drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/major.h>
  48 #include <linux/slab.h>
  49 #include <linux/poll.h>
  50 #include <linux/fcntl.h>
  51 #include <linux/init.h>
  52 #include <linux/skbuff.h>
  53 #include <linux/netdevice.h>
  54 #include <linux/etherdevice.h>
  55 #include <linux/miscdevice.h>
  56 #include <linux/ethtool.h>
  57 #include <linux/rtnetlink.h>
  58 #include <linux/compat.h>
  59 #include <linux/if.h>
  60 #include <linux/if_arp.h>
  61 #include <linux/if_ether.h>
  62 #include <linux/if_tun.h>
  63 #include <linux/crc32.h>
  64 #include <linux/nsproxy.h>
  65 #include <linux/virtio_net.h>
  66 #include <linux/rcupdate.h>
  67 #include <net/net_namespace.h>
  68 #include <net/netns/generic.h>
  69 #include <net/rtnetlink.h>
  70 #include <net/sock.h>
  71
  72 #include <asm/uaccess.h>
  73
  74 /* Uncomment to enable debugging */
  75 /* #define TUN_DEBUG 1 */
  76
  77 #ifdef TUN_DEBUG
  78 static int debug;
  79
  80 #define tun_debug(level, tun, fmt, args...)                     \
  81 do {                                                            \
  82         if (tun->debug)                                         \
  83                 netdev_printk(level, tun->dev, fmt, ##args);    \
  84 } while (0)
  85 #define DBG1(level, fmt, args...)                               \
  86 do {                                                            \
  87         if (debug == 2)                                         \
  88                 printk(level fmt, ##args);                      \
  89 } while (0)
  90 #else
  91 #define tun_debug(level, tun, fmt, args...)                     \
  92 do {                                                            \
  93         if (0)                                                  \
  94                 netdev_printk(level, tun->dev, fmt, ##args);    \
  95 } while (0)
  96 #define DBG1(level, fmt, args...)                               \
  97 do {                                                            \
  98         if (0)                                                  \
  99                 printk(level fmt, ##args);                      \
 100 } while (0)
 101 #endif
 102
 103 #define GOODCOPY_LEN 128
 104
 105 #define FLT_EXACT_COUNT 8
 106 struct tap_filter {
 107         unsigned int    count;    /* Number of addrs. Zero means disabled */
 108         u32             mask[2];  /* Mask of the hashed addrs */
 109         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 110 };
 111
 112 /* 1024 is probably a high enough limit: modern hypervisors seem to support on
 113  * the order of 100-200 CPUs so this leaves us some breathing space if we want
 114  * to match a queue per guest CPU.
 115  */
 116 #define MAX_TAP_QUEUES 1024
 117
 118 #define TUN_FLOW_EXPIRE (3 * HZ)
 119
 120 /* A tun_file connects an open character device to a tuntap netdevice. It
 121  * also contains all socket related strctures (except sock_fprog and tap_filter)
 122  * to serve as one transmit queue for tuntap device. The sock_fprog and
 123  * tap_filter were kept in tun_struct since they were used for filtering for the
 124  * netdevice not for a specific queue (at least I didn't see the requirement for
 125  * this).
 126  *
 127  * RCU usage:
 128  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 129  * other can only be read while rcu_read_lock or rtnl_lock is held.
 130  */
 131 struct tun_file {
 132         struct sock sk;
 133         struct socket socket;
 134         struct socket_wq wq;
 135         struct tun_struct __rcu *tun;
 136         struct net *net;
 137         struct fasync_struct *fasync;
 138         /* only used for fasnyc */
 139         unsigned int flags;
 140         u16 queue_index;
 141         struct list_head next;
 142         struct tun_struct *detached;
 143 };
 144
 145 struct tun_flow_entry {
 146         struct hlist_node hash_link;
 147         struct rcu_head rcu;
 148         struct tun_struct *tun;
 149
 150         u32 rxhash;
 151         int queue_index;
 152         unsigned long updated;
 153 };
 154
 155 #define TUN_NUM_FLOW_ENTRIES 1024
 156
 157 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 158  * device, socket filter, sndbuf and vnet header size were restore when the
 159  * file were attached to a persist device.
 160  */
 161 struct tun_struct {
 162         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 163         unsigned int            numqueues;
 164         unsigned int            flags;
 165         kuid_t                  owner;
 166         kgid_t                  group;
 167
 168         struct net_device       *dev;
 169         netdev_features_t       set_features;
 170 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 171                           NETIF_F_TSO6|NETIF_F_UFO)
 172
 173         int                     vnet_hdr_sz;
 174         int                     sndbuf;
 175         struct tap_filter       txflt;
 176         struct sock_fprog       fprog;
 177         /* protected by rtnl lock */
 178         bool                    filter_attached;
 179 #ifdef TUN_DEBUG
 180         int debug;
 181 #endif
 182         spinlock_t lock;
 183         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 184         struct timer_list flow_gc_timer;
 185         unsigned long ageing_time;
 186         unsigned int numdisabled;
 187         struct list_head disabled;
 188         void *security;
 189 };
 190
 191 static inline u32 tun_hashfn(u32 rxhash)
 192 {
 193         return rxhash & 0x3ff;
 194 }
 195
 196 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 197 {
 198         struct tun_flow_entry *e;
 199         struct hlist_node *n;
 200
 201         hlist_for_each_entry_rcu(e, n, head, hash_link) {
 202                 if (e->rxhash == rxhash)
 203                         return e;
 204         }
 205         return NULL;
 206 }
 207
 208 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 209                                               struct hlist_head *head,
 210                                               u32 rxhash, u16 queue_index)
 211 {
 212         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 213
 214         if (e) {
 215                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 216                           rxhash, queue_index);
 217                 e->updated = jiffies;
 218                 e->rxhash = rxhash;
 219                 e->queue_index = queue_index;
 220                 e->tun = tun;
 221                 hlist_add_head_rcu(&e->hash_link, head);
 222         }
 223         return e;
 224 }
 225
 226 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 227 {
 228         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 229                   e->rxhash, e->queue_index);
 230         hlist_del_rcu(&e->hash_link);
 231         kfree_rcu(e, rcu);
 232 }
 233
 234 static void tun_flow_flush(struct tun_struct *tun)
 235 {
 236         int i;
 237
 238         spin_lock_bh(&tun->lock);
 239         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 240                 struct tun_flow_entry *e;
 241                 struct hlist_node *h, *n;
 242
 243                 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link)
 244                         tun_flow_delete(tun, e);
 245         }
 246         spin_unlock_bh(&tun->lock);
 247 }
 248
 249 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 250 {
 251         int i;
 252
 253         spin_lock_bh(&tun->lock);
 254         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 255                 struct tun_flow_entry *e;
 256                 struct hlist_node *h, *n;
 257
 258                 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
 259                         if (e->queue_index == queue_index)
 260                                 tun_flow_delete(tun, e);
 261                 }
 262         }
 263         spin_unlock_bh(&tun->lock);
 264 }
 265
 266 static void tun_flow_cleanup(unsigned long data)
 267 {
 268         struct tun_struct *tun = (struct tun_struct *)data;
 269         unsigned long delay = tun->ageing_time;
 270         unsigned long next_timer = jiffies + delay;
 271         unsigned long count = 0;
 272         int i;
 273
 274         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 275
 276         spin_lock_bh(&tun->lock);
 277         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 278                 struct tun_flow_entry *e;
 279                 struct hlist_node *h, *n;
 280
 281                 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
 282                         unsigned long this_timer;
 283                         count++;
 284                         this_timer = e->updated + delay;
 285                         if (time_before_eq(this_timer, jiffies))
 286                                 tun_flow_delete(tun, e);
 287                         else if (time_before(this_timer, next_timer))
 288                                 next_timer = this_timer;
 289                 }
 290         }
 291
 292         if (count)
 293                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 294         spin_unlock_bh(&tun->lock);
 295 }
 296
 297 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 298                             u16 queue_index)
 299 {
 300         struct hlist_head *head;
 301         struct tun_flow_entry *e;
 302         unsigned long delay = tun->ageing_time;
 303
 304         if (!rxhash)
 305                 return;
 306         else
 307                 head = &tun->flows[tun_hashfn(rxhash)];
 308
 309         rcu_read_lock();
 310
 311         if (tun->numqueues == 1)
 312                 goto unlock;
 313
 314         e = tun_flow_find(head, rxhash);
 315         if (likely(e)) {
 316                 /* TODO: keep queueing to old queue until it's empty? */
 317                 e->queue_index = queue_index;
 318                 e->updated = jiffies;
 319         } else {
 320                 spin_lock_bh(&tun->lock);
 321                 if (!tun_flow_find(head, rxhash))
 322                         tun_flow_create(tun, head, rxhash, queue_index);
 323
 324                 if (!timer_pending(&tun->flow_gc_timer))
 325                         mod_timer(&tun->flow_gc_timer,
 326                                   round_jiffies_up(jiffies + delay));
 327                 spin_unlock_bh(&tun->lock);
 328         }
 329
 330 unlock:
 331         rcu_read_unlock();
 332 }
 333
 334 /* We try to identify a flow through its rxhash first. The reason that
 335  * we do not check rxq no. is becuase some cards(e.g 82599), chooses
 336  * the rxq based on the txq where the last packet of the flow comes. As
 337  * the userspace application move between processors, we may get a
 338  * different rxq no. here. If we could not get rxhash, then we would
 339  * hope the rxq no. may help here.
