drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <[email protected]>
   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 <[email protected]> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <[email protected]>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <[email protected]>  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 <[email protected]>
  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 <[email protected]>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/major.h>
  48 #include <linux/slab.h>
  49 #include <linux/poll.h>
  50 #include <linux/fcntl.h>
  51 #include <linux/init.h>
  52 #include <linux/skbuff.h>
  53 #include <linux/netdevice.h>
  54 #include <linux/etherdevice.h>
  55 #include <linux/miscdevice.h>
  56 #include <linux/ethtool.h>
  57 #include <linux/rtnetlink.h>
  58 #include <linux/compat.h>
  59 #include <linux/if.h>
  60 #include <linux/if_arp.h>
  61 #include <linux/if_ether.h>
  62 #include <linux/if_tun.h>
  63 #include <linux/if_vlan.h>
  64 #include <linux/crc32.h>
  65 #include <linux/nsproxy.h>
  66 #include <linux/virtio_net.h>
  67 #include <linux/rcupdate.h>
  68 #include <net/net_namespace.h>
  69 #include <net/netns/generic.h>
  70 #include <net/rtnetlink.h>
  71 #include <net/sock.h>
  72 #include <linux/seq_file.h>
  73 #include <linux/uio.h>
  74 #include <linux/skb_array.h>
  75
  76 #include <asm/uaccess.h>
  77
  78 /* Uncomment to enable debugging */
  79 /* #define TUN_DEBUG 1 */
  80
  81 #ifdef TUN_DEBUG
  82 static int debug;
  83
  84 #define tun_debug(level, tun, fmt, args...)                     \
  85 do {                                                            \
  86         if (tun->debug)                                         \
  87                 netdev_printk(level, tun->dev, fmt, ##args);    \
  88 } while (0)
  89 #define DBG1(level, fmt, args...)                               \
  90 do {                                                            \
  91         if (debug == 2)                                         \
  92                 printk(level fmt, ##args);                      \
  93 } while (0)
  94 #else
  95 #define tun_debug(level, tun, fmt, args...)                     \
  96 do {                                                            \
  97         if (0)                                                  \
  98                 netdev_printk(level, tun->dev, fmt, ##args);    \
  99 } while (0)
 100 #define DBG1(level, fmt, args...)                               \
 101 do {                                                            \
 102         if (0)                                                  \
 103                 printk(level fmt, ##args);                      \
 104 } while (0)
 105 #endif
 106
 107 /* TUN device flags */
 108
 109 /* IFF_ATTACH_QUEUE is never stored in device flags,
 110  * overload it to mean fasync when stored there.
 111  */
 112 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 113 /* High bits in flags field are unused. */
 114 #define TUN_VNET_LE     0x80000000
 115 #define TUN_VNET_BE     0x40000000
 116
 117 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 118                       IFF_MULTI_QUEUE)
 119 #define GOODCOPY_LEN 128
 120
 121 #define FLT_EXACT_COUNT 8
 122 struct tap_filter {
 123         unsigned int    count;    /* Number of addrs. Zero means disabled */
 124         u32             mask[2];  /* Mask of the hashed addrs */
 125         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 126 };
 127
 128 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 129  * to max number of VCPUs in guest. */
 130 #define MAX_TAP_QUEUES 256
 131 #define MAX_TAP_FLOWS  4096
 132
 133 #define TUN_FLOW_EXPIRE (3 * HZ)
 134
 135 struct tun_pcpu_stats {
 136         u64 rx_packets;
 137         u64 rx_bytes;
 138         u64 tx_packets;
 139         u64 tx_bytes;
 140         struct u64_stats_sync syncp;
 141         u32 rx_dropped;
 142         u32 tx_dropped;
 143         u32 rx_frame_errors;
 144 };
 145
 146 /* A tun_file connects an open character device to a tuntap netdevice. It
 147  * also contains all socket related structures (except sock_fprog and tap_filter)
 148  * to serve as one transmit queue for tuntap device. The sock_fprog and
 149  * tap_filter were kept in tun_struct since they were used for filtering for the
 150  * netdevice not for a specific queue (at least I didn't see the requirement for
 151  * this).
 152  *
 153  * RCU usage:
 154  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 155  * other can only be read while rcu_read_lock or rtnl_lock is held.
 156  */
 157 struct tun_file {
 158         struct sock sk;
 159         struct socket socket;
 160         struct socket_wq wq;
 161         struct tun_struct __rcu *tun;
 162         struct fasync_struct *fasync;
 163         /* only used for fasnyc */
 164         unsigned int flags;
 165         union {
 166                 u16 queue_index;
 167                 unsigned int ifindex;
 168         };
 169         struct list_head next;
 170         struct tun_struct *detached;
 171         struct skb_array tx_array;
 172 };
 173
 174 struct tun_flow_entry {
 175         struct hlist_node hash_link;
 176         struct rcu_head rcu;
 177         struct tun_struct *tun;
 178
 179         u32 rxhash;
 180         u32 rps_rxhash;
 181         int queue_index;
 182         unsigned long updated;
 183 };
 184
 185 #define TUN_NUM_FLOW_ENTRIES 1024
 186
 187 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 188  * device, socket filter, sndbuf and vnet header size were restore when the
 189  * file were attached to a persist device.
 190  */
 191 struct tun_struct {
 192         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 193         unsigned int            numqueues;
 194         unsigned int            flags;
 195         kuid_t                  owner;
 196         kgid_t                  group;
 197
 198         struct net_device       *dev;
 199         netdev_features_t       set_features;
 200 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 201                           NETIF_F_TSO6|NETIF_F_UFO)
 202
 203         int                     align;
 204         int                     vnet_hdr_sz;
 205         int                     sndbuf;
 206         struct tap_filter       txflt;
 207         struct sock_fprog       fprog;
 208         /* protected by rtnl lock */
 209         bool                    filter_attached;
 210 #ifdef TUN_DEBUG
 211         int debug;
 212 #endif
 213         spinlock_t lock;
 214         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 215         struct timer_list flow_gc_timer;
 216         unsigned long ageing_time;
 217         unsigned int numdisabled;
 218         struct list_head disabled;
 219         void *security;
 220         u32 flow_count;
 221         struct tun_pcpu_stats __percpu *pcpu_stats;
 222 };
 223
 224 #ifdef CONFIG_TUN_VNET_CROSS_LE
 225 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 226 {
 227         return tun->flags & TUN_VNET_BE ? false :
 228                 virtio_legacy_is_little_endian();
 229 }
 230
 231 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 232 {
 233         int be = !!(tun->flags & TUN_VNET_BE);
 234
 235         if (put_user(be, argp))
 236                 return -EFAULT;
 237
 238         return 0;
 239 }
 240
 241 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 242 {
 243         int be;
 244
 245         if (get_user(be, argp))
 246                 return -EFAULT;
 247
 248         if (be)
 249                 tun->flags |= TUN_VNET_BE;
 250         else
 251                 tun->flags &= ~TUN_VNET_BE;
 252
 253         return 0;
 254 }
 255 #else
 256 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 257 {
 258         return virtio_legacy_is_little_endian();
 259 }
 260
 261 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 262 {
 263         return -EINVAL;
 264 }
 265
 266 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 267 {
 268         return -EINVAL;
 269 }
 270 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 271
 272 static inline bool tun_is_little_endian(struct tun_struct *tun)
 273 {
 274         return tun->flags & TUN_VNET_LE ||
 275                 tun_legacy_is_little_endian(tun);
 276 }
 277
 278 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 279 {
 280         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 281 }
 282
 283 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 284 {
 285         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 286 }
 287
 288 static inline u32 tun_hashfn(u32 rxhash)
 289 {
 290         return rxhash & 0x3ff;
 291 }
 292
 293 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 294 {
 295         struct tun_flow_entry *e;
 296
 297         hlist_for_each_entry_rcu(e, head, hash_link) {
 298                 if (e->rxhash == rxhash)
 299                         return e;
 300         }
 301         return NULL;
 302 }
 303
 304 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 305                                               struct hlist_head *head,
 306                                               u32 rxhash, u16 queue_index)
 307 {
 308         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 309
 310         if (e) {
 311                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 312                           rxhash, queue_index);
 313                 e->updated = jiffies;
 314                 e->rxhash = rxhash;
 315                 e->rps_rxhash = 0;
 316                 e->queue_index = queue_index;
 317                 e->tun = tun;
 318                 hlist_add_head_rcu(&e->hash_link, head);
 319                 ++tun->flow_count;
 320         }
 321         return e;
 322 }
 323
 324 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 325 {
 326         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 327                   e->rxhash, e->queue_index);
 328         hlist_del_rcu(&e->hash_link);
 329         kfree_rcu(e, rcu);
 330         --tun->flow_count;
 331 }
 332
 333 static void tun_flow_flush(struct tun_struct *tun)
 334 {
 335         int i;
 336
 337         spin_lock_bh(&tun->lock);
 338         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 339                 struct tun_flow_entry *e;
 340                 struct hlist_node *n;
 341
 342                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 343                         tun_flow_delete(tun, e);
 344         }
 345         spin_unlock_bh(&tun->lock);
 346 }
 347
 348 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 349 {
 350         int i;
 351
 352         spin_lock_bh(&tun->lock);
 353         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 354                 struct tun_flow_entry *e;
 355                 struct hlist_node *n;
 356
 357                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 358                         if (e->queue_index == queue_index)
 359                                 tun_flow_delete(tun, e);
 360                 }
 361         }
 362         spin_unlock_bh(&tun->lock);
 363 }
 364
 365 static void tun_flow_cleanup(unsigned long data)
 366 {
 367         struct tun_struct *tun = (struct tun_struct *)data;
 368         unsigned long delay = tun->ageing_time;
 369         unsigned long next_timer = jiffies + delay;
 370         unsigned long count = 0;
 371         int i;
 372
 373         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 374
 375         spin_lock_bh(&tun->lock);
 376         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 377                 struct tun_flow_entry *e;
 378                 struct hlist_node *n;
 379
 380                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 381                         unsigned long this_timer;
 382                         count++;
 383                         this_timer = e->updated + delay;
 384                         if (time_before_eq(this_timer, jiffies))
 385                                 tun_flow_delete(tun, e);
 386                         else if (time_before(this_timer, next_timer))
 387                                 next_timer = this_timer;
 388                 }
 389         }
 390
 391         if (count)
 392                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 393         spin_unlock_bh(&tun->lock);
 394 }
 395
 396 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 397                             struct tun_file *tfile)
 398 {
 399         struct hlist_head *head;
 400         struct tun_flow_entry *e;
 401         unsigned long delay = tun->ageing_time;
 402         u16 queue_index = tfile->queue_index;
 403
 404         if (!rxhash)
 405                 return;
 406         else
 407                 head = &tun->flows[tun_hashfn(rxhash)];
 408
 409         rcu_read_lock();
 410
 411         /* We may get a very small possibility of OOO during switching, not
 412          * worth to optimize.*/
 413         if (tun->numqueues == 1 || tfile->detached)
 414                 goto unlock;
 415
 416         e = tun_flow_find(head, rxhash);
 417         if (likely(e)) {
 418                 /* TODO: keep queueing to old queue until it's empty? */
 419                 e->queue_index = queue_index;
 420                 e->updated = jiffies;
 421                 sock_rps_record_flow_hash(e->rps_rxhash);
 422         } else {
 423                 spin_lock_bh(&tun->lock);
 424                 if (!tun_flow_find(head, rxhash) &&
 425                     tun->flow_count < MAX_TAP_FLOWS)
 426                         tun_flow_create(tun, head, rxhash, queue_index);
 427
 428                 if (!timer_pending(&tun->flow_gc_timer))
 429                         mod_timer(&tun->flow_gc_timer,
 430                                   round_jiffies_up(jiffies + delay));
 431                 spin_unlock_bh(&tun->lock);
 432         }
 433
 434 unlock:
 435         rcu_read_unlock();
 436 }
 437
 438 /**
 439  * Save the hash received in the stack receive path and update the
 440  * flow_hash table accordingly.
