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         struct tun_struct *tun = netdev_priv(dev);
 823         int i;
 824
 825         netif_tx_start_all_queues(dev);
 826
 827         for (i = 0; i < tun->numqueues; i++) {
 828                 struct tun_file *tfile;
 829
 830                 tfile = rtnl_dereference(tun->tfiles[i]);
 831                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
 832         }
 833
 834         return 0;
 835 }
 836
 837 /* Net device close. */
 838 static int tun_net_close(struct net_device *dev)
 839 {
 840         netif_tx_stop_all_queues(dev);
 841         return 0;
 842 }
 843
 844 /* Net device start xmit */
 845 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 846 {
 847         struct tun_struct *tun = netdev_priv(dev);
 848         int txq = skb->queue_mapping;
 849         struct tun_file *tfile;
 850         u32 numqueues = 0;
 851
 852         rcu_read_lock();
 853         tfile = rcu_dereference(tun->tfiles[txq]);
 854         numqueues = ACCESS_ONCE(tun->numqueues);
 855
 856         /* Drop packet if interface is not attached */
 857         if (txq >= numqueues)
 858                 goto drop;
 859
 860 #ifdef CONFIG_RPS
 861         if (numqueues == 1 && static_key_false(&rps_needed)) {
 862                 /* Select queue was not called for the skbuff, so we extract the
 863                  * RPS hash and save it into the flow_table here.
 864                  */
 865                 __u32 rxhash;
 866
 867                 rxhash = skb_get_hash(skb);
 868                 if (rxhash) {
 869                         struct tun_flow_entry *e;
 870                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 871                                         rxhash);
 872                         if (e)
 873                                 tun_flow_save_rps_rxhash(e, rxhash);
 874                 }
 875         }
 876 #endif
 877
 878         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 879
 880         BUG_ON(!tfile);
 881
 882         /* Drop if the filter does not like it.
 883          * This is a noop if the filter is disabled.
 884          * Filter can be enabled only for the TAP devices. */
 885         if (!check_filter(&tun->txflt, skb))
 886                 goto drop;
 887
 888         if (tfile->socket.sk->sk_filter &&
 889             sk_filter(tfile->socket.sk, skb))
 890                 goto drop;
 891
 892         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 893                 goto drop;
 894
 895         skb_tx_timestamp(skb);
 896
 897         /* Orphan the skb - required as we might hang on to it
 898          * for indefinite time.
 899          */
 900         skb_orphan(skb);
 901
 902         nf_reset(skb);
 903
 904         if (skb_array_produce(&tfile->tx_array, skb))
 905                 goto drop;
 906
 907         /* Notify and wake up reader process */
 908         if (tfile->flags & TUN_FASYNC)
 909                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 910         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 911
 912         rcu_read_unlock();
 913         return NETDEV_TX_OK;
 914
 915 drop:
 916         this_cpu_inc(tun->pcpu_stats->tx_dropped);
 917         skb_tx_error(skb);
 918         kfree_skb(skb);
 919         rcu_read_unlock();
 920         return NET_XMIT_DROP;
 921 }
 922
 923 static void tun_net_mclist(struct net_device *dev)
 924 {
 925         /*
 926          * This callback is supposed to deal with mc filter in
 927          * _rx_ path and has nothing to do with the _tx_ path.
 928          * In rx path we always accept everything userspace gives us.
 929          */
 930 }
 931
 932 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 933         netdev_features_t features)
 934 {
 935         struct tun_struct *tun = netdev_priv(dev);
 936
 937         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 938 }
 939 #ifdef CONFIG_NET_POLL_CONTROLLER
 940 static void tun_poll_controller(struct net_device *dev)
 941 {
 942         /*
 943          * Tun only receives frames when:
 944          * 1) the char device endpoint gets data from user space
 945          * 2) the tun socket gets a sendmsg call from user space
 946          * Since both of those are synchronous operations, we are guaranteed
 947          * never to have pending data when we poll for it
 948          * so there is nothing to do here but return.
 949          * We need this though so netpoll recognizes us as an interface that
 950          * supports polling, which enables bridge devices in virt setups to
 951          * still use netconsole
 952          */
 953         return;
 954 }
 955 #endif
 956
 957 static void tun_set_headroom(struct net_device *dev, int new_hr)
 958 {
 959         struct tun_struct *tun = netdev_priv(dev);
 960
 961         if (new_hr < NET_SKB_PAD)
 962                 new_hr = NET_SKB_PAD;
 963
 964         tun->align = new_hr;
 965 }
 966
 967 static struct rtnl_link_stats64 *
 968 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
 969 {
 970         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
 971         struct tun_struct *tun = netdev_priv(dev);
 972         struct tun_pcpu_stats *p;
 973         int i;
 974
 975         for_each_possible_cpu(i) {
 976                 u64 rxpackets, rxbytes, txpackets, txbytes;
 977                 unsigned int start;
 978
 979                 p = per_cpu_ptr(tun->pcpu_stats, i);
 980                 do {
 981                         start = u64_stats_fetch_begin(&p->syncp);
 982                         rxpackets       = p->rx_packets;
 983                         rxbytes         = p->rx_bytes;
 984                         txpackets       = p->tx_packets;
 985                         txbytes         = p->tx_bytes;
 986                 } while (u64_stats_fetch_retry(&p->syncp, start));
 987
 988                 stats->rx_packets       += rxpackets;
 989                 stats->rx_bytes         += rxbytes;
 990                 stats->tx_packets       += txpackets;
 991                 stats->tx_bytes         += txbytes;
 992
 993                 /* u32 counters */
 994                 rx_dropped      += p->rx_dropped;
 995                 rx_frame_errors += p->rx_frame_errors;
 996                 tx_dropped      += p->tx_dropped;
 997         }
 998         stats->rx_dropped  = rx_dropped;
 999         stats->rx_frame_errors = rx_frame_errors;
1000         stats->tx_dropped = tx_dropped;
1001         return stats;
1002 }
1003
1004 static const struct net_device_ops tun_netdev_ops = {
1005         .ndo_uninit             = tun_net_uninit,
1006         .ndo_open               = tun_net_open,
1007         .ndo_stop               = tun_net_close,
1008         .ndo_start_xmit         = tun_net_xmit,
1009         .ndo_fix_features       = tun_net_fix_features,
1010         .ndo_select_queue       = tun_select_queue,
1011 #ifdef CONFIG_NET_POLL_CONTROLLER
1012         .ndo_poll_controller    = tun_poll_controller,
1013 #endif
1014         .ndo_set_rx_headroom    = tun_set_headroom,
1015         .ndo_get_stats64        = tun_net_get_stats64,
1016 };
1017
1018 static const struct net_device_ops tap_netdev_ops = {
1019         .ndo_uninit             = tun_net_uninit,
