drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/major.h>
  48 #include <linux/slab.h>
  49 #include <linux/poll.h>
  50 #include <linux/fcntl.h>
  51 #include <linux/init.h>
  52 #include <linux/skbuff.h>
  53 #include <linux/netdevice.h>
  54 #include <linux/etherdevice.h>
  55 #include <linux/miscdevice.h>
  56 #include <linux/ethtool.h>
  57 #include <linux/rtnetlink.h>
  58 #include <linux/compat.h>
  59 #include <linux/if.h>
  60 #include <linux/if_arp.h>
  61 #include <linux/if_ether.h>
  62 #include <linux/if_tun.h>
  63 #include <linux/if_vlan.h>
  64 #include <linux/crc32.h>
  65 #include <linux/nsproxy.h>
  66 #include <linux/virtio_net.h>
  67 #include <linux/rcupdate.h>
  68 #include <net/net_namespace.h>
  69 #include <net/netns/generic.h>
  70 #include <net/rtnetlink.h>
  71 #include <net/sock.h>
  72 #include <linux/seq_file.h>
  73 #include <linux/uio.h>
  74 #include <linux/skb_array.h>
  75
  76 #include <linux/uaccess.h>
  77
  78 /* Uncomment to enable debugging */
  79 /* #define TUN_DEBUG 1 */
  80
  81 #ifdef TUN_DEBUG
  82 static int debug;
  83
  84 #define tun_debug(level, tun, fmt, args...)                     \
  85 do {                                                            \
  86         if (tun->debug)                                         \
  87                 netdev_printk(level, tun->dev, fmt, ##args);    \
  88 } while (0)
  89 #define DBG1(level, fmt, args...)                               \
  90 do {                                                            \
  91         if (debug == 2)                                         \
  92                 printk(level fmt, ##args);                      \
  93 } while (0)
  94 #else
  95 #define tun_debug(level, tun, fmt, args...)                     \
  96 do {                                                            \
  97         if (0)                                                  \
  98                 netdev_printk(level, tun->dev, fmt, ##args);    \
  99 } while (0)
 100 #define DBG1(level, fmt, args...)                               \
 101 do {                                                            \
 102         if (0)                                                  \
 103                 printk(level fmt, ##args);                      \
 104 } while (0)
 105 #endif
 106
 107 /* TUN device flags */
 108
 109 /* IFF_ATTACH_QUEUE is never stored in device flags,
 110  * overload it to mean fasync when stored there.
 111  */
 112 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 113 /* High bits in flags field are unused. */
 114 #define TUN_VNET_LE     0x80000000
 115 #define TUN_VNET_BE     0x40000000
 116
 117 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 118                       IFF_MULTI_QUEUE)
 119 #define GOODCOPY_LEN 128
 120
 121 #define FLT_EXACT_COUNT 8
 122 struct tap_filter {
 123         unsigned int    count;    /* Number of addrs. Zero means disabled */
 124         u32             mask[2];  /* Mask of the hashed addrs */
 125         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 126 };
 127
 128 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 129  * to max number of VCPUs in guest. */
 130 #define MAX_TAP_QUEUES 256
 131 #define MAX_TAP_FLOWS  4096
 132
 133 #define TUN_FLOW_EXPIRE (3 * HZ)
 134
 135 struct tun_pcpu_stats {
 136         u64 rx_packets;
 137         u64 rx_bytes;
 138         u64 tx_packets;
 139         u64 tx_bytes;
 140         struct u64_stats_sync syncp;
 141         u32 rx_dropped;
 142         u32 tx_dropped;
 143         u32 rx_frame_errors;
 144 };
 145
 146 /* A tun_file connects an open character device to a tuntap netdevice. It
 147  * also contains all socket related structures (except sock_fprog and tap_filter)
 148  * to serve as one transmit queue for tuntap device. The sock_fprog and
 149  * tap_filter were kept in tun_struct since they were used for filtering for the
 150  * netdevice not for a specific queue (at least I didn't see the requirement for
 151  * this).
 152  *
 153  * RCU usage:
 154  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 155  * other can only be read while rcu_read_lock or rtnl_lock is held.
 156  */
 157 struct tun_file {
 158         struct sock sk;
 159         struct socket socket;
 160         struct socket_wq wq;
 161         struct tun_struct __rcu *tun;
 162         struct fasync_struct *fasync;
 163         /* only used for fasnyc */
 164         unsigned int flags;
 165         union {
 166                 u16 queue_index;
 167                 unsigned int ifindex;
 168         };
 169         struct list_head next;
 170         struct tun_struct *detached;
 171         struct skb_array tx_array;
 172 };
 173
 174 struct tun_flow_entry {
 175         struct hlist_node hash_link;
 176         struct rcu_head rcu;
 177         struct tun_struct *tun;
 178
 179         u32 rxhash;
 180         u32 rps_rxhash;
 181         int queue_index;
 182         unsigned long updated;
 183 };
 184
 185 #define TUN_NUM_FLOW_ENTRIES 1024
 186
 187 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 188  * device, socket filter, sndbuf and vnet header size were restore when the
 189  * file were attached to a persist device.
 190  */
 191 struct tun_struct {
 192         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 193         unsigned int            numqueues;
 194         unsigned int            flags;
 195         kuid_t                  owner;
 196         kgid_t                  group;
 197
 198         struct net_device       *dev;
 199         netdev_features_t       set_features;
 200 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 201                           NETIF_F_TSO6|NETIF_F_UFO)
 202
 203         int                     align;
 204         int                     vnet_hdr_sz;
 205         int                     sndbuf;
 206         struct tap_filter       txflt;
 207         struct sock_fprog       fprog;
 208         /* protected by rtnl lock */
 209         bool                    filter_attached;
 210 #ifdef TUN_DEBUG
 211         int debug;
 212 #endif
 213         spinlock_t lock;
 214         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 215         struct timer_list flow_gc_timer;
 216         unsigned long ageing_time;
 217         unsigned int numdisabled;
 218         struct list_head disabled;
 219         void *security;
 220         u32 flow_count;
 221         u32 rx_batched;
 222         struct tun_pcpu_stats __percpu *pcpu_stats;
 223 };
 224
 225 #ifdef CONFIG_TUN_VNET_CROSS_LE
 226 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 227 {
 228         return tun->flags & TUN_VNET_BE ? false :
 229                 virtio_legacy_is_little_endian();
 230 }
 231
 232 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 233 {
 234         int be = !!(tun->flags & TUN_VNET_BE);
 235
 236         if (put_user(be, argp))
 237                 return -EFAULT;
 238
 239         return 0;
 240 }
 241
 242 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 243 {
 244         int be;
 245
 246         if (get_user(be, argp))
 247                 return -EFAULT;
 248
 249         if (be)
 250                 tun->flags |= TUN_VNET_BE;
 251         else
 252                 tun->flags &= ~TUN_VNET_BE;
 253
 254         return 0;
 255 }
 256 #else
 257 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 258 {
 259         return virtio_legacy_is_little_endian();
 260 }
 261
 262 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 263 {
 264         return -EINVAL;
 265 }
 266
 267 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 268 {
 269         return -EINVAL;
 270 }
 271 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 272
 273 static inline bool tun_is_little_endian(struct tun_struct *tun)
 274 {
 275         return tun->flags & TUN_VNET_LE ||
 276                 tun_legacy_is_little_endian(tun);
 277 }
 278
 279 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 280 {
 281         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 282 }
 283
 284 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 285 {
 286         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 287 }
 288
 289 static inline u32 tun_hashfn(u32 rxhash)
 290 {
 291         return rxhash & 0x3ff;
 292 }
 293
 294 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 295 {
 296         struct tun_flow_entry *e;
 297
 298         hlist_for_each_entry_rcu(e, head, hash_link) {
 299                 if (e->rxhash == rxhash)
 300                         return e;
 301         }
 302         return NULL;
 303 }
 304
 305 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 306                                               struct hlist_head *head,
 307                                               u32 rxhash, u16 queue_index)
 308 {
 309         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 310
 311         if (e) {
 312                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 313                           rxhash, queue_index);
 314                 e->updated = jiffies;
 315                 e->rxhash = rxhash;
 316                 e->rps_rxhash = 0;
 317                 e->queue_index = queue_index;
 318                 e->tun = tun;
 319                 hlist_add_head_rcu(&e->hash_link, head);
 320                 ++tun->flow_count;
 321         }
 322         return e;
 323 }
 324
 325 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 326 {
 327         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 328                   e->rxhash, e->queue_index);
 329         hlist_del_rcu(&e->hash_link);
 330         kfree_rcu(e, rcu);
 331         --tun->flow_count;
 332 }
 333
 334 static void tun_flow_flush(struct tun_struct *tun)
 335 {
 336         int i;
 337
 338         spin_lock_bh(&tun->lock);
 339         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 340                 struct tun_flow_entry *e;
 341                 struct hlist_node *n;
 342
 343                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 344                         tun_flow_delete(tun, e);
 345         }
 346         spin_unlock_bh(&tun->lock);
 347 }
 348
 349 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 350 {
 351         int i;
 352
 353         spin_lock_bh(&tun->lock);
 354         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 355                 struct tun_flow_entry *e;
 356                 struct hlist_node *n;
 357
 358                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 359                         if (e->queue_index == queue_index)
 360                                 tun_flow_delete(tun, e);
 361                 }
 362         }
 363         spin_unlock_bh(&tun->lock);
 364 }
 365
 366 static void tun_flow_cleanup(unsigned long data)
 367 {
 368         struct tun_struct *tun = (struct tun_struct *)data;
 369         unsigned long delay = tun->ageing_time;
 370         unsigned long next_timer = jiffies + delay;
 371         unsigned long count = 0;
 372         int i;
 373
 374         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 375
 376         spin_lock_bh(&tun->lock);
 377         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 378                 struct tun_flow_entry *e;
 379                 struct hlist_node *n;
 380
 381                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 382                         unsigned long this_timer;
 383                         count++;
 384                         this_timer = e->updated + delay;
 385                         if (time_before_eq(this_timer, jiffies))
 386                                 tun_flow_delete(tun, e);
 387                         else if (time_before(this_timer, next_timer))
 388                                 next_timer = this_timer;
 389                 }
 390         }
 391
 392         if (count)
 393                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 394         spin_unlock_bh(&tun->lock);
 395 }
 396
 397 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 398                             struct tun_file *tfile)
 399 {
 400         struct hlist_head *head;
 401         struct tun_flow_entry *e;
 402         unsigned long delay = tun->ageing_time;
 403         u16 queue_index = tfile->queue_index;
 404
 405         if (!rxhash)
 406                 return;
 407         else
 408                 head = &tun->flows[tun_hashfn(rxhash)];
 409
 410         rcu_read_lock();
 411
 412         /* We may get a very small possibility of OOO during switching, not
 413          * worth to optimize.*/
 414         if (tun->numqueues == 1 || tfile->detached)
 415                 goto unlock;
 416
 417         e = tun_flow_find(head, rxhash);
 418         if (likely(e)) {
 419                 /* TODO: keep queueing to old queue until it's empty? */
 420                 e->queue_index = queue_index;
 421                 e->updated = jiffies;
 422                 sock_rps_record_flow_hash(e->rps_rxhash);
 423         } else {
 424                 spin_lock_bh(&tun->lock);
 425                 if (!tun_flow_find(head, rxhash) &&
 426                     tun->flow_count < MAX_TAP_FLOWS)
 427                         tun_flow_create(tun, head, rxhash, queue_index);
 428
 429                 if (!timer_pending(&tun->flow_gc_timer))
 430                         mod_timer(&tun->flow_gc_timer,
 431                                   round_jiffies_up(jiffies + delay));
 432                 spin_unlock_bh(&tun->lock);
 433         }
 434
 435 unlock:
 436         rcu_read_unlock();
 437 }
 438
 439 /**
 440  * Save the hash received in the stack receive path and update the
 441  * flow_hash table accordingly.
