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 #include <net/flow_keys.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 = 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         /* Orphan the skb - required as we might hang on to it
 744          * for indefinite time. */
 745         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 746                 goto drop;
 747         skb_orphan(skb);
 748
 749         nf_reset(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         struct flow_keys keys;
1054
1055         if (!(tun->flags & TUN_NO_PI)) {
1056                 if ((len -= sizeof(pi)) > total_len)
1057                         return -EINVAL;
1058
1059                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1060                         return -EFAULT;
1061                 offset += sizeof(pi);
1062         }
1063
1064         if (tun->flags & TUN_VNET_HDR) {
1065                 if ((len -= tun->vnet_hdr_sz) > total_len)
1066                         return -EINVAL;
1067
1068                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1069                         return -EFAULT;
1070
1071                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1072                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1073                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1074
1075                 if (gso.hdr_len > len)
1076                         return -EINVAL;
1077                 offset += tun->vnet_hdr_sz;
1078         }
1079
1080         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1081                 align += NET_IP_ALIGN;
1082                 if (unlikely(len < ETH_HLEN ||
1083                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1084                         return -EINVAL;
1085         }
1086
1087         if (msg_control)
1088                 zerocopy = true;
1089
1090         if (zerocopy) {
1091                 /* Userspace may produce vectors with count greater than
1092                  * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1093                  * to let the rest of data to be fit in the frags.
1094                  */
1095                 if (count > MAX_SKB_FRAGS) {
1096                         copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1097                         if (copylen < offset)
1098                                 copylen = 0;
1099                         else
1100                                 copylen -= offset;
1101                 } else
1102                                 copylen = 0;
1103                 /* There are 256 bytes to be copied in skb, so there is enough
1104                  * room for skb expand head in case it is used.
1105                  * The rest of the buffer is mapped from userspace.
1106                  */
1107                 if (copylen < gso.hdr_len)
1108                         copylen = gso.hdr_len;
1109                 if (!copylen)
1110                         copylen = GOODCOPY_LEN;
1111         } else
1112                 copylen = len;
1113
1114         skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
1115         if (IS_ERR(skb)) {
1116                 if (PTR_ERR(skb) != -EAGAIN)
1117                         tun->dev->stats.rx_dropped++;
1118                 return PTR_ERR(skb);
1119         }
1120
1121         if (zerocopy)
1122                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1123         else
1124                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1125
1126         if (err) {
1127                 tun->dev->stats.rx_dropped++;
1128                 kfree_skb(skb);
1129                 return -EFAULT;
1130         }
1131
1132         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1133                 if (!skb_partial_csum_set(skb, gso.csum_start,
1134                                           gso.csum_offset)) {
1135                         tun->dev->stats.rx_frame_errors++;
1136                         kfree_skb(skb);
1137                         return -EINVAL;
1138                 }
1139         }
1140
1141         switch (tun->flags & TUN_TYPE_MASK) {
1142         case TUN_TUN_DEV:
1143                 if (tun->flags & TUN_NO_PI) {
1144                         switch (skb->data[0] & 0xf0) {
1145                         case 0x40:
1146                                 pi.proto = htons(ETH_P_IP);
1147                                 break;
1148                         case 0x60:
1149                                 pi.proto = htons(ETH_P_IPV6);
1150                                 break;
1151                         default:
1152                                 tun->dev->stats.rx_dropped++;
1153                                 kfree_skb(skb);
1154                                 return -EINVAL;
1155                         }
1156                 }
1157
1158                 skb_reset_mac_header(skb);
1159                 skb->protocol = pi.proto;
1160                 skb->dev = tun->dev;
1161                 break;
1162         case TUN_TAP_DEV:
1163                 skb->protocol = eth_type_trans(skb, tun->dev);
1164                 break;
1165         }
1166
1167         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1168                 pr_debug("GSO!\n");
1169                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1170                 case VIRTIO_NET_HDR_GSO_TCPV4:
1171                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1172                         break;
1173                 case VIRTIO_NET_HDR_GSO_TCPV6:
1174                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1175                         break;
1176                 case VIRTIO_NET_HDR_GSO_UDP:
1177                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1178                         break;
1179                 default:
1180                         tun->dev->stats.rx_frame_errors++;
1181                         kfree_skb(skb);
1182                         return -EINVAL;
1183                 }
1184
1185                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1186                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1187
1188                 skb_shinfo(skb)->gso_size = gso.gso_size;
1189                 if (skb_shinfo(skb)->gso_size == 0) {
