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/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/ipv6.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <linux/seq_file.h>
  74
  75 #include <asm/uaccess.h>
  76
  77 /* Uncomment to enable debugging */
  78 /* #define TUN_DEBUG 1 */
  79
  80 #ifdef TUN_DEBUG
  81 static int debug;
  82
  83 #define tun_debug(level, tun, fmt, args...)                     \
  84 do {                                                            \
  85         if (tun->debug)                                         \
  86                 netdev_printk(level, tun->dev, fmt, ##args);    \
  87 } while (0)
  88 #define DBG1(level, fmt, args...)                               \
  89 do {                                                            \
  90         if (debug == 2)                                         \
  91                 printk(level fmt, ##args);                      \
  92 } while (0)
  93 #else
  94 #define tun_debug(level, tun, fmt, args...)                     \
  95 do {                                                            \
  96         if (0)                                                  \
  97                 netdev_printk(level, tun->dev, fmt, ##args);    \
  98 } while (0)
  99 #define DBG1(level, fmt, args...)                               \
 100 do {                                                            \
 101         if (0)                                                  \
 102                 printk(level fmt, ##args);                      \
 103 } while (0)
 104 #endif
 105
 106 #define GOODCOPY_LEN 128
 107
 108 #define FLT_EXACT_COUNT 8
 109 struct tap_filter {
 110         unsigned int    count;    /* Number of addrs. Zero means disabled */
 111         u32             mask[2];  /* Mask of the hashed addrs */
 112         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 113 };
 114
 115 /* DEFAULT_MAX_NUM_RSS_QUEUES were chosen to let the rx/tx queues allocated for
 116  * the netdevice to be fit in one page. So we can make sure the success of
 117  * memory allocation. TODO: increase the limit. */
 118 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
 119 #define MAX_TAP_FLOWS  4096
 120
 121 #define TUN_FLOW_EXPIRE (3 * HZ)
 122
 123 /* A tun_file connects an open character device to a tuntap netdevice. It
 124  * also contains all socket related structures (except sock_fprog and tap_filter)
 125  * to serve as one transmit queue for tuntap device. The sock_fprog and
 126  * tap_filter were kept in tun_struct since they were used for filtering for the
 127  * netdevice not for a specific queue (at least I didn't see the requirement for
 128  * this).
 129  *
 130  * RCU usage:
 131  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 132  * other can only be read while rcu_read_lock or rtnl_lock is held.
 133  */
 134 struct tun_file {
 135         struct sock sk;
 136         struct socket socket;
 137         struct socket_wq wq;
 138         struct tun_struct __rcu *tun;
 139         struct net *net;
 140         struct fasync_struct *fasync;
 141         /* only used for fasnyc */
 142         unsigned int flags;
 143         union {
 144                 u16 queue_index;
 145                 unsigned int ifindex;
 146         };
 147         struct list_head next;
 148         struct tun_struct *detached;
 149 };
 150
 151 struct tun_flow_entry {
 152         struct hlist_node hash_link;
 153         struct rcu_head rcu;
 154         struct tun_struct *tun;
 155
 156         u32 rxhash;
 157         u32 rps_rxhash;
 158         int queue_index;
 159         unsigned long updated;
 160 };
 161
 162 #define TUN_NUM_FLOW_ENTRIES 1024
 163
 164 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 165  * device, socket filter, sndbuf and vnet header size were restore when the
 166  * file were attached to a persist device.
 167  */
 168 struct tun_struct {
 169         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 170         unsigned int            numqueues;
 171         unsigned int            flags;
 172         kuid_t                  owner;
 173         kgid_t                  group;
 174
 175         struct net_device       *dev;
 176         netdev_features_t       set_features;
 177 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 178                           NETIF_F_TSO6)
 179
 180         int                     vnet_hdr_sz;
 181         int                     sndbuf;
 182         struct tap_filter       txflt;
 183         struct sock_fprog       fprog;
 184         /* protected by rtnl lock */
 185         bool                    filter_attached;
 186 #ifdef TUN_DEBUG
 187         int debug;
 188 #endif
 189         spinlock_t lock;
 190         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 191         struct timer_list flow_gc_timer;
 192         unsigned long ageing_time;
 193         unsigned int numdisabled;
 194         struct list_head disabled;
 195         void *security;
 196         u32 flow_count;
 197 };
 198
 199 static inline u32 tun_hashfn(u32 rxhash)
 200 {
 201         return rxhash & 0x3ff;
 202 }
 203
 204 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 205 {
 206         struct tun_flow_entry *e;
 207
 208         hlist_for_each_entry_rcu(e, head, hash_link) {
 209                 if (e->rxhash == rxhash)
 210                         return e;
 211         }
 212         return NULL;
 213 }
 214
 215 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 216                                               struct hlist_head *head,
 217                                               u32 rxhash, u16 queue_index)
 218 {
 219         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 220
 221         if (e) {
 222                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 223                           rxhash, queue_index);
 224                 e->updated = jiffies;
 225                 e->rxhash = rxhash;
 226                 e->rps_rxhash = 0;
 227                 e->queue_index = queue_index;
 228                 e->tun = tun;
 229                 hlist_add_head_rcu(&e->hash_link, head);
 230                 ++tun->flow_count;
 231         }
 232         return e;
 233 }
 234
 235 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 236 {
 237         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 238                   e->rxhash, e->queue_index);
 239         sock_rps_reset_flow_hash(e->rps_rxhash);
 240         hlist_del_rcu(&e->hash_link);
 241         kfree_rcu(e, rcu);
 242         --tun->flow_count;
 243 }
 244
 245 static void tun_flow_flush(struct tun_struct *tun)
 246 {
 247         int i;
 248
 249         spin_lock_bh(&tun->lock);
 250         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 251                 struct tun_flow_entry *e;
 252                 struct hlist_node *n;
 253
 254                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 255                         tun_flow_delete(tun, e);
 256         }
 257         spin_unlock_bh(&tun->lock);
 258 }
 259
 260 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 261 {
 262         int i;
 263
 264         spin_lock_bh(&tun->lock);
 265         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 266                 struct tun_flow_entry *e;
 267                 struct hlist_node *n;
 268
 269                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 270                         if (e->queue_index == queue_index)
 271                                 tun_flow_delete(tun, e);
 272                 }
 273         }
 274         spin_unlock_bh(&tun->lock);
 275 }
 276
 277 static void tun_flow_cleanup(unsigned long data)
 278 {
 279         struct tun_struct *tun = (struct tun_struct *)data;
 280         unsigned long delay = tun->ageing_time;
 281         unsigned long next_timer = jiffies + delay;
 282         unsigned long count = 0;
 283         int i;
 284
 285         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 286
 287         spin_lock_bh(&tun->lock);
 288         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 289                 struct tun_flow_entry *e;
 290                 struct hlist_node *n;
 291
 292                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 293                         unsigned long this_timer;
 294                         count++;
 295                         this_timer = e->updated + delay;
 296                         if (time_before_eq(this_timer, jiffies))
 297                                 tun_flow_delete(tun, e);
 298                         else if (time_before(this_timer, next_timer))
 299                                 next_timer = this_timer;
 300                 }
 301         }
 302
 303         if (count)
 304                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 305         spin_unlock_bh(&tun->lock);
 306 }
 307
 308 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 309                             struct tun_file *tfile)
 310 {
 311         struct hlist_head *head;
 312         struct tun_flow_entry *e;
 313         unsigned long delay = tun->ageing_time;
 314         u16 queue_index = tfile->queue_index;
 315
 316         if (!rxhash)
 317                 return;
 318         else
 319                 head = &tun->flows[tun_hashfn(rxhash)];
 320
 321         rcu_read_lock();
 322
 323         /* We may get a very small possibility of OOO during switching, not
 324          * worth to optimize.*/
 325         if (tun->numqueues == 1 || tfile->detached)
 326                 goto unlock;
 327
 328         e = tun_flow_find(head, rxhash);
 329         if (likely(e)) {
 330                 /* TODO: keep queueing to old queue until it's empty? */
 331                 e->queue_index = queue_index;
 332                 e->updated = jiffies;
 333                 sock_rps_record_flow_hash(e->rps_rxhash);
 334         } else {
 335                 spin_lock_bh(&tun->lock);
 336                 if (!tun_flow_find(head, rxhash) &&
 337                     tun->flow_count < MAX_TAP_FLOWS)
 338                         tun_flow_create(tun, head, rxhash, queue_index);
 339
 340                 if (!timer_pending(&tun->flow_gc_timer))
 341                         mod_timer(&tun->flow_gc_timer,
 342                                   round_jiffies_up(jiffies + delay));
 343                 spin_unlock_bh(&tun->lock);
 344         }
 345
 346 unlock:
 347         rcu_read_unlock();
 348 }
 349
 350 /**
 351  * Save the hash received in the stack receive path and update the
 352  * flow_hash table accordingly.
