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/skb_array.h>
  75
  76 #include <linux/uaccess.h>
  77
  78 /* Uncomment to enable debugging */
  79 /* #define TUN_DEBUG 1 */
  80
  81 #ifdef TUN_DEBUG
  82 static int debug;
  83
  84 #define tun_debug(level, tun, fmt, args...)                     \
  85 do {                                                            \
  86         if (tun->debug)                                         \
  87                 netdev_printk(level, tun->dev, fmt, ##args);    \
  88 } while (0)
  89 #define DBG1(level, fmt, args...)                               \
  90 do {                                                            \
  91         if (debug == 2)                                         \
  92                 printk(level fmt, ##args);                      \
  93 } while (0)
  94 #else
  95 #define tun_debug(level, tun, fmt, args...)                     \
  96 do {                                                            \
  97         if (0)                                                  \
  98                 netdev_printk(level, tun->dev, fmt, ##args);    \
  99 } while (0)
 100 #define DBG1(level, fmt, args...)                               \
 101 do {                                                            \
 102         if (0)                                                  \
 103                 printk(level fmt, ##args);                      \
 104 } while (0)
 105 #endif
 106
 107 /* TUN device flags */
 108
 109 /* IFF_ATTACH_QUEUE is never stored in device flags,
 110  * overload it to mean fasync when stored there.
 111  */
 112 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 113 /* High bits in flags field are unused. */
 114 #define TUN_VNET_LE     0x80000000
 115 #define TUN_VNET_BE     0x40000000
 116
 117 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 118                       IFF_MULTI_QUEUE)
 119 #define GOODCOPY_LEN 128
 120
 121 #define FLT_EXACT_COUNT 8
 122 struct tap_filter {
 123         unsigned int    count;    /* Number of addrs. Zero means disabled */
 124         u32             mask[2];  /* Mask of the hashed addrs */
 125         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 126 };
 127
 128 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 129  * to max number of VCPUs in guest. */
 130 #define MAX_TAP_QUEUES 256
 131 #define MAX_TAP_FLOWS  4096
 132
 133 #define TUN_FLOW_EXPIRE (3 * HZ)
 134
 135 struct tun_pcpu_stats {
 136         u64 rx_packets;
 137         u64 rx_bytes;
 138         u64 tx_packets;
 139         u64 tx_bytes;
 140         struct u64_stats_sync syncp;
 141         u32 rx_dropped;
 142         u32 tx_dropped;
 143         u32 rx_frame_errors;
 144 };
 145
 146 /* A tun_file connects an open character device to a tuntap netdevice. It
 147  * also contains all socket related structures (except sock_fprog and tap_filter)
 148  * to serve as one transmit queue for tuntap device. The sock_fprog and
 149  * tap_filter were kept in tun_struct since they were used for filtering for the
 150  * netdevice not for a specific queue (at least I didn't see the requirement for
 151  * this).
 152  *
 153  * RCU usage:
 154  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 155  * other can only be read while rcu_read_lock or rtnl_lock is held.
 156  */
 157 struct tun_file {
 158         struct sock sk;
 159         struct socket socket;
 160         struct socket_wq wq;
 161         struct tun_struct __rcu *tun;
 162         struct fasync_struct *fasync;
 163         /* only used for fasnyc */
 164         unsigned int flags;
 165         union {
 166                 u16 queue_index;
 167                 unsigned int ifindex;
 168         };
 169         struct list_head next;
 170         struct tun_struct *detached;
 171         struct skb_array tx_array;
 172 };
 173
 174 struct tun_flow_entry {
 175         struct hlist_node hash_link;
 176         struct rcu_head rcu;
 177         struct tun_struct *tun;
 178
 179         u32 rxhash;
 180         u32 rps_rxhash;
 181         int queue_index;
 182         unsigned long updated;
 183 };
 184
 185 #define TUN_NUM_FLOW_ENTRIES 1024
 186
 187 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 188  * device, socket filter, sndbuf and vnet header size were restore when the
 189  * file were attached to a persist device.
 190  */
 191 struct tun_struct {
 192         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 193         unsigned int            numqueues;
 194         unsigned int            flags;
 195         kuid_t                  owner;
 196         kgid_t                  group;
 197
 198         struct net_device       *dev;
 199         netdev_features_t       set_features;
 200 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 201                           NETIF_F_TSO6|NETIF_F_UFO)
 202
 203         int                     align;
 204         int                     vnet_hdr_sz;
 205         int                     sndbuf;
 206         struct tap_filter       txflt;
 207         struct sock_fprog       fprog;
 208         /* protected by rtnl lock */
 209         bool                    filter_attached;
 210 #ifdef TUN_DEBUG
 211         int debug;
 212 #endif
 213         spinlock_t lock;
 214         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 215         struct timer_list flow_gc_timer;
 216         unsigned long ageing_time;
 217         unsigned int numdisabled;
 218         struct list_head disabled;
 219         void *security;
 220         u32 flow_count;
 221         struct tun_pcpu_stats __percpu *pcpu_stats;
 222 };
 223
 224 #ifdef CONFIG_TUN_VNET_CROSS_LE
 225 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 226 {
 227         return tun->flags & TUN_VNET_BE ? false :
 228                 virtio_legacy_is_little_endian();
 229 }
 230
 231 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 232 {
 233         int be = !!(tun->flags & TUN_VNET_BE);
 234
 235         if (put_user(be, argp))
 236                 return -EFAULT;
 237
 238         return 0;
 239 }
 240
 241 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 242 {
 243         int be;
 244
 245         if (get_user(be, argp))
 246                 return -EFAULT;
 247
 248         if (be)
 249                 tun->flags |= TUN_VNET_BE;
 250         else
 251                 tun->flags &= ~TUN_VNET_BE;
 252
 253         return 0;
 254 }
 255 #else
 256 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 257 {
 258         return virtio_legacy_is_little_endian();
 259 }
 260
 261 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 262 {
 263         return -EINVAL;
 264 }
 265
 266 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 267 {
 268         return -EINVAL;
 269 }
 270 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 271
 272 static inline bool tun_is_little_endian(struct tun_struct *tun)
 273 {
 274         return tun->flags & TUN_VNET_LE ||
 275                 tun_legacy_is_little_endian(tun);
 276 }
 277
 278 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 279 {
 280         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 281 }
 282
 283 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 284 {
 285         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 286 }
 287
 288 static inline u32 tun_hashfn(u32 rxhash)
 289 {
 290         return rxhash & 0x3ff;
 291 }
 292
 293 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 294 {
 295         struct tun_flow_entry *e;
 296
 297         hlist_for_each_entry_rcu(e, head, hash_link) {
 298                 if (e->rxhash == rxhash)
 299                         return e;
 300         }
 301         return NULL;
 302 }
 303
 304 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 305                                               struct hlist_head *head,
 306                                               u32 rxhash, u16 queue_index)
 307 {
 308         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 309
 310         if (e) {
 311                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 312                           rxhash, queue_index);
 313                 e->updated = jiffies;
 314                 e->rxhash = rxhash;
 315                 e->rps_rxhash = 0;
 316                 e->queue_index = queue_index;
 317                 e->tun = tun;
 318                 hlist_add_head_rcu(&e->hash_link, head);
 319                 ++tun->flow_count;
 320         }
 321         return e;
 322 }
 323
 324 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 325 {
 326         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 327                   e->rxhash, e->queue_index);
 328         hlist_del_rcu(&e->hash_link);
 329         kfree_rcu(e, rcu);
 330         --tun->flow_count;
 331 }
 332
 333 static void tun_flow_flush(struct tun_struct *tun)
 334 {
 335         int i;
 336
 337         spin_lock_bh(&tun->lock);
 338         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 339                 struct tun_flow_entry *e;
 340                 struct hlist_node *n;
 341
 342                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 343                         tun_flow_delete(tun, e);
 344         }
 345         spin_unlock_bh(&tun->lock);
 346 }
 347
 348 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 349 {
 350         int i;
 351
 352         spin_lock_bh(&tun->lock);
 353         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 354                 struct tun_flow_entry *e;
 355                 struct hlist_node *n;
 356
 357                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 358                         if (e->queue_index == queue_index)
 359                                 tun_flow_delete(tun, e);
 360                 }
 361         }
 362         spin_unlock_bh(&tun->lock);
 363 }
 364
 365 static void tun_flow_cleanup(unsigned long data)
 366 {
 367         struct tun_struct *tun = (struct tun_struct *)data;
 368         unsigned long delay = tun->ageing_time;
 369         unsigned long next_timer = jiffies + delay;
 370         unsigned long count = 0;
 371         int i;
 372
 373         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 374
 375         spin_lock_bh(&tun->lock);
 376         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 377                 struct tun_flow_entry *e;
 378                 struct hlist_node *n;
 379
 380                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 381                         unsigned long this_timer;
 382                         count++;
 383                         this_timer = e->updated + delay;
 384                         if (time_before_eq(this_timer, jiffies))
 385                                 tun_flow_delete(tun, e);
 386                         else if (time_before(this_timer, next_timer))
 387                                 next_timer = this_timer;
 388                 }
 389         }
 390
 391         if (count)
 392                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 393         spin_unlock_bh(&tun->lock);
 394 }
 395
 396 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 397                             struct tun_file *tfile)
 398 {
 399         struct hlist_head *head;
 400         struct tun_flow_entry *e;
 401         unsigned long delay = tun->ageing_time;
 402         u16 queue_index = tfile->queue_index;
 403
 404         if (!rxhash)
 405                 return;
 406         else
 407                 head = &tun->flows[tun_hashfn(rxhash)];
 408
 409         rcu_read_lock();
 410
 411         /* We may get a very small possibility of OOO during switching, not
 412          * worth to optimize.*/
 413         if (tun->numqueues == 1 || tfile->detached)
 414                 goto unlock;
 415
 416         e = tun_flow_find(head, rxhash);
 417         if (likely(e)) {
 418                 /* TODO: keep queueing to old queue until it's empty? */
 419                 e->queue_index = queue_index;
 420                 e->updated = jiffies;
 421                 sock_rps_record_flow_hash(e->rps_rxhash);
 422         } else {
 423                 spin_lock_bh(&tun->lock);
 424                 if (!tun_flow_find(head, rxhash) &&
 425                     tun->flow_count < MAX_TAP_FLOWS)
 426                         tun_flow_create(tun, head, rxhash, queue_index);
 427
 428                 if (!timer_pending(&tun->flow_gc_timer))
 429                         mod_timer(&tun->flow_gc_timer,
 430                                   round_jiffies_up(jiffies + delay));
 431                 spin_unlock_bh(&tun->lock);
 432         }
 433
 434 unlock:
 435         rcu_read_unlock();
 436 }
 437
 438 /**
 439  * Save the hash received in the stack receive path and update the
 440  * flow_hash table accordingly.