 442  */
 443 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 444 {
 445         if (unlikely(e->rps_rxhash != hash))
 446                 e->rps_rxhash = hash;
 447 }
 448
 449 /* We try to identify a flow through its rxhash first. The reason that
 450  * we do not check rxq no. is because some cards(e.g 82599), chooses
 451  * the rxq based on the txq where the last packet of the flow comes. As
 452  * the userspace application move between processors, we may get a
 453  * different rxq no. here. If we could not get rxhash, then we would
 454  * hope the rxq no. may help here.
 455  */
 456 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 457                             void *accel_priv, select_queue_fallback_t fallback)
 458 {
 459         struct tun_struct *tun = netdev_priv(dev);
 460         struct tun_flow_entry *e;
 461         u32 txq = 0;
 462         u32 numqueues = 0;
 463
 464         rcu_read_lock();
 465         numqueues = ACCESS_ONCE(tun->numqueues);
 466
 467         txq = skb_get_hash(skb);
 468         if (txq) {
 469                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 470                 if (e) {
 471                         tun_flow_save_rps_rxhash(e, txq);
 472                         txq = e->queue_index;
 473                 } else
 474                         /* use multiply and shift instead of expensive divide */
 475                         txq = ((u64)txq * numqueues) >> 32;
 476         } else if (likely(skb_rx_queue_recorded(skb))) {
 477                 txq = skb_get_rx_queue(skb);
 478                 while (unlikely(txq >= numqueues))
 479                         txq -= numqueues;
 480         }
 481
 482         rcu_read_unlock();
 483         return txq;
 484 }
 485
 486 static inline bool tun_not_capable(struct tun_struct *tun)
 487 {
 488         const struct cred *cred = current_cred();
 489         struct net *net = dev_net(tun->dev);
 490
 491         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 492                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 493                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 494 }
 495
 496 static void tun_set_real_num_queues(struct tun_struct *tun)
 497 {
 498         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 499         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 500 }
 501
 502 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 503 {
 504         tfile->detached = tun;
 505         list_add_tail(&tfile->next, &tun->disabled);
 506         ++tun->numdisabled;
 507 }
 508
 509 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 510 {
 511         struct tun_struct *tun = tfile->detached;
 512
 513         tfile->detached = NULL;
 514         list_del_init(&tfile->next);
 515         --tun->numdisabled;
 516         return tun;
 517 }
 518
 519 static void tun_queue_purge(struct tun_file *tfile)
 520 {
 521         struct sk_buff *skb;
 522
 523         while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
 524                 kfree_skb(skb);
 525
 526         skb_queue_purge(&tfile->sk.sk_write_queue);
 527         skb_queue_purge(&tfile->sk.sk_error_queue);
 528 }
 529
 530 static void __tun_detach(struct tun_file *tfile, bool clean)
 531 {
 532         struct tun_file *ntfile;
 533         struct tun_struct *tun;
 534
 535         tun = rtnl_dereference(tfile->tun);
 536
 537         if (tun && !tfile->detached) {
 538                 u16 index = tfile->queue_index;
 539                 BUG_ON(index >= tun->numqueues);
 540
 541                 rcu_assign_pointer(tun->tfiles[index],
 542                                    tun->tfiles[tun->numqueues - 1]);
 543                 ntfile = rtnl_dereference(tun->tfiles[index]);
 544                 ntfile->queue_index = index;
 545
 546                 --tun->numqueues;
 547                 if (clean) {
 548                         RCU_INIT_POINTER(tfile->tun, NULL);
 549                         sock_put(&tfile->sk);
 550                 } else
 551                         tun_disable_queue(tun, tfile);
 552
 553                 synchronize_net();
 554                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 555                 /* Drop read queue */
 556                 tun_queue_purge(tfile);
 557                 tun_set_real_num_queues(tun);
 558         } else if (tfile->detached && clean) {
 559                 tun = tun_enable_queue(tfile);
 560                 sock_put(&tfile->sk);
 561         }
 562
 563         if (clean) {
 564                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 565                         netif_carrier_off(tun->dev);
 566
 567                         if (!(tun->flags & IFF_PERSIST) &&
 568                             tun->dev->reg_state == NETREG_REGISTERED)
 569                                 unregister_netdevice(tun->dev);
 570                 }
 571                 if (tun)
 572                         skb_array_cleanup(&tfile->tx_array);
 573                 sock_put(&tfile->sk);
 574         }
 575 }
 576
 577 static void tun_detach(struct tun_file *tfile, bool clean)
 578 {
 579         rtnl_lock();
 580         __tun_detach(tfile, clean);
 581         rtnl_unlock();
 582 }
 583
 584 static void tun_detach_all(struct net_device *dev)
 585 {
 586         struct tun_struct *tun = netdev_priv(dev);
 587         struct tun_file *tfile, *tmp;
 588         int i, n = tun->numqueues;
 589
 590         for (i = 0; i < n; i++) {
 591                 tfile = rtnl_dereference(tun->tfiles[i]);
 592                 BUG_ON(!tfile);
 593                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 594                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 595                 RCU_INIT_POINTER(tfile->tun, NULL);
 596                 --tun->numqueues;
 597         }
 598         list_for_each_entry(tfile, &tun->disabled, next) {
 599                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 600                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 601                 RCU_INIT_POINTER(tfile->tun, NULL);
 602         }
 603         BUG_ON(tun->numqueues != 0);
 604
 605         synchronize_net();
 606         for (i = 0; i < n; i++) {
 607                 tfile = rtnl_dereference(tun->tfiles[i]);
 608                 /* Drop read queue */
 609                 tun_queue_purge(tfile);
 610                 sock_put(&tfile->sk);
 611         }
 612         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 613                 tun_enable_queue(tfile);
 614                 tun_queue_purge(tfile);
 615                 sock_put(&tfile->sk);
 616         }
 617         BUG_ON(tun->numdisabled != 0);
 618
 619         if (tun->flags & IFF_PERSIST)
 620                 module_put(THIS_MODULE);
 621 }
 622
 623 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 624 {
 625         struct tun_file *tfile = file->private_data;
 626         struct net_device *dev = tun->dev;
 627         int err;
 628
 629         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 630         if (err < 0)
 631                 goto out;
 632
 633         err = -EINVAL;
 634         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 635                 goto out;
 636
 637         err = -EBUSY;
 638         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 639                 goto out;
 640
 641         err = -E2BIG;
 642         if (!tfile->detached &&
 643             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 644                 goto out;
 645
 646         err = 0;
 647
 648         /* Re-attach the filter to persist device */
 649         if (!skip_filter && (tun->filter_attached == true)) {
 650                 lock_sock(tfile->socket.sk);
 651                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 652                 release_sock(tfile->socket.sk);
 653                 if (!err)
 654                         goto out;
 655         }
 656
 657         if (!tfile->detached &&
 658             skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
 659                 err = -ENOMEM;
 660                 goto out;
 661         }
 662
 663         tfile->queue_index = tun->numqueues;
 664         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 665         rcu_assign_pointer(tfile->tun, tun);
 666         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 667         tun->numqueues++;
 668
 669         if (tfile->detached)
 670                 tun_enable_queue(tfile);
 671         else
 672                 sock_hold(&tfile->sk);
 673
 674         tun_set_real_num_queues(tun);
 675
 676         /* device is allowed to go away first, so no need to hold extra
 677          * refcnt.
