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                 skb_shinfo(skb)->gso_type |= SKB_GSO_SHARED_FRAG;
1013                 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
1014                 while (len) {
1015                         int off = base & ~PAGE_MASK;
1016                         int size = min_t(int, len, PAGE_SIZE - off);
1017                         __skb_fill_page_desc(skb, i, page[i], off, size);
1018                         skb_shinfo(skb)->nr_frags++;
1019                         /* increase sk_wmem_alloc */
1020                         base += size;
1021                         len -= size;
1022                         i++;
1023                 }
1024                 offset = 0;
1025                 ++from;
1026         }
1027         return 0;
1028 }
1029
1030 /* Get packet from user space buffer */
1031 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1032                             void *msg_control, const struct iovec *iv,
1033                             size_t total_len, size_t count, int noblock)
1034 {
1035         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1036         struct sk_buff *skb;
1037         size_t len = total_len, align = NET_SKB_PAD;
1038         struct virtio_net_hdr gso = { 0 };
1039         int offset = 0;
1040         int copylen;
1041         bool zerocopy = false;
1042         int err;
1043         u32 rxhash;
1044
1045         if (!(tun->flags & TUN_NO_PI)) {
1046                 if ((len -= sizeof(pi)) > total_len)
1047                         return -EINVAL;
1048
1049                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1050                         return -EFAULT;
1051                 offset += sizeof(pi);
1052         }
1053
1054         if (tun->flags & TUN_VNET_HDR) {
1055                 if ((len -= tun->vnet_hdr_sz) > total_len)
1056                         return -EINVAL;
1057
1058                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1059                         return -EFAULT;
1060
1061                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1062                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1063                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1064
1065                 if (gso.hdr_len > len)
1066                         return -EINVAL;
1067                 offset += tun->vnet_hdr_sz;
1068         }
1069
1070         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1071                 align += NET_IP_ALIGN;
1072                 if (unlikely(len < ETH_HLEN ||
1073                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1074                         return -EINVAL;
1075         }
1076
1077         if (msg_control)
1078                 zerocopy = true;
1079
1080         if (zerocopy) {
1081                 /* Userspace may produce vectors with count greater than
1082                  * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1083                  * to let the rest of data to be fit in the frags.
1084                  */
1085                 if (count > MAX_SKB_FRAGS) {
1086                         copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1087                         if (copylen < offset)
1088                                 copylen = 0;
1089                         else
1090                                 copylen -= offset;
1091                 } else
1092                                 copylen = 0;
1093                 /* There are 256 bytes to be copied in skb, so there is enough
1094                  * room for skb expand head in case it is used.
1095                  * The rest of the buffer is mapped from userspace.
1096                  */
1097                 if (copylen < gso.hdr_len)
1098                         copylen = gso.hdr_len;
1099                 if (!copylen)
1100                         copylen = GOODCOPY_LEN;
1101         } else
1102                 copylen = len;
1103
1104         skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
1105         if (IS_ERR(skb)) {
1106                 if (PTR_ERR(skb) != -EAGAIN)
1107                         tun->dev->stats.rx_dropped++;
1108                 return PTR_ERR(skb);
1109         }
1110
1111         if (zerocopy)
1112                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1113         else
1114                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1115
1116         if (err) {
1117                 tun->dev->stats.rx_dropped++;
1118                 kfree_skb(skb);
1119                 return -EFAULT;
1120         }
1121
1122         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1123                 if (!skb_partial_csum_set(skb, gso.csum_start,
1124                                           gso.csum_offset)) {
1125                         tun->dev->stats.rx_frame_errors++;
1126                         kfree_skb(skb);
1127                         return -EINVAL;
1128                 }
1129         }
1130
1131         switch (tun->flags & TUN_TYPE_MASK) {
1132         case TUN_TUN_DEV:
1133                 if (tun->flags & TUN_NO_PI) {
1134                         switch (skb->data[0] & 0xf0) {
1135                         case 0x40:
1136                                 pi.proto = htons(ETH_P_IP);
1137                                 break;
1138                         case 0x60:
1139                                 pi.proto = htons(ETH_P_IPV6);
1140                                 break;
1141                         default:
1142                                 tun->dev->stats.rx_dropped++;
1143                                 kfree_skb(skb);
1144                                 return -EINVAL;
1145                         }
1146                 }
1147
1148                 skb_reset_mac_header(skb);
1149                 skb->protocol = pi.proto;
1150                 skb->dev = tun->dev;
1151                 break;
1152         case TUN_TAP_DEV:
1153                 skb->protocol = eth_type_trans(skb, tun->dev);
1154                 break;
1155         }
1156
1157         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1158                 unsigned short gso_type = 0;
1159
1160                 pr_debug("GSO!\n");
1161                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1162                 case VIRTIO_NET_HDR_GSO_TCPV4:
1163                         gso_type = SKB_GSO_TCPV4;
1164                         break;
1165                 case VIRTIO_NET_HDR_GSO_TCPV6:
