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
 201         hlist_for_each_entry_rcu(e, head, hash_link) {
 202                 if (e->rxhash == rxhash)
 203                         return e;
 204         }
 205         return NULL;
 206 }
 207
 208 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 209                                               struct hlist_head *head,
 210                                               u32 rxhash, u16 queue_index)
 211 {
 212         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 213
 214         if (e) {
 215                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 216                           rxhash, queue_index);
 217                 e->updated = jiffies;
 218                 e->rxhash = rxhash;
 219                 e->queue_index = queue_index;
 220                 e->tun = tun;
 221                 hlist_add_head_rcu(&e->hash_link, head);
 222                 ++tun->flow_count;
 223         }
 224         return e;
 225 }
 226
 227 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 228 {
 229         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 230                   e->rxhash, e->queue_index);
 231         hlist_del_rcu(&e->hash_link);
 232         kfree_rcu(e, rcu);
 233         --tun->flow_count;
 234 }
 235
 236 static void tun_flow_flush(struct tun_struct *tun)
 237 {
 238         int i;
 239
 240         spin_lock_bh(&tun->lock);
 241         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 242                 struct tun_flow_entry *e;
 243                 struct hlist_node *n;
 244
 245                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 246                         tun_flow_delete(tun, e);
 247         }
 248         spin_unlock_bh(&tun->lock);
 249 }
 250
 251 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 252 {
 253         int i;
 254
 255         spin_lock_bh(&tun->lock);
 256         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 257                 struct tun_flow_entry *e;
 258                 struct hlist_node *n;
 259
 260                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 261                         if (e->queue_index == queue_index)
 262                                 tun_flow_delete(tun, e);
 263                 }
 264         }
 265         spin_unlock_bh(&tun->lock);
 266 }
 267
 268 static void tun_flow_cleanup(unsigned long data)
 269 {
 270         struct tun_struct *tun = (struct tun_struct *)data;
 271         unsigned long delay = tun->ageing_time;
 272         unsigned long next_timer = jiffies + delay;
 273         unsigned long count = 0;
 274         int i;
 275
 276         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 277
 278         spin_lock_bh(&tun->lock);
 279         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 280                 struct tun_flow_entry *e;
 281                 struct hlist_node *n;
 282
 283                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 284                         unsigned long this_timer;
 285                         count++;
 286                         this_timer = e->updated + delay;
 287                         if (time_before_eq(this_timer, jiffies))
 288                                 tun_flow_delete(tun, e);
 289                         else if (time_before(this_timer, next_timer))
 290                                 next_timer = this_timer;
 291                 }
 292         }
 293
 294         if (count)
 295                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 296         spin_unlock_bh(&tun->lock);
 297 }
 298
 299 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 300                             struct tun_file *tfile)
 301 {
 302         struct hlist_head *head;
 303         struct tun_flow_entry *e;
 304         unsigned long delay = tun->ageing_time;
 305         u16 queue_index = tfile->queue_index;
 306
 307         if (!rxhash)
 308                 return;
 309         else
 310                 head = &tun->flows[tun_hashfn(rxhash)];
 311
 312         rcu_read_lock();
 313
 314         /* We may get a very small possibility of OOO during switching, not
 315          * worth to optimize.*/
 316         if (tun->numqueues == 1 || tfile->detached)
 317                 goto unlock;
 318
 319         e = tun_flow_find(head, rxhash);
 320         if (likely(e)) {
 321                 /* TODO: keep queueing to old queue until it's empty? */
 322                 e->queue_index = queue_index;
 323                 e->updated = jiffies;
 324         } else {
 325                 spin_lock_bh(&tun->lock);
 326                 if (!tun_flow_find(head, rxhash) &&
 327                     tun->flow_count < MAX_TAP_FLOWS)
 328                         tun_flow_create(tun, head, rxhash, queue_index);
 329
 330                 if (!timer_pending(&tun->flow_gc_timer))
 331                         mod_timer(&tun->flow_gc_timer,
 332                                   round_jiffies_up(jiffies + delay));
 333                 spin_unlock_bh(&tun->lock);
 334         }
 335
 336 unlock:
 337         rcu_read_unlock();
 338 }
 339
 340 /* We try to identify a flow through its rxhash first. The reason that
 341  * we do not check rxq no. is becuase some cards(e.g 82599), chooses
 342  * the rxq based on the txq where the last packet of the flow comes. As
 343  * the userspace application move between processors, we may get a
 344  * different rxq no. here. If we could not get rxhash, then we would
 345  * hope the rxq no. may help here.
 346  */
 347 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
 348 {
 349         struct tun_struct *tun = netdev_priv(dev);
 350         struct tun_flow_entry *e;
 351         u32 txq = 0;
 352         u32 numqueues = 0;
 353
 354         rcu_read_lock();
 355         numqueues = ACCESS_ONCE(tun->numqueues);
 356
 357         txq = skb_get_rxhash(skb);
 358         if (txq) {
 359                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 360                 if (e)
 361                         txq = e->queue_index;
 362                 else
 363                         /* use multiply and shift instead of expensive divide */
 364                         txq = ((u64)txq * numqueues) >> 32;
 365         } else if (likely(skb_rx_queue_recorded(skb))) {
 366                 txq = skb_get_rx_queue(skb);
 367                 while (unlikely(txq >= numqueues))
 368                         txq -= numqueues;
 369         }
 370
 371         rcu_read_unlock();
 372         return txq;
 373 }
 374
 375 static inline bool tun_not_capable(struct tun_struct *tun)
 376 {
 377         const struct cred *cred = current_cred();
 378         struct net *net = dev_net(tun->dev);
 379
 380         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 381                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 382                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 383 }
 384
 385 static void tun_set_real_num_queues(struct tun_struct *tun)
 386 {
 387         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 388         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 389 }
 390
 391 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 392 {
 393         tfile->detached = tun;
 394         list_add_tail(&tfile->next, &tun->disabled);
 395         ++tun->numdisabled;
 396 }
 397
 398 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 399 {
 400         struct tun_struct *tun = tfile->detached;
 401
 402         tfile->detached = NULL;
 403         list_del_init(&tfile->next);
 404         --tun->numdisabled;
 405         return tun;
 406 }
 407
 408 static void __tun_detach(struct tun_file *tfile, bool clean)
 409 {
 410         struct tun_file *ntfile;
 411         struct tun_struct *tun;
 412
 413         tun = rtnl_dereference(tfile->tun);
 414
 415         if (tun && !tfile->detached) {
 416                 u16 index = tfile->queue_index;
 417                 BUG_ON(index >= tun->numqueues);
 418
 419                 rcu_assign_pointer(tun->tfiles[index],
 420                                    tun->tfiles[tun->numqueues - 1]);
 421                 ntfile = rtnl_dereference(tun->tfiles[index]);
 422                 ntfile->queue_index = index;
 423
 424                 --tun->numqueues;
 425                 if (clean) {
 426                         rcu_assign_pointer(tfile->tun, NULL);
 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                         netif_carrier_off(tun->dev);
 444
 445                         if (!(tun->flags & TUN_PERSIST) &&
 446                             tun->dev->reg_state == NETREG_REGISTERED)
 447                                 unregister_netdevice(tun->dev);
 448                 }
 449
 450                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 451                                  &tfile->socket.flags));
 452                 sk_release_kernel(&tfile->sk);
 453         }
 454 }
 455
 456 static void tun_detach(struct tun_file *tfile, bool clean)
 457 {
 458         rtnl_lock();
 459         __tun_detach(tfile, clean);
 460         rtnl_unlock();
 461 }
 462
 463 static void tun_detach_all(struct net_device *dev)
 464 {
 465         struct tun_struct *tun = netdev_priv(dev);
 466         struct tun_file *tfile, *tmp;
 467         int i, n = tun->numqueues;
 468
 469         for (i = 0; i < n; i++) {
 470                 tfile = rtnl_dereference(tun->tfiles[i]);
 471                 BUG_ON(!tfile);
 472                 wake_up_all(&tfile->wq.wait);
 473                 rcu_assign_pointer(tfile->tun, NULL);
 474                 --tun->numqueues;
 475         }
 476         list_for_each_entry(tfile, &tun->disabled, next) {
 477                 wake_up_all(&tfile->wq.wait);
 478                 rcu_assign_pointer(tfile->tun, NULL);
 479         }
 480         BUG_ON(tun->numqueues != 0);
 481
 482         synchronize_net();
 483         for (i = 0; i < n; i++) {
 484                 tfile = rtnl_dereference(tun->tfiles[i]);
 485                 /* Drop read queue */
 486                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 487                 sock_put(&tfile->sk);
 488         }
 489         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 490                 tun_enable_queue(tfile);
 491                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 492                 sock_put(&tfile->sk);
 493         }
 494         BUG_ON(tun->numdisabled != 0);
 495
 496         if (tun->flags & TUN_PERSIST)
 497                 module_put(THIS_MODULE);
 498 }
 499
 500 static int tun_attach(struct tun_struct *tun, struct file *file)
 501 {
 502         struct tun_file *tfile = file->private_data;
 503         int err;
 504
 505         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 506         if (err < 0)
 507                 goto out;
 508
 509         err = -EINVAL;
 510         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 511                 goto out;
 512
 513         err = -EBUSY;
 514         if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 515                 goto out;
 516
 517         err = -E2BIG;
 518         if (!tfile->detached &&
 519             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 520                 goto out;
 521
 522         err = 0;
 523
 524         /* Re-attach the filter to presist device */
 525         if (tun->filter_attached == true) {
 526                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 527                 if (!err)
 528                         goto out;
 529         }
 530         tfile->queue_index = tun->numqueues;
 531         rcu_assign_pointer(tfile->tun, tun);
 532         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 533         tun->numqueues++;
 534
 535         if (tfile->detached)
 536                 tun_enable_queue(tfile);
 537         else
 538                 sock_hold(&tfile->sk);
 539
 540         tun_set_real_num_queues(tun);
 541
 542         /* device is allowed to go away first, so no need to hold extra
 543          * refcnt.
