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/net_namespace.h>
  69 #include <net/netns/generic.h>
  70 #include <net/rtnetlink.h>
  71 #include <net/sock.h>
  72 #include <linux/seq_file.h>
  73 #include <linux/uio.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 /* TUN device flags */
 107
 108 /* IFF_ATTACH_QUEUE is never stored in device flags,
 109  * overload it to mean fasync when stored there.
 110  */
 111 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 112 /* High bits in flags field are unused. */
 113 #define TUN_VNET_LE     0x80000000
 114 #define TUN_VNET_BE     0x40000000
 115
 116 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 117                       IFF_MULTI_QUEUE)
 118 #define GOODCOPY_LEN 128
 119
 120 #define FLT_EXACT_COUNT 8
 121 struct tap_filter {
 122         unsigned int    count;    /* Number of addrs. Zero means disabled */
 123         u32             mask[2];  /* Mask of the hashed addrs */
 124         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 125 };
 126
 127 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 128  * to max number of VCPUs in guest. */
 129 #define MAX_TAP_QUEUES 256
 130 #define MAX_TAP_FLOWS  4096
 131
 132 #define TUN_FLOW_EXPIRE (3 * HZ)
 133
 134 /* A tun_file connects an open character device to a tuntap netdevice. It
 135  * also contains all socket related structures (except sock_fprog and tap_filter)
 136  * to serve as one transmit queue for tuntap device. The sock_fprog and
 137  * tap_filter were kept in tun_struct since they were used for filtering for the
 138  * netdevice not for a specific queue (at least I didn't see the requirement for
 139  * this).
 140  *
 141  * RCU usage:
 142  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 143  * other can only be read while rcu_read_lock or rtnl_lock is held.
 144  */
 145 struct tun_file {
 146         struct sock sk;
 147         struct socket socket;
 148         struct socket_wq wq;
 149         struct tun_struct __rcu *tun;
 150         struct fasync_struct *fasync;
 151         /* only used for fasnyc */
 152         unsigned int flags;
 153         union {
 154                 u16 queue_index;
 155                 unsigned int ifindex;
 156         };
 157         struct list_head next;
 158         struct tun_struct *detached;
 159 };
 160
 161 struct tun_flow_entry {
 162         struct hlist_node hash_link;
 163         struct rcu_head rcu;
 164         struct tun_struct *tun;
 165
 166         u32 rxhash;
 167         u32 rps_rxhash;
 168         int queue_index;
 169         unsigned long updated;
 170 };
 171
 172 #define TUN_NUM_FLOW_ENTRIES 1024
 173
 174 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 175  * device, socket filter, sndbuf and vnet header size were restore when the
 176  * file were attached to a persist device.
 177  */
 178 struct tun_struct {
 179         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 180         unsigned int            numqueues;
 181         unsigned int            flags;
 182         kuid_t                  owner;
 183         kgid_t                  group;
 184
 185         struct net_device       *dev;
 186         netdev_features_t       set_features;
 187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 188                           NETIF_F_TSO6|NETIF_F_UFO)
 189
 190         int                     align;
 191         int                     vnet_hdr_sz;
 192         int                     sndbuf;
 193         struct tap_filter       txflt;
 194         struct sock_fprog       fprog;
 195         /* protected by rtnl lock */
 196         bool                    filter_attached;
 197 #ifdef TUN_DEBUG
 198         int debug;
 199 #endif
 200         spinlock_t lock;
 201         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 202         struct timer_list flow_gc_timer;
 203         unsigned long ageing_time;
 204         unsigned int numdisabled;
 205         struct list_head disabled;
 206         void *security;
 207         u32 flow_count;
 208 };
 209
 210 #ifdef CONFIG_TUN_VNET_CROSS_LE
 211 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 212 {
 213         return tun->flags & TUN_VNET_BE ? false :
 214                 virtio_legacy_is_little_endian();
 215 }
 216
 217 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 218 {
 219         int be = !!(tun->flags & TUN_VNET_BE);
 220
 221         if (put_user(be, argp))
 222                 return -EFAULT;
 223
 224         return 0;
 225 }
 226
 227 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 228 {
 229         int be;
 230
 231         if (get_user(be, argp))
 232                 return -EFAULT;
 233
 234         if (be)
 235                 tun->flags |= TUN_VNET_BE;
 236         else
 237                 tun->flags &= ~TUN_VNET_BE;
 238
 239         return 0;
 240 }
 241 #else
 242 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 243 {
 244         return virtio_legacy_is_little_endian();
 245 }
 246
 247 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 248 {
 249         return -EINVAL;
 250 }
 251
 252 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 253 {
 254         return -EINVAL;
 255 }
 256 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 257
 258 static inline bool tun_is_little_endian(struct tun_struct *tun)
 259 {
 260         return tun->flags & TUN_VNET_LE ||
 261                 tun_legacy_is_little_endian(tun);
 262 }
 263
 264 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 265 {
 266         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 267 }
 268
 269 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 270 {
 271         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 272 }
 273
 274 static inline u32 tun_hashfn(u32 rxhash)
 275 {
 276         return rxhash & 0x3ff;
 277 }
 278
 279 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 280 {
 281         struct tun_flow_entry *e;
 282
 283         hlist_for_each_entry_rcu(e, head, hash_link) {
 284                 if (e->rxhash == rxhash)
 285                         return e;
 286         }
 287         return NULL;
 288 }
 289
 290 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 291                                               struct hlist_head *head,
 292                                               u32 rxhash, u16 queue_index)
 293 {
 294         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 295
 296         if (e) {
 297                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 298                           rxhash, queue_index);
 299                 e->updated = jiffies;
 300                 e->rxhash = rxhash;
 301                 e->rps_rxhash = 0;
 302                 e->queue_index = queue_index;
 303                 e->tun = tun;
 304                 hlist_add_head_rcu(&e->hash_link, head);
 305                 ++tun->flow_count;
 306         }
 307         return e;
 308 }
 309
 310 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 311 {
 312         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 313                   e->rxhash, e->queue_index);
 314         hlist_del_rcu(&e->hash_link);
 315         kfree_rcu(e, rcu);
 316         --tun->flow_count;
 317 }
 318
 319 static void tun_flow_flush(struct tun_struct *tun)
 320 {
 321         int i;
 322
 323         spin_lock_bh(&tun->lock);
 324         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 325                 struct tun_flow_entry *e;
 326                 struct hlist_node *n;
 327
 328                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 329                         tun_flow_delete(tun, e);
 330         }
 331         spin_unlock_bh(&tun->lock);
 332 }
 333
 334 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 335 {
 336         int i;
 337
 338         spin_lock_bh(&tun->lock);
 339         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 340                 struct tun_flow_entry *e;
 341                 struct hlist_node *n;
 342
 343                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 344                         if (e->queue_index == queue_index)
 345                                 tun_flow_delete(tun, e);
 346                 }
 347         }
 348         spin_unlock_bh(&tun->lock);
 349 }
 350
 351 static void tun_flow_cleanup(unsigned long data)
 352 {
 353         struct tun_struct *tun = (struct tun_struct *)data;
 354         unsigned long delay = tun->ageing_time;
 355         unsigned long next_timer = jiffies + delay;
 356         unsigned long count = 0;
 357         int i;
 358
 359         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 360
 361         spin_lock_bh(&tun->lock);
 362         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 363                 struct tun_flow_entry *e;
 364                 struct hlist_node *n;
 365
 366                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 367                         unsigned long this_timer;
 368                         count++;
 369                         this_timer = e->updated + delay;
 370                         if (time_before_eq(this_timer, jiffies))
 371                                 tun_flow_delete(tun, e);
 372                         else if (time_before(this_timer, next_timer))
 373                                 next_timer = this_timer;
 374                 }
 375         }
 376
 377         if (count)
 378                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 379         spin_unlock_bh(&tun->lock);
 380 }
 381
 382 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 383                             struct tun_file *tfile)
 384 {
 385         struct hlist_head *head;
 386         struct tun_flow_entry *e;
 387         unsigned long delay = tun->ageing_time;
 388         u16 queue_index = tfile->queue_index;
 389
 390         if (!rxhash)
 391                 return;
 392         else
 393                 head = &tun->flows[tun_hashfn(rxhash)];
 394
 395         rcu_read_lock();
 396
 397         /* We may get a very small possibility of OOO during switching, not
 398          * worth to optimize.*/
 399         if (tun->numqueues == 1 || tfile->detached)
 400                 goto unlock;
 401
 402         e = tun_flow_find(head, rxhash);
 403         if (likely(e)) {
 404                 /* TODO: keep queueing to old queue until it's empty? */
 405                 e->queue_index = queue_index;
 406                 e->updated = jiffies;
 407                 sock_rps_record_flow_hash(e->rps_rxhash);
 408         } else {
 409                 spin_lock_bh(&tun->lock);
 410                 if (!tun_flow_find(head, rxhash) &&
 411                     tun->flow_count < MAX_TAP_FLOWS)
 412                         tun_flow_create(tun, head, rxhash, queue_index);
 413
 414                 if (!timer_pending(&tun->flow_gc_timer))
 415                         mod_timer(&tun->flow_gc_timer,
 416                                   round_jiffies_up(jiffies + delay));
 417                 spin_unlock_bh(&tun->lock);
 418         }
 419
 420 unlock:
 421         rcu_read_unlock();
 422 }
 423
 424 /**
 425  * Save the hash received in the stack receive path and update the
 426  * flow_hash table accordingly.
