drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <[email protected]>
   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 <[email protected]> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <[email protected]>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <[email protected]>  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 <[email protected]>
  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 <[email protected]>"
  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/if_vlan.h>
  64 #include <linux/crc32.h>
  65 #include <linux/nsproxy.h>
  66 #include <linux/virtio_net.h>
  67 #include <linux/rcupdate.h>
  68 #include <net/net_namespace.h>
  69 #include <net/netns/generic.h>
  70 #include <net/rtnetlink.h>
  71 #include <net/sock.h>
  72
  73 #include <asm/uaccess.h>
  74
  75 /* Uncomment to enable debugging */
  76 /* #define TUN_DEBUG 1 */
  77
  78 #ifdef TUN_DEBUG
  79 static int debug;
  80
  81 #define tun_debug(level, tun, fmt, args...)                     \
  82 do {                                                            \
  83         if (tun->debug)                                         \
  84                 netdev_printk(level, tun->dev, fmt, ##args);    \
  85 } while (0)
  86 #define DBG1(level, fmt, args...)                               \
  87 do {                                                            \
  88         if (debug == 2)                                         \
  89                 printk(level fmt, ##args);                      \
  90 } while (0)
  91 #else
  92 #define tun_debug(level, tun, fmt, args...)                     \
  93 do {                                                            \
  94         if (0)                                                  \
  95                 netdev_printk(level, tun->dev, fmt, ##args);    \
  96 } while (0)
  97 #define DBG1(level, fmt, args...)                               \
  98 do {                                                            \
  99         if (0)                                                  \
 100                 printk(level fmt, ##args);                      \
 101 } while (0)
 102 #endif
 103
 104 #define GOODCOPY_LEN 128
 105
 106 #define FLT_EXACT_COUNT 8
 107 struct tap_filter {
 108         unsigned int    count;    /* Number of addrs. Zero means disabled */
 109         u32             mask[2];  /* Mask of the hashed addrs */
 110         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 111 };
 112
 113 /* DEFAULT_MAX_NUM_RSS_QUEUES were choosed to let the rx/tx queues allocated for
 114  * the netdevice to be fit in one page. So we can make sure the success of
 115  * memory allocation. TODO: increase the limit. */
 116 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
 117 #define MAX_TAP_FLOWS  4096
 118
 119 #define TUN_FLOW_EXPIRE (3 * HZ)
 120
 121 /* A tun_file connects an open character device to a tuntap netdevice. It
 122  * also contains all socket related strctures (except sock_fprog and tap_filter)
 123  * to serve as one transmit queue for tuntap device. The sock_fprog and
 124  * tap_filter were kept in tun_struct since they were used for filtering for the
 125  * netdevice not for a specific queue (at least I didn't see the requirement for
 126  * this).
 127  *
 128  * RCU usage:
 129  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 130  * other can only be read while rcu_read_lock or rtnl_lock is held.
 131  */
 132 struct tun_file {
 133         struct sock sk;
 134         struct socket socket;
 135         struct socket_wq wq;
 136         struct tun_struct __rcu *tun;
 137         struct net *net;
 138         struct fasync_struct *fasync;
 139         /* only used for fasnyc */
 140         unsigned int flags;
 141         union {
 142                 u16 queue_index;
 143                 unsigned int ifindex;
 144         };
 145         struct list_head next;
 146         struct tun_struct *detached;
 147 };
 148
 149 struct tun_flow_entry {
 150         struct hlist_node hash_link;
 151         struct rcu_head rcu;
 152         struct tun_struct *tun;
 153
 154         u32 rxhash;
 155         int queue_index;
 156         unsigned long updated;
 157 };
 158
 159 #define TUN_NUM_FLOW_ENTRIES 1024
 160
 161 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 162  * device, socket filter, sndbuf and vnet header size were restore when the
 163  * file were attached to a persist device.
 164  */
 165 struct tun_struct {
 166         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 167         unsigned int            numqueues;
 168         unsigned int            flags;
 169         kuid_t                  owner;
 170         kgid_t                  group;
 171
 172         struct net_device       *dev;
 173         netdev_features_t       set_features;
 174 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 175                           NETIF_F_TSO6|NETIF_F_UFO)
 176
 177         int                     vnet_hdr_sz;
 178         int                     sndbuf;
 179         struct tap_filter       txflt;
 180         struct sock_fprog       fprog;
 181         /* protected by rtnl lock */
 182         bool                    filter_attached;
 183 #ifdef TUN_DEBUG
 184         int debug;
 185 #endif
 186         spinlock_t lock;
 187         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 188         struct timer_list flow_gc_timer;
 189         unsigned long ageing_time;
 190         unsigned int numdisabled;
 191         struct list_head disabled;
 192         void *security;
 193         u32 flow_count;
 194 };
 195
 196 static inline u32 tun_hashfn(u32 rxhash)
 197 {
 198         return rxhash & 0x3ff;
 199 }
 200
 201 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 202 {
 203         struct tun_flow_entry *e;
 204
 205         hlist_for_each_entry_rcu(e, head, hash_link) {
 206                 if (e->rxhash == rxhash)
 207                         return e;
 208         }
 209         return NULL;
 210 }
 211
 212 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 213                                               struct hlist_head *head,
 214                                               u32 rxhash, u16 queue_index)
 215 {
 216         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 217
 218         if (e) {
 219                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 220                           rxhash, queue_index);
 221                 e->updated = jiffies;
 222                 e->rxhash = rxhash;
 223                 e->queue_index = queue_index;
 224                 e->tun = tun;
 225                 hlist_add_head_rcu(&e->hash_link, head);
 226                 ++tun->flow_count;
 227         }
 228         return e;
 229 }
 230
 231 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 232 {
 233         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 234                   e->rxhash, e->queue_index);
 235         hlist_del_rcu(&e->hash_link);
 236         kfree_rcu(e, rcu);
 237         --tun->flow_count;
 238 }
 239
 240 static void tun_flow_flush(struct tun_struct *tun)
 241 {
 242         int i;
 243
 244         spin_lock_bh(&tun->lock);
 245         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 246                 struct tun_flow_entry *e;
 247                 struct hlist_node *n;
 248
 249                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 250                         tun_flow_delete(tun, e);
 251         }
 252         spin_unlock_bh(&tun->lock);
 253 }
 254
 255 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 256 {
 257         int i;
 258
 259         spin_lock_bh(&tun->lock);
 260         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 261                 struct tun_flow_entry *e;
 262                 struct hlist_node *n;
 263
 264                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 265                         if (e->queue_index == queue_index)
 266                                 tun_flow_delete(tun, e);
 267                 }
 268         }
 269         spin_unlock_bh(&tun->lock);
 270 }
 271
 272 static void tun_flow_cleanup(unsigned long data)
 273 {
 274         struct tun_struct *tun = (struct tun_struct *)data;
 275         unsigned long delay = tun->ageing_time;
 276         unsigned long next_timer = jiffies + delay;
 277         unsigned long count = 0;
 278         int i;
 279
 280         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 281
 282         spin_lock_bh(&tun->lock);
 283         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 284                 struct tun_flow_entry *e;
 285                 struct hlist_node *n;
 286
 287                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 288                         unsigned long this_timer;
 289                         count++;
 290                         this_timer = e->updated + delay;
 291                         if (time_before_eq(this_timer, jiffies))
 292                                 tun_flow_delete(tun, e);
 293                         else if (time_before(this_timer, next_timer))
 294                                 next_timer = this_timer;
 295                 }
 296         }
 297
 298         if (count)
 299                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 300         spin_unlock_bh(&tun->lock);
 301 }
 302
 303 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 304                             struct tun_file *tfile)
 305 {
 306         struct hlist_head *head;
 307         struct tun_flow_entry *e;
 308         unsigned long delay = tun->ageing_time;
 309         u16 queue_index = tfile->queue_index;
 310
 311         if (!rxhash)
 312                 return;
 313         else
 314                 head = &tun->flows[tun_hashfn(rxhash)];
 315
 316         rcu_read_lock();
 317
 318         /* We may get a very small possibility of OOO during switching, not
 319          * worth to optimize.*/
 320         if (tun->numqueues == 1 || tfile->detached)
 321                 goto unlock;
 322
 323         e = tun_flow_find(head, rxhash);
 324         if (likely(e)) {
 325                 /* TODO: keep queueing to old queue until it's empty? */
 326                 e->queue_index = queue_index;
 327                 e->updated = jiffies;
 328         } else {
 329                 spin_lock_bh(&tun->lock);
 330                 if (!tun_flow_find(head, rxhash) &&
 331                     tun->flow_count < MAX_TAP_FLOWS)
 332                         tun_flow_create(tun, head, rxhash, queue_index);
 333
 334                 if (!timer_pending(&tun->flow_gc_timer))
 335                         mod_timer(&tun->flow_gc_timer,
 336                                   round_jiffies_up(jiffies + delay));
 337                 spin_unlock_bh(&tun->lock);
 338         }
 339
 340 unlock:
 341         rcu_read_unlock();
 342 }
 343
 344 /* We try to identify a flow through its rxhash first. The reason that
 345  * we do not check rxq no. is becuase some cards(e.g 82599), chooses
 346  * the rxq based on the txq where the last packet of the flow comes. As
 347  * the userspace application move between processors, we may get a
 348  * different rxq no. here. If we could not get rxhash, then we would
 349  * hope the rxq no. may help here.