1166                         gso_type = SKB_GSO_TCPV6;
1167                         break;
1168                 case VIRTIO_NET_HDR_GSO_UDP:
1169                         gso_type = SKB_GSO_UDP;
1170                         break;
1171                 default:
1172                         tun->dev->stats.rx_frame_errors++;
1173                         kfree_skb(skb);
1174                         return -EINVAL;
1175                 }
1176
1177                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1178                         gso_type |= SKB_GSO_TCP_ECN;
1179
1180                 skb_shinfo(skb)->gso_size = gso.gso_size;
1181                 skb_shinfo(skb)->gso_type |= gso_type;
1182                 if (skb_shinfo(skb)->gso_size == 0) {
1183                         tun->dev->stats.rx_frame_errors++;
1184                         kfree_skb(skb);
1185                         return -EINVAL;
1186                 }
1187
1188                 /* Header must be checked, and gso_segs computed. */
1189                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1190                 skb_shinfo(skb)->gso_segs = 0;
1191         }
1192
1193         /* copy skb_ubuf_info for callback when skb has no error */
1194         if (zerocopy) {
1195                 skb_shinfo(skb)->destructor_arg = msg_control;
1196                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1197         }
1198
1199         skb_reset_network_header(skb);
1200         rxhash = skb_get_rxhash(skb);
1201         netif_rx_ni(skb);
1202
1203         tun->dev->stats.rx_packets++;
1204         tun->dev->stats.rx_bytes += len;
1205
1206         tun_flow_update(tun, rxhash, tfile->queue_index);
1207         return total_len;
1208 }
1209
1210 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1211                               unsigned long count, loff_t pos)
1212 {
1213         struct file *file = iocb->ki_filp;
1214         struct tun_struct *tun = tun_get(file);
1215         struct tun_file *tfile = file->private_data;
1216         ssize_t result;
1217
1218         if (!tun)
1219                 return -EBADFD;
1220
1221         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1222
1223         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1224                               count, file->f_flags & O_NONBLOCK);
1225
1226         tun_put(tun);
1227         return result;
1228 }
1229
1230 /* Put packet to the user space buffer */
1231 static ssize_t tun_put_user(struct tun_struct *tun,
1232                             struct tun_file *tfile,
1233                             struct sk_buff *skb,
1234                             const struct iovec *iv, int len)
1235 {
1236         struct tun_pi pi = { 0, skb->protocol };
1237         ssize_t total = 0;
1238
1239         if (!(tun->flags & TUN_NO_PI)) {
1240                 if ((len -= sizeof(pi)) < 0)
1241                         return -EINVAL;
1242
1243                 if (len < skb->len) {
1244                         /* Packet will be striped */
1245                         pi.flags |= TUN_PKT_STRIP;
1246                 }
1247
1248                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1249                         return -EFAULT;
1250                 total += sizeof(pi);
1251         }
1252
1253         if (tun->flags & TUN_VNET_HDR) {
1254                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1255                 if ((len -= tun->vnet_hdr_sz) < 0)
1256                         return -EINVAL;
1257
1258                 if (skb_is_gso(skb)) {
1259                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1260
1261                         /* This is a hint as to how much should be linear. */
1262                         gso.hdr_len = skb_headlen(skb);
1263                         gso.gso_size = sinfo->gso_size;
1264                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1265                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1266                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1267                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1268                         else if (sinfo->gso_type & SKB_GSO_UDP)
1269                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1270                         else {
1271                                 pr_err("unexpected GSO type: "
1272                                        "0x%x, gso_size %d, hdr_len %d\n",
1273                                        sinfo->gso_type, gso.gso_size,
1274                                        gso.hdr_len);
1275                                 print_hex_dump(KERN_ERR, "tun: ",
1276                                                DUMP_PREFIX_NONE,
1277                                                16, 1, skb->head,
1278                                                min((int)gso.hdr_len, 64), true);
1279                                 WARN_ON_ONCE(1);
1280                                 return -EINVAL;
1281                         }
1282                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1283                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1284                 } else
1285                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1286
1287                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1288                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1289                         gso.csum_start = skb_checksum_start_offset(skb);
1290                         gso.csum_offset = skb->csum_offset;
1291                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1292                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1293                 } /* else everything is zero */
1294
1295                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1296                                                sizeof(gso))))
1297                         return -EFAULT;
1298                 total += tun->vnet_hdr_sz;
1299         }
1300
1301         len = min_t(int, skb->len, len);
1302
1303         skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1304         total += skb->len;
1305
1306         tun->dev->stats.tx_packets++;
1307         tun->dev->stats.tx_bytes += len;
1308
1309         return total;
1310 }
1311
1312 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1313                            struct kiocb *iocb, const struct iovec *iv,
1314                            ssize_t len, int noblock)
1315 {