1190                         tun->dev->stats.rx_frame_errors++;
1191                         kfree_skb(skb);
1192                         return -EINVAL;
1193                 }
1194
1195                 /* Header must be checked, and gso_segs computed. */
1196                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1197                 skb_shinfo(skb)->gso_segs = 0;
1198         }
1199
1200         /* copy skb_ubuf_info for callback when skb has no error */
1201         if (zerocopy) {
1202                 skb_shinfo(skb)->destructor_arg = msg_control;
1203                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1204                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1205         }
1206
1207         skb_reset_network_header(skb);
1208
1209         if (skb->ip_summed == CHECKSUM_PARTIAL)
1210                 skb_set_transport_header(skb, skb_checksum_start_offset(skb));
1211         else if (skb_flow_dissect(skb, &keys))
1212                 skb_set_transport_header(skb, keys.thoff);
1213         else
1214                 skb_reset_transport_header(skb);
1215
1216         rxhash = skb_get_rxhash(skb);
1217         netif_rx_ni(skb);
1218
1219         tun->dev->stats.rx_packets++;
1220         tun->dev->stats.rx_bytes += len;
1221
1222         tun_flow_update(tun, rxhash, tfile);
1223         return total_len;
1224 }
1225
1226 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1227                               unsigned long count, loff_t pos)
1228 {
1229         struct file *file = iocb->ki_filp;
1230         struct tun_struct *tun = tun_get(file);
1231         struct tun_file *tfile = file->private_data;
1232         ssize_t result;
1233
1234         if (!tun)
1235                 return -EBADFD;
1236
1237         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1238
1239         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1240                               count, file->f_flags & O_NONBLOCK);
1241
1242         tun_put(tun);
1243         return result;
1244 }
1245
1246 /* Put packet to the user space buffer */
1247 static ssize_t tun_put_user(struct tun_struct *tun,
1248                             struct tun_file *tfile,
1249                             struct sk_buff *skb,
1250                             const struct iovec *iv, int len)
1251 {
1252         struct tun_pi pi = { 0, skb->protocol };
1253         ssize_t total = 0;
1254
1255         if (!(tun->flags & TUN_NO_PI)) {
1256                 if ((len -= sizeof(pi)) < 0)
1257                         return -EINVAL;
1258
1259                 if (len < skb->len) {
1260                         /* Packet will be striped */
1261                         pi.flags |= TUN_PKT_STRIP;
1262                 }
1263
1264                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1265                         return -EFAULT;
1266                 total += sizeof(pi);
1267         }
1268
1269         if (tun->flags & TUN_VNET_HDR) {
1270                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1271                 if ((len -= tun->vnet_hdr_sz) < 0)
1272                         return -EINVAL;
1273
1274                 if (skb_is_gso(skb)) {
1275                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1276
1277                         /* This is a hint as to how much should be linear. */
1278                         gso.hdr_len = skb_headlen(skb);
1279                         gso.gso_size = sinfo->gso_size;
1280                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1281                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1282                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1283                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1284                         else if (sinfo->gso_type & SKB_GSO_UDP)
1285                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1286                         else {
1287                                 pr_err("unexpected GSO type: "
1288                                        "0x%x, gso_size %d, hdr_len %d\n",
1289                                        sinfo->gso_type, gso.gso_size,
1290                                        gso.hdr_len);
1291                                 print_hex_dump(KERN_ERR, "tun: ",
1292                                                DUMP_PREFIX_NONE,
1293                                                16, 1, skb->head,
1294                                                min((int)gso.hdr_len, 64), true);
1295                                 WARN_ON_ONCE(1);
1296                                 return -EINVAL;
1297                         }
1298                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1299                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1300                 } else
1301                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1302
1303                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1304                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1305                         gso.csum_start = skb_checksum_start_offset(skb);
1306                         gso.csum_offset = skb->csum_offset;
1307                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1308                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1309                 } /* else everything is zero */
1310
1311                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1312                                                sizeof(gso))))
1313                         return -EFAULT;
1314                 total += tun->vnet_hdr_sz;
1315         }
1316
1317         len = min_t(int, skb->len, len);
1318
1319         skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1320         total += skb->len;
1321
1322         tun->dev->stats.tx_packets++;
1323         tun->dev->stats.tx_bytes += len;
1324
1325         return total;
1326 }
1327
1328 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1329                            struct kiocb *iocb, const struct iovec *iv,
1330                            ssize_t len, int noblock)
1331 {
1332         DECLARE_WAITQUEUE(wait, current);
1333         struct sk_buff *skb;
1334         ssize_t ret = 0;
1335
1336         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1337
