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/sched/signal.h>
  48 #include <linux/major.h>
  49 #include <linux/slab.h>
  50 #include <linux/poll.h>
  51 #include <linux/fcntl.h>
  52 #include <linux/init.h>
  53 #include <linux/skbuff.h>
  54 #include <linux/netdevice.h>
  55 #include <linux/etherdevice.h>
  56 #include <linux/miscdevice.h>
  57 #include <linux/ethtool.h>
  58 #include <linux/rtnetlink.h>
  59 #include <linux/compat.h>
  60 #include <linux/if.h>
  61 #include <linux/if_arp.h>
  62 #include <linux/if_ether.h>
  63 #include <linux/if_tun.h>
  64 #include <linux/if_vlan.h>
  65 #include <linux/crc32.h>
  66 #include <linux/nsproxy.h>
  67 #include <linux/virtio_net.h>
  68 #include <linux/rcupdate.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <linux/seq_file.h>
  74 #include <linux/uio.h>
  75 #include <linux/skb_array.h>
  76
  77 #include <linux/uaccess.h>
  78
  79 /* Uncomment to enable debugging */
  80 /* #define TUN_DEBUG 1 */
  81
  82 #ifdef TUN_DEBUG
  83 static int debug;
  84
  85 #define tun_debug(level, tun, fmt, args...)                     \
  86 do {                                                            \
  87         if (tun->debug)                                         \
  88                 netdev_printk(level, tun->dev, fmt, ##args);    \
  89 } while (0)
  90 #define DBG1(level, fmt, args...)                               \
  91 do {                                                            \
  92         if (debug == 2)                                         \
  93                 printk(level fmt, ##args);                      \
  94 } while (0)
  95 #else
  96 #define tun_debug(level, tun, fmt, args...)                     \
  97 do {                                                            \
  98         if (0)                                                  \
  99                 netdev_printk(level, tun->dev, fmt, ##args);    \
 100 } while (0)
 101 #define DBG1(level, fmt, args...)                               \
 102 do {                                                            \
 103         if (0)                                                  \
 104                 printk(level fmt, ##args);                      \
 105 } while (0)
 106 #endif
 107
 108 /* TUN device flags */
 109
 110 /* IFF_ATTACH_QUEUE is never stored in device flags,
 111  * overload it to mean fasync when stored there.
 112  */
 113 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 114 /* High bits in flags field are unused. */
 115 #define TUN_VNET_LE     0x80000000
 116 #define TUN_VNET_BE     0x40000000
 117
 118 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 119                       IFF_MULTI_QUEUE)
 120 #define GOODCOPY_LEN 128
 121
 122 #define FLT_EXACT_COUNT 8
 123 struct tap_filter {
 124         unsigned int    count;    /* Number of addrs. Zero means disabled */
 125         u32             mask[2];  /* Mask of the hashed addrs */
 126         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 127 };
 128
 129 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 130  * to max number of VCPUs in guest. */
 131 #define MAX_TAP_QUEUES 256
 132 #define MAX_TAP_FLOWS  4096
 133
 134 #define TUN_FLOW_EXPIRE (3 * HZ)
 135
 136 struct tun_pcpu_stats {
 137         u64 rx_packets;
 138         u64 rx_bytes;
 139         u64 tx_packets;
 140         u64 tx_bytes;
 141         struct u64_stats_sync syncp;
 142         u32 rx_dropped;
 143         u32 tx_dropped;
 144         u32 rx_frame_errors;
 145 };
 146
 147 /* A tun_file connects an open character device to a tuntap netdevice. It
 148  * also contains all socket related structures (except sock_fprog and tap_filter)
 149  * to serve as one transmit queue for tuntap device. The sock_fprog and
 150  * tap_filter were kept in tun_struct since they were used for filtering for the
 151  * netdevice not for a specific queue (at least I didn't see the requirement for
 152  * this).
 153  *
 154  * RCU usage:
 155  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 156  * other can only be read while rcu_read_lock or rtnl_lock is held.
 157  */
 158 struct tun_file {
 159         struct sock sk;
 160         struct socket socket;
 161         struct socket_wq wq;
 162         struct tun_struct __rcu *tun;
 163         struct fasync_struct *fasync;
 164         /* only used for fasnyc */
 165         unsigned int flags;
 166         union {
 167                 u16 queue_index;
 168                 unsigned int ifindex;
 169         };
 170         struct list_head next;
 171         struct tun_struct *detached;
 172         struct skb_array tx_array;
 173 };
 174
 175 struct tun_flow_entry {
 176         struct hlist_node hash_link;
 177         struct rcu_head rcu;
 178         struct tun_struct *tun;
 179
 180         u32 rxhash;
 181         u32 rps_rxhash;
 182         int queue_index;
 183         unsigned long updated;
 184 };
 185
 186 #define TUN_NUM_FLOW_ENTRIES 1024
 187
 188 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 189  * device, socket filter, sndbuf and vnet header size were restore when the
 190  * file were attached to a persist device.
 191  */
 192 struct tun_struct {
 193         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 194         unsigned int            numqueues;
 195         unsigned int            flags;
 196         kuid_t                  owner;
 197         kgid_t                  group;
 198
 199         struct net_device       *dev;
 200         netdev_features_t       set_features;
 201 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 202                           NETIF_F_TSO6|NETIF_F_UFO)
 203
 204         int                     align;
 205         int                     vnet_hdr_sz;
 206         int                     sndbuf;
 207         struct tap_filter       txflt;
 208         struct sock_fprog       fprog;
 209         /* protected by rtnl lock */
 210         bool                    filter_attached;
 211 #ifdef TUN_DEBUG
 212         int debug;
 213 #endif
 214         spinlock_t lock;
 215         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 216         struct timer_list flow_gc_timer;
 217         unsigned long ageing_time;
 218         unsigned int numdisabled;
 219         struct list_head disabled;
 220         void *security;
 221         u32 flow_count;
 222         u32 rx_batched;
 223         struct tun_pcpu_stats __percpu *pcpu_stats;
 224 };
 225
 226 #ifdef CONFIG_TUN_VNET_CROSS_LE
 227 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 228 {
 229         return tun->flags & TUN_VNET_BE ? false :
 230                 virtio_legacy_is_little_endian();
 231 }
 232
 233 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 234 {
 235         int be = !!(tun->flags & TUN_VNET_BE);
 236
 237         if (put_user(be, argp))
 238                 return -EFAULT;
 239
 240         return 0;
 241 }
 242
 243 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 244 {
 245         int be;
 246
 247         if (get_user(be, argp))
 248                 return -EFAULT;
 249
 250         if (be)
 251                 tun->flags |= TUN_VNET_BE;
 252         else
 253                 tun->flags &= ~TUN_VNET_BE;
 254
 255         return 0;
 256 }
 257 #else
 258 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 259 {
 260         return virtio_legacy_is_little_endian();
 261 }
 262
 263 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 264 {
 265         return -EINVAL;
 266 }
 267
 268 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 269 {
 270         return -EINVAL;
 271 }
 272 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 273
 274 static inline bool tun_is_little_endian(struct tun_struct *tun)
 275 {
 276         return tun->flags & TUN_VNET_LE ||
 277                 tun_legacy_is_little_endian(tun);
 278 }
 279
 280 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 281 {
 282         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 283 }
 284
 285 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 286 {
 287         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 288 }
 289
 290 static inline u32 tun_hashfn(u32 rxhash)
 291 {
 292         return rxhash & 0x3ff;
 293 }
 294
 295 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 296 {
 297         struct tun_flow_entry *e;
 298
 299         hlist_for_each_entry_rcu(e, head, hash_link) {
 300                 if (e->rxhash == rxhash)
 301                         return e;
 302         }
 303         return NULL;
 304 }
 305
 306 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 307                                               struct hlist_head *head,
 308                                               u32 rxhash, u16 queue_index)
 309 {
 310         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 311
 312         if (e) {
 313                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 314                           rxhash, queue_index);
 315                 e->updated = jiffies;
 316                 e->rxhash = rxhash;
 317                 e->rps_rxhash = 0;
 318                 e->queue_index = queue_index;
 319                 e->tun = tun;
 320                 hlist_add_head_rcu(&e->hash_link, head);
 321                 ++tun->flow_count;
 322         }
 323         return e;
 324 }
 325
 326 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 327 {
 328         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 329                   e->rxhash, e->queue_index);
 330         hlist_del_rcu(&e->hash_link);
 331         kfree_rcu(e, rcu);
 332         --tun->flow_count;
 333 }
 334
 335 static void tun_flow_flush(struct tun_struct *tun)
 336 {
 337         int i;
 338
 339         spin_lock_bh(&tun->lock);
 340         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 341                 struct tun_flow_entry *e;
 342                 struct hlist_node *n;
 343
 344                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 345                         tun_flow_delete(tun, e);
 346         }
 347         spin_unlock_bh(&tun->lock);
 348 }
 349
 350 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 351 {
 352         int i;
 353
 354         spin_lock_bh(&tun->lock);
 355         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 356                 struct tun_flow_entry *e;
 357                 struct hlist_node *n;
 358
 359                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 360                         if (e->queue_index == queue_index)
 361                                 tun_flow_delete(tun, e);
 362                 }
 363         }
 364         spin_unlock_bh(&tun->lock);
 365 }
 366
 367 static void tun_flow_cleanup(unsigned long data)
 368 {
 369         struct tun_struct *tun = (struct tun_struct *)data;
 370         unsigned long delay = tun->ageing_time;
 371         unsigned long next_timer = jiffies + delay;
 372         unsigned long count = 0;
 373         int i;
 374
 375         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 376
 377         spin_lock_bh(&tun->lock);
 378         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 379                 struct tun_flow_entry *e;
 380                 struct hlist_node *n;
 381
 382                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 383                         unsigned long this_timer;
 384                         count++;
 385                         this_timer = e->updated + delay;
 386                         if (time_before_eq(this_timer, jiffies))
 387                                 tun_flow_delete(tun, e);
 388                         else if (time_before(this_timer, next_timer))
 389                                 next_timer = this_timer;
 390                 }
 391         }
 392
 393         if (count)
 394                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 395         spin_unlock_bh(&tun->lock);
 396 }
 397
 398 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 399                             struct tun_file *tfile)
 400 {
 401         struct hlist_head *head;
 402         struct tun_flow_entry *e;
 403         unsigned long delay = tun->ageing_time;
 404         u16 queue_index = tfile->queue_index;
 405
 406         if (!rxhash)
 407                 return;
 408         else
 409                 head = &tun->flows[tun_hashfn(rxhash)];
 410
 411         rcu_read_lock();
 412
 413         /* We may get a very small possibility of OOO during switching, not
 414          * worth to optimize.*/
 415         if (tun->numqueues == 1 || tfile->detached)
 416                 goto unlock;
 417
 418         e = tun_flow_find(head, rxhash);
 419         if (likely(e)) {
 420                 /* TODO: keep queueing to old queue until it's empty? */
 421                 e->queue_index = queue_index;
 422                 e->updated = jiffies;
 423                 sock_rps_record_flow_hash(e->rps_rxhash);
 424         } else {
 425                 spin_lock_bh(&tun->lock);
 426                 if (!tun_flow_find(head, rxhash) &&
 427                     tun->flow_count < MAX_TAP_FLOWS)
 428                         tun_flow_create(tun, head, rxhash, queue_index);
 429
 430                 if (!timer_pending(&tun->flow_gc_timer))
 431                         mod_timer(&tun->flow_gc_timer,
 432                                   round_jiffies_up(jiffies + delay));
 433                 spin_unlock_bh(&tun->lock);
 434         }
 435
 436 unlock:
 437         rcu_read_unlock();
 438 }
 439
 440 /**
 441  * Save the hash received in the stack receive path and update the
 442  * flow_hash table accordingly.
