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                 lock_sock(tfile->socket.sk);
 626                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 627                 release_sock(tfile->socket.sk);
 628                 if (!err)
 629                         goto out;
 630         }
 631         tfile->queue_index = tun->numqueues;
 632         rcu_assign_pointer(tfile->tun, tun);
 633         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 634         tun->numqueues++;
 635
 636         if (tfile->detached)
 637                 tun_enable_queue(tfile);
 638         else
 639                 sock_hold(&tfile->sk);
 640
 641         tun_set_real_num_queues(tun);
 642
 643         /* device is allowed to go away first, so no need to hold extra
 644          * refcnt.
 645          */
 646
 647 out:
 648         return err;
 649 }
 650
 651 static struct tun_struct *__tun_get(struct tun_file *tfile)
 652 {
 653         struct tun_struct *tun;
 654
 655         rcu_read_lock();
 656         tun = rcu_dereference(tfile->tun);
 657         if (tun)
 658                 dev_hold(tun->dev);
 659         rcu_read_unlock();
 660
 661         return tun;
 662 }
 663
 664 static struct tun_struct *tun_get(struct file *file)
 665 {
 666         return __tun_get(file->private_data);
 667 }
 668
 669 static void tun_put(struct tun_struct *tun)
 670 {
 671         dev_put(tun->dev);
 672 }
 673
 674 /* TAP filtering */
 675 static void addr_hash_set(u32 *mask, const u8 *addr)
 676 {
 677         int n = ether_crc(ETH_ALEN, addr) >> 26;
 678         mask[n >> 5] |= (1 << (n & 31));
 679 }
 680
 681 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 682 {
 683         int n = ether_crc(ETH_ALEN, addr) >> 26;
 684         return mask[n >> 5] & (1 << (n & 31));
 685 }
 686
 687 static int update_filter(struct tap_filter *filter, void __user *arg)
 688 {
 689         struct { u8 u[ETH_ALEN]; } *addr;
 690         struct tun_filter uf;
 691         int err, alen, n, nexact;
 692
 693         if (copy_from_user(&uf, arg, sizeof(uf)))
 694                 return -EFAULT;
 695
 696         if (!uf.count) {
 697                 /* Disabled */
 698                 filter->count = 0;
 699                 return 0;
 700         }
 701
 702         alen = ETH_ALEN * uf.count;
 703         addr = kmalloc(alen, GFP_KERNEL);
 704         if (!addr)
 705                 return -ENOMEM;
 706
 707         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 708                 err = -EFAULT;
 709                 goto done;
 710         }
 711
 712         /* The filter is updated without holding any locks. Which is
 713          * perfectly safe. We disable it first and in the worst
 714          * case we'll accept a few undesired packets. */
 715         filter->count = 0;
 716         wmb();
 717
 718         /* Use first set of addresses as an exact filter */
 719         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 720                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 721
 722         nexact = n;
 723
 724         /* Remaining multicast addresses are hashed,
 725          * unicast will leave the filter disabled. */
 726         memset(filter->mask, 0, sizeof(filter->mask));
 727         for (; n < uf.count; n++) {
 728                 if (!is_multicast_ether_addr(addr[n].u)) {
 729                         err = 0; /* no filter */
 730                         goto done;
 731                 }
 732                 addr_hash_set(filter->mask, addr[n].u);
 733         }
 734
 735         /* For ALLMULTI just set the mask to all ones.
 736          * This overrides the mask populated above. */
 737         if ((uf.flags & TUN_FLT_ALLMULTI))
 738                 memset(filter->mask, ~0, sizeof(filter->mask));
 739
 740         /* Now enable the filter */
 741         wmb();
 742         filter->count = nexact;
 743
 744         /* Return the number of exact filters */
 745         err = nexact;
 746
 747 done:
 748         kfree(addr);
 749         return err;
 750 }
 751
 752 /* Returns: 0 - drop, !=0 - accept */
 753 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 754 {
 755         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 756          * at this point. */
 757         struct ethhdr *eh = (struct ethhdr *) skb->data;
 758         int i;
 759
 760         /* Exact match */
 761         for (i = 0; i < filter->count; i++)
 762                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 763                         return 1;
 764
 765         /* Inexact match (multicast only) */
 766         if (is_multicast_ether_addr(eh->h_dest))
 767                 return addr_hash_test(filter->mask, eh->h_dest);
 768
 769         return 0;
 770 }
 771
 772 /*
 773  * Checks whether the packet is accepted or not.