1338         if (unlikely(!noblock))
1339                 add_wait_queue(&tfile->wq.wait, &wait);
1340         while (len) {
1341                 current->state = TASK_INTERRUPTIBLE;
1342
1343                 /* Read frames from the queue */
1344                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1345                         if (noblock) {
1346                                 ret = -EAGAIN;
1347                                 break;
1348                         }
1349                         if (signal_pending(current)) {
1350                                 ret = -ERESTARTSYS;
1351                                 break;
1352                         }
1353                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1354                                 ret = -EIO;
1355                                 break;
1356                         }
1357
1358                         /* Nothing to read, let's sleep */
1359                         schedule();
1360                         continue;
1361                 }
1362
1363                 ret = tun_put_user(tun, tfile, skb, iv, len);
1364                 kfree_skb(skb);
1365                 break;
1366         }
1367
1368         current->state = TASK_RUNNING;
1369         if (unlikely(!noblock))
1370                 remove_wait_queue(&tfile->wq.wait, &wait);
1371
1372         return ret;
1373 }
1374
1375 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1376                             unsigned long count, loff_t pos)
1377 {
1378         struct file *file = iocb->ki_filp;
1379         struct tun_file *tfile = file->private_data;
1380         struct tun_struct *tun = __tun_get(tfile);
1381         ssize_t len, ret;
1382
1383         if (!tun)
1384                 return -EBADFD;
1385         len = iov_length(iv, count);
1386         if (len < 0) {
1387                 ret = -EINVAL;
1388                 goto out;
1389         }
1390
1391         ret = tun_do_read(tun, tfile, iocb, iv, len,
1392                           file->f_flags & O_NONBLOCK);
1393         ret = min_t(ssize_t, ret, len);
1394 out:
1395         tun_put(tun);
1396         return ret;
1397 }
1398
1399 static void tun_free_netdev(struct net_device *dev)
1400 {
1401         struct tun_struct *tun = netdev_priv(dev);
1402
1403         BUG_ON(!(list_empty(&tun->disabled)));
1404         tun_flow_uninit(tun);
1405         security_tun_dev_free_security(tun->security);
1406         free_netdev(dev);
1407 }
1408
1409 static void tun_setup(struct net_device *dev)
1410 {
1411         struct tun_struct *tun = netdev_priv(dev);
1412
1413         tun->owner = INVALID_UID;
1414         tun->group = INVALID_GID;
1415
1416         dev->ethtool_ops = &tun_ethtool_ops;
1417         dev->destructor = tun_free_netdev;
1418 }
1419
1420 /* Trivial set of netlink ops to allow deleting tun or tap
1421  * device with netlink.
1422  */
1423 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1424 {
1425         return -EINVAL;
1426 }
1427
1428 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1429         .kind           = DRV_NAME,
1430         .priv_size      = sizeof(struct tun_struct),
1431         .setup          = tun_setup,
1432         .validate       = tun_validate,
1433 };
1434
1435 static void tun_sock_write_space(struct sock *sk)
1436 {
1437         struct tun_file *tfile;
1438         wait_queue_head_t *wqueue;
1439
1440         if (!sock_writeable(sk))
1441                 return;
1442
1443         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1444                 return;
1445
1446         wqueue = sk_sleep(sk);
1447         if (wqueue && waitqueue_active(wqueue))
1448                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1449                                                 POLLWRNORM | POLLWRBAND);
1450
1451         tfile = container_of(sk, struct tun_file, sk);
1452         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1453 }
1454
1455 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1456                        struct msghdr *m, size_t total_len)
1457 {
1458         int ret;
1459         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1460         struct tun_struct *tun = __tun_get(tfile);
1461
1462         if (!tun)
1463                 return -EBADFD;
1464         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1465                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1466         tun_put(tun);
1467         return ret;
1468 }
1469
1470
1471 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1472                        struct msghdr *m, size_t total_len,
1473                        int flags)
1474 {
1475         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1476         struct tun_struct *tun = __tun_get(tfile);
1477         int ret;
1478
1479         if (!tun)
1480                 return -EBADFD;
1481
1482         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1483                 return -EINVAL;
1484         ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1485                           flags & MSG_DONTWAIT);
1486         if (ret > total_len) {
1487                 m->msg_flags |= MSG_TRUNC;
1488                 ret = flags & MSG_TRUNC ? ret : total_len;
1489         }
1490         tun_put(tun);
1491         return ret;
1492 }
1493
1494 static int tun_release(struct socket *sock)
1495 {
1496         if (sock->sk)
1497                 sock_put(sock->sk);
1498         return 0;
1499 }
1500
1501 /* Ops structure to mimic raw sockets with tun */
1502 static const struct proto_ops tun_socket_ops = {
1503         .sendmsg = tun_sendmsg,
1504         .recvmsg = tun_recvmsg,
1505         .release = tun_release,
1506 };
1507
1508 static struct proto tun_proto = {
1509         .name           = "tun",
1510         .owner          = THIS_MODULE,
1511         .obj_size       = sizeof(struct tun_file),
1512 };
1513
1514 static int tun_flags(struct tun_struct *tun)
1515 {
1516         int flags = 0;
1517
1518         if (tun->flags & TUN_TUN_DEV)
1519                 flags |= IFF_TUN;
1520         else
1521                 flags |= IFF_TAP;
1522
1523         if (tun->flags & TUN_NO_PI)
1524                 flags |= IFF_NO_PI;
1525
1526         /* This flag has no real effect.  We track the value for backwards
1527          * compatibility.
