1501                               char *buf)
1502 {
1503         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1504         return uid_valid(tun->owner)?
1505                 sprintf(buf, "%u\n",
1506                         from_kuid_munged(current_user_ns(), tun->owner)):
1507                 sprintf(buf, "-1\n");
1508 }
1509
1510 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1511                               char *buf)
1512 {
1513         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1514         return gid_valid(tun->group) ?
1515                 sprintf(buf, "%u\n",
1516                         from_kgid_munged(current_user_ns(), tun->group)):
1517                 sprintf(buf, "-1\n");
1518 }
1519
1520 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1521 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1522 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1523
1524 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1525 {
1526         struct tun_struct *tun;
1527         struct tun_file *tfile = file->private_data;
1528         struct net_device *dev;
1529         int err;
1530
1531         if (tfile->detached)
1532                 return -EINVAL;
1533
1534         dev = __dev_get_by_name(net, ifr->ifr_name);
1535         if (dev) {
1536                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1537                         return -EBUSY;
1538                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1539                         tun = netdev_priv(dev);
1540                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1541                         tun = netdev_priv(dev);
1542                 else
1543                         return -EINVAL;
1544
1545                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1546                     !!(tun->flags & TUN_TAP_MQ))
1547                         return -EINVAL;
1548
1549                 if (tun_not_capable(tun))
1550                         return -EPERM;
1551                 err = security_tun_dev_open(tun->security);
1552                 if (err < 0)
1553                         return err;
1554
1555                 err = tun_attach(tun, file);
1556                 if (err < 0)
1557                         return err;
1558
1559                 if (tun->flags & TUN_TAP_MQ &&
1560                     (tun->numqueues + tun->numdisabled > 1)) {
1561                         /* One or more queue has already been attached, no need
1562                          * to initialize the device again.
1563                          */
1564                         return 0;
1565                 }
1566         }
1567         else {
1568                 char *name;
1569                 unsigned long flags = 0;
1570                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1571                              MAX_TAP_QUEUES : 1;
1572
1573                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1574                         return -EPERM;
1575                 err = security_tun_dev_create();
1576                 if (err < 0)
1577                         return err;
1578
1579                 /* Set dev type */
1580                 if (ifr->ifr_flags & IFF_TUN) {
1581                         /* TUN device */
1582                         flags |= TUN_TUN_DEV;
1583                         name = "tun%d";
1584                 } else if (ifr->ifr_flags & IFF_TAP) {
1585                         /* TAP device */
1586                         flags |= TUN_TAP_DEV;
1587                         name = "tap%d";
1588                 } else
1589                         return -EINVAL;
1590
1591                 if (*ifr->ifr_name)
1592                         name = ifr->ifr_name;
1593
1594                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1595                                        tun_setup, queues, queues);
1596
1597                 if (!dev)
1598                         return -ENOMEM;
1599
1600                 dev_net_set(dev, net);
1601                 dev->rtnl_link_ops = &tun_link_ops;
1602
1603                 tun = netdev_priv(dev);
1604                 tun->dev = dev;
1605                 tun->flags = flags;
1606                 tun->txflt.count = 0;
1607                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1608
1609                 tun->filter_attached = false;
1610                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1611
1612                 spin_lock_init(&tun->lock);
1613
1614                 err = security_tun_dev_alloc_security(&tun->security);
1615                 if (err < 0)
1616                         goto err_free_dev;
1617
1618                 tun_net_init(dev);
1619                 tun_flow_init(tun);
1620
1621                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1622                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1623                                    NETIF_F_HW_VLAN_STAG_TX;
1624                 dev->features = dev->hw_features;
1625                 dev->vlan_features = dev->features;
1626
1627                 INIT_LIST_HEAD(&tun->disabled);
1628                 err = tun_attach(tun, file);
1629                 if (err < 0)
1630                         goto err_free_dev;
1631
1632                 err = register_netdevice(tun->dev);
1633                 if (err < 0)
1634                         goto err_free_dev;
1635
1636                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1637                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1638                     device_create_file(&tun->dev->dev, &dev_attr_group))
1639                         pr_err("Failed to create tun sysfs files\n");
1640         }
1641
1642         netif_carrier_on(tun->dev);
1643
1644         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1645
1646         if (ifr->ifr_flags & IFF_NO_PI)
1647                 tun->flags |= TUN_NO_PI;
1648         else
1649                 tun->flags &= ~TUN_NO_PI;
1650
1651         /* This flag has no real effect.  We track the value for backwards
1652          * compatibility.