 544          */
 545
 546 out:
 547         return err;
 548 }
 549
 550 static struct tun_struct *__tun_get(struct tun_file *tfile)
 551 {
 552         struct tun_struct *tun;
 553
 554         rcu_read_lock();
 555         tun = rcu_dereference(tfile->tun);
 556         if (tun)
 557                 dev_hold(tun->dev);
 558         rcu_read_unlock();
 559
 560         return tun;
 561 }
 562
 563 static struct tun_struct *tun_get(struct file *file)
 564 {
 565         return __tun_get(file->private_data);
 566 }
 567
 568 static void tun_put(struct tun_struct *tun)
 569 {
 570         dev_put(tun->dev);
 571 }
 572
 573 /* TAP filtering */
 574 static void addr_hash_set(u32 *mask, const u8 *addr)
 575 {
 576         int n = ether_crc(ETH_ALEN, addr) >> 26;
 577         mask[n >> 5] |= (1 << (n & 31));
 578 }
 579
 580 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 581 {
 582         int n = ether_crc(ETH_ALEN, addr) >> 26;
 583         return mask[n >> 5] & (1 << (n & 31));
 584 }
 585
 586 static int update_filter(struct tap_filter *filter, void __user *arg)
 587 {
 588         struct { u8 u[ETH_ALEN]; } *addr;
 589         struct tun_filter uf;
 590         int err, alen, n, nexact;
 591
 592         if (copy_from_user(&uf, arg, sizeof(uf)))
 593                 return -EFAULT;
 594
 595         if (!uf.count) {
 596                 /* Disabled */
 597                 filter->count = 0;
 598                 return 0;
 599         }
 600
 601         alen = ETH_ALEN * uf.count;
 602         addr = kmalloc(alen, GFP_KERNEL);
 603         if (!addr)
 604                 return -ENOMEM;
 605
 606         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 607                 err = -EFAULT;
 608                 goto done;
 609         }
 610
 611         /* The filter is updated without holding any locks. Which is
 612          * perfectly safe. We disable it first and in the worst
 613          * case we'll accept a few undesired packets. */
 614         filter->count = 0;
 615         wmb();
 616
 617         /* Use first set of addresses as an exact filter */
 618         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 619                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 620
 621         nexact = n;
 622
 623         /* Remaining multicast addresses are hashed,
 624          * unicast will leave the filter disabled. */
 625         memset(filter->mask, 0, sizeof(filter->mask));
 626         for (; n < uf.count; n++) {
 627                 if (!is_multicast_ether_addr(addr[n].u)) {
 628                         err = 0; /* no filter */
 629                         goto done;
 630                 }
 631                 addr_hash_set(filter->mask, addr[n].u);
 632         }
 633
 634         /* For ALLMULTI just set the mask to all ones.
 635          * This overrides the mask populated above. */
 636         if ((uf.flags & TUN_FLT_ALLMULTI))
 637                 memset(filter->mask, ~0, sizeof(filter->mask));
 638
 639         /* Now enable the filter */
 640         wmb();
 641         filter->count = nexact;
 642
 643         /* Return the number of exact filters */
 644         err = nexact;
 645
 646 done:
 647         kfree(addr);
 648         return err;
 649 }
 650
 651 /* Returns: 0 - drop, !=0 - accept */
 652 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 653 {
 654         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 655          * at this point. */
 656         struct ethhdr *eh = (struct ethhdr *) skb->data;
 657         int i;
 658
 659         /* Exact match */
 660         for (i = 0; i < filter->count; i++)
 661                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 662                         return 1;
 663
 664         /* Inexact match (multicast only) */
 665         if (is_multicast_ether_addr(eh->h_dest))
 666                 return addr_hash_test(filter->mask, eh->h_dest);
 667
 668         return 0;
 669 }
 670
 671 /*
 672  * Checks whether the packet is accepted or not.
 673  * Returns: 0 - drop, !=0 - accept
 674  */
 675 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 676 {
 677         if (!filter->count)
 678                 return 1;
 679
 680         return run_filter(filter, skb);
 681 }
 682
 683 /* Network device part of the driver */
 684
 685 static const struct ethtool_ops tun_ethtool_ops;
 686
 687 /* Net device detach from fd. */
 688 static void tun_net_uninit(struct net_device *dev)
 689 {
 690         tun_detach_all(dev);
 691 }
 692
 693 /* Net device open. */
 694 static int tun_net_open(struct net_device *dev)
 695 {
 696         netif_tx_start_all_queues(dev);
 697         return 0;
 698 }
 699
 700 /* Net device close. */
 701 static int tun_net_close(struct net_device *dev)
 702 {
 703         netif_tx_stop_all_queues(dev);
 704         return 0;
 705 }
 706
 707 /* Net device start xmit */
 708 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 709 {
 710         struct tun_struct *tun = netdev_priv(dev);
 711         int txq = skb->queue_mapping;
 712         struct tun_file *tfile;
 713
 714         rcu_read_lock();
 715         tfile = rcu_dereference(tun->tfiles[txq]);
 716
 717         /* Drop packet if interface is not attached */
 718         if (txq >= tun->numqueues)
 719                 goto drop;
 720
 721         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 722
 723         BUG_ON(!tfile);
 724
 725         /* Drop if the filter does not like it.
 726          * This is a noop if the filter is disabled.
 727          * Filter can be enabled only for the TAP devices. */
 728         if (!check_filter(&tun->txflt, skb))
 729                 goto drop;
 730
 731         if (tfile->socket.sk->sk_filter &&
 732             sk_filter(tfile->socket.sk, skb))
 733                 goto drop;
 734
 735         /* Limit the number of packets queued by dividing txq length with the
 736          * number of queues.
