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
  75 #include <asm/uaccess.h>
  76
  77 /* Uncomment to enable debugging */
  78 /* #define TUN_DEBUG 1 */
  79
  80 #ifdef TUN_DEBUG
  81 static int debug;
  82
  83 #define tun_debug(level, tun, fmt, args...)                     \
  84 do {                                                            \
  85         if (tun->debug)                                         \
  86                 netdev_printk(level, tun->dev, fmt, ##args);    \
  87 } while (0)
  88 #define DBG1(level, fmt, args...)                               \
  89 do {                                                            \
  90         if (debug == 2)                                         \
  91                 printk(level fmt, ##args);                      \
  92 } while (0)
  93 #else
  94 #define tun_debug(level, tun, fmt, args...)                     \
  95 do {                                                            \
  96         if (0)                                                  \
  97                 netdev_printk(level, tun->dev, fmt, ##args);    \
  98 } while (0)
  99 #define DBG1(level, fmt, args...)                               \
 100 do {                                                            \
 101         if (0)                                                  \
 102                 printk(level fmt, ##args);                      \
 103 } while (0)
 104 #endif
 105
 106 /* TUN device flags */
 107
 108 /* IFF_ATTACH_QUEUE is never stored in device flags,
 109  * overload it to mean fasync when stored there.
 110  */
 111 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 112 /* High bits in flags field are unused. */
 113 #define TUN_VNET_LE     0x80000000
 114 #define TUN_VNET_BE     0x40000000
 115
 116 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 117                       IFF_MULTI_QUEUE)
 118 #define GOODCOPY_LEN 128
 119
 120 #define FLT_EXACT_COUNT 8
 121 struct tap_filter {
 122         unsigned int    count;    /* Number of addrs. Zero means disabled */
 123         u32             mask[2];  /* Mask of the hashed addrs */
 124         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 125 };
 126
 127 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 128  * to max number of VCPUs in guest. */
 129 #define MAX_TAP_QUEUES 256
 130 #define MAX_TAP_FLOWS  4096
 131
 132 #define TUN_FLOW_EXPIRE (3 * HZ)
 133
 134 /* A tun_file connects an open character device to a tuntap netdevice. It
 135  * also contains all socket related structures (except sock_fprog and tap_filter)
 136  * to serve as one transmit queue for tuntap device. The sock_fprog and
 137  * tap_filter were kept in tun_struct since they were used for filtering for the
 138  * netdevice not for a specific queue (at least I didn't see the requirement for
 139  * this).
 140  *
 141  * RCU usage:
 142  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 143  * other can only be read while rcu_read_lock or rtnl_lock is held.
 144  */
 145 struct tun_file {
 146         struct sock sk;
 147         struct socket socket;
 148         struct socket_wq wq;
 149         struct tun_struct __rcu *tun;
 150         struct fasync_struct *fasync;
 151         /* only used for fasnyc */
 152         unsigned int flags;
 153         union {
 154                 u16 queue_index;
 155                 unsigned int ifindex;
 156         };
 157         struct list_head next;
 158         struct tun_struct *detached;
 159 };
 160
 161 struct tun_flow_entry {
 162         struct hlist_node hash_link;
 163         struct rcu_head rcu;
 164         struct tun_struct *tun;
 165
 166         u32 rxhash;
 167         u32 rps_rxhash;
 168         int queue_index;
 169         unsigned long updated;
 170 };
 171
 172 #define TUN_NUM_FLOW_ENTRIES 1024
 173
 174 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 175  * device, socket filter, sndbuf and vnet header size were restore when the
 176  * file were attached to a persist device.
 177  */
 178 struct tun_struct {
 179         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 180         unsigned int            numqueues;
 181         unsigned int            flags;
 182         kuid_t                  owner;
 183         kgid_t                  group;
 184
 185         struct net_device       *dev;
 186         netdev_features_t       set_features;
 187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 188                           NETIF_F_TSO6|NETIF_F_UFO)
 189
 190         int                     align;
 191         int                     vnet_hdr_sz;
 192         int                     sndbuf;
 193         struct tap_filter       txflt;
 194         struct sock_fprog       fprog;
 195         /* protected by rtnl lock */
 196         bool                    filter_attached;
 197 #ifdef TUN_DEBUG
 198         int debug;
 199 #endif
 200         spinlock_t lock;
 201         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 202         struct timer_list flow_gc_timer;
 203         unsigned long ageing_time;
 204         unsigned int numdisabled;
 205         struct list_head disabled;
 206         void *security;
 207         u32 flow_count;
 208 };
 209
 210 #ifdef CONFIG_TUN_VNET_CROSS_LE
 211 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 212 {
 213         return tun->flags & TUN_VNET_BE ? false :
 214                 virtio_legacy_is_little_endian();
 215 }
 216
 217 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 218 {
 219         int be = !!(tun->flags & TUN_VNET_BE);
 220
 221         if (put_user(be, argp))
 222                 return -EFAULT;
 223
 224         return 0;
 225 }
 226
 227 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 228 {
 229         int be;
 230
 231         if (get_user(be, argp))
 232                 return -EFAULT;
 233
 234         if (be)
 235                 tun->flags |= TUN_VNET_BE;
 236         else
 237                 tun->flags &= ~TUN_VNET_BE;
 238
 239         return 0;
 240 }
 241 #else
 242 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 243 {
 244         return virtio_legacy_is_little_endian();
 245 }
 246
 247 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 248 {
 249         return -EINVAL;
 250 }
 251
 252 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 253 {
 254         return -EINVAL;
 255 }
 256 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 257
 258 static inline bool tun_is_little_endian(struct tun_struct *tun)
 259 {
 260         return tun->flags & TUN_VNET_LE ||
 261                 tun_legacy_is_little_endian(tun);
 262 }
 263
 264 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 265 {
 266         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 267 }
 268
 269 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 270 {
 271         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 272 }
 273
 274 static inline u32 tun_hashfn(u32 rxhash)
 275 {
 276         return rxhash & 0x3ff;
 277 }
 278
 279 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 280 {
 281         struct tun_flow_entry *e;
 282
 283         hlist_for_each_entry_rcu(e, head, hash_link) {
 284                 if (e->rxhash == rxhash)
 285                         return e;
 286         }
 287         return NULL;
 288 }
 289
 290 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 291                                               struct hlist_head *head,
 292                                               u32 rxhash, u16 queue_index)
 293 {
 294         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 295
 296         if (e) {
 297                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 298                           rxhash, queue_index);
 299                 e->updated = jiffies;
 300                 e->rxhash = rxhash;
 301                 e->rps_rxhash = 0;
 302                 e->queue_index = queue_index;
 303                 e->tun = tun;
 304                 hlist_add_head_rcu(&e->hash_link, head);
 305                 ++tun->flow_count;
 306         }
 307         return e;
 308 }
 309
 310 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 311 {
 312         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 313                   e->rxhash, e->queue_index);
 314         hlist_del_rcu(&e->hash_link);
 315         kfree_rcu(e, rcu);
 316         --tun->flow_count;
 317 }
 318
 319 static void tun_flow_flush(struct tun_struct *tun)
 320 {
 321         int i;
 322
 323         spin_lock_bh(&tun->lock);
 324         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 325                 struct tun_flow_entry *e;
 326                 struct hlist_node *n;
 327
 328                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 329                         tun_flow_delete(tun, e);
 330         }
 331         spin_unlock_bh(&tun->lock);
 332 }
 333
 334 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 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                         if (e->queue_index == queue_index)
 345                                 tun_flow_delete(tun, e);
 346                 }
 347         }
 348         spin_unlock_bh(&tun->lock);
 349 }
 350
 351 static void tun_flow_cleanup(unsigned long data)
 352 {
 353         struct tun_struct *tun = (struct tun_struct *)data;
 354         unsigned long delay = tun->ageing_time;
 355         unsigned long next_timer = jiffies + delay;
 356         unsigned long count = 0;
 357         int i;
 358
 359         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 360
 361         spin_lock_bh(&tun->lock);
 362         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 363                 struct tun_flow_entry *e;
 364                 struct hlist_node *n;
 365
 366                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 367                         unsigned long this_timer;
 368                         count++;
 369                         this_timer = e->updated + delay;
 370                         if (time_before_eq(this_timer, jiffies))
 371                                 tun_flow_delete(tun, e);
 372                         else if (time_before(this_timer, next_timer))
 373                                 next_timer = this_timer;
 374                 }
 375         }
 376
 377         if (count)
 378                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 379         spin_unlock_bh(&tun->lock);
 380 }
 381
 382 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 383                             struct tun_file *tfile)
 384 {
 385         struct hlist_head *head;
 386         struct tun_flow_entry *e;
 387         unsigned long delay = tun->ageing_time;
 388         u16 queue_index = tfile->queue_index;
 389
 390         if (!rxhash)
 391                 return;
 392         else
 393                 head = &tun->flows[tun_hashfn(rxhash)];
 394
 395         rcu_read_lock();
 396
 397         /* We may get a very small possibility of OOO during switching, not
 398          * worth to optimize.*/
 399         if (tun->numqueues == 1 || tfile->detached)
 400                 goto unlock;
 401
 402         e = tun_flow_find(head, rxhash);
 403         if (likely(e)) {
 404                 /* TODO: keep queueing to old queue until it's empty? */
 405                 e->queue_index = queue_index;
 406                 e->updated = jiffies;
 407                 sock_rps_record_flow_hash(e->rps_rxhash);
 408         } else {
 409                 spin_lock_bh(&tun->lock);
 410                 if (!tun_flow_find(head, rxhash) &&
 411                     tun->flow_count < MAX_TAP_FLOWS)
 412                         tun_flow_create(tun, head, rxhash, queue_index);
 413
 414                 if (!timer_pending(&tun->flow_gc_timer))
 415                         mod_timer(&tun->flow_gc_timer,
 416                                   round_jiffies_up(jiffies + delay));
 417                 spin_unlock_bh(&tun->lock);
 418         }
 419
 420 unlock:
 421         rcu_read_unlock();
 422 }
 423
 424 /**
 425  * Save the hash received in the stack receive path and update the
 426  * flow_hash table accordingly.
