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 #include <linux/ieee802154.h>
  75 #include <linux/if_ltalk.h>
  76 #include <uapi/linux/if_fddi.h>
  77 #include <uapi/linux/if_hippi.h>
  78 #include <uapi/linux/if_fc.h>
  79 #include <net/ax25.h>
  80 #include <net/rose.h>
  81 #include <net/6lowpan.h>
  82
  83 #include <asm/uaccess.h>
  84
  85 /* Uncomment to enable debugging */
  86 /* #define TUN_DEBUG 1 */
  87
  88 #ifdef TUN_DEBUG
  89 static int debug;
  90
  91 #define tun_debug(level, tun, fmt, args...)                     \
  92 do {                                                            \
  93         if (tun->debug)                                         \
  94                 netdev_printk(level, tun->dev, fmt, ##args);    \
  95 } while (0)
  96 #define DBG1(level, fmt, args...)                               \
  97 do {                                                            \
  98         if (debug == 2)                                         \
  99                 printk(level fmt, ##args);                      \
 100 } while (0)
 101 #else
 102 #define tun_debug(level, tun, fmt, args...)                     \
 103 do {                                                            \
 104         if (0)                                                  \
 105                 netdev_printk(level, tun->dev, fmt, ##args);    \
 106 } while (0)
 107 #define DBG1(level, fmt, args...)                               \
 108 do {                                                            \
 109         if (0)                                                  \
 110                 printk(level fmt, ##args);                      \
 111 } while (0)
 112 #endif
 113
 114 /* TUN device flags */
 115
 116 /* IFF_ATTACH_QUEUE is never stored in device flags,
 117  * overload it to mean fasync when stored there.
 118  */
 119 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 120 /* High bits in flags field are unused. */
 121 #define TUN_VNET_LE     0x80000000
 122 #define TUN_VNET_BE     0x40000000
 123
 124 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 125                       IFF_MULTI_QUEUE)
 126 #define GOODCOPY_LEN 128
 127
 128 #define FLT_EXACT_COUNT 8
 129 struct tap_filter {
 130         unsigned int    count;    /* Number of addrs. Zero means disabled */
 131         u32             mask[2];  /* Mask of the hashed addrs */
 132         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 133 };
 134
 135 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 136  * to max number of VCPUs in guest. */
 137 #define MAX_TAP_QUEUES 256
 138 #define MAX_TAP_FLOWS  4096
 139
 140 #define TUN_FLOW_EXPIRE (3 * HZ)
 141
 142 /* A tun_file connects an open character device to a tuntap netdevice. It
 143  * also contains all socket related structures (except sock_fprog and tap_filter)
 144  * to serve as one transmit queue for tuntap device. The sock_fprog and
 145  * tap_filter were kept in tun_struct since they were used for filtering for the
 146  * netdevice not for a specific queue (at least I didn't see the requirement for
 147  * this).
 148  *
 149  * RCU usage:
 150  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 151  * other can only be read while rcu_read_lock or rtnl_lock is held.
 152  */
 153 struct tun_file {
 154         struct sock sk;
 155         struct socket socket;
 156         struct socket_wq wq;
 157         struct tun_struct __rcu *tun;
 158         struct fasync_struct *fasync;
 159         /* only used for fasnyc */
 160         unsigned int flags;
 161         union {
 162                 u16 queue_index;
 163                 unsigned int ifindex;
 164         };
 165         struct list_head next;
 166         struct tun_struct *detached;
 167 };
 168
 169 struct tun_flow_entry {
 170         struct hlist_node hash_link;
 171         struct rcu_head rcu;
 172         struct tun_struct *tun;
 173
 174         u32 rxhash;
 175         u32 rps_rxhash;
 176         int queue_index;
 177         unsigned long updated;
 178 };
 179
 180 #define TUN_NUM_FLOW_ENTRIES 1024
 181
 182 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 183  * device, socket filter, sndbuf and vnet header size were restore when the
 184  * file were attached to a persist device.
 185  */
 186 struct tun_struct {
 187         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 188         unsigned int            numqueues;
 189         unsigned int            flags;
 190         kuid_t                  owner;
 191         kgid_t                  group;
 192
 193         struct net_device       *dev;
 194         netdev_features_t       set_features;
 195 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 196                           NETIF_F_TSO6|NETIF_F_UFO)
 197
 198         int                     vnet_hdr_sz;
 199         int                     sndbuf;
 200         struct tap_filter       txflt;
 201         struct sock_fprog       fprog;
 202         /* protected by rtnl lock */
 203         bool                    filter_attached;
 204 #ifdef TUN_DEBUG
 205         int debug;
 206 #endif
 207         spinlock_t lock;
 208         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 209         struct timer_list flow_gc_timer;
 210         unsigned long ageing_time;
 211         unsigned int numdisabled;
 212         struct list_head disabled;
 213         void *security;
 214         u32 flow_count;
 215 };
 216
 217 #ifdef CONFIG_TUN_VNET_CROSS_LE
 218 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 219 {
 220         return tun->flags & TUN_VNET_BE ? false :
 221                 virtio_legacy_is_little_endian();
 222 }
 223
 224 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 225 {
 226         int be = !!(tun->flags & TUN_VNET_BE);
 227
 228         if (put_user(be, argp))
 229                 return -EFAULT;
 230
 231         return 0;
 232 }
 233
 234 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 235 {
 236         int be;
 237
 238         if (get_user(be, argp))
 239                 return -EFAULT;
 240
 241         if (be)
 242                 tun->flags |= TUN_VNET_BE;
 243         else
 244                 tun->flags &= ~TUN_VNET_BE;
 245
 246         return 0;
 247 }
 248 #else
 249 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 250 {
 251         return virtio_legacy_is_little_endian();
 252 }
 253
 254 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 255 {
 256         return -EINVAL;
 257 }
 258
 259 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 260 {
 261         return -EINVAL;
 262 }
 263 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 264
 265 static inline bool tun_is_little_endian(struct tun_struct *tun)
 266 {
 267         return tun->flags & TUN_VNET_LE ||
 268                 tun_legacy_is_little_endian(tun);
 269 }
 270
 271 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 272 {
 273         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 274 }
 275
 276 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 277 {
 278         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 279 }
 280
 281 static inline u32 tun_hashfn(u32 rxhash)
 282 {
 283         return rxhash & 0x3ff;
 284 }
 285
 286 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 287 {
 288         struct tun_flow_entry *e;
 289
 290         hlist_for_each_entry_rcu(e, head, hash_link) {
 291                 if (e->rxhash == rxhash)
 292                         return e;
 293         }
 294         return NULL;
 295 }
 296
 297 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 298                                               struct hlist_head *head,
 299                                               u32 rxhash, u16 queue_index)
 300 {
 301         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 302
 303         if (e) {
 304                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 305                           rxhash, queue_index);
 306                 e->updated = jiffies;
 307                 e->rxhash = rxhash;
 308                 e->rps_rxhash = 0;
 309                 e->queue_index = queue_index;
 310                 e->tun = tun;
 311                 hlist_add_head_rcu(&e->hash_link, head);
 312                 ++tun->flow_count;
 313         }
 314         return e;
 315 }
 316
 317 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 318 {
 319         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 320                   e->rxhash, e->queue_index);
 321         hlist_del_rcu(&e->hash_link);
 322         kfree_rcu(e, rcu);
 323         --tun->flow_count;
 324 }
 325
 326 static void tun_flow_flush(struct tun_struct *tun)
 327 {
 328         int i;
 329
 330         spin_lock_bh(&tun->lock);
 331         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 332                 struct tun_flow_entry *e;
 333                 struct hlist_node *n;
 334
 335                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 336                         tun_flow_delete(tun, e);
 337         }
 338         spin_unlock_bh(&tun->lock);
 339 }
 340
 341 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 342 {
 343         int i;
 344
 345         spin_lock_bh(&tun->lock);
 346         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 347                 struct tun_flow_entry *e;
 348                 struct hlist_node *n;
 349
 350                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 351                         if (e->queue_index == queue_index)
 352                                 tun_flow_delete(tun, e);
 353                 }
 354         }
 355         spin_unlock_bh(&tun->lock);
 356 }
 357
 358 static void tun_flow_cleanup(unsigned long data)
 359 {
 360         struct tun_struct *tun = (struct tun_struct *)data;
 361         unsigned long delay = tun->ageing_time;
 362         unsigned long next_timer = jiffies + delay;
 363         unsigned long count = 0;
 364         int i;
 365
 366         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 367
 368         spin_lock_bh(&tun->lock);
 369         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 370                 struct tun_flow_entry *e;
 371                 struct hlist_node *n;
 372
 373                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 374                         unsigned long this_timer;
 375                         count++;
 376                         this_timer = e->updated + delay;