 678          */
 679
 680 out:
 681         return err;
 682 }
 683
 684 static struct tun_struct *__tun_get(struct tun_file *tfile)
 685 {
 686         struct tun_struct *tun;
 687
 688         rcu_read_lock();
 689         tun = rcu_dereference(tfile->tun);
 690         if (tun)
 691                 dev_hold(tun->dev);
 692         rcu_read_unlock();
 693
 694         return tun;
 695 }
 696
 697 static struct tun_struct *tun_get(struct file *file)
 698 {
 699         return __tun_get(file->private_data);
 700 }
 701
 702 static void tun_put(struct tun_struct *tun)
 703 {
 704         dev_put(tun->dev);
 705 }
 706
 707 /* TAP filtering */
 708 static void addr_hash_set(u32 *mask, const u8 *addr)
 709 {
 710         int n = ether_crc(ETH_ALEN, addr) >> 26;
 711         mask[n >> 5] |= (1 << (n & 31));
 712 }
 713
 714 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 715 {
 716         int n = ether_crc(ETH_ALEN, addr) >> 26;
 717         return mask[n >> 5] & (1 << (n & 31));
 718 }
 719
 720 static int update_filter(struct tap_filter *filter, void __user *arg)
 721 {
 722         struct { u8 u[ETH_ALEN]; } *addr;
 723         struct tun_filter uf;
 724         int err, alen, n, nexact;
 725
 726         if (copy_from_user(&uf, arg, sizeof(uf)))
 727                 return -EFAULT;
 728
 729         if (!uf.count) {
 730                 /* Disabled */
 731                 filter->count = 0;
 732                 return 0;
 733         }
 734
 735         alen = ETH_ALEN * uf.count;
 736         addr = memdup_user(arg + sizeof(uf), alen);
 737         if (IS_ERR(addr))
 738                 return PTR_ERR(addr);
 739
 740         /* The filter is updated without holding any locks. Which is
 741          * perfectly safe. We disable it first and in the worst
 742          * case we'll accept a few undesired packets. */
 743         filter->count = 0;
 744         wmb();
 745
 746         /* Use first set of addresses as an exact filter */
 747         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 748                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 749
 750         nexact = n;
 751
 752         /* Remaining multicast addresses are hashed,
 753          * unicast will leave the filter disabled. */
 754         memset(filter->mask, 0, sizeof(filter->mask));
 755         for (; n < uf.count; n++) {
 756                 if (!is_multicast_ether_addr(addr[n].u)) {
 757                         err = 0; /* no filter */
 758                         goto free_addr;
 759                 }
 760                 addr_hash_set(filter->mask, addr[n].u);
 761         }
 762
 763         /* For ALLMULTI just set the mask to all ones.
 764          * This overrides the mask populated above. */
 765         if ((uf.flags & TUN_FLT_ALLMULTI))
 766                 memset(filter->mask, ~0, sizeof(filter->mask));
 767
 768         /* Now enable the filter */
 769         wmb();
 770         filter->count = nexact;
 771
 772         /* Return the number of exact filters */
 773         err = nexact;
 774 free_addr:
 775         kfree(addr);
 776         return err;
 777 }
 778
 779 /* Returns: 0 - drop, !=0 - accept */
 780 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 781 {
 782         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 783          * at this point. */
 784         struct ethhdr *eh = (struct ethhdr *) skb->data;
 785         int i;
 786
 787         /* Exact match */
 788         for (i = 0; i < filter->count; i++)
 789                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 790                         return 1;
 791
 792         /* Inexact match (multicast only) */
 793         if (is_multicast_ether_addr(eh->h_dest))
 794                 return addr_hash_test(filter->mask, eh->h_dest);
 795
 796         return 0;
 797 }
 798
 799 /*
 800  * Checks whether the packet is accepted or not.
 801  * Returns: 0 - drop, !=0 - accept
 802  */
 803 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 804 {
 805         if (!filter->count)
 806                 return 1;
 807
 808         return run_filter(filter, skb);
 809 }
 810
 811 /* Network device part of the driver */
 812
 813 static const struct ethtool_ops tun_ethtool_ops;
 814
 815 /* Net device detach from fd. */
 816 static void tun_net_uninit(struct net_device *dev)
 817 {
 818         tun_detach_all(dev);
 819 }
 820
 821 /* Net device open. */
 822 static int tun_net_open(struct net_device *dev)
 823 {
 824         netif_tx_start_all_queues(dev);
 825         return 0;
 826 }
 827
 828 /* Net device close. */
 829 static int tun_net_close(struct net_device *dev)
 830 {
 831         netif_tx_stop_all_queues(dev);
 832         return 0;
 833 }
 834
 835 /* Net device start xmit */
 836 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 837 {
 838         struct tun_struct *tun = netdev_priv(dev);
 839         int txq = skb->queue_mapping;
 840         struct tun_file *tfile;
 841         u32 numqueues = 0;
 842
 843         rcu_read_lock();
 844         tfile = rcu_dereference(tun->tfiles[txq]);
 845         numqueues = ACCESS_ONCE(tun->numqueues);
 846
 847         /* Drop packet if interface is not attached */
 848         if (txq >= numqueues)
 849                 goto drop;
 850
 851 #ifdef CONFIG_RPS
 852         if (numqueues == 1 && static_key_false(&rps_needed)) {
 853                 /* Select queue was not called for the skbuff, so we extract the
 854                  * RPS hash and save it into the flow_table here.
 855                  */
 856                 __u32 rxhash;
 857
 858                 rxhash = skb_get_hash(skb);
 859                 if (rxhash) {
 860                         struct tun_flow_entry *e;
 861                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 862                                         rxhash);
 863                         if (e)
 864                                 tun_flow_save_rps_rxhash(e, rxhash);
 865                 }
 866         }
 867 #endif
 868
 869         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 870
 871         BUG_ON(!tfile);
 872
 873         /* Drop if the filter does not like it.
 874          * This is a noop if the filter is disabled.
 875          * Filter can be enabled only for the TAP devices. */
 876         if (!check_filter(&tun->txflt, skb))
 877                 goto drop;
 878
 879         if (tfile->socket.sk->sk_filter &&
 880             sk_filter(tfile->socket.sk, skb))
 881                 goto drop;
 882
 883         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 884                 goto drop;
 885
 886         skb_tx_timestamp(skb);
 887
 888         /* Orphan the skb - required as we might hang on to it
 889          * for indefinite time.
 890          */
 891         skb_orphan(skb);
 892
 893         nf_reset(skb);
 894
 895         if (skb_array_produce(&tfile->tx_array, skb))
 896                 goto drop;
 897
 898         /* Notify and wake up reader process */
 899         if (tfile->flags & TUN_FASYNC)
 900                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 901         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 902
 903         rcu_read_unlock();
 904         return NETDEV_TX_OK;
 905
 906 drop:
 907         this_cpu_inc(tun->pcpu_stats->tx_dropped);
 908         skb_tx_error(skb);
 909         kfree_skb(skb);
 910         rcu_read_unlock();
 911         return NET_XMIT_DROP;
 912 }
 913
 914 static void tun_net_mclist(struct net_device *dev)
 915 {
 916         /*
 917          * This callback is supposed to deal with mc filter in
 918          * _rx_ path and has nothing to do with the _tx_ path.
 919          * In rx path we always accept everything userspace gives us.
 920          */
 921 }
 922
 923 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 924         netdev_features_t features)
 925 {
 926         struct tun_struct *tun = netdev_priv(dev);
 927
 928         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 929 }
 930 #ifdef CONFIG_NET_POLL_CONTROLLER
 931 static void tun_poll_controller(struct net_device *dev)
 932 {
 933         /*
 934          * Tun only receives frames when:
 935          * 1) the char device endpoint gets data from user space
 936          * 2) the tun socket gets a sendmsg call from user space
 937          * Since both of those are synchronous operations, we are guaranteed
 938          * never to have pending data when we poll for it
 939          * so there is nothing to do here but return.