 427  */
 428 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 429 {
 430         if (unlikely(e->rps_rxhash != hash))
 431                 e->rps_rxhash = hash;
 432 }
 433
 434 /* We try to identify a flow through its rxhash first. The reason that
 435  * we do not check rxq no. is because some cards(e.g 82599), chooses
 436  * the rxq based on the txq where the last packet of the flow comes. As
 437  * the userspace application move between processors, we may get a
 438  * different rxq no. here. If we could not get rxhash, then we would
 439  * hope the rxq no. may help here.
 440  */
 441 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 442                             void *accel_priv, select_queue_fallback_t fallback)
 443 {
 444         struct tun_struct *tun = netdev_priv(dev);
 445         struct tun_flow_entry *e;
 446         u32 txq = 0;
 447         u32 numqueues = 0;
 448
 449         rcu_read_lock();
 450         numqueues = ACCESS_ONCE(tun->numqueues);
 451
 452         txq = skb_get_hash(skb);
 453         if (txq) {
 454                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 455                 if (e) {
 456                         tun_flow_save_rps_rxhash(e, txq);
 457                         txq = e->queue_index;
 458                 } else
 459                         /* use multiply and shift instead of expensive divide */
 460                         txq = ((u64)txq * numqueues) >> 32;
 461         } else if (likely(skb_rx_queue_recorded(skb))) {
 462                 txq = skb_get_rx_queue(skb);
 463                 while (unlikely(txq >= numqueues))
 464                         txq -= numqueues;
 465         }
 466
 467         rcu_read_unlock();
 468         return txq;
 469 }
 470
 471 static inline bool tun_not_capable(struct tun_struct *tun)
 472 {
 473         const struct cred *cred = current_cred();
 474         struct net *net = dev_net(tun->dev);
 475
 476         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 477                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 478                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 479 }
 480
 481 static void tun_set_real_num_queues(struct tun_struct *tun)
 482 {
 483         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 484         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 485 }
 486
 487 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 488 {
 489         tfile->detached = tun;
 490         list_add_tail(&tfile->next, &tun->disabled);
 491         ++tun->numdisabled;
 492 }
 493
 494 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 495 {
 496         struct tun_struct *tun = tfile->detached;
 497
 498         tfile->detached = NULL;
 499         list_del_init(&tfile->next);
 500         --tun->numdisabled;
 501         return tun;
 502 }
 503
 504 static void tun_queue_purge(struct tun_file *tfile)
 505 {
 506         skb_queue_purge(&tfile->sk.sk_receive_queue);
 507         skb_queue_purge(&tfile->sk.sk_error_queue);
 508 }
 509
 510 static void __tun_detach(struct tun_file *tfile, bool clean)
 511 {
 512         struct tun_file *ntfile;
 513         struct tun_struct *tun;
 514
 515         tun = rtnl_dereference(tfile->tun);
 516
 517         if (tun && !tfile->detached) {
 518                 u16 index = tfile->queue_index;
 519                 BUG_ON(index >= tun->numqueues);
 520
 521                 rcu_assign_pointer(tun->tfiles[index],
 522                                    tun->tfiles[tun->numqueues - 1]);
 523                 ntfile = rtnl_dereference(tun->tfiles[index]);
 524                 ntfile->queue_index = index;
 525
 526                 --tun->numqueues;
 527                 if (clean) {
 528                         RCU_INIT_POINTER(tfile->tun, NULL);
 529                         sock_put(&tfile->sk);
 530                 } else
 531                         tun_disable_queue(tun, tfile);
 532
 533                 synchronize_net();
 534                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 535                 /* Drop read queue */
 536                 tun_queue_purge(tfile);
 537                 tun_set_real_num_queues(tun);
 538         } else if (tfile->detached && clean) {
 539                 tun = tun_enable_queue(tfile);
 540                 sock_put(&tfile->sk);
 541         }
 542
 543         if (clean) {
 544                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 545                         netif_carrier_off(tun->dev);
 546
 547                         if (!(tun->flags & IFF_PERSIST) &&
 548                             tun->dev->reg_state == NETREG_REGISTERED)
 549                                 unregister_netdevice(tun->dev);
 550                 }
 551                 sock_put(&tfile->sk);
 552         }
 553 }
 554
 555 static void tun_detach(struct tun_file *tfile, bool clean)
 556 {
 557         rtnl_lock();
 558         __tun_detach(tfile, clean);
 559         rtnl_unlock();
 560 }
 561
 562 static void tun_detach_all(struct net_device *dev)
 563 {
 564         struct tun_struct *tun = netdev_priv(dev);
 565         struct tun_file *tfile, *tmp;
 566         int i, n = tun->numqueues;
 567
 568         for (i = 0; i < n; i++) {
 569                 tfile = rtnl_dereference(tun->tfiles[i]);
 570                 BUG_ON(!tfile);
 571                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 572                 RCU_INIT_POINTER(tfile->tun, NULL);
 573                 --tun->numqueues;
 574         }
 575         list_for_each_entry(tfile, &tun->disabled, next) {
 576                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 577                 RCU_INIT_POINTER(tfile->tun, NULL);
 578         }
 579         BUG_ON(tun->numqueues != 0);
 580
 581         synchronize_net();
 582         for (i = 0; i < n; i++) {
 583                 tfile = rtnl_dereference(tun->tfiles[i]);
 584                 /* Drop read queue */
 585                 tun_queue_purge(tfile);
 586                 sock_put(&tfile->sk);
 587         }
 588         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 589                 tun_enable_queue(tfile);
 590                 tun_queue_purge(tfile);
 591                 sock_put(&tfile->sk);
 592         }
 593         BUG_ON(tun->numdisabled != 0);
 594
 595         if (tun->flags & IFF_PERSIST)
 596                 module_put(THIS_MODULE);
 597 }
 598
 599 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 600 {
 601         struct tun_file *tfile = file->private_data;
 602         int err;
 603
 604         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 605         if (err < 0)
 606                 goto out;
 607
 608         err = -EINVAL;
 609         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 610                 goto out;
 611
 612         err = -EBUSY;
 613         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 614                 goto out;
 615
 616         err = -E2BIG;
 617         if (!tfile->detached &&
 618             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 619                 goto out;
 620
 621         err = 0;
 622
 623         /* Re-attach the filter to persist device */
 624         if (!skip_filter && (tun->filter_attached == true)) {
 625                 err = __sk_attach_filter(&tun->fprog, tfile->socket.sk,
 626                                          lockdep_rtnl_is_held());
 627                 if (!err)
 628                         goto out;
 629         }
 630         tfile->queue_index = tun->numqueues;
 631         rcu_assign_pointer(tfile->tun, tun);
 632         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 633         tun->numqueues++;
 634
 635         if (tfile->detached)
 636                 tun_enable_queue(tfile);
 637         else
 638                 sock_hold(&tfile->sk);
 639
 640         tun_set_real_num_queues(tun);
 641
 642         /* device is allowed to go away first, so no need to hold extra
 643          * refcnt.