 353  */
 354 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 355 {
 356         if (unlikely(e->rps_rxhash != hash)) {
 357                 sock_rps_reset_flow_hash(e->rps_rxhash);
 358                 e->rps_rxhash = hash;
 359         }
 360 }
 361
 362 /* We try to identify a flow through its rxhash first. The reason that
 363  * we do not check rxq no. is because some cards(e.g 82599), chooses
 364  * the rxq based on the txq where the last packet of the flow comes. As
 365  * the userspace application move between processors, we may get a
 366  * different rxq no. here. If we could not get rxhash, then we would
 367  * hope the rxq no. may help here.
 368  */
 369 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 370                             void *accel_priv, select_queue_fallback_t fallback)
 371 {
 372         struct tun_struct *tun = netdev_priv(dev);
 373         struct tun_flow_entry *e;
 374         u32 txq = 0;
 375         u32 numqueues = 0;
 376
 377         rcu_read_lock();
 378         numqueues = ACCESS_ONCE(tun->numqueues);
 379
 380         txq = skb_get_hash(skb);
 381         if (txq) {
 382                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 383                 if (e) {
 384                         tun_flow_save_rps_rxhash(e, txq);
 385                         txq = e->queue_index;
 386                 } else
 387                         /* use multiply and shift instead of expensive divide */
 388                         txq = ((u64)txq * numqueues) >> 32;
 389         } else if (likely(skb_rx_queue_recorded(skb))) {
 390                 txq = skb_get_rx_queue(skb);
 391                 while (unlikely(txq >= numqueues))
 392                         txq -= numqueues;
 393         }
 394
 395         rcu_read_unlock();
 396         return txq;
 397 }
 398
 399 static inline bool tun_not_capable(struct tun_struct *tun)
 400 {
 401         const struct cred *cred = current_cred();
 402         struct net *net = dev_net(tun->dev);
 403
 404         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 405                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 406                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 407 }
 408
 409 static void tun_set_real_num_queues(struct tun_struct *tun)
 410 {
 411         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 412         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 413 }
 414
 415 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 416 {
 417         tfile->detached = tun;
 418         list_add_tail(&tfile->next, &tun->disabled);
 419         ++tun->numdisabled;
 420 }
 421
 422 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 423 {
 424         struct tun_struct *tun = tfile->detached;
 425
 426         tfile->detached = NULL;
 427         list_del_init(&tfile->next);
 428         --tun->numdisabled;
 429         return tun;
 430 }
 431
 432 static void tun_queue_purge(struct tun_file *tfile)
 433 {
 434         skb_queue_purge(&tfile->sk.sk_receive_queue);
 435         skb_queue_purge(&tfile->sk.sk_error_queue);
 436 }
 437
 438 static void __tun_detach(struct tun_file *tfile, bool clean)
 439 {
 440         struct tun_file *ntfile;
 441         struct tun_struct *tun;
 442
 443         tun = rtnl_dereference(tfile->tun);
 444
 445         if (tun && !tfile->detached) {
 446                 u16 index = tfile->queue_index;
 447                 BUG_ON(index >= tun->numqueues);
 448
 449                 rcu_assign_pointer(tun->tfiles[index],
 450                                    tun->tfiles[tun->numqueues - 1]);
 451                 ntfile = rtnl_dereference(tun->tfiles[index]);
 452                 ntfile->queue_index = index;
 453
 454                 --tun->numqueues;
 455                 if (clean) {
 456                         rcu_assign_pointer(tfile->tun, NULL);
 457                         sock_put(&tfile->sk);
 458                 } else
 459                         tun_disable_queue(tun, tfile);
 460
 461                 synchronize_net();
 462                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 463                 /* Drop read queue */
 464                 tun_queue_purge(tfile);
 465                 tun_set_real_num_queues(tun);
 466         } else if (tfile->detached && clean) {
 467                 tun = tun_enable_queue(tfile);
 468                 sock_put(&tfile->sk);
 469         }
 470
 471         if (clean) {
 472                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 473                         netif_carrier_off(tun->dev);
 474
 475                         if (!(tun->flags & TUN_PERSIST) &&
 476                             tun->dev->reg_state == NETREG_REGISTERED)
 477                                 unregister_netdevice(tun->dev);
 478                 }
 479
 480                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 481                                  &tfile->socket.flags));
 482                 sk_release_kernel(&tfile->sk);
 483         }
 484 }
 485
 486 static void tun_detach(struct tun_file *tfile, bool clean)
 487 {
 488         rtnl_lock();
 489         __tun_detach(tfile, clean);
 490         rtnl_unlock();
 491 }
 492
 493 static void tun_detach_all(struct net_device *dev)
 494 {
 495         struct tun_struct *tun = netdev_priv(dev);
 496         struct tun_file *tfile, *tmp;
 497         int i, n = tun->numqueues;
 498
 499         for (i = 0; i < n; i++) {
 500                 tfile = rtnl_dereference(tun->tfiles[i]);
 501                 BUG_ON(!tfile);
 502                 wake_up_all(&tfile->wq.wait);
 503                 rcu_assign_pointer(tfile->tun, NULL);
 504                 --tun->numqueues;
 505         }
 506         list_for_each_entry(tfile, &tun->disabled, next) {
 507                 wake_up_all(&tfile->wq.wait);
 508                 rcu_assign_pointer(tfile->tun, NULL);
 509         }
 510         BUG_ON(tun->numqueues != 0);
 511
 512         synchronize_net();
 513         for (i = 0; i < n; i++) {
 514                 tfile = rtnl_dereference(tun->tfiles[i]);
 515                 /* Drop read queue */
 516                 tun_queue_purge(tfile);
 517                 sock_put(&tfile->sk);
 518         }
 519         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 520                 tun_enable_queue(tfile);
 521                 tun_queue_purge(tfile);
 522                 sock_put(&tfile->sk);
 523         }
 524         BUG_ON(tun->numdisabled != 0);
 525
 526         if (tun->flags & TUN_PERSIST)
 527                 module_put(THIS_MODULE);
 528 }
 529
 530 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 531 {
 532         struct tun_file *tfile = file->private_data;
 533         int err;
 534
 535         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 536         if (err < 0)
 537                 goto out;
 538
 539         err = -EINVAL;
 540         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 541                 goto out;
 542
 543         err = -EBUSY;
 544         if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 545                 goto out;
 546
 547         err = -E2BIG;
 548         if (!tfile->detached &&
 549             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 550                 goto out;
 551
 552         err = 0;
 553
 554         /* Re-attach the filter to persist device */
 555         if (!skip_filter && (tun->filter_attached == true)) {
 556                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 557                 if (!err)
 558                         goto out;
 559         }
 560         tfile->queue_index = tun->numqueues;
 561         rcu_assign_pointer(tfile->tun, tun);
 562         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 563         tun->numqueues++;
 564
 565         if (tfile->detached)
 566                 tun_enable_queue(tfile);
 567         else
 568                 sock_hold(&tfile->sk);
 569
 570         tun_set_real_num_queues(tun);
 571
 572         /* device is allowed to go away first, so no need to hold extra
 573          * refcnt.
 574          */
 575
 576 out:
 577         return err;
 578 }
 579
 580 static struct tun_struct *__tun_get(struct tun_file *tfile)
 581 {
 582         struct tun_struct *tun;
 583
 584         rcu_read_lock();
 585         tun = rcu_dereference(tfile->tun);
 586         if (tun)
 587                 dev_hold(tun->dev);
 588         rcu_read_unlock();
 589
 590         return tun;
 591 }
 592
 593 static struct tun_struct *tun_get(struct file *file)
 594 {
 595         return __tun_get(file->private_data);
 596 }
 597
 598 static void tun_put(struct tun_struct *tun)
 599 {
 600         dev_put(tun->dev);
 601 }
 602
 603 /* TAP filtering */
 604 static void addr_hash_set(u32 *mask, const u8 *addr)
 605 {
 606         int n = ether_crc(ETH_ALEN, addr) >> 26;
 607         mask[n >> 5] |= (1 << (n & 31));
 608 }
 609
 610 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 611 {
 612         int n = ether_crc(ETH_ALEN, addr) >> 26;
 613         return mask[n >> 5] & (1 << (n & 31));
 614 }
 615
 616 static int update_filter(struct tap_filter *filter, void __user *arg)
 617 {
 618         struct { u8 u[ETH_ALEN]; } *addr;
 619         struct tun_filter uf;
 620         int err, alen, n, nexact;
 621
 622         if (copy_from_user(&uf, arg, sizeof(uf)))
 623                 return -EFAULT;
 624
 625         if (!uf.count) {
 626                 /* Disabled */
 627                 filter->count = 0;
 628                 return 0;
 629         }
 630
 631         alen = ETH_ALEN * uf.count;
 632         addr = kmalloc(alen, GFP_KERNEL);
 633         if (!addr)
 634                 return -ENOMEM;
 635
 636         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 637                 err = -EFAULT;
 638                 goto done;
 639         }
 640
 641         /* The filter is updated without holding any locks. Which is
 642          * perfectly safe. We disable it first and in the worst
 643          * case we'll accept a few undesired packets. */
 644         filter->count = 0;
 645         wmb();
 646
 647         /* Use first set of addresses as an exact filter */
 648         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 649                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 650
 651         nexact = n;
 652
 653         /* Remaining multicast addresses are hashed,
 654          * unicast will leave the filter disabled. */
 655         memset(filter->mask, 0, sizeof(filter->mask));
 656         for (; n < uf.count; n++) {
 657                 if (!is_multicast_ether_addr(addr[n].u)) {
 658                         err = 0; /* no filter */
 659                         goto done;
 660                 }
 661                 addr_hash_set(filter->mask, addr[n].u);
 662         }
 663
 664         /* For ALLMULTI just set the mask to all ones.