 441  */
 442 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 443 {
 444         if (unlikely(e->rps_rxhash != hash))
 445                 e->rps_rxhash = hash;
 446 }
 447
 448 /* We try to identify a flow through its rxhash first. The reason that
 449  * we do not check rxq no. is because some cards(e.g 82599), chooses
 450  * the rxq based on the txq where the last packet of the flow comes. As
 451  * the userspace application move between processors, we may get a
 452  * different rxq no. here. If we could not get rxhash, then we would
 453  * hope the rxq no. may help here.
 454  */
 455 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 456                             void *accel_priv, select_queue_fallback_t fallback)
 457 {
 458         struct tun_struct *tun = netdev_priv(dev);
 459         struct tun_flow_entry *e;
 460         u32 txq = 0;
 461         u32 numqueues = 0;
 462
 463         rcu_read_lock();
 464         numqueues = ACCESS_ONCE(tun->numqueues);
 465
 466         txq = skb_get_hash(skb);
 467         if (txq) {
 468                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 469                 if (e) {
 470                         tun_flow_save_rps_rxhash(e, txq);
 471                         txq = e->queue_index;
 472                 } else
 473                         /* use multiply and shift instead of expensive divide */
 474                         txq = ((u64)txq * numqueues) >> 32;
 475         } else if (likely(skb_rx_queue_recorded(skb))) {
 476                 txq = skb_get_rx_queue(skb);
 477                 while (unlikely(txq >= numqueues))
 478                         txq -= numqueues;
 479         }
 480
 481         rcu_read_unlock();
 482         return txq;
 483 }
 484
 485 static inline bool tun_not_capable(struct tun_struct *tun)
 486 {
 487         const struct cred *cred = current_cred();
 488         struct net *net = dev_net(tun->dev);
 489
 490         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 491                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 492                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 493 }
 494
 495 static void tun_set_real_num_queues(struct tun_struct *tun)
 496 {
 497         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 498         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 499 }
 500
 501 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 502 {
 503         tfile->detached = tun;
 504         list_add_tail(&tfile->next, &tun->disabled);
 505         ++tun->numdisabled;
 506 }
 507
 508 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 509 {
 510         struct tun_struct *tun = tfile->detached;
 511
 512         tfile->detached = NULL;
 513         list_del_init(&tfile->next);
 514         --tun->numdisabled;
 515         return tun;
 516 }
 517
 518 static void tun_queue_purge(struct tun_file *tfile)
 519 {
 520         struct sk_buff *skb;
 521
 522         while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
 523                 kfree_skb(skb);
 524
 525         skb_queue_purge(&tfile->sk.sk_error_queue);
 526 }
 527
 528 static void __tun_detach(struct tun_file *tfile, bool clean)
 529 {
 530         struct tun_file *ntfile;
 531         struct tun_struct *tun;
 532
 533         tun = rtnl_dereference(tfile->tun);
 534
 535         if (tun && !tfile->detached) {
 536                 u16 index = tfile->queue_index;
 537                 BUG_ON(index >= tun->numqueues);
 538
 539                 rcu_assign_pointer(tun->tfiles[index],
 540                                    tun->tfiles[tun->numqueues - 1]);
 541                 ntfile = rtnl_dereference(tun->tfiles[index]);
 542                 ntfile->queue_index = index;
 543
 544                 --tun->numqueues;
 545                 if (clean) {
 546                         RCU_INIT_POINTER(tfile->tun, NULL);
 547                         sock_put(&tfile->sk);
 548                 } else
 549                         tun_disable_queue(tun, tfile);
 550
 551                 synchronize_net();
 552                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 553                 /* Drop read queue */
 554                 tun_queue_purge(tfile);
 555                 tun_set_real_num_queues(tun);
 556         } else if (tfile->detached && clean) {
 557                 tun = tun_enable_queue(tfile);
 558                 sock_put(&tfile->sk);
 559         }
 560
 561         if (clean) {
 562                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 563                         netif_carrier_off(tun->dev);
 564
 565                         if (!(tun->flags & IFF_PERSIST) &&
 566                             tun->dev->reg_state == NETREG_REGISTERED)
 567                                 unregister_netdevice(tun->dev);
 568                 }
 569                 if (tun)
 570                         skb_array_cleanup(&tfile->tx_array);
 571                 sock_put(&tfile->sk);
 572         }
 573 }
 574
 575 static void tun_detach(struct tun_file *tfile, bool clean)
 576 {
 577         rtnl_lock();
 578         __tun_detach(tfile, clean);
 579         rtnl_unlock();
 580 }
 581
 582 static void tun_detach_all(struct net_device *dev)
 583 {
 584         struct tun_struct *tun = netdev_priv(dev);
 585         struct tun_file *tfile, *tmp;
 586         int i, n = tun->numqueues;
 587
 588         for (i = 0; i < n; i++) {
 589                 tfile = rtnl_dereference(tun->tfiles[i]);
 590                 BUG_ON(!tfile);
 591                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 592                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 593                 RCU_INIT_POINTER(tfile->tun, NULL);
 594                 --tun->numqueues;
 595         }
 596         list_for_each_entry(tfile, &tun->disabled, next) {
 597                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 598                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 599                 RCU_INIT_POINTER(tfile->tun, NULL);
 600         }
 601         BUG_ON(tun->numqueues != 0);
 602
 603         synchronize_net();
 604         for (i = 0; i < n; i++) {
 605                 tfile = rtnl_dereference(tun->tfiles[i]);
 606                 /* Drop read queue */
 607                 tun_queue_purge(tfile);
 608                 sock_put(&tfile->sk);
 609         }
 610         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 611                 tun_enable_queue(tfile);
 612                 tun_queue_purge(tfile);
 613                 sock_put(&tfile->sk);
 614         }
 615         BUG_ON(tun->numdisabled != 0);
 616
 617         if (tun->flags & IFF_PERSIST)
 618                 module_put(THIS_MODULE);
 619 }
 620
 621 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 622 {
 623         struct tun_file *tfile = file->private_data;
 624         struct net_device *dev = tun->dev;
 625         int err;
 626
 627         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 628         if (err < 0)
 629                 goto out;
 630
 631         err = -EINVAL;
 632         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 633                 goto out;
 634
 635         err = -EBUSY;
 636         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 637                 goto out;
 638
 639         err = -E2BIG;
 640         if (!tfile->detached &&
 641             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 642                 goto out;
 643
 644         err = 0;
 645
 646         /* Re-attach the filter to persist device */
 647         if (!skip_filter && (tun->filter_attached == true)) {
 648                 lock_sock(tfile->socket.sk);
 649                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 650                 release_sock(tfile->socket.sk);
 651                 if (!err)
 652                         goto out;
 653         }
 654
 655         if (!tfile->detached &&
 656             skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
 657                 err = -ENOMEM;
 658                 goto out;
 659         }
 660
 661         tfile->queue_index = tun->numqueues;
 662         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 663         rcu_assign_pointer(tfile->tun, tun);
 664         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 665         tun->numqueues++;
 666
 667         if (tfile->detached)
 668                 tun_enable_queue(tfile);
 669         else
 670                 sock_hold(&tfile->sk);
 671
 672         tun_set_real_num_queues(tun);
 673
 674         /* device is allowed to go away first, so no need to hold extra
 675          * refcnt.
