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 #include <linux/seq_file.h>
  73 #include <linux/uio.h>
  74 #include <linux/skb_array.h>
  75 #include <linux/ieee802154.h>
  76 #include <linux/if_ltalk.h>
  77 #include <uapi/linux/if_fddi.h>
  78 #include <uapi/linux/if_hippi.h>
  79 #include <uapi/linux/if_fc.h>
  80 #include <net/ax25.h>
  81 #include <net/rose.h>
  82 #include <net/6lowpan.h>
  83
  84 #include <asm/uaccess.h>
  85
  86 /* Uncomment to enable debugging */
  87 /* #define TUN_DEBUG 1 */
  88
  89 #ifdef TUN_DEBUG
  90 static int debug;
  91
  92 #define tun_debug(level, tun, fmt, args...)                     \
  93 do {                                                            \
  94         if (tun->debug)                                         \
  95                 netdev_printk(level, tun->dev, fmt, ##args);    \
  96 } while (0)
  97 #define DBG1(level, fmt, args...)                               \
  98 do {                                                            \
  99         if (debug == 2)                                         \
 100                 printk(level fmt, ##args);                      \
 101 } while (0)
 102 #else
 103 #define tun_debug(level, tun, fmt, args...)                     \
 104 do {                                                            \
 105         if (0)                                                  \
 106                 netdev_printk(level, tun->dev, fmt, ##args);    \
 107 } while (0)
 108 #define DBG1(level, fmt, args...)                               \
 109 do {                                                            \
 110         if (0)                                                  \
 111                 printk(level fmt, ##args);                      \
 112 } while (0)
 113 #endif
 114
 115 /* TUN device flags */
 116
 117 /* IFF_ATTACH_QUEUE is never stored in device flags,
 118  * overload it to mean fasync when stored there.
 119  */
 120 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 121 /* High bits in flags field are unused. */
 122 #define TUN_VNET_LE     0x80000000
 123 #define TUN_VNET_BE     0x40000000
 124
 125 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 126                       IFF_MULTI_QUEUE)
 127 #define GOODCOPY_LEN 128
 128
 129 #define FLT_EXACT_COUNT 8
 130 struct tap_filter {
 131         unsigned int    count;    /* Number of addrs. Zero means disabled */
 132         u32             mask[2];  /* Mask of the hashed addrs */
 133         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 134 };
 135
 136 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 137  * to max number of VCPUs in guest. */
 138 #define MAX_TAP_QUEUES 256
 139 #define MAX_TAP_FLOWS  4096
 140
 141 #define TUN_FLOW_EXPIRE (3 * HZ)
 142
 143 struct tun_pcpu_stats {
 144         u64 rx_packets;
 145         u64 rx_bytes;
 146         u64 tx_packets;
 147         u64 tx_bytes;
 148         struct u64_stats_sync syncp;
 149         u32 rx_dropped;
 150         u32 tx_dropped;
 151         u32 rx_frame_errors;
 152 };
 153
 154 /* A tun_file connects an open character device to a tuntap netdevice. It
 155  * also contains all socket related structures (except sock_fprog and tap_filter)
 156  * to serve as one transmit queue for tuntap device. The sock_fprog and
 157  * tap_filter were kept in tun_struct since they were used for filtering for the
 158  * netdevice not for a specific queue (at least I didn't see the requirement for
 159  * this).
 160  *
 161  * RCU usage:
 162  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 163  * other can only be read while rcu_read_lock or rtnl_lock is held.
 164  */
 165 struct tun_file {
 166         struct sock sk;
 167         struct socket socket;
 168         struct socket_wq wq;
 169         struct tun_struct __rcu *tun;
 170         struct fasync_struct *fasync;
 171         /* only used for fasnyc */
 172         unsigned int flags;
 173         union {
 174                 u16 queue_index;
 175                 unsigned int ifindex;
 176         };
 177         struct list_head next;
 178         struct tun_struct *detached;
 179         struct skb_array tx_array;
 180 };
 181
 182 struct tun_flow_entry {
 183         struct hlist_node hash_link;
 184         struct rcu_head rcu;
 185         struct tun_struct *tun;
 186
 187         u32 rxhash;
 188         u32 rps_rxhash;
 189         int queue_index;
 190         unsigned long updated;
 191 };
 192
 193 #define TUN_NUM_FLOW_ENTRIES 1024
 194
 195 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 196  * device, socket filter, sndbuf and vnet header size were restore when the
 197  * file were attached to a persist device.
 198  */
 199 struct tun_struct {
 200         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 201         unsigned int            numqueues;
 202         unsigned int            flags;
 203         kuid_t                  owner;
 204         kgid_t                  group;
 205
 206         struct net_device       *dev;
 207         netdev_features_t       set_features;
 208 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 209                           NETIF_F_TSO6|NETIF_F_UFO)
 210
 211         int                     align;
 212         int                     vnet_hdr_sz;
 213         int                     sndbuf;
 214         struct tap_filter       txflt;
 215         struct sock_fprog       fprog;
 216         /* protected by rtnl lock */
 217         bool                    filter_attached;
 218 #ifdef TUN_DEBUG
 219         int debug;
 220 #endif
 221         spinlock_t lock;
 222         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 223         struct timer_list flow_gc_timer;
 224         unsigned long ageing_time;
 225         unsigned int numdisabled;
 226         struct list_head disabled;
 227         void *security;
 228         u32 flow_count;
 229         struct tun_pcpu_stats __percpu *pcpu_stats;
 230 };
 231
 232 #ifdef CONFIG_TUN_VNET_CROSS_LE
 233 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 234 {
 235         return tun->flags & TUN_VNET_BE ? false :
 236                 virtio_legacy_is_little_endian();
 237 }
 238
 239 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 240 {
 241         int be = !!(tun->flags & TUN_VNET_BE);
 242
 243         if (put_user(be, argp))
 244                 return -EFAULT;
 245
 246         return 0;
 247 }
 248
 249 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 250 {
 251         int be;
 252
 253         if (get_user(be, argp))
 254                 return -EFAULT;
 255
 256         if (be)
 257                 tun->flags |= TUN_VNET_BE;
 258         else
 259                 tun->flags &= ~TUN_VNET_BE;
 260
 261         return 0;
 262 }
 263 #else
 264 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 265 {
 266         return virtio_legacy_is_little_endian();
 267 }
 268
 269 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 270 {
 271         return -EINVAL;
 272 }
 273
 274 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 275 {
 276         return -EINVAL;
 277 }
 278 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 279
 280 static inline bool tun_is_little_endian(struct tun_struct *tun)
 281 {
 282         return tun->flags & TUN_VNET_LE ||
 283                 tun_legacy_is_little_endian(tun);
 284 }
 285
 286 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 287 {
 288         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 289 }
 290
 291 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 292 {
 293         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 294 }
 295
 296 static inline u32 tun_hashfn(u32 rxhash)
 297 {
 298         return rxhash & 0x3ff;
 299 }
 300
 301 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 302 {
 303         struct tun_flow_entry *e;
 304
 305         hlist_for_each_entry_rcu(e, head, hash_link) {
 306                 if (e->rxhash == rxhash)
 307                         return e;
 308         }
 309         return NULL;
 310 }
 311
 312 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 313                                               struct hlist_head *head,
 314                                               u32 rxhash, u16 queue_index)
 315 {
 316         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 317
 318         if (e) {
 319                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 320                           rxhash, queue_index);
 321                 e->updated = jiffies;
 322                 e->rxhash = rxhash;
 323                 e->rps_rxhash = 0;
 324                 e->queue_index = queue_index;
 325                 e->tun = tun;
 326                 hlist_add_head_rcu(&e->hash_link, head);
 327                 ++tun->flow_count;
 328         }
 329         return e;
 330 }
 331
 332 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 333 {
 334         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 335                   e->rxhash, e->queue_index);
 336         hlist_del_rcu(&e->hash_link);
 337         kfree_rcu(e, rcu);
 338         --tun->flow_count;
 339 }
 340
 341 static void tun_flow_flush(struct tun_struct *tun)
 342 {
 343         int i;
 344
 345         spin_lock_bh(&tun->lock);
 346         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 347                 struct tun_flow_entry *e;
 348                 struct hlist_node *n;
 349
 350                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 351                         tun_flow_delete(tun, e);
 352         }
 353         spin_unlock_bh(&tun->lock);
 354 }
 355
 356 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 357 {
 358         int i;
 359
 360         spin_lock_bh(&tun->lock);
 361         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 362                 struct tun_flow_entry *e;
 363                 struct hlist_node *n;
 364
 365                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 366                         if (e->queue_index == queue_index)
 367                                 tun_flow_delete(tun, e);
 368                 }
 369         }
 370         spin_unlock_bh(&tun->lock);
 371 }
 372
 373 static void tun_flow_cleanup(unsigned long data)
 374 {
 375         struct tun_struct *tun = (struct tun_struct *)data;
 376         unsigned long delay = tun->ageing_time;
 377         unsigned long next_timer = jiffies + delay;
 378         unsigned long count = 0;
 379         int i;
 380
 381         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 382
 383         spin_lock_bh(&tun->lock);
 384         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 385                 struct tun_flow_entry *e;
 386                 struct hlist_node *n;
 387
 388                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 389                         unsigned long this_timer;
 390                         count++;
 391                         this_timer = e->updated + delay;
 392                         if (time_before_eq(this_timer, jiffies))
 393                                 tun_flow_delete(tun, e);
 394                         else if (time_before(this_timer, next_timer))
 395                                 next_timer = this_timer;
 396                 }
 397         }
 398
 399         if (count)
 400                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 401         spin_unlock_bh(&tun->lock);
 402 }
 403
 404 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 405                             struct tun_file *tfile)
 406 {
 407         struct hlist_head *head;
 408         struct tun_flow_entry *e;
 409         unsigned long delay = tun->ageing_time;
 410         u16 queue_index = tfile->queue_index;
 411
 412         if (!rxhash)
 413                 return;
 414         else
 415                 head = &tun->flows[tun_hashfn(rxhash)];
 416
 417         rcu_read_lock();
 418
 419         /* We may get a very small possibility of OOO during switching, not
 420          * worth to optimize.*/
 421         if (tun->numqueues == 1 || tfile->detached)
 422                 goto unlock;
 423
 424         e = tun_flow_find(head, rxhash);
 425         if (likely(e)) {
 426                 /* TODO: keep queueing to old queue until it's empty? */
 427                 e->queue_index = queue_index;
 428                 e->updated = jiffies;
 429                 sock_rps_record_flow_hash(e->rps_rxhash);
 430         } else {
 431                 spin_lock_bh(&tun->lock);
 432                 if (!tun_flow_find(head, rxhash) &&
 433                     tun->flow_count < MAX_TAP_FLOWS)
 434                         tun_flow_create(tun, head, rxhash, queue_index);
 435
 436                 if (!timer_pending(&tun->flow_gc_timer))
 437                         mod_timer(&tun->flow_gc_timer,
 438                                   round_jiffies_up(jiffies + delay));
 439                 spin_unlock_bh(&tun->lock);
 440         }
 441
 442 unlock:
 443         rcu_read_unlock();
 444 }
 445
 446 /**
 447  * Save the hash received in the stack receive path and update the
 448  * flow_hash table accordingly.