 443  */
 444 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 445 {
 446         if (unlikely(e->rps_rxhash != hash))
 447                 e->rps_rxhash = hash;
 448 }
 449
 450 /* We try to identify a flow through its rxhash first. The reason that
 451  * we do not check rxq no. is because some cards(e.g 82599), chooses
 452  * the rxq based on the txq where the last packet of the flow comes. As
 453  * the userspace application move between processors, we may get a
 454  * different rxq no. here. If we could not get rxhash, then we would
 455  * hope the rxq no. may help here.
 456  */
 457 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 458                             void *accel_priv, select_queue_fallback_t fallback)
 459 {
 460         struct tun_struct *tun = netdev_priv(dev);
 461         struct tun_flow_entry *e;
 462         u32 txq = 0;
 463         u32 numqueues = 0;
 464
 465         rcu_read_lock();
 466         numqueues = ACCESS_ONCE(tun->numqueues);
 467
 468         txq = skb_get_hash(skb);
 469         if (txq) {
 470                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 471                 if (e) {
 472                         tun_flow_save_rps_rxhash(e, txq);
 473                         txq = e->queue_index;
 474                 } else
 475                         /* use multiply and shift instead of expensive divide */
 476                         txq = ((u64)txq * numqueues) >> 32;
 477         } else if (likely(skb_rx_queue_recorded(skb))) {
 478                 txq = skb_get_rx_queue(skb);
 479                 while (unlikely(txq >= numqueues))
 480                         txq -= numqueues;
 481         }
 482
 483         rcu_read_unlock();
 484         return txq;
 485 }
 486
 487 static inline bool tun_not_capable(struct tun_struct *tun)
 488 {
 489         const struct cred *cred = current_cred();
 490         struct net *net = dev_net(tun->dev);
 491
 492         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 493                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 494                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 495 }
 496
 497 static void tun_set_real_num_queues(struct tun_struct *tun)
 498 {
 499         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 500         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 501 }
 502
 503 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 504 {
 505         tfile->detached = tun;
 506         list_add_tail(&tfile->next, &tun->disabled);
 507         ++tun->numdisabled;
 508 }
 509
 510 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 511 {
 512         struct tun_struct *tun = tfile->detached;
 513
 514         tfile->detached = NULL;
 515         list_del_init(&tfile->next);
 516         --tun->numdisabled;
 517         return tun;
 518 }
 519
 520 static void tun_queue_purge(struct tun_file *tfile)
 521 {
 522         struct sk_buff *skb;
 523
 524         while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
 525                 kfree_skb(skb);
 526
 527         skb_queue_purge(&tfile->sk.sk_write_queue);
 528         skb_queue_purge(&tfile->sk.sk_error_queue);
 529 }
 530
 531 static void __tun_detach(struct tun_file *tfile, bool clean)
 532 {
 533         struct tun_file *ntfile;
 534         struct tun_struct *tun;
 535
 536         tun = rtnl_dereference(tfile->tun);
 537
 538         if (tun && !tfile->detached) {
 539                 u16 index = tfile->queue_index;
 540                 BUG_ON(index >= tun->numqueues);
 541
 542                 rcu_assign_pointer(tun->tfiles[index],
 543                                    tun->tfiles[tun->numqueues - 1]);
 544                 ntfile = rtnl_dereference(tun->tfiles[index]);
 545                 ntfile->queue_index = index;
 546
 547                 --tun->numqueues;
 548                 if (clean) {
 549                         RCU_INIT_POINTER(tfile->tun, NULL);
 550                         sock_put(&tfile->sk);
 551                 } else
 552                         tun_disable_queue(tun, tfile);
 553
 554                 synchronize_net();
 555                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 556                 /* Drop read queue */
 557                 tun_queue_purge(tfile);
 558                 tun_set_real_num_queues(tun);
 559         } else if (tfile->detached && clean) {
 560                 tun = tun_enable_queue(tfile);
 561                 sock_put(&tfile->sk);
 562         }
 563
 564         if (clean) {
 565                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 566                         netif_carrier_off(tun->dev);
 567
 568                         if (!(tun->flags & IFF_PERSIST) &&
 569                             tun->dev->reg_state == NETREG_REGISTERED)
 570                                 unregister_netdevice(tun->dev);
 571                 }
 572                 if (tun)
 573                         skb_array_cleanup(&tfile->tx_array);
 574                 sock_put(&tfile->sk);
 575         }
 576 }
 577
 578 static void tun_detach(struct tun_file *tfile, bool clean)
 579 {
 580         rtnl_lock();
 581         __tun_detach(tfile, clean);
 582         rtnl_unlock();
 583 }
 584
 585 static void tun_detach_all(struct net_device *dev)
 586 {
 587         struct tun_struct *tun = netdev_priv(dev);
 588         struct tun_file *tfile, *tmp;
 589         int i, n = tun->numqueues;
 590
 591         for (i = 0; i < n; i++) {
 592                 tfile = rtnl_dereference(tun->tfiles[i]);
 593                 BUG_ON(!tfile);
 594                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 595                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 596                 RCU_INIT_POINTER(tfile->tun, NULL);
 597                 --tun->numqueues;
 598         }
 599         list_for_each_entry(tfile, &tun->disabled, next) {
 600                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 601                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 602                 RCU_INIT_POINTER(tfile->tun, NULL);
 603         }
 604         BUG_ON(tun->numqueues != 0);
 605
 606         synchronize_net();
 607         for (i = 0; i < n; i++) {
 608                 tfile = rtnl_dereference(tun->tfiles[i]);
 609                 /* Drop read queue */
 610                 tun_queue_purge(tfile);
 611                 sock_put(&tfile->sk);
 612         }
 613         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 614                 tun_enable_queue(tfile);
 615                 tun_queue_purge(tfile);
 616                 sock_put(&tfile->sk);
 617         }
 618         BUG_ON(tun->numdisabled != 0);
 619
 620         if (tun->flags & IFF_PERSIST)
 621                 module_put(THIS_MODULE);
 622 }
 623
 624 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 625 {
 626         struct tun_file *tfile = file->private_data;
 627         struct net_device *dev = tun->dev;
 628         int err;
 629
 630         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 631         if (err < 0)
 632                 goto out;
 633
 634         err = -EINVAL;
 635         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 636                 goto out;
 637
 638         err = -EBUSY;
 639         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 640                 goto out;
 641
 642         err = -E2BIG;
 643         if (!tfile->detached &&
 644             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 645                 goto out;
 646
 647         err = 0;
 648
 649         /* Re-attach the filter to persist device */
 650         if (!skip_filter && (tun->filter_attached == true)) {
 651                 lock_sock(tfile->socket.sk);
 652                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 653                 release_sock(tfile->socket.sk);
 654                 if (!err)
 655                         goto out;
 656         }
 657
 658         if (!tfile->detached &&
 659             skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
 660                 err = -ENOMEM;
 661                 goto out;
 662         }
 663
 664         tfile->queue_index = tun->numqueues;
 665         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 666         rcu_assign_pointer(tfile->tun, tun);
 667         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 668         tun->numqueues++;
 669
 670         if (tfile->detached)
 671                 tun_enable_queue(tfile);
 672         else
 673                 sock_hold(&tfile->sk);
 674
 675         tun_set_real_num_queues(tun);
 676
 677         /* device is allowed to go away first, so no need to hold extra
 678          * refcnt.
