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 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, 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         struct net *net = dev_net(tun->dev);
 377
 378         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 379                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 380                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 381 }
 382
 383 static void tun_set_real_num_queues(struct tun_struct *tun)
 384 {
 385         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 386         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 387 }
 388
 389 static void __tun_detach(struct tun_file *tfile, bool clean)
 390 {
 391         struct tun_file *ntfile;
 392         struct tun_struct *tun;
 393         struct net_device *dev;
 394
 395         tun = rcu_dereference_protected(tfile->tun,
 396                                         lockdep_rtnl_is_held());
 397         if (tun) {
 398                 u16 index = tfile->queue_index;
 399                 BUG_ON(index >= tun->numqueues);
 400                 dev = tun->dev;
 401
 402                 rcu_assign_pointer(tun->tfiles[index],
 403                                    tun->tfiles[tun->numqueues - 1]);
 404                 rcu_assign_pointer(tfile->tun, NULL);
 405                 ntfile = rcu_dereference_protected(tun->tfiles[index],
 406                                                    lockdep_rtnl_is_held());
 407                 ntfile->queue_index = index;
 408
 409                 --tun->numqueues;
 410                 sock_put(&tfile->sk);
 411
 412                 synchronize_net();
 413                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 414                 /* Drop read queue */
 415                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 416                 tun_set_real_num_queues(tun);
 417
 418                 if (tun->numqueues == 0 && !(tun->flags & TUN_PERSIST))
 419                         if (dev->reg_state == NETREG_REGISTERED)
 420                                 unregister_netdevice(dev);
 421         }
 422
 423         if (clean) {
 424                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 425                                  &tfile->socket.flags));
 426                 sk_release_kernel(&tfile->sk);
 427         }
 428 }
 429
 430 static void tun_detach(struct tun_file *tfile, bool clean)
 431 {
 432         rtnl_lock();
 433         __tun_detach(tfile, clean);
 434         rtnl_unlock();
 435 }
 436
 437 static void tun_detach_all(struct net_device *dev)
 438 {
 439         struct tun_struct *tun = netdev_priv(dev);
 440         struct tun_file *tfile;
 441         int i, n = tun->numqueues;
 442
 443         for (i = 0; i < n; i++) {
 444                 tfile = rcu_dereference_protected(tun->tfiles[i],
 445                                                   lockdep_rtnl_is_held());
 446                 BUG_ON(!tfile);
 447                 wake_up_all(&tfile->wq.wait);
 448                 rcu_assign_pointer(tfile->tun, NULL);
 449                 --tun->numqueues;
 450         }
 451         BUG_ON(tun->numqueues != 0);
 452
 453         synchronize_net();
 454         for (i = 0; i < n; i++) {
 455                 tfile = rcu_dereference_protected(tun->tfiles[i],
 456                                                   lockdep_rtnl_is_held());
 457                 /* Drop read queue */
 458                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 459                 sock_put(&tfile->sk);
 460         }
 461 }
 462
 463 static int tun_attach(struct tun_struct *tun, struct file *file)
 464 {
 465         struct tun_file *tfile = file->private_data;
 466         int err;
 467
 468         err = -EINVAL;
 469         if (rcu_dereference_protected(tfile->tun, lockdep_rtnl_is_held()))
 470                 goto out;
 471
 472         err = -EBUSY;
 473         if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 474                 goto out;
 475
 476         err = -E2BIG;
 477         if (tun->numqueues == MAX_TAP_QUEUES)
 478                 goto out;
 479
 480         err = 0;
 481
 482         /* Re-attach the filter to presist device */
 483         if (tun->filter_attached == true) {
 484                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 485                 if (!err)
 486                         goto out;
 487         }
 488         tfile->queue_index = tun->numqueues;
 489         rcu_assign_pointer(tfile->tun, tun);
 490         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 491         sock_hold(&tfile->sk);
 492         tun->numqueues++;
 493
 494         tun_set_real_num_queues(tun);
 495
 496         /* device is allowed to go away first, so no need to hold extra
 497          * refcnt.
 498          */
 499
 500 out:
 501         return err;
 502 }
 503
 504 static struct tun_struct *__tun_get(struct tun_file *tfile)
 505 {
 506         struct tun_struct *tun;
 507
 508         rcu_read_lock();
 509         tun = rcu_dereference(tfile->tun);
 510         if (tun)
 511                 dev_hold(tun->dev);
 512         rcu_read_unlock();
 513
 514         return tun;
 515 }
 516
 517 static struct tun_struct *tun_get(struct file *file)
 518 {
 519         return __tun_get(file->private_data);
 520 }
 521
 522 static void tun_put(struct tun_struct *tun)
 523 {
 524         dev_put(tun->dev);
 525 }
 526
 527 /* TAP filtering */
 528 static void addr_hash_set(u32 *mask, const u8 *addr)
 529 {
 530         int n = ether_crc(ETH_ALEN, addr) >> 26;
 531         mask[n >> 5] |= (1 << (n & 31));
 532 }
 533
 534 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 535 {
 536         int n = ether_crc(ETH_ALEN, addr) >> 26;
 537         return mask[n >> 5] & (1 << (n & 31));
 538 }
 539
 540 static int update_filter(struct tap_filter *filter, void __user *arg)
 541 {
 542         struct { u8 u[ETH_ALEN]; } *addr;
 543         struct tun_filter uf;
 544         int err, alen, n, nexact;
 545
 546         if (copy_from_user(&uf, arg, sizeof(uf)))
 547                 return -EFAULT;
 548
 549         if (!uf.count) {
 550                 /* Disabled */
 551                 filter->count = 0;
 552                 return 0;
 553         }
 554
 555         alen = ETH_ALEN * uf.count;
 556         addr = kmalloc(alen, GFP_KERNEL);
 557         if (!addr)
 558                 return -ENOMEM;
 559
 560         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 561                 err = -EFAULT;
 562                 goto done;
 563         }
 564
 565         /* The filter is updated without holding any locks. Which is
 566          * perfectly safe. We disable it first and in the worst
 567          * case we'll accept a few undesired packets. */
 568         filter->count = 0;
 569         wmb();
 570
 571         /* Use first set of addresses as an exact filter */
 572         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 573                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 574
 575         nexact = n;
 576
 577         /* Remaining multicast addresses are hashed,
 578          * unicast will leave the filter disabled. */
 579         memset(filter->mask, 0, sizeof(filter->mask));
 580         for (; n < uf.count; n++) {
 581                 if (!is_multicast_ether_addr(addr[n].u)) {
 582                         err = 0; /* no filter */
 583                         goto done;
 584                 }
 585                 addr_hash_set(filter->mask, addr[n].u);
 586         }
 587
 588         /* For ALLMULTI just set the mask to all ones.