 449  */
 450 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 451 {
 452         if (unlikely(e->rps_rxhash != hash))
 453                 e->rps_rxhash = hash;
 454 }
 455
 456 /* We try to identify a flow through its rxhash first. The reason that
 457  * we do not check rxq no. is because some cards(e.g 82599), chooses
 458  * the rxq based on the txq where the last packet of the flow comes. As
 459  * the userspace application move between processors, we may get a
 460  * different rxq no. here. If we could not get rxhash, then we would
 461  * hope the rxq no. may help here.
 462  */
 463 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 464                             void *accel_priv, select_queue_fallback_t fallback)
 465 {
 466         struct tun_struct *tun = netdev_priv(dev);
 467         struct tun_flow_entry *e;
 468         u32 txq = 0;
 469         u32 numqueues = 0;
 470
 471         rcu_read_lock();
 472         numqueues = ACCESS_ONCE(tun->numqueues);
 473
 474         txq = skb_get_hash(skb);
 475         if (txq) {
 476                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 477                 if (e) {
 478                         tun_flow_save_rps_rxhash(e, txq);
 479                         txq = e->queue_index;
 480                 } else
 481                         /* use multiply and shift instead of expensive divide */
 482                         txq = ((u64)txq * numqueues) >> 32;
 483         } else if (likely(skb_rx_queue_recorded(skb))) {
 484                 txq = skb_get_rx_queue(skb);
 485                 while (unlikely(txq >= numqueues))
 486                         txq -= numqueues;
 487         }
 488
 489         rcu_read_unlock();
 490         return txq;
 491 }
 492
 493 static inline bool tun_not_capable(struct tun_struct *tun)
 494 {
 495         const struct cred *cred = current_cred();
 496         struct net *net = dev_net(tun->dev);
 497
 498         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 499                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 500                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 501 }
 502
 503 static void tun_set_real_num_queues(struct tun_struct *tun)
 504 {
 505         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 506         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 507 }
 508
 509 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 510 {
 511         tfile->detached = tun;
 512         list_add_tail(&tfile->next, &tun->disabled);
 513         ++tun->numdisabled;
 514 }
 515
 516 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 517 {
 518         struct tun_struct *tun = tfile->detached;
 519
 520         tfile->detached = NULL;
 521         list_del_init(&tfile->next);
 522         --tun->numdisabled;
 523         return tun;
 524 }
 525
 526 static void tun_queue_purge(struct tun_file *tfile)
 527 {
 528         struct sk_buff *skb;
 529
 530         while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
 531                 kfree_skb(skb);
 532
 533         skb_queue_purge(&tfile->sk.sk_error_queue);
 534 }
 535
 536 static void tun_cleanup_tx_array(struct tun_file *tfile)
 537 {
 538         if (tfile->tx_array.ring.queue) {
 539                 skb_array_cleanup(&tfile->tx_array);
 540                 memset(&tfile->tx_array, 0, sizeof(tfile->tx_array));
 541         }
 542 }
 543
 544 static void __tun_detach(struct tun_file *tfile, bool clean)
 545 {
 546         struct tun_file *ntfile;
 547         struct tun_struct *tun;
 548
 549         tun = rtnl_dereference(tfile->tun);
 550
 551         if (tun && !tfile->detached) {
 552                 u16 index = tfile->queue_index;
 553                 BUG_ON(index >= tun->numqueues);
 554
 555                 rcu_assign_pointer(tun->tfiles[index],
 556                                    tun->tfiles[tun->numqueues - 1]);
 557                 ntfile = rtnl_dereference(tun->tfiles[index]);
 558                 ntfile->queue_index = index;
 559
 560                 --tun->numqueues;
 561                 if (clean) {
 562                         RCU_INIT_POINTER(tfile->tun, NULL);
 563                         sock_put(&tfile->sk);
 564                 } else
 565                         tun_disable_queue(tun, tfile);
 566
 567                 synchronize_net();
 568                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 569                 /* Drop read queue */
 570                 tun_queue_purge(tfile);
 571                 tun_set_real_num_queues(tun);
 572         } else if (tfile->detached && clean) {
 573                 tun = tun_enable_queue(tfile);
 574                 sock_put(&tfile->sk);
 575         }
 576
 577         if (clean) {
 578                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 579                         netif_carrier_off(tun->dev);
 580
 581                         if (!(tun->flags & IFF_PERSIST) &&
 582                             tun->dev->reg_state == NETREG_REGISTERED)
 583                                 unregister_netdevice(tun->dev);
 584                 }
 585                 tun_cleanup_tx_array(tfile);
 586                 sock_put(&tfile->sk);
 587         }
 588 }
 589
 590 static void tun_detach(struct tun_file *tfile, bool clean)
 591 {
 592         rtnl_lock();
 593         __tun_detach(tfile, clean);
 594         rtnl_unlock();
 595 }
 596
 597 static void tun_detach_all(struct net_device *dev)
 598 {
 599         struct tun_struct *tun = netdev_priv(dev);
 600         struct tun_file *tfile, *tmp;
 601         int i, n = tun->numqueues;
 602
 603         for (i = 0; i < n; i++) {
 604                 tfile = rtnl_dereference(tun->tfiles[i]);
 605                 BUG_ON(!tfile);
 606                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 607                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 608                 RCU_INIT_POINTER(tfile->tun, NULL);
 609                 --tun->numqueues;
 610         }
 611         list_for_each_entry(tfile, &tun->disabled, next) {
 612                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 613                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 614                 RCU_INIT_POINTER(tfile->tun, NULL);
 615         }
 616         BUG_ON(tun->numqueues != 0);
 617
 618         synchronize_net();
 619         for (i = 0; i < n; i++) {
 620                 tfile = rtnl_dereference(tun->tfiles[i]);
 621                 /* Drop read queue */
 622                 tun_queue_purge(tfile);
 623                 sock_put(&tfile->sk);
 624                 tun_cleanup_tx_array(tfile);
 625         }
 626         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 627                 tun_enable_queue(tfile);
 628                 tun_queue_purge(tfile);
 629                 sock_put(&tfile->sk);
 630                 tun_cleanup_tx_array(tfile);
 631         }
 632         BUG_ON(tun->numdisabled != 0);
 633
 634         if (tun->flags & IFF_PERSIST)
 635                 module_put(THIS_MODULE);
 636 }
 637
 638 static int tun_attach(struct tun_struct *tun, struct file *file,
 639                       bool skip_filter, bool publish_tun)
 640 {
 641         struct tun_file *tfile = file->private_data;
 642         struct net_device *dev = tun->dev;
 643         int err;
 644
 645         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 646         if (err < 0)
 647                 goto out;
 648
 649         err = -EINVAL;
 650         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 651                 goto out;
 652
 653         err = -EBUSY;
 654         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 655                 goto out;
 656
 657         err = -E2BIG;
 658         if (!tfile->detached &&
 659             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 660                 goto out;
 661
 662         err = 0;
 663
 664         /* Re-attach the filter to persist device */
 665         if (!skip_filter && (tun->filter_attached == true)) {
 666                 lock_sock(tfile->socket.sk);
 667                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 668                 release_sock(tfile->socket.sk);
 669                 if (!err)
 670                         goto out;
 671         }
 672
 673         if (!tfile->detached &&
 674             skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
 675                 err = -ENOMEM;
 676                 goto out;
 677         }
 678
 679         tfile->queue_index = tun->numqueues;
 680         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 681         if (publish_tun)
 682                 rcu_assign_pointer(tfile->tun, tun);
 683         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 684         tun->numqueues++;
 685
 686         if (tfile->detached)
 687                 tun_enable_queue(tfile);
 688         else
 689                 sock_hold(&tfile->sk);
 690
 691         tun_set_real_num_queues(tun);
 692
 693         /* device is allowed to go away first, so no need to hold extra
 694          * refcnt.