 940          * We need this though so netpoll recognizes us as an interface that
 941          * supports polling, which enables bridge devices in virt setups to
 942          * still use netconsole
 943          */
 944         return;
 945 }
 946 #endif
 947
 948 static void tun_set_headroom(struct net_device *dev, int new_hr)
 949 {
 950         struct tun_struct *tun = netdev_priv(dev);
 951
 952         if (new_hr < NET_SKB_PAD)
 953                 new_hr = NET_SKB_PAD;
 954
 955         tun->align = new_hr;
 956 }
 957
 958 static void
 959 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
 960 {
 961         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
 962         struct tun_struct *tun = netdev_priv(dev);
 963         struct tun_pcpu_stats *p;
 964         int i;
 965
 966         for_each_possible_cpu(i) {
 967                 u64 rxpackets, rxbytes, txpackets, txbytes;
 968                 unsigned int start;
 969
 970                 p = per_cpu_ptr(tun->pcpu_stats, i);
 971                 do {
 972                         start = u64_stats_fetch_begin(&p->syncp);
 973                         rxpackets       = p->rx_packets;
 974                         rxbytes         = p->rx_bytes;
 975                         txpackets       = p->tx_packets;
 976                         txbytes         = p->tx_bytes;
 977                 } while (u64_stats_fetch_retry(&p->syncp, start));
 978
 979                 stats->rx_packets       += rxpackets;
 980                 stats->rx_bytes         += rxbytes;
 981                 stats->tx_packets       += txpackets;
 982                 stats->tx_bytes         += txbytes;
 983
 984                 /* u32 counters */
 985                 rx_dropped      += p->rx_dropped;
 986                 rx_frame_errors += p->rx_frame_errors;
 987                 tx_dropped      += p->tx_dropped;
 988         }
 989         stats->rx_dropped  = rx_dropped;
 990         stats->rx_frame_errors = rx_frame_errors;
 991         stats->tx_dropped = tx_dropped;
 992 }
 993
 994 static const struct net_device_ops tun_netdev_ops = {
 995         .ndo_uninit             = tun_net_uninit,
 996         .ndo_open               = tun_net_open,
 997         .ndo_stop               = tun_net_close,
 998         .ndo_start_xmit         = tun_net_xmit,
 999         .ndo_fix_features       = tun_net_fix_features,
1000         .ndo_select_queue       = tun_select_queue,
1001 #ifdef CONFIG_NET_POLL_CONTROLLER
1002         .ndo_poll_controller    = tun_poll_controller,
1003 #endif
1004         .ndo_set_rx_headroom    = tun_set_headroom,
1005         .ndo_get_stats64        = tun_net_get_stats64,
1006 };
1007
1008 static const struct net_device_ops tap_netdev_ops = {
1009         .ndo_uninit             = tun_net_uninit,
1010         .ndo_open               = tun_net_open,
1011         .ndo_stop               = tun_net_close,
1012         .ndo_start_xmit         = tun_net_xmit,
1013         .ndo_fix_features       = tun_net_fix_features,
1014         .ndo_set_rx_mode        = tun_net_mclist,
1015         .ndo_set_mac_address    = eth_mac_addr,
1016         .ndo_validate_addr      = eth_validate_addr,
1017         .ndo_select_queue       = tun_select_queue,
1018 #ifdef CONFIG_NET_POLL_CONTROLLER
1019         .ndo_poll_controller    = tun_poll_controller,
1020 #endif
1021         .ndo_features_check     = passthru_features_check,
1022         .ndo_set_rx_headroom    = tun_set_headroom,
1023         .ndo_get_stats64        = tun_net_get_stats64,
1024 };
1025
1026 static void tun_flow_init(struct tun_struct *tun)
1027 {
1028         int i;
1029
1030         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1031                 INIT_HLIST_HEAD(&tun->flows[i]);
1032
1033         tun->ageing_time = TUN_FLOW_EXPIRE;
1034         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1035         mod_timer(&tun->flow_gc_timer,
1036                   round_jiffies_up(jiffies + tun->ageing_time));
1037 }
1038
1039 static void tun_flow_uninit(struct tun_struct *tun)
1040 {
1041         del_timer_sync(&tun->flow_gc_timer);
1042         tun_flow_flush(tun);
1043 }
1044
1045 #define MIN_MTU 68
1046 #define MAX_MTU 65535
1047
1048 /* Initialize net device. */
1049 static void tun_net_init(struct net_device *dev)
1050 {
1051         struct tun_struct *tun = netdev_priv(dev);
1052
1053         switch (tun->flags & TUN_TYPE_MASK) {
1054         case IFF_TUN:
1055                 dev->netdev_ops = &tun_netdev_ops;
1056
1057                 /* Point-to-Point TUN Device */
1058                 dev->hard_header_len = 0;
1059                 dev->addr_len = 0;
1060                 dev->mtu = 1500;
1061
1062                 /* Zero header length */
1063                 dev->type = ARPHRD_NONE;
1064                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1065                 break;
1066
1067         case IFF_TAP:
1068                 dev->netdev_ops = &tap_netdev_ops;
1069                 /* Ethernet TAP Device */
1070                 ether_setup(dev);
1071                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1072                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1073
1074                 eth_hw_addr_random(dev);
1075
1076                 break;
1077         }
1078
1079         dev->min_mtu = MIN_MTU;
1080         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1081 }
1082
1083 /* Character device part */
1084
1085 /* Poll */
1086 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1087 {
1088         struct tun_file *tfile = file->private_data;
1089         struct tun_struct *tun = __tun_get(tfile);
1090         struct sock *sk;
1091         unsigned int mask = 0;
1092
1093         if (!tun)
1094                 return POLLERR;
1095
1096         sk = tfile->socket.sk;
1097
1098         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1099
1100         poll_wait(file, sk_sleep(sk), wait);
1101
1102         if (!skb_array_empty(&tfile->tx_array))
1103                 mask |= POLLIN | POLLRDNORM;
1104
1105         if (sock_writeable(sk) ||
1106             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1107              sock_writeable(sk)))
1108                 mask |= POLLOUT | POLLWRNORM;
1109
1110         if (tun->dev->reg_state != NETREG_REGISTERED)
1111                 mask = POLLERR;
1112
1113         tun_put(tun);
1114         return mask;
1115 }
1116
1117 /* prepad is the amount to reserve at front.  len is length after that.
1118  * linear is a hint as to how much to copy (usually headers). */
1119 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1120                                      size_t prepad, size_t len,
1121                                      size_t linear, int noblock)
1122 {
1123         struct sock *sk = tfile->socket.sk;
1124         struct sk_buff *skb;
1125         int err;
1126
1127         /* Under a page?  Don't bother with paged skb. */
1128         if (prepad + len < PAGE_SIZE || !linear)
1129                 linear = len;
1130
1131         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1132                                    &err, 0);
1133         if (!skb)
1134                 return ERR_PTR(err);
1135
1136         skb_reserve(skb, prepad);
1137         skb_put(skb, linear);
1138         skb->data_len = len - linear;
1139         skb->len += len - linear;
1140
1141         return skb;
1142 }
1143
1144 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1145                            struct sk_buff *skb, int more)
1146 {
1147         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1148         struct sk_buff_head process_queue;
1149         u32 rx_batched = tun->rx_batched;
1150         bool rcv = false;
1151
1152         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1153                 local_bh_disable();
1154                 netif_receive_skb(skb);
1155                 local_bh_enable();
1156                 return;
1157         }
1158
1159         spin_lock(&queue->lock);
1160         if (!more || skb_queue_len(queue) == rx_batched) {
1161                 __skb_queue_head_init(&process_queue);
1162                 skb_queue_splice_tail_init(queue, &process_queue);
1163                 rcv = true;
1164         } else {
1165                 __skb_queue_tail(queue, skb);
1166         }
1167         spin_unlock(&queue->lock);
1168
1169         if (rcv) {
1170                 struct sk_buff *nskb;
1171
1172                 local_bh_disable();
1173                 while ((nskb = __skb_dequeue(&process_queue)))
1174                         netif_receive_skb(nskb);
1175                 netif_receive_skb(skb);
1176                 local_bh_enable();
1177         }
1178 }
1179
1180 /* Get packet from user space buffer */
1181 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1182                             void *msg_control, struct iov_iter *from,
1183                             int noblock, bool more)
1184 {
1185         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1186         struct sk_buff *skb;
1187         size_t total_len = iov_iter_count(from);
1188         size_t len = total_len, align = tun->align, linear;
1189         struct virtio_net_hdr gso = { 0 };
1190         struct tun_pcpu_stats *stats;
1191         int good_linear;
1192         int copylen;
1193         bool zerocopy = false;
1194         int err;
1195         u32 rxhash;
1196
1197         if (!(tun->dev->flags & IFF_UP))
1198                 return -EIO;
1199
1200         if (!(tun->flags & IFF_NO_PI)) {
1201                 if (len < sizeof(pi))
1202                         return -EINVAL;
1203                 len -= sizeof(pi);
1204
1205                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1206                         return -EFAULT;
1207         }
1208
1209         if (tun->flags & IFF_VNET_HDR) {
1210                 if (len < tun->vnet_hdr_sz)
1211                         return -EINVAL;
1212                 len -= tun->vnet_hdr_sz;
1213
1214                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1215                         return -EFAULT;
1216
1217                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1218                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1219                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1220
1221                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1222                         return -EINVAL;
1223                 iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1224         }
1225
1226         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1227                 align += NET_IP_ALIGN;
1228                 if (unlikely(len < ETH_HLEN ||
1229                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1230                         return -EINVAL;
1231         }
1232
1233         good_linear = SKB_MAX_HEAD(align);
1234
1235         if (msg_control) {
1236                 struct iov_iter i = *from;
1237
1238                 /* There are 256 bytes to be copied in skb, so there is
1239                  * enough room for skb expand head in case it is used.
1240                  * The rest of the buffer is mapped from userspace.