 350  */
 351 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 352                             void *accel_priv)
 353 {
 354         struct tun_struct *tun = netdev_priv(dev);
 355         struct tun_flow_entry *e;
 356         u32 txq = 0;
 357         u32 numqueues = 0;
 358
 359         rcu_read_lock();
 360         numqueues = ACCESS_ONCE(tun->numqueues);
 361
 362         txq = skb_get_rxhash(skb);
 363         if (txq) {
 364                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 365                 if (e)
 366                         txq = e->queue_index;
 367                 else
 368                         /* use multiply and shift instead of expensive divide */
 369                         txq = ((u64)txq * numqueues) >> 32;
 370         } else if (likely(skb_rx_queue_recorded(skb))) {
 371                 txq = skb_get_rx_queue(skb);
 372                 while (unlikely(txq >= numqueues))
 373                         txq -= numqueues;
 374         }
 375
 376         rcu_read_unlock();
 377         return txq;
 378 }
 379
 380 static inline bool tun_not_capable(struct tun_struct *tun)
 381 {
 382         const struct cred *cred = current_cred();
 383         struct net *net = dev_net(tun->dev);
 384
 385         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 386                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 387                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 388 }
 389
 390 static void tun_set_real_num_queues(struct tun_struct *tun)
 391 {
 392         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 393         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 394 }
 395
 396 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 397 {
 398         tfile->detached = tun;
 399         list_add_tail(&tfile->next, &tun->disabled);
 400         ++tun->numdisabled;
 401 }
 402
 403 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 404 {
 405         struct tun_struct *tun = tfile->detached;
 406
 407         tfile->detached = NULL;
 408         list_del_init(&tfile->next);
 409         --tun->numdisabled;
 410         return tun;
 411 }
 412
 413 static void tun_queue_purge(struct tun_file *tfile)
 414 {
 415         skb_queue_purge(&tfile->sk.sk_receive_queue);
 416         skb_queue_purge(&tfile->sk.sk_error_queue);
 417 }
 418
 419 static void __tun_detach(struct tun_file *tfile, bool clean)
 420 {
 421         struct tun_file *ntfile;
 422         struct tun_struct *tun;
 423
 424         tun = rtnl_dereference(tfile->tun);
 425
 426         if (tun && !tfile->detached) {
 427                 u16 index = tfile->queue_index;
 428                 BUG_ON(index >= tun->numqueues);
 429
 430                 rcu_assign_pointer(tun->tfiles[index],
 431                                    tun->tfiles[tun->numqueues - 1]);
 432                 ntfile = rtnl_dereference(tun->tfiles[index]);
 433                 ntfile->queue_index = index;
 434
 435                 --tun->numqueues;
 436                 if (clean) {
 437                         rcu_assign_pointer(tfile->tun, NULL);
 438                         sock_put(&tfile->sk);
 439                 } else
 440                         tun_disable_queue(tun, tfile);
 441
 442                 synchronize_net();
 443                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 444                 /* Drop read queue */
 445                 tun_queue_purge(tfile);
 446                 tun_set_real_num_queues(tun);
 447         } else if (tfile->detached && clean) {
 448                 tun = tun_enable_queue(tfile);
 449                 sock_put(&tfile->sk);
 450         }
 451
 452         if (clean) {
 453                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 454                         netif_carrier_off(tun->dev);
 455
 456                         if (!(tun->flags & TUN_PERSIST) &&
 457                             tun->dev->reg_state == NETREG_REGISTERED)
 458                                 unregister_netdevice(tun->dev);
 459                 }
 460
 461                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 462                                  &tfile->socket.flags));
 463                 sk_release_kernel(&tfile->sk);
 464         }
 465 }
 466
 467 static void tun_detach(struct tun_file *tfile, bool clean)
 468 {
 469         rtnl_lock();
 470         __tun_detach(tfile, clean);
 471         rtnl_unlock();
 472 }
 473
 474 static void tun_detach_all(struct net_device *dev)
 475 {
 476         struct tun_struct *tun = netdev_priv(dev);
 477         struct tun_file *tfile, *tmp;
 478         int i, n = tun->numqueues;
 479
 480         for (i = 0; i < n; i++) {
 481                 tfile = rtnl_dereference(tun->tfiles[i]);
 482                 BUG_ON(!tfile);
 483                 wake_up_all(&tfile->wq.wait);
 484                 rcu_assign_pointer(tfile->tun, NULL);
 485                 --tun->numqueues;
 486         }
 487         list_for_each_entry(tfile, &tun->disabled, next) {
 488                 wake_up_all(&tfile->wq.wait);
 489                 rcu_assign_pointer(tfile->tun, NULL);
 490         }
 491         BUG_ON(tun->numqueues != 0);
 492
 493         synchronize_net();
 494         for (i = 0; i < n; i++) {
 495                 tfile = rtnl_dereference(tun->tfiles[i]);
 496                 /* Drop read queue */
 497                 tun_queue_purge(tfile);
 498                 sock_put(&tfile->sk);
 499         }
 500         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 501                 tun_enable_queue(tfile);
 502                 tun_queue_purge(tfile);
 503                 sock_put(&tfile->sk);
 504         }
 505         BUG_ON(tun->numdisabled != 0);
 506
 507         if (tun->flags & TUN_PERSIST)
 508                 module_put(THIS_MODULE);
 509 }
 510
 511 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 512 {
 513         struct tun_file *tfile = file->private_data;
 514         int err;
 515
 516         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 517         if (err < 0)
 518                 goto out;
 519
 520         err = -EINVAL;
 521         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 522                 goto out;
 523
 524         err = -EBUSY;
 525         if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 526                 goto out;
 527
 528         err = -E2BIG;
 529         if (!tfile->detached &&
 530             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 531                 goto out;
 532
 533         err = 0;
 534
 535         /* Re-attach the filter to presist device */
 536         if (!skip_filter && (tun->filter_attached == true)) {
 537                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 538                 if (!err)
 539                         goto out;
 540         }
 541         tfile->queue_index = tun->numqueues;
 542         rcu_assign_pointer(tfile->tun, tun);
 543         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 544         tun->numqueues++;
 545
 546         if (tfile->detached)
 547                 tun_enable_queue(tfile);
 548         else
 549                 sock_hold(&tfile->sk);
 550
 551         tun_set_real_num_queues(tun);
 552
 553         /* device is allowed to go away first, so no need to hold extra
 554          * refcnt.
 555          */
 556
 557 out:
 558         return err;
 559 }
 560
 561 static struct tun_struct *__tun_get(struct tun_file *tfile)
 562 {
 563         struct tun_struct *tun;
 564
 565         rcu_read_lock();
 566         tun = rcu_dereference(tfile->tun);
 567         if (tun)
 568                 dev_hold(tun->dev);
 569         rcu_read_unlock();
 570
 571         return tun;
 572 }
 573
 574 static struct tun_struct *tun_get(struct file *file)
 575 {
 576         return __tun_get(file->private_data);
 577 }
 578
 579 static void tun_put(struct tun_struct *tun)
 580 {
 581         dev_put(tun->dev);
 582 }
 583
 584 /* TAP filtering */
 585 static void addr_hash_set(u32 *mask, const u8 *addr)
 586 {
 587         int n = ether_crc(ETH_ALEN, addr) >> 26;
 588         mask[n >> 5] |= (1 << (n & 31));
 589 }
 590
 591 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 592 {
 593         int n = ether_crc(ETH_ALEN, addr) >> 26;
 594         return mask[n >> 5] & (1 << (n & 31));
 595 }
 596
 597 static int update_filter(struct tap_filter *filter, void __user *arg)
 598 {
 599         struct { u8 u[ETH_ALEN]; } *addr;
 600         struct tun_filter uf;
 601         int err, alen, n, nexact;
 602
 603         if (copy_from_user(&uf, arg, sizeof(uf)))
 604                 return -EFAULT;
 605
 606         if (!uf.count) {
 607                 /* Disabled */
 608                 filter->count = 0;
 609                 return 0;
 610         }
 611
 612         alen = ETH_ALEN * uf.count;
 613         addr = kmalloc(alen, GFP_KERNEL);
 614         if (!addr)
 615                 return -ENOMEM;
 616
 617         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 618                 err = -EFAULT;
 619                 goto done;
 620         }
 621
 622         /* The filter is updated without holding any locks. Which is
 623          * perfectly safe. We disable it first and in the worst
 624          * case we'll accept a few undesired packets. */
 625         filter->count = 0;
 626         wmb();
 627
 628         /* Use first set of addresses as an exact filter */
 629         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 630                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 631
 632         nexact = n;
 633
 634         /* Remaining multicast addresses are hashed,
 635          * unicast will leave the filter disabled. */
 636         memset(filter->mask, 0, sizeof(filter->mask));
 637         for (; n < uf.count; n++) {
 638                 if (!is_multicast_ether_addr(addr[n].u)) {
 639                         err = 0; /* no filter */
 640                         goto done;
 641                 }
 642                 addr_hash_set(filter->mask, addr[n].u);
 643         }
 644
 645         /* For ALLMULTI just set the mask to all ones.
