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 /* DEFAULT_MAX_NUM_RSS_QUEUES were choosed to let the rx/tx queues allocated for
 113  * the netdevice to be fit in one page. So we can make sure the success of
 114  * memory allocation. TODO: increase the limit. */
 115 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
 116 #define MAX_TAP_FLOWS  4096
 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         struct list_head next;
 142         struct tun_struct *detached;
 143 };
 144
 145 struct tun_flow_entry {
 146         struct hlist_node hash_link;
 147         struct rcu_head rcu;
 148         struct tun_struct *tun;
 149
 150         u32 rxhash;
 151         int queue_index;
 152         unsigned long updated;
 153 };
 154
 155 #define TUN_NUM_FLOW_ENTRIES 1024
 156
 157 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 158  * device, socket filter, sndbuf and vnet header size were restore when the
 159  * file were attached to a persist device.
 160  */
 161 struct tun_struct {
 162         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 163         unsigned int            numqueues;
 164         unsigned int            flags;
 165         kuid_t                  owner;
 166         kgid_t                  group;
 167
 168         struct net_device       *dev;
 169         netdev_features_t       set_features;
 170 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 171                           NETIF_F_TSO6|NETIF_F_UFO)
 172
 173         int                     vnet_hdr_sz;
 174         int                     sndbuf;
 175         struct tap_filter       txflt;
 176         struct sock_fprog       fprog;
 177         /* protected by rtnl lock */
 178         bool                    filter_attached;
 179 #ifdef TUN_DEBUG
 180         int debug;
 181 #endif
 182         spinlock_t lock;
 183         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 184         struct timer_list flow_gc_timer;
 185         unsigned long ageing_time;
 186         unsigned int numdisabled;
 187         struct list_head disabled;
 188         void *security;
 189         u32 flow_count;
 190 };
 191
 192 static inline u32 tun_hashfn(u32 rxhash)
 193 {
 194         return rxhash & 0x3ff;
 195 }
 196
 197 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 198 {
 199         struct tun_flow_entry *e;
 200         struct hlist_node *n;
 201
 202         hlist_for_each_entry_rcu(e, n, head, hash_link) {
 203                 if (e->rxhash == rxhash)
 204                         return e;
 205         }
 206         return NULL;
 207 }
 208
 209 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 210                                               struct hlist_head *head,
 211                                               u32 rxhash, u16 queue_index)
 212 {
 213         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 214
 215         if (e) {
 216                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 217                           rxhash, queue_index);
 218                 e->updated = jiffies;
 219                 e->rxhash = rxhash;
 220                 e->queue_index = queue_index;
 221                 e->tun = tun;
 222                 hlist_add_head_rcu(&e->hash_link, head);
 223                 ++tun->flow_count;
 224         }
 225         return e;
 226 }
 227
 228 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 229 {
 230         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 231                   e->rxhash, e->queue_index);
 232         hlist_del_rcu(&e->hash_link);
 233         kfree_rcu(e, rcu);
 234         --tun->flow_count;
 235 }
 236
 237 static void tun_flow_flush(struct tun_struct *tun)
 238 {
 239         int i;
 240
 241         spin_lock_bh(&tun->lock);
 242         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 243                 struct tun_flow_entry *e;
 244                 struct hlist_node *h, *n;
 245
 246                 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link)
 247                         tun_flow_delete(tun, e);
 248         }
 249         spin_unlock_bh(&tun->lock);
 250 }
 251
 252 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 253 {
 254         int i;
 255
 256         spin_lock_bh(&tun->lock);
 257         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 258                 struct tun_flow_entry *e;
 259                 struct hlist_node *h, *n;
 260
 261                 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
 262                         if (e->queue_index == queue_index)
 263                                 tun_flow_delete(tun, e);
 264                 }
 265         }
 266         spin_unlock_bh(&tun->lock);
 267 }
 268
 269 static void tun_flow_cleanup(unsigned long data)
 270 {
 271         struct tun_struct *tun = (struct tun_struct *)data;
 272         unsigned long delay = tun->ageing_time;
 273         unsigned long next_timer = jiffies + delay;
 274         unsigned long count = 0;
 275         int i;
 276
 277         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 278
 279         spin_lock_bh(&tun->lock);
 280         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 281                 struct tun_flow_entry *e;
 282                 struct hlist_node *h, *n;
 283
 284                 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
 285                         unsigned long this_timer;
 286                         count++;
 287                         this_timer = e->updated + delay;
 288                         if (time_before_eq(this_timer, jiffies))
 289                                 tun_flow_delete(tun, e);
 290                         else if (time_before(this_timer, next_timer))
 291                                 next_timer = this_timer;
 292                 }
 293         }
 294
 295         if (count)
 296                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 297         spin_unlock_bh(&tun->lock);
 298 }
 299
 300 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 301                             u16 queue_index)
 302 {
 303         struct hlist_head *head;
 304         struct tun_flow_entry *e;
 305         unsigned long delay = tun->ageing_time;
 306
 307         if (!rxhash)
 308                 return;
 309         else
 310                 head = &tun->flows[tun_hashfn(rxhash)];
 311
 312         rcu_read_lock();
 313
 314         if (tun->numqueues == 1)
 315                 goto unlock;
 316
 317         e = tun_flow_find(head, rxhash);
 318         if (likely(e)) {
 319                 /* TODO: keep queueing to old queue until it's empty? */
 320                 e->queue_index = queue_index;
 321                 e->updated = jiffies;
 322         } else {
 323                 spin_lock_bh(&tun->lock);
 324                 if (!tun_flow_find(head, rxhash) &&
 325                     tun->flow_count < MAX_TAP_FLOWS)
 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_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 390 {
 391         tfile->detached = tun;
 392         list_add_tail(&tfile->next, &tun->disabled);
 393         ++tun->numdisabled;
 394 }
 395
 396 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 397 {
 398         struct tun_struct *tun = tfile->detached;
 399
 400         tfile->detached = NULL;
 401         list_del_init(&tfile->next);
 402         --tun->numdisabled;
 403         return tun;
 404 }
 405
 406 static void __tun_detach(struct tun_file *tfile, bool clean)
 407 {
 408         struct tun_file *ntfile;
 409         struct tun_struct *tun;
 410         struct net_device *dev;
 411
 412         tun = rtnl_dereference(tfile->tun);
 413
 414         if (tun) {
 415                 u16 index = tfile->queue_index;
 416                 BUG_ON(index >= tun->numqueues);
 417                 dev = tun->dev;
 418
 419                 rcu_assign_pointer(tun->tfiles[index],
 420                                    tun->tfiles[tun->numqueues - 1]);
 421                 rcu_assign_pointer(tfile->tun, NULL);
 422                 ntfile = rtnl_dereference(tun->tfiles[index]);
 423                 ntfile->queue_index = index;
 424
 425                 --tun->numqueues;
 426                 if (clean)
 427                         sock_put(&tfile->sk);
 428                 else
 429                         tun_disable_queue(tun, tfile);
 430
 431                 synchronize_net();
 432                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 433                 /* Drop read queue */
 434                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 435                 tun_set_real_num_queues(tun);
 436         } else if (tfile->detached && clean) {
 437                 tun = tun_enable_queue(tfile);
 438                 sock_put(&tfile->sk);
 439         }
 440
 441         if (clean) {
 442                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0 &&
 443                     !(tun->flags & TUN_PERSIST))
 444                         if (tun->dev->reg_state == NETREG_REGISTERED)
 445                                 unregister_netdevice(tun->dev);
 446
 447                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 448                                  &tfile->socket.flags));
 449                 sk_release_kernel(&tfile->sk);
 450         }
 451 }
 452
 453 static void tun_detach(struct tun_file *tfile, bool clean)
 454 {
 455         rtnl_lock();
 456         __tun_detach(tfile, clean);
 457         rtnl_unlock();
 458 }
 459
 460 static void tun_detach_all(struct net_device *dev)
 461 {
 462         struct tun_struct *tun = netdev_priv(dev);
 463         struct tun_file *tfile, *tmp;
 464         int i, n = tun->numqueues;
 465
 466         for (i = 0; i < n; i++) {
 467                 tfile = rtnl_dereference(tun->tfiles[i]);
 468                 BUG_ON(!tfile);
 469                 wake_up_all(&tfile->wq.wait);
 470                 rcu_assign_pointer(tfile->tun, NULL);
 471                 --tun->numqueues;
 472         }
 473         BUG_ON(tun->numqueues != 0);
 474
 475         synchronize_net();
 476         for (i = 0; i < n; i++) {
 477                 tfile = rtnl_dereference(tun->tfiles[i]);
 478                 /* Drop read queue */
 479                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 480                 sock_put(&tfile->sk);
 481         }
 482         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 483                 tun_enable_queue(tfile);
 484                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 485                 sock_put(&tfile->sk);
 486         }
 487         BUG_ON(tun->numdisabled != 0);
 488
 489         if (tun->flags & TUN_PERSIST)
 490                 module_put(THIS_MODULE);
 491 }
 492
 493 static int tun_attach(struct tun_struct *tun, struct file *file)
 494 {
 495         struct tun_file *tfile = file->private_data;
 496         int err;
 497
 498         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 499         if (err < 0)
 500                 goto out;
 501
 502         err = -EINVAL;
 503         if (rtnl_dereference(tfile->tun))
 504                 goto out;
 505
 506         err = -EBUSY;
 507         if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 508                 goto out;
 509
 510         err = -E2BIG;
 511         if (!tfile->detached &&
 512             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 513                 goto out;
 514
 515         err = 0;
 516
 517         /* Re-attach the filter to presist device */
 518         if (tun->filter_attached == true) {
 519                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 520                 if (!err)
 521                         goto out;
 522         }
 523         tfile->queue_index = tun->numqueues;
 524         rcu_assign_pointer(tfile->tun, tun);
 525         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 526         tun->numqueues++;
 527
 528         if (tfile->detached)
 529                 tun_enable_queue(tfile);
 530         else
 531                 sock_hold(&tfile->sk);
 532
 533         tun_set_real_num_queues(tun);
 534
 535         /* device is allowed to go away first, so no need to hold extra
 536          * refcnt.