 665          * This overrides the mask populated above. */
 666         if ((uf.flags & TUN_FLT_ALLMULTI))
 667                 memset(filter->mask, ~0, sizeof(filter->mask));
 668
 669         /* Now enable the filter */
 670         wmb();
 671         filter->count = nexact;
 672
 673         /* Return the number of exact filters */
 674         err = nexact;
 675
 676 done:
 677         kfree(addr);
 678         return err;
 679 }
 680
 681 /* Returns: 0 - drop, !=0 - accept */
 682 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 683 {
 684         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 685          * at this point. */
 686         struct ethhdr *eh = (struct ethhdr *) skb->data;
 687         int i;
 688
 689         /* Exact match */
 690         for (i = 0; i < filter->count; i++)
 691                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 692                         return 1;
 693
 694         /* Inexact match (multicast only) */
 695         if (is_multicast_ether_addr(eh->h_dest))
 696                 return addr_hash_test(filter->mask, eh->h_dest);
 697
 698         return 0;
 699 }
 700
 701 /*
 702  * Checks whether the packet is accepted or not.
 703  * Returns: 0 - drop, !=0 - accept
 704  */
 705 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 706 {
 707         if (!filter->count)
 708                 return 1;
 709
 710         return run_filter(filter, skb);
 711 }
 712
 713 /* Network device part of the driver */
 714
 715 static const struct ethtool_ops tun_ethtool_ops;
 716
 717 /* Net device detach from fd. */
 718 static void tun_net_uninit(struct net_device *dev)
 719 {
 720         tun_detach_all(dev);
 721 }
 722
 723 /* Net device open. */
 724 static int tun_net_open(struct net_device *dev)
 725 {
 726         netif_tx_start_all_queues(dev);
 727         return 0;
 728 }
 729
 730 /* Net device close. */
 731 static int tun_net_close(struct net_device *dev)
 732 {
 733         netif_tx_stop_all_queues(dev);
 734         return 0;
 735 }
 736
 737 /* Net device start xmit */
 738 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 739 {
 740         struct tun_struct *tun = netdev_priv(dev);
 741         int txq = skb->queue_mapping;
 742         struct tun_file *tfile;
 743         u32 numqueues = 0;
 744
 745         rcu_read_lock();
 746         tfile = rcu_dereference(tun->tfiles[txq]);
 747         numqueues = ACCESS_ONCE(tun->numqueues);
 748
 749         /* Drop packet if interface is not attached */
 750         if (txq >= numqueues)
 751                 goto drop;
 752
 753         if (numqueues == 1) {
 754                 /* Select queue was not called for the skbuff, so we extract the
 755                  * RPS hash and save it into the flow_table here.
 756                  */
 757                 __u32 rxhash;
 758
 759                 rxhash = skb_get_hash(skb);
 760                 if (rxhash) {
 761                         struct tun_flow_entry *e;
 762                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 763                                         rxhash);
 764                         if (e)
 765                                 tun_flow_save_rps_rxhash(e, rxhash);
 766                 }
 767         }
 768
 769         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 770
 771         BUG_ON(!tfile);
 772
 773         /* Drop if the filter does not like it.
 774          * This is a noop if the filter is disabled.
 775          * Filter can be enabled only for the TAP devices. */
 776         if (!check_filter(&tun->txflt, skb))
 777                 goto drop;
 778
 779         if (tfile->socket.sk->sk_filter &&
 780             sk_filter(tfile->socket.sk, skb))
 781                 goto drop;
 782
 783         /* Limit the number of packets queued by dividing txq length with the
 784          * number of queues.
 785          */
 786         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
 787                           >= dev->tx_queue_len)
 788                 goto drop;
 789
 790         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 791                 goto drop;
 792
 793         if (skb->sk) {
 794                 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
 795                 sw_tx_timestamp(skb);
 796         }
 797
 798         /* Orphan the skb - required as we might hang on to it
 799          * for indefinite time.
 800          */
 801         skb_orphan(skb);
 802
 803         nf_reset(skb);
 804
 805         /* Enqueue packet */
 806         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 807
 808         /* Notify and wake up reader process */
 809         if (tfile->flags & TUN_FASYNC)
 810                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 811         wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
 812                                    POLLRDNORM | POLLRDBAND);
 813
 814         rcu_read_unlock();
 815         return NETDEV_TX_OK;
 816
 817 drop:
 818         dev->stats.tx_dropped++;
 819         skb_tx_error(skb);
 820         kfree_skb(skb);
 821         rcu_read_unlock();
 822         return NETDEV_TX_OK;
 823 }
 824
 825 static void tun_net_mclist(struct net_device *dev)
 826 {
 827         /*
 828          * This callback is supposed to deal with mc filter in
 829          * _rx_ path and has nothing to do with the _tx_ path.
 830          * In rx path we always accept everything userspace gives us.
 831          */
 832 }
 833
 834 #define MIN_MTU 68
 835 #define MAX_MTU 65535
 836
 837 static int
 838 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 839 {
 840         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 841                 return -EINVAL;
 842         dev->mtu = new_mtu;
 843         return 0;
 844 }
 845
 846 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 847         netdev_features_t features)
 848 {
 849         struct tun_struct *tun = netdev_priv(dev);
 850
 851         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 852 }
 853 #ifdef CONFIG_NET_POLL_CONTROLLER
 854 static void tun_poll_controller(struct net_device *dev)
 855 {
 856         /*
 857          * Tun only receives frames when:
 858          * 1) the char device endpoint gets data from user space
 859          * 2) the tun socket gets a sendmsg call from user space
 860          * Since both of those are synchronous operations, we are guaranteed
 861          * never to have pending data when we poll for it
 862          * so there is nothing to do here but return.
 863          * We need this though so netpoll recognizes us as an interface that
 864          * supports polling, which enables bridge devices in virt setups to
 865          * still use netconsole
 866          */
 867         return;
 868 }
 869 #endif
 870 static const struct net_device_ops tun_netdev_ops = {
 871         .ndo_uninit             = tun_net_uninit,
 872         .ndo_open               = tun_net_open,
 873         .ndo_stop               = tun_net_close,
 874         .ndo_start_xmit         = tun_net_xmit,
 875         .ndo_change_mtu         = tun_net_change_mtu,
 876         .ndo_fix_features       = tun_net_fix_features,
 877         .ndo_select_queue       = tun_select_queue,
 878 #ifdef CONFIG_NET_POLL_CONTROLLER
 879         .ndo_poll_controller    = tun_poll_controller,
 880 #endif
 881 };
 882
 883 static const struct net_device_ops tap_netdev_ops = {
 884         .ndo_uninit             = tun_net_uninit,
 885         .ndo_open               = tun_net_open,
 886         .ndo_stop               = tun_net_close,
 887         .ndo_start_xmit         = tun_net_xmit,
 888         .ndo_change_mtu         = tun_net_change_mtu,
 889         .ndo_fix_features       = tun_net_fix_features,
 890         .ndo_set_rx_mode        = tun_net_mclist,
 891         .ndo_set_mac_address    = eth_mac_addr,
 892         .ndo_validate_addr      = eth_validate_addr,
 893         .ndo_select_queue       = tun_select_queue,
 894 #ifdef CONFIG_NET_POLL_CONTROLLER
 895         .ndo_poll_controller    = tun_poll_controller,
 896 #endif
 897 };
 898
 899 static void tun_flow_init(struct tun_struct *tun)
 900 {
 901         int i;
 902
 903         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 904                 INIT_HLIST_HEAD(&tun->flows[i]);
 905
 906         tun->ageing_time = TUN_FLOW_EXPIRE;
 907         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 908         mod_timer(&tun->flow_gc_timer,
 909                   round_jiffies_up(jiffies + tun->ageing_time));
 910 }
 911
 912 static void tun_flow_uninit(struct tun_struct *tun)
 913 {
 914         del_timer_sync(&tun->flow_gc_timer);
 915         tun_flow_flush(tun);
 916 }
 917
 918 /* Initialize net device. */
 919 static void tun_net_init(struct net_device *dev)
 920 {
 921         struct tun_struct *tun = netdev_priv(dev);
 922
 923         switch (tun->flags & TUN_TYPE_MASK) {
 924         case TUN_TUN_DEV:
 925                 dev->netdev_ops = &tun_netdev_ops;
 926
 927                 /* Point-to-Point TUN Device */
 928                 dev->hard_header_len = 0;
 929                 dev->addr_len = 0;
 930                 dev->mtu = 1500;
 931
 932                 /* Zero header length */
 933                 dev->type = ARPHRD_NONE;
 934                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 935                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 936                 break;
 937
 938         case TUN_TAP_DEV:
 939                 dev->netdev_ops = &tap_netdev_ops;
 940                 /* Ethernet TAP Device */
 941                 ether_setup(dev);
 942                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 943                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 944
 945                 eth_hw_addr_random(dev);
 946
 947                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 948                 break;
 949         }
 950 }
 951
 952 /* Character device part */
 953
 954 /* Poll */
 955 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 956 {
 957         struct tun_file *tfile = file->private_data;
 958         struct tun_struct *tun = __tun_get(tfile);
 959         struct sock *sk;
 960         unsigned int mask = 0;
 961
 962         if (!tun)
 963                 return POLLERR;
 964
 965         sk = tfile->socket.sk;
 966
 967         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 968
 969         poll_wait(file, &tfile->wq.wait, wait);
 970
 971         if (!skb_queue_empty(&sk->sk_receive_queue))
 972                 mask |= POLLIN | POLLRDNORM;
 973
 974         if (sock_writeable(sk) ||
 975             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 976              sock_writeable(sk)))
 977                 mask |= POLLOUT | POLLWRNORM;
 978
 979         if (tun->dev->reg_state != NETREG_REGISTERED)
 980                 mask = POLLERR;
 981
 982         tun_put(tun);
 983         return mask;
 984 }
 985
 986 /* prepad is the amount to reserve at front.  len is length after that.