 589          * This overrides the mask populated above. */
 590         if ((uf.flags & TUN_FLT_ALLMULTI))
 591                 memset(filter->mask, ~0, sizeof(filter->mask));
 592
 593         /* Now enable the filter */
 594         wmb();
 595         filter->count = nexact;
 596
 597         /* Return the number of exact filters */
 598         err = nexact;
 599
 600 done:
 601         kfree(addr);
 602         return err;
 603 }
 604
 605 /* Returns: 0 - drop, !=0 - accept */
 606 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 607 {
 608         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 609          * at this point. */
 610         struct ethhdr *eh = (struct ethhdr *) skb->data;
 611         int i;
 612
 613         /* Exact match */
 614         for (i = 0; i < filter->count; i++)
 615                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 616                         return 1;
 617
 618         /* Inexact match (multicast only) */
 619         if (is_multicast_ether_addr(eh->h_dest))
 620                 return addr_hash_test(filter->mask, eh->h_dest);
 621
 622         return 0;
 623 }
 624
 625 /*
 626  * Checks whether the packet is accepted or not.
 627  * Returns: 0 - drop, !=0 - accept
 628  */
 629 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 630 {
 631         if (!filter->count)
 632                 return 1;
 633
 634         return run_filter(filter, skb);
 635 }
 636
 637 /* Network device part of the driver */
 638
 639 static const struct ethtool_ops tun_ethtool_ops;
 640
 641 /* Net device detach from fd. */
 642 static void tun_net_uninit(struct net_device *dev)
 643 {
 644         tun_detach_all(dev);
 645 }
 646
 647 /* Net device open. */
 648 static int tun_net_open(struct net_device *dev)
 649 {
 650         netif_tx_start_all_queues(dev);
 651         return 0;
 652 }
 653
 654 /* Net device close. */
 655 static int tun_net_close(struct net_device *dev)
 656 {
 657         netif_tx_stop_all_queues(dev);
 658         return 0;
 659 }
 660
 661 /* Net device start xmit */
 662 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 663 {
 664         struct tun_struct *tun = netdev_priv(dev);
 665         int txq = skb->queue_mapping;
 666         struct tun_file *tfile;
 667
 668         rcu_read_lock();
 669         tfile = rcu_dereference(tun->tfiles[txq]);
 670
 671         /* Drop packet if interface is not attached */
 672         if (txq >= tun->numqueues)
 673                 goto drop;
 674
 675         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 676
 677         BUG_ON(!tfile);
 678
 679         /* Drop if the filter does not like it.
 680          * This is a noop if the filter is disabled.
 681          * Filter can be enabled only for the TAP devices. */
 682         if (!check_filter(&tun->txflt, skb))
 683                 goto drop;
 684
 685         if (tfile->socket.sk->sk_filter &&
 686             sk_filter(tfile->socket.sk, skb))
 687                 goto drop;
 688
 689         /* Limit the number of packets queued by dividing txq length with the
 690          * number of queues.
 691          */
 692         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
 693                           >= dev->tx_queue_len / tun->numqueues){
 694                 if (!(tun->flags & TUN_ONE_QUEUE)) {
 695                         /* Normal queueing mode. */
 696                         /* Packet scheduler handles dropping of further packets. */
 697                         netif_stop_subqueue(dev, txq);
 698
 699                         /* We won't see all dropped packets individually, so overrun
 700                          * error is more appropriate. */
 701                         dev->stats.tx_fifo_errors++;
 702                 } else {
 703                         /* Single queue mode.
 704                          * Driver handles dropping of all packets itself. */
 705                         goto drop;
 706                 }
 707         }
 708
 709         /* Orphan the skb - required as we might hang on to it
 710          * for indefinite time. */
 711         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 712                 goto drop;
 713         skb_orphan(skb);
 714
 715         /* Enqueue packet */
 716         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 717
 718         /* Notify and wake up reader process */
 719         if (tfile->flags & TUN_FASYNC)
 720                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 721         wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
 722                                    POLLRDNORM | POLLRDBAND);
 723
 724         rcu_read_unlock();
 725         return NETDEV_TX_OK;
 726
 727 drop:
 728         dev->stats.tx_dropped++;
 729         skb_tx_error(skb);
 730         kfree_skb(skb);
 731         rcu_read_unlock();
 732         return NETDEV_TX_OK;
 733 }
 734
 735 static void tun_net_mclist(struct net_device *dev)
 736 {
 737         /*
 738          * This callback is supposed to deal with mc filter in
 739          * _rx_ path and has nothing to do with the _tx_ path.
 740          * In rx path we always accept everything userspace gives us.
 741          */
 742 }
 743
 744 #define MIN_MTU 68
 745 #define MAX_MTU 65535
 746
 747 static int
 748 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 749 {
 750         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 751                 return -EINVAL;
 752         dev->mtu = new_mtu;
 753         return 0;
 754 }
 755
 756 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 757         netdev_features_t features)
 758 {
 759         struct tun_struct *tun = netdev_priv(dev);
 760
 761         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 762 }
 763 #ifdef CONFIG_NET_POLL_CONTROLLER
 764 static void tun_poll_controller(struct net_device *dev)
 765 {
 766         /*
 767          * Tun only receives frames when:
 768          * 1) the char device endpoint gets data from user space
 769          * 2) the tun socket gets a sendmsg call from user space
 770          * Since both of those are syncronous operations, we are guaranteed
 771          * never to have pending data when we poll for it
 772          * so theres nothing to do here but return.