 537          */
 538
 539 out:
 540         return err;
 541 }
 542
 543 static struct tun_struct *__tun_get(struct tun_file *tfile)
 544 {
 545         struct tun_struct *tun;
 546
 547         rcu_read_lock();
 548         tun = rcu_dereference(tfile->tun);
 549         if (tun)
 550                 dev_hold(tun->dev);
 551         rcu_read_unlock();
 552
 553         return tun;
 554 }
 555
 556 static struct tun_struct *tun_get(struct file *file)
 557 {
 558         return __tun_get(file->private_data);
 559 }
 560
 561 static void tun_put(struct tun_struct *tun)
 562 {
 563         dev_put(tun->dev);
 564 }
 565
 566 /* TAP filtering */
 567 static void addr_hash_set(u32 *mask, const u8 *addr)
 568 {
 569         int n = ether_crc(ETH_ALEN, addr) >> 26;
 570         mask[n >> 5] |= (1 << (n & 31));
 571 }
 572
 573 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 574 {
 575         int n = ether_crc(ETH_ALEN, addr) >> 26;
 576         return mask[n >> 5] & (1 << (n & 31));
 577 }
 578
 579 static int update_filter(struct tap_filter *filter, void __user *arg)
 580 {
 581         struct { u8 u[ETH_ALEN]; } *addr;
 582         struct tun_filter uf;
 583         int err, alen, n, nexact;
 584
 585         if (copy_from_user(&uf, arg, sizeof(uf)))
 586                 return -EFAULT;
 587
 588         if (!uf.count) {
 589                 /* Disabled */
 590                 filter->count = 0;
 591                 return 0;
 592         }
 593
 594         alen = ETH_ALEN * uf.count;
 595         addr = kmalloc(alen, GFP_KERNEL);
 596         if (!addr)
 597                 return -ENOMEM;
 598
 599         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 600                 err = -EFAULT;
 601                 goto done;
 602         }
 603
 604         /* The filter is updated without holding any locks. Which is
 605          * perfectly safe. We disable it first and in the worst
 606          * case we'll accept a few undesired packets. */
 607         filter->count = 0;
 608         wmb();
 609
 610         /* Use first set of addresses as an exact filter */
 611         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 612                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 613
 614         nexact = n;
 615
 616         /* Remaining multicast addresses are hashed,
 617          * unicast will leave the filter disabled. */
 618         memset(filter->mask, 0, sizeof(filter->mask));
 619         for (; n < uf.count; n++) {
 620                 if (!is_multicast_ether_addr(addr[n].u)) {
 621                         err = 0; /* no filter */
 622                         goto done;
 623                 }
 624                 addr_hash_set(filter->mask, addr[n].u);
 625         }
 626
 627         /* For ALLMULTI just set the mask to all ones.
 628          * This overrides the mask populated above. */
 629         if ((uf.flags & TUN_FLT_ALLMULTI))
 630                 memset(filter->mask, ~0, sizeof(filter->mask));
 631
 632         /* Now enable the filter */
 633         wmb();
 634         filter->count = nexact;
 635
 636         /* Return the number of exact filters */
 637         err = nexact;
 638
 639 done:
 640         kfree(addr);
 641         return err;
 642 }
 643
 644 /* Returns: 0 - drop, !=0 - accept */
 645 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 646 {
 647         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 648          * at this point. */
 649         struct ethhdr *eh = (struct ethhdr *) skb->data;
 650         int i;
 651
 652         /* Exact match */
 653         for (i = 0; i < filter->count; i++)
 654                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 655                         return 1;
 656
 657         /* Inexact match (multicast only) */
 658         if (is_multicast_ether_addr(eh->h_dest))
 659                 return addr_hash_test(filter->mask, eh->h_dest);
 660
 661         return 0;
 662 }
 663
 664 /*
 665  * Checks whether the packet is accepted or not.
 666  * Returns: 0 - drop, !=0 - accept
 667  */
 668 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 669 {
 670         if (!filter->count)
 671                 return 1;
 672
 673         return run_filter(filter, skb);
 674 }
 675
 676 /* Network device part of the driver */
 677
 678 static const struct ethtool_ops tun_ethtool_ops;
 679
 680 /* Net device detach from fd. */
 681 static void tun_net_uninit(struct net_device *dev)
 682 {
 683         tun_detach_all(dev);
 684 }
 685
 686 /* Net device open. */
 687 static int tun_net_open(struct net_device *dev)
 688 {
 689         netif_tx_start_all_queues(dev);
 690         return 0;
 691 }
 692
 693 /* Net device close. */
 694 static int tun_net_close(struct net_device *dev)
 695 {
 696         netif_tx_stop_all_queues(dev);
 697         return 0;
 698 }
 699
 700 /* Net device start xmit */
 701 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 702 {
 703         struct tun_struct *tun = netdev_priv(dev);
 704         int txq = skb->queue_mapping;
 705         struct tun_file *tfile;
 706
 707         rcu_read_lock();
 708         tfile = rcu_dereference(tun->tfiles[txq]);
 709
 710         /* Drop packet if interface is not attached */
 711         if (txq >= tun->numqueues)
 712                 goto drop;
 713
 714         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 715
 716         BUG_ON(!tfile);
 717
 718         /* Drop if the filter does not like it.
 719          * This is a noop if the filter is disabled.
 720          * Filter can be enabled only for the TAP devices. */
 721         if (!check_filter(&tun->txflt, skb))
 722                 goto drop;
 723
 724         if (tfile->socket.sk->sk_filter &&
 725             sk_filter(tfile->socket.sk, skb))
 726                 goto drop;
 727
 728         /* Limit the number of packets queued by dividing txq length with the
 729          * number of queues.