 676          */
 677
 678 out:
 679         return err;
 680 }
 681
 682 static struct tun_struct *__tun_get(struct tun_file *tfile)
 683 {
 684         struct tun_struct *tun;
 685
 686         rcu_read_lock();
 687         tun = rcu_dereference(tfile->tun);
 688         if (tun)
 689                 dev_hold(tun->dev);
 690         rcu_read_unlock();
 691
 692         return tun;
 693 }
 694
 695 static struct tun_struct *tun_get(struct file *file)
 696 {
 697         return __tun_get(file->private_data);
 698 }
 699
 700 static void tun_put(struct tun_struct *tun)
 701 {
 702         dev_put(tun->dev);
 703 }
 704
 705 /* TAP filtering */
 706 static void addr_hash_set(u32 *mask, const u8 *addr)
 707 {
 708         int n = ether_crc(ETH_ALEN, addr) >> 26;
 709         mask[n >> 5] |= (1 << (n & 31));
 710 }
 711
 712 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 713 {
 714         int n = ether_crc(ETH_ALEN, addr) >> 26;
 715         return mask[n >> 5] & (1 << (n & 31));
 716 }
 717
 718 static int update_filter(struct tap_filter *filter, void __user *arg)
 719 {
 720         struct { u8 u[ETH_ALEN]; } *addr;
 721         struct tun_filter uf;
 722         int err, alen, n, nexact;
 723
 724         if (copy_from_user(&uf, arg, sizeof(uf)))
 725                 return -EFAULT;
 726
 727         if (!uf.count) {
 728                 /* Disabled */
 729                 filter->count = 0;
 730                 return 0;
 731         }
 732
 733         alen = ETH_ALEN * uf.count;
 734         addr = memdup_user(arg + sizeof(uf), alen);
 735         if (IS_ERR(addr))
 736                 return PTR_ERR(addr);
 737
 738         /* The filter is updated without holding any locks. Which is
 739          * perfectly safe. We disable it first and in the worst
 740          * case we'll accept a few undesired packets. */
 741         filter->count = 0;
 742         wmb();
 743
 744         /* Use first set of addresses as an exact filter */
 745         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 746                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 747
 748         nexact = n;
 749
 750         /* Remaining multicast addresses are hashed,
 751          * unicast will leave the filter disabled. */
 752         memset(filter->mask, 0, sizeof(filter->mask));
 753         for (; n < uf.count; n++) {
 754                 if (!is_multicast_ether_addr(addr[n].u)) {
 755                         err = 0; /* no filter */
 756                         goto free_addr;
 757                 }
 758                 addr_hash_set(filter->mask, addr[n].u);
 759         }
 760
 761         /* For ALLMULTI just set the mask to all ones.
 762          * This overrides the mask populated above. */
 763         if ((uf.flags & TUN_FLT_ALLMULTI))
 764                 memset(filter->mask, ~0, sizeof(filter->mask));
 765
 766         /* Now enable the filter */
 767         wmb();
 768         filter->count = nexact;
 769
 770         /* Return the number of exact filters */
 771         err = nexact;
 772 free_addr:
 773         kfree(addr);
 774         return err;
 775 }
 776
 777 /* Returns: 0 - drop, !=0 - accept */
 778 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 779 {
 780         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 781          * at this point. */
 782         struct ethhdr *eh = (struct ethhdr *) skb->data;
 783         int i;
 784
 785         /* Exact match */
 786         for (i = 0; i < filter->count; i++)
 787                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 788                         return 1;
 789
 790         /* Inexact match (multicast only) */
 791         if (is_multicast_ether_addr(eh->h_dest))
 792                 return addr_hash_test(filter->mask, eh->h_dest);
 793
 794         return 0;
 795 }
 796
 797 /*
 798  * Checks whether the packet is accepted or not.
 799  * Returns: 0 - drop, !=0 - accept
 800  */
 801 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 802 {
 803         if (!filter->count)
 804                 return 1;
 805
 806         return run_filter(filter, skb);
 807 }
 808
 809 /* Network device part of the driver */
 810
 811 static const struct ethtool_ops tun_ethtool_ops;
 812
 813 /* Net device detach from fd. */
 814 static void tun_net_uninit(struct net_device *dev)
 815 {
 816         tun_detach_all(dev);
 817 }
 818
 819 /* Net device open. */
 820 static int tun_net_open(struct net_device *dev)
 821 {
 822         netif_tx_start_all_queues(dev);
 823         return 0;
 824 }
 825
 826 /* Net device close. */
 827 static int tun_net_close(struct net_device *dev)
 828 {
 829         netif_tx_stop_all_queues(dev);
 830         return 0;
 831 }
 832
 833 /* Net device start xmit */
 834 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 835 {
 836         struct tun_struct *tun = netdev_priv(dev);
 837         int txq = skb->queue_mapping;
 838         struct tun_file *tfile;
 839         u32 numqueues = 0;
 840
 841         rcu_read_lock();
 842         tfile = rcu_dereference(tun->tfiles[txq]);
 843         numqueues = ACCESS_ONCE(tun->numqueues);
 844
 845         /* Drop packet if interface is not attached */
 846         if (txq >= numqueues)
 847                 goto drop;
 848
 849 #ifdef CONFIG_RPS
 850         if (numqueues == 1 && static_key_false(&rps_needed)) {
 851                 /* Select queue was not called for the skbuff, so we extract the
 852                  * RPS hash and save it into the flow_table here.
 853                  */
 854                 __u32 rxhash;
 855
 856                 rxhash = skb_get_hash(skb);
 857                 if (rxhash) {
 858                         struct tun_flow_entry *e;
 859                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 860                                         rxhash);
 861                         if (e)
 862                                 tun_flow_save_rps_rxhash(e, rxhash);
 863                 }
 864         }
 865 #endif
 866
 867         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 868
 869         BUG_ON(!tfile);
 870
 871         /* Drop if the filter does not like it.
 872          * This is a noop if the filter is disabled.
 873          * Filter can be enabled only for the TAP devices. */
 874         if (!check_filter(&tun->txflt, skb))
 875                 goto drop;
 876
 877         if (tfile->socket.sk->sk_filter &&
 878             sk_filter(tfile->socket.sk, skb))
 879                 goto drop;
 880
 881         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 882                 goto drop;
 883
 884         skb_tx_timestamp(skb);
 885
 886         /* Orphan the skb - required as we might hang on to it
 887          * for indefinite time.
 888          */
 889         skb_orphan(skb);
 890
 891         nf_reset(skb);
 892
 893         if (skb_array_produce(&tfile->tx_array, skb))
 894                 goto drop;
 895
 896         /* Notify and wake up reader process */
 897         if (tfile->flags & TUN_FASYNC)
 898                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 899         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 900
 901         rcu_read_unlock();
 902         return NETDEV_TX_OK;
 903
 904 drop:
 905         this_cpu_inc(tun->pcpu_stats->tx_dropped);
 906         skb_tx_error(skb);
 907         kfree_skb(skb);
 908         rcu_read_unlock();
 909         return NET_XMIT_DROP;
 910 }
 911
 912 static void tun_net_mclist(struct net_device *dev)
 913 {
 914         /*
 915          * This callback is supposed to deal with mc filter in
 916          * _rx_ path and has nothing to do with the _tx_ path.
 917          * In rx path we always accept everything userspace gives us.
 918          */
 919 }
 920
 921 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 922         netdev_features_t features)
 923 {
 924         struct tun_struct *tun = netdev_priv(dev);
 925
 926         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 927 }
 928 #ifdef CONFIG_NET_POLL_CONTROLLER
 929 static void tun_poll_controller(struct net_device *dev)
 930 {
 931         /*
 932          * Tun only receives frames when:
 933          * 1) the char device endpoint gets data from user space
 934          * 2) the tun socket gets a sendmsg call from user space
 935          * Since both of those are synchronous operations, we are guaranteed
 936          * never to have pending data when we poll for it
 937          * so there is nothing to do here but return.