 441  */
 442 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 443 {
 444         if (unlikely(e->rps_rxhash != hash))
 445                 e->rps_rxhash = hash;
 446 }
 447
 448 /* We try to identify a flow through its rxhash first. The reason that
 449  * we do not check rxq no. is because some cards(e.g 82599), chooses
 450  * the rxq based on the txq where the last packet of the flow comes. As
 451  * the userspace application move between processors, we may get a
 452  * different rxq no. here. If we could not get rxhash, then we would
 453  * hope the rxq no. may help here.
 454  */
 455 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 456                             void *accel_priv, select_queue_fallback_t fallback)
 457 {
 458         struct tun_struct *tun = netdev_priv(dev);
 459         struct tun_flow_entry *e;
 460         u32 txq = 0;
 461         u32 numqueues = 0;
 462
 463         rcu_read_lock();
 464         numqueues = ACCESS_ONCE(tun->numqueues);
 465
 466         txq = skb_get_hash(skb);
 467         if (txq) {
 468                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 469                 if (e) {
 470                         tun_flow_save_rps_rxhash(e, txq);
 471                         txq = e->queue_index;
 472                 } else
 473                         /* use multiply and shift instead of expensive divide */
 474                         txq = ((u64)txq * numqueues) >> 32;
 475         } else if (likely(skb_rx_queue_recorded(skb))) {
 476                 txq = skb_get_rx_queue(skb);
 477                 while (unlikely(txq >= numqueues))
 478                         txq -= numqueues;
 479         }
 480
 481         rcu_read_unlock();
 482         return txq;
 483 }
 484
 485 static inline bool tun_not_capable(struct tun_struct *tun)
 486 {
 487         const struct cred *cred = current_cred();
 488         struct net *net = dev_net(tun->dev);
 489
 490         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 491                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 492                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 493 }
 494
 495 static void tun_set_real_num_queues(struct tun_struct *tun)
 496 {
 497         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 498         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 499 }
 500
 501 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 502 {
 503         tfile->detached = tun;
 504         list_add_tail(&tfile->next, &tun->disabled);
 505         ++tun->numdisabled;
 506 }
 507
 508 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 509 {
 510         struct tun_struct *tun = tfile->detached;
 511
 512         tfile->detached = NULL;
 513         list_del_init(&tfile->next);
 514         --tun->numdisabled;
 515         return tun;
 516 }
 517
 518 static void tun_queue_purge(struct tun_file *tfile)
 519 {
 520         struct sk_buff *skb;
 521
 522         while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
 523                 kfree_skb(skb);
 524
 525         skb_queue_purge(&tfile->sk.sk_error_queue);
 526 }
 527
 528 static void __tun_detach(struct tun_file *tfile, bool clean)
 529 {
 530         struct tun_file *ntfile;
 531         struct tun_struct *tun;
 532
 533         tun = rtnl_dereference(tfile->tun);
 534
 535         if (tun && !tfile->detached) {
 536                 u16 index = tfile->queue_index;
 537                 BUG_ON(index >= tun->numqueues);
 538
 539                 rcu_assign_pointer(tun->tfiles[index],
 540                                    tun->tfiles[tun->numqueues - 1]);
 541                 ntfile = rtnl_dereference(tun->tfiles[index]);
 542                 ntfile->queue_index = index;
 543
 544                 --tun->numqueues;
 545                 if (clean) {
 546                         RCU_INIT_POINTER(tfile->tun, NULL);
 547                         sock_put(&tfile->sk);
 548                 } else
 549                         tun_disable_queue(tun, tfile);
 550
 551                 synchronize_net();
 552                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 553                 /* Drop read queue */
 554                 tun_queue_purge(tfile);
 555                 tun_set_real_num_queues(tun);
 556         } else if (tfile->detached && clean) {
 557                 tun = tun_enable_queue(tfile);
 558                 sock_put(&tfile->sk);
 559         }
 560
 561         if (clean) {
 562                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 563                         netif_carrier_off(tun->dev);
 564
 565                         if (!(tun->flags & IFF_PERSIST) &&
 566                             tun->dev->reg_state == NETREG_REGISTERED)
 567                                 unregister_netdevice(tun->dev);
 568                 }
 569                 if (tun)
 570                         skb_array_cleanup(&tfile->tx_array);
 571                 sock_put(&tfile->sk);
 572         }
 573 }
 574
 575 static void tun_detach(struct tun_file *tfile, bool clean)
 576 {
 577         rtnl_lock();
 578         __tun_detach(tfile, clean);
 579         rtnl_unlock();
 580 }
 581
 582 static void tun_detach_all(struct net_device *dev)
 583 {
 584         struct tun_struct *tun = netdev_priv(dev);
 585         struct tun_file *tfile, *tmp;
 586         int i, n = tun->numqueues;
 587
 588         for (i = 0; i < n; i++) {
 589                 tfile = rtnl_dereference(tun->tfiles[i]);
 590                 BUG_ON(!tfile);
 591                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 592                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 593                 RCU_INIT_POINTER(tfile->tun, NULL);
 594                 --tun->numqueues;
 595         }
 596         list_for_each_entry(tfile, &tun->disabled, next) {
 597                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 598                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 599                 RCU_INIT_POINTER(tfile->tun, NULL);
 600         }
 601         BUG_ON(tun->numqueues != 0);
 602
 603         synchronize_net();
 604         for (i = 0; i < n; i++) {
 605                 tfile = rtnl_dereference(tun->tfiles[i]);
 606                 /* Drop read queue */
 607                 tun_queue_purge(tfile);
 608                 sock_put(&tfile->sk);
 609         }
 610         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 611                 tun_enable_queue(tfile);
 612                 tun_queue_purge(tfile);
 613                 sock_put(&tfile->sk);
 614         }
 615         BUG_ON(tun->numdisabled != 0);
 616
 617         if (tun->flags & IFF_PERSIST)
 618                 module_put(THIS_MODULE);
 619 }
 620
 621 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 622 {
 623         struct tun_file *tfile = file->private_data;
 624         struct net_device *dev = tun->dev;
 625         int err;
 626
 627         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 628         if (err < 0)
 629                 goto out;
 630
 631         err = -EINVAL;
 632         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 633                 goto out;
 634
 635         err = -EBUSY;
 636         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 637                 goto out;
 638
 639         err = -E2BIG;
 640         if (!tfile->detached &&
 641             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 642                 goto out;
 643
 644         err = 0;
 645
 646         /* Re-attach the filter to persist device */
 647         if (!skip_filter && (tun->filter_attached == true)) {
 648                 lock_sock(tfile->socket.sk);
 649                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 650                 release_sock(tfile->socket.sk);
 651                 if (!err)
 652                         goto out;
 653         }
 654
 655         if (!tfile->detached &&
 656             skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
 657                 err = -ENOMEM;
 658                 goto out;
 659         }
 660
 661         tfile->queue_index = tun->numqueues;
 662         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 663         rcu_assign_pointer(tfile->tun, tun);
 664         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 665         tun->numqueues++;
 666
 667         if (tfile->detached)
 668                 tun_enable_queue(tfile);
 669         else
 670                 sock_hold(&tfile->sk);
 671
 672         tun_set_real_num_queues(tun);
 673
 674         /* device is allowed to go away first, so no need to hold extra
 675          * refcnt.