 774  * Returns: 0 - drop, !=0 - accept
 775  */
 776 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 777 {
 778         if (!filter->count)
 779                 return 1;
 780
 781         return run_filter(filter, skb);
 782 }
 783
 784 /* Network device part of the driver */
 785
 786 static const struct ethtool_ops tun_ethtool_ops;
 787
 788 /* Net device detach from fd. */
 789 static void tun_net_uninit(struct net_device *dev)
 790 {
 791         tun_detach_all(dev);
 792 }
 793
 794 /* Net device open. */
 795 static int tun_net_open(struct net_device *dev)
 796 {
 797         netif_tx_start_all_queues(dev);
 798         return 0;
 799 }
 800
 801 /* Net device close. */
 802 static int tun_net_close(struct net_device *dev)
 803 {
 804         netif_tx_stop_all_queues(dev);
 805         return 0;
 806 }
 807
 808 /* Net device start xmit */
 809 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 810 {
 811         struct tun_struct *tun = netdev_priv(dev);
 812         int txq = skb->queue_mapping;
 813         struct tun_file *tfile;
 814         u32 numqueues = 0;
 815
 816         rcu_read_lock();
 817         tfile = rcu_dereference(tun->tfiles[txq]);
 818         numqueues = ACCESS_ONCE(tun->numqueues);
 819
 820         /* Drop packet if interface is not attached */
 821         if (txq >= numqueues)
 822                 goto drop;
 823
 824         if (numqueues == 1) {
 825                 /* Select queue was not called for the skbuff, so we extract the
 826                  * RPS hash and save it into the flow_table here.
 827                  */
 828                 __u32 rxhash;
 829
 830                 rxhash = skb_get_hash(skb);
 831                 if (rxhash) {
 832                         struct tun_flow_entry *e;
 833                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 834                                         rxhash);
 835                         if (e)
 836                                 tun_flow_save_rps_rxhash(e, rxhash);
 837                 }
 838         }
 839
 840         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 841
 842         BUG_ON(!tfile);
 843
 844         /* Drop if the filter does not like it.
 845          * This is a noop if the filter is disabled.
 846          * Filter can be enabled only for the TAP devices. */
 847         if (!check_filter(&tun->txflt, skb))
 848                 goto drop;
 849
 850         if (tfile->socket.sk->sk_filter &&
 851             sk_filter(tfile->socket.sk, skb))
 852                 goto drop;
 853
 854         /* Limit the number of packets queued by dividing txq length with the
 855          * number of queues.
 856          */
 857         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
 858                           >= dev->tx_queue_len)
 859                 goto drop;
 860
 861         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 862                 goto drop;
 863
 864         if (skb->sk && sk_fullsock(skb->sk)) {
 865                 sock_tx_timestamp(skb->sk, skb->sk->sk_tsflags,
 866                                   &skb_shinfo(skb)->tx_flags);
 867                 sw_tx_timestamp(skb);
 868         }
 869
 870         /* Orphan the skb - required as we might hang on to it
 871          * for indefinite time.
 872          */
 873         skb_orphan(skb);
 874
 875         nf_reset(skb);
 876
 877         /* Enqueue packet */
 878         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 879
 880         /* Notify and wake up reader process */
 881         if (tfile->flags & TUN_FASYNC)
 882                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 883         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 884
 885         rcu_read_unlock();
 886         return NETDEV_TX_OK;
 887
 888 drop:
 889         dev->stats.tx_dropped++;
 890         skb_tx_error(skb);
 891         kfree_skb(skb);
 892         rcu_read_unlock();
 893         return NET_XMIT_DROP;
 894 }
 895
 896 static void tun_net_mclist(struct net_device *dev)
 897 {
 898         /*
 899          * This callback is supposed to deal with mc filter in
 900          * _rx_ path and has nothing to do with the _tx_ path.