 340  */
 341 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
 342 {
 343         struct tun_struct *tun = netdev_priv(dev);
 344         struct tun_flow_entry *e;
 345         u32 txq = 0;
 346         u32 numqueues = 0;
 347
 348         rcu_read_lock();
 349         numqueues = tun->numqueues;
 350
 351         txq = skb_get_rxhash(skb);
 352         if (txq) {
 353                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 354                 if (e)
 355                         txq = e->queue_index;
 356                 else
 357                         /* use multiply and shift instead of expensive divide */
 358                         txq = ((u64)txq * numqueues) >> 32;
 359         } else if (likely(skb_rx_queue_recorded(skb))) {
 360                 txq = skb_get_rx_queue(skb);
 361                 while (unlikely(txq >= numqueues))
 362                         txq -= numqueues;
 363         }
 364
 365         rcu_read_unlock();
 366         return txq;
 367 }
 368
 369 static inline bool tun_not_capable(struct tun_struct *tun)
 370 {
 371         const struct cred *cred = current_cred();
 372         struct net *net = dev_net(tun->dev);
 373
 374         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 375                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 376                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 377 }
 378
 379 static void tun_set_real_num_queues(struct tun_struct *tun)
 380 {
 381         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 382         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 383 }
 384
 385 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 386 {
 387         tfile->detached = tun;
 388         list_add_tail(&tfile->next, &tun->disabled);
 389         ++tun->numdisabled;
 390 }
 391
 392 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 393 {
 394         struct tun_struct *tun = tfile->detached;
 395
 396         tfile->detached = NULL;
 397         list_del_init(&tfile->next);
 398         --tun->numdisabled;
 399         return tun;
 400 }
 401
 402 static void __tun_detach(struct tun_file *tfile, bool clean)
 403 {
 404         struct tun_file *ntfile;
 405         struct tun_struct *tun;
 406         struct net_device *dev;
 407
 408         tun = rtnl_dereference(tfile->tun);
 409
 410         if (tun) {
 411                 u16 index = tfile->queue_index;
 412                 BUG_ON(index >= tun->numqueues);
 413                 dev = tun->dev;
 414
 415                 rcu_assign_pointer(tun->tfiles[index],
 416                                    tun->tfiles[tun->numqueues - 1]);
 417                 rcu_assign_pointer(tfile->tun, NULL);
 418                 ntfile = rtnl_dereference(tun->tfiles[index]);
 419                 ntfile->queue_index = index;
 420
 421                 --tun->numqueues;
 422                 if (clean)
 423                         sock_put(&tfile->sk);
 424                 else
 425                         tun_disable_queue(tun, tfile);
 426
 427                 synchronize_net();
 428                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 429                 /* Drop read queue */
 430                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 431                 tun_set_real_num_queues(tun);
 432         } else if (tfile->detached && clean) {
 433                 tun = tun_enable_queue(tfile);
 434                 sock_put(&tfile->sk);
 435         }
 436
 437         if (clean) {
 438                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0 &&
 439                     !(tun->flags & TUN_PERSIST))
 440                         if (tun->dev->reg_state == NETREG_REGISTERED)
 441                                 unregister_netdevice(tun->dev);
 442
 443                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 444                                  &tfile->socket.flags));
 445                 sk_release_kernel(&tfile->sk);
 446         }
 447 }
 448
 449 static void tun_detach(struct tun_file *tfile, bool clean)
 450 {
 451         rtnl_lock();
 452         __tun_detach(tfile, clean);
 453         rtnl_unlock();
 454 }
 455
 456 static void tun_detach_all(struct net_device *dev)
 457 {
 458         struct tun_struct *tun = netdev_priv(dev);
 459         struct tun_file *tfile, *tmp;
 460         int i, n = tun->numqueues;
 461
 462         for (i = 0; i < n; i++) {
 463                 tfile = rtnl_dereference(tun->tfiles[i]);
 464                 BUG_ON(!tfile);
 465                 wake_up_all(&tfile->wq.wait);
 466                 rcu_assign_pointer(tfile->tun, NULL);
 467                 --tun->numqueues;
 468         }
 469         BUG_ON(tun->numqueues != 0);
 470
 471         synchronize_net();
 472         for (i = 0; i < n; i++) {
 473                 tfile = rtnl_dereference(tun->tfiles[i]);
 474                 /* Drop read queue */
 475                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 476                 sock_put(&tfile->sk);
 477         }
 478         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 479                 tun_enable_queue(tfile);
 480                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 481                 sock_put(&tfile->sk);
 482         }
 483         BUG_ON(tun->numdisabled != 0);
 484
 485         if (tun->flags & TUN_PERSIST)
 486                 module_put(THIS_MODULE);
 487 }
 488
 489 static int tun_attach(struct tun_struct *tun, struct file *file)
 490 {
 491         struct tun_file *tfile = file->private_data;
 492         int err;
 493
 494         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 495         if (err < 0)
 496                 goto out;
 497
 498         err = -EINVAL;
 499         if (rtnl_dereference(tfile->tun))
 500                 goto out;
 501
 502         err = -EBUSY;
 503         if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 504                 goto out;
 505
 506         err = -E2BIG;
 507         if (!tfile->detached &&
 508             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 509                 goto out;
 510
 511         err = 0;
 512
 513         /* Re-attach the filter to presist device */
 514         if (tun->filter_attached == true) {
 515                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 516                 if (!err)
 517                         goto out;
 518         }
 519         tfile->queue_index = tun->numqueues;
 520         rcu_assign_pointer(tfile->tun, tun);
 521         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 522         tun->numqueues++;
 523
 524         if (tfile->detached)
 525                 tun_enable_queue(tfile);
 526         else
 527                 sock_hold(&tfile->sk);
 528
 529         tun_set_real_num_queues(tun);
 530
 531         /* device is allowed to go away first, so no need to hold extra
 532          * refcnt.
 533          */
 534
 535 out:
 536         return err;
 537 }
 538
 539 static struct tun_struct *__tun_get(struct tun_file *tfile)
 540 {
 541         struct tun_struct *tun;
 542
 543         rcu_read_lock();
 544         tun = rcu_dereference(tfile->tun);
 545         if (tun)
 546                 dev_hold(tun->dev);
 547         rcu_read_unlock();
 548
 549         return tun;
 550 }
 551
 552 static struct tun_struct *tun_get(struct file *file)
 553 {
 554         return __tun_get(file->private_data);
 555 }
 556
 557 static void tun_put(struct tun_struct *tun)
 558 {
 559         dev_put(tun->dev);
 560 }
 561
 562 /* TAP filtering */
 563 static void addr_hash_set(u32 *mask, const u8 *addr)
 564 {
 565         int n = ether_crc(ETH_ALEN, addr) >> 26;
 566         mask[n >> 5] |= (1 << (n & 31));
 567 }
 568
 569 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 570 {
 571         int n = ether_crc(ETH_ALEN, addr) >> 26;
 572         return mask[n >> 5] & (1 << (n & 31));
 573 }
 574
 575 static int update_filter(struct tap_filter *filter, void __user *arg)
 576 {
 577         struct { u8 u[ETH_ALEN]; } *addr;
 578         struct tun_filter uf;
 579         int err, alen, n, nexact;
 580
 581         if (copy_from_user(&uf, arg, sizeof(uf)))
 582                 return -EFAULT;
 583
 584         if (!uf.count) {
 585                 /* Disabled */
 586                 filter->count = 0;
 587                 return 0;
 588         }
 589
 590         alen = ETH_ALEN * uf.count;
 591         addr = kmalloc(alen, GFP_KERNEL);
 592         if (!addr)
 593                 return -ENOMEM;
 594
 595         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 596                 err = -EFAULT;
 597                 goto done;
 598         }
 599
 600         /* The filter is updated without holding any locks. Which is
 601          * perfectly safe. We disable it first and in the worst
 602          * case we'll accept a few undesired packets. */
 603         filter->count = 0;
 604         wmb();
 605
 606         /* Use first set of addresses as an exact filter */
 607         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 608                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 609
 610         nexact = n;
 611
 612         /* Remaining multicast addresses are hashed,
 613          * unicast will leave the filter disabled. */
 614         memset(filter->mask, 0, sizeof(filter->mask));
 615         for (; n < uf.count; n++) {
 616                 if (!is_multicast_ether_addr(addr[n].u)) {
 617                         err = 0; /* no filter */
 618                         goto done;
 619                 }
 620                 addr_hash_set(filter->mask, addr[n].u);
 621         }
 622
 623         /* For ALLMULTI just set the mask to all ones.
