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

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/major.h>
  48 #include <linux/slab.h>
  49 #include <linux/poll.h>
  50 #include <linux/fcntl.h>
  51 #include <linux/init.h>
  52 #include <linux/skbuff.h>
  53 #include <linux/netdevice.h>
  54 #include <linux/etherdevice.h>
  55 #include <linux/miscdevice.h>
  56 #include <linux/ethtool.h>
  57 #include <linux/rtnetlink.h>
  58 #include <linux/compat.h>
  59 #include <linux/if.h>
  60 #include <linux/if_arp.h>
  61 #include <linux/if_ether.h>
  62 #include <linux/if_tun.h>
  63 #include <linux/crc32.h>
  64 #include <linux/nsproxy.h>
  65 #include <linux/virtio_net.h>
  66 #include <linux/rcupdate.h>
  67 #include <net/net_namespace.h>
  68 #include <net/netns/generic.h>
  69 #include <net/rtnetlink.h>
  70 #include <net/sock.h>
  71
  72 #include <asm/uaccess.h>
  73
  74 /* Uncomment to enable debugging */
  75 /* #define TUN_DEBUG 1 */
  76
  77 #ifdef TUN_DEBUG
  78 static int debug;
  79
  80 #define tun_debug(level, tun, fmt, args...)                     \
  81 do {                                                            \
  82         if (tun->debug)                                         \
  83                 netdev_printk(level, tun->dev, fmt, ##args);    \
  84 } while (0)
  85 #define DBG1(level, fmt, args...)                               \
  86 do {                                                            \
  87         if (debug == 2)                                         \
  88                 printk(level fmt, ##args);                      \
  89 } while (0)
  90 #else
  91 #define tun_debug(level, tun, fmt, args...)                     \
  92 do {                                                            \
  93         if (0)                                                  \
  94                 netdev_printk(level, tun->dev, fmt, ##args);    \
  95 } while (0)
  96 #define DBG1(level, fmt, args...)                               \
  97 do {                                                            \
  98         if (0)                                                  \
  99                 printk(level fmt, ##args);                      \
 100 } while (0)
 101 #endif
 102
 103 #define GOODCOPY_LEN 128
 104
 105 #define FLT_EXACT_COUNT 8
 106 struct tap_filter {
 107         unsigned int    count;    /* Number of addrs. Zero means disabled */
 108         u32             mask[2];  /* Mask of the hashed addrs */
 109         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 110 };
 111
 112 /* 1024 is probably a high enough limit: modern hypervisors seem to support on
 113  * the order of 100-200 CPUs so this leaves us some breathing space if we want
 114  * to match a queue per guest CPU.
 115  */
 116 #define MAX_TAP_QUEUES 1024
 117
 118 #define TUN_FLOW_EXPIRE (3 * HZ)
 119
 120 /* A tun_file connects an open character device to a tuntap netdevice. It
 121  * also contains all socket related strctures (except sock_fprog and tap_filter)
 122  * to serve as one transmit queue for tuntap device. The sock_fprog and
 123  * tap_filter were kept in tun_struct since they were used for filtering for the
 124  * netdevice not for a specific queue (at least I didn't see the reqirement for
 125  * this).
 126  *
 127  * RCU usage:
 128  * The tun_file and tun_struct are loosely coupled, the pointer from on to the
 129  * other can only be read while rcu_read_lock or rtnl_lock is held.
 130  */
 131 struct tun_file {
 132         struct sock sk;
 133         struct socket socket;
 134         struct socket_wq wq;
 135         struct tun_struct __rcu *tun;
 136         struct net *net;
 137         struct fasync_struct *fasync;
 138         /* only used for fasnyc */
 139         unsigned int flags;
 140         u16 queue_index;
 141 };
 142
 143 struct tun_flow_entry {
 144         struct hlist_node hash_link;
 145         struct rcu_head rcu;
 146         struct tun_struct *tun;
 147
 148         u32 rxhash;
 149         int queue_index;
 150         unsigned long updated;
 151 };
 152
 153 #define TUN_NUM_FLOW_ENTRIES 1024
 154
 155 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 156  * device, socket fileter, sndbuf and vnet header size were restore when the
 157  * file were attached to a persist device.
 158  */
 159 struct tun_struct {
 160         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 161         unsigned int            numqueues;
 162         unsigned int            flags;
 163         kuid_t                  owner;
 164         kgid_t                  group;
 165
 166         struct net_device       *dev;
 167         netdev_features_t       set_features;
 168 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 169                           NETIF_F_TSO6|NETIF_F_UFO)
 170
 171         int                     vnet_hdr_sz;
 172         int                     sndbuf;
 173         struct tap_filter       txflt;
 174         struct sock_fprog       fprog;
 175         /* protected by rtnl lock */
 176         bool                    filter_attached;
 177 #ifdef TUN_DEBUG
 178         int debug;
 179 #endif
 180         spinlock_t lock;
 181         struct kmem_cache *flow_cache;
 182         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 183         struct timer_list flow_gc_timer;
 184         unsigned long ageing_time;
 185 };
 186
 187 static inline u32 tun_hashfn(u32 rxhash)
 188 {
 189         return rxhash & 0x3ff;
 190 }
 191
 192 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 193 {
 194         struct tun_flow_entry *e;
 195         struct hlist_node *n;
 196
 197         hlist_for_each_entry_rcu(e, n, head, hash_link) {
 198                 if (e->rxhash == rxhash)
 199                         return e;
 200         }
 201         return NULL;
 202 }
 203
 204 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 205                                               struct hlist_head *head,
 206                                               u32 rxhash, u16 queue_index)
 207 {
 208         struct tun_flow_entry *e = kmem_cache_alloc(tun->flow_cache,
 209                                                     GFP_ATOMIC);
 210         if (e) {
 211                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 212                           rxhash, queue_index);
 213                 e->updated = jiffies;
 214                 e->rxhash = rxhash;
 215                 e->queue_index = queue_index;
 216                 e->tun = tun;
 217                 hlist_add_head_rcu(&e->hash_link, head);
 218         }
 219         return e;
 220 }
 221
 222 static void tun_flow_free(struct rcu_head *head)
 223 {
 224         struct tun_flow_entry *e
 225                 = container_of(head, struct tun_flow_entry, rcu);
 226         kmem_cache_free(e->tun->flow_cache, e);
 227 }
 228
 229 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 230 {
 231         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 232                   e->rxhash, e->queue_index);
 233         hlist_del_rcu(&e->hash_link);
 234         call_rcu(&e->rcu, tun_flow_free);
 235 }
 236
 237 static void tun_flow_flush(struct tun_struct *tun)
 238 {
 239         int i;
 240
 241         spin_lock_bh(&tun->lock);
 242         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 243                 struct tun_flow_entry *e;
 244                 struct hlist_node *h, *n;
 245
 246                 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link)
 247                         tun_flow_delete(tun, e);
 248         }
 249         spin_unlock_bh(&tun->lock);
 250 }
 251
 252 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 253 {
 254         int i;
 255
 256         spin_lock_bh(&tun->lock);
 257         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 258                 struct tun_flow_entry *e;
 259                 struct hlist_node *h, *n;
 260
 261                 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
 262                         if (e->queue_index == queue_index)
 263                                 tun_flow_delete(tun, e);
 264                 }
 265         }
 266         spin_unlock_bh(&tun->lock);
 267 }
 268
 269 static void tun_flow_cleanup(unsigned long data)
 270 {
 271         struct tun_struct *tun = (struct tun_struct *)data;
 272         unsigned long delay = tun->ageing_time;
 273         unsigned long next_timer = jiffies + delay;
 274         unsigned long count = 0;
 275         int i;
 276
 277         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 278
 279         spin_lock_bh(&tun->lock);
 280         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 281                 struct tun_flow_entry *e;
 282                 struct hlist_node *h, *n;
 283
 284                 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
 285                         unsigned long this_timer;
 286                         count++;
 287                         this_timer = e->updated + delay;
 288                         if (time_before_eq(this_timer, jiffies))
 289                                 tun_flow_delete(tun, e);
 290                         else if (time_before(this_timer, next_timer))
 291                                 next_timer = this_timer;
 292                 }
 293         }
 294
 295         if (count)
 296                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 297         spin_unlock_bh(&tun->lock);
 298 }
 299
 300 static void tun_flow_update(struct tun_struct *tun, struct sk_buff *skb,
 301                             u16 queue_index)
 302 {
 303         struct hlist_head *head;
 304         struct tun_flow_entry *e;
 305         unsigned long delay = tun->ageing_time;
 306         u32 rxhash = skb_get_rxhash(skb);
 307
 308         if (!rxhash)
 309                 return;
 310         else
 311                 head = &tun->flows[tun_hashfn(rxhash)];
 312
 313         rcu_read_lock();
 314
 315         if (tun->numqueues == 1)
 316                 goto unlock;
 317
 318         e = tun_flow_find(head, rxhash);
 319         if (likely(e)) {
 320                 /* TODO: keep queueing to old queue until it's empty? */
 321                 e->queue_index = queue_index;
 322                 e->updated = jiffies;
 323         } else {
 324                 spin_lock_bh(&tun->lock);
 325                 if (!tun_flow_find(head, rxhash))
 326                         tun_flow_create(tun, head, rxhash, queue_index);
 327
 328                 if (!timer_pending(&tun->flow_gc_timer))
 329                         mod_timer(&tun->flow_gc_timer,
 330                                   round_jiffies_up(jiffies + delay));
 331                 spin_unlock_bh(&tun->lock);
 332         }
 333
 334 unlock:
 335         rcu_read_unlock();
 336 }
 337
 338 /* We try to identify a flow through its rxhash first. The reason that
 339  * we do not check rxq no. is becuase some cards(e.g 82599), chooses
 340  * the rxq based on the txq where the last packet of the flow comes. As
 341  * the userspace application move between processors, we may get a
 342  * different rxq no. here. If we could not get rxhash, then we would
 343  * hope the rxq no. may help here.