 773          * We need this though so netpoll recognizes us as an interface that
 774          * supports polling, which enables bridge devices in virt setups to
 775          * still use netconsole
 776          */
 777         return;
 778 }
 779 #endif
 780 static const struct net_device_ops tun_netdev_ops = {
 781         .ndo_uninit             = tun_net_uninit,
 782         .ndo_open               = tun_net_open,
 783         .ndo_stop               = tun_net_close,
 784         .ndo_start_xmit         = tun_net_xmit,
 785         .ndo_change_mtu         = tun_net_change_mtu,
 786         .ndo_fix_features       = tun_net_fix_features,
 787         .ndo_select_queue       = tun_select_queue,
 788 #ifdef CONFIG_NET_POLL_CONTROLLER
 789         .ndo_poll_controller    = tun_poll_controller,
 790 #endif
 791 };
 792
 793 static const struct net_device_ops tap_netdev_ops = {
 794         .ndo_uninit             = tun_net_uninit,
 795         .ndo_open               = tun_net_open,
 796         .ndo_stop               = tun_net_close,
 797         .ndo_start_xmit         = tun_net_xmit,
 798         .ndo_change_mtu         = tun_net_change_mtu,
 799         .ndo_fix_features       = tun_net_fix_features,
 800         .ndo_set_rx_mode        = tun_net_mclist,
 801         .ndo_set_mac_address    = eth_mac_addr,
 802         .ndo_validate_addr      = eth_validate_addr,
 803         .ndo_select_queue       = tun_select_queue,
 804 #ifdef CONFIG_NET_POLL_CONTROLLER
 805         .ndo_poll_controller    = tun_poll_controller,
 806 #endif
 807 };
 808
 809 static int tun_flow_init(struct tun_struct *tun)
 810 {
 811         int i;
 812
 813         tun->flow_cache = kmem_cache_create("tun_flow_cache",
 814                                             sizeof(struct tun_flow_entry), 0, 0,
 815                                             NULL);
 816         if (!tun->flow_cache)
 817                 return -ENOMEM;
 818
 819         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 820                 INIT_HLIST_HEAD(&tun->flows[i]);
 821
 822         tun->ageing_time = TUN_FLOW_EXPIRE;
 823         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 824         mod_timer(&tun->flow_gc_timer,
 825                   round_jiffies_up(jiffies + tun->ageing_time));
 826
 827         return 0;
 828 }
 829
 830 static void tun_flow_uninit(struct tun_struct *tun)
 831 {
 832         del_timer_sync(&tun->flow_gc_timer);
 833         tun_flow_flush(tun);
 834
 835         /* Wait for completion of call_rcu()'s */
 836         rcu_barrier();
 837         kmem_cache_destroy(tun->flow_cache);
 838 }
 839
 840 /* Initialize net device. */
 841 static void tun_net_init(struct net_device *dev)
 842 {
 843         struct tun_struct *tun = netdev_priv(dev);
 844
 845         switch (tun->flags & TUN_TYPE_MASK) {
 846         case TUN_TUN_DEV:
 847                 dev->netdev_ops = &tun_netdev_ops;
 848
 849                 /* Point-to-Point TUN Device */
 850                 dev->hard_header_len = 0;
 851                 dev->addr_len = 0;
 852                 dev->mtu = 1500;
 853
 854                 /* Zero header length */
 855                 dev->type = ARPHRD_NONE;
 856                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 857                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 858                 break;
 859
 860         case TUN_TAP_DEV:
 861                 dev->netdev_ops = &tap_netdev_ops;
 862                 /* Ethernet TAP Device */
 863                 ether_setup(dev);
 864                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 865
 866                 eth_hw_addr_random(dev);
 867
 868                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 869                 break;
 870         }
 871 }
 872
 873 /* Character device part */
 874
 875 /* Poll */
 876 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 877 {
 878         struct tun_file *tfile = file->private_data;
 879         struct tun_struct *tun = __tun_get(tfile);
 880         struct sock *sk;
 881         unsigned int mask = 0;
 882
 883         if (!tun)
 884                 return POLLERR;
 885
 886         sk = tfile->socket.sk;
 887
 888         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 889
 890         poll_wait(file, &tfile->wq.wait, wait);
 891
 892         if (!skb_queue_empty(&sk->sk_receive_queue))
 893                 mask |= POLLIN | POLLRDNORM;
 894
 895         if (sock_writeable(sk) ||
 896             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 897              sock_writeable(sk)))
 898                 mask |= POLLOUT | POLLWRNORM;
 899
 900         if (tun->dev->reg_state != NETREG_REGISTERED)
 901                 mask = POLLERR;
 902
 903         tun_put(tun);
 904         return mask;
 905 }
 906
 907 /* prepad is the amount to reserve at front.  len is length after that.
 908  * linear is a hint as to how much to copy (usually headers). */
 909 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 910                                      size_t prepad, size_t len,
 911                                      size_t linear, int noblock)
 912 {
 913         struct sock *sk = tfile->socket.sk;
 914         struct sk_buff *skb;
 915         int err;
 916
 917         /* Under a page?  Don't bother with paged skb. */
 918         if (prepad + len < PAGE_SIZE || !linear)
 919                 linear = len;
 920
 921         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 922                                    &err);
 923         if (!skb)
 924                 return ERR_PTR(err);
 925
 926         skb_reserve(skb, prepad);
 927         skb_put(skb, linear);
 928         skb->data_len = len - linear;
 929         skb->len += len - linear;
 930
 931         return skb;
 932 }
 933
 934 /* set skb frags from iovec, this can move to core network code for reuse */
 935 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 936                                   int offset, size_t count)
 937 {
 938         int len = iov_length(from, count) - offset;
 939         int copy = skb_headlen(skb);
 940         int size, offset1 = 0;
 941         int i = 0;
 942
 943         /* Skip over from offset */
 944         while (count && (offset >= from->iov_len)) {
 945                 offset -= from->iov_len;
 946                 ++from;
 947                 --count;
 948         }
 949
 950         /* copy up to skb headlen */
 951         while (count && (copy > 0)) {
 952                 size = min_t(unsigned int, copy, from->iov_len - offset);
 953                 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
 954                                    size))
 955                         return -EFAULT;
 956                 if (copy > size) {
 957                         ++from;
 958                         --count;
 959                         offset = 0;
 960                 } else
 961                         offset += size;
 962                 copy -= size;
 963                 offset1 += size;
 964         }
 965
 966         if (len == offset1)
 967                 return 0;
 968
 969         while (count--) {
 970                 struct page *page[MAX_SKB_FRAGS];
 971                 int num_pages;
 972                 unsigned long base;
 973                 unsigned long truesize;
 974
 975                 len = from->iov_len - offset;
 976                 if (!len) {
 977                         offset = 0;
 978                         ++from;
 979                         continue;
 980                 }
 981                 base = (unsigned long)from->iov_base + offset;
 982                 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
 983                 if (i + size > MAX_SKB_FRAGS)
 984                         return -EMSGSIZE;
 985                 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
 986                 if (num_pages != size) {
 987                         for (i = 0; i < num_pages; i++)