 624          * This overrides the mask populated above. */
 625         if ((uf.flags & TUN_FLT_ALLMULTI))
 626                 memset(filter->mask, ~0, sizeof(filter->mask));
 627
 628         /* Now enable the filter */
 629         wmb();
 630         filter->count = nexact;
 631
 632         /* Return the number of exact filters */
 633         err = nexact;
 634
 635 done:
 636         kfree(addr);
 637         return err;
 638 }
 639
 640 /* Returns: 0 - drop, !=0 - accept */
 641 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 642 {
 643         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 644          * at this point. */
 645         struct ethhdr *eh = (struct ethhdr *) skb->data;
 646         int i;
 647
 648         /* Exact match */
 649         for (i = 0; i < filter->count; i++)
 650                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 651                         return 1;
 652
 653         /* Inexact match (multicast only) */
 654         if (is_multicast_ether_addr(eh->h_dest))
 655                 return addr_hash_test(filter->mask, eh->h_dest);
 656
 657         return 0;
 658 }
 659
 660 /*
 661  * Checks whether the packet is accepted or not.
 662  * Returns: 0 - drop, !=0 - accept
 663  */
 664 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 665 {
 666         if (!filter->count)
 667                 return 1;
 668
 669         return run_filter(filter, skb);
 670 }
 671
 672 /* Network device part of the driver */
 673
 674 static const struct ethtool_ops tun_ethtool_ops;
 675
 676 /* Net device detach from fd. */
 677 static void tun_net_uninit(struct net_device *dev)
 678 {
 679         tun_detach_all(dev);
 680 }
 681
 682 /* Net device open. */
 683 static int tun_net_open(struct net_device *dev)
 684 {
 685         netif_tx_start_all_queues(dev);
 686         return 0;
 687 }
 688
 689 /* Net device close. */
 690 static int tun_net_close(struct net_device *dev)
 691 {
 692         netif_tx_stop_all_queues(dev);
 693         return 0;
 694 }
 695
 696 /* Net device start xmit */
 697 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 698 {
 699         struct tun_struct *tun = netdev_priv(dev);
 700         int txq = skb->queue_mapping;
 701         struct tun_file *tfile;
 702
 703         rcu_read_lock();
 704         tfile = rcu_dereference(tun->tfiles[txq]);
 705
 706         /* Drop packet if interface is not attached */
 707         if (txq >= tun->numqueues)
 708                 goto drop;
 709
 710         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 711
 712         BUG_ON(!tfile);
 713
 714         /* Drop if the filter does not like it.
 715          * This is a noop if the filter is disabled.
 716          * Filter can be enabled only for the TAP devices. */
 717         if (!check_filter(&tun->txflt, skb))
 718                 goto drop;
 719
 720         if (tfile->socket.sk->sk_filter &&
 721             sk_filter(tfile->socket.sk, skb))
 722                 goto drop;
 723
 724         /* Limit the number of packets queued by dividing txq length with the
 725          * number of queues.
 726          */
 727         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
 728                           >= dev->tx_queue_len / tun->numqueues)
 729                 goto drop;
 730
 731         /* Orphan the skb - required as we might hang on to it
 732          * for indefinite time. */
 733         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 734                 goto drop;
 735         skb_orphan(skb);
 736
 737         /* Enqueue packet */
 738         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 739
 740         /* Notify and wake up reader process */
 741         if (tfile->flags & TUN_FASYNC)
 742                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 743         wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
 744                                    POLLRDNORM | POLLRDBAND);
 745
 746         rcu_read_unlock();
 747         return NETDEV_TX_OK;
 748
 749 drop:
 750         dev->stats.tx_dropped++;
 751         skb_tx_error(skb);
 752         kfree_skb(skb);
 753         rcu_read_unlock();
 754         return NETDEV_TX_OK;
 755 }
 756
 757 static void tun_net_mclist(struct net_device *dev)
 758 {
 759         /*
 760          * This callback is supposed to deal with mc filter in
 761          * _rx_ path and has nothing to do with the _tx_ path.
 762          * In rx path we always accept everything userspace gives us.
 763          */
 764 }
 765
 766 #define MIN_MTU 68
 767 #define MAX_MTU 65535
 768
 769 static int
 770 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 771 {
 772         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 773                 return -EINVAL;
 774         dev->mtu = new_mtu;
 775         return 0;
 776 }
 777
 778 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 779         netdev_features_t features)
 780 {
 781         struct tun_struct *tun = netdev_priv(dev);
 782
 783         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 784 }
 785 #ifdef CONFIG_NET_POLL_CONTROLLER
 786 static void tun_poll_controller(struct net_device *dev)
 787 {
 788         /*
 789          * Tun only receives frames when:
 790          * 1) the char device endpoint gets data from user space
 791          * 2) the tun socket gets a sendmsg call from user space
 792          * Since both of those are syncronous operations, we are guaranteed
 793          * never to have pending data when we poll for it
 794          * so theres nothing to do here but return.
 795          * We need this though so netpoll recognizes us as an interface that
 796          * supports polling, which enables bridge devices in virt setups to
 797          * still use netconsole
 798          */
 799         return;
 800 }
 801 #endif
 802 static const struct net_device_ops tun_netdev_ops = {
 803         .ndo_uninit             = tun_net_uninit,
 804         .ndo_open               = tun_net_open,
 805         .ndo_stop               = tun_net_close,
 806         .ndo_start_xmit         = tun_net_xmit,
 807         .ndo_change_mtu         = tun_net_change_mtu,
 808         .ndo_fix_features       = tun_net_fix_features,
 809         .ndo_select_queue       = tun_select_queue,
 810 #ifdef CONFIG_NET_POLL_CONTROLLER
 811         .ndo_poll_controller    = tun_poll_controller,
 812 #endif
 813 };
 814
 815 static const struct net_device_ops tap_netdev_ops = {
 816         .ndo_uninit             = tun_net_uninit,
 817         .ndo_open               = tun_net_open,
 818         .ndo_stop               = tun_net_close,
 819         .ndo_start_xmit         = tun_net_xmit,
 820         .ndo_change_mtu         = tun_net_change_mtu,
 821         .ndo_fix_features       = tun_net_fix_features,
 822         .ndo_set_rx_mode        = tun_net_mclist,
 823         .ndo_set_mac_address    = eth_mac_addr,
 824         .ndo_validate_addr      = eth_validate_addr,
 825         .ndo_select_queue       = tun_select_queue,
 826 #ifdef CONFIG_NET_POLL_CONTROLLER
 827         .ndo_poll_controller    = tun_poll_controller,
 828 #endif
 829 };
 830
 831 static int tun_flow_init(struct tun_struct *tun)
 832 {
 833         int i;
 834
 835         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 836                 INIT_HLIST_HEAD(&tun->flows[i]);
 837
 838         tun->ageing_time = TUN_FLOW_EXPIRE;
 839         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 840         mod_timer(&tun->flow_gc_timer,
 841                   round_jiffies_up(jiffies + tun->ageing_time));
 842
 843         return 0;
 844 }
 845
 846 static void tun_flow_uninit(struct tun_struct *tun)
 847 {
 848         del_timer_sync(&tun->flow_gc_timer);
 849         tun_flow_flush(tun);
 850 }
 851
 852 /* Initialize net device. */
 853 static void tun_net_init(struct net_device *dev)
 854 {
 855         struct tun_struct *tun = netdev_priv(dev);
 856
 857         switch (tun->flags & TUN_TYPE_MASK) {
 858         case TUN_TUN_DEV:
 859                 dev->netdev_ops = &tun_netdev_ops;
 860
 861                 /* Point-to-Point TUN Device */
 862                 dev->hard_header_len = 0;
 863                 dev->addr_len = 0;
 864                 dev->mtu = 1500;
 865
 866                 /* Zero header length */
 867                 dev->type = ARPHRD_NONE;
 868                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 869                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 870                 break;
 871
 872         case TUN_TAP_DEV:
 873                 dev->netdev_ops = &tap_netdev_ops;
 874                 /* Ethernet TAP Device */
 875                 ether_setup(dev);
 876                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 877                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 878
 879                 eth_hw_addr_random(dev);
 880
 881                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 882                 break;
 883         }
 884 }
 885
 886 /* Character device part */
 887
 888 /* Poll */
 889 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 890 {
 891         struct tun_file *tfile = file->private_data;
 892         struct tun_struct *tun = __tun_get(tfile);
 893         struct sock *sk;
 894         unsigned int mask = 0;
 895
 896         if (!tun)
 897                 return POLLERR;
 898
 899         sk = tfile->socket.sk;
 900
 901         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 902
 903         poll_wait(file, &tfile->wq.wait, wait);
 904
 905         if (!skb_queue_empty(&sk->sk_receive_queue))
 906                 mask |= POLLIN | POLLRDNORM;
 907
 908         if (sock_writeable(sk) ||
 909             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 910              sock_writeable(sk)))
 911                 mask |= POLLOUT | POLLWRNORM;
 912
 913         if (tun->dev->reg_state != NETREG_REGISTERED)
 914                 mask = POLLERR;
 915
 916         tun_put(tun);
 917         return mask;
 918 }
 919
 920 /* prepad is the amount to reserve at front.  len is length after that.