 377                         if (time_before_eq(this_timer, jiffies))
 378                                 tun_flow_delete(tun, e);
 379                         else if (time_before(this_timer, next_timer))
 380                                 next_timer = this_timer;
 381                 }
 382         }
 383
 384         if (count)
 385                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 386         spin_unlock_bh(&tun->lock);
 387 }
 388
 389 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 390                             struct tun_file *tfile)
 391 {
 392         struct hlist_head *head;
 393         struct tun_flow_entry *e;
 394         unsigned long delay = tun->ageing_time;
 395         u16 queue_index = tfile->queue_index;
 396
 397         if (!rxhash)
 398                 return;
 399         else
 400                 head = &tun->flows[tun_hashfn(rxhash)];
 401
 402         rcu_read_lock();
 403
 404         /* We may get a very small possibility of OOO during switching, not
 405          * worth to optimize.*/
 406         if (tun->numqueues == 1 || tfile->detached)
 407                 goto unlock;
 408
 409         e = tun_flow_find(head, rxhash);
 410         if (likely(e)) {
 411                 /* TODO: keep queueing to old queue until it's empty? */
 412                 e->queue_index = queue_index;
 413                 e->updated = jiffies;
 414                 sock_rps_record_flow_hash(e->rps_rxhash);
 415         } else {
 416                 spin_lock_bh(&tun->lock);
 417                 if (!tun_flow_find(head, rxhash) &&
 418                     tun->flow_count < MAX_TAP_FLOWS)
 419                         tun_flow_create(tun, head, rxhash, queue_index);
 420
 421                 if (!timer_pending(&tun->flow_gc_timer))
 422                         mod_timer(&tun->flow_gc_timer,
 423                                   round_jiffies_up(jiffies + delay));
 424                 spin_unlock_bh(&tun->lock);
 425         }
 426
 427 unlock:
 428         rcu_read_unlock();
 429 }
 430
 431 /**
 432  * Save the hash received in the stack receive path and update the
 433  * flow_hash table accordingly.
 434  */
 435 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 436 {
 437         if (unlikely(e->rps_rxhash != hash))
 438                 e->rps_rxhash = hash;
 439 }
 440
 441 /* We try to identify a flow through its rxhash first. The reason that
 442  * we do not check rxq no. is because some cards(e.g 82599), chooses
 443  * the rxq based on the txq where the last packet of the flow comes. As
 444  * the userspace application move between processors, we may get a
 445  * different rxq no. here. If we could not get rxhash, then we would
 446  * hope the rxq no. may help here.
 447  */
 448 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 449                             void *accel_priv, select_queue_fallback_t fallback)
 450 {
 451         struct tun_struct *tun = netdev_priv(dev);
 452         struct tun_flow_entry *e;
 453         u32 txq = 0;
 454         u32 numqueues = 0;
 455
 456         rcu_read_lock();
 457         numqueues = ACCESS_ONCE(tun->numqueues);
 458
 459         txq = skb_get_hash(skb);
 460         if (txq) {
 461                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 462                 if (e) {
 463                         tun_flow_save_rps_rxhash(e, txq);
 464                         txq = e->queue_index;
 465                 } else
 466                         /* use multiply and shift instead of expensive divide */
 467                         txq = ((u64)txq * numqueues) >> 32;
 468         } else if (likely(skb_rx_queue_recorded(skb))) {
 469                 txq = skb_get_rx_queue(skb);
 470                 while (unlikely(txq >= numqueues))
 471                         txq -= numqueues;
 472         }
 473
 474         rcu_read_unlock();
 475         return txq;
 476 }
 477
 478 static inline bool tun_not_capable(struct tun_struct *tun)
 479 {
 480         const struct cred *cred = current_cred();
 481         struct net *net = dev_net(tun->dev);
 482
 483         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 484                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 485                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 486 }
 487
 488 static void tun_set_real_num_queues(struct tun_struct *tun)
 489 {
 490         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 491         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 492 }
 493
 494 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 495 {
 496         tfile->detached = tun;
 497         list_add_tail(&tfile->next, &tun->disabled);
 498         ++tun->numdisabled;
 499 }
 500
 501 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 502 {
 503         struct tun_struct *tun = tfile->detached;
 504
 505         tfile->detached = NULL;
 506         list_del_init(&tfile->next);
 507         --tun->numdisabled;
 508         return tun;
 509 }
 510
 511 static void tun_queue_purge(struct tun_file *tfile)
 512 {
 513         skb_queue_purge(&tfile->sk.sk_receive_queue);
 514         skb_queue_purge(&tfile->sk.sk_error_queue);
 515 }
 516
 517 static void __tun_detach(struct tun_file *tfile, bool clean)
 518 {
 519         struct tun_file *ntfile;
 520         struct tun_struct *tun;
 521
 522         tun = rtnl_dereference(tfile->tun);
 523
 524         if (tun && !tfile->detached) {
 525                 u16 index = tfile->queue_index;
 526                 BUG_ON(index >= tun->numqueues);
 527
 528                 rcu_assign_pointer(tun->tfiles[index],
 529                                    tun->tfiles[tun->numqueues - 1]);
 530                 ntfile = rtnl_dereference(tun->tfiles[index]);
 531                 ntfile->queue_index = index;
 532
 533                 --tun->numqueues;
 534                 if (clean) {
 535                         RCU_INIT_POINTER(tfile->tun, NULL);
 536                         sock_put(&tfile->sk);
 537                 } else
 538                         tun_disable_queue(tun, tfile);
 539
 540                 synchronize_net();
 541                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 542                 /* Drop read queue */
 543                 tun_queue_purge(tfile);
 544                 tun_set_real_num_queues(tun);
 545         } else if (tfile->detached && clean) {
 546                 tun = tun_enable_queue(tfile);
 547                 sock_put(&tfile->sk);
 548         }
 549
 550         if (clean) {
 551                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 552                         netif_carrier_off(tun->dev);
 553
 554                         if (!(tun->flags & IFF_PERSIST) &&
 555                             tun->dev->reg_state == NETREG_REGISTERED)
 556                                 unregister_netdevice(tun->dev);
 557                 }
 558                 sock_put(&tfile->sk);
 559         }
 560 }
 561
 562 static void tun_detach(struct tun_file *tfile, bool clean)
 563 {
 564         rtnl_lock();
 565         __tun_detach(tfile, clean);
 566         rtnl_unlock();
 567 }
 568
 569 static void tun_detach_all(struct net_device *dev)
 570 {
 571         struct tun_struct *tun = netdev_priv(dev);
 572         struct tun_file *tfile, *tmp;
 573         int i, n = tun->numqueues;
 574
 575         for (i = 0; i < n; i++) {
 576                 tfile = rtnl_dereference(tun->tfiles[i]);
 577                 BUG_ON(!tfile);
 578                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 579                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 580                 RCU_INIT_POINTER(tfile->tun, NULL);
 581                 --tun->numqueues;
 582         }
 583         list_for_each_entry(tfile, &tun->disabled, next) {
 584                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 585                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 586                 RCU_INIT_POINTER(tfile->tun, NULL);
 587         }
 588         BUG_ON(tun->numqueues != 0);
 589
 590         synchronize_net();
 591         for (i = 0; i < n; i++) {
 592                 tfile = rtnl_dereference(tun->tfiles[i]);
 593                 /* Drop read queue */
 594                 tun_queue_purge(tfile);
 595                 sock_put(&tfile->sk);
 596         }
 597         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 598                 tun_enable_queue(tfile);
 599                 tun_queue_purge(tfile);
 600                 sock_put(&tfile->sk);
 601         }
 602         BUG_ON(tun->numdisabled != 0);
 603
 604         if (tun->flags & IFF_PERSIST)
 605                 module_put(THIS_MODULE);
 606 }
 607
 608 static int tun_attach(struct tun_struct *tun, struct file *file,
 609                       bool skip_filter, bool publish_tun)
 610 {
 611         struct tun_file *tfile = file->private_data;
 612         int err;
 613
 614         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 615         if (err < 0)
 616                 goto out;
 617
 618         err = -EINVAL;
 619         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 620                 goto out;
 621
 622         err = -EBUSY;
 623         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 624                 goto out;
 625
 626         err = -E2BIG;
 627         if (!tfile->detached &&
 628             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 629                 goto out;
 630
 631         err = 0;
 632
 633         /* Re-attach the filter to persist device */
 634         if (!skip_filter && (tun->filter_attached == true)) {
 635                 err = __sk_attach_filter(&tun->fprog, tfile->socket.sk,
 636                                          lockdep_rtnl_is_held());
 637                 if (!err)
 638                         goto out;
 639         }
 640         tfile->queue_index = tun->numqueues;
 641         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 642         if (publish_tun)
 643                 rcu_assign_pointer(tfile->tun, tun);
 644         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 645         tun->numqueues++;
 646
 647         if (tfile->detached)
 648                 tun_enable_queue(tfile);
 649         else
 650                 sock_hold(&tfile->sk);
 651
 652         tun_set_real_num_queues(tun);
 653
 654         /* device is allowed to go away first, so no need to hold extra
 655          * refcnt.
