Pileus Git - ~andy/linux/blob - drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/major.h>
  48 #include <linux/slab.h>
  49 #include <linux/poll.h>
  50 #include <linux/fcntl.h>
  51 #include <linux/init.h>
  52 #include <linux/skbuff.h>
  53 #include <linux/netdevice.h>
  54 #include <linux/etherdevice.h>
  55 #include <linux/miscdevice.h>
  56 #include <linux/ethtool.h>
  57 #include <linux/rtnetlink.h>
  58 #include <linux/compat.h>
  59 #include <linux/if.h>
  60 #include <linux/if_arp.h>
  61 #include <linux/if_ether.h>
  62 #include <linux/if_tun.h>
  63 #include <linux/if_vlan.h>
  64 #include <linux/crc32.h>
  65 #include <linux/nsproxy.h>
  66 #include <linux/virtio_net.h>
  67 #include <linux/rcupdate.h>
  68 #include <net/net_namespace.h>
  69 #include <net/netns/generic.h>
  70 #include <net/rtnetlink.h>
  71 #include <net/sock.h>
  72
  73 #include <asm/uaccess.h>
  74
  75 /* Uncomment to enable debugging */
  76 /* #define TUN_DEBUG 1 */
  77
  78 #ifdef TUN_DEBUG
  79 static int debug;
  80
  81 #define tun_debug(level, tun, fmt, args...)                     \
  82 do {                                                            \
  83         if (tun->debug)                                         \
  84                 netdev_printk(level, tun->dev, fmt, ##args);    \
  85 } while (0)
  86 #define DBG1(level, fmt, args...)                               \
  87 do {                                                            \
  88         if (debug == 2)                                         \
  89                 printk(level fmt, ##args);                      \
  90 } while (0)
  91 #else
  92 #define tun_debug(level, tun, fmt, args...)                     \
  93 do {                                                            \
  94         if (0)                                                  \
  95                 netdev_printk(level, tun->dev, fmt, ##args);    \
  96 } while (0)
  97 #define DBG1(level, fmt, args...)                               \
  98 do {                                                            \
  99         if (0)                                                  \
 100                 printk(level fmt, ##args);                      \
 101 } while (0)
 102 #endif
 103
 104 #define GOODCOPY_LEN 128
 105
 106 #define FLT_EXACT_COUNT 8
 107 struct tap_filter {
 108         unsigned int    count;    /* Number of addrs. Zero means disabled */
 109         u32             mask[2];  /* Mask of the hashed addrs */
 110         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 111 };
 112
 113 /* DEFAULT_MAX_NUM_RSS_QUEUES were choosed to let the rx/tx queues allocated for
 114  * the netdevice to be fit in one page. So we can make sure the success of
 115  * memory allocation. TODO: increase the limit. */
 116 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
 117 #define MAX_TAP_FLOWS  4096
 118
 119 #define TUN_FLOW_EXPIRE (3 * HZ)
 120
 121 /* A tun_file connects an open character device to a tuntap netdevice. It
 122  * also contains all socket related strctures (except sock_fprog and tap_filter)
 123  * to serve as one transmit queue for tuntap device. The sock_fprog and
 124  * tap_filter were kept in tun_struct since they were used for filtering for the
 125  * netdevice not for a specific queue (at least I didn't see the requirement for
 126  * this).
 127  *
 128  * RCU usage:
 129  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 130  * other can only be read while rcu_read_lock or rtnl_lock is held.
 131  */
 132 struct tun_file {
 133         struct sock sk;
 134         struct socket socket;
 135         struct socket_wq wq;
 136         struct tun_struct __rcu *tun;
 137         struct net *net;
 138         struct fasync_struct *fasync;
 139         /* only used for fasnyc */
 140         unsigned int flags;
 141         u16 queue_index;
 142         struct list_head next;
 143         struct tun_struct *detached;
 144 };
 145
 146 struct tun_flow_entry {
 147         struct hlist_node hash_link;
 148         struct rcu_head rcu;
 149         struct tun_struct *tun;
 150
 151         u32 rxhash;
 152         int queue_index;
 153         unsigned long updated;
 154 };
 155
 156 #define TUN_NUM_FLOW_ENTRIES 1024
 157
 158 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 159  * device, socket filter, sndbuf and vnet header size were restore when the
 160  * file were attached to a persist device.
 161  */
 162 struct tun_struct {
 163         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 164         unsigned int            numqueues;
 165         unsigned int            flags;
 166         kuid_t                  owner;
 167         kgid_t                  group;
 168
 169         struct net_device       *dev;
 170         netdev_features_t       set_features;
 171 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 172                           NETIF_F_TSO6|NETIF_F_UFO)
 173
 174         int                     vnet_hdr_sz;
 175         int                     sndbuf;
 176         struct tap_filter       txflt;
 177         struct sock_fprog       fprog;
 178         /* protected by rtnl lock */
 179         bool                    filter_attached;
 180 #ifdef TUN_DEBUG
 181         int debug;
 182 #endif
 183         spinlock_t lock;
 184         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 185         struct timer_list flow_gc_timer;
 186         unsigned long ageing_time;
 187         unsigned int numdisabled;
 188         struct list_head disabled;
 189         void *security;
 190         u32 flow_count;
 191 };
 192
 193 static inline u32 tun_hashfn(u32 rxhash)
 194 {
 195         return rxhash & 0x3ff;
 196 }
 197
 198 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 199 {
 200         struct tun_flow_entry *e;
 201
 202         hlist_for_each_entry_rcu(e, head, hash_link) {
 203                 if (e->rxhash == rxhash)
 204                         return e;
 205         }
 206         return NULL;
 207 }
 208
 209 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 210                                               struct hlist_head *head,
 211                                               u32 rxhash, u16 queue_index)
 212 {
 213         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 214
 215         if (e) {
 216                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 217                           rxhash, queue_index);
 218                 e->updated = jiffies;
 219                 e->rxhash = rxhash;
 220                 e->queue_index = queue_index;
 221                 e->tun = tun;
 222                 hlist_add_head_rcu(&e->hash_link, head);
 223                 ++tun->flow_count;
 224         }
 225         return e;
 226 }
 227
 228 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 229 {
 230         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 231                   e->rxhash, e->queue_index);
 232         hlist_del_rcu(&e->hash_link);
 233         kfree_rcu(e, rcu);
 234         --tun->flow_count;
 235 }
 236
 237 static void tun_flow_flush(struct tun_struct *tun)
 238 {
 239         int i;
 240
 241         spin_lock_bh(&tun->lock);
 242         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 243                 struct tun_flow_entry *e;
 244                 struct hlist_node *n;
 245
 246                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 247                         tun_flow_delete(tun, e);
 248         }
 249         spin_unlock_bh(&tun->lock);
 250 }
 251
 252 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 253 {
 254         int i;
 255
 256         spin_lock_bh(&tun->lock);
 257         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 258                 struct tun_flow_entry *e;
 259                 struct hlist_node *n;
 260
 261                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 262                         if (e->queue_index == queue_index)
 263                                 tun_flow_delete(tun, e);
 264                 }
 265         }
 266         spin_unlock_bh(&tun->lock);
 267 }
 268
 269 static void tun_flow_cleanup(unsigned long data)
 270 {
 271         struct tun_struct *tun = (struct tun_struct *)data;
 272         unsigned long delay = tun->ageing_time;
 273         unsigned long next_timer = jiffies + delay;
 274         unsigned long count = 0;
 275         int i;
 276
 277         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 278
 279         spin_lock_bh(&tun->lock);
 280         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 281                 struct tun_flow_entry *e;
 282                 struct hlist_node *n;
 283
 284                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 285                         unsigned long this_timer;
 286                         count++;
 287                         this_timer = e->updated + delay;
 288                         if (time_before_eq(this_timer, jiffies))
 289                                 tun_flow_delete(tun, e);
 290                         else if (time_before(this_timer, next_timer))
 291                                 next_timer = this_timer;
 292                 }
 293         }
 294
 295         if (count)
 296                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 297         spin_unlock_bh(&tun->lock);
 298 }
 299
 300 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 301                             struct tun_file *tfile)
 302 {
 303         struct hlist_head *head;
 304         struct tun_flow_entry *e;
 305         unsigned long delay = tun->ageing_time;
 306         u16 queue_index = tfile->queue_index;
 307
 308         if (!rxhash)
 309                 return;
 310         else
 311                 head = &tun->flows[tun_hashfn(rxhash)];
 312
 313         rcu_read_lock();
 314
 315         /* We may get a very small possibility of OOO during switching, not
 316          * worth to optimize.*/
 317         if (tun->numqueues == 1 || tfile->detached)
 318                 goto unlock;
 319
 320         e = tun_flow_find(head, rxhash);
 321         if (likely(e)) {
 322                 /* TODO: keep queueing to old queue until it's empty? */
 323                 e->queue_index = queue_index;
 324                 e->updated = jiffies;
 325         } else {
 326                 spin_lock_bh(&tun->lock);
 327                 if (!tun_flow_find(head, rxhash) &&
 328                     tun->flow_count < MAX_TAP_FLOWS)
 329                         tun_flow_create(tun, head, rxhash, queue_index);
 330
 331                 if (!timer_pending(&tun->flow_gc_timer))
 332                         mod_timer(&tun->flow_gc_timer,
 333                                   round_jiffies_up(jiffies + delay));
 334                 spin_unlock_bh(&tun->lock);
 335         }
 336
 337 unlock:
 338         rcu_read_unlock();
 339 }
 340
 341 /* We try to identify a flow through its rxhash first. The reason that
 342  * we do not check rxq no. is becuase some cards(e.g 82599), chooses
 343  * the rxq based on the txq where the last packet of the flow comes. As
 344  * the userspace application move between processors, we may get a
 345  * different rxq no. here. If we could not get rxhash, then we would
 346  * hope the rxq no. may help here.