 921  * linear is a hint as to how much to copy (usually headers). */
 922 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 923                                      size_t prepad, size_t len,
 924                                      size_t linear, int noblock)
 925 {
 926         struct sock *sk = tfile->socket.sk;
 927         struct sk_buff *skb;
 928         int err;
 929
 930         /* Under a page?  Don't bother with paged skb. */
 931         if (prepad + len < PAGE_SIZE || !linear)
 932                 linear = len;
 933
 934         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 935                                    &err);
 936         if (!skb)
 937                 return ERR_PTR(err);
 938
 939         skb_reserve(skb, prepad);
 940         skb_put(skb, linear);
 941         skb->data_len = len - linear;
 942         skb->len += len - linear;
 943
 944         return skb;
 945 }
 946
 947 /* set skb frags from iovec, this can move to core network code for reuse */
 948 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 949                                   int offset, size_t count)
 950 {
 951         int len = iov_length(from, count) - offset;
 952         int copy = skb_headlen(skb);
 953         int size, offset1 = 0;
 954         int i = 0;
 955
 956         /* Skip over from offset */
 957         while (count && (offset >= from->iov_len)) {
 958                 offset -= from->iov_len;
 959                 ++from;
 960                 --count;
 961         }
 962
 963         /* copy up to skb headlen */
 964         while (count && (copy > 0)) {
 965                 size = min_t(unsigned int, copy, from->iov_len - offset);
 966                 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
 967                                    size))
 968                         return -EFAULT;
 969                 if (copy > size) {
 970                         ++from;
 971                         --count;
 972                         offset = 0;
 973                 } else
 974                         offset += size;
 975                 copy -= size;
 976                 offset1 += size;
 977         }
 978
 979         if (len == offset1)
 980                 return 0;
 981
 982         while (count--) {
 983                 struct page *page[MAX_SKB_FRAGS];
 984                 int num_pages;
 985                 unsigned long base;
 986                 unsigned long truesize;
 987
 988                 len = from->iov_len - offset;
 989                 if (!len) {
 990                         offset = 0;
 991                         ++from;
 992                         continue;
 993                 }
 994                 base = (unsigned long)from->iov_base + offset;
 995                 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
 996                 if (i + size > MAX_SKB_FRAGS)
 997                         return -EMSGSIZE;
 998                 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
 999                 if (num_pages != size) {
1000                         for (i = 0; i < num_pages; i++)
1001                                 put_page(page[i]);
1002                         return -EFAULT;
1003                 }
1004                 truesize = size * PAGE_SIZE;
1005                 skb->data_len += len;
1006                 skb->len += len;
1007                 skb->truesize += truesize;
1008                 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
1009                 while (len) {
1010                         int off = base & ~PAGE_MASK;
1011                         int size = min_t(int, len, PAGE_SIZE - off);
1012                         __skb_fill_page_desc(skb, i, page[i], off, size);
1013                         skb_shinfo(skb)->nr_frags++;
1014                         /* increase sk_wmem_alloc */
1015                         base += size;
1016                         len -= size;
1017                         i++;
1018                 }
1019                 offset = 0;
1020                 ++from;
1021         }
1022         return 0;
1023 }
1024
1025 /* Get packet from user space buffer */
1026 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1027                             void *msg_control, const struct iovec *iv,
1028                             size_t total_len, size_t count, int noblock)
1029 {
1030         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1031         struct sk_buff *skb;
1032         size_t len = total_len, align = NET_SKB_PAD;
1033         struct virtio_net_hdr gso = { 0 };
1034         int offset = 0;
1035         int copylen;
1036         bool zerocopy = false;
1037         int err;
1038         u32 rxhash;
1039
1040         if (!(tun->flags & TUN_NO_PI)) {
1041                 if ((len -= sizeof(pi)) > total_len)
1042                         return -EINVAL;
1043
1044                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1045                         return -EFAULT;
1046                 offset += sizeof(pi);
1047         }
1048
1049         if (tun->flags & TUN_VNET_HDR) {
1050                 if ((len -= tun->vnet_hdr_sz) > total_len)
1051                         return -EINVAL;
1052
1053                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1054                         return -EFAULT;
1055
1056                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1057                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1058                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1059
1060                 if (gso.hdr_len > len)
1061                         return -EINVAL;
1062                 offset += tun->vnet_hdr_sz;
1063         }
1064
1065         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1066                 align += NET_IP_ALIGN;
1067                 if (unlikely(len < ETH_HLEN ||
1068                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1069                         return -EINVAL;
1070         }
1071
1072         if (msg_control)
1073                 zerocopy = true;
1074
1075         if (zerocopy) {
1076                 /* Userspace may produce vectors with count greater than
1077                  * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1078                  * to let the rest of data to be fit in the frags.
1079                  */
1080                 if (count > MAX_SKB_FRAGS) {
1081                         copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1082                         if (copylen < offset)
1083                                 copylen = 0;
1084                         else
1085                                 copylen -= offset;
1086                 } else
1087                                 copylen = 0;
1088                 /* There are 256 bytes to be copied in skb, so there is enough
1089                  * room for skb expand head in case it is used.
1090                  * The rest of the buffer is mapped from userspace.