 656          */
 657
 658 out:
 659         return err;
 660 }
 661
 662 static struct tun_struct *__tun_get(struct tun_file *tfile)
 663 {
 664         struct tun_struct *tun;
 665
 666         rcu_read_lock();
 667         tun = rcu_dereference(tfile->tun);
 668         if (tun)
 669                 dev_hold(tun->dev);
 670         rcu_read_unlock();
 671
 672         return tun;
 673 }
 674
 675 static struct tun_struct *tun_get(struct file *file)
 676 {
 677         return __tun_get(file->private_data);
 678 }
 679
 680 static void tun_put(struct tun_struct *tun)
 681 {
 682         dev_put(tun->dev);
 683 }
 684
 685 /* TAP filtering */
 686 static void addr_hash_set(u32 *mask, const u8 *addr)
 687 {
 688         int n = ether_crc(ETH_ALEN, addr) >> 26;
 689         mask[n >> 5] |= (1 << (n & 31));
 690 }
 691
 692 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 693 {
 694         int n = ether_crc(ETH_ALEN, addr) >> 26;
 695         return mask[n >> 5] & (1 << (n & 31));
 696 }
 697
 698 static int update_filter(struct tap_filter *filter, void __user *arg)
 699 {
 700         struct { u8 u[ETH_ALEN]; } *addr;
 701         struct tun_filter uf;
 702         int err, alen, n, nexact;
 703
 704         if (copy_from_user(&uf, arg, sizeof(uf)))
 705                 return -EFAULT;
 706
 707         if (!uf.count) {
 708                 /* Disabled */
 709                 filter->count = 0;
 710                 return 0;
 711         }
 712
 713         alen = ETH_ALEN * uf.count;
 714         addr = kmalloc(alen, GFP_KERNEL);
 715         if (!addr)
 716                 return -ENOMEM;
 717
 718         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 719                 err = -EFAULT;
 720                 goto done;
 721         }
 722
 723         /* The filter is updated without holding any locks. Which is
 724          * perfectly safe. We disable it first and in the worst
 725          * case we'll accept a few undesired packets. */
 726         filter->count = 0;
 727         wmb();
 728
 729         /* Use first set of addresses as an exact filter */
 730         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 731                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 732
 733         nexact = n;
 734
 735         /* Remaining multicast addresses are hashed,
 736          * unicast will leave the filter disabled. */
 737         memset(filter->mask, 0, sizeof(filter->mask));
 738         for (; n < uf.count; n++) {
 739                 if (!is_multicast_ether_addr(addr[n].u)) {
 740                         err = 0; /* no filter */
 741                         goto done;
 742                 }
 743                 addr_hash_set(filter->mask, addr[n].u);
 744         }
 745
 746         /* For ALLMULTI just set the mask to all ones.
 747          * This overrides the mask populated above. */
 748         if ((uf.flags & TUN_FLT_ALLMULTI))
 749                 memset(filter->mask, ~0, sizeof(filter->mask));
 750
 751         /* Now enable the filter */
 752         wmb();
 753         filter->count = nexact;
 754
 755         /* Return the number of exact filters */
 756         err = nexact;
 757
 758 done:
 759         kfree(addr);
 760         return err;
 761 }
 762
 763 /* Returns: 0 - drop, !=0 - accept */
 764 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 765 {
 766         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 767          * at this point. */
 768         struct ethhdr *eh = (struct ethhdr *) skb->data;
 769         int i;
 770
 771         /* Exact match */
 772         for (i = 0; i < filter->count; i++)
 773                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 774                         return 1;
 775
 776         /* Inexact match (multicast only) */
 777         if (is_multicast_ether_addr(eh->h_dest))
 778                 return addr_hash_test(filter->mask, eh->h_dest);
 779
 780         return 0;
 781 }
 782
 783 /*
 784  * Checks whether the packet is accepted or not.
 785  * Returns: 0 - drop, !=0 - accept
 786  */
 787 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 788 {
 789         if (!filter->count)
 790                 return 1;
 791
 792         return run_filter(filter, skb);
 793 }
 794
 795 /* Network device part of the driver */
 796
 797 static const struct ethtool_ops tun_ethtool_ops;
 798
 799 /* Net device detach from fd. */
 800 static void tun_net_uninit(struct net_device *dev)
 801 {
 802         tun_detach_all(dev);
 803 }
 804
 805 /* Net device open. */
 806 static int tun_net_open(struct net_device *dev)
 807 {
 808         netif_tx_start_all_queues(dev);
 809         return 0;
 810 }
 811
 812 /* Net device close. */
 813 static int tun_net_close(struct net_device *dev)
 814 {
 815         netif_tx_stop_all_queues(dev);
 816         return 0;
 817 }
 818
 819 /* Net device start xmit */
 820 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 821 {
 822         struct tun_struct *tun = netdev_priv(dev);
 823         int txq = skb->queue_mapping;
 824         struct tun_file *tfile;
 825         u32 numqueues = 0;
 826
 827         rcu_read_lock();
 828         tfile = rcu_dereference(tun->tfiles[txq]);
 829         numqueues = ACCESS_ONCE(tun->numqueues);
 830
 831         /* Drop packet if interface is not attached */
 832         if (txq >= numqueues)
 833                 goto drop;
 834
 835         if (numqueues == 1) {
 836                 /* Select queue was not called for the skbuff, so we extract the
 837                  * RPS hash and save it into the flow_table here.
 838                  */
 839                 __u32 rxhash;
 840
 841                 rxhash = skb_get_hash(skb);
 842                 if (rxhash) {
 843                         struct tun_flow_entry *e;
 844                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 845                                         rxhash);
 846                         if (e)
 847                                 tun_flow_save_rps_rxhash(e, rxhash);
 848                 }
 849         }
 850
 851         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 852
 853         BUG_ON(!tfile);
 854
 855         /* Drop if the filter does not like it.
 856          * This is a noop if the filter is disabled.
 857          * Filter can be enabled only for the TAP devices. */
 858         if (!check_filter(&tun->txflt, skb))
 859                 goto drop;
 860
 861         if (tfile->socket.sk->sk_filter &&
 862             sk_filter(tfile->socket.sk, skb))
 863                 goto drop;
 864
 865         /* Limit the number of packets queued by dividing txq length with the
 866          * number of queues.
 867          */
 868         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
 869                           >= dev->tx_queue_len)
 870                 goto drop;
 871
 872         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 873                 goto drop;
 874
 875         if (skb->sk && sk_fullsock(skb->sk)) {
 876                 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
 877                 sw_tx_timestamp(skb);
 878         }
 879
 880         /* Orphan the skb - required as we might hang on to it
 881          * for indefinite time.