 695          */
 696
 697 out:
 698         return err;
 699 }
 700
 701 static struct tun_struct *__tun_get(struct tun_file *tfile)
 702 {
 703         struct tun_struct *tun;
 704
 705         rcu_read_lock();
 706         tun = rcu_dereference(tfile->tun);
 707         if (tun)
 708                 dev_hold(tun->dev);
 709         rcu_read_unlock();
 710
 711         return tun;
 712 }
 713
 714 static struct tun_struct *tun_get(struct file *file)
 715 {
 716         return __tun_get(file->private_data);
 717 }
 718
 719 static void tun_put(struct tun_struct *tun)
 720 {
 721         dev_put(tun->dev);
 722 }
 723
 724 /* TAP filtering */
 725 static void addr_hash_set(u32 *mask, const u8 *addr)
 726 {
 727         int n = ether_crc(ETH_ALEN, addr) >> 26;
 728         mask[n >> 5] |= (1 << (n & 31));
 729 }
 730
 731 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 732 {
 733         int n = ether_crc(ETH_ALEN, addr) >> 26;
 734         return mask[n >> 5] & (1 << (n & 31));
 735 }
 736
 737 static int update_filter(struct tap_filter *filter, void __user *arg)
 738 {
 739         struct { u8 u[ETH_ALEN]; } *addr;
 740         struct tun_filter uf;
 741         int err, alen, n, nexact;
 742
 743         if (copy_from_user(&uf, arg, sizeof(uf)))
 744                 return -EFAULT;
 745
 746         if (!uf.count) {
 747                 /* Disabled */
 748                 filter->count = 0;
 749                 return 0;
 750         }
 751
 752         alen = ETH_ALEN * uf.count;
 753         addr = memdup_user(arg + sizeof(uf), alen);
 754         if (IS_ERR(addr))
 755                 return PTR_ERR(addr);
 756
 757         /* The filter is updated without holding any locks. Which is
 758          * perfectly safe. We disable it first and in the worst
 759          * case we'll accept a few undesired packets. */
 760         filter->count = 0;
 761         wmb();
 762
 763         /* Use first set of addresses as an exact filter */
 764         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 765                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 766
 767         nexact = n;
 768
 769         /* Remaining multicast addresses are hashed,
 770          * unicast will leave the filter disabled. */
 771         memset(filter->mask, 0, sizeof(filter->mask));
 772         for (; n < uf.count; n++) {
 773                 if (!is_multicast_ether_addr(addr[n].u)) {
 774                         err = 0; /* no filter */
 775                         goto free_addr;
 776                 }
 777                 addr_hash_set(filter->mask, addr[n].u);
 778         }
 779
 780         /* For ALLMULTI just set the mask to all ones.
 781          * This overrides the mask populated above. */
 782         if ((uf.flags & TUN_FLT_ALLMULTI))
 783                 memset(filter->mask, ~0, sizeof(filter->mask));
 784
 785         /* Now enable the filter */
 786         wmb();
 787         filter->count = nexact;
 788
 789         /* Return the number of exact filters */
 790         err = nexact;
 791 free_addr:
 792         kfree(addr);
 793         return err;
 794 }
 795
 796 /* Returns: 0 - drop, !=0 - accept */
 797 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 798 {
 799         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 800          * at this point. */
 801         struct ethhdr *eh = (struct ethhdr *) skb->data;
 802         int i;
 803
 804         /* Exact match */
 805         for (i = 0; i < filter->count; i++)
 806                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 807                         return 1;
 808
 809         /* Inexact match (multicast only) */
 810         if (is_multicast_ether_addr(eh->h_dest))
 811                 return addr_hash_test(filter->mask, eh->h_dest);
 812
 813         return 0;
 814 }
 815
 816 /*
 817  * Checks whether the packet is accepted or not.
 818  * Returns: 0 - drop, !=0 - accept
 819  */
 820 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 821 {
 822         if (!filter->count)
 823                 return 1;
 824
 825         return run_filter(filter, skb);
 826 }
 827
 828 /* Network device part of the driver */
 829
 830 static const struct ethtool_ops tun_ethtool_ops;
 831
 832 /* Net device detach from fd. */
 833 static void tun_net_uninit(struct net_device *dev)
 834 {
 835         tun_detach_all(dev);
 836 }
 837
 838 /* Net device open. */
 839 static int tun_net_open(struct net_device *dev)
 840 {
 841         netif_tx_start_all_queues(dev);
 842
 843         return 0;
 844 }
 845
 846 /* Net device close. */
 847 static int tun_net_close(struct net_device *dev)
 848 {
 849         netif_tx_stop_all_queues(dev);
 850         return 0;
 851 }
 852
 853 /* Net device start xmit */
 854 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 855 {
 856         struct tun_struct *tun = netdev_priv(dev);
 857         int txq = skb->queue_mapping;
 858         struct netdev_queue *queue;
 859         struct tun_file *tfile;
 860         u32 numqueues = 0;
 861
 862         rcu_read_lock();
 863         tfile = rcu_dereference(tun->tfiles[txq]);
 864         numqueues = ACCESS_ONCE(tun->numqueues);
 865
 866         /* Drop packet if interface is not attached */
 867         if (txq >= numqueues)
 868                 goto drop;
 869
 870 #ifdef CONFIG_RPS
 871         if (numqueues == 1 && static_key_false(&rps_needed)) {
 872                 /* Select queue was not called for the skbuff, so we extract the
 873                  * RPS hash and save it into the flow_table here.
 874                  */
 875                 __u32 rxhash;
 876
 877                 rxhash = skb_get_hash(skb);
 878                 if (rxhash) {
 879                         struct tun_flow_entry *e;
 880                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 881                                         rxhash);
 882                         if (e)
 883                                 tun_flow_save_rps_rxhash(e, rxhash);
 884                 }
 885         }
 886 #endif
 887
 888         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 889
 890         BUG_ON(!tfile);
 891
 892         /* Drop if the filter does not like it.
 893          * This is a noop if the filter is disabled.
 894          * Filter can be enabled only for the TAP devices. */
 895         if (!check_filter(&tun->txflt, skb))
 896                 goto drop;
 897
 898         if (tfile->socket.sk->sk_filter &&
 899             sk_filter(tfile->socket.sk, skb))
 900                 goto drop;
 901
 902         /* Limit the number of packets queued by dividing txq length with the
 903          * number of queues.
 904          */
 905         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
 906                           >= dev->tx_queue_len)
 907                 goto drop;
 908
 909         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 910                 goto drop;
 911
 912         skb_tx_timestamp(skb);
 913
 914         /* Orphan the skb - required as we might hang on to it
 915          * for indefinite time.
 916          */
 917         skb_orphan(skb);
 918
 919         nf_reset(skb);
 920
 921         if (skb_array_produce(&tfile->tx_array, skb))
 922                 goto drop;
 923
 924         /* NETIF_F_LLTX requires to do our own update of trans_start */
 925         queue = netdev_get_tx_queue(dev, txq);
 926         queue->trans_start = jiffies;
 927
 928         /* Notify and wake up reader process */
 929         if (tfile->flags & TUN_FASYNC)
 930                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 931         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 932
 933         rcu_read_unlock();
 934         return NETDEV_TX_OK;
 935
 936 drop:
 937         this_cpu_inc(tun->pcpu_stats->tx_dropped);
 938         skb_tx_error(skb);
 939         kfree_skb(skb);
 940         rcu_read_unlock();
 941         return NET_XMIT_DROP;
 942 }
 943
 944 static void tun_net_mclist(struct net_device *dev)
 945 {
 946         /*
 947          * This callback is supposed to deal with mc filter in
 948          * _rx_ path and has nothing to do with the _tx_ path.
 949          * In rx path we always accept everything userspace gives us.
 950          */
 951 }
 952
 953 #define MIN_MTU 68
 954 #define MAX_MTU 65535
 955
 956 static int
 957 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 958 {
 959         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 960                 return -EINVAL;
 961         dev->mtu = new_mtu;
 962         return 0;
 963 }
 964
 965 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 966         netdev_features_t features)
 967 {
 968         struct tun_struct *tun = netdev_priv(dev);
 969
 970         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 971 }
 972 #ifdef CONFIG_NET_POLL_CONTROLLER
 973 static void tun_poll_controller(struct net_device *dev)
 974 {
 975         /*
 976          * Tun only receives frames when:
 977          * 1) the char device endpoint gets data from user space
 978          * 2) the tun socket gets a sendmsg call from user space
 979          * Since both of those are synchronous operations, we are guaranteed
 980          * never to have pending data when we poll for it
 981          * so there is nothing to do here but return.