 988                                 put_page(page[i]);
 989                         return -EFAULT;
 990                 }
 991                 truesize = size * PAGE_SIZE;
 992                 skb->data_len += len;
 993                 skb->len += len;
 994                 skb->truesize += truesize;
 995                 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
 996                 while (len) {
 997                         int off = base & ~PAGE_MASK;
 998                         int size = min_t(int, len, PAGE_SIZE - off);
 999                         __skb_fill_page_desc(skb, i, page[i], off, size);
1000                         skb_shinfo(skb)->nr_frags++;
1001                         /* increase sk_wmem_alloc */
1002                         base += size;
1003                         len -= size;
1004                         i++;
1005                 }
1006                 offset = 0;
1007                 ++from;
1008         }
1009         return 0;
1010 }
1011
1012 /* Get packet from user space buffer */
1013 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1014                             void *msg_control, const struct iovec *iv,
1015                             size_t total_len, size_t count, int noblock)
1016 {
1017         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1018         struct sk_buff *skb;
1019         size_t len = total_len, align = NET_SKB_PAD;
1020         struct virtio_net_hdr gso = { 0 };
1021         int offset = 0;
1022         int copylen;
1023         bool zerocopy = false;
1024         int err;
1025
1026         if (!(tun->flags & TUN_NO_PI)) {
1027                 if ((len -= sizeof(pi)) > total_len)
1028                         return -EINVAL;
1029
1030                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1031                         return -EFAULT;
1032                 offset += sizeof(pi);
1033         }
1034
1035         if (tun->flags & TUN_VNET_HDR) {
1036                 if ((len -= tun->vnet_hdr_sz) > total_len)
1037                         return -EINVAL;
1038
1039                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1040                         return -EFAULT;
1041
1042                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1043                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1044                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1045
1046                 if (gso.hdr_len > len)
1047                         return -EINVAL;
1048                 offset += tun->vnet_hdr_sz;
1049         }
1050
1051         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1052                 align += NET_IP_ALIGN;
1053                 if (unlikely(len < ETH_HLEN ||
1054                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1055                         return -EINVAL;
1056         }
1057
1058         if (msg_control)
1059                 zerocopy = true;
1060
1061         if (zerocopy) {
1062                 /* Userspace may produce vectors with count greater than
1063                  * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1064                  * to let the rest of data to be fit in the frags.
1065                  */
1066                 if (count > MAX_SKB_FRAGS) {
1067                         copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1068                         if (copylen < offset)
1069                                 copylen = 0;
1070                         else
1071                                 copylen -= offset;
1072                 } else
1073                                 copylen = 0;
1074                 /* There are 256 bytes to be copied in skb, so there is enough
1075                  * room for skb expand head in case it is used.
1076                  * The rest of the buffer is mapped from userspace.
1077                  */
1078                 if (copylen < gso.hdr_len)
1079                         copylen = gso.hdr_len;
1080                 if (!copylen)
1081                         copylen = GOODCOPY_LEN;
1082         } else
1083                 copylen = len;
1084
1085         skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
1086         if (IS_ERR(skb)) {
1087                 if (PTR_ERR(skb) != -EAGAIN)
1088                         tun->dev->stats.rx_dropped++;
1089                 return PTR_ERR(skb);
1090         }
1091
1092         if (zerocopy)
1093                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1094         else
1095                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1096
1097         if (err) {
1098                 tun->dev->stats.rx_dropped++;
1099                 kfree_skb(skb);
1100                 return -EFAULT;
1101         }
1102
1103         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1104                 if (!skb_partial_csum_set(skb, gso.csum_start,
1105                                           gso.csum_offset)) {
1106                         tun->dev->stats.rx_frame_errors++;
1107                         kfree_skb(skb);
1108                         return -EINVAL;
1109                 }
1110         }
1111
1112         switch (tun->flags & TUN_TYPE_MASK) {
1113         case TUN_TUN_DEV:
1114                 if (tun->flags & TUN_NO_PI) {
1115                         switch (skb->data[0] & 0xf0) {
1116                         case 0x40:
1117                                 pi.proto = htons(ETH_P_IP);
1118                                 break;
1119                         case 0x60:
1120                                 pi.proto = htons(ETH_P_IPV6);
1121                                 break;
1122                         default:
1123                                 tun->dev->stats.rx_dropped++;
1124                                 kfree_skb(skb);
1125                                 return -EINVAL;
1126                         }
1127                 }
1128
1129                 skb_reset_mac_header(skb);
1130                 skb->protocol = pi.proto;
1131                 skb->dev = tun->dev;
1132                 break;
1133         case TUN_TAP_DEV:
1134                 skb->protocol = eth_type_trans(skb, tun->dev);
1135                 break;
1136         }
1137
1138         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1139                 pr_debug("GSO!\n");
1140                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1141                 case VIRTIO_NET_HDR_GSO_TCPV4:
1142                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1143                         break;
1144                 case VIRTIO_NET_HDR_GSO_TCPV6:
1145                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1146                         break;
1147                 case VIRTIO_NET_HDR_GSO_UDP:
1148                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1149                         break;
1150                 default:
1151                         tun->dev->stats.rx_frame_errors++;
1152                         kfree_skb(skb);
1153                         return -EINVAL;
1154                 }
1155
1156                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1157                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1158
1159                 skb_shinfo(skb)->gso_size = gso.gso_size;
1160                 if (skb_shinfo(skb)->gso_size == 0) {
1161                         tun->dev->stats.rx_frame_errors++;
1162                         kfree_skb(skb);
1163                         return -EINVAL;
1164                 }
1165
1166                 /* Header must be checked, and gso_segs computed. */
1167                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1168                 skb_shinfo(skb)->gso_segs = 0;
1169         }
1170
1171         /* copy skb_ubuf_info for callback when skb has no error */
1172         if (zerocopy) {
1173                 skb_shinfo(skb)->destructor_arg = msg_control;
1174                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1175         }
1176