 730          */
 731         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
 732                           >= dev->tx_queue_len / tun->numqueues)
 733                 goto drop;
 734
 735         /* Orphan the skb - required as we might hang on to it
 736          * for indefinite time. */
 737         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 738                 goto drop;
 739         skb_orphan(skb);
 740
 741         /* Enqueue packet */
 742         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 743
 744         /* Notify and wake up reader process */
 745         if (tfile->flags & TUN_FASYNC)
 746                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 747         wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
 748                                    POLLRDNORM | POLLRDBAND);
 749
 750         rcu_read_unlock();
 751         return NETDEV_TX_OK;
 752
 753 drop:
 754         dev->stats.tx_dropped++;
 755         skb_tx_error(skb);
 756         kfree_skb(skb);
 757         rcu_read_unlock();
 758         return NETDEV_TX_OK;
 759 }
 760
 761 static void tun_net_mclist(struct net_device *dev)
 762 {
 763         /*
 764          * This callback is supposed to deal with mc filter in
 765          * _rx_ path and has nothing to do with the _tx_ path.
 766          * In rx path we always accept everything userspace gives us.
 767          */
 768 }
 769
 770 #define MIN_MTU 68
 771 #define MAX_MTU 65535
 772
 773 static int
 774 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 775 {
 776         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 777                 return -EINVAL;
 778         dev->mtu = new_mtu;
 779         return 0;
 780 }
 781
 782 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 783         netdev_features_t features)
 784 {
 785         struct tun_struct *tun = netdev_priv(dev);
 786
 787         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 788 }
 789 #ifdef CONFIG_NET_POLL_CONTROLLER
 790 static void tun_poll_controller(struct net_device *dev)
 791 {
 792         /*
 793          * Tun only receives frames when:
 794          * 1) the char device endpoint gets data from user space
 795          * 2) the tun socket gets a sendmsg call from user space
 796          * Since both of those are syncronous operations, we are guaranteed
 797          * never to have pending data when we poll for it
 798          * so theres nothing to do here but return.
 799          * We need this though so netpoll recognizes us as an interface that
 800          * supports polling, which enables bridge devices in virt setups to
 801          * still use netconsole
 802          */
 803         return;
 804 }
 805 #endif
 806 static const struct net_device_ops tun_netdev_ops = {
 807         .ndo_uninit             = tun_net_uninit,
 808         .ndo_open               = tun_net_open,
 809         .ndo_stop               = tun_net_close,
 810         .ndo_start_xmit         = tun_net_xmit,
 811         .ndo_change_mtu         = tun_net_change_mtu,
 812         .ndo_fix_features       = tun_net_fix_features,
 813         .ndo_select_queue       = tun_select_queue,
 814 #ifdef CONFIG_NET_POLL_CONTROLLER
 815         .ndo_poll_controller    = tun_poll_controller,
 816 #endif
 817 };
 818
 819 static const struct net_device_ops tap_netdev_ops = {
 820         .ndo_uninit             = tun_net_uninit,
 821         .ndo_open               = tun_net_open,
 822         .ndo_stop               = tun_net_close,
 823         .ndo_start_xmit         = tun_net_xmit,
 824         .ndo_change_mtu         = tun_net_change_mtu,
 825         .ndo_fix_features       = tun_net_fix_features,
 826         .ndo_set_rx_mode        = tun_net_mclist,
 827         .ndo_set_mac_address    = eth_mac_addr,
 828         .ndo_validate_addr      = eth_validate_addr,
 829         .ndo_select_queue       = tun_select_queue,
 830 #ifdef CONFIG_NET_POLL_CONTROLLER
 831         .ndo_poll_controller    = tun_poll_controller,
 832 #endif
 833 };
 834
 835 static int tun_flow_init(struct tun_struct *tun)
 836 {
 837         int i;
 838
 839         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 840                 INIT_HLIST_HEAD(&tun->flows[i]);
 841
 842         tun->ageing_time = TUN_FLOW_EXPIRE;
 843         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 844         mod_timer(&tun->flow_gc_timer,
 845                   round_jiffies_up(jiffies + tun->ageing_time));
 846
 847         return 0;
 848 }
 849
 850 static void tun_flow_uninit(struct tun_struct *tun)
 851 {
 852         del_timer_sync(&tun->flow_gc_timer);
 853         tun_flow_flush(tun);
 854 }
 855
 856 /* Initialize net device. */
 857 static void tun_net_init(struct net_device *dev)
 858 {
 859         struct tun_struct *tun = netdev_priv(dev);
 860
 861         switch (tun->flags & TUN_TYPE_MASK) {
 862         case TUN_TUN_DEV:
 863                 dev->netdev_ops = &tun_netdev_ops;
 864
 865                 /* Point-to-Point TUN Device */
 866                 dev->hard_header_len = 0;
 867                 dev->addr_len = 0;
 868                 dev->mtu = 1500;
 869
 870                 /* Zero header length */
 871                 dev->type = ARPHRD_NONE;
 872                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 873                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 874                 break;
 875
 876         case TUN_TAP_DEV:
 877                 dev->netdev_ops = &tap_netdev_ops;
 878                 /* Ethernet TAP Device */
 879                 ether_setup(dev);
 880                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 881                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 882
 883                 eth_hw_addr_random(dev);
 884
 885                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 886                 break;
 887         }
 888 }
 889
 890 /* Character device part */
 891
 892 /* Poll */
 893 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 894 {
 895         struct tun_file *tfile = file->private_data;
 896         struct tun_struct *tun = __tun_get(tfile);
 897         struct sock *sk;
 898         unsigned int mask = 0;
 899
 900         if (!tun)
 901                 return POLLERR;
 902
 903         sk = tfile->socket.sk;
 904
 905         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 906
 907         poll_wait(file, &tfile->wq.wait, wait);
 908
 909         if (!skb_queue_empty(&sk->sk_receive_queue))
 910                 mask |= POLLIN | POLLRDNORM;
 911
 912         if (sock_writeable(sk) ||
 913             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 914              sock_writeable(sk)))
 915                 mask |= POLLOUT | POLLWRNORM;
 916
 917         if (tun->dev->reg_state != NETREG_REGISTERED)
 918                 mask = POLLERR;
 919
 920         tun_put(tun);
 921         return mask;
 922 }
 923
 924 /* prepad is the amount to reserve at front.  len is length after that.