 982          * We need this though so netpoll recognizes us as an interface that
 983          * supports polling, which enables bridge devices in virt setups to
 984          * still use netconsole
 985          */
 986         return;
 987 }
 988 #endif
 989
 990 static void tun_set_headroom(struct net_device *dev, int new_hr)
 991 {
 992         struct tun_struct *tun = netdev_priv(dev);
 993
 994         if (new_hr < NET_SKB_PAD)
 995                 new_hr = NET_SKB_PAD;
 996
 997         tun->align = new_hr;
 998 }
 999
1000 static struct rtnl_link_stats64 *
1001 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1002 {
1003         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1004         struct tun_struct *tun = netdev_priv(dev);
1005         struct tun_pcpu_stats *p;
1006         int i;
1007
1008         for_each_possible_cpu(i) {
1009                 u64 rxpackets, rxbytes, txpackets, txbytes;
1010                 unsigned int start;
1011
1012                 p = per_cpu_ptr(tun->pcpu_stats, i);
1013                 do {
1014                         start = u64_stats_fetch_begin(&p->syncp);
1015                         rxpackets       = p->rx_packets;
1016                         rxbytes         = p->rx_bytes;
1017                         txpackets       = p->tx_packets;
1018                         txbytes         = p->tx_bytes;
1019                 } while (u64_stats_fetch_retry(&p->syncp, start));
1020
1021                 stats->rx_packets       += rxpackets;
1022                 stats->rx_bytes         += rxbytes;
1023                 stats->tx_packets       += txpackets;
1024                 stats->tx_bytes         += txbytes;
1025
1026                 /* u32 counters */
1027                 rx_dropped      += p->rx_dropped;
1028                 rx_frame_errors += p->rx_frame_errors;
1029                 tx_dropped      += p->tx_dropped;
1030         }
1031         stats->rx_dropped  = rx_dropped;
1032         stats->rx_frame_errors = rx_frame_errors;
1033         stats->tx_dropped = tx_dropped;
1034         return stats;
1035 }
1036
1037 static const struct net_device_ops tun_netdev_ops = {
1038         .ndo_uninit             = tun_net_uninit,
1039         .ndo_open               = tun_net_open,
1040         .ndo_stop               = tun_net_close,
1041         .ndo_start_xmit         = tun_net_xmit,
1042         .ndo_change_mtu         = tun_net_change_mtu,
1043         .ndo_fix_features       = tun_net_fix_features,
1044         .ndo_select_queue       = tun_select_queue,
1045 #ifdef CONFIG_NET_POLL_CONTROLLER
1046         .ndo_poll_controller    = tun_poll_controller,
1047 #endif
1048         .ndo_set_rx_headroom    = tun_set_headroom,
1049         .ndo_get_stats64        = tun_net_get_stats64,
1050 };
1051
1052 static const struct net_device_ops tap_netdev_ops = {
1053         .ndo_uninit             = tun_net_uninit,
1054         .ndo_open               = tun_net_open,
1055         .ndo_stop               = tun_net_close,
1056         .ndo_start_xmit         = tun_net_xmit,
1057         .ndo_change_mtu         = tun_net_change_mtu,
1058         .ndo_fix_features       = tun_net_fix_features,
1059         .ndo_set_rx_mode        = tun_net_mclist,
1060         .ndo_set_mac_address    = eth_mac_addr,
1061         .ndo_validate_addr      = eth_validate_addr,
1062         .ndo_select_queue       = tun_select_queue,
1063 #ifdef CONFIG_NET_POLL_CONTROLLER
1064         .ndo_poll_controller    = tun_poll_controller,
1065 #endif
1066         .ndo_features_check     = passthru_features_check,
1067         .ndo_set_rx_headroom    = tun_set_headroom,
1068         .ndo_get_stats64        = tun_net_get_stats64,
1069 };
1070
1071 static void tun_flow_init(struct tun_struct *tun)
1072 {
1073         int i;
1074
1075         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1076                 INIT_HLIST_HEAD(&tun->flows[i]);
1077
1078         tun->ageing_time = TUN_FLOW_EXPIRE;
1079         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1080         mod_timer(&tun->flow_gc_timer,
1081                   round_jiffies_up(jiffies + tun->ageing_time));
1082 }
1083
1084 static void tun_flow_uninit(struct tun_struct *tun)
1085 {
1086         del_timer_sync(&tun->flow_gc_timer);
1087         tun_flow_flush(tun);
1088 }
1089
1090 /* Initialize net device. */
1091 static void tun_net_init(struct net_device *dev)
1092 {
1093         struct tun_struct *tun = netdev_priv(dev);
1094
1095         switch (tun->flags & TUN_TYPE_MASK) {
1096         case IFF_TUN:
1097                 dev->netdev_ops = &tun_netdev_ops;
1098
1099                 /* Point-to-Point TUN Device */
1100                 dev->hard_header_len = 0;
1101                 dev->addr_len = 0;
1102                 dev->mtu = 1500;
1103
1104                 /* Zero header length */
1105                 dev->type = ARPHRD_NONE;
1106                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1107                 break;
1108
1109         case IFF_TAP:
1110                 dev->netdev_ops = &tap_netdev_ops;
1111                 /* Ethernet TAP Device */
1112                 ether_setup(dev);
1113                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1114                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1115
1116                 eth_hw_addr_random(dev);
1117
1118                 break;
1119         }
1120 }
1121
1122 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1123 {
1124         struct sock *sk = tfile->socket.sk;
1125
1126         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1127 }
1128
1129 /* Character device part */
1130
1131 /* Poll */
1132 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1133 {
1134         struct tun_file *tfile = file->private_data;
1135         struct tun_struct *tun = __tun_get(tfile);
1136         struct sock *sk;
1137         unsigned int mask = 0;
1138
1139         if (!tun)
1140                 return POLLERR;
1141
1142         sk = tfile->socket.sk;
1143
1144         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1145
1146         poll_wait(file, sk_sleep(sk), wait);
1147
1148         if (!skb_array_empty(&tfile->tx_array))
1149                 mask |= POLLIN | POLLRDNORM;
1150
1151         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1152          * guarantee EPOLLOUT to be raised by either here or
1153          * tun_sock_write_space(). Then process could get notification
1154          * after it writes to a down device and meets -EIO.
1155          */
1156         if (tun_sock_writeable(tun, tfile) ||
1157             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1158              tun_sock_writeable(tun, tfile)))
1159                 mask |= POLLOUT | POLLWRNORM;
1160
1161         if (tun->dev->reg_state != NETREG_REGISTERED)
1162                 mask = POLLERR;
1163
1164         tun_put(tun);
1165         return mask;
1166 }
1167
1168 /* prepad is the amount to reserve at front.  len is length after that.
1169  * linear is a hint as to how much to copy (usually headers). */
1170 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1171                                      size_t prepad, size_t len,
1172                                      size_t linear, int noblock)
1173 {
1174         struct sock *sk = tfile->socket.sk;
1175         struct sk_buff *skb;
1176         int err;
1177
1178         /* Under a page?  Don't bother with paged skb. */
1179         if (prepad + len < PAGE_SIZE || !linear)
1180                 linear = len;
1181
1182         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1183                                    &err, 0);
1184         if (!skb)
1185                 return ERR_PTR(err);
1186
1187         skb_reserve(skb, prepad);
1188         skb_put(skb, linear);
1189         skb->data_len = len - linear;
1190         skb->len += len - linear;
1191
1192         return skb;
1193 }
1194
1195 /* Get packet from user space buffer */
1196 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1197                             void *msg_control, struct iov_iter *from,
1198                             int noblock)
1199 {
1200         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1201         struct sk_buff *skb;
1202         size_t total_len = iov_iter_count(from);
1203         size_t len = total_len, align = tun->align, linear;
1204         struct virtio_net_hdr gso = { 0 };
1205         struct tun_pcpu_stats *stats;
1206         int good_linear;
1207         int copylen;
1208         bool zerocopy = false;
1209         int err;
1210         u32 rxhash;
1211         ssize_t n;
1212
1213         if (!(tun->flags & IFF_NO_PI)) {
1214                 if (len < sizeof(pi))
1215                         return -EINVAL;
1216                 len -= sizeof(pi);
1217
1218                 n = copy_from_iter(&pi, sizeof(pi), from);
1219                 if (n != sizeof(pi))
1220                         return -EFAULT;
1221         }
1222
1223         if (tun->flags & IFF_VNET_HDR) {
1224                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1225
1226                 if (len < vnet_hdr_sz)
1227                         return -EINVAL;
1228                 len -= vnet_hdr_sz;
1229
1230                 n = copy_from_iter(&gso, sizeof(gso), from);
1231                 if (n != sizeof(gso))
1232                         return -EFAULT;
1233
1234                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1235                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1236                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1237
1238                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1239                         return -EINVAL;
1240                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1241         }
1242
1243         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1244                 align += NET_IP_ALIGN;
1245                 if (unlikely(len < ETH_HLEN ||
1246                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1247                         return -EINVAL;
1248         }
1249
1250         good_linear = SKB_MAX_HEAD(align);
1251
1252         if (msg_control) {
1253                 struct iov_iter i = *from;
1254
1255                 /* There are 256 bytes to be copied in skb, so there is
1256                  * enough room for skb expand head in case it is used.
1257                  * The rest of the buffer is mapped from userspace.