 347  */
 348 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
 349 {
 350         struct tun_struct *tun = netdev_priv(dev);
 351         struct tun_flow_entry *e;
 352         u32 txq = 0;
 353         u32 numqueues = 0;
 354
 355         rcu_read_lock();
 356         numqueues = ACCESS_ONCE(tun->numqueues);
 357
 358         txq = skb_get_rxhash(skb);
 359         if (txq) {
 360                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 361                 if (e)
 362                         txq = e->queue_index;
 363                 else
 364                         /* use multiply and shift instead of expensive divide */
 365                         txq = ((u64)txq * numqueues) >> 32;
 366         } else if (likely(skb_rx_queue_recorded(skb))) {
 367                 txq = skb_get_rx_queue(skb);
 368                 while (unlikely(txq >= numqueues))
 369                         txq -= numqueues;
 370         }
 371
 372         rcu_read_unlock();
 373         return txq;
 374 }
 375
 376 static inline bool tun_not_capable(struct tun_struct *tun)
 377 {
 378         const struct cred *cred = current_cred();
 379         struct net *net = dev_net(tun->dev);
 380
 381         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 382                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 383                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 384 }
 385
 386 static void tun_set_real_num_queues(struct tun_struct *tun)
 387 {
 388         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 389         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 390 }
 391
 392 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 393 {
 394         tfile->detached = tun;
 395         list_add_tail(&tfile->next, &tun->disabled);
 396         ++tun->numdisabled;
 397 }
 398
 399 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 400 {
 401         struct tun_struct *tun = tfile->detached;
 402
 403         tfile->detached = NULL;
 404         list_del_init(&tfile->next);
 405         --tun->numdisabled;
 406         return tun;
 407 }
 408
 409 static void __tun_detach(struct tun_file *tfile, bool clean)
 410 {
 411         struct tun_file *ntfile;
 412         struct tun_struct *tun;
 413
 414         tun = rtnl_dereference(tfile->tun);
 415
 416         if (tun && !tfile->detached) {
 417                 u16 index = tfile->queue_index;
 418                 BUG_ON(index >= tun->numqueues);
 419
 420                 rcu_assign_pointer(tun->tfiles[index],
 421                                    tun->tfiles[tun->numqueues - 1]);
 422                 ntfile = rtnl_dereference(tun->tfiles[index]);
 423                 ntfile->queue_index = index;
 424
 425                 --tun->numqueues;
 426                 if (clean) {
 427                         rcu_assign_pointer(tfile->tun, NULL);
 428                         sock_put(&tfile->sk);
 429                 } else
 430                         tun_disable_queue(tun, tfile);
 431
 432                 synchronize_net();
 433                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 434                 /* Drop read queue */
 435                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 436                 tun_set_real_num_queues(tun);
 437         } else if (tfile->detached && clean) {
 438                 tun = tun_enable_queue(tfile);
 439                 sock_put(&tfile->sk);
 440         }
 441
 442         if (clean) {
 443                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 444                         netif_carrier_off(tun->dev);
 445
 446                         if (!(tun->flags & TUN_PERSIST) &&
 447                             tun->dev->reg_state == NETREG_REGISTERED)
 448                                 unregister_netdevice(tun->dev);
 449                 }
 450
 451                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 452                                  &tfile->socket.flags));
 453                 sk_release_kernel(&tfile->sk);
 454         }
 455 }
 456
 457 static void tun_detach(struct tun_file *tfile, bool clean)
 458 {
 459         rtnl_lock();
 460         __tun_detach(tfile, clean);
 461         rtnl_unlock();
 462 }
 463
 464 static void tun_detach_all(struct net_device *dev)
 465 {
 466         struct tun_struct *tun = netdev_priv(dev);
 467         struct tun_file *tfile, *tmp;
 468         int i, n = tun->numqueues;
 469
 470         for (i = 0; i < n; i++) {
 471                 tfile = rtnl_dereference(tun->tfiles[i]);
 472                 BUG_ON(!tfile);
 473                 wake_up_all(&tfile->wq.wait);
 474                 rcu_assign_pointer(tfile->tun, NULL);
 475                 --tun->numqueues;
 476         }
 477         list_for_each_entry(tfile, &tun->disabled, next) {
 478                 wake_up_all(&tfile->wq.wait);
 479                 rcu_assign_pointer(tfile->tun, NULL);
 480         }
 481         BUG_ON(tun->numqueues != 0);
 482
 483         synchronize_net();
 484         for (i = 0; i < n; i++) {
 485                 tfile = rtnl_dereference(tun->tfiles[i]);
 486                 /* Drop read queue */
 487                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 488                 sock_put(&tfile->sk);
 489         }
 490         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 491                 tun_enable_queue(tfile);
 492                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 493                 sock_put(&tfile->sk);
 494         }
 495         BUG_ON(tun->numdisabled != 0);
 496
 497         if (tun->flags & TUN_PERSIST)
 498                 module_put(THIS_MODULE);
 499 }
 500
 501 static int tun_attach(struct tun_struct *tun, struct file *file)
 502 {
 503         struct tun_file *tfile = file->private_data;
 504         int err;
 505
 506         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 507         if (err < 0)
 508                 goto out;
 509
 510         err = -EINVAL;
 511         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 512                 goto out;
 513
 514         err = -EBUSY;
 515         if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 516                 goto out;
 517
 518         err = -E2BIG;
 519         if (!tfile->detached &&
 520             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 521                 goto out;
 522
 523         err = 0;
 524
 525         /* Re-attach the filter to presist device */
 526         if (tun->filter_attached == true) {
 527                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 528                 if (!err)
 529                         goto out;
 530         }
 531         tfile->queue_index = tun->numqueues;
 532         rcu_assign_pointer(tfile->tun, tun);
 533         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 534         tun->numqueues++;
 535
 536         if (tfile->detached)
 537                 tun_enable_queue(tfile);
 538         else
 539                 sock_hold(&tfile->sk);
 540
 541         tun_set_real_num_queues(tun);
 542
 543         /* device is allowed to go away first, so no need to hold extra
 544          * refcnt.
 545          */
 546
 547 out:
 548         return err;
 549 }
 550
 551 static struct tun_struct *__tun_get(struct tun_file *tfile)
 552 {
 553         struct tun_struct *tun;
 554
 555         rcu_read_lock();
 556         tun = rcu_dereference(tfile->tun);
 557         if (tun)
 558                 dev_hold(tun->dev);
 559         rcu_read_unlock();
 560
 561         return tun;
 562 }
 563
 564 static struct tun_struct *tun_get(struct file *file)
 565 {
 566         return __tun_get(file->private_data);
 567 }
 568
 569 static void tun_put(struct tun_struct *tun)
 570 {
 571         dev_put(tun->dev);
 572 }
 573
 574 /* TAP filtering */
 575 static void addr_hash_set(u32 *mask, const u8 *addr)
 576 {
 577         int n = ether_crc(ETH_ALEN, addr) >> 26;
 578         mask[n >> 5] |= (1 << (n & 31));
 579 }
 580
 581 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 582 {
 583         int n = ether_crc(ETH_ALEN, addr) >> 26;
 584         return mask[n >> 5] & (1 << (n & 31));
 585 }
 586
 587 static int update_filter(struct tap_filter *filter, void __user *arg)
 588 {
 589         struct { u8 u[ETH_ALEN]; } *addr;
 590         struct tun_filter uf;
 591         int err, alen, n, nexact;
 592
 593         if (copy_from_user(&uf, arg, sizeof(uf)))
 594                 return -EFAULT;
 595
 596         if (!uf.count) {
 597                 /* Disabled */
 598                 filter->count = 0;
 599                 return 0;
 600         }
 601
 602         alen = ETH_ALEN * uf.count;
 603         addr = kmalloc(alen, GFP_KERNEL);
 604         if (!addr)
 605                 return -ENOMEM;
 606
 607         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 608                 err = -EFAULT;
 609                 goto done;
 610         }
 611
 612         /* The filter is updated without holding any locks. Which is
 613          * perfectly safe. We disable it first and in the worst
 614          * case we'll accept a few undesired packets. */
 615         filter->count = 0;
 616         wmb();
 617
 618         /* Use first set of addresses as an exact filter */
 619         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 620                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 621
 622         nexact = n;
 623
 624         /* Remaining multicast addresses are hashed,
 625          * unicast will leave the filter disabled. */
 626         memset(filter->mask, 0, sizeof(filter->mask));
 627         for (; n < uf.count; n++) {
 628                 if (!is_multicast_ether_addr(addr[n].u)) {
 629                         err = 0; /* no filter */
 630                         goto done;
 631                 }
 632                 addr_hash_set(filter->mask, addr[n].u);
 633         }
 634
 635         /* For ALLMULTI just set the mask to all ones.