 644          */
 645
 646 out:
 647         return err;
 648 }
 649
 650 static struct tun_struct *__tun_get(struct tun_file *tfile)
 651 {
 652         struct tun_struct *tun;
 653
 654         rcu_read_lock();
 655         tun = rcu_dereference(tfile->tun);
 656         if (tun)
 657                 dev_hold(tun->dev);
 658         rcu_read_unlock();
 659
 660         return tun;
 661 }
 662
 663 static struct tun_struct *tun_get(struct file *file)
 664 {
 665         return __tun_get(file->private_data);
 666 }
 667
 668 static void tun_put(struct tun_struct *tun)
 669 {
 670         dev_put(tun->dev);
 671 }
 672
 673 /* TAP filtering */
 674 static void addr_hash_set(u32 *mask, const u8 *addr)
 675 {
 676         int n = ether_crc(ETH_ALEN, addr) >> 26;
 677         mask[n >> 5] |= (1 << (n & 31));
 678 }
 679
 680 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 681 {
 682         int n = ether_crc(ETH_ALEN, addr) >> 26;
 683         return mask[n >> 5] & (1 << (n & 31));
 684 }
 685
 686 static int update_filter(struct tap_filter *filter, void __user *arg)
 687 {
 688         struct { u8 u[ETH_ALEN]; } *addr;
 689         struct tun_filter uf;
 690         int err, alen, n, nexact;
 691
 692         if (copy_from_user(&uf, arg, sizeof(uf)))
 693                 return -EFAULT;
 694
 695         if (!uf.count) {
 696                 /* Disabled */
 697                 filter->count = 0;
 698                 return 0;
 699         }
 700
 701         alen = ETH_ALEN * uf.count;
 702         addr = kmalloc(alen, GFP_KERNEL);
 703         if (!addr)
 704                 return -ENOMEM;
 705
 706         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 707                 err = -EFAULT;
 708                 goto done;
 709         }
 710
 711         /* The filter is updated without holding any locks. Which is
 712          * perfectly safe. We disable it first and in the worst
 713          * case we'll accept a few undesired packets. */
 714         filter->count = 0;
 715         wmb();
 716
 717         /* Use first set of addresses as an exact filter */
 718         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 719                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 720
 721         nexact = n;
 722
 723         /* Remaining multicast addresses are hashed,
 724          * unicast will leave the filter disabled. */
 725         memset(filter->mask, 0, sizeof(filter->mask));
 726         for (; n < uf.count; n++) {
 727                 if (!is_multicast_ether_addr(addr[n].u)) {
 728                         err = 0; /* no filter */
 729                         goto done;
 730                 }
 731                 addr_hash_set(filter->mask, addr[n].u);
 732         }
 733
 734         /* For ALLMULTI just set the mask to all ones.
 735          * This overrides the mask populated above. */
 736         if ((uf.flags & TUN_FLT_ALLMULTI))
 737                 memset(filter->mask, ~0, sizeof(filter->mask));
 738
 739         /* Now enable the filter */
 740         wmb();
 741         filter->count = nexact;
 742
 743         /* Return the number of exact filters */
 744         err = nexact;
 745
 746 done:
 747         kfree(addr);
 748         return err;
 749 }
 750
 751 /* Returns: 0 - drop, !=0 - accept */
 752 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 753 {
 754         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 755          * at this point. */
 756         struct ethhdr *eh = (struct ethhdr *) skb->data;
 757         int i;
 758
 759         /* Exact match */
 760         for (i = 0; i < filter->count; i++)
 761                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 762                         return 1;
 763
 764         /* Inexact match (multicast only) */
 765         if (is_multicast_ether_addr(eh->h_dest))
 766                 return addr_hash_test(filter->mask, eh->h_dest);
 767
 768         return 0;
 769 }
 770
 771 /*
 772  * Checks whether the packet is accepted or not.
 773  * Returns: 0 - drop, !=0 - accept
 774  */
 775 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 776 {
 777         if (!filter->count)
 778                 return 1;
 779
 780         return run_filter(filter, skb);
 781 }
 782
 783 /* Network device part of the driver */
 784
 785 static const struct ethtool_ops tun_ethtool_ops;
 786
 787 /* Net device detach from fd. */
 788 static void tun_net_uninit(struct net_device *dev)
 789 {
 790         tun_detach_all(dev);
 791 }
 792
 793 /* Net device open. */
 794 static int tun_net_open(struct net_device *dev)
 795 {
 796         netif_tx_start_all_queues(dev);
 797         return 0;
 798 }
 799
 800 /* Net device close. */
 801 static int tun_net_close(struct net_device *dev)
 802 {
 803         netif_tx_stop_all_queues(dev);
 804         return 0;
 805 }
 806
 807 /* Net device start xmit */
 808 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 809 {
 810         struct tun_struct *tun = netdev_priv(dev);
 811         int txq = skb->queue_mapping;
 812         struct tun_file *tfile;
 813         u32 numqueues = 0;
 814
 815         rcu_read_lock();
 816         tfile = rcu_dereference(tun->tfiles[txq]);
 817         numqueues = ACCESS_ONCE(tun->numqueues);
 818
 819         /* Drop packet if interface is not attached */
 820         if (txq >= numqueues)
 821                 goto drop;
 822
 823         if (numqueues == 1) {
 824                 /* Select queue was not called for the skbuff, so we extract the
 825                  * RPS hash and save it into the flow_table here.
 826                  */
 827                 __u32 rxhash;
 828
 829                 rxhash = skb_get_hash(skb);
 830                 if (rxhash) {
 831                         struct tun_flow_entry *e;
 832                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 833                                         rxhash);
 834                         if (e)
 835                                 tun_flow_save_rps_rxhash(e, rxhash);
 836                 }
 837         }
 838
 839         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 840
 841         BUG_ON(!tfile);
 842
 843         /* Drop if the filter does not like it.
 844          * This is a noop if the filter is disabled.
 845          * Filter can be enabled only for the TAP devices. */
 846         if (!check_filter(&tun->txflt, skb))
 847                 goto drop;
 848
 849         if (tfile->socket.sk->sk_filter &&
 850             sk_filter(tfile->socket.sk, skb))
 851                 goto drop;
 852
 853         /* Limit the number of packets queued by dividing txq length with the
 854          * number of queues.
 855          */
 856         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
 857                           >= dev->tx_queue_len)
 858                 goto drop;
 859
 860         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 861                 goto drop;
 862
 863         if (skb->sk && sk_fullsock(skb->sk)) {
 864                 sock_tx_timestamp(skb->sk, skb->sk->sk_tsflags,
 865                                   &skb_shinfo(skb)->tx_flags);
 866                 sw_tx_timestamp(skb);
 867         }
 868
 869         /* Orphan the skb - required as we might hang on to it
 870          * for indefinite time.
 871          */
 872         skb_orphan(skb);
 873
 874         nf_reset(skb);
 875
 876         /* Enqueue packet */
 877         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 878
 879         /* Notify and wake up reader process */
 880         if (tfile->flags & TUN_FASYNC)
 881                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 882         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 883
 884         rcu_read_unlock();
 885         return NETDEV_TX_OK;
 886
 887 drop:
 888         dev->stats.tx_dropped++;
 889         skb_tx_error(skb);
 890         kfree_skb(skb);
 891         rcu_read_unlock();
 892         return NET_XMIT_DROP;
 893 }
 894
 895 static void tun_net_mclist(struct net_device *dev)
 896 {
 897         /*
 898          * This callback is supposed to deal with mc filter in
 899          * _rx_ path and has nothing to do with the _tx_ path.
 900          * In rx path we always accept everything userspace gives us.
 901          */
 902 }
 903
 904 #define MIN_MTU 68
 905 #define MAX_MTU 65535
 906
 907 static int
 908 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 909 {
 910         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 911                 return -EINVAL;
 912         dev->mtu = new_mtu;
 913         return 0;
 914 }
 915
 916 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 917         netdev_features_t features)
 918 {
 919         struct tun_struct *tun = netdev_priv(dev);
 920
 921         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 922 }
 923 #ifdef CONFIG_NET_POLL_CONTROLLER
 924 static void tun_poll_controller(struct net_device *dev)
 925 {
 926         /*
 927          * Tun only receives frames when:
 928          * 1) the char device endpoint gets data from user space
 929          * 2) the tun socket gets a sendmsg call from user space
 930          * Since both of those are synchronous operations, we are guaranteed
 931          * never to have pending data when we poll for it
 932          * so there is nothing to do here but return.