 882          */
 883         skb_orphan(skb);
 884
 885         nf_reset(skb);
 886
 887         /* Enqueue packet */
 888         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 889
 890         /* Notify and wake up reader process */
 891         if (tfile->flags & TUN_FASYNC)
 892                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 893         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 894
 895         rcu_read_unlock();
 896         return NETDEV_TX_OK;
 897
 898 drop:
 899         dev->stats.tx_dropped++;
 900         skb_tx_error(skb);
 901         kfree_skb(skb);
 902         rcu_read_unlock();
 903         return NET_XMIT_DROP;
 904 }
 905
 906 static void tun_net_mclist(struct net_device *dev)
 907 {
 908         /*
 909          * This callback is supposed to deal with mc filter in
 910          * _rx_ path and has nothing to do with the _tx_ path.
 911          * In rx path we always accept everything userspace gives us.
 912          */
 913 }
 914
 915 #define MIN_MTU 68
 916 #define MAX_MTU 65535
 917
 918 static int
 919 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 920 {
 921         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 922                 return -EINVAL;
 923         dev->mtu = new_mtu;
 924         return 0;
 925 }
 926
 927 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 928         netdev_features_t features)
 929 {
 930         struct tun_struct *tun = netdev_priv(dev);
 931
 932         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 933 }
 934 #ifdef CONFIG_NET_POLL_CONTROLLER
 935 static void tun_poll_controller(struct net_device *dev)
 936 {
 937         /*
 938          * Tun only receives frames when:
 939          * 1) the char device endpoint gets data from user space
 940          * 2) the tun socket gets a sendmsg call from user space
 941          * Since both of those are synchronous operations, we are guaranteed
 942          * never to have pending data when we poll for it
 943          * so there is nothing to do here but return.
 944          * We need this though so netpoll recognizes us as an interface that
 945          * supports polling, which enables bridge devices in virt setups to
 946          * still use netconsole
 947          */
 948         return;
 949 }
 950 #endif
 951 static const struct net_device_ops tun_netdev_ops = {
 952         .ndo_uninit             = tun_net_uninit,
 953         .ndo_open               = tun_net_open,
 954         .ndo_stop               = tun_net_close,
 955         .ndo_start_xmit         = tun_net_xmit,
 956         .ndo_change_mtu         = tun_net_change_mtu,
 957         .ndo_fix_features       = tun_net_fix_features,
 958         .ndo_select_queue       = tun_select_queue,
 959 #ifdef CONFIG_NET_POLL_CONTROLLER
 960         .ndo_poll_controller    = tun_poll_controller,
 961 #endif
 962 };
 963
 964 static const struct net_device_ops tap_netdev_ops = {
 965         .ndo_uninit             = tun_net_uninit,
 966         .ndo_open               = tun_net_open,
 967         .ndo_stop               = tun_net_close,
 968         .ndo_start_xmit         = tun_net_xmit,
 969         .ndo_change_mtu         = tun_net_change_mtu,
 970         .ndo_fix_features       = tun_net_fix_features,
 971         .ndo_set_rx_mode        = tun_net_mclist,
 972         .ndo_set_mac_address    = eth_mac_addr,
 973         .ndo_validate_addr      = eth_validate_addr,
 974         .ndo_select_queue       = tun_select_queue,
 975 #ifdef CONFIG_NET_POLL_CONTROLLER
 976         .ndo_poll_controller    = tun_poll_controller,
 977 #endif
 978         .ndo_features_check     = passthru_features_check,
 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                 break;
1018
1019         case IFF_TAP:
1020                 dev->netdev_ops = &tap_netdev_ops;
1021                 /* Ethernet TAP Device */
1022                 ether_setup(dev);
1023                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1024                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1025
1026                 eth_hw_addr_random(dev);
1027
1028                 break;
1029         }
1030 }
1031
1032 /* Character device part */
1033
1034 /* Poll */
1035 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1036 {
1037         struct tun_file *tfile = file->private_data;
1038         struct tun_struct *tun = __tun_get(tfile);
1039         struct sock *sk;
1040         unsigned int mask = 0;
1041
1042         if (!tun)
1043                 return POLLERR;
1044
1045         sk = tfile->socket.sk;
1046
1047         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1048
1049         poll_wait(file, sk_sleep(sk), wait);
1050
1051         if (!skb_queue_empty(&sk->sk_receive_queue))
1052                 mask |= POLLIN | POLLRDNORM;
1053
1054         if (sock_writeable(sk) ||
1055             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1056              sock_writeable(sk)))
1057                 mask |= POLLOUT | POLLWRNORM;
1058
1059         if (tun->dev->reg_state != NETREG_REGISTERED)
1060                 mask = POLLERR;
1061
1062         tun_put(tun);
1063         return mask;
1064 }
1065
1066 /* prepad is the amount to reserve at front.  len is length after that.
1067  * linear is a hint as to how much to copy (usually headers). */
1068 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1069                                      size_t prepad, size_t len,
1070                                      size_t linear, int noblock)
1071 {
1072         struct sock *sk = tfile->socket.sk;
1073         struct sk_buff *skb;
1074         int err;
1075
1076         /* Under a page?  Don't bother with paged skb. */
1077         if (prepad + len < PAGE_SIZE || !linear)
1078                 linear = len;
1079
1080         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1081                                    &err, 0);
1082         if (!skb)
1083                 return ERR_PTR(err);
1084
1085         skb_reserve(skb, prepad);
1086         skb_put(skb, linear);
1087         skb->data_len = len - linear;
1088         skb->len += len - linear;
1089
1090         return skb;
1091 }
1092
1093 /* Get packet from user space buffer */
1094 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1095                             void *msg_control, struct iov_iter *from,
1096                             int noblock)
1097 {
1098         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1099         struct sk_buff *skb;
1100         size_t total_len = iov_iter_count(from);
1101         size_t len = total_len, align = NET_SKB_PAD, linear;
1102         struct virtio_net_hdr gso = { 0 };
1103         int good_linear;
1104         int copylen;
1105         bool zerocopy = false;
1106         int err;
1107         u32 rxhash;
1108         ssize_t n;
1109
1110         if (!(tun->flags & IFF_NO_PI)) {
1111                 if (len < sizeof(pi))
1112                         return -EINVAL;
1113                 len -= sizeof(pi);
1114
1115                 n = copy_from_iter(&pi, sizeof(pi), from);
1116                 if (n != sizeof(pi))
1117                         return -EFAULT;
1118         }
1119
1120         if (tun->flags & IFF_VNET_HDR) {
1121                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1122
1123                 if (len < vnet_hdr_sz)
1124                         return -EINVAL;
1125                 len -= vnet_hdr_sz;
1126
1127                 n = copy_from_iter(&gso, sizeof(gso), from);
1128                 if (n != sizeof(gso))
1129                         return -EFAULT;
1130
1131                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1132                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1133                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1134
1135                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1136                         return -EINVAL;
1137                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1138         }
1139
1140         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1141                 align += NET_IP_ALIGN;
1142                 if (unlikely(len < ETH_HLEN ||
1143                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1144                         return -EINVAL;
1145         }
1146
1147         good_linear = SKB_MAX_HEAD(align);
1148
1149         if (msg_control) {
1150                 struct iov_iter i = *from;
1151
1152                 /* There are 256 bytes to be copied in skb, so there is
1153                  * enough room for skb expand head in case it is used.
1154                  * The rest of the buffer is mapped from userspace.