 676          */
 677
 678 out:
 679         return err;
 680 }
 681
 682 static struct tun_struct *__tun_get(struct tun_file *tfile)
 683 {
 684         struct tun_struct *tun;
 685
 686         rcu_read_lock();
 687         tun = rcu_dereference(tfile->tun);
 688         if (tun)
 689                 dev_hold(tun->dev);
 690         rcu_read_unlock();
 691
 692         return tun;
 693 }
 694
 695 static struct tun_struct *tun_get(struct file *file)
 696 {
 697         return __tun_get(file->private_data);
 698 }
 699
 700 static void tun_put(struct tun_struct *tun)
 701 {
 702         dev_put(tun->dev);
 703 }
 704
 705 /* TAP filtering */
 706 static void addr_hash_set(u32 *mask, const u8 *addr)
 707 {
 708         int n = ether_crc(ETH_ALEN, addr) >> 26;
 709         mask[n >> 5] |= (1 << (n & 31));
 710 }
 711
 712 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 713 {
 714         int n = ether_crc(ETH_ALEN, addr) >> 26;
 715         return mask[n >> 5] & (1 << (n & 31));
 716 }
 717
 718 static int update_filter(struct tap_filter *filter, void __user *arg)
 719 {
 720         struct { u8 u[ETH_ALEN]; } *addr;
 721         struct tun_filter uf;
 722         int err, alen, n, nexact;
 723
 724         if (copy_from_user(&uf, arg, sizeof(uf)))
 725                 return -EFAULT;
 726
 727         if (!uf.count) {
 728                 /* Disabled */
 729                 filter->count = 0;
 730                 return 0;
 731         }
 732
 733         alen = ETH_ALEN * uf.count;
 734         addr = kmalloc(alen, GFP_KERNEL);
 735         if (!addr)
 736                 return -ENOMEM;
 737
 738         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 739                 err = -EFAULT;
 740                 goto done;
 741         }
 742
 743         /* The filter is updated without holding any locks. Which is
 744          * perfectly safe. We disable it first and in the worst
 745          * case we'll accept a few undesired packets. */
 746         filter->count = 0;
 747         wmb();
 748
 749         /* Use first set of addresses as an exact filter */
 750         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 751                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 752
 753         nexact = n;
 754
 755         /* Remaining multicast addresses are hashed,
 756          * unicast will leave the filter disabled. */
 757         memset(filter->mask, 0, sizeof(filter->mask));
 758         for (; n < uf.count; n++) {
 759                 if (!is_multicast_ether_addr(addr[n].u)) {
 760                         err = 0; /* no filter */
 761                         goto done;
 762                 }
 763                 addr_hash_set(filter->mask, addr[n].u);
 764         }
 765
 766         /* For ALLMULTI just set the mask to all ones.
 767          * This overrides the mask populated above. */
 768         if ((uf.flags & TUN_FLT_ALLMULTI))
 769                 memset(filter->mask, ~0, sizeof(filter->mask));
 770
 771         /* Now enable the filter */
 772         wmb();
 773         filter->count = nexact;
 774
 775         /* Return the number of exact filters */
 776         err = nexact;
 777
 778 done:
 779         kfree(addr);
 780         return err;
 781 }
 782
 783 /* Returns: 0 - drop, !=0 - accept */
 784 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 785 {
 786         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 787          * at this point. */
 788         struct ethhdr *eh = (struct ethhdr *) skb->data;
 789         int i;
 790
 791         /* Exact match */
 792         for (i = 0; i < filter->count; i++)
 793                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 794                         return 1;
 795
 796         /* Inexact match (multicast only) */
 797         if (is_multicast_ether_addr(eh->h_dest))
 798                 return addr_hash_test(filter->mask, eh->h_dest);
 799
 800         return 0;
 801 }
 802
 803 /*
 804  * Checks whether the packet is accepted or not.
 805  * Returns: 0 - drop, !=0 - accept
 806  */
 807 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 808 {
 809         if (!filter->count)
 810                 return 1;
 811
 812         return run_filter(filter, skb);
 813 }
 814
 815 /* Network device part of the driver */
 816
 817 static const struct ethtool_ops tun_ethtool_ops;
 818
 819 /* Net device detach from fd. */
 820 static void tun_net_uninit(struct net_device *dev)
 821 {
 822         tun_detach_all(dev);
 823 }
 824
 825 /* Net device open. */
 826 static int tun_net_open(struct net_device *dev)
 827 {
 828         netif_tx_start_all_queues(dev);
 829         return 0;
 830 }
 831
 832 /* Net device close. */
 833 static int tun_net_close(struct net_device *dev)
 834 {
 835         netif_tx_stop_all_queues(dev);
 836         return 0;
 837 }
 838
 839 /* Net device start xmit */
 840 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 841 {
 842         struct tun_struct *tun = netdev_priv(dev);
 843         int txq = skb->queue_mapping;
 844         struct tun_file *tfile;
 845         u32 numqueues = 0;
 846
 847         rcu_read_lock();
 848         tfile = rcu_dereference(tun->tfiles[txq]);
 849         numqueues = ACCESS_ONCE(tun->numqueues);
 850
 851         /* Drop packet if interface is not attached */
 852         if (txq >= numqueues)
 853                 goto drop;
 854
 855 #ifdef CONFIG_RPS
 856         if (numqueues == 1 && static_key_false(&rps_needed)) {
 857                 /* Select queue was not called for the skbuff, so we extract the
 858                  * RPS hash and save it into the flow_table here.
 859                  */
 860                 __u32 rxhash;
 861
 862                 rxhash = skb_get_hash(skb);
 863                 if (rxhash) {
 864                         struct tun_flow_entry *e;
 865                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 866                                         rxhash);
 867                         if (e)
 868                                 tun_flow_save_rps_rxhash(e, rxhash);
 869                 }
 870         }
 871 #endif
 872
 873         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 874
 875         BUG_ON(!tfile);
 876
 877         /* Drop if the filter does not like it.
 878          * This is a noop if the filter is disabled.
 879          * Filter can be enabled only for the TAP devices. */
 880         if (!check_filter(&tun->txflt, skb))
 881                 goto drop;
 882
 883         if (tfile->socket.sk->sk_filter &&
 884             sk_filter(tfile->socket.sk, skb))
 885                 goto drop;
 886
 887         /* Limit the number of packets queued by dividing txq length with the
 888          * number of queues.
 889          */
 890         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
 891                           >= dev->tx_queue_len)
 892                 goto drop;
 893
 894         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 895                 goto drop;
 896
 897         if (skb->sk && sk_fullsock(skb->sk)) {
 898                 sock_tx_timestamp(skb->sk, skb->sk->sk_tsflags,
 899                                   &skb_shinfo(skb)->tx_flags);
 900                 sw_tx_timestamp(skb);
 901         }
 902
 903         /* Orphan the skb - required as we might hang on to it
 904          * for indefinite time.
 905          */
 906         skb_orphan(skb);
 907
 908         nf_reset(skb);
 909
 910         if (skb_array_produce(&tfile->tx_array, skb))
 911                 goto drop;
 912
 913         /* Notify and wake up reader process */
 914         if (tfile->flags & TUN_FASYNC)
 915                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 916         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 917
 918         rcu_read_unlock();
 919         return NETDEV_TX_OK;
 920
 921 drop:
 922         this_cpu_inc(tun->pcpu_stats->tx_dropped);
 923         skb_tx_error(skb);
 924         kfree_skb(skb);
 925         rcu_read_unlock();
 926         return NET_XMIT_DROP;
 927 }
 928
 929 static void tun_net_mclist(struct net_device *dev)
 930 {
 931         /*
 932          * This callback is supposed to deal with mc filter in
 933          * _rx_ path and has nothing to do with the _tx_ path.
 934          * In rx path we always accept everything userspace gives us.
 935          */
 936 }
 937
 938 #define MIN_MTU 68
 939 #define MAX_MTU 65535
 940
 941 static int
 942 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 943 {
 944         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 945                 return -EINVAL;
 946         dev->mtu = new_mtu;
 947         return 0;
 948 }
 949
 950 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 951         netdev_features_t features)
 952 {
 953         struct tun_struct *tun = netdev_priv(dev);
 954
 955         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 956 }
 957 #ifdef CONFIG_NET_POLL_CONTROLLER
 958 static void tun_poll_controller(struct net_device *dev)
 959 {
 960         /*
 961          * Tun only receives frames when:
 962          * 1) the char device endpoint gets data from user space
 963          * 2) the tun socket gets a sendmsg call from user space
 964          * Since both of those are synchronous operations, we are guaranteed
 965          * never to have pending data when we poll for it
 966          * so there is nothing to do here but return.