1316         DECLARE_WAITQUEUE(wait, current);
1317         struct sk_buff *skb;
1318         ssize_t ret = 0;
1319
1320         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1321
1322         if (unlikely(!noblock))
1323                 add_wait_queue(&tfile->wq.wait, &wait);
1324         while (len) {
1325                 current->state = TASK_INTERRUPTIBLE;
1326
1327                 /* Read frames from the queue */
1328                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1329                         if (noblock) {
1330                                 ret = -EAGAIN;
1331                                 break;
1332                         }
1333                         if (signal_pending(current)) {
1334                                 ret = -ERESTARTSYS;
1335                                 break;
1336                         }
1337                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1338                                 ret = -EIO;
1339                                 break;
1340                         }
1341
1342                         /* Nothing to read, let's sleep */
1343                         schedule();
1344                         continue;
1345                 }
1346
1347                 ret = tun_put_user(tun, tfile, skb, iv, len);
1348                 kfree_skb(skb);
1349                 break;
1350         }
1351
1352         current->state = TASK_RUNNING;
1353         if (unlikely(!noblock))
1354                 remove_wait_queue(&tfile->wq.wait, &wait);
1355
1356         return ret;
1357 }
1358
1359 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1360                             unsigned long count, loff_t pos)
1361 {
1362         struct file *file = iocb->ki_filp;
1363         struct tun_file *tfile = file->private_data;
1364         struct tun_struct *tun = __tun_get(tfile);
1365         ssize_t len, ret;
1366
1367         if (!tun)
1368                 return -EBADFD;
1369         len = iov_length(iv, count);
1370         if (len < 0) {
1371                 ret = -EINVAL;
1372                 goto out;
1373         }
1374
1375         ret = tun_do_read(tun, tfile, iocb, iv, len,
1376                           file->f_flags & O_NONBLOCK);
1377         ret = min_t(ssize_t, ret, len);
1378 out:
1379         tun_put(tun);
1380         return ret;
1381 }
1382
1383 static void tun_free_netdev(struct net_device *dev)
1384 {
1385         struct tun_struct *tun = netdev_priv(dev);
1386
1387         BUG_ON(!(list_empty(&tun->disabled)));
1388         tun_flow_uninit(tun);
1389         security_tun_dev_free_security(tun->security);
1390         free_netdev(dev);
1391 }
1392
1393 static void tun_setup(struct net_device *dev)
1394 {
1395         struct tun_struct *tun = netdev_priv(dev);
1396
1397         tun->owner = INVALID_UID;
1398         tun->group = INVALID_GID;
1399
1400         dev->ethtool_ops = &tun_ethtool_ops;
1401         dev->destructor = tun_free_netdev;
1402 }
1403
1404 /* Trivial set of netlink ops to allow deleting tun or tap
1405  * device with netlink.
1406  */
1407 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1408 {
1409         return -EINVAL;
1410 }
1411
1412 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1413         .kind           = DRV_NAME,
1414         .priv_size      = sizeof(struct tun_struct),
1415         .setup          = tun_setup,
1416         .validate       = tun_validate,
1417 };
1418
1419 static void tun_sock_write_space(struct sock *sk)
1420 {
1421         struct tun_file *tfile;
1422         wait_queue_head_t *wqueue;
1423
1424         if (!sock_writeable(sk))
1425                 return;
1426
1427         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1428                 return;
1429
1430         wqueue = sk_sleep(sk);
1431         if (wqueue && waitqueue_active(wqueue))
1432                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1433                                                 POLLWRNORM | POLLWRBAND);
1434
1435         tfile = container_of(sk, struct tun_file, sk);
1436         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1437 }
1438
1439 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1440                        struct msghdr *m, size_t total_len)
1441 {
1442         int ret;
1443         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1444         struct tun_struct *tun = __tun_get(tfile);
1445
1446         if (!tun)
1447                 return -EBADFD;
1448         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1449                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1450         tun_put(tun);
1451         return ret;
1452 }
1453
1454
1455 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1456                        struct msghdr *m, size_t total_len,
1457                        int flags)
1458 {
1459         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1460         struct tun_struct *tun = __tun_get(tfile);
1461         int ret;
1462
1463         if (!tun)
1464                 return -EBADFD;
1465
1466         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1467                 return -EINVAL;
1468         ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1469                           flags & MSG_DONTWAIT);
1470         if (ret > total_len) {
1471                 m->msg_flags |= MSG_TRUNC;
1472                 ret = flags & MSG_TRUNC ? ret : total_len;
1473         }
1474         tun_put(tun);
1475         return ret;
1476 }
1477
1478 static int tun_release(struct socket *sock)
1479 {
1480         if (sock->sk)
1481                 sock_put(sock->sk);
1482         return 0;
1483 }
1484
1485 /* Ops structure to mimic raw sockets with tun */
1486 static const struct proto_ops tun_socket_ops = {
1487         .sendmsg = tun_sendmsg,
1488         .recvmsg = tun_recvmsg,
1489         .release = tun_release,
1490 };
1491
1492 static struct proto tun_proto = {
1493         .name           = "tun",
1494         .owner          = THIS_MODULE,
1495         .obj_size       = sizeof(struct tun_file),
1496 };
1497
1498 static int tun_flags(struct tun_struct *tun)
1499 {
1500         int flags = 0;
1501
1502         if (tun->flags & TUN_TUN_DEV)
1503                 flags |= IFF_TUN;
1504         else
1505                 flags |= IFF_TAP;
1506
1507         if (tun->flags & TUN_NO_PI)
1508                 flags |= IFF_NO_PI;
1509
1510         /* This flag has no real effect.  We track the value for backwards
1511          * compatibility.