 933          * We need this though so netpoll recognizes us as an interface that
 934          * supports polling, which enables bridge devices in virt setups to
 935          * still use netconsole
 936          */
 937         return;
 938 }
 939 #endif
 940
 941 static void tun_set_headroom(struct net_device *dev, int new_hr)
 942 {
 943         struct tun_struct *tun = netdev_priv(dev);
 944
 945         if (new_hr < NET_SKB_PAD)
 946                 new_hr = NET_SKB_PAD;
 947
 948         tun->align = new_hr;
 949 }
 950
 951 static const struct net_device_ops tun_netdev_ops = {
 952         .ndo_uninit             = tun_net_uninit,
 953         .ndo_open               = tun_net_open,
 954         .ndo_stop               = tun_net_close,
 955         .ndo_start_xmit         = tun_net_xmit,
 956         .ndo_change_mtu         = tun_net_change_mtu,
 957         .ndo_fix_features       = tun_net_fix_features,
 958         .ndo_select_queue       = tun_select_queue,
 959 #ifdef CONFIG_NET_POLL_CONTROLLER
 960         .ndo_poll_controller    = tun_poll_controller,
 961 #endif
 962         .ndo_set_rx_headroom    = tun_set_headroom,
 963 };
 964
 965 static const struct net_device_ops tap_netdev_ops = {
 966         .ndo_uninit             = tun_net_uninit,
 967         .ndo_open               = tun_net_open,
 968         .ndo_stop               = tun_net_close,
 969         .ndo_start_xmit         = tun_net_xmit,
 970         .ndo_change_mtu         = tun_net_change_mtu,
 971         .ndo_fix_features       = tun_net_fix_features,
 972         .ndo_set_rx_mode        = tun_net_mclist,
 973         .ndo_set_mac_address    = eth_mac_addr,
 974         .ndo_validate_addr      = eth_validate_addr,
 975         .ndo_select_queue       = tun_select_queue,
 976 #ifdef CONFIG_NET_POLL_CONTROLLER
 977         .ndo_poll_controller    = tun_poll_controller,
 978 #endif
 979         .ndo_features_check     = passthru_features_check,
 980         .ndo_set_rx_headroom    = tun_set_headroom,
 981 };
 982
 983 static void tun_flow_init(struct tun_struct *tun)
 984 {
 985         int i;
 986
 987         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 988                 INIT_HLIST_HEAD(&tun->flows[i]);
 989
 990         tun->ageing_time = TUN_FLOW_EXPIRE;
 991         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 992         mod_timer(&tun->flow_gc_timer,
 993                   round_jiffies_up(jiffies + tun->ageing_time));
 994 }
 995
 996 static void tun_flow_uninit(struct tun_struct *tun)
 997 {
 998         del_timer_sync(&tun->flow_gc_timer);
 999         tun_flow_flush(tun);
1000 }
1001
1002 /* Initialize net device. */
1003 static void tun_net_init(struct net_device *dev)
1004 {
1005         struct tun_struct *tun = netdev_priv(dev);
1006
1007         switch (tun->flags & TUN_TYPE_MASK) {
1008         case IFF_TUN:
1009                 dev->netdev_ops = &tun_netdev_ops;
1010
1011                 /* Point-to-Point TUN Device */
1012                 dev->hard_header_len = 0;
1013                 dev->addr_len = 0;
1014                 dev->mtu = 1500;
1015
1016                 /* Zero header length */
1017                 dev->type = ARPHRD_NONE;
1018                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1019                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
1020                 break;
1021
1022         case IFF_TAP:
1023                 dev->netdev_ops = &tap_netdev_ops;
1024                 /* Ethernet TAP Device */
1025                 ether_setup(dev);
1026                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1027                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1028
1029                 eth_hw_addr_random(dev);
1030
1031                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
1032                 break;
1033         }
1034 }
1035
1036 /* Character device part */
1037
1038 /* Poll */
1039 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1040 {
1041         struct tun_file *tfile = file->private_data;
1042         struct tun_struct *tun = __tun_get(tfile);
1043         struct sock *sk;
1044         unsigned int mask = 0;
1045
1046         if (!tun)
1047                 return POLLERR;
1048
1049         sk = tfile->socket.sk;
1050
1051         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1052
1053         poll_wait(file, sk_sleep(sk), wait);
1054
1055         if (!skb_queue_empty(&sk->sk_receive_queue))
1056                 mask |= POLLIN | POLLRDNORM;
1057
1058         if (sock_writeable(sk) ||
1059             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1060              sock_writeable(sk)))
1061                 mask |= POLLOUT | POLLWRNORM;
1062
1063         if (tun->dev->reg_state != NETREG_REGISTERED)
1064                 mask = POLLERR;
1065
1066         tun_put(tun);
1067         return mask;
1068 }
1069
1070 /* prepad is the amount to reserve at front.  len is length after that.
1071  * linear is a hint as to how much to copy (usually headers). */
1072 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1073                                      size_t prepad, size_t len,
1074                                      size_t linear, int noblock)
1075 {
1076         struct sock *sk = tfile->socket.sk;
1077         struct sk_buff *skb;
1078         int err;
1079
1080         /* Under a page?  Don't bother with paged skb. */
1081         if (prepad + len < PAGE_SIZE || !linear)
1082                 linear = len;
1083
1084         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1085                                    &err, 0);
1086         if (!skb)
1087                 return ERR_PTR(err);
1088
1089         skb_reserve(skb, prepad);
1090         skb_put(skb, linear);
1091         skb->data_len = len - linear;
1092         skb->len += len - linear;
1093
1094         return skb;
1095 }
1096
1097 /* Get packet from user space buffer */
1098 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1099                             void *msg_control, struct iov_iter *from,
1100                             int noblock)
1101 {
1102         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1103         struct sk_buff *skb;
1104         size_t total_len = iov_iter_count(from);
1105         size_t len = total_len, align = tun->align, linear;
1106         struct virtio_net_hdr gso = { 0 };
1107         int good_linear;
1108         int copylen;
1109         bool zerocopy = false;
1110         int err;
1111         u32 rxhash;
1112         ssize_t n;
1113
1114         if (!(tun->dev->flags & IFF_UP))
1115                 return -EIO;
1116
1117         if (!(tun->flags & IFF_NO_PI)) {
1118                 if (len < sizeof(pi))
1119                         return -EINVAL;
1120                 len -= sizeof(pi);
1121
1122                 n = copy_from_iter(&pi, sizeof(pi), from);
1123                 if (n != sizeof(pi))
1124                         return -EFAULT;
1125         }
1126
1127         if (tun->flags & IFF_VNET_HDR) {
1128                 if (len < tun->vnet_hdr_sz)
1129                         return -EINVAL;
1130                 len -= tun->vnet_hdr_sz;
1131
1132                 n = copy_from_iter(&gso, sizeof(gso), from);
1133                 if (n != sizeof(gso))
1134                         return -EFAULT;
1135
1136                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1137                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1138                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1139
1140                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1141                         return -EINVAL;
1142                 iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1143         }
1144
1145         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1146                 align += NET_IP_ALIGN;
1147                 if (unlikely(len < ETH_HLEN ||
1148                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1149                         return -EINVAL;
1150         }
1151
1152         good_linear = SKB_MAX_HEAD(align);
1153
1154         if (msg_control) {
1155                 struct iov_iter i = *from;
1156
1157                 /* There are 256 bytes to be copied in skb, so there is
1158                  * enough room for skb expand head in case it is used.
1159                  * The rest of the buffer is mapped from userspace.