1653          */
1654         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1655                 tun->flags |= TUN_ONE_QUEUE;
1656         else
1657                 tun->flags &= ~TUN_ONE_QUEUE;
1658
1659         if (ifr->ifr_flags & IFF_VNET_HDR)
1660                 tun->flags |= TUN_VNET_HDR;
1661         else
1662                 tun->flags &= ~TUN_VNET_HDR;
1663
1664         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1665                 tun->flags |= TUN_TAP_MQ;
1666         else
1667                 tun->flags &= ~TUN_TAP_MQ;
1668
1669         /* Make sure persistent devices do not get stuck in
1670          * xoff state.
1671          */
1672         if (netif_running(tun->dev))
1673                 netif_tx_wake_all_queues(tun->dev);
1674
1675         strcpy(ifr->ifr_name, tun->dev->name);
1676         return 0;
1677
1678  err_free_dev:
1679         free_netdev(dev);
1680         return err;
1681 }
1682
1683 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1684                        struct ifreq *ifr)
1685 {
1686         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1687
1688         strcpy(ifr->ifr_name, tun->dev->name);
1689
1690         ifr->ifr_flags = tun_flags(tun);
1691
1692 }
1693
1694 /* This is like a cut-down ethtool ops, except done via tun fd so no
1695  * privs required. */
1696 static int set_offload(struct tun_struct *tun, unsigned long arg)
1697 {
1698         netdev_features_t features = 0;
1699
1700         if (arg & TUN_F_CSUM) {
1701                 features |= NETIF_F_HW_CSUM;
1702                 arg &= ~TUN_F_CSUM;
1703
1704                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1705                         if (arg & TUN_F_TSO_ECN) {
1706                                 features |= NETIF_F_TSO_ECN;
1707                                 arg &= ~TUN_F_TSO_ECN;
1708                         }
1709                         if (arg & TUN_F_TSO4)
1710                                 features |= NETIF_F_TSO;
1711                         if (arg & TUN_F_TSO6)
1712                                 features |= NETIF_F_TSO6;
1713                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1714                 }
1715
1716                 if (arg & TUN_F_UFO) {
1717                         features |= NETIF_F_UFO;
1718                         arg &= ~TUN_F_UFO;
1719                 }
1720         }
1721
1722         /* This gives the user a way to test for new features in future by
1723          * trying to set them. */
1724         if (arg)
1725                 return -EINVAL;
1726
1727         tun->set_features = features;
1728         netdev_update_features(tun->dev);
1729
1730         return 0;
1731 }
1732
1733 static void tun_detach_filter(struct tun_struct *tun, int n)
1734 {
1735         int i;
1736         struct tun_file *tfile;
1737
1738         for (i = 0; i < n; i++) {
1739                 tfile = rtnl_dereference(tun->tfiles[i]);
1740                 sk_detach_filter(tfile->socket.sk);
1741         }
1742
1743         tun->filter_attached = false;
1744 }
1745
1746 static int tun_attach_filter(struct tun_struct *tun)
1747 {
1748         int i, ret = 0;
1749         struct tun_file *tfile;
1750
1751         for (i = 0; i < tun->numqueues; i++) {
1752                 tfile = rtnl_dereference(tun->tfiles[i]);
1753                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1754                 if (ret) {
1755                         tun_detach_filter(tun, i);
1756                         return ret;
1757                 }
1758         }
1759
1760         tun->filter_attached = true;
1761         return ret;
1762 }
1763
1764 static void tun_set_sndbuf(struct tun_struct *tun)
1765 {
1766         struct tun_file *tfile;
1767         int i;
1768
1769         for (i = 0; i < tun->numqueues; i++) {
1770                 tfile = rtnl_dereference(tun->tfiles[i]);