1091                  */
1092                 if (copylen < gso.hdr_len)
1093                         copylen = gso.hdr_len;
1094                 if (!copylen)
1095                         copylen = GOODCOPY_LEN;
1096         } else
1097                 copylen = len;
1098
1099         skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
1100         if (IS_ERR(skb)) {
1101                 if (PTR_ERR(skb) != -EAGAIN)
1102                         tun->dev->stats.rx_dropped++;
1103                 return PTR_ERR(skb);
1104         }
1105
1106         if (zerocopy)
1107                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1108         else
1109                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1110
1111         if (err) {
1112                 tun->dev->stats.rx_dropped++;
1113                 kfree_skb(skb);
1114                 return -EFAULT;
1115         }
1116
1117         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1118                 if (!skb_partial_csum_set(skb, gso.csum_start,
1119                                           gso.csum_offset)) {
1120                         tun->dev->stats.rx_frame_errors++;
1121                         kfree_skb(skb);
1122                         return -EINVAL;
1123                 }
1124         }
1125
1126         switch (tun->flags & TUN_TYPE_MASK) {
1127         case TUN_TUN_DEV:
1128                 if (tun->flags & TUN_NO_PI) {
1129                         switch (skb->data[0] & 0xf0) {
1130                         case 0x40:
1131                                 pi.proto = htons(ETH_P_IP);
1132                                 break;
1133                         case 0x60:
1134                                 pi.proto = htons(ETH_P_IPV6);
1135                                 break;
1136                         default:
1137                                 tun->dev->stats.rx_dropped++;
1138                                 kfree_skb(skb);
1139                                 return -EINVAL;
1140                         }
1141                 }
1142
1143                 skb_reset_mac_header(skb);
1144                 skb->protocol = pi.proto;
1145                 skb->dev = tun->dev;
1146                 break;
1147         case TUN_TAP_DEV:
1148                 skb->protocol = eth_type_trans(skb, tun->dev);
1149                 break;
1150         }
1151
1152         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1153                 pr_debug("GSO!\n");
1154                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1155                 case VIRTIO_NET_HDR_GSO_TCPV4:
1156                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1157                         break;
1158                 case VIRTIO_NET_HDR_GSO_TCPV6:
1159                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1160                         break;
1161                 case VIRTIO_NET_HDR_GSO_UDP:
1162                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1163                         break;
1164                 default:
1165                         tun->dev->stats.rx_frame_errors++;
1166                         kfree_skb(skb);
1167                         return -EINVAL;
1168                 }
1169
1170                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1171                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1172
1173                 skb_shinfo(skb)->gso_size = gso.gso_size;
1174                 if (skb_shinfo(skb)->gso_size == 0) {
1175                         tun->dev->stats.rx_frame_errors++;
1176                         kfree_skb(skb);
1177                         return -EINVAL;
1178                 }
1179
1180                 /* Header must be checked, and gso_segs computed. */
1181                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1182                 skb_shinfo(skb)->gso_segs = 0;
1183         }
1184
1185         /* copy skb_ubuf_info for callback when skb has no error */
1186         if (zerocopy) {
1187                 skb_shinfo(skb)->destructor_arg = msg_control;
1188                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1189         }
1190
1191         skb_reset_network_header(skb);
1192         rxhash = skb_get_rxhash(skb);
1193         netif_rx_ni(skb);
1194
1195         tun->dev->stats.rx_packets++;
1196         tun->dev->stats.rx_bytes += len;
1197
1198         tun_flow_update(tun, rxhash, tfile->queue_index);
1199         return total_len;
1200 }
1201
1202 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1203                               unsigned long count, loff_t pos)
1204 {
1205         struct file *file = iocb->ki_filp;
1206         struct tun_struct *tun = tun_get(file);
1207         struct tun_file *tfile = file->private_data;
1208         ssize_t result;
1209
1210         if (!tun)
1211                 return -EBADFD;
1212
1213         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1214
1215         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1216                               count, file->f_flags & O_NONBLOCK);
1217
1218         tun_put(tun);
1219         return result;
1220 }
1221
1222 /* Put packet to the user space buffer */
1223 static ssize_t tun_put_user(struct tun_struct *tun,
1224                             struct tun_file *tfile,
1225                             struct sk_buff *skb,
1226                             const struct iovec *iv, int len)
1227 {
1228         struct tun_pi pi = { 0, skb->protocol };
1229         ssize_t total = 0;
1230
1231         if (!(tun->flags & TUN_NO_PI)) {
1232                 if ((len -= sizeof(pi)) < 0)
1233                         return -EINVAL;
1234
1235                 if (len < skb->len) {
1236                         /* Packet will be striped */
1237                         pi.flags |= TUN_PKT_STRIP;
1238                 }
1239
1240                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1241                         return -EFAULT;
1242                 total += sizeof(pi);
1243         }
1244
1245         if (tun->flags & TUN_VNET_HDR) {
1246                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1247                 if ((len -= tun->vnet_hdr_sz) < 0)
1248                         return -EINVAL;
1249
1250                 if (skb_is_gso(skb)) {
1251                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1252
1253                         /* This is a hint as to how much should be linear. */
1254                         gso.hdr_len = skb_headlen(skb);
1255                         gso.gso_size = sinfo->gso_size;
1256                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1257                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1258                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1259                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1260                         else if (sinfo->gso_type & SKB_GSO_UDP)
1261                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1262                         else {
1263                                 pr_err("unexpected GSO type: "
1264                                        "0x%x, gso_size %d, hdr_len %d\n",
1265                                        sinfo->gso_type, gso.gso_size,
1266                                        gso.hdr_len);
1267                                 print_hex_dump(KERN_ERR, "tun: ",
1268                                                DUMP_PREFIX_NONE,
1269                                                16, 1, skb->head,
1270                                                min((int)gso.hdr_len, 64), true);
1271                                 WARN_ON_ONCE(1);
1272                                 return -EINVAL;
1273                         }
1274                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1275                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1276                 } else
1277                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1278
1279                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1280                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1281                         gso.csum_start = skb_checksum_start_offset(skb);
1282                         gso.csum_offset = skb->csum_offset;
1283                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1284                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1285                 } /* else everything is zero */
1286
1287                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1288                                                sizeof(gso))))
1289                         return -EFAULT;
1290                 total += tun->vnet_hdr_sz;
1291         }
1292
1293         len = min_t(int, skb->len, len);
1294
1295         skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1296         total += skb->len;
1297
1298         tun->dev->stats.tx_packets++;
1299         tun->dev->stats.tx_bytes += len;
1300
1301         return total;
1302 }
1303
1304 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1305                            struct kiocb *iocb, const struct iovec *iv,
1306                            ssize_t len, int noblock)
1307 {
1308         DECLARE_WAITQUEUE(wait, current);
1309         struct sk_buff *skb;
1310         ssize_t ret = 0;
1311
1312         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1313
1314         if (unlikely(!noblock))
1315                 add_wait_queue(&tfile->wq.wait, &wait);
1316         while (len) {
1317                 current->state = TASK_INTERRUPTIBLE;
1318
1319                 /* Read frames from the queue */
1320                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1321                         if (noblock) {
1322                                 ret = -EAGAIN;
1323                                 break;
1324                         }
1325                         if (signal_pending(current)) {
1326                                 ret = -ERESTARTSYS;
1327                                 break;
1328                         }
1329                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1330                                 ret = -EIO;
1331                                 break;
1332                         }
1333
1334                         /* Nothing to read, let's sleep */
1335                         schedule();
1336                         continue;
1337                 }
1338
1339                 ret = tun_put_user(tun, tfile, skb, iv, len);
1340                 kfree_skb(skb);
1341                 break;
1342         }
1343
1344         current->state = TASK_RUNNING;
1345         if (unlikely(!noblock))
1346                 remove_wait_queue(&tfile->wq.wait, &wait);
1347
1348         return ret;
1349 }
1350
1351 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1352                             unsigned long count, loff_t pos)
1353 {
1354         struct file *file = iocb->ki_filp;
1355         struct tun_file *tfile = file->private_data;
1356         struct tun_struct *tun = __tun_get(tfile);
1357         ssize_t len, ret;
1358
1359         if (!tun)
1360                 return -EBADFD;
1361         len = iov_length(iv, count);
1362         if (len < 0) {
1363                 ret = -EINVAL;
1364                 goto out;
1365         }
1366
1367         ret = tun_do_read(tun, tfile, iocb, iv, len,
1368                           file->f_flags & O_NONBLOCK);
1369         ret = min_t(ssize_t, ret, len);
1370 out:
1371         tun_put(tun);
1372         return ret;
1373 }
1374
1375 static void tun_free_netdev(struct net_device *dev)
1376 {
1377         struct tun_struct *tun = netdev_priv(dev);
1378
1379         BUG_ON(!(list_empty(&tun->disabled)));
1380         tun_flow_uninit(tun);
1381         security_tun_dev_free_security(tun->security);
1382         free_netdev(dev);
1383 }
1384
1385 static void tun_setup(struct net_device *dev)
1386 {
1387         struct tun_struct *tun = netdev_priv(dev);
1388
1389         tun->owner = INVALID_UID;
1390         tun->group = INVALID_GID;
1391
1392         dev->ethtool_ops = &tun_ethtool_ops;
1393         dev->destructor = tun_free_netdev;
1394 }
1395
1396 /* Trivial set of netlink ops to allow deleting tun or tap
1397  * device with netlink.