1241                  */
1242                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1243                 if (copylen > good_linear)
1244                         copylen = good_linear;
1245                 linear = copylen;
1246                 iov_iter_advance(&i, copylen);
1247                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1248                         zerocopy = true;
1249         }
1250
1251         if (!zerocopy) {
1252                 copylen = len;
1253                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1254                         linear = good_linear;
1255                 else
1256                         linear = tun16_to_cpu(tun, gso.hdr_len);
1257         }
1258
1259         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1260         if (IS_ERR(skb)) {
1261                 if (PTR_ERR(skb) != -EAGAIN)
1262                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1263                 return PTR_ERR(skb);
1264         }
1265
1266         if (zerocopy)
1267                 err = zerocopy_sg_from_iter(skb, from);
1268         else
1269                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1270
1271         if (err) {
1272                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1273                 kfree_skb(skb);
1274                 return -EFAULT;
1275         }
1276
1277         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1278                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1279                 kfree_skb(skb);
1280                 return -EINVAL;
1281         }
1282
1283         switch (tun->flags & TUN_TYPE_MASK) {
1284         case IFF_TUN:
1285                 if (tun->flags & IFF_NO_PI) {
1286                         switch (skb->data[0] & 0xf0) {
1287                         case 0x40:
1288                                 pi.proto = htons(ETH_P_IP);
1289                                 break;
1290                         case 0x60:
1291                                 pi.proto = htons(ETH_P_IPV6);
1292                                 break;
1293                         default:
1294                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1295                                 kfree_skb(skb);
1296                                 return -EINVAL;
1297                         }
1298                 }
1299
1300                 skb_reset_mac_header(skb);
1301                 skb->protocol = pi.proto;
1302                 skb->dev = tun->dev;
1303                 break;
1304         case IFF_TAP:
1305                 skb->protocol = eth_type_trans(skb, tun->dev);
1306                 break;
1307         }
1308
1309         /* copy skb_ubuf_info for callback when skb has no error */
1310         if (zerocopy) {
1311                 skb_shinfo(skb)->destructor_arg = msg_control;
1312                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1313                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1314         } else if (msg_control) {
1315                 struct ubuf_info *uarg = msg_control;
1316                 uarg->callback(uarg, false);
1317         }
1318
1319         skb_reset_network_header(skb);
1320         skb_probe_transport_header(skb, 0);
1321
1322         rxhash = skb_get_hash(skb);
1323 #ifndef CONFIG_4KSTACKS
1324         tun_rx_batched(tun, tfile, skb, more);
1325 #else
1326         netif_rx_ni(skb);
1327 #endif
1328
1329         stats = get_cpu_ptr(tun->pcpu_stats);
1330         u64_stats_update_begin(&stats->syncp);
1331         stats->rx_packets++;
1332         stats->rx_bytes += len;
1333         u64_stats_update_end(&stats->syncp);
1334         put_cpu_ptr(stats);
1335
1336         tun_flow_update(tun, rxhash, tfile);
1337         return total_len;
1338 }
1339
1340 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1341 {
1342         struct file *file = iocb->ki_filp;
1343         struct tun_struct *tun = tun_get(file);
1344         struct tun_file *tfile = file->private_data;
1345         ssize_t result;
1346
1347         if (!tun)
1348                 return -EBADFD;
1349
1350         result = tun_get_user(tun, tfile, NULL, from,
1351                               file->f_flags & O_NONBLOCK, false);
1352
1353         tun_put(tun);
1354         return result;
1355 }
1356
1357 /* Put packet to the user space buffer */
1358 static ssize_t tun_put_user(struct tun_struct *tun,
1359                             struct tun_file *tfile,
1360                             struct sk_buff *skb,
1361                             struct iov_iter *iter)
1362 {
1363         struct tun_pi pi = { 0, skb->protocol };
1364         struct tun_pcpu_stats *stats;
1365         ssize_t total;
1366         int vlan_offset = 0;
1367         int vlan_hlen = 0;
1368         int vnet_hdr_sz = 0;
1369
1370         if (skb_vlan_tag_present(skb))
1371                 vlan_hlen = VLAN_HLEN;
1372
1373         if (tun->flags & IFF_VNET_HDR)
1374                 vnet_hdr_sz = tun->vnet_hdr_sz;
1375
1376         total = skb->len + vlan_hlen + vnet_hdr_sz;
1377
1378         if (!(tun->flags & IFF_NO_PI)) {
1379                 if (iov_iter_count(iter) < sizeof(pi))
1380                         return -EINVAL;
1381
1382                 total += sizeof(pi);
1383                 if (iov_iter_count(iter) < total) {
1384                         /* Packet will be striped */
1385                         pi.flags |= TUN_PKT_STRIP;
1386                 }
1387
1388                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1389                         return -EFAULT;
1390         }
1391
1392         if (vnet_hdr_sz) {
1393                 struct virtio_net_hdr gso;
1394
1395                 if (iov_iter_count(iter) < vnet_hdr_sz)
1396                         return -EINVAL;
1397
1398                 if (virtio_net_hdr_from_skb(skb, &gso,
1399                                             tun_is_little_endian(tun), true)) {
1400                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1401                         pr_err("unexpected GSO type: "
1402                                "0x%x, gso_size %d, hdr_len %d\n",
1403                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1404                                tun16_to_cpu(tun, gso.hdr_len));
1405                         print_hex_dump(KERN_ERR, "tun: ",
1406                                        DUMP_PREFIX_NONE,
1407                                        16, 1, skb->head,
1408                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1409                         WARN_ON_ONCE(1);
1410                         return -EINVAL;
1411                 }
1412
1413                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1414                         return -EFAULT;
1415
1416                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1417         }
1418
1419         if (vlan_hlen) {
1420                 int ret;
1421                 struct {
1422                         __be16 h_vlan_proto;
1423                         __be16 h_vlan_TCI;
1424                 } veth;
1425
1426                 veth.h_vlan_proto = skb->vlan_proto;
1427                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1428
1429                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1430
1431                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1432                 if (ret || !iov_iter_count(iter))
1433                         goto done;
1434
1435                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1436                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1437                         goto done;
1438         }
1439
1440         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1441
1442 done:
1443         /* caller is in process context, */
1444         stats = get_cpu_ptr(tun->pcpu_stats);
1445         u64_stats_update_begin(&stats->syncp);
1446         stats->tx_packets++;
1447         stats->tx_bytes += skb->len + vlan_hlen;
1448         u64_stats_update_end(&stats->syncp);
1449         put_cpu_ptr(tun->pcpu_stats);
1450
1451         return total;
1452 }
1453
1454 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1455                                      int *err)
1456 {
1457         DECLARE_WAITQUEUE(wait, current);
1458         struct sk_buff *skb = NULL;
1459         int error = 0;
1460
1461         skb = skb_array_consume(&tfile->tx_array);
1462         if (skb)
1463                 goto out;
1464         if (noblock) {
1465                 error = -EAGAIN;
1466                 goto out;
1467         }
1468
1469         add_wait_queue(&tfile->wq.wait, &wait);
1470         current->state = TASK_INTERRUPTIBLE;
1471
1472         while (1) {
1473                 skb = skb_array_consume(&tfile->tx_array);
1474                 if (skb)
1475                         break;
1476                 if (signal_pending(current)) {
1477                         error = -ERESTARTSYS;
1478                         break;
1479                 }
1480                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1481                         error = -EFAULT;
1482                         break;
1483                 }
1484
1485                 schedule();
1486         }
1487
1488         current->state = TASK_RUNNING;
1489         remove_wait_queue(&tfile->wq.wait, &wait);
1490
1491 out:
1492         *err = error;
1493         return skb;
1494 }
1495
1496 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1497                            struct iov_iter *to,
1498                            int noblock)
1499 {
1500         struct sk_buff *skb;
1501         ssize_t ret;
1502         int err;
1503
1504         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1505
1506         if (!iov_iter_count(to))
1507                 return 0;
1508
1509         /* Read frames from ring */
1510         skb = tun_ring_recv(tfile, noblock, &err);
1511         if (!skb)
1512                 return err;
1513
1514         ret = tun_put_user(tun, tfile, skb, to);
1515         if (unlikely(ret < 0))
1516                 kfree_skb(skb);
1517         else
1518                 consume_skb(skb);
1519
1520         return ret;
1521 }
1522
1523 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1524 {
1525         struct file *file = iocb->ki_filp;
1526         struct tun_file *tfile = file->private_data;
1527         struct tun_struct *tun = __tun_get(tfile);
1528         ssize_t len = iov_iter_count(to), ret;
1529
1530         if (!tun)
1531                 return -EBADFD;
1532         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1533         ret = min_t(ssize_t, ret, len);
1534         if (ret > 0)
1535                 iocb->ki_pos = ret;
1536         tun_put(tun);
1537         return ret;
1538 }
1539
1540 static void tun_free_netdev(struct net_device *dev)
1541 {
1542         struct tun_struct *tun = netdev_priv(dev);
1543
1544         BUG_ON(!(list_empty(&tun->disabled)));
1545         free_percpu(tun->pcpu_stats);
1546         tun_flow_uninit(tun);
1547         security_tun_dev_free_security(tun->security);
1548         free_netdev(dev);
1549 }
1550
1551 static void tun_setup(struct net_device *dev)
1552 {
1553         struct tun_struct *tun = netdev_priv(dev);
1554
1555         tun->owner = INVALID_UID;
1556         tun->group = INVALID_GID;
1557
1558         dev->ethtool_ops = &tun_ethtool_ops;
1559         dev->destructor = tun_free_netdev;
1560         /* We prefer our own queue length */
1561         dev->tx_queue_len = TUN_READQ_SIZE;
1562 }
1563
1564 /* Trivial set of netlink ops to allow deleting tun or tap
1565  * device with netlink.
1566  */
1567 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1568 {
1569         return -EINVAL;
1570 }
1571
1572 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1573         .kind           = DRV_NAME,
1574         .priv_size      = sizeof(struct tun_struct),
1575         .setup          = tun_setup,
1576         .validate       = tun_validate,
1577 };
1578
1579 static void tun_sock_write_space(struct sock *sk)
1580 {
1581         struct tun_file *tfile;
1582         wait_queue_head_t *wqueue;
1583
1584         if (!sock_writeable(sk))
1585                 return;
1586
1587         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1588                 return;
1589
1590         wqueue = sk_sleep(sk);
1591         if (wqueue && waitqueue_active(wqueue))
1592                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1593                                                 POLLWRNORM | POLLWRBAND);
1594
1595         tfile = container_of(sk, struct tun_file, sk);
1596         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1597 }
1598
1599 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1600 {
1601         int ret;
1602         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1603         struct tun_struct *tun = __tun_get(tfile);
1604
1605         if (!tun)
1606                 return -EBADFD;
1607
1608         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1609                            m->msg_flags & MSG_DONTWAIT,
1610                            m->msg_flags & MSG_MORE);
1611         tun_put(tun);
1612         return ret;
1613 }
1614
1615 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1616                        int flags)
1617 {
1618         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1619         struct tun_struct *tun = __tun_get(tfile);
1620         int ret;
1621
1622         if (!tun)
1623                 return -EBADFD;
1624
1625         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1626                 ret = -EINVAL;
1627                 goto out;
1628         }
1629         if (flags & MSG_ERRQUEUE) {
1630                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1631                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1632                 goto out;
1633         }
1634         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1635         if (ret > (ssize_t)total_len) {
1636                 m->msg_flags |= MSG_TRUNC;
1637                 ret = flags & MSG_TRUNC ? ret : total_len;
1638         }
1639 out:
1640         tun_put(tun);
1641         return ret;
1642 }
1643
1644 static int tun_peek_len(struct socket *sock)
1645 {
1646         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1647         struct tun_struct *tun;
1648         int ret = 0;
1649
1650         tun = __tun_get(tfile);
1651         if (!tun)
1652                 return 0;
1653
1654         ret = skb_array_peek_len(&tfile->tx_array);
1655         tun_put(tun);
1656
1657         return ret;
1658 }
1659
1660 /* Ops structure to mimic raw sockets with tun */
1661 static const struct proto_ops tun_socket_ops = {
1662         .peek_len = tun_peek_len,
1663         .sendmsg = tun_sendmsg,
1664         .recvmsg = tun_recvmsg,
1665 };
1666
1667 static struct proto tun_proto = {
1668         .name           = "tun",
1669         .owner          = THIS_MODULE,
1670         .obj_size       = sizeof(struct tun_file),
1671 };
1672
1673 static int tun_flags(struct tun_struct *tun)
1674 {
1675         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1676 }
1677
1678 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1679                               char *buf)
1680 {
1681         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1682         return sprintf(buf, "0x%x\n", tun_flags(tun));
1683 }
1684
1685 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1686                               char *buf)
1687 {
1688         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1689         return uid_valid(tun->owner)?