 901          * In rx path we always accept everything userspace gives us.
 902          */
 903 }
 904
 905 #define MIN_MTU 68
 906 #define MAX_MTU 65535
 907
 908 static int
 909 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 910 {
 911         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 912                 return -EINVAL;
 913         dev->mtu = new_mtu;
 914         return 0;
 915 }
 916
 917 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 918         netdev_features_t features)
 919 {
 920         struct tun_struct *tun = netdev_priv(dev);
 921
 922         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 923 }
 924 #ifdef CONFIG_NET_POLL_CONTROLLER
 925 static void tun_poll_controller(struct net_device *dev)
 926 {
 927         /*
 928          * Tun only receives frames when:
 929          * 1) the char device endpoint gets data from user space
 930          * 2) the tun socket gets a sendmsg call from user space
 931          * Since both of those are synchronous operations, we are guaranteed
 932          * never to have pending data when we poll for it
 933          * so there is nothing to do here but return.
 934          * We need this though so netpoll recognizes us as an interface that
 935          * supports polling, which enables bridge devices in virt setups to
 936          * still use netconsole
 937          */
 938         return;
 939 }
 940 #endif
 941
 942 static void tun_set_headroom(struct net_device *dev, int new_hr)
 943 {
 944         struct tun_struct *tun = netdev_priv(dev);
 945
 946         if (new_hr < NET_SKB_PAD)
 947                 new_hr = NET_SKB_PAD;
 948
 949         tun->align = new_hr;
 950 }
 951
 952 static const struct net_device_ops tun_netdev_ops = {
 953         .ndo_uninit             = tun_net_uninit,
 954         .ndo_open               = tun_net_open,
 955         .ndo_stop               = tun_net_close,
 956         .ndo_start_xmit         = tun_net_xmit,
 957         .ndo_change_mtu         = tun_net_change_mtu,
 958         .ndo_fix_features       = tun_net_fix_features,
 959         .ndo_select_queue       = tun_select_queue,
 960 #ifdef CONFIG_NET_POLL_CONTROLLER
 961         .ndo_poll_controller    = tun_poll_controller,
 962 #endif
 963         .ndo_set_rx_headroom    = tun_set_headroom,
 964 };
 965
 966 static const struct net_device_ops tap_netdev_ops = {
 967         .ndo_uninit             = tun_net_uninit,
 968         .ndo_open               = tun_net_open,
 969         .ndo_stop               = tun_net_close,
 970         .ndo_start_xmit         = tun_net_xmit,
 971         .ndo_change_mtu         = tun_net_change_mtu,
 972         .ndo_fix_features       = tun_net_fix_features,
 973         .ndo_set_rx_mode        = tun_net_mclist,
 974         .ndo_set_mac_address    = eth_mac_addr,
 975         .ndo_validate_addr      = eth_validate_addr,
 976         .ndo_select_queue       = tun_select_queue,
 977 #ifdef CONFIG_NET_POLL_CONTROLLER
 978         .ndo_poll_controller    = tun_poll_controller,
 979 #endif
 980         .ndo_features_check     = passthru_features_check,
 981         .ndo_set_rx_headroom    = tun_set_headroom,
 982 };
 983
 984 static void tun_flow_init(struct tun_struct *tun)
 985 {
 986         int i;
 987
 988         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 989                 INIT_HLIST_HEAD(&tun->flows[i]);
 990
 991         tun->ageing_time = TUN_FLOW_EXPIRE;
 992         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 993         mod_timer(&tun->flow_gc_timer,
 994                   round_jiffies_up(jiffies + tun->ageing_time));
 995 }
 996
 997 static void tun_flow_uninit(struct tun_struct *tun)
 998 {
 999         del_timer_sync(&tun->flow_gc_timer);
1000         tun_flow_flush(tun);
1001 }
1002
1003 /* Initialize net device. */
1004 static void tun_net_init(struct net_device *dev)
1005 {
1006         struct tun_struct *tun = netdev_priv(dev);
1007
1008         switch (tun->flags & TUN_TYPE_MASK) {
1009         case IFF_TUN:
1010                 dev->netdev_ops = &tun_netdev_ops;
1011
1012                 /* Point-to-Point TUN Device */
1013                 dev->hard_header_len = 0;
1014                 dev->addr_len = 0;
1015                 dev->mtu = 1500;
1016
1017                 /* Zero header length */
1018                 dev->type = ARPHRD_NONE;
1019                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
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                 break;