 987  * linear is a hint as to how much to copy (usually headers). */
 988 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 989                                      size_t prepad, size_t len,
 990                                      size_t linear, int noblock)
 991 {
 992         struct sock *sk = tfile->socket.sk;
 993         struct sk_buff *skb;
 994         int err;
 995
 996         /* Under a page?  Don't bother with paged skb. */
 997         if (prepad + len < PAGE_SIZE || !linear)
 998                 linear = len;
 999
1000         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1001                                    &err, 0);
1002         if (!skb)
1003                 return ERR_PTR(err);
1004
1005         skb_reserve(skb, prepad);
1006         skb_put(skb, linear);
1007         skb->data_len = len - linear;
1008         skb->len += len - linear;
1009
1010         return skb;
1011 }
1012
1013 /* Get packet from user space buffer */
1014 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1015                             void *msg_control, const struct iovec *iv,
1016                             size_t total_len, size_t count, int noblock)
1017 {
1018         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1019         struct sk_buff *skb;
1020         size_t len = total_len, align = NET_SKB_PAD, linear;
1021         struct virtio_net_hdr gso = { 0 };
1022         int good_linear;
1023         int offset = 0;
1024         int copylen;
1025         bool zerocopy = false;
1026         int err;
1027         u32 rxhash;
1028
1029         if (!(tun->flags & TUN_NO_PI)) {
1030                 if (len < sizeof(pi))
1031                         return -EINVAL;
1032                 len -= sizeof(pi);
1033
1034                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1035                         return -EFAULT;
1036                 offset += sizeof(pi);
1037         }
1038
1039         if (tun->flags & TUN_VNET_HDR) {
1040                 if (len < tun->vnet_hdr_sz)
1041                         return -EINVAL;
1042                 len -= tun->vnet_hdr_sz;
1043
1044                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1045                         return -EFAULT;
1046
1047                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1048                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1049                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1050
1051                 if (gso.hdr_len > len)
1052                         return -EINVAL;
1053                 offset += tun->vnet_hdr_sz;
1054         }
1055
1056         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1057                 align += NET_IP_ALIGN;
1058                 if (unlikely(len < ETH_HLEN ||
1059                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1060                         return -EINVAL;
1061         }
1062
1063         good_linear = SKB_MAX_HEAD(align);
1064
1065         if (msg_control) {
1066                 /* There are 256 bytes to be copied in skb, so there is
1067                  * enough room for skb expand head in case it is used.
1068                  * The rest of the buffer is mapped from userspace.
1069                  */
1070                 copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1071                 if (copylen > good_linear)
1072                         copylen = good_linear;
1073                 linear = copylen;
1074                 if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1075                         zerocopy = true;
1076         }
1077
1078         if (!zerocopy) {
1079                 copylen = len;
1080                 if (gso.hdr_len > good_linear)
1081                         linear = good_linear;
1082                 else
1083                         linear = gso.hdr_len;
1084         }
1085
1086         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1087         if (IS_ERR(skb)) {
1088                 if (PTR_ERR(skb) != -EAGAIN)
1089                         tun->dev->stats.rx_dropped++;
1090                 return PTR_ERR(skb);
1091         }
1092
1093         if (zerocopy)
1094                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1095         else {
1096                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1097                 if (!err && msg_control) {
1098                         struct ubuf_info *uarg = msg_control;
1099                         uarg->callback(uarg, false);
1100                 }
1101         }
1102
1103         if (err) {
1104                 tun->dev->stats.rx_dropped++;
1105                 kfree_skb(skb);
1106                 return -EFAULT;
1107         }
1108
1109         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1110                 if (!skb_partial_csum_set(skb, gso.csum_start,
1111                                           gso.csum_offset)) {
1112                         tun->dev->stats.rx_frame_errors++;
1113                         kfree_skb(skb);
1114                         return -EINVAL;
1115                 }
1116         }
1117
1118         switch (tun->flags & TUN_TYPE_MASK) {
1119         case TUN_TUN_DEV:
1120                 if (tun->flags & TUN_NO_PI) {
1121                         switch (skb->data[0] & 0xf0) {
1122                         case 0x40:
1123                                 pi.proto = htons(ETH_P_IP);
1124                                 break;
1125                         case 0x60:
1126                                 pi.proto = htons(ETH_P_IPV6);
1127                                 break;
1128                         default:
1129                                 tun->dev->stats.rx_dropped++;
1130                                 kfree_skb(skb);
1131                                 return -EINVAL;
1132                         }
1133                 }
1134
1135                 skb_reset_mac_header(skb);
1136                 skb->protocol = pi.proto;
1137                 skb->dev = tun->dev;
1138                 break;
1139         case TUN_TAP_DEV:
1140                 skb->protocol = eth_type_trans(skb, tun->dev);
1141                 break;
1142         }
1143
1144         skb_reset_network_header(skb);
1145
1146         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1147                 pr_debug("GSO!\n");
1148                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1149                 case VIRTIO_NET_HDR_GSO_TCPV4:
1150                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1151                         break;
1152                 case VIRTIO_NET_HDR_GSO_TCPV6:
1153                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1154                         break;
1155                 case VIRTIO_NET_HDR_GSO_UDP:
1156                 {
1157                         static bool warned;
1158
1159                         if (!warned) {
1160                                 warned = true;
1161                                 netdev_warn(tun->dev,
1162                                             "%s: using disabled UFO feature; please fix this program\n",
1163                                             current->comm);
1164                         }
1165                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1166                         if (skb->protocol == htons(ETH_P_IPV6))
1167                                 ipv6_proxy_select_ident(skb);
1168                         break;
1169                 }
1170                 default:
1171                         tun->dev->stats.rx_frame_errors++;
1172                         kfree_skb(skb);
1173                         return -EINVAL;
1174                 }
1175
1176                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1177                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1178
1179                 skb_shinfo(skb)->gso_size = gso.gso_size;
1180                 if (skb_shinfo(skb)->gso_size == 0) {
1181                         tun->dev->stats.rx_frame_errors++;
1182                         kfree_skb(skb);
1183                         return -EINVAL;
1184                 }
1185
1186                 /* Header must be checked, and gso_segs computed. */
1187                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1188                 skb_shinfo(skb)->gso_segs = 0;
1189         }
1190
1191         /* copy skb_ubuf_info for callback when skb has no error */
1192         if (zerocopy) {
1193                 skb_shinfo(skb)->destructor_arg = msg_control;
1194                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1195                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1196         }
1197
1198         skb_probe_transport_header(skb, 0);
1199
1200         rxhash = skb_get_hash(skb);
1201         netif_rx_ni(skb);
1202
1203         tun->dev->stats.rx_packets++;
1204         tun->dev->stats.rx_bytes += len;
1205
1206         tun_flow_update(tun, rxhash, tfile);
1207         return total_len;
1208 }
1209
1210 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1211                               unsigned long count, loff_t pos)
1212 {
1213         struct file *file = iocb->ki_filp;
1214         struct tun_struct *tun = tun_get(file);
1215         struct tun_file *tfile = file->private_data;
1216         ssize_t result;
1217
1218         if (!tun)
1219                 return -EBADFD;
1220
1221         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1222
1223         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1224                               count, file->f_flags & O_NONBLOCK);
1225
1226         tun_put(tun);
1227         return result;
1228 }
1229
1230 /* Put packet to the user space buffer */
1231 static ssize_t tun_put_user(struct tun_struct *tun,
1232                             struct tun_file *tfile,
1233                             struct sk_buff *skb,
1234                             const struct iovec *iv, int len)
1235 {
1236         struct tun_pi pi = { 0, skb->protocol };
1237         ssize_t total = 0;
1238         int vlan_offset = 0, copied;
1239
1240         if (!(tun->flags & TUN_NO_PI)) {
1241                 if ((len -= sizeof(pi)) < 0)
1242                         return -EINVAL;
1243
1244                 if (len < skb->len) {
1245                         /* Packet will be striped */
1246                         pi.flags |= TUN_PKT_STRIP;
1247                 }
1248
1249                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1250                         return -EFAULT;
1251                 total += sizeof(pi);
1252         }
1253
1254         if (tun->flags & TUN_VNET_HDR) {
1255                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1256                 if ((len -= tun->vnet_hdr_sz) < 0)
1257                         return -EINVAL;
1258
1259                 if (skb_is_gso(skb)) {