 427  */
 428 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 429 {
 430         if (unlikely(e->rps_rxhash != hash))
 431                 e->rps_rxhash = hash;
 432 }
 433
 434 /* We try to identify a flow through its rxhash first. The reason that
 435  * we do not check rxq no. is because some cards(e.g 82599), chooses
 436  * the rxq based on the txq where the last packet of the flow comes. As
 437  * the userspace application move between processors, we may get a
 438  * different rxq no. here. If we could not get rxhash, then we would
 439  * hope the rxq no. may help here.
 440  */
 441 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 442                             void *accel_priv, select_queue_fallback_t fallback)
 443 {
 444         struct tun_struct *tun = netdev_priv(dev);
 445         struct tun_flow_entry *e;
 446         u32 txq = 0;
 447         u32 numqueues = 0;
 448
 449         rcu_read_lock();
 450         numqueues = ACCESS_ONCE(tun->numqueues);
 451
 452         txq = skb_get_hash(skb);
 453         if (txq) {
 454                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 455                 if (e) {
 456                         tun_flow_save_rps_rxhash(e, txq);
 457                         txq = e->queue_index;
 458                 } else
 459                         /* use multiply and shift instead of expensive divide */
 460                         txq = ((u64)txq * numqueues) >> 32;
 461         } else if (likely(skb_rx_queue_recorded(skb))) {
 462                 txq = skb_get_rx_queue(skb);
 463                 while (unlikely(txq >= numqueues))
 464                         txq -= numqueues;
 465         }
 466
 467         rcu_read_unlock();
 468         return txq;
 469 }
 470
 471 static inline bool tun_not_capable(struct tun_struct *tun)
 472 {
 473         const struct cred *cred = current_cred();
 474         struct net *net = dev_net(tun->dev);
 475
 476         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 477                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 478                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 479 }
 480
 481 static void tun_set_real_num_queues(struct tun_struct *tun)
 482 {
 483         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 484         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 485 }
 486
 487 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 488 {
 489         tfile->detached = tun;
 490         list_add_tail(&tfile->next, &tun->disabled);
 491         ++tun->numdisabled;
 492 }
 493
 494 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 495 {
 496         struct tun_struct *tun = tfile->detached;
 497
 498         tfile->detached = NULL;
 499         list_del_init(&tfile->next);
 500         --tun->numdisabled;
 501         return tun;
 502 }
 503
 504 static void tun_queue_purge(struct tun_file *tfile)
 505 {
 506         skb_queue_purge(&tfile->sk.sk_receive_queue);
 507         skb_queue_purge(&tfile->sk.sk_error_queue);
 508 }
 509
 510 static void __tun_detach(struct tun_file *tfile, bool clean)
 511 {
 512         struct tun_file *ntfile;
 513         struct tun_struct *tun;
 514
 515         tun = rtnl_dereference(tfile->tun);
 516
 517         if (tun && !tfile->detached) {
 518                 u16 index = tfile->queue_index;
 519                 BUG_ON(index >= tun->numqueues);
 520
 521                 rcu_assign_pointer(tun->tfiles[index],
 522                                    tun->tfiles[tun->numqueues - 1]);
 523                 ntfile = rtnl_dereference(tun->tfiles[index]);
 524                 ntfile->queue_index = index;
 525
 526                 --tun->numqueues;
 527                 if (clean) {
 528                         RCU_INIT_POINTER(tfile->tun, NULL);
 529                         sock_put(&tfile->sk);
 530                 } else
 531                         tun_disable_queue(tun, tfile);
 532
 533                 synchronize_net();
 534                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 535                 /* Drop read queue */
 536                 tun_queue_purge(tfile);
 537                 tun_set_real_num_queues(tun);
 538         } else if (tfile->detached && clean) {
 539                 tun = tun_enable_queue(tfile);
 540                 sock_put(&tfile->sk);
 541         }
 542
 543         if (clean) {
 544                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 545                         netif_carrier_off(tun->dev);
 546
 547                         if (!(tun->flags & IFF_PERSIST) &&
 548                             tun->dev->reg_state == NETREG_REGISTERED)
 549                                 unregister_netdevice(tun->dev);
 550                 }
 551                 sock_put(&tfile->sk);
 552         }
 553 }
 554
 555 static void tun_detach(struct tun_file *tfile, bool clean)
 556 {
 557         rtnl_lock();
 558         __tun_detach(tfile, clean);
 559         rtnl_unlock();
 560 }
 561
 562 static void tun_detach_all(struct net_device *dev)
 563 {
 564         struct tun_struct *tun = netdev_priv(dev);
 565         struct tun_file *tfile, *tmp;
 566         int i, n = tun->numqueues;
 567
 568         for (i = 0; i < n; i++) {
 569                 tfile = rtnl_dereference(tun->tfiles[i]);
 570                 BUG_ON(!tfile);
 571                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 572                 RCU_INIT_POINTER(tfile->tun, NULL);
 573                 --tun->numqueues;
 574         }
 575         list_for_each_entry(tfile, &tun->disabled, next) {
 576                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 577                 RCU_INIT_POINTER(tfile->tun, NULL);
 578         }
 579         BUG_ON(tun->numqueues != 0);
 580
 581         synchronize_net();
 582         for (i = 0; i < n; i++) {
 583                 tfile = rtnl_dereference(tun->tfiles[i]);
 584                 /* Drop read queue */
 585                 tun_queue_purge(tfile);
 586                 sock_put(&tfile->sk);
 587         }
 588         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 589                 tun_enable_queue(tfile);
 590                 tun_queue_purge(tfile);
 591                 sock_put(&tfile->sk);
 592         }
 593         BUG_ON(tun->numdisabled != 0);
 594
 595         if (tun->flags & IFF_PERSIST)
 596                 module_put(THIS_MODULE);
 597 }
 598
 599 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 600 {
 601         struct tun_file *tfile = file->private_data;
 602         int err;
 603
 604         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 605         if (err < 0)
 606                 goto out;
 607
 608         err = -EINVAL;
 609         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 610                 goto out;
 611
 612         err = -EBUSY;
 613         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 614                 goto out;
 615
 616         err = -E2BIG;
 617         if (!tfile->detached &&
 618             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 619                 goto out;
 620
 621         err = 0;
 622
 623         /* Re-attach the filter to persist device */
 624         if (!skip_filter && (tun->filter_attached == true)) {
 625                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 626                 if (!err)
 627                         goto out;
 628         }
 629         tfile->queue_index = tun->numqueues;
 630         rcu_assign_pointer(tfile->tun, tun);
 631         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 632         tun->numqueues++;
 633
 634         if (tfile->detached)
 635                 tun_enable_queue(tfile);
 636         else
 637                 sock_hold(&tfile->sk);
 638
 639         tun_set_real_num_queues(tun);
 640
 641         /* device is allowed to go away first, so no need to hold extra
 642          * refcnt.