1512          */
1513         if (tun->flags & TUN_ONE_QUEUE)
1514                 flags |= IFF_ONE_QUEUE;
1515
1516         if (tun->flags & TUN_VNET_HDR)
1517                 flags |= IFF_VNET_HDR;
1518
1519         if (tun->flags & TUN_TAP_MQ)
1520                 flags |= IFF_MULTI_QUEUE;
1521
1522         return flags;
1523 }
1524
1525 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1526                               char *buf)
1527 {
1528         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1529         return sprintf(buf, "0x%x\n", tun_flags(tun));
1530 }
1531
1532 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1533                               char *buf)
1534 {
1535         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1536         return uid_valid(tun->owner)?
1537                 sprintf(buf, "%u\n",
1538                         from_kuid_munged(current_user_ns(), tun->owner)):
1539                 sprintf(buf, "-1\n");
1540 }
1541
1542 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1543                               char *buf)
1544 {
1545         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1546         return gid_valid(tun->group) ?
1547                 sprintf(buf, "%u\n",
1548                         from_kgid_munged(current_user_ns(), tun->group)):
1549                 sprintf(buf, "-1\n");
1550 }
1551
1552 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1553 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1554 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1555
1556 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1557 {
1558         struct tun_struct *tun;
1559         struct tun_file *tfile = file->private_data;
1560         struct net_device *dev;
1561         int err;
1562
1563         if (tfile->detached)
1564                 return -EINVAL;
1565
1566         dev = __dev_get_by_name(net, ifr->ifr_name);
1567         if (dev) {
1568                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1569                         return -EBUSY;
1570                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1571                         tun = netdev_priv(dev);
1572                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1573                         tun = netdev_priv(dev);
1574                 else
1575                         return -EINVAL;
1576
1577                 if (tun_not_capable(tun))
1578                         return -EPERM;
1579                 err = security_tun_dev_open(tun->security);
1580                 if (err < 0)
1581                         return err;
1582
1583                 err = tun_attach(tun, file);
1584                 if (err < 0)
1585                         return err;
1586
1587                 if (tun->flags & TUN_TAP_MQ &&
1588                     (tun->numqueues + tun->numdisabled > 1))
1589                         return err;
1590         }
1591         else {
1592                 char *name;
1593                 unsigned long flags = 0;
1594                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1595                              MAX_TAP_QUEUES : 1;
1596
1597                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1598                         return -EPERM;
1599                 err = security_tun_dev_create();
1600                 if (err < 0)
1601                         return err;
1602
1603                 /* Set dev type */
1604                 if (ifr->ifr_flags & IFF_TUN) {
1605                         /* TUN device */
1606                         flags |= TUN_TUN_DEV;
1607                         name = "tun%d";
1608                 } else if (ifr->ifr_flags & IFF_TAP) {
1609                         /* TAP device */
1610                         flags |= TUN_TAP_DEV;
1611                         name = "tap%d";
1612                 } else
1613                         return -EINVAL;
1614
1615                 if (*ifr->ifr_name)
1616                         name = ifr->ifr_name;
1617
1618                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1619                                        tun_setup, queues, queues);
1620
1621                 if (!dev)
1622                         return -ENOMEM;
1623
1624                 dev_net_set(dev, net);
1625                 dev->rtnl_link_ops = &tun_link_ops;
1626
1627                 tun = netdev_priv(dev);
1628                 tun->dev = dev;
1629                 tun->flags = flags;
1630                 tun->txflt.count = 0;
1631                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1632
1633                 tun->filter_attached = false;
1634                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1635
1636                 spin_lock_init(&tun->lock);
1637
1638                 err = security_tun_dev_alloc_security(&tun->security);
1639                 if (err < 0)
1640                         goto err_free_dev;
1641
1642                 tun_net_init(dev);
1643
1644                 err = tun_flow_init(tun);
1645                 if (err < 0)
1646                         goto err_free_dev;
1647
1648                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1649                         TUN_USER_FEATURES;
1650                 dev->features = dev->hw_features;
1651
1652                 INIT_LIST_HEAD(&tun->disabled);
1653                 err = tun_attach(tun, file);
1654                 if (err < 0)
1655                         goto err_free_dev;
1656
1657                 err = register_netdevice(tun->dev);
1658                 if (err < 0)
1659                         goto err_free_dev;
1660
1661                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1662                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1663                     device_create_file(&tun->dev->dev, &dev_attr_group))
1664                         pr_err("Failed to create tun sysfs files\n");
1665
1666                 netif_carrier_on(tun->dev);
1667         }
1668
1669         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1670
1671         if (ifr->ifr_flags & IFF_NO_PI)
1672                 tun->flags |= TUN_NO_PI;
1673         else
1674                 tun->flags &= ~TUN_NO_PI;
1675
1676         /* This flag has no real effect.  We track the value for backwards
1677          * compatibility.