 679          */
 680
 681 out:
 682         return err;
 683 }
 684
 685 static struct tun_struct *__tun_get(struct tun_file *tfile)
 686 {
 687         struct tun_struct *tun;
 688
 689         rcu_read_lock();
 690         tun = rcu_dereference(tfile->tun);
 691         if (tun)
 692                 dev_hold(tun->dev);
 693         rcu_read_unlock();
 694
 695         return tun;
 696 }
 697
 698 static struct tun_struct *tun_get(struct file *file)
 699 {
 700         return __tun_get(file->private_data);
 701 }
 702
 703 static void tun_put(struct tun_struct *tun)
 704 {
 705         dev_put(tun->dev);
 706 }
 707
 708 /* TAP filtering */
 709 static void addr_hash_set(u32 *mask, const u8 *addr)
 710 {
 711         int n = ether_crc(ETH_ALEN, addr) >> 26;
 712         mask[n >> 5] |= (1 << (n & 31));
 713 }
 714
 715 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 716 {
 717         int n = ether_crc(ETH_ALEN, addr) >> 26;
 718         return mask[n >> 5] & (1 << (n & 31));
 719 }
 720
 721 static int update_filter(struct tap_filter *filter, void __user *arg)
 722 {
 723         struct { u8 u[ETH_ALEN]; } *addr;
 724         struct tun_filter uf;
 725         int err, alen, n, nexact;
 726
 727         if (copy_from_user(&uf, arg, sizeof(uf)))
 728                 return -EFAULT;
 729
 730         if (!uf.count) {
 731                 /* Disabled */
 732                 filter->count = 0;
 733                 return 0;
 734         }
 735
 736         alen = ETH_ALEN * uf.count;
 737         addr = memdup_user(arg + sizeof(uf), alen);
 738         if (IS_ERR(addr))
 739                 return PTR_ERR(addr);
 740
 741         /* The filter is updated without holding any locks. Which is
 742          * perfectly safe. We disable it first and in the worst
 743          * case we'll accept a few undesired packets. */
 744         filter->count = 0;
 745         wmb();
 746
 747         /* Use first set of addresses as an exact filter */
 748         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 749                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 750
 751         nexact = n;
 752
 753         /* Remaining multicast addresses are hashed,
 754          * unicast will leave the filter disabled. */
 755         memset(filter->mask, 0, sizeof(filter->mask));
 756         for (; n < uf.count; n++) {
 757                 if (!is_multicast_ether_addr(addr[n].u)) {
 758                         err = 0; /* no filter */
 759                         goto free_addr;
 760                 }
 761                 addr_hash_set(filter->mask, addr[n].u);
 762         }
 763
 764         /* For ALLMULTI just set the mask to all ones.
 765          * This overrides the mask populated above. */
 766         if ((uf.flags & TUN_FLT_ALLMULTI))
 767                 memset(filter->mask, ~0, sizeof(filter->mask));
 768
 769         /* Now enable the filter */
 770         wmb();
 771         filter->count = nexact;
 772
 773         /* Return the number of exact filters */
 774         err = nexact;
 775 free_addr:
 776         kfree(addr);
 777         return err;
 778 }
 779
 780 /* Returns: 0 - drop, !=0 - accept */
 781 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 782 {
 783         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 784          * at this point. */
 785         struct ethhdr *eh = (struct ethhdr *) skb->data;
 786         int i;
 787
 788         /* Exact match */
 789         for (i = 0; i < filter->count; i++)
 790                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 791                         return 1;
 792
 793         /* Inexact match (multicast only) */
 794         if (is_multicast_ether_addr(eh->h_dest))
 795                 return addr_hash_test(filter->mask, eh->h_dest);
 796
 797         return 0;
 798 }
 799
 800 /*
 801  * Checks whether the packet is accepted or not.
 802  * Returns: 0 - drop, !=0 - accept
 803  */
 804 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 805 {
 806         if (!filter->count)
 807                 return 1;
 808
 809         return run_filter(filter, skb);
 810 }
 811
 812 /* Network device part of the driver */
 813
 814 static const struct ethtool_ops tun_ethtool_ops;
 815
 816 /* Net device detach from fd. */
 817 static void tun_net_uninit(struct net_device *dev)
 818 {
 819         tun_detach_all(dev);
 820 }
 821
 822 /* Net device open. */
 823 static int tun_net_open(struct net_device *dev)
 824 {
 825         netif_tx_start_all_queues(dev);
 826         return 0;
 827 }
 828
 829 /* Net device close. */
 830 static int tun_net_close(struct net_device *dev)
 831 {
 832         netif_tx_stop_all_queues(dev);
 833         return 0;
 834 }
 835
 836 /* Net device start xmit */
 837 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 838 {
 839         struct tun_struct *tun = netdev_priv(dev);
 840         int txq = skb->queue_mapping;
 841         struct tun_file *tfile;
 842         u32 numqueues = 0;
 843
 844         rcu_read_lock();
 845         tfile = rcu_dereference(tun->tfiles[txq]);
 846         numqueues = ACCESS_ONCE(tun->numqueues);
 847
 848         /* Drop packet if interface is not attached */
 849         if (txq >= numqueues)
 850                 goto drop;
 851
 852 #ifdef CONFIG_RPS
 853         if (numqueues == 1 && static_key_false(&rps_needed)) {
 854                 /* Select queue was not called for the skbuff, so we extract the
 855                  * RPS hash and save it into the flow_table here.
 856                  */
 857                 __u32 rxhash;
 858
 859                 rxhash = skb_get_hash(skb);
 860                 if (rxhash) {
 861                         struct tun_flow_entry *e;
 862                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 863                                         rxhash);
 864                         if (e)
 865                                 tun_flow_save_rps_rxhash(e, rxhash);
 866                 }
 867         }
 868 #endif
 869
 870         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 871
 872         BUG_ON(!tfile);
 873
 874         /* Drop if the filter does not like it.
 875          * This is a noop if the filter is disabled.
 876          * Filter can be enabled only for the TAP devices. */
 877         if (!check_filter(&tun->txflt, skb))
 878                 goto drop;
 879
 880         if (tfile->socket.sk->sk_filter &&
 881             sk_filter(tfile->socket.sk, skb))
 882                 goto drop;
 883
 884         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 885                 goto drop;
 886
 887         skb_tx_timestamp(skb);
 888
 889         /* Orphan the skb - required as we might hang on to it
 890          * for indefinite time.
 891          */
 892         skb_orphan(skb);
 893
 894         nf_reset(skb);
 895
 896         if (skb_array_produce(&tfile->tx_array, skb))
 897                 goto drop;
 898
 899         /* Notify and wake up reader process */
 900         if (tfile->flags & TUN_FASYNC)
 901                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 902         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 903
 904         rcu_read_unlock();
 905         return NETDEV_TX_OK;
 906
 907 drop:
 908         this_cpu_inc(tun->pcpu_stats->tx_dropped);
 909         skb_tx_error(skb);
 910         kfree_skb(skb);
 911         rcu_read_unlock();
 912         return NET_XMIT_DROP;
 913 }
 914
 915 static void tun_net_mclist(struct net_device *dev)
 916 {
 917         /*
 918          * This callback is supposed to deal with mc filter in
 919          * _rx_ path and has nothing to do with the _tx_ path.
 920          * In rx path we always accept everything userspace gives us.
 921          */
 922 }
 923
 924 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 925         netdev_features_t features)
 926 {
 927         struct tun_struct *tun = netdev_priv(dev);
 928
 929         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 930 }
 931 #ifdef CONFIG_NET_POLL_CONTROLLER
 932 static void tun_poll_controller(struct net_device *dev)
 933 {
 934         /*
 935          * Tun only receives frames when:
 936          * 1) the char device endpoint gets data from user space
 937          * 2) the tun socket gets a sendmsg call from user space
 938          * Since both of those are synchronous operations, we are guaranteed
 939          * never to have pending data when we poll for it
 940          * so there is nothing to do here but return.