 643          */
 644
 645 out:
 646         return err;
 647 }
 648
 649 static struct tun_struct *__tun_get(struct tun_file *tfile)
 650 {
 651         struct tun_struct *tun;
 652
 653         rcu_read_lock();
 654         tun = rcu_dereference(tfile->tun);
 655         if (tun)
 656                 dev_hold(tun->dev);
 657         rcu_read_unlock();
 658
 659         return tun;
 660 }
 661
 662 static struct tun_struct *tun_get(struct file *file)
 663 {
 664         return __tun_get(file->private_data);
 665 }
 666
 667 static void tun_put(struct tun_struct *tun)
 668 {
 669         dev_put(tun->dev);
 670 }
 671
 672 /* TAP filtering */
 673 static void addr_hash_set(u32 *mask, const u8 *addr)
 674 {
 675         int n = ether_crc(ETH_ALEN, addr) >> 26;
 676         mask[n >> 5] |= (1 << (n & 31));
 677 }
 678
 679 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 680 {
 681         int n = ether_crc(ETH_ALEN, addr) >> 26;
 682         return mask[n >> 5] & (1 << (n & 31));
 683 }
 684
 685 static int update_filter(struct tap_filter *filter, void __user *arg)
 686 {
 687         struct { u8 u[ETH_ALEN]; } *addr;
 688         struct tun_filter uf;
 689         int err, alen, n, nexact;
 690
 691         if (copy_from_user(&uf, arg, sizeof(uf)))
 692                 return -EFAULT;
 693
 694         if (!uf.count) {
 695                 /* Disabled */
 696                 filter->count = 0;
 697                 return 0;
 698         }
 699
 700         alen = ETH_ALEN * uf.count;
 701         addr = kmalloc(alen, GFP_KERNEL);
 702         if (!addr)
 703                 return -ENOMEM;
 704
 705         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 706                 err = -EFAULT;
 707                 goto done;
 708         }
 709
 710         /* The filter is updated without holding any locks. Which is
 711          * perfectly safe. We disable it first and in the worst
 712          * case we'll accept a few undesired packets. */
 713         filter->count = 0;
 714         wmb();
 715
 716         /* Use first set of addresses as an exact filter */
 717         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 718                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 719
 720         nexact = n;
 721
 722         /* Remaining multicast addresses are hashed,
 723          * unicast will leave the filter disabled. */
 724         memset(filter->mask, 0, sizeof(filter->mask));
 725         for (; n < uf.count; n++) {
 726                 if (!is_multicast_ether_addr(addr[n].u)) {
 727                         err = 0; /* no filter */
 728                         goto done;
 729                 }
 730                 addr_hash_set(filter->mask, addr[n].u);
 731         }
 732
 733         /* For ALLMULTI just set the mask to all ones.
 734          * This overrides the mask populated above. */
 735         if ((uf.flags & TUN_FLT_ALLMULTI))
 736                 memset(filter->mask, ~0, sizeof(filter->mask));
 737
 738         /* Now enable the filter */
 739         wmb();
 740         filter->count = nexact;
 741
 742         /* Return the number of exact filters */
 743         err = nexact;
 744
 745 done:
 746         kfree(addr);
 747         return err;
 748 }
 749
 750 /* Returns: 0 - drop, !=0 - accept */
 751 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 752 {
 753         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 754          * at this point. */
 755         struct ethhdr *eh = (struct ethhdr *) skb->data;
 756         int i;
 757
 758         /* Exact match */
 759         for (i = 0; i < filter->count; i++)
 760                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 761                         return 1;
 762
 763         /* Inexact match (multicast only) */
 764         if (is_multicast_ether_addr(eh->h_dest))
 765                 return addr_hash_test(filter->mask, eh->h_dest);
 766
 767         return 0;
 768 }
 769
 770 /*
 771  * Checks whether the packet is accepted or not.
 772  * Returns: 0 - drop, !=0 - accept
 773  */
 774 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 775 {
 776         if (!filter->count)
 777                 return 1;
 778
 779         return run_filter(filter, skb);
 780 }
 781
 782 /* Network device part of the driver */
 783
 784 static const struct ethtool_ops tun_ethtool_ops;
 785
 786 /* Net device detach from fd. */
 787 static void tun_net_uninit(struct net_device *dev)
 788 {
 789         tun_detach_all(dev);
 790 }
 791
 792 /* Net device open. */
 793 static int tun_net_open(struct net_device *dev)
 794 {
 795         netif_tx_start_all_queues(dev);
 796         return 0;
 797 }
 798
 799 /* Net device close. */
 800 static int tun_net_close(struct net_device *dev)
 801 {
 802         netif_tx_stop_all_queues(dev);
 803         return 0;
 804 }
 805
 806 /* Net device start xmit */
 807 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 808 {
 809         struct tun_struct *tun = netdev_priv(dev);
 810         int txq = skb->queue_mapping;
 811         struct tun_file *tfile;
 812         u32 numqueues = 0;
 813
 814         rcu_read_lock();
 815         tfile = rcu_dereference(tun->tfiles[txq]);
 816         numqueues = ACCESS_ONCE(tun->numqueues);
 817
 818         /* Drop packet if interface is not attached */
 819         if (txq >= numqueues)
 820                 goto drop;
 821
 822         if (numqueues == 1) {
 823                 /* Select queue was not called for the skbuff, so we extract the
 824                  * RPS hash and save it into the flow_table here.
 825                  */
 826                 __u32 rxhash;
 827
 828                 rxhash = skb_get_hash(skb);
 829                 if (rxhash) {
 830                         struct tun_flow_entry *e;
 831                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 832                                         rxhash);
 833                         if (e)
 834                                 tun_flow_save_rps_rxhash(e, rxhash);
 835                 }
 836         }
 837
 838         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 839
 840         BUG_ON(!tfile);
 841
 842         /* Drop if the filter does not like it.
 843          * This is a noop if the filter is disabled.
 844          * Filter can be enabled only for the TAP devices. */
 845         if (!check_filter(&tun->txflt, skb))
 846                 goto drop;
 847
 848         if (tfile->socket.sk->sk_filter &&
 849             sk_filter(tfile->socket.sk, skb))
 850                 goto drop;
 851
 852         /* Limit the number of packets queued by dividing txq length with the
 853          * number of queues.
 854          */
 855         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
 856                           >= dev->tx_queue_len)
 857                 goto drop;
 858
 859         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 860                 goto drop;
 861
 862         if (skb->sk && sk_fullsock(skb->sk)) {
 863                 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
 864                 sw_tx_timestamp(skb);
 865         }
 866
 867         /* Orphan the skb - required as we might hang on to it
 868          * for indefinite time.
 869          */
 870         skb_orphan(skb);
 871
 872         nf_reset(skb);
 873
 874         /* Enqueue packet */
 875         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 876
 877         /* Notify and wake up reader process */
 878         if (tfile->flags & TUN_FASYNC)
 879                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 880         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 881
 882         rcu_read_unlock();
 883         return NETDEV_TX_OK;
 884
 885 drop:
 886         dev->stats.tx_dropped++;
 887         skb_tx_error(skb);
 888         kfree_skb(skb);
 889         rcu_read_unlock();
 890         return NET_XMIT_DROP;
 891 }
 892
 893 static void tun_net_mclist(struct net_device *dev)
 894 {
 895         /*
 896          * This callback is supposed to deal with mc filter in
 897          * _rx_ path and has nothing to do with the _tx_ path.
 898          * In rx path we always accept everything userspace gives us.
 899          */
 900 }
 901
 902 #define MIN_MTU 68
 903 #define MAX_MTU 65535
 904
 905 static int
 906 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 907 {
 908         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 909                 return -EINVAL;
 910         dev->mtu = new_mtu;
 911         return 0;
 912 }
 913
 914 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 915         netdev_features_t features)
 916 {
 917         struct tun_struct *tun = netdev_priv(dev);
 918
 919         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 920 }
 921 #ifdef CONFIG_NET_POLL_CONTROLLER
 922 static void tun_poll_controller(struct net_device *dev)
 923 {
 924         /*
 925          * Tun only receives frames when:
 926          * 1) the char device endpoint gets data from user space
 927          * 2) the tun socket gets a sendmsg call from user space
 928          * Since both of those are synchronous operations, we are guaranteed
 929          * never to have pending data when we poll for it
 930          * so there is nothing to do here but return.