1528          */
1529         if (tun->flags & TUN_ONE_QUEUE)
1530                 flags |= IFF_ONE_QUEUE;
1531
1532         if (tun->flags & TUN_VNET_HDR)
1533                 flags |= IFF_VNET_HDR;
1534
1535         if (tun->flags & TUN_TAP_MQ)
1536                 flags |= IFF_MULTI_QUEUE;
1537
1538         return flags;
1539 }
1540
1541 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1542                               char *buf)
1543 {
1544         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1545         return sprintf(buf, "0x%x\n", tun_flags(tun));
1546 }
1547
1548 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1549                               char *buf)
1550 {
1551         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1552         return uid_valid(tun->owner)?
1553                 sprintf(buf, "%u\n",
1554                         from_kuid_munged(current_user_ns(), tun->owner)):
1555                 sprintf(buf, "-1\n");
1556 }
1557
1558 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1559                               char *buf)
1560 {
1561         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1562         return gid_valid(tun->group) ?
1563                 sprintf(buf, "%u\n",
1564                         from_kgid_munged(current_user_ns(), tun->group)):
1565                 sprintf(buf, "-1\n");
1566 }
1567
1568 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1569 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1570 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1571
1572 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1573 {
1574         struct tun_struct *tun;
1575         struct tun_file *tfile = file->private_data;
1576         struct net_device *dev;
1577         int err;
1578
1579         if (tfile->detached)
1580                 return -EINVAL;
1581
1582         dev = __dev_get_by_name(net, ifr->ifr_name);
1583         if (dev) {
1584                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1585                         return -EBUSY;
1586                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1587                         tun = netdev_priv(dev);
1588                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1589                         tun = netdev_priv(dev);
1590                 else
1591                         return -EINVAL;
1592
1593                 if (tun_not_capable(tun))
1594                         return -EPERM;
1595                 err = security_tun_dev_open(tun->security);
1596                 if (err < 0)
1597                         return err;
1598
1599                 err = tun_attach(tun, file);
1600                 if (err < 0)
1601                         return err;
1602
1603                 if (tun->flags & TUN_TAP_MQ &&
1604                     (tun->numqueues + tun->numdisabled > 1))
1605                         return err;
1606         }
1607         else {
1608                 char *name;
1609                 unsigned long flags = 0;
1610                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1611                              MAX_TAP_QUEUES : 1;
1612
1613                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1614                         return -EPERM;
1615                 err = security_tun_dev_create();
1616                 if (err < 0)
1617                         return err;
1618
1619                 /* Set dev type */
1620                 if (ifr->ifr_flags & IFF_TUN) {
1621                         /* TUN device */
1622                         flags |= TUN_TUN_DEV;
1623                         name = "tun%d";
1624                 } else if (ifr->ifr_flags & IFF_TAP) {
1625                         /* TAP device */
1626                         flags |= TUN_TAP_DEV;
1627                         name = "tap%d";
1628                 } else
1629                         return -EINVAL;
1630
1631                 if (*ifr->ifr_name)
1632                         name = ifr->ifr_name;
1633
1634                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1635                                        tun_setup, queues, queues);
1636
1637                 if (!dev)
1638                         return -ENOMEM;
1639
1640                 dev_net_set(dev, net);
1641                 dev->rtnl_link_ops = &tun_link_ops;
1642
1643                 tun = netdev_priv(dev);
1644                 tun->dev = dev;
1645                 tun->flags = flags;
1646                 tun->txflt.count = 0;
1647                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1648
1649                 tun->filter_attached = false;
1650                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1651
1652                 spin_lock_init(&tun->lock);
1653
1654                 err = security_tun_dev_alloc_security(&tun->security);
1655                 if (err < 0)
1656                         goto err_free_dev;
1657
1658                 tun_net_init(dev);
1659
1660                 err = tun_flow_init(tun);
1661                 if (err < 0)
1662                         goto err_free_dev;
1663
1664                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1665                         TUN_USER_FEATURES;
1666                 dev->features = dev->hw_features;
1667
1668                 INIT_LIST_HEAD(&tun->disabled);
1669                 err = tun_attach(tun, file);
1670                 if (err < 0)
1671                         goto err_free_dev;
1672
1673                 err = register_netdevice(tun->dev);
1674                 if (err < 0)
1675                         goto err_free_dev;
1676
1677                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1678                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1679                     device_create_file(&tun->dev->dev, &dev_attr_group))
1680                         pr_err("Failed to create tun sysfs files\n");
1681         }
1682
1683         netif_carrier_on(tun->dev);
1684
1685         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1686
1687         if (ifr->ifr_flags & IFF_NO_PI)
1688                 tun->flags |= TUN_NO_PI;
1689         else
1690                 tun->flags &= ~TUN_NO_PI;
1691
1692         /* This flag has no real effect.  We track the value for backwards
1693          * compatibility.