1260                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1261
1262                         /* This is a hint as to how much should be linear. */
1263                         gso.hdr_len = skb_headlen(skb);
1264                         gso.gso_size = sinfo->gso_size;
1265                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1266                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1267                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1268                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1269                         else {
1270                                 pr_err("unexpected GSO type: "
1271                                        "0x%x, gso_size %d, hdr_len %d\n",
1272                                        sinfo->gso_type, gso.gso_size,
1273                                        gso.hdr_len);
1274                                 print_hex_dump(KERN_ERR, "tun: ",
1275                                                DUMP_PREFIX_NONE,
1276                                                16, 1, skb->head,
1277                                                min((int)gso.hdr_len, 64), true);
1278                                 WARN_ON_ONCE(1);
1279                                 return -EINVAL;
1280                         }
1281                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1282                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1283                 } else
1284                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1285
1286                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1287                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1288                         gso.csum_start = skb_checksum_start_offset(skb);
1289                         gso.csum_offset = skb->csum_offset;
1290                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1291                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1292                 } /* else everything is zero */
1293
1294                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1295                                                sizeof(gso))))
1296                         return -EFAULT;
1297                 total += tun->vnet_hdr_sz;
1298         }
1299
1300         copied = total;
1301         total += skb->len;
1302         if (!vlan_tx_tag_present(skb)) {
1303                 len = min_t(int, skb->len, len);
1304         } else {
1305                 int copy, ret;
1306                 struct {
1307                         __be16 h_vlan_proto;
1308                         __be16 h_vlan_TCI;
1309                 } veth;
1310
1311                 veth.h_vlan_proto = skb->vlan_proto;
1312                 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1313
1314                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1315                 len = min_t(int, skb->len + VLAN_HLEN, len);
1316                 total += VLAN_HLEN;
1317
1318                 copy = min_t(int, vlan_offset, len);
1319                 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
1320                 len -= copy;
1321                 copied += copy;
1322                 if (ret || !len)
1323                         goto done;
1324
1325                 copy = min_t(int, sizeof(veth), len);
1326                 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
1327                 len -= copy;
1328                 copied += copy;
1329                 if (ret || !len)
1330                         goto done;
1331         }
1332
1333         skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
1334
1335 done:
1336         tun->dev->stats.tx_packets++;
1337         tun->dev->stats.tx_bytes += len;
1338
1339         return total;
1340 }
1341
1342 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1343                            const struct iovec *iv, ssize_t len, int noblock)
1344 {
1345         DECLARE_WAITQUEUE(wait, current);
1346         struct sk_buff *skb;
1347         ssize_t ret = 0;
1348
1349         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1350
1351         if (unlikely(!noblock))
1352                 add_wait_queue(&tfile->wq.wait, &wait);
1353         while (len) {
1354                 if (unlikely(!noblock))
1355                         current->state = TASK_INTERRUPTIBLE;
1356
1357                 /* Read frames from the queue */
1358                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1359                         if (noblock) {
1360                                 ret = -EAGAIN;
1361                                 break;
1362                         }
1363                         if (signal_pending(current)) {
1364                                 ret = -ERESTARTSYS;
1365                                 break;
1366                         }
1367                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1368                                 ret = -EIO;
1369                                 break;
1370                         }
1371
1372                         /* Nothing to read, let's sleep */
1373                         schedule();
1374                         continue;
1375                 }
1376
1377                 ret = tun_put_user(tun, tfile, skb, iv, len);
1378                 kfree_skb(skb);
1379                 break;
1380         }
1381
1382         if (unlikely(!noblock)) {
1383                 current->state = TASK_RUNNING;
1384                 remove_wait_queue(&tfile->wq.wait, &wait);
1385         }
1386
1387         return ret;
1388 }
1389
1390 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1391                             unsigned long count, loff_t pos)
1392 {
1393         struct file *file = iocb->ki_filp;
1394         struct tun_file *tfile = file->private_data;
1395         struct tun_struct *tun = __tun_get(tfile);
1396         ssize_t len, ret;
1397
1398         if (!tun)
1399                 return -EBADFD;
1400         len = iov_length(iv, count);
1401         if (len < 0) {
1402                 ret = -EINVAL;
1403                 goto out;
1404         }
1405
1406         ret = tun_do_read(tun, tfile, iv, len,
1407                           file->f_flags & O_NONBLOCK);
1408         ret = min_t(ssize_t, ret, len);
1409         if (ret > 0)
1410                 iocb->ki_pos = ret;
1411 out:
1412         tun_put(tun);
1413         return ret;
1414 }
1415
1416 static void tun_free_netdev(struct net_device *dev)
1417 {
1418         struct tun_struct *tun = netdev_priv(dev);
1419
1420         BUG_ON(!(list_empty(&tun->disabled)));
1421         tun_flow_uninit(tun);
1422         security_tun_dev_free_security(tun->security);
1423         free_netdev(dev);
1424 }
1425
1426 static void tun_setup(struct net_device *dev)
1427 {
1428         struct tun_struct *tun = netdev_priv(dev);
1429
1430         tun->owner = INVALID_UID;
1431         tun->group = INVALID_GID;
1432
1433         dev->ethtool_ops = &tun_ethtool_ops;
1434         dev->destructor = tun_free_netdev;
1435 }
1436
1437 /* Trivial set of netlink ops to allow deleting tun or tap
1438  * device with netlink.
1439  */
1440 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1441 {
1442         return -EINVAL;
1443 }
1444
1445 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1446         .kind           = DRV_NAME,
1447         .priv_size      = sizeof(struct tun_struct),
1448         .setup          = tun_setup,
1449         .validate       = tun_validate,
1450 };
1451
1452 static void tun_sock_write_space(struct sock *sk)
1453 {
1454         struct tun_file *tfile;
1455         wait_queue_head_t *wqueue;
1456
1457         if (!sock_writeable(sk))
1458                 return;
1459
1460         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1461                 return;
1462
1463         wqueue = sk_sleep(sk);
1464         if (wqueue && waitqueue_active(wqueue))
1465                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1466                                                 POLLWRNORM | POLLWRBAND);
1467
1468         tfile = container_of(sk, struct tun_file, sk);
1469         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1470 }
1471
1472 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1473                        struct msghdr *m, size_t total_len)
1474 {
1475         int ret;
1476         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1477         struct tun_struct *tun = __tun_get(tfile);
1478
1479         if (!tun)
1480                 return -EBADFD;
1481         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1482                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1483         tun_put(tun);
1484         return ret;
1485 }
1486
1487 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1488                        struct msghdr *m, size_t total_len,
1489                        int flags)
1490 {
1491         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1492         struct tun_struct *tun = __tun_get(tfile);
1493         int ret;
1494
1495         if (!tun)
1496                 return -EBADFD;
1497
1498         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1499                 ret = -EINVAL;
1500                 goto out;
1501         }
1502         if (flags & MSG_ERRQUEUE) {
1503                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1504                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1505                 goto out;
1506         }
1507         ret = tun_do_read(tun, tfile, m->msg_iov, total_len,
1508                           flags & MSG_DONTWAIT);
1509         if (ret > total_len) {
1510                 m->msg_flags |= MSG_TRUNC;
1511                 ret = flags & MSG_TRUNC ? ret : total_len;
1512         }
1513 out:
1514         tun_put(tun);
1515         return ret;
1516 }
1517
1518 static int tun_release(struct socket *sock)
1519 {
1520         if (sock->sk)
1521                 sock_put(sock->sk);
1522         return 0;
1523 }
1524
1525 /* Ops structure to mimic raw sockets with tun */
1526 static const struct proto_ops tun_socket_ops = {
1527         .sendmsg = tun_sendmsg,
1528         .recvmsg = tun_recvmsg,
1529         .release = tun_release,
1530 };
1531
1532 static struct proto tun_proto = {
1533         .name           = "tun",
1534         .owner          = THIS_MODULE,
1535         .obj_size       = sizeof(struct tun_file),
1536 };
1537
1538 static int tun_flags(struct tun_struct *tun)
1539 {
1540         int flags = 0;
1541
1542         if (tun->flags & TUN_TUN_DEV)
1543                 flags |= IFF_TUN;
1544         else
1545                 flags |= IFF_TAP;
1546
1547         if (tun->flags & TUN_NO_PI)
1548                 flags |= IFF_NO_PI;
1549
1550         /* This flag has no real effect.  We track the value for backwards
1551          * compatibility.