 646          * This overrides the mask populated above. */
 647         if ((uf.flags & TUN_FLT_ALLMULTI))
 648                 memset(filter->mask, ~0, sizeof(filter->mask));
 649
 650         /* Now enable the filter */
 651         wmb();
 652         filter->count = nexact;
 653
 654         /* Return the number of exact filters */
 655         err = nexact;
 656
 657 done:
 658         kfree(addr);
 659         return err;
 660 }
 661
 662 /* Returns: 0 - drop, !=0 - accept */
 663 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 664 {
 665         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 666          * at this point. */
 667         struct ethhdr *eh = (struct ethhdr *) skb->data;
 668         int i;
 669
 670         /* Exact match */
 671         for (i = 0; i < filter->count; i++)
 672                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 673                         return 1;
 674
 675         /* Inexact match (multicast only) */
 676         if (is_multicast_ether_addr(eh->h_dest))
 677                 return addr_hash_test(filter->mask, eh->h_dest);
 678
 679         return 0;
 680 }
 681
 682 /*
 683  * Checks whether the packet is accepted or not.
 684  * Returns: 0 - drop, !=0 - accept
 685  */
 686 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 687 {
 688         if (!filter->count)
 689                 return 1;
 690
 691         return run_filter(filter, skb);
 692 }
 693
 694 /* Network device part of the driver */
 695
 696 static const struct ethtool_ops tun_ethtool_ops;
 697
 698 /* Net device detach from fd. */
 699 static void tun_net_uninit(struct net_device *dev)
 700 {
 701         tun_detach_all(dev);
 702 }
 703
 704 /* Net device open. */
 705 static int tun_net_open(struct net_device *dev)
 706 {
 707         netif_tx_start_all_queues(dev);
 708         return 0;
 709 }
 710
 711 /* Net device close. */
 712 static int tun_net_close(struct net_device *dev)
 713 {
 714         netif_tx_stop_all_queues(dev);
 715         return 0;
 716 }
 717
 718 /* Net device start xmit */
 719 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 720 {
 721         struct tun_struct *tun = netdev_priv(dev);
 722         int txq = skb->queue_mapping;
 723         struct tun_file *tfile;
 724
 725         rcu_read_lock();
 726         tfile = rcu_dereference(tun->tfiles[txq]);
 727
 728         /* Drop packet if interface is not attached */
 729         if (txq >= tun->numqueues)
 730                 goto drop;
 731
 732         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 733
 734         BUG_ON(!tfile);
 735
 736         /* Drop if the filter does not like it.
 737          * This is a noop if the filter is disabled.
 738          * Filter can be enabled only for the TAP devices. */
 739         if (!check_filter(&tun->txflt, skb))
 740                 goto drop;
 741
 742         if (tfile->socket.sk->sk_filter &&
 743             sk_filter(tfile->socket.sk, skb))
 744                 goto drop;
 745
 746         /* Limit the number of packets queued by dividing txq length with the
 747          * number of queues.
 748          */
 749         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
 750                           >= dev->tx_queue_len / tun->numqueues)
 751                 goto drop;
 752
 753         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 754                 goto drop;
 755
 756         if (skb->sk) {
 757                 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
 758                 sw_tx_timestamp(skb);
 759         }
 760
 761         /* Orphan the skb - required as we might hang on to it
 762          * for indefinite time.
 763          */
 764         skb_orphan(skb);
 765
 766         nf_reset(skb);
 767
 768         /* Enqueue packet */
 769         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 770
 771         /* Notify and wake up reader process */
 772         if (tfile->flags & TUN_FASYNC)
 773                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 774         wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
 775                                    POLLRDNORM | POLLRDBAND);
 776
 777         rcu_read_unlock();
 778         return NETDEV_TX_OK;
 779
 780 drop:
 781         dev->stats.tx_dropped++;
 782         skb_tx_error(skb);
 783         kfree_skb(skb);
 784         rcu_read_unlock();
 785         return NETDEV_TX_OK;
 786 }
 787
 788 static void tun_net_mclist(struct net_device *dev)
 789 {
 790         /*
 791          * This callback is supposed to deal with mc filter in
 792          * _rx_ path and has nothing to do with the _tx_ path.
 793          * In rx path we always accept everything userspace gives us.
 794          */
 795 }
 796
 797 #define MIN_MTU 68
 798 #define MAX_MTU 65535
 799
 800 static int
 801 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 802 {
 803         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 804                 return -EINVAL;
 805         dev->mtu = new_mtu;
 806         return 0;
 807 }
 808
 809 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 810         netdev_features_t features)
 811 {
 812         struct tun_struct *tun = netdev_priv(dev);
 813
 814         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 815 }
 816 #ifdef CONFIG_NET_POLL_CONTROLLER
 817 static void tun_poll_controller(struct net_device *dev)
 818 {
 819         /*
 820          * Tun only receives frames when:
 821          * 1) the char device endpoint gets data from user space
 822          * 2) the tun socket gets a sendmsg call from user space
 823          * Since both of those are syncronous operations, we are guaranteed
 824          * never to have pending data when we poll for it
 825          * so theres nothing to do here but return.
 826          * We need this though so netpoll recognizes us as an interface that
 827          * supports polling, which enables bridge devices in virt setups to
 828          * still use netconsole
 829          */
 830         return;
 831 }
 832 #endif
 833 static const struct net_device_ops tun_netdev_ops = {
 834         .ndo_uninit             = tun_net_uninit,
 835         .ndo_open               = tun_net_open,
 836         .ndo_stop               = tun_net_close,
 837         .ndo_start_xmit         = tun_net_xmit,
 838         .ndo_change_mtu         = tun_net_change_mtu,
 839         .ndo_fix_features       = tun_net_fix_features,
 840         .ndo_select_queue       = tun_select_queue,
 841 #ifdef CONFIG_NET_POLL_CONTROLLER
 842         .ndo_poll_controller    = tun_poll_controller,
 843 #endif
 844 };
 845
 846 static const struct net_device_ops tap_netdev_ops = {
 847         .ndo_uninit             = tun_net_uninit,
 848         .ndo_open               = tun_net_open,
 849         .ndo_stop               = tun_net_close,
 850         .ndo_start_xmit         = tun_net_xmit,
 851         .ndo_change_mtu         = tun_net_change_mtu,
 852         .ndo_fix_features       = tun_net_fix_features,
 853         .ndo_set_rx_mode        = tun_net_mclist,
 854         .ndo_set_mac_address    = eth_mac_addr,
 855         .ndo_validate_addr      = eth_validate_addr,
 856         .ndo_select_queue       = tun_select_queue,
 857 #ifdef CONFIG_NET_POLL_CONTROLLER
 858         .ndo_poll_controller    = tun_poll_controller,
 859 #endif
 860 };
 861
 862 static void tun_flow_init(struct tun_struct *tun)
 863 {
 864         int i;
 865
 866         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 867                 INIT_HLIST_HEAD(&tun->flows[i]);
 868
 869         tun->ageing_time = TUN_FLOW_EXPIRE;
 870         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 871         mod_timer(&tun->flow_gc_timer,
 872                   round_jiffies_up(jiffies + tun->ageing_time));
 873 }
 874
 875 static void tun_flow_uninit(struct tun_struct *tun)
 876 {
 877         del_timer_sync(&tun->flow_gc_timer);
 878         tun_flow_flush(tun);
 879 }
 880
 881 /* Initialize net device. */
 882 static void tun_net_init(struct net_device *dev)
 883 {
 884         struct tun_struct *tun = netdev_priv(dev);
 885
 886         switch (tun->flags & TUN_TYPE_MASK) {
 887         case TUN_TUN_DEV:
 888                 dev->netdev_ops = &tun_netdev_ops;
 889
 890                 /* Point-to-Point TUN Device */
 891                 dev->hard_header_len = 0;
 892                 dev->addr_len = 0;
 893                 dev->mtu = 1500;
 894
 895                 /* Zero header length */
 896                 dev->type = ARPHRD_NONE;
 897                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 898                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 899                 break;
 900
 901         case TUN_TAP_DEV:
 902                 dev->netdev_ops = &tap_netdev_ops;
 903                 /* Ethernet TAP Device */
 904                 ether_setup(dev);
 905                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 906                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 907
 908                 eth_hw_addr_random(dev);
 909
 910                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 911                 break;
 912         }
 913 }
 914
 915 /* Character device part */
 916
 917 /* Poll */
 918 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 919 {
 920         struct tun_file *tfile = file->private_data;
 921         struct tun_struct *tun = __tun_get(tfile);
 922         struct sock *sk;
 923         unsigned int mask = 0;
 924
 925         if (!tun)
 926                 return POLLERR;
 927
 928         sk = tfile->socket.sk;
 929
 930         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 931
 932         poll_wait(file, &tfile->wq.wait, wait);
 933
 934         if (!skb_queue_empty(&sk->sk_receive_queue))
 935                 mask |= POLLIN | POLLRDNORM;
 936
 937         if (sock_writeable(sk) ||
 938             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 939              sock_writeable(sk)))
 940                 mask |= POLLOUT | POLLWRNORM;
 941
 942         if (tun->dev->reg_state != NETREG_REGISTERED)
 943                 mask = POLLERR;
 944
 945         tun_put(tun);
 946         return mask;
 947 }
 948
 949 /* prepad is the amount to reserve at front.  len is length after that.