1694          */
1695         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1696                 tun->flags |= TUN_ONE_QUEUE;
1697         else
1698                 tun->flags &= ~TUN_ONE_QUEUE;
1699
1700         if (ifr->ifr_flags & IFF_VNET_HDR)
1701                 tun->flags |= TUN_VNET_HDR;
1702         else
1703                 tun->flags &= ~TUN_VNET_HDR;
1704
1705         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1706                 tun->flags |= TUN_TAP_MQ;
1707         else
1708                 tun->flags &= ~TUN_TAP_MQ;
1709
1710         /* Make sure persistent devices do not get stuck in
1711          * xoff state.
1712          */
1713         if (netif_running(tun->dev))
1714                 netif_tx_wake_all_queues(tun->dev);
1715
1716         strcpy(ifr->ifr_name, tun->dev->name);
1717         return 0;
1718
1719  err_free_dev:
1720         free_netdev(dev);
1721         return err;
1722 }
1723
1724 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1725                        struct ifreq *ifr)
1726 {
1727         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1728
1729         strcpy(ifr->ifr_name, tun->dev->name);
1730
1731         ifr->ifr_flags = tun_flags(tun);
1732
1733 }
1734
1735 /* This is like a cut-down ethtool ops, except done via tun fd so no
1736  * privs required. */
1737 static int set_offload(struct tun_struct *tun, unsigned long arg)
1738 {
1739         netdev_features_t features = 0;
1740
1741         if (arg & TUN_F_CSUM) {
1742                 features |= NETIF_F_HW_CSUM;
1743                 arg &= ~TUN_F_CSUM;
1744
1745                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1746                         if (arg & TUN_F_TSO_ECN) {
1747                                 features |= NETIF_F_TSO_ECN;
1748                                 arg &= ~TUN_F_TSO_ECN;
1749                         }
1750                         if (arg & TUN_F_TSO4)
1751                                 features |= NETIF_F_TSO;
1752                         if (arg & TUN_F_TSO6)
1753                                 features |= NETIF_F_TSO6;
1754                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1755                 }
1756
1757                 if (arg & TUN_F_UFO) {
1758                         features |= NETIF_F_UFO;
1759                         arg &= ~TUN_F_UFO;
1760                 }
1761         }
1762
1763         /* This gives the user a way to test for new features in future by
1764          * trying to set them. */
1765         if (arg)
1766                 return -EINVAL;
1767
1768         tun->set_features = features;
1769         netdev_update_features(tun->dev);
1770
1771         return 0;
1772 }
1773
1774 static void tun_detach_filter(struct tun_struct *tun, int n)
1775 {
1776         int i;
1777         struct tun_file *tfile;
1778
1779         for (i = 0; i < n; i++) {
1780                 tfile = rtnl_dereference(tun->tfiles[i]);
1781                 sk_detach_filter(tfile->socket.sk);
1782         }
1783
1784         tun->filter_attached = false;
1785 }
1786
1787 static int tun_attach_filter(struct tun_struct *tun)
1788 {
1789         int i, ret = 0;
1790         struct tun_file *tfile;
1791
1792         for (i = 0; i < tun->numqueues; i++) {
1793                 tfile = rtnl_dereference(tun->tfiles[i]);
1794                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1795                 if (ret) {
1796                         tun_detach_filter(tun, i);
1797                         return ret;
1798                 }
1799         }
1800
1801         tun->filter_attached = true;
1802         return ret;
1803 }
1804
1805 static void tun_set_sndbuf(struct tun_struct *tun)
1806 {
1807         struct tun_file *tfile;
1808         int i;
1809
1810         for (i = 0; i < tun->numqueues; i++) {
1811                 tfile = rtnl_dereference(tun->tfiles[i]);
1812                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1813         }
1814 }
1815
1816 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1817 {
1818         struct tun_file *tfile = file->private_data;
1819         struct tun_struct *tun;
1820         int ret = 0;
1821
1822         rtnl_lock();
1823
1824         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1825                 tun = tfile->detached;
1826                 if (!tun) {
1827                         ret = -EINVAL;
1828                         goto unlock;
1829                 }
1830                 ret = security_tun_dev_attach_queue(tun->security);
1831                 if (ret < 0)
1832                         goto unlock;
1833                 ret = tun_attach(tun, file);
1834         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1835                 tun = rtnl_dereference(tfile->tun);
1836                 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1837                         ret = -EINVAL;
1838                 else
1839                         __tun_detach(tfile, false);
1840         } else
1841                 ret = -EINVAL;
1842
1843 unlock:
1844         rtnl_unlock();
1845         return ret;
1846 }
1847
1848 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1849                             unsigned long arg, int ifreq_len)
1850 {
1851         struct tun_file *tfile = file->private_data;
1852         struct tun_struct *tun;
1853         void __user* argp = (void __user*)arg;
1854         struct ifreq ifr;
1855         kuid_t owner;
1856         kgid_t group;
1857         int sndbuf;
1858         int vnet_hdr_sz;
1859         int ret;
1860
1861         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1862                 if (copy_from_user(&ifr, argp, ifreq_len))
1863                         return -EFAULT;
1864         } else {
1865                 memset(&ifr, 0, sizeof(ifr));
1866         }
1867         if (cmd == TUNGETFEATURES) {
1868                 /* Currently this just means: "what IFF flags are valid?".