 931          * We need this though so netpoll recognizes us as an interface that
 932          * supports polling, which enables bridge devices in virt setups to
 933          * still use netconsole
 934          */
 935         return;
 936 }
 937 #endif
 938
 939 static void tun_set_headroom(struct net_device *dev, int new_hr)
 940 {
 941         struct tun_struct *tun = netdev_priv(dev);
 942
 943         if (new_hr < NET_SKB_PAD)
 944                 new_hr = NET_SKB_PAD;
 945
 946         tun->align = new_hr;
 947 }
 948
 949 static const struct net_device_ops tun_netdev_ops = {
 950         .ndo_uninit             = tun_net_uninit,
 951         .ndo_open               = tun_net_open,
 952         .ndo_stop               = tun_net_close,
 953         .ndo_start_xmit         = tun_net_xmit,
 954         .ndo_change_mtu         = tun_net_change_mtu,
 955         .ndo_fix_features       = tun_net_fix_features,
 956         .ndo_select_queue       = tun_select_queue,
 957 #ifdef CONFIG_NET_POLL_CONTROLLER
 958         .ndo_poll_controller    = tun_poll_controller,
 959 #endif
 960         .ndo_set_rx_headroom    = tun_set_headroom,
 961 };
 962
 963 static const struct net_device_ops tap_netdev_ops = {
 964         .ndo_uninit             = tun_net_uninit,
 965         .ndo_open               = tun_net_open,
 966         .ndo_stop               = tun_net_close,
 967         .ndo_start_xmit         = tun_net_xmit,
 968         .ndo_change_mtu         = tun_net_change_mtu,
 969         .ndo_fix_features       = tun_net_fix_features,
 970         .ndo_set_rx_mode        = tun_net_mclist,
 971         .ndo_set_mac_address    = eth_mac_addr,
 972         .ndo_validate_addr      = eth_validate_addr,
 973         .ndo_select_queue       = tun_select_queue,
 974 #ifdef CONFIG_NET_POLL_CONTROLLER
 975         .ndo_poll_controller    = tun_poll_controller,
 976 #endif
 977         .ndo_features_check     = passthru_features_check,
 978         .ndo_set_rx_headroom    = tun_set_headroom,
 979 };
 980
 981 static void tun_flow_init(struct tun_struct *tun)
 982 {
 983         int i;
 984
 985         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 986                 INIT_HLIST_HEAD(&tun->flows[i]);
 987
 988         tun->ageing_time = TUN_FLOW_EXPIRE;
 989         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 990         mod_timer(&tun->flow_gc_timer,
 991                   round_jiffies_up(jiffies + tun->ageing_time));
 992 }
 993
 994 static void tun_flow_uninit(struct tun_struct *tun)
 995 {
 996         del_timer_sync(&tun->flow_gc_timer);
 997         tun_flow_flush(tun);
 998 }
 999
1000 /* Initialize net device. */
1001 static void tun_net_init(struct net_device *dev)
1002 {
1003         struct tun_struct *tun = netdev_priv(dev);
1004
1005         switch (tun->flags & TUN_TYPE_MASK) {
1006         case IFF_TUN:
1007                 dev->netdev_ops = &tun_netdev_ops;
1008
1009                 /* Point-to-Point TUN Device */
1010                 dev->hard_header_len = 0;
1011                 dev->addr_len = 0;
1012                 dev->mtu = 1500;
1013
1014                 /* Zero header length */
1015                 dev->type = ARPHRD_NONE;
1016                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1017                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
1018                 break;
1019
1020         case IFF_TAP:
1021                 dev->netdev_ops = &tap_netdev_ops;
1022                 /* Ethernet TAP Device */
1023                 ether_setup(dev);
1024                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1025                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1026
1027                 eth_hw_addr_random(dev);
1028
1029                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
1030                 break;
1031         }
1032 }
1033
1034 /* Character device part */
1035
1036 /* Poll */
1037 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1038 {
1039         struct tun_file *tfile = file->private_data;
1040         struct tun_struct *tun = __tun_get(tfile);
1041         struct sock *sk;
1042         unsigned int mask = 0;
1043
1044         if (!tun)
1045                 return POLLERR;
1046
1047         sk = tfile->socket.sk;
1048
1049         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1050
1051         poll_wait(file, sk_sleep(sk), wait);
1052
1053         if (!skb_queue_empty(&sk->sk_receive_queue))
1054                 mask |= POLLIN | POLLRDNORM;
1055
1056         if (sock_writeable(sk) ||
1057             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1058              sock_writeable(sk)))
1059                 mask |= POLLOUT | POLLWRNORM;
1060
1061         if (tun->dev->reg_state != NETREG_REGISTERED)
1062                 mask = POLLERR;
1063
1064         tun_put(tun);
1065         return mask;
1066 }
1067
1068 /* prepad is the amount to reserve at front.  len is length after that.
1069  * linear is a hint as to how much to copy (usually headers). */
1070 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1071                                      size_t prepad, size_t len,
1072                                      size_t linear, int noblock)
1073 {
1074         struct sock *sk = tfile->socket.sk;
1075         struct sk_buff *skb;
1076         int err;
1077
1078         /* Under a page?  Don't bother with paged skb. */
1079         if (prepad + len < PAGE_SIZE || !linear)
1080                 linear = len;
1081
1082         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1083                                    &err, 0);
1084         if (!skb)
1085                 return ERR_PTR(err);
1086
1087         skb_reserve(skb, prepad);
1088         skb_put(skb, linear);
1089         skb->data_len = len - linear;
1090         skb->len += len - linear;
1091
1092         return skb;
1093 }
1094
1095 /* Get packet from user space buffer */
1096 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1097                             void *msg_control, struct iov_iter *from,
1098                             int noblock)
1099 {
1100         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1101         struct sk_buff *skb;
1102         size_t total_len = iov_iter_count(from);
1103         size_t len = total_len, align = tun->align, linear;
1104         struct virtio_net_hdr gso = { 0 };
1105         int good_linear;
1106         int copylen;
1107         bool zerocopy = false;
1108         int err;
1109         u32 rxhash;
1110         ssize_t n;
1111
1112         if (!(tun->dev->flags & IFF_UP))
1113                 return -EIO;
1114
1115         if (!(tun->flags & IFF_NO_PI)) {
1116                 if (len < sizeof(pi))
1117                         return -EINVAL;
1118                 len -= sizeof(pi);
1119
1120                 n = copy_from_iter(&pi, sizeof(pi), from);
1121                 if (n != sizeof(pi))
1122                         return -EFAULT;
1123         }
1124
1125         if (tun->flags & IFF_VNET_HDR) {
1126                 if (len < tun->vnet_hdr_sz)
1127                         return -EINVAL;
1128                 len -= tun->vnet_hdr_sz;
1129
1130                 n = copy_from_iter(&gso, sizeof(gso), from);
1131                 if (n != sizeof(gso))
1132                         return -EFAULT;
1133
1134                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1135                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1136                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1137
1138                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1139                         return -EINVAL;
1140                 iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1141         }
1142
1143         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1144                 align += NET_IP_ALIGN;
1145                 if (unlikely(len < ETH_HLEN ||
1146                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1147                         return -EINVAL;
1148         }
1149
1150         good_linear = SKB_MAX_HEAD(align);
1151
1152         if (msg_control) {
1153                 struct iov_iter i = *from;
1154
1155                 /* There are 256 bytes to be copied in skb, so there is
1156                  * enough room for skb expand head in case it is used.
1157                  * The rest of the buffer is mapped from userspace.
1158                  */
1159                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1160                 if (copylen > good_linear)
1161                         copylen = good_linear;
1162                 linear = copylen;
1163                 iov_iter_advance(&i, copylen);
1164                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1165                         zerocopy = true;
1166         }
1167
1168         if (!zerocopy) {
1169                 copylen = len;
1170                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1171                         linear = good_linear;
1172                 else
1173                         linear = tun16_to_cpu(tun, gso.hdr_len);
1174         }
1175
1176         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1177         if (IS_ERR(skb)) {
1178                 if (PTR_ERR(skb) != -EAGAIN)
1179                         tun->dev->stats.rx_dropped++;
1180                 return PTR_ERR(skb);
1181         }
1182
1183         if (zerocopy)
1184                 err = zerocopy_sg_from_iter(skb, from);
1185         else {
1186                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1187                 if (!err && msg_control) {
1188                         struct ubuf_info *uarg = msg_control;
1189                         uarg->callback(uarg, false);
1190                 }
1191         }
1192
1193         if (err) {
1194                 tun->dev->stats.rx_dropped++;
1195                 kfree_skb(skb);
1196                 return -EFAULT;
1197         }
1198
1199         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1200                 if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start),
1201                                           tun16_to_cpu(tun, gso.csum_offset))) {
1202                         tun->dev->stats.rx_frame_errors++;
1203                         kfree_skb(skb);
1204                         return -EINVAL;
1205                 }
1206         }
1207
1208         switch (tun->flags & TUN_TYPE_MASK) {
1209         case IFF_TUN:
1210                 if (tun->flags & IFF_NO_PI) {
1211                         switch (skb->data[0] & 0xf0) {
1212                         case 0x40:
1213                                 pi.proto = htons(ETH_P_IP);
1214                                 break;
1215                         case 0x60:
1216                                 pi.proto = htons(ETH_P_IPV6);
1217                                 break;
1218                         default:
1219                                 tun->dev->stats.rx_dropped++;
1220                                 kfree_skb(skb);
1221                                 return -EINVAL;
1222                         }
1223                 }
1224
1225                 skb_reset_mac_header(skb);
1226                 skb->protocol = pi.proto;
1227                 skb->dev = tun->dev;
1228                 break;
1229         case IFF_TAP:
1230                 skb->protocol = eth_type_trans(skb, tun->dev);
1231                 break;
1232         }
1233
1234         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1235                 pr_debug("GSO!\n");
1236                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1237                 case VIRTIO_NET_HDR_GSO_TCPV4:
1238                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1239                         break;
1240                 case VIRTIO_NET_HDR_GSO_TCPV6:
1241                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1242                         break;
1243                 case VIRTIO_NET_HDR_GSO_UDP:
1244                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1245                         break;
1246                 default:
1247                         tun->dev->stats.rx_frame_errors++;
1248                         kfree_skb(skb);
1249                         return -EINVAL;
1250                 }
1251
1252                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1253                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1254
1255                 skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size);
