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

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