 950  * linear is a hint as to how much to copy (usually headers). */
 951 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 952                                      size_t prepad, size_t len,
 953                                      size_t linear, int noblock)
 954 {
 955         struct sock *sk = tfile->socket.sk;
 956         struct sk_buff *skb;
 957         int err;
 958
 959         /* Under a page?  Don't bother with paged skb. */
 960         if (prepad + len < PAGE_SIZE || !linear)
 961                 linear = len;
 962
 963         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 964                                    &err, 0);
 965         if (!skb)
 966                 return ERR_PTR(err);
 967
 968         skb_reserve(skb, prepad);
 969         skb_put(skb, linear);
 970         skb->data_len = len - linear;
 971         skb->len += len - linear;
 972
 973         return skb;
 974 }
 975
 976 /* Get packet from user space buffer */
 977 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
 978                             void *msg_control, const struct iovec *iv,
 979                             size_t total_len, size_t count, int noblock)
 980 {
 981         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
 982         struct sk_buff *skb;
 983         size_t len = total_len, align = NET_SKB_PAD, linear;
 984         struct virtio_net_hdr gso = { 0 };
 985         int good_linear;
 986         int offset = 0;
 987         int copylen;
 988         bool zerocopy = false;
 989         int err;
 990         u32 rxhash;
 991
 992         if (!(tun->flags & TUN_NO_PI)) {
 993                 if (len < sizeof(pi))
 994                         return -EINVAL;
 995                 len -= sizeof(pi);
 996
 997                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
 998                         return -EFAULT;
 999                 offset += sizeof(pi);
1000         }
1001
1002         if (tun->flags & TUN_VNET_HDR) {
1003                 if (len < tun->vnet_hdr_sz)
1004                         return -EINVAL;
1005                 len -= tun->vnet_hdr_sz;
1006
1007                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1008                         return -EFAULT;
1009
1010                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1011                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1012                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1013
1014                 if (gso.hdr_len > len)
1015                         return -EINVAL;
1016                 offset += tun->vnet_hdr_sz;
1017         }
1018
1019         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1020                 align += NET_IP_ALIGN;
1021                 if (unlikely(len < ETH_HLEN ||
1022                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1023                         return -EINVAL;
1024         }
1025
1026         good_linear = SKB_MAX_HEAD(align);
1027
1028         if (msg_control) {
1029                 /* There are 256 bytes to be copied in skb, so there is
1030                  * enough room for skb expand head in case it is used.
1031                  * The rest of the buffer is mapped from userspace.
1032                  */
1033                 copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1034                 if (copylen > good_linear)
1035                         copylen = good_linear;
1036                 linear = copylen;
1037                 if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1038                         zerocopy = true;
1039         }
1040
1041         if (!zerocopy) {
1042                 copylen = len;
1043                 if (gso.hdr_len > good_linear)
1044                         linear = good_linear;
1045                 else
1046                         linear = gso.hdr_len;
1047         }
1048
1049         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1050         if (IS_ERR(skb)) {
1051                 if (PTR_ERR(skb) != -EAGAIN)
1052                         tun->dev->stats.rx_dropped++;
1053                 return PTR_ERR(skb);
1054         }
1055
1056         if (zerocopy)
1057                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1058         else {
1059                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1060                 if (!err && msg_control) {
1061                         struct ubuf_info *uarg = msg_control;
1062                         uarg->callback(uarg, false);
1063                 }
1064         }
1065
1066         if (err) {
1067                 tun->dev->stats.rx_dropped++;
1068                 kfree_skb(skb);
1069                 return -EFAULT;
1070         }
1071
1072         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1073                 if (!skb_partial_csum_set(skb, gso.csum_start,
1074                                           gso.csum_offset)) {
1075                         tun->dev->stats.rx_frame_errors++;
1076                         kfree_skb(skb);
1077                         return -EINVAL;
1078                 }
1079         }
1080
1081         switch (tun->flags & TUN_TYPE_MASK) {
1082         case TUN_TUN_DEV:
1083                 if (tun->flags & TUN_NO_PI) {
1084                         switch (skb->data[0] & 0xf0) {
1085                         case 0x40:
1086                                 pi.proto = htons(ETH_P_IP);
1087                                 break;
1088                         case 0x60:
1089                                 pi.proto = htons(ETH_P_IPV6);
1090                                 break;
1091                         default:
1092                                 tun->dev->stats.rx_dropped++;
1093                                 kfree_skb(skb);
1094                                 return -EINVAL;
1095                         }
1096                 }
1097
1098                 skb_reset_mac_header(skb);
1099                 skb->protocol = pi.proto;
1100                 skb->dev = tun->dev;
1101                 break;
1102         case TUN_TAP_DEV:
1103                 skb->protocol = eth_type_trans(skb, tun->dev);
1104                 break;
1105         }
1106
1107         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1108                 pr_debug("GSO!\n");
1109                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1110                 case VIRTIO_NET_HDR_GSO_TCPV4:
1111                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1112                         break;
1113                 case VIRTIO_NET_HDR_GSO_TCPV6:
1114                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1115                         break;
1116                 case VIRTIO_NET_HDR_GSO_UDP:
1117                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1118                         break;
1119                 default:
1120                         tun->dev->stats.rx_frame_errors++;
1121                         kfree_skb(skb);
1122                         return -EINVAL;
1123                 }
1124
1125                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1126                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1127
1128                 skb_shinfo(skb)->gso_size = gso.gso_size;
1129                 if (skb_shinfo(skb)->gso_size == 0) {
1130                         tun->dev->stats.rx_frame_errors++;
1131                         kfree_skb(skb);
1132                         return -EINVAL;
1133                 }
1134
1135                 /* Header must be checked, and gso_segs computed. */
1136                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1137                 skb_shinfo(skb)->gso_segs = 0;
1138         }
1139
1140         /* copy skb_ubuf_info for callback when skb has no error */
1141         if (zerocopy) {
1142                 skb_shinfo(skb)->destructor_arg = msg_control;
1143                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1144                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1145         }
1146
1147         skb_reset_network_header(skb);
1148         skb_probe_transport_header(skb, 0);
1149
1150         rxhash = skb_get_rxhash(skb);
1151         netif_rx_ni(skb);
1152
1153         tun->dev->stats.rx_packets++;
1154         tun->dev->stats.rx_bytes += len;
1155
1156         tun_flow_update(tun, rxhash, tfile);
1157         return total_len;
1158 }
1159
1160 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1161                               unsigned long count, loff_t pos)
1162 {
1163         struct file *file = iocb->ki_filp;
1164         struct tun_struct *tun = tun_get(file);
1165         struct tun_file *tfile = file->private_data;
1166         ssize_t result;
1167
1168         if (!tun)
1169                 return -EBADFD;
1170
1171         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1172
1173         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1174                               count, file->f_flags & O_NONBLOCK);
1175
1176         tun_put(tun);
1177         return result;
1178 }
1179
1180 /* Put packet to the user space buffer */
1181 static ssize_t tun_put_user(struct tun_struct *tun,
1182                             struct tun_file *tfile,
1183                             struct sk_buff *skb,
1184                             const struct iovec *iv, int len)
1185 {
1186         struct tun_pi pi = { 0, skb->protocol };
1187         ssize_t total = 0;
1188         int vlan_offset = 0, copied;
1189
1190         if (!(tun->flags & TUN_NO_PI)) {
1191                 if ((len -= sizeof(pi)) < 0)
1192                         return -EINVAL;
1193
1194                 if (len < skb->len) {
1195                         /* Packet will be striped */
1196                         pi.flags |= TUN_PKT_STRIP;
1197                 }
1198
1199                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1200                         return -EFAULT;
1201                 total += sizeof(pi);
1202         }
1203
1204         if (tun->flags & TUN_VNET_HDR) {
1205                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1206                 if ((len -= tun->vnet_hdr_sz) < 0)
1207                         return -EINVAL;
1208
1209                 if (skb_is_gso(skb)) {
1210                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1211
1212                         /* This is a hint as to how much should be linear. */
1213                         gso.hdr_len = skb_headlen(skb);
1214                         gso.gso_size = sinfo->gso_size;