 925  * linear is a hint as to how much to copy (usually headers). */
 926 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 927                                      size_t prepad, size_t len,
 928                                      size_t linear, int noblock)
 929 {
 930         struct sock *sk = tfile->socket.sk;
 931         struct sk_buff *skb;
 932         int err;
 933
 934         /* Under a page?  Don't bother with paged skb. */
 935         if (prepad + len < PAGE_SIZE || !linear)
 936                 linear = len;
 937
 938         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 939                                    &err);
 940         if (!skb)
 941                 return ERR_PTR(err);
 942
 943         skb_reserve(skb, prepad);
 944         skb_put(skb, linear);
 945         skb->data_len = len - linear;
 946         skb->len += len - linear;
 947
 948         return skb;
 949 }
 950
 951 /* set skb frags from iovec, this can move to core network code for reuse */
 952 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 953                                   int offset, size_t count)
 954 {
 955         int len = iov_length(from, count) - offset;
 956         int copy = skb_headlen(skb);
 957         int size, offset1 = 0;
 958         int i = 0;
 959
 960         /* Skip over from offset */
 961         while (count && (offset >= from->iov_len)) {
 962                 offset -= from->iov_len;
 963                 ++from;
 964                 --count;
 965         }
 966
 967         /* copy up to skb headlen */
 968         while (count && (copy > 0)) {
 969                 size = min_t(unsigned int, copy, from->iov_len - offset);
 970                 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
 971                                    size))
 972                         return -EFAULT;
 973                 if (copy > size) {
 974                         ++from;
 975                         --count;
 976                         offset = 0;
 977                 } else
 978                         offset += size;
 979                 copy -= size;
 980                 offset1 += size;
 981         }
 982
 983         if (len == offset1)
 984                 return 0;
 985
 986         while (count--) {
 987                 struct page *page[MAX_SKB_FRAGS];
 988                 int num_pages;
 989                 unsigned long base;
 990                 unsigned long truesize;
 991
 992                 len = from->iov_len - offset;
 993                 if (!len) {
 994                         offset = 0;
 995                         ++from;
 996                         continue;
 997                 }
 998                 base = (unsigned long)from->iov_base + offset;
 999                 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
1000                 if (i + size > MAX_SKB_FRAGS)
1001                         return -EMSGSIZE;
1002                 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
1003                 if (num_pages != size) {
1004                         for (i = 0; i < num_pages; i++)
1005                                 put_page(page[i]);
1006                         return -EFAULT;
1007                 }
1008                 truesize = size * PAGE_SIZE;
1009                 skb->data_len += len;
1010                 skb->len += len;
1011                 skb->truesize += truesize;
1012                 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
1013                 while (len) {
1014                         int off = base & ~PAGE_MASK;
1015                         int size = min_t(int, len, PAGE_SIZE - off);
1016                         __skb_fill_page_desc(skb, i, page[i], off, size);
1017                         skb_shinfo(skb)->nr_frags++;
1018                         /* increase sk_wmem_alloc */
1019                         base += size;
1020                         len -= size;
1021                         i++;
1022                 }
1023                 offset = 0;
1024                 ++from;
1025         }
1026         return 0;
1027 }
1028
1029 /* Get packet from user space buffer */
1030 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1031                             void *msg_control, const struct iovec *iv,
1032                             size_t total_len, size_t count, int noblock)
1033 {
1034         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1035         struct sk_buff *skb;
1036         size_t len = total_len, align = NET_SKB_PAD;
1037         struct virtio_net_hdr gso = { 0 };
1038         int offset = 0;
1039         int copylen;
1040         bool zerocopy = false;
1041         int err;
1042         u32 rxhash;
1043
1044         if (!(tun->flags & TUN_NO_PI)) {
1045                 if ((len -= sizeof(pi)) > total_len)
1046                         return -EINVAL;
1047
1048                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1049                         return -EFAULT;
1050                 offset += sizeof(pi);
1051         }
1052
1053         if (tun->flags & TUN_VNET_HDR) {
1054                 if ((len -= tun->vnet_hdr_sz) > total_len)
1055                         return -EINVAL;
1056
1057                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1058                         return -EFAULT;
1059
1060                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1061                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1062                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1063
1064                 if (gso.hdr_len > len)
1065                         return -EINVAL;
1066                 offset += tun->vnet_hdr_sz;
1067         }
1068
1069         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1070                 align += NET_IP_ALIGN;
1071                 if (unlikely(len < ETH_HLEN ||
1072                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1073                         return -EINVAL;
1074         }
1075
1076         if (msg_control)
1077                 zerocopy = true;
1078
1079         if (zerocopy) {
1080                 /* Userspace may produce vectors with count greater than
1081                  * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1082                  * to let the rest of data to be fit in the frags.
1083                  */
1084                 if (count > MAX_SKB_FRAGS) {
1085                         copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1086                         if (copylen < offset)
1087                                 copylen = 0;
1088                         else
1089                                 copylen -= offset;
1090                 } else
1091                                 copylen = 0;
1092                 /* There are 256 bytes to be copied in skb, so there is enough
1093                  * room for skb expand head in case it is used.
1094                  * The rest of the buffer is mapped from userspace.
1095                  */
1096                 if (copylen < gso.hdr_len)
1097                         copylen = gso.hdr_len;
1098                 if (!copylen)
1099                         copylen = GOODCOPY_LEN;
1100         } else
1101                 copylen = len;
1102
1103         skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
1104         if (IS_ERR(skb)) {
1105                 if (PTR_ERR(skb) != -EAGAIN)
1106                         tun->dev->stats.rx_dropped++;
1107                 return PTR_ERR(skb);
1108         }
1109
1110         if (zerocopy)
1111                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1112         else
1113                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1114
1115         if (err) {
1116                 tun->dev->stats.rx_dropped++;
1117                 kfree_skb(skb);
1118                 return -EFAULT;
1119         }
1120
1121         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1122                 if (!skb_partial_csum_set(skb, gso.csum_start,
1123                                           gso.csum_offset)) {
1124                         tun->dev->stats.rx_frame_errors++;
1125                         kfree_skb(skb);
1126                         return -EINVAL;
1127                 }
1128         }
1129
1130         switch (tun->flags & TUN_TYPE_MASK) {
1131         case TUN_TUN_DEV:
1132                 if (tun->flags & TUN_NO_PI) {
1133                         switch (skb->data[0] & 0xf0) {
1134                         case 0x40:
1135                                 pi.proto = htons(ETH_P_IP);
1136                                 break;
1137                         case 0x60:
1138                                 pi.proto = htons(ETH_P_IPV6);
1139                                 break;
1140                         default:
1141                                 tun->dev->stats.rx_dropped++;
1142                                 kfree_skb(skb);
1143                                 return -EINVAL;
1144                         }
1145                 }
1146
1147                 skb_reset_mac_header(skb);
1148                 skb->protocol = pi.proto;
1149                 skb->dev = tun->dev;
1150                 break;
1151         case TUN_TAP_DEV:
1152                 skb->protocol = eth_type_trans(skb, tun->dev);
1153                 break;
1154         }
1155
1156         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1157                 pr_debug("GSO!\n");
1158                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1159                 case VIRTIO_NET_HDR_GSO_TCPV4:
1160                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1161                         break;
1162                 case VIRTIO_NET_HDR_GSO_TCPV6:
1163                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1164                         break;
1165                 case VIRTIO_NET_HDR_GSO_UDP:
1166                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1167                         break;
1168                 default:
1169                         tun->dev->stats.rx_frame_errors++;
1170                         kfree_skb(skb);
1171                         return -EINVAL;
1172                 }
1173
1174                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1175                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1176
1177                 skb_shinfo(skb)->gso_size = gso.gso_size;
1178                 if (skb_shinfo(skb)->gso_size == 0) {
1179                         tun->dev->stats.rx_frame_errors++;
1180                         kfree_skb(skb);
1181                         return -EINVAL;
1182                 }
1183
1184                 /* Header must be checked, and gso_segs computed. */
1185                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1186                 skb_shinfo(skb)->gso_segs = 0;
1187         }
1188
1189         /* copy skb_ubuf_info for callback when skb has no error */
1190         if (zerocopy) {
1191                 skb_shinfo(skb)->destructor_arg = msg_control;
1192                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1193         }
1194
1195         skb_reset_network_header(skb);
1196         rxhash = skb_get_rxhash(skb);
1197         netif_rx_ni(skb);
1198
1199         tun->dev->stats.rx_packets++;
1200         tun->dev->stats.rx_bytes += len;
1201