1020         .ndo_open               = tun_net_open,
1021         .ndo_stop               = tun_net_close,
1022         .ndo_start_xmit         = tun_net_xmit,
1023         .ndo_fix_features       = tun_net_fix_features,
1024         .ndo_set_rx_mode        = tun_net_mclist,
1025         .ndo_set_mac_address    = eth_mac_addr,
1026         .ndo_validate_addr      = eth_validate_addr,
1027         .ndo_select_queue       = tun_select_queue,
1028 #ifdef CONFIG_NET_POLL_CONTROLLER
1029         .ndo_poll_controller    = tun_poll_controller,
1030 #endif
1031         .ndo_features_check     = passthru_features_check,
1032         .ndo_set_rx_headroom    = tun_set_headroom,
1033         .ndo_get_stats64        = tun_net_get_stats64,
1034 };
1035
1036 static void tun_flow_init(struct tun_struct *tun)
1037 {
1038         int i;
1039
1040         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1041                 INIT_HLIST_HEAD(&tun->flows[i]);
1042
1043         tun->ageing_time = TUN_FLOW_EXPIRE;
1044         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1045         mod_timer(&tun->flow_gc_timer,
1046                   round_jiffies_up(jiffies + tun->ageing_time));
1047 }
1048
1049 static void tun_flow_uninit(struct tun_struct *tun)
1050 {
1051         del_timer_sync(&tun->flow_gc_timer);
1052         tun_flow_flush(tun);
1053 }
1054
1055 #define MIN_MTU 68
1056 #define MAX_MTU 65535
1057
1058 /* Initialize net device. */
1059 static void tun_net_init(struct net_device *dev)
1060 {
1061         struct tun_struct *tun = netdev_priv(dev);
1062
1063         switch (tun->flags & TUN_TYPE_MASK) {
1064         case IFF_TUN:
1065                 dev->netdev_ops = &tun_netdev_ops;
1066
1067                 /* Point-to-Point TUN Device */
1068                 dev->hard_header_len = 0;
1069                 dev->addr_len = 0;
1070                 dev->mtu = 1500;
1071
1072                 /* Zero header length */
1073                 dev->type = ARPHRD_NONE;
1074                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1075                 break;
1076
1077         case IFF_TAP:
1078                 dev->netdev_ops = &tap_netdev_ops;
1079                 /* Ethernet TAP Device */
1080                 ether_setup(dev);
1081                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1082                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1083
1084                 eth_hw_addr_random(dev);
1085
1086                 break;
1087         }
1088
1089         dev->min_mtu = MIN_MTU;
1090         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1091 }
1092
1093 /* Character device part */
1094
1095 /* Poll */
1096 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1097 {
1098         struct tun_file *tfile = file->private_data;
1099         struct tun_struct *tun = __tun_get(tfile);
1100         struct sock *sk;
1101         unsigned int mask = 0;
1102
1103         if (!tun)
1104                 return POLLERR;
1105
1106         sk = tfile->socket.sk;
1107
1108         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1109
1110         poll_wait(file, sk_sleep(sk), wait);
1111
1112         if (!skb_array_empty(&tfile->tx_array))
1113                 mask |= POLLIN | POLLRDNORM;
1114
1115         if (tun->dev->flags & IFF_UP &&
1116             (sock_writeable(sk) ||
1117              (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1118               sock_writeable(sk))))
1119                 mask |= POLLOUT | POLLWRNORM;
1120
1121         if (tun->dev->reg_state != NETREG_REGISTERED)
1122                 mask = POLLERR;
1123
1124         tun_put(tun);
1125         return mask;
1126 }
1127
1128 /* prepad is the amount to reserve at front.  len is length after that.
1129  * linear is a hint as to how much to copy (usually headers). */
1130 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1131                                      size_t prepad, size_t len,
1132                                      size_t linear, int noblock)
1133 {
1134         struct sock *sk = tfile->socket.sk;
1135         struct sk_buff *skb;
1136         int err;
1137
1138         /* Under a page?  Don't bother with paged skb. */
1139         if (prepad + len < PAGE_SIZE || !linear)
1140                 linear = len;
1141
1142         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1143                                    &err, 0);
1144         if (!skb)
1145                 return ERR_PTR(err);
1146
1147         skb_reserve(skb, prepad);
1148         skb_put(skb, linear);
1149         skb->data_len = len - linear;
1150         skb->len += len - linear;
1151
1152         return skb;
1153 }
1154
1155 /* Get packet from user space buffer */
1156 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1157                             void *msg_control, struct iov_iter *from,
1158                             int noblock)
1159 {
1160         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1161         struct sk_buff *skb;
1162         size_t total_len = iov_iter_count(from);
1163         size_t len = total_len, align = tun->align, linear;
1164         struct virtio_net_hdr gso = { 0 };
1165         struct tun_pcpu_stats *stats;
1166         int good_linear;
1167         int copylen;
1168         bool zerocopy = false;
1169         int err;
1170         u32 rxhash;
1171
1172         if (!(tun->dev->flags & IFF_UP))
1173                 return -EIO;
1174
1175         if (!(tun->flags & IFF_NO_PI)) {
1176                 if (len < sizeof(pi))
1177                         return -EINVAL;
1178                 len -= sizeof(pi);
1179
1180                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1181                         return -EFAULT;
1182         }
1183
1184         if (tun->flags & IFF_VNET_HDR) {
1185                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1186
1187                 if (len < vnet_hdr_sz)
1188                         return -EINVAL;
1189                 len -= vnet_hdr_sz;
1190
1191                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1192                         return -EFAULT;
1193
1194                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1195                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1196                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1197
1198                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1199                         return -EINVAL;
1200                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1201         }
1202
1203         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1204                 align += NET_IP_ALIGN;
1205                 if (unlikely(len < ETH_HLEN ||
1206                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1207                         return -EINVAL;
1208         }
1209
1210         good_linear = SKB_MAX_HEAD(align);
1211
1212         if (msg_control) {
1213                 struct iov_iter i = *from;
1214
1215                 /* There are 256 bytes to be copied in skb, so there is
1216                  * enough room for skb expand head in case it is used.
1217                  * The rest of the buffer is mapped from userspace.