1552          */
1553         if (tun->flags & TUN_ONE_QUEUE)
1554                 flags |= IFF_ONE_QUEUE;
1555
1556         if (tun->flags & TUN_VNET_HDR)
1557                 flags |= IFF_VNET_HDR;
1558
1559         if (tun->flags & TUN_TAP_MQ)
1560                 flags |= IFF_MULTI_QUEUE;
1561
1562         if (tun->flags & TUN_PERSIST)
1563                 flags |= IFF_PERSIST;
1564
1565         return flags;
1566 }
1567
1568 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1569                               char *buf)
1570 {
1571         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1572         return sprintf(buf, "0x%x\n", tun_flags(tun));
1573 }
1574
1575 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1576                               char *buf)
1577 {
1578         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1579         return uid_valid(tun->owner)?
1580                 sprintf(buf, "%u\n",
1581                         from_kuid_munged(current_user_ns(), tun->owner)):
1582                 sprintf(buf, "-1\n");
1583 }
1584
1585 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1586                               char *buf)
1587 {
1588         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1589         return gid_valid(tun->group) ?
1590                 sprintf(buf, "%u\n",
1591                         from_kgid_munged(current_user_ns(), tun->group)):
1592                 sprintf(buf, "-1\n");
1593 }
1594
1595 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1596 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1597 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1598
1599 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1600 {
1601         struct tun_struct *tun;
1602         struct tun_file *tfile = file->private_data;
1603         struct net_device *dev;
1604         int err;
1605
1606         if (tfile->detached)
1607                 return -EINVAL;
1608
1609         dev = __dev_get_by_name(net, ifr->ifr_name);
1610         if (dev) {
1611                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1612                         return -EBUSY;
1613                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1614                         tun = netdev_priv(dev);
1615                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1616                         tun = netdev_priv(dev);
1617                 else
1618                         return -EINVAL;
1619
1620                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1621                     !!(tun->flags & TUN_TAP_MQ))
1622                         return -EINVAL;
1623
1624                 if (tun_not_capable(tun))
1625                         return -EPERM;
1626                 err = security_tun_dev_open(tun->security);
1627                 if (err < 0)
1628                         return err;
1629
1630                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1631                 if (err < 0)
1632                         return err;
1633
1634                 if (tun->flags & TUN_TAP_MQ &&
1635                     (tun->numqueues + tun->numdisabled > 1)) {
1636                         /* One or more queue has already been attached, no need
1637                          * to initialize the device again.
1638                          */
1639                         return 0;
1640                 }
1641         }
1642         else {
1643                 char *name;
1644                 unsigned long flags = 0;
1645                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1646                              MAX_TAP_QUEUES : 1;
1647
1648                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1649                         return -EPERM;
1650                 err = security_tun_dev_create();
1651                 if (err < 0)
1652                         return err;
1653
1654                 /* Set dev type */
1655                 if (ifr->ifr_flags & IFF_TUN) {
1656                         /* TUN device */
1657                         flags |= TUN_TUN_DEV;
1658                         name = "tun%d";
1659                 } else if (ifr->ifr_flags & IFF_TAP) {
1660                         /* TAP device */
1661                         flags |= TUN_TAP_DEV;
1662                         name = "tap%d";
1663                 } else
1664                         return -EINVAL;
1665
1666                 if (*ifr->ifr_name)
1667                         name = ifr->ifr_name;
1668
1669                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1670                                        tun_setup, queues, queues);
1671
1672                 if (!dev)
1673                         return -ENOMEM;
1674
1675                 dev_net_set(dev, net);
1676                 dev->rtnl_link_ops = &tun_link_ops;
1677                 dev->ifindex = tfile->ifindex;
1678
1679                 tun = netdev_priv(dev);
1680                 tun->dev = dev;
1681                 tun->flags = flags;
1682                 tun->txflt.count = 0;
1683                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1684
1685                 tun->filter_attached = false;
1686                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1687
1688                 spin_lock_init(&tun->lock);
1689
1690                 err = security_tun_dev_alloc_security(&tun->security);
1691                 if (err < 0)
1692                         goto err_free_dev;
1693
1694                 tun_net_init(dev);
1695                 tun_flow_init(tun);
1696
1697                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1698                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1699                                    NETIF_F_HW_VLAN_STAG_TX;
1700                 dev->features = dev->hw_features;
1701                 dev->vlan_features = dev->features &
1702                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1703                                        NETIF_F_HW_VLAN_STAG_TX);
1704
1705                 INIT_LIST_HEAD(&tun->disabled);
1706                 err = tun_attach(tun, file, false);
1707                 if (err < 0)
1708                         goto err_free_flow;
1709
1710                 err = register_netdevice(tun->dev);
1711                 if (err < 0)
1712                         goto err_detach;
1713
1714                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1715                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1716                     device_create_file(&tun->dev->dev, &dev_attr_group))
1717                         pr_err("Failed to create tun sysfs files\n");
1718         }
1719
1720         netif_carrier_on(tun->dev);
1721
1722         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1723
1724         if (ifr->ifr_flags & IFF_NO_PI)
1725                 tun->flags |= TUN_NO_PI;
1726         else
1727                 tun->flags &= ~TUN_NO_PI;
1728
1729         /* This flag has no real effect.  We track the value for backwards
1730          * compatibility.
1731          */
1732         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1733                 tun->flags |= TUN_ONE_QUEUE;
1734         else
1735                 tun->flags &= ~TUN_ONE_QUEUE;
1736
1737         if (ifr->ifr_flags & IFF_VNET_HDR)
1738                 tun->flags |= TUN_VNET_HDR;
1739         else
1740                 tun->flags &= ~TUN_VNET_HDR;
1741
1742         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1743                 tun->flags |= TUN_TAP_MQ;
1744         else
1745                 tun->flags &= ~TUN_TAP_MQ;
1746
1747         /* Make sure persistent devices do not get stuck in
1748          * xoff state.