 344  */
 345 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
 346 {
 347         struct tun_struct *tun = netdev_priv(dev);
 348         struct tun_flow_entry *e;
 349         u32 txq = 0;
 350         u32 numqueues = 0;
 351
 352         rcu_read_lock();
 353         numqueues = tun->numqueues;
 354
 355         txq = skb_get_rxhash(skb);
 356         if (txq) {
 357                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 358                 if (e)
 359                         txq = e->queue_index;
 360                 else
 361                         /* use multiply and shift instead of expensive divide */
 362                         txq = ((u64)txq * numqueues) >> 32;
 363         } else if (likely(skb_rx_queue_recorded(skb))) {
 364                 txq = skb_get_rx_queue(skb);
 365                 while (unlikely(txq >= numqueues))
 366                         txq -= numqueues;
 367         }
 368
 369         rcu_read_unlock();
 370         return txq;
 371 }
 372
 373 static inline bool tun_not_capable(struct tun_struct *tun)
 374 {
 375         const struct cred *cred = current_cred();
 376
 377         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 378                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 379                 !capable(CAP_NET_ADMIN);
 380 }
 381
 382 static void tun_set_real_num_queues(struct tun_struct *tun)
 383 {
 384         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 385         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 386 }
 387
 388 static void __tun_detach(struct tun_file *tfile, bool clean)
 389 {
 390         struct tun_file *ntfile;
 391         struct tun_struct *tun;
 392         struct net_device *dev;
 393
 394         tun = rcu_dereference_protected(tfile->tun,
 395                                         lockdep_rtnl_is_held());
 396         if (tun) {
 397                 u16 index = tfile->queue_index;
 398                 BUG_ON(index >= tun->numqueues);
 399                 dev = tun->dev;
 400
 401                 rcu_assign_pointer(tun->tfiles[index],
 402                                    tun->tfiles[tun->numqueues - 1]);
 403                 rcu_assign_pointer(tfile->tun, NULL);
 404                 ntfile = rcu_dereference_protected(tun->tfiles[index],
 405                                                    lockdep_rtnl_is_held());
 406                 ntfile->queue_index = index;
 407
 408                 --tun->numqueues;
 409                 sock_put(&tfile->sk);
 410
 411                 synchronize_net();
 412                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 413                 /* Drop read queue */
 414                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 415                 tun_set_real_num_queues(tun);
 416
 417                 if (tun->numqueues == 0 && !(tun->flags & TUN_PERSIST))
 418                         if (dev->reg_state == NETREG_REGISTERED)
 419                                 unregister_netdevice(dev);
 420         }
 421
 422         if (clean) {
 423                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 424                                  &tfile->socket.flags));
 425                 sk_release_kernel(&tfile->sk);
 426         }
 427 }
 428
 429 static void tun_detach(struct tun_file *tfile, bool clean)
 430 {
 431         rtnl_lock();
 432         __tun_detach(tfile, clean);
 433         rtnl_unlock();
 434 }
 435
 436 static void tun_detach_all(struct net_device *dev)
 437 {
 438         struct tun_struct *tun = netdev_priv(dev);
 439         struct tun_file *tfile;
 440         int i, n = tun->numqueues;
 441
 442         for (i = 0; i < n; i++) {
 443                 tfile = rcu_dereference_protected(tun->tfiles[i],
 444                                                   lockdep_rtnl_is_held());
 445                 BUG_ON(!tfile);
 446                 wake_up_all(&tfile->wq.wait);
 447                 rcu_assign_pointer(tfile->tun, NULL);
 448                 --tun->numqueues;
 449         }
 450         BUG_ON(tun->numqueues != 0);
 451
 452         synchronize_net();
 453         for (i = 0; i < n; i++) {
 454                 tfile = rcu_dereference_protected(tun->tfiles[i],
 455                                                   lockdep_rtnl_is_held());
 456                 /* Drop read queue */
 457                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 458                 sock_put(&tfile->sk);
 459         }
 460 }
 461
 462 static int tun_attach(struct tun_struct *tun, struct file *file)
 463 {
 464         struct tun_file *tfile = file->private_data;
 465         int err;
 466
 467         err = -EINVAL;
 468         if (rcu_dereference_protected(tfile->tun, lockdep_rtnl_is_held()))
 469                 goto out;
 470
 471         err = -EBUSY;
 472         if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 473                 goto out;
 474
 475         err = -E2BIG;
 476         if (tun->numqueues == MAX_TAP_QUEUES)
 477                 goto out;
 478
 479         err = 0;
 480
 481         /* Re-attach the filter to presist device */
 482         if (tun->filter_attached == true) {
 483                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 484                 if (!err)
 485                         goto out;
 486         }
 487         tfile->queue_index = tun->numqueues;
 488         rcu_assign_pointer(tfile->tun, tun);
 489         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 490         sock_hold(&tfile->sk);
 491         tun->numqueues++;
 492
 493         tun_set_real_num_queues(tun);
 494
 495         if (tun->numqueues == 1)
 496                 netif_carrier_on(tun->dev);
 497
 498         /* device is allowed to go away first, so no need to hold extra
 499          * refcnt.
 500          */
 501
 502 out:
 503         return err;
 504 }
 505
 506 static struct tun_struct *__tun_get(struct tun_file *tfile)
 507 {
 508         struct tun_struct *tun;
 509
 510         rcu_read_lock();
 511         tun = rcu_dereference(tfile->tun);
 512         if (tun)
 513                 dev_hold(tun->dev);
 514         rcu_read_unlock();
 515
 516         return tun;
 517 }
 518
 519 static struct tun_struct *tun_get(struct file *file)
 520 {
 521         return __tun_get(file->private_data);
 522 }
 523
 524 static void tun_put(struct tun_struct *tun)
 525 {
 526         dev_put(tun->dev);
 527 }
 528
 529 /* TAP filtering */
 530 static void addr_hash_set(u32 *mask, const u8 *addr)
 531 {
 532         int n = ether_crc(ETH_ALEN, addr) >> 26;
 533         mask[n >> 5] |= (1 << (n & 31));
 534 }
 535
 536 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 537 {
 538         int n = ether_crc(ETH_ALEN, addr) >> 26;
 539         return mask[n >> 5] & (1 << (n & 31));
 540 }
 541
 542 static int update_filter(struct tap_filter *filter, void __user *arg)
 543 {
 544         struct { u8 u[ETH_ALEN]; } *addr;
 545         struct tun_filter uf;
 546         int err, alen, n, nexact;
 547
 548         if (copy_from_user(&uf, arg, sizeof(uf)))
 549                 return -EFAULT;
 550
 551         if (!uf.count) {
 552                 /* Disabled */
 553                 filter->count = 0;
 554                 return 0;
 555         }
 556
 557         alen = ETH_ALEN * uf.count;
 558         addr = kmalloc(alen, GFP_KERNEL);
 559         if (!addr)
 560                 return -ENOMEM;
 561
 562         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 563                 err = -EFAULT;
 564                 goto done;
 565         }
 566
 567         /* The filter is updated without holding any locks. Which is
 568          * perfectly safe. We disable it first and in the worst
 569          * case we'll accept a few undesired packets. */
 570         filter->count = 0;
 571         wmb();
 572
 573         /* Use first set of addresses as an exact filter */
 574         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 575                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 576
 577         nexact = n;
 578
 579         /* Remaining multicast addresses are hashed,
 580          * unicast will leave the filter disabled. */
 581         memset(filter->mask, 0, sizeof(filter->mask));
 582         for (; n < uf.count; n++) {
 583                 if (!is_multicast_ether_addr(addr[n].u)) {
 584                         err = 0; /* no filter */
 585                         goto done;
 586                 }
 587                 addr_hash_set(filter->mask, addr[n].u);
 588         }
 589
 590         /* For ALLMULTI just set the mask to all ones.