1202         tun_flow_update(tun, rxhash, tfile->queue_index);
1203         return total_len;
1204 }
1205
1206 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1207                               unsigned long count, loff_t pos)
1208 {
1209         struct file *file = iocb->ki_filp;
1210         struct tun_struct *tun = tun_get(file);
1211         struct tun_file *tfile = file->private_data;
1212         ssize_t result;
1213
1214         if (!tun)
1215                 return -EBADFD;
1216
1217         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1218
1219         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1220                               count, file->f_flags & O_NONBLOCK);
1221
1222         tun_put(tun);
1223         return result;
1224 }
1225
1226 /* Put packet to the user space buffer */
1227 static ssize_t tun_put_user(struct tun_struct *tun,
1228                             struct tun_file *tfile,
1229                             struct sk_buff *skb,
1230                             const struct iovec *iv, int len)
1231 {
1232         struct tun_pi pi = { 0, skb->protocol };
1233         ssize_t total = 0;
1234
1235         if (!(tun->flags & TUN_NO_PI)) {
1236                 if ((len -= sizeof(pi)) < 0)
1237                         return -EINVAL;
1238
1239                 if (len < skb->len) {
1240                         /* Packet will be striped */
1241                         pi.flags |= TUN_PKT_STRIP;
1242                 }
1243
1244                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1245                         return -EFAULT;
1246                 total += sizeof(pi);
1247         }
1248
1249         if (tun->flags & TUN_VNET_HDR) {
1250                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1251                 if ((len -= tun->vnet_hdr_sz) < 0)
1252                         return -EINVAL;
1253
1254                 if (skb_is_gso(skb)) {
1255                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1256
1257                         /* This is a hint as to how much should be linear. */
1258                         gso.hdr_len = skb_headlen(skb);
1259                         gso.gso_size = sinfo->gso_size;
1260                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1261                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1262                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1263                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1264                         else if (sinfo->gso_type & SKB_GSO_UDP)
1265                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1266                         else {
1267                                 pr_err("unexpected GSO type: "
1268                                        "0x%x, gso_size %d, hdr_len %d\n",
1269                                        sinfo->gso_type, gso.gso_size,
1270                                        gso.hdr_len);
1271                                 print_hex_dump(KERN_ERR, "tun: ",
1272                                                DUMP_PREFIX_NONE,
1273                                                16, 1, skb->head,
1274                                                min((int)gso.hdr_len, 64), true);
1275                                 WARN_ON_ONCE(1);
1276                                 return -EINVAL;
1277                         }
1278                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1279                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1280                 } else
1281                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1282
1283                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1284                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1285                         gso.csum_start = skb_checksum_start_offset(skb);
1286                         gso.csum_offset = skb->csum_offset;
1287                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1288                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1289                 } /* else everything is zero */
1290
1291                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1292                                                sizeof(gso))))
1293                         return -EFAULT;
1294                 total += tun->vnet_hdr_sz;
1295         }
1296
1297         len = min_t(int, skb->len, len);
1298
1299         skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1300         total += skb->len;
1301
1302         tun->dev->stats.tx_packets++;
1303         tun->dev->stats.tx_bytes += len;
1304
1305         return total;
1306 }
1307
1308 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1309                            struct kiocb *iocb, const struct iovec *iv,
1310                            ssize_t len, int noblock)
1311 {
1312         DECLARE_WAITQUEUE(wait, current);
1313         struct sk_buff *skb;
1314         ssize_t ret = 0;
1315
1316         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1317
1318         if (unlikely(!noblock))
1319                 add_wait_queue(&tfile->wq.wait, &wait);
1320         while (len) {
1321                 current->state = TASK_INTERRUPTIBLE;
1322
1323                 /* Read frames from the queue */
1324                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1325                         if (noblock) {
1326                                 ret = -EAGAIN;
1327                                 break;
1328                         }
1329                         if (signal_pending(current)) {
1330                                 ret = -ERESTARTSYS;
1331                                 break;
1332                         }
1333                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1334                                 ret = -EIO;
1335                                 break;
1336                         }
1337
1338                         /* Nothing to read, let's sleep */
1339                         schedule();
1340                         continue;
1341                 }
1342
1343                 ret = tun_put_user(tun, tfile, skb, iv, len);
1344                 kfree_skb(skb);
1345                 break;
1346         }
1347
1348         current->state = TASK_RUNNING;
1349         if (unlikely(!noblock))
1350                 remove_wait_queue(&tfile->wq.wait, &wait);
1351
1352         return ret;
1353 }
1354
1355 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1356                             unsigned long count, loff_t pos)
1357 {
1358         struct file *file = iocb->ki_filp;
1359         struct tun_file *tfile = file->private_data;
1360         struct tun_struct *tun = __tun_get(tfile);
1361         ssize_t len, ret;
1362
1363         if (!tun)
1364                 return -EBADFD;
1365         len = iov_length(iv, count);
1366         if (len < 0) {
1367                 ret = -EINVAL;
1368                 goto out;
1369         }
1370
1371         ret = tun_do_read(tun, tfile, iocb, iv, len,
1372                           file->f_flags & O_NONBLOCK);
1373         ret = min_t(ssize_t, ret, len);
1374 out:
1375         tun_put(tun);
1376         return ret;
1377 }
1378
1379 static void tun_free_netdev(struct net_device *dev)
1380 {
1381         struct tun_struct *tun = netdev_priv(dev);
1382
1383         BUG_ON(!(list_empty(&tun->disabled)));
1384         tun_flow_uninit(tun);
1385         security_tun_dev_free_security(tun->security);
1386         free_netdev(dev);
1387 }
1388
1389 static void tun_setup(struct net_device *dev)
1390 {
1391         struct tun_struct *tun = netdev_priv(dev);
1392
1393         tun->owner = INVALID_UID;
1394         tun->group = INVALID_GID;
1395
1396         dev->ethtool_ops = &tun_ethtool_ops;
1397         dev->destructor = tun_free_netdev;
1398 }
1399
1400 /* Trivial set of netlink ops to allow deleting tun or tap
1401  * device with netlink.
1402  */
1403 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1404 {
1405         return -EINVAL;
1406 }
1407
1408 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1409         .kind           = DRV_NAME,
1410         .priv_size      = sizeof(struct tun_struct),
1411         .setup          = tun_setup,
1412         .validate       = tun_validate,
1413 };
1414
1415 static void tun_sock_write_space(struct sock *sk)
1416 {
1417         struct tun_file *tfile;
1418         wait_queue_head_t *wqueue;
1419
1420         if (!sock_writeable(sk))
1421                 return;
1422
1423         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1424                 return;
1425
1426         wqueue = sk_sleep(sk);
1427         if (wqueue && waitqueue_active(wqueue))
1428                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1429                                                 POLLWRNORM | POLLWRBAND);
1430
1431         tfile = container_of(sk, struct tun_file, sk);
1432         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1433 }
1434
1435 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1436                        struct msghdr *m, size_t total_len)
1437 {
1438         int ret;
1439         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1440         struct tun_struct *tun = __tun_get(tfile);
1441
1442         if (!tun)
1443                 return -EBADFD;
1444         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1445                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1446         tun_put(tun);
1447         return ret;
1448 }
1449
1450
1451 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1452                        struct msghdr *m, size_t total_len,
1453                        int flags)
1454 {
1455         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1456         struct tun_struct *tun = __tun_get(tfile);
1457         int ret;
1458
1459         if (!tun)
1460                 return -EBADFD;
1461
1462         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1463                 return -EINVAL;
1464         ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1465                           flags & MSG_DONTWAIT);
1466         if (ret > total_len) {
1467                 m->msg_flags |= MSG_TRUNC;
1468                 ret = flags & MSG_TRUNC ? ret : total_len;
1469         }
1470         tun_put(tun);
1471         return ret;
1472 }
1473
1474 static int tun_release(struct socket *sock)
1475 {
1476         if (sock->sk)
1477                 sock_put(sock->sk);
1478         return 0;
1479 }
1480
1481 /* Ops structure to mimic raw sockets with tun */
1482 static const struct proto_ops tun_socket_ops = {
1483         .sendmsg = tun_sendmsg,
1484         .recvmsg = tun_recvmsg,
1485         .release = tun_release,
1486 };
1487
1488 static struct proto tun_proto = {
1489         .name           = "tun",
1490         .owner          = THIS_MODULE,
1491         .obj_size       = sizeof(struct tun_file),
1492 };
1493
1494 static int tun_flags(struct tun_struct *tun)
1495 {
1496         int flags = 0;
1497
1498         if (tun->flags & TUN_TUN_DEV)
1499                 flags |= IFF_TUN;
1500         else
1501                 flags |= IFF_TAP;
1502
1503         if (tun->flags & TUN_NO_PI)
1504                 flags |= IFF_NO_PI;
1505
1506         /* This flag has no real effect.  We track the value for backwards
1507          * compatibility.