1160                  */
1161                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1162                 if (copylen > good_linear)
1163                         copylen = good_linear;
1164                 linear = copylen;
1165                 iov_iter_advance(&i, copylen);
1166                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1167                         zerocopy = true;
1168         }
1169
1170         if (!zerocopy) {
1171                 copylen = len;
1172                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1173                         linear = good_linear;
1174                 else
1175                         linear = tun16_to_cpu(tun, gso.hdr_len);
1176         }
1177
1178         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1179         if (IS_ERR(skb)) {
1180                 if (PTR_ERR(skb) != -EAGAIN)
1181                         tun->dev->stats.rx_dropped++;
1182                 return PTR_ERR(skb);
1183         }
1184
1185         if (zerocopy)
1186                 err = zerocopy_sg_from_iter(skb, from);
1187         else {
1188                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1189                 if (!err && msg_control) {
1190                         struct ubuf_info *uarg = msg_control;
1191                         uarg->callback(uarg, false);
1192                 }
1193         }
1194
1195         if (err) {
1196                 tun->dev->stats.rx_dropped++;
1197                 kfree_skb(skb);
1198                 return -EFAULT;
1199         }
1200
1201         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1202                 if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start),
1203                                           tun16_to_cpu(tun, gso.csum_offset))) {
1204                         tun->dev->stats.rx_frame_errors++;
1205                         kfree_skb(skb);
1206                         return -EINVAL;
1207                 }
1208         }
1209
1210         switch (tun->flags & TUN_TYPE_MASK) {
1211         case IFF_TUN:
1212                 if (tun->flags & IFF_NO_PI) {
1213                         switch (skb->data[0] & 0xf0) {
1214                         case 0x40:
1215                                 pi.proto = htons(ETH_P_IP);
1216                                 break;
1217                         case 0x60:
1218                                 pi.proto = htons(ETH_P_IPV6);
1219                                 break;
1220                         default:
1221                                 tun->dev->stats.rx_dropped++;
1222                                 kfree_skb(skb);
1223                                 return -EINVAL;
1224                         }
1225                 }
1226
1227                 skb_reset_mac_header(skb);
1228                 skb->protocol = pi.proto;
1229                 skb->dev = tun->dev;
1230                 break;
1231         case IFF_TAP:
1232                 skb->protocol = eth_type_trans(skb, tun->dev);
1233                 break;
1234         }
1235
1236         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1237                 pr_debug("GSO!\n");
1238                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1239                 case VIRTIO_NET_HDR_GSO_TCPV4:
1240                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1241                         break;
1242                 case VIRTIO_NET_HDR_GSO_TCPV6:
1243                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1244                         break;
1245                 case VIRTIO_NET_HDR_GSO_UDP:
1246                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1247                         break;
1248                 default:
1249                         tun->dev->stats.rx_frame_errors++;
1250                         kfree_skb(skb);
1251                         return -EINVAL;
1252                 }
1253
1254                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1255                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1256
1257                 skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size);
1258                 if (skb_shinfo(skb)->gso_size == 0) {
1259                         tun->dev->stats.rx_frame_errors++;
1260                         kfree_skb(skb);
1261                         return -EINVAL;
1262                 }
1263
1264                 /* Header must be checked, and gso_segs computed. */
1265                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1266                 skb_shinfo(skb)->gso_segs = 0;
1267         }
1268
1269         /* copy skb_ubuf_info for callback when skb has no error */
1270         if (zerocopy) {
1271                 skb_shinfo(skb)->destructor_arg = msg_control;
1272                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1273                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1274         }
1275
1276         skb_reset_network_header(skb);
1277         skb_probe_transport_header(skb, 0);
1278
1279         rxhash = skb_get_hash(skb);
1280         netif_rx_ni(skb);
1281
1282         tun->dev->stats.rx_packets++;
1283         tun->dev->stats.rx_bytes += len;
1284
1285         tun_flow_update(tun, rxhash, tfile);
1286         return total_len;
1287 }
1288
1289 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1290 {
1291         struct file *file = iocb->ki_filp;
1292         struct tun_struct *tun = tun_get(file);
1293         struct tun_file *tfile = file->private_data;
1294         ssize_t result;
1295
1296         if (!tun)
1297                 return -EBADFD;
1298
1299         result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1300
1301         tun_put(tun);
1302         return result;
1303 }
1304
1305 /* Put packet to the user space buffer */
1306 static ssize_t tun_put_user(struct tun_struct *tun,
1307                             struct tun_file *tfile,
1308                             struct sk_buff *skb,
1309                             struct iov_iter *iter)
1310 {
1311         struct tun_pi pi = { 0, skb->protocol };
1312         ssize_t total;
1313         int vlan_offset = 0;
1314         int vlan_hlen = 0;
1315         int vnet_hdr_sz = 0;
1316
1317         if (skb_vlan_tag_present(skb))
1318                 vlan_hlen = VLAN_HLEN;
1319
1320         if (tun->flags & IFF_VNET_HDR)
1321                 vnet_hdr_sz = tun->vnet_hdr_sz;
1322
1323         total = skb->len + vlan_hlen + vnet_hdr_sz;
1324
1325         if (!(tun->flags & IFF_NO_PI)) {
1326                 if (iov_iter_count(iter) < sizeof(pi))
1327                         return -EINVAL;
1328
1329                 total += sizeof(pi);
1330                 if (iov_iter_count(iter) < total) {
1331                         /* Packet will be striped */
1332                         pi.flags |= TUN_PKT_STRIP;
1333                 }
1334
1335                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1336                         return -EFAULT;
1337         }
1338
1339         if (vnet_hdr_sz) {
1340                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1341                 if (iov_iter_count(iter) < vnet_hdr_sz)
1342                         return -EINVAL;
1343
1344                 if (skb_is_gso(skb)) {
1345                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1346
1347                         /* This is a hint as to how much should be linear. */
1348                         gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb));
1349                         gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size);
1350                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1351                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1352                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1353                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1354                         else if (sinfo->gso_type & SKB_GSO_UDP)
1355                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1356                         else {
1357                                 pr_err("unexpected GSO type: "
1358                                        "0x%x, gso_size %d, hdr_len %d\n",
1359                                        sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1360                                        tun16_to_cpu(tun, gso.hdr_len));
1361                                 print_hex_dump(KERN_ERR, "tun: ",
1362                                                DUMP_PREFIX_NONE,
1363                                                16, 1, skb->head,
1364                                                min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1365                                 WARN_ON_ONCE(1);
1366                                 return -EINVAL;
1367                         }
1368                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1369                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1370                 } else
1371                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1372
1373                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1374                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1375                         gso.csum_start = cpu_to_tun16(tun, skb_checksum_start_offset(skb) +
1376                                                       vlan_hlen);
1377                         gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset);
1378                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1379                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1380                 } /* else everything is zero */
1381
1382                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1383                         return -EFAULT;
1384
1385                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1386         }
1387
1388         if (vlan_hlen) {
1389                 int ret;
1390                 struct {
1391                         __be16 h_vlan_proto;
1392                         __be16 h_vlan_TCI;
1393                 } veth;
1394
1395                 veth.h_vlan_proto = skb->vlan_proto;
1396                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1397
1398                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1399
1400                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1401                 if (ret || !iov_iter_count(iter))
1402                         goto done;
1403
1404                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1405                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1406                         goto done;
1407         }
1408
1409         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1410
1411 done:
1412         tun->dev->stats.tx_packets++;
1413         tun->dev->stats.tx_bytes += skb->len + vlan_hlen;
1414
1415         return total;
1416 }
1417
1418 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1419                            struct iov_iter *to,
1420                            int noblock)
1421 {
1422         struct sk_buff *skb;
1423         ssize_t ret;
1424         int peeked, err, off = 0;
1425
1426         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1427
1428         if (!iov_iter_count(to))
1429                 return 0;
1430
1431         if (tun->dev->reg_state != NETREG_REGISTERED)
1432                 return -EIO;
1433
1434         /* Read frames from queue */
1435         skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1436                                   &peeked, &off, &err);
1437         if (!skb)
1438                 return err;
1439
1440         ret = tun_put_user(tun, tfile, skb, to);
1441         if (unlikely(ret < 0))
1442                 kfree_skb(skb);
1443         else
1444                 consume_skb(skb);
1445
1446         return ret;
1447 }
1448
1449 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1450 {
1451         struct file *file = iocb->ki_filp;
1452         struct tun_file *tfile = file->private_data;
1453         struct tun_struct *tun = __tun_get(tfile);
1454         ssize_t len = iov_iter_count(to), ret;
1455
1456         if (!tun)
1457                 return -EBADFD;
1458         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1459         ret = min_t(ssize_t, ret, len);
1460         if (ret > 0)
1461                 iocb->ki_pos = ret;
1462         tun_put(tun);
1463         return ret;
1464 }
1465
1466 static void tun_free_netdev(struct net_device *dev)
1467 {
1468         struct tun_struct *tun = netdev_priv(dev);
1469
1470         BUG_ON(!(list_empty(&tun->disabled)));
1471         tun_flow_uninit(tun);
1472         security_tun_dev_free_security(tun->security);
1473         free_netdev(dev);
1474 }
1475
1476 static void tun_setup(struct net_device *dev)
1477 {
1478         struct tun_struct *tun = netdev_priv(dev);
1479
1480         tun->owner = INVALID_UID;
1481         tun->group = INVALID_GID;
1482
1483         dev->ethtool_ops = &tun_ethtool_ops;
1484         dev->destructor = tun_free_netdev;
1485 }
1486
1487 /* Trivial set of netlink ops to allow deleting tun or tap
1488  * device with netlink.
1489  */
1490 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1491 {
1492         return -EINVAL;
1493 }
1494
1495 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1496         .kind           = DRV_NAME,
1497         .priv_size      = sizeof(struct tun_struct),
1498         .setup          = tun_setup,
1499         .validate       = tun_validate,
1500 };
1501
1502 static void tun_sock_write_space(struct sock *sk)
1503 {
1504         struct tun_file *tfile;
1505         wait_queue_head_t *wqueue;
1506
1507         if (!sock_writeable(sk))
1508                 return;
1509
1510         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1511                 return;
1512
1513         wqueue = sk_sleep(sk);
1514         if (wqueue && waitqueue_active(wqueue))
1515                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1516                                                 POLLWRNORM | POLLWRBAND);
1517
1518         tfile = container_of(sk, struct tun_file, sk);
1519         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1520 }
1521
1522 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1523 {
1524         int ret;
1525         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1526         struct tun_struct *tun = __tun_get(tfile);
1527
1528         if (!tun)
1529                 return -EBADFD;
1530
1531         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1532                            m->msg_flags & MSG_DONTWAIT);
1533         tun_put(tun);
1534         return ret;
1535 }
1536
1537 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1538                        int flags)
1539 {
1540         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1541         struct tun_struct *tun = __tun_get(tfile);
1542         int ret;
1543
1544         if (!tun)
1545                 return -EBADFD;
1546
1547         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1548                 ret = -EINVAL;
1549                 goto out;
1550         }
1551         if (flags & MSG_ERRQUEUE) {
1552                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1553                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1554                 goto out;
1555         }
1556         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1557         if (ret > (ssize_t)total_len) {
1558                 m->msg_flags |= MSG_TRUNC;
1559                 ret = flags & MSG_TRUNC ? ret : total_len;
1560         }
1561 out:
1562         tun_put(tun);
1563         return ret;
1564 }
1565
1566 /* Ops structure to mimic raw sockets with tun */
1567 static const struct proto_ops tun_socket_ops = {
1568         .sendmsg = tun_sendmsg,
1569         .recvmsg = tun_recvmsg,
1570 };
1571
1572 static struct proto tun_proto = {
1573         .name           = "tun",
1574         .owner          = THIS_MODULE,
1575         .obj_size       = sizeof(struct tun_file),
1576 };
1577
1578 static int tun_flags(struct tun_struct *tun)
1579 {
1580         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1581 }
1582
1583 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1584                               char *buf)
1585 {
1586         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1587         return sprintf(buf, "0x%x\n", tun_flags(tun));
1588 }
1589
1590 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1591                               char *buf)
1592 {
1593         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1594         return uid_valid(tun->owner)?