1215                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1216                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1217                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1218                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1219                         else if (sinfo->gso_type & SKB_GSO_UDP)
1220                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1221                         else {
1222                                 pr_err("unexpected GSO type: "
1223                                        "0x%x, gso_size %d, hdr_len %d\n",
1224                                        sinfo->gso_type, gso.gso_size,
1225                                        gso.hdr_len);
1226                                 print_hex_dump(KERN_ERR, "tun: ",
1227                                                DUMP_PREFIX_NONE,
1228                                                16, 1, skb->head,
1229                                                min((int)gso.hdr_len, 64), true);
1230                                 WARN_ON_ONCE(1);
1231                                 return -EINVAL;
1232                         }
1233                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1234                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1235                 } else
1236                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1237
1238                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1239                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1240                         gso.csum_start = skb_checksum_start_offset(skb);
1241                         gso.csum_offset = skb->csum_offset;
1242                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1243                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1244                 } /* else everything is zero */
1245
1246                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1247                                                sizeof(gso))))
1248                         return -EFAULT;
1249                 total += tun->vnet_hdr_sz;
1250         }
1251
1252         copied = total;
1253         total += skb->len;
1254         if (!vlan_tx_tag_present(skb)) {
1255                 len = min_t(int, skb->len, len);
1256         } else {
1257                 int copy, ret;
1258                 struct {
1259                         __be16 h_vlan_proto;
1260                         __be16 h_vlan_TCI;
1261                 } veth;
1262
1263                 veth.h_vlan_proto = skb->vlan_proto;
1264                 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1265
1266                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1267                 len = min_t(int, skb->len + VLAN_HLEN, len);
1268                 total += VLAN_HLEN;
1269
1270                 copy = min_t(int, vlan_offset, len);
1271                 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
1272                 len -= copy;
1273                 copied += copy;
1274                 if (ret || !len)
1275                         goto done;
1276
1277                 copy = min_t(int, sizeof(veth), len);
1278                 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
1279                 len -= copy;
1280                 copied += copy;
1281                 if (ret || !len)
1282                         goto done;
1283         }
1284
1285         skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
1286
1287 done:
1288         tun->dev->stats.tx_packets++;
1289         tun->dev->stats.tx_bytes += len;
1290
1291         return total;
1292 }
1293
1294 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1295                            struct kiocb *iocb, const struct iovec *iv,
1296                            ssize_t len, int noblock)
1297 {
1298         DECLARE_WAITQUEUE(wait, current);
1299         struct sk_buff *skb;
1300         ssize_t ret = 0;
1301
1302         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1303
1304         if (unlikely(!noblock))
1305                 add_wait_queue(&tfile->wq.wait, &wait);
1306         while (len) {
1307                 if (unlikely(!noblock))
1308                         current->state = TASK_INTERRUPTIBLE;
1309
1310                 /* Read frames from the queue */
1311                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1312                         if (noblock) {
1313                                 ret = -EAGAIN;
1314                                 break;
1315                         }
1316                         if (signal_pending(current)) {
1317                                 ret = -ERESTARTSYS;
1318                                 break;
1319                         }
1320                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1321                                 ret = -EIO;
1322                                 break;
1323                         }
1324
1325                         /* Nothing to read, let's sleep */
1326                         schedule();
1327                         continue;
1328                 }
1329
1330                 ret = tun_put_user(tun, tfile, skb, iv, len);
1331                 kfree_skb(skb);
1332                 break;
1333         }
1334
1335         if (unlikely(!noblock)) {
1336                 current->state = TASK_RUNNING;
1337                 remove_wait_queue(&tfile->wq.wait, &wait);
1338         }
1339
1340         return ret;
1341 }
1342
1343 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1344                             unsigned long count, loff_t pos)
1345 {
1346         struct file *file = iocb->ki_filp;
1347         struct tun_file *tfile = file->private_data;
1348         struct tun_struct *tun = __tun_get(tfile);
1349         ssize_t len, ret;
1350
1351         if (!tun)
1352                 return -EBADFD;
1353         len = iov_length(iv, count);
1354         if (len < 0) {
1355                 ret = -EINVAL;
1356                 goto out;
1357         }
1358
1359         ret = tun_do_read(tun, tfile, iocb, iv, len,
1360                           file->f_flags & O_NONBLOCK);
1361         ret = min_t(ssize_t, ret, len);
1362         if (ret > 0)
1363                 iocb->ki_pos = ret;
1364 out:
1365         tun_put(tun);
1366         return ret;
1367 }
1368
1369 static void tun_free_netdev(struct net_device *dev)
1370 {
1371         struct tun_struct *tun = netdev_priv(dev);
1372
1373         BUG_ON(!(list_empty(&tun->disabled)));
1374         tun_flow_uninit(tun);
1375         security_tun_dev_free_security(tun->security);
1376         free_netdev(dev);
1377 }
1378
1379 static void tun_setup(struct net_device *dev)
1380 {
1381         struct tun_struct *tun = netdev_priv(dev);
1382
1383         tun->owner = INVALID_UID;
1384         tun->group = INVALID_GID;
1385
1386         dev->ethtool_ops = &tun_ethtool_ops;
1387         dev->destructor = tun_free_netdev;
1388 }
1389
1390 /* Trivial set of netlink ops to allow deleting tun or tap
1391  * device with netlink.
1392  */
1393 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1394 {
1395         return -EINVAL;
1396 }
1397
1398 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1399         .kind           = DRV_NAME,
1400         .priv_size      = sizeof(struct tun_struct),
1401         .setup          = tun_setup,
1402         .validate       = tun_validate,
1403 };
1404
1405 static void tun_sock_write_space(struct sock *sk)
1406 {
1407         struct tun_file *tfile;
1408         wait_queue_head_t *wqueue;
1409
1410         if (!sock_writeable(sk))
1411                 return;
1412
1413         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1414                 return;
1415
1416         wqueue = sk_sleep(sk);
1417         if (wqueue && waitqueue_active(wqueue))
1418                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1419                                                 POLLWRNORM | POLLWRBAND);
1420
1421         tfile = container_of(sk, struct tun_file, sk);
1422         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1423 }
1424
1425 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1426                        struct msghdr *m, size_t total_len)
1427 {
1428         int ret;
1429         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1430         struct tun_struct *tun = __tun_get(tfile);
1431
1432         if (!tun)
1433                 return -EBADFD;
1434         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1435                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1436         tun_put(tun);
1437         return ret;
1438 }
1439
1440 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1441                        struct msghdr *m, size_t total_len,
1442                        int flags)
1443 {
1444         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1445         struct tun_struct *tun = __tun_get(tfile);
1446         int ret;
1447
1448         if (!tun)
1449                 return -EBADFD;
1450
1451         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1452                 ret = -EINVAL;
1453                 goto out;
1454         }
1455         if (flags & MSG_ERRQUEUE) {
1456                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1457                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1458                 goto out;
1459         }
1460         ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1461                           flags & MSG_DONTWAIT);
1462         if (ret > total_len) {
1463                 m->msg_flags |= MSG_TRUNC;
1464                 ret = flags & MSG_TRUNC ? ret : total_len;
1465         }
1466 out:
1467         tun_put(tun);
1468         return ret;
1469 }
1470
1471 static int tun_release(struct socket *sock)
1472 {
1473         if (sock->sk)
1474                 sock_put(sock->sk);
1475         return 0;
1476 }
1477
1478 /* Ops structure to mimic raw sockets with tun */
1479 static const struct proto_ops tun_socket_ops = {
1480         .sendmsg = tun_sendmsg,
1481         .recvmsg = tun_recvmsg,
1482         .release = tun_release,
1483 };
1484
1485 static struct proto tun_proto = {
1486         .name           = "tun",
1487         .owner          = THIS_MODULE,
1488         .obj_size       = sizeof(struct tun_file),
1489 };
1490
1491 static int tun_flags(struct tun_struct *tun)
1492 {
1493         int flags = 0;
1494
1495         if (tun->flags & TUN_TUN_DEV)
1496                 flags |= IFF_TUN;
1497         else
1498                 flags |= IFF_TAP;
1499
1500         if (tun->flags & TUN_NO_PI)
1501                 flags |= IFF_NO_PI;
1502
1503         /* This flag has no real effect.  We track the value for backwards
1504          * compatibility.