 938          * We need this though so netpoll recognizes us as an interface that
 939          * supports polling, which enables bridge devices in virt setups to
 940          * still use netconsole
 941          */
 942         return;
 943 }
 944 #endif
 945
 946 static void tun_set_headroom(struct net_device *dev, int new_hr)
 947 {
 948         struct tun_struct *tun = netdev_priv(dev);
 949
 950         if (new_hr < NET_SKB_PAD)
 951                 new_hr = NET_SKB_PAD;
 952
 953         tun->align = new_hr;
 954 }
 955
 956 static struct rtnl_link_stats64 *
 957 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
 958 {
 959         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
 960         struct tun_struct *tun = netdev_priv(dev);
 961         struct tun_pcpu_stats *p;
 962         int i;
 963
 964         for_each_possible_cpu(i) {
 965                 u64 rxpackets, rxbytes, txpackets, txbytes;
 966                 unsigned int start;
 967
 968                 p = per_cpu_ptr(tun->pcpu_stats, i);
 969                 do {
 970                         start = u64_stats_fetch_begin(&p->syncp);
 971                         rxpackets       = p->rx_packets;
 972                         rxbytes         = p->rx_bytes;
 973                         txpackets       = p->tx_packets;
 974                         txbytes         = p->tx_bytes;
 975                 } while (u64_stats_fetch_retry(&p->syncp, start));
 976
 977                 stats->rx_packets       += rxpackets;
 978                 stats->rx_bytes         += rxbytes;
 979                 stats->tx_packets       += txpackets;
 980                 stats->tx_bytes         += txbytes;
 981
 982                 /* u32 counters */
 983                 rx_dropped      += p->rx_dropped;
 984                 rx_frame_errors += p->rx_frame_errors;
 985                 tx_dropped      += p->tx_dropped;
 986         }
 987         stats->rx_dropped  = rx_dropped;
 988         stats->rx_frame_errors = rx_frame_errors;
 989         stats->tx_dropped = tx_dropped;
 990         return stats;
 991 }
 992
 993 static const struct net_device_ops tun_netdev_ops = {
 994         .ndo_uninit             = tun_net_uninit,
 995         .ndo_open               = tun_net_open,
 996         .ndo_stop               = tun_net_close,
 997         .ndo_start_xmit         = tun_net_xmit,
 998         .ndo_fix_features       = tun_net_fix_features,
 999         .ndo_select_queue       = tun_select_queue,
1000 #ifdef CONFIG_NET_POLL_CONTROLLER
1001         .ndo_poll_controller    = tun_poll_controller,
1002 #endif
1003         .ndo_set_rx_headroom    = tun_set_headroom,
1004         .ndo_get_stats64        = tun_net_get_stats64,
1005 };
1006
1007 static const struct net_device_ops tap_netdev_ops = {
1008         .ndo_uninit             = tun_net_uninit,
1009         .ndo_open               = tun_net_open,
1010         .ndo_stop               = tun_net_close,
1011         .ndo_start_xmit         = tun_net_xmit,
1012         .ndo_fix_features       = tun_net_fix_features,
1013         .ndo_set_rx_mode        = tun_net_mclist,
1014         .ndo_set_mac_address    = eth_mac_addr,
1015         .ndo_validate_addr      = eth_validate_addr,
1016         .ndo_select_queue       = tun_select_queue,
1017 #ifdef CONFIG_NET_POLL_CONTROLLER
1018         .ndo_poll_controller    = tun_poll_controller,
1019 #endif
1020         .ndo_features_check     = passthru_features_check,
1021         .ndo_set_rx_headroom    = tun_set_headroom,
1022         .ndo_get_stats64        = tun_net_get_stats64,
1023 };
1024
1025 static void tun_flow_init(struct tun_struct *tun)
1026 {
1027         int i;
1028
1029         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1030                 INIT_HLIST_HEAD(&tun->flows[i]);
1031
1032         tun->ageing_time = TUN_FLOW_EXPIRE;
1033         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1034         mod_timer(&tun->flow_gc_timer,
1035                   round_jiffies_up(jiffies + tun->ageing_time));
1036 }
1037
1038 static void tun_flow_uninit(struct tun_struct *tun)
1039 {
1040         del_timer_sync(&tun->flow_gc_timer);
1041         tun_flow_flush(tun);
1042 }
1043
1044 #define MIN_MTU 68
1045 #define MAX_MTU 65535
1046
1047 /* Initialize net device. */
1048 static void tun_net_init(struct net_device *dev)
1049 {
1050         struct tun_struct *tun = netdev_priv(dev);
1051
1052         switch (tun->flags & TUN_TYPE_MASK) {
1053         case IFF_TUN:
1054                 dev->netdev_ops = &tun_netdev_ops;
1055
1056                 /* Point-to-Point TUN Device */
1057                 dev->hard_header_len = 0;
1058                 dev->addr_len = 0;
1059                 dev->mtu = 1500;
1060
1061                 /* Zero header length */
1062                 dev->type = ARPHRD_NONE;
1063                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1064                 break;
1065
1066         case IFF_TAP:
1067                 dev->netdev_ops = &tap_netdev_ops;
1068                 /* Ethernet TAP Device */
1069                 ether_setup(dev);
1070                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1071                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1072
1073                 eth_hw_addr_random(dev);
1074
1075                 break;
1076         }
1077
1078         dev->min_mtu = MIN_MTU;
1079         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1080 }
1081
1082 /* Character device part */
1083
1084 /* Poll */
1085 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1086 {
1087         struct tun_file *tfile = file->private_data;
1088         struct tun_struct *tun = __tun_get(tfile);
1089         struct sock *sk;
1090         unsigned int mask = 0;
1091
1092         if (!tun)
1093                 return POLLERR;
1094
1095         sk = tfile->socket.sk;
1096
1097         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1098
1099         poll_wait(file, sk_sleep(sk), wait);
1100
1101         if (!skb_array_empty(&tfile->tx_array))
1102                 mask |= POLLIN | POLLRDNORM;
1103
1104         if (sock_writeable(sk) ||
1105             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1106              sock_writeable(sk)))
1107                 mask |= POLLOUT | POLLWRNORM;
1108
1109         if (tun->dev->reg_state != NETREG_REGISTERED)
1110                 mask = POLLERR;
1111
1112         tun_put(tun);
1113         return mask;
1114 }
1115
1116 /* prepad is the amount to reserve at front.  len is length after that.
1117  * linear is a hint as to how much to copy (usually headers). */
1118 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1119                                      size_t prepad, size_t len,
1120                                      size_t linear, int noblock)
1121 {
1122         struct sock *sk = tfile->socket.sk;
1123         struct sk_buff *skb;
1124         int err;
1125
1126         /* Under a page?  Don't bother with paged skb. */
1127         if (prepad + len < PAGE_SIZE || !linear)
1128                 linear = len;
1129
1130         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1131                                    &err, 0);
1132         if (!skb)
1133                 return ERR_PTR(err);
1134
1135         skb_reserve(skb, prepad);
1136         skb_put(skb, linear);
1137         skb->data_len = len - linear;
1138         skb->len += len - linear;
1139
1140         return skb;
1141 }
1142
1143 /* Get packet from user space buffer */
1144 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1145                             void *msg_control, struct iov_iter *from,
1146                             int noblock)
1147 {
1148         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1149         struct sk_buff *skb;
1150         size_t total_len = iov_iter_count(from);
1151         size_t len = total_len, align = tun->align, linear;
1152         struct virtio_net_hdr gso = { 0 };
1153         struct tun_pcpu_stats *stats;
1154         int good_linear;
1155         int copylen;
1156         bool zerocopy = false;
1157         int err;
1158         u32 rxhash;
1159
1160         if (!(tun->dev->flags & IFF_UP))
1161                 return -EIO;
1162
1163         if (!(tun->flags & IFF_NO_PI)) {
1164                 if (len < sizeof(pi))
1165                         return -EINVAL;
1166                 len -= sizeof(pi);
1167
1168                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1169                         return -EFAULT;
1170         }
1171
1172         if (tun->flags & IFF_VNET_HDR) {
1173                 if (len < tun->vnet_hdr_sz)
1174                         return -EINVAL;
1175                 len -= tun->vnet_hdr_sz;
1176
1177                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1178                         return -EFAULT;
1179
1180                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1181                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1182                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1183
1184                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1185                         return -EINVAL;
1186                 iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1187         }
1188
1189         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1190                 align += NET_IP_ALIGN;
1191                 if (unlikely(len < ETH_HLEN ||
1192                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1193                         return -EINVAL;
1194         }
1195
1196         good_linear = SKB_MAX_HEAD(align);
1197
1198         if (msg_control) {
1199                 struct iov_iter i = *from;
1200
1201                 /* There are 256 bytes to be copied in skb, so there is
1202                  * enough room for skb expand head in case it is used.
1203                  * The rest of the buffer is mapped from userspace.