 737          */
 738         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
 739                           >= dev->tx_queue_len / tun->numqueues)
 740                 goto drop;
 741
 742         /* Orphan the skb - required as we might hang on to it
 743          * for indefinite time. */
 744         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 745                 goto drop;
 746         skb_orphan(skb);
 747
 748         nf_reset(skb);
 749
 750         /* Enqueue packet */
 751         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 752
 753         /* Notify and wake up reader process */
 754         if (tfile->flags & TUN_FASYNC)
 755                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 756         wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
 757                                    POLLRDNORM | POLLRDBAND);
 758
 759         rcu_read_unlock();
 760         return NETDEV_TX_OK;
 761
 762 drop:
 763         dev->stats.tx_dropped++;
 764         skb_tx_error(skb);
 765         kfree_skb(skb);
 766         rcu_read_unlock();
 767         return NETDEV_TX_OK;
 768 }
 769
 770 static void tun_net_mclist(struct net_device *dev)
 771 {
 772         /*
 773          * This callback is supposed to deal with mc filter in
 774          * _rx_ path and has nothing to do with the _tx_ path.
 775          * In rx path we always accept everything userspace gives us.
 776          */
 777 }
 778
 779 #define MIN_MTU 68
 780 #define MAX_MTU 65535
 781
 782 static int
 783 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 784 {
 785         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 786                 return -EINVAL;
 787         dev->mtu = new_mtu;
 788         return 0;
 789 }
 790
 791 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 792         netdev_features_t features)
 793 {
 794         struct tun_struct *tun = netdev_priv(dev);
 795
 796         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 797 }
 798 #ifdef CONFIG_NET_POLL_CONTROLLER
 799 static void tun_poll_controller(struct net_device *dev)
 800 {
 801         /*
 802          * Tun only receives frames when:
 803          * 1) the char device endpoint gets data from user space
 804          * 2) the tun socket gets a sendmsg call from user space
 805          * Since both of those are syncronous operations, we are guaranteed
 806          * never to have pending data when we poll for it
 807          * so theres nothing to do here but return.
 808          * We need this though so netpoll recognizes us as an interface that
 809          * supports polling, which enables bridge devices in virt setups to
 810          * still use netconsole
 811          */
 812         return;
 813 }
 814 #endif
 815 static const struct net_device_ops tun_netdev_ops = {
 816         .ndo_uninit             = tun_net_uninit,
 817         .ndo_open               = tun_net_open,
 818         .ndo_stop               = tun_net_close,
 819         .ndo_start_xmit         = tun_net_xmit,
 820         .ndo_change_mtu         = tun_net_change_mtu,
 821         .ndo_fix_features       = tun_net_fix_features,
 822         .ndo_select_queue       = tun_select_queue,
 823 #ifdef CONFIG_NET_POLL_CONTROLLER
 824         .ndo_poll_controller    = tun_poll_controller,
 825 #endif
 826 };
 827
 828 static const struct net_device_ops tap_netdev_ops = {
 829         .ndo_uninit             = tun_net_uninit,
 830         .ndo_open               = tun_net_open,
 831         .ndo_stop               = tun_net_close,
 832         .ndo_start_xmit         = tun_net_xmit,
 833         .ndo_change_mtu         = tun_net_change_mtu,
 834         .ndo_fix_features       = tun_net_fix_features,
 835         .ndo_set_rx_mode        = tun_net_mclist,
 836         .ndo_set_mac_address    = eth_mac_addr,
 837         .ndo_validate_addr      = eth_validate_addr,
 838         .ndo_select_queue       = tun_select_queue,
 839 #ifdef CONFIG_NET_POLL_CONTROLLER
 840         .ndo_poll_controller    = tun_poll_controller,
 841 #endif
 842 };
 843
 844 static void tun_flow_init(struct tun_struct *tun)
 845 {
 846         int i;
 847
 848         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 849                 INIT_HLIST_HEAD(&tun->flows[i]);
 850
 851         tun->ageing_time = TUN_FLOW_EXPIRE;
 852         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 853         mod_timer(&tun->flow_gc_timer,
 854                   round_jiffies_up(jiffies + tun->ageing_time));
 855 }
 856
 857 static void tun_flow_uninit(struct tun_struct *tun)
 858 {
 859         del_timer_sync(&tun->flow_gc_timer);
 860         tun_flow_flush(tun);
 861 }
 862
 863 /* Initialize net device. */
 864 static void tun_net_init(struct net_device *dev)
 865 {
 866         struct tun_struct *tun = netdev_priv(dev);
 867
 868         switch (tun->flags & TUN_TYPE_MASK) {
 869         case TUN_TUN_DEV:
 870                 dev->netdev_ops = &tun_netdev_ops;
 871
 872                 /* Point-to-Point TUN Device */
 873                 dev->hard_header_len = 0;
 874                 dev->addr_len = 0;
 875                 dev->mtu = 1500;
 876
 877                 /* Zero header length */
 878                 dev->type = ARPHRD_NONE;
 879                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 880                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 881                 break;
 882
 883         case TUN_TAP_DEV:
 884                 dev->netdev_ops = &tap_netdev_ops;
 885                 /* Ethernet TAP Device */
 886                 ether_setup(dev);
 887                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 888                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 889
 890                 eth_hw_addr_random(dev);
 891
 892                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 893                 break;
 894         }
 895 }
 896
 897 /* Character device part */
 898
 899 /* Poll */
 900 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 901 {
 902         struct tun_file *tfile = file->private_data;
 903         struct tun_struct *tun = __tun_get(tfile);
 904         struct sock *sk;
 905         unsigned int mask = 0;
 906
 907         if (!tun)
 908                 return POLLERR;
 909
 910         sk = tfile->socket.sk;
 911
 912         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 913
 914         poll_wait(file, &tfile->wq.wait, wait);
 915
 916         if (!skb_queue_empty(&sk->sk_receive_queue))
 917                 mask |= POLLIN | POLLRDNORM;
 918
 919         if (sock_writeable(sk) ||
 920             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 921              sock_writeable(sk)))
 922                 mask |= POLLOUT | POLLWRNORM;
 923
 924         if (tun->dev->reg_state != NETREG_REGISTERED)
 925                 mask = POLLERR;
 926
 927         tun_put(tun);
 928         return mask;
 929 }
 930
 931 /* prepad is the amount to reserve at front.  len is length after that.