1771                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1772         }
1773 }
1774
1775 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1776 {
1777         struct tun_file *tfile = file->private_data;
1778         struct tun_struct *tun;
1779         int ret = 0;
1780
1781         rtnl_lock();
1782
1783         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1784                 tun = tfile->detached;
1785                 if (!tun) {
1786                         ret = -EINVAL;
1787                         goto unlock;
1788                 }
1789                 ret = security_tun_dev_attach_queue(tun->security);
1790                 if (ret < 0)
1791                         goto unlock;
1792                 ret = tun_attach(tun, file);
1793         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1794                 tun = rtnl_dereference(tfile->tun);
1795                 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1796                         ret = -EINVAL;
1797                 else
1798                         __tun_detach(tfile, false);
1799         } else
1800                 ret = -EINVAL;
1801
1802 unlock:
1803         rtnl_unlock();
1804         return ret;
1805 }
1806
1807 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1808                             unsigned long arg, int ifreq_len)
1809 {
1810         struct tun_file *tfile = file->private_data;
1811         struct tun_struct *tun;
1812         void __user* argp = (void __user*)arg;
1813         struct ifreq ifr;
1814         kuid_t owner;
1815         kgid_t group;
1816         int sndbuf;
1817         int vnet_hdr_sz;
1818         int ret;
1819
1820         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1821                 if (copy_from_user(&ifr, argp, ifreq_len))
1822                         return -EFAULT;
1823         } else {
1824                 memset(&ifr, 0, sizeof(ifr));
1825         }
1826         if (cmd == TUNGETFEATURES) {
1827                 /* Currently this just means: "what IFF flags are valid?".
1828                  * This is needed because we never checked for invalid flags on
1829                  * TUNSETIFF. */
1830                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1831                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1832                                 (unsigned int __user*)argp);
1833         } else if (cmd == TUNSETQUEUE)
1834                 return tun_set_queue(file, &ifr);
1835
1836         ret = 0;
1837         rtnl_lock();
1838
1839         tun = __tun_get(tfile);
1840         if (cmd == TUNSETIFF && !tun) {
1841                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1842
1843                 ret = tun_set_iff(tfile->net, file, &ifr);
1844
1845                 if (ret)
1846                         goto unlock;
1847
1848                 if (copy_to_user(argp, &ifr, ifreq_len))
1849                         ret = -EFAULT;
1850                 goto unlock;
1851         }
1852
1853         ret = -EBADFD;
1854         if (!tun)
1855                 goto unlock;
1856
1857         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1858
1859         ret = 0;
1860         switch (cmd) {
1861         case TUNGETIFF:
1862                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1863
1864                 if (copy_to_user(argp, &ifr, ifreq_len))
1865                         ret = -EFAULT;
1866                 break;
1867
1868         case TUNSETNOCSUM:
1869                 /* Disable/Enable checksum */
1870
1871                 /* [unimplemented] */
1872                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1873                           arg ? "disabled" : "enabled");
1874                 break;
1875
1876         case TUNSETPERSIST:
1877                 /* Disable/Enable persist mode. Keep an extra reference to the
1878                  * module to prevent the module being unprobed.
