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