1398  */
1399 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1400 {
1401         return -EINVAL;
1402 }
1403
1404 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1405         .kind           = DRV_NAME,
1406         .priv_size      = sizeof(struct tun_struct),
1407         .setup          = tun_setup,
1408         .validate       = tun_validate,
1409 };
1410
1411 static void tun_sock_write_space(struct sock *sk)
1412 {
1413         struct tun_file *tfile;
1414         wait_queue_head_t *wqueue;
1415
1416         if (!sock_writeable(sk))
1417                 return;
1418
1419         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1420                 return;
1421
1422         wqueue = sk_sleep(sk);
1423         if (wqueue && waitqueue_active(wqueue))
1424                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1425                                                 POLLWRNORM | POLLWRBAND);
1426
1427         tfile = container_of(sk, struct tun_file, sk);
1428         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1429 }
1430
1431 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1432                        struct msghdr *m, size_t total_len)
1433 {
1434         int ret;
1435         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1436         struct tun_struct *tun = __tun_get(tfile);
1437
1438         if (!tun)
1439                 return -EBADFD;
1440         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1441                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1442         tun_put(tun);
1443         return ret;
1444 }
1445
1446
1447 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1448                        struct msghdr *m, size_t total_len,
1449                        int flags)
1450 {
1451         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1452         struct tun_struct *tun = __tun_get(tfile);
1453         int ret;
1454
1455         if (!tun)
1456                 return -EBADFD;
1457
1458         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1459                 return -EINVAL;
1460         ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1461                           flags & MSG_DONTWAIT);
1462         if (ret > total_len) {
1463                 m->msg_flags |= MSG_TRUNC;
1464                 ret = flags & MSG_TRUNC ? ret : total_len;
1465         }
1466         tun_put(tun);
1467         return ret;
1468 }
1469
1470 static int tun_release(struct socket *sock)
1471 {
1472         if (sock->sk)
1473                 sock_put(sock->sk);
1474         return 0;
1475 }
1476
1477 /* Ops structure to mimic raw sockets with tun */
1478 static const struct proto_ops tun_socket_ops = {
1479         .sendmsg = tun_sendmsg,
1480         .recvmsg = tun_recvmsg,
1481         .release = tun_release,
1482 };
1483
1484 static struct proto tun_proto = {
1485         .name           = "tun",
1486         .owner          = THIS_MODULE,
1487         .obj_size       = sizeof(struct tun_file),
1488 };
1489
1490 static int tun_flags(struct tun_struct *tun)
1491 {
1492         int flags = 0;
1493
1494         if (tun->flags & TUN_TUN_DEV)
1495                 flags |= IFF_TUN;
1496         else
1497                 flags |= IFF_TAP;
1498
1499         if (tun->flags & TUN_NO_PI)
1500                 flags |= IFF_NO_PI;
1501
1502         /* This flag has no real effect.  We track the value for backwards
1503          * compatibility.
1504          */
1505         if (tun->flags & TUN_ONE_QUEUE)
1506                 flags |= IFF_ONE_QUEUE;
1507
1508         if (tun->flags & TUN_VNET_HDR)
1509                 flags |= IFF_VNET_HDR;
1510
1511         if (tun->flags & TUN_TAP_MQ)
1512                 flags |= IFF_MULTI_QUEUE;
1513
1514         return flags;
1515 }
1516
1517 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1518                               char *buf)
1519 {
1520         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1521         return sprintf(buf, "0x%x\n", tun_flags(tun));
1522 }
1523
1524 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1525                               char *buf)
1526 {
1527         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1528         return uid_valid(tun->owner)?
1529                 sprintf(buf, "%u\n",
1530                         from_kuid_munged(current_user_ns(), tun->owner)):
1531                 sprintf(buf, "-1\n");
1532 }
1533
1534 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1535                               char *buf)
1536 {
1537         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1538         return gid_valid(tun->group) ?
1539                 sprintf(buf, "%u\n",
1540                         from_kgid_munged(current_user_ns(), tun->group)):
1541                 sprintf(buf, "-1\n");
1542 }
1543
1544 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1545 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1546 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1547
1548 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1549 {
1550         struct tun_struct *tun;
1551         struct tun_file *tfile = file->private_data;
1552         struct net_device *dev;
1553         int err;
1554
1555         if (tfile->detached)
1556                 return -EINVAL;
1557
1558         dev = __dev_get_by_name(net, ifr->ifr_name);
1559         if (dev) {
1560                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1561                         return -EBUSY;
1562                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1563                         tun = netdev_priv(dev);
1564                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1565                         tun = netdev_priv(dev);
1566                 else
1567                         return -EINVAL;
1568
1569                 if (tun_not_capable(tun))
1570                         return -EPERM;
1571                 err = security_tun_dev_open(tun->security);
1572                 if (err < 0)
1573                         return err;
1574
1575                 err = tun_attach(tun, file);
1576                 if (err < 0)
1577                         return err;
1578
1579                 if (tun->flags & TUN_TAP_MQ &&
1580                     (tun->numqueues + tun->numdisabled > 1))
1581                         return err;
1582         }
1583         else {
1584                 char *name;
1585                 unsigned long flags = 0;
1586
1587                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1588                         return -EPERM;
1589                 err = security_tun_dev_create();
1590                 if (err < 0)
1591                         return err;
1592
1593                 /* Set dev type */
1594                 if (ifr->ifr_flags & IFF_TUN) {
1595                         /* TUN device */
1596                         flags |= TUN_TUN_DEV;
1597                         name = "tun%d";
1598                 } else if (ifr->ifr_flags & IFF_TAP) {
1599                         /* TAP device */
1600                         flags |= TUN_TAP_DEV;
1601                         name = "tap%d";
1602                 } else
1603                         return -EINVAL;
1604
1605                 if (*ifr->ifr_name)
1606                         name = ifr->ifr_name;
1607
1608                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1609                                        tun_setup,
1610                                        MAX_TAP_QUEUES, MAX_TAP_QUEUES);
1611                 if (!dev)
1612                         return -ENOMEM;
1613
1614                 dev_net_set(dev, net);
1615                 dev->rtnl_link_ops = &tun_link_ops;
1616
1617                 tun = netdev_priv(dev);
1618                 tun->dev = dev;
1619                 tun->flags = flags;
1620                 tun->txflt.count = 0;
1621                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1622
1623                 tun->filter_attached = false;
1624                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1625
1626                 spin_lock_init(&tun->lock);
1627
1628                 err = security_tun_dev_alloc_security(&tun->security);
1629                 if (err < 0)
1630                         goto err_free_dev;
1631
1632                 tun_net_init(dev);
1633
1634                 err = tun_flow_init(tun);
1635                 if (err < 0)
1636                         goto err_free_dev;
1637
1638                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1639                         TUN_USER_FEATURES;
1640                 dev->features = dev->hw_features;
1641
1642                 INIT_LIST_HEAD(&tun->disabled);
1643                 err = tun_attach(tun, file);
1644                 if (err < 0)
1645                         goto err_free_dev;
1646
1647                 err = register_netdevice(tun->dev);
1648                 if (err < 0)
1649                         goto err_free_dev;
1650
1651                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1652                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1653                     device_create_file(&tun->dev->dev, &dev_attr_group))
1654                         pr_err("Failed to create tun sysfs files\n");
1655
1656                 netif_carrier_on(tun->dev);
1657         }
1658
1659         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1660
1661         if (ifr->ifr_flags & IFF_NO_PI)
1662                 tun->flags |= TUN_NO_PI;
1663         else
1664                 tun->flags &= ~TUN_NO_PI;
1665
1666         /* This flag has no real effect.  We track the value for backwards
1667          * compatibility.
1668          */
1669         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1670                 tun->flags |= TUN_ONE_QUEUE;
1671         else
1672                 tun->flags &= ~TUN_ONE_QUEUE;
1673
1674         if (ifr->ifr_flags & IFF_VNET_HDR)
1675                 tun->flags |= TUN_VNET_HDR;
1676         else
1677                 tun->flags &= ~TUN_VNET_HDR;
1678
1679         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1680                 tun->flags |= TUN_TAP_MQ;
1681         else
1682                 tun->flags &= ~TUN_TAP_MQ;
1683
1684         /* Make sure persistent devices do not get stuck in
1685          * xoff state.