 636          * This overrides the mask populated above. */
 637         if ((uf.flags & TUN_FLT_ALLMULTI))
 638                 memset(filter->mask, ~0, sizeof(filter->mask));
 639
 640         /* Now enable the filter */
 641         wmb();
 642         filter->count = nexact;
 643
 644         /* Return the number of exact filters */
 645         err = nexact;
 646
 647 done:
 648         kfree(addr);
 649         return err;
 650 }
 651
 652 /* Returns: 0 - drop, !=0 - accept */
 653 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 654 {
 655         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 656          * at this point. */
 657         struct ethhdr *eh = (struct ethhdr *) skb->data;
 658         int i;
 659
 660         /* Exact match */
 661         for (i = 0; i < filter->count; i++)
 662                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 663                         return 1;
 664
 665         /* Inexact match (multicast only) */
 666         if (is_multicast_ether_addr(eh->h_dest))
 667                 return addr_hash_test(filter->mask, eh->h_dest);
 668
 669         return 0;
 670 }
 671
 672 /*
 673  * Checks whether the packet is accepted or not.
 674  * Returns: 0 - drop, !=0 - accept
 675  */
 676 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 677 {
 678         if (!filter->count)
 679                 return 1;
 680
 681         return run_filter(filter, skb);
 682 }
 683
 684 /* Network device part of the driver */
 685
 686 static const struct ethtool_ops tun_ethtool_ops;
 687
 688 /* Net device detach from fd. */
 689 static void tun_net_uninit(struct net_device *dev)
 690 {
 691         tun_detach_all(dev);
 692 }
 693
 694 /* Net device open. */
 695 static int tun_net_open(struct net_device *dev)
 696 {
 697         netif_tx_start_all_queues(dev);
 698         return 0;
 699 }
 700
 701 /* Net device close. */
 702 static int tun_net_close(struct net_device *dev)
 703 {
 704         netif_tx_stop_all_queues(dev);
 705         return 0;
 706 }
 707
 708 /* Net device start xmit */
 709 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 710 {
 711         struct tun_struct *tun = netdev_priv(dev);
 712         int txq = skb->queue_mapping;
 713         struct tun_file *tfile;
 714
 715         rcu_read_lock();
 716         tfile = rcu_dereference(tun->tfiles[txq]);
 717
 718         /* Drop packet if interface is not attached */
 719         if (txq >= tun->numqueues)
 720                 goto drop;
 721
 722         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 723
 724         BUG_ON(!tfile);
 725
 726         /* Drop if the filter does not like it.
 727          * This is a noop if the filter is disabled.
 728          * Filter can be enabled only for the TAP devices. */
 729         if (!check_filter(&tun->txflt, skb))
 730                 goto drop;
 731
 732         if (tfile->socket.sk->sk_filter &&
 733             sk_filter(tfile->socket.sk, skb))
 734                 goto drop;
 735
 736         /* Limit the number of packets queued by dividing txq length with the
 737          * number of queues.
 738          */
 739         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
 740                           >= dev->tx_queue_len / tun->numqueues)
 741                 goto drop;
 742
 743         if (skb->sk) {
 744                 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
 745                 sw_tx_timestamp(skb);
 746         }
 747
 748         /* Orphan the skb - required as we might hang on to it
 749          * for indefinite time. */
 750         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 751                 goto drop;
 752         skb_orphan(skb);
 753
 754         nf_reset(skb);
 755
 756         /* Enqueue packet */
 757         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 758
 759         /* Notify and wake up reader process */
 760         if (tfile->flags & TUN_FASYNC)
 761                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 762         wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
 763                                    POLLRDNORM | POLLRDBAND);
 764
 765         rcu_read_unlock();
 766         return NETDEV_TX_OK;
 767
 768 drop:
 769         dev->stats.tx_dropped++;
 770         skb_tx_error(skb);
 771         kfree_skb(skb);
 772         rcu_read_unlock();
 773         return NETDEV_TX_OK;
 774 }
 775
 776 static void tun_net_mclist(struct net_device *dev)
 777 {
 778         /*
 779          * This callback is supposed to deal with mc filter in
 780          * _rx_ path and has nothing to do with the _tx_ path.
 781          * In rx path we always accept everything userspace gives us.
 782          */
 783 }
 784
 785 #define MIN_MTU 68
 786 #define MAX_MTU 65535
 787
 788 static int
 789 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 790 {
 791         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 792                 return -EINVAL;
 793         dev->mtu = new_mtu;
 794         return 0;
 795 }
 796
 797 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 798         netdev_features_t features)
 799 {
 800         struct tun_struct *tun = netdev_priv(dev);
 801
 802         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 803 }
 804 #ifdef CONFIG_NET_POLL_CONTROLLER
 805 static void tun_poll_controller(struct net_device *dev)
 806 {
 807         /*
 808          * Tun only receives frames when:
 809          * 1) the char device endpoint gets data from user space
 810          * 2) the tun socket gets a sendmsg call from user space
 811          * Since both of those are syncronous operations, we are guaranteed
 812          * never to have pending data when we poll for it
 813          * so theres nothing to do here but return.
 814          * We need this though so netpoll recognizes us as an interface that
 815          * supports polling, which enables bridge devices in virt setups to
 816          * still use netconsole
 817          */
 818         return;
 819 }
 820 #endif
 821 static const struct net_device_ops tun_netdev_ops = {
 822         .ndo_uninit             = tun_net_uninit,
 823         .ndo_open               = tun_net_open,
 824         .ndo_stop               = tun_net_close,
 825         .ndo_start_xmit         = tun_net_xmit,
 826         .ndo_change_mtu         = tun_net_change_mtu,
 827         .ndo_fix_features       = tun_net_fix_features,
 828         .ndo_select_queue       = tun_select_queue,
 829 #ifdef CONFIG_NET_POLL_CONTROLLER
 830         .ndo_poll_controller    = tun_poll_controller,
 831 #endif
 832 };
 833
 834 static const struct net_device_ops tap_netdev_ops = {
 835         .ndo_uninit             = tun_net_uninit,
 836         .ndo_open               = tun_net_open,
 837         .ndo_stop               = tun_net_close,
 838         .ndo_start_xmit         = tun_net_xmit,
 839         .ndo_change_mtu         = tun_net_change_mtu,
 840         .ndo_fix_features       = tun_net_fix_features,
 841         .ndo_set_rx_mode        = tun_net_mclist,
 842         .ndo_set_mac_address    = eth_mac_addr,
 843         .ndo_validate_addr      = eth_validate_addr,
 844         .ndo_select_queue       = tun_select_queue,
 845 #ifdef CONFIG_NET_POLL_CONTROLLER
 846         .ndo_poll_controller    = tun_poll_controller,
 847 #endif
 848 };
 849
 850 static void tun_flow_init(struct tun_struct *tun)
 851 {
 852         int i;
 853
 854         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 855                 INIT_HLIST_HEAD(&tun->flows[i]);
 856
 857         tun->ageing_time = TUN_FLOW_EXPIRE;
 858         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 859         mod_timer(&tun->flow_gc_timer,
 860                   round_jiffies_up(jiffies + tun->ageing_time));
 861 }
 862
 863 static void tun_flow_uninit(struct tun_struct *tun)
 864 {
 865         del_timer_sync(&tun->flow_gc_timer);
 866         tun_flow_flush(tun);
 867 }
 868
 869 /* Initialize net device. */
 870 static void tun_net_init(struct net_device *dev)
 871 {
 872         struct tun_struct *tun = netdev_priv(dev);
 873
 874         switch (tun->flags & TUN_TYPE_MASK) {
 875         case TUN_TUN_DEV:
 876                 dev->netdev_ops = &tun_netdev_ops;
 877
 878                 /* Point-to-Point TUN Device */
 879                 dev->hard_header_len = 0;
 880                 dev->addr_len = 0;
 881                 dev->mtu = 1500;
 882
 883                 /* Zero header length */
 884                 dev->type = ARPHRD_NONE;
 885                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 886                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 887                 break;
 888
 889         case TUN_TAP_DEV:
 890                 dev->netdev_ops = &tap_netdev_ops;
 891                 /* Ethernet TAP Device */
 892                 ether_setup(dev);
 893                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 894                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 895
 896                 eth_hw_addr_random(dev);
 897
 898                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 899                 break;
 900         }
 901 }
 902
 903 /* Character device part */
 904
 905 /* Poll */
 906 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 907 {
 908         struct tun_file *tfile = file->private_data;
 909         struct tun_struct *tun = __tun_get(tfile);
 910         struct sock *sk;
 911         unsigned int mask = 0;
 912
 913         if (!tun)
 914                 return POLLERR;
 915
 916         sk = tfile->socket.sk;
 917
 918         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 919
 920         poll_wait(file, &tfile->wq.wait, wait);
 921
 922         if (!skb_queue_empty(&sk->sk_receive_queue))
 923                 mask |= POLLIN | POLLRDNORM;
 924
 925         if (sock_writeable(sk) ||
 926             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 927              sock_writeable(sk)))
 928                 mask |= POLLOUT | POLLWRNORM;
 929
 930         if (tun->dev->reg_state != NETREG_REGISTERED)
 931                 mask = POLLERR;
 932
 933         tun_put(tun);
 934         return mask;
 935 }
 936
 937 /* prepad is the amount to reserve at front.  len is length after that.