1258                  */
1259                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1260                 if (copylen > good_linear)
1261                         copylen = good_linear;
1262                 linear = copylen;
1263                 iov_iter_advance(&i, copylen);
1264                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1265                         zerocopy = true;
1266         }
1267
1268         if (!zerocopy) {
1269                 copylen = len;
1270                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1271                         linear = good_linear;
1272                 else
1273                         linear = tun16_to_cpu(tun, gso.hdr_len);
1274         }
1275
1276         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1277         if (IS_ERR(skb)) {
1278                 if (PTR_ERR(skb) != -EAGAIN)
1279                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1280                 return PTR_ERR(skb);
1281         }
1282
1283         if (zerocopy)
1284                 err = zerocopy_sg_from_iter(skb, from);
1285         else
1286                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1287
1288         if (err) {
1289                 err = -EFAULT;
1290 drop:
1291                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1292                 kfree_skb(skb);
1293                 return err;
1294         }
1295
1296         err = virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun));
1297         if (err) {
1298                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1299                 kfree_skb(skb);
1300                 return -EINVAL;
1301         }
1302
1303         switch (tun->flags & TUN_TYPE_MASK) {
1304         case IFF_TUN:
1305                 if (tun->flags & IFF_NO_PI) {
1306                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1307
1308                         switch (ip_version) {
1309                         case 4:
1310                                 pi.proto = htons(ETH_P_IP);
1311                                 break;
1312                         case 6:
1313                                 pi.proto = htons(ETH_P_IPV6);
1314                                 break;
1315                         default:
1316                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1317                                 kfree_skb(skb);
1318                                 return -EINVAL;
1319                         }
1320                 }
1321
1322                 skb_reset_mac_header(skb);
1323                 skb->protocol = pi.proto;
1324                 skb->dev = tun->dev;
1325                 break;
1326         case IFF_TAP:
1327                 skb->protocol = eth_type_trans(skb, tun->dev);
1328                 break;
1329         }
1330
1331         /* copy skb_ubuf_info for callback when skb has no error */
1332         if (zerocopy) {
1333                 skb_shinfo(skb)->destructor_arg = msg_control;
1334                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1335                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1336         } else if (msg_control) {
1337                 struct ubuf_info *uarg = msg_control;
1338                 uarg->callback(uarg, false);
1339         }
1340
1341         skb_reset_network_header(skb);
1342         skb_probe_transport_header(skb, 0);
1343
1344         rxhash = skb_get_hash(skb);
1345
1346         rcu_read_lock();
1347         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1348                 err = -EIO;
1349                 rcu_read_unlock();
1350                 goto drop;
1351         }
1352
1353         netif_rx_ni(skb);
1354         rcu_read_unlock();
1355
1356         stats = get_cpu_ptr(tun->pcpu_stats);
1357         u64_stats_update_begin(&stats->syncp);
1358         stats->rx_packets++;
1359         stats->rx_bytes += len;
1360         u64_stats_update_end(&stats->syncp);
1361         put_cpu_ptr(stats);
1362
1363         tun_flow_update(tun, rxhash, tfile);
1364         return total_len;
1365 }
1366
1367 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1368 {
1369         struct file *file = iocb->ki_filp;
1370         struct tun_struct *tun = tun_get(file);
1371         struct tun_file *tfile = file->private_data;
1372         ssize_t result;
1373
1374         if (!tun)
1375                 return -EBADFD;
1376
1377         result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1378
1379         tun_put(tun);
1380         return result;
1381 }
1382
1383 /* Put packet to the user space buffer */
1384 static ssize_t tun_put_user(struct tun_struct *tun,
1385                             struct tun_file *tfile,
1386                             struct sk_buff *skb,
1387                             struct iov_iter *iter)
1388 {
1389         struct tun_pi pi = { 0, skb->protocol };
1390         struct tun_pcpu_stats *stats;
1391         ssize_t total;
1392         int vlan_offset = 0;
1393         int vlan_hlen = 0;
1394         int vnet_hdr_sz = 0;
1395
1396         if (skb_vlan_tag_present(skb))
1397                 vlan_hlen = VLAN_HLEN;
1398
1399         if (tun->flags & IFF_VNET_HDR)
1400                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1401
1402         total = skb->len + vlan_hlen + vnet_hdr_sz;
1403
1404         if (!(tun->flags & IFF_NO_PI)) {
1405                 if (iov_iter_count(iter) < sizeof(pi))
1406                         return -EINVAL;
1407
1408                 total += sizeof(pi);
1409                 if (iov_iter_count(iter) < total) {
1410                         /* Packet will be striped */
1411                         pi.flags |= TUN_PKT_STRIP;
1412                 }
1413
1414                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1415                         return -EFAULT;
1416         }
1417
1418         if (vnet_hdr_sz) {
1419                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1420                 int ret;
1421
1422                 if (iov_iter_count(iter) < vnet_hdr_sz)
1423                         return -EINVAL;
1424
1425                 ret = virtio_net_hdr_from_skb(skb, &gso,
1426                                               tun_is_little_endian(tun), true);
1427                 if (ret) {
1428                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1429                         pr_err("unexpected GSO type: "
1430                                "0x%x, gso_size %d, hdr_len %d\n",
1431                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1432                                tun16_to_cpu(tun, gso.hdr_len));
1433                         print_hex_dump(KERN_ERR, "tun: ",
1434                                        DUMP_PREFIX_NONE,
1435                                        16, 1, skb->head,
1436                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1437                         WARN_ON_ONCE(1);
1438                         return -EINVAL;
1439                 }
1440
1441                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1442                         return -EFAULT;
1443
1444                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1445         }
1446
1447         if (vlan_hlen) {
1448                 int ret;
1449                 struct {
1450                         __be16 h_vlan_proto;
1451                         __be16 h_vlan_TCI;
1452                 } veth;
1453
1454                 veth.h_vlan_proto = skb->vlan_proto;
1455                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1456
1457                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1458
1459                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1460                 if (ret || !iov_iter_count(iter))
1461                         goto done;
1462
1463                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1464                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1465                         goto done;
1466         }
1467
1468         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1469
1470 done:
1471         /* caller is in process context, */
1472         stats = get_cpu_ptr(tun->pcpu_stats);
1473         u64_stats_update_begin(&stats->syncp);
1474         stats->tx_packets++;
1475         stats->tx_bytes += skb->len + vlan_hlen;
1476         u64_stats_update_end(&stats->syncp);
1477         put_cpu_ptr(tun->pcpu_stats);
1478
1479         return total;
1480 }
1481
1482 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1483                                      int *err)
1484 {
1485         DECLARE_WAITQUEUE(wait, current);
1486         struct sk_buff *skb = NULL;
1487         int error = 0;
1488
1489         skb = skb_array_consume(&tfile->tx_array);
1490         if (skb)
1491                 goto out;
1492         if (noblock) {
1493                 error = -EAGAIN;
1494                 goto out;
1495         }
1496
1497         add_wait_queue(&tfile->wq.wait, &wait);
1498
1499         while (1) {
1500                 set_current_state(TASK_INTERRUPTIBLE);
1501                 skb = skb_array_consume(&tfile->tx_array);
1502                 if (skb)
1503                         break;
1504                 if (signal_pending(current)) {
1505                         error = -ERESTARTSYS;
1506                         break;
1507                 }
1508                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1509                         error = -EFAULT;
1510                         break;
1511                 }
1512
1513                 schedule();
1514         }
1515
1516         __set_current_state(TASK_RUNNING);
1517         remove_wait_queue(&tfile->wq.wait, &wait);
1518
1519 out:
1520         *err = error;
1521         return skb;
1522 }
1523
1524 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1525                            struct iov_iter *to,
1526                            int noblock)
1527 {
1528         struct sk_buff *skb;
1529         ssize_t ret;
1530         int err;
1531
1532         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1533
1534         if (!iov_iter_count(to))
1535                 return 0;
1536
1537         /* Read frames from ring */
1538         skb = tun_ring_recv(tfile, noblock, &err);
1539         if (!skb)
1540                 return err;
1541
1542         ret = tun_put_user(tun, tfile, skb, to);
1543         if (unlikely(ret < 0))
1544                 kfree_skb(skb);
1545         else
1546                 consume_skb(skb);
1547
1548         return ret;
1549 }
1550
1551 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1552 {
1553         struct file *file = iocb->ki_filp;
1554         struct tun_file *tfile = file->private_data;
1555         struct tun_struct *tun = __tun_get(tfile);
1556         ssize_t len = iov_iter_count(to), ret;
1557
1558         if (!tun)
1559                 return -EBADFD;
1560         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1561         ret = min_t(ssize_t, ret, len);
1562         if (ret > 0)
1563                 iocb->ki_pos = ret;
1564         tun_put(tun);
1565         return ret;
1566 }
1567
1568 static void tun_free_netdev(struct net_device *dev)
1569 {
1570         struct tun_struct *tun = netdev_priv(dev);
1571
1572         BUG_ON(!(list_empty(&tun->disabled)));
1573         free_percpu(tun->pcpu_stats);
1574         tun_flow_uninit(tun);
1575         security_tun_dev_free_security(tun->security);
1576         free_netdev(dev);
1577 }
1578
1579 static void tun_setup(struct net_device *dev)
1580 {
1581         struct tun_struct *tun = netdev_priv(dev);
1582
1583         tun->owner = INVALID_UID;
1584         tun->group = INVALID_GID;
1585
1586         dev->ethtool_ops = &tun_ethtool_ops;
1587         dev->destructor = tun_free_netdev;
1588         /* We prefer our own queue length */
1589         dev->tx_queue_len = TUN_READQ_SIZE;
1590 }
1591
1592 /* Trivial set of netlink ops to allow deleting tun or tap
1593  * device with netlink.
1594  */
1595 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1596 {
1597         /* NL_SET_ERR_MSG(extack,
1598                        "tun/tap creation via rtnetlink is not supported."); */
1599         return -EOPNOTSUPP;
1600 }
1601
1602 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1603         .kind           = DRV_NAME,
1604         .priv_size      = sizeof(struct tun_struct),
1605         .setup          = tun_setup,
1606         .validate       = tun_validate,
1607 };
1608
1609 static void tun_sock_write_space(struct sock *sk)
1610 {
1611         struct tun_file *tfile;
1612         wait_queue_head_t *wqueue;
1613
1614         if (!sock_writeable(sk))
1615                 return;
1616
1617         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1618                 return;
1619
1620         wqueue = sk_sleep(sk);
1621         if (wqueue && waitqueue_active(wqueue))
1622                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1623                                                 POLLWRNORM | POLLWRBAND);
1624
1625         tfile = container_of(sk, struct tun_file, sk);
1626         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1627 }
1628
1629 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1630 {
1631         int ret;
1632         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1633         struct tun_struct *tun = __tun_get(tfile);
1634
1635         if (!tun)
1636                 return -EBADFD;
1637
1638         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1639                            m->msg_flags & MSG_DONTWAIT);
1640         tun_put(tun);
1641         return ret;
1642 }
1643
1644 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1645                        int flags)
1646 {
1647         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1648         struct tun_struct *tun = __tun_get(tfile);
1649         int ret;
1650
1651         if (!tun)
1652                 return -EBADFD;
1653
1654         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1655                 ret = -EINVAL;
1656                 goto out;
1657         }
1658         if (flags & MSG_ERRQUEUE) {
1659                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1660                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1661                 goto out;
1662         }
1663         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1664         if (ret > (ssize_t)total_len) {
1665                 m->msg_flags |= MSG_TRUNC;
1666                 ret = flags & MSG_TRUNC ? ret : total_len;
1667         }
1668 out:
1669         tun_put(tun);
1670         return ret;
1671 }
1672
1673 static int tun_peek_len(struct socket *sock)
1674 {
1675         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1676         struct tun_struct *tun;
1677         int ret = 0;
1678
1679         tun = __tun_get(tfile);
1680         if (!tun)
1681                 return 0;
1682
1683         ret = skb_array_peek_len(&tfile->tx_array);
1684         tun_put(tun);
1685
1686         return ret;
1687 }
1688
1689 /* Ops structure to mimic raw sockets with tun */
1690 static const struct proto_ops tun_socket_ops = {
1691         .peek_len = tun_peek_len,
1692         .sendmsg = tun_sendmsg,
1693         .recvmsg = tun_recvmsg,
1694 };
1695
1696 static struct proto tun_proto = {
1697         .name           = "tun",
1698         .owner          = THIS_MODULE,
1699         .obj_size       = sizeof(struct tun_file),
1700 };
1701
1702 static int tun_flags(struct tun_struct *tun)
1703 {
1704         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1705 }
1706
1707 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1708                               char *buf)
1709 {
1710         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1711         return sprintf(buf, "0x%x\n", tun_flags(tun));
1712 }
1713
1714 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1715                               char *buf)
1716 {
1717         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1718         return uid_valid(tun->owner)?