1218                  */
1219                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1220                 if (copylen > good_linear)
1221                         copylen = good_linear;
1222                 linear = copylen;
1223                 iov_iter_advance(&i, copylen);
1224                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1225                         zerocopy = true;
1226         }
1227
1228         if (!zerocopy) {
1229                 copylen = len;
1230                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1231                         linear = good_linear;
1232                 else
1233                         linear = tun16_to_cpu(tun, gso.hdr_len);
1234         }
1235
1236         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1237         if (IS_ERR(skb)) {
1238                 if (PTR_ERR(skb) != -EAGAIN)
1239                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1240                 return PTR_ERR(skb);
1241         }
1242
1243         if (zerocopy)
1244                 err = zerocopy_sg_from_iter(skb, from);
1245         else
1246                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1247
1248         if (err) {
1249                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1250                 kfree_skb(skb);
1251                 return -EFAULT;
1252         }
1253
1254         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1255                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1256                 kfree_skb(skb);
1257                 return -EINVAL;
1258         }
1259
1260         switch (tun->flags & TUN_TYPE_MASK) {
1261         case IFF_TUN:
1262                 if (tun->flags & IFF_NO_PI) {
1263                         switch (skb->data[0] & 0xf0) {
1264                         case 0x40:
1265                                 pi.proto = htons(ETH_P_IP);
1266                                 break;
1267                         case 0x60:
1268                                 pi.proto = htons(ETH_P_IPV6);
1269                                 break;
1270                         default:
1271                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1272                                 kfree_skb(skb);
1273                                 return -EINVAL;
1274                         }
1275                 }
1276
1277                 skb_reset_mac_header(skb);
1278                 skb->protocol = pi.proto;
1279                 skb->dev = tun->dev;
1280                 break;
1281         case IFF_TAP:
1282                 skb->protocol = eth_type_trans(skb, tun->dev);
1283                 break;
1284         }
1285
1286         /* copy skb_ubuf_info for callback when skb has no error */
1287         if (zerocopy) {
1288                 skb_shinfo(skb)->destructor_arg = msg_control;
1289                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1290                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1291         } else if (msg_control) {
1292                 struct ubuf_info *uarg = msg_control;
1293                 uarg->callback(uarg, false);
1294         }
1295
1296         skb_reset_network_header(skb);
1297         skb_probe_transport_header(skb, 0);
1298
1299         rxhash = skb_get_hash(skb);
1300 #ifndef CONFIG_4KSTACKS
1301         local_bh_disable();
1302         netif_receive_skb(skb);
1303         local_bh_enable();
1304 #else
1305         netif_rx_ni(skb);
1306 #endif
1307
1308         stats = get_cpu_ptr(tun->pcpu_stats);
1309         u64_stats_update_begin(&stats->syncp);
1310         stats->rx_packets++;
1311         stats->rx_bytes += len;
1312         u64_stats_update_end(&stats->syncp);
1313         put_cpu_ptr(stats);
1314
1315         tun_flow_update(tun, rxhash, tfile);
1316         return total_len;
1317 }
1318
1319 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1320 {
1321         struct file *file = iocb->ki_filp;
1322         struct tun_struct *tun = tun_get(file);
1323         struct tun_file *tfile = file->private_data;
1324         ssize_t result;
1325
1326         if (!tun)
1327                 return -EBADFD;
1328
1329         result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1330
1331         tun_put(tun);
1332         return result;
1333 }
1334
1335 /* Put packet to the user space buffer */
1336 static ssize_t tun_put_user(struct tun_struct *tun,
1337                             struct tun_file *tfile,
1338                             struct sk_buff *skb,
1339                             struct iov_iter *iter)
1340 {
1341         struct tun_pi pi = { 0, skb->protocol };
1342         struct tun_pcpu_stats *stats;
1343         ssize_t total;
1344         int vlan_offset = 0;
1345         int vlan_hlen = 0;
1346         int vnet_hdr_sz = 0;
1347
1348         if (skb_vlan_tag_present(skb))
1349                 vlan_hlen = VLAN_HLEN;
1350
1351         if (tun->flags & IFF_VNET_HDR)
1352                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1353
1354         total = skb->len + vlan_hlen + vnet_hdr_sz;
1355
1356         if (!(tun->flags & IFF_NO_PI)) {
1357                 if (iov_iter_count(iter) < sizeof(pi))
1358                         return -EINVAL;
1359
1360                 total += sizeof(pi);
1361                 if (iov_iter_count(iter) < total) {
1362                         /* Packet will be striped */
1363                         pi.flags |= TUN_PKT_STRIP;
1364                 }
1365
1366                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1367                         return -EFAULT;
1368         }
1369
1370         if (vnet_hdr_sz) {
1371                 struct virtio_net_hdr gso;
1372
1373                 if (iov_iter_count(iter) < vnet_hdr_sz)
1374                         return -EINVAL;
1375
1376                 if (virtio_net_hdr_from_skb(skb, &gso,
1377                                             tun_is_little_endian(tun), true)) {
1378                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1379                         pr_err("unexpected GSO type: "
1380                                "0x%x, gso_size %d, hdr_len %d\n",
1381                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1382                                tun16_to_cpu(tun, gso.hdr_len));
1383                         print_hex_dump(KERN_ERR, "tun: ",
1384                                        DUMP_PREFIX_NONE,
1385                                        16, 1, skb->head,
1386                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1387                         WARN_ON_ONCE(1);
1388                         return -EINVAL;
1389                 }
1390
1391                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1392                         return -EFAULT;
1393
1394                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1395         }
1396
1397         if (vlan_hlen) {
1398                 int ret;
1399                 struct {
1400                         __be16 h_vlan_proto;
1401                         __be16 h_vlan_TCI;
1402                 } veth;
1403
1404                 veth.h_vlan_proto = skb->vlan_proto;
1405                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1406
1407                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1408
1409                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1410                 if (ret || !iov_iter_count(iter))
1411                         goto done;
1412
1413                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1414                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1415                         goto done;
1416         }
1417
1418         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1419
1420 done:
1421         /* caller is in process context, */
1422         stats = get_cpu_ptr(tun->pcpu_stats);
1423         u64_stats_update_begin(&stats->syncp);
1424         stats->tx_packets++;
1425         stats->tx_bytes += skb->len + vlan_hlen;