1595                 sprintf(buf, "%u\n",
1596                         from_kuid_munged(current_user_ns(), tun->owner)):
1597                 sprintf(buf, "-1\n");
1598 }
1599
1600 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1601                               char *buf)
1602 {
1603         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1604         return gid_valid(tun->group) ?
1605                 sprintf(buf, "%u\n",
1606                         from_kgid_munged(current_user_ns(), tun->group)):
1607                 sprintf(buf, "-1\n");
1608 }
1609
1610 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1611 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1612 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1613
1614 static struct attribute *tun_dev_attrs[] = {
1615         &dev_attr_tun_flags.attr,
1616         &dev_attr_owner.attr,
1617         &dev_attr_group.attr,
1618         NULL
1619 };
1620
1621 static const struct attribute_group tun_attr_group = {
1622         .attrs = tun_dev_attrs
1623 };
1624
1625 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1626 {
1627         struct tun_struct *tun;
1628         struct tun_file *tfile = file->private_data;
1629         struct net_device *dev;
1630         int err;
1631
1632         if (tfile->detached)
1633                 return -EINVAL;
1634
1635         dev = __dev_get_by_name(net, ifr->ifr_name);
1636         if (dev) {
1637                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1638                         return -EBUSY;
1639                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1640                         tun = netdev_priv(dev);
1641                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1642                         tun = netdev_priv(dev);
1643                 else
1644                         return -EINVAL;
1645
1646                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1647                     !!(tun->flags & IFF_MULTI_QUEUE))
1648                         return -EINVAL;
1649
1650                 if (tun_not_capable(tun))
1651                         return -EPERM;
1652                 err = security_tun_dev_open(tun->security);
1653                 if (err < 0)
1654                         return err;
1655
1656                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1657                 if (err < 0)
1658                         return err;
1659
1660                 if (tun->flags & IFF_MULTI_QUEUE &&
1661                     (tun->numqueues + tun->numdisabled > 1)) {
1662                         /* One or more queue has already been attached, no need
1663                          * to initialize the device again.
1664                          */
1665                         return 0;
1666                 }
1667         }
1668         else {
1669                 char *name;
1670                 unsigned long flags = 0;
1671                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1672                              MAX_TAP_QUEUES : 1;
1673
1674                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1675                         return -EPERM;
1676                 err = security_tun_dev_create();
1677                 if (err < 0)
1678                         return err;
1679
1680                 /* Set dev type */
1681                 if (ifr->ifr_flags & IFF_TUN) {
1682                         /* TUN device */
1683                         flags |= IFF_TUN;
1684                         name = "tun%d";
1685                 } else if (ifr->ifr_flags & IFF_TAP) {
1686                         /* TAP device */
1687                         flags |= IFF_TAP;
1688                         name = "tap%d";
1689                 } else
1690                         return -EINVAL;
1691
1692                 if (*ifr->ifr_name)
1693                         name = ifr->ifr_name;
1694
1695                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1696                                        NET_NAME_UNKNOWN, tun_setup, queues,
1697                                        queues);
1698
1699                 if (!dev)
1700                         return -ENOMEM;
1701
1702                 dev_net_set(dev, net);
1703                 dev->rtnl_link_ops = &tun_link_ops;
1704                 dev->ifindex = tfile->ifindex;
1705                 dev->sysfs_groups[0] = &tun_attr_group;
1706
1707                 tun = netdev_priv(dev);
1708                 tun->dev = dev;
1709                 tun->flags = flags;
1710                 tun->txflt.count = 0;
1711                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1712
1713                 tun->align = NET_SKB_PAD;
1714                 tun->filter_attached = false;
1715                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1716
1717                 spin_lock_init(&tun->lock);
1718
1719                 err = security_tun_dev_alloc_security(&tun->security);
1720                 if (err < 0)
1721                         goto err_free_dev;
1722
1723                 tun_net_init(dev);
1724                 tun_flow_init(tun);
1725
1726                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1727                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1728                                    NETIF_F_HW_VLAN_STAG_TX;
1729                 dev->features = dev->hw_features;
1730                 dev->vlan_features = dev->features &
1731                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1732                                        NETIF_F_HW_VLAN_STAG_TX);
1733
1734                 INIT_LIST_HEAD(&tun->disabled);
1735                 err = tun_attach(tun, file, false);
1736                 if (err < 0)
1737                         goto err_free_flow;
1738
1739                 err = register_netdevice(tun->dev);
1740                 if (err < 0)
1741                         goto err_detach;
1742         }
1743
1744         netif_carrier_on(tun->dev);
1745
1746         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1747
1748         tun->flags = (tun->flags & ~TUN_FEATURES) |
1749                 (ifr->ifr_flags & TUN_FEATURES);
1750
1751         /* Make sure persistent devices do not get stuck in
1752          * xoff state.
1753          */
1754         if (netif_running(tun->dev))
1755                 netif_tx_wake_all_queues(tun->dev);
1756
1757         strcpy(ifr->ifr_name, tun->dev->name);
1758         return 0;
1759
1760 err_detach:
1761         tun_detach_all(dev);
1762 err_free_flow:
1763         tun_flow_uninit(tun);
1764         security_tun_dev_free_security(tun->security);
1765 err_free_dev:
1766         free_netdev(dev);
1767         return err;
1768 }
1769
1770 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1771                        struct ifreq *ifr)
1772 {
1773         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1774
1775         strcpy(ifr->ifr_name, tun->dev->name);
1776
1777         ifr->ifr_flags = tun_flags(tun);
1778
1779 }
1780
1781 /* This is like a cut-down ethtool ops, except done via tun fd so no
1782  * privs required. */
1783 static int set_offload(struct tun_struct *tun, unsigned long arg)
1784 {
1785         netdev_features_t features = 0;
1786
1787         if (arg & TUN_F_CSUM) {
1788                 features |= NETIF_F_HW_CSUM;
1789                 arg &= ~TUN_F_CSUM;
1790
1791                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1792                         if (arg & TUN_F_TSO_ECN) {
1793                                 features |= NETIF_F_TSO_ECN;
1794                                 arg &= ~TUN_F_TSO_ECN;
1795                         }
1796                         if (arg & TUN_F_TSO4)
1797                                 features |= NETIF_F_TSO;
1798                         if (arg & TUN_F_TSO6)
1799                                 features |= NETIF_F_TSO6;
1800                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1801                 }
1802
1803                 if (arg & TUN_F_UFO) {
1804                         features |= NETIF_F_UFO;
1805                         arg &= ~TUN_F_UFO;
1806                 }
1807         }
1808
1809         /* This gives the user a way to test for new features in future by
1810          * trying to set them. */
1811         if (arg)
1812                 return -EINVAL;
1813
1814         tun->set_features = features;
1815         netdev_update_features(tun->dev);
1816
1817         return 0;
1818 }
1819
1820 static void tun_detach_filter(struct tun_struct *tun, int n)
1821 {
1822         int i;
1823         struct tun_file *tfile;
1824
1825         for (i = 0; i < n; i++) {
1826                 tfile = rtnl_dereference(tun->tfiles[i]);
1827                 __sk_detach_filter(tfile->socket.sk, lockdep_rtnl_is_held());
1828         }
1829
1830         tun->filter_attached = false;
1831 }
1832
1833 static int tun_attach_filter(struct tun_struct *tun)
1834 {
1835         int i, ret = 0;
1836         struct tun_file *tfile;
1837
1838         for (i = 0; i < tun->numqueues; i++) {
1839                 tfile = rtnl_dereference(tun->tfiles[i]);
1840                 ret = __sk_attach_filter(&tun->fprog, tfile->socket.sk,
1841                                          lockdep_rtnl_is_held());
1842                 if (ret) {
1843                         tun_detach_filter(tun, i);
1844                         return ret;
1845                 }
1846         }
1847
1848         tun->filter_attached = true;
1849         return ret;
1850 }
1851
1852 static void tun_set_sndbuf(struct tun_struct *tun)
1853 {
1854         struct tun_file *tfile;
1855         int i;
1856
1857         for (i = 0; i < tun->numqueues; i++) {
1858                 tfile = rtnl_dereference(tun->tfiles[i]);
1859                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1860         }
1861 }
1862
1863 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1864 {
1865         struct tun_file *tfile = file->private_data;
1866         struct tun_struct *tun;
1867         int ret = 0;
1868
1869         rtnl_lock();
1870
1871         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1872                 tun = tfile->detached;
1873                 if (!tun) {
1874                         ret = -EINVAL;
1875                         goto unlock;
1876                 }
1877                 ret = security_tun_dev_attach_queue(tun->security);
1878                 if (ret < 0)
1879                         goto unlock;
1880                 ret = tun_attach(tun, file, false);
1881         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1882                 tun = rtnl_dereference(tfile->tun);
1883                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1884                         ret = -EINVAL;
1885                 else
1886                         __tun_detach(tfile, false);
1887         } else
1888                 ret = -EINVAL;
1889
1890 unlock:
1891         rtnl_unlock();
1892         return ret;
1893 }
1894
1895 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1896                             unsigned long arg, int ifreq_len)
1897 {
1898         struct tun_file *tfile = file->private_data;
1899         struct tun_struct *tun;
1900         void __user* argp = (void __user*)arg;
1901         struct ifreq ifr;
1902         kuid_t owner;
1903         kgid_t group;
1904         int sndbuf;
1905         int vnet_hdr_sz;
1906         unsigned int ifindex;
1907         int le;
1908         int ret;
1909
1910         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1911                 if (copy_from_user(&ifr, argp, ifreq_len))
1912                         return -EFAULT;
1913         } else {
1914                 memset(&ifr, 0, sizeof(ifr));
1915         }
1916         if (cmd == TUNGETFEATURES) {
1917                 /* Currently this just means: "what IFF flags are valid?".