 967          * We need this though so netpoll recognizes us as an interface that
 968          * supports polling, which enables bridge devices in virt setups to
 969          * still use netconsole
 970          */
 971         return;
 972 }
 973 #endif
 974
 975 static void tun_set_headroom(struct net_device *dev, int new_hr)
 976 {
 977         struct tun_struct *tun = netdev_priv(dev);
 978
 979         if (new_hr < NET_SKB_PAD)
 980                 new_hr = NET_SKB_PAD;
 981
 982         tun->align = new_hr;
 983 }
 984
 985 static struct rtnl_link_stats64 *
 986 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
 987 {
 988         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
 989         struct tun_struct *tun = netdev_priv(dev);
 990         struct tun_pcpu_stats *p;
 991         int i;
 992
 993         for_each_possible_cpu(i) {
 994                 u64 rxpackets, rxbytes, txpackets, txbytes;
 995                 unsigned int start;
 996
 997                 p = per_cpu_ptr(tun->pcpu_stats, i);
 998                 do {
 999                         start = u64_stats_fetch_begin(&p->syncp);
1000                         rxpackets       = p->rx_packets;
1001                         rxbytes         = p->rx_bytes;
1002                         txpackets       = p->tx_packets;
1003                         txbytes         = p->tx_bytes;
1004                 } while (u64_stats_fetch_retry(&p->syncp, start));
1005
1006                 stats->rx_packets       += rxpackets;
1007                 stats->rx_bytes         += rxbytes;
1008                 stats->tx_packets       += txpackets;
1009                 stats->tx_bytes         += txbytes;
1010
1011                 /* u32 counters */
1012                 rx_dropped      += p->rx_dropped;
1013                 rx_frame_errors += p->rx_frame_errors;
1014                 tx_dropped      += p->tx_dropped;
1015         }
1016         stats->rx_dropped  = rx_dropped;
1017         stats->rx_frame_errors = rx_frame_errors;
1018         stats->tx_dropped = tx_dropped;
1019         return stats;
1020 }
1021
1022 static const struct net_device_ops tun_netdev_ops = {
1023         .ndo_uninit             = tun_net_uninit,
1024         .ndo_open               = tun_net_open,
1025         .ndo_stop               = tun_net_close,
1026         .ndo_start_xmit         = tun_net_xmit,
1027         .ndo_change_mtu         = tun_net_change_mtu,
1028         .ndo_fix_features       = tun_net_fix_features,
1029         .ndo_select_queue       = tun_select_queue,
1030 #ifdef CONFIG_NET_POLL_CONTROLLER
1031         .ndo_poll_controller    = tun_poll_controller,
1032 #endif
1033         .ndo_set_rx_headroom    = tun_set_headroom,
1034         .ndo_get_stats64        = tun_net_get_stats64,
1035 };
1036
1037 static const struct net_device_ops tap_netdev_ops = {
1038         .ndo_uninit             = tun_net_uninit,
1039         .ndo_open               = tun_net_open,
1040         .ndo_stop               = tun_net_close,
1041         .ndo_start_xmit         = tun_net_xmit,
1042         .ndo_change_mtu         = tun_net_change_mtu,
1043         .ndo_fix_features       = tun_net_fix_features,
1044         .ndo_set_rx_mode        = tun_net_mclist,
1045         .ndo_set_mac_address    = eth_mac_addr,
1046         .ndo_validate_addr      = eth_validate_addr,
1047         .ndo_select_queue       = tun_select_queue,
1048 #ifdef CONFIG_NET_POLL_CONTROLLER
1049         .ndo_poll_controller    = tun_poll_controller,
1050 #endif
1051         .ndo_features_check     = passthru_features_check,
1052         .ndo_set_rx_headroom    = tun_set_headroom,
1053         .ndo_get_stats64        = tun_net_get_stats64,
1054 };
1055
1056 static void tun_flow_init(struct tun_struct *tun)
1057 {
1058         int i;
1059
1060         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1061                 INIT_HLIST_HEAD(&tun->flows[i]);
1062
1063         tun->ageing_time = TUN_FLOW_EXPIRE;
1064         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1065         mod_timer(&tun->flow_gc_timer,
1066                   round_jiffies_up(jiffies + tun->ageing_time));
1067 }
1068
1069 static void tun_flow_uninit(struct tun_struct *tun)
1070 {
1071         del_timer_sync(&tun->flow_gc_timer);
1072         tun_flow_flush(tun);
1073 }
1074
1075 /* Initialize net device. */
1076 static void tun_net_init(struct net_device *dev)
1077 {
1078         struct tun_struct *tun = netdev_priv(dev);
1079
1080         switch (tun->flags & TUN_TYPE_MASK) {
1081         case IFF_TUN:
1082                 dev->netdev_ops = &tun_netdev_ops;
1083
1084                 /* Point-to-Point TUN Device */
1085                 dev->hard_header_len = 0;
1086                 dev->addr_len = 0;
1087                 dev->mtu = 1500;
1088
1089                 /* Zero header length */
1090                 dev->type = ARPHRD_NONE;
1091                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1092                 break;
1093
1094         case IFF_TAP:
1095                 dev->netdev_ops = &tap_netdev_ops;
1096                 /* Ethernet TAP Device */
1097                 ether_setup(dev);
1098                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1099                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1100
1101                 eth_hw_addr_random(dev);
1102
1103                 break;
1104         }
1105 }
1106
1107 /* Character device part */
1108
1109 /* Poll */
1110 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1111 {
1112         struct tun_file *tfile = file->private_data;
1113         struct tun_struct *tun = __tun_get(tfile);
1114         struct sock *sk;
1115         unsigned int mask = 0;
1116
1117         if (!tun)
1118                 return POLLERR;
1119
1120         sk = tfile->socket.sk;
1121
1122         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1123
1124         poll_wait(file, sk_sleep(sk), wait);
1125
1126         if (!skb_array_empty(&tfile->tx_array))
1127                 mask |= POLLIN | POLLRDNORM;
1128
1129         if (sock_writeable(sk) ||
1130             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1131              sock_writeable(sk)))
1132                 mask |= POLLOUT | POLLWRNORM;
1133
1134         if (tun->dev->reg_state != NETREG_REGISTERED)
1135                 mask = POLLERR;
1136
1137         tun_put(tun);
1138         return mask;
1139 }
1140
1141 /* prepad is the amount to reserve at front.  len is length after that.
1142  * linear is a hint as to how much to copy (usually headers). */
1143 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1144                                      size_t prepad, size_t len,
1145                                      size_t linear, int noblock)
1146 {
1147         struct sock *sk = tfile->socket.sk;
1148         struct sk_buff *skb;
1149         int err;
1150
1151         /* Under a page?  Don't bother with paged skb. */
1152         if (prepad + len < PAGE_SIZE || !linear)
1153                 linear = len;
1154
1155         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1156                                    &err, 0);
1157         if (!skb)
1158                 return ERR_PTR(err);
1159
1160         skb_reserve(skb, prepad);
1161         skb_put(skb, linear);
1162         skb->data_len = len - linear;
1163         skb->len += len - linear;
1164
1165         return skb;
1166 }
1167
1168 /* Get packet from user space buffer */
1169 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1170                             void *msg_control, struct iov_iter *from,
1171                             int noblock)
1172 {
1173         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1174         struct sk_buff *skb;
1175         size_t total_len = iov_iter_count(from);
1176         size_t len = total_len, align = tun->align, linear;
1177         struct virtio_net_hdr gso = { 0 };
1178         struct tun_pcpu_stats *stats;
1179         int good_linear;
1180         int copylen;
1181         bool zerocopy = false;
1182         int err;
1183         u32 rxhash;
1184         ssize_t n;
1185
1186         if (!(tun->dev->flags & IFF_UP))
1187                 return -EIO;
1188
1189         if (!(tun->flags & IFF_NO_PI)) {
1190                 if (len < sizeof(pi))
1191                         return -EINVAL;
1192                 len -= sizeof(pi);
1193
1194                 n = copy_from_iter(&pi, sizeof(pi), from);
1195                 if (n != sizeof(pi))
1196                         return -EFAULT;
1197         }
1198
1199         if (tun->flags & IFF_VNET_HDR) {
1200                 if (len < tun->vnet_hdr_sz)
1201                         return -EINVAL;
1202                 len -= tun->vnet_hdr_sz;
1203
1204                 n = copy_from_iter(&gso, sizeof(gso), from);
1205                 if (n != sizeof(gso))
1206                         return -EFAULT;
1207
1208                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1209                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1210                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1211
1212                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1213                         return -EINVAL;
1214                 iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1215         }
1216
1217         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1218                 align += NET_IP_ALIGN;
1219                 if (unlikely(len < ETH_HLEN ||
1220                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1221                         return -EINVAL;
1222         }
1223
1224         good_linear = SKB_MAX_HEAD(align);
1225
1226         if (msg_control) {
1227                 struct iov_iter i = *from;
1228
1229                 /* There are 256 bytes to be copied in skb, so there is
1230                  * enough room for skb expand head in case it is used.
1231                  * The rest of the buffer is mapped from userspace.