1686          */
1687         if (netif_running(tun->dev))
1688                 netif_tx_wake_all_queues(tun->dev);
1689
1690         strcpy(ifr->ifr_name, tun->dev->name);
1691         return 0;
1692
1693  err_free_dev:
1694         free_netdev(dev);
1695         return err;
1696 }
1697
1698 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1699                        struct ifreq *ifr)
1700 {
1701         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1702
1703         strcpy(ifr->ifr_name, tun->dev->name);
1704
1705         ifr->ifr_flags = tun_flags(tun);
1706
1707 }
1708
1709 /* This is like a cut-down ethtool ops, except done via tun fd so no
1710  * privs required. */
1711 static int set_offload(struct tun_struct *tun, unsigned long arg)
1712 {
1713         netdev_features_t features = 0;
1714
1715         if (arg & TUN_F_CSUM) {
1716                 features |= NETIF_F_HW_CSUM;
1717                 arg &= ~TUN_F_CSUM;
1718
1719                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1720                         if (arg & TUN_F_TSO_ECN) {
1721                                 features |= NETIF_F_TSO_ECN;
1722                                 arg &= ~TUN_F_TSO_ECN;
1723                         }
1724                         if (arg & TUN_F_TSO4)
1725                                 features |= NETIF_F_TSO;
1726                         if (arg & TUN_F_TSO6)
1727                                 features |= NETIF_F_TSO6;
1728                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1729                 }
1730
1731                 if (arg & TUN_F_UFO) {
1732                         features |= NETIF_F_UFO;
1733                         arg &= ~TUN_F_UFO;
1734                 }
1735         }
1736
1737         /* This gives the user a way to test for new features in future by
1738          * trying to set them. */
1739         if (arg)
1740                 return -EINVAL;
1741
1742         tun->set_features = features;
1743         netdev_update_features(tun->dev);
1744
1745         return 0;
1746 }
1747
1748 static void tun_detach_filter(struct tun_struct *tun, int n)
1749 {
1750         int i;
1751         struct tun_file *tfile;
1752
1753         for (i = 0; i < n; i++) {
1754                 tfile = rtnl_dereference(tun->tfiles[i]);
1755                 sk_detach_filter(tfile->socket.sk);
1756         }
1757
1758         tun->filter_attached = false;
1759 }
1760
1761 static int tun_attach_filter(struct tun_struct *tun)
1762 {
1763         int i, ret = 0;
1764         struct tun_file *tfile;
1765
1766         for (i = 0; i < tun->numqueues; i++) {
1767                 tfile = rtnl_dereference(tun->tfiles[i]);
1768                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1769                 if (ret) {
1770                         tun_detach_filter(tun, i);
1771                         return ret;
1772                 }
1773         }
1774
1775         tun->filter_attached = true;
1776         return ret;
1777 }
1778
1779 static void tun_set_sndbuf(struct tun_struct *tun)
1780 {
1781         struct tun_file *tfile;
1782         int i;
1783
1784         for (i = 0; i < tun->numqueues; i++) {
1785                 tfile = rtnl_dereference(tun->tfiles[i]);
1786                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1787         }
1788 }
1789
1790 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1791 {
1792         struct tun_file *tfile = file->private_data;
1793         struct tun_struct *tun;
1794         int ret = 0;
1795
1796         rtnl_lock();
1797
1798         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1799                 tun = tfile->detached;
1800                 if (!tun) {
1801                         ret = -EINVAL;
1802                         goto unlock;
1803                 }
1804                 ret = security_tun_dev_attach_queue(tun->security);
1805                 if (ret < 0)
1806                         goto unlock;
1807                 ret = tun_attach(tun, file);
1808         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1809                 tun = rtnl_dereference(tfile->tun);
1810                 if (!tun || !(tun->flags & TUN_TAP_MQ))
1811                         ret = -EINVAL;
1812                 else
1813                         __tun_detach(tfile, false);
1814         } else
1815                 ret = -EINVAL;
1816
1817 unlock:
1818         rtnl_unlock();
1819         return ret;
1820 }
1821
1822 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1823                             unsigned long arg, int ifreq_len)
1824 {
1825         struct tun_file *tfile = file->private_data;
1826         struct tun_struct *tun;
1827         void __user* argp = (void __user*)arg;
1828         struct ifreq ifr;
1829         kuid_t owner;
1830         kgid_t group;
1831         int sndbuf;
1832         int vnet_hdr_sz;
1833         int ret;
1834
1835         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1836                 if (copy_from_user(&ifr, argp, ifreq_len))
1837                         return -EFAULT;
1838         } else {
1839                 memset(&ifr, 0, sizeof(ifr));
1840         }
1841         if (cmd == TUNGETFEATURES) {
1842                 /* Currently this just means: "what IFF flags are valid?".
1843                  * This is needed because we never checked for invalid flags on
1844                  * TUNSETIFF. */
1845                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1846                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1847                                 (unsigned int __user*)argp);
1848         } else if (cmd == TUNSETQUEUE)
1849                 return tun_set_queue(file, &ifr);
1850
1851         ret = 0;
1852         rtnl_lock();
1853
1854         tun = __tun_get(tfile);
1855         if (cmd == TUNSETIFF && !tun) {
1856                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1857
1858                 ret = tun_set_iff(tfile->net, file, &ifr);
1859
1860                 if (ret)
1861                         goto unlock;
1862
1863                 if (copy_to_user(argp, &ifr, ifreq_len))
1864                         ret = -EFAULT;
1865                 goto unlock;
1866         }
1867
1868         ret = -EBADFD;
1869         if (!tun)
1870                 goto unlock;
1871
1872         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1873
1874         ret = 0;
1875         switch (cmd) {
1876         case TUNGETIFF:
1877                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1878
1879                 if (copy_to_user(argp, &ifr, ifreq_len))
1880                         ret = -EFAULT;
1881                 break;
1882
1883         case TUNSETNOCSUM:
1884                 /* Disable/Enable checksum */
1885
1886                 /* [unimplemented] */
1887                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1888                           arg ? "disabled" : "enabled");
1889                 break;
1890
1891         case TUNSETPERSIST:
1892                 /* Disable/Enable persist mode. Keep an extra reference to the
1893                  * module to prevent the module being unprobed.
1894                  */
1895                 if (arg && !(tun->flags & TUN_PERSIST)) {
1896                         tun->flags |= TUN_PERSIST;
1897                         __module_get(THIS_MODULE);
1898                 }
1899                 if (!arg && (tun->flags & TUN_PERSIST)) {
1900                         tun->flags &= ~TUN_PERSIST;
1901                         module_put(THIS_MODULE);
1902                 }
1903
1904                 tun_debug(KERN_INFO, tun, "persist %s\n",
1905                           arg ? "enabled" : "disabled");
1906                 break;
1907
1908         case TUNSETOWNER:
1909                 /* Set owner of the device */
1910                 owner = make_kuid(current_user_ns(), arg);
1911                 if (!uid_valid(owner)) {
1912                         ret = -EINVAL;
1913                         break;
1914                 }
1915                 tun->owner = owner;
1916                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1917                           from_kuid(&init_user_ns, tun->owner));
1918                 break;
1919
1920         case TUNSETGROUP:
1921                 /* Set group of the device */
1922                 group = make_kgid(current_user_ns(), arg);
1923                 if (!gid_valid(group)) {
1924                         ret = -EINVAL;
1925                         break;
1926                 }
1927                 tun->group = group;
1928                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1929                           from_kgid(&init_user_ns, tun->group));
1930                 break;
1931
1932         case TUNSETLINK:
1933                 /* Only allow setting the type when the interface is down */
1934                 if (tun->dev->flags & IFF_UP) {
1935                         tun_debug(KERN_INFO, tun,
1936                                   "Linktype set failed because interface is up\n");
1937                         ret = -EBUSY;
1938                 } else {
1939                         tun->dev->type = (int) arg;
1940                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1941                                   tun->dev->type);
1942                         ret = 0;
1943                 }
1944                 break;
1945
1946 #ifdef TUN_DEBUG
1947         case TUNSETDEBUG:
1948                 tun->debug = arg;
1949                 break;
1950 #endif
1951         case TUNSETOFFLOAD:
1952                 ret = set_offload(tun, arg);
1953                 break;
1954
1955         case TUNSETTXFILTER:
1956                 /* Can be set only for TAPs */
1957                 ret = -EINVAL;
1958                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1959                         break;
1960                 ret = update_filter(&tun->txflt, (void __user *)arg);
1961                 break;
1962
1963         case SIOCGIFHWADDR:
1964                 /* Get hw address */
1965                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1966                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1967                 if (copy_to_user(argp, &ifr, ifreq_len))
1968                         ret = -EFAULT;
1969                 break;
1970
1971         case SIOCSIFHWADDR:
1972                 /* Set hw address */
1973                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1974                           ifr.ifr_hwaddr.sa_data);
1975
1976                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1977                 break;
1978
1979         case TUNGETSNDBUF:
1980                 sndbuf = tfile->socket.sk->sk_sndbuf;
1981                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1982                         ret = -EFAULT;
1983                 break;
1984
1985         case TUNSETSNDBUF:
1986                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1987                         ret = -EFAULT;
1988                         break;
1989                 }
1990
1991                 tun->sndbuf = sndbuf;
1992                 tun_set_sndbuf(tun);
1993                 break;
1994
1995         case TUNGETVNETHDRSZ:
1996                 vnet_hdr_sz = tun->vnet_hdr_sz;
1997                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1998                         ret = -EFAULT;
1999                 break;
2000
2001         case TUNSETVNETHDRSZ:
2002                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2003                         ret = -EFAULT;
2004                         break;
2005                 }
2006                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2007                         ret = -EINVAL;
2008                         break;
2009                 }
2010
2011                 tun->vnet_hdr_sz = vnet_hdr_sz;
2012                 break;
2013
2014         case TUNATTACHFILTER:
2015                 /* Can be set only for TAPs */
2016                 ret = -EINVAL;
2017                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2018                         break;
2019                 ret = -EFAULT;
2020                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2021                         break;
2022
2023                 ret = tun_attach_filter(tun);
2024                 break;
2025
2026         case TUNDETACHFILTER:
2027                 /* Can be set only for TAPs */
2028                 ret = -EINVAL;
2029                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2030                         break;
2031                 ret = 0;
2032                 tun_detach_filter(tun, tun->numqueues);
2033                 break;
2034
2035         default:
2036                 ret = -EINVAL;
2037                 break;
2038         }
2039
2040 unlock:
2041         rtnl_unlock();
2042         if (tun)
2043                 tun_put(tun);
2044         return ret;
2045 }
2046
2047 static long tun_chr_ioctl(struct file *file,
2048                           unsigned int cmd, unsigned long arg)
2049 {
2050         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2051 }
2052
2053 #ifdef CONFIG_COMPAT
2054 static long tun_chr_compat_ioctl(struct file *file,
2055                          unsigned int cmd, unsigned long arg)
2056 {
2057         switch (cmd) {
2058         case TUNSETIFF:
2059         case TUNGETIFF:
2060         case TUNSETTXFILTER:
2061         case TUNGETSNDBUF:
2062         case TUNSETSNDBUF:
2063         case SIOCGIFHWADDR:
2064         case SIOCSIFHWADDR:
2065                 arg = (unsigned long)compat_ptr(arg);
2066                 break;
2067         default:
2068                 arg = (compat_ulong_t)arg;
2069                 break;
2070         }
2071
2072         /*
2073          * compat_ifreq is shorter than ifreq, so we must not access beyond
2074          * the end of that structure. All fields that are used in this
2075          * driver are compatible though, we don't need to convert the
2076          * contents.