 938  * linear is a hint as to how much to copy (usually headers). */
 939 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 940                                      size_t prepad, size_t len,
 941                                      size_t linear, int noblock)
 942 {
 943         struct sock *sk = tfile->socket.sk;
 944         struct sk_buff *skb;
 945         int err;
 946
 947         /* Under a page?  Don't bother with paged skb. */
 948         if (prepad + len < PAGE_SIZE || !linear)
 949                 linear = len;
 950
 951         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 952                                    &err);
 953         if (!skb)
 954                 return ERR_PTR(err);
 955
 956         skb_reserve(skb, prepad);
 957         skb_put(skb, linear);
 958         skb->data_len = len - linear;
 959         skb->len += len - linear;
 960
 961         return skb;
 962 }
 963
 964 /* set skb frags from iovec, this can move to core network code for reuse */
 965 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 966                                   int offset, size_t count)
 967 {
 968         int len = iov_length(from, count) - offset;
 969         int copy = skb_headlen(skb);
 970         int size, offset1 = 0;
 971         int i = 0;
 972
 973         /* Skip over from offset */
 974         while (count && (offset >= from->iov_len)) {
 975                 offset -= from->iov_len;
 976                 ++from;
 977                 --count;
 978         }
 979
 980         /* copy up to skb headlen */
 981         while (count && (copy > 0)) {
 982                 size = min_t(unsigned int, copy, from->iov_len - offset);
 983                 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
 984                                    size))
 985                         return -EFAULT;
 986                 if (copy > size) {
 987                         ++from;
 988                         --count;
 989                         offset = 0;
 990                 } else
 991                         offset += size;
 992                 copy -= size;
 993                 offset1 += size;
 994         }
 995
 996         if (len == offset1)
 997                 return 0;
 998
 999         while (count--) {
1000                 struct page *page[MAX_SKB_FRAGS];
1001                 int num_pages;
1002                 unsigned long base;
1003                 unsigned long truesize;
1004
1005                 len = from->iov_len - offset;
1006                 if (!len) {
1007                         offset = 0;
1008                         ++from;
1009                         continue;
1010                 }
1011                 base = (unsigned long)from->iov_base + offset;
1012                 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
1013                 if (i + size > MAX_SKB_FRAGS)
1014                         return -EMSGSIZE;
1015                 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
1016                 if (num_pages != size) {
1017                         int j;
1018
1019                         for (j = 0; j < num_pages; j++)
1020                                 put_page(page[i + j]);
1021                         return -EFAULT;
1022                 }
1023                 truesize = size * PAGE_SIZE;
1024                 skb->data_len += len;
1025                 skb->len += len;
1026                 skb->truesize += truesize;
1027                 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
1028                 while (len) {
1029                         int off = base & ~PAGE_MASK;
1030                         int size = min_t(int, len, PAGE_SIZE - off);
1031                         __skb_fill_page_desc(skb, i, page[i], off, size);
1032                         skb_shinfo(skb)->nr_frags++;
1033                         /* increase sk_wmem_alloc */
1034                         base += size;
1035                         len -= size;
1036                         i++;
1037                 }
1038                 offset = 0;
1039                 ++from;
1040         }
1041         return 0;
1042 }
1043
1044 /* Get packet from user space buffer */
1045 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1046                             void *msg_control, const struct iovec *iv,
1047                             size_t total_len, size_t count, int noblock)
1048 {
1049         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1050         struct sk_buff *skb;
1051         size_t len = total_len, align = NET_SKB_PAD, linear;
1052         struct virtio_net_hdr gso = { 0 };
1053         int offset = 0;
1054         int copylen;
1055         bool zerocopy = false;
1056         int err;
1057         u32 rxhash;
1058
1059         if (!(tun->flags & TUN_NO_PI)) {
1060                 if ((len -= sizeof(pi)) > total_len)
1061                         return -EINVAL;
1062
1063                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1064                         return -EFAULT;
1065                 offset += sizeof(pi);
1066         }
1067
1068         if (tun->flags & TUN_VNET_HDR) {
1069                 if ((len -= tun->vnet_hdr_sz) > total_len)
1070                         return -EINVAL;
1071
1072                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1073                         return -EFAULT;
1074
1075                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1076                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1077                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1078
1079                 if (gso.hdr_len > len)
1080                         return -EINVAL;
1081                 offset += tun->vnet_hdr_sz;
1082         }
1083
1084         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1085                 align += NET_IP_ALIGN;
1086                 if (unlikely(len < ETH_HLEN ||
1087                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1088                         return -EINVAL;
1089         }
1090
1091         if (msg_control) {
1092                 /* There are 256 bytes to be copied in skb, so there is
1093                  * enough room for skb expand head in case it is used.
1094                  * The rest of the buffer is mapped from userspace.