 932  * linear is a hint as to how much to copy (usually headers). */
 933 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 934                                      size_t prepad, size_t len,
 935                                      size_t linear, int noblock)
 936 {
 937         struct sock *sk = tfile->socket.sk;
 938         struct sk_buff *skb;
 939         int err;
 940
 941         /* Under a page?  Don't bother with paged skb. */
 942         if (prepad + len < PAGE_SIZE || !linear)
 943                 linear = len;
 944
 945         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 946                                    &err);
 947         if (!skb)
 948                 return ERR_PTR(err);
 949
 950         skb_reserve(skb, prepad);
 951         skb_put(skb, linear);
 952         skb->data_len = len - linear;
 953         skb->len += len - linear;
 954
 955         return skb;
 956 }
 957
 958 /* set skb frags from iovec, this can move to core network code for reuse */
 959 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 960                                   int offset, size_t count)
 961 {
 962         int len = iov_length(from, count) - offset;
 963         int copy = skb_headlen(skb);
 964         int size, offset1 = 0;
 965         int i = 0;
 966
 967         /* Skip over from offset */
 968         while (count && (offset >= from->iov_len)) {
 969                 offset -= from->iov_len;
 970                 ++from;
 971                 --count;
 972         }
 973
 974         /* copy up to skb headlen */
 975         while (count && (copy > 0)) {
 976                 size = min_t(unsigned int, copy, from->iov_len - offset);
 977                 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
 978                                    size))
 979                         return -EFAULT;
 980                 if (copy > size) {
 981                         ++from;
 982                         --count;
 983                         offset = 0;
 984                 } else
 985                         offset += size;
 986                 copy -= size;
 987                 offset1 += size;
 988         }
 989
 990         if (len == offset1)
 991                 return 0;
 992
 993         while (count--) {
 994                 struct page *page[MAX_SKB_FRAGS];
 995                 int num_pages;
 996                 unsigned long base;
 997                 unsigned long truesize;
 998
 999                 len = from->iov_len - offset;
1000                 if (!len) {
1001                         offset = 0;
1002                         ++from;
1003                         continue;
1004                 }
1005                 base = (unsigned long)from->iov_base + offset;
1006                 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
1007                 if (i + size > MAX_SKB_FRAGS)
1008                         return -EMSGSIZE;
1009                 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
1010                 if (num_pages != size) {
1011                         int j;
1012
1013                         for (j = 0; j < num_pages; j++)
1014                                 put_page(page[i + j]);
1015                         return -EFAULT;
1016                 }
1017                 truesize = size * PAGE_SIZE;
1018                 skb->data_len += len;
1019                 skb->len += len;
1020                 skb->truesize += truesize;
1021                 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
1022                 while (len) {
1023                         int off = base & ~PAGE_MASK;
1024                         int size = min_t(int, len, PAGE_SIZE - off);
1025                         __skb_fill_page_desc(skb, i, page[i], off, size);
1026                         skb_shinfo(skb)->nr_frags++;
1027                         /* increase sk_wmem_alloc */
1028                         base += size;
1029                         len -= size;
1030                         i++;
1031                 }
1032                 offset = 0;
1033                 ++from;
1034         }
1035         return 0;
1036 }
1037
1038 /* Get packet from user space buffer */
1039 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1040                             void *msg_control, const struct iovec *iv,
1041                             size_t total_len, size_t count, int noblock)
1042 {
1043         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1044         struct sk_buff *skb;
1045         size_t len = total_len, align = NET_SKB_PAD, linear;
1046         struct virtio_net_hdr gso = { 0 };
1047         int offset = 0;
1048         int copylen;
1049         bool zerocopy = false;
1050         int err;
1051         u32 rxhash;
1052
1053         if (!(tun->flags & TUN_NO_PI)) {
1054                 if ((len -= sizeof(pi)) > total_len)
1055                         return -EINVAL;
1056
1057                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1058                         return -EFAULT;
1059                 offset += sizeof(pi);
1060         }
1061
1062         if (tun->flags & TUN_VNET_HDR) {
1063                 if ((len -= tun->vnet_hdr_sz) > total_len)
1064                         return -EINVAL;
1065
1066                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1067                         return -EFAULT;
1068
1069                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1070                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1071                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1072
1073                 if (gso.hdr_len > len)
1074                         return -EINVAL;
1075                 offset += tun->vnet_hdr_sz;
1076         }
1077
1078         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1079                 align += NET_IP_ALIGN;
1080                 if (unlikely(len < ETH_HLEN ||
1081                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1082                         return -EINVAL;
1083         }
1084
1085         if (msg_control)
1086                 zerocopy = true;
1087
1088         if (zerocopy) {
1089                 /* Userspace may produce vectors with count greater than
1090                  * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1091                  * to let the rest of data to be fit in the frags.
1092                  */
1093                 if (count > MAX_SKB_FRAGS) {
1094                         copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1095                         if (copylen < offset)
1096                                 copylen = 0;
1097                         else
1098                                 copylen -= offset;
1099                 } else
1100                                 copylen = 0;
1101                 /* There are 256 bytes to be copied in skb, so there is enough
1102                  * room for skb expand head in case it is used.
1103                  * The rest of the buffer is mapped from userspace.
1104                  */
1105                 if (copylen < gso.hdr_len)
1106                         copylen = gso.hdr_len;
1107                 if (!copylen)
1108                         copylen = GOODCOPY_LEN;
1109                 linear = copylen;
1110         } else {
1111                 copylen = len;
1112                 linear = gso.hdr_len;
1113         }
1114
1115         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1116         if (IS_ERR(skb)) {
1117                 if (PTR_ERR(skb) != -EAGAIN)
1118                         tun->dev->stats.rx_dropped++;
1119                 return PTR_ERR(skb);
1120         }
1121
1122         if (zerocopy)
1123                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1124         else
1125                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1126
1127         if (err) {
1128                 tun->dev->stats.rx_dropped++;
1129                 kfree_skb(skb);
1130                 return -EFAULT;
1131         }
1132
1133         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1134                 if (!skb_partial_csum_set(skb, gso.csum_start,
1135                                           gso.csum_offset)) {
1136                         tun->dev->stats.rx_frame_errors++;
1137                         kfree_skb(skb);
1138                         return -EINVAL;
1139                 }
1140         }
1141
1142         switch (tun->flags & TUN_TYPE_MASK) {
1143         case TUN_TUN_DEV:
1144                 if (tun->flags & TUN_NO_PI) {
1145                         switch (skb->data[0] & 0xf0) {
1146                         case 0x40:
1147                                 pi.proto = htons(ETH_P_IP);
1148                                 break;
1149                         case 0x60:
1150                                 pi.proto = htons(ETH_P_IPV6);
1151                                 break;
1152                         default:
1153                                 tun->dev->stats.rx_dropped++;
1154                                 kfree_skb(skb);
1155                                 return -EINVAL;
1156                         }
1157                 }
1158
1159                 skb_reset_mac_header(skb);
1160                 skb->protocol = pi.proto;
1161                 skb->dev = tun->dev;
1162                 break;
1163         case TUN_TAP_DEV:
1164                 skb->protocol = eth_type_trans(skb, tun->dev);
1165                 break;
1166         }
1167
1168         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1169                 pr_debug("GSO!\n");
1170                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1171                 case VIRTIO_NET_HDR_GSO_TCPV4:
1172                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1173                         break;
1174                 case VIRTIO_NET_HDR_GSO_TCPV6:
1175                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1176                         break;
1177                 case VIRTIO_NET_HDR_GSO_UDP:
1178                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1179                         break;
1180                 default:
1181                         tun->dev->stats.rx_frame_errors++;
1182                         kfree_skb(skb);
1183                         return -EINVAL;
1184                 }
1185
1186                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1187                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1188
1189                 skb_shinfo(skb)->gso_size = gso.gso_size;
1190                 if (skb_shinfo(skb)->gso_size == 0) {
1191                         tun->dev->stats.rx_frame_errors++;
1192                         kfree_skb(skb);
1193                         return -EINVAL;
1194                 }
1195
1196                 /* Header must be checked, and gso_segs computed. */
1197                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1198                 skb_shinfo(skb)->gso_segs = 0;
1199         }
1200
1201         /* copy skb_ubuf_info for callback when skb has no error */
1202         if (zerocopy) {
1203                 skb_shinfo(skb)->destructor_arg = msg_control;
1204                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1205                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1206         }
1207
1208         skb_reset_network_header(skb);
1209         skb_probe_transport_header(skb, 0);
1210
1211         rxhash = skb_get_rxhash(skb);
1212         netif_rx_ni(skb);