 941          * We need this though so netpoll recognizes us as an interface that
 942          * supports polling, which enables bridge devices in virt setups to
 943          * still use netconsole
 944          */
 945         return;
 946 }
 947 #endif
 948
 949 static void tun_set_headroom(struct net_device *dev, int new_hr)
 950 {
 951         struct tun_struct *tun = netdev_priv(dev);
 952
 953         if (new_hr < NET_SKB_PAD)
 954                 new_hr = NET_SKB_PAD;
 955
 956         tun->align = new_hr;
 957 }
 958
 959 static void
 960 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
 961 {
 962         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
 963         struct tun_struct *tun = netdev_priv(dev);
 964         struct tun_pcpu_stats *p;
 965         int i;
 966
 967         for_each_possible_cpu(i) {
 968                 u64 rxpackets, rxbytes, txpackets, txbytes;
 969                 unsigned int start;
 970
 971                 p = per_cpu_ptr(tun->pcpu_stats, i);
 972                 do {
 973                         start = u64_stats_fetch_begin(&p->syncp);
 974                         rxpackets       = p->rx_packets;
 975                         rxbytes         = p->rx_bytes;
 976                         txpackets       = p->tx_packets;
 977                         txbytes         = p->tx_bytes;
 978                 } while (u64_stats_fetch_retry(&p->syncp, start));
 979
 980                 stats->rx_packets       += rxpackets;
 981                 stats->rx_bytes         += rxbytes;
 982                 stats->tx_packets       += txpackets;
 983                 stats->tx_bytes         += txbytes;
 984
 985                 /* u32 counters */
 986                 rx_dropped      += p->rx_dropped;
 987                 rx_frame_errors += p->rx_frame_errors;
 988                 tx_dropped      += p->tx_dropped;
 989         }
 990         stats->rx_dropped  = rx_dropped;
 991         stats->rx_frame_errors = rx_frame_errors;
 992         stats->tx_dropped = tx_dropped;
 993 }
 994
 995 static const struct net_device_ops tun_netdev_ops = {
 996         .ndo_uninit             = tun_net_uninit,
 997         .ndo_open               = tun_net_open,
 998         .ndo_stop               = tun_net_close,
 999         .ndo_start_xmit         = tun_net_xmit,
1000         .ndo_fix_features       = tun_net_fix_features,
1001         .ndo_select_queue       = tun_select_queue,
1002 #ifdef CONFIG_NET_POLL_CONTROLLER
1003         .ndo_poll_controller    = tun_poll_controller,
1004 #endif
1005         .ndo_set_rx_headroom    = tun_set_headroom,
1006         .ndo_get_stats64        = tun_net_get_stats64,
1007 };
1008
1009 static const struct net_device_ops tap_netdev_ops = {
1010         .ndo_uninit             = tun_net_uninit,
1011         .ndo_open               = tun_net_open,
1012         .ndo_stop               = tun_net_close,
1013         .ndo_start_xmit         = tun_net_xmit,
1014         .ndo_fix_features       = tun_net_fix_features,
1015         .ndo_set_rx_mode        = tun_net_mclist,
1016         .ndo_set_mac_address    = eth_mac_addr,
1017         .ndo_validate_addr      = eth_validate_addr,
1018         .ndo_select_queue       = tun_select_queue,
1019 #ifdef CONFIG_NET_POLL_CONTROLLER
1020         .ndo_poll_controller    = tun_poll_controller,
1021 #endif
1022         .ndo_features_check     = passthru_features_check,
1023         .ndo_set_rx_headroom    = tun_set_headroom,
1024         .ndo_get_stats64        = tun_net_get_stats64,
1025 };
1026
1027 static void tun_flow_init(struct tun_struct *tun)
1028 {
1029         int i;
1030
1031         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1032                 INIT_HLIST_HEAD(&tun->flows[i]);
1033
1034         tun->ageing_time = TUN_FLOW_EXPIRE;
1035         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1036         mod_timer(&tun->flow_gc_timer,
1037                   round_jiffies_up(jiffies + tun->ageing_time));
1038 }
1039
1040 static void tun_flow_uninit(struct tun_struct *tun)
1041 {
1042         del_timer_sync(&tun->flow_gc_timer);
1043         tun_flow_flush(tun);
1044 }
1045
1046 #define MIN_MTU 68
1047 #define MAX_MTU 65535
1048
1049 /* Initialize net device. */
1050 static void tun_net_init(struct net_device *dev)
1051 {
1052         struct tun_struct *tun = netdev_priv(dev);
1053
1054         switch (tun->flags & TUN_TYPE_MASK) {
1055         case IFF_TUN:
1056                 dev->netdev_ops = &tun_netdev_ops;
1057
1058                 /* Point-to-Point TUN Device */
1059                 dev->hard_header_len = 0;
1060                 dev->addr_len = 0;
1061                 dev->mtu = 1500;
1062
1063                 /* Zero header length */
1064                 dev->type = ARPHRD_NONE;
1065                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1066                 break;
1067
1068         case IFF_TAP:
1069                 dev->netdev_ops = &tap_netdev_ops;
1070                 /* Ethernet TAP Device */
1071                 ether_setup(dev);
1072                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1073                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1074
1075                 eth_hw_addr_random(dev);
1076
1077                 break;
1078         }
1079
1080         dev->min_mtu = MIN_MTU;
1081         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1082 }
1083
1084 /* Character device part */
1085
1086 /* Poll */
1087 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1088 {
1089         struct tun_file *tfile = file->private_data;
1090         struct tun_struct *tun = __tun_get(tfile);
1091         struct sock *sk;
1092         unsigned int mask = 0;
1093
1094         if (!tun)
1095                 return POLLERR;
1096
1097         sk = tfile->socket.sk;
1098
1099         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1100
1101         poll_wait(file, sk_sleep(sk), wait);
1102
1103         if (!skb_array_empty(&tfile->tx_array))
1104                 mask |= POLLIN | POLLRDNORM;
1105
1106         if (sock_writeable(sk) ||
1107             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1108              sock_writeable(sk)))
1109                 mask |= POLLOUT | POLLWRNORM;
1110
1111         if (tun->dev->reg_state != NETREG_REGISTERED)
1112                 mask = POLLERR;
1113
1114         tun_put(tun);
1115         return mask;
1116 }
1117
1118 /* prepad is the amount to reserve at front.  len is length after that.
1119  * linear is a hint as to how much to copy (usually headers). */
1120 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1121                                      size_t prepad, size_t len,
1122                                      size_t linear, int noblock)
1123 {
1124         struct sock *sk = tfile->socket.sk;
1125         struct sk_buff *skb;
1126         int err;
1127
1128         /* Under a page?  Don't bother with paged skb. */
1129         if (prepad + len < PAGE_SIZE || !linear)
1130                 linear = len;
1131
1132         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1133                                    &err, 0);
1134         if (!skb)
1135                 return ERR_PTR(err);
1136
1137         skb_reserve(skb, prepad);
1138         skb_put(skb, linear);
1139         skb->data_len = len - linear;
1140         skb->len += len - linear;
1141
1142         return skb;
1143 }
1144
1145 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1146                            struct sk_buff *skb, int more)
1147 {
1148         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1149         struct sk_buff_head process_queue;
1150         u32 rx_batched = tun->rx_batched;
1151         bool rcv = false;
1152
1153         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1154                 local_bh_disable();
1155                 netif_receive_skb(skb);
1156                 local_bh_enable();
1157                 return;
1158         }
1159
1160         spin_lock(&queue->lock);
1161         if (!more || skb_queue_len(queue) == rx_batched) {
1162                 __skb_queue_head_init(&process_queue);
1163                 skb_queue_splice_tail_init(queue, &process_queue);
1164                 rcv = true;
1165         } else {
1166                 __skb_queue_tail(queue, skb);
1167         }
1168         spin_unlock(&queue->lock);
1169
1170         if (rcv) {
1171                 struct sk_buff *nskb;
1172
1173                 local_bh_disable();
1174                 while ((nskb = __skb_dequeue(&process_queue)))
1175                         netif_receive_skb(nskb);
1176                 netif_receive_skb(skb);
1177                 local_bh_enable();
1178         }
1179 }
1180
1181 /* Get packet from user space buffer */
1182 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1183                             void *msg_control, struct iov_iter *from,
1184                             int noblock, bool more)
1185 {
1186         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1187         struct sk_buff *skb;
1188         size_t total_len = iov_iter_count(from);
1189         size_t len = total_len, align = tun->align, linear;
1190         struct virtio_net_hdr gso = { 0 };
1191         struct tun_pcpu_stats *stats;
1192         int good_linear;
1193         int copylen;
1194         bool zerocopy = false;
1195         int err;
1196         u32 rxhash;
1197
1198         if (!(tun->dev->flags & IFF_UP))
1199                 return -EIO;
1200
1201         if (!(tun->flags & IFF_NO_PI)) {
1202                 if (len < sizeof(pi))
1203                         return -EINVAL;
1204                 len -= sizeof(pi);
1205
1206                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1207                         return -EFAULT;
1208         }
1209
1210         if (tun->flags & IFF_VNET_HDR) {
1211                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1212
1213                 if (len < vnet_hdr_sz)
1214                         return -EINVAL;
1215                 len -= vnet_hdr_sz;
1216
1217                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1218                         return -EFAULT;
1219
1220                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1221                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1222                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1223
1224                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1225                         return -EINVAL;
1226                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1227         }
1228
1229         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1230                 align += NET_IP_ALIGN;
1231                 if (unlikely(len < ETH_HLEN ||
1232                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1233                         return -EINVAL;
1234         }
1235
1236         good_linear = SKB_MAX_HEAD(align);
1237
1238         if (msg_control) {
1239                 struct iov_iter i = *from;
1240
1241                 /* There are 256 bytes to be copied in skb, so there is
1242                  * enough room for skb expand head in case it is used.
1243                  * The rest of the buffer is mapped from userspace.