1095                  */
1096                 copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1097                 linear = copylen;
1098                 if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1099                         zerocopy = true;
1100         }
1101
1102         if (!zerocopy) {
1103                 copylen = len;
1104                 linear = gso.hdr_len;
1105         }
1106
1107         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1108         if (IS_ERR(skb)) {
1109                 if (PTR_ERR(skb) != -EAGAIN)
1110                         tun->dev->stats.rx_dropped++;
1111                 return PTR_ERR(skb);
1112         }
1113
1114         if (zerocopy)
1115                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1116         else {
1117                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1118                 if (!err && msg_control) {
1119                         struct ubuf_info *uarg = msg_control;
1120                         uarg->callback(uarg, false);
1121                 }
1122         }
1123
1124         if (err) {
1125                 tun->dev->stats.rx_dropped++;
1126                 kfree_skb(skb);
1127                 return -EFAULT;
1128         }
1129
1130         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1131                 if (!skb_partial_csum_set(skb, gso.csum_start,
1132                                           gso.csum_offset)) {
1133                         tun->dev->stats.rx_frame_errors++;
1134                         kfree_skb(skb);
1135                         return -EINVAL;
1136                 }
1137         }
1138
1139         switch (tun->flags & TUN_TYPE_MASK) {
1140         case TUN_TUN_DEV:
1141                 if (tun->flags & TUN_NO_PI) {
1142                         switch (skb->data[0] & 0xf0) {
1143                         case 0x40:
1144                                 pi.proto = htons(ETH_P_IP);
1145                                 break;
1146                         case 0x60:
1147                                 pi.proto = htons(ETH_P_IPV6);
1148                                 break;
1149                         default:
1150                                 tun->dev->stats.rx_dropped++;
1151                                 kfree_skb(skb);
1152                                 return -EINVAL;
1153                         }
1154                 }
1155
1156                 skb_reset_mac_header(skb);
1157                 skb->protocol = pi.proto;
1158                 skb->dev = tun->dev;
1159                 break;
1160         case TUN_TAP_DEV:
1161                 skb->protocol = eth_type_trans(skb, tun->dev);
1162                 break;
1163         }
1164
1165         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1166                 pr_debug("GSO!\n");
1167                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1168                 case VIRTIO_NET_HDR_GSO_TCPV4:
1169                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1170                         break;
1171                 case VIRTIO_NET_HDR_GSO_TCPV6:
1172                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1173                         break;
1174                 case VIRTIO_NET_HDR_GSO_UDP:
1175                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1176                         break;
1177                 default:
1178                         tun->dev->stats.rx_frame_errors++;
1179                         kfree_skb(skb);
1180                         return -EINVAL;
1181                 }
1182
1183                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1184                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1185
1186                 skb_shinfo(skb)->gso_size = gso.gso_size;
1187                 if (skb_shinfo(skb)->gso_size == 0) {
1188                         tun->dev->stats.rx_frame_errors++;
1189                         kfree_skb(skb);
1190                         return -EINVAL;
1191                 }
1192
1193                 /* Header must be checked, and gso_segs computed. */
1194                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1195                 skb_shinfo(skb)->gso_segs = 0;
1196         }
1197
1198         /* copy skb_ubuf_info for callback when skb has no error */
1199         if (zerocopy) {
1200                 skb_shinfo(skb)->destructor_arg = msg_control;
1201                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1202                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1203         }
1204
1205         skb_reset_network_header(skb);
1206         skb_probe_transport_header(skb, 0);
1207
1208         rxhash = skb_get_rxhash(skb);
1209         netif_rx_ni(skb);
1210
1211         tun->dev->stats.rx_packets++;
1212         tun->dev->stats.rx_bytes += len;
1213
1214         tun_flow_update(tun, rxhash, tfile);
1215         return total_len;
1216 }
1217
1218 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1219                               unsigned long count, loff_t pos)
1220 {
1221         struct file *file = iocb->ki_filp;
1222         struct tun_struct *tun = tun_get(file);
1223         struct tun_file *tfile = file->private_data;
1224         ssize_t result;
1225
1226         if (!tun)
1227                 return -EBADFD;
1228
1229         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1230
1231         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1232                               count, file->f_flags & O_NONBLOCK);
1233
1234         tun_put(tun);
1235         return result;
1236 }
1237
1238 /* Put packet to the user space buffer */
1239 static ssize_t tun_put_user(struct tun_struct *tun,
1240                             struct tun_file *tfile,
1241                             struct sk_buff *skb,
1242                             const struct iovec *iv, int len)
1243 {
1244         struct tun_pi pi = { 0, skb->protocol };
1245         ssize_t total = 0;
1246         int vlan_offset = 0;
1247
1248         if (!(tun->flags & TUN_NO_PI)) {
1249                 if ((len -= sizeof(pi)) < 0)
1250                         return -EINVAL;
1251
1252                 if (len < skb->len) {
1253                         /* Packet will be striped */
1254                         pi.flags |= TUN_PKT_STRIP;
1255                 }
1256
1257                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1258                         return -EFAULT;
1259                 total += sizeof(pi);
1260         }
1261
1262         if (tun->flags & TUN_VNET_HDR) {
1263                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1264                 if ((len -= tun->vnet_hdr_sz) < 0)
1265                         return -EINVAL;
1266
1267                 if (skb_is_gso(skb)) {
1268                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1269
1270                         /* This is a hint as to how much should be linear. */
1271                         gso.hdr_len = skb_headlen(skb);
1272                         gso.gso_size = sinfo->gso_size;
1273                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1274                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1275                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1276                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1277                         else if (sinfo->gso_type & SKB_GSO_UDP)
1278                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1279                         else {
1280                                 pr_err("unexpected GSO type: "
1281                                        "0x%x, gso_size %d, hdr_len %d\n",
1282                                        sinfo->gso_type, gso.gso_size,
1283                                        gso.hdr_len);
1284                                 print_hex_dump(KERN_ERR, "tun: ",
1285                                                DUMP_PREFIX_NONE,
1286                                                16, 1, skb->head,
1287                                                min((int)gso.hdr_len, 64), true);
1288                                 WARN_ON_ONCE(1);
1289                                 return -EINVAL;
1290                         }
1291                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1292                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1293                 } else
1294                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1295
1296                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1297                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1298                         gso.csum_start = skb_checksum_start_offset(skb);
1299                         gso.csum_offset = skb->csum_offset;
1300                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1301                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1302                 } /* else everything is zero */
1303
1304                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1305                                                sizeof(gso))))
1306                         return -EFAULT;
1307                 total += tun->vnet_hdr_sz;
1308         }
1309
1310         if (!vlan_tx_tag_present(skb)) {
1311                 len = min_t(int, skb->len, len);
1312         } else {
1313                 int copy, ret;
1314                 struct {
1315                         __be16 h_vlan_proto;
1316                         __be16 h_vlan_TCI;
1317                 } veth;
1318
1319                 veth.h_vlan_proto = skb->vlan_proto;
1320                 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1321
1322                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1323                 len = min_t(int, skb->len + VLAN_HLEN, len);
1324
1325                 copy = min_t(int, vlan_offset, len);
1326                 ret = skb_copy_datagram_const_iovec(skb, 0, iv, total, copy);
1327                 len -= copy;
1328                 total += copy;
1329                 if (ret || !len)
1330                         goto done;
1331
1332                 copy = min_t(int, sizeof(veth), len);
1333                 ret = memcpy_toiovecend(iv, (void *)&veth, total, copy);
1334                 len -= copy;
1335                 total += copy;
1336                 if (ret || !len)
1337                         goto done;
1338         }
1339
1340         skb_copy_datagram_const_iovec(skb, vlan_offset, iv, total, len);
1341         total += len;
1342
1343 done:
1344         tun->dev->stats.tx_packets++;
1345         tun->dev->stats.tx_bytes += len;
1346
1347         return total;
1348 }
1349
1350 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1351                            struct kiocb *iocb, const struct iovec *iv,
1352                            ssize_t len, int noblock)
1353 {
1354         DECLARE_WAITQUEUE(wait, current);
1355         struct sk_buff *skb;
1356         ssize_t ret = 0;
1357
1358         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1359
1360         if (unlikely(!noblock))
1361                 add_wait_queue(&tfile->wq.wait, &wait);
1362         while (len) {
1363                 current->state = TASK_INTERRUPTIBLE;
1364
1365                 /* Read frames from the queue */
1366                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1367                         if (noblock) {
1368                                 ret = -EAGAIN;
1369                                 break;
1370                         }
1371                         if (signal_pending(current)) {
1372                                 ret = -ERESTARTSYS;
1373                                 break;
1374                         }
1375                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1376                                 ret = -EIO;
1377                                 break;
1378                         }
1379
1380                         /* Nothing to read, let's sleep */
1381                         schedule();
1382                         continue;
1383                 }
1384
1385                 ret = tun_put_user(tun, tfile, skb, iv, len);
1386                 kfree_skb(skb);
1387                 break;
1388         }
1389
1390         current->state = TASK_RUNNING;
1391         if (unlikely(!noblock))
1392                 remove_wait_queue(&tfile->wq.wait, &wait);
1393
1394         return ret;
1395 }
1396
1397 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1398                             unsigned long count, loff_t pos)
1399 {
1400         struct file *file = iocb->ki_filp;
1401         struct tun_file *tfile = file->private_data;
1402         struct tun_struct *tun = __tun_get(tfile);
1403         ssize_t len, ret;
1404
1405         if (!tun)
1406                 return -EBADFD;
1407         len = iov_length(iv, count);
1408         if (len < 0) {
1409                 ret = -EINVAL;
1410                 goto out;
1411         }
1412
1413         ret = tun_do_read(tun, tfile, iocb, iv, len,
1414                           file->f_flags & O_NONBLOCK);
1415         ret = min_t(ssize_t, ret, len);
1416 out:
1417         tun_put(tun);
1418         return ret;
1419 }
1420
1421 static void tun_free_netdev(struct net_device *dev)
1422 {
1423         struct tun_struct *tun = netdev_priv(dev);
1424
1425         BUG_ON(!(list_empty(&tun->disabled)));
1426         tun_flow_uninit(tun);
1427         security_tun_dev_free_security(tun->security);
1428         free_netdev(dev);
1429 }
1430
1431 static void tun_setup(struct net_device *dev)
1432 {
1433         struct tun_struct *tun = netdev_priv(dev);
1434
1435         tun->owner = INVALID_UID;
1436         tun->group = INVALID_GID;
1437
1438         dev->ethtool_ops = &tun_ethtool_ops;
1439         dev->destructor = tun_free_netdev;
1440 }
1441
1442 /* Trivial set of netlink ops to allow deleting tun or tap
1443  * device with netlink.