1869                  * This is needed because we never checked for invalid flags on
1870                  * TUNSETIFF. */
1871                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1872                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1873                                 (unsigned int __user*)argp);
1874         } else if (cmd == TUNSETQUEUE)
1875                 return tun_set_queue(file, &ifr);
1876
1877         ret = 0;
1878         rtnl_lock();
1879
1880         tun = __tun_get(tfile);
1881         if (cmd == TUNSETIFF && !tun) {
1882                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1883
1884                 ret = tun_set_iff(tfile->net, file, &ifr);
1885
1886                 if (ret)
1887                         goto unlock;
1888
1889                 if (copy_to_user(argp, &ifr, ifreq_len))
1890                         ret = -EFAULT;
1891                 goto unlock;
1892         }
1893
1894         ret = -EBADFD;
1895         if (!tun)
1896                 goto unlock;
1897
1898         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1899
1900         ret = 0;
1901         switch (cmd) {
1902         case TUNGETIFF:
1903                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1904
1905                 if (copy_to_user(argp, &ifr, ifreq_len))
1906                         ret = -EFAULT;
1907                 break;
1908
1909         case TUNSETNOCSUM:
1910                 /* Disable/Enable checksum */
1911
1912                 /* [unimplemented] */
1913                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1914                           arg ? "disabled" : "enabled");
1915                 break;
1916
1917         case TUNSETPERSIST:
1918                 /* Disable/Enable persist mode. Keep an extra reference to the
1919                  * module to prevent the module being unprobed.
1920                  */
1921                 if (arg && !(tun->flags & TUN_PERSIST)) {
1922                         tun->flags |= TUN_PERSIST;
1923                         __module_get(THIS_MODULE);
1924                 }
1925                 if (!arg && (tun->flags & TUN_PERSIST)) {
1926                         tun->flags &= ~TUN_PERSIST;
1927                         module_put(THIS_MODULE);
1928                 }
1929
1930                 tun_debug(KERN_INFO, tun, "persist %s\n",
1931                           arg ? "enabled" : "disabled");
1932                 break;
1933
1934         case TUNSETOWNER:
1935                 /* Set owner of the device */
1936                 owner = make_kuid(current_user_ns(), arg);
1937                 if (!uid_valid(owner)) {
1938                         ret = -EINVAL;
1939                         break;
1940                 }
1941                 tun->owner = owner;
1942                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1943                           from_kuid(&init_user_ns, tun->owner));
1944                 break;
1945
1946         case TUNSETGROUP:
1947                 /* Set group of the device */
1948                 group = make_kgid(current_user_ns(), arg);
1949                 if (!gid_valid(group)) {
1950                         ret = -EINVAL;
1951                         break;
1952                 }
1953                 tun->group = group;
1954                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1955                           from_kgid(&init_user_ns, tun->group));
1956                 break;
1957
1958         case TUNSETLINK:
1959                 /* Only allow setting the type when the interface is down */
1960                 if (tun->dev->flags & IFF_UP) {
1961                         tun_debug(KERN_INFO, tun,
1962                                   "Linktype set failed because interface is up\n");
1963                         ret = -EBUSY;
1964                 } else {
1965                         tun->dev->type = (int) arg;
1966                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1967                                   tun->dev->type);
1968                         ret = 0;
1969                 }
1970                 break;
1971
1972 #ifdef TUN_DEBUG
1973         case TUNSETDEBUG:
1974                 tun->debug = arg;
1975                 break;
1976 #endif
1977         case TUNSETOFFLOAD:
1978                 ret = set_offload(tun, arg);
1979                 break;
1980
1981         case TUNSETTXFILTER:
1982                 /* Can be set only for TAPs */
1983                 ret = -EINVAL;
1984                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1985                         break;
1986                 ret = update_filter(&tun->txflt, (void __user *)arg);
1987                 break;
1988
1989         case SIOCGIFHWADDR:
1990                 /* Get hw address */
1991                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1992                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1993                 if (copy_to_user(argp, &ifr, ifreq_len))
1994                         ret = -EFAULT;
1995                 break;
1996
1997         case SIOCSIFHWADDR:
1998                 /* Set hw address */