1879                  */
1880                 if (arg && !(tun->flags & TUN_PERSIST)) {
1881                         tun->flags |= TUN_PERSIST;
1882                         __module_get(THIS_MODULE);
1883                 }
1884                 if (!arg && (tun->flags & TUN_PERSIST)) {
1885                         tun->flags &= ~TUN_PERSIST;
1886                         module_put(THIS_MODULE);
1887                 }
1888
1889                 tun_debug(KERN_INFO, tun, "persist %s\n",
1890                           arg ? "enabled" : "disabled");
1891                 break;
1892
1893         case TUNSETOWNER:
1894                 /* Set owner of the device */
1895                 owner = make_kuid(current_user_ns(), arg);
1896                 if (!uid_valid(owner)) {
1897                         ret = -EINVAL;
1898                         break;
1899                 }
1900                 tun->owner = owner;
1901                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1902                           from_kuid(&init_user_ns, tun->owner));
1903                 break;
1904
1905         case TUNSETGROUP:
1906                 /* Set group of the device */
1907                 group = make_kgid(current_user_ns(), arg);
1908                 if (!gid_valid(group)) {
1909                         ret = -EINVAL;
1910                         break;
1911                 }
1912                 tun->group = group;
1913                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1914                           from_kgid(&init_user_ns, tun->group));
1915                 break;
1916
1917         case TUNSETLINK:
1918                 /* Only allow setting the type when the interface is down */
1919                 if (tun->dev->flags & IFF_UP) {
1920                         tun_debug(KERN_INFO, tun,
1921                                   "Linktype set failed because interface is up\n");
1922                         ret = -EBUSY;
1923                 } else {
1924                         tun->dev->type = (int) arg;
1925                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1926                                   tun->dev->type);
1927                         ret = 0;
1928                 }
1929                 break;
1930
1931 #ifdef TUN_DEBUG
1932         case TUNSETDEBUG:
1933                 tun->debug = arg;
1934                 break;
1935 #endif
1936         case TUNSETOFFLOAD:
1937                 ret = set_offload(tun, arg);
1938                 break;
1939
1940         case TUNSETTXFILTER:
1941                 /* Can be set only for TAPs */
1942                 ret = -EINVAL;
1943                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1944                         break;
1945                 ret = update_filter(&tun->txflt, (void __user *)arg);
1946                 break;
1947
1948         case SIOCGIFHWADDR:
1949                 /* Get hw address */
1950                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1951                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1952                 if (copy_to_user(argp, &ifr, ifreq_len))
1953                         ret = -EFAULT;
1954                 break;
1955
1956         case SIOCSIFHWADDR:
1957                 /* Set hw address */
1958                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1959                           ifr.ifr_hwaddr.sa_data);
1960
1961                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1962                 break;
1963
1964         case TUNGETSNDBUF:
1965                 sndbuf = tfile->socket.sk->sk_sndbuf;
1966                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1967                         ret = -EFAULT;
1968                 break;
1969
1970         case TUNSETSNDBUF:
1971                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1972                         ret = -EFAULT;
1973                         break;
1974                 }
1975
1976                 tun->sndbuf = sndbuf;
1977                 tun_set_sndbuf(tun);
1978                 break;
1979
1980         case TUNGETVNETHDRSZ:
1981                 vnet_hdr_sz = tun->vnet_hdr_sz;
1982                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1983                         ret = -EFAULT;
1984                 break;
1985
1986         case TUNSETVNETHDRSZ:
1987                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1988                         ret = -EFAULT;
1989                         break;
1990                 }
1991                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1992                         ret = -EINVAL;