1426         u64_stats_update_end(&stats->syncp);
1427         put_cpu_ptr(tun->pcpu_stats);
1428
1429         return total;
1430 }
1431
1432 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1433                                      int *err)
1434 {
1435         DECLARE_WAITQUEUE(wait, current);
1436         struct sk_buff *skb = NULL;
1437         int error = 0;
1438
1439         skb = skb_array_consume(&tfile->tx_array);
1440         if (skb)
1441                 goto out;
1442         if (noblock) {
1443                 error = -EAGAIN;
1444                 goto out;
1445         }
1446
1447         add_wait_queue(&tfile->wq.wait, &wait);
1448         current->state = TASK_INTERRUPTIBLE;
1449
1450         while (1) {
1451                 skb = skb_array_consume(&tfile->tx_array);
1452                 if (skb)
1453                         break;
1454                 if (signal_pending(current)) {
1455                         error = -ERESTARTSYS;
1456                         break;
1457                 }
1458                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1459                         error = -EFAULT;
1460                         break;
1461                 }
1462
1463                 schedule();
1464         }
1465
1466         current->state = TASK_RUNNING;
1467         remove_wait_queue(&tfile->wq.wait, &wait);
1468
1469 out:
1470         *err = error;
1471         return skb;
1472 }
1473
1474 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1475                            struct iov_iter *to,
1476                            int noblock)
1477 {
1478         struct sk_buff *skb;
1479         ssize_t ret;
1480         int err;
1481
1482         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1483
1484         if (!iov_iter_count(to))
1485                 return 0;
1486
1487         /* Read frames from ring */
1488         skb = tun_ring_recv(tfile, noblock, &err);
1489         if (!skb)
1490                 return err;
1491
1492         ret = tun_put_user(tun, tfile, skb, to);
1493         if (unlikely(ret < 0))
1494                 kfree_skb(skb);
1495         else
1496                 consume_skb(skb);
1497
1498         return ret;
1499 }
1500
1501 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1502 {
1503         struct file *file = iocb->ki_filp;
1504         struct tun_file *tfile = file->private_data;
1505         struct tun_struct *tun = __tun_get(tfile);
1506         ssize_t len = iov_iter_count(to), ret;
1507
1508         if (!tun)
1509                 return -EBADFD;
1510         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1511         ret = min_t(ssize_t, ret, len);
1512         if (ret > 0)
1513                 iocb->ki_pos = ret;
1514         tun_put(tun);
1515         return ret;
1516 }
1517
1518 static void tun_free_netdev(struct net_device *dev)
1519 {
1520         struct tun_struct *tun = netdev_priv(dev);
1521
1522         BUG_ON(!(list_empty(&tun->disabled)));
1523         free_percpu(tun->pcpu_stats);
1524         tun_flow_uninit(tun);
1525         security_tun_dev_free_security(tun->security);
1526         free_netdev(dev);
1527 }
1528
1529 static void tun_setup(struct net_device *dev)
1530 {
1531         struct tun_struct *tun = netdev_priv(dev);
1532
1533         tun->owner = INVALID_UID;
1534         tun->group = INVALID_GID;
1535
1536         dev->ethtool_ops = &tun_ethtool_ops;
1537         dev->destructor = tun_free_netdev;
1538         /* We prefer our own queue length */
1539         dev->tx_queue_len = TUN_READQ_SIZE;
1540 }
1541
1542 /* Trivial set of netlink ops to allow deleting tun or tap
1543  * device with netlink.
1544  */
1545 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1546 {
1547         return -EINVAL;
1548 }
1549
1550 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1551         .kind           = DRV_NAME,
1552         .priv_size      = sizeof(struct tun_struct),
1553         .setup          = tun_setup,
1554         .validate       = tun_validate,
1555 };
1556
1557 static void tun_sock_write_space(struct sock *sk)
1558 {
1559         struct tun_file *tfile;
1560         wait_queue_head_t *wqueue;
1561
1562         if (!sock_writeable(sk))
1563                 return;
1564
1565         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1566                 return;
1567
1568         wqueue = sk_sleep(sk);
1569         if (wqueue && waitqueue_active(wqueue))
1570                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1571                                                 POLLWRNORM | POLLWRBAND);
1572
1573         tfile = container_of(sk, struct tun_file, sk);
1574         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1575 }
1576
1577 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1578 {
1579         int ret;
1580         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1581         struct tun_struct *tun = __tun_get(tfile);
1582
1583         if (!tun)
1584                 return -EBADFD;
1585
1586         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1587                            m->msg_flags & MSG_DONTWAIT);
1588         tun_put(tun);
1589         return ret;
1590 }
1591
1592 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1593                        int flags)
1594 {
1595         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1596         struct tun_struct *tun = __tun_get(tfile);
1597         int ret;
1598
1599         if (!tun)
1600                 return -EBADFD;
1601
1602         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1603                 ret = -EINVAL;
1604                 goto out;
1605         }
1606         if (flags & MSG_ERRQUEUE) {
1607                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1608                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1609                 goto out;
1610         }
1611         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1612         if (ret > (ssize_t)total_len) {
1613                 m->msg_flags |= MSG_TRUNC;
1614                 ret = flags & MSG_TRUNC ? ret : total_len;
1615         }
1616 out:
1617         tun_put(tun);
1618         return ret;
1619 }
1620
1621 static int tun_peek_len(struct socket *sock)
1622 {
1623         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1624         struct tun_struct *tun;
1625         int ret = 0;
1626
1627         tun = __tun_get(tfile);
1628         if (!tun)
1629                 return 0;
1630
1631         ret = skb_array_peek_len(&tfile->tx_array);
1632         tun_put(tun);
1633
1634         return ret;
1635 }
1636
1637 /* Ops structure to mimic raw sockets with tun */
1638 static const struct proto_ops tun_socket_ops = {
1639         .peek_len = tun_peek_len,
1640         .sendmsg = tun_sendmsg,
1641         .recvmsg = tun_recvmsg,
1642 };
1643
1644 static struct proto tun_proto = {
1645         .name           = "tun",
1646         .owner          = THIS_MODULE,
1647         .obj_size       = sizeof(struct tun_file),
1648 };
1649
1650 static int tun_flags(struct tun_struct *tun)
1651 {
1652         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1653 }
1654
1655 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1656                               char *buf)
1657 {
1658         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1659         return sprintf(buf, "0x%x\n", tun_flags(tun));
1660 }
1661
1662 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1663                               char *buf)
1664 {
1665         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1666         return uid_valid(tun->owner)?
1667                 sprintf(buf, "%u\n",
1668                         from_kuid_munged(current_user_ns(), tun->owner)):
1669                 sprintf(buf, "-1\n");
1670 }
1671
1672 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1673                               char *buf)
1674 {
1675         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1676         return gid_valid(tun->group) ?