1155                  */
1156                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1157                 if (copylen > good_linear)
1158                         copylen = good_linear;
1159                 linear = copylen;
1160                 iov_iter_advance(&i, copylen);
1161                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1162                         zerocopy = true;
1163         }
1164
1165         if (!zerocopy) {
1166                 copylen = len;
1167                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1168                         linear = good_linear;
1169                 else
1170                         linear = tun16_to_cpu(tun, gso.hdr_len);
1171         }
1172
1173         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1174         if (IS_ERR(skb)) {
1175                 if (PTR_ERR(skb) != -EAGAIN)
1176                         tun->dev->stats.rx_dropped++;
1177                 return PTR_ERR(skb);
1178         }
1179
1180         if (zerocopy)
1181                 err = zerocopy_sg_from_iter(skb, from);
1182         else {
1183                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1184                 if (!err && msg_control) {
1185                         struct ubuf_info *uarg = msg_control;
1186                         uarg->callback(uarg, false);
1187                 }
1188         }
1189
1190         if (err) {
1191                 tun->dev->stats.rx_dropped++;
1192                 kfree_skb(skb);
1193                 return -EFAULT;
1194         }
1195
1196         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1197                 if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start),
1198                                           tun16_to_cpu(tun, gso.csum_offset))) {
1199                         tun->dev->stats.rx_frame_errors++;
1200                         kfree_skb(skb);
1201                         return -EINVAL;
1202                 }
1203         }
1204
1205         switch (tun->flags & TUN_TYPE_MASK) {
1206         case IFF_TUN:
1207                 if (tun->flags & IFF_NO_PI) {
1208                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1209
1210                         switch (ip_version) {
1211                         case 4:
1212                                 pi.proto = htons(ETH_P_IP);
1213                                 break;
1214                         case 6:
1215                                 pi.proto = htons(ETH_P_IPV6);
1216                                 break;
1217                         default:
1218                                 tun->dev->stats.rx_dropped++;
1219                                 kfree_skb(skb);
1220                                 return -EINVAL;
1221                         }
1222                 }
1223
1224                 skb_reset_mac_header(skb);
1225                 skb->protocol = pi.proto;
1226                 skb->dev = tun->dev;
1227                 break;
1228         case IFF_TAP:
1229                 skb->protocol = eth_type_trans(skb, tun->dev);
1230                 break;
1231         }
1232
1233         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1234                 pr_debug("GSO!\n");
1235                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1236                 case VIRTIO_NET_HDR_GSO_TCPV4:
1237                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1238                         break;
1239                 case VIRTIO_NET_HDR_GSO_TCPV6:
1240                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1241                         break;
1242                 case VIRTIO_NET_HDR_GSO_UDP:
1243                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1244                         break;
1245                 default:
1246                         tun->dev->stats.rx_frame_errors++;
1247                         kfree_skb(skb);
1248                         return -EINVAL;
1249                 }
1250
1251                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1252                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1253
1254                 skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size);
1255                 if (skb_shinfo(skb)->gso_size == 0) {
1256                         tun->dev->stats.rx_frame_errors++;
1257                         kfree_skb(skb);
1258                         return -EINVAL;
1259                 }
1260
1261                 /* Header must be checked, and gso_segs computed. */
1262                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1263                 skb_shinfo(skb)->gso_segs = 0;
1264         }
1265
1266         /* copy skb_ubuf_info for callback when skb has no error */
1267         if (zerocopy) {
1268                 skb_shinfo(skb)->destructor_arg = msg_control;
1269                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1270                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1271         }
1272
1273         skb_reset_network_header(skb);
1274         skb_probe_transport_header(skb, 0);
1275
1276         rxhash = skb_get_hash(skb);
1277         netif_rx_ni(skb);
1278
1279         tun->dev->stats.rx_packets++;
1280         tun->dev->stats.rx_bytes += len;
1281
1282         tun_flow_update(tun, rxhash, tfile);
1283         return total_len;
1284 }
1285
1286 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1287 {
1288         struct file *file = iocb->ki_filp;
1289         struct tun_struct *tun = tun_get(file);
1290         struct tun_file *tfile = file->private_data;
1291         ssize_t result;
1292
1293         if (!tun)
1294                 return -EBADFD;
1295
1296         result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1297
1298         tun_put(tun);
1299         return result;
1300 }
1301
1302 /* Put packet to the user space buffer */
1303 static ssize_t tun_put_user(struct tun_struct *tun,
1304                             struct tun_file *tfile,
1305                             struct sk_buff *skb,
1306                             struct iov_iter *iter)
1307 {
1308         struct tun_pi pi = { 0, skb->protocol };
1309         ssize_t total;
1310         int vlan_offset = 0;
1311         int vlan_hlen = 0;
1312         int vnet_hdr_sz = 0;
1313
1314         if (skb_vlan_tag_present(skb))
1315                 vlan_hlen = VLAN_HLEN;
1316
1317         if (tun->flags & IFF_VNET_HDR)
1318                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1319
1320         total = skb->len + vlan_hlen + vnet_hdr_sz;
1321
1322         if (!(tun->flags & IFF_NO_PI)) {
1323                 if (iov_iter_count(iter) < sizeof(pi))
1324                         return -EINVAL;
1325
1326                 total += sizeof(pi);
1327                 if (iov_iter_count(iter) < total) {
1328                         /* Packet will be striped */
1329                         pi.flags |= TUN_PKT_STRIP;
1330                 }
1331
1332                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1333                         return -EFAULT;
1334         }
1335
1336         if (vnet_hdr_sz) {
1337                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1338                 if (iov_iter_count(iter) < vnet_hdr_sz)
1339                         return -EINVAL;
1340
1341                 if (skb_is_gso(skb)) {
1342                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1343
1344                         /* This is a hint as to how much should be linear. */
1345                         gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb));
1346                         gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size);
1347                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1348                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1349                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1350                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1351                         else if (sinfo->gso_type & SKB_GSO_UDP)
1352                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1353                         else {
1354                                 pr_err("unexpected GSO type: "
1355                                        "0x%x, gso_size %d, hdr_len %d\n",
1356                                        sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1357                                        tun16_to_cpu(tun, gso.hdr_len));
1358                                 print_hex_dump(KERN_ERR, "tun: ",
1359                                                DUMP_PREFIX_NONE,
1360                                                16, 1, skb->head,
1361                                                min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1362                                 WARN_ON_ONCE(1);
1363                                 return -EINVAL;
1364                         }
1365                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1366                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1367                 } else
1368                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1369
1370                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1371                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1372                         gso.csum_start = cpu_to_tun16(tun, skb_checksum_start_offset(skb) +
1373                                                       vlan_hlen);
1374                         gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset);
1375                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1376                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1377                 } /* else everything is zero */
1378
1379                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1380                         return -EFAULT;
1381
1382                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1383         }
1384
1385         if (vlan_hlen) {
1386                 int ret;
1387                 struct {
1388                         __be16 h_vlan_proto;
1389                         __be16 h_vlan_TCI;
1390                 } veth;
1391
1392                 veth.h_vlan_proto = skb->vlan_proto;
1393                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1394
1395                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1396
1397                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1398                 if (ret || !iov_iter_count(iter))
1399                         goto done;
1400
1401                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1402                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1403                         goto done;
1404         }
1405
1406         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1407
1408 done:
1409         tun->dev->stats.tx_packets++;
1410         tun->dev->stats.tx_bytes += skb->len + vlan_hlen;
1411
1412         return total;
1413 }
1414
1415 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1416                            struct iov_iter *to,
1417                            int noblock)
1418 {
1419         struct sk_buff *skb;
1420         ssize_t ret;
1421         int peeked, err, off = 0;
1422
1423         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1424
1425         if (!iov_iter_count(to))
1426                 return 0;
1427
1428         /* Read frames from queue */
1429         skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1430                                   &peeked, &off, &err);
1431         if (!skb)
1432                 return err;
1433
1434         ret = tun_put_user(tun, tfile, skb, to);
1435         if (unlikely(ret < 0))
1436                 kfree_skb(skb);
1437         else
1438                 consume_skb(skb);
1439
1440         return ret;
1441 }
1442
1443 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1444 {
1445         struct file *file = iocb->ki_filp;
1446         struct tun_file *tfile = file->private_data;
1447         struct tun_struct *tun = __tun_get(tfile);
1448         ssize_t len = iov_iter_count(to), ret;
1449
1450         if (!tun)
1451                 return -EBADFD;
1452         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1453         ret = min_t(ssize_t, ret, len);
1454         if (ret > 0)
1455                 iocb->ki_pos = ret;
1456         tun_put(tun);
1457         return ret;
1458 }
1459
1460 static void tun_free_netdev(struct net_device *dev)
1461 {
1462         struct tun_struct *tun = netdev_priv(dev);
1463
1464         BUG_ON(!(list_empty(&tun->disabled)));
1465         tun_flow_uninit(tun);
1466         security_tun_dev_free_security(tun->security);
1467         free_netdev(dev);
1468 }
1469
1470 static void tun_setup(struct net_device *dev)
1471 {
1472         struct tun_struct *tun = netdev_priv(dev);
1473
1474         tun->owner = INVALID_UID;
1475         tun->group = INVALID_GID;
1476
1477         dev->ethtool_ops = &tun_ethtool_ops;
1478         dev->destructor = tun_free_netdev;
1479         /* We prefer our own queue length */
1480         dev->tx_queue_len = TUN_READQ_SIZE;
1481 }
1482
1483 /* Trivial set of netlink ops to allow deleting tun or tap
1484  * device with netlink.