1244                  */
1245                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1246                 if (copylen > good_linear)
1247                         copylen = good_linear;
1248                 linear = copylen;
1249                 iov_iter_advance(&i, copylen);
1250                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1251                         zerocopy = true;
1252         }
1253
1254         if (!zerocopy) {
1255                 copylen = len;
1256                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1257                         linear = good_linear;
1258                 else
1259                         linear = tun16_to_cpu(tun, gso.hdr_len);
1260         }
1261
1262         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1263         if (IS_ERR(skb)) {
1264                 if (PTR_ERR(skb) != -EAGAIN)
1265                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1266                 return PTR_ERR(skb);
1267         }
1268
1269         if (zerocopy)
1270                 err = zerocopy_sg_from_iter(skb, from);
1271         else
1272                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1273
1274         if (err) {
1275                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1276                 kfree_skb(skb);
1277                 return -EFAULT;
1278         }
1279
1280         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1281                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1282                 kfree_skb(skb);
1283                 return -EINVAL;
1284         }
1285
1286         switch (tun->flags & TUN_TYPE_MASK) {
1287         case IFF_TUN:
1288                 if (tun->flags & IFF_NO_PI) {
1289                         switch (skb->data[0] & 0xf0) {
1290                         case 0x40:
1291                                 pi.proto = htons(ETH_P_IP);
1292                                 break;
1293                         case 0x60:
1294                                 pi.proto = htons(ETH_P_IPV6);
1295                                 break;
1296                         default:
1297                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1298                                 kfree_skb(skb);
1299                                 return -EINVAL;
1300                         }
1301                 }
1302
1303                 skb_reset_mac_header(skb);
1304                 skb->protocol = pi.proto;
1305                 skb->dev = tun->dev;
1306                 break;
1307         case IFF_TAP:
1308                 skb->protocol = eth_type_trans(skb, tun->dev);
1309                 break;
1310         }
1311
1312         /* copy skb_ubuf_info for callback when skb has no error */
1313         if (zerocopy) {
1314                 skb_shinfo(skb)->destructor_arg = msg_control;
1315                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1316                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1317         } else if (msg_control) {
1318                 struct ubuf_info *uarg = msg_control;
1319                 uarg->callback(uarg, false);
1320         }
1321
1322         skb_reset_network_header(skb);
1323         skb_probe_transport_header(skb, 0);
1324
1325         rxhash = skb_get_hash(skb);
1326 #ifndef CONFIG_4KSTACKS
1327         tun_rx_batched(tun, tfile, skb, more);
1328 #else
1329         netif_rx_ni(skb);
1330 #endif
1331
1332         stats = get_cpu_ptr(tun->pcpu_stats);
1333         u64_stats_update_begin(&stats->syncp);
1334         stats->rx_packets++;
1335         stats->rx_bytes += len;
1336         u64_stats_update_end(&stats->syncp);
1337         put_cpu_ptr(stats);
1338
1339         tun_flow_update(tun, rxhash, tfile);
1340         return total_len;
1341 }
1342
1343 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1344 {
1345         struct file *file = iocb->ki_filp;
1346         struct tun_struct *tun = tun_get(file);
1347         struct tun_file *tfile = file->private_data;
1348         ssize_t result;
1349
1350         if (!tun)
1351                 return -EBADFD;
1352
1353         result = tun_get_user(tun, tfile, NULL, from,
1354                               file->f_flags & O_NONBLOCK, false);
1355
1356         tun_put(tun);
1357         return result;
1358 }
1359
1360 /* Put packet to the user space buffer */
1361 static ssize_t tun_put_user(struct tun_struct *tun,
1362                             struct tun_file *tfile,
1363                             struct sk_buff *skb,
1364                             struct iov_iter *iter)
1365 {
1366         struct tun_pi pi = { 0, skb->protocol };
1367         struct tun_pcpu_stats *stats;
1368         ssize_t total;
1369         int vlan_offset = 0;
1370         int vlan_hlen = 0;
1371         int vnet_hdr_sz = 0;
1372
1373         if (skb_vlan_tag_present(skb))
1374                 vlan_hlen = VLAN_HLEN;
1375
1376         if (tun->flags & IFF_VNET_HDR)
1377                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1378
1379         total = skb->len + vlan_hlen + vnet_hdr_sz;
1380
1381         if (!(tun->flags & IFF_NO_PI)) {
1382                 if (iov_iter_count(iter) < sizeof(pi))
1383                         return -EINVAL;
1384
1385                 total += sizeof(pi);
1386                 if (iov_iter_count(iter) < total) {
1387                         /* Packet will be striped */
1388                         pi.flags |= TUN_PKT_STRIP;
1389                 }
1390
1391                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1392                         return -EFAULT;
1393         }
1394
1395         if (vnet_hdr_sz) {
1396                 struct virtio_net_hdr gso;
1397
1398                 if (iov_iter_count(iter) < vnet_hdr_sz)
1399                         return -EINVAL;
1400
1401                 if (virtio_net_hdr_from_skb(skb, &gso,
1402                                             tun_is_little_endian(tun), true)) {
1403                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1404                         pr_err("unexpected GSO type: "
1405                                "0x%x, gso_size %d, hdr_len %d\n",
1406                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1407                                tun16_to_cpu(tun, gso.hdr_len));
1408                         print_hex_dump(KERN_ERR, "tun: ",
1409                                        DUMP_PREFIX_NONE,
1410                                        16, 1, skb->head,
1411                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1412                         WARN_ON_ONCE(1);
1413                         return -EINVAL;
1414                 }
1415
1416                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1417                         return -EFAULT;
1418
1419                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1420         }
1421
1422         if (vlan_hlen) {
1423                 int ret;
1424                 struct {
1425                         __be16 h_vlan_proto;
1426                         __be16 h_vlan_TCI;
1427                 } veth;
1428
1429                 veth.h_vlan_proto = skb->vlan_proto;
1430                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1431
1432                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1433
1434                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1435                 if (ret || !iov_iter_count(iter))
1436                         goto done;
1437
1438                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1439                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1440                         goto done;
1441         }
1442
1443         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1444
1445 done:
1446         /* caller is in process context, */
1447         stats = get_cpu_ptr(tun->pcpu_stats);
1448         u64_stats_update_begin(&stats->syncp);
1449         stats->tx_packets++;
1450         stats->tx_bytes += skb->len + vlan_hlen;
1451         u64_stats_update_end(&stats->syncp);
1452         put_cpu_ptr(tun->pcpu_stats);
1453
1454         return total;
1455 }
1456
1457 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1458                                      int *err)
1459 {
1460         DECLARE_WAITQUEUE(wait, current);
1461         struct sk_buff *skb = NULL;
1462         int error = 0;
1463
1464         skb = skb_array_consume(&tfile->tx_array);
1465         if (skb)
1466                 goto out;
1467         if (noblock) {
1468                 error = -EAGAIN;
1469                 goto out;
1470         }
1471
1472         add_wait_queue(&tfile->wq.wait, &wait);
1473         current->state = TASK_INTERRUPTIBLE;
1474
1475         while (1) {
1476                 skb = skb_array_consume(&tfile->tx_array);
1477                 if (skb)
1478                         break;
1479                 if (signal_pending(current)) {
1480                         error = -ERESTARTSYS;
1481                         break;
1482                 }
1483                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1484                         error = -EFAULT;
1485                         break;
1486                 }
1487
1488                 schedule();
1489         }
1490
1491         current->state = TASK_RUNNING;
1492         remove_wait_queue(&tfile->wq.wait, &wait);
1493
1494 out:
1495         *err = error;
1496         return skb;
1497 }
1498
1499 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1500                            struct iov_iter *to,
1501                            int noblock)
1502 {
1503         struct sk_buff *skb;
1504         ssize_t ret;
1505         int err;
1506
1507         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1508
1509         if (!iov_iter_count(to))
1510                 return 0;
1511
1512         /* Read frames from ring */
1513         skb = tun_ring_recv(tfile, noblock, &err);
1514         if (!skb)
1515                 return err;
1516
1517         ret = tun_put_user(tun, tfile, skb, to);
1518         if (unlikely(ret < 0))
1519                 kfree_skb(skb);
1520         else
1521                 consume_skb(skb);
1522
1523         return ret;
1524 }
1525
1526 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1527 {
1528         struct file *file = iocb->ki_filp;
1529         struct tun_file *tfile = file->private_data;
1530         struct tun_struct *tun = __tun_get(tfile);
1531         ssize_t len = iov_iter_count(to), ret;
1532
1533         if (!tun)
1534                 return -EBADFD;
1535         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1536         ret = min_t(ssize_t, ret, len);
1537         if (ret > 0)
1538                 iocb->ki_pos = ret;
1539         tun_put(tun);
1540         return ret;
1541 }
1542
1543 static void tun_free_netdev(struct net_device *dev)
1544 {
1545         struct tun_struct *tun = netdev_priv(dev);
1546
1547         BUG_ON(!(list_empty(&tun->disabled)));
1548         free_percpu(tun->pcpu_stats);
1549         tun_flow_uninit(tun);
1550         security_tun_dev_free_security(tun->security);
1551         free_netdev(dev);
1552 }
1553
1554 static void tun_setup(struct net_device *dev)
1555 {
1556         struct tun_struct *tun = netdev_priv(dev);
1557
1558         tun->owner = INVALID_UID;
1559         tun->group = INVALID_GID;
1560
1561         dev->ethtool_ops = &tun_ethtool_ops;
1562         dev->destructor = tun_free_netdev;
1563         /* We prefer our own queue length */
1564         dev->tx_queue_len = TUN_READQ_SIZE;
1565 }
1566
1567 /* Trivial set of netlink ops to allow deleting tun or tap
1568  * device with netlink.
1569  */
1570 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1571 {
1572         return -EINVAL;
1573 }
1574
1575 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1576         .kind           = DRV_NAME,
1577         .priv_size      = sizeof(struct tun_struct),
1578         .setup          = tun_setup,
1579         .validate       = tun_validate,
1580 };
1581
1582 static void tun_sock_write_space(struct sock *sk)
1583 {
1584         struct tun_file *tfile;
1585         wait_queue_head_t *wqueue;
1586
1587         if (!sock_writeable(sk))
1588                 return;
1589
1590         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1591                 return;
1592
1593         wqueue = sk_sleep(sk);
1594         if (wqueue && waitqueue_active(wqueue))
1595                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1596                                                 POLLWRNORM | POLLWRBAND);
1597
1598         tfile = container_of(sk, struct tun_file, sk);
1599         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1600 }
1601
1602 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1603 {
1604         int ret;
1605         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1606         struct tun_struct *tun = __tun_get(tfile);
1607
1608         if (!tun)
1609                 return -EBADFD;
1610
1611         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1612                            m->msg_flags & MSG_DONTWAIT,
1613                            m->msg_flags & MSG_MORE);
1614         tun_put(tun);
1615         return ret;
1616 }
1617
1618 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1619                        int flags)
1620 {
1621         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1622         struct tun_struct *tun = __tun_get(tfile);
1623         int ret;
1624
1625         if (!tun)
1626                 return -EBADFD;
1627
1628         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1629                 ret = -EINVAL;
1630                 goto out;
1631         }
1632         if (flags & MSG_ERRQUEUE) {
1633                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1634                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1635                 goto out;
1636         }
1637         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1638         if (ret > (ssize_t)total_len) {
1639                 m->msg_flags |= MSG_TRUNC;
1640                 ret = flags & MSG_TRUNC ? ret : total_len;
1641         }
1642 out:
1643         tun_put(tun);
1644         return ret;
1645 }
1646
1647 static int tun_peek_len(struct socket *sock)
1648 {
1649         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1650         struct tun_struct *tun;
1651         int ret = 0;
1652
1653         tun = __tun_get(tfile);
1654         if (!tun)
1655                 return 0;
1656
1657         ret = skb_array_peek_len(&tfile->tx_array);
1658         tun_put(tun);
1659
1660         return ret;
1661 }
1662
1663 /* Ops structure to mimic raw sockets with tun */
1664 static const struct proto_ops tun_socket_ops = {
1665         .peek_len = tun_peek_len,
1666         .sendmsg = tun_sendmsg,
1667         .recvmsg = tun_recvmsg,
1668 };
1669
1670 static struct proto tun_proto = {
1671         .name           = "tun",
1672         .owner          = THIS_MODULE,
1673         .obj_size       = sizeof(struct tun_file),
1674 };
1675
1676 static int tun_flags(struct tun_struct *tun)
1677 {
1678         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1679 }
1680
1681 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1682                               char *buf)
1683 {
1684         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1685         return sprintf(buf, "0x%x\n", tun_flags(tun));
1686 }
1687
1688 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1689                               char *buf)
1690 {
1691         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1692         return uid_valid(tun->owner)?