1690                 sprintf(buf, "%u\n",
1691                         from_kuid_munged(current_user_ns(), tun->owner)):
1692                 sprintf(buf, "-1\n");
1693 }
1694
1695 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1696                               char *buf)
1697 {
1698         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1699         return gid_valid(tun->group) ?
1700                 sprintf(buf, "%u\n",
1701                         from_kgid_munged(current_user_ns(), tun->group)):
1702                 sprintf(buf, "-1\n");
1703 }
1704
1705 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1706 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1707 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1708
1709 static struct attribute *tun_dev_attrs[] = {
1710         &dev_attr_tun_flags.attr,
1711         &dev_attr_owner.attr,
1712         &dev_attr_group.attr,
1713         NULL
1714 };
1715
1716 static const struct attribute_group tun_attr_group = {
1717         .attrs = tun_dev_attrs
1718 };
1719
1720 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1721 {
1722         struct tun_struct *tun;
1723         struct tun_file *tfile = file->private_data;
1724         struct net_device *dev;
1725         int err;
1726
1727         if (tfile->detached)
1728                 return -EINVAL;
1729
1730         dev = __dev_get_by_name(net, ifr->ifr_name);
1731         if (dev) {
1732                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1733                         return -EBUSY;
1734                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1735                         tun = netdev_priv(dev);
1736                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1737                         tun = netdev_priv(dev);
1738                 else
1739                         return -EINVAL;
1740
1741                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1742                     !!(tun->flags & IFF_MULTI_QUEUE))
1743                         return -EINVAL;
1744
1745                 if (tun_not_capable(tun))
1746                         return -EPERM;
1747                 err = security_tun_dev_open(tun->security);
1748                 if (err < 0)
1749                         return err;
1750
1751                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1752                 if (err < 0)
1753                         return err;
1754
1755                 if (tun->flags & IFF_MULTI_QUEUE &&
1756                     (tun->numqueues + tun->numdisabled > 1)) {
1757                         /* One or more queue has already been attached, no need
1758                          * to initialize the device again.
1759                          */
1760                         return 0;
1761                 }
1762         }
1763         else {
1764                 char *name;
1765                 unsigned long flags = 0;
1766                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1767                              MAX_TAP_QUEUES : 1;
1768
1769                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1770                         return -EPERM;
1771                 err = security_tun_dev_create();
1772                 if (err < 0)
1773                         return err;
1774
1775                 /* Set dev type */
1776                 if (ifr->ifr_flags & IFF_TUN) {
1777                         /* TUN device */
1778                         flags |= IFF_TUN;
1779                         name = "tun%d";
1780                 } else if (ifr->ifr_flags & IFF_TAP) {
1781                         /* TAP device */
1782                         flags |= IFF_TAP;
1783                         name = "tap%d";
1784                 } else
1785                         return -EINVAL;
1786
1787                 if (*ifr->ifr_name)
1788                         name = ifr->ifr_name;
1789
1790                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1791                                        NET_NAME_UNKNOWN, tun_setup, queues,
1792                                        queues);
1793
1794                 if (!dev)
1795                         return -ENOMEM;
1796
1797                 dev_net_set(dev, net);
1798                 dev->rtnl_link_ops = &tun_link_ops;
1799                 dev->ifindex = tfile->ifindex;
1800                 dev->sysfs_groups[0] = &tun_attr_group;
1801
1802                 tun = netdev_priv(dev);
1803                 tun->dev = dev;
1804                 tun->flags = flags;
1805                 tun->txflt.count = 0;
1806                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1807
1808                 tun->align = NET_SKB_PAD;
1809                 tun->filter_attached = false;
1810                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1811                 tun->rx_batched = 0;
1812
1813                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
1814                 if (!tun->pcpu_stats) {
1815                         err = -ENOMEM;
1816                         goto err_free_dev;
1817                 }
1818
1819                 spin_lock_init(&tun->lock);
1820
1821                 err = security_tun_dev_alloc_security(&tun->security);
1822                 if (err < 0)
1823                         goto err_free_stat;
1824
1825                 tun_net_init(dev);
1826                 tun_flow_init(tun);
1827
1828                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1829                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1830                                    NETIF_F_HW_VLAN_STAG_TX;
1831                 dev->features = dev->hw_features | NETIF_F_LLTX;
1832                 dev->vlan_features = dev->features &
1833                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1834                                        NETIF_F_HW_VLAN_STAG_TX);
1835
1836                 INIT_LIST_HEAD(&tun->disabled);
1837                 err = tun_attach(tun, file, false);
1838                 if (err < 0)
1839                         goto err_free_flow;
1840
1841                 err = register_netdevice(tun->dev);
1842                 if (err < 0)
1843                         goto err_detach;
1844         }
1845
1846         netif_carrier_on(tun->dev);
1847
1848         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1849
1850         tun->flags = (tun->flags & ~TUN_FEATURES) |
1851                 (ifr->ifr_flags & TUN_FEATURES);
1852
1853         /* Make sure persistent devices do not get stuck in
1854          * xoff state.
1855          */
1856         if (netif_running(tun->dev))
1857                 netif_tx_wake_all_queues(tun->dev);
1858
1859         strcpy(ifr->ifr_name, tun->dev->name);
1860         return 0;
1861
1862 err_detach:
1863         tun_detach_all(dev);
1864 err_free_flow:
1865         tun_flow_uninit(tun);
1866         security_tun_dev_free_security(tun->security);
1867 err_free_stat:
1868         free_percpu(tun->pcpu_stats);
1869 err_free_dev:
1870         free_netdev(dev);
1871         return err;
1872 }
1873
1874 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1875                        struct ifreq *ifr)
1876 {
1877         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1878
1879         strcpy(ifr->ifr_name, tun->dev->name);
1880
1881         ifr->ifr_flags = tun_flags(tun);
1882
1883 }
1884
1885 /* This is like a cut-down ethtool ops, except done via tun fd so no
1886  * privs required. */
1887 static int set_offload(struct tun_struct *tun, unsigned long arg)
1888 {
1889         netdev_features_t features = 0;
1890
1891         if (arg & TUN_F_CSUM) {
1892                 features |= NETIF_F_HW_CSUM;
1893                 arg &= ~TUN_F_CSUM;
1894
1895                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1896                         if (arg & TUN_F_TSO_ECN) {
1897                                 features |= NETIF_F_TSO_ECN;
1898                                 arg &= ~TUN_F_TSO_ECN;
1899                         }
1900                         if (arg & TUN_F_TSO4)
1901                                 features |= NETIF_F_TSO;
1902                         if (arg & TUN_F_TSO6)
1903                                 features |= NETIF_F_TSO6;
1904                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1905                 }
1906
1907                 if (arg & TUN_F_UFO) {
1908                         features |= NETIF_F_UFO;
1909                         arg &= ~TUN_F_UFO;
1910                 }
1911         }
1912
1913         /* This gives the user a way to test for new features in future by
1914          * trying to set them. */
1915         if (arg)
1916                 return -EINVAL;
1917
1918         tun->set_features = features;
1919         netdev_update_features(tun->dev);
1920
1921         return 0;
1922 }
1923
1924 static void tun_detach_filter(struct tun_struct *tun, int n)
1925 {
1926         int i;
1927         struct tun_file *tfile;
1928
1929         for (i = 0; i < n; i++) {
1930                 tfile = rtnl_dereference(tun->tfiles[i]);
1931                 lock_sock(tfile->socket.sk);
1932                 sk_detach_filter(tfile->socket.sk);
1933                 release_sock(tfile->socket.sk);
1934         }
1935
1936         tun->filter_attached = false;
1937 }
1938
1939 static int tun_attach_filter(struct tun_struct *tun)
1940 {
1941         int i, ret = 0;
1942         struct tun_file *tfile;
1943
1944         for (i = 0; i < tun->numqueues; i++) {
1945                 tfile = rtnl_dereference(tun->tfiles[i]);
1946                 lock_sock(tfile->socket.sk);
1947                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1948                 release_sock(tfile->socket.sk);
1949                 if (ret) {
1950                         tun_detach_filter(tun, i);
1951                         return ret;
1952                 }
1953         }
1954
1955         tun->filter_attached = true;
1956         return ret;
1957 }
1958
1959 static void tun_set_sndbuf(struct tun_struct *tun)
1960 {
1961         struct tun_file *tfile;
1962         int i;
1963
1964         for (i = 0; i < tun->numqueues; i++) {
1965                 tfile = rtnl_dereference(tun->tfiles[i]);
1966                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1967         }
1968 }
1969
1970 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1971 {
1972         struct tun_file *tfile = file->private_data;
1973         struct tun_struct *tun;
1974         int ret = 0;
1975
1976         rtnl_lock();
1977
1978         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1979                 tun = tfile->detached;
1980                 if (!tun) {
1981                         ret = -EINVAL;
1982                         goto unlock;
1983                 }
1984                 ret = security_tun_dev_attach_queue(tun->security);
1985                 if (ret < 0)
1986                         goto unlock;
1987                 ret = tun_attach(tun, file, false);
1988         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1989                 tun = rtnl_dereference(tfile->tun);
1990                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1991                         ret = -EINVAL;
1992                 else
1993                         __tun_detach(tfile, false);
1994         } else
1995                 ret = -EINVAL;
1996
1997 unlock:
1998         rtnl_unlock();
1999         return ret;
2000 }
2001
2002 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2003                             unsigned long arg, int ifreq_len)
2004 {
2005         struct tun_file *tfile = file->private_data;
2006         struct tun_struct *tun;
2007         void __user* argp = (void __user*)arg;
2008         struct ifreq ifr;
2009         kuid_t owner;
2010         kgid_t group;
2011         int sndbuf;
2012         int vnet_hdr_sz;
2013         unsigned int ifindex;
2014         int le;
2015         int ret;
2016
2017         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2018                 if (copy_from_user(&ifr, argp, ifreq_len))
2019                         return -EFAULT;
2020         } else {
2021                 memset(&ifr, 0, sizeof(ifr));
2022         }
2023         if (cmd == TUNGETFEATURES) {
2024                 /* Currently this just means: "what IFF flags are valid?".