1256                 if (skb_shinfo(skb)->gso_size == 0) {
1257                         tun->dev->stats.rx_frame_errors++;
1258                         kfree_skb(skb);
1259                         return -EINVAL;
1260                 }
1261
1262                 /* Header must be checked, and gso_segs computed. */
1263                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1264                 skb_shinfo(skb)->gso_segs = 0;
1265         }
1266
1267         /* copy skb_ubuf_info for callback when skb has no error */
1268         if (zerocopy) {
1269                 skb_shinfo(skb)->destructor_arg = msg_control;
1270                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1271                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1272         }
1273
1274         skb_reset_network_header(skb);
1275         skb_probe_transport_header(skb, 0);
1276
1277         rxhash = skb_get_hash(skb);
1278         netif_rx_ni(skb);
1279
1280         tun->dev->stats.rx_packets++;
1281         tun->dev->stats.rx_bytes += len;
1282
1283         tun_flow_update(tun, rxhash, tfile);
1284         return total_len;
1285 }
1286
1287 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1288 {
1289         struct file *file = iocb->ki_filp;
1290         struct tun_struct *tun = tun_get(file);
1291         struct tun_file *tfile = file->private_data;
1292         ssize_t result;
1293
1294         if (!tun)
1295                 return -EBADFD;
1296
1297         result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1298
1299         tun_put(tun);
1300         return result;
1301 }
1302
1303 /* Put packet to the user space buffer */
1304 static ssize_t tun_put_user(struct tun_struct *tun,
1305                             struct tun_file *tfile,
1306                             struct sk_buff *skb,
1307                             struct iov_iter *iter)
1308 {
1309         struct tun_pi pi = { 0, skb->protocol };
1310         ssize_t total;
1311         int vlan_offset = 0;
1312         int vlan_hlen = 0;
1313         int vnet_hdr_sz = 0;
1314
1315         if (skb_vlan_tag_present(skb))
1316                 vlan_hlen = VLAN_HLEN;
1317
1318         if (tun->flags & IFF_VNET_HDR)
1319                 vnet_hdr_sz = tun->vnet_hdr_sz;
1320
1321         total = skb->len + vlan_hlen + vnet_hdr_sz;
1322
1323         if (!(tun->flags & IFF_NO_PI)) {
1324                 if (iov_iter_count(iter) < sizeof(pi))
1325                         return -EINVAL;
1326
1327                 total += sizeof(pi);
1328                 if (iov_iter_count(iter) < total) {
1329                         /* Packet will be striped */
1330                         pi.flags |= TUN_PKT_STRIP;
1331                 }
1332
1333                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1334                         return -EFAULT;
1335         }
1336
1337         if (vnet_hdr_sz) {
1338                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1339                 if (iov_iter_count(iter) < vnet_hdr_sz)
1340                         return -EINVAL;
1341
1342                 if (skb_is_gso(skb)) {
1343                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1344
1345                         /* This is a hint as to how much should be linear. */
1346                         gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb));
1347                         gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size);
1348                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1349                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1350                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1351                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1352                         else if (sinfo->gso_type & SKB_GSO_UDP)
1353                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1354                         else {
1355                                 pr_err("unexpected GSO type: "
1356                                        "0x%x, gso_size %d, hdr_len %d\n",
1357                                        sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1358                                        tun16_to_cpu(tun, gso.hdr_len));
1359                                 print_hex_dump(KERN_ERR, "tun: ",
1360                                                DUMP_PREFIX_NONE,
1361                                                16, 1, skb->head,
1362                                                min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1363                                 WARN_ON_ONCE(1);
1364                                 return -EINVAL;
1365                         }
1366                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1367                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1368                 } else
1369                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1370
1371                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1372                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1373                         gso.csum_start = cpu_to_tun16(tun, skb_checksum_start_offset(skb) +
1374                                                       vlan_hlen);
1375                         gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset);
1376                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1377                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1378                 } /* else everything is zero */
1379
1380                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1381                         return -EFAULT;
1382
1383                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1384         }
1385
1386         if (vlan_hlen) {
1387                 int ret;
1388                 struct {
1389                         __be16 h_vlan_proto;
1390                         __be16 h_vlan_TCI;
1391                 } veth;
1392
1393                 veth.h_vlan_proto = skb->vlan_proto;
1394                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1395
1396                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1397
1398                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1399                 if (ret || !iov_iter_count(iter))
1400                         goto done;
1401
1402                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1403                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1404                         goto done;
1405         }
1406
1407         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1408
1409 done:
1410         tun->dev->stats.tx_packets++;
1411         tun->dev->stats.tx_bytes += skb->len + vlan_hlen;
1412
1413         return total;
1414 }
1415
1416 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1417                            struct iov_iter *to,
1418                            int noblock)
1419 {
1420         struct sk_buff *skb;
1421         ssize_t ret;
1422         int peeked, err, off = 0;
1423
1424         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1425
1426         if (!iov_iter_count(to))
1427                 return 0;
1428
1429         if (tun->dev->reg_state != NETREG_REGISTERED)
1430                 return -EIO;
1431
1432         /* Read frames from queue */
1433         skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1434                                   &peeked, &off, &err);
1435         if (!skb)
1436                 return err;
1437
1438         ret = tun_put_user(tun, tfile, skb, to);
1439         if (unlikely(ret < 0))
1440                 kfree_skb(skb);
1441         else
1442                 consume_skb(skb);
1443
1444         return ret;
1445 }
1446
1447 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1448 {
1449         struct file *file = iocb->ki_filp;
1450         struct tun_file *tfile = file->private_data;
1451         struct tun_struct *tun = __tun_get(tfile);
1452         ssize_t len = iov_iter_count(to), ret;
1453
1454         if (!tun)
1455                 return -EBADFD;
1456         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1457         ret = min_t(ssize_t, ret, len);
1458         if (ret > 0)
1459                 iocb->ki_pos = ret;
1460         tun_put(tun);
1461         return ret;
1462 }
1463
1464 static void tun_free_netdev(struct net_device *dev)
1465 {
1466         struct tun_struct *tun = netdev_priv(dev);
1467
1468         BUG_ON(!(list_empty(&tun->disabled)));
1469         tun_flow_uninit(tun);
1470         security_tun_dev_free_security(tun->security);
1471         free_netdev(dev);
1472 }
1473
1474 static void tun_setup(struct net_device *dev)
1475 {
1476         struct tun_struct *tun = netdev_priv(dev);
1477
1478         tun->owner = INVALID_UID;
1479         tun->group = INVALID_GID;
1480
1481         dev->ethtool_ops = &tun_ethtool_ops;
1482         dev->destructor = tun_free_netdev;
1483 }
1484
1485 /* Trivial set of netlink ops to allow deleting tun or tap
1486  * device with netlink.
1487  */
1488 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1489 {
1490         return -EINVAL;
1491 }
1492
1493 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1494         .kind           = DRV_NAME,
1495         .priv_size      = sizeof(struct tun_struct),
1496         .setup          = tun_setup,
1497         .validate       = tun_validate,
1498 };
1499
1500 static void tun_sock_write_space(struct sock *sk)
1501 {
1502         struct tun_file *tfile;
1503         wait_queue_head_t *wqueue;
1504
1505         if (!sock_writeable(sk))
1506                 return;
1507
1508         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1509                 return;
1510
1511         wqueue = sk_sleep(sk);
1512         if (wqueue && waitqueue_active(wqueue))
1513                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1514                                                 POLLWRNORM | POLLWRBAND);
1515
1516         tfile = container_of(sk, struct tun_file, sk);
1517         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1518 }
1519
1520 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1521 {
1522         int ret;
1523         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1524         struct tun_struct *tun = __tun_get(tfile);
1525
1526         if (!tun)
1527                 return -EBADFD;
1528
1529         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1530                            m->msg_flags & MSG_DONTWAIT);
1531         tun_put(tun);
1532         return ret;
1533 }
1534
1535 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1536                        int flags)
1537 {
1538         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1539         struct tun_struct *tun = __tun_get(tfile);
1540         int ret;
1541
1542         if (!tun)
1543                 return -EBADFD;
1544
1545         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1546                 ret = -EINVAL;
1547                 goto out;
1548         }
1549         if (flags & MSG_ERRQUEUE) {
1550                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1551                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1552                 goto out;
1553         }
1554         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1555         if (ret > (ssize_t)total_len) {
1556                 m->msg_flags |= MSG_TRUNC;
1557                 ret = flags & MSG_TRUNC ? ret : total_len;
1558         }
1559 out:
1560         tun_put(tun);
1561         return ret;
1562 }
1563
1564 /* Ops structure to mimic raw sockets with tun */
1565 static const struct proto_ops tun_socket_ops = {
1566         .sendmsg = tun_sendmsg,
1567         .recvmsg = tun_recvmsg,
1568 };
1569
1570 static struct proto tun_proto = {
1571         .name           = "tun",
1572         .owner          = THIS_MODULE,
1573         .obj_size       = sizeof(struct tun_file),
1574 };
1575
1576 static int tun_flags(struct tun_struct *tun)
1577 {
1578         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1579 }
1580
1581 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1582                               char *buf)
1583 {
1584         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1585         return sprintf(buf, "0x%x\n", tun_flags(tun));
1586 }
1587
1588 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1589                               char *buf)
1590 {
1591         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1592         return uid_valid(tun->owner)?