1693                 sprintf(buf, "%u\n",
1694                         from_kuid_munged(current_user_ns(), tun->owner)):
1695                 sprintf(buf, "-1\n");
1696 }
1697
1698 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1699                               char *buf)
1700 {
1701         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1702         return gid_valid(tun->group) ?
1703                 sprintf(buf, "%u\n",
1704                         from_kgid_munged(current_user_ns(), tun->group)):
1705                 sprintf(buf, "-1\n");
1706 }
1707
1708 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1709 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1710 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1711
1712 static struct attribute *tun_dev_attrs[] = {
1713         &dev_attr_tun_flags.attr,
1714         &dev_attr_owner.attr,
1715         &dev_attr_group.attr,
1716         NULL
1717 };
1718
1719 static const struct attribute_group tun_attr_group = {
1720         .attrs = tun_dev_attrs
1721 };
1722
1723 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1724 {
1725         struct tun_struct *tun;
1726         struct tun_file *tfile = file->private_data;
1727         struct net_device *dev;
1728         int err;
1729
1730         if (tfile->detached)
1731                 return -EINVAL;
1732
1733         dev = __dev_get_by_name(net, ifr->ifr_name);
1734         if (dev) {
1735                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1736                         return -EBUSY;
1737                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1738                         tun = netdev_priv(dev);
1739                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1740                         tun = netdev_priv(dev);
1741                 else
1742                         return -EINVAL;
1743
1744                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1745                     !!(tun->flags & IFF_MULTI_QUEUE))
1746                         return -EINVAL;
1747
1748                 if (tun_not_capable(tun))
1749                         return -EPERM;
1750                 err = security_tun_dev_open(tun->security);
1751                 if (err < 0)
1752                         return err;
1753
1754                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1755                 if (err < 0)
1756                         return err;
1757
1758                 if (tun->flags & IFF_MULTI_QUEUE &&
1759                     (tun->numqueues + tun->numdisabled > 1)) {
1760                         /* One or more queue has already been attached, no need
1761                          * to initialize the device again.
1762                          */
1763                         return 0;
1764                 }
1765         }
1766         else {
1767                 char *name;
1768                 unsigned long flags = 0;
1769                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1770                              MAX_TAP_QUEUES : 1;
1771
1772                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1773                         return -EPERM;
1774                 err = security_tun_dev_create();
1775                 if (err < 0)
1776                         return err;
1777
1778                 /* Set dev type */
1779                 if (ifr->ifr_flags & IFF_TUN) {
1780                         /* TUN device */
1781                         flags |= IFF_TUN;
1782                         name = "tun%d";
1783                 } else if (ifr->ifr_flags & IFF_TAP) {
1784                         /* TAP device */
1785                         flags |= IFF_TAP;
1786                         name = "tap%d";
1787                 } else
1788                         return -EINVAL;
1789
1790                 if (*ifr->ifr_name)
1791                         name = ifr->ifr_name;
1792
1793                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1794                                        NET_NAME_UNKNOWN, tun_setup, queues,
1795                                        queues);
1796
1797                 if (!dev)
1798                         return -ENOMEM;
1799
1800                 dev_net_set(dev, net);
1801                 dev->rtnl_link_ops = &tun_link_ops;
1802                 dev->ifindex = tfile->ifindex;
1803                 dev->sysfs_groups[0] = &tun_attr_group;
1804
1805                 tun = netdev_priv(dev);
1806                 tun->dev = dev;
1807                 tun->flags = flags;
1808                 tun->txflt.count = 0;
1809                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1810
1811                 tun->align = NET_SKB_PAD;
1812                 tun->filter_attached = false;
1813                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1814                 tun->rx_batched = 0;
1815
1816                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
1817                 if (!tun->pcpu_stats) {
1818                         err = -ENOMEM;
1819                         goto err_free_dev;
1820                 }
1821
1822                 spin_lock_init(&tun->lock);
1823
1824                 err = security_tun_dev_alloc_security(&tun->security);
1825                 if (err < 0)
1826                         goto err_free_stat;
1827
1828                 tun_net_init(dev);
1829                 tun_flow_init(tun);
1830
1831                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1832                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1833                                    NETIF_F_HW_VLAN_STAG_TX;
1834                 dev->features = dev->hw_features | NETIF_F_LLTX;
1835                 dev->vlan_features = dev->features &
1836                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1837                                        NETIF_F_HW_VLAN_STAG_TX);
1838
1839                 INIT_LIST_HEAD(&tun->disabled);
1840                 err = tun_attach(tun, file, false);
1841                 if (err < 0)
1842                         goto err_free_flow;
1843
1844                 err = register_netdevice(tun->dev);
1845                 if (err < 0)
1846                         goto err_detach;
1847         }
1848
1849         netif_carrier_on(tun->dev);
1850
1851         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1852
1853         tun->flags = (tun->flags & ~TUN_FEATURES) |
1854                 (ifr->ifr_flags & TUN_FEATURES);
1855
1856         /* Make sure persistent devices do not get stuck in
1857          * xoff state.
1858          */
1859         if (netif_running(tun->dev))
1860                 netif_tx_wake_all_queues(tun->dev);
1861
1862         strcpy(ifr->ifr_name, tun->dev->name);
1863         return 0;
1864
1865 err_detach:
1866         tun_detach_all(dev);
1867 err_free_flow:
1868         tun_flow_uninit(tun);
1869         security_tun_dev_free_security(tun->security);
1870 err_free_stat:
1871         free_percpu(tun->pcpu_stats);
1872 err_free_dev:
1873         free_netdev(dev);
1874         return err;
1875 }
1876
1877 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1878                        struct ifreq *ifr)
1879 {
1880         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1881
1882         strcpy(ifr->ifr_name, tun->dev->name);
1883
1884         ifr->ifr_flags = tun_flags(tun);
1885
1886 }
1887
1888 /* This is like a cut-down ethtool ops, except done via tun fd so no
1889  * privs required. */
1890 static int set_offload(struct tun_struct *tun, unsigned long arg)
1891 {
1892         netdev_features_t features = 0;
1893
1894         if (arg & TUN_F_CSUM) {
1895                 features |= NETIF_F_HW_CSUM;
1896                 arg &= ~TUN_F_CSUM;
1897
1898                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1899                         if (arg & TUN_F_TSO_ECN) {
1900                                 features |= NETIF_F_TSO_ECN;
1901                                 arg &= ~TUN_F_TSO_ECN;
1902                         }
1903                         if (arg & TUN_F_TSO4)
1904                                 features |= NETIF_F_TSO;
1905                         if (arg & TUN_F_TSO6)
1906                                 features |= NETIF_F_TSO6;
1907                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1908                 }
1909
1910                 if (arg & TUN_F_UFO) {
1911                         features |= NETIF_F_UFO;
1912                         arg &= ~TUN_F_UFO;
1913                 }
1914         }
1915
1916         /* This gives the user a way to test for new features in future by
1917          * trying to set them. */
1918         if (arg)
1919                 return -EINVAL;
1920
1921         tun->set_features = features;
1922         netdev_update_features(tun->dev);
1923
1924         return 0;
1925 }
1926
1927 static void tun_detach_filter(struct tun_struct *tun, int n)
1928 {
1929         int i;
1930         struct tun_file *tfile;
1931
1932         for (i = 0; i < n; i++) {
1933                 tfile = rtnl_dereference(tun->tfiles[i]);
1934                 lock_sock(tfile->socket.sk);
1935                 sk_detach_filter(tfile->socket.sk);
1936                 release_sock(tfile->socket.sk);
1937         }
1938
1939         tun->filter_attached = false;
1940 }
1941
1942 static int tun_attach_filter(struct tun_struct *tun)
1943 {
1944         int i, ret = 0;
1945         struct tun_file *tfile;
1946
1947         for (i = 0; i < tun->numqueues; i++) {
1948                 tfile = rtnl_dereference(tun->tfiles[i]);
1949                 lock_sock(tfile->socket.sk);
1950                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1951                 release_sock(tfile->socket.sk);
1952                 if (ret) {
1953                         tun_detach_filter(tun, i);
1954                         return ret;
1955                 }
1956         }
1957
1958         tun->filter_attached = true;
1959         return ret;
1960 }
1961
1962 static void tun_set_sndbuf(struct tun_struct *tun)
1963 {
1964         struct tun_file *tfile;
1965         int i;
1966
1967         for (i = 0; i < tun->numqueues; i++) {
1968                 tfile = rtnl_dereference(tun->tfiles[i]);
1969                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1970         }
1971 }
1972
1973 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1974 {
1975         struct tun_file *tfile = file->private_data;
1976         struct tun_struct *tun;
1977         int ret = 0;
1978
1979         rtnl_lock();
1980
1981         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1982                 tun = tfile->detached;
1983                 if (!tun) {
1984                         ret = -EINVAL;
1985                         goto unlock;
1986                 }
1987                 ret = security_tun_dev_attach_queue(tun->security);
1988                 if (ret < 0)
1989                         goto unlock;
1990                 ret = tun_attach(tun, file, false);
1991         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1992                 tun = rtnl_dereference(tfile->tun);
1993                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1994                         ret = -EINVAL;
1995                 else
1996                         __tun_detach(tfile, false);
1997         } else
1998                 ret = -EINVAL;
1999
2000 unlock:
2001         rtnl_unlock();
2002         return ret;
2003 }
2004
2005 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2006                             unsigned long arg, int ifreq_len)
2007 {
2008         struct tun_file *tfile = file->private_data;
2009         struct tun_struct *tun;
2010         void __user* argp = (void __user*)arg;
2011         struct ifreq ifr;
2012         kuid_t owner;
2013         kgid_t group;
2014         int sndbuf;
2015         int vnet_hdr_sz;
2016         unsigned int ifindex;
2017         int le;
2018         int ret;
2019
2020         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2021                 if (copy_from_user(&ifr, argp, ifreq_len))
2022                         return -EFAULT;
2023         } else {
2024                 memset(&ifr, 0, sizeof(ifr));
2025         }
2026         if (cmd == TUNGETFEATURES) {
2027                 /* Currently this just means: "what IFF flags are valid?".