1444  */
1445 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1446 {
1447         return -EINVAL;
1448 }
1449
1450 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1451         .kind           = DRV_NAME,
1452         .priv_size      = sizeof(struct tun_struct),
1453         .setup          = tun_setup,
1454         .validate       = tun_validate,
1455 };
1456
1457 static void tun_sock_write_space(struct sock *sk)
1458 {
1459         struct tun_file *tfile;
1460         wait_queue_head_t *wqueue;
1461
1462         if (!sock_writeable(sk))
1463                 return;
1464
1465         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1466                 return;
1467
1468         wqueue = sk_sleep(sk);
1469         if (wqueue && waitqueue_active(wqueue))
1470                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1471                                                 POLLWRNORM | POLLWRBAND);
1472
1473         tfile = container_of(sk, struct tun_file, sk);
1474         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1475 }
1476
1477 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1478                        struct msghdr *m, size_t total_len)
1479 {
1480         int ret;
1481         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1482         struct tun_struct *tun = __tun_get(tfile);
1483
1484         if (!tun)
1485                 return -EBADFD;
1486         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1487                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1488         tun_put(tun);
1489         return ret;
1490 }
1491
1492 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1493                        struct msghdr *m, size_t total_len,
1494                        int flags)
1495 {
1496         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1497         struct tun_struct *tun = __tun_get(tfile);
1498         int ret;
1499
1500         if (!tun)
1501                 return -EBADFD;
1502
1503         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1504                 ret = -EINVAL;
1505                 goto out;
1506         }
1507         if (flags & MSG_ERRQUEUE) {
1508                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1509                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1510                 goto out;
1511         }
1512         ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1513                           flags & MSG_DONTWAIT);
1514         if (ret > total_len) {
1515                 m->msg_flags |= MSG_TRUNC;
1516                 ret = flags & MSG_TRUNC ? ret : total_len;
1517         }
1518 out:
1519         tun_put(tun);
1520         return ret;
1521 }
1522
1523 static int tun_release(struct socket *sock)
1524 {
1525         if (sock->sk)
1526                 sock_put(sock->sk);
1527         return 0;
1528 }
1529
1530 /* Ops structure to mimic raw sockets with tun */
1531 static const struct proto_ops tun_socket_ops = {
1532         .sendmsg = tun_sendmsg,
1533         .recvmsg = tun_recvmsg,
1534         .release = tun_release,
1535 };
1536
1537 static struct proto tun_proto = {
1538         .name           = "tun",
1539         .owner          = THIS_MODULE,
1540         .obj_size       = sizeof(struct tun_file),
1541 };
1542
1543 static int tun_flags(struct tun_struct *tun)
1544 {
1545         int flags = 0;
1546
1547         if (tun->flags & TUN_TUN_DEV)
1548                 flags |= IFF_TUN;
1549         else
1550                 flags |= IFF_TAP;
1551
1552         if (tun->flags & TUN_NO_PI)
1553                 flags |= IFF_NO_PI;
1554
1555         /* This flag has no real effect.  We track the value for backwards
1556          * compatibility.
1557          */
1558         if (tun->flags & TUN_ONE_QUEUE)
1559                 flags |= IFF_ONE_QUEUE;
1560
1561         if (tun->flags & TUN_VNET_HDR)
1562                 flags |= IFF_VNET_HDR;
1563
1564         if (tun->flags & TUN_TAP_MQ)
1565                 flags |= IFF_MULTI_QUEUE;
1566
1567         if (tun->flags & TUN_PERSIST)
1568                 flags |= IFF_PERSIST;
1569
1570         return flags;
1571 }
1572
1573 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1574                               char *buf)
1575 {
1576         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1577         return sprintf(buf, "0x%x\n", tun_flags(tun));
1578 }
1579
1580 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1581                               char *buf)
1582 {
1583         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1584         return uid_valid(tun->owner)?
1585                 sprintf(buf, "%u\n",
1586                         from_kuid_munged(current_user_ns(), tun->owner)):
1587                 sprintf(buf, "-1\n");
1588 }
1589
1590 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1591                               char *buf)
1592 {
1593         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1594         return gid_valid(tun->group) ?
1595                 sprintf(buf, "%u\n",
1596                         from_kgid_munged(current_user_ns(), tun->group)):
1597                 sprintf(buf, "-1\n");
1598 }
1599
1600 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1601 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1602 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1603
1604 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1605 {
1606         struct tun_struct *tun;
1607         struct tun_file *tfile = file->private_data;
1608         struct net_device *dev;
1609         int err;
1610
1611         if (tfile->detached)
1612                 return -EINVAL;
1613
1614         dev = __dev_get_by_name(net, ifr->ifr_name);
1615         if (dev) {
1616                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1617                         return -EBUSY;
1618                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1619                         tun = netdev_priv(dev);
1620                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1621                         tun = netdev_priv(dev);
1622                 else
1623                         return -EINVAL;
1624
1625                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1626                     !!(tun->flags & TUN_TAP_MQ))
1627                         return -EINVAL;
1628
1629                 if (tun_not_capable(tun))
1630                         return -EPERM;
1631                 err = security_tun_dev_open(tun->security);
1632                 if (err < 0)
1633                         return err;
1634
1635                 err = tun_attach(tun, file);
1636                 if (err < 0)
1637                         return err;
1638
1639                 if (tun->flags & TUN_TAP_MQ &&
1640                     (tun->numqueues + tun->numdisabled > 1)) {
1641                         /* One or more queue has already been attached, no need
1642                          * to initialize the device again.
1643                          */
1644                         return 0;
1645                 }
1646         }
1647         else {
1648                 char *name;
1649                 unsigned long flags = 0;
1650                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1651                              MAX_TAP_QUEUES : 1;
1652
1653                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1654                         return -EPERM;
1655                 err = security_tun_dev_create();
1656                 if (err < 0)
1657                         return err;
1658
1659                 /* Set dev type */
1660                 if (ifr->ifr_flags & IFF_TUN) {
1661                         /* TUN device */
1662                         flags |= TUN_TUN_DEV;
1663                         name = "tun%d";
1664                 } else if (ifr->ifr_flags & IFF_TAP) {
1665                         /* TAP device */
1666                         flags |= TUN_TAP_DEV;
1667                         name = "tap%d";
1668                 } else
1669                         return -EINVAL;
1670
1671                 if (*ifr->ifr_name)
1672                         name = ifr->ifr_name;
1673
1674                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1675                                        tun_setup, queues, queues);
1676
1677                 if (!dev)
1678                         return -ENOMEM;
1679
1680                 dev_net_set(dev, net);
1681                 dev->rtnl_link_ops = &tun_link_ops;
1682
1683                 tun = netdev_priv(dev);
1684                 tun->dev = dev;
1685                 tun->flags = flags;
1686                 tun->txflt.count = 0;
1687                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1688
1689                 tun->filter_attached = false;
1690                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1691
1692                 spin_lock_init(&tun->lock);
1693
1694                 err = security_tun_dev_alloc_security(&tun->security);
1695                 if (err < 0)
1696                         goto err_free_dev;
1697
1698                 tun_net_init(dev);
1699                 tun_flow_init(tun);
1700
1701                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1702                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1703                                    NETIF_F_HW_VLAN_STAG_TX;
1704                 dev->features = dev->hw_features;
1705                 dev->vlan_features = dev->features;
1706
1707                 INIT_LIST_HEAD(&tun->disabled);
1708                 err = tun_attach(tun, file);
1709                 if (err < 0)
1710                         goto err_free_dev;
1711
1712                 err = register_netdevice(tun->dev);
1713                 if (err < 0)
1714                         goto err_free_dev;
1715
1716                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1717                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1718                     device_create_file(&tun->dev->dev, &dev_attr_group))
1719                         pr_err("Failed to create tun sysfs files\n");
1720         }
1721
1722         netif_carrier_on(tun->dev);
1723
1724         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1725
1726         if (ifr->ifr_flags & IFF_NO_PI)
1727                 tun->flags |= TUN_NO_PI;
1728         else
1729                 tun->flags &= ~TUN_NO_PI;
1730
1731         /* This flag has no real effect.  We track the value for backwards
1732          * compatibility.
1733          */
1734         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1735                 tun->flags |= TUN_ONE_QUEUE;
1736         else
1737                 tun->flags &= ~TUN_ONE_QUEUE;
1738
1739         if (ifr->ifr_flags & IFF_VNET_HDR)
1740                 tun->flags |= TUN_VNET_HDR;
1741         else
1742                 tun->flags &= ~TUN_VNET_HDR;
1743
1744         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1745                 tun->flags |= TUN_TAP_MQ;
1746         else
1747                 tun->flags &= ~TUN_TAP_MQ;
1748
1749         /* Make sure persistent devices do not get stuck in
1750          * xoff state.