2077          */
2078         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2079 }
2080 #endif /* CONFIG_COMPAT */
2081
2082 static int tun_chr_fasync(int fd, struct file *file, int on)
2083 {
2084         struct tun_file *tfile = file->private_data;
2085         int ret;
2086
2087         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2088                 goto out;
2089
2090         if (on) {
2091                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2092                 if (ret)
2093                         goto out;
2094                 tfile->flags |= TUN_FASYNC;
2095         } else
2096                 tfile->flags &= ~TUN_FASYNC;
2097         ret = 0;
2098 out:
2099         return ret;
2100 }
2101
2102 static int tun_chr_open(struct inode *inode, struct file * file)
2103 {
2104         struct tun_file *tfile;
2105
2106         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2107
2108         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2109                                             &tun_proto);
2110         if (!tfile)
2111                 return -ENOMEM;
2112         rcu_assign_pointer(tfile->tun, NULL);
2113         tfile->net = get_net(current->nsproxy->net_ns);
2114         tfile->flags = 0;
2115
2116         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2117         init_waitqueue_head(&tfile->wq.wait);
2118
2119         tfile->socket.file = file;
2120         tfile->socket.ops = &tun_socket_ops;
2121
2122         sock_init_data(&tfile->socket, &tfile->sk);
2123         sk_change_net(&tfile->sk, tfile->net);
2124
2125         tfile->sk.sk_write_space = tun_sock_write_space;
2126         tfile->sk.sk_sndbuf = INT_MAX;
2127
2128         file->private_data = tfile;
2129         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2130         INIT_LIST_HEAD(&tfile->next);
2131
2132         return 0;
2133 }
2134
2135 static int tun_chr_close(struct inode *inode, struct file *file)
2136 {
2137         struct tun_file *tfile = file->private_data;
2138         struct net *net = tfile->net;
2139
2140         tun_detach(tfile, true);
2141         put_net(net);
2142
2143         return 0;
2144 }
2145
2146 static const struct file_operations tun_fops = {
2147         .owner  = THIS_MODULE,
2148         .llseek = no_llseek,
2149         .read  = do_sync_read,
2150         .aio_read  = tun_chr_aio_read,
2151         .write = do_sync_write,
2152         .aio_write = tun_chr_aio_write,
2153         .poll   = tun_chr_poll,
2154         .unlocked_ioctl = tun_chr_ioctl,
2155 #ifdef CONFIG_COMPAT
2156         .compat_ioctl = tun_chr_compat_ioctl,
2157 #endif
2158         .open   = tun_chr_open,
2159         .release = tun_chr_close,
2160         .fasync = tun_chr_fasync
2161 };
2162
2163 static struct miscdevice tun_miscdev = {
2164         .minor = TUN_MINOR,
2165         .name = "tun",
2166         .nodename = "net/tun",
2167         .fops = &tun_fops,
2168 };
2169
2170 /* ethtool interface */
2171
2172 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2173 {
2174         cmd->supported          = 0;
2175         cmd->advertising        = 0;
2176         ethtool_cmd_speed_set(cmd, SPEED_10);
2177         cmd->duplex             = DUPLEX_FULL;
2178         cmd->port               = PORT_TP;
2179         cmd->phy_address        = 0;
2180         cmd->transceiver        = XCVR_INTERNAL;
2181         cmd->autoneg            = AUTONEG_DISABLE;
2182         cmd->maxtxpkt           = 0;
2183         cmd->maxrxpkt           = 0;
2184         return 0;
2185 }
2186
2187 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2188 {
2189         struct tun_struct *tun = netdev_priv(dev);
2190
2191         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2192         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2193
2194         switch (tun->flags & TUN_TYPE_MASK) {
2195         case TUN_TUN_DEV:
2196                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2197                 break;
2198         case TUN_TAP_DEV:
2199                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2200                 break;
2201         }
2202 }
2203
2204 static u32 tun_get_msglevel(struct net_device *dev)
2205 {
2206 #ifdef TUN_DEBUG
2207         struct tun_struct *tun = netdev_priv(dev);
2208         return tun->debug;
2209 #else
2210         return -EOPNOTSUPP;
2211 #endif
2212 }
2213
2214 static void tun_set_msglevel(struct net_device *dev, u32 value)
2215 {
2216 #ifdef TUN_DEBUG
2217         struct tun_struct *tun = netdev_priv(dev);
2218         tun->debug = value;
2219 #endif
2220 }
2221
2222 static const struct ethtool_ops tun_ethtool_ops = {
2223         .get_settings   = tun_get_settings,
2224         .get_drvinfo    = tun_get_drvinfo,
2225         .get_msglevel   = tun_get_msglevel,
2226         .set_msglevel   = tun_set_msglevel,
2227         .get_link       = ethtool_op_get_link,
2228 };
2229
2230
2231 static int __init tun_init(void)
2232 {
2233         int ret = 0;
2234
2235         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2236         pr_info("%s\n", DRV_COPYRIGHT);
2237
2238         ret = rtnl_link_register(&tun_link_ops);
2239         if (ret) {
2240                 pr_err("Can't register link_ops\n");
2241                 goto err_linkops;
2242         }
2243
2244         ret = misc_register(&tun_miscdev);
2245         if (ret) {
2246                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2247                 goto err_misc;
2248         }
2249         return  0;
2250 err_misc:
2251         rtnl_link_unregister(&tun_link_ops);
2252 err_linkops:
2253         return ret;
2254 }
2255
2256 static void tun_cleanup(void)
2257 {
2258         misc_deregister(&tun_miscdev);
2259         rtnl_link_unregister(&tun_link_ops);
2260 }
2261
2262 /* Get an underlying socket object from tun file.  Returns error unless file is
2263  * attached to a device.  The returned object works like a packet socket, it
2264  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2265  * holding a reference to the file for as long as the socket is in use. */
2266 struct socket *tun_get_socket(struct file *file)
2267 {
2268         struct tun_file *tfile;
2269         if (file->f_op != &tun_fops)
2270                 return ERR_PTR(-EINVAL);
2271         tfile = file->private_data;
2272         if (!tfile)
2273                 return ERR_PTR(-EBADFD);
2274         return &tfile->socket;
2275 }
2276 EXPORT_SYMBOL_GPL(tun_get_socket);
2277
2278 module_init(tun_init);
2279 module_exit(tun_cleanup);
2280 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2281 MODULE_AUTHOR(DRV_COPYRIGHT);
2282 MODULE_LICENSE("GPL");
2283 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2284 MODULE_ALIAS("devname:net/tun");