1593                 sprintf(buf, "%u\n",
1594                         from_kuid_munged(current_user_ns(), tun->owner)):
1595                 sprintf(buf, "-1\n");
1596 }
1597
1598 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1599                               char *buf)
1600 {
1601         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1602         return gid_valid(tun->group) ?
1603                 sprintf(buf, "%u\n",
1604                         from_kgid_munged(current_user_ns(), tun->group)):
1605                 sprintf(buf, "-1\n");
1606 }
1607
1608 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1609 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1610 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1611
1612 static struct attribute *tun_dev_attrs[] = {
1613         &dev_attr_tun_flags.attr,
1614         &dev_attr_owner.attr,
1615         &dev_attr_group.attr,
1616         NULL
1617 };
1618
1619 static const struct attribute_group tun_attr_group = {
1620         .attrs = tun_dev_attrs
1621 };
1622
1623 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1624 {
1625         struct tun_struct *tun;
1626         struct tun_file *tfile = file->private_data;
1627         struct net_device *dev;
1628         int err;
1629
1630         if (tfile->detached)
1631                 return -EINVAL;
1632
1633         dev = __dev_get_by_name(net, ifr->ifr_name);
1634         if (dev) {
1635                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1636                         return -EBUSY;
1637                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1638                         tun = netdev_priv(dev);
1639                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1640                         tun = netdev_priv(dev);
1641                 else
1642                         return -EINVAL;
1643
1644                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1645                     !!(tun->flags & IFF_MULTI_QUEUE))
1646                         return -EINVAL;
1647
1648                 if (tun_not_capable(tun))
1649                         return -EPERM;
1650                 err = security_tun_dev_open(tun->security);
1651                 if (err < 0)
1652                         return err;
1653
1654                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1655                 if (err < 0)
1656                         return err;
1657
1658                 if (tun->flags & IFF_MULTI_QUEUE &&
1659                     (tun->numqueues + tun->numdisabled > 1)) {
1660                         /* One or more queue has already been attached, no need
1661                          * to initialize the device again.
1662                          */
1663                         return 0;
1664                 }
1665         }
1666         else {
1667                 char *name;
1668                 unsigned long flags = 0;
1669                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1670                              MAX_TAP_QUEUES : 1;
1671
1672                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1673                         return -EPERM;
1674                 err = security_tun_dev_create();
1675                 if (err < 0)
1676                         return err;
1677
1678                 /* Set dev type */
1679                 if (ifr->ifr_flags & IFF_TUN) {
1680                         /* TUN device */
1681                         flags |= IFF_TUN;
1682                         name = "tun%d";
1683                 } else if (ifr->ifr_flags & IFF_TAP) {
1684                         /* TAP device */
1685                         flags |= IFF_TAP;
1686                         name = "tap%d";
1687                 } else
1688                         return -EINVAL;
1689
1690                 if (*ifr->ifr_name)
1691                         name = ifr->ifr_name;
1692
1693                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1694                                        NET_NAME_UNKNOWN, tun_setup, queues,
1695                                        queues);
1696
1697                 if (!dev)
1698                         return -ENOMEM;
1699
1700                 dev_net_set(dev, net);
1701                 dev->rtnl_link_ops = &tun_link_ops;
1702                 dev->ifindex = tfile->ifindex;
1703                 dev->sysfs_groups[0] = &tun_attr_group;
1704
1705                 tun = netdev_priv(dev);
1706                 tun->dev = dev;
1707                 tun->flags = flags;
1708                 tun->txflt.count = 0;
1709                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1710
1711                 tun->align = NET_SKB_PAD;
1712                 tun->filter_attached = false;
1713                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1714
1715                 spin_lock_init(&tun->lock);
1716
1717                 err = security_tun_dev_alloc_security(&tun->security);
1718                 if (err < 0)
1719                         goto err_free_dev;
1720
1721                 tun_net_init(dev);
1722                 tun_flow_init(tun);
1723
1724                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1725                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1726                                    NETIF_F_HW_VLAN_STAG_TX;
1727                 dev->features = dev->hw_features;
1728                 dev->vlan_features = dev->features &
1729                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1730                                        NETIF_F_HW_VLAN_STAG_TX);
1731
1732                 INIT_LIST_HEAD(&tun->disabled);
1733                 err = tun_attach(tun, file, false);
1734                 if (err < 0)
1735                         goto err_free_flow;
1736
1737                 err = register_netdevice(tun->dev);
1738                 if (err < 0)
1739                         goto err_detach;
1740         }
1741
1742         netif_carrier_on(tun->dev);
1743
1744         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1745
1746         tun->flags = (tun->flags & ~TUN_FEATURES) |
1747                 (ifr->ifr_flags & TUN_FEATURES);
1748
1749         /* Make sure persistent devices do not get stuck in
1750          * xoff state.
1751          */
1752         if (netif_running(tun->dev))
1753                 netif_tx_wake_all_queues(tun->dev);
1754
1755         strcpy(ifr->ifr_name, tun->dev->name);
1756         return 0;
1757
1758 err_detach:
1759         tun_detach_all(dev);
1760 err_free_flow:
1761         tun_flow_uninit(tun);
1762         security_tun_dev_free_security(tun->security);
1763 err_free_dev:
1764         free_netdev(dev);
1765         return err;
1766 }
1767
1768 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1769                        struct ifreq *ifr)
1770 {
1771         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1772
1773         strcpy(ifr->ifr_name, tun->dev->name);
1774
1775         ifr->ifr_flags = tun_flags(tun);
1776
1777 }
1778
1779 /* This is like a cut-down ethtool ops, except done via tun fd so no
1780  * privs required. */
1781 static int set_offload(struct tun_struct *tun, unsigned long arg)
1782 {
1783         netdev_features_t features = 0;
1784
1785         if (arg & TUN_F_CSUM) {
1786                 features |= NETIF_F_HW_CSUM;
1787                 arg &= ~TUN_F_CSUM;
1788
1789                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1790                         if (arg & TUN_F_TSO_ECN) {
1791                                 features |= NETIF_F_TSO_ECN;
1792                                 arg &= ~TUN_F_TSO_ECN;
1793                         }
1794                         if (arg & TUN_F_TSO4)
1795                                 features |= NETIF_F_TSO;
1796                         if (arg & TUN_F_TSO6)
1797                                 features |= NETIF_F_TSO6;
1798                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1799                 }
1800
1801                 if (arg & TUN_F_UFO) {
1802                         features |= NETIF_F_UFO;
1803                         arg &= ~TUN_F_UFO;
1804                 }
1805         }
1806
1807         /* This gives the user a way to test for new features in future by
1808          * trying to set them. */
1809         if (arg)
1810                 return -EINVAL;
1811
1812         tun->set_features = features;
1813         netdev_update_features(tun->dev);
1814
1815         return 0;
1816 }
1817
1818 static void tun_detach_filter(struct tun_struct *tun, int n)
1819 {
1820         int i;
1821         struct tun_file *tfile;
1822
1823         for (i = 0; i < n; i++) {
1824                 tfile = rtnl_dereference(tun->tfiles[i]);
1825                 sk_detach_filter(tfile->socket.sk);
1826         }
1827
1828         tun->filter_attached = false;
1829 }
1830
1831 static int tun_attach_filter(struct tun_struct *tun)
1832 {
1833         int i, ret = 0;
1834         struct tun_file *tfile;
1835
1836         for (i = 0; i < tun->numqueues; i++) {
1837                 tfile = rtnl_dereference(tun->tfiles[i]);
1838                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1839                 if (ret) {
1840                         tun_detach_filter(tun, i);
1841                         return ret;
1842                 }
1843         }
1844
1845         tun->filter_attached = true;
1846         return ret;
1847 }
1848
1849 static void tun_set_sndbuf(struct tun_struct *tun)
1850 {
1851         struct tun_file *tfile;
1852         int i;
1853
1854         for (i = 0; i < tun->numqueues; i++) {
1855                 tfile = rtnl_dereference(tun->tfiles[i]);
1856                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1857         }
1858 }
1859
1860 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1861 {
1862         struct tun_file *tfile = file->private_data;
1863         struct tun_struct *tun;
1864         int ret = 0;
1865
1866         rtnl_lock();
1867
1868         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1869                 tun = tfile->detached;
1870                 if (!tun) {
1871                         ret = -EINVAL;
1872                         goto unlock;
1873                 }
1874                 ret = security_tun_dev_attach_queue(tun->security);
1875                 if (ret < 0)
1876                         goto unlock;
1877                 ret = tun_attach(tun, file, false);
1878         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1879                 tun = rtnl_dereference(tfile->tun);
1880                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1881                         ret = -EINVAL;
1882                 else
1883                         __tun_detach(tfile, false);
1884         } else
1885                 ret = -EINVAL;
1886
1887 unlock:
1888         rtnl_unlock();
1889         return ret;
1890 }
1891
1892 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1893                             unsigned long arg, int ifreq_len)
1894 {
1895         struct tun_file *tfile = file->private_data;
1896         struct tun_struct *tun;
1897         void __user* argp = (void __user*)arg;
1898         struct ifreq ifr;
1899         kuid_t owner;
1900         kgid_t group;
1901         int sndbuf;
1902         int vnet_hdr_sz;
1903         unsigned int ifindex;
1904         int le;
1905         int ret;
1906
1907         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1908                 if (copy_from_user(&ifr, argp, ifreq_len))
1909                         return -EFAULT;
1910         } else {
1911                 memset(&ifr, 0, sizeof(ifr));
1912         }
1913         if (cmd == TUNGETFEATURES) {
1914                 /* Currently this just means: "what IFF flags are valid?".