1204                  */
1205                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1206                 if (copylen > good_linear)
1207                         copylen = good_linear;
1208                 linear = copylen;
1209                 iov_iter_advance(&i, copylen);
1210                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1211                         zerocopy = true;
1212         }
1213
1214         if (!zerocopy) {
1215                 copylen = len;
1216                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1217                         linear = good_linear;
1218                 else
1219                         linear = tun16_to_cpu(tun, gso.hdr_len);
1220         }
1221
1222         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1223         if (IS_ERR(skb)) {
1224                 if (PTR_ERR(skb) != -EAGAIN)
1225                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1226                 return PTR_ERR(skb);
1227         }
1228
1229         if (zerocopy)
1230                 err = zerocopy_sg_from_iter(skb, from);
1231         else
1232                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1233
1234         if (err) {
1235                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1236                 kfree_skb(skb);
1237                 return -EFAULT;
1238         }
1239
1240         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1241                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1242                 kfree_skb(skb);
1243                 return -EINVAL;
1244         }
1245
1246         switch (tun->flags & TUN_TYPE_MASK) {
1247         case IFF_TUN:
1248                 if (tun->flags & IFF_NO_PI) {
1249                         switch (skb->data[0] & 0xf0) {
1250                         case 0x40:
1251                                 pi.proto = htons(ETH_P_IP);
1252                                 break;
1253                         case 0x60:
1254                                 pi.proto = htons(ETH_P_IPV6);
1255                                 break;
1256                         default:
1257                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1258                                 kfree_skb(skb);
1259                                 return -EINVAL;
1260                         }
1261                 }
1262
1263                 skb_reset_mac_header(skb);
1264                 skb->protocol = pi.proto;
1265                 skb->dev = tun->dev;
1266                 break;
1267         case IFF_TAP:
1268                 skb->protocol = eth_type_trans(skb, tun->dev);
1269                 break;
1270         }
1271
1272         /* copy skb_ubuf_info for callback when skb has no error */
1273         if (zerocopy) {
1274                 skb_shinfo(skb)->destructor_arg = msg_control;
1275                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1276                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1277         } else if (msg_control) {
1278                 struct ubuf_info *uarg = msg_control;
1279                 uarg->callback(uarg, false);
1280         }
1281
1282         skb_reset_network_header(skb);
1283         skb_probe_transport_header(skb, 0);
1284
1285         rxhash = skb_get_hash(skb);
1286 #ifndef CONFIG_4KSTACKS
1287         local_bh_disable();
1288         netif_receive_skb(skb);
1289         local_bh_enable();
1290 #else
1291         netif_rx_ni(skb);
1292 #endif
1293
1294         stats = get_cpu_ptr(tun->pcpu_stats);
1295         u64_stats_update_begin(&stats->syncp);
1296         stats->rx_packets++;
1297         stats->rx_bytes += len;
1298         u64_stats_update_end(&stats->syncp);
1299         put_cpu_ptr(stats);
1300
1301         tun_flow_update(tun, rxhash, tfile);
1302         return total_len;
1303 }
1304
1305 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1306 {
1307         struct file *file = iocb->ki_filp;
1308         struct tun_struct *tun = tun_get(file);
1309         struct tun_file *tfile = file->private_data;
1310         ssize_t result;
1311
1312         if (!tun)
1313                 return -EBADFD;
1314
1315         result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1316
1317         tun_put(tun);
1318         return result;
1319 }
1320
1321 /* Put packet to the user space buffer */
1322 static ssize_t tun_put_user(struct tun_struct *tun,
1323                             struct tun_file *tfile,
1324                             struct sk_buff *skb,
1325                             struct iov_iter *iter)
1326 {
1327         struct tun_pi pi = { 0, skb->protocol };
1328         struct tun_pcpu_stats *stats;
1329         ssize_t total;
1330         int vlan_offset = 0;
1331         int vlan_hlen = 0;
1332         int vnet_hdr_sz = 0;
1333
1334         if (skb_vlan_tag_present(skb))
1335                 vlan_hlen = VLAN_HLEN;
1336
1337         if (tun->flags & IFF_VNET_HDR)
1338                 vnet_hdr_sz = tun->vnet_hdr_sz;
1339
1340         total = skb->len + vlan_hlen + vnet_hdr_sz;
1341
1342         if (!(tun->flags & IFF_NO_PI)) {
1343                 if (iov_iter_count(iter) < sizeof(pi))
1344                         return -EINVAL;
1345
1346                 total += sizeof(pi);
1347                 if (iov_iter_count(iter) < total) {
1348                         /* Packet will be striped */
1349                         pi.flags |= TUN_PKT_STRIP;
1350                 }
1351
1352                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1353                         return -EFAULT;
1354         }
1355
1356         if (vnet_hdr_sz) {
1357                 struct virtio_net_hdr gso;
1358
1359                 if (iov_iter_count(iter) < vnet_hdr_sz)
1360                         return -EINVAL;
1361
1362                 if (virtio_net_hdr_from_skb(skb, &gso,
1363                                             tun_is_little_endian(tun))) {
1364                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1365                         pr_err("unexpected GSO type: "
1366                                "0x%x, gso_size %d, hdr_len %d\n",
1367                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1368                                tun16_to_cpu(tun, gso.hdr_len));
1369                         print_hex_dump(KERN_ERR, "tun: ",
1370                                        DUMP_PREFIX_NONE,
1371                                        16, 1, skb->head,
1372                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1373                         WARN_ON_ONCE(1);
1374                         return -EINVAL;
1375                 }
1376
1377                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1378                         return -EFAULT;
1379
1380                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1381         }
1382
1383         if (vlan_hlen) {
1384                 int ret;
1385                 struct {
1386                         __be16 h_vlan_proto;
1387                         __be16 h_vlan_TCI;
1388                 } veth;
1389
1390                 veth.h_vlan_proto = skb->vlan_proto;
1391                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1392
1393                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1394
1395                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1396                 if (ret || !iov_iter_count(iter))
1397                         goto done;
1398
1399                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1400                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1401                         goto done;
1402         }
1403
1404         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1405
1406 done:
1407         /* caller is in process context, */
1408         stats = get_cpu_ptr(tun->pcpu_stats);
1409         u64_stats_update_begin(&stats->syncp);
1410         stats->tx_packets++;
1411         stats->tx_bytes += skb->len + vlan_hlen;
1412         u64_stats_update_end(&stats->syncp);
1413         put_cpu_ptr(tun->pcpu_stats);
1414
1415         return total;
1416 }
1417
1418 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1419                                      int *err)
1420 {
1421         DECLARE_WAITQUEUE(wait, current);
1422         struct sk_buff *skb = NULL;
1423         int error = 0;
1424
1425         skb = skb_array_consume(&tfile->tx_array);
1426         if (skb)
1427                 goto out;
1428         if (noblock) {
1429                 error = -EAGAIN;
1430                 goto out;
1431         }
1432
1433         add_wait_queue(&tfile->wq.wait, &wait);
1434         current->state = TASK_INTERRUPTIBLE;
1435
1436         while (1) {
1437                 skb = skb_array_consume(&tfile->tx_array);
1438                 if (skb)
1439                         break;
1440                 if (signal_pending(current)) {
1441                         error = -ERESTARTSYS;
1442                         break;
1443                 }
1444                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1445                         error = -EFAULT;
1446                         break;
1447                 }
1448
1449                 schedule();
1450         }
1451
1452         current->state = TASK_RUNNING;
1453         remove_wait_queue(&tfile->wq.wait, &wait);
1454
1455 out:
1456         *err = error;
1457         return skb;
1458 }
1459
1460 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1461                            struct iov_iter *to,
1462                            int noblock)
1463 {
1464         struct sk_buff *skb;
1465         ssize_t ret;
1466         int err;
1467
1468         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1469
1470         if (!iov_iter_count(to))
1471                 return 0;
1472
1473         /* Read frames from ring */
1474         skb = tun_ring_recv(tfile, noblock, &err);
1475         if (!skb)
1476                 return err;
1477
1478         ret = tun_put_user(tun, tfile, skb, to);
1479         if (unlikely(ret < 0))
1480                 kfree_skb(skb);
1481         else
1482                 consume_skb(skb);
1483
1484         return ret;
1485 }
1486
1487 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1488 {
1489         struct file *file = iocb->ki_filp;
1490         struct tun_file *tfile = file->private_data;
1491         struct tun_struct *tun = __tun_get(tfile);
1492         ssize_t len = iov_iter_count(to), ret;
1493
1494         if (!tun)
1495                 return -EBADFD;
1496         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1497         ret = min_t(ssize_t, ret, len);
1498         if (ret > 0)
1499                 iocb->ki_pos = ret;
1500         tun_put(tun);
1501         return ret;
1502 }
1503
1504 static void tun_free_netdev(struct net_device *dev)
1505 {
1506         struct tun_struct *tun = netdev_priv(dev);
1507
1508         BUG_ON(!(list_empty(&tun->disabled)));
1509         free_percpu(tun->pcpu_stats);
1510         tun_flow_uninit(tun);
1511         security_tun_dev_free_security(tun->security);
1512         free_netdev(dev);
1513 }
1514
1515 static void tun_setup(struct net_device *dev)
1516 {
1517         struct tun_struct *tun = netdev_priv(dev);
1518
1519         tun->owner = INVALID_UID;
1520         tun->group = INVALID_GID;
1521
1522         dev->ethtool_ops = &tun_ethtool_ops;
1523         dev->destructor = tun_free_netdev;
1524         /* We prefer our own queue length */
1525         dev->tx_queue_len = TUN_READQ_SIZE;
1526 }
1527
1528 /* Trivial set of netlink ops to allow deleting tun or tap
1529  * device with netlink.
1530  */
1531 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1532 {
1533         return -EINVAL;
1534 }
1535
1536 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1537         .kind           = DRV_NAME,
1538         .priv_size      = sizeof(struct tun_struct),
1539         .setup          = tun_setup,
1540         .validate       = tun_validate,
1541 };
1542
1543 static void tun_sock_write_space(struct sock *sk)
1544 {
1545         struct tun_file *tfile;
1546         wait_queue_head_t *wqueue;
1547
1548         if (!sock_writeable(sk))
1549                 return;
1550
1551         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1552                 return;
1553
1554         wqueue = sk_sleep(sk);
1555         if (wqueue && waitqueue_active(wqueue))
1556                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1557                                                 POLLWRNORM | POLLWRBAND);
1558
1559         tfile = container_of(sk, struct tun_file, sk);
1560         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1561 }
1562
1563 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1564 {
1565         int ret;
1566         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1567         struct tun_struct *tun = __tun_get(tfile);
1568
1569         if (!tun)
1570                 return -EBADFD;
1571
1572         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1573                            m->msg_flags & MSG_DONTWAIT);
1574         tun_put(tun);
1575         return ret;
1576 }
1577
1578 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1579                        int flags)
1580 {
1581         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1582         struct tun_struct *tun = __tun_get(tfile);
1583         int ret;
1584
1585         if (!tun)
1586                 return -EBADFD;
1587
1588         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1589                 ret = -EINVAL;
1590                 goto out;
1591         }
1592         if (flags & MSG_ERRQUEUE) {
1593                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1594                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1595                 goto out;
1596         }
1597         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1598         if (ret > (ssize_t)total_len) {
1599                 m->msg_flags |= MSG_TRUNC;
1600                 ret = flags & MSG_TRUNC ? ret : total_len;
1601         }
1602 out:
1603         tun_put(tun);
1604         return ret;
1605 }
1606
1607 static int tun_peek_len(struct socket *sock)
1608 {
1609         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1610         struct tun_struct *tun;
1611         int ret = 0;
1612
1613         tun = __tun_get(tfile);
1614         if (!tun)
1615                 return 0;
1616
1617         ret = skb_array_peek_len(&tfile->tx_array);
1618         tun_put(tun);
1619
1620         return ret;
1621 }
1622
1623 /* Ops structure to mimic raw sockets with tun */
1624 static const struct proto_ops tun_socket_ops = {
1625         .peek_len = tun_peek_len,
1626         .sendmsg = tun_sendmsg,
1627         .recvmsg = tun_recvmsg,
1628 };
1629
1630 static struct proto tun_proto = {
1631         .name           = "tun",
1632         .owner          = THIS_MODULE,
1633         .obj_size       = sizeof(struct tun_file),
1634 };
1635
1636 static int tun_flags(struct tun_struct *tun)
1637 {
1638         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1639 }
1640
1641 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1642                               char *buf)
1643 {
1644         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1645         return sprintf(buf, "0x%x\n", tun_flags(tun));
1646 }
1647
1648 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1649                               char *buf)
1650 {
1651         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1652         return uid_valid(tun->owner)?