1032         }
1033 }
1034
1035 /* Character device part */
1036
1037 /* Poll */
1038 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1039 {
1040         struct tun_file *tfile = file->private_data;
1041         struct tun_struct *tun = __tun_get(tfile);
1042         struct sock *sk;
1043         unsigned int mask = 0;
1044
1045         if (!tun)
1046                 return POLLERR;
1047
1048         sk = tfile->socket.sk;
1049
1050         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1051
1052         poll_wait(file, sk_sleep(sk), wait);
1053
1054         if (!skb_queue_empty(&sk->sk_receive_queue))
1055                 mask |= POLLIN | POLLRDNORM;
1056
1057         if (sock_writeable(sk) ||
1058             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1059              sock_writeable(sk)))
1060                 mask |= POLLOUT | POLLWRNORM;
1061
1062         if (tun->dev->reg_state != NETREG_REGISTERED)
1063                 mask = POLLERR;
1064
1065         tun_put(tun);
1066         return mask;
1067 }
1068
1069 /* prepad is the amount to reserve at front.  len is length after that.
1070  * linear is a hint as to how much to copy (usually headers). */
1071 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1072                                      size_t prepad, size_t len,
1073                                      size_t linear, int noblock)
1074 {
1075         struct sock *sk = tfile->socket.sk;
1076         struct sk_buff *skb;
1077         int err;
1078
1079         /* Under a page?  Don't bother with paged skb. */
1080         if (prepad + len < PAGE_SIZE || !linear)
1081                 linear = len;
1082
1083         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1084                                    &err, 0);
1085         if (!skb)
1086                 return ERR_PTR(err);
1087
1088         skb_reserve(skb, prepad);
1089         skb_put(skb, linear);
1090         skb->data_len = len - linear;
1091         skb->len += len - linear;
1092
1093         return skb;
1094 }
1095
1096 /* Get packet from user space buffer */
1097 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1098                             void *msg_control, struct iov_iter *from,
1099                             int noblock)
1100 {
1101         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1102         struct sk_buff *skb;
1103         size_t total_len = iov_iter_count(from);
1104         size_t len = total_len, align = tun->align, linear;
1105         struct virtio_net_hdr gso = { 0 };
1106         int good_linear;
1107         int copylen;
1108         bool zerocopy = false;
1109         int err;
1110         u32 rxhash;
1111         ssize_t n;
1112
1113         if (!(tun->dev->flags & IFF_UP))
1114                 return -EIO;
1115
1116         if (!(tun->flags & IFF_NO_PI)) {
1117                 if (len < sizeof(pi))
1118                         return -EINVAL;
1119                 len -= sizeof(pi);
1120
1121                 n = copy_from_iter(&pi, sizeof(pi), from);
1122                 if (n != sizeof(pi))
1123                         return -EFAULT;
1124         }
1125
1126         if (tun->flags & IFF_VNET_HDR) {
1127                 if (len < tun->vnet_hdr_sz)
1128                         return -EINVAL;
1129                 len -= tun->vnet_hdr_sz;
1130
1131                 n = copy_from_iter(&gso, sizeof(gso), from);
1132                 if (n != sizeof(gso))
1133                         return -EFAULT;
1134
1135                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1136                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1137                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1138
1139                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1140                         return -EINVAL;
1141                 iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1142         }
1143
1144         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1145                 align += NET_IP_ALIGN;
1146                 if (unlikely(len < ETH_HLEN ||
1147                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1148                         return -EINVAL;
1149         }
1150
1151         good_linear = SKB_MAX_HEAD(align);
1152
1153         if (msg_control) {
1154                 struct iov_iter i = *from;
1155
1156                 /* There are 256 bytes to be copied in skb, so there is
1157                  * enough room for skb expand head in case it is used.