1678          */
1679         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1680                 tun->flags |= TUN_ONE_QUEUE;
1681         else
1682                 tun->flags &= ~TUN_ONE_QUEUE;
1683
1684         if (ifr->ifr_flags & IFF_VNET_HDR)
1685                 tun->flags |= TUN_VNET_HDR;
1686         else
1687                 tun->flags &= ~TUN_VNET_HDR;
1688
1689         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1690                 tun->flags |= TUN_TAP_MQ;
1691         else
1692                 tun->flags &= ~TUN_TAP_MQ;
1693
1694         /* Make sure persistent devices do not get stuck in
1695          * xoff state.
1696          */
1697         if (netif_running(tun->dev))
1698                 netif_tx_wake_all_queues(tun->dev);
1699
1700         strcpy(ifr->ifr_name, tun->dev->name);
1701         return 0;
1702
1703  err_free_dev:
1704         free_netdev(dev);
1705         return err;
1706 }
1707
1708 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1709                        struct ifreq *ifr)
1710 {
1711         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1712
1713         strcpy(ifr->ifr_name, tun->dev->name);
1714
1715         ifr->ifr_flags = tun_flags(tun);
1716
1717 }
1718
1719 /* This is like a cut-down ethtool ops, except done via tun fd so no
1720  * privs required. */
1721 static int set_offload(struct tun_struct *tun, unsigned long arg)
1722 {
1723         netdev_features_t features = 0;
1724
1725         if (arg & TUN_F_CSUM) {
1726                 features |= NETIF_F_HW_CSUM;
1727                 arg &= ~TUN_F_CSUM;
1728
1729                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1730                         if (arg & TUN_F_TSO_ECN) {
1731                                 features |= NETIF_F_TSO_ECN;
1732                                 arg &= ~TUN_F_TSO_ECN;
1733                         }
1734                         if (arg & TUN_F_TSO4)
1735                                 features |= NETIF_F_TSO;
1736                         if (arg & TUN_F_TSO6)
1737                                 features |= NETIF_F_TSO6;
1738                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1739                 }
1740
1741                 if (arg & TUN_F_UFO) {
1742                         features |= NETIF_F_UFO;
1743                         arg &= ~TUN_F_UFO;
1744                 }
1745         }
1746
1747         /* This gives the user a way to test for new features in future by
1748          * trying to set them. */
1749         if (arg)
1750                 return -EINVAL;
1751
1752         tun->set_features = features;
1753         netdev_update_features(tun->dev);
1754
1755         return 0;
1756 }
1757
1758 static void tun_detach_filter(struct tun_struct *tun, int n)
1759 {
1760         int i;
1761         struct tun_file *tfile;
1762
1763         for (i = 0; i < n; i++) {
1764                 tfile = rtnl_dereference(tun->tfiles[i]);
1765                 sk_detach_filter(tfile->socket.sk);
1766         }
1767
1768         tun->filter_attached = false;
1769 }
1770
1771 static int tun_attach_filter(struct tun_struct *tun)
1772 {
1773         int i, ret = 0;
1774         struct tun_file *tfile;
1775
1776         for (i = 0; i < tun->numqueues; i++) {
1777                 tfile = rtnl_dereference(tun->tfiles[i]);
1778                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1779                 if (ret) {
1780                         tun_detach_filter(tun, i);
1781                         return ret;
1782                 }
1783         }
1784
1785         tun->filter_attached = true;
1786         return ret;
1787 }
1788
1789 static void tun_set_sndbuf(struct tun_struct *tun)
1790 {
1791         struct tun_file *tfile;
1792         int i;
1793
1794         for (i = 0; i < tun->numqueues; i++) {
1795                 tfile = rtnl_dereference(tun->tfiles[i]);
1796                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1797         }
1798 }
1799
1800 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1801 {
1802         struct tun_file *tfile = file->private_data;
1803         struct tun_struct *tun;
1804         int ret = 0;
1805
1806         rtnl_lock();
1807
1808         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1809                 tun = tfile->detached;
1810                 if (!tun) {
1811                         ret = -EINVAL;
1812                         goto unlock;
1813                 }
1814                 ret = security_tun_dev_attach_queue(tun->security);
1815                 if (ret < 0)
1816                         goto unlock;
1817                 ret = tun_attach(tun, file);
1818         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1819                 tun = rtnl_dereference(tfile->tun);
1820                 if (!tun || !(tun->flags & TUN_TAP_MQ))
1821                         ret = -EINVAL;
1822                 else
1823                         __tun_detach(tfile, false);
1824         } else
1825                 ret = -EINVAL;
1826
1827 unlock:
1828         rtnl_unlock();
1829         return ret;
1830 }
1831
1832 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1833                             unsigned long arg, int ifreq_len)
1834 {
1835         struct tun_file *tfile = file->private_data;
1836         struct tun_struct *tun;
1837         void __user* argp = (void __user*)arg;
1838         struct ifreq ifr;
1839         kuid_t owner;
1840         kgid_t group;
1841         int sndbuf;
1842         int vnet_hdr_sz;
1843         int ret;
1844
1845         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1846                 if (copy_from_user(&ifr, argp, ifreq_len))
1847                         return -EFAULT;
1848         } else {
1849                 memset(&ifr, 0, sizeof(ifr));
1850         }
1851         if (cmd == TUNGETFEATURES) {
1852                 /* Currently this just means: "what IFF flags are valid?".