1177         netif_rx_ni(skb);
1178
1179         tun->dev->stats.rx_packets++;
1180         tun->dev->stats.rx_bytes += len;
1181
1182         tun_flow_update(tun, skb, tfile->queue_index);
1183         return total_len;
1184 }
1185
1186 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1187                               unsigned long count, loff_t pos)
1188 {
1189         struct file *file = iocb->ki_filp;
1190         struct tun_struct *tun = tun_get(file);
1191         struct tun_file *tfile = file->private_data;
1192         ssize_t result;
1193
1194         if (!tun)
1195                 return -EBADFD;
1196
1197         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1198
1199         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1200                               count, file->f_flags & O_NONBLOCK);
1201
1202         tun_put(tun);
1203         return result;
1204 }
1205
1206 /* Put packet to the user space buffer */
1207 static ssize_t tun_put_user(struct tun_struct *tun,
1208                             struct tun_file *tfile,
1209                             struct sk_buff *skb,
1210                             const struct iovec *iv, int len)
1211 {
1212         struct tun_pi pi = { 0, skb->protocol };
1213         ssize_t total = 0;
1214
1215         if (!(tun->flags & TUN_NO_PI)) {
1216                 if ((len -= sizeof(pi)) < 0)
1217                         return -EINVAL;
1218
1219                 if (len < skb->len) {
1220                         /* Packet will be striped */
1221                         pi.flags |= TUN_PKT_STRIP;
1222                 }
1223
1224                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1225                         return -EFAULT;
1226                 total += sizeof(pi);
1227         }
1228
1229         if (tun->flags & TUN_VNET_HDR) {
1230                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1231                 if ((len -= tun->vnet_hdr_sz) < 0)
1232                         return -EINVAL;
1233
1234                 if (skb_is_gso(skb)) {
1235                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1236
1237                         /* This is a hint as to how much should be linear. */
1238                         gso.hdr_len = skb_headlen(skb);
1239                         gso.gso_size = sinfo->gso_size;
1240                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1241                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1242                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1243                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1244                         else if (sinfo->gso_type & SKB_GSO_UDP)
1245                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1246                         else {
1247                                 pr_err("unexpected GSO type: "
1248                                        "0x%x, gso_size %d, hdr_len %d\n",
1249                                        sinfo->gso_type, gso.gso_size,
1250                                        gso.hdr_len);
1251                                 print_hex_dump(KERN_ERR, "tun: ",
1252                                                DUMP_PREFIX_NONE,
1253                                                16, 1, skb->head,
1254                                                min((int)gso.hdr_len, 64), true);
1255                                 WARN_ON_ONCE(1);
1256                                 return -EINVAL;
1257                         }
1258                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1259                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1260                 } else
1261                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1262
1263                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1264                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1265                         gso.csum_start = skb_checksum_start_offset(skb);
1266                         gso.csum_offset = skb->csum_offset;
1267                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1268                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1269                 } /* else everything is zero */
1270
1271                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1272                                                sizeof(gso))))
1273                         return -EFAULT;
1274                 total += tun->vnet_hdr_sz;
1275         }
1276
1277         len = min_t(int, skb->len, len);
1278
1279         skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1280         total += skb->len;
1281
1282         tun->dev->stats.tx_packets++;
1283         tun->dev->stats.tx_bytes += len;
1284
1285         return total;
1286 }
1287
1288 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1289                            struct kiocb *iocb, const struct iovec *iv,
1290                            ssize_t len, int noblock)
1291 {
1292         DECLARE_WAITQUEUE(wait, current);
1293         struct sk_buff *skb;
1294         ssize_t ret = 0;
1295
1296         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1297
1298         if (unlikely(!noblock))
1299                 add_wait_queue(&tfile->wq.wait, &wait);
1300         while (len) {
1301                 current->state = TASK_INTERRUPTIBLE;
1302
1303                 /* Read frames from the queue */
1304                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1305                         if (noblock) {
1306                                 ret = -EAGAIN;
1307                                 break;
1308                         }
1309                         if (signal_pending(current)) {
1310                                 ret = -ERESTARTSYS;
1311                                 break;
1312                         }
1313                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1314                                 ret = -EIO;
1315                                 break;
1316                         }
1317
1318                         /* Nothing to read, let's sleep */
1319                         schedule();
1320                         continue;
1321                 }
1322                 netif_wake_subqueue(tun->dev, tfile->queue_index);
1323
1324                 ret = tun_put_user(tun, tfile, skb, iv, len);
1325                 kfree_skb(skb);
1326                 break;
1327         }
1328
1329         current->state = TASK_RUNNING;
1330         if (unlikely(!noblock))
1331                 remove_wait_queue(&tfile->wq.wait, &wait);
1332
1333         return ret;
1334 }
1335
1336 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1337                             unsigned long count, loff_t pos)
1338 {
1339         struct file *file = iocb->ki_filp;
1340         struct tun_file *tfile = file->private_data;
1341         struct tun_struct *tun = __tun_get(tfile);
1342         ssize_t len, ret;
1343
1344         if (!tun)
1345                 return -EBADFD;
1346         len = iov_length(iv, count);
1347         if (len < 0) {
1348                 ret = -EINVAL;
1349                 goto out;
1350         }
1351
1352         ret = tun_do_read(tun, tfile, iocb, iv, len,
1353                           file->f_flags & O_NONBLOCK);
1354         ret = min_t(ssize_t, ret, len);
1355 out:
1356         tun_put(tun);
1357         return ret;
1358 }
1359
1360 static void tun_free_netdev(struct net_device *dev)
1361 {
1362         struct tun_struct *tun = netdev_priv(dev);
1363
1364         tun_flow_uninit(tun);
1365         free_netdev(dev);
1366 }
1367
1368 static void tun_setup(struct net_device *dev)
1369 {
1370         struct tun_struct *tun = netdev_priv(dev);
1371
1372         tun->owner = INVALID_UID;
1373         tun->group = INVALID_GID;
1374
1375         dev->ethtool_ops = &tun_ethtool_ops;
1376         dev->destructor = tun_free_netdev;
1377 }
1378
1379 /* Trivial set of netlink ops to allow deleting tun or tap
1380  * device with netlink.