1999                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2000                           ifr.ifr_hwaddr.sa_data);
2001
2002                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2003                 break;
2004
2005         case TUNGETSNDBUF:
2006                 sndbuf = tfile->socket.sk->sk_sndbuf;
2007                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2008                         ret = -EFAULT;
2009                 break;
2010
2011         case TUNSETSNDBUF:
2012                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2013                         ret = -EFAULT;
2014                         break;
2015                 }
2016
2017                 tun->sndbuf = sndbuf;
2018                 tun_set_sndbuf(tun);
2019                 break;
2020
2021         case TUNGETVNETHDRSZ:
2022                 vnet_hdr_sz = tun->vnet_hdr_sz;
2023                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2024                         ret = -EFAULT;
2025                 break;
2026
2027         case TUNSETVNETHDRSZ:
2028                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2029                         ret = -EFAULT;
2030                         break;
2031                 }
2032                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2033                         ret = -EINVAL;
2034                         break;
2035                 }
2036
2037                 tun->vnet_hdr_sz = vnet_hdr_sz;
2038                 break;
2039
2040         case TUNATTACHFILTER:
2041                 /* Can be set only for TAPs */
2042                 ret = -EINVAL;
2043                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2044                         break;
2045                 ret = -EFAULT;
2046                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2047                         break;
2048
2049                 ret = tun_attach_filter(tun);
2050                 break;
2051
2052         case TUNDETACHFILTER:
2053                 /* Can be set only for TAPs */
2054                 ret = -EINVAL;
2055                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2056                         break;
2057                 ret = 0;
2058                 tun_detach_filter(tun, tun->numqueues);
2059                 break;
2060
2061         default:
2062                 ret = -EINVAL;
2063                 break;
2064         }
2065
2066 unlock:
2067         rtnl_unlock();
2068         if (tun)
2069                 tun_put(tun);
2070         return ret;
2071 }
2072
2073 static long tun_chr_ioctl(struct file *file,
2074                           unsigned int cmd, unsigned long arg)
2075 {
2076         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2077 }
2078
2079 #ifdef CONFIG_COMPAT
2080 static long tun_chr_compat_ioctl(struct file *file,
2081                          unsigned int cmd, unsigned long arg)
2082 {
2083         switch (cmd) {
2084         case TUNSETIFF:
2085         case TUNGETIFF:
2086         case TUNSETTXFILTER:
2087         case TUNGETSNDBUF:
2088         case TUNSETSNDBUF:
2089         case SIOCGIFHWADDR:
2090         case SIOCSIFHWADDR:
2091                 arg = (unsigned long)compat_ptr(arg);
2092                 break;
2093         default:
2094                 arg = (compat_ulong_t)arg;
2095                 break;
2096         }
2097
2098         /*
2099          * compat_ifreq is shorter than ifreq, so we must not access beyond
2100          * the end of that structure. All fields that are used in this
2101          * driver are compatible though, we don't need to convert the
2102          * contents.
2103          */
2104         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2105 }
2106 #endif /* CONFIG_COMPAT */
2107
2108 static int tun_chr_fasync(int fd, struct file *file, int on)
2109 {
2110         struct tun_file *tfile = file->private_data;
2111         int ret;
2112
2113         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2114                 goto out;
2115
2116         if (on) {
2117                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2118                 if (ret)
2119                         goto out;
2120                 tfile->flags |= TUN_FASYNC;
2121         } else
2122                 tfile->flags &= ~TUN_FASYNC;
2123         ret = 0;
2124 out:
2125         return ret;
2126 }
2127
2128 static int tun_chr_open(struct inode *inode, struct file * file)
2129 {
2130         struct tun_file *tfile;
2131
2132         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2133
2134         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2135                                             &tun_proto);
2136         if (!tfile)
2137                 return -ENOMEM;
2138         rcu_assign_pointer(tfile->tun, NULL);
2139         tfile->net = get_net(current->nsproxy->net_ns);
2140         tfile->flags = 0;
2141
2142         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2143         init_waitqueue_head(&tfile->wq.wait);
2144
2145         tfile->socket.file = file;
2146         tfile->socket.ops = &tun_socket_ops;
2147
2148         sock_init_data(&tfile->socket, &tfile->sk);
2149         sk_change_net(&tfile->sk, tfile->net);