1158                  * The rest of the buffer is mapped from userspace.
1159                  */
1160                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1161                 if (copylen > good_linear)
1162                         copylen = good_linear;
1163                 linear = copylen;
1164                 iov_iter_advance(&i, copylen);
1165                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1166                         zerocopy = true;
1167         }
1168
1169         if (!zerocopy) {
1170                 copylen = len;
1171                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1172                         linear = good_linear;
1173                 else
1174                         linear = tun16_to_cpu(tun, gso.hdr_len);
1175         }
1176
1177         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1178         if (IS_ERR(skb)) {
1179                 if (PTR_ERR(skb) != -EAGAIN)
1180                         tun->dev->stats.rx_dropped++;
1181                 return PTR_ERR(skb);
1182         }
1183
1184         if (zerocopy)
1185                 err = zerocopy_sg_from_iter(skb, from);
1186         else {
1187                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1188                 if (!err && msg_control) {
1189                         struct ubuf_info *uarg = msg_control;
1190                         uarg->callback(uarg, false);
1191                 }
1192         }
1193
1194         if (err) {
1195                 tun->dev->stats.rx_dropped++;
1196                 kfree_skb(skb);
1197                 return -EFAULT;
1198         }
1199
1200         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1201                 if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start),
1202                                           tun16_to_cpu(tun, gso.csum_offset))) {
1203                         tun->dev->stats.rx_frame_errors++;
1204                         kfree_skb(skb);
1205                         return -EINVAL;
1206                 }
1207         }
1208
1209         switch (tun->flags & TUN_TYPE_MASK) {
1210         case IFF_TUN:
1211                 if (tun->flags & IFF_NO_PI) {
1212                         switch (skb->data[0] & 0xf0) {
1213                         case 0x40:
1214                                 pi.proto = htons(ETH_P_IP);
1215                                 break;
1216                         case 0x60:
1217                                 pi.proto = htons(ETH_P_IPV6);
1218                                 break;
1219                         default:
1220                                 tun->dev->stats.rx_dropped++;
1221                                 kfree_skb(skb);
1222                                 return -EINVAL;
1223                         }
1224                 }
1225
1226                 skb_reset_mac_header(skb);
1227                 skb->protocol = pi.proto;
1228                 skb->dev = tun->dev;
1229                 break;
1230         case IFF_TAP:
1231                 skb->protocol = eth_type_trans(skb, tun->dev);
1232                 break;
1233         }
1234
1235         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1236                 pr_debug("GSO!\n");
1237                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1238                 case VIRTIO_NET_HDR_GSO_TCPV4:
1239                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1240                         break;
1241                 case VIRTIO_NET_HDR_GSO_TCPV6:
1242                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1243                         break;
1244                 case VIRTIO_NET_HDR_GSO_UDP:
1245                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1246                         break;
1247                 default:
1248                         tun->dev->stats.rx_frame_errors++;
1249                         kfree_skb(skb);
1250                         return -EINVAL;
1251                 }
1252
1253                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1254                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1255
1256                 skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size);
1257                 if (skb_shinfo(skb)->gso_size == 0) {
1258                         tun->dev->stats.rx_frame_errors++;
1259                         kfree_skb(skb);
1260                         return -EINVAL;
1261                 }
1262
1263                 /* Header must be checked, and gso_segs computed. */
1264                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1265                 skb_shinfo(skb)->gso_segs = 0;
1266         }
1267
1268         /* copy skb_ubuf_info for callback when skb has no error */
1269         if (zerocopy) {
1270                 skb_shinfo(skb)->destructor_arg = msg_control;
1271                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1272                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1273         }
1274
1275         skb_reset_network_header(skb);