1213
1214         tun->dev->stats.rx_packets++;
1215         tun->dev->stats.rx_bytes += len;
1216
1217         tun_flow_update(tun, rxhash, tfile);
1218         return total_len;
1219 }
1220
1221 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1222                               unsigned long count, loff_t pos)
1223 {
1224         struct file *file = iocb->ki_filp;
1225         struct tun_struct *tun = tun_get(file);
1226         struct tun_file *tfile = file->private_data;
1227         ssize_t result;
1228
1229         if (!tun)
1230                 return -EBADFD;
1231
1232         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1233
1234         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1235                               count, file->f_flags & O_NONBLOCK);
1236
1237         tun_put(tun);
1238         return result;
1239 }
1240
1241 /* Put packet to the user space buffer */
1242 static ssize_t tun_put_user(struct tun_struct *tun,
1243                             struct tun_file *tfile,
1244                             struct sk_buff *skb,
1245                             const struct iovec *iv, int len)
1246 {
1247         struct tun_pi pi = { 0, skb->protocol };
1248         ssize_t total = 0;
1249
1250         if (!(tun->flags & TUN_NO_PI)) {
1251                 if ((len -= sizeof(pi)) < 0)
1252                         return -EINVAL;
1253
1254                 if (len < skb->len) {
1255                         /* Packet will be striped */
1256                         pi.flags |= TUN_PKT_STRIP;
1257                 }
1258
1259                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1260                         return -EFAULT;
1261                 total += sizeof(pi);
1262         }
1263
1264         if (tun->flags & TUN_VNET_HDR) {
1265                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1266                 if ((len -= tun->vnet_hdr_sz) < 0)
1267                         return -EINVAL;
1268
1269                 if (skb_is_gso(skb)) {
1270                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1271
1272                         /* This is a hint as to how much should be linear. */
1273                         gso.hdr_len = skb_headlen(skb);
1274                         gso.gso_size = sinfo->gso_size;
1275                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1276                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1277                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1278                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1279                         else if (sinfo->gso_type & SKB_GSO_UDP)
1280                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1281                         else {
1282                                 pr_err("unexpected GSO type: "
1283                                        "0x%x, gso_size %d, hdr_len %d\n",
1284                                        sinfo->gso_type, gso.gso_size,
1285                                        gso.hdr_len);
1286                                 print_hex_dump(KERN_ERR, "tun: ",
1287                                                DUMP_PREFIX_NONE,
1288                                                16, 1, skb->head,
1289                                                min((int)gso.hdr_len, 64), true);
1290                                 WARN_ON_ONCE(1);
1291                                 return -EINVAL;
1292                         }
1293                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1294                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1295                 } else
1296                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1297
1298                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1299                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1300                         gso.csum_start = skb_checksum_start_offset(skb);
1301                         gso.csum_offset = skb->csum_offset;
1302                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1303                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1304                 } /* else everything is zero */
1305
1306                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1307                                                sizeof(gso))))
1308                         return -EFAULT;
1309                 total += tun->vnet_hdr_sz;
1310         }
1311
1312         len = min_t(int, skb->len, len);
1313
1314         skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1315         total += skb->len;
1316
1317         tun->dev->stats.tx_packets++;
1318         tun->dev->stats.tx_bytes += len;
1319
1320         return total;
1321 }
1322
1323 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1324                            struct kiocb *iocb, const struct iovec *iv,
1325                            ssize_t len, int noblock)
1326 {
1327         DECLARE_WAITQUEUE(wait, current);
1328         struct sk_buff *skb;
1329         ssize_t ret = 0;
1330
1331         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1332
1333         if (unlikely(!noblock))
1334                 add_wait_queue(&tfile->wq.wait, &wait);
1335         while (len) {
1336                 current->state = TASK_INTERRUPTIBLE;
1337
1338                 /* Read frames from the queue */
1339                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1340                         if (noblock) {
1341                                 ret = -EAGAIN;
1342                                 break;
1343                         }
1344                         if (signal_pending(current)) {
1345                                 ret = -ERESTARTSYS;
1346                                 break;
1347                         }
1348                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1349                                 ret = -EIO;
1350                                 break;
1351                         }
1352
1353                         /* Nothing to read, let's sleep */
1354                         schedule();
1355                         continue;
1356                 }
1357
1358                 ret = tun_put_user(tun, tfile, skb, iv, len);
1359                 kfree_skb(skb);
1360                 break;
1361         }
1362
1363         current->state = TASK_RUNNING;
1364         if (unlikely(!noblock))
1365                 remove_wait_queue(&tfile->wq.wait, &wait);
1366
1367         return ret;
1368 }
1369
1370 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1371                             unsigned long count, loff_t pos)
1372 {
1373         struct file *file = iocb->ki_filp;
1374         struct tun_file *tfile = file->private_data;
1375         struct tun_struct *tun = __tun_get(tfile);
1376         ssize_t len, ret;
1377
1378         if (!tun)
1379                 return -EBADFD;
1380         len = iov_length(iv, count);
1381         if (len < 0) {
1382                 ret = -EINVAL;
1383                 goto out;
1384         }
1385
1386         ret = tun_do_read(tun, tfile, iocb, iv, len,
1387                           file->f_flags & O_NONBLOCK);
1388         ret = min_t(ssize_t, ret, len);
1389 out:
1390         tun_put(tun);
1391         return ret;
1392 }
1393
1394 static void tun_free_netdev(struct net_device *dev)
1395 {
1396         struct tun_struct *tun = netdev_priv(dev);
1397
1398         BUG_ON(!(list_empty(&tun->disabled)));
1399         tun_flow_uninit(tun);
1400         security_tun_dev_free_security(tun->security);
1401         free_netdev(dev);
1402 }
1403
1404 static void tun_setup(struct net_device *dev)
1405 {
1406         struct tun_struct *tun = netdev_priv(dev);
1407
1408         tun->owner = INVALID_UID;
1409         tun->group = INVALID_GID;
1410
1411         dev->ethtool_ops = &tun_ethtool_ops;
1412         dev->destructor = tun_free_netdev;
1413 }
1414
1415 /* Trivial set of netlink ops to allow deleting tun or tap
1416  * device with netlink.
1417  */
1418 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1419 {
1420         return -EINVAL;
1421 }
1422
1423 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1424         .kind           = DRV_NAME,
1425         .priv_size      = sizeof(struct tun_struct),
1426         .setup          = tun_setup,
1427         .validate       = tun_validate,
1428 };
1429
1430 static void tun_sock_write_space(struct sock *sk)
1431 {
1432         struct tun_file *tfile;
1433         wait_queue_head_t *wqueue;
1434
1435         if (!sock_writeable(sk))
1436                 return;
1437
1438         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1439                 return;
1440
1441         wqueue = sk_sleep(sk);
1442         if (wqueue && waitqueue_active(wqueue))
1443                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1444                                                 POLLWRNORM | POLLWRBAND);
1445
1446         tfile = container_of(sk, struct tun_file, sk);
1447         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1448 }
1449
1450 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1451                        struct msghdr *m, size_t total_len)
1452 {
1453         int ret;
1454         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1455         struct tun_struct *tun = __tun_get(tfile);
1456
1457         if (!tun)
1458                 return -EBADFD;
1459         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1460                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1461         tun_put(tun);
1462         return ret;
1463 }
1464
1465
1466 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1467                        struct msghdr *m, size_t total_len,
1468                        int flags)
1469 {
1470         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1471         struct tun_struct *tun = __tun_get(tfile);
1472         int ret;
1473
1474         if (!tun)
1475                 return -EBADFD;
1476
1477         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) {
1478                 ret = -EINVAL;
1479                 goto out;
1480         }
1481         ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1482                           flags & MSG_DONTWAIT);
1483         if (ret > total_len) {
1484                 m->msg_flags |= MSG_TRUNC;
1485                 ret = flags & MSG_TRUNC ? ret : total_len;
1486         }
1487 out:
1488         tun_put(tun);
1489         return ret;
1490 }
1491
1492 static int tun_release(struct socket *sock)
1493 {
1494         if (sock->sk)
1495                 sock_put(sock->sk);
1496         return 0;
1497 }
1498
1499 /* Ops structure to mimic raw sockets with tun */
1500 static const struct proto_ops tun_socket_ops = {
1501         .sendmsg = tun_sendmsg,
1502         .recvmsg = tun_recvmsg,
1503         .release = tun_release,
1504 };
1505
1506 static struct proto tun_proto = {
1507         .name           = "tun",
1508         .owner          = THIS_MODULE,
1509         .obj_size       = sizeof(struct tun_file),
1510 };
1511
1512 static int tun_flags(struct tun_struct *tun)
1513 {
1514         int flags = 0;
1515
1516         if (tun->flags & TUN_TUN_DEV)
1517                 flags |= IFF_TUN;
1518         else
1519                 flags |= IFF_TAP;
1520
1521         if (tun->flags & TUN_NO_PI)
1522                 flags |= IFF_NO_PI;
1523
1524         /* This flag has no real effect.  We track the value for backwards
1525          * compatibility.