1915                  * This is needed because we never checked for invalid flags on
1916                  * TUNSETIFF.
1917                  */
1918                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1919                                 (unsigned int __user*)argp);
1920         } else if (cmd == TUNSETQUEUE)
1921                 return tun_set_queue(file, &ifr);
1922
1923         ret = 0;
1924         rtnl_lock();
1925
1926         tun = __tun_get(tfile);
1927         if (cmd == TUNSETIFF && !tun) {
1928                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1929
1930                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
1931
1932                 if (ret)
1933                         goto unlock;
1934
1935                 if (copy_to_user(argp, &ifr, ifreq_len))
1936                         ret = -EFAULT;
1937                 goto unlock;
1938         }
1939         if (cmd == TUNSETIFINDEX) {
1940                 ret = -EPERM;
1941                 if (tun)
1942                         goto unlock;
1943
1944                 ret = -EFAULT;
1945                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1946                         goto unlock;
1947
1948                 ret = 0;
1949                 tfile->ifindex = ifindex;
1950                 goto unlock;
1951         }
1952
1953         ret = -EBADFD;
1954         if (!tun)
1955                 goto unlock;
1956
1957         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1958
1959         ret = 0;
1960         switch (cmd) {
1961         case TUNGETIFF:
1962                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1963
1964                 if (tfile->detached)
1965                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
1966                 if (!tfile->socket.sk->sk_filter)
1967                         ifr.ifr_flags |= IFF_NOFILTER;
1968
1969                 if (copy_to_user(argp, &ifr, ifreq_len))
1970                         ret = -EFAULT;
1971                 break;
1972
1973         case TUNSETNOCSUM:
1974                 /* Disable/Enable checksum */
1975
1976                 /* [unimplemented] */
1977                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1978                           arg ? "disabled" : "enabled");
1979                 break;
1980
1981         case TUNSETPERSIST:
1982                 /* Disable/Enable persist mode. Keep an extra reference to the
1983                  * module to prevent the module being unprobed.
1984                  */
1985                 if (arg && !(tun->flags & IFF_PERSIST)) {
1986                         tun->flags |= IFF_PERSIST;
1987                         __module_get(THIS_MODULE);
1988                 }
1989                 if (!arg && (tun->flags & IFF_PERSIST)) {
1990                         tun->flags &= ~IFF_PERSIST;
1991                         module_put(THIS_MODULE);
1992                 }
1993
1994                 tun_debug(KERN_INFO, tun, "persist %s\n",
1995                           arg ? "enabled" : "disabled");
1996                 break;
1997
1998         case TUNSETOWNER:
1999                 /* Set owner of the device */
2000                 owner = make_kuid(current_user_ns(), arg);
2001                 if (!uid_valid(owner)) {
2002                         ret = -EINVAL;
2003                         break;
2004                 }
2005                 tun->owner = owner;
2006                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2007                           from_kuid(&init_user_ns, tun->owner));
2008                 break;
2009
2010         case TUNSETGROUP:
2011                 /* Set group of the device */
2012                 group = make_kgid(current_user_ns(), arg);
2013                 if (!gid_valid(group)) {
2014                         ret = -EINVAL;
2015                         break;
2016                 }
2017                 tun->group = group;
2018                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2019                           from_kgid(&init_user_ns, tun->group));
2020                 break;
2021
2022         case TUNSETLINK:
2023                 /* Only allow setting the type when the interface is down */
2024                 if (tun->dev->flags & IFF_UP) {
2025                         tun_debug(KERN_INFO, tun,
2026                                   "Linktype set failed because interface is up\n");
2027                         ret = -EBUSY;
2028                 } else {
2029                         tun->dev->type = (int) arg;
2030                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2031                                   tun->dev->type);
2032                         ret = 0;
2033                 }
2034                 break;
2035
2036 #ifdef TUN_DEBUG
2037         case TUNSETDEBUG:
2038                 tun->debug = arg;
2039                 break;
2040 #endif
2041         case TUNSETOFFLOAD:
2042                 ret = set_offload(tun, arg);
2043                 break;
2044
2045         case TUNSETTXFILTER:
2046                 /* Can be set only for TAPs */
2047                 ret = -EINVAL;
2048                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2049                         break;
2050                 ret = update_filter(&tun->txflt, (void __user *)arg);
2051                 break;
2052
2053         case SIOCGIFHWADDR:
2054                 /* Get hw address */
2055                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2056                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2057                 if (copy_to_user(argp, &ifr, ifreq_len))
2058                         ret = -EFAULT;
2059                 break;
2060
2061         case SIOCSIFHWADDR:
2062                 /* Set hw address */
2063                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2064                           ifr.ifr_hwaddr.sa_data);
2065
2066                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2067                 break;
2068
2069         case TUNGETSNDBUF:
2070                 sndbuf = tfile->socket.sk->sk_sndbuf;
2071                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2072                         ret = -EFAULT;
2073                 break;
2074
2075         case TUNSETSNDBUF:
2076                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2077                         ret = -EFAULT;
2078                         break;
2079                 }
2080
2081                 tun->sndbuf = sndbuf;
2082                 tun_set_sndbuf(tun);
2083                 break;
2084
2085         case TUNGETVNETHDRSZ:
2086                 vnet_hdr_sz = tun->vnet_hdr_sz;
2087                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2088                         ret = -EFAULT;
2089                 break;
2090
2091         case TUNSETVNETHDRSZ:
2092                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2093                         ret = -EFAULT;
2094                         break;
2095                 }
2096                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2097                         ret = -EINVAL;
2098                         break;
2099                 }
2100
2101                 tun->vnet_hdr_sz = vnet_hdr_sz;
2102                 break;
2103
2104         case TUNGETVNETLE:
2105                 le = !!(tun->flags & TUN_VNET_LE);
2106                 if (put_user(le, (int __user *)argp))
2107                         ret = -EFAULT;
2108                 break;
2109
2110         case TUNSETVNETLE:
2111                 if (get_user(le, (int __user *)argp)) {
2112                         ret = -EFAULT;
2113                         break;
2114                 }
2115                 if (le)
2116                         tun->flags |= TUN_VNET_LE;
2117                 else
2118                         tun->flags &= ~TUN_VNET_LE;
2119                 break;
2120
2121         case TUNGETVNETBE:
2122                 ret = tun_get_vnet_be(tun, argp);
2123                 break;
2124
2125         case TUNSETVNETBE:
2126                 ret = tun_set_vnet_be(tun, argp);
2127                 break;
2128
2129         case TUNATTACHFILTER:
2130                 /* Can be set only for TAPs */
2131                 ret = -EINVAL;
2132                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2133                         break;
2134                 ret = -EFAULT;
2135                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2136                         break;
2137
2138                 ret = tun_attach_filter(tun);
2139                 break;
2140
2141         case TUNDETACHFILTER:
2142                 /* Can be set only for TAPs */
2143                 ret = -EINVAL;
2144                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2145                         break;
2146                 ret = 0;
2147                 tun_detach_filter(tun, tun->numqueues);
2148                 break;
2149
2150         case TUNGETFILTER:
2151                 ret = -EINVAL;
2152                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2153                         break;
2154                 ret = -EFAULT;
2155                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2156                         break;
2157                 ret = 0;
2158                 break;
2159
2160         default:
2161                 ret = -EINVAL;
2162                 break;
2163         }
2164
2165 unlock:
2166         rtnl_unlock();
2167         if (tun)
2168                 tun_put(tun);
2169         return ret;
2170 }
2171
2172 static long tun_chr_ioctl(struct file *file,
2173                           unsigned int cmd, unsigned long arg)
2174 {
2175         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2176 }
2177
2178 #ifdef CONFIG_COMPAT
2179 static long tun_chr_compat_ioctl(struct file *file,
2180                          unsigned int cmd, unsigned long arg)
2181 {
2182         switch (cmd) {
2183         case TUNSETIFF:
2184         case TUNGETIFF:
2185         case TUNSETTXFILTER:
2186         case TUNGETSNDBUF:
2187         case TUNSETSNDBUF:
2188         case SIOCGIFHWADDR:
2189         case SIOCSIFHWADDR:
2190                 arg = (unsigned long)compat_ptr(arg);
2191                 break;
2192         default:
2193                 arg = (compat_ulong_t)arg;
2194                 break;
2195         }
2196
2197         /*
2198          * compat_ifreq is shorter than ifreq, so we must not access beyond
2199          * the end of that structure. All fields that are used in this
2200          * driver are compatible though, we don't need to convert the
2201          * contents.