1677                 sprintf(buf, "%u\n",
1678                         from_kgid_munged(current_user_ns(), tun->group)):
1679                 sprintf(buf, "-1\n");
1680 }
1681
1682 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1683 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1684 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1685
1686 static struct attribute *tun_dev_attrs[] = {
1687         &dev_attr_tun_flags.attr,
1688         &dev_attr_owner.attr,
1689         &dev_attr_group.attr,
1690         NULL
1691 };
1692
1693 static const struct attribute_group tun_attr_group = {
1694         .attrs = tun_dev_attrs
1695 };
1696
1697 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1698 {
1699         struct tun_struct *tun;
1700         struct tun_file *tfile = file->private_data;
1701         struct net_device *dev;
1702         int err;
1703
1704         if (tfile->detached)
1705                 return -EINVAL;
1706
1707         dev = __dev_get_by_name(net, ifr->ifr_name);
1708         if (dev) {
1709                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1710                         return -EBUSY;
1711                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1712                         tun = netdev_priv(dev);
1713                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1714                         tun = netdev_priv(dev);
1715                 else
1716                         return -EINVAL;
1717
1718                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1719                     !!(tun->flags & IFF_MULTI_QUEUE))
1720                         return -EINVAL;
1721
1722                 if (tun_not_capable(tun))
1723                         return -EPERM;
1724                 err = security_tun_dev_open(tun->security);
1725                 if (err < 0)
1726                         return err;
1727
1728                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1729                 if (err < 0)
1730                         return err;
1731
1732                 if (tun->flags & IFF_MULTI_QUEUE &&
1733                     (tun->numqueues + tun->numdisabled > 1)) {
1734                         /* One or more queue has already been attached, no need
1735                          * to initialize the device again.
1736                          */
1737                         return 0;
1738                 }
1739         }
1740         else {
1741                 char *name;
1742                 unsigned long flags = 0;
1743                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1744                              MAX_TAP_QUEUES : 1;
1745
1746                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1747                         return -EPERM;
1748                 err = security_tun_dev_create();
1749                 if (err < 0)
1750                         return err;
1751
1752                 /* Set dev type */
1753                 if (ifr->ifr_flags & IFF_TUN) {
1754                         /* TUN device */
1755                         flags |= IFF_TUN;
1756                         name = "tun%d";
1757                 } else if (ifr->ifr_flags & IFF_TAP) {
1758                         /* TAP device */
1759                         flags |= IFF_TAP;
1760                         name = "tap%d";
1761                 } else
1762                         return -EINVAL;
1763
1764                 if (*ifr->ifr_name)
1765                         name = ifr->ifr_name;
1766
1767                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1768                                        NET_NAME_UNKNOWN, tun_setup, queues,
1769                                        queues);
1770
1771                 if (!dev)
1772                         return -ENOMEM;
1773
1774                 dev_net_set(dev, net);
1775                 dev->rtnl_link_ops = &tun_link_ops;
1776                 dev->ifindex = tfile->ifindex;
1777                 dev->sysfs_groups[0] = &tun_attr_group;
1778
1779                 tun = netdev_priv(dev);
1780                 tun->dev = dev;
1781                 tun->flags = flags;
1782                 tun->txflt.count = 0;
1783                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1784
1785                 tun->align = NET_SKB_PAD;
1786                 tun->filter_attached = false;
1787                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1788
1789                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
1790                 if (!tun->pcpu_stats) {
1791                         err = -ENOMEM;
1792                         goto err_free_dev;
1793                 }
1794
1795                 spin_lock_init(&tun->lock);
1796
1797                 err = security_tun_dev_alloc_security(&tun->security);
1798                 if (err < 0)
1799                         goto err_free_stat;
1800
1801                 tun_net_init(dev);
1802                 tun_flow_init(tun);
1803
1804                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1805                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1806                                    NETIF_F_HW_VLAN_STAG_TX;
1807                 dev->features = dev->hw_features | NETIF_F_LLTX;
1808                 dev->vlan_features = dev->features &
1809                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1810                                        NETIF_F_HW_VLAN_STAG_TX);
1811
1812                 INIT_LIST_HEAD(&tun->disabled);
1813                 err = tun_attach(tun, file, false);
1814                 if (err < 0)
1815                         goto err_free_flow;
1816
1817                 err = register_netdevice(tun->dev);
1818                 if (err < 0)
1819                         goto err_detach;
1820         }
1821
1822         netif_carrier_on(tun->dev);
1823
1824         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1825
1826         tun->flags = (tun->flags & ~TUN_FEATURES) |
1827                 (ifr->ifr_flags & TUN_FEATURES);
1828
1829         /* Make sure persistent devices do not get stuck in
1830          * xoff state.
1831          */
1832         if (netif_running(tun->dev))
1833                 netif_tx_wake_all_queues(tun->dev);
1834
1835         strcpy(ifr->ifr_name, tun->dev->name);
1836         return 0;
1837
1838 err_detach:
1839         tun_detach_all(dev);
1840 err_free_flow:
1841         tun_flow_uninit(tun);
1842         security_tun_dev_free_security(tun->security);
1843 err_free_stat:
1844         free_percpu(tun->pcpu_stats);
1845 err_free_dev:
1846         free_netdev(dev);
1847         return err;
1848 }
1849
1850 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1851                        struct ifreq *ifr)
1852 {
1853         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1854
1855         strcpy(ifr->ifr_name, tun->dev->name);
1856
1857         ifr->ifr_flags = tun_flags(tun);
1858
1859 }
1860
1861 /* This is like a cut-down ethtool ops, except done via tun fd so no
1862  * privs required. */
1863 static int set_offload(struct tun_struct *tun, unsigned long arg)
1864 {
1865         netdev_features_t features = 0;
1866
1867         if (arg & TUN_F_CSUM) {
1868                 features |= NETIF_F_HW_CSUM;
1869                 arg &= ~TUN_F_CSUM;
1870
1871                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1872                         if (arg & TUN_F_TSO_ECN) {
1873                                 features |= NETIF_F_TSO_ECN;
1874                                 arg &= ~TUN_F_TSO_ECN;
1875                         }
1876                         if (arg & TUN_F_TSO4)
1877                                 features |= NETIF_F_TSO;
1878                         if (arg & TUN_F_TSO6)
1879                                 features |= NETIF_F_TSO6;
1880                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1881                 }
1882
1883                 if (arg & TUN_F_UFO) {