1381  */
1382 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1383 {
1384         return -EINVAL;
1385 }
1386
1387 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1388         .kind           = DRV_NAME,
1389         .priv_size      = sizeof(struct tun_struct),
1390         .setup          = tun_setup,
1391         .validate       = tun_validate,
1392 };
1393
1394 static void tun_sock_write_space(struct sock *sk)
1395 {
1396         struct tun_file *tfile;
1397         wait_queue_head_t *wqueue;
1398
1399         if (!sock_writeable(sk))
1400                 return;
1401
1402         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1403                 return;
1404
1405         wqueue = sk_sleep(sk);
1406         if (wqueue && waitqueue_active(wqueue))
1407                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1408                                                 POLLWRNORM | POLLWRBAND);
1409
1410         tfile = container_of(sk, struct tun_file, sk);
1411         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1412 }
1413
1414 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1415                        struct msghdr *m, size_t total_len)
1416 {
1417         int ret;
1418         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1419         struct tun_struct *tun = __tun_get(tfile);
1420
1421         if (!tun)
1422                 return -EBADFD;
1423         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1424                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1425         tun_put(tun);
1426         return ret;
1427 }
1428
1429
1430 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1431                        struct msghdr *m, size_t total_len,
1432                        int flags)
1433 {
1434         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1435         struct tun_struct *tun = __tun_get(tfile);
1436         int ret;
1437
1438         if (!tun)
1439                 return -EBADFD;
1440
1441         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1442                 return -EINVAL;
1443         ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1444                           flags & MSG_DONTWAIT);
1445         if (ret > total_len) {
1446                 m->msg_flags |= MSG_TRUNC;
1447                 ret = flags & MSG_TRUNC ? ret : total_len;
1448         }
1449         tun_put(tun);
1450         return ret;
1451 }
1452
1453 static int tun_release(struct socket *sock)
1454 {
1455         if (sock->sk)
1456                 sock_put(sock->sk);
1457         return 0;
1458 }
1459
1460 /* Ops structure to mimic raw sockets with tun */
1461 static const struct proto_ops tun_socket_ops = {
1462         .sendmsg = tun_sendmsg,
1463         .recvmsg = tun_recvmsg,
1464         .release = tun_release,
1465 };
1466
1467 static struct proto tun_proto = {
1468         .name           = "tun",
1469         .owner          = THIS_MODULE,
1470         .obj_size       = sizeof(struct tun_file),
1471 };
1472
1473 static int tun_flags(struct tun_struct *tun)
1474 {
1475         int flags = 0;
1476
1477         if (tun->flags & TUN_TUN_DEV)
1478                 flags |= IFF_TUN;
1479         else
1480                 flags |= IFF_TAP;
1481
1482         if (tun->flags & TUN_NO_PI)
1483                 flags |= IFF_NO_PI;
1484
1485         if (tun->flags & TUN_ONE_QUEUE)
1486                 flags |= IFF_ONE_QUEUE;
1487
1488         if (tun->flags & TUN_VNET_HDR)
1489                 flags |= IFF_VNET_HDR;
1490
1491         if (tun->flags & TUN_TAP_MQ)
1492                 flags |= IFF_MULTI_QUEUE;
1493
1494         return flags;
1495 }
1496
1497 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1498                               char *buf)
1499 {
1500         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1501         return sprintf(buf, "0x%x\n", tun_flags(tun));
1502 }
1503
1504 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1505                               char *buf)
1506 {
1507         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1508         return uid_valid(tun->owner)?
1509                 sprintf(buf, "%u\n",
1510                         from_kuid_munged(current_user_ns(), tun->owner)):
1511                 sprintf(buf, "-1\n");
1512 }
1513
1514 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1515                               char *buf)
1516 {
1517         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1518         return gid_valid(tun->group) ?
1519                 sprintf(buf, "%u\n",
1520                         from_kgid_munged(current_user_ns(), tun->group)):
1521                 sprintf(buf, "-1\n");
1522 }
1523
1524 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1525 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1526 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1527
1528 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1529 {
1530         struct tun_struct *tun;
1531         struct tun_file *tfile = file->private_data;
1532         struct net_device *dev;
1533         int err;
1534
1535         dev = __dev_get_by_name(net, ifr->ifr_name);
1536         if (dev) {
1537                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1538                         return -EBUSY;
1539                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1540                         tun = netdev_priv(dev);
1541                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1542                         tun = netdev_priv(dev);
1543                 else
1544                         return -EINVAL;
1545
1546                 if (tun_not_capable(tun))
1547                         return -EPERM;
1548                 err = security_tun_dev_attach(tfile->socket.sk);
1549                 if (err < 0)
1550                         return err;
1551
1552                 err = tun_attach(tun, file);
1553                 if (err < 0)
1554                         return err;
1555         }
1556         else {
1557                 char *name;
1558                 unsigned long flags = 0;
1559
1560                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1561                         return -EPERM;
1562                 err = security_tun_dev_create();
1563                 if (err < 0)
1564                         return err;
1565
1566                 /* Set dev type */
1567                 if (ifr->ifr_flags & IFF_TUN) {
1568                         /* TUN device */
1569                         flags |= TUN_TUN_DEV;
1570                         name = "tun%d";
1571                 } else if (ifr->ifr_flags & IFF_TAP) {
1572                         /* TAP device */
1573                         flags |= TUN_TAP_DEV;
1574                         name = "tap%d";
1575                 } else
1576                         return -EINVAL;
1577
1578                 if (*ifr->ifr_name)
1579                         name = ifr->ifr_name;
1580
1581                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1582                                        tun_setup,
1583                                        MAX_TAP_QUEUES, MAX_TAP_QUEUES);
1584                 if (!dev)
1585                         return -ENOMEM;
1586
1587                 dev_net_set(dev, net);
1588                 dev->rtnl_link_ops = &tun_link_ops;
1589
1590                 tun = netdev_priv(dev);
1591                 tun->dev = dev;
1592                 tun->flags = flags;
1593                 tun->txflt.count = 0;
1594                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1595
1596                 tun->filter_attached = false;
1597                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1598
1599                 spin_lock_init(&tun->lock);
1600
1601                 security_tun_dev_post_create(&tfile->sk);
1602
1603                 tun_net_init(dev);
1604
1605                 if (tun_flow_init(tun))
1606                         goto err_free_dev;
1607
1608                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1609                         TUN_USER_FEATURES;
1610                 dev->features = dev->hw_features;
1611
1612                 err = tun_attach(tun, file);
1613                 if (err < 0)
1614                         goto err_free_dev;
1615
1616                 err = register_netdevice(tun->dev);
1617                 if (err < 0)
1618                         goto err_free_dev;
1619
1620                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1621                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1622                     device_create_file(&tun->dev->dev, &dev_attr_group))
1623                         pr_err("Failed to create tun sysfs files\n");
1624
1625                 netif_carrier_on(tun->dev);
1626         }
1627
1628         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1629
1630         if (ifr->ifr_flags & IFF_NO_PI)
1631                 tun->flags |= TUN_NO_PI;
1632         else
1633                 tun->flags &= ~TUN_NO_PI;
1634
1635         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1636                 tun->flags |= TUN_ONE_QUEUE;
1637         else
1638                 tun->flags &= ~TUN_ONE_QUEUE;
1639
1640         if (ifr->ifr_flags & IFF_VNET_HDR)
1641                 tun->flags |= TUN_VNET_HDR;
1642         else
1643                 tun->flags &= ~TUN_VNET_HDR;
1644
1645         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1646                 tun->flags |= TUN_TAP_MQ;
1647         else
1648                 tun->flags &= ~TUN_TAP_MQ;
1649
1650         /* Make sure persistent devices do not get stuck in
1651          * xoff state.