1853                  * This is needed because we never checked for invalid flags on
1854                  * TUNSETIFF. */
1855                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1856                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1857                                 (unsigned int __user*)argp);
1858         } else if (cmd == TUNSETQUEUE)
1859                 return tun_set_queue(file, &ifr);
1860
1861         ret = 0;
1862         rtnl_lock();
1863
1864         tun = __tun_get(tfile);
1865         if (cmd == TUNSETIFF && !tun) {
1866                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1867
1868                 ret = tun_set_iff(tfile->net, file, &ifr);
1869
1870                 if (ret)
1871                         goto unlock;
1872
1873                 if (copy_to_user(argp, &ifr, ifreq_len))
1874                         ret = -EFAULT;
1875                 goto unlock;
1876         }
1877
1878         ret = -EBADFD;
1879         if (!tun)
1880                 goto unlock;
1881
1882         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1883
1884         ret = 0;
1885         switch (cmd) {
1886         case TUNGETIFF:
1887                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1888
1889                 if (copy_to_user(argp, &ifr, ifreq_len))
1890                         ret = -EFAULT;
1891                 break;
1892
1893         case TUNSETNOCSUM:
1894                 /* Disable/Enable checksum */
1895
1896                 /* [unimplemented] */
1897                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1898                           arg ? "disabled" : "enabled");
1899                 break;
1900
1901         case TUNSETPERSIST:
1902                 /* Disable/Enable persist mode. Keep an extra reference to the
1903                  * module to prevent the module being unprobed.
1904                  */
1905                 if (arg && !(tun->flags & TUN_PERSIST)) {
1906                         tun->flags |= TUN_PERSIST;
1907                         __module_get(THIS_MODULE);
1908                 }
1909                 if (!arg && (tun->flags & TUN_PERSIST)) {
1910                         tun->flags &= ~TUN_PERSIST;
1911                         module_put(THIS_MODULE);
1912                 }
1913
1914                 tun_debug(KERN_INFO, tun, "persist %s\n",
1915                           arg ? "enabled" : "disabled");
1916                 break;
1917
1918         case TUNSETOWNER:
1919                 /* Set owner of the device */
1920                 owner = make_kuid(current_user_ns(), arg);
1921                 if (!uid_valid(owner)) {
1922                         ret = -EINVAL;
1923                         break;
1924                 }
1925                 tun->owner = owner;
1926                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1927                           from_kuid(&init_user_ns, tun->owner));
1928                 break;
1929
1930         case TUNSETGROUP:
1931                 /* Set group of the device */
1932                 group = make_kgid(current_user_ns(), arg);
1933                 if (!gid_valid(group)) {
1934                         ret = -EINVAL;
1935                         break;
1936                 }
1937                 tun->group = group;
1938                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1939                           from_kgid(&init_user_ns, tun->group));
1940                 break;
1941
1942         case TUNSETLINK:
1943                 /* Only allow setting the type when the interface is down */
1944                 if (tun->dev->flags & IFF_UP) {
1945                         tun_debug(KERN_INFO, tun,
1946                                   "Linktype set failed because interface is up\n");
1947                         ret = -EBUSY;
1948                 } else {
1949                         tun->dev->type = (int) arg;
1950                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1951                                   tun->dev->type);
1952                         ret = 0;
1953                 }
1954                 break;
1955
1956 #ifdef TUN_DEBUG
1957         case TUNSETDEBUG:
1958                 tun->debug = arg;
1959                 break;
1960 #endif
1961         case TUNSETOFFLOAD:
1962                 ret = set_offload(tun, arg);
1963                 break;
1964
1965         case TUNSETTXFILTER:
1966                 /* Can be set only for TAPs */
1967                 ret = -EINVAL;
1968                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1969                         break;
1970                 ret = update_filter(&tun->txflt, (void __user *)arg);
1971                 break;
1972
1973         case SIOCGIFHWADDR:
1974                 /* Get hw address */
1975                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1976                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1977                 if (copy_to_user(argp, &ifr, ifreq_len))
1978                         ret = -EFAULT;
1979                 break;
1980
1981         case SIOCSIFHWADDR:
1982                 /* Set hw address */