1276         skb_probe_transport_header(skb, 0);
1277
1278         rxhash = skb_get_hash(skb);
1279         netif_rx_ni(skb);
1280
1281         tun->dev->stats.rx_packets++;
1282         tun->dev->stats.rx_bytes += len;
1283
1284         tun_flow_update(tun, rxhash, tfile);
1285         return total_len;
1286 }
1287
1288 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1289 {
1290         struct file *file = iocb->ki_filp;
1291         struct tun_struct *tun = tun_get(file);
1292         struct tun_file *tfile = file->private_data;
1293         ssize_t result;
1294
1295         if (!tun)
1296                 return -EBADFD;
1297
1298         result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1299
1300         tun_put(tun);
1301         return result;
1302 }
1303
1304 /* Put packet to the user space buffer */
1305 static ssize_t tun_put_user(struct tun_struct *tun,
1306                             struct tun_file *tfile,
1307                             struct sk_buff *skb,
1308                             struct iov_iter *iter)
1309 {
1310         struct tun_pi pi = { 0, skb->protocol };
1311         ssize_t total;
1312         int vlan_offset = 0;
1313         int vlan_hlen = 0;
1314         int vnet_hdr_sz = 0;
1315
1316         if (skb_vlan_tag_present(skb))
1317                 vlan_hlen = VLAN_HLEN;
1318
1319         if (tun->flags & IFF_VNET_HDR)
1320                 vnet_hdr_sz = tun->vnet_hdr_sz;
1321
1322         total = skb->len + vlan_hlen + vnet_hdr_sz;
1323
1324         if (!(tun->flags & IFF_NO_PI)) {
1325                 if (iov_iter_count(iter) < sizeof(pi))
1326                         return -EINVAL;
1327
1328                 total += sizeof(pi);
1329                 if (iov_iter_count(iter) < total) {
1330                         /* Packet will be striped */
1331                         pi.flags |= TUN_PKT_STRIP;
1332                 }
1333
1334                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1335                         return -EFAULT;
1336         }
1337
1338         if (vnet_hdr_sz) {
1339                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1340                 if (iov_iter_count(iter) < vnet_hdr_sz)
1341                         return -EINVAL;
1342
1343                 if (skb_is_gso(skb)) {
1344                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1345
1346                         /* This is a hint as to how much should be linear. */
1347                         gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb));
1348                         gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size);
1349                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1350                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1351                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1352                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1353                         else if (sinfo->gso_type & SKB_GSO_UDP)
1354                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1355                         else {
1356                                 pr_err("unexpected GSO type: "
1357                                        "0x%x, gso_size %d, hdr_len %d\n",
1358                                        sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1359                                        tun16_to_cpu(tun, gso.hdr_len));
1360                                 print_hex_dump(KERN_ERR, "tun: ",
1361                                                DUMP_PREFIX_NONE,
1362                                                16, 1, skb->head,
1363                                                min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1364                                 WARN_ON_ONCE(1);
1365                                 return -EINVAL;
1366                         }
1367                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1368                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1369                 } else
1370                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1371
1372                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1373                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1374                         gso.csum_start = cpu_to_tun16(tun, skb_checksum_start_offset(skb) +
1375                                                       vlan_hlen);
1376                         gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset);
1377                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1378                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1379                 } /* else everything is zero */
1380
1381                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1382                         return -EFAULT;
1383
1384                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1385         }
1386
1387         if (vlan_hlen) {
1388                 int ret;
1389                 struct {