1749          */
1750         if (netif_running(tun->dev))
1751                 netif_tx_wake_all_queues(tun->dev);
1752
1753         strcpy(ifr->ifr_name, tun->dev->name);
1754         return 0;
1755
1756 err_detach:
1757         tun_detach_all(dev);
1758 err_free_flow:
1759         tun_flow_uninit(tun);
1760         security_tun_dev_free_security(tun->security);
1761 err_free_dev:
1762         free_netdev(dev);
1763         return err;
1764 }
1765
1766 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1767                        struct ifreq *ifr)
1768 {
1769         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1770
1771         strcpy(ifr->ifr_name, tun->dev->name);
1772
1773         ifr->ifr_flags = tun_flags(tun);
1774
1775 }
1776
1777 /* This is like a cut-down ethtool ops, except done via tun fd so no
1778  * privs required. */
1779 static int set_offload(struct tun_struct *tun, unsigned long arg)
1780 {
1781         netdev_features_t features = 0;
1782
1783         if (arg & TUN_F_CSUM) {
1784                 features |= NETIF_F_HW_CSUM;
1785                 arg &= ~TUN_F_CSUM;
1786
1787                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1788                         if (arg & TUN_F_TSO_ECN) {
1789                                 features |= NETIF_F_TSO_ECN;
1790                                 arg &= ~TUN_F_TSO_ECN;
1791                         }
1792                         if (arg & TUN_F_TSO4)
1793                                 features |= NETIF_F_TSO;
1794                         if (arg & TUN_F_TSO6)
1795                                 features |= NETIF_F_TSO6;
1796                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1797                 }
1798         }
1799
1800         /* This gives the user a way to test for new features in future by
1801          * trying to set them. */
1802         if (arg)
1803                 return -EINVAL;
1804
1805         tun->set_features = features;
1806         netdev_update_features(tun->dev);
1807
1808         return 0;
1809 }
1810
1811 static void tun_detach_filter(struct tun_struct *tun, int n)
1812 {
1813         int i;
1814         struct tun_file *tfile;
1815
1816         for (i = 0; i < n; i++) {
1817                 tfile = rtnl_dereference(tun->tfiles[i]);
1818                 sk_detach_filter(tfile->socket.sk);
1819         }
1820
1821         tun->filter_attached = false;
1822 }
1823
1824 static int tun_attach_filter(struct tun_struct *tun)
1825 {
1826         int i, ret = 0;
1827         struct tun_file *tfile;
1828
1829         for (i = 0; i < tun->numqueues; i++) {
1830                 tfile = rtnl_dereference(tun->tfiles[i]);
1831                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1832                 if (ret) {
1833                         tun_detach_filter(tun, i);
1834                         return ret;
1835                 }
1836         }
1837
1838         tun->filter_attached = true;
1839         return ret;
1840 }
1841
1842 static void tun_set_sndbuf(struct tun_struct *tun)
1843 {
1844         struct tun_file *tfile;
1845         int i;
1846
1847         for (i = 0; i < tun->numqueues; i++) {
1848                 tfile = rtnl_dereference(tun->tfiles[i]);
1849                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1850         }
1851 }
1852
1853 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1854 {
1855         struct tun_file *tfile = file->private_data;
1856         struct tun_struct *tun;
1857         int ret = 0;
1858
1859         rtnl_lock();
1860
1861         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1862                 tun = tfile->detached;
1863                 if (!tun) {
1864                         ret = -EINVAL;
1865                         goto unlock;
1866                 }
1867                 ret = security_tun_dev_attach_queue(tun->security);
1868                 if (ret < 0)
1869                         goto unlock;
1870                 ret = tun_attach(tun, file, false);
1871         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1872                 tun = rtnl_dereference(tfile->tun);
1873                 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1874                         ret = -EINVAL;
1875                 else
1876                         __tun_detach(tfile, false);
1877         } else
1878                 ret = -EINVAL;
1879
1880 unlock:
1881         rtnl_unlock();
1882         return ret;
1883 }
1884
1885 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1886                             unsigned long arg, int ifreq_len)
1887 {
1888         struct tun_file *tfile = file->private_data;
1889         struct tun_struct *tun;
1890         void __user* argp = (void __user*)arg;
1891         struct ifreq ifr;
1892         kuid_t owner;
1893         kgid_t group;
1894         int sndbuf;
1895         int vnet_hdr_sz;
1896         unsigned int ifindex;
1897         int ret;
1898
1899         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1900                 if (copy_from_user(&ifr, argp, ifreq_len))
1901                         return -EFAULT;
1902         } else {
1903                 memset(&ifr, 0, sizeof(ifr));
1904         }
1905         if (cmd == TUNGETFEATURES) {
1906                 /* Currently this just means: "what IFF flags are valid?".
1907                  * This is needed because we never checked for invalid flags on
1908                  * TUNSETIFF. */
1909                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1910                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1911                                 (unsigned int __user*)argp);
1912         } else if (cmd == TUNSETQUEUE)
1913                 return tun_set_queue(file, &ifr);
1914
1915         ret = 0;
1916         rtnl_lock();
1917
1918         tun = __tun_get(tfile);
1919         if (cmd == TUNSETIFF && !tun) {
1920                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1921
1922                 ret = tun_set_iff(tfile->net, file, &ifr);
1923
1924                 if (ret)
1925                         goto unlock;
1926
1927                 if (copy_to_user(argp, &ifr, ifreq_len))
1928                         ret = -EFAULT;
1929                 goto unlock;
1930         }
1931         if (cmd == TUNSETIFINDEX) {
1932                 ret = -EPERM;
1933                 if (tun)
1934                         goto unlock;
1935
1936                 ret = -EFAULT;
1937                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1938                         goto unlock;
1939
1940                 ret = 0;
1941                 tfile->ifindex = ifindex;
1942                 goto unlock;
1943         }
1944
1945         ret = -EBADFD;
1946         if (!tun)
1947                 goto unlock;
1948
1949         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1950
1951         ret = 0;
1952         switch (cmd) {
1953         case TUNGETIFF:
1954                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1955
1956                 if (tfile->detached)
1957                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
1958                 if (!tfile->socket.sk->sk_filter)
1959                         ifr.ifr_flags |= IFF_NOFILTER;
1960
1961                 if (copy_to_user(argp, &ifr, ifreq_len))
1962                         ret = -EFAULT;
1963                 break;
1964
1965         case TUNSETNOCSUM:
1966                 /* Disable/Enable checksum */
1967
1968                 /* [unimplemented] */
1969                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1970                           arg ? "disabled" : "enabled");
1971                 break;
1972
1973         case TUNSETPERSIST:
1974                 /* Disable/Enable persist mode. Keep an extra reference to the
1975                  * module to prevent the module being unprobed.
1976                  */
1977                 if (arg && !(tun->flags & TUN_PERSIST)) {
1978                         tun->flags |= TUN_PERSIST;
1979                         __module_get(THIS_MODULE);
1980                 }
1981                 if (!arg && (tun->flags & TUN_PERSIST)) {
1982                         tun->flags &= ~TUN_PERSIST;
1983                         module_put(THIS_MODULE);
1984                 }
1985
1986                 tun_debug(KERN_INFO, tun, "persist %s\n",
1987                           arg ? "enabled" : "disabled");
1988                 break;
1989
1990         case TUNSETOWNER:
1991                 /* Set owner of the device */
1992                 owner = make_kuid(current_user_ns(), arg);
1993                 if (!uid_valid(owner)) {
1994                         ret = -EINVAL;
1995                         break;
1996                 }
1997                 tun->owner = owner;
1998                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1999                           from_kuid(&init_user_ns, tun->owner));
2000                 break;
2001
2002         case TUNSETGROUP:
2003                 /* Set group of the device */
2004                 group = make_kgid(current_user_ns(), arg);
2005                 if (!gid_valid(group)) {
2006                         ret = -EINVAL;
2007                         break;
2008                 }
2009                 tun->group = group;
2010                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2011                           from_kgid(&init_user_ns, tun->group));
2012                 break;
2013
2014         case TUNSETLINK:
2015                 /* Only allow setting the type when the interface is down */
2016                 if (tun->dev->flags & IFF_UP) {
2017                         tun_debug(KERN_INFO, tun,
2018                                   "Linktype set failed because interface is up\n");
2019                         ret = -EBUSY;
2020                 } else {
2021                         tun->dev->type = (int) arg;
2022                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2023                                   tun->dev->type);
2024                         ret = 0;
2025                 }
2026                 break;
2027
2028 #ifdef TUN_DEBUG
2029         case TUNSETDEBUG:
2030                 tun->debug = arg;
2031                 break;
2032 #endif
2033         case TUNSETOFFLOAD:
2034                 ret = set_offload(tun, arg);
2035                 break;
2036
2037         case TUNSETTXFILTER:
2038                 /* Can be set only for TAPs */
2039                 ret = -EINVAL;
2040                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2041                         break;
2042                 ret = update_filter(&tun->txflt, (void __user *)arg);
2043                 break;
2044
2045         case SIOCGIFHWADDR:
2046                 /* Get hw address */
2047                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2048                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2049                 if (copy_to_user(argp, &ifr, ifreq_len))
2050                         ret = -EFAULT;
2051                 break;
2052
2053         case SIOCSIFHWADDR:
2054                 /* Set hw address */
2055                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2056                           ifr.ifr_hwaddr.sa_data);
2057
2058                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2059                 break;
2060
2061         case TUNGETSNDBUF:
2062                 sndbuf = tfile->socket.sk->sk_sndbuf;
2063                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2064                         ret = -EFAULT;
2065                 break;
2066
2067         case TUNSETSNDBUF:
2068                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2069                         ret = -EFAULT;
2070                         break;
2071                 }
2072
2073                 tun->sndbuf = sndbuf;
2074                 tun_set_sndbuf(tun);
2075                 break;
2076
2077         case TUNGETVNETHDRSZ:
2078                 vnet_hdr_sz = tun->vnet_hdr_sz;
2079                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2080                         ret = -EFAULT;
2081                 break;
2082
2083         case TUNSETVNETHDRSZ:
2084                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2085                         ret = -EFAULT;
2086                         break;
2087                 }
2088                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2089                         ret = -EINVAL;
2090                         break;
2091                 }
2092
2093                 tun->vnet_hdr_sz = vnet_hdr_sz;
2094                 break;
2095
2096         case TUNATTACHFILTER:
2097                 /* Can be set only for TAPs */
2098                 ret = -EINVAL;
2099                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2100                         break;
2101                 ret = -EFAULT;
2102                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2103                         break;
2104
2105                 ret = tun_attach_filter(tun);
2106                 break;
2107
2108         case TUNDETACHFILTER:
2109                 /* Can be set only for TAPs */
2110                 ret = -EINVAL;
2111                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2112                         break;
2113                 ret = 0;
2114                 tun_detach_filter(tun, tun->numqueues);
2115                 break;
2116
2117         case TUNGETFILTER:
2118                 ret = -EINVAL;
2119                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2120                         break;
2121                 ret = -EFAULT;
2122                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2123                         break;
2124                 ret = 0;
2125                 break;
2126
2127         default:
2128                 ret = -EINVAL;
2129                 break;
2130         }
2131
2132 unlock:
2133         rtnl_unlock();
2134         if (tun)
2135                 tun_put(tun);
2136         return ret;
2137 }
2138
2139 static long tun_chr_ioctl(struct file *file,
2140                           unsigned int cmd, unsigned long arg)
2141 {
2142         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2143 }
2144
2145 #ifdef CONFIG_COMPAT
2146 static long tun_chr_compat_ioctl(struct file *file,
2147                          unsigned int cmd, unsigned long arg)
2148 {
2149         switch (cmd) {
2150         case TUNSETIFF:
2151         case TUNGETIFF:
2152         case TUNSETTXFILTER:
2153         case TUNGETSNDBUF:
2154         case TUNSETSNDBUF:
2155         case SIOCGIFHWADDR:
2156         case SIOCSIFHWADDR:
2157                 arg = (unsigned long)compat_ptr(arg);
2158                 break;
2159         default:
2160                 arg = (compat_ulong_t)arg;
2161                 break;
2162         }
2163
2164         /*
2165          * compat_ifreq is shorter than ifreq, so we must not access beyond
2166          * the end of that structure. All fields that are used in this
2167          * driver are compatible though, we don't need to convert the
2168          * contents.