1526          */
1527         if (tun->flags & TUN_ONE_QUEUE)
1528                 flags |= IFF_ONE_QUEUE;
1529
1530         if (tun->flags & TUN_VNET_HDR)
1531                 flags |= IFF_VNET_HDR;
1532
1533         if (tun->flags & TUN_TAP_MQ)
1534                 flags |= IFF_MULTI_QUEUE;
1535
1536         if (tun->flags & TUN_PERSIST)
1537                 flags |= IFF_PERSIST;
1538
1539         return flags;
1540 }
1541
1542 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1543                               char *buf)
1544 {
1545         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1546         return sprintf(buf, "0x%x\n", tun_flags(tun));
1547 }
1548
1549 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1550                               char *buf)
1551 {
1552         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1553         return uid_valid(tun->owner)?
1554                 sprintf(buf, "%u\n",
1555                         from_kuid_munged(current_user_ns(), tun->owner)):
1556                 sprintf(buf, "-1\n");
1557 }
1558
1559 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1560                               char *buf)
1561 {
1562         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1563         return gid_valid(tun->group) ?
1564                 sprintf(buf, "%u\n",
1565                         from_kgid_munged(current_user_ns(), tun->group)):
1566                 sprintf(buf, "-1\n");
1567 }
1568
1569 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1570 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1571 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1572
1573 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1574 {
1575         struct tun_struct *tun;
1576         struct tun_file *tfile = file->private_data;
1577         struct net_device *dev;
1578         int err;
1579
1580         if (tfile->detached)
1581                 return -EINVAL;
1582
1583         dev = __dev_get_by_name(net, ifr->ifr_name);
1584         if (dev) {
1585                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1586                         return -EBUSY;
1587                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1588                         tun = netdev_priv(dev);
1589                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1590                         tun = netdev_priv(dev);
1591                 else
1592                         return -EINVAL;
1593
1594                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1595                     !!(tun->flags & TUN_TAP_MQ))
1596                         return -EINVAL;
1597
1598                 if (tun_not_capable(tun))
1599                         return -EPERM;
1600                 err = security_tun_dev_open(tun->security);
1601                 if (err < 0)
1602                         return err;
1603
1604                 err = tun_attach(tun, file);
1605                 if (err < 0)
1606                         return err;
1607
1608                 if (tun->flags & TUN_TAP_MQ &&
1609                     (tun->numqueues + tun->numdisabled > 1)) {
1610                         /* One or more queue has already been attached, no need
1611                          * to initialize the device again.
1612                          */
1613                         return 0;
1614                 }
1615         }
1616         else {
1617                 char *name;
1618                 unsigned long flags = 0;
1619                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1620                              MAX_TAP_QUEUES : 1;
1621
1622                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1623                         return -EPERM;
1624                 err = security_tun_dev_create();
1625                 if (err < 0)
1626                         return err;
1627
1628                 /* Set dev type */
1629                 if (ifr->ifr_flags & IFF_TUN) {
1630                         /* TUN device */
1631                         flags |= TUN_TUN_DEV;
1632                         name = "tun%d";
1633                 } else if (ifr->ifr_flags & IFF_TAP) {
1634                         /* TAP device */
1635                         flags |= TUN_TAP_DEV;
1636                         name = "tap%d";
1637                 } else
1638                         return -EINVAL;
1639
1640                 if (*ifr->ifr_name)
1641                         name = ifr->ifr_name;
1642
1643                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1644                                        tun_setup, queues, queues);
1645
1646                 if (!dev)
1647                         return -ENOMEM;
1648
1649                 dev_net_set(dev, net);
1650                 dev->rtnl_link_ops = &tun_link_ops;
1651
1652                 tun = netdev_priv(dev);
1653                 tun->dev = dev;
1654                 tun->flags = flags;
1655                 tun->txflt.count = 0;
1656                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1657
1658                 tun->filter_attached = false;
1659                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1660
1661                 spin_lock_init(&tun->lock);
1662
1663                 err = security_tun_dev_alloc_security(&tun->security);
1664                 if (err < 0)
1665                         goto err_free_dev;
1666
1667                 tun_net_init(dev);
1668                 tun_flow_init(tun);
1669
1670                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1671                         TUN_USER_FEATURES;
1672                 dev->features = dev->hw_features;
1673                 dev->vlan_features = dev->features;
1674
1675                 INIT_LIST_HEAD(&tun->disabled);
1676                 err = tun_attach(tun, file);
1677                 if (err < 0)
1678                         goto err_free_dev;
1679
1680                 err = register_netdevice(tun->dev);
1681                 if (err < 0)
1682                         goto err_free_dev;
1683
1684                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1685                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1686                     device_create_file(&tun->dev->dev, &dev_attr_group))
1687                         pr_err("Failed to create tun sysfs files\n");
1688         }
1689
1690         netif_carrier_on(tun->dev);
1691
1692         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1693
1694         if (ifr->ifr_flags & IFF_NO_PI)
1695                 tun->flags |= TUN_NO_PI;
1696         else
1697                 tun->flags &= ~TUN_NO_PI;
1698
1699         /* This flag has no real effect.  We track the value for backwards
1700          * compatibility.
1701          */
1702         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1703                 tun->flags |= TUN_ONE_QUEUE;
1704         else
1705                 tun->flags &= ~TUN_ONE_QUEUE;
1706
1707         if (ifr->ifr_flags & IFF_VNET_HDR)
1708                 tun->flags |= TUN_VNET_HDR;
1709         else
1710                 tun->flags &= ~TUN_VNET_HDR;
1711
1712         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1713                 tun->flags |= TUN_TAP_MQ;
1714         else
1715                 tun->flags &= ~TUN_TAP_MQ;
1716
1717         /* Make sure persistent devices do not get stuck in
1718          * xoff state.