1505          */
1506         if (tun->flags & TUN_ONE_QUEUE)
1507                 flags |= IFF_ONE_QUEUE;
1508
1509         if (tun->flags & TUN_VNET_HDR)
1510                 flags |= IFF_VNET_HDR;
1511
1512         if (tun->flags & TUN_TAP_MQ)
1513                 flags |= IFF_MULTI_QUEUE;
1514
1515         if (tun->flags & TUN_PERSIST)
1516                 flags |= IFF_PERSIST;
1517
1518         return flags;
1519 }
1520
1521 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1522                               char *buf)
1523 {
1524         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1525         return sprintf(buf, "0x%x\n", tun_flags(tun));
1526 }
1527
1528 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1529                               char *buf)
1530 {
1531         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1532         return uid_valid(tun->owner)?
1533                 sprintf(buf, "%u\n",
1534                         from_kuid_munged(current_user_ns(), tun->owner)):
1535                 sprintf(buf, "-1\n");
1536 }
1537
1538 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1539                               char *buf)
1540 {
1541         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1542         return gid_valid(tun->group) ?
1543                 sprintf(buf, "%u\n",
1544                         from_kgid_munged(current_user_ns(), tun->group)):
1545                 sprintf(buf, "-1\n");
1546 }
1547
1548 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1549 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1550 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1551
1552 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1553 {
1554         struct tun_struct *tun;
1555         struct tun_file *tfile = file->private_data;
1556         struct net_device *dev;
1557         int err;
1558
1559         if (tfile->detached)
1560                 return -EINVAL;
1561
1562         dev = __dev_get_by_name(net, ifr->ifr_name);
1563         if (dev) {
1564                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1565                         return -EBUSY;
1566                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1567                         tun = netdev_priv(dev);
1568                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1569                         tun = netdev_priv(dev);
1570                 else
1571                         return -EINVAL;
1572
1573                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1574                     !!(tun->flags & TUN_TAP_MQ))
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, ifr->ifr_flags & IFF_NOFILTER);
1584                 if (err < 0)
1585                         return err;
1586
1587                 if (tun->flags & TUN_TAP_MQ &&
1588                     (tun->numqueues + tun->numdisabled > 1)) {
1589                         /* One or more queue has already been attached, no need
1590                          * to initialize the device again.
1591                          */
1592                         return 0;
1593                 }
1594         }
1595         else {
1596                 char *name;
1597                 unsigned long flags = 0;
1598                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1599                              MAX_TAP_QUEUES : 1;
1600
1601                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1602                         return -EPERM;
1603                 err = security_tun_dev_create();
1604                 if (err < 0)
1605                         return err;
1606
1607                 /* Set dev type */
1608                 if (ifr->ifr_flags & IFF_TUN) {
1609                         /* TUN device */
1610                         flags |= TUN_TUN_DEV;
1611                         name = "tun%d";
1612                 } else if (ifr->ifr_flags & IFF_TAP) {
1613                         /* TAP device */
1614                         flags |= TUN_TAP_DEV;
1615                         name = "tap%d";
1616                 } else
1617                         return -EINVAL;
1618
1619                 if (*ifr->ifr_name)
1620                         name = ifr->ifr_name;
1621
1622                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1623                                        tun_setup, queues, queues);
1624
1625                 if (!dev)
1626                         return -ENOMEM;
1627
1628                 dev_net_set(dev, net);
1629                 dev->rtnl_link_ops = &tun_link_ops;
1630                 dev->ifindex = tfile->ifindex;
1631
1632                 tun = netdev_priv(dev);
1633                 tun->dev = dev;
1634                 tun->flags = flags;
1635                 tun->txflt.count = 0;
1636                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1637
1638                 tun->filter_attached = false;
1639                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1640
1641                 spin_lock_init(&tun->lock);
1642
1643                 err = security_tun_dev_alloc_security(&tun->security);
1644                 if (err < 0)
1645                         goto err_free_dev;
1646
1647                 tun_net_init(dev);
1648                 tun_flow_init(tun);
1649
1650                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1651                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1652                                    NETIF_F_HW_VLAN_STAG_TX;
1653                 dev->features = dev->hw_features;
1654                 dev->vlan_features = dev->features;
1655
1656                 INIT_LIST_HEAD(&tun->disabled);
1657                 err = tun_attach(tun, file, false);
1658                 if (err < 0)
1659                         goto err_free_flow;
1660
1661                 err = register_netdevice(tun->dev);
1662                 if (err < 0)
1663                         goto err_detach;
1664
1665                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1666                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1667                     device_create_file(&tun->dev->dev, &dev_attr_group))
1668                         pr_err("Failed to create tun sysfs files\n");
1669         }
1670
1671         netif_carrier_on(tun->dev);
1672
1673         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1674
1675         if (ifr->ifr_flags & IFF_NO_PI)
1676                 tun->flags |= TUN_NO_PI;
1677         else
1678                 tun->flags &= ~TUN_NO_PI;
1679
1680         /* This flag has no real effect.  We track the value for backwards
1681          * compatibility.
1682          */
1683         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1684                 tun->flags |= TUN_ONE_QUEUE;
1685         else
1686                 tun->flags &= ~TUN_ONE_QUEUE;
1687
1688         if (ifr->ifr_flags & IFF_VNET_HDR)
1689                 tun->flags |= TUN_VNET_HDR;
1690         else
1691                 tun->flags &= ~TUN_VNET_HDR;
1692
1693         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1694                 tun->flags |= TUN_TAP_MQ;
1695         else
1696                 tun->flags &= ~TUN_TAP_MQ;
1697
1698         /* Make sure persistent devices do not get stuck in
1699          * xoff state.