2025                  * This is needed because we never checked for invalid flags on
2026                  * TUNSETIFF.
2027                  */
2028                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2029                                 (unsigned int __user*)argp);
2030         } else if (cmd == TUNSETQUEUE)
2031                 return tun_set_queue(file, &ifr);
2032
2033         ret = 0;
2034         rtnl_lock();
2035
2036         tun = __tun_get(tfile);
2037         if (cmd == TUNSETIFF) {
2038                 ret = -EEXIST;
2039                 if (tun)
2040                         goto unlock;
2041
2042                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2043
2044                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2045
2046                 if (ret)
2047                         goto unlock;
2048
2049                 if (copy_to_user(argp, &ifr, ifreq_len))
2050                         ret = -EFAULT;
2051                 goto unlock;
2052         }
2053         if (cmd == TUNSETIFINDEX) {
2054                 ret = -EPERM;
2055                 if (tun)
2056                         goto unlock;
2057
2058                 ret = -EFAULT;
2059                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2060                         goto unlock;
2061
2062                 ret = 0;
2063                 tfile->ifindex = ifindex;
2064                 goto unlock;
2065         }
2066
2067         ret = -EBADFD;
2068         if (!tun)
2069                 goto unlock;
2070
2071         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2072
2073         ret = 0;
2074         switch (cmd) {
2075         case TUNGETIFF:
2076                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2077
2078                 if (tfile->detached)
2079                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2080                 if (!tfile->socket.sk->sk_filter)
2081                         ifr.ifr_flags |= IFF_NOFILTER;
2082
2083                 if (copy_to_user(argp, &ifr, ifreq_len))
2084                         ret = -EFAULT;
2085                 break;
2086
2087         case TUNSETNOCSUM:
2088                 /* Disable/Enable checksum */
2089
2090                 /* [unimplemented] */
2091                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2092                           arg ? "disabled" : "enabled");
2093                 break;
2094
2095         case TUNSETPERSIST:
2096                 /* Disable/Enable persist mode. Keep an extra reference to the
2097                  * module to prevent the module being unprobed.
2098                  */
2099                 if (arg && !(tun->flags & IFF_PERSIST)) {
2100                         tun->flags |= IFF_PERSIST;
2101                         __module_get(THIS_MODULE);
2102                 }
2103                 if (!arg && (tun->flags & IFF_PERSIST)) {
2104                         tun->flags &= ~IFF_PERSIST;
2105                         module_put(THIS_MODULE);
2106                 }
2107
2108                 tun_debug(KERN_INFO, tun, "persist %s\n",
2109                           arg ? "enabled" : "disabled");
2110                 break;
2111
2112         case TUNSETOWNER:
2113                 /* Set owner of the device */
2114                 owner = make_kuid(current_user_ns(), arg);
2115                 if (!uid_valid(owner)) {
2116                         ret = -EINVAL;
2117                         break;
2118                 }
2119                 tun->owner = owner;
2120                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2121                           from_kuid(&init_user_ns, tun->owner));
2122                 break;
2123
2124         case TUNSETGROUP:
2125                 /* Set group of the device */
2126                 group = make_kgid(current_user_ns(), arg);
2127                 if (!gid_valid(group)) {
2128                         ret = -EINVAL;
2129                         break;
2130                 }
2131                 tun->group = group;
2132                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2133                           from_kgid(&init_user_ns, tun->group));
2134                 break;
2135
2136         case TUNSETLINK:
2137                 /* Only allow setting the type when the interface is down */
2138                 if (tun->dev->flags & IFF_UP) {
2139                         tun_debug(KERN_INFO, tun,
2140                                   "Linktype set failed because interface is up\n");
2141                         ret = -EBUSY;
2142                 } else {
2143                         tun->dev->type = (int) arg;
2144                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2145                                   tun->dev->type);
2146                         ret = 0;
2147                 }
2148                 break;
2149
2150 #ifdef TUN_DEBUG
2151         case TUNSETDEBUG:
2152                 tun->debug = arg;
2153                 break;
2154 #endif
2155         case TUNSETOFFLOAD:
2156                 ret = set_offload(tun, arg);
2157                 break;
2158
2159         case TUNSETTXFILTER:
2160                 /* Can be set only for TAPs */
2161                 ret = -EINVAL;
2162                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2163                         break;
2164                 ret = update_filter(&tun->txflt, (void __user *)arg);
2165                 break;
2166
2167         case SIOCGIFHWADDR:
2168                 /* Get hw address */
2169                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2170                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2171                 if (copy_to_user(argp, &ifr, ifreq_len))
2172                         ret = -EFAULT;
2173                 break;
2174
2175         case SIOCSIFHWADDR:
2176                 /* Set hw address */
2177                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2178                           ifr.ifr_hwaddr.sa_data);
2179
2180                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2181                 break;
2182
2183         case TUNGETSNDBUF:
2184                 sndbuf = tfile->socket.sk->sk_sndbuf;
2185                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2186                         ret = -EFAULT;
2187                 break;
2188
2189         case TUNSETSNDBUF:
2190                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2191                         ret = -EFAULT;
2192                         break;
2193                 }
2194
2195                 tun->sndbuf = sndbuf;
2196                 tun_set_sndbuf(tun);
2197                 break;
2198
2199         case TUNGETVNETHDRSZ:
2200                 vnet_hdr_sz = tun->vnet_hdr_sz;
2201                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2202                         ret = -EFAULT;
2203                 break;
2204
2205         case TUNSETVNETHDRSZ:
2206                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2207                         ret = -EFAULT;
2208                         break;
2209                 }
2210                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2211                         ret = -EINVAL;
2212                         break;
2213                 }
2214
2215                 tun->vnet_hdr_sz = vnet_hdr_sz;
2216                 break;
2217
2218         case TUNGETVNETLE:
2219                 le = !!(tun->flags & TUN_VNET_LE);
2220                 if (put_user(le, (int __user *)argp))
2221                         ret = -EFAULT;
2222                 break;
2223
2224         case TUNSETVNETLE:
2225                 if (get_user(le, (int __user *)argp)) {
2226                         ret = -EFAULT;
2227                         break;
2228                 }
2229                 if (le)
2230                         tun->flags |= TUN_VNET_LE;
2231                 else
2232                         tun->flags &= ~TUN_VNET_LE;
2233                 break;
2234
2235         case TUNGETVNETBE:
2236                 ret = tun_get_vnet_be(tun, argp);
2237                 break;
2238
2239         case TUNSETVNETBE:
2240                 ret = tun_set_vnet_be(tun, argp);
2241                 break;
2242
2243         case TUNATTACHFILTER:
2244                 /* Can be set only for TAPs */
2245                 ret = -EINVAL;
2246                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2247                         break;
2248                 ret = -EFAULT;
2249                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2250                         break;
2251
2252                 ret = tun_attach_filter(tun);
2253                 break;
2254
2255         case TUNDETACHFILTER:
2256                 /* Can be set only for TAPs */
2257                 ret = -EINVAL;
2258                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2259                         break;
2260                 ret = 0;
2261                 tun_detach_filter(tun, tun->numqueues);
2262                 break;
2263
2264         case TUNGETFILTER:
2265                 ret = -EINVAL;
2266                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2267                         break;
2268                 ret = -EFAULT;
2269                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2270                         break;
2271                 ret = 0;
2272                 break;
2273
2274         default:
2275                 ret = -EINVAL;
2276                 break;
2277         }
2278
2279 unlock:
2280         rtnl_unlock();
2281         if (tun)
2282                 tun_put(tun);
2283         return ret;
2284 }
2285
2286 static long tun_chr_ioctl(struct file *file,
2287                           unsigned int cmd, unsigned long arg)
2288 {
2289         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2290 }
2291
2292 #ifdef CONFIG_COMPAT
2293 static long tun_chr_compat_ioctl(struct file *file,
2294                          unsigned int cmd, unsigned long arg)
2295 {
2296         switch (cmd) {
2297         case TUNSETIFF:
2298         case TUNGETIFF:
2299         case TUNSETTXFILTER:
2300         case TUNGETSNDBUF:
2301         case TUNSETSNDBUF:
2302         case SIOCGIFHWADDR:
2303         case SIOCSIFHWADDR:
2304                 arg = (unsigned long)compat_ptr(arg);
2305                 break;
2306         default:
2307                 arg = (compat_ulong_t)arg;
2308                 break;
2309         }
2310
2311         /*
2312          * compat_ifreq is shorter than ifreq, so we must not access beyond
2313          * the end of that structure. All fields that are used in this
2314          * driver are compatible though, we don't need to convert the
2315          * contents.