2150
2151         tfile->sk.sk_write_space = tun_sock_write_space;
2152         tfile->sk.sk_sndbuf = INT_MAX;
2153
2154         file->private_data = tfile;
2155         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2156         INIT_LIST_HEAD(&tfile->next);
2157
2158         return 0;
2159 }
2160
2161 static int tun_chr_close(struct inode *inode, struct file *file)
2162 {
2163         struct tun_file *tfile = file->private_data;
2164         struct net *net = tfile->net;
2165
2166         tun_detach(tfile, true);
2167         put_net(net);
2168
2169         return 0;
2170 }
2171
2172 static const struct file_operations tun_fops = {
2173         .owner  = THIS_MODULE,
2174         .llseek = no_llseek,
2175         .read  = do_sync_read,
2176         .aio_read  = tun_chr_aio_read,
2177         .write = do_sync_write,
2178         .aio_write = tun_chr_aio_write,
2179         .poll   = tun_chr_poll,
2180         .unlocked_ioctl = tun_chr_ioctl,
2181 #ifdef CONFIG_COMPAT
2182         .compat_ioctl = tun_chr_compat_ioctl,
2183 #endif
2184         .open   = tun_chr_open,
2185         .release = tun_chr_close,
2186         .fasync = tun_chr_fasync
2187 };
2188
2189 static struct miscdevice tun_miscdev = {
2190         .minor = TUN_MINOR,
2191         .name = "tun",
2192         .nodename = "net/tun",
2193         .fops = &tun_fops,
2194 };
2195
2196 /* ethtool interface */
2197
2198 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2199 {
2200         cmd->supported          = 0;
2201         cmd->advertising        = 0;
2202         ethtool_cmd_speed_set(cmd, SPEED_10);
2203         cmd->duplex             = DUPLEX_FULL;
2204         cmd->port               = PORT_TP;
2205         cmd->phy_address        = 0;
2206         cmd->transceiver        = XCVR_INTERNAL;
2207         cmd->autoneg            = AUTONEG_DISABLE;
2208         cmd->maxtxpkt           = 0;
2209         cmd->maxrxpkt           = 0;
2210         return 0;
2211 }
2212
2213 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2214 {
2215         struct tun_struct *tun = netdev_priv(dev);
2216
2217         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2218         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2219
2220         switch (tun->flags & TUN_TYPE_MASK) {
2221         case TUN_TUN_DEV:
2222                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2223                 break;
2224         case TUN_TAP_DEV:
2225                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2226                 break;
2227         }
2228 }
2229
2230 static u32 tun_get_msglevel(struct net_device *dev)
2231 {
2232 #ifdef TUN_DEBUG
2233         struct tun_struct *tun = netdev_priv(dev);
2234         return tun->debug;
2235 #else
2236         return -EOPNOTSUPP;
2237 #endif
2238 }
2239
2240 static void tun_set_msglevel(struct net_device *dev, u32 value)
2241 {
2242 #ifdef TUN_DEBUG
2243         struct tun_struct *tun = netdev_priv(dev);
2244         tun->debug = value;
2245 #endif
2246 }
2247
2248 static const struct ethtool_ops tun_ethtool_ops = {
2249         .get_settings   = tun_get_settings,
2250         .get_drvinfo    = tun_get_drvinfo,
2251         .get_msglevel   = tun_get_msglevel,
2252         .set_msglevel   = tun_set_msglevel,
2253         .get_link       = ethtool_op_get_link,
2254 };
2255
2256
2257 static int __init tun_init(void)
2258 {
2259         int ret = 0;
2260
2261         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2262         pr_info("%s\n", DRV_COPYRIGHT);
2263
2264         ret = rtnl_link_register(&tun_link_ops);
2265         if (ret) {
2266                 pr_err("Can't register link_ops\n");
2267                 goto err_linkops;
2268         }
2269
2270         ret = misc_register(&tun_miscdev);
2271         if (ret) {
2272                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2273                 goto err_misc;
2274         }
2275         return  0;
2276 err_misc:
2277         rtnl_link_unregister(&tun_link_ops);
2278 err_linkops:
2279         return ret;
2280 }
2281
2282 static void tun_cleanup(void)
2283 {
2284         misc_deregister(&tun_miscdev);
2285         rtnl_link_unregister(&tun_link_ops);
2286 }
2287
2288 /* Get an underlying socket object from tun file.  Returns error unless file is
2289  * attached to a device.  The returned object works like a packet socket, it
2290  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2291  * holding a reference to the file for as long as the socket is in use. */
2292 struct socket *tun_get_socket(struct file *file)
2293 {
2294         struct tun_file *tfile;
2295         if (file->f_op != &tun_fops)
2296                 return ERR_PTR(-EINVAL);
2297         tfile = file->private_data;
2298         if (!tfile)
2299                 return ERR_PTR(-EBADFD);
2300         return &tfile->socket;
2301 }
2302 EXPORT_SYMBOL_GPL(tun_get_socket);
2303
2304 module_init(tun_init);
2305 module_exit(tun_cleanup);
2306 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2307 MODULE_AUTHOR(DRV_COPYRIGHT);
2308 MODULE_LICENSE("GPL");
2309 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2310 MODULE_ALIAS("devname:net/tun");