 591          * This overrides the mask populated above. */
 592         if ((uf.flags & TUN_FLT_ALLMULTI))
 593                 memset(filter->mask, ~0, sizeof(filter->mask));
 594
 595         /* Now enable the filter */
 596         wmb();
 597         filter->count = nexact;
 598
 599         /* Return the number of exact filters */
 600         err = nexact;
 601
 602 done:
 603         kfree(addr);
 604         return err;
 605 }
 606
 607 /* Returns: 0 - drop, !=0 - accept */
 608 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 609 {
 610         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 611          * at this point. */
 612         struct ethhdr *eh = (struct ethhdr *) skb->data;
 613         int i;
 614
 615         /* Exact match */
 616         for (i = 0; i < filter->count; i++)
 617                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 618                         return 1;
 619
 620         /* Inexact match (multicast only) */
 621         if (is_multicast_ether_addr(eh->h_dest))
 622                 return addr_hash_test(filter->mask, eh->h_dest);
 623
 624         return 0;
 625 }
 626
 627 /*
 628  * Checks whether the packet is accepted or not.
 629  * Returns: 0 - drop, !=0 - accept
 630  */
 631 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 632 {
 633         if (!filter->count)
 634                 return 1;
 635
 636         return run_filter(filter, skb);
 637 }
 638
 639 /* Network device part of the driver */
 640
 641 static const struct ethtool_ops tun_ethtool_ops;
 642
 643 /* Net device detach from fd. */
 644 static void tun_net_uninit(struct net_device *dev)
 645 {
 646         tun_detach_all(dev);
 647 }
 648
 649 /* Net device open. */
 650 static int tun_net_open(struct net_device *dev)
 651 {
 652         netif_tx_start_all_queues(dev);
 653         return 0;
 654 }
 655
 656 /* Net device close. */
 657 static int tun_net_close(struct net_device *dev)
 658 {
 659         netif_tx_stop_all_queues(dev);
 660         return 0;
 661 }
 662
 663 /* Net device start xmit */
 664 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 665 {
 666         struct tun_struct *tun = netdev_priv(dev);
 667         int txq = skb->queue_mapping;
 668         struct tun_file *tfile;
 669
 670         rcu_read_lock();
 671         tfile = rcu_dereference(tun->tfiles[txq]);
 672
 673         /* Drop packet if interface is not attached */
 674         if (txq >= tun->numqueues)
 675                 goto drop;
 676
 677         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 678
 679         BUG_ON(!tfile);
 680
 681         /* Drop if the filter does not like it.
 682          * This is a noop if the filter is disabled.
 683          * Filter can be enabled only for the TAP devices. */
 684         if (!check_filter(&tun->txflt, skb))
 685                 goto drop;
 686
 687         if (tfile->socket.sk->sk_filter &&
 688             sk_filter(tfile->socket.sk, skb))
 689                 goto drop;
 690
 691         /* Limit the number of packets queued by divining txq length with the
 692          * number of queues.
 693          */
 694         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
 695                           >= dev->tx_queue_len / tun->numqueues){
 696                 if (!(tun->flags & TUN_ONE_QUEUE)) {
 697                         /* Normal queueing mode. */
 698                         /* Packet scheduler handles dropping of further packets. */
 699                         netif_stop_subqueue(dev, txq);
 700
 701                         /* We won't see all dropped packets individually, so overrun
 702                          * error is more appropriate. */
 703                         dev->stats.tx_fifo_errors++;
 704                 } else {
 705                         /* Single queue mode.
 706                          * Driver handles dropping of all packets itself. */
 707                         goto drop;
 708                 }
 709         }
 710
 711         /* Orphan the skb - required as we might hang on to it
 712          * for indefinite time. */
 713         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 714                 goto drop;
 715         skb_orphan(skb);
 716
 717         /* Enqueue packet */
 718         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 719
 720         /* Notify and wake up reader process */
 721         if (tfile->flags & TUN_FASYNC)
 722                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 723         wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
 724                                    POLLRDNORM | POLLRDBAND);
 725
 726         rcu_read_unlock();
 727         return NETDEV_TX_OK;
 728
 729 drop:
 730         dev->stats.tx_dropped++;
 731         skb_tx_error(skb);
 732         kfree_skb(skb);
 733         rcu_read_unlock();
 734         return NETDEV_TX_OK;
 735 }
 736
 737 static void tun_net_mclist(struct net_device *dev)
 738 {
 739         /*
 740          * This callback is supposed to deal with mc filter in
 741          * _rx_ path and has nothing to do with the _tx_ path.
 742          * In rx path we always accept everything userspace gives us.
 743          */
 744 }
 745
 746 #define MIN_MTU 68
 747 #define MAX_MTU 65535
 748
 749 static int
 750 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 751 {
 752         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 753                 return -EINVAL;
 754         dev->mtu = new_mtu;
 755         return 0;
 756 }
 757
 758 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 759         netdev_features_t features)
 760 {
 761         struct tun_struct *tun = netdev_priv(dev);
 762
 763         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 764 }
 765 #ifdef CONFIG_NET_POLL_CONTROLLER
 766 static void tun_poll_controller(struct net_device *dev)
 767 {
 768         /*
 769          * Tun only receives frames when:
 770          * 1) the char device endpoint gets data from user space
 771          * 2) the tun socket gets a sendmsg call from user space
 772          * Since both of those are syncronous operations, we are guaranteed
 773          * never to have pending data when we poll for it
 774          * so theres nothing to do here but return.