2169          */
2170         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2171 }
2172 #endif /* CONFIG_COMPAT */
2173
2174 static int tun_chr_fasync(int fd, struct file *file, int on)
2175 {
2176         struct tun_file *tfile = file->private_data;
2177         int ret;
2178
2179         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2180                 goto out;
2181
2182         if (on) {
2183                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2184                 if (ret)
2185                         goto out;
2186                 tfile->flags |= TUN_FASYNC;
2187         } else
2188                 tfile->flags &= ~TUN_FASYNC;
2189         ret = 0;
2190 out:
2191         return ret;
2192 }
2193
2194 static int tun_chr_open(struct inode *inode, struct file * file)
2195 {
2196         struct tun_file *tfile;
2197
2198         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2199
2200         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2201                                             &tun_proto);
2202         if (!tfile)
2203                 return -ENOMEM;
2204         rcu_assign_pointer(tfile->tun, NULL);
2205         tfile->net = get_net(current->nsproxy->net_ns);
2206         tfile->flags = 0;
2207         tfile->ifindex = 0;
2208
2209         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2210         init_waitqueue_head(&tfile->wq.wait);
2211
2212         tfile->socket.file = file;
2213         tfile->socket.ops = &tun_socket_ops;
2214
2215         sock_init_data(&tfile->socket, &tfile->sk);
2216         sk_change_net(&tfile->sk, tfile->net);
2217
2218         tfile->sk.sk_write_space = tun_sock_write_space;
2219         tfile->sk.sk_sndbuf = INT_MAX;
2220
2221         file->private_data = tfile;
2222         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2223         INIT_LIST_HEAD(&tfile->next);
2224
2225         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2226
2227         return 0;
2228 }
2229
2230 static int tun_chr_close(struct inode *inode, struct file *file)
2231 {
2232         struct tun_file *tfile = file->private_data;
2233         struct net *net = tfile->net;
2234
2235         tun_detach(tfile, true);
2236         put_net(net);
2237
2238         return 0;
2239 }
2240
2241 #ifdef CONFIG_PROC_FS
2242 static int tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2243 {
2244         struct tun_struct *tun;
2245         struct ifreq ifr;
2246
2247         memset(&ifr, 0, sizeof(ifr));
2248
2249         rtnl_lock();
2250         tun = tun_get(f);
2251         if (tun)
2252                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2253         rtnl_unlock();
2254
2255         if (tun)
2256                 tun_put(tun);
2257
2258         return seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2259 }
2260 #endif
2261
2262 static const struct file_operations tun_fops = {
2263         .owner  = THIS_MODULE,
2264         .llseek = no_llseek,
2265         .read  = do_sync_read,
2266         .aio_read  = tun_chr_aio_read,
2267         .write = do_sync_write,
2268         .aio_write = tun_chr_aio_write,
2269         .poll   = tun_chr_poll,
2270         .unlocked_ioctl = tun_chr_ioctl,
2271 #ifdef CONFIG_COMPAT
2272         .compat_ioctl = tun_chr_compat_ioctl,
2273 #endif
2274         .open   = tun_chr_open,
2275         .release = tun_chr_close,
2276         .fasync = tun_chr_fasync,
2277 #ifdef CONFIG_PROC_FS
2278         .show_fdinfo = tun_chr_show_fdinfo,
2279 #endif
2280 };
2281
2282 static struct miscdevice tun_miscdev = {
2283         .minor = TUN_MINOR,
2284         .name = "tun",
2285         .nodename = "net/tun",
2286         .fops = &tun_fops,
2287 };
2288
2289 /* ethtool interface */
2290
2291 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2292 {
2293         cmd->supported          = 0;
2294         cmd->advertising        = 0;
2295         ethtool_cmd_speed_set(cmd, SPEED_10);
2296         cmd->duplex             = DUPLEX_FULL;
2297         cmd->port               = PORT_TP;
2298         cmd->phy_address        = 0;
2299         cmd->transceiver        = XCVR_INTERNAL;
2300         cmd->autoneg            = AUTONEG_DISABLE;
2301         cmd->maxtxpkt           = 0;
2302         cmd->maxrxpkt           = 0;
2303         return 0;
2304 }
2305
2306 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2307 {
2308         struct tun_struct *tun = netdev_priv(dev);
2309
2310         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2311         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2312
2313         switch (tun->flags & TUN_TYPE_MASK) {
2314         case TUN_TUN_DEV:
2315                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2316                 break;
2317         case TUN_TAP_DEV:
2318                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2319                 break;
2320         }
2321 }
2322
2323 static u32 tun_get_msglevel(struct net_device *dev)
2324 {
2325 #ifdef TUN_DEBUG
2326         struct tun_struct *tun = netdev_priv(dev);
2327         return tun->debug;
2328 #else
2329         return -EOPNOTSUPP;
2330 #endif
2331 }
2332
2333 static void tun_set_msglevel(struct net_device *dev, u32 value)
2334 {
2335 #ifdef TUN_DEBUG
2336         struct tun_struct *tun = netdev_priv(dev);
2337         tun->debug = value;
2338 #endif
2339 }
2340
2341 static const struct ethtool_ops tun_ethtool_ops = {
2342         .get_settings   = tun_get_settings,
2343         .get_drvinfo    = tun_get_drvinfo,
2344         .get_msglevel   = tun_get_msglevel,
2345         .set_msglevel   = tun_set_msglevel,
2346         .get_link       = ethtool_op_get_link,
2347         .get_ts_info    = ethtool_op_get_ts_info,
2348 };
2349
2350
2351 static int __init tun_init(void)
2352 {
2353         int ret = 0;
2354
2355         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2356         pr_info("%s\n", DRV_COPYRIGHT);
2357
2358         ret = rtnl_link_register(&tun_link_ops);
2359         if (ret) {
2360                 pr_err("Can't register link_ops\n");
2361                 goto err_linkops;
2362         }
2363
2364         ret = misc_register(&tun_miscdev);
2365         if (ret) {
2366                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2367                 goto err_misc;
2368         }
2369         return  0;
2370 err_misc:
2371         rtnl_link_unregister(&tun_link_ops);
2372 err_linkops:
2373         return ret;
2374 }
2375
2376 static void tun_cleanup(void)
2377 {
2378         misc_deregister(&tun_miscdev);
2379         rtnl_link_unregister(&tun_link_ops);
2380 }
2381
2382 /* Get an underlying socket object from tun file.  Returns error unless file is
2383  * attached to a device.  The returned object works like a packet socket, it
2384  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2385  * holding a reference to the file for as long as the socket is in use. */
2386 struct socket *tun_get_socket(struct file *file)
2387 {
2388         struct tun_file *tfile;
2389         if (file->f_op != &tun_fops)
2390                 return ERR_PTR(-EINVAL);
2391         tfile = file->private_data;
2392         if (!tfile)
2393                 return ERR_PTR(-EBADFD);
2394         return &tfile->socket;
2395 }
2396 EXPORT_SYMBOL_GPL(tun_get_socket);
2397
2398 module_init(tun_init);
2399 module_exit(tun_cleanup);
2400 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2401 MODULE_AUTHOR(DRV_COPYRIGHT);
2402 MODULE_LICENSE("GPL");
2403 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2404 MODULE_ALIAS("devname:net/tun");