drivers/net/tun.c

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