1751          */
1752         if (netif_running(tun->dev))
1753                 netif_tx_wake_all_queues(tun->dev);
1754
1755         strcpy(ifr->ifr_name, tun->dev->name);
1756         return 0;
1757
1758  err_free_dev:
1759         free_netdev(dev);
1760         return err;
1761 }
1762
1763 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1764                        struct ifreq *ifr)
1765 {
1766         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1767
1768         strcpy(ifr->ifr_name, tun->dev->name);
1769
1770         ifr->ifr_flags = tun_flags(tun);
1771
1772 }
1773
1774 /* This is like a cut-down ethtool ops, except done via tun fd so no
1775  * privs required. */
1776 static int set_offload(struct tun_struct *tun, unsigned long arg)
1777 {
1778         netdev_features_t features = 0;
1779
1780         if (arg & TUN_F_CSUM) {
1781                 features |= NETIF_F_HW_CSUM;
1782                 arg &= ~TUN_F_CSUM;
1783
1784                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1785                         if (arg & TUN_F_TSO_ECN) {
1786                                 features |= NETIF_F_TSO_ECN;
1787                                 arg &= ~TUN_F_TSO_ECN;
1788                         }
1789                         if (arg & TUN_F_TSO4)
1790                                 features |= NETIF_F_TSO;
1791                         if (arg & TUN_F_TSO6)
1792                                 features |= NETIF_F_TSO6;
1793                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1794                 }
1795
1796                 if (arg & TUN_F_UFO) {
1797                         features |= NETIF_F_UFO;
1798                         arg &= ~TUN_F_UFO;
1799                 }
1800         }
1801
1802         /* This gives the user a way to test for new features in future by
1803          * trying to set them. */
1804         if (arg)
1805                 return -EINVAL;
1806
1807         tun->set_features = features;
1808         netdev_update_features(tun->dev);
1809
1810         return 0;
1811 }
1812
1813 static void tun_detach_filter(struct tun_struct *tun, int n)
1814 {
1815         int i;
1816         struct tun_file *tfile;
1817
1818         for (i = 0; i < n; i++) {
1819                 tfile = rtnl_dereference(tun->tfiles[i]);
1820                 sk_detach_filter(tfile->socket.sk);
1821         }
1822
1823         tun->filter_attached = false;
1824 }
1825
1826 static int tun_attach_filter(struct tun_struct *tun)
1827 {
1828         int i, ret = 0;
1829         struct tun_file *tfile;
1830
1831         for (i = 0; i < tun->numqueues; i++) {
1832                 tfile = rtnl_dereference(tun->tfiles[i]);
1833                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1834                 if (ret) {
1835                         tun_detach_filter(tun, i);
1836                         return ret;
1837                 }
1838         }
1839
1840         tun->filter_attached = true;
1841         return ret;
1842 }
1843
1844 static void tun_set_sndbuf(struct tun_struct *tun)
1845 {
1846         struct tun_file *tfile;
1847         int i;
1848
1849         for (i = 0; i < tun->numqueues; i++) {
1850                 tfile = rtnl_dereference(tun->tfiles[i]);
1851                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1852         }
1853 }
1854
1855 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1856 {
1857         struct tun_file *tfile = file->private_data;
1858         struct tun_struct *tun;
1859         int ret = 0;
1860
1861         rtnl_lock();
1862
1863         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1864                 tun = tfile->detached;
1865                 if (!tun) {
1866                         ret = -EINVAL;
1867                         goto unlock;
1868                 }
1869                 ret = security_tun_dev_attach_queue(tun->security);
1870                 if (ret < 0)
1871                         goto unlock;
1872                 ret = tun_attach(tun, file);
1873         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1874                 tun = rtnl_dereference(tfile->tun);
1875                 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1876                         ret = -EINVAL;
1877                 else
1878                         __tun_detach(tfile, false);
1879         } else
1880                 ret = -EINVAL;
1881
1882 unlock:
1883         rtnl_unlock();
1884         return ret;
1885 }
1886
1887 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1888                             unsigned long arg, int ifreq_len)
1889 {
1890         struct tun_file *tfile = file->private_data;
1891         struct tun_struct *tun;
1892         void __user* argp = (void __user*)arg;
1893         struct ifreq ifr;
1894         kuid_t owner;
1895         kgid_t group;
1896         int sndbuf;
1897         int vnet_hdr_sz;
1898         int ret;
1899
1900         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1901                 if (copy_from_user(&ifr, argp, ifreq_len))
1902                         return -EFAULT;
1903         } else {
1904                 memset(&ifr, 0, sizeof(ifr));
1905         }
1906         if (cmd == TUNGETFEATURES) {
1907                 /* Currently this just means: "what IFF flags are valid?".
1908                  * This is needed because we never checked for invalid flags on
1909                  * TUNSETIFF. */
1910                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1911                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1912                                 (unsigned int __user*)argp);
1913         } else if (cmd == TUNSETQUEUE)
1914                 return tun_set_queue(file, &ifr);
1915
1916         ret = 0;
1917         rtnl_lock();
1918
1919         tun = __tun_get(tfile);
1920         if (cmd == TUNSETIFF && !tun) {
1921                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1922
1923                 ret = tun_set_iff(tfile->net, file, &ifr);
1924
1925                 if (ret)
1926                         goto unlock;
1927
1928                 if (copy_to_user(argp, &ifr, ifreq_len))
1929                         ret = -EFAULT;
1930                 goto unlock;
1931         }
1932
1933         ret = -EBADFD;
1934         if (!tun)
1935                 goto unlock;
1936
1937         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1938
1939         ret = 0;
1940         switch (cmd) {
1941         case TUNGETIFF:
1942                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1943
1944                 if (copy_to_user(argp, &ifr, ifreq_len))
1945                         ret = -EFAULT;
1946                 break;
1947
1948         case TUNSETNOCSUM:
1949                 /* Disable/Enable checksum */
1950
1951                 /* [unimplemented] */
1952                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1953                           arg ? "disabled" : "enabled");
1954                 break;
1955
1956         case TUNSETPERSIST:
1957                 /* Disable/Enable persist mode. Keep an extra reference to the
1958                  * module to prevent the module being unprobed.
1959                  */
1960                 if (arg && !(tun->flags & TUN_PERSIST)) {
1961                         tun->flags |= TUN_PERSIST;
1962                         __module_get(THIS_MODULE);
1963                 }
1964                 if (!arg && (tun->flags & TUN_PERSIST)) {
1965                         tun->flags &= ~TUN_PERSIST;
1966                         module_put(THIS_MODULE);
1967                 }
1968
1969                 tun_debug(KERN_INFO, tun, "persist %s\n",
1970                           arg ? "enabled" : "disabled");
1971                 break;
1972
1973         case TUNSETOWNER:
1974                 /* Set owner of the device */
1975                 owner = make_kuid(current_user_ns(), arg);
1976                 if (!uid_valid(owner)) {
1977                         ret = -EINVAL;
1978                         break;
1979                 }
1980                 tun->owner = owner;
1981                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1982                           from_kuid(&init_user_ns, tun->owner));
1983                 break;
1984
1985         case TUNSETGROUP:
1986                 /* Set group of the device */
1987                 group = make_kgid(current_user_ns(), arg);
1988                 if (!gid_valid(group)) {
1989                         ret = -EINVAL;
1990                         break;
1991                 }
1992                 tun->group = group;
1993                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1994                           from_kgid(&init_user_ns, tun->group));
1995                 break;
1996
1997         case TUNSETLINK:
1998                 /* Only allow setting the type when the interface is down */
1999                 if (tun->dev->flags & IFF_UP) {
2000                         tun_debug(KERN_INFO, tun,
2001                                   "Linktype set failed because interface is up\n");
2002                         ret = -EBUSY;
2003                 } else {
2004                         tun->dev->type = (int) arg;
2005                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2006                                   tun->dev->type);
2007                         ret = 0;
2008                 }
2009                 break;
2010
2011 #ifdef TUN_DEBUG
2012         case TUNSETDEBUG:
2013                 tun->debug = arg;
2014                 break;
2015 #endif
2016         case TUNSETOFFLOAD:
2017                 ret = set_offload(tun, arg);
2018                 break;
2019
2020         case TUNSETTXFILTER:
2021                 /* Can be set only for TAPs */
2022                 ret = -EINVAL;
2023                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2024                         break;
2025                 ret = update_filter(&tun->txflt, (void __user *)arg);
2026                 break;
2027
2028         case SIOCGIFHWADDR:
2029                 /* Get hw address */
2030                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2031                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2032                 if (copy_to_user(argp, &ifr, ifreq_len))
2033                         ret = -EFAULT;
2034                 break;
2035
2036         case SIOCSIFHWADDR:
2037                 /* Set hw address */
2038                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2039                           ifr.ifr_hwaddr.sa_data);
2040
2041                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2042                 break;
2043
2044         case TUNGETSNDBUF:
2045                 sndbuf = tfile->socket.sk->sk_sndbuf;
2046                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2047                         ret = -EFAULT;
2048                 break;
2049
2050         case TUNSETSNDBUF:
2051                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2052                         ret = -EFAULT;
2053                         break;
2054                 }
2055
2056                 tun->sndbuf = sndbuf;
2057                 tun_set_sndbuf(tun);
2058                 break;
2059
2060         case TUNGETVNETHDRSZ:
2061                 vnet_hdr_sz = tun->vnet_hdr_sz;
2062                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2063                         ret = -EFAULT;
2064                 break;
2065
2066         case TUNSETVNETHDRSZ:
2067                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2068                         ret = -EFAULT;
2069                         break;
2070                 }
2071                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2072                         ret = -EINVAL;
2073                         break;
2074                 }
2075
2076                 tun->vnet_hdr_sz = vnet_hdr_sz;
2077                 break;
2078
2079         case TUNATTACHFILTER:
2080                 /* Can be set only for TAPs */
2081                 ret = -EINVAL;
2082                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2083                         break;
2084                 ret = -EFAULT;
2085                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2086                         break;
2087
2088                 ret = tun_attach_filter(tun);
2089                 break;
2090
2091         case TUNDETACHFILTER:
2092                 /* Can be set only for TAPs */
2093                 ret = -EINVAL;
2094                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2095                         break;
2096                 ret = 0;
2097                 tun_detach_filter(tun, tun->numqueues);
2098                 break;
2099
2100         default:
2101                 ret = -EINVAL;
2102                 break;
2103         }
2104
2105 unlock:
2106         rtnl_unlock();
2107         if (tun)
2108                 tun_put(tun);
2109         return ret;
2110 }
2111
2112 static long tun_chr_ioctl(struct file *file,
2113                           unsigned int cmd, unsigned long arg)
2114 {
2115         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2116 }
2117
2118 #ifdef CONFIG_COMPAT
2119 static long tun_chr_compat_ioctl(struct file *file,
2120                          unsigned int cmd, unsigned long arg)
2121 {
2122         switch (cmd) {
2123         case TUNSETIFF:
2124         case TUNGETIFF:
2125         case TUNSETTXFILTER:
2126         case TUNGETSNDBUF:
2127         case TUNSETSNDBUF:
2128         case SIOCGIFHWADDR:
2129         case SIOCSIFHWADDR:
2130                 arg = (unsigned long)compat_ptr(arg);
2131                 break;
2132         default:
2133                 arg = (compat_ulong_t)arg;
2134                 break;
2135         }
2136
2137         /*
2138          * compat_ifreq is shorter than ifreq, so we must not access beyond
2139          * the end of that structure. All fields that are used in this
2140          * driver are compatible though, we don't need to convert the
2141          * contents.