1993                         break;
1994                 }
1995
1996                 tun->vnet_hdr_sz = vnet_hdr_sz;
1997                 break;
1998
1999         case TUNATTACHFILTER:
2000                 /* Can be set only for TAPs */
2001                 ret = -EINVAL;
2002                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2003                         break;
2004                 ret = -EFAULT;
2005                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2006                         break;
2007
2008                 ret = tun_attach_filter(tun);
2009                 break;
2010
2011         case TUNDETACHFILTER:
2012                 /* Can be set only for TAPs */
2013                 ret = -EINVAL;
2014                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2015                         break;
2016                 ret = 0;
2017                 tun_detach_filter(tun, tun->numqueues);
2018                 break;
2019
2020         default:
2021                 ret = -EINVAL;
2022                 break;
2023         }
2024
2025 unlock:
2026         rtnl_unlock();
2027         if (tun)
2028                 tun_put(tun);
2029         return ret;
2030 }
2031
2032 static long tun_chr_ioctl(struct file *file,
2033                           unsigned int cmd, unsigned long arg)
2034 {
2035         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2036 }
2037
2038 #ifdef CONFIG_COMPAT
2039 static long tun_chr_compat_ioctl(struct file *file,
2040                          unsigned int cmd, unsigned long arg)
2041 {
2042         switch (cmd) {
2043         case TUNSETIFF:
2044         case TUNGETIFF:
2045         case TUNSETTXFILTER:
2046         case TUNGETSNDBUF:
2047         case TUNSETSNDBUF:
2048         case SIOCGIFHWADDR:
2049         case SIOCSIFHWADDR:
2050                 arg = (unsigned long)compat_ptr(arg);
2051                 break;
2052         default:
2053                 arg = (compat_ulong_t)arg;
2054                 break;
2055         }
2056
2057         /*
2058          * compat_ifreq is shorter than ifreq, so we must not access beyond
2059          * the end of that structure. All fields that are used in this
2060          * driver are compatible though, we don't need to convert the
2061          * contents.
2062          */
2063         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2064 }
2065 #endif /* CONFIG_COMPAT */
2066
2067 static int tun_chr_fasync(int fd, struct file *file, int on)
2068 {
2069         struct tun_file *tfile = file->private_data;
2070         int ret;
2071
2072         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2073                 goto out;
2074
2075         if (on) {
2076                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2077                 if (ret)
2078                         goto out;
2079                 tfile->flags |= TUN_FASYNC;
2080         } else
2081                 tfile->flags &= ~TUN_FASYNC;
2082         ret = 0;
2083 out:
2084         return ret;
2085 }
2086
2087 static int tun_chr_open(struct inode *inode, struct file * file)
2088 {
2089         struct tun_file *tfile;
2090
2091         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2092
2093         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2094                                             &tun_proto);
2095         if (!tfile)
2096                 return -ENOMEM;
2097         rcu_assign_pointer(tfile->tun, NULL);
2098         tfile->net = get_net(current->nsproxy->net_ns);
2099         tfile->flags = 0;
2100
2101         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2102         init_waitqueue_head(&tfile->wq.wait);
2103
2104         tfile->socket.file = file;
2105         tfile->socket.ops = &tun_socket_ops;
2106
2107         sock_init_data(&tfile->socket, &tfile->sk);
2108         sk_change_net(&tfile->sk, tfile->net);
2109
2110         tfile->sk.sk_write_space = tun_sock_write_space;
2111         tfile->sk.sk_sndbuf = INT_MAX;
2112
2113         file->private_data = tfile;
2114         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2115         INIT_LIST_HEAD(&tfile->next);
2116
2117         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2118
2119         return 0;
2120 }
2121
2122 static int tun_chr_close(struct inode *inode, struct file *file)
2123 {
2124         struct tun_file *tfile = file->private_data;
2125         struct net *net = tfile->net;
2126
2127         tun_detach(tfile, true);
2128         put_net(net);
2129
2130         return 0;
2131 }
2132