1700          */
1701         if (netif_running(tun->dev))
1702                 netif_tx_wake_all_queues(tun->dev);
1703
1704         strcpy(ifr->ifr_name, tun->dev->name);
1705         return 0;
1706
1707 err_detach:
1708         tun_detach_all(dev);
1709 err_free_flow:
1710         tun_flow_uninit(tun);
1711         security_tun_dev_free_security(tun->security);
1712 err_free_dev:
1713         free_netdev(dev);
1714         return err;
1715 }
1716
1717 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1718                        struct ifreq *ifr)
1719 {
1720         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1721
1722         strcpy(ifr->ifr_name, tun->dev->name);
1723
1724         ifr->ifr_flags = tun_flags(tun);
1725
1726 }
1727
1728 /* This is like a cut-down ethtool ops, except done via tun fd so no
1729  * privs required. */
1730 static int set_offload(struct tun_struct *tun, unsigned long arg)
1731 {
1732         netdev_features_t features = 0;
1733
1734         if (arg & TUN_F_CSUM) {
1735                 features |= NETIF_F_HW_CSUM;
1736                 arg &= ~TUN_F_CSUM;
1737
1738                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1739                         if (arg & TUN_F_TSO_ECN) {
1740                                 features |= NETIF_F_TSO_ECN;
1741                                 arg &= ~TUN_F_TSO_ECN;
1742                         }
1743                         if (arg & TUN_F_TSO4)
1744                                 features |= NETIF_F_TSO;
1745                         if (arg & TUN_F_TSO6)
1746                                 features |= NETIF_F_TSO6;
1747                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1748                 }
1749
1750                 if (arg & TUN_F_UFO) {
1751                         features |= NETIF_F_UFO;
1752                         arg &= ~TUN_F_UFO;
1753                 }
1754         }
1755
1756         /* This gives the user a way to test for new features in future by
1757          * trying to set them. */
1758         if (arg)
1759                 return -EINVAL;
1760
1761         tun->set_features = features;
1762         netdev_update_features(tun->dev);
1763
1764         return 0;
1765 }
1766
1767 static void tun_detach_filter(struct tun_struct *tun, int n)
1768 {
1769         int i;
1770         struct tun_file *tfile;
1771
1772         for (i = 0; i < n; i++) {
1773                 tfile = rtnl_dereference(tun->tfiles[i]);
1774                 sk_detach_filter(tfile->socket.sk);
1775         }
1776
1777         tun->filter_attached = false;
1778 }
1779
1780 static int tun_attach_filter(struct tun_struct *tun)
1781 {
1782         int i, ret = 0;
1783         struct tun_file *tfile;
1784
1785         for (i = 0; i < tun->numqueues; i++) {
1786                 tfile = rtnl_dereference(tun->tfiles[i]);
1787                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1788                 if (ret) {
1789                         tun_detach_filter(tun, i);
1790                         return ret;
1791                 }
1792         }
1793
1794         tun->filter_attached = true;
1795         return ret;
1796 }
1797
1798 static void tun_set_sndbuf(struct tun_struct *tun)
1799 {
1800         struct tun_file *tfile;
1801         int i;
1802
1803         for (i = 0; i < tun->numqueues; i++) {
1804                 tfile = rtnl_dereference(tun->tfiles[i]);
1805                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1806         }
1807 }
1808
1809 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1810 {
1811         struct tun_file *tfile = file->private_data;
1812         struct tun_struct *tun;
1813         int ret = 0;
1814
1815         rtnl_lock();
1816
1817         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1818                 tun = tfile->detached;
1819                 if (!tun) {
1820                         ret = -EINVAL;
1821                         goto unlock;
1822                 }
1823                 ret = security_tun_dev_attach_queue(tun->security);
1824                 if (ret < 0)
1825                         goto unlock;
1826                 ret = tun_attach(tun, file, false);
1827         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1828                 tun = rtnl_dereference(tfile->tun);
1829                 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1830                         ret = -EINVAL;
1831                 else
1832                         __tun_detach(tfile, false);
1833         } else
1834                 ret = -EINVAL;
1835
1836 unlock:
1837         rtnl_unlock();
1838         return ret;
1839 }
1840
1841 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1842                             unsigned long arg, int ifreq_len)
1843 {
1844         struct tun_file *tfile = file->private_data;
1845         struct tun_struct *tun;
1846         void __user* argp = (void __user*)arg;
1847         struct ifreq ifr;
1848         kuid_t owner;
1849         kgid_t group;
1850         int sndbuf;
1851         int vnet_hdr_sz;
1852         unsigned int ifindex;
1853         int ret;
1854
1855         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1856                 if (copy_from_user(&ifr, argp, ifreq_len))
1857                         return -EFAULT;
1858         } else {
1859                 memset(&ifr, 0, sizeof(ifr));
1860         }
1861         if (cmd == TUNGETFEATURES) {
1862                 /* Currently this just means: "what IFF flags are valid?".
1863                  * This is needed because we never checked for invalid flags on
1864                  * TUNSETIFF. */
1865                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1866                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1867                                 (unsigned int __user*)argp);
1868         } else if (cmd == TUNSETQUEUE)
1869                 return tun_set_queue(file, &ifr);
1870
1871         ret = 0;
1872         rtnl_lock();
1873
1874         tun = __tun_get(tfile);
1875         if (cmd == TUNSETIFF && !tun) {
1876                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1877
1878                 ret = tun_set_iff(tfile->net, file, &ifr);
1879
1880                 if (ret)
1881                         goto unlock;
1882
1883                 if (copy_to_user(argp, &ifr, ifreq_len))
1884                         ret = -EFAULT;
1885                 goto unlock;
1886         }
1887         if (cmd == TUNSETIFINDEX) {
1888                 ret = -EPERM;
1889                 if (tun)
1890                         goto unlock;
1891
1892                 ret = -EFAULT;
1893                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1894                         goto unlock;
1895
1896                 ret = 0;
1897                 tfile->ifindex = ifindex;
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 (tfile->detached)
1913                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
1914                 if (!tfile->socket.sk->sk_filter)
1915                         ifr.ifr_flags |= IFF_NOFILTER;
1916
1917                 if (copy_to_user(argp, &ifr, ifreq_len))
1918                         ret = -EFAULT;
1919                 break;
1920
1921         case TUNSETNOCSUM:
1922                 /* Disable/Enable checksum */
1923
1924                 /* [unimplemented] */
1925                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1926                           arg ? "disabled" : "enabled");
1927                 break;
1928
1929         case TUNSETPERSIST:
1930                 /* Disable/Enable persist mode. Keep an extra reference to the
1931                  * module to prevent the module being unprobed.
1932                  */
1933                 if (arg && !(tun->flags & TUN_PERSIST)) {
1934                         tun->flags |= TUN_PERSIST;
1935                         __module_get(THIS_MODULE);
1936                 }
1937                 if (!arg && (tun->flags & TUN_PERSIST)) {
1938                         tun->flags &= ~TUN_PERSIST;
1939                         module_put(THIS_MODULE);
1940                 }
1941
1942                 tun_debug(KERN_INFO, tun, "persist %s\n",
1943                           arg ? "enabled" : "disabled");
1944                 break;
1945
1946         case TUNSETOWNER:
1947                 /* Set owner of the device */
1948                 owner = make_kuid(current_user_ns(), arg);
1949                 if (!uid_valid(owner)) {
1950                         ret = -EINVAL;
1951                         break;
1952                 }
1953                 tun->owner = owner;
1954                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1955                           from_kuid(&init_user_ns, tun->owner));
1956                 break;
1957
1958         case TUNSETGROUP:
1959                 /* Set group of the device */
1960                 group = make_kgid(current_user_ns(), arg);
1961                 if (!gid_valid(group)) {
1962                         ret = -EINVAL;
1963                         break;
1964                 }
1965                 tun->group = group;
1966                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1967                           from_kgid(&init_user_ns, tun->group));
1968                 break;
1969
1970         case TUNSETLINK:
1971                 /* Only allow setting the type when the interface is down */
1972                 if (tun->dev->flags & IFF_UP) {
1973                         tun_debug(KERN_INFO, tun,
1974                                   "Linktype set failed because interface is up\n");
1975                         ret = -EBUSY;
1976                 } else {
1977                         tun->dev->type = (int) arg;
1978                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1979                                   tun->dev->type);
1980                         ret = 0;
1981                 }
1982                 break;
1983
1984 #ifdef TUN_DEBUG
1985         case TUNSETDEBUG:
1986                 tun->debug = arg;
1987                 break;
1988 #endif
1989         case TUNSETOFFLOAD:
1990                 ret = set_offload(tun, arg);
1991                 break;
1992
1993         case TUNSETTXFILTER:
1994                 /* Can be set only for TAPs */
1995                 ret = -EINVAL;
1996                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1997                         break;
1998                 ret = update_filter(&tun->txflt, (void __user *)arg);
1999                 break;
2000
2001         case SIOCGIFHWADDR:
2002                 /* Get hw address */
2003                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2004                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2005                 if (copy_to_user(argp, &ifr, ifreq_len))
2006                         ret = -EFAULT;
2007                 break;
2008
2009         case SIOCSIFHWADDR:
2010                 /* Set hw address */
2011                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2012                           ifr.ifr_hwaddr.sa_data);
2013
2014                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2015                 break;
2016
2017         case TUNGETSNDBUF:
2018                 sndbuf = tfile->socket.sk->sk_sndbuf;
2019                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2020                         ret = -EFAULT;
2021                 break;
2022
2023         case TUNSETSNDBUF:
2024                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2025                         ret = -EFAULT;
2026                         break;
2027                 }
2028
2029                 tun->sndbuf = sndbuf;
2030                 tun_set_sndbuf(tun);
2031                 break;
2032
2033         case TUNGETVNETHDRSZ:
2034                 vnet_hdr_sz = tun->vnet_hdr_sz;
2035                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2036                         ret = -EFAULT;
2037                 break;
2038
2039         case TUNSETVNETHDRSZ:
2040                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2041                         ret = -EFAULT;
2042                         break;
2043                 }
2044                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2045                         ret = -EINVAL;
2046                         break;
2047                 }
2048
2049                 tun->vnet_hdr_sz = vnet_hdr_sz;
2050                 break;
2051
2052         case TUNATTACHFILTER:
2053                 /* Can be set only for TAPs */
2054                 ret = -EINVAL;
2055                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2056                         break;
2057                 ret = -EFAULT;
2058                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2059                         break;
2060
2061                 ret = tun_attach_filter(tun);
2062                 break;
2063
2064         case TUNDETACHFILTER:
2065                 /* Can be set only for TAPs */
2066                 ret = -EINVAL;
2067                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2068                         break;
2069                 ret = 0;
2070                 tun_detach_filter(tun, tun->numqueues);
2071                 break;
2072
2073         case TUNGETFILTER:
2074                 ret = -EINVAL;
2075                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2076                         break;
2077                 ret = -EFAULT;
2078                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2079                         break;
2080                 ret = 0;
2081                 break;
2082
2083         default:
2084                 ret = -EINVAL;
2085                 break;
2086         }
2087
2088 unlock:
2089         rtnl_unlock();
2090         if (tun)
2091                 tun_put(tun);
2092         return ret;
2093 }
2094
2095 static long tun_chr_ioctl(struct file *file,
2096                           unsigned int cmd, unsigned long arg)
2097 {
2098         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2099 }
2100
2101 #ifdef CONFIG_COMPAT
2102 static long tun_chr_compat_ioctl(struct file *file,
2103                          unsigned int cmd, unsigned long arg)
2104 {
2105         switch (cmd) {
2106         case TUNSETIFF:
2107         case TUNGETIFF:
2108         case TUNSETTXFILTER:
2109         case TUNGETSNDBUF:
2110         case TUNSETSNDBUF:
2111         case SIOCGIFHWADDR:
2112         case SIOCSIFHWADDR:
2113                 arg = (unsigned long)compat_ptr(arg);
2114                 break;
2115         default:
2116                 arg = (compat_ulong_t)arg;
2117                 break;
2118         }
2119
2120         /*
2121          * compat_ifreq is shorter than ifreq, so we must not access beyond
2122          * the end of that structure. All fields that are used in this
2123          * driver are compatible though, we don't need to convert the
2124          * contents.