1884                         features |= NETIF_F_UFO;
1885                         arg &= ~TUN_F_UFO;
1886                 }
1887         }
1888
1889         /* This gives the user a way to test for new features in future by
1890          * trying to set them. */
1891         if (arg)
1892                 return -EINVAL;
1893
1894         tun->set_features = features;
1895         netdev_update_features(tun->dev);
1896
1897         return 0;
1898 }
1899
1900 static void tun_detach_filter(struct tun_struct *tun, int n)
1901 {
1902         int i;
1903         struct tun_file *tfile;
1904
1905         for (i = 0; i < n; i++) {
1906                 tfile = rtnl_dereference(tun->tfiles[i]);
1907                 lock_sock(tfile->socket.sk);
1908                 sk_detach_filter(tfile->socket.sk);
1909                 release_sock(tfile->socket.sk);
1910         }
1911
1912         tun->filter_attached = false;
1913 }
1914
1915 static int tun_attach_filter(struct tun_struct *tun)
1916 {
1917         int i, ret = 0;
1918         struct tun_file *tfile;
1919
1920         for (i = 0; i < tun->numqueues; i++) {
1921                 tfile = rtnl_dereference(tun->tfiles[i]);
1922                 lock_sock(tfile->socket.sk);
1923                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1924                 release_sock(tfile->socket.sk);
1925                 if (ret) {
1926                         tun_detach_filter(tun, i);
1927                         return ret;
1928                 }
1929         }
1930
1931         tun->filter_attached = true;
1932         return ret;
1933 }
1934
1935 static void tun_set_sndbuf(struct tun_struct *tun)
1936 {
1937         struct tun_file *tfile;
1938         int i;
1939
1940         for (i = 0; i < tun->numqueues; i++) {
1941                 tfile = rtnl_dereference(tun->tfiles[i]);
1942                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1943         }
1944 }
1945
1946 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1947 {
1948         struct tun_file *tfile = file->private_data;
1949         struct tun_struct *tun;
1950         int ret = 0;
1951
1952         rtnl_lock();
1953
1954         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1955                 tun = tfile->detached;
1956                 if (!tun) {
1957                         ret = -EINVAL;
1958                         goto unlock;
1959                 }
1960                 ret = security_tun_dev_attach_queue(tun->security);
1961                 if (ret < 0)
1962                         goto unlock;
1963                 ret = tun_attach(tun, file, false);
1964         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1965                 tun = rtnl_dereference(tfile->tun);
1966                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1967                         ret = -EINVAL;
1968                 else
1969                         __tun_detach(tfile, false);
1970         } else
1971                 ret = -EINVAL;
1972
1973 unlock:
1974         rtnl_unlock();
1975         return ret;
1976 }
1977
1978 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1979                             unsigned long arg, int ifreq_len)
1980 {
1981         struct tun_file *tfile = file->private_data;
1982         struct tun_struct *tun;
1983         void __user* argp = (void __user*)arg;
1984         struct ifreq ifr;
1985         kuid_t owner;
1986         kgid_t group;
1987         int sndbuf;
1988         int vnet_hdr_sz;
1989         unsigned int ifindex;
1990         int le;
1991         int ret;
1992
1993         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
1994                 if (copy_from_user(&ifr, argp, ifreq_len))
1995                         return -EFAULT;
1996         } else {
1997                 memset(&ifr, 0, sizeof(ifr));
1998         }
1999         if (cmd == TUNGETFEATURES) {
2000                 /* Currently this just means: "what IFF flags are valid?".
2001                  * This is needed because we never checked for invalid flags on
2002                  * TUNSETIFF.
2003                  */
2004                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2005                                 (unsigned int __user*)argp);
2006         } else if (cmd == TUNSETQUEUE)
2007                 return tun_set_queue(file, &ifr);
2008
2009         ret = 0;
2010         rtnl_lock();
2011
2012         tun = __tun_get(tfile);
2013         if (cmd == TUNSETIFF) {
2014                 ret = -EEXIST;
2015                 if (tun)
2016                         goto unlock;
2017
2018                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2019
2020                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2021
2022                 if (ret)
2023                         goto unlock;
2024
2025                 if (copy_to_user(argp, &ifr, ifreq_len))
2026                         ret = -EFAULT;
2027                 goto unlock;
2028         }
2029         if (cmd == TUNSETIFINDEX) {
2030                 ret = -EPERM;
2031                 if (tun)
2032                         goto unlock;
2033
2034                 ret = -EFAULT;
2035                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2036                         goto unlock;
2037
2038                 ret = 0;
2039                 tfile->ifindex = ifindex;
2040                 goto unlock;
2041         }
2042
2043         ret = -EBADFD;
2044         if (!tun)
2045                 goto unlock;
2046
2047         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2048
2049         ret = 0;
2050         switch (cmd) {
2051         case TUNGETIFF:
2052                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2053
2054                 if (tfile->detached)
2055                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2056                 if (!tfile->socket.sk->sk_filter)
2057                         ifr.ifr_flags |= IFF_NOFILTER;
2058
2059                 if (copy_to_user(argp, &ifr, ifreq_len))
2060                         ret = -EFAULT;
2061                 break;
2062
2063         case TUNSETNOCSUM:
2064                 /* Disable/Enable checksum */
2065
2066                 /* [unimplemented] */
2067                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2068                           arg ? "disabled" : "enabled");
2069                 break;
2070
2071         case TUNSETPERSIST:
2072                 /* Disable/Enable persist mode. Keep an extra reference to the
2073                  * module to prevent the module being unprobed.
2074                  */
2075                 if (arg && !(tun->flags & IFF_PERSIST)) {
2076                         tun->flags |= IFF_PERSIST;
2077                         __module_get(THIS_MODULE);
2078                 }
2079                 if (!arg && (tun->flags & IFF_PERSIST)) {
2080                         tun->flags &= ~IFF_PERSIST;
2081                         module_put(THIS_MODULE);
2082                 }
2083
2084                 tun_debug(KERN_INFO, tun, "persist %s\n",
2085                           arg ? "enabled" : "disabled");
2086                 break;
2087
2088         case TUNSETOWNER:
2089                 /* Set owner of the device */
2090                 owner = make_kuid(current_user_ns(), arg);
2091                 if (!uid_valid(owner)) {
2092                         ret = -EINVAL;
2093                         break;
2094                 }
2095                 tun->owner = owner;
2096                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2097                           from_kuid(&init_user_ns, tun->owner));
2098                 break;
2099
2100         case TUNSETGROUP:
2101                 /* Set group of the device */
2102                 group = make_kgid(current_user_ns(), arg);
2103                 if (!gid_valid(group)) {
2104                         ret = -EINVAL;
2105                         break;