 775          * We need this though so netpoll recognizes us as an interface that
 776          * supports polling, which enables bridge devices in virt setups to
 777          * still use netconsole
 778          */
 779         return;
 780 }
 781 #endif
 782 static const struct net_device_ops tun_netdev_ops = {
 783         .ndo_uninit             = tun_net_uninit,
 784         .ndo_open               = tun_net_open,
 785         .ndo_stop               = tun_net_close,
 786         .ndo_start_xmit         = tun_net_xmit,
 787         .ndo_change_mtu         = tun_net_change_mtu,
 788         .ndo_fix_features       = tun_net_fix_features,
 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 const struct net_device_ops tap_netdev_ops = {
 796         .ndo_uninit             = tun_net_uninit,
 797         .ndo_open               = tun_net_open,
 798         .ndo_stop               = tun_net_close,
 799         .ndo_start_xmit         = tun_net_xmit,
 800         .ndo_change_mtu         = tun_net_change_mtu,
 801         .ndo_fix_features       = tun_net_fix_features,
 802         .ndo_set_rx_mode        = tun_net_mclist,
 803         .ndo_set_mac_address    = eth_mac_addr,
 804         .ndo_validate_addr      = eth_validate_addr,
 805         .ndo_select_queue       = tun_select_queue,
 806 #ifdef CONFIG_NET_POLL_CONTROLLER
 807         .ndo_poll_controller    = tun_poll_controller,
 808 #endif
 809 };
 810
 811 static int tun_flow_init(struct tun_struct *tun)
 812 {
 813         int i;
 814
 815         tun->flow_cache = kmem_cache_create("tun_flow_cache",
 816                                             sizeof(struct tun_flow_entry), 0, 0,
 817                                             NULL);
 818         if (!tun->flow_cache)
 819                 return -ENOMEM;
 820
 821         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 822                 INIT_HLIST_HEAD(&tun->flows[i]);
 823
 824         tun->ageing_time = TUN_FLOW_EXPIRE;
 825         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 826         mod_timer(&tun->flow_gc_timer,
 827                   round_jiffies_up(jiffies + tun->ageing_time));
 828
 829         return 0;
 830 }
 831
 832 static void tun_flow_uninit(struct tun_struct *tun)
 833 {
 834         del_timer_sync(&tun->flow_gc_timer);
 835         tun_flow_flush(tun);
 836
 837         /* Wait for completion of call_rcu()'s */
 838         rcu_barrier();
 839         kmem_cache_destroy(tun->flow_cache);
 840 }
 841
 842 /* Initialize net device. */
 843 static void tun_net_init(struct net_device *dev)
 844 {
 845         struct tun_struct *tun = netdev_priv(dev);
 846
 847         switch (tun->flags & TUN_TYPE_MASK) {
 848         case TUN_TUN_DEV:
 849                 dev->netdev_ops = &tun_netdev_ops;
 850
 851                 /* Point-to-Point TUN Device */
 852                 dev->hard_header_len = 0;
 853                 dev->addr_len = 0;
 854                 dev->mtu = 1500;
 855
 856                 /* Zero header length */
 857                 dev->type = ARPHRD_NONE;
 858                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 859                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 860                 break;
 861
 862         case TUN_TAP_DEV:
 863                 dev->netdev_ops = &tap_netdev_ops;
 864                 /* Ethernet TAP Device */
 865                 ether_setup(dev);
 866                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 867
 868                 eth_hw_addr_random(dev);
 869
 870                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 871                 break;
 872         }
 873 }
 874
 875 /* Character device part */
 876
 877 /* Poll */
 878 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 879 {
 880         struct tun_file *tfile = file->private_data;
 881         struct tun_struct *tun = __tun_get(tfile);
 882         struct sock *sk;
 883         unsigned int mask = 0;
 884
 885         if (!tun)
 886                 return POLLERR;
 887
 888         sk = tfile->socket.sk;
 889
 890         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 891
 892         poll_wait(file, &tfile->wq.wait, wait);
 893
 894         if (!skb_queue_empty(&sk->sk_receive_queue))
 895                 mask |= POLLIN | POLLRDNORM;
 896
 897         if (sock_writeable(sk) ||
 898             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 899              sock_writeable(sk)))
 900                 mask |= POLLOUT | POLLWRNORM;
 901
 902         if (tun->dev->reg_state != NETREG_REGISTERED)
 903                 mask = POLLERR;
 904
 905         tun_put(tun);
 906         return mask;
 907 }
 908
 909 /* prepad is the amount to reserve at front.  len is length after that.
 910  * linear is a hint as to how much to copy (usually headers). */
 911 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 912                                      size_t prepad, size_t len,
 913                                      size_t linear, int noblock)
 914 {
 915         struct sock *sk = tfile->socket.sk;
 916         struct sk_buff *skb;
 917         int err;
 918
 919         /* Under a page?  Don't bother with paged skb. */
 920         if (prepad + len < PAGE_SIZE || !linear)
 921                 linear = len;
 922
 923         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 924                                    &err);
 925         if (!skb)
 926                 return ERR_PTR(err);
 927
 928         skb_reserve(skb, prepad);
 929         skb_put(skb, linear);
 930         skb->data_len = len - linear;
 931         skb->len += len - linear;
 932
 933         return skb;
 934 }
 935
 936 /* set skb frags from iovec, this can move to core network code for reuse */
 937 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 938                                   int offset, size_t count)
 939 {
 940         int len = iov_length(from, count) - offset;
 941         int copy = skb_headlen(skb);
 942         int size, offset1 = 0;
 943         int i = 0;
 944
 945         /* Skip over from offset */
 946         while (count && (offset >= from->iov_len)) {
 947                 offset -= from->iov_len;
 948                 ++from;
 949                 --count;
 950         }
 951
 952         /* copy up to skb headlen */
 953         while (count && (copy > 0)) {
 954                 size = min_t(unsigned int, copy, from->iov_len - offset);
 955                 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
 956                                    size))
 957                         return -EFAULT;
 958                 if (copy > size) {
 959                         ++from;
 960                         --count;
 961                         offset = 0;
 962                 } else
 963                         offset += size;
 964                 copy -= size;
 965                 offset1 += size;
 966         }
 967
 968         if (len == offset1)
 969                 return 0;
 970
 971         while (count--) {
 972                 struct page *page[MAX_SKB_FRAGS];
 973                 int num_pages;
 974                 unsigned long base;
 975                 unsigned long truesize;
 976
 977                 len = from->iov_len - offset;
 978                 if (!len) {
 979                         offset = 0;
 980                         ++from;
 981                         continue;
 982                 }
 983                 base = (unsigned long)from->iov_base + offset;
 984                 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
 985                 if (i + size > MAX_SKB_FRAGS)
 986                         return -EMSGSIZE;
 987                 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
 988                 if (num_pages != size) {
 989                         for (i = 0; i < num_pages; i++)
 990                                 put_page(page[i]);
 991                         return -EFAULT;
 992                 }
 993                 truesize = size * PAGE_SIZE;
 994                 skb->data_len += len;
 995                 skb->len += len;
 996                 skb->truesize += truesize;
 997                 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
 998                 while (len) {
 999                         int off = base & ~PAGE_MASK;
1000                         int size = min_t(int, len, PAGE_SIZE - off);
1001                         __skb_fill_page_desc(skb, i, page[i], off, size);
1002                         skb_shinfo(skb)->nr_frags++;
1003                         /* increase sk_wmem_alloc */
1004                         base += size;
1005                         len -= size;
1006                         i++;
1007                 }
1008                 offset = 0;
1009                 ++from;
1010         }
1011         return 0;
1012 }
1013
1014 /* Get packet from user space buffer */
1015 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1016                             void *msg_control, const struct iovec *iv,
1017                             size_t total_len, size_t count, int noblock)
1018 {
1019         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1020         struct sk_buff *skb;
1021         size_t len = total_len, align = NET_SKB_PAD;
1022         struct virtio_net_hdr gso = { 0 };
1023         int offset = 0;
1024         int copylen;
1025         bool zerocopy = false;
1026         int err;
1027
1028         if (!(tun->flags & TUN_NO_PI)) {
1029                 if ((len -= sizeof(pi)) > total_len)
1030                         return -EINVAL;
1031
1032                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1033                         return -EFAULT;
1034                 offset += sizeof(pi);
1035         }
1036
1037         if (tun->flags & TUN_VNET_HDR) {
1038                 if ((len -= tun->vnet_hdr_sz) > total_len)
1039                         return -EINVAL;
1040
1041                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1042                         return -EFAULT;
1043
1044                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1045                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1046                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1047
1048                 if (gso.hdr_len > len)
1049                         return -EINVAL;
1050                 offset += tun->vnet_hdr_sz;
1051         }
1052
1053         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1054                 align += NET_IP_ALIGN;
1055                 if (unlikely(len < ETH_HLEN ||
1056                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1057                         return -EINVAL;
1058         }
1059
1060         if (msg_control)
1061                 zerocopy = true;
1062
1063         if (zerocopy) {
1064                 /* Userspace may produce vectors with count greater than
1065                  * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1066                  * to let the rest of data to be fit in the frags.
1067                  */
1068                 if (count > MAX_SKB_FRAGS) {
1069                         copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1070                         if (copylen < offset)
1071                                 copylen = 0;
1072                         else
1073                                 copylen -= offset;
1074                 } else
1075                                 copylen = 0;
1076                 /* There are 256 bytes to be copied in skb, so there is enough
1077                  * room for skb expand head in case it is used.
1078                  * The rest of the buffer is mapped from userspace.