1653                 sprintf(buf, "%u\n",
1654                         from_kuid_munged(current_user_ns(), tun->owner)):
1655                 sprintf(buf, "-1\n");
1656 }
1657
1658 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1659                               char *buf)
1660 {
1661         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1662         return gid_valid(tun->group) ?
1663                 sprintf(buf, "%u\n",
1664                         from_kgid_munged(current_user_ns(), tun->group)):
1665                 sprintf(buf, "-1\n");
1666 }
1667
1668 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1669 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1670 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1671
1672 static struct attribute *tun_dev_attrs[] = {
1673         &dev_attr_tun_flags.attr,
1674         &dev_attr_owner.attr,
1675         &dev_attr_group.attr,
1676         NULL
1677 };
1678
1679 static const struct attribute_group tun_attr_group = {
1680         .attrs = tun_dev_attrs
1681 };
1682
1683 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1684 {
1685         struct tun_struct *tun;
1686         struct tun_file *tfile = file->private_data;
1687         struct net_device *dev;
1688         int err;
1689
1690         if (tfile->detached)
1691                 return -EINVAL;
1692
1693         dev = __dev_get_by_name(net, ifr->ifr_name);
1694         if (dev) {
1695                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1696                         return -EBUSY;
1697                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1698                         tun = netdev_priv(dev);
1699                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1700                         tun = netdev_priv(dev);
1701                 else
1702                         return -EINVAL;
1703
1704                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1705                     !!(tun->flags & IFF_MULTI_QUEUE))
1706                         return -EINVAL;
1707
1708                 if (tun_not_capable(tun))
1709                         return -EPERM;
1710                 err = security_tun_dev_open(tun->security);
1711                 if (err < 0)
1712                         return err;
1713
1714                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1715                 if (err < 0)
1716                         return err;
1717
1718                 if (tun->flags & IFF_MULTI_QUEUE &&
1719                     (tun->numqueues + tun->numdisabled > 1)) {
1720                         /* One or more queue has already been attached, no need
1721                          * to initialize the device again.
1722                          */
1723                         return 0;
1724                 }
1725         }
1726         else {
1727                 char *name;
1728                 unsigned long flags = 0;
1729                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1730                              MAX_TAP_QUEUES : 1;
1731
1732                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1733                         return -EPERM;
1734                 err = security_tun_dev_create();
1735                 if (err < 0)
1736                         return err;
1737
1738                 /* Set dev type */
1739                 if (ifr->ifr_flags & IFF_TUN) {
1740                         /* TUN device */
1741                         flags |= IFF_TUN;
1742                         name = "tun%d";
1743                 } else if (ifr->ifr_flags & IFF_TAP) {
1744                         /* TAP device */
1745                         flags |= IFF_TAP;
1746                         name = "tap%d";
1747                 } else
1748                         return -EINVAL;
1749
1750                 if (*ifr->ifr_name)
1751                         name = ifr->ifr_name;
1752
1753                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1754                                        NET_NAME_UNKNOWN, tun_setup, queues,
1755                                        queues);
1756
1757                 if (!dev)
1758                         return -ENOMEM;
1759
1760                 dev_net_set(dev, net);
1761                 dev->rtnl_link_ops = &tun_link_ops;
1762                 dev->ifindex = tfile->ifindex;
1763                 dev->sysfs_groups[0] = &tun_attr_group;
1764
1765                 tun = netdev_priv(dev);
1766                 tun->dev = dev;
1767                 tun->flags = flags;
1768                 tun->txflt.count = 0;
1769                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1770
1771                 tun->align = NET_SKB_PAD;
1772                 tun->filter_attached = false;
1773                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1774
1775                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
1776                 if (!tun->pcpu_stats) {
1777                         err = -ENOMEM;
1778                         goto err_free_dev;
1779                 }
1780
1781                 spin_lock_init(&tun->lock);
1782
1783                 err = security_tun_dev_alloc_security(&tun->security);
1784                 if (err < 0)
1785                         goto err_free_stat;
1786
1787                 tun_net_init(dev);
1788                 tun_flow_init(tun);
1789
1790                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1791                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1792                                    NETIF_F_HW_VLAN_STAG_TX;
1793                 dev->features = dev->hw_features | NETIF_F_LLTX;
1794                 dev->vlan_features = dev->features &
1795                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1796                                        NETIF_F_HW_VLAN_STAG_TX);
1797
1798                 INIT_LIST_HEAD(&tun->disabled);
1799                 err = tun_attach(tun, file, false);
1800                 if (err < 0)
1801                         goto err_free_flow;
1802
1803                 err = register_netdevice(tun->dev);
1804                 if (err < 0)
1805                         goto err_detach;
1806         }
1807
1808         netif_carrier_on(tun->dev);
1809
1810         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1811
1812         tun->flags = (tun->flags & ~TUN_FEATURES) |
1813                 (ifr->ifr_flags & TUN_FEATURES);
1814
1815         /* Make sure persistent devices do not get stuck in
1816          * xoff state.
1817          */
1818         if (netif_running(tun->dev))
1819                 netif_tx_wake_all_queues(tun->dev);
1820
1821         strcpy(ifr->ifr_name, tun->dev->name);
1822         return 0;
1823
1824 err_detach:
1825         tun_detach_all(dev);
1826 err_free_flow:
1827         tun_flow_uninit(tun);
1828         security_tun_dev_free_security(tun->security);
1829 err_free_stat:
1830         free_percpu(tun->pcpu_stats);
1831 err_free_dev:
1832         free_netdev(dev);
1833         return err;
1834 }
1835
1836 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1837                        struct ifreq *ifr)
1838 {
1839         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1840
1841         strcpy(ifr->ifr_name, tun->dev->name);
1842
1843         ifr->ifr_flags = tun_flags(tun);
1844
1845 }
1846
1847 /* This is like a cut-down ethtool ops, except done via tun fd so no
1848  * privs required. */
1849 static int set_offload(struct tun_struct *tun, unsigned long arg)
1850 {
1851         netdev_features_t features = 0;
1852
1853         if (arg & TUN_F_CSUM) {
1854                 features |= NETIF_F_HW_CSUM;
1855                 arg &= ~TUN_F_CSUM;
1856
1857                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1858                         if (arg & TUN_F_TSO_ECN) {
1859                                 features |= NETIF_F_TSO_ECN;
1860                                 arg &= ~TUN_F_TSO_ECN;
1861                         }
1862                         if (arg & TUN_F_TSO4)
1863                                 features |= NETIF_F_TSO;
1864                         if (arg & TUN_F_TSO6)
1865                                 features |= NETIF_F_TSO6;
1866                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1867                 }
1868
1869                 if (arg & TUN_F_UFO) {
1870                         features |= NETIF_F_UFO;
1871                         arg &= ~TUN_F_UFO;
1872                 }
1873         }
1874
1875         /* This gives the user a way to test for new features in future by
1876          * trying to set them. */
1877         if (arg)
1878                 return -EINVAL;
1879
1880         tun->set_features = features;
1881         netdev_update_features(tun->dev);
1882
1883         return 0;
1884 }
1885
1886 static void tun_detach_filter(struct tun_struct *tun, int n)
1887 {
1888         int i;
1889         struct tun_file *tfile;
1890
1891         for (i = 0; i < n; i++) {
1892                 tfile = rtnl_dereference(tun->tfiles[i]);
1893                 lock_sock(tfile->socket.sk);
1894                 sk_detach_filter(tfile->socket.sk);
1895                 release_sock(tfile->socket.sk);
1896         }
1897
1898         tun->filter_attached = false;
1899 }
1900
1901 static int tun_attach_filter(struct tun_struct *tun)
1902 {
1903         int i, ret = 0;
1904         struct tun_file *tfile;
1905
1906         for (i = 0; i < tun->numqueues; i++) {
1907                 tfile = rtnl_dereference(tun->tfiles[i]);
1908                 lock_sock(tfile->socket.sk);
1909                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1910                 release_sock(tfile->socket.sk);
1911                 if (ret) {
1912                         tun_detach_filter(tun, i);
1913                         return ret;
1914                 }
1915         }
1916
1917         tun->filter_attached = true;
1918         return ret;
1919 }
1920
1921 static void tun_set_sndbuf(struct tun_struct *tun)
1922 {
1923         struct tun_file *tfile;
1924         int i;
1925
1926         for (i = 0; i < tun->numqueues; i++) {
1927                 tfile = rtnl_dereference(tun->tfiles[i]);
1928                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1929         }
1930 }
1931
1932 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1933 {
1934         struct tun_file *tfile = file->private_data;
1935         struct tun_struct *tun;
1936         int ret = 0;
1937
1938         rtnl_lock();
1939
1940         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1941                 tun = tfile->detached;
1942                 if (!tun) {
1943                         ret = -EINVAL;
1944                         goto unlock;
1945                 }
1946                 ret = security_tun_dev_attach_queue(tun->security);
1947                 if (ret < 0)
1948                         goto unlock;
1949                 ret = tun_attach(tun, file, false);
1950         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1951                 tun = rtnl_dereference(tfile->tun);
1952                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1953                         ret = -EINVAL;
1954                 else
1955                         __tun_detach(tfile, false);
1956         } else
1957                 ret = -EINVAL;
1958
1959 unlock:
1960         rtnl_unlock();
1961         return ret;
1962 }
1963
1964 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1965                             unsigned long arg, int ifreq_len)
1966 {
1967         struct tun_file *tfile = file->private_data;
1968         struct tun_struct *tun;
1969         void __user* argp = (void __user*)arg;
1970         struct ifreq ifr;
1971         kuid_t owner;
1972         kgid_t group;
1973         int sndbuf;
1974         int vnet_hdr_sz;
1975         unsigned int ifindex;
1976         int le;
1977         int ret;
1978
1979         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
1980                 if (copy_from_user(&ifr, argp, ifreq_len))
1981                         return -EFAULT;
1982         } else {
1983                 memset(&ifr, 0, sizeof(ifr));
1984         }
1985         if (cmd == TUNGETFEATURES) {
1986                 /* Currently this just means: "what IFF flags are valid?".