2106                 }
2107                 tun->group = group;
2108                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2109                           from_kgid(&init_user_ns, tun->group));
2110                 break;
2111
2112         case TUNSETLINK:
2113                 /* Only allow setting the type when the interface is down */
2114                 if (tun->dev->flags & IFF_UP) {
2115                         tun_debug(KERN_INFO, tun,
2116                                   "Linktype set failed because interface is up\n");
2117                         ret = -EBUSY;
2118                 } else {
2119                         tun->dev->type = (int) arg;
2120                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2121                                   tun->dev->type);
2122                         ret = 0;
2123                 }
2124                 break;
2125
2126 #ifdef TUN_DEBUG
2127         case TUNSETDEBUG:
2128                 tun->debug = arg;
2129                 break;
2130 #endif
2131         case TUNSETOFFLOAD:
2132                 ret = set_offload(tun, arg);
2133                 break;
2134
2135         case TUNSETTXFILTER:
2136                 /* Can be set only for TAPs */
2137                 ret = -EINVAL;
2138                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2139                         break;
2140                 ret = update_filter(&tun->txflt, (void __user *)arg);
2141                 break;
2142
2143         case SIOCGIFHWADDR:
2144                 /* Get hw address */
2145                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2146                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2147                 if (copy_to_user(argp, &ifr, ifreq_len))
2148                         ret = -EFAULT;
2149                 break;
2150
2151         case SIOCSIFHWADDR:
2152                 /* Set hw address */
2153                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2154                           ifr.ifr_hwaddr.sa_data);
2155
2156                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2157                 break;
2158
2159         case TUNGETSNDBUF:
2160                 sndbuf = tfile->socket.sk->sk_sndbuf;
2161                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2162                         ret = -EFAULT;
2163                 break;
2164
2165         case TUNSETSNDBUF:
2166                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2167                         ret = -EFAULT;
2168                         break;
2169                 }
2170
2171                 tun->sndbuf = sndbuf;
2172                 tun_set_sndbuf(tun);
2173                 break;
2174
2175         case TUNGETVNETHDRSZ:
2176                 vnet_hdr_sz = tun->vnet_hdr_sz;
2177                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2178                         ret = -EFAULT;
2179                 break;
2180
2181         case TUNSETVNETHDRSZ:
2182                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2183                         ret = -EFAULT;
2184                         break;
2185                 }
2186                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2187                         ret = -EINVAL;
2188                         break;
2189                 }
2190
2191                 tun->vnet_hdr_sz = vnet_hdr_sz;
2192                 break;
2193
2194         case TUNGETVNETLE:
2195                 le = !!(tun->flags & TUN_VNET_LE);
2196                 if (put_user(le, (int __user *)argp))
2197                         ret = -EFAULT;
2198                 break;
2199
2200         case TUNSETVNETLE:
2201                 if (get_user(le, (int __user *)argp)) {
2202                         ret = -EFAULT;
2203                         break;
2204                 }
2205                 if (le)
2206                         tun->flags |= TUN_VNET_LE;
2207                 else
2208                         tun->flags &= ~TUN_VNET_LE;
2209                 break;
2210
2211         case TUNGETVNETBE:
2212                 ret = tun_get_vnet_be(tun, argp);
2213                 break;
2214
2215         case TUNSETVNETBE:
2216                 ret = tun_set_vnet_be(tun, argp);
2217                 break;
2218
2219         case TUNATTACHFILTER:
2220                 /* Can be set only for TAPs */
2221                 ret = -EINVAL;
2222                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2223                         break;
2224                 ret = -EFAULT;
2225                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2226                         break;
2227
2228                 ret = tun_attach_filter(tun);
2229                 break;
2230
2231         case TUNDETACHFILTER:
2232                 /* Can be set only for TAPs */
2233                 ret = -EINVAL;
2234                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2235                         break;
2236                 ret = 0;
2237                 tun_detach_filter(tun, tun->numqueues);
2238                 break;
2239
2240         case TUNGETFILTER:
2241                 ret = -EINVAL;
2242                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2243                         break;
2244                 ret = -EFAULT;
2245                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2246                         break;
2247                 ret = 0;
2248                 break;
2249
2250         default:
2251                 ret = -EINVAL;
2252                 break;
2253         }
2254
2255 unlock:
2256         rtnl_unlock();
2257         if (tun)
2258                 tun_put(tun);
2259         return ret;
2260 }
2261
2262 static long tun_chr_ioctl(struct file *file,
2263                           unsigned int cmd, unsigned long arg)
2264 {
2265         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2266 }
2267
2268 #ifdef CONFIG_COMPAT
2269 static long tun_chr_compat_ioctl(struct file *file,
2270                          unsigned int cmd, unsigned long arg)
2271 {
2272         switch (cmd) {
2273         case TUNSETIFF:
2274         case TUNGETIFF:
2275         case TUNSETTXFILTER:
2276         case TUNGETSNDBUF:
2277         case TUNSETSNDBUF:
2278         case SIOCGIFHWADDR:
2279         case SIOCSIFHWADDR:
2280                 arg = (unsigned long)compat_ptr(arg);
2281                 break;
2282         default:
2283                 arg = (compat_ulong_t)arg;
2284                 break;
2285         }
2286
2287         /*
2288          * compat_ifreq is shorter than ifreq, so we must not access beyond
2289          * the end of that structure. All fields that are used in this
2290          * driver are compatible though, we don't need to convert the
2291          * contents.
2292          */
2293         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2294 }
2295 #endif /* CONFIG_COMPAT */
2296
2297 static int tun_chr_fasync(int fd, struct file *file, int on)
2298 {
2299         struct tun_file *tfile = file->private_data;
2300         int ret;
2301
2302         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2303                 goto out;
2304
2305         if (on) {
2306                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2307                 tfile->flags |= TUN_FASYNC;
2308         } else
2309                 tfile->flags &= ~TUN_FASYNC;
2310         ret = 0;
2311 out:
2312         return ret;
2313 }
2314
2315 static int tun_chr_open(struct inode *inode, struct file * file)
2316 {
2317         struct net *net = current->nsproxy->net_ns;
2318         struct tun_file *tfile;
2319
2320         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2321
2322         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2323                                             &tun_proto, 0);
2324         if (!tfile)
2325                 return -ENOMEM;