1719                 sprintf(buf, "%u\n",
1720                         from_kuid_munged(current_user_ns(), tun->owner)):
1721                 sprintf(buf, "-1\n");
1722 }
1723
1724 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1725                               char *buf)
1726 {
1727         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1728         return gid_valid(tun->group) ?
1729                 sprintf(buf, "%u\n",
1730                         from_kgid_munged(current_user_ns(), tun->group)):
1731                 sprintf(buf, "-1\n");
1732 }
1733
1734 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1735 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1736 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1737
1738 static struct attribute *tun_dev_attrs[] = {
1739         &dev_attr_tun_flags.attr,
1740         &dev_attr_owner.attr,
1741         &dev_attr_group.attr,
1742         NULL
1743 };
1744
1745 static const struct attribute_group tun_attr_group = {
1746         .attrs = tun_dev_attrs
1747 };
1748
1749 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1750 {
1751         struct tun_struct *tun;
1752         struct tun_file *tfile = file->private_data;
1753         struct net_device *dev;
1754         int err;
1755
1756         if (tfile->detached)
1757                 return -EINVAL;
1758
1759         dev = __dev_get_by_name(net, ifr->ifr_name);
1760         if (dev) {
1761                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1762                         return -EBUSY;
1763                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1764                         tun = netdev_priv(dev);
1765                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1766                         tun = netdev_priv(dev);
1767                 else
1768                         return -EINVAL;
1769
1770                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1771                     !!(tun->flags & IFF_MULTI_QUEUE))
1772                         return -EINVAL;
1773
1774                 if (tun_not_capable(tun))
1775                         return -EPERM;
1776                 err = security_tun_dev_open(tun->security);
1777                 if (err < 0)
1778                         return err;
1779
1780                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER, true);
1781                 if (err < 0)
1782                         return err;
1783
1784                 if (tun->flags & IFF_MULTI_QUEUE &&
1785                     (tun->numqueues + tun->numdisabled > 1)) {
1786                         /* One or more queue has already been attached, no need
1787                          * to initialize the device again.
1788                          */
1789                         return 0;
1790                 }
1791         }
1792         else {
1793                 char *name;
1794                 unsigned long flags = 0;
1795                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1796                              MAX_TAP_QUEUES : 1;
1797
1798                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1799                         return -EPERM;
1800                 err = security_tun_dev_create();
1801                 if (err < 0)
1802                         return err;
1803
1804                 /* Set dev type */
1805                 if (ifr->ifr_flags & IFF_TUN) {
1806                         /* TUN device */
1807                         flags |= IFF_TUN;
1808                         name = "tun%d";
1809                 } else if (ifr->ifr_flags & IFF_TAP) {
1810                         /* TAP device */
1811                         flags |= IFF_TAP;
1812                         name = "tap%d";
1813                 } else
1814                         return -EINVAL;
1815
1816                 if (*ifr->ifr_name)
1817                         name = ifr->ifr_name;
1818
1819                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1820                                        NET_NAME_UNKNOWN, tun_setup, queues,
1821                                        queues);
1822
1823                 if (!dev)
1824                         return -ENOMEM;
1825                 err = dev_get_valid_name(net, dev, name);
1826                 if (err < 0)
1827                         goto err_free_dev;
1828
1829                 dev_net_set(dev, net);
1830                 dev->rtnl_link_ops = &tun_link_ops;
1831                 dev->ifindex = tfile->ifindex;
1832                 dev->sysfs_groups[0] = &tun_attr_group;
1833
1834                 tun = netdev_priv(dev);
1835                 tun->dev = dev;
1836                 tun->flags = flags;
1837                 tun->txflt.count = 0;
1838                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1839
1840                 tun->align = NET_SKB_PAD;
1841                 tun->filter_attached = false;
1842                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1843
1844                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
1845                 if (!tun->pcpu_stats) {
1846                         err = -ENOMEM;
1847                         goto err_free_dev;
1848                 }
1849
1850                 spin_lock_init(&tun->lock);
1851
1852                 err = security_tun_dev_alloc_security(&tun->security);
1853                 if (err < 0)
1854                         goto err_free_stat;
1855
1856                 tun_net_init(dev);
1857                 tun_flow_init(tun);
1858
1859                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1860                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1861                                    NETIF_F_HW_VLAN_STAG_TX;
1862                 dev->features = dev->hw_features | NETIF_F_LLTX;
1863                 dev->vlan_features = dev->features &
1864                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1865                                        NETIF_F_HW_VLAN_STAG_TX);
1866
1867                 INIT_LIST_HEAD(&tun->disabled);
1868                 err = tun_attach(tun, file, false, false);
1869                 if (err < 0)
1870                         goto err_free_flow;
1871
1872                 err = register_netdevice(tun->dev);
1873                 if (err < 0)
1874                         goto err_detach;
1875                 /* free_netdev() won't check refcnt, to aovid race
1876                  * with dev_put() we need publish tun after registration.
1877                  */
1878                 rcu_assign_pointer(tfile->tun, tun);
1879         }
1880
1881         netif_carrier_on(tun->dev);
1882
1883         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1884
1885         tun->flags = (tun->flags & ~TUN_FEATURES) |
1886                 (ifr->ifr_flags & TUN_FEATURES);
1887
1888         /* Make sure persistent devices do not get stuck in
1889          * xoff state.
1890          */
1891         if (netif_running(tun->dev))
1892                 netif_tx_wake_all_queues(tun->dev);
1893
1894         strcpy(ifr->ifr_name, tun->dev->name);
1895         return 0;
1896
1897 err_detach:
1898         tun_detach_all(dev);
1899 err_free_flow:
1900         tun_flow_uninit(tun);
1901         security_tun_dev_free_security(tun->security);
1902 err_free_stat:
1903         free_percpu(tun->pcpu_stats);
1904 err_free_dev:
1905         free_netdev(dev);
1906         return err;
1907 }
1908
1909 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1910                        struct ifreq *ifr)
1911 {
1912         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1913
1914         strcpy(ifr->ifr_name, tun->dev->name);
1915
1916         ifr->ifr_flags = tun_flags(tun);
1917
1918 }
1919
1920 /* This is like a cut-down ethtool ops, except done via tun fd so no
1921  * privs required. */
1922 static int set_offload(struct tun_struct *tun, unsigned long arg)
1923 {
1924         netdev_features_t features = 0;
1925
1926         if (arg & TUN_F_CSUM) {
1927                 features |= NETIF_F_HW_CSUM;
1928                 arg &= ~TUN_F_CSUM;
1929
1930                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1931                         if (arg & TUN_F_TSO_ECN) {
1932                                 features |= NETIF_F_TSO_ECN;
1933                                 arg &= ~TUN_F_TSO_ECN;
1934                         }
1935                         if (arg & TUN_F_TSO4)
1936                                 features |= NETIF_F_TSO;
1937                         if (arg & TUN_F_TSO6)
1938                                 features |= NETIF_F_TSO6;
1939                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1940                 }
1941
1942                 if (arg & TUN_F_UFO) {
1943                         features |= NETIF_F_UFO;
1944                         arg &= ~TUN_F_UFO;
1945                 }
1946         }
1947
1948         /* This gives the user a way to test for new features in future by
1949          * trying to set them. */
1950         if (arg)
1951                 return -EINVAL;
1952
1953         tun->set_features = features;
1954         netdev_update_features(tun->dev);
1955
1956         return 0;
1957 }
1958
1959 static void tun_detach_filter(struct tun_struct *tun, int n)
1960 {
1961         int i;
1962         struct tun_file *tfile;
1963
1964         for (i = 0; i < n; i++) {
1965                 tfile = rtnl_dereference(tun->tfiles[i]);
1966                 lock_sock(tfile->socket.sk);
1967                 sk_detach_filter(tfile->socket.sk);
1968                 release_sock(tfile->socket.sk);
1969         }
1970
1971         tun->filter_attached = false;
1972 }
1973
1974 static int tun_attach_filter(struct tun_struct *tun)
1975 {
1976         int i, ret = 0;
1977         struct tun_file *tfile;
1978
1979         for (i = 0; i < tun->numqueues; i++) {
1980                 tfile = rtnl_dereference(tun->tfiles[i]);
1981                 lock_sock(tfile->socket.sk);
1982                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1983                 release_sock(tfile->socket.sk);
1984                 if (ret) {
1985                         tun_detach_filter(tun, i);
1986                         return ret;
1987                 }
1988         }
1989
1990         tun->filter_attached = true;
1991         return ret;
1992 }
1993
1994 static void tun_set_sndbuf(struct tun_struct *tun)
1995 {
1996         struct tun_file *tfile;
1997         int i;
1998
1999         for (i = 0; i < tun->numqueues; i++) {
2000                 tfile = rtnl_dereference(tun->tfiles[i]);
2001                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2002         }
2003 }
2004
2005 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2006 {
2007         struct tun_file *tfile = file->private_data;
2008         struct tun_struct *tun;
2009         int ret = 0;
2010
2011         rtnl_lock();
2012
2013         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2014                 tun = tfile->detached;
2015                 if (!tun) {
2016                         ret = -EINVAL;
2017                         goto unlock;
2018                 }
2019                 ret = security_tun_dev_attach_queue(tun->security);
2020                 if (ret < 0)
2021                         goto unlock;
2022                 ret = tun_attach(tun, file, false, true);
2023         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2024                 tun = rtnl_dereference(tfile->tun);
2025                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2026                         ret = -EINVAL;
2027                 else
2028                         __tun_detach(tfile, false);
2029         } else
2030                 ret = -EINVAL;
2031
2032 unlock:
2033         rtnl_unlock();
2034         return ret;
2035 }
2036
2037 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
2038 static unsigned char tun_get_addr_len(unsigned short type)
2039 {
2040         switch (type) {
2041         case ARPHRD_IP6GRE:
2042         case ARPHRD_TUNNEL6:
2043                 return sizeof(struct in6_addr);
2044         case ARPHRD_IPGRE:
2045         case ARPHRD_TUNNEL:
2046         case ARPHRD_SIT:
2047                 return 4;
2048         case ARPHRD_ETHER:
2049                 return ETH_ALEN;
2050         case ARPHRD_IEEE802154:
2051         case ARPHRD_IEEE802154_MONITOR:
2052                 return IEEE802154_EXTENDED_ADDR_LEN;
2053         case ARPHRD_PHONET_PIPE:
2054         case ARPHRD_PPP:
2055         case ARPHRD_NONE:
2056                 return 0;
2057         case ARPHRD_6LOWPAN:
2058                 return EUI64_ADDR_LEN;
2059         case ARPHRD_FDDI:
2060                 return FDDI_K_ALEN;
2061         case ARPHRD_HIPPI:
2062                 return HIPPI_ALEN;
2063         case ARPHRD_IEEE802:
2064                 return FC_ALEN;
2065         case ARPHRD_ROSE:
2066                 return ROSE_ADDR_LEN;
2067         case ARPHRD_NETROM:
2068                 return AX25_ADDR_LEN;
2069         case ARPHRD_LOCALTLK:
2070                 return LTALK_ALEN;
2071         default:
2072                 return 0;
2073         }
2074 }
2075
2076 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2077                             unsigned long arg, int ifreq_len)
2078 {
2079         struct tun_file *tfile = file->private_data;
2080         struct tun_struct *tun;
2081         void __user* argp = (void __user*)arg;
2082         struct ifreq ifr;
2083         kuid_t owner;
2084         kgid_t group;
2085         int sndbuf;
2086         int vnet_hdr_sz;
2087         unsigned int ifindex;
2088         int le;
2089         int ret;
2090
2091         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
2092                 if (copy_from_user(&ifr, argp, ifreq_len))
2093                         return -EFAULT;
2094         } else {
2095                 memset(&ifr, 0, sizeof(ifr));
2096         }
2097         if (cmd == TUNGETFEATURES) {
2098                 /* Currently this just means: "what IFF flags are valid?".