1983                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1984                           ifr.ifr_hwaddr.sa_data);
1985
1986                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1987                 break;
1988
1989         case TUNGETSNDBUF:
1990                 sndbuf = tfile->socket.sk->sk_sndbuf;
1991                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1992                         ret = -EFAULT;
1993                 break;
1994
1995         case TUNSETSNDBUF:
1996                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1997                         ret = -EFAULT;
1998                         break;
1999                 }
2000
2001                 tun->sndbuf = sndbuf;
2002                 tun_set_sndbuf(tun);
2003                 break;
2004
2005         case TUNGETVNETHDRSZ:
2006                 vnet_hdr_sz = tun->vnet_hdr_sz;
2007                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2008                         ret = -EFAULT;
2009                 break;
2010
2011         case TUNSETVNETHDRSZ:
2012                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2013                         ret = -EFAULT;
2014                         break;
2015                 }
2016                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2017                         ret = -EINVAL;
2018                         break;
2019                 }
2020
2021                 tun->vnet_hdr_sz = vnet_hdr_sz;
2022                 break;
2023
2024         case TUNATTACHFILTER:
2025                 /* Can be set only for TAPs */
2026                 ret = -EINVAL;
2027                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2028                         break;
2029                 ret = -EFAULT;
2030                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2031                         break;
2032
2033                 ret = tun_attach_filter(tun);
2034                 break;
2035
2036         case TUNDETACHFILTER:
2037                 /* Can be set only for TAPs */
2038                 ret = -EINVAL;
2039                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2040                         break;
2041                 ret = 0;
2042                 tun_detach_filter(tun, tun->numqueues);
2043                 break;
2044
2045         default:
2046                 ret = -EINVAL;
2047                 break;
2048         }
2049
2050 unlock:
2051         rtnl_unlock();
2052         if (tun)
2053                 tun_put(tun);
2054         return ret;
2055 }
2056
2057 static long tun_chr_ioctl(struct file *file,
2058                           unsigned int cmd, unsigned long arg)
2059 {
2060         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2061 }
2062
2063 #ifdef CONFIG_COMPAT
2064 static long tun_chr_compat_ioctl(struct file *file,
2065                          unsigned int cmd, unsigned long arg)
2066 {
2067         switch (cmd) {
2068         case TUNSETIFF:
2069         case TUNGETIFF:
2070         case TUNSETTXFILTER:
2071         case TUNGETSNDBUF:
2072         case TUNSETSNDBUF:
2073         case SIOCGIFHWADDR:
2074         case SIOCSIFHWADDR:
2075                 arg = (unsigned long)compat_ptr(arg);
2076                 break;
2077         default:
2078                 arg = (compat_ulong_t)arg;
2079                 break;
2080         }
2081
2082         /*
2083          * compat_ifreq is shorter than ifreq, so we must not access beyond
2084          * the end of that structure. All fields that are used in this
2085          * driver are compatible though, we don't need to convert the
2086          * contents.
2087          */
2088         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2089 }
2090 #endif /* CONFIG_COMPAT */
2091
2092 static int tun_chr_fasync(int fd, struct file *file, int on)
2093 {
2094         struct tun_file *tfile = file->private_data;
2095         int ret;
2096
2097         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2098                 goto out;
2099
2100         if (on) {
2101                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2102                 if (ret)
2103                         goto out;
2104                 tfile->flags |= TUN_FASYNC;
2105         } else
2106                 tfile->flags &= ~TUN_FASYNC;
2107         ret = 0;
2108 out:
2109         return ret;
2110 }
2111
2112 static int tun_chr_open(struct inode *inode, struct file * file)
2113 {
2114         struct tun_file *tfile;
2115
2116         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2117
2118         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2119                                             &tun_proto);
2120         if (!tfile)
2121                 return -ENOMEM;
2122         rcu_assign_pointer(tfile->tun, NULL);
2123         tfile->net = get_net(current->nsproxy->net_ns);
2124         tfile->flags = 0;
2125
2126         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2127         init_waitqueue_head(&tfile->wq.wait);
2128
2129         tfile->socket.file = file;
2130         tfile->socket.ops = &tun_socket_ops;
2131
2132         sock_init_data(&tfile->socket, &tfile->sk);
2133         sk_change_net(&tfile->sk, tfile->net);