1485  */
1486 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1487 {
1488         /* NL_SET_ERR_MSG(extack,
1489                        "tun/tap creation via rtnetlink is not supported."); */
1490         return -EOPNOTSUPP;
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, true);
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                 err = dev_get_valid_name(net, dev, name);
1700                 if (err < 0)
1701                         goto err_free_dev;
1702
1703                 dev_net_set(dev, net);
1704                 dev->rtnl_link_ops = &tun_link_ops;
1705                 dev->ifindex = tfile->ifindex;
1706                 dev->sysfs_groups[0] = &tun_attr_group;
1707
1708                 tun = netdev_priv(dev);
1709                 tun->dev = dev;
1710                 tun->flags = flags;
1711                 tun->txflt.count = 0;
1712                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1713
1714                 tun->filter_attached = false;
1715                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1716
1717                 spin_lock_init(&tun->lock);
1718
1719                 err = security_tun_dev_alloc_security(&tun->security);
1720                 if (err < 0)
1721                         goto err_free_dev;
1722
1723                 tun_net_init(dev);
1724                 tun_flow_init(tun);
1725
1726                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1727                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1728                                    NETIF_F_HW_VLAN_STAG_TX;
1729                 dev->features = dev->hw_features;
1730                 dev->vlan_features = dev->features &
1731                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1732                                        NETIF_F_HW_VLAN_STAG_TX);
1733
1734                 INIT_LIST_HEAD(&tun->disabled);
1735                 err = tun_attach(tun, file, false, false);
1736                 if (err < 0)
1737                         goto err_free_flow;
1738
1739                 err = register_netdevice(tun->dev);
1740                 if (err < 0)
1741                         goto err_detach;
1742                 /* free_netdev() won't check refcnt, to aovid race
1743                  * with dev_put() we need publish tun after registration.
1744                  */
1745                 rcu_assign_pointer(tfile->tun, tun);
1746         }
1747
1748         netif_carrier_on(tun->dev);
1749
1750         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1751
1752         tun->flags = (tun->flags & ~TUN_FEATURES) |
1753                 (ifr->ifr_flags & TUN_FEATURES);
1754
1755         /* Make sure persistent devices do not get stuck in
1756          * xoff state.
1757          */
1758         if (netif_running(tun->dev))
1759                 netif_tx_wake_all_queues(tun->dev);
1760
1761         strcpy(ifr->ifr_name, tun->dev->name);
1762         return 0;
1763
1764 err_detach:
1765         tun_detach_all(dev);
1766 err_free_flow:
1767         tun_flow_uninit(tun);
1768         security_tun_dev_free_security(tun->security);
1769 err_free_dev:
1770         free_netdev(dev);
1771         return err;
1772 }
1773
1774 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1775                        struct ifreq *ifr)
1776 {
1777         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1778
1779         strcpy(ifr->ifr_name, tun->dev->name);
1780
1781         ifr->ifr_flags = tun_flags(tun);
1782
1783 }
1784
1785 /* This is like a cut-down ethtool ops, except done via tun fd so no
1786  * privs required. */
1787 static int set_offload(struct tun_struct *tun, unsigned long arg)
1788 {
1789         netdev_features_t features = 0;
1790
1791         if (arg & TUN_F_CSUM) {
1792                 features |= NETIF_F_HW_CSUM;
1793                 arg &= ~TUN_F_CSUM;
1794
1795                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1796                         if (arg & TUN_F_TSO_ECN) {
1797                                 features |= NETIF_F_TSO_ECN;
1798                                 arg &= ~TUN_F_TSO_ECN;
1799                         }
1800                         if (arg & TUN_F_TSO4)
1801                                 features |= NETIF_F_TSO;
1802                         if (arg & TUN_F_TSO6)
1803                                 features |= NETIF_F_TSO6;
1804                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1805                 }
1806
1807                 if (arg & TUN_F_UFO) {
1808                         features |= NETIF_F_UFO;
1809                         arg &= ~TUN_F_UFO;
1810                 }
1811         }
1812
1813         /* This gives the user a way to test for new features in future by
1814          * trying to set them. */
1815         if (arg)
1816                 return -EINVAL;
1817
1818         tun->set_features = features;
1819         netdev_update_features(tun->dev);
1820
1821         return 0;
1822 }
1823
1824 static void tun_detach_filter(struct tun_struct *tun, int n)
1825 {
1826         int i;
1827         struct tun_file *tfile;
1828
1829         for (i = 0; i < n; i++) {
1830                 tfile = rtnl_dereference(tun->tfiles[i]);
1831                 __sk_detach_filter(tfile->socket.sk, lockdep_rtnl_is_held());
1832         }
1833
1834         tun->filter_attached = false;
1835 }
1836
1837 static int tun_attach_filter(struct tun_struct *tun)
1838 {
1839         int i, ret = 0;
1840         struct tun_file *tfile;
1841
1842         for (i = 0; i < tun->numqueues; i++) {
1843                 tfile = rtnl_dereference(tun->tfiles[i]);
1844                 ret = __sk_attach_filter(&tun->fprog, tfile->socket.sk,
1845                                          lockdep_rtnl_is_held());
1846                 if (ret) {
1847                         tun_detach_filter(tun, i);
1848                         return ret;
1849                 }
1850         }
1851
1852         tun->filter_attached = true;
1853         return ret;
1854 }
1855
1856 static void tun_set_sndbuf(struct tun_struct *tun)
1857 {
1858         struct tun_file *tfile;
1859         int i;
1860
1861         for (i = 0; i < tun->numqueues; i++) {
1862                 tfile = rtnl_dereference(tun->tfiles[i]);
1863                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1864         }
1865 }
1866
1867 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1868 {
1869         struct tun_file *tfile = file->private_data;
1870         struct tun_struct *tun;
1871         int ret = 0;
1872
1873         rtnl_lock();
1874
1875         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1876                 tun = tfile->detached;
1877                 if (!tun) {
1878                         ret = -EINVAL;
1879                         goto unlock;
1880                 }
1881                 ret = security_tun_dev_attach_queue(tun->security);
1882                 if (ret < 0)
1883                         goto unlock;
1884                 ret = tun_attach(tun, file, false, true);
1885         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1886                 tun = rtnl_dereference(tfile->tun);
1887                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1888                         ret = -EINVAL;
1889                 else
1890                         __tun_detach(tfile, false);
1891         } else
1892                 ret = -EINVAL;
1893
1894 unlock:
1895         rtnl_unlock();
1896         return ret;
1897 }
1898
1899 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
1900 static unsigned char tun_get_addr_len(unsigned short type)
1901 {
1902         switch (type) {
1903         case ARPHRD_IP6GRE:
1904         case ARPHRD_TUNNEL6:
1905                 return sizeof(struct in6_addr);
1906         case ARPHRD_IPGRE:
1907         case ARPHRD_TUNNEL:
1908         case ARPHRD_SIT:
1909                 return 4;
1910         case ARPHRD_ETHER:
1911                 return ETH_ALEN;
1912         case ARPHRD_IEEE802154:
1913         case ARPHRD_IEEE802154_MONITOR:
1914                 return IEEE802154_EXTENDED_ADDR_LEN;
1915         case ARPHRD_PHONET_PIPE:
1916         case ARPHRD_PPP:
1917         case ARPHRD_NONE:
1918                 return 0;
1919         case ARPHRD_6LOWPAN:
1920                 return EUI64_ADDR_LEN;
1921         case ARPHRD_FDDI:
1922                 return FDDI_K_ALEN;
1923         case ARPHRD_HIPPI:
1924                 return HIPPI_ALEN;
1925         case ARPHRD_IEEE802:
1926                 return FC_ALEN;
1927         case ARPHRD_ROSE:
1928                 return ROSE_ADDR_LEN;
1929         case ARPHRD_NETROM:
1930                 return AX25_ADDR_LEN;
1931         case ARPHRD_LOCALTLK:
1932                 return LTALK_ALEN;
1933         default:
1934                 return 0;
1935         }
1936 }
1937
1938 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1939                             unsigned long arg, int ifreq_len)
1940 {
1941         struct tun_file *tfile = file->private_data;
1942         struct tun_struct *tun;
1943         void __user* argp = (void __user*)arg;
1944         struct ifreq ifr;
1945         kuid_t owner;
1946         kgid_t group;
1947         int sndbuf;
1948         int vnet_hdr_sz;
1949         unsigned int ifindex;
1950         int le;
1951         int ret;
1952
1953         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1954                 if (copy_from_user(&ifr, argp, ifreq_len))
1955                         return -EFAULT;
1956         } else {
1957                 memset(&ifr, 0, sizeof(ifr));
1958         }
1959         if (cmd == TUNGETFEATURES) {
1960                 /* Currently this just means: "what IFF flags are valid?".