1987                  * This is needed because we never checked for invalid flags on
1988                  * TUNSETIFF.
1989                  */
1990                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1991                                 (unsigned int __user*)argp);
1992         } else if (cmd == TUNSETQUEUE)
1993                 return tun_set_queue(file, &ifr);
1994
1995         ret = 0;
1996         rtnl_lock();
1997
1998         tun = __tun_get(tfile);
1999         if (cmd == TUNSETIFF) {
2000                 ret = -EEXIST;
2001                 if (tun)
2002                         goto unlock;
2003
2004                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2005
2006                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2007
2008                 if (ret)
2009                         goto unlock;
2010
2011                 if (copy_to_user(argp, &ifr, ifreq_len))
2012                         ret = -EFAULT;
2013                 goto unlock;
2014         }
2015         if (cmd == TUNSETIFINDEX) {
2016                 ret = -EPERM;
2017                 if (tun)
2018                         goto unlock;
2019
2020                 ret = -EFAULT;
2021                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2022                         goto unlock;
2023
2024                 ret = 0;
2025                 tfile->ifindex = ifindex;
2026                 goto unlock;
2027         }
2028
2029         ret = -EBADFD;
2030         if (!tun)
2031                 goto unlock;
2032
2033         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2034
2035         ret = 0;
2036         switch (cmd) {
2037         case TUNGETIFF:
2038                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2039
2040                 if (tfile->detached)
2041                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2042                 if (!tfile->socket.sk->sk_filter)
2043                         ifr.ifr_flags |= IFF_NOFILTER;
2044
2045                 if (copy_to_user(argp, &ifr, ifreq_len))
2046                         ret = -EFAULT;
2047                 break;
2048
2049         case TUNSETNOCSUM:
2050                 /* Disable/Enable checksum */
2051
2052                 /* [unimplemented] */
2053                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2054                           arg ? "disabled" : "enabled");
2055                 break;
2056
2057         case TUNSETPERSIST:
2058                 /* Disable/Enable persist mode. Keep an extra reference to the
2059                  * module to prevent the module being unprobed.
2060                  */
2061                 if (arg && !(tun->flags & IFF_PERSIST)) {
2062                         tun->flags |= IFF_PERSIST;
2063                         __module_get(THIS_MODULE);
2064                 }
2065                 if (!arg && (tun->flags & IFF_PERSIST)) {
2066                         tun->flags &= ~IFF_PERSIST;
2067                         module_put(THIS_MODULE);
2068                 }
2069
2070                 tun_debug(KERN_INFO, tun, "persist %s\n",
2071                           arg ? "enabled" : "disabled");
2072                 break;
2073
2074         case TUNSETOWNER:
2075                 /* Set owner of the device */
2076                 owner = make_kuid(current_user_ns(), arg);
2077                 if (!uid_valid(owner)) {
2078                         ret = -EINVAL;
2079                         break;
2080                 }
2081                 tun->owner = owner;
2082                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2083                           from_kuid(&init_user_ns, tun->owner));
2084                 break;
2085
2086         case TUNSETGROUP:
2087                 /* Set group of the device */
2088                 group = make_kgid(current_user_ns(), arg);
2089                 if (!gid_valid(group)) {
2090                         ret = -EINVAL;
2091                         break;
2092                 }
2093                 tun->group = group;
2094                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2095                           from_kgid(&init_user_ns, tun->group));
2096                 break;
2097
2098         case TUNSETLINK:
2099                 /* Only allow setting the type when the interface is down */
2100                 if (tun->dev->flags & IFF_UP) {
2101                         tun_debug(KERN_INFO, tun,
2102                                   "Linktype set failed because interface is up\n");
2103                         ret = -EBUSY;
2104                 } else {
2105                         tun->dev->type = (int) arg;
2106                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2107                                   tun->dev->type);
2108                         ret = 0;
2109                 }
2110                 break;
2111
2112 #ifdef TUN_DEBUG
2113         case TUNSETDEBUG:
2114                 tun->debug = arg;
2115                 break;
2116 #endif
2117         case TUNSETOFFLOAD:
2118                 ret = set_offload(tun, arg);
2119                 break;
2120
2121         case TUNSETTXFILTER:
2122                 /* Can be set only for TAPs */
2123                 ret = -EINVAL;
2124                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2125                         break;
2126                 ret = update_filter(&tun->txflt, (void __user *)arg);
2127                 break;
2128
2129         case SIOCGIFHWADDR:
2130                 /* Get hw address */
2131                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2132                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2133                 if (copy_to_user(argp, &ifr, ifreq_len))
2134                         ret = -EFAULT;
2135                 break;
2136
2137         case SIOCSIFHWADDR:
2138                 /* Set hw address */
2139                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2140                           ifr.ifr_hwaddr.sa_data);
2141
2142                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2143                 break;
2144
2145         case TUNGETSNDBUF:
2146                 sndbuf = tfile->socket.sk->sk_sndbuf;
2147                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2148                         ret = -EFAULT;
2149                 break;
2150
2151         case TUNSETSNDBUF:
2152                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2153                         ret = -EFAULT;
2154                         break;
2155                 }
2156
2157                 tun->sndbuf = sndbuf;
2158                 tun_set_sndbuf(tun);
2159                 break;
2160
2161         case TUNGETVNETHDRSZ:
2162                 vnet_hdr_sz = tun->vnet_hdr_sz;
2163                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2164                         ret = -EFAULT;
2165                 break;
2166
2167         case TUNSETVNETHDRSZ:
2168                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2169                         ret = -EFAULT;
2170                         break;
2171                 }
2172                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2173                         ret = -EINVAL;
2174                         break;
2175                 }
2176
2177                 tun->vnet_hdr_sz = vnet_hdr_sz;
2178                 break;
2179
2180         case TUNGETVNETLE:
2181                 le = !!(tun->flags & TUN_VNET_LE);
2182                 if (put_user(le, (int __user *)argp))
2183                         ret = -EFAULT;
2184                 break;
2185
2186         case TUNSETVNETLE:
2187                 if (get_user(le, (int __user *)argp)) {
2188                         ret = -EFAULT;
2189                         break;
2190                 }
2191                 if (le)
2192                         tun->flags |= TUN_VNET_LE;
2193                 else
2194                         tun->flags &= ~TUN_VNET_LE;
2195                 break;
2196
2197         case TUNGETVNETBE:
2198                 ret = tun_get_vnet_be(tun, argp);
2199                 break;
2200
2201         case TUNSETVNETBE:
2202                 ret = tun_set_vnet_be(tun, argp);
2203                 break;
2204
2205         case TUNATTACHFILTER:
2206                 /* Can be set only for TAPs */
2207                 ret = -EINVAL;
2208                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2209                         break;
2210                 ret = -EFAULT;
2211                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2212                         break;
2213
2214                 ret = tun_attach_filter(tun);
2215                 break;
2216
2217         case TUNDETACHFILTER:
2218                 /* Can be set only for TAPs */
2219                 ret = -EINVAL;
2220                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2221                         break;
2222                 ret = 0;
2223                 tun_detach_filter(tun, tun->numqueues);
2224                 break;
2225
2226         case TUNGETFILTER:
2227                 ret = -EINVAL;
2228                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2229                         break;
2230                 ret = -EFAULT;
2231                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2232                         break;
2233                 ret = 0;
2234                 break;
2235
2236         default:
2237                 ret = -EINVAL;
2238                 break;
2239         }
2240
2241 unlock:
2242         rtnl_unlock();
2243         if (tun)
2244                 tun_put(tun);
2245         return ret;
2246 }
2247
2248 static long tun_chr_ioctl(struct file *file,
2249                           unsigned int cmd, unsigned long arg)
2250 {
2251         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2252 }
2253
2254 #ifdef CONFIG_COMPAT
2255 static long tun_chr_compat_ioctl(struct file *file,
2256                          unsigned int cmd, unsigned long arg)
2257 {
2258         switch (cmd) {
2259         case TUNSETIFF:
2260         case TUNGETIFF:
2261         case TUNSETTXFILTER:
2262         case TUNGETSNDBUF:
2263         case TUNSETSNDBUF:
2264         case SIOCGIFHWADDR:
2265         case SIOCSIFHWADDR:
2266                 arg = (unsigned long)compat_ptr(arg);
2267                 break;
2268         default:
2269                 arg = (compat_ulong_t)arg;
2270                 break;
2271         }
2272
2273         /*
2274          * compat_ifreq is shorter than ifreq, so we must not access beyond
2275          * the end of that structure. All fields that are used in this
2276          * driver are compatible though, we don't need to convert the
2277          * contents.