2134
2135         tfile->sk.sk_write_space = tun_sock_write_space;
2136         tfile->sk.sk_sndbuf = INT_MAX;
2137
2138         file->private_data = tfile;
2139         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2140         INIT_LIST_HEAD(&tfile->next);
2141
2142         return 0;
2143 }
2144
2145 static int tun_chr_close(struct inode *inode, struct file *file)
2146 {
2147         struct tun_file *tfile = file->private_data;
2148         struct net *net = tfile->net;
2149
2150         tun_detach(tfile, true);
2151         put_net(net);
2152
2153         return 0;
2154 }
2155
2156 static const struct file_operations tun_fops = {
2157         .owner  = THIS_MODULE,
2158         .llseek = no_llseek,
2159         .read  = do_sync_read,
2160         .aio_read  = tun_chr_aio_read,
2161         .write = do_sync_write,
2162         .aio_write = tun_chr_aio_write,
2163         .poll   = tun_chr_poll,
2164         .unlocked_ioctl = tun_chr_ioctl,
2165 #ifdef CONFIG_COMPAT
2166         .compat_ioctl = tun_chr_compat_ioctl,
2167 #endif
2168         .open   = tun_chr_open,
2169         .release = tun_chr_close,
2170         .fasync = tun_chr_fasync
2171 };
2172
2173 static struct miscdevice tun_miscdev = {
2174         .minor = TUN_MINOR,
2175         .name = "tun",
2176         .nodename = "net/tun",
2177         .fops = &tun_fops,
2178 };
2179
2180 /* ethtool interface */
2181
2182 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2183 {
2184         cmd->supported          = 0;
2185         cmd->advertising        = 0;
2186         ethtool_cmd_speed_set(cmd, SPEED_10);
2187         cmd->duplex             = DUPLEX_FULL;
2188         cmd->port               = PORT_TP;
2189         cmd->phy_address        = 0;
2190         cmd->transceiver        = XCVR_INTERNAL;
2191         cmd->autoneg            = AUTONEG_DISABLE;
2192         cmd->maxtxpkt           = 0;
2193         cmd->maxrxpkt           = 0;
2194         return 0;
2195 }
2196
2197 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2198 {
2199         struct tun_struct *tun = netdev_priv(dev);
2200
2201         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2202         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2203
2204         switch (tun->flags & TUN_TYPE_MASK) {
2205         case TUN_TUN_DEV:
2206                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2207                 break;
2208         case TUN_TAP_DEV:
2209                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2210                 break;
2211         }
2212 }
2213
2214 static u32 tun_get_msglevel(struct net_device *dev)
2215 {
2216 #ifdef TUN_DEBUG
2217         struct tun_struct *tun = netdev_priv(dev);
2218         return tun->debug;
2219 #else
2220         return -EOPNOTSUPP;
2221 #endif
2222 }
2223
2224 static void tun_set_msglevel(struct net_device *dev, u32 value)
2225 {
2226 #ifdef TUN_DEBUG
2227         struct tun_struct *tun = netdev_priv(dev);
2228         tun->debug = value;
2229 #endif
2230 }
2231
2232 static const struct ethtool_ops tun_ethtool_ops = {
2233         .get_settings   = tun_get_settings,
2234         .get_drvinfo    = tun_get_drvinfo,
2235         .get_msglevel   = tun_get_msglevel,
2236         .set_msglevel   = tun_set_msglevel,
2237         .get_link       = ethtool_op_get_link,
2238 };
2239
2240
2241 static int __init tun_init(void)
2242 {
2243         int ret = 0;
2244
2245         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2246         pr_info("%s\n", DRV_COPYRIGHT);
2247
2248         ret = rtnl_link_register(&tun_link_ops);
2249         if (ret) {
2250                 pr_err("Can't register link_ops\n");
2251                 goto err_linkops;
2252         }
2253
2254         ret = misc_register(&tun_miscdev);
2255         if (ret) {
2256                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2257                 goto err_misc;
2258         }
2259         return  0;
2260 err_misc:
2261         rtnl_link_unregister(&tun_link_ops);
2262 err_linkops:
2263         return ret;
2264 }
2265
2266 static void tun_cleanup(void)
2267 {
2268         misc_deregister(&tun_miscdev);
2269         rtnl_link_unregister(&tun_link_ops);
2270 }
2271
2272 /* Get an underlying socket object from tun file.  Returns error unless file is
2273  * attached to a device.  The returned object works like a packet socket, it
2274  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2275  * holding a reference to the file for as long as the socket is in use. */
2276 struct socket *tun_get_socket(struct file *file)
2277 {
2278         struct tun_file *tfile;
2279         if (file->f_op != &tun_fops)
2280                 return ERR_PTR(-EINVAL);
2281         tfile = file->private_data;
2282         if (!tfile)
2283                 return ERR_PTR(-EBADFD);
2284         return &tfile->socket;
2285 }
2286 EXPORT_SYMBOL_GPL(tun_get_socket);
2287
2288 module_init(tun_init);
2289 module_exit(tun_cleanup);
2290 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2291 MODULE_AUTHOR(DRV_COPYRIGHT);
2292 MODULE_LICENSE("GPL");
2293 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2294 MODULE_ALIAS("devname:net/tun");