1918                  * This is needed because we never checked for invalid flags on
1919                  * TUNSETIFF.
1920                  */
1921                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1922                                 (unsigned int __user*)argp);
1923         } else if (cmd == TUNSETQUEUE)
1924                 return tun_set_queue(file, &ifr);
1925
1926         ret = 0;
1927         rtnl_lock();
1928
1929         tun = __tun_get(tfile);
1930         if (cmd == TUNSETIFF && !tun) {
1931                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1932
1933                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
1934
1935                 if (ret)
1936                         goto unlock;
1937
1938                 if (copy_to_user(argp, &ifr, ifreq_len))
1939                         ret = -EFAULT;
1940                 goto unlock;
1941         }
1942         if (cmd == TUNSETIFINDEX) {
1943                 ret = -EPERM;
1944                 if (tun)
1945                         goto unlock;
1946
1947                 ret = -EFAULT;
1948                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1949                         goto unlock;
1950
1951                 ret = 0;
1952                 tfile->ifindex = ifindex;
1953                 goto unlock;
1954         }
1955
1956         ret = -EBADFD;
1957         if (!tun)
1958                 goto unlock;
1959
1960         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1961
1962         ret = 0;
1963         switch (cmd) {
1964         case TUNGETIFF:
1965                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1966
1967                 if (tfile->detached)
1968                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
1969                 if (!tfile->socket.sk->sk_filter)
1970                         ifr.ifr_flags |= IFF_NOFILTER;
1971
1972                 if (copy_to_user(argp, &ifr, ifreq_len))
1973                         ret = -EFAULT;
1974                 break;
1975
1976         case TUNSETNOCSUM:
1977                 /* Disable/Enable checksum */
1978
1979                 /* [unimplemented] */
1980                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1981                           arg ? "disabled" : "enabled");
1982                 break;
1983
1984         case TUNSETPERSIST:
1985                 /* Disable/Enable persist mode. Keep an extra reference to the
1986                  * module to prevent the module being unprobed.
1987                  */
1988                 if (arg && !(tun->flags & IFF_PERSIST)) {
1989                         tun->flags |= IFF_PERSIST;
1990                         __module_get(THIS_MODULE);
1991                 }
1992                 if (!arg && (tun->flags & IFF_PERSIST)) {
1993                         tun->flags &= ~IFF_PERSIST;
1994                         module_put(THIS_MODULE);
1995                 }
1996
1997                 tun_debug(KERN_INFO, tun, "persist %s\n",
1998                           arg ? "enabled" : "disabled");
1999                 break;
2000
2001         case TUNSETOWNER:
2002                 /* Set owner of the device */
2003                 owner = make_kuid(current_user_ns(), arg);
2004                 if (!uid_valid(owner)) {
2005                         ret = -EINVAL;
2006                         break;
2007                 }
2008                 tun->owner = owner;
2009                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2010                           from_kuid(&init_user_ns, tun->owner));
2011                 break;
2012
2013         case TUNSETGROUP:
2014                 /* Set group of the device */
2015                 group = make_kgid(current_user_ns(), arg);
2016                 if (!gid_valid(group)) {
2017                         ret = -EINVAL;
2018                         break;
2019                 }
2020                 tun->group = group;
2021                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2022                           from_kgid(&init_user_ns, tun->group));
2023                 break;
2024
2025         case TUNSETLINK:
2026                 /* Only allow setting the type when the interface is down */
2027                 if (tun->dev->flags & IFF_UP) {
2028                         tun_debug(KERN_INFO, tun,
2029                                   "Linktype set failed because interface is up\n");
2030                         ret = -EBUSY;
2031                 } else {
2032                         tun->dev->type = (int) arg;
2033                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2034                                   tun->dev->type);
2035                         ret = 0;
2036                 }
2037                 break;
2038
2039 #ifdef TUN_DEBUG
2040         case TUNSETDEBUG:
2041                 tun->debug = arg;
2042                 break;
2043 #endif
2044         case TUNSETOFFLOAD:
2045                 ret = set_offload(tun, arg);
2046                 break;
2047
2048         case TUNSETTXFILTER:
2049                 /* Can be set only for TAPs */
2050                 ret = -EINVAL;
2051                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2052                         break;
2053                 ret = update_filter(&tun->txflt, (void __user *)arg);
2054                 break;
2055
2056         case SIOCGIFHWADDR:
2057                 /* Get hw address */
2058                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2059                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2060                 if (copy_to_user(argp, &ifr, ifreq_len))
2061                         ret = -EFAULT;
2062                 break;
2063
2064         case SIOCSIFHWADDR:
2065                 /* Set hw address */
2066                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2067                           ifr.ifr_hwaddr.sa_data);
2068
2069                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2070                 break;
2071
2072         case TUNGETSNDBUF:
2073                 sndbuf = tfile->socket.sk->sk_sndbuf;
2074                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2075                         ret = -EFAULT;
2076                 break;
2077
2078         case TUNSETSNDBUF:
2079                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2080                         ret = -EFAULT;
2081                         break;
2082                 }
2083
2084                 tun->sndbuf = sndbuf;
2085                 tun_set_sndbuf(tun);
2086                 break;
2087
2088         case TUNGETVNETHDRSZ:
2089                 vnet_hdr_sz = tun->vnet_hdr_sz;
2090                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2091                         ret = -EFAULT;
2092                 break;
2093
2094         case TUNSETVNETHDRSZ:
2095                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2096                         ret = -EFAULT;
2097                         break;
2098                 }
2099                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2100                         ret = -EINVAL;
2101                         break;
2102                 }
2103
2104                 tun->vnet_hdr_sz = vnet_hdr_sz;
2105                 break;
2106
2107         case TUNGETVNETLE:
2108                 le = !!(tun->flags & TUN_VNET_LE);
2109                 if (put_user(le, (int __user *)argp))
2110                         ret = -EFAULT;
2111                 break;
2112
2113         case TUNSETVNETLE:
2114                 if (get_user(le, (int __user *)argp)) {
2115                         ret = -EFAULT;
2116                         break;
2117                 }
2118                 if (le)
2119                         tun->flags |= TUN_VNET_LE;
2120                 else
2121                         tun->flags &= ~TUN_VNET_LE;
2122                 break;
2123
2124         case TUNGETVNETBE:
2125                 ret = tun_get_vnet_be(tun, argp);
2126                 break;
2127
2128         case TUNSETVNETBE:
2129                 ret = tun_set_vnet_be(tun, argp);
2130                 break;
2131
2132         case TUNATTACHFILTER:
2133                 /* Can be set only for TAPs */
2134                 ret = -EINVAL;
2135                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2136                         break;
2137                 ret = -EFAULT;
2138                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2139                         break;
2140
2141                 ret = tun_attach_filter(tun);
2142                 break;
2143
2144         case TUNDETACHFILTER:
2145                 /* Can be set only for TAPs */
2146                 ret = -EINVAL;
2147                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2148                         break;
2149                 ret = 0;
2150                 tun_detach_filter(tun, tun->numqueues);
2151                 break;
2152
2153         case TUNGETFILTER:
2154                 ret = -EINVAL;
2155                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2156                         break;
2157                 ret = -EFAULT;
2158                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2159                         break;
2160                 ret = 0;
2161                 break;
2162
2163         default:
2164                 ret = -EINVAL;
2165                 break;
2166         }
2167
2168 unlock:
2169         rtnl_unlock();
2170         if (tun)
2171                 tun_put(tun);
2172         return ret;
2173 }
2174
2175 static long tun_chr_ioctl(struct file *file,
2176                           unsigned int cmd, unsigned long arg)
2177 {
2178         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2179 }
2180
2181 #ifdef CONFIG_COMPAT
2182 static long tun_chr_compat_ioctl(struct file *file,
2183                          unsigned int cmd, unsigned long arg)
2184 {
2185         switch (cmd) {
2186         case TUNSETIFF:
2187         case TUNGETIFF:
2188         case TUNSETTXFILTER:
2189         case TUNGETSNDBUF:
2190         case TUNSETSNDBUF:
2191         case SIOCGIFHWADDR:
2192         case SIOCSIFHWADDR:
2193                 arg = (unsigned long)compat_ptr(arg);
2194                 break;
2195         default:
2196                 arg = (compat_ulong_t)arg;
2197                 break;
2198         }
2199
2200         /*
2201          * compat_ifreq is shorter than ifreq, so we must not access beyond
2202          * the end of that structure. All fields that are used in this
2203          * driver are compatible though, we don't need to convert the
2204          * contents.