1508          */
1509         if (tun->flags & TUN_ONE_QUEUE)
1510                 flags |= IFF_ONE_QUEUE;
1511
1512         if (tun->flags & TUN_VNET_HDR)
1513                 flags |= IFF_VNET_HDR;
1514
1515         if (tun->flags & TUN_TAP_MQ)
1516                 flags |= IFF_MULTI_QUEUE;
1517
1518         return flags;
1519 }
1520
1521 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1522                               char *buf)
1523 {
1524         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1525         return sprintf(buf, "0x%x\n", tun_flags(tun));
1526 }
1527
1528 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1529                               char *buf)
1530 {
1531         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1532         return uid_valid(tun->owner)?
1533                 sprintf(buf, "%u\n",
1534                         from_kuid_munged(current_user_ns(), tun->owner)):
1535                 sprintf(buf, "-1\n");
1536 }
1537
1538 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1539                               char *buf)
1540 {
1541         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1542         return gid_valid(tun->group) ?
1543                 sprintf(buf, "%u\n",
1544                         from_kgid_munged(current_user_ns(), tun->group)):
1545                 sprintf(buf, "-1\n");
1546 }
1547
1548 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1549 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1550 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1551
1552 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1553 {
1554         struct tun_struct *tun;
1555         struct tun_file *tfile = file->private_data;
1556         struct net_device *dev;
1557         int err;
1558
1559         if (tfile->detached)
1560                 return -EINVAL;
1561
1562         dev = __dev_get_by_name(net, ifr->ifr_name);
1563         if (dev) {
1564                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1565                         return -EBUSY;
1566                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1567                         tun = netdev_priv(dev);
1568                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1569                         tun = netdev_priv(dev);
1570                 else
1571                         return -EINVAL;
1572
1573                 if (tun_not_capable(tun))
1574                         return -EPERM;
1575                 err = security_tun_dev_open(tun->security);
1576                 if (err < 0)
1577                         return err;
1578
1579                 err = tun_attach(tun, file);
1580                 if (err < 0)
1581                         return err;
1582
1583                 if (tun->flags & TUN_TAP_MQ &&
1584                     (tun->numqueues + tun->numdisabled > 1))
1585                         return err;
1586         }
1587         else {
1588                 char *name;
1589                 unsigned long flags = 0;
1590                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1591                              MAX_TAP_QUEUES : 1;
1592
1593                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1594                         return -EPERM;
1595                 err = security_tun_dev_create();
1596                 if (err < 0)
1597                         return err;
1598
1599                 /* Set dev type */
1600                 if (ifr->ifr_flags & IFF_TUN) {
1601                         /* TUN device */
1602                         flags |= TUN_TUN_DEV;
1603                         name = "tun%d";
1604                 } else if (ifr->ifr_flags & IFF_TAP) {
1605                         /* TAP device */
1606                         flags |= TUN_TAP_DEV;
1607                         name = "tap%d";
1608                 } else
1609                         return -EINVAL;
1610
1611                 if (*ifr->ifr_name)
1612                         name = ifr->ifr_name;
1613
1614                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1615                                        tun_setup, queues, queues);
1616
1617                 if (!dev)
1618                         return -ENOMEM;
1619
1620                 dev_net_set(dev, net);
1621                 dev->rtnl_link_ops = &tun_link_ops;
1622
1623                 tun = netdev_priv(dev);
1624                 tun->dev = dev;
1625                 tun->flags = flags;
1626                 tun->txflt.count = 0;
1627                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1628
1629                 tun->filter_attached = false;
1630                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1631
1632                 spin_lock_init(&tun->lock);
1633
1634                 err = security_tun_dev_alloc_security(&tun->security);
1635                 if (err < 0)
1636                         goto err_free_dev;
1637
1638                 tun_net_init(dev);
1639
1640                 err = tun_flow_init(tun);
1641                 if (err < 0)
1642                         goto err_free_dev;
1643
1644                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1645                         TUN_USER_FEATURES;
1646                 dev->features = dev->hw_features;
1647
1648                 INIT_LIST_HEAD(&tun->disabled);
1649                 err = tun_attach(tun, file);
1650                 if (err < 0)
1651                         goto err_free_dev;
1652
1653                 err = register_netdevice(tun->dev);
1654                 if (err < 0)
1655                         goto err_free_dev;
1656
1657                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1658                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1659                     device_create_file(&tun->dev->dev, &dev_attr_group))
1660                         pr_err("Failed to create tun sysfs files\n");
1661
1662                 netif_carrier_on(tun->dev);
1663         }
1664
1665         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1666
1667         if (ifr->ifr_flags & IFF_NO_PI)
1668                 tun->flags |= TUN_NO_PI;
1669         else
1670                 tun->flags &= ~TUN_NO_PI;
1671
1672         /* This flag has no real effect.  We track the value for backwards
1673          * compatibility.
1674          */
1675         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1676                 tun->flags |= TUN_ONE_QUEUE;
1677         else
1678                 tun->flags &= ~TUN_ONE_QUEUE;
1679
1680         if (ifr->ifr_flags & IFF_VNET_HDR)
1681                 tun->flags |= TUN_VNET_HDR;
1682         else
1683                 tun->flags &= ~TUN_VNET_HDR;
1684
1685         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1686                 tun->flags |= TUN_TAP_MQ;
1687         else
1688                 tun->flags &= ~TUN_TAP_MQ;
1689
1690         /* Make sure persistent devices do not get stuck in
1691          * xoff state.
1692          */
1693         if (netif_running(tun->dev))
1694                 netif_tx_wake_all_queues(tun->dev);
1695
1696         strcpy(ifr->ifr_name, tun->dev->name);
1697         return 0;
1698
1699  err_free_dev:
1700         free_netdev(dev);
1701         return err;
1702 }
1703
1704 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1705                        struct ifreq *ifr)
1706 {
1707         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1708
1709         strcpy(ifr->ifr_name, tun->dev->name);
1710
1711         ifr->ifr_flags = tun_flags(tun);
1712
1713 }
1714
1715 /* This is like a cut-down ethtool ops, except done via tun fd so no
1716  * privs required. */
1717 static int set_offload(struct tun_struct *tun, unsigned long arg)
1718 {
1719         netdev_features_t features = 0;
1720
1721         if (arg & TUN_F_CSUM) {
1722                 features |= NETIF_F_HW_CSUM;
1723                 arg &= ~TUN_F_CSUM;
1724
1725                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1726                         if (arg & TUN_F_TSO_ECN) {
1727                                 features |= NETIF_F_TSO_ECN;
1728                                 arg &= ~TUN_F_TSO_ECN;
1729                         }
1730                         if (arg & TUN_F_TSO4)
1731                                 features |= NETIF_F_TSO;
1732                         if (arg & TUN_F_TSO6)
1733                                 features |= NETIF_F_TSO6;
1734                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1735                 }
1736
1737                 if (arg & TUN_F_UFO) {
1738                         features |= NETIF_F_UFO;
1739                         arg &= ~TUN_F_UFO;
1740                 }
1741         }
1742
1743         /* This gives the user a way to test for new features in future by
1744          * trying to set them. */
1745         if (arg)
1746                 return -EINVAL;
1747
1748         tun->set_features = features;
1749         netdev_update_features(tun->dev);
1750
1751         return 0;
1752 }
1753
1754 static void tun_detach_filter(struct tun_struct *tun, int n)
1755 {
1756         int i;
1757         struct tun_file *tfile;
1758
1759         for (i = 0; i < n; i++) {
1760                 tfile = rtnl_dereference(tun->tfiles[i]);
1761                 sk_detach_filter(tfile->socket.sk);
1762         }
1763
1764         tun->filter_attached = false;
1765 }
1766
1767 static int tun_attach_filter(struct tun_struct *tun)
1768 {
1769         int i, ret = 0;
1770         struct tun_file *tfile;
1771
1772         for (i = 0; i < tun->numqueues; i++) {
1773                 tfile = rtnl_dereference(tun->tfiles[i]);
1774                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1775                 if (ret) {
1776                         tun_detach_filter(tun, i);
1777                         return ret;
1778                 }
1779         }
1780
1781         tun->filter_attached = true;
1782         return ret;
1783 }
1784
1785 static void tun_set_sndbuf(struct tun_struct *tun)
1786 {
1787         struct tun_file *tfile;
1788         int i;
1789
1790         for (i = 0; i < tun->numqueues; i++) {
1791                 tfile = rtnl_dereference(tun->tfiles[i]);
1792                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1793         }
1794 }
1795
1796 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1797 {
1798         struct tun_file *tfile = file->private_data;
1799         struct tun_struct *tun;
1800         int ret = 0;
1801
1802         rtnl_lock();
1803
1804         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1805                 tun = tfile->detached;
1806                 if (!tun) {
1807                         ret = -EINVAL;
1808                         goto unlock;
1809                 }
1810                 ret = security_tun_dev_attach_queue(tun->security);
1811                 if (ret < 0)
1812                         goto unlock;
1813                 ret = tun_attach(tun, file);
1814         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1815                 tun = rtnl_dereference(tfile->tun);
1816                 if (!tun || !(tun->flags & TUN_TAP_MQ))
1817                         ret = -EINVAL;
1818                 else
1819                         __tun_detach(tfile, false);
1820         } else
1821                 ret = -EINVAL;
1822
1823 unlock:
1824         rtnl_unlock();
1825         return ret;
1826 }
1827
1828 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1829                             unsigned long arg, int ifreq_len)
1830 {
1831         struct tun_file *tfile = file->private_data;
1832         struct tun_struct *tun;
1833         void __user* argp = (void __user*)arg;
1834         struct ifreq ifr;
1835         kuid_t owner;
1836         kgid_t group;
1837         int sndbuf;
1838         int vnet_hdr_sz;
1839         int ret;
1840
1841         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1842                 if (copy_from_user(&ifr, argp, ifreq_len))
1843                         return -EFAULT;
1844         } else {
1845                 memset(&ifr, 0, sizeof(ifr));
1846         }
1847         if (cmd == TUNGETFEATURES) {
1848                 /* Currently this just means: "what IFF flags are valid?".