2125          */
2126         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2127 }
2128 #endif /* CONFIG_COMPAT */
2129
2130 static int tun_chr_fasync(int fd, struct file *file, int on)
2131 {
2132         struct tun_file *tfile = file->private_data;
2133         int ret;
2134
2135         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2136                 goto out;
2137
2138         if (on) {
2139                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2140                 if (ret)
2141                         goto out;
2142                 tfile->flags |= TUN_FASYNC;
2143         } else
2144                 tfile->flags &= ~TUN_FASYNC;
2145         ret = 0;
2146 out:
2147         return ret;
2148 }
2149
2150 static int tun_chr_open(struct inode *inode, struct file * file)
2151 {
2152         struct tun_file *tfile;
2153
2154         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2155
2156         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2157                                             &tun_proto);
2158         if (!tfile)
2159                 return -ENOMEM;
2160         rcu_assign_pointer(tfile->tun, NULL);
2161         tfile->net = get_net(current->nsproxy->net_ns);
2162         tfile->flags = 0;
2163         tfile->ifindex = 0;
2164
2165         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2166         init_waitqueue_head(&tfile->wq.wait);
2167
2168         tfile->socket.file = file;
2169         tfile->socket.ops = &tun_socket_ops;
2170
2171         sock_init_data(&tfile->socket, &tfile->sk);
2172         sk_change_net(&tfile->sk, tfile->net);
2173
2174         tfile->sk.sk_write_space = tun_sock_write_space;
2175         tfile->sk.sk_sndbuf = INT_MAX;
2176
2177         file->private_data = tfile;
2178         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2179         INIT_LIST_HEAD(&tfile->next);
2180
2181         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2182
2183         return 0;
2184 }
2185
2186 static int tun_chr_close(struct inode *inode, struct file *file)
2187 {
2188         struct tun_file *tfile = file->private_data;
2189         struct net *net = tfile->net;
2190
2191         tun_detach(tfile, true);
2192         put_net(net);
2193
2194         return 0;
2195 }
2196
2197 static const struct file_operations tun_fops = {
2198         .owner  = THIS_MODULE,
2199         .llseek = no_llseek,
2200         .read  = do_sync_read,
2201         .aio_read  = tun_chr_aio_read,
2202         .write = do_sync_write,
2203         .aio_write = tun_chr_aio_write,
2204         .poll   = tun_chr_poll,
2205         .unlocked_ioctl = tun_chr_ioctl,
2206 #ifdef CONFIG_COMPAT
2207         .compat_ioctl = tun_chr_compat_ioctl,
2208 #endif
2209         .open   = tun_chr_open,
2210         .release = tun_chr_close,
2211         .fasync = tun_chr_fasync
2212 };
2213
2214 static struct miscdevice tun_miscdev = {
2215         .minor = TUN_MINOR,
2216         .name = "tun",
2217         .nodename = "net/tun",
2218         .fops = &tun_fops,
2219 };
2220
2221 /* ethtool interface */
2222
2223 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2224 {
2225         cmd->supported          = 0;
2226         cmd->advertising        = 0;
2227         ethtool_cmd_speed_set(cmd, SPEED_10);
2228         cmd->duplex             = DUPLEX_FULL;
2229         cmd->port               = PORT_TP;
2230         cmd->phy_address        = 0;
2231         cmd->transceiver        = XCVR_INTERNAL;
2232         cmd->autoneg            = AUTONEG_DISABLE;
2233         cmd->maxtxpkt           = 0;
2234         cmd->maxrxpkt           = 0;
2235         return 0;
2236 }
2237
2238 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2239 {
2240         struct tun_struct *tun = netdev_priv(dev);
2241
2242         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2243         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2244
2245         switch (tun->flags & TUN_TYPE_MASK) {
2246         case TUN_TUN_DEV:
2247                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2248                 break;
2249         case TUN_TAP_DEV:
2250                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2251                 break;
2252         }
2253 }
2254
2255 static u32 tun_get_msglevel(struct net_device *dev)
2256 {
2257 #ifdef TUN_DEBUG
2258         struct tun_struct *tun = netdev_priv(dev);
2259         return tun->debug;
2260 #else
2261         return -EOPNOTSUPP;
2262 #endif
2263 }
2264
2265 static void tun_set_msglevel(struct net_device *dev, u32 value)
2266 {
2267 #ifdef TUN_DEBUG
2268         struct tun_struct *tun = netdev_priv(dev);
2269         tun->debug = value;
2270 #endif
2271 }
2272
2273 static const struct ethtool_ops tun_ethtool_ops = {
2274         .get_settings   = tun_get_settings,
2275         .get_drvinfo    = tun_get_drvinfo,
2276         .get_msglevel   = tun_get_msglevel,
2277         .set_msglevel   = tun_set_msglevel,
2278         .get_link       = ethtool_op_get_link,
2279         .get_ts_info    = ethtool_op_get_ts_info,
2280 };
2281
2282
2283 static int __init tun_init(void)
2284 {
2285         int ret = 0;
2286
2287         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2288         pr_info("%s\n", DRV_COPYRIGHT);
2289
2290         ret = rtnl_link_register(&tun_link_ops);
2291         if (ret) {
2292                 pr_err("Can't register link_ops\n");
2293                 goto err_linkops;
2294         }
2295
2296         ret = misc_register(&tun_miscdev);
2297         if (ret) {
2298                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2299                 goto err_misc;
2300         }
2301         return  0;
2302 err_misc:
2303         rtnl_link_unregister(&tun_link_ops);
2304 err_linkops:
2305         return ret;
2306 }
2307
2308 static void tun_cleanup(void)
2309 {
2310         misc_deregister(&tun_miscdev);
2311         rtnl_link_unregister(&tun_link_ops);
2312 }
2313
2314 /* Get an underlying socket object from tun file.  Returns error unless file is
2315  * attached to a device.  The returned object works like a packet socket, it
2316  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2317  * holding a reference to the file for as long as the socket is in use. */
2318 struct socket *tun_get_socket(struct file *file)
2319 {
2320         struct tun_file *tfile;
2321         if (file->f_op != &tun_fops)
2322                 return ERR_PTR(-EINVAL);
2323         tfile = file->private_data;
2324         if (!tfile)
2325                 return ERR_PTR(-EBADFD);
2326         return &tfile->socket;
2327 }
2328 EXPORT_SYMBOL_GPL(tun_get_socket);
2329
2330 module_init(tun_init);
2331 module_exit(tun_cleanup);
2332 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2333 MODULE_AUTHOR(DRV_COPYRIGHT);
2334 MODULE_LICENSE("GPL");
2335 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2336 MODULE_ALIAS("devname:net/tun");