2099                  * This is needed because we never checked for invalid flags on
2100                  * TUNSETIFF.
2101                  */
2102                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2103                                 (unsigned int __user*)argp);
2104         } else if (cmd == TUNSETQUEUE)
2105                 return tun_set_queue(file, &ifr);
2106
2107         ret = 0;
2108         rtnl_lock();
2109
2110         tun = __tun_get(tfile);
2111         if (cmd == TUNSETIFF && !tun) {
2112                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2113
2114                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2115
2116                 if (ret)
2117                         goto unlock;
2118
2119                 if (copy_to_user(argp, &ifr, ifreq_len))
2120                         ret = -EFAULT;
2121                 goto unlock;
2122         }
2123         if (cmd == TUNSETIFINDEX) {
2124                 ret = -EPERM;
2125                 if (tun)
2126                         goto unlock;
2127
2128                 ret = -EFAULT;
2129                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2130                         goto unlock;
2131
2132                 ret = 0;
2133                 tfile->ifindex = ifindex;
2134                 goto unlock;
2135         }
2136
2137         ret = -EBADFD;
2138         if (!tun)
2139                 goto unlock;
2140
2141         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2142
2143         ret = 0;
2144         switch (cmd) {
2145         case TUNGETIFF:
2146                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2147
2148                 if (tfile->detached)
2149                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2150                 if (!tfile->socket.sk->sk_filter)
2151                         ifr.ifr_flags |= IFF_NOFILTER;
2152
2153                 if (copy_to_user(argp, &ifr, ifreq_len))
2154                         ret = -EFAULT;
2155                 break;
2156
2157         case TUNSETNOCSUM:
2158                 /* Disable/Enable checksum */
2159
2160                 /* [unimplemented] */
2161                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2162                           arg ? "disabled" : "enabled");
2163                 break;
2164
2165         case TUNSETPERSIST:
2166                 /* Disable/Enable persist mode. Keep an extra reference to the
2167                  * module to prevent the module being unprobed.
2168                  */
2169                 if (arg && !(tun->flags & IFF_PERSIST)) {
2170                         tun->flags |= IFF_PERSIST;
2171                         __module_get(THIS_MODULE);
2172                 }
2173                 if (!arg && (tun->flags & IFF_PERSIST)) {
2174                         tun->flags &= ~IFF_PERSIST;
2175                         module_put(THIS_MODULE);
2176                 }
2177
2178                 tun_debug(KERN_INFO, tun, "persist %s\n",
2179                           arg ? "enabled" : "disabled");
2180                 break;
2181
2182         case TUNSETOWNER:
2183                 /* Set owner of the device */
2184                 owner = make_kuid(current_user_ns(), arg);
2185                 if (!uid_valid(owner)) {
2186                         ret = -EINVAL;
2187                         break;
2188                 }
2189                 tun->owner = owner;
2190                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2191                           from_kuid(&init_user_ns, tun->owner));
2192                 break;
2193
2194         case TUNSETGROUP:
2195                 /* Set group of the device */
2196                 group = make_kgid(current_user_ns(), arg);
2197                 if (!gid_valid(group)) {
2198                         ret = -EINVAL;
2199                         break;
2200                 }
2201                 tun->group = group;
2202                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2203                           from_kgid(&init_user_ns, tun->group));
2204                 break;
2205
2206         case TUNSETLINK:
2207                 /* Only allow setting the type when the interface is down */
2208                 if (tun->dev->flags & IFF_UP) {
2209                         tun_debug(KERN_INFO, tun,
2210                                   "Linktype set failed because interface is up\n");
2211                         ret = -EBUSY;
2212                 } else {
2213                         tun->dev->type = (int) arg;
2214                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
2215                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2216                                   tun->dev->type);
2217                         ret = 0;
2218                 }
2219                 break;
2220
2221 #ifdef TUN_DEBUG
2222         case TUNSETDEBUG:
2223                 tun->debug = arg;
2224                 break;
2225 #endif
2226         case TUNSETOFFLOAD:
2227                 ret = set_offload(tun, arg);
2228                 break;
2229
2230         case TUNSETTXFILTER:
2231                 /* Can be set only for TAPs */
2232                 ret = -EINVAL;
2233                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2234                         break;
2235                 ret = update_filter(&tun->txflt, (void __user *)arg);
2236                 break;
2237
2238         case SIOCGIFHWADDR:
2239                 /* Get hw address */
2240                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2241                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2242                 if (copy_to_user(argp, &ifr, ifreq_len))
2243                         ret = -EFAULT;
2244                 break;
2245
2246         case SIOCSIFHWADDR:
2247                 /* Set hw address */
2248                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2249                           ifr.ifr_hwaddr.sa_data);
2250
2251                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2252                 break;
2253
2254         case TUNGETSNDBUF:
2255                 sndbuf = tfile->socket.sk->sk_sndbuf;
2256                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2257                         ret = -EFAULT;
2258                 break;
2259
2260         case TUNSETSNDBUF:
2261                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2262                         ret = -EFAULT;
2263                         break;
2264                 }
2265                 if (sndbuf <= 0) {
2266                         ret = -EINVAL;
2267                         break;
2268                 }
2269
2270                 tun->sndbuf = sndbuf;
2271                 tun_set_sndbuf(tun);
2272                 break;
2273
2274         case TUNGETVNETHDRSZ:
2275                 vnet_hdr_sz = tun->vnet_hdr_sz;
2276                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2277                         ret = -EFAULT;
2278                 break;
2279
2280         case TUNSETVNETHDRSZ:
2281                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2282                         ret = -EFAULT;
2283                         break;
2284                 }
2285                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2286                         ret = -EINVAL;
2287                         break;
2288                 }
2289
2290                 tun->vnet_hdr_sz = vnet_hdr_sz;
2291                 break;
2292
2293         case TUNGETVNETLE:
2294                 le = !!(tun->flags & TUN_VNET_LE);
2295                 if (put_user(le, (int __user *)argp))
2296                         ret = -EFAULT;
2297                 break;
2298
2299         case TUNSETVNETLE:
2300                 if (get_user(le, (int __user *)argp)) {
2301                         ret = -EFAULT;
2302                         break;
2303                 }
2304                 if (le)
2305                         tun->flags |= TUN_VNET_LE;
2306                 else
2307                         tun->flags &= ~TUN_VNET_LE;
2308                 break;
2309
2310         case TUNGETVNETBE:
2311                 ret = tun_get_vnet_be(tun, argp);
2312                 break;
2313
2314         case TUNSETVNETBE:
2315                 ret = tun_set_vnet_be(tun, argp);
2316                 break;
2317
2318         case TUNATTACHFILTER:
2319                 /* Can be set only for TAPs */
2320                 ret = -EINVAL;
2321                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2322                         break;
2323                 ret = -EFAULT;
2324                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2325                         break;
2326
2327                 ret = tun_attach_filter(tun);
2328                 break;
2329
2330         case TUNDETACHFILTER:
2331                 /* Can be set only for TAPs */
2332                 ret = -EINVAL;
2333                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2334                         break;
2335                 ret = 0;
2336                 tun_detach_filter(tun, tun->numqueues);
2337                 break;
2338
2339         case TUNGETFILTER:
2340                 ret = -EINVAL;
2341                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2342                         break;
2343                 ret = -EFAULT;
2344                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2345                         break;
2346                 ret = 0;
2347                 break;
2348
2349         default:
2350                 ret = -EINVAL;
2351                 break;
2352         }
2353
2354 unlock:
2355         rtnl_unlock();
2356         if (tun)
2357                 tun_put(tun);
2358         return ret;
2359 }
2360
2361 static long tun_chr_ioctl(struct file *file,
2362                           unsigned int cmd, unsigned long arg)
2363 {
2364         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2365 }
2366
2367 #ifdef CONFIG_COMPAT
2368 static long tun_chr_compat_ioctl(struct file *file,
2369                          unsigned int cmd, unsigned long arg)
2370 {
2371         switch (cmd) {
2372         case TUNSETIFF:
2373         case TUNGETIFF:
2374         case TUNSETTXFILTER:
2375         case TUNGETSNDBUF:
2376         case TUNSETSNDBUF:
2377         case SIOCGIFHWADDR:
2378         case SIOCSIFHWADDR:
2379                 arg = (unsigned long)compat_ptr(arg);
2380                 break;
2381         default:
2382                 arg = (compat_ulong_t)arg;
2383                 break;
2384         }
2385
2386         /*
2387          * compat_ifreq is shorter than ifreq, so we must not access beyond
2388          * the end of that structure. All fields that are used in this
2389          * driver are compatible though, we don't need to convert the
2390          * contents.