1719          */
1720         if (netif_running(tun->dev))
1721                 netif_tx_wake_all_queues(tun->dev);
1722
1723         strcpy(ifr->ifr_name, tun->dev->name);
1724         return 0;
1725
1726  err_free_dev:
1727         free_netdev(dev);
1728         return err;
1729 }
1730
1731 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1732                        struct ifreq *ifr)
1733 {
1734         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1735
1736         strcpy(ifr->ifr_name, tun->dev->name);
1737
1738         ifr->ifr_flags = tun_flags(tun);
1739
1740 }
1741
1742 /* This is like a cut-down ethtool ops, except done via tun fd so no
1743  * privs required. */
1744 static int set_offload(struct tun_struct *tun, unsigned long arg)
1745 {
1746         netdev_features_t features = 0;
1747
1748         if (arg & TUN_F_CSUM) {
1749                 features |= NETIF_F_HW_CSUM;
1750                 arg &= ~TUN_F_CSUM;
1751
1752                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1753                         if (arg & TUN_F_TSO_ECN) {
1754                                 features |= NETIF_F_TSO_ECN;
1755                                 arg &= ~TUN_F_TSO_ECN;
1756                         }
1757                         if (arg & TUN_F_TSO4)
1758                                 features |= NETIF_F_TSO;
1759                         if (arg & TUN_F_TSO6)
1760                                 features |= NETIF_F_TSO6;
1761                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1762                 }
1763
1764                 if (arg & TUN_F_UFO) {
1765                         features |= NETIF_F_UFO;
1766                         arg &= ~TUN_F_UFO;
1767                 }
1768         }
1769
1770         /* This gives the user a way to test for new features in future by
1771          * trying to set them. */
1772         if (arg)
1773                 return -EINVAL;
1774
1775         tun->set_features = features;
1776         netdev_update_features(tun->dev);
1777
1778         return 0;
1779 }
1780
1781 static void tun_detach_filter(struct tun_struct *tun, int n)
1782 {
1783         int i;
1784         struct tun_file *tfile;
1785
1786         for (i = 0; i < n; i++) {
1787                 tfile = rtnl_dereference(tun->tfiles[i]);
1788                 sk_detach_filter(tfile->socket.sk);
1789         }
1790
1791         tun->filter_attached = false;
1792 }
1793
1794 static int tun_attach_filter(struct tun_struct *tun)
1795 {
1796         int i, ret = 0;
1797         struct tun_file *tfile;
1798
1799         for (i = 0; i < tun->numqueues; i++) {
1800                 tfile = rtnl_dereference(tun->tfiles[i]);
1801                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1802                 if (ret) {
1803                         tun_detach_filter(tun, i);
1804                         return ret;
1805                 }
1806         }
1807
1808         tun->filter_attached = true;
1809         return ret;
1810 }
1811
1812 static void tun_set_sndbuf(struct tun_struct *tun)
1813 {
1814         struct tun_file *tfile;
1815         int i;
1816
1817         for (i = 0; i < tun->numqueues; i++) {
1818                 tfile = rtnl_dereference(tun->tfiles[i]);
1819                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1820         }
1821 }
1822
1823 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1824 {
1825         struct tun_file *tfile = file->private_data;
1826         struct tun_struct *tun;
1827         int ret = 0;
1828
1829         rtnl_lock();
1830
1831         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1832                 tun = tfile->detached;
1833                 if (!tun) {
1834                         ret = -EINVAL;
1835                         goto unlock;
1836                 }
1837                 ret = security_tun_dev_attach_queue(tun->security);
1838                 if (ret < 0)
1839                         goto unlock;
1840                 ret = tun_attach(tun, file);
1841         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1842                 tun = rtnl_dereference(tfile->tun);
1843                 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1844                         ret = -EINVAL;
1845                 else
1846                         __tun_detach(tfile, false);
1847         } else
1848                 ret = -EINVAL;
1849
1850 unlock:
1851         rtnl_unlock();
1852         return ret;
1853 }
1854
1855 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1856                             unsigned long arg, int ifreq_len)
1857 {
1858         struct tun_file *tfile = file->private_data;
1859         struct tun_struct *tun;
1860         void __user* argp = (void __user*)arg;
1861         struct ifreq ifr;
1862         kuid_t owner;
1863         kgid_t group;
1864         int sndbuf;
1865         int vnet_hdr_sz;
1866         int ret;
1867
1868         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1869                 if (copy_from_user(&ifr, argp, ifreq_len))
1870                         return -EFAULT;
1871         } else {
1872                 memset(&ifr, 0, sizeof(ifr));
1873         }
1874         if (cmd == TUNGETFEATURES) {
1875                 /* Currently this just means: "what IFF flags are valid?".
1876                  * This is needed because we never checked for invalid flags on
1877                  * TUNSETIFF. */
1878                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1879                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1880                                 (unsigned int __user*)argp);
1881         } else if (cmd == TUNSETQUEUE)
1882                 return tun_set_queue(file, &ifr);
1883
1884         ret = 0;
1885         rtnl_lock();
1886
1887         tun = __tun_get(tfile);
1888         if (cmd == TUNSETIFF && !tun) {
1889                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1890
1891                 ret = tun_set_iff(tfile->net, file, &ifr);
1892
1893                 if (ret)
1894                         goto unlock;
1895
1896                 if (copy_to_user(argp, &ifr, ifreq_len))
1897                         ret = -EFAULT;
1898                 goto unlock;
1899         }
1900
1901         ret = -EBADFD;
1902         if (!tun)
1903                 goto unlock;
1904
1905         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1906
1907         ret = 0;
1908         switch (cmd) {
1909         case TUNGETIFF:
1910                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1911
1912                 if (copy_to_user(argp, &ifr, ifreq_len))
1913                         ret = -EFAULT;
1914                 break;
1915
1916         case TUNSETNOCSUM:
1917                 /* Disable/Enable checksum */
1918
1919                 /* [unimplemented] */
1920                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1921                           arg ? "disabled" : "enabled");
1922                 break;
1923
1924         case TUNSETPERSIST:
1925                 /* Disable/Enable persist mode. Keep an extra reference to the
1926                  * module to prevent the module being unprobed.
1927                  */
1928                 if (arg && !(tun->flags & TUN_PERSIST)) {
1929                         tun->flags |= TUN_PERSIST;
1930                         __module_get(THIS_MODULE);
1931                 }
1932                 if (!arg && (tun->flags & TUN_PERSIST)) {
1933                         tun->flags &= ~TUN_PERSIST;
1934                         module_put(THIS_MODULE);
1935                 }
1936
1937                 tun_debug(KERN_INFO, tun, "persist %s\n",
1938                           arg ? "enabled" : "disabled");
1939                 break;
1940
1941         case TUNSETOWNER:
1942                 /* Set owner of the device */
1943                 owner = make_kuid(current_user_ns(), arg);
1944                 if (!uid_valid(owner)) {
1945                         ret = -EINVAL;
1946                         break;
1947                 }
1948                 tun->owner = owner;
1949                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1950                           from_kuid(&init_user_ns, tun->owner));
1951                 break;
1952
1953         case TUNSETGROUP:
1954                 /* Set group of the device */
1955                 group = make_kgid(current_user_ns(), arg);
1956                 if (!gid_valid(group)) {
1957                         ret = -EINVAL;
1958                         break;
1959                 }
1960                 tun->group = group;
1961                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1962                           from_kgid(&init_user_ns, tun->group));
1963                 break;
1964
1965         case TUNSETLINK:
1966                 /* Only allow setting the type when the interface is down */
1967                 if (tun->dev->flags & IFF_UP) {
1968                         tun_debug(KERN_INFO, tun,
1969                                   "Linktype set failed because interface is up\n");
1970                         ret = -EBUSY;
1971                 } else {
1972                         tun->dev->type = (int) arg;
1973                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1974                                   tun->dev->type);
1975                         ret = 0;
1976                 }
1977                 break;
1978
1979 #ifdef TUN_DEBUG
1980         case TUNSETDEBUG:
1981                 tun->debug = arg;
1982                 break;
1983 #endif
1984         case TUNSETOFFLOAD:
1985                 ret = set_offload(tun, arg);
1986                 break;
1987
1988         case TUNSETTXFILTER:
1989                 /* Can be set only for TAPs */
1990                 ret = -EINVAL;
1991                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1992                         break;
1993                 ret = update_filter(&tun->txflt, (void __user *)arg);
1994                 break;
1995
1996         case SIOCGIFHWADDR:
1997                 /* Get hw address */
1998                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1999                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2000                 if (copy_to_user(argp, &ifr, ifreq_len))
2001                         ret = -EFAULT;
2002                 break;
2003
2004         case SIOCSIFHWADDR:
2005                 /* Set hw address */
2006                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2007                           ifr.ifr_hwaddr.sa_data);
2008
2009                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2010                 break;
2011
2012         case TUNGETSNDBUF:
2013                 sndbuf = tfile->socket.sk->sk_sndbuf;
2014                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2015                         ret = -EFAULT;
2016                 break;
2017
2018         case TUNSETSNDBUF:
2019                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2020                         ret = -EFAULT;
2021                         break;
2022                 }
2023
2024                 tun->sndbuf = sndbuf;
2025                 tun_set_sndbuf(tun);
2026                 break;
2027
2028         case TUNGETVNETHDRSZ:
2029                 vnet_hdr_sz = tun->vnet_hdr_sz;
2030                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2031                         ret = -EFAULT;
2032                 break;
2033
2034         case TUNSETVNETHDRSZ:
2035                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2036                         ret = -EFAULT;
2037                         break;
2038                 }
2039                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2040                         ret = -EINVAL;
2041                         break;
2042                 }
2043
2044                 tun->vnet_hdr_sz = vnet_hdr_sz;
2045                 break;
2046
2047         case TUNATTACHFILTER:
2048                 /* Can be set only for TAPs */
2049                 ret = -EINVAL;
2050                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2051                         break;
2052                 ret = -EFAULT;
2053                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2054                         break;
2055
2056                 ret = tun_attach_filter(tun);
2057                 break;
2058
2059         case TUNDETACHFILTER:
2060                 /* Can be set only for TAPs */
2061                 ret = -EINVAL;
2062                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2063                         break;
2064                 ret = 0;
2065                 tun_detach_filter(tun, tun->numqueues);
2066                 break;
2067
2068         default:
2069                 ret = -EINVAL;
2070                 break;
2071         }
2072
2073 unlock:
2074         rtnl_unlock();
2075         if (tun)
2076                 tun_put(tun);
2077         return ret;
2078 }
2079
2080 static long tun_chr_ioctl(struct file *file,
2081                           unsigned int cmd, unsigned long arg)
2082 {
2083         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2084 }
2085
2086 #ifdef CONFIG_COMPAT
2087 static long tun_chr_compat_ioctl(struct file *file,
2088                          unsigned int cmd, unsigned long arg)
2089 {
2090         switch (cmd) {
2091         case TUNSETIFF:
2092         case TUNGETIFF:
2093         case TUNSETTXFILTER:
2094         case TUNGETSNDBUF:
2095         case TUNSETSNDBUF:
2096         case SIOCGIFHWADDR:
2097         case SIOCSIFHWADDR:
2098                 arg = (unsigned long)compat_ptr(arg);
2099                 break;
2100         default:
2101                 arg = (compat_ulong_t)arg;
2102                 break;
2103         }
2104
2105         /*
2106          * compat_ifreq is shorter than ifreq, so we must not access beyond
2107          * the end of that structure. All fields that are used in this
2108          * driver are compatible though, we don't need to convert the
2109          * contents.