2278          */
2279         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2280 }
2281 #endif /* CONFIG_COMPAT */
2282
2283 static int tun_chr_fasync(int fd, struct file *file, int on)
2284 {
2285         struct tun_file *tfile = file->private_data;
2286         int ret;
2287
2288         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2289                 goto out;
2290
2291         if (on) {
2292                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2293                 tfile->flags |= TUN_FASYNC;
2294         } else
2295                 tfile->flags &= ~TUN_FASYNC;
2296         ret = 0;
2297 out:
2298         return ret;
2299 }
2300
2301 static int tun_chr_open(struct inode *inode, struct file * file)
2302 {
2303         struct net *net = current->nsproxy->net_ns;
2304         struct tun_file *tfile;
2305
2306         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2307
2308         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2309                                             &tun_proto, 0);
2310         if (!tfile)
2311                 return -ENOMEM;
2312         RCU_INIT_POINTER(tfile->tun, NULL);
2313         tfile->flags = 0;
2314         tfile->ifindex = 0;
2315
2316         init_waitqueue_head(&tfile->wq.wait);
2317         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2318
2319         tfile->socket.file = file;
2320         tfile->socket.ops = &tun_socket_ops;
2321
2322         sock_init_data(&tfile->socket, &tfile->sk);
2323
2324         tfile->sk.sk_write_space = tun_sock_write_space;
2325         tfile->sk.sk_sndbuf = INT_MAX;
2326
2327         file->private_data = tfile;
2328         INIT_LIST_HEAD(&tfile->next);
2329
2330         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2331
2332         return 0;
2333 }
2334
2335 static int tun_chr_close(struct inode *inode, struct file *file)
2336 {
2337         struct tun_file *tfile = file->private_data;
2338
2339         tun_detach(tfile, true);
2340
2341         return 0;
2342 }
2343
2344 #ifdef CONFIG_PROC_FS
2345 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2346 {
2347         struct tun_struct *tun;
2348         struct ifreq ifr;
2349
2350         memset(&ifr, 0, sizeof(ifr));
2351
2352         rtnl_lock();
2353         tun = tun_get(f);
2354         if (tun)
2355                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2356         rtnl_unlock();
2357
2358         if (tun)
2359                 tun_put(tun);
2360
2361         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2362 }
2363 #endif
2364
2365 static const struct file_operations tun_fops = {
2366         .owner  = THIS_MODULE,
2367         .llseek = no_llseek,
2368         .read_iter  = tun_chr_read_iter,
2369         .write_iter = tun_chr_write_iter,
2370         .poll   = tun_chr_poll,
2371         .unlocked_ioctl = tun_chr_ioctl,
2372 #ifdef CONFIG_COMPAT
2373         .compat_ioctl = tun_chr_compat_ioctl,
2374 #endif
2375         .open   = tun_chr_open,
2376         .release = tun_chr_close,
2377         .fasync = tun_chr_fasync,
2378 #ifdef CONFIG_PROC_FS
2379         .show_fdinfo = tun_chr_show_fdinfo,
2380 #endif
2381 };
2382
2383 static struct miscdevice tun_miscdev = {
2384         .minor = TUN_MINOR,
2385         .name = "tun",
2386         .nodename = "net/tun",
2387         .fops = &tun_fops,
2388 };
2389
2390 /* ethtool interface */
2391
2392 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2393 {
2394         cmd->supported          = 0;
2395         cmd->advertising        = 0;
2396         ethtool_cmd_speed_set(cmd, SPEED_10);
2397         cmd->duplex             = DUPLEX_FULL;
2398         cmd->port               = PORT_TP;
2399         cmd->phy_address        = 0;
2400         cmd->transceiver        = XCVR_INTERNAL;
2401         cmd->autoneg            = AUTONEG_DISABLE;
2402         cmd->maxtxpkt           = 0;
2403         cmd->maxrxpkt           = 0;
2404         return 0;
2405 }
2406
2407 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2408 {
2409         struct tun_struct *tun = netdev_priv(dev);
2410
2411         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2412         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2413
2414         switch (tun->flags & TUN_TYPE_MASK) {
2415         case IFF_TUN:
2416                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2417                 break;
2418         case IFF_TAP:
2419                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2420                 break;
2421         }
2422 }
2423
2424 static u32 tun_get_msglevel(struct net_device *dev)
2425 {
2426 #ifdef TUN_DEBUG
2427         struct tun_struct *tun = netdev_priv(dev);
2428         return tun->debug;
2429 #else
2430         return -EOPNOTSUPP;
2431 #endif
2432 }
2433
2434 static void tun_set_msglevel(struct net_device *dev, u32 value)
2435 {
2436 #ifdef TUN_DEBUG
2437         struct tun_struct *tun = netdev_priv(dev);
2438         tun->debug = value;
2439 #endif
2440 }
2441
2442 static const struct ethtool_ops tun_ethtool_ops = {
2443         .get_settings   = tun_get_settings,
2444         .get_drvinfo    = tun_get_drvinfo,
2445         .get_msglevel   = tun_get_msglevel,
2446         .set_msglevel   = tun_set_msglevel,
2447         .get_link       = ethtool_op_get_link,
2448         .get_ts_info    = ethtool_op_get_ts_info,
2449 };
2450
2451 static int tun_queue_resize(struct tun_struct *tun)
2452 {
2453         struct net_device *dev = tun->dev;
2454         struct tun_file *tfile;
2455         struct skb_array **arrays;
2456         int n = tun->numqueues + tun->numdisabled;
2457         int ret, i;
2458
2459         arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL);
2460         if (!arrays)
2461                 return -ENOMEM;
2462
2463         for (i = 0; i < tun->numqueues; i++) {
2464                 tfile = rtnl_dereference(tun->tfiles[i]);
2465                 arrays[i] = &tfile->tx_array;
2466         }
2467         list_for_each_entry(tfile, &tun->disabled, next)
2468                 arrays[i++] = &tfile->tx_array;
2469
2470         ret = skb_array_resize_multiple(arrays, n,
2471                                         dev->tx_queue_len, GFP_KERNEL);
2472
2473         kfree(arrays);
2474         return ret;
2475 }
2476
2477 static int tun_device_event(struct notifier_block *unused,
2478                             unsigned long event, void *ptr)
2479 {
2480         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2481         struct tun_struct *tun = netdev_priv(dev);
2482
2483         if (dev->rtnl_link_ops != &tun_link_ops)
2484                 return NOTIFY_DONE;
2485
2486         switch (event) {
2487         case NETDEV_CHANGE_TX_QUEUE_LEN:
2488                 if (tun_queue_resize(tun))
2489                         return NOTIFY_BAD;
2490                 break;
2491         default:
2492                 break;
2493         }
2494
2495         return NOTIFY_DONE;
2496 }
2497
2498 static struct notifier_block tun_notifier_block __read_mostly = {
2499         .notifier_call  = tun_device_event,
2500 };
2501
2502 static int __init tun_init(void)
2503 {
2504         int ret = 0;
2505
2506         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2507         pr_info("%s\n", DRV_COPYRIGHT);
2508
2509         ret = rtnl_link_register(&tun_link_ops);
2510         if (ret) {
2511                 pr_err("Can't register link_ops\n");
2512                 goto err_linkops;
2513         }
2514
2515         ret = misc_register(&tun_miscdev);
2516         if (ret) {
2517                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2518                 goto err_misc;
2519         }
2520
2521         register_netdevice_notifier(&tun_notifier_block);
2522         return  0;
2523 err_misc:
2524         rtnl_link_unregister(&tun_link_ops);
2525 err_linkops:
2526         return ret;
2527 }
2528
2529 static void tun_cleanup(void)
2530 {
2531         misc_deregister(&tun_miscdev);
2532         rtnl_link_unregister(&tun_link_ops);
2533         unregister_netdevice_notifier(&tun_notifier_block);
2534 }
2535
2536 /* Get an underlying socket object from tun file.  Returns error unless file is
2537  * attached to a device.  The returned object works like a packet socket, it
2538  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2539  * holding a reference to the file for as long as the socket is in use. */
2540 struct socket *tun_get_socket(struct file *file)
2541 {
2542         struct tun_file *tfile;
2543         if (file->f_op != &tun_fops)
2544                 return ERR_PTR(-EINVAL);
2545         tfile = file->private_data;
2546         if (!tfile)
2547                 return ERR_PTR(-EBADFD);
2548         return &tfile->socket;
2549 }
2550 EXPORT_SYMBOL_GPL(tun_get_socket);
2551
2552 module_init(tun_init);
2553 module_exit(tun_cleanup);
2554 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2555 MODULE_AUTHOR(DRV_COPYRIGHT);
2556 MODULE_LICENSE("GPL");
2557 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2558 MODULE_ALIAS("devname:net/tun");