2142          */
2143         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2144 }
2145 #endif /* CONFIG_COMPAT */
2146
2147 static int tun_chr_fasync(int fd, struct file *file, int on)
2148 {
2149         struct tun_file *tfile = file->private_data;
2150         int ret;
2151
2152         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2153                 goto out;
2154
2155         if (on) {
2156                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2157                 if (ret)
2158                         goto out;
2159                 tfile->flags |= TUN_FASYNC;
2160         } else
2161                 tfile->flags &= ~TUN_FASYNC;
2162         ret = 0;
2163 out:
2164         return ret;
2165 }
2166
2167 static int tun_chr_open(struct inode *inode, struct file * file)
2168 {
2169         struct tun_file *tfile;
2170
2171         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2172
2173         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2174                                             &tun_proto);
2175         if (!tfile)
2176                 return -ENOMEM;
2177         rcu_assign_pointer(tfile->tun, NULL);
2178         tfile->net = get_net(current->nsproxy->net_ns);
2179         tfile->flags = 0;
2180
2181         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2182         init_waitqueue_head(&tfile->wq.wait);
2183
2184         tfile->socket.file = file;
2185         tfile->socket.ops = &tun_socket_ops;
2186
2187         sock_init_data(&tfile->socket, &tfile->sk);
2188         sk_change_net(&tfile->sk, tfile->net);
2189
2190         tfile->sk.sk_write_space = tun_sock_write_space;
2191         tfile->sk.sk_sndbuf = INT_MAX;
2192
2193         file->private_data = tfile;
2194         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2195         INIT_LIST_HEAD(&tfile->next);
2196
2197         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2198
2199         return 0;
2200 }
2201
2202 static int tun_chr_close(struct inode *inode, struct file *file)
2203 {
2204         struct tun_file *tfile = file->private_data;
2205         struct net *net = tfile->net;
2206
2207         tun_detach(tfile, true);
2208         put_net(net);
2209
2210         return 0;
2211 }
2212
2213 static const struct file_operations tun_fops = {
2214         .owner  = THIS_MODULE,
2215         .llseek = no_llseek,
2216         .read  = do_sync_read,
2217         .aio_read  = tun_chr_aio_read,
2218         .write = do_sync_write,
2219         .aio_write = tun_chr_aio_write,
2220         .poll   = tun_chr_poll,
2221         .unlocked_ioctl = tun_chr_ioctl,
2222 #ifdef CONFIG_COMPAT
2223         .compat_ioctl = tun_chr_compat_ioctl,
2224 #endif
2225         .open   = tun_chr_open,
2226         .release = tun_chr_close,
2227         .fasync = tun_chr_fasync
2228 };
2229
2230 static struct miscdevice tun_miscdev = {
2231         .minor = TUN_MINOR,
2232         .name = "tun",
2233         .nodename = "net/tun",
2234         .fops = &tun_fops,
2235 };
2236
2237 /* ethtool interface */
2238
2239 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2240 {
2241         cmd->supported          = 0;
2242         cmd->advertising        = 0;
2243         ethtool_cmd_speed_set(cmd, SPEED_10);
2244         cmd->duplex             = DUPLEX_FULL;
2245         cmd->port               = PORT_TP;
2246         cmd->phy_address        = 0;
2247         cmd->transceiver        = XCVR_INTERNAL;
2248         cmd->autoneg            = AUTONEG_DISABLE;
2249         cmd->maxtxpkt           = 0;
2250         cmd->maxrxpkt           = 0;
2251         return 0;
2252 }
2253
2254 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2255 {
2256         struct tun_struct *tun = netdev_priv(dev);
2257
2258         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2259         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2260
2261         switch (tun->flags & TUN_TYPE_MASK) {
2262         case TUN_TUN_DEV:
2263                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2264                 break;
2265         case TUN_TAP_DEV:
2266                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2267                 break;
2268         }
2269 }
2270
2271 static u32 tun_get_msglevel(struct net_device *dev)
2272 {
2273 #ifdef TUN_DEBUG
2274         struct tun_struct *tun = netdev_priv(dev);
2275         return tun->debug;
2276 #else
2277         return -EOPNOTSUPP;
2278 #endif
2279 }
2280
2281 static void tun_set_msglevel(struct net_device *dev, u32 value)
2282 {
2283 #ifdef TUN_DEBUG
2284         struct tun_struct *tun = netdev_priv(dev);
2285         tun->debug = value;
2286 #endif
2287 }
2288
2289 static const struct ethtool_ops tun_ethtool_ops = {
2290         .get_settings   = tun_get_settings,
2291         .get_drvinfo    = tun_get_drvinfo,
2292         .get_msglevel   = tun_get_msglevel,
2293         .set_msglevel   = tun_set_msglevel,
2294         .get_link       = ethtool_op_get_link,
2295         .get_ts_info    = ethtool_op_get_ts_info,
2296 };
2297
2298
2299 static int __init tun_init(void)
2300 {
2301         int ret = 0;
2302
2303         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2304         pr_info("%s\n", DRV_COPYRIGHT);
2305
2306         ret = rtnl_link_register(&tun_link_ops);
2307         if (ret) {
2308                 pr_err("Can't register link_ops\n");
2309                 goto err_linkops;
2310         }
2311
2312         ret = misc_register(&tun_miscdev);
2313         if (ret) {
2314                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2315                 goto err_misc;
2316         }
2317         return  0;
2318 err_misc:
2319         rtnl_link_unregister(&tun_link_ops);
2320 err_linkops:
2321         return ret;
2322 }
2323
2324 static void tun_cleanup(void)
2325 {
2326         misc_deregister(&tun_miscdev);
2327         rtnl_link_unregister(&tun_link_ops);
2328 }
2329
2330 /* Get an underlying socket object from tun file.  Returns error unless file is
2331  * attached to a device.  The returned object works like a packet socket, it
2332  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2333  * holding a reference to the file for as long as the socket is in use. */
2334 struct socket *tun_get_socket(struct file *file)
2335 {
2336         struct tun_file *tfile;
2337         if (file->f_op != &tun_fops)
2338                 return ERR_PTR(-EINVAL);
2339         tfile = file->private_data;
2340         if (!tfile)
2341                 return ERR_PTR(-EBADFD);
2342         return &tfile->socket;
2343 }
2344 EXPORT_SYMBOL_GPL(tun_get_socket);
2345
2346 module_init(tun_init);
2347 module_exit(tun_cleanup);
2348 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2349 MODULE_AUTHOR(DRV_COPYRIGHT);
2350 MODULE_LICENSE("GPL");
2351 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2352 MODULE_ALIAS("devname:net/tun");