2205          */
2206         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2207 }
2208 #endif /* CONFIG_COMPAT */
2209
2210 static int tun_chr_fasync(int fd, struct file *file, int on)
2211 {
2212         struct tun_file *tfile = file->private_data;
2213         int ret;
2214
2215         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2216                 goto out;
2217
2218         if (on) {
2219                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2220                 tfile->flags |= TUN_FASYNC;
2221         } else
2222                 tfile->flags &= ~TUN_FASYNC;
2223         ret = 0;
2224 out:
2225         return ret;
2226 }
2227
2228 static int tun_chr_open(struct inode *inode, struct file * file)
2229 {
2230         struct net *net = current->nsproxy->net_ns;
2231         struct tun_file *tfile;
2232
2233         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2234
2235         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2236                                             &tun_proto, 0);
2237         if (!tfile)
2238                 return -ENOMEM;
2239         RCU_INIT_POINTER(tfile->tun, NULL);
2240         tfile->flags = 0;
2241         tfile->ifindex = 0;
2242
2243         init_waitqueue_head(&tfile->wq.wait);
2244         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2245
2246         tfile->socket.file = file;
2247         tfile->socket.ops = &tun_socket_ops;
2248
2249         sock_init_data(&tfile->socket, &tfile->sk);
2250
2251         tfile->sk.sk_write_space = tun_sock_write_space;
2252         tfile->sk.sk_sndbuf = INT_MAX;
2253
2254         file->private_data = tfile;
2255         INIT_LIST_HEAD(&tfile->next);
2256
2257         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2258
2259         return 0;
2260 }
2261
2262 static int tun_chr_close(struct inode *inode, struct file *file)
2263 {
2264         struct tun_file *tfile = file->private_data;
2265
2266         tun_detach(tfile, true);
2267
2268         return 0;
2269 }
2270
2271 #ifdef CONFIG_PROC_FS
2272 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2273 {
2274         struct tun_struct *tun;
2275         struct ifreq ifr;
2276
2277         memset(&ifr, 0, sizeof(ifr));
2278
2279         rtnl_lock();
2280         tun = tun_get(f);
2281         if (tun)
2282                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2283         rtnl_unlock();
2284
2285         if (tun)
2286                 tun_put(tun);
2287
2288         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2289 }
2290 #endif
2291
2292 static const struct file_operations tun_fops = {
2293         .owner  = THIS_MODULE,
2294         .llseek = no_llseek,
2295         .read_iter  = tun_chr_read_iter,
2296         .write_iter = tun_chr_write_iter,
2297         .poll   = tun_chr_poll,
2298         .unlocked_ioctl = tun_chr_ioctl,
2299 #ifdef CONFIG_COMPAT
2300         .compat_ioctl = tun_chr_compat_ioctl,
2301 #endif
2302         .open   = tun_chr_open,
2303         .release = tun_chr_close,
2304         .fasync = tun_chr_fasync,
2305 #ifdef CONFIG_PROC_FS
2306         .show_fdinfo = tun_chr_show_fdinfo,
2307 #endif
2308 };
2309
2310 static struct miscdevice tun_miscdev = {
2311         .minor = TUN_MINOR,
2312         .name = "tun",
2313         .nodename = "net/tun",
2314         .fops = &tun_fops,
2315 };
2316
2317 /* ethtool interface */
2318
2319 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2320 {
2321         cmd->supported          = 0;
2322         cmd->advertising        = 0;
2323         ethtool_cmd_speed_set(cmd, SPEED_10);
2324         cmd->duplex             = DUPLEX_FULL;
2325         cmd->port               = PORT_TP;
2326         cmd->phy_address        = 0;
2327         cmd->transceiver        = XCVR_INTERNAL;
2328         cmd->autoneg            = AUTONEG_DISABLE;
2329         cmd->maxtxpkt           = 0;
2330         cmd->maxrxpkt           = 0;
2331         return 0;
2332 }
2333
2334 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2335 {
2336         struct tun_struct *tun = netdev_priv(dev);
2337
2338         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2339         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2340
2341         switch (tun->flags & TUN_TYPE_MASK) {
2342         case IFF_TUN:
2343                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2344                 break;
2345         case IFF_TAP:
2346                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2347                 break;
2348         }
2349 }
2350
2351 static u32 tun_get_msglevel(struct net_device *dev)
2352 {
2353 #ifdef TUN_DEBUG
2354         struct tun_struct *tun = netdev_priv(dev);
2355         return tun->debug;
2356 #else
2357         return -EOPNOTSUPP;
2358 #endif
2359 }
2360
2361 static void tun_set_msglevel(struct net_device *dev, u32 value)
2362 {
2363 #ifdef TUN_DEBUG
2364         struct tun_struct *tun = netdev_priv(dev);
2365         tun->debug = value;
2366 #endif
2367 }
2368
2369 static const struct ethtool_ops tun_ethtool_ops = {
2370         .get_settings   = tun_get_settings,
2371         .get_drvinfo    = tun_get_drvinfo,
2372         .get_msglevel   = tun_get_msglevel,
2373         .set_msglevel   = tun_set_msglevel,
2374         .get_link       = ethtool_op_get_link,
2375         .get_ts_info    = ethtool_op_get_ts_info,
2376 };
2377
2378
2379 static int __init tun_init(void)
2380 {
2381         int ret = 0;
2382
2383         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2384         pr_info("%s\n", DRV_COPYRIGHT);
2385
2386         ret = rtnl_link_register(&tun_link_ops);
2387         if (ret) {
2388                 pr_err("Can't register link_ops\n");
2389                 goto err_linkops;
2390         }
2391
2392         ret = misc_register(&tun_miscdev);
2393         if (ret) {
2394                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2395                 goto err_misc;
2396         }
2397         return  0;
2398 err_misc:
2399         rtnl_link_unregister(&tun_link_ops);
2400 err_linkops:
2401         return ret;
2402 }
2403
2404 static void tun_cleanup(void)
2405 {
2406         misc_deregister(&tun_miscdev);
2407         rtnl_link_unregister(&tun_link_ops);
2408 }
2409
2410 /* Get an underlying socket object from tun file.  Returns error unless file is
2411  * attached to a device.  The returned object works like a packet socket, it
2412  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2413  * holding a reference to the file for as long as the socket is in use. */
2414 struct socket *tun_get_socket(struct file *file)
2415 {
2416         struct tun_file *tfile;
2417         if (file->f_op != &tun_fops)
2418                 return ERR_PTR(-EINVAL);
2419         tfile = file->private_data;
2420         if (!tfile)
2421                 return ERR_PTR(-EBADFD);
2422         return &tfile->socket;
2423 }
2424 EXPORT_SYMBOL_GPL(tun_get_socket);
2425
2426 module_init(tun_init);
2427 module_exit(tun_cleanup);
2428 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2429 MODULE_AUTHOR(DRV_COPYRIGHT);
2430 MODULE_LICENSE("GPL");
2431 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2432 MODULE_ALIAS("devname:net/tun");