1652          */
1653         if (netif_running(tun->dev))
1654                 netif_tx_wake_all_queues(tun->dev);
1655
1656         strcpy(ifr->ifr_name, tun->dev->name);
1657         return 0;
1658
1659  err_free_dev:
1660         free_netdev(dev);
1661         return err;
1662 }
1663
1664 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1665                        struct ifreq *ifr)
1666 {
1667         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1668
1669         strcpy(ifr->ifr_name, tun->dev->name);
1670
1671         ifr->ifr_flags = tun_flags(tun);
1672
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                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1849
1850                 if (copy_to_user(argp, &ifr, ifreq_len))
1851                         ret = -EFAULT;
1852                 break;
1853
1854         case TUNSETNOCSUM:
1855                 /* Disable/Enable checksum */
1856
1857                 /* [unimplemented] */
1858                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1859                           arg ? "disabled" : "enabled");
1860                 break;
1861
1862         case TUNSETPERSIST:
1863                 /* Disable/Enable persist mode. Keep an extra reference to the
1864                  * module to prevent the module being unprobed.
1865                  */
1866                 if (arg) {
1867                         tun->flags |= TUN_PERSIST;
1868                         __module_get(THIS_MODULE);
1869                 } else {
1870                         tun->flags &= ~TUN_PERSIST;
1871                         module_put(THIS_MODULE);
1872                 }
1873
1874                 tun_debug(KERN_INFO, tun, "persist %s\n",
1875                           arg ? "enabled" : "disabled");
1876                 break;
1877
1878         case TUNSETOWNER:
1879                 /* Set owner of the device */
1880                 owner = make_kuid(current_user_ns(), arg);
1881                 if (!uid_valid(owner)) {
1882                         ret = -EINVAL;
1883                         break;
1884                 }
1885                 tun->owner = owner;
1886                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1887                           from_kuid(&init_user_ns, tun->owner));
1888                 break;
1889
1890         case TUNSETGROUP:
1891                 /* Set group of the device */
1892                 group = make_kgid(current_user_ns(), arg);
1893                 if (!gid_valid(group)) {
1894                         ret = -EINVAL;
1895                         break;
1896                 }
1897                 tun->group = group;
1898                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1899                           from_kgid(&init_user_ns, tun->group));
1900                 break;
1901
1902         case TUNSETLINK:
1903                 /* Only allow setting the type when the interface is down */
1904                 if (tun->dev->flags & IFF_UP) {
1905                         tun_debug(KERN_INFO, tun,
1906                                   "Linktype set failed because interface is up\n");
1907                         ret = -EBUSY;
1908                 } else {
1909                         tun->dev->type = (int) arg;
1910                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1911                                   tun->dev->type);
1912                         ret = 0;
1913                 }
1914                 break;
1915
1916 #ifdef TUN_DEBUG
1917         case TUNSETDEBUG:
1918                 tun->debug = arg;
1919                 break;
1920 #endif
1921         case TUNSETOFFLOAD:
1922                 ret = set_offload(tun, arg);
1923                 break;
1924
1925         case TUNSETTXFILTER:
1926                 /* Can be set only for TAPs */
1927                 ret = -EINVAL;
1928                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1929                         break;
1930                 ret = update_filter(&tun->txflt, (void __user *)arg);
1931                 break;
1932
1933         case SIOCGIFHWADDR:
1934                 /* Get hw address */
1935                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1936                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1937                 if (copy_to_user(argp, &ifr, ifreq_len))
1938                         ret = -EFAULT;
1939                 break;
1940
1941         case SIOCSIFHWADDR:
1942                 /* Set hw address */
1943                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1944                           ifr.ifr_hwaddr.sa_data);
1945
1946                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1947                 break;
1948
1949         case TUNGETSNDBUF:
1950                 sndbuf = tfile->socket.sk->sk_sndbuf;
1951                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1952                         ret = -EFAULT;
1953                 break;
1954
1955         case TUNSETSNDBUF:
1956                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1957                         ret = -EFAULT;
1958                         break;
1959                 }
1960
1961                 tun->sndbuf = sndbuf;
1962                 tun_set_sndbuf(tun);
1963                 break;
1964
1965         case TUNGETVNETHDRSZ:
1966                 vnet_hdr_sz = tun->vnet_hdr_sz;
1967                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1968                         ret = -EFAULT;
1969                 break;
1970
1971         case TUNSETVNETHDRSZ:
1972                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1973                         ret = -EFAULT;
1974                         break;
1975                 }
1976                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1977                         ret = -EINVAL;
1978                         break;
1979                 }
1980
1981                 tun->vnet_hdr_sz = vnet_hdr_sz;
1982                 break;
1983
1984         case TUNATTACHFILTER:
1985                 /* Can be set only for TAPs */
1986                 ret = -EINVAL;
1987                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1988                         break;
1989                 ret = -EFAULT;
1990                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
1991                         break;
1992
1993                 ret = tun_attach_filter(tun);
1994                 break;
1995
1996         case TUNDETACHFILTER:
1997                 /* Can be set only for TAPs */
1998                 ret = -EINVAL;
1999                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2000                         break;
2001                 ret = 0;
2002                 tun_detach_filter(tun, tun->numqueues);
2003                 break;
2004
2005         default:
2006                 ret = -EINVAL;
2007                 break;
2008         }
2009
2010 unlock:
2011         rtnl_unlock();
2012         if (tun)
2013                 tun_put(tun);
2014         return ret;
2015 }
2016
2017 static long tun_chr_ioctl(struct file *file,
2018                           unsigned int cmd, unsigned long arg)
2019 {
2020         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2021 }
2022
2023 #ifdef CONFIG_COMPAT
2024 static long tun_chr_compat_ioctl(struct file *file,
2025                          unsigned int cmd, unsigned long arg)
2026 {
2027         switch (cmd) {
2028         case TUNSETIFF:
2029         case TUNGETIFF:
2030         case TUNSETTXFILTER:
2031         case TUNGETSNDBUF:
2032         case TUNSETSNDBUF:
2033         case SIOCGIFHWADDR:
2034         case SIOCSIFHWADDR:
2035                 arg = (unsigned long)compat_ptr(arg);
2036                 break;
2037         default:
2038                 arg = (compat_ulong_t)arg;
2039                 break;
2040         }
2041
2042         /*
2043          * compat_ifreq is shorter than ifreq, so we must not access beyond
2044          * the end of that structure. All fields that are used in this
2045          * driver are compatible though, we don't need to convert the
2046          * contents.