2202          */
2203         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2204 }
2205 #endif /* CONFIG_COMPAT */
2206
2207 static int tun_chr_fasync(int fd, struct file *file, int on)
2208 {
2209         struct tun_file *tfile = file->private_data;
2210         int ret;
2211
2212         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2213                 goto out;
2214
2215         if (on) {
2216                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2217                 tfile->flags |= TUN_FASYNC;
2218         } else
2219                 tfile->flags &= ~TUN_FASYNC;
2220         ret = 0;
2221 out:
2222         return ret;
2223 }
2224
2225 static int tun_chr_open(struct inode *inode, struct file * file)
2226 {
2227         struct net *net = current->nsproxy->net_ns;
2228         struct tun_file *tfile;
2229
2230         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2231
2232         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2233                                             &tun_proto, 0);
2234         if (!tfile)
2235                 return -ENOMEM;
2236         RCU_INIT_POINTER(tfile->tun, NULL);
2237         tfile->flags = 0;
2238         tfile->ifindex = 0;
2239
2240         init_waitqueue_head(&tfile->wq.wait);
2241         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2242
2243         tfile->socket.file = file;
2244         tfile->socket.ops = &tun_socket_ops;
2245
2246         sock_init_data(&tfile->socket, &tfile->sk);
2247
2248         tfile->sk.sk_write_space = tun_sock_write_space;
2249         tfile->sk.sk_sndbuf = INT_MAX;
2250
2251         file->private_data = tfile;
2252         INIT_LIST_HEAD(&tfile->next);
2253
2254         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2255
2256         return 0;
2257 }
2258
2259 static int tun_chr_close(struct inode *inode, struct file *file)
2260 {
2261         struct tun_file *tfile = file->private_data;
2262
2263         tun_detach(tfile, true);
2264
2265         return 0;
2266 }
2267
2268 #ifdef CONFIG_PROC_FS
2269 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2270 {
2271         struct tun_struct *tun;
2272         struct ifreq ifr;
2273
2274         memset(&ifr, 0, sizeof(ifr));
2275
2276         rtnl_lock();
2277         tun = tun_get(f);
2278         if (tun)
2279                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2280         rtnl_unlock();
2281
2282         if (tun)
2283                 tun_put(tun);
2284
2285         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2286 }
2287 #endif
2288
2289 static const struct file_operations tun_fops = {
2290         .owner  = THIS_MODULE,
2291         .llseek = no_llseek,
2292         .read_iter  = tun_chr_read_iter,
2293         .write_iter = tun_chr_write_iter,
2294         .poll   = tun_chr_poll,
2295         .unlocked_ioctl = tun_chr_ioctl,
2296 #ifdef CONFIG_COMPAT
2297         .compat_ioctl = tun_chr_compat_ioctl,
2298 #endif
2299         .open   = tun_chr_open,
2300         .release = tun_chr_close,
2301         .fasync = tun_chr_fasync,
2302 #ifdef CONFIG_PROC_FS
2303         .show_fdinfo = tun_chr_show_fdinfo,
2304 #endif
2305 };
2306
2307 static struct miscdevice tun_miscdev = {
2308         .minor = TUN_MINOR,
2309         .name = "tun",
2310         .nodename = "net/tun",
2311         .fops = &tun_fops,
2312 };
2313
2314 /* ethtool interface */
2315
2316 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2317 {
2318         cmd->supported          = 0;
2319         cmd->advertising        = 0;
2320         ethtool_cmd_speed_set(cmd, SPEED_10);
2321         cmd->duplex             = DUPLEX_FULL;
2322         cmd->port               = PORT_TP;
2323         cmd->phy_address        = 0;
2324         cmd->transceiver        = XCVR_INTERNAL;
2325         cmd->autoneg            = AUTONEG_DISABLE;
2326         cmd->maxtxpkt           = 0;
2327         cmd->maxrxpkt           = 0;
2328         return 0;
2329 }
2330
2331 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2332 {
2333         struct tun_struct *tun = netdev_priv(dev);
2334
2335         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2336         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2337
2338         switch (tun->flags & TUN_TYPE_MASK) {
2339         case IFF_TUN:
2340                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2341                 break;
2342         case IFF_TAP:
2343                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2344                 break;
2345         }
2346 }
2347
2348 static u32 tun_get_msglevel(struct net_device *dev)
2349 {
2350 #ifdef TUN_DEBUG
2351         struct tun_struct *tun = netdev_priv(dev);
2352         return tun->debug;
2353 #else
2354         return -EOPNOTSUPP;
2355 #endif
2356 }
2357
2358 static void tun_set_msglevel(struct net_device *dev, u32 value)
2359 {
2360 #ifdef TUN_DEBUG
2361         struct tun_struct *tun = netdev_priv(dev);
2362         tun->debug = value;
2363 #endif
2364 }
2365
2366 static const struct ethtool_ops tun_ethtool_ops = {
2367         .get_settings   = tun_get_settings,
2368         .get_drvinfo    = tun_get_drvinfo,
2369         .get_msglevel   = tun_get_msglevel,
2370         .set_msglevel   = tun_set_msglevel,
2371         .get_link       = ethtool_op_get_link,
2372         .get_ts_info    = ethtool_op_get_ts_info,
2373 };
2374
2375
2376 static int __init tun_init(void)
2377 {
2378         int ret = 0;
2379
2380         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2381         pr_info("%s\n", DRV_COPYRIGHT);
2382
2383         ret = rtnl_link_register(&tun_link_ops);
2384         if (ret) {
2385                 pr_err("Can't register link_ops\n");
2386                 goto err_linkops;
2387         }
2388
2389         ret = misc_register(&tun_miscdev);
2390         if (ret) {
2391                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2392                 goto err_misc;
2393         }
2394         return  0;
2395 err_misc:
2396         rtnl_link_unregister(&tun_link_ops);
2397 err_linkops:
2398         return ret;
2399 }
2400
2401 static void tun_cleanup(void)
2402 {
2403         misc_deregister(&tun_miscdev);
2404         rtnl_link_unregister(&tun_link_ops);
2405 }
2406
2407 /* Get an underlying socket object from tun file.  Returns error unless file is
2408  * attached to a device.  The returned object works like a packet socket, it
2409  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2410  * holding a reference to the file for as long as the socket is in use. */
2411 struct socket *tun_get_socket(struct file *file)
2412 {
2413         struct tun_file *tfile;
2414         if (file->f_op != &tun_fops)
2415                 return ERR_PTR(-EINVAL);
2416         tfile = file->private_data;
2417         if (!tfile)
2418                 return ERR_PTR(-EBADFD);
2419         return &tfile->socket;
2420 }
2421 EXPORT_SYMBOL_GPL(tun_get_socket);
2422
2423 module_init(tun_init);
2424 module_exit(tun_cleanup);
2425 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2426 MODULE_AUTHOR(DRV_COPYRIGHT);
2427 MODULE_LICENSE("GPL");
2428 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2429 MODULE_ALIAS("devname:net/tun");