1079                  */
1080                 if (copylen < gso.hdr_len)
1081                         copylen = gso.hdr_len;
1082                 if (!copylen)
1083                         copylen = GOODCOPY_LEN;
1084         } else
1085                 copylen = len;
1086
1087         skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
1088         if (IS_ERR(skb)) {
1089                 if (PTR_ERR(skb) != -EAGAIN)
1090                         tun->dev->stats.rx_dropped++;
1091                 return PTR_ERR(skb);
1092         }
1093
1094         if (zerocopy)
1095                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1096         else
1097                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1098
1099         if (err) {
1100                 tun->dev->stats.rx_dropped++;
1101                 kfree_skb(skb);
1102                 return -EFAULT;
1103         }
1104
1105         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1106                 if (!skb_partial_csum_set(skb, gso.csum_start,
1107                                           gso.csum_offset)) {
1108                         tun->dev->stats.rx_frame_errors++;
1109                         kfree_skb(skb);
1110                         return -EINVAL;
1111                 }
1112         }
1113
1114         switch (tun->flags & TUN_TYPE_MASK) {
1115         case TUN_TUN_DEV:
1116                 if (tun->flags & TUN_NO_PI) {
1117                         switch (skb->data[0] & 0xf0) {
1118                         case 0x40:
1119                                 pi.proto = htons(ETH_P_IP);
1120                                 break;
1121                         case 0x60:
1122                                 pi.proto = htons(ETH_P_IPV6);
1123                                 break;
1124                         default:
1125                                 tun->dev->stats.rx_dropped++;
1126                                 kfree_skb(skb);
1127                                 return -EINVAL;
1128                         }
1129                 }
1130
1131                 skb_reset_mac_header(skb);
1132                 skb->protocol = pi.proto;
1133                 skb->dev = tun->dev;
1134                 break;
1135         case TUN_TAP_DEV:
1136                 skb->protocol = eth_type_trans(skb, tun->dev);
1137                 break;
1138         }
1139
1140         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1141                 pr_debug("GSO!\n");
1142                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1143                 case VIRTIO_NET_HDR_GSO_TCPV4:
1144                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1145                         break;
1146                 case VIRTIO_NET_HDR_GSO_TCPV6:
1147                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1148                         break;
1149                 case VIRTIO_NET_HDR_GSO_UDP:
1150                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1151                         break;
1152                 default:
1153                         tun->dev->stats.rx_frame_errors++;
1154                         kfree_skb(skb);
1155                         return -EINVAL;
1156                 }
1157
1158                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1159                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1160
1161                 skb_shinfo(skb)->gso_size = gso.gso_size;
1162                 if (skb_shinfo(skb)->gso_size == 0) {
1163                         tun->dev->stats.rx_frame_errors++;
1164                         kfree_skb(skb);
1165                         return -EINVAL;
1166                 }
1167
1168                 /* Header must be checked, and gso_segs computed. */
1169                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1170                 skb_shinfo(skb)->gso_segs = 0;
1171         }
1172
1173         /* copy skb_ubuf_info for callback when skb has no error */
1174         if (zerocopy) {
1175                 skb_shinfo(skb)->destructor_arg = msg_control;
1176                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1177         }
1178
1179         netif_rx_ni(skb);
1180
1181         tun->dev->stats.rx_packets++;
1182         tun->dev->stats.rx_bytes += len;
1183
1184         tun_flow_update(tun, skb, tfile->queue_index);
1185         return total_len;
1186 }
1187
1188 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1189                               unsigned long count, loff_t pos)
1190 {
1191         struct file *file = iocb->ki_filp;
1192         struct tun_struct *tun = tun_get(file);
1193         struct tun_file *tfile = file->private_data;
1194         ssize_t result;
1195
1196         if (!tun)
1197                 return -EBADFD;
1198
1199         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1200
1201         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1202                               count, file->f_flags & O_NONBLOCK);
1203
1204         tun_put(tun);
1205         return result;
1206 }
1207
1208 /* Put packet to the user space buffer */
1209 static ssize_t tun_put_user(struct tun_struct *tun,
1210                             struct tun_file *tfile,
1211                             struct sk_buff *skb,
1212                             const struct iovec *iv, int len)
1213 {
1214         struct tun_pi pi = { 0, skb->protocol };
1215         ssize_t total = 0;
1216
1217         if (!(tun->flags & TUN_NO_PI)) {
1218                 if ((len -= sizeof(pi)) < 0)
1219                         return -EINVAL;
1220
1221                 if (len < skb->len) {
1222                         /* Packet will be striped */
1223                         pi.flags |= TUN_PKT_STRIP;
1224                 }
1225
1226                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1227                         return -EFAULT;
1228                 total += sizeof(pi);
1229         }
1230
1231         if (tun->flags & TUN_VNET_HDR) {
1232                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1233                 if ((len -= tun->vnet_hdr_sz) < 0)
1234                         return -EINVAL;
1235
1236                 if (skb_is_gso(skb)) {
1237                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1238
1239                         /* This is a hint as to how much should be linear. */
1240                         gso.hdr_len = skb_headlen(skb);
1241                         gso.gso_size = sinfo->gso_size;
1242                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1243                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1244                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1245                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1246                         else if (sinfo->gso_type & SKB_GSO_UDP)
1247                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1248                         else {
1249                                 pr_err("unexpected GSO type: "
1250                                        "0x%x, gso_size %d, hdr_len %d\n",
1251                                        sinfo->gso_type, gso.gso_size,
1252                                        gso.hdr_len);
1253                                 print_hex_dump(KERN_ERR, "tun: ",
1254                                                DUMP_PREFIX_NONE,
1255                                                16, 1, skb->head,
1256                                                min((int)gso.hdr_len, 64), true);
1257                                 WARN_ON_ONCE(1);
1258                                 return -EINVAL;
1259                         }
1260                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1261                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1262                 } else
1263                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1264
1265                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1266                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1267                         gso.csum_start = skb_checksum_start_offset(skb);
1268                         gso.csum_offset = skb->csum_offset;
1269                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1270                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1271                 } /* else everything is zero */
1272
1273                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1274                                                sizeof(gso))))
1275                         return -EFAULT;
1276                 total += tun->vnet_hdr_sz;
1277         }
1278
1279         len = min_t(int, skb->len, len);
1280
1281         skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1282         total += skb->len;
1283
1284         tun->dev->stats.tx_packets++;
1285         tun->dev->stats.tx_bytes += len;
1286
1287         return total;
1288 }
1289
1290 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1291                            struct kiocb *iocb, const struct iovec *iv,
1292                            ssize_t len, int noblock)
1293 {
1294         DECLARE_WAITQUEUE(wait, current);
1295         struct sk_buff *skb;
1296         ssize_t ret = 0;
1297
1298         tun_debug(KERN_INFO, tun, "tun_chr_read\n");
1299
1300         if (unlikely(!noblock))
1301                 add_wait_queue(&tfile->wq.wait, &wait);
1302         while (len) {
1303                 current->state = TASK_INTERRUPTIBLE;
1304
1305                 /* Read frames from the queue */
1306                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1307                         if (noblock) {
1308                                 ret = -EAGAIN;
1309                                 break;
1310                         }
1311                         if (signal_pending(current)) {
1312                                 ret = -ERESTARTSYS;
1313                                 break;
1314                         }
1315                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1316                                 ret = -EIO;
1317                                 break;
1318                         }
1319
1320                         /* Nothing to read, let's sleep */
1321                         schedule();
1322                         continue;
1323                 }
1324                 netif_wake_subqueue(tun->dev, tfile->queue_index);
1325
1326                 ret = tun_put_user(tun, tfile, skb, iv, len);
1327                 kfree_skb(skb);
1328                 break;
1329         }
1330
1331         current->state = TASK_RUNNING;
1332         if (unlikely(!noblock))
1333                 remove_wait_queue(&tfile->wq.wait, &wait);
1334
1335         return ret;
1336 }
1337
1338 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1339                             unsigned long count, loff_t pos)
1340 {
1341         struct file *file = iocb->ki_filp;
1342         struct tun_file *tfile = file->private_data;
1343         struct tun_struct *tun = __tun_get(tfile);
1344         ssize_t len, ret;
1345
1346         if (!tun)
1347                 return -EBADFD;
1348         len = iov_length(iv, count);
1349         if (len < 0) {
1350                 ret = -EINVAL;
1351                 goto out;
1352         }
1353
1354         ret = tun_do_read(tun, tfile, iocb, iv, len,
1355                           file->f_flags & O_NONBLOCK);
1356         ret = min_t(ssize_t, ret, len);
1357 out:
1358         tun_put(tun);
1359         return ret;
1360 }
1361
1362 static void tun_free_netdev(struct net_device *dev)
1363 {
1364         struct tun_struct *tun = netdev_priv(dev);
1365
1366         tun_flow_uninit(tun);
1367         free_netdev(dev);
1368 }
1369
1370 static void tun_setup(struct net_device *dev)
1371 {
1372         struct tun_struct *tun = netdev_priv(dev);
1373
1374         tun->owner = INVALID_UID;
1375         tun->group = INVALID_GID;
1376
1377         dev->ethtool_ops = &tun_ethtool_ops;
1378         dev->destructor = tun_free_netdev;
1379 }
1380
1381 /* Trivial set of netlink ops to allow deleting tun or tap
1382  * device with netlink.