1961                  * This is needed because we never checked for invalid flags on
1962                  * TUNSETIFF.
1963                  */
1964                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1965                                 (unsigned int __user*)argp);
1966         } else if (cmd == TUNSETQUEUE)
1967                 return tun_set_queue(file, &ifr);
1968
1969         ret = 0;
1970         rtnl_lock();
1971
1972         tun = __tun_get(tfile);
1973         if (cmd == TUNSETIFF && !tun) {
1974                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1975
1976                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
1977
1978                 if (ret)
1979                         goto unlock;
1980
1981                 if (copy_to_user(argp, &ifr, ifreq_len))
1982                         ret = -EFAULT;
1983                 goto unlock;
1984         }
1985         if (cmd == TUNSETIFINDEX) {
1986                 ret = -EPERM;
1987                 if (tun)
1988                         goto unlock;
1989
1990                 ret = -EFAULT;
1991                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1992                         goto unlock;
1993
1994                 ret = 0;
1995                 tfile->ifindex = ifindex;
1996                 goto unlock;
1997         }
1998
1999         ret = -EBADFD;
2000         if (!tun)
2001                 goto unlock;
2002
2003         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2004
2005         ret = 0;
2006         switch (cmd) {
2007         case TUNGETIFF:
2008                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2009
2010                 if (tfile->detached)
2011                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2012                 if (!tfile->socket.sk->sk_filter)
2013                         ifr.ifr_flags |= IFF_NOFILTER;
2014
2015                 if (copy_to_user(argp, &ifr, ifreq_len))
2016                         ret = -EFAULT;
2017                 break;
2018
2019         case TUNSETNOCSUM:
2020                 /* Disable/Enable checksum */
2021
2022                 /* [unimplemented] */
2023                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2024                           arg ? "disabled" : "enabled");
2025                 break;
2026
2027         case TUNSETPERSIST:
2028                 /* Disable/Enable persist mode. Keep an extra reference to the
2029                  * module to prevent the module being unprobed.
2030                  */
2031                 if (arg && !(tun->flags & IFF_PERSIST)) {
2032                         tun->flags |= IFF_PERSIST;
2033                         __module_get(THIS_MODULE);
2034                 }
2035                 if (!arg && (tun->flags & IFF_PERSIST)) {
2036                         tun->flags &= ~IFF_PERSIST;
2037                         module_put(THIS_MODULE);
2038                 }
2039
2040                 tun_debug(KERN_INFO, tun, "persist %s\n",
2041                           arg ? "enabled" : "disabled");
2042                 break;
2043
2044         case TUNSETOWNER:
2045                 /* Set owner of the device */
2046                 owner = make_kuid(current_user_ns(), arg);
2047                 if (!uid_valid(owner)) {
2048                         ret = -EINVAL;
2049                         break;
2050                 }
2051                 tun->owner = owner;
2052                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2053                           from_kuid(&init_user_ns, tun->owner));
2054                 break;
2055
2056         case TUNSETGROUP:
2057                 /* Set group of the device */
2058                 group = make_kgid(current_user_ns(), arg);
2059                 if (!gid_valid(group)) {
2060                         ret = -EINVAL;
2061                         break;
2062                 }
2063                 tun->group = group;
2064                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2065                           from_kgid(&init_user_ns, tun->group));
2066                 break;
2067
2068         case TUNSETLINK:
2069                 /* Only allow setting the type when the interface is down */
2070                 if (tun->dev->flags & IFF_UP) {
2071                         tun_debug(KERN_INFO, tun,
2072                                   "Linktype set failed because interface is up\n");
2073                         ret = -EBUSY;
2074                 } else {
2075                         tun->dev->type = (int) arg;
2076                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
2077                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2078                                   tun->dev->type);
2079                         ret = 0;
2080                 }
2081                 break;
2082
2083 #ifdef TUN_DEBUG
2084         case TUNSETDEBUG:
2085                 tun->debug = arg;
2086                 break;
2087 #endif
2088         case TUNSETOFFLOAD:
2089                 ret = set_offload(tun, arg);
2090                 break;
2091
2092         case TUNSETTXFILTER:
2093                 /* Can be set only for TAPs */
2094                 ret = -EINVAL;
2095                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2096                         break;
2097                 ret = update_filter(&tun->txflt, (void __user *)arg);
2098                 break;
2099
2100         case SIOCGIFHWADDR:
2101                 /* Get hw address */
2102                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2103                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2104                 if (copy_to_user(argp, &ifr, ifreq_len))
2105                         ret = -EFAULT;
2106                 break;
2107
2108         case SIOCSIFHWADDR:
2109                 /* Set hw address */
2110                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2111                           ifr.ifr_hwaddr.sa_data);
2112
2113                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2114                 break;
2115
2116         case TUNGETSNDBUF:
2117                 sndbuf = tfile->socket.sk->sk_sndbuf;
2118                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2119                         ret = -EFAULT;
2120                 break;
2121
2122         case TUNSETSNDBUF:
2123                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2124                         ret = -EFAULT;
2125                         break;
2126                 }
2127                 if (sndbuf <= 0) {
2128                         ret = -EINVAL;
2129                         break;
2130                 }
2131
2132                 tun->sndbuf = sndbuf;
2133                 tun_set_sndbuf(tun);
2134                 break;
2135
2136         case TUNGETVNETHDRSZ:
2137                 vnet_hdr_sz = tun->vnet_hdr_sz;
2138                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2139                         ret = -EFAULT;
2140                 break;
2141
2142         case TUNSETVNETHDRSZ:
2143                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2144                         ret = -EFAULT;
2145                         break;
2146                 }
2147                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2148                         ret = -EINVAL;
2149                         break;
2150                 }
2151
2152                 tun->vnet_hdr_sz = vnet_hdr_sz;
2153                 break;
2154
2155         case TUNGETVNETLE:
2156                 le = !!(tun->flags & TUN_VNET_LE);
2157                 if (put_user(le, (int __user *)argp))
2158                         ret = -EFAULT;
2159                 break;
2160
2161         case TUNSETVNETLE:
2162                 if (get_user(le, (int __user *)argp)) {
2163                         ret = -EFAULT;
2164                         break;
2165                 }
2166                 if (le)
2167                         tun->flags |= TUN_VNET_LE;
2168                 else
2169                         tun->flags &= ~TUN_VNET_LE;
2170                 break;
2171
2172         case TUNGETVNETBE:
2173                 ret = tun_get_vnet_be(tun, argp);
2174                 break;
2175
2176         case TUNSETVNETBE:
2177                 ret = tun_set_vnet_be(tun, argp);
2178                 break;
2179
2180         case TUNATTACHFILTER:
2181                 /* Can be set only for TAPs */
2182                 ret = -EINVAL;
2183                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2184                         break;
2185                 ret = -EFAULT;
2186                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2187                         break;
2188
2189                 ret = tun_attach_filter(tun);
2190                 break;
2191
2192         case TUNDETACHFILTER:
2193                 /* Can be set only for TAPs */
2194                 ret = -EINVAL;
2195                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2196                         break;
2197                 ret = 0;
2198                 tun_detach_filter(tun, tun->numqueues);
2199                 break;
2200
2201         case TUNGETFILTER:
2202                 ret = -EINVAL;
2203                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2204                         break;
2205                 ret = -EFAULT;
2206                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2207                         break;
2208                 ret = 0;
2209                 break;
2210
2211         default:
2212                 ret = -EINVAL;
2213                 break;
2214         }
2215
2216 unlock:
2217         rtnl_unlock();
2218         if (tun)
2219                 tun_put(tun);
2220         return ret;
2221 }
2222
2223 static long tun_chr_ioctl(struct file *file,
2224                           unsigned int cmd, unsigned long arg)
2225 {
2226         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2227 }
2228
2229 #ifdef CONFIG_COMPAT
2230 static long tun_chr_compat_ioctl(struct file *file,
2231                          unsigned int cmd, unsigned long arg)
2232 {
2233         switch (cmd) {
2234         case TUNSETIFF:
2235         case TUNGETIFF:
2236         case TUNSETTXFILTER:
2237         case TUNGETSNDBUF:
2238         case TUNSETSNDBUF:
2239         case SIOCGIFHWADDR:
2240         case SIOCSIFHWADDR:
2241                 arg = (unsigned long)compat_ptr(arg);
2242                 break;
2243         default:
2244                 arg = (compat_ulong_t)arg;
2245                 break;
2246         }
2247
2248         /*
2249          * compat_ifreq is shorter than ifreq, so we must not access beyond
2250          * the end of that structure. All fields that are used in this
2251          * driver are compatible though, we don't need to convert the
2252          * contents.