1232                  */
1233                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1234                 if (copylen > good_linear)
1235                         copylen = good_linear;
1236                 linear = copylen;
1237                 iov_iter_advance(&i, copylen);
1238                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1239                         zerocopy = true;
1240         }
1241
1242         if (!zerocopy) {
1243                 copylen = len;
1244                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1245                         linear = good_linear;
1246                 else
1247                         linear = tun16_to_cpu(tun, gso.hdr_len);
1248         }
1249
1250         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1251         if (IS_ERR(skb)) {
1252                 if (PTR_ERR(skb) != -EAGAIN)
1253                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1254                 return PTR_ERR(skb);
1255         }
1256
1257         if (zerocopy)
1258                 err = zerocopy_sg_from_iter(skb, from);
1259         else {
1260                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1261                 if (!err && msg_control) {
1262                         struct ubuf_info *uarg = msg_control;
1263                         uarg->callback(uarg, false);
1264                 }
1265         }
1266
1267         if (err) {
1268                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1269                 kfree_skb(skb);
1270                 return -EFAULT;
1271         }
1272
1273         err = virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun));
1274         if (err) {
1275                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1276                 kfree_skb(skb);
1277                 return -EINVAL;
1278         }
1279
1280         switch (tun->flags & TUN_TYPE_MASK) {
1281         case IFF_TUN:
1282                 if (tun->flags & IFF_NO_PI) {
1283                         switch (skb->data[0] & 0xf0) {
1284                         case 0x40:
1285                                 pi.proto = htons(ETH_P_IP);
1286                                 break;
1287                         case 0x60:
1288                                 pi.proto = htons(ETH_P_IPV6);
1289                                 break;
1290                         default:
1291                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1292                                 kfree_skb(skb);
1293                                 return -EINVAL;
1294                         }
1295                 }
1296
1297                 skb_reset_mac_header(skb);
1298                 skb->protocol = pi.proto;
1299                 skb->dev = tun->dev;
1300                 break;
1301         case IFF_TAP:
1302                 skb->protocol = eth_type_trans(skb, tun->dev);
1303                 break;
1304         }
1305
1306         /* copy skb_ubuf_info for callback when skb has no error */
1307         if (zerocopy) {
1308                 skb_shinfo(skb)->destructor_arg = msg_control;
1309                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1310                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1311         }
1312
1313         skb_reset_network_header(skb);
1314         skb_probe_transport_header(skb, 0);
1315
1316         rxhash = skb_get_hash(skb);
1317         netif_rx_ni(skb);
1318
1319         stats = get_cpu_ptr(tun->pcpu_stats);
1320         u64_stats_update_begin(&stats->syncp);
1321         stats->rx_packets++;
1322         stats->rx_bytes += len;
1323         u64_stats_update_end(&stats->syncp);
1324         put_cpu_ptr(stats);
1325
1326         tun_flow_update(tun, rxhash, tfile);
1327         return total_len;
1328 }
1329
1330 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1331 {
1332         struct file *file = iocb->ki_filp;
1333         struct tun_struct *tun = tun_get(file);
1334         struct tun_file *tfile = file->private_data;
1335         ssize_t result;
1336
1337         if (!tun)
1338                 return -EBADFD;
1339
1340         result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1341
1342         tun_put(tun);
1343         return result;
1344 }
1345
1346 /* Put packet to the user space buffer */
1347 static ssize_t tun_put_user(struct tun_struct *tun,
1348                             struct tun_file *tfile,
1349                             struct sk_buff *skb,
1350                             struct iov_iter *iter)
1351 {
1352         struct tun_pi pi = { 0, skb->protocol };
1353         struct tun_pcpu_stats *stats;
1354         ssize_t total;
1355         int vlan_offset = 0;
1356         int vlan_hlen = 0;
1357         int vnet_hdr_sz = 0;
1358
1359         if (skb_vlan_tag_present(skb))
1360                 vlan_hlen = VLAN_HLEN;
1361
1362         if (tun->flags & IFF_VNET_HDR)
1363                 vnet_hdr_sz = tun->vnet_hdr_sz;
1364
1365         total = skb->len + vlan_hlen + vnet_hdr_sz;
1366
1367         if (!(tun->flags & IFF_NO_PI)) {
1368                 if (iov_iter_count(iter) < sizeof(pi))
1369                         return -EINVAL;
1370
1371                 total += sizeof(pi);
1372                 if (iov_iter_count(iter) < total) {
1373                         /* Packet will be striped */
1374                         pi.flags |= TUN_PKT_STRIP;
1375                 }
1376
1377                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1378                         return -EFAULT;
1379         }
1380
1381         if (vnet_hdr_sz) {
1382                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1383                 int ret;
1384
1385                 if (iov_iter_count(iter) < vnet_hdr_sz)
1386                         return -EINVAL;
1387
1388                 ret = virtio_net_hdr_from_skb(skb, &gso,
1389                                               tun_is_little_endian(tun));
1390                 if (ret) {
1391                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1392                         pr_err("unexpected GSO type: "
1393                                "0x%x, gso_size %d, hdr_len %d\n",
1394                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1395                                tun16_to_cpu(tun, gso.hdr_len));
1396                         print_hex_dump(KERN_ERR, "tun: ",
1397                                        DUMP_PREFIX_NONE,
1398                                        16, 1, skb->head,
1399                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1400                         WARN_ON_ONCE(1);
1401                         return -EINVAL;
1402                 }
1403
1404                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1405                         return -EFAULT;
1406
1407                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1408         }
1409
1410         if (vlan_hlen) {
1411                 int ret;
1412                 struct {
1413                         __be16 h_vlan_proto;
1414                         __be16 h_vlan_TCI;
1415                 } veth;
1416
1417                 veth.h_vlan_proto = skb->vlan_proto;
1418                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1419
1420                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1421
1422                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1423                 if (ret || !iov_iter_count(iter))
1424                         goto done;
1425
1426                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1427                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1428                         goto done;
1429         }
1430
1431         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1432
1433 done:
1434         /* caller is in process context, */
1435         stats = get_cpu_ptr(tun->pcpu_stats);
1436         u64_stats_update_begin(&stats->syncp);
1437         stats->tx_packets++;
1438         stats->tx_bytes += skb->len + vlan_hlen;
1439         u64_stats_update_end(&stats->syncp);
1440         put_cpu_ptr(tun->pcpu_stats);
1441
1442         return total;
1443 }
1444
1445 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1446                                      int *err)
1447 {
1448         DECLARE_WAITQUEUE(wait, current);
1449         struct sk_buff *skb = NULL;
1450
1451         skb = skb_array_consume(&tfile->tx_array);
1452         if (skb)
1453                 goto out;
1454         if (noblock) {
1455                 *err = -EAGAIN;
1456                 goto out;
1457         }
1458
1459         add_wait_queue(&tfile->wq.wait, &wait);
1460         current->state = TASK_INTERRUPTIBLE;
1461
1462         while (1) {
1463                 skb = skb_array_consume(&tfile->tx_array);
1464                 if (skb)
1465                         break;
1466                 if (signal_pending(current)) {
1467                         *err = -ERESTARTSYS;
1468                         break;
1469                 }
1470                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1471                         *err = -EFAULT;
1472                         break;
1473                 }
1474
1475                 schedule();
1476         }
1477
1478         current->state = TASK_RUNNING;
1479         remove_wait_queue(&tfile->wq.wait, &wait);
1480
1481 out:
1482         return skb;
1483 }
1484
1485 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1486                            struct iov_iter *to,
1487                            int noblock)
1488 {
1489         struct sk_buff *skb;
1490         ssize_t ret;
1491         int err;
1492
1493         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1494
1495         if (!iov_iter_count(to))
1496                 return 0;
1497
1498         /* Read frames from ring */
1499         skb = tun_ring_recv(tfile, noblock, &err);
1500         if (!skb)
1501                 return err;
1502
1503         ret = tun_put_user(tun, tfile, skb, to);
1504         if (unlikely(ret < 0))
1505                 kfree_skb(skb);
1506         else
1507                 consume_skb(skb);
1508
1509         return ret;
1510 }
1511
1512 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1513 {
1514         struct file *file = iocb->ki_filp;
1515         struct tun_file *tfile = file->private_data;
1516         struct tun_struct *tun = __tun_get(tfile);
1517         ssize_t len = iov_iter_count(to), ret;
1518
1519         if (!tun)
1520                 return -EBADFD;
1521         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1522         ret = min_t(ssize_t, ret, len);
1523         if (ret > 0)
1524                 iocb->ki_pos = ret;
1525         tun_put(tun);
1526         return ret;
1527 }
1528
1529 static void tun_free_netdev(struct net_device *dev)
1530 {
1531         struct tun_struct *tun = netdev_priv(dev);
1532
1533         BUG_ON(!(list_empty(&tun->disabled)));
1534         free_percpu(tun->pcpu_stats);
1535         tun_flow_uninit(tun);
1536         security_tun_dev_free_security(tun->security);
1537         free_netdev(dev);
1538 }
1539
1540 static void tun_setup(struct net_device *dev)
1541 {
1542         struct tun_struct *tun = netdev_priv(dev);
1543
1544         tun->owner = INVALID_UID;
1545         tun->group = INVALID_GID;
1546
1547         dev->ethtool_ops = &tun_ethtool_ops;
1548         dev->destructor = tun_free_netdev;
1549         /* We prefer our own queue length */
1550         dev->tx_queue_len = TUN_READQ_SIZE;
1551 }
1552
1553 /* Trivial set of netlink ops to allow deleting tun or tap
1554  * device with netlink.