2326         RCU_INIT_POINTER(tfile->tun, NULL);
2327         tfile->flags = 0;
2328         tfile->ifindex = 0;
2329
2330         init_waitqueue_head(&tfile->wq.wait);
2331         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2332
2333         tfile->socket.file = file;
2334         tfile->socket.ops = &tun_socket_ops;
2335
2336         sock_init_data(&tfile->socket, &tfile->sk);
2337
2338         tfile->sk.sk_write_space = tun_sock_write_space;
2339         tfile->sk.sk_sndbuf = INT_MAX;
2340
2341         file->private_data = tfile;
2342         INIT_LIST_HEAD(&tfile->next);
2343
2344         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2345
2346         return 0;
2347 }
2348
2349 static int tun_chr_close(struct inode *inode, struct file *file)
2350 {
2351         struct tun_file *tfile = file->private_data;
2352
2353         tun_detach(tfile, true);
2354
2355         return 0;
2356 }
2357
2358 #ifdef CONFIG_PROC_FS
2359 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2360 {
2361         struct tun_struct *tun;
2362         struct ifreq ifr;
2363
2364         memset(&ifr, 0, sizeof(ifr));
2365
2366         rtnl_lock();
2367         tun = tun_get(f);
2368         if (tun)
2369                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2370         rtnl_unlock();
2371
2372         if (tun)
2373                 tun_put(tun);
2374
2375         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2376 }
2377 #endif
2378
2379 static const struct file_operations tun_fops = {
2380         .owner  = THIS_MODULE,
2381         .llseek = no_llseek,
2382         .read_iter  = tun_chr_read_iter,
2383         .write_iter = tun_chr_write_iter,
2384         .poll   = tun_chr_poll,
2385         .unlocked_ioctl = tun_chr_ioctl,
2386 #ifdef CONFIG_COMPAT
2387         .compat_ioctl = tun_chr_compat_ioctl,
2388 #endif
2389         .open   = tun_chr_open,
2390         .release = tun_chr_close,
2391         .fasync = tun_chr_fasync,
2392 #ifdef CONFIG_PROC_FS
2393         .show_fdinfo = tun_chr_show_fdinfo,
2394 #endif
2395 };
2396
2397 static struct miscdevice tun_miscdev = {
2398         .minor = TUN_MINOR,
2399         .name = "tun",
2400         .nodename = "net/tun",
2401         .fops = &tun_fops,
2402 };
2403
2404 /* ethtool interface */
2405
2406 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2407 {
2408         cmd->supported          = 0;
2409         cmd->advertising        = 0;
2410         ethtool_cmd_speed_set(cmd, SPEED_10);
2411         cmd->duplex             = DUPLEX_FULL;
2412         cmd->port               = PORT_TP;
2413         cmd->phy_address        = 0;
2414         cmd->transceiver        = XCVR_INTERNAL;
2415         cmd->autoneg            = AUTONEG_DISABLE;
2416         cmd->maxtxpkt           = 0;
2417         cmd->maxrxpkt           = 0;
2418         return 0;
2419 }
2420
2421 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2422 {
2423         struct tun_struct *tun = netdev_priv(dev);
2424
2425         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2426         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2427
2428         switch (tun->flags & TUN_TYPE_MASK) {
2429         case IFF_TUN:
2430                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2431                 break;
2432         case IFF_TAP:
2433                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2434                 break;
2435         }
2436 }
2437
2438 static u32 tun_get_msglevel(struct net_device *dev)
2439 {
2440 #ifdef TUN_DEBUG
2441         struct tun_struct *tun = netdev_priv(dev);
2442         return tun->debug;
2443 #else
2444         return -EOPNOTSUPP;
2445 #endif
2446 }
2447
2448 static void tun_set_msglevel(struct net_device *dev, u32 value)
2449 {
2450 #ifdef TUN_DEBUG
2451         struct tun_struct *tun = netdev_priv(dev);
2452         tun->debug = value;
2453 #endif
2454 }
2455
2456 static const struct ethtool_ops tun_ethtool_ops = {
2457         .get_settings   = tun_get_settings,
2458         .get_drvinfo    = tun_get_drvinfo,
2459         .get_msglevel   = tun_get_msglevel,
2460         .set_msglevel   = tun_set_msglevel,
2461         .get_link       = ethtool_op_get_link,
2462         .get_ts_info    = ethtool_op_get_ts_info,
2463 };
2464
2465 static int tun_queue_resize(struct tun_struct *tun)
2466 {
2467         struct net_device *dev = tun->dev;
2468         struct tun_file *tfile;
2469         struct skb_array **arrays;
2470         int n = tun->numqueues + tun->numdisabled;
2471         int ret, i;
2472
2473         arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL);
2474         if (!arrays)
2475                 return -ENOMEM;
2476
2477         for (i = 0; i < tun->numqueues; i++) {
2478                 tfile = rtnl_dereference(tun->tfiles[i]);
2479                 arrays[i] = &tfile->tx_array;
2480         }
2481         list_for_each_entry(tfile, &tun->disabled, next)
2482                 arrays[i++] = &tfile->tx_array;
2483
2484         ret = skb_array_resize_multiple(arrays, n,
2485                                         dev->tx_queue_len, GFP_KERNEL);
2486
2487         kfree(arrays);
2488         return ret;
2489 }
2490
2491 static int tun_device_event(struct notifier_block *unused,
2492                             unsigned long event, void *ptr)
2493 {
2494         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2495         struct tun_struct *tun = netdev_priv(dev);
2496
2497         if (dev->rtnl_link_ops != &tun_link_ops)
2498                 return NOTIFY_DONE;
2499
2500         switch (event) {
2501         case NETDEV_CHANGE_TX_QUEUE_LEN:
2502                 if (tun_queue_resize(tun))
2503                         return NOTIFY_BAD;
2504                 break;
2505         default:
2506                 break;
2507         }
2508
2509         return NOTIFY_DONE;
2510 }
2511
2512 static struct notifier_block tun_notifier_block __read_mostly = {
2513         .notifier_call  = tun_device_event,
2514 };
2515
2516 static int __init tun_init(void)
2517 {
2518         int ret = 0;
2519
2520         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2521         pr_info("%s\n", DRV_COPYRIGHT);
2522
2523         ret = rtnl_link_register(&tun_link_ops);
2524         if (ret) {
2525                 pr_err("Can't register link_ops\n");
2526                 goto err_linkops;
2527         }
2528
2529         ret = misc_register(&tun_miscdev);
2530         if (ret) {
2531                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2532                 goto err_misc;
2533         }
2534
2535         register_netdevice_notifier(&tun_notifier_block);
2536         return  0;
2537 err_misc:
2538         rtnl_link_unregister(&tun_link_ops);
2539 err_linkops:
2540         return ret;
2541 }
2542
2543 static void tun_cleanup(void)
2544 {
2545         misc_deregister(&tun_miscdev);
2546         rtnl_link_unregister(&tun_link_ops);
2547         unregister_netdevice_notifier(&tun_notifier_block);
2548 }
2549
2550 /* Get an underlying socket object from tun file.  Returns error unless file is
2551  * attached to a device.  The returned object works like a packet socket, it
2552  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2553  * holding a reference to the file for as long as the socket is in use. */
2554 struct socket *tun_get_socket(struct file *file)
2555 {
2556         struct tun_file *tfile;
2557         if (file->f_op != &tun_fops)
2558                 return ERR_PTR(-EINVAL);
2559         tfile = file->private_data;
2560         if (!tfile)
2561                 return ERR_PTR(-EBADFD);
2562         return &tfile->socket;
2563 }
2564 EXPORT_SYMBOL_GPL(tun_get_socket);
2565
2566 module_init(tun_init);
2567 module_exit(tun_cleanup);
2568 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2569 MODULE_AUTHOR(DRV_COPYRIGHT);
2570 MODULE_LICENSE("GPL");
2571 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2572 MODULE_ALIAS("devname:net/tun");