2028                  * This is needed because we never checked for invalid flags on
2029                  * TUNSETIFF.
2030                  */
2031                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2032                                 (unsigned int __user*)argp);
2033         } else if (cmd == TUNSETQUEUE)
2034                 return tun_set_queue(file, &ifr);
2035
2036         ret = 0;
2037         rtnl_lock();
2038
2039         tun = __tun_get(tfile);
2040         if (cmd == TUNSETIFF) {
2041                 ret = -EEXIST;
2042                 if (tun)
2043                         goto unlock;
2044
2045                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2046
2047                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2048
2049                 if (ret)
2050                         goto unlock;
2051
2052                 if (copy_to_user(argp, &ifr, ifreq_len))
2053                         ret = -EFAULT;
2054                 goto unlock;
2055         }
2056         if (cmd == TUNSETIFINDEX) {
2057                 ret = -EPERM;
2058                 if (tun)
2059                         goto unlock;
2060
2061                 ret = -EFAULT;
2062                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2063                         goto unlock;
2064
2065                 ret = 0;
2066                 tfile->ifindex = ifindex;
2067                 goto unlock;
2068         }
2069
2070         ret = -EBADFD;
2071         if (!tun)
2072                 goto unlock;
2073
2074         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2075
2076         ret = 0;
2077         switch (cmd) {
2078         case TUNGETIFF:
2079                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2080
2081                 if (tfile->detached)
2082                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2083                 if (!tfile->socket.sk->sk_filter)
2084                         ifr.ifr_flags |= IFF_NOFILTER;
2085
2086                 if (copy_to_user(argp, &ifr, ifreq_len))
2087                         ret = -EFAULT;
2088                 break;
2089
2090         case TUNSETNOCSUM:
2091                 /* Disable/Enable checksum */
2092
2093                 /* [unimplemented] */
2094                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2095                           arg ? "disabled" : "enabled");
2096                 break;
2097
2098         case TUNSETPERSIST:
2099                 /* Disable/Enable persist mode. Keep an extra reference to the
2100                  * module to prevent the module being unprobed.
2101                  */
2102                 if (arg && !(tun->flags & IFF_PERSIST)) {
2103                         tun->flags |= IFF_PERSIST;
2104                         __module_get(THIS_MODULE);
2105                 }
2106                 if (!arg && (tun->flags & IFF_PERSIST)) {
2107                         tun->flags &= ~IFF_PERSIST;
2108                         module_put(THIS_MODULE);
2109                 }
2110
2111                 tun_debug(KERN_INFO, tun, "persist %s\n",
2112                           arg ? "enabled" : "disabled");
2113                 break;
2114
2115         case TUNSETOWNER:
2116                 /* Set owner of the device */
2117                 owner = make_kuid(current_user_ns(), arg);
2118                 if (!uid_valid(owner)) {
2119                         ret = -EINVAL;
2120                         break;
2121                 }
2122                 tun->owner = owner;
2123                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2124                           from_kuid(&init_user_ns, tun->owner));
2125                 break;
2126
2127         case TUNSETGROUP:
2128                 /* Set group of the device */
2129                 group = make_kgid(current_user_ns(), arg);
2130                 if (!gid_valid(group)) {
2131                         ret = -EINVAL;
2132                         break;
2133                 }
2134                 tun->group = group;
2135                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2136                           from_kgid(&init_user_ns, tun->group));
2137                 break;
2138
2139         case TUNSETLINK:
2140                 /* Only allow setting the type when the interface is down */
2141                 if (tun->dev->flags & IFF_UP) {
2142                         tun_debug(KERN_INFO, tun,
2143                                   "Linktype set failed because interface is up\n");
2144                         ret = -EBUSY;
2145                 } else {
2146                         tun->dev->type = (int) arg;
2147                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2148                                   tun->dev->type);
2149                         ret = 0;
2150                 }
2151                 break;
2152
2153 #ifdef TUN_DEBUG
2154         case TUNSETDEBUG:
2155                 tun->debug = arg;
2156                 break;
2157 #endif
2158         case TUNSETOFFLOAD:
2159                 ret = set_offload(tun, arg);
2160                 break;
2161
2162         case TUNSETTXFILTER:
2163                 /* Can be set only for TAPs */
2164                 ret = -EINVAL;
2165                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2166                         break;
2167                 ret = update_filter(&tun->txflt, (void __user *)arg);
2168                 break;
2169
2170         case SIOCGIFHWADDR:
2171                 /* Get hw address */
2172                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2173                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2174                 if (copy_to_user(argp, &ifr, ifreq_len))
2175                         ret = -EFAULT;
2176                 break;
2177
2178         case SIOCSIFHWADDR:
2179                 /* Set hw address */
2180                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2181                           ifr.ifr_hwaddr.sa_data);
2182
2183                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2184                 break;
2185
2186         case TUNGETSNDBUF:
2187                 sndbuf = tfile->socket.sk->sk_sndbuf;
2188                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2189                         ret = -EFAULT;
2190                 break;
2191
2192         case TUNSETSNDBUF:
2193                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2194                         ret = -EFAULT;
2195                         break;
2196                 }
2197
2198                 tun->sndbuf = sndbuf;
2199                 tun_set_sndbuf(tun);
2200                 break;
2201
2202         case TUNGETVNETHDRSZ:
2203                 vnet_hdr_sz = tun->vnet_hdr_sz;
2204                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2205                         ret = -EFAULT;
2206                 break;
2207
2208         case TUNSETVNETHDRSZ:
2209                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2210                         ret = -EFAULT;
2211                         break;
2212                 }
2213                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2214                         ret = -EINVAL;
2215                         break;
2216                 }
2217
2218                 tun->vnet_hdr_sz = vnet_hdr_sz;
2219                 break;
2220
2221         case TUNGETVNETLE:
2222                 le = !!(tun->flags & TUN_VNET_LE);
2223                 if (put_user(le, (int __user *)argp))
2224                         ret = -EFAULT;
2225                 break;
2226
2227         case TUNSETVNETLE:
2228                 if (get_user(le, (int __user *)argp)) {
2229                         ret = -EFAULT;
2230                         break;
2231                 }
2232                 if (le)
2233                         tun->flags |= TUN_VNET_LE;
2234                 else
2235                         tun->flags &= ~TUN_VNET_LE;
2236                 break;
2237
2238         case TUNGETVNETBE:
2239                 ret = tun_get_vnet_be(tun, argp);
2240                 break;
2241
2242         case TUNSETVNETBE:
2243                 ret = tun_set_vnet_be(tun, argp);
2244                 break;
2245
2246         case TUNATTACHFILTER:
2247                 /* Can be set only for TAPs */
2248                 ret = -EINVAL;
2249                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2250                         break;
2251                 ret = -EFAULT;
2252                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2253                         break;
2254
2255                 ret = tun_attach_filter(tun);
2256                 break;
2257
2258         case TUNDETACHFILTER:
2259                 /* Can be set only for TAPs */
2260                 ret = -EINVAL;
2261                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2262                         break;
2263                 ret = 0;
2264                 tun_detach_filter(tun, tun->numqueues);
2265                 break;
2266
2267         case TUNGETFILTER:
2268                 ret = -EINVAL;
2269                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2270                         break;
2271                 ret = -EFAULT;
2272                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2273                         break;
2274                 ret = 0;
2275                 break;
2276
2277         default:
2278                 ret = -EINVAL;
2279                 break;
2280         }
2281
2282 unlock:
2283         rtnl_unlock();
2284         if (tun)
2285                 tun_put(tun);
2286         return ret;
2287 }
2288
2289 static long tun_chr_ioctl(struct file *file,
2290                           unsigned int cmd, unsigned long arg)
2291 {
2292         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2293 }
2294
2295 #ifdef CONFIG_COMPAT
2296 static long tun_chr_compat_ioctl(struct file *file,
2297                          unsigned int cmd, unsigned long arg)
2298 {
2299         switch (cmd) {
2300         case TUNSETIFF:
2301         case TUNGETIFF:
2302         case TUNSETTXFILTER:
2303         case TUNGETSNDBUF:
2304         case TUNSETSNDBUF:
2305         case SIOCGIFHWADDR:
2306         case SIOCSIFHWADDR:
2307                 arg = (unsigned long)compat_ptr(arg);
2308                 break;
2309         default:
2310                 arg = (compat_ulong_t)arg;
2311                 break;
2312         }
2313
2314         /*
2315          * compat_ifreq is shorter than ifreq, so we must not access beyond
2316          * the end of that structure. All fields that are used in this
2317          * driver are compatible though, we don't need to convert the
2318          * contents.