2047          */
2048         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2049 }
2050 #endif /* CONFIG_COMPAT */
2051
2052 static int tun_chr_fasync(int fd, struct file *file, int on)
2053 {
2054         struct tun_file *tfile = file->private_data;
2055         int ret;
2056
2057         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2058                 goto out;
2059
2060         if (on) {
2061                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2062                 if (ret)
2063                         goto out;
2064                 tfile->flags |= TUN_FASYNC;
2065         } else
2066                 tfile->flags &= ~TUN_FASYNC;
2067         ret = 0;
2068 out:
2069         return ret;
2070 }
2071
2072 static int tun_chr_open(struct inode *inode, struct file * file)
2073 {
2074         struct tun_file *tfile;
2075
2076         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2077
2078         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2079                                             &tun_proto);
2080         if (!tfile)
2081                 return -ENOMEM;
2082         rcu_assign_pointer(tfile->tun, NULL);
2083         tfile->net = get_net(current->nsproxy->net_ns);
2084         tfile->flags = 0;
2085
2086         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2087         init_waitqueue_head(&tfile->wq.wait);
2088
2089         tfile->socket.file = file;
2090         tfile->socket.ops = &tun_socket_ops;
2091
2092         sock_init_data(&tfile->socket, &tfile->sk);
2093         sk_change_net(&tfile->sk, tfile->net);
2094
2095         tfile->sk.sk_write_space = tun_sock_write_space;
2096         tfile->sk.sk_sndbuf = INT_MAX;
2097
2098         file->private_data = tfile;
2099         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2100
2101         return 0;
2102 }
2103
2104 static int tun_chr_close(struct inode *inode, struct file *file)
2105 {
2106         struct tun_file *tfile = file->private_data;
2107         struct net *net = tfile->net;
2108
2109         tun_detach(tfile, true);
2110         put_net(net);
2111
2112         return 0;
2113 }
2114
2115 static const struct file_operations tun_fops = {
2116         .owner  = THIS_MODULE,
2117         .llseek = no_llseek,
2118         .read  = do_sync_read,
2119         .aio_read  = tun_chr_aio_read,
2120         .write = do_sync_write,
2121         .aio_write = tun_chr_aio_write,
2122         .poll   = tun_chr_poll,
2123         .unlocked_ioctl = tun_chr_ioctl,
2124 #ifdef CONFIG_COMPAT
2125         .compat_ioctl = tun_chr_compat_ioctl,
2126 #endif
2127         .open   = tun_chr_open,
2128         .release = tun_chr_close,
2129         .fasync = tun_chr_fasync
2130 };
2131
2132 static struct miscdevice tun_miscdev = {
2133         .minor = TUN_MINOR,
2134         .name = "tun",
2135         .nodename = "net/tun",
2136         .fops = &tun_fops,
2137 };
2138
2139 /* ethtool interface */
2140
2141 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2142 {
2143         cmd->supported          = 0;
2144         cmd->advertising        = 0;
2145         ethtool_cmd_speed_set(cmd, SPEED_10);
2146         cmd->duplex             = DUPLEX_FULL;
2147         cmd->port               = PORT_TP;
2148         cmd->phy_address        = 0;
2149         cmd->transceiver        = XCVR_INTERNAL;
2150         cmd->autoneg            = AUTONEG_DISABLE;
2151         cmd->maxtxpkt           = 0;
2152         cmd->maxrxpkt           = 0;
2153         return 0;
2154 }
2155
2156 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2157 {
2158         struct tun_struct *tun = netdev_priv(dev);
2159
2160         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2161         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2162
2163         switch (tun->flags & TUN_TYPE_MASK) {
2164         case TUN_TUN_DEV:
2165                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2166                 break;
2167         case TUN_TAP_DEV:
2168                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2169                 break;
2170         }
2171 }
2172
2173 static u32 tun_get_msglevel(struct net_device *dev)
2174 {
2175 #ifdef TUN_DEBUG
2176         struct tun_struct *tun = netdev_priv(dev);
2177         return tun->debug;
2178 #else
2179         return -EOPNOTSUPP;
2180 #endif
2181 }
2182
2183 static void tun_set_msglevel(struct net_device *dev, u32 value)
2184 {
2185 #ifdef TUN_DEBUG
2186         struct tun_struct *tun = netdev_priv(dev);
2187         tun->debug = value;
2188 #endif
2189 }
2190
2191 static const struct ethtool_ops tun_ethtool_ops = {
2192         .get_settings   = tun_get_settings,
2193         .get_drvinfo    = tun_get_drvinfo,
2194         .get_msglevel   = tun_get_msglevel,
2195         .set_msglevel   = tun_set_msglevel,
2196         .get_link       = ethtool_op_get_link,
2197 };
2198
2199
2200 static int __init tun_init(void)
2201 {
2202         int ret = 0;
2203
2204         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2205         pr_info("%s\n", DRV_COPYRIGHT);
2206
2207         ret = rtnl_link_register(&tun_link_ops);
2208         if (ret) {
2209                 pr_err("Can't register link_ops\n");
2210                 goto err_linkops;
2211         }
2212
2213         ret = misc_register(&tun_miscdev);
2214         if (ret) {
2215                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2216                 goto err_misc;
2217         }
2218         return  0;
2219 err_misc:
2220         rtnl_link_unregister(&tun_link_ops);
2221 err_linkops:
2222         return ret;
2223 }
2224
2225 static void tun_cleanup(void)
2226 {
2227         misc_deregister(&tun_miscdev);
2228         rtnl_link_unregister(&tun_link_ops);
2229 }
2230
2231 /* Get an underlying socket object from tun file.  Returns error unless file is
2232  * attached to a device.  The returned object works like a packet socket, it
2233  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2234  * holding a reference to the file for as long as the socket is in use. */
2235 struct socket *tun_get_socket(struct file *file)
2236 {
2237         struct tun_file *tfile;
2238         if (file->f_op != &tun_fops)
2239                 return ERR_PTR(-EINVAL);
2240         tfile = file->private_data;
2241         if (!tfile)
2242                 return ERR_PTR(-EBADFD);
2243         return &tfile->socket;
2244 }
2245 EXPORT_SYMBOL_GPL(tun_get_socket);
2246
2247 module_init(tun_init);
2248 module_exit(tun_cleanup);
2249 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2250 MODULE_AUTHOR(DRV_COPYRIGHT);
2251 MODULE_LICENSE("GPL");
2252 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2253 MODULE_ALIAS("devname:net/tun");