1555  */
1556 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1557 {
1558         return -EINVAL;
1559 }
1560
1561 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1562         .kind           = DRV_NAME,
1563         .priv_size      = sizeof(struct tun_struct),
1564         .setup          = tun_setup,
1565         .validate       = tun_validate,
1566 };
1567
1568 static void tun_sock_write_space(struct sock *sk)
1569 {
1570         struct tun_file *tfile;
1571         wait_queue_head_t *wqueue;
1572
1573         if (!sock_writeable(sk))
1574                 return;
1575
1576         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1577                 return;
1578
1579         wqueue = sk_sleep(sk);
1580         if (wqueue && waitqueue_active(wqueue))
1581                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1582                                                 POLLWRNORM | POLLWRBAND);
1583
1584         tfile = container_of(sk, struct tun_file, sk);
1585         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1586 }
1587
1588 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1589 {
1590         int ret;
1591         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1592         struct tun_struct *tun = __tun_get(tfile);
1593
1594         if (!tun)
1595                 return -EBADFD;
1596
1597         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1598                            m->msg_flags & MSG_DONTWAIT);
1599         tun_put(tun);
1600         return ret;
1601 }
1602
1603 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1604                        int flags)
1605 {
1606         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1607         struct tun_struct *tun = __tun_get(tfile);
1608         int ret;
1609
1610         if (!tun)
1611                 return -EBADFD;
1612
1613         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1614                 ret = -EINVAL;
1615                 goto out;
1616         }
1617         if (flags & MSG_ERRQUEUE) {
1618                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1619                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1620                 goto out;
1621         }
1622         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1623         if (ret > (ssize_t)total_len) {
1624                 m->msg_flags |= MSG_TRUNC;
1625                 ret = flags & MSG_TRUNC ? ret : total_len;
1626         }
1627 out:
1628         tun_put(tun);
1629         return ret;
1630 }
1631
1632 static int tun_peek_len(struct socket *sock)
1633 {
1634         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1635         struct tun_struct *tun;
1636         int ret = 0;
1637
1638         tun = __tun_get(tfile);
1639         if (!tun)
1640                 return 0;
1641
1642         ret = skb_array_peek_len(&tfile->tx_array);
1643         tun_put(tun);
1644
1645         return ret;
1646 }
1647
1648 /* Ops structure to mimic raw sockets with tun */
1649 static const struct proto_ops tun_socket_ops = {
1650         .peek_len = tun_peek_len,
1651         .sendmsg = tun_sendmsg,
1652         .recvmsg = tun_recvmsg,
1653 };
1654
1655 static struct proto tun_proto = {
1656         .name           = "tun",
1657         .owner          = THIS_MODULE,
1658         .obj_size       = sizeof(struct tun_file),
1659 };
1660
1661 static int tun_flags(struct tun_struct *tun)
1662 {
1663         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1664 }
1665
1666 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1667                               char *buf)
1668 {
1669         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1670         return sprintf(buf, "0x%x\n", tun_flags(tun));
1671 }
1672
1673 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1674                               char *buf)
1675 {
1676         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1677         return uid_valid(tun->owner)?
1678                 sprintf(buf, "%u\n",
1679                         from_kuid_munged(current_user_ns(), tun->owner)):
1680                 sprintf(buf, "-1\n");
1681 }
1682
1683 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1684                               char *buf)
1685 {
1686         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1687         return gid_valid(tun->group) ?
1688                 sprintf(buf, "%u\n",
1689                         from_kgid_munged(current_user_ns(), tun->group)):
1690                 sprintf(buf, "-1\n");
1691 }
1692
1693 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1694 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1695 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1696
1697 static struct attribute *tun_dev_attrs[] = {
1698         &dev_attr_tun_flags.attr,
1699         &dev_attr_owner.attr,
1700         &dev_attr_group.attr,
1701         NULL
1702 };
1703
1704 static const struct attribute_group tun_attr_group = {
1705         .attrs = tun_dev_attrs
1706 };
1707
1708 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1709 {
1710         struct tun_struct *tun;
1711         struct tun_file *tfile = file->private_data;
1712         struct net_device *dev;
1713         int err;
1714
1715         if (tfile->detached)
1716                 return -EINVAL;
1717
1718         dev = __dev_get_by_name(net, ifr->ifr_name);
1719         if (dev) {
1720                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1721                         return -EBUSY;
1722                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1723                         tun = netdev_priv(dev);
1724                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1725                         tun = netdev_priv(dev);
1726                 else
1727                         return -EINVAL;
1728
1729                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1730                     !!(tun->flags & IFF_MULTI_QUEUE))
1731                         return -EINVAL;
1732
1733                 if (tun_not_capable(tun))
1734                         return -EPERM;
1735                 err = security_tun_dev_open(tun->security);
1736                 if (err < 0)
1737                         return err;
1738
1739                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1740                 if (err < 0)
1741                         return err;
1742
1743                 if (tun->flags & IFF_MULTI_QUEUE &&
1744                     (tun->numqueues + tun->numdisabled > 1)) {
1745                         /* One or more queue has already been attached, no need
1746                          * to initialize the device again.
1747                          */
1748                         return 0;
1749                 }
1750         }
1751         else {
1752                 char *name;
1753                 unsigned long flags = 0;
1754                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1755                              MAX_TAP_QUEUES : 1;
1756
1757                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1758                         return -EPERM;
1759                 err = security_tun_dev_create();
1760                 if (err < 0)
1761                         return err;
1762
1763                 /* Set dev type */
1764                 if (ifr->ifr_flags & IFF_TUN) {
1765                         /* TUN device */
1766                         flags |= IFF_TUN;
1767                         name = "tun%d";
1768                 } else if (ifr->ifr_flags & IFF_TAP) {
1769                         /* TAP device */
1770                         flags |= IFF_TAP;
1771                         name = "tap%d";
1772                 } else
1773                         return -EINVAL;
1774
1775                 if (*ifr->ifr_name)
1776                         name = ifr->ifr_name;
1777
1778                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1779                                        NET_NAME_UNKNOWN, tun_setup, queues,
1780                                        queues);
1781
1782                 if (!dev)
1783                         return -ENOMEM;
1784
1785                 dev_net_set(dev, net);
1786                 dev->rtnl_link_ops = &tun_link_ops;
1787                 dev->ifindex = tfile->ifindex;
1788                 dev->sysfs_groups[0] = &tun_attr_group;
1789
1790                 tun = netdev_priv(dev);
1791                 tun->dev = dev;
1792                 tun->flags = flags;
1793                 tun->txflt.count = 0;
1794                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1795
1796                 tun->align = NET_SKB_PAD;
1797                 tun->filter_attached = false;
1798                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1799
1800                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
1801                 if (!tun->pcpu_stats) {
1802                         err = -ENOMEM;
1803                         goto err_free_dev;
1804                 }
1805
1806                 spin_lock_init(&tun->lock);
1807
1808                 err = security_tun_dev_alloc_security(&tun->security);
1809                 if (err < 0)
1810                         goto err_free_stat;
1811
1812                 tun_net_init(dev);
1813                 tun_flow_init(tun);
1814
1815                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1816                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1817                                    NETIF_F_HW_VLAN_STAG_TX;
1818                 dev->features = dev->hw_features | NETIF_F_LLTX;
1819                 dev->vlan_features = dev->features &
1820                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1821                                        NETIF_F_HW_VLAN_STAG_TX);
1822
1823                 INIT_LIST_HEAD(&tun->disabled);
1824                 err = tun_attach(tun, file, false);
1825                 if (err < 0)
1826                         goto err_free_flow;
1827
1828                 err = register_netdevice(tun->dev);
1829                 if (err < 0)
1830                         goto err_detach;
1831         }
1832
1833         netif_carrier_on(tun->dev);
1834
1835         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1836
1837         tun->flags = (tun->flags & ~TUN_FEATURES) |
1838                 (ifr->ifr_flags & TUN_FEATURES);
1839
1840         /* Make sure persistent devices do not get stuck in
1841          * xoff state.
1842          */
1843         if (netif_running(tun->dev))
1844                 netif_tx_wake_all_queues(tun->dev);
1845
1846         strcpy(ifr->ifr_name, tun->dev->name);
1847         return 0;
1848
1849 err_detach:
1850         tun_detach_all(dev);
1851 err_free_flow:
1852         tun_flow_uninit(tun);
1853         security_tun_dev_free_security(tun->security);
1854 err_free_stat:
1855         free_percpu(tun->pcpu_stats);
1856 err_free_dev:
1857         free_netdev(dev);
1858         return err;
1859 }
1860
1861 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1862                        struct ifreq *ifr)
1863 {
1864         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1865
1866         strcpy(ifr->ifr_name, tun->dev->name);
1867
1868         ifr->ifr_flags = tun_flags(tun);
1869
1870 }
1871
1872 /* This is like a cut-down ethtool ops, except done via tun fd so no
1873  * privs required. */
1874 static int set_offload(struct tun_struct *tun, unsigned long arg)
1875 {
1876         netdev_features_t features = 0;
1877
1878         if (arg & TUN_F_CSUM) {
1879                 features |= NETIF_F_HW_CSUM;
1880                 arg &= ~TUN_F_CSUM;
1881
1882                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1883                         if (arg & TUN_F_TSO_ECN) {
1884                                 features |= NETIF_F_TSO_ECN;
1885                                 arg &= ~TUN_F_TSO_ECN;
1886                         }
1887                         if (arg & TUN_F_TSO4)
1888                                 features |= NETIF_F_TSO;
1889                         if (arg & TUN_F_TSO6)
1890                                 features |= NETIF_F_TSO6;
1891                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1892                 }
1893
1894                 if (arg & TUN_F_UFO) {
1895                         features |= NETIF_F_UFO;
1896                         arg &= ~TUN_F_UFO;
1897                 }
1898         }
1899
1900         /* This gives the user a way to test for new features in future by
1901          * trying to set them. */
1902         if (arg)
1903                 return -EINVAL;
1904
1905         tun->set_features = features;
1906         netdev_update_features(tun->dev);
1907
1908         return 0;
1909 }
1910
1911 static void tun_detach_filter(struct tun_struct *tun, int n)
1912 {
1913         int i;
1914         struct tun_file *tfile;
1915
1916         for (i = 0; i < n; i++) {
1917                 tfile = rtnl_dereference(tun->tfiles[i]);
1918                 lock_sock(tfile->socket.sk);
1919                 sk_detach_filter(tfile->socket.sk);
1920                 release_sock(tfile->socket.sk);
1921         }
1922
1923         tun->filter_attached = false;
1924 }
1925
1926 static int tun_attach_filter(struct tun_struct *tun)
1927 {
1928         int i, ret = 0;
1929         struct tun_file *tfile;
1930
1931         for (i = 0; i < tun->numqueues; i++) {
1932                 tfile = rtnl_dereference(tun->tfiles[i]);
1933                 lock_sock(tfile->socket.sk);
1934                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1935                 release_sock(tfile->socket.sk);
1936                 if (ret) {
1937                         tun_detach_filter(tun, i);
1938                         return ret;
1939                 }
1940         }
1941
1942         tun->filter_attached = true;
1943         return ret;
1944 }
1945
1946 static void tun_set_sndbuf(struct tun_struct *tun)
1947 {
1948         struct tun_file *tfile;
1949         int i;
1950
1951         for (i = 0; i < tun->numqueues; i++) {
1952                 tfile = rtnl_dereference(tun->tfiles[i]);
1953                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1954         }
1955 }
1956
1957 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1958 {
1959         struct tun_file *tfile = file->private_data;
1960         struct tun_struct *tun;
1961         int ret = 0;
1962
1963         rtnl_lock();
1964
1965         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1966                 tun = tfile->detached;
1967                 if (!tun) {
1968                         ret = -EINVAL;
1969                         goto unlock;
1970                 }
1971                 ret = security_tun_dev_attach_queue(tun->security);
1972                 if (ret < 0)
1973                         goto unlock;
1974                 ret = tun_attach(tun, file, false);
1975         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1976                 tun = rtnl_dereference(tfile->tun);
1977                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1978                         ret = -EINVAL;
1979                 else
1980                         __tun_detach(tfile, false);
1981         } else
1982                 ret = -EINVAL;
1983
1984 unlock:
1985         rtnl_unlock();
1986         return ret;
1987 }
1988
1989 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1990                             unsigned long arg, int ifreq_len)
1991 {
1992         struct tun_file *tfile = file->private_data;
1993         struct tun_struct *tun;
1994         void __user* argp = (void __user*)arg;
1995         struct ifreq ifr;
1996         kuid_t owner;
1997         kgid_t group;
1998         int sndbuf;
1999         int vnet_hdr_sz;
2000         unsigned int ifindex;
2001         int le;
2002         int ret;
2003
2004         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
2005                 if (copy_from_user(&ifr, argp, ifreq_len))
2006                         return -EFAULT;
2007         } else {
2008                 memset(&ifr, 0, sizeof(ifr));
2009         }
2010         if (cmd == TUNGETFEATURES) {
2011                 /* Currently this just means: "what IFF flags are valid?".