2133 static const struct file_operations tun_fops = {
2134         .owner  = THIS_MODULE,
2135         .llseek = no_llseek,
2136         .read  = do_sync_read,
2137         .aio_read  = tun_chr_aio_read,
2138         .write = do_sync_write,
2139         .aio_write = tun_chr_aio_write,
2140         .poll   = tun_chr_poll,
2141         .unlocked_ioctl = tun_chr_ioctl,
2142 #ifdef CONFIG_COMPAT
2143         .compat_ioctl = tun_chr_compat_ioctl,
2144 #endif
2145         .open   = tun_chr_open,
2146         .release = tun_chr_close,
2147         .fasync = tun_chr_fasync
2148 };
2149
2150 static struct miscdevice tun_miscdev = {
2151         .minor = TUN_MINOR,
2152         .name = "tun",
2153         .nodename = "net/tun",
2154         .fops = &tun_fops,
2155 };
2156
2157 /* ethtool interface */
2158
2159 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2160 {
2161         cmd->supported          = 0;
2162         cmd->advertising        = 0;
2163         ethtool_cmd_speed_set(cmd, SPEED_10);
2164         cmd->duplex             = DUPLEX_FULL;
2165         cmd->port               = PORT_TP;
2166         cmd->phy_address        = 0;
2167         cmd->transceiver        = XCVR_INTERNAL;
2168         cmd->autoneg            = AUTONEG_DISABLE;
2169         cmd->maxtxpkt           = 0;
2170         cmd->maxrxpkt           = 0;
2171         return 0;
2172 }
2173
2174 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2175 {
2176         struct tun_struct *tun = netdev_priv(dev);
2177
2178         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2179         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2180
2181         switch (tun->flags & TUN_TYPE_MASK) {
2182         case TUN_TUN_DEV:
2183                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2184                 break;
2185         case TUN_TAP_DEV:
2186                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2187                 break;
2188         }
2189 }
2190
2191 static u32 tun_get_msglevel(struct net_device *dev)
2192 {
2193 #ifdef TUN_DEBUG
2194         struct tun_struct *tun = netdev_priv(dev);
2195         return tun->debug;
2196 #else
2197         return -EOPNOTSUPP;
2198 #endif
2199 }
2200
2201 static void tun_set_msglevel(struct net_device *dev, u32 value)
2202 {
2203 #ifdef TUN_DEBUG
2204         struct tun_struct *tun = netdev_priv(dev);
2205         tun->debug = value;
2206 #endif
2207 }
2208
2209 static const struct ethtool_ops tun_ethtool_ops = {
2210         .get_settings   = tun_get_settings,
2211         .get_drvinfo    = tun_get_drvinfo,
2212         .get_msglevel   = tun_get_msglevel,
2213         .set_msglevel   = tun_set_msglevel,
2214         .get_link       = ethtool_op_get_link,
2215         .get_ts_info    = ethtool_op_get_ts_info,
2216 };
2217
2218
2219 static int __init tun_init(void)
2220 {
2221         int ret = 0;
2222
2223         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2224         pr_info("%s\n", DRV_COPYRIGHT);
2225
2226         ret = rtnl_link_register(&tun_link_ops);
2227         if (ret) {
2228                 pr_err("Can't register link_ops\n");
2229                 goto err_linkops;
2230         }
2231
2232         ret = misc_register(&tun_miscdev);
2233         if (ret) {
2234                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2235                 goto err_misc;
2236         }
2237         return  0;
2238 err_misc:
2239         rtnl_link_unregister(&tun_link_ops);
2240 err_linkops:
2241         return ret;
2242 }
2243
2244 static void tun_cleanup(void)
2245 {
2246         misc_deregister(&tun_miscdev);
2247         rtnl_link_unregister(&tun_link_ops);
2248 }
2249
2250 /* Get an underlying socket object from tun file.  Returns error unless file is
2251  * attached to a device.  The returned object works like a packet socket, it
2252  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2253  * holding a reference to the file for as long as the socket is in use. */
2254 struct socket *tun_get_socket(struct file *file)
2255 {
2256         struct tun_file *tfile;
2257         if (file->f_op != &tun_fops)
2258                 return ERR_PTR(-EINVAL);
2259         tfile = file->private_data;
2260         if (!tfile)
2261                 return ERR_PTR(-EBADFD);
2262         return &tfile->socket;
2263 }
2264 EXPORT_SYMBOL_GPL(tun_get_socket);
2265
2266 module_init(tun_init);
2267 module_exit(tun_cleanup);
2268 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2269 MODULE_AUTHOR(DRV_COPYRIGHT);
2270 MODULE_LICENSE("GPL");
2271 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2272 MODULE_ALIAS("devname:net/tun");