1383  */
1384 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1385 {
1386         return -EINVAL;
1387 }
1388
1389 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1390         .kind           = DRV_NAME,
1391         .priv_size      = sizeof(struct tun_struct),
1392         .setup          = tun_setup,
1393         .validate       = tun_validate,
1394 };
1395
1396 static void tun_sock_write_space(struct sock *sk)
1397 {
1398         struct tun_file *tfile;
1399         wait_queue_head_t *wqueue;
1400
1401         if (!sock_writeable(sk))
1402                 return;
1403
1404         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1405                 return;
1406
1407         wqueue = sk_sleep(sk);
1408         if (wqueue && waitqueue_active(wqueue))
1409                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1410                                                 POLLWRNORM | POLLWRBAND);
1411
1412         tfile = container_of(sk, struct tun_file, sk);
1413         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1414 }
1415
1416 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1417                        struct msghdr *m, size_t total_len)
1418 {
1419         int ret;
1420         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1421         struct tun_struct *tun = __tun_get(tfile);
1422
1423         if (!tun)
1424                 return -EBADFD;
1425         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1426                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1427         tun_put(tun);
1428         return ret;
1429 }
1430
1431
1432 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1433                        struct msghdr *m, size_t total_len,
1434                        int flags)
1435 {
1436         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1437         struct tun_struct *tun = __tun_get(tfile);
1438         int ret;
1439
1440         if (!tun)
1441                 return -EBADFD;
1442
1443         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1444                 return -EINVAL;
1445         ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1446                           flags & MSG_DONTWAIT);
1447         if (ret > total_len) {
1448                 m->msg_flags |= MSG_TRUNC;
1449                 ret = flags & MSG_TRUNC ? ret : total_len;
1450         }
1451         tun_put(tun);
1452         return ret;
1453 }
1454
1455 static int tun_release(struct socket *sock)
1456 {
1457         if (sock->sk)
1458                 sock_put(sock->sk);
1459         return 0;
1460 }
1461
1462 /* Ops structure to mimic raw sockets with tun */
1463 static const struct proto_ops tun_socket_ops = {
1464         .sendmsg = tun_sendmsg,
1465         .recvmsg = tun_recvmsg,
1466         .release = tun_release,
1467 };
1468
1469 static struct proto tun_proto = {
1470         .name           = "tun",
1471         .owner          = THIS_MODULE,
1472         .obj_size       = sizeof(struct tun_file),
1473 };
1474
1475 static int tun_flags(struct tun_struct *tun)
1476 {
1477         int flags = 0;
1478
1479         if (tun->flags & TUN_TUN_DEV)
1480                 flags |= IFF_TUN;
1481         else
1482                 flags |= IFF_TAP;
1483
1484         if (tun->flags & TUN_NO_PI)
1485                 flags |= IFF_NO_PI;
1486
1487         if (tun->flags & TUN_ONE_QUEUE)
1488                 flags |= IFF_ONE_QUEUE;
1489
1490         if (tun->flags & TUN_VNET_HDR)
1491                 flags |= IFF_VNET_HDR;
1492
1493         if (tun->flags & TUN_TAP_MQ)
1494                 flags |= IFF_MULTI_QUEUE;
1495
1496         return flags;
1497 }
1498
1499 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1500                               char *buf)
1501 {
1502         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1503         return sprintf(buf, "0x%x\n", tun_flags(tun));
1504 }
1505
1506 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1507                               char *buf)
1508 {
1509         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1510         return uid_valid(tun->owner)?
1511                 sprintf(buf, "%u\n",
1512                         from_kuid_munged(current_user_ns(), tun->owner)):
1513                 sprintf(buf, "-1\n");
1514 }
1515
1516 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1517                               char *buf)
1518 {
1519         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1520         return gid_valid(tun->group) ?
1521                 sprintf(buf, "%u\n",
1522                         from_kgid_munged(current_user_ns(), tun->group)):
1523                 sprintf(buf, "-1\n");
1524 }
1525
1526 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1527 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1528 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1529
1530 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1531 {
1532         struct tun_struct *tun;
1533         struct tun_file *tfile = file->private_data;
1534         struct net_device *dev;
1535         int err;
1536
1537         dev = __dev_get_by_name(net, ifr->ifr_name);
1538         if (dev) {
1539                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1540                         return -EBUSY;
1541                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1542                         tun = netdev_priv(dev);
1543                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1544                         tun = netdev_priv(dev);
1545                 else
1546                         return -EINVAL;
1547
1548                 if (tun_not_capable(tun))
1549                         return -EPERM;
1550                 err = security_tun_dev_attach(tfile->socket.sk);
1551                 if (err < 0)
1552                         return err;
1553
1554                 err = tun_attach(tun, file);
1555                 if (err < 0)
1556                         return err;
1557         }
1558         else {
1559                 char *name;
1560                 unsigned long flags = 0;
1561
1562                 if (!capable(CAP_NET_ADMIN))
1563                         return -EPERM;
1564                 err = security_tun_dev_create();
1565                 if (err < 0)
1566                         return err;
1567
1568                 /* Set dev type */
1569                 if (ifr->ifr_flags & IFF_TUN) {
1570                         /* TUN device */
1571                         flags |= TUN_TUN_DEV;
1572                         name = "tun%d";
1573                 } else if (ifr->ifr_flags & IFF_TAP) {
1574                         /* TAP device */
1575                         flags |= TUN_TAP_DEV;
1576                         name = "tap%d";
1577                 } else
1578                         return -EINVAL;
1579
1580                 if (*ifr->ifr_name)
1581                         name = ifr->ifr_name;
1582
1583                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1584                                        tun_setup,
1585                                        MAX_TAP_QUEUES, MAX_TAP_QUEUES);
1586                 if (!dev)
1587                         return -ENOMEM;
1588
1589                 dev_net_set(dev, net);
1590                 dev->rtnl_link_ops = &tun_link_ops;
1591
1592                 tun = netdev_priv(dev);
1593                 tun->dev = dev;
1594                 tun->flags = flags;
1595                 tun->txflt.count = 0;
1596                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1597
1598                 tun->filter_attached = false;
1599                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1600
1601                 spin_lock_init(&tun->lock);
1602
1603                 security_tun_dev_post_create(&tfile->sk);
1604
1605                 tun_net_init(dev);
1606
1607                 if (tun_flow_init(tun))
1608                         goto err_free_dev;
1609
1610                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1611                         TUN_USER_FEATURES;
1612                 dev->features = dev->hw_features;
1613
1614                 err = register_netdevice(tun->dev);
1615                 if (err < 0)
1616                         goto err_free_dev;
1617
1618                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1619                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1620                     device_create_file(&tun->dev->dev, &dev_attr_group))
1621                         pr_err("Failed to create tun sysfs files\n");
1622
1623                 err = tun_attach(tun, file);
1624                 if (err < 0)
1625                         goto err_free_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 int 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         return 0;
1674 }
1675
1676 /* This is like a cut-down ethtool ops, except done via tun fd so no
1677  * privs required. */
1678 static int set_offload(struct tun_struct *tun, unsigned long arg)
1679 {
1680         netdev_features_t features = 0;
1681
1682         if (arg & TUN_F_CSUM) {
1683                 features |= NETIF_F_HW_CSUM;
1684                 arg &= ~TUN_F_CSUM;
1685
1686                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1687                         if (arg & TUN_F_TSO_ECN) {
1688                                 features |= NETIF_F_TSO_ECN;
1689                                 arg &= ~TUN_F_TSO_ECN;
1690                         }
1691                         if (arg & TUN_F_TSO4)
1692                                 features |= NETIF_F_TSO;
1693                         if (arg & TUN_F_TSO6)
1694                                 features |= NETIF_F_TSO6;
1695                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1696                 }
1697
1698                 if (arg & TUN_F_UFO) {
1699                         features |= NETIF_F_UFO;
1700                         arg &= ~TUN_F_UFO;
1701                 }
1702         }
1703
1704         /* This gives the user a way to test for new features in future by
1705          * trying to set them. */
1706         if (arg)
1707                 return -EINVAL;
1708
1709         tun->set_features = features;
1710         netdev_update_features(tun->dev);
1711
1712         return 0;
1713 }
1714
1715 static void tun_detach_filter(struct tun_struct *tun, int n)
1716 {
1717         int i;
1718         struct tun_file *tfile;
1719
1720         for (i = 0; i < n; i++) {
1721                 tfile = rcu_dereference_protected(tun->tfiles[i],
1722                                                   lockdep_rtnl_is_held());
1723                 sk_detach_filter(tfile->socket.sk);
1724         }
1725
1726         tun->filter_attached = false;
1727 }
1728
1729 static int tun_attach_filter(struct tun_struct *tun)
1730 {
1731         int i, ret = 0;
1732         struct tun_file *tfile;
1733
1734         for (i = 0; i < tun->numqueues; i++) {
1735                 tfile = rcu_dereference_protected(tun->tfiles[i],
1736                                                   lockdep_rtnl_is_held());
1737                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1738                 if (ret) {
1739                         tun_detach_filter(tun, i);
1740                         return ret;
1741                 }