1849                  * This is needed because we never checked for invalid flags on
1850                  * TUNSETIFF. */
1851                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1852                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1853                                 (unsigned int __user*)argp);
1854         } else if (cmd == TUNSETQUEUE)
1855                 return tun_set_queue(file, &ifr);
1856
1857         ret = 0;
1858         rtnl_lock();
1859
1860         tun = __tun_get(tfile);
1861         if (cmd == TUNSETIFF && !tun) {
1862                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1863
1864                 ret = tun_set_iff(tfile->net, file, &ifr);
1865
1866                 if (ret)
1867                         goto unlock;
1868
1869                 if (copy_to_user(argp, &ifr, ifreq_len))
1870                         ret = -EFAULT;
1871                 goto unlock;
1872         }
1873
1874         ret = -EBADFD;
1875         if (!tun)
1876                 goto unlock;
1877
1878         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1879
1880         ret = 0;
1881         switch (cmd) {
1882         case TUNGETIFF:
1883                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1884
1885                 if (copy_to_user(argp, &ifr, ifreq_len))
1886                         ret = -EFAULT;
1887                 break;
1888
1889         case TUNSETNOCSUM:
1890                 /* Disable/Enable checksum */
1891
1892                 /* [unimplemented] */
1893                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1894                           arg ? "disabled" : "enabled");
1895                 break;
1896
1897         case TUNSETPERSIST:
1898                 /* Disable/Enable persist mode. Keep an extra reference to the
1899                  * module to prevent the module being unprobed.
1900                  */
1901                 if (arg && !(tun->flags & TUN_PERSIST)) {
1902                         tun->flags |= TUN_PERSIST;
1903                         __module_get(THIS_MODULE);
1904                 }
1905                 if (!arg && (tun->flags & TUN_PERSIST)) {
1906                         tun->flags &= ~TUN_PERSIST;
1907                         module_put(THIS_MODULE);
1908                 }
1909
1910                 tun_debug(KERN_INFO, tun, "persist %s\n",
1911                           arg ? "enabled" : "disabled");
1912                 break;
1913
1914         case TUNSETOWNER:
1915                 /* Set owner of the device */
1916                 owner = make_kuid(current_user_ns(), arg);
1917                 if (!uid_valid(owner)) {
1918                         ret = -EINVAL;
1919                         break;
1920                 }
1921                 tun->owner = owner;
1922                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1923                           from_kuid(&init_user_ns, tun->owner));
1924                 break;
1925
1926         case TUNSETGROUP:
1927                 /* Set group of the device */
1928                 group = make_kgid(current_user_ns(), arg);
1929                 if (!gid_valid(group)) {
1930                         ret = -EINVAL;
1931                         break;
1932                 }
1933                 tun->group = group;
1934                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1935                           from_kgid(&init_user_ns, tun->group));
1936                 break;
1937
1938         case TUNSETLINK:
1939                 /* Only allow setting the type when the interface is down */
1940                 if (tun->dev->flags & IFF_UP) {
1941                         tun_debug(KERN_INFO, tun,
1942                                   "Linktype set failed because interface is up\n");
1943                         ret = -EBUSY;
1944                 } else {
1945                         tun->dev->type = (int) arg;
1946                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1947                                   tun->dev->type);
1948                         ret = 0;
1949                 }
1950                 break;
1951
1952 #ifdef TUN_DEBUG
1953         case TUNSETDEBUG:
1954                 tun->debug = arg;
1955                 break;
1956 #endif
1957         case TUNSETOFFLOAD:
1958                 ret = set_offload(tun, arg);
1959                 break;
1960
1961         case TUNSETTXFILTER:
1962                 /* Can be set only for TAPs */
1963                 ret = -EINVAL;
1964                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1965                         break;
1966                 ret = update_filter(&tun->txflt, (void __user *)arg);
1967                 break;
1968
1969         case SIOCGIFHWADDR:
1970                 /* Get hw address */
1971                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1972                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1973                 if (copy_to_user(argp, &ifr, ifreq_len))
1974                         ret = -EFAULT;
1975                 break;
1976
1977         case SIOCSIFHWADDR:
1978                 /* Set hw address */
1979                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1980                           ifr.ifr_hwaddr.sa_data);
1981
1982                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1983                 break;
1984
1985         case TUNGETSNDBUF:
1986                 sndbuf = tfile->socket.sk->sk_sndbuf;
1987                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1988                         ret = -EFAULT;
1989                 break;
1990
1991         case TUNSETSNDBUF:
1992                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1993                         ret = -EFAULT;
1994                         break;
1995                 }
1996
1997                 tun->sndbuf = sndbuf;
1998                 tun_set_sndbuf(tun);
1999                 break;
2000
2001         case TUNGETVNETHDRSZ:
2002                 vnet_hdr_sz = tun->vnet_hdr_sz;
2003                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2004                         ret = -EFAULT;
2005                 break;
2006
2007         case TUNSETVNETHDRSZ:
2008                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2009                         ret = -EFAULT;
2010                         break;
2011                 }
2012                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2013                         ret = -EINVAL;
2014                         break;
2015                 }
2016
2017                 tun->vnet_hdr_sz = vnet_hdr_sz;
2018                 break;
2019
2020         case TUNATTACHFILTER:
2021                 /* Can be set only for TAPs */
2022                 ret = -EINVAL;
2023                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2024                         break;
2025                 ret = -EFAULT;
2026                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2027                         break;
2028
2029                 ret = tun_attach_filter(tun);
2030                 break;
2031
2032         case TUNDETACHFILTER:
2033                 /* Can be set only for TAPs */
2034                 ret = -EINVAL;
2035                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2036                         break;
2037                 ret = 0;
2038                 tun_detach_filter(tun, tun->numqueues);
2039                 break;
2040
2041         default:
2042                 ret = -EINVAL;
2043                 break;
2044         }
2045
2046 unlock:
2047         rtnl_unlock();
2048         if (tun)
2049                 tun_put(tun);
2050         return ret;
2051 }
2052
2053 static long tun_chr_ioctl(struct file *file,
2054                           unsigned int cmd, unsigned long arg)
2055 {
2056         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2057 }
2058
2059 #ifdef CONFIG_COMPAT
2060 static long tun_chr_compat_ioctl(struct file *file,
2061                          unsigned int cmd, unsigned long arg)
2062 {
2063         switch (cmd) {
2064         case TUNSETIFF:
2065         case TUNGETIFF:
2066         case TUNSETTXFILTER:
2067         case TUNGETSNDBUF:
2068         case TUNSETSNDBUF:
2069         case SIOCGIFHWADDR:
2070         case SIOCSIFHWADDR:
2071                 arg = (unsigned long)compat_ptr(arg);
2072                 break;
2073         default:
2074                 arg = (compat_ulong_t)arg;
2075                 break;
2076         }
2077
2078         /*
2079          * compat_ifreq is shorter than ifreq, so we must not access beyond
2080          * the end of that structure. All fields that are used in this
2081          * driver are compatible though, we don't need to convert the
2082          * contents.