2012                  * This is needed because we never checked for invalid flags on
2013                  * TUNSETIFF.
2014                  */
2015                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2016                                 (unsigned int __user*)argp);
2017         } else if (cmd == TUNSETQUEUE)
2018                 return tun_set_queue(file, &ifr);
2019
2020         ret = 0;
2021         rtnl_lock();
2022
2023         tun = __tun_get(tfile);
2024         if (cmd == TUNSETIFF && !tun) {
2025                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2026
2027                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2028
2029                 if (ret)
2030                         goto unlock;
2031
2032                 if (copy_to_user(argp, &ifr, ifreq_len))
2033                         ret = -EFAULT;
2034                 goto unlock;
2035         }
2036         if (cmd == TUNSETIFINDEX) {
2037                 ret = -EPERM;
2038                 if (tun)
2039                         goto unlock;
2040
2041                 ret = -EFAULT;
2042                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2043                         goto unlock;
2044
2045                 ret = 0;
2046                 tfile->ifindex = ifindex;
2047                 goto unlock;
2048         }
2049
2050         ret = -EBADFD;
2051         if (!tun)
2052                 goto unlock;
2053
2054         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2055
2056         ret = 0;
2057         switch (cmd) {
2058         case TUNGETIFF:
2059                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2060
2061                 if (tfile->detached)
2062                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2063                 if (!tfile->socket.sk->sk_filter)
2064                         ifr.ifr_flags |= IFF_NOFILTER;
2065
2066                 if (copy_to_user(argp, &ifr, ifreq_len))
2067                         ret = -EFAULT;
2068                 break;
2069
2070         case TUNSETNOCSUM:
2071                 /* Disable/Enable checksum */
2072
2073                 /* [unimplemented] */
2074                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2075                           arg ? "disabled" : "enabled");
2076                 break;
2077
2078         case TUNSETPERSIST:
2079                 /* Disable/Enable persist mode. Keep an extra reference to the
2080                  * module to prevent the module being unprobed.
2081                  */
2082                 if (arg && !(tun->flags & IFF_PERSIST)) {
2083                         tun->flags |= IFF_PERSIST;
2084                         __module_get(THIS_MODULE);
2085                 }
2086                 if (!arg && (tun->flags & IFF_PERSIST)) {
2087                         tun->flags &= ~IFF_PERSIST;
2088                         module_put(THIS_MODULE);
2089                 }
2090
2091                 tun_debug(KERN_INFO, tun, "persist %s\n",
2092                           arg ? "enabled" : "disabled");
2093                 break;
2094
2095         case TUNSETOWNER:
2096                 /* Set owner of the device */
2097                 owner = make_kuid(current_user_ns(), arg);
2098                 if (!uid_valid(owner)) {
2099                         ret = -EINVAL;
2100                         break;
2101                 }
2102                 tun->owner = owner;
2103                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2104                           from_kuid(&init_user_ns, tun->owner));
2105                 break;
2106
2107         case TUNSETGROUP:
2108                 /* Set group of the device */
2109                 group = make_kgid(current_user_ns(), arg);
2110                 if (!gid_valid(group)) {
2111                         ret = -EINVAL;
2112                         break;
2113                 }
2114                 tun->group = group;
2115                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2116                           from_kgid(&init_user_ns, tun->group));
2117                 break;
2118
2119         case TUNSETLINK:
2120                 /* Only allow setting the type when the interface is down */
2121                 if (tun->dev->flags & IFF_UP) {
2122                         tun_debug(KERN_INFO, tun,
2123                                   "Linktype set failed because interface is up\n");
2124                         ret = -EBUSY;
2125                 } else {
2126                         tun->dev->type = (int) arg;
2127                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2128                                   tun->dev->type);
2129                         ret = 0;
2130                 }
2131                 break;
2132
2133 #ifdef TUN_DEBUG
2134         case TUNSETDEBUG:
2135                 tun->debug = arg;
2136                 break;
2137 #endif
2138         case TUNSETOFFLOAD:
2139                 ret = set_offload(tun, arg);
2140                 break;
2141
2142         case TUNSETTXFILTER:
2143                 /* Can be set only for TAPs */
2144                 ret = -EINVAL;
2145                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2146                         break;
2147                 ret = update_filter(&tun->txflt, (void __user *)arg);
2148                 break;
2149
2150         case SIOCGIFHWADDR:
2151                 /* Get hw address */
2152                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2153                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2154                 if (copy_to_user(argp, &ifr, ifreq_len))
2155                         ret = -EFAULT;
2156                 break;
2157
2158         case SIOCSIFHWADDR:
2159                 /* Set hw address */
2160                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2161                           ifr.ifr_hwaddr.sa_data);
2162
2163                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2164                 break;
2165
2166         case TUNGETSNDBUF:
2167                 sndbuf = tfile->socket.sk->sk_sndbuf;
2168                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2169                         ret = -EFAULT;
2170                 break;
2171
2172         case TUNSETSNDBUF:
2173                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2174                         ret = -EFAULT;
2175                         break;
2176                 }
2177
2178                 tun->sndbuf = sndbuf;
2179                 tun_set_sndbuf(tun);
2180                 break;
2181
2182         case TUNGETVNETHDRSZ:
2183                 vnet_hdr_sz = tun->vnet_hdr_sz;
2184                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2185                         ret = -EFAULT;
2186                 break;
2187
2188         case TUNSETVNETHDRSZ:
2189                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2190                         ret = -EFAULT;
2191                         break;
2192                 }
2193                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2194                         ret = -EINVAL;
2195                         break;
2196                 }
2197
2198                 tun->vnet_hdr_sz = vnet_hdr_sz;
2199                 break;
2200
2201         case TUNGETVNETLE:
2202                 le = !!(tun->flags & TUN_VNET_LE);
2203                 if (put_user(le, (int __user *)argp))
2204                         ret = -EFAULT;
2205                 break;
2206
2207         case TUNSETVNETLE:
2208                 if (get_user(le, (int __user *)argp)) {
2209                         ret = -EFAULT;
2210                         break;
2211                 }
2212                 if (le)
2213                         tun->flags |= TUN_VNET_LE;
2214                 else
2215                         tun->flags &= ~TUN_VNET_LE;
2216                 break;
2217
2218         case TUNGETVNETBE:
2219                 ret = tun_get_vnet_be(tun, argp);
2220                 break;
2221
2222         case TUNSETVNETBE:
2223                 ret = tun_set_vnet_be(tun, argp);
2224                 break;
2225
2226         case TUNATTACHFILTER:
2227                 /* Can be set only for TAPs */
2228                 ret = -EINVAL;
2229                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2230                         break;
2231                 ret = -EFAULT;
2232                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2233                         break;
2234
2235                 ret = tun_attach_filter(tun);
2236                 break;
2237
2238         case TUNDETACHFILTER:
2239                 /* Can be set only for TAPs */
2240                 ret = -EINVAL;
2241                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2242                         break;
2243                 ret = 0;
2244                 tun_detach_filter(tun, tun->numqueues);
2245                 break;
2246
2247         case TUNGETFILTER:
2248                 ret = -EINVAL;
2249                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2250                         break;
2251                 ret = -EFAULT;
2252                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2253                         break;
2254                 ret = 0;
2255                 break;
2256
2257         default:
2258                 ret = -EINVAL;
2259                 break;
2260         }
2261
2262 unlock:
2263         rtnl_unlock();
2264         if (tun)
2265                 tun_put(tun);
2266         return ret;
2267 }
2268
2269 static long tun_chr_ioctl(struct file *file,
2270                           unsigned int cmd, unsigned long arg)
2271 {
2272         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2273 }
2274
2275 #ifdef CONFIG_COMPAT
2276 static long tun_chr_compat_ioctl(struct file *file,
2277                          unsigned int cmd, unsigned long arg)
2278 {
2279         switch (cmd) {
2280         case TUNSETIFF:
2281         case TUNGETIFF:
2282         case TUNSETTXFILTER:
2283         case TUNGETSNDBUF:
2284         case TUNSETSNDBUF:
2285         case SIOCGIFHWADDR:
2286         case SIOCSIFHWADDR:
2287                 arg = (unsigned long)compat_ptr(arg);
2288                 break;
2289         default:
2290                 arg = (compat_ulong_t)arg;
2291                 break;
2292         }
2293
2294         /*
2295          * compat_ifreq is shorter than ifreq, so we must not access beyond
2296          * the end of that structure. All fields that are used in this
2297          * driver are compatible though, we don't need to convert the
2298          * contents.