2391          */
2392         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2393 }
2394 #endif /* CONFIG_COMPAT */
2395
2396 static int tun_chr_fasync(int fd, struct file *file, int on)
2397 {
2398         struct tun_file *tfile = file->private_data;
2399         int ret;
2400
2401         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2402                 goto out;
2403
2404         if (on) {
2405                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2406                 tfile->flags |= TUN_FASYNC;
2407         } else
2408                 tfile->flags &= ~TUN_FASYNC;
2409         ret = 0;
2410 out:
2411         return ret;
2412 }
2413
2414 static int tun_chr_open(struct inode *inode, struct file * file)
2415 {
2416         struct net *net = current->nsproxy->net_ns;
2417         struct tun_file *tfile;
2418
2419         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2420
2421         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2422                                             &tun_proto, 0);
2423         if (!tfile)
2424                 return -ENOMEM;
2425         RCU_INIT_POINTER(tfile->tun, NULL);
2426         tfile->flags = 0;
2427         tfile->ifindex = 0;
2428
2429         init_waitqueue_head(&tfile->wq.wait);
2430         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2431
2432         tfile->socket.file = file;
2433         tfile->socket.ops = &tun_socket_ops;
2434
2435         sock_init_data(&tfile->socket, &tfile->sk);
2436
2437         tfile->sk.sk_write_space = tun_sock_write_space;
2438         tfile->sk.sk_sndbuf = INT_MAX;
2439
2440         file->private_data = tfile;
2441         INIT_LIST_HEAD(&tfile->next);
2442
2443         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2444
2445         memset(&tfile->tx_array, 0, sizeof(tfile->tx_array));
2446
2447         return 0;
2448 }
2449
2450 static int tun_chr_close(struct inode *inode, struct file *file)
2451 {
2452         struct tun_file *tfile = file->private_data;
2453
2454         tun_detach(tfile, true);
2455
2456         return 0;
2457 }
2458
2459 #ifdef CONFIG_PROC_FS
2460 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2461 {
2462         struct tun_struct *tun;
2463         struct ifreq ifr;
2464
2465         memset(&ifr, 0, sizeof(ifr));
2466
2467         rtnl_lock();
2468         tun = tun_get(f);
2469         if (tun)
2470                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2471         rtnl_unlock();
2472
2473         if (tun)
2474                 tun_put(tun);
2475
2476         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2477 }
2478 #endif
2479
2480 static const struct file_operations tun_fops = {
2481         .owner  = THIS_MODULE,
2482         .llseek = no_llseek,
2483         .read_iter  = tun_chr_read_iter,
2484         .write_iter = tun_chr_write_iter,
2485         .poll   = tun_chr_poll,
2486         .unlocked_ioctl = tun_chr_ioctl,
2487 #ifdef CONFIG_COMPAT
2488         .compat_ioctl = tun_chr_compat_ioctl,
2489 #endif
2490         .open   = tun_chr_open,
2491         .release = tun_chr_close,
2492         .fasync = tun_chr_fasync,
2493 #ifdef CONFIG_PROC_FS
2494         .show_fdinfo = tun_chr_show_fdinfo,
2495 #endif
2496 };
2497
2498 static struct miscdevice tun_miscdev = {
2499         .minor = TUN_MINOR,
2500         .name = "tun",
2501         .nodename = "net/tun",
2502         .fops = &tun_fops,
2503 };
2504
2505 /* ethtool interface */
2506
2507 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2508 {
2509         cmd->supported          = 0;
2510         cmd->advertising        = 0;
2511         ethtool_cmd_speed_set(cmd, SPEED_10);
2512         cmd->duplex             = DUPLEX_FULL;
2513         cmd->port               = PORT_TP;
2514         cmd->phy_address        = 0;
2515         cmd->transceiver        = XCVR_INTERNAL;
2516         cmd->autoneg            = AUTONEG_DISABLE;
2517         cmd->maxtxpkt           = 0;
2518         cmd->maxrxpkt           = 0;
2519         return 0;
2520 }
2521
2522 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2523 {
2524         struct tun_struct *tun = netdev_priv(dev);
2525
2526         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2527         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2528
2529         switch (tun->flags & TUN_TYPE_MASK) {
2530         case IFF_TUN:
2531                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2532                 break;
2533         case IFF_TAP:
2534                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2535                 break;
2536         }
2537 }
2538
2539 static u32 tun_get_msglevel(struct net_device *dev)
2540 {
2541 #ifdef TUN_DEBUG
2542         struct tun_struct *tun = netdev_priv(dev);
2543         return tun->debug;
2544 #else
2545         return -EOPNOTSUPP;
2546 #endif
2547 }
2548
2549 static void tun_set_msglevel(struct net_device *dev, u32 value)
2550 {
2551 #ifdef TUN_DEBUG
2552         struct tun_struct *tun = netdev_priv(dev);
2553         tun->debug = value;
2554 #endif
2555 }
2556
2557 static const struct ethtool_ops tun_ethtool_ops = {
2558         .get_settings   = tun_get_settings,
2559         .get_drvinfo    = tun_get_drvinfo,
2560         .get_msglevel   = tun_get_msglevel,
2561         .set_msglevel   = tun_set_msglevel,
2562         .get_link       = ethtool_op_get_link,
2563         .get_ts_info    = ethtool_op_get_ts_info,
2564 };
2565
2566 static int tun_queue_resize(struct tun_struct *tun)
2567 {
2568         struct net_device *dev = tun->dev;
2569         struct tun_file *tfile;
2570         struct skb_array **arrays;
2571         int n = tun->numqueues + tun->numdisabled;
2572         int ret, i;
2573
2574         arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL);
2575         if (!arrays)
2576                 return -ENOMEM;
2577
2578         for (i = 0; i < tun->numqueues; i++) {
2579                 tfile = rtnl_dereference(tun->tfiles[i]);
2580                 arrays[i] = &tfile->tx_array;
2581         }
2582         list_for_each_entry(tfile, &tun->disabled, next)
2583                 arrays[i++] = &tfile->tx_array;
2584
2585         ret = skb_array_resize_multiple(arrays, n,
2586                                         dev->tx_queue_len, GFP_KERNEL);
2587
2588         kfree(arrays);
2589         return ret;
2590 }
2591
2592 static int tun_device_event(struct notifier_block *unused,
2593                             unsigned long event, void *ptr)
2594 {
2595         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2596         struct tun_struct *tun = netdev_priv(dev);
2597         int i;
2598
2599         if (dev->rtnl_link_ops != &tun_link_ops)
2600                 return NOTIFY_DONE;
2601
2602         switch (event) {
2603         case NETDEV_CHANGE_TX_QUEUE_LEN:
2604                 if (tun_queue_resize(tun))
2605                         return NOTIFY_BAD;
2606                 break;
2607         case NETDEV_UP:
2608                 for (i = 0; i < tun->numqueues; i++) {
2609                         struct tun_file *tfile;
2610
2611                         tfile = rtnl_dereference(tun->tfiles[i]);
2612                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
2613                 }
2614                 break;
2615         default:
2616                 break;
2617         }
2618
2619         return NOTIFY_DONE;
2620 }
2621
2622 static struct notifier_block tun_notifier_block __read_mostly = {
2623         .notifier_call  = tun_device_event,
2624 };
2625
2626 static int __init tun_init(void)
2627 {
2628         int ret = 0;
2629
2630         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2631         pr_info("%s\n", DRV_COPYRIGHT);
2632
2633         ret = rtnl_link_register(&tun_link_ops);
2634         if (ret) {
2635                 pr_err("Can't register link_ops\n");
2636                 goto err_linkops;
2637         }
2638
2639         ret = misc_register(&tun_miscdev);
2640         if (ret) {
2641                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2642                 goto err_misc;
2643         }
2644
2645         register_netdevice_notifier(&tun_notifier_block);
2646         return  0;
2647 err_misc:
2648         rtnl_link_unregister(&tun_link_ops);
2649 err_linkops:
2650         return ret;
2651 }
2652
2653 static void tun_cleanup(void)
2654 {
2655         misc_deregister(&tun_miscdev);
2656         rtnl_link_unregister(&tun_link_ops);
2657         unregister_netdevice_notifier(&tun_notifier_block);
2658 }
2659
2660 /* Get an underlying socket object from tun file.  Returns error unless file is
2661  * attached to a device.  The returned object works like a packet socket, it
2662  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2663  * holding a reference to the file for as long as the socket is in use. */
2664 struct socket *tun_get_socket(struct file *file)
2665 {
2666         struct tun_file *tfile;
2667         if (file->f_op != &tun_fops)
2668                 return ERR_PTR(-EINVAL);
2669         tfile = file->private_data;
2670         if (!tfile)
2671                 return ERR_PTR(-EBADFD);
2672         return &tfile->socket;
2673 }
2674 EXPORT_SYMBOL_GPL(tun_get_socket);
2675
2676 module_init(tun_init);
2677 module_exit(tun_cleanup);
2678 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2679 MODULE_AUTHOR(DRV_COPYRIGHT);
2680 MODULE_LICENSE("GPL");
2681 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2682 MODULE_ALIAS("devname:net/tun");