2319          */
2320         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2321 }
2322 #endif /* CONFIG_COMPAT */
2323
2324 static int tun_chr_fasync(int fd, struct file *file, int on)
2325 {
2326         struct tun_file *tfile = file->private_data;
2327         int ret;
2328
2329         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2330                 goto out;
2331
2332         if (on) {
2333                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2334                 tfile->flags |= TUN_FASYNC;
2335         } else
2336                 tfile->flags &= ~TUN_FASYNC;
2337         ret = 0;
2338 out:
2339         return ret;
2340 }
2341
2342 static int tun_chr_open(struct inode *inode, struct file * file)
2343 {
2344         struct net *net = current->nsproxy->net_ns;
2345         struct tun_file *tfile;
2346
2347         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2348
2349         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2350                                             &tun_proto, 0);
2351         if (!tfile)
2352                 return -ENOMEM;
2353         RCU_INIT_POINTER(tfile->tun, NULL);
2354         tfile->flags = 0;
2355         tfile->ifindex = 0;
2356
2357         init_waitqueue_head(&tfile->wq.wait);
2358         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2359
2360         tfile->socket.file = file;
2361         tfile->socket.ops = &tun_socket_ops;
2362
2363         sock_init_data(&tfile->socket, &tfile->sk);
2364
2365         tfile->sk.sk_write_space = tun_sock_write_space;
2366         tfile->sk.sk_sndbuf = INT_MAX;
2367
2368         file->private_data = tfile;
2369         INIT_LIST_HEAD(&tfile->next);
2370
2371         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2372
2373         return 0;
2374 }
2375
2376 static int tun_chr_close(struct inode *inode, struct file *file)
2377 {
2378         struct tun_file *tfile = file->private_data;
2379
2380         tun_detach(tfile, true);
2381
2382         return 0;
2383 }
2384
2385 #ifdef CONFIG_PROC_FS
2386 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2387 {
2388         struct tun_struct *tun;
2389         struct ifreq ifr;
2390
2391         memset(&ifr, 0, sizeof(ifr));
2392
2393         rtnl_lock();
2394         tun = tun_get(f);
2395         if (tun)
2396                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2397         rtnl_unlock();
2398
2399         if (tun)
2400                 tun_put(tun);
2401
2402         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2403 }
2404 #endif
2405
2406 static const struct file_operations tun_fops = {
2407         .owner  = THIS_MODULE,
2408         .llseek = no_llseek,
2409         .read_iter  = tun_chr_read_iter,
2410         .write_iter = tun_chr_write_iter,
2411         .poll   = tun_chr_poll,
2412         .unlocked_ioctl = tun_chr_ioctl,
2413 #ifdef CONFIG_COMPAT
2414         .compat_ioctl = tun_chr_compat_ioctl,
2415 #endif
2416         .open   = tun_chr_open,
2417         .release = tun_chr_close,
2418         .fasync = tun_chr_fasync,
2419 #ifdef CONFIG_PROC_FS
2420         .show_fdinfo = tun_chr_show_fdinfo,
2421 #endif
2422 };
2423
2424 static struct miscdevice tun_miscdev = {
2425         .minor = TUN_MINOR,
2426         .name = "tun",
2427         .nodename = "net/tun",
2428         .fops = &tun_fops,
2429 };
2430
2431 /* ethtool interface */
2432
2433 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2434 {
2435         cmd->supported          = 0;
2436         cmd->advertising        = 0;
2437         ethtool_cmd_speed_set(cmd, SPEED_10);
2438         cmd->duplex             = DUPLEX_FULL;
2439         cmd->port               = PORT_TP;
2440         cmd->phy_address        = 0;
2441         cmd->transceiver        = XCVR_INTERNAL;
2442         cmd->autoneg            = AUTONEG_DISABLE;
2443         cmd->maxtxpkt           = 0;
2444         cmd->maxrxpkt           = 0;
2445         return 0;
2446 }
2447
2448 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2449 {
2450         struct tun_struct *tun = netdev_priv(dev);
2451
2452         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2453         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2454
2455         switch (tun->flags & TUN_TYPE_MASK) {
2456         case IFF_TUN:
2457                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2458                 break;
2459         case IFF_TAP:
2460                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2461                 break;
2462         }
2463 }
2464
2465 static u32 tun_get_msglevel(struct net_device *dev)
2466 {
2467 #ifdef TUN_DEBUG
2468         struct tun_struct *tun = netdev_priv(dev);
2469         return tun->debug;
2470 #else
2471         return -EOPNOTSUPP;
2472 #endif
2473 }
2474
2475 static void tun_set_msglevel(struct net_device *dev, u32 value)
2476 {
2477 #ifdef TUN_DEBUG
2478         struct tun_struct *tun = netdev_priv(dev);
2479         tun->debug = value;
2480 #endif
2481 }
2482
2483 static int tun_get_coalesce(struct net_device *dev,
2484                             struct ethtool_coalesce *ec)
2485 {
2486         struct tun_struct *tun = netdev_priv(dev);
2487
2488         ec->rx_max_coalesced_frames = tun->rx_batched;
2489
2490         return 0;
2491 }
2492
2493 static int tun_set_coalesce(struct net_device *dev,
2494                             struct ethtool_coalesce *ec)
2495 {
2496         struct tun_struct *tun = netdev_priv(dev);
2497
2498         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
2499                 tun->rx_batched = NAPI_POLL_WEIGHT;
2500         else
2501                 tun->rx_batched = ec->rx_max_coalesced_frames;
2502
2503         return 0;
2504 }
2505
2506 static const struct ethtool_ops tun_ethtool_ops = {
2507         .get_settings   = tun_get_settings,
2508         .get_drvinfo    = tun_get_drvinfo,
2509         .get_msglevel   = tun_get_msglevel,
2510         .set_msglevel   = tun_set_msglevel,
2511         .get_link       = ethtool_op_get_link,
2512         .get_ts_info    = ethtool_op_get_ts_info,
2513         .get_coalesce   = tun_get_coalesce,
2514         .set_coalesce   = tun_set_coalesce,
2515 };
2516
2517 static int tun_queue_resize(struct tun_struct *tun)
2518 {
2519         struct net_device *dev = tun->dev;
2520         struct tun_file *tfile;
2521         struct skb_array **arrays;
2522         int n = tun->numqueues + tun->numdisabled;
2523         int ret, i;
2524
2525         arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL);
2526         if (!arrays)
2527                 return -ENOMEM;
2528
2529         for (i = 0; i < tun->numqueues; i++) {
2530                 tfile = rtnl_dereference(tun->tfiles[i]);
2531                 arrays[i] = &tfile->tx_array;
2532         }
2533         list_for_each_entry(tfile, &tun->disabled, next)
2534                 arrays[i++] = &tfile->tx_array;
2535
2536         ret = skb_array_resize_multiple(arrays, n,
2537                                         dev->tx_queue_len, GFP_KERNEL);
2538
2539         kfree(arrays);
2540         return ret;
2541 }
2542
2543 static int tun_device_event(struct notifier_block *unused,
2544                             unsigned long event, void *ptr)
2545 {
2546         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2547         struct tun_struct *tun = netdev_priv(dev);
2548
2549         if (dev->rtnl_link_ops != &tun_link_ops)
2550                 return NOTIFY_DONE;
2551
2552         switch (event) {
2553         case NETDEV_CHANGE_TX_QUEUE_LEN:
2554                 if (tun_queue_resize(tun))
2555                         return NOTIFY_BAD;
2556                 break;
2557         default:
2558                 break;
2559         }
2560
2561         return NOTIFY_DONE;
2562 }
2563
2564 static struct notifier_block tun_notifier_block __read_mostly = {
2565         .notifier_call  = tun_device_event,
2566 };
2567
2568 static int __init tun_init(void)
2569 {
2570         int ret = 0;
2571
2572         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2573         pr_info("%s\n", DRV_COPYRIGHT);
2574
2575         ret = rtnl_link_register(&tun_link_ops);
2576         if (ret) {
2577                 pr_err("Can't register link_ops\n");
2578                 goto err_linkops;
2579         }
2580
2581         ret = misc_register(&tun_miscdev);
2582         if (ret) {
2583                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2584                 goto err_misc;
2585         }
2586
2587         register_netdevice_notifier(&tun_notifier_block);
2588         return  0;
2589 err_misc:
2590         rtnl_link_unregister(&tun_link_ops);
2591 err_linkops:
2592         return ret;
2593 }
2594
2595 static void tun_cleanup(void)
2596 {
2597         misc_deregister(&tun_miscdev);
2598         rtnl_link_unregister(&tun_link_ops);
2599         unregister_netdevice_notifier(&tun_notifier_block);
2600 }
2601
2602 /* Get an underlying socket object from tun file.  Returns error unless file is
2603  * attached to a device.  The returned object works like a packet socket, it
2604  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2605  * holding a reference to the file for as long as the socket is in use. */
2606 struct socket *tun_get_socket(struct file *file)
2607 {
2608         struct tun_file *tfile;
2609         if (file->f_op != &tun_fops)
2610                 return ERR_PTR(-EINVAL);
2611         tfile = file->private_data;
2612         if (!tfile)
2613                 return ERR_PTR(-EBADFD);
2614         return &tfile->socket;
2615 }
2616 EXPORT_SYMBOL_GPL(tun_get_socket);
2617
2618 module_init(tun_init);
2619 module_exit(tun_cleanup);
2620 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2621 MODULE_AUTHOR(DRV_COPYRIGHT);
2622 MODULE_LICENSE("GPL");
2623 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2624 MODULE_ALIAS("devname:net/tun");