2083          */
2084         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2085 }
2086 #endif /* CONFIG_COMPAT */
2087
2088 static int tun_chr_fasync(int fd, struct file *file, int on)
2089 {
2090         struct tun_file *tfile = file->private_data;
2091         int ret;
2092
2093         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2094                 goto out;
2095
2096         if (on) {
2097                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2098                 if (ret)
2099                         goto out;
2100                 tfile->flags |= TUN_FASYNC;
2101         } else
2102                 tfile->flags &= ~TUN_FASYNC;
2103         ret = 0;
2104 out:
2105         return ret;
2106 }
2107
2108 static int tun_chr_open(struct inode *inode, struct file * file)
2109 {
2110         struct tun_file *tfile;
2111
2112         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2113
2114         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2115                                             &tun_proto);
2116         if (!tfile)
2117                 return -ENOMEM;
2118         rcu_assign_pointer(tfile->tun, NULL);
2119         tfile->net = get_net(current->nsproxy->net_ns);
2120         tfile->flags = 0;
2121
2122         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2123         init_waitqueue_head(&tfile->wq.wait);
2124
2125         tfile->socket.file = file;
2126         tfile->socket.ops = &tun_socket_ops;
2127
2128         sock_init_data(&tfile->socket, &tfile->sk);
2129         sk_change_net(&tfile->sk, tfile->net);
2130
2131         tfile->sk.sk_write_space = tun_sock_write_space;
2132         tfile->sk.sk_sndbuf = INT_MAX;
2133
2134         file->private_data = tfile;
2135         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2136         INIT_LIST_HEAD(&tfile->next);
2137
2138         return 0;
2139 }
2140
2141 static int tun_chr_close(struct inode *inode, struct file *file)
2142 {
2143         struct tun_file *tfile = file->private_data;
2144         struct net *net = tfile->net;
2145
2146         tun_detach(tfile, true);
2147         put_net(net);
2148
2149         return 0;
2150 }
2151
2152 static const struct file_operations tun_fops = {
2153         .owner  = THIS_MODULE,
2154         .llseek = no_llseek,
2155         .read  = do_sync_read,
2156         .aio_read  = tun_chr_aio_read,
2157         .write = do_sync_write,
2158         .aio_write = tun_chr_aio_write,
2159         .poll   = tun_chr_poll,
2160         .unlocked_ioctl = tun_chr_ioctl,
2161 #ifdef CONFIG_COMPAT
2162         .compat_ioctl = tun_chr_compat_ioctl,
2163 #endif
2164         .open   = tun_chr_open,
2165         .release = tun_chr_close,
2166         .fasync = tun_chr_fasync
2167 };
2168
2169 static struct miscdevice tun_miscdev = {
2170         .minor = TUN_MINOR,
2171         .name = "tun",
2172         .nodename = "net/tun",
2173         .fops = &tun_fops,
2174 };
2175
2176 /* ethtool interface */
2177
2178 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2179 {
2180         cmd->supported          = 0;
2181         cmd->advertising        = 0;
2182         ethtool_cmd_speed_set(cmd, SPEED_10);
2183         cmd->duplex             = DUPLEX_FULL;
2184         cmd->port               = PORT_TP;
2185         cmd->phy_address        = 0;
2186         cmd->transceiver        = XCVR_INTERNAL;
2187         cmd->autoneg            = AUTONEG_DISABLE;
2188         cmd->maxtxpkt           = 0;
2189         cmd->maxrxpkt           = 0;
2190         return 0;
2191 }
2192
2193 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2194 {
2195         struct tun_struct *tun = netdev_priv(dev);
2196
2197         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2198         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2199
2200         switch (tun->flags & TUN_TYPE_MASK) {
2201         case TUN_TUN_DEV:
2202                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2203                 break;
2204         case TUN_TAP_DEV:
2205                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2206                 break;
2207         }
2208 }
2209
2210 static u32 tun_get_msglevel(struct net_device *dev)
2211 {
2212 #ifdef TUN_DEBUG
2213         struct tun_struct *tun = netdev_priv(dev);
2214         return tun->debug;
2215 #else
2216         return -EOPNOTSUPP;
2217 #endif
2218 }
2219
2220 static void tun_set_msglevel(struct net_device *dev, u32 value)
2221 {
2222 #ifdef TUN_DEBUG
2223         struct tun_struct *tun = netdev_priv(dev);
2224         tun->debug = value;
2225 #endif
2226 }
2227
2228 static const struct ethtool_ops tun_ethtool_ops = {
2229         .get_settings   = tun_get_settings,
2230         .get_drvinfo    = tun_get_drvinfo,
2231         .get_msglevel   = tun_get_msglevel,
2232         .set_msglevel   = tun_set_msglevel,
2233         .get_link       = ethtool_op_get_link,
2234 };
2235
2236
2237 static int __init tun_init(void)
2238 {
2239         int ret = 0;
2240
2241         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2242         pr_info("%s\n", DRV_COPYRIGHT);
2243
2244         ret = rtnl_link_register(&tun_link_ops);
2245         if (ret) {
2246                 pr_err("Can't register link_ops\n");
2247                 goto err_linkops;
2248         }
2249
2250         ret = misc_register(&tun_miscdev);
2251         if (ret) {
2252                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2253                 goto err_misc;
2254         }
2255         return  0;
2256 err_misc:
2257         rtnl_link_unregister(&tun_link_ops);
2258 err_linkops:
2259         return ret;
2260 }
2261
2262 static void tun_cleanup(void)
2263 {
2264         misc_deregister(&tun_miscdev);
2265         rtnl_link_unregister(&tun_link_ops);
2266 }
2267
2268 /* Get an underlying socket object from tun file.  Returns error unless file is
2269  * attached to a device.  The returned object works like a packet socket, it
2270  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2271  * holding a reference to the file for as long as the socket is in use. */
2272 struct socket *tun_get_socket(struct file *file)
2273 {
2274         struct tun_file *tfile;
2275         if (file->f_op != &tun_fops)
2276                 return ERR_PTR(-EINVAL);
2277         tfile = file->private_data;
2278         if (!tfile)
2279                 return ERR_PTR(-EBADFD);
2280         return &tfile->socket;
2281 }
2282 EXPORT_SYMBOL_GPL(tun_get_socket);
2283
2284 module_init(tun_init);
2285 module_exit(tun_cleanup);
2286 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2287 MODULE_AUTHOR(DRV_COPYRIGHT);
2288 MODULE_LICENSE("GPL");
2289 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2290 MODULE_ALIAS("devname:net/tun");