2253          */
2254         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2255 }
2256 #endif /* CONFIG_COMPAT */
2257
2258 static int tun_chr_fasync(int fd, struct file *file, int on)
2259 {
2260         struct tun_file *tfile = file->private_data;
2261         int ret;
2262
2263         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2264                 goto out;
2265
2266         if (on) {
2267                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2268                 tfile->flags |= TUN_FASYNC;
2269         } else
2270                 tfile->flags &= ~TUN_FASYNC;
2271         ret = 0;
2272 out:
2273         return ret;
2274 }
2275
2276 static int tun_chr_open(struct inode *inode, struct file * file)
2277 {
2278         struct net *net = current->nsproxy->net_ns;
2279         struct tun_file *tfile;
2280
2281         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2282
2283         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2284                                             &tun_proto, 0);
2285         if (!tfile)
2286                 return -ENOMEM;
2287         RCU_INIT_POINTER(tfile->tun, NULL);
2288         tfile->flags = 0;
2289         tfile->ifindex = 0;
2290
2291         init_waitqueue_head(&tfile->wq.wait);
2292         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2293
2294         tfile->socket.file = file;
2295         tfile->socket.ops = &tun_socket_ops;
2296
2297         sock_init_data(&tfile->socket, &tfile->sk);
2298
2299         tfile->sk.sk_write_space = tun_sock_write_space;
2300         tfile->sk.sk_sndbuf = INT_MAX;
2301
2302         file->private_data = tfile;
2303         INIT_LIST_HEAD(&tfile->next);
2304
2305         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2306
2307         return 0;
2308 }
2309
2310 static int tun_chr_close(struct inode *inode, struct file *file)
2311 {
2312         struct tun_file *tfile = file->private_data;
2313
2314         tun_detach(tfile, true);
2315
2316         return 0;
2317 }
2318
2319 #ifdef CONFIG_PROC_FS
2320 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2321 {
2322         struct tun_struct *tun;
2323         struct ifreq ifr;
2324
2325         memset(&ifr, 0, sizeof(ifr));
2326
2327         rtnl_lock();
2328         tun = tun_get(f);
2329         if (tun)
2330                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2331         rtnl_unlock();
2332
2333         if (tun)
2334                 tun_put(tun);
2335
2336         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2337 }
2338 #endif
2339
2340 static const struct file_operations tun_fops = {
2341         .owner  = THIS_MODULE,
2342         .llseek = no_llseek,
2343         .read_iter  = tun_chr_read_iter,
2344         .write_iter = tun_chr_write_iter,
2345         .poll   = tun_chr_poll,
2346         .unlocked_ioctl = tun_chr_ioctl,
2347 #ifdef CONFIG_COMPAT
2348         .compat_ioctl = tun_chr_compat_ioctl,
2349 #endif
2350         .open   = tun_chr_open,
2351         .release = tun_chr_close,
2352         .fasync = tun_chr_fasync,
2353 #ifdef CONFIG_PROC_FS
2354         .show_fdinfo = tun_chr_show_fdinfo,
2355 #endif
2356 };
2357
2358 static struct miscdevice tun_miscdev = {
2359         .minor = TUN_MINOR,
2360         .name = "tun",
2361         .nodename = "net/tun",
2362         .fops = &tun_fops,
2363 };
2364
2365 /* ethtool interface */
2366
2367 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2368 {
2369         cmd->supported          = 0;
2370         cmd->advertising        = 0;
2371         ethtool_cmd_speed_set(cmd, SPEED_10);
2372         cmd->duplex             = DUPLEX_FULL;
2373         cmd->port               = PORT_TP;
2374         cmd->phy_address        = 0;
2375         cmd->transceiver        = XCVR_INTERNAL;
2376         cmd->autoneg            = AUTONEG_DISABLE;
2377         cmd->maxtxpkt           = 0;
2378         cmd->maxrxpkt           = 0;
2379         return 0;
2380 }
2381
2382 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2383 {
2384         struct tun_struct *tun = netdev_priv(dev);
2385
2386         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2387         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2388
2389         switch (tun->flags & TUN_TYPE_MASK) {
2390         case IFF_TUN:
2391                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2392                 break;
2393         case IFF_TAP:
2394                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2395                 break;
2396         }
2397 }
2398
2399 static u32 tun_get_msglevel(struct net_device *dev)
2400 {
2401 #ifdef TUN_DEBUG
2402         struct tun_struct *tun = netdev_priv(dev);
2403         return tun->debug;
2404 #else
2405         return -EOPNOTSUPP;
2406 #endif
2407 }
2408
2409 static void tun_set_msglevel(struct net_device *dev, u32 value)
2410 {
2411 #ifdef TUN_DEBUG
2412         struct tun_struct *tun = netdev_priv(dev);
2413         tun->debug = value;
2414 #endif
2415 }
2416
2417 static const struct ethtool_ops tun_ethtool_ops = {
2418         .get_settings   = tun_get_settings,
2419         .get_drvinfo    = tun_get_drvinfo,
2420         .get_msglevel   = tun_get_msglevel,
2421         .set_msglevel   = tun_set_msglevel,
2422         .get_link       = ethtool_op_get_link,
2423         .get_ts_info    = ethtool_op_get_ts_info,
2424 };
2425
2426
2427 static int __init tun_init(void)
2428 {
2429         int ret = 0;
2430
2431         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2432         pr_info("%s\n", DRV_COPYRIGHT);
2433
2434         ret = rtnl_link_register(&tun_link_ops);
2435         if (ret) {
2436                 pr_err("Can't register link_ops\n");
2437                 goto err_linkops;
2438         }
2439
2440         ret = misc_register(&tun_miscdev);
2441         if (ret) {
2442                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2443                 goto err_misc;
2444         }
2445         return  0;
2446 err_misc:
2447         rtnl_link_unregister(&tun_link_ops);
2448 err_linkops:
2449         return ret;
2450 }
2451
2452 static void tun_cleanup(void)
2453 {
2454         misc_deregister(&tun_miscdev);
2455         rtnl_link_unregister(&tun_link_ops);
2456 }
2457
2458 /* Get an underlying socket object from tun file.  Returns error unless file is
2459  * attached to a device.  The returned object works like a packet socket, it
2460  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2461  * holding a reference to the file for as long as the socket is in use. */
2462 struct socket *tun_get_socket(struct file *file)
2463 {
2464         struct tun_file *tfile;
2465         if (file->f_op != &tun_fops)
2466                 return ERR_PTR(-EINVAL);
2467         tfile = file->private_data;
2468         if (!tfile)
2469                 return ERR_PTR(-EBADFD);
2470         return &tfile->socket;
2471 }
2472 EXPORT_SYMBOL_GPL(tun_get_socket);
2473
2474 module_init(tun_init);
2475 module_exit(tun_cleanup);
2476 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2477 MODULE_AUTHOR(DRV_COPYRIGHT);
2478 MODULE_LICENSE("GPL");
2479 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2480 MODULE_ALIAS("devname:net/tun");