1742         }
1743
1744         tun->filter_attached = true;
1745         return ret;
1746 }
1747
1748 static void tun_set_sndbuf(struct tun_struct *tun)
1749 {
1750         struct tun_file *tfile;
1751         int i;
1752
1753         for (i = 0; i < tun->numqueues; i++) {
1754                 tfile = rcu_dereference_protected(tun->tfiles[i],
1755                                                 lockdep_rtnl_is_held());
1756                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1757         }
1758 }
1759
1760 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1761 {
1762         struct tun_file *tfile = file->private_data;
1763         struct tun_struct *tun;
1764         struct net_device *dev;
1765         int ret = 0;
1766
1767         rtnl_lock();
1768
1769         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1770                 dev = __dev_get_by_name(tfile->net, ifr->ifr_name);
1771                 if (!dev) {
1772                         ret = -EINVAL;
1773                         goto unlock;
1774                 }
1775
1776                 tun = netdev_priv(dev);
1777                 if (dev->netdev_ops != &tap_netdev_ops &&
1778                         dev->netdev_ops != &tun_netdev_ops)
1779                         ret = -EINVAL;
1780                 else if (tun_not_capable(tun))
1781                         ret = -EPERM;
1782                 else
1783                         ret = tun_attach(tun, file);
1784         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE)
1785                 __tun_detach(tfile, false);
1786         else
1787                 ret = -EINVAL;
1788
1789 unlock:
1790         rtnl_unlock();
1791         return ret;
1792 }
1793
1794 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1795                             unsigned long arg, int ifreq_len)
1796 {
1797         struct tun_file *tfile = file->private_data;
1798         struct tun_struct *tun;
1799         void __user* argp = (void __user*)arg;
1800         struct ifreq ifr;
1801         kuid_t owner;
1802         kgid_t group;
1803         int sndbuf;
1804         int vnet_hdr_sz;
1805         int ret;
1806
1807         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1808                 if (copy_from_user(&ifr, argp, ifreq_len))
1809                         return -EFAULT;
1810         } else {
1811                 memset(&ifr, 0, sizeof(ifr));
1812         }
1813         if (cmd == TUNGETFEATURES) {
1814                 /* Currently this just means: "what IFF flags are valid?".
1815                  * This is needed because we never checked for invalid flags on
1816                  * TUNSETIFF. */
1817                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1818                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1819                                 (unsigned int __user*)argp);
1820         } else if (cmd == TUNSETQUEUE)
1821                 return tun_set_queue(file, &ifr);
1822
1823         ret = 0;
1824         rtnl_lock();
1825
1826         tun = __tun_get(tfile);
1827         if (cmd == TUNSETIFF && !tun) {
1828                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1829
1830                 ret = tun_set_iff(tfile->net, file, &ifr);
1831
1832                 if (ret)
1833                         goto unlock;
1834
1835                 if (copy_to_user(argp, &ifr, ifreq_len))
1836                         ret = -EFAULT;
1837                 goto unlock;
1838         }
1839
1840         ret = -EBADFD;
1841         if (!tun)
1842                 goto unlock;
1843
1844         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1845
1846         ret = 0;
1847         switch (cmd) {
1848         case TUNGETIFF:
1849                 ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1850                 if (ret)
1851                         break;
1852
1853                 if (copy_to_user(argp, &ifr, ifreq_len))
1854                         ret = -EFAULT;
1855                 break;
1856
1857         case TUNSETNOCSUM:
1858                 /* Disable/Enable checksum */
1859
1860                 /* [unimplemented] */
1861                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1862                           arg ? "disabled" : "enabled");
1863                 break;
1864
1865         case TUNSETPERSIST:
1866                 /* Disable/Enable persist mode. Keep an extra reference to the
1867                  * module to prevent the module being unprobed.
1868                  */
1869                 if (arg) {
1870                         tun->flags |= TUN_PERSIST;
1871                         __module_get(THIS_MODULE);
1872                 } else {
1873                         tun->flags &= ~TUN_PERSIST;
1874                         module_put(THIS_MODULE);
1875                 }
1876
1877                 tun_debug(KERN_INFO, tun, "persist %s\n",
1878                           arg ? "enabled" : "disabled");
1879                 break;
1880
1881         case TUNSETOWNER:
1882                 /* Set owner of the device */
1883                 owner = make_kuid(current_user_ns(), arg);
1884                 if (!uid_valid(owner)) {
1885                         ret = -EINVAL;
1886                         break;
1887                 }
1888                 tun->owner = owner;
1889                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1890                           from_kuid(&init_user_ns, tun->owner));
1891                 break;
1892
1893         case TUNSETGROUP:
1894                 /* Set group of the device */
1895                 group = make_kgid(current_user_ns(), arg);
1896                 if (!gid_valid(group)) {
1897                         ret = -EINVAL;
1898                         break;
1899                 }
1900                 tun->group = group;
1901                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1902                           from_kgid(&init_user_ns, tun->group));
1903                 break;
1904
1905         case TUNSETLINK:
1906                 /* Only allow setting the type when the interface is down */
1907                 if (tun->dev->flags & IFF_UP) {
1908                         tun_debug(KERN_INFO, tun,
1909                                   "Linktype set failed because interface is up\n");
1910                         ret = -EBUSY;
1911                 } else {
1912                         tun->dev->type = (int) arg;
1913                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1914                                   tun->dev->type);
1915                         ret = 0;
1916                 }
1917                 break;
1918
1919 #ifdef TUN_DEBUG
1920         case TUNSETDEBUG:
1921                 tun->debug = arg;
1922                 break;
1923 #endif
1924         case TUNSETOFFLOAD:
1925                 ret = set_offload(tun, arg);
1926                 break;
1927
1928         case TUNSETTXFILTER:
1929                 /* Can be set only for TAPs */
1930                 ret = -EINVAL;
1931                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1932                         break;
1933                 ret = update_filter(&tun->txflt, (void __user *)arg);
1934                 break;
1935
1936         case SIOCGIFHWADDR:
1937                 /* Get hw address */
1938                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1939                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1940                 if (copy_to_user(argp, &ifr, ifreq_len))
1941                         ret = -EFAULT;
1942                 break;
1943
1944         case SIOCSIFHWADDR:
1945                 /* Set hw address */
1946                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1947                           ifr.ifr_hwaddr.sa_data);
1948
1949                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1950                 break;
1951
1952         case TUNGETSNDBUF:
1953                 sndbuf = tfile->socket.sk->sk_sndbuf;
1954                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1955                         ret = -EFAULT;
1956                 break;
1957
1958         case TUNSETSNDBUF:
1959                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1960                         ret = -EFAULT;
1961                         break;
1962                 }
1963
1964                 tun->sndbuf = sndbuf;
1965                 tun_set_sndbuf(tun);
1966                 break;
1967
1968         case TUNGETVNETHDRSZ:
1969                 vnet_hdr_sz = tun->vnet_hdr_sz;
1970                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1971                         ret = -EFAULT;
1972                 break;
1973
1974         case TUNSETVNETHDRSZ:
1975                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1976                         ret = -EFAULT;
1977                         break;
1978                 }
1979                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1980                         ret = -EINVAL;
1981                         break;
1982                 }
1983
1984                 tun->vnet_hdr_sz = vnet_hdr_sz;
1985                 break;
1986
1987         case TUNATTACHFILTER:
1988                 /* Can be set only for TAPs */
1989                 ret = -EINVAL;
1990                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1991                         break;
1992                 ret = -EFAULT;
1993                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
1994                         break;
1995
1996                 ret = tun_attach_filter(tun);
1997                 break;
1998
1999         case TUNDETACHFILTER:
2000                 /* Can be set only for TAPs */
2001                 ret = -EINVAL;
2002                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2003                         break;
2004                 ret = 0;
2005                 tun_detach_filter(tun, tun->numqueues);
2006                 break;
2007
2008         default:
2009                 ret = -EINVAL;
2010                 break;
2011         }
2012
2013 unlock:
2014         rtnl_unlock();
2015         if (tun)
2016                 tun_put(tun);
2017         return ret;
2018 }
2019
2020 static long tun_chr_ioctl(struct file *file,
2021                           unsigned int cmd, unsigned long arg)
2022 {
2023         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2024 }
2025
2026 #ifdef CONFIG_COMPAT
2027 static long tun_chr_compat_ioctl(struct file *file,
2028                          unsigned int cmd, unsigned long arg)
2029 {
2030         switch (cmd) {
2031         case TUNSETIFF:
2032         case TUNGETIFF:
2033         case TUNSETTXFILTER:
2034         case TUNGETSNDBUF:
2035         case TUNSETSNDBUF:
2036         case SIOCGIFHWADDR:
2037         case SIOCSIFHWADDR:
2038                 arg = (unsigned long)compat_ptr(arg);
2039                 break;
2040         default:
2041                 arg = (compat_ulong_t)arg;
2042                 break;
2043         }
2044
2045         /*
2046          * compat_ifreq is shorter than ifreq, so we must not access beyond
2047          * the end of that structure. All fields that are used in this
2048          * driver are compatible though, we don't need to convert the
2049          * contents.