2299          */
2300         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2301 }
2302 #endif /* CONFIG_COMPAT */
2303
2304 static int tun_chr_fasync(int fd, struct file *file, int on)
2305 {
2306         struct tun_file *tfile = file->private_data;
2307         int ret;
2308
2309         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2310                 goto out;
2311
2312         if (on) {
2313                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2314                 tfile->flags |= TUN_FASYNC;
2315         } else
2316                 tfile->flags &= ~TUN_FASYNC;
2317         ret = 0;
2318 out:
2319         return ret;
2320 }
2321
2322 static int tun_chr_open(struct inode *inode, struct file * file)
2323 {
2324         struct net *net = current->nsproxy->net_ns;
2325         struct tun_file *tfile;
2326
2327         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2328
2329         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2330                                             &tun_proto, 0);
2331         if (!tfile)
2332                 return -ENOMEM;
2333         RCU_INIT_POINTER(tfile->tun, NULL);
2334         tfile->flags = 0;
2335         tfile->ifindex = 0;
2336
2337         init_waitqueue_head(&tfile->wq.wait);
2338         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2339
2340         tfile->socket.file = file;
2341         tfile->socket.ops = &tun_socket_ops;
2342
2343         sock_init_data(&tfile->socket, &tfile->sk);
2344
2345         tfile->sk.sk_write_space = tun_sock_write_space;
2346         tfile->sk.sk_sndbuf = INT_MAX;
2347
2348         file->private_data = tfile;
2349         INIT_LIST_HEAD(&tfile->next);
2350
2351         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2352
2353         return 0;
2354 }
2355
2356 static int tun_chr_close(struct inode *inode, struct file *file)
2357 {
2358         struct tun_file *tfile = file->private_data;
2359
2360         tun_detach(tfile, true);
2361
2362         return 0;
2363 }
2364
2365 #ifdef CONFIG_PROC_FS
2366 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2367 {
2368         struct tun_struct *tun;
2369         struct ifreq ifr;
2370
2371         memset(&ifr, 0, sizeof(ifr));
2372
2373         rtnl_lock();
2374         tun = tun_get(f);
2375         if (tun)
2376                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2377         rtnl_unlock();
2378
2379         if (tun)
2380                 tun_put(tun);
2381
2382         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2383 }
2384 #endif
2385
2386 static const struct file_operations tun_fops = {
2387         .owner  = THIS_MODULE,
2388         .llseek = no_llseek,
2389         .read_iter  = tun_chr_read_iter,
2390         .write_iter = tun_chr_write_iter,
2391         .poll   = tun_chr_poll,
2392         .unlocked_ioctl = tun_chr_ioctl,
2393 #ifdef CONFIG_COMPAT
2394         .compat_ioctl = tun_chr_compat_ioctl,
2395 #endif
2396         .open   = tun_chr_open,
2397         .release = tun_chr_close,
2398         .fasync = tun_chr_fasync,
2399 #ifdef CONFIG_PROC_FS
2400         .show_fdinfo = tun_chr_show_fdinfo,
2401 #endif
2402 };
2403
2404 static struct miscdevice tun_miscdev = {
2405         .minor = TUN_MINOR,
2406         .name = "tun",
2407         .nodename = "net/tun",
2408         .fops = &tun_fops,
2409 };
2410
2411 /* ethtool interface */
2412
2413 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2414 {
2415         cmd->supported          = 0;
2416         cmd->advertising        = 0;
2417         ethtool_cmd_speed_set(cmd, SPEED_10);
2418         cmd->duplex             = DUPLEX_FULL;
2419         cmd->port               = PORT_TP;
2420         cmd->phy_address        = 0;
2421         cmd->transceiver        = XCVR_INTERNAL;
2422         cmd->autoneg            = AUTONEG_DISABLE;
2423         cmd->maxtxpkt           = 0;
2424         cmd->maxrxpkt           = 0;
2425         return 0;
2426 }
2427
2428 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2429 {
2430         struct tun_struct *tun = netdev_priv(dev);
2431
2432         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2433         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2434
2435         switch (tun->flags & TUN_TYPE_MASK) {
2436         case IFF_TUN:
2437                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2438                 break;
2439         case IFF_TAP:
2440                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2441                 break;
2442         }
2443 }
2444
2445 static u32 tun_get_msglevel(struct net_device *dev)
2446 {
2447 #ifdef TUN_DEBUG
2448         struct tun_struct *tun = netdev_priv(dev);
2449         return tun->debug;
2450 #else
2451         return -EOPNOTSUPP;
2452 #endif
2453 }
2454
2455 static void tun_set_msglevel(struct net_device *dev, u32 value)
2456 {
2457 #ifdef TUN_DEBUG
2458         struct tun_struct *tun = netdev_priv(dev);
2459         tun->debug = value;
2460 #endif
2461 }
2462
2463 static const struct ethtool_ops tun_ethtool_ops = {
2464         .get_settings   = tun_get_settings,
2465         .get_drvinfo    = tun_get_drvinfo,
2466         .get_msglevel   = tun_get_msglevel,
2467         .set_msglevel   = tun_set_msglevel,
2468         .get_link       = ethtool_op_get_link,
2469         .get_ts_info    = ethtool_op_get_ts_info,
2470 };
2471
2472 static int tun_queue_resize(struct tun_struct *tun)
2473 {
2474         struct net_device *dev = tun->dev;
2475         struct tun_file *tfile;
2476         struct skb_array **arrays;
2477         int n = tun->numqueues + tun->numdisabled;
2478         int ret, i;
2479
2480         arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL);
2481         if (!arrays)
2482                 return -ENOMEM;
2483
2484         for (i = 0; i < tun->numqueues; i++) {
2485                 tfile = rtnl_dereference(tun->tfiles[i]);
2486                 arrays[i] = &tfile->tx_array;
2487         }
2488         list_for_each_entry(tfile, &tun->disabled, next)
2489                 arrays[i++] = &tfile->tx_array;
2490
2491         ret = skb_array_resize_multiple(arrays, n,
2492                                         dev->tx_queue_len, GFP_KERNEL);
2493
2494         kfree(arrays);
2495         return ret;
2496 }
2497
2498 static int tun_device_event(struct notifier_block *unused,
2499                             unsigned long event, void *ptr)
2500 {
2501         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2502         struct tun_struct *tun = netdev_priv(dev);
2503
2504         switch (event) {
2505         case NETDEV_CHANGE_TX_QUEUE_LEN:
2506                 if (tun_queue_resize(tun))
2507                         return NOTIFY_BAD;
2508                 break;
2509         default:
2510                 break;
2511         }
2512
2513         return NOTIFY_DONE;
2514 }
2515
2516 static struct notifier_block tun_notifier_block __read_mostly = {
2517         .notifier_call  = tun_device_event,
2518 };
2519
2520 static int __init tun_init(void)
2521 {
2522         int ret = 0;
2523
2524         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2525         pr_info("%s\n", DRV_COPYRIGHT);
2526
2527         ret = rtnl_link_register(&tun_link_ops);
2528         if (ret) {
2529                 pr_err("Can't register link_ops\n");
2530                 goto err_linkops;
2531         }
2532
2533         ret = misc_register(&tun_miscdev);
2534         if (ret) {
2535                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2536                 goto err_misc;
2537         }
2538
2539         register_netdevice_notifier(&tun_notifier_block);
2540         return  0;
2541 err_misc:
2542         rtnl_link_unregister(&tun_link_ops);
2543 err_linkops:
2544         return ret;
2545 }
2546
2547 static void tun_cleanup(void)
2548 {
2549         misc_deregister(&tun_miscdev);
2550         rtnl_link_unregister(&tun_link_ops);
2551         unregister_netdevice_notifier(&tun_notifier_block);
2552 }
2553
2554 /* Get an underlying socket object from tun file.  Returns error unless file is
2555  * attached to a device.  The returned object works like a packet socket, it
2556  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2557  * holding a reference to the file for as long as the socket is in use. */
2558 struct socket *tun_get_socket(struct file *file)
2559 {
2560         struct tun_file *tfile;
2561         if (file->f_op != &tun_fops)
2562                 return ERR_PTR(-EINVAL);
2563         tfile = file->private_data;
2564         if (!tfile)
2565                 return ERR_PTR(-EBADFD);
2566         return &tfile->socket;
2567 }
2568 EXPORT_SYMBOL_GPL(tun_get_socket);
2569
2570 module_init(tun_init);
2571 module_exit(tun_cleanup);
2572 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2573 MODULE_AUTHOR(DRV_COPYRIGHT);
2574 MODULE_LICENSE("GPL");
2575 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2576 MODULE_ALIAS("devname:net/tun");