2316          */
2317         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2318 }
2319 #endif /* CONFIG_COMPAT */
2320
2321 static int tun_chr_fasync(int fd, struct file *file, int on)
2322 {
2323         struct tun_file *tfile = file->private_data;
2324         int ret;
2325
2326         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2327                 goto out;
2328
2329         if (on) {
2330                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2331                 tfile->flags |= TUN_FASYNC;
2332         } else
2333                 tfile->flags &= ~TUN_FASYNC;
2334         ret = 0;
2335 out:
2336         return ret;
2337 }
2338
2339 static int tun_chr_open(struct inode *inode, struct file * file)
2340 {
2341         struct net *net = current->nsproxy->net_ns;
2342         struct tun_file *tfile;
2343
2344         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2345
2346         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2347                                             &tun_proto, 0);
2348         if (!tfile)
2349                 return -ENOMEM;
2350         RCU_INIT_POINTER(tfile->tun, NULL);
2351         tfile->flags = 0;
2352         tfile->ifindex = 0;
2353
2354         init_waitqueue_head(&tfile->wq.wait);
2355         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2356
2357         tfile->socket.file = file;
2358         tfile->socket.ops = &tun_socket_ops;
2359
2360         sock_init_data(&tfile->socket, &tfile->sk);
2361
2362         tfile->sk.sk_write_space = tun_sock_write_space;
2363         tfile->sk.sk_sndbuf = INT_MAX;
2364
2365         file->private_data = tfile;
2366         INIT_LIST_HEAD(&tfile->next);
2367
2368         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2369
2370         return 0;
2371 }
2372
2373 static int tun_chr_close(struct inode *inode, struct file *file)
2374 {
2375         struct tun_file *tfile = file->private_data;
2376
2377         tun_detach(tfile, true);
2378
2379         return 0;
2380 }
2381
2382 #ifdef CONFIG_PROC_FS
2383 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2384 {
2385         struct tun_struct *tun;
2386         struct ifreq ifr;
2387
2388         memset(&ifr, 0, sizeof(ifr));
2389
2390         rtnl_lock();
2391         tun = tun_get(f);
2392         if (tun)
2393                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2394         rtnl_unlock();
2395
2396         if (tun)
2397                 tun_put(tun);
2398
2399         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2400 }
2401 #endif
2402
2403 static const struct file_operations tun_fops = {
2404         .owner  = THIS_MODULE,
2405         .llseek = no_llseek,
2406         .read_iter  = tun_chr_read_iter,
2407         .write_iter = tun_chr_write_iter,
2408         .poll   = tun_chr_poll,
2409         .unlocked_ioctl = tun_chr_ioctl,
2410 #ifdef CONFIG_COMPAT
2411         .compat_ioctl = tun_chr_compat_ioctl,
2412 #endif
2413         .open   = tun_chr_open,
2414         .release = tun_chr_close,
2415         .fasync = tun_chr_fasync,
2416 #ifdef CONFIG_PROC_FS
2417         .show_fdinfo = tun_chr_show_fdinfo,
2418 #endif
2419 };
2420
2421 static struct miscdevice tun_miscdev = {
2422         .minor = TUN_MINOR,
2423         .name = "tun",
2424         .nodename = "net/tun",
2425         .fops = &tun_fops,
2426 };
2427
2428 /* ethtool interface */
2429
2430 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2431 {
2432         cmd->supported          = 0;
2433         cmd->advertising        = 0;
2434         ethtool_cmd_speed_set(cmd, SPEED_10);
2435         cmd->duplex             = DUPLEX_FULL;
2436         cmd->port               = PORT_TP;
2437         cmd->phy_address        = 0;
2438         cmd->transceiver        = XCVR_INTERNAL;
2439         cmd->autoneg            = AUTONEG_DISABLE;
2440         cmd->maxtxpkt           = 0;
2441         cmd->maxrxpkt           = 0;
2442         return 0;
2443 }
2444
2445 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2446 {
2447         struct tun_struct *tun = netdev_priv(dev);
2448
2449         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2450         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2451
2452         switch (tun->flags & TUN_TYPE_MASK) {
2453         case IFF_TUN:
2454                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2455                 break;
2456         case IFF_TAP:
2457                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2458                 break;
2459         }
2460 }
2461
2462 static u32 tun_get_msglevel(struct net_device *dev)
2463 {
2464 #ifdef TUN_DEBUG
2465         struct tun_struct *tun = netdev_priv(dev);
2466         return tun->debug;
2467 #else
2468         return -EOPNOTSUPP;
2469 #endif
2470 }
2471
2472 static void tun_set_msglevel(struct net_device *dev, u32 value)
2473 {
2474 #ifdef TUN_DEBUG
2475         struct tun_struct *tun = netdev_priv(dev);
2476         tun->debug = value;
2477 #endif
2478 }
2479
2480 static int tun_get_coalesce(struct net_device *dev,
2481                             struct ethtool_coalesce *ec)
2482 {
2483         struct tun_struct *tun = netdev_priv(dev);
2484
2485         ec->rx_max_coalesced_frames = tun->rx_batched;
2486
2487         return 0;
2488 }
2489
2490 static int tun_set_coalesce(struct net_device *dev,
2491                             struct ethtool_coalesce *ec)
2492 {
2493         struct tun_struct *tun = netdev_priv(dev);
2494
2495         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
2496                 tun->rx_batched = NAPI_POLL_WEIGHT;
2497         else
2498                 tun->rx_batched = ec->rx_max_coalesced_frames;
2499
2500         return 0;
2501 }
2502
2503 static const struct ethtool_ops tun_ethtool_ops = {
2504         .get_settings   = tun_get_settings,
2505         .get_drvinfo    = tun_get_drvinfo,
2506         .get_msglevel   = tun_get_msglevel,
2507         .set_msglevel   = tun_set_msglevel,
2508         .get_link       = ethtool_op_get_link,
2509         .get_ts_info    = ethtool_op_get_ts_info,
2510         .get_coalesce   = tun_get_coalesce,
2511         .set_coalesce   = tun_set_coalesce,
2512 };
2513
2514 static int tun_queue_resize(struct tun_struct *tun)
2515 {
2516         struct net_device *dev = tun->dev;
2517         struct tun_file *tfile;
2518         struct skb_array **arrays;
2519         int n = tun->numqueues + tun->numdisabled;
2520         int ret, i;
2521
2522         arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL);
2523         if (!arrays)
2524                 return -ENOMEM;
2525
2526         for (i = 0; i < tun->numqueues; i++) {
2527                 tfile = rtnl_dereference(tun->tfiles[i]);
2528                 arrays[i] = &tfile->tx_array;
2529         }
2530         list_for_each_entry(tfile, &tun->disabled, next)
2531                 arrays[i++] = &tfile->tx_array;
2532
2533         ret = skb_array_resize_multiple(arrays, n,
2534                                         dev->tx_queue_len, GFP_KERNEL);
2535
2536         kfree(arrays);
2537         return ret;
2538 }
2539
2540 static int tun_device_event(struct notifier_block *unused,
2541                             unsigned long event, void *ptr)
2542 {
2543         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2544         struct tun_struct *tun = netdev_priv(dev);
2545
2546         if (dev->rtnl_link_ops != &tun_link_ops)
2547                 return NOTIFY_DONE;
2548
2549         switch (event) {
2550         case NETDEV_CHANGE_TX_QUEUE_LEN:
2551                 if (tun_queue_resize(tun))
2552                         return NOTIFY_BAD;
2553                 break;
2554         default:
2555                 break;
2556         }
2557
2558         return NOTIFY_DONE;
2559 }
2560
2561 static struct notifier_block tun_notifier_block __read_mostly = {
2562         .notifier_call  = tun_device_event,
2563 };
2564
2565 static int __init tun_init(void)
2566 {
2567         int ret = 0;
2568
2569         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2570         pr_info("%s\n", DRV_COPYRIGHT);
2571
2572         ret = rtnl_link_register(&tun_link_ops);
2573         if (ret) {
2574                 pr_err("Can't register link_ops\n");
2575                 goto err_linkops;
2576         }
2577
2578         ret = misc_register(&tun_miscdev);
2579         if (ret) {
2580                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2581                 goto err_misc;
2582         }
2583
2584         register_netdevice_notifier(&tun_notifier_block);
2585         return  0;
2586 err_misc:
2587         rtnl_link_unregister(&tun_link_ops);
2588 err_linkops:
2589         return ret;
2590 }
2591
2592 static void tun_cleanup(void)
2593 {
2594         misc_deregister(&tun_miscdev);
2595         rtnl_link_unregister(&tun_link_ops);
2596         unregister_netdevice_notifier(&tun_notifier_block);
2597 }
2598
2599 /* Get an underlying socket object from tun file.  Returns error unless file is
2600  * attached to a device.  The returned object works like a packet socket, it
2601  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2602  * holding a reference to the file for as long as the socket is in use. */
2603 struct socket *tun_get_socket(struct file *file)
2604 {
2605         struct tun_file *tfile;
2606         if (file->f_op != &tun_fops)
2607                 return ERR_PTR(-EINVAL);
2608         tfile = file->private_data;
2609         if (!tfile)
2610                 return ERR_PTR(-EBADFD);
2611         return &tfile->socket;
2612 }
2613 EXPORT_SYMBOL_GPL(tun_get_socket);
2614
2615 module_init(tun_init);
2616 module_exit(tun_cleanup);
2617 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2618 MODULE_AUTHOR(DRV_COPYRIGHT);
2619 MODULE_LICENSE("GPL");
2620 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2621 MODULE_ALIAS("devname:net/tun");