2050          */
2051         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2052 }
2053 #endif /* CONFIG_COMPAT */
2054
2055 static int tun_chr_fasync(int fd, struct file *file, int on)
2056 {
2057         struct tun_file *tfile = file->private_data;
2058         int ret;
2059
2060         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2061                 goto out;
2062
2063         if (on) {
2064                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2065                 if (ret)
2066                         goto out;
2067                 tfile->flags |= TUN_FASYNC;
2068         } else
2069                 tfile->flags &= ~TUN_FASYNC;
2070         ret = 0;
2071 out:
2072         return ret;
2073 }
2074
2075 static int tun_chr_open(struct inode *inode, struct file * file)
2076 {
2077         struct tun_file *tfile;
2078
2079         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2080
2081         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2082                                             &tun_proto);
2083         if (!tfile)
2084                 return -ENOMEM;
2085         rcu_assign_pointer(tfile->tun, NULL);
2086         tfile->net = get_net(current->nsproxy->net_ns);
2087         tfile->flags = 0;
2088
2089         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2090         init_waitqueue_head(&tfile->wq.wait);
2091
2092         tfile->socket.file = file;
2093         tfile->socket.ops = &tun_socket_ops;
2094
2095         sock_init_data(&tfile->socket, &tfile->sk);
2096         sk_change_net(&tfile->sk, tfile->net);
2097
2098         tfile->sk.sk_write_space = tun_sock_write_space;
2099         tfile->sk.sk_sndbuf = INT_MAX;
2100
2101         file->private_data = tfile;
2102         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2103
2104         return 0;
2105 }
2106
2107 static int tun_chr_close(struct inode *inode, struct file *file)
2108 {
2109         struct tun_file *tfile = file->private_data;
2110         struct net *net = tfile->net;
2111
2112         tun_detach(tfile, true);
2113         put_net(net);
2114
2115         return 0;
2116 }
2117
2118 static const struct file_operations tun_fops = {
2119         .owner  = THIS_MODULE,
2120         .llseek = no_llseek,
2121         .read  = do_sync_read,
2122         .aio_read  = tun_chr_aio_read,
2123         .write = do_sync_write,
2124         .aio_write = tun_chr_aio_write,
2125         .poll   = tun_chr_poll,
2126         .unlocked_ioctl = tun_chr_ioctl,
2127 #ifdef CONFIG_COMPAT
2128         .compat_ioctl = tun_chr_compat_ioctl,
2129 #endif
2130         .open   = tun_chr_open,
2131         .release = tun_chr_close,
2132         .fasync = tun_chr_fasync
2133 };
2134
2135 static struct miscdevice tun_miscdev = {
2136         .minor = TUN_MINOR,
2137         .name = "tun",
2138         .nodename = "net/tun",
2139         .fops = &tun_fops,
2140 };
2141
2142 /* ethtool interface */
2143
2144 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2145 {
2146         cmd->supported          = 0;
2147         cmd->advertising        = 0;
2148         ethtool_cmd_speed_set(cmd, SPEED_10);
2149         cmd->duplex             = DUPLEX_FULL;
2150         cmd->port               = PORT_TP;
2151         cmd->phy_address        = 0;
2152         cmd->transceiver        = XCVR_INTERNAL;
2153         cmd->autoneg            = AUTONEG_DISABLE;
2154         cmd->maxtxpkt           = 0;
2155         cmd->maxrxpkt           = 0;
2156         return 0;
2157 }
2158
2159 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2160 {
2161         struct tun_struct *tun = netdev_priv(dev);
2162
2163         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2164         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2165
2166         switch (tun->flags & TUN_TYPE_MASK) {
2167         case TUN_TUN_DEV:
2168                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2169                 break;
2170         case TUN_TAP_DEV:
2171                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2172                 break;
2173         }
2174 }
2175
2176 static u32 tun_get_msglevel(struct net_device *dev)
2177 {
2178 #ifdef TUN_DEBUG
2179         struct tun_struct *tun = netdev_priv(dev);
2180         return tun->debug;
2181 #else
2182         return -EOPNOTSUPP;
2183 #endif
2184 }
2185
2186 static void tun_set_msglevel(struct net_device *dev, u32 value)
2187 {
2188 #ifdef TUN_DEBUG
2189         struct tun_struct *tun = netdev_priv(dev);
2190         tun->debug = value;
2191 #endif
2192 }
2193
2194 static const struct ethtool_ops tun_ethtool_ops = {
2195         .get_settings   = tun_get_settings,
2196         .get_drvinfo    = tun_get_drvinfo,
2197         .get_msglevel   = tun_get_msglevel,
2198         .set_msglevel   = tun_set_msglevel,
2199         .get_link       = ethtool_op_get_link,
2200 };
2201
2202
2203 static int __init tun_init(void)
2204 {
2205         int ret = 0;
2206
2207         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2208         pr_info("%s\n", DRV_COPYRIGHT);
2209
2210         ret = rtnl_link_register(&tun_link_ops);
2211         if (ret) {
2212                 pr_err("Can't register link_ops\n");
2213                 goto err_linkops;
2214         }
2215
2216         ret = misc_register(&tun_miscdev);
2217         if (ret) {
2218                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2219                 goto err_misc;
2220         }
2221         return  0;
2222 err_misc:
2223         rtnl_link_unregister(&tun_link_ops);
2224 err_linkops:
2225         return ret;
2226 }
2227
2228 static void tun_cleanup(void)
2229 {
2230         misc_deregister(&tun_miscdev);
2231         rtnl_link_unregister(&tun_link_ops);
2232 }
2233
2234 /* Get an underlying socket object from tun file.  Returns error unless file is
2235  * attached to a device.  The returned object works like a packet socket, it
2236  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2237  * holding a reference to the file for as long as the socket is in use. */
2238 struct socket *tun_get_socket(struct file *file)
2239 {
2240         struct tun_file *tfile;
2241         if (file->f_op != &tun_fops)
2242                 return ERR_PTR(-EINVAL);
2243         tfile = file->private_data;
2244         if (!tfile)
2245                 return ERR_PTR(-EBADFD);
2246         return &tfile->socket;
2247 }
2248 EXPORT_SYMBOL_GPL(tun_get_socket);
2249
2250 module_init(tun_init);
2251 module_exit(tun_cleanup);
2252 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2253 MODULE_AUTHOR(DRV_COPYRIGHT);
2254 MODULE_LICENSE("GPL");
2255 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2256 MODULE_ALIAS("devname:net/tun");