2110          */
2111         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2112 }
2113 #endif /* CONFIG_COMPAT */
2114
2115 static int tun_chr_fasync(int fd, struct file *file, int on)
2116 {
2117         struct tun_file *tfile = file->private_data;
2118         int ret;
2119
2120         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2121                 goto out;
2122
2123         if (on) {
2124                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2125                 if (ret)
2126                         goto out;
2127                 tfile->flags |= TUN_FASYNC;
2128         } else
2129                 tfile->flags &= ~TUN_FASYNC;
2130         ret = 0;
2131 out:
2132         return ret;
2133 }
2134
2135 static int tun_chr_open(struct inode *inode, struct file * file)
2136 {
2137         struct tun_file *tfile;
2138
2139         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2140
2141         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2142                                             &tun_proto);
2143         if (!tfile)
2144                 return -ENOMEM;
2145         rcu_assign_pointer(tfile->tun, NULL);
2146         tfile->net = get_net(current->nsproxy->net_ns);
2147         tfile->flags = 0;
2148
2149         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2150         init_waitqueue_head(&tfile->wq.wait);
2151
2152         tfile->socket.file = file;
2153         tfile->socket.ops = &tun_socket_ops;
2154
2155         sock_init_data(&tfile->socket, &tfile->sk);
2156         sk_change_net(&tfile->sk, tfile->net);
2157
2158         tfile->sk.sk_write_space = tun_sock_write_space;
2159         tfile->sk.sk_sndbuf = INT_MAX;
2160
2161         file->private_data = tfile;
2162         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2163         INIT_LIST_HEAD(&tfile->next);
2164
2165         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2166
2167         return 0;
2168 }
2169
2170 static int tun_chr_close(struct inode *inode, struct file *file)
2171 {
2172         struct tun_file *tfile = file->private_data;
2173         struct net *net = tfile->net;
2174
2175         tun_detach(tfile, true);
2176         put_net(net);
2177
2178         return 0;
2179 }
2180
2181 static const struct file_operations tun_fops = {
2182         .owner  = THIS_MODULE,
2183         .llseek = no_llseek,
2184         .read  = do_sync_read,
2185         .aio_read  = tun_chr_aio_read,
2186         .write = do_sync_write,
2187         .aio_write = tun_chr_aio_write,
2188         .poll   = tun_chr_poll,
2189         .unlocked_ioctl = tun_chr_ioctl,
2190 #ifdef CONFIG_COMPAT
2191         .compat_ioctl = tun_chr_compat_ioctl,
2192 #endif
2193         .open   = tun_chr_open,
2194         .release = tun_chr_close,
2195         .fasync = tun_chr_fasync
2196 };
2197
2198 static struct miscdevice tun_miscdev = {
2199         .minor = TUN_MINOR,
2200         .name = "tun",
2201         .nodename = "net/tun",
2202         .fops = &tun_fops,
2203 };
2204
2205 /* ethtool interface */
2206
2207 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2208 {
2209         cmd->supported          = 0;
2210         cmd->advertising        = 0;
2211         ethtool_cmd_speed_set(cmd, SPEED_10);
2212         cmd->duplex             = DUPLEX_FULL;
2213         cmd->port               = PORT_TP;
2214         cmd->phy_address        = 0;
2215         cmd->transceiver        = XCVR_INTERNAL;
2216         cmd->autoneg            = AUTONEG_DISABLE;
2217         cmd->maxtxpkt           = 0;
2218         cmd->maxrxpkt           = 0;
2219         return 0;
2220 }
2221
2222 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2223 {
2224         struct tun_struct *tun = netdev_priv(dev);
2225
2226         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2227         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2228
2229         switch (tun->flags & TUN_TYPE_MASK) {
2230         case TUN_TUN_DEV:
2231                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2232                 break;
2233         case TUN_TAP_DEV:
2234                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2235                 break;
2236         }
2237 }
2238
2239 static u32 tun_get_msglevel(struct net_device *dev)
2240 {
2241 #ifdef TUN_DEBUG
2242         struct tun_struct *tun = netdev_priv(dev);
2243         return tun->debug;
2244 #else
2245         return -EOPNOTSUPP;
2246 #endif
2247 }
2248
2249 static void tun_set_msglevel(struct net_device *dev, u32 value)
2250 {
2251 #ifdef TUN_DEBUG
2252         struct tun_struct *tun = netdev_priv(dev);
2253         tun->debug = value;
2254 #endif
2255 }
2256
2257 static const struct ethtool_ops tun_ethtool_ops = {
2258         .get_settings   = tun_get_settings,
2259         .get_drvinfo    = tun_get_drvinfo,
2260         .get_msglevel   = tun_get_msglevel,
2261         .set_msglevel   = tun_set_msglevel,
2262         .get_link       = ethtool_op_get_link,
2263 };
2264
2265
2266 static int __init tun_init(void)
2267 {
2268         int ret = 0;
2269
2270         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2271         pr_info("%s\n", DRV_COPYRIGHT);
2272
2273         ret = rtnl_link_register(&tun_link_ops);
2274         if (ret) {
2275                 pr_err("Can't register link_ops\n");
2276                 goto err_linkops;
2277         }
2278
2279         ret = misc_register(&tun_miscdev);
2280         if (ret) {
2281                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2282                 goto err_misc;
2283         }
2284         return  0;
2285 err_misc:
2286         rtnl_link_unregister(&tun_link_ops);
2287 err_linkops:
2288         return ret;
2289 }
2290
2291 static void tun_cleanup(void)
2292 {
2293         misc_deregister(&tun_miscdev);
2294         rtnl_link_unregister(&tun_link_ops);
2295 }
2296
2297 /* Get an underlying socket object from tun file.  Returns error unless file is
2298  * attached to a device.  The returned object works like a packet socket, it
2299  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2300  * holding a reference to the file for as long as the socket is in use. */
2301 struct socket *tun_get_socket(struct file *file)
2302 {
2303         struct tun_file *tfile;
2304         if (file->f_op != &tun_fops)
2305                 return ERR_PTR(-EINVAL);
2306         tfile = file->private_data;
2307         if (!tfile)
2308                 return ERR_PTR(-EBADFD);
2309         return &tfile->socket;
2310 }
2311 EXPORT_SYMBOL_GPL(tun_get_socket);
2312
2313 module_init(tun_init);
2314 module_exit(tun_cleanup);
2315 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2316 MODULE_AUTHOR(DRV_COPYRIGHT);
2317 MODULE_LICENSE("GPL");
2318 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2319 MODULE_ALIAS("devname:net/tun");