1390                         __be16 h_vlan_proto;
1391                         __be16 h_vlan_TCI;
1392                 } veth;
1393
1394                 veth.h_vlan_proto = skb->vlan_proto;
1395                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1396
1397                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1398
1399                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1400                 if (ret || !iov_iter_count(iter))
1401                         goto done;
1402
1403                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1404                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1405                         goto done;
1406         }
1407
1408         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1409
1410 done:
1411         tun->dev->stats.tx_packets++;
1412         tun->dev->stats.tx_bytes += skb->len + vlan_hlen;
1413
1414         return total;
1415 }
1416
1417 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1418                            struct iov_iter *to,
1419                            int noblock)
1420 {
1421         struct sk_buff *skb;
1422         ssize_t ret;
1423         int peeked, err, off = 0;
1424
1425         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1426
1427         if (!iov_iter_count(to))
1428                 return 0;
1429
1430         if (tun->dev->reg_state != NETREG_REGISTERED)
1431                 return -EIO;
1432
1433         /* Read frames from queue */
1434         skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1435                                   &peeked, &off, &err);
1436         if (!skb)
1437                 return err;
1438
1439         ret = tun_put_user(tun, tfile, skb, to);
1440         if (unlikely(ret < 0))
1441                 kfree_skb(skb);
1442         else
1443                 consume_skb(skb);
1444
1445         return ret;
1446 }
1447
1448 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1449 {
1450         struct file *file = iocb->ki_filp;
1451         struct tun_file *tfile = file->private_data;
1452         struct tun_struct *tun = __tun_get(tfile);
1453         ssize_t len = iov_iter_count(to), ret;
1454
1455         if (!tun)
1456                 return -EBADFD;
1457         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1458         ret = min_t(ssize_t, ret, len);
1459         if (ret > 0)
1460                 iocb->ki_pos = ret;
1461         tun_put(tun);
1462         return ret;
1463 }
1464
1465 static void tun_free_netdev(struct net_device *dev)
1466 {
1467         struct tun_struct *tun = netdev_priv(dev);
1468
1469         BUG_ON(!(list_empty(&tun->disabled)));
1470         tun_flow_uninit(tun);
1471         security_tun_dev_free_security(tun->security);
1472         free_netdev(dev);
1473 }
1474
1475 static void tun_setup(struct net_device *dev)
1476 {
1477         struct tun_struct *tun = netdev_priv(dev);
1478
1479         tun->owner = INVALID_UID;
1480         tun->group = INVALID_GID;
1481
1482         dev->ethtool_ops = &tun_ethtool_ops;
1483         dev->destructor = tun_free_netdev;
1484         /* We prefer our own queue length */
1485         dev->tx_queue_len = TUN_READQ_SIZE;
1486 }
1487
1488 /* Trivial set of netlink ops to allow deleting tun or tap
1489  * device with netlink.
1490  */
1491 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1492 {
1493         return -EINVAL;
1494 }
1495
1496 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1497         .kind           = DRV_NAME,
1498         .priv_size      = sizeof(struct tun_struct),
1499         .setup          = tun_setup,
1500         .validate       = tun_validate,
1501 };
1502
1503 static void tun_sock_write_space(struct sock *sk)
1504 {
1505         struct tun_file *tfile;
1506         wait_queue_head_t *wqueue;
1507
1508         if (!sock_writeable(sk))
1509                 return;
1510
1511         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1512                 return;
1513
1514         wqueue = sk_sleep(sk);
1515         if (wqueue && waitqueue_active(wqueue))
1516                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1517                                                 POLLWRNORM | POLLWRBAND);
1518
1519         tfile = container_of(sk, struct tun_file, sk);
1520         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1521 }
1522
1523 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1524 {
1525         int ret;
1526         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1527         struct tun_struct *tun = __tun_get(tfile);
1528
1529         if (!tun)
1530                 return -EBADFD;
1531
1532         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1533                            m->msg_flags & MSG_DONTWAIT);
1534         tun_put(tun);
1535         return ret;
1536 }
1537
1538 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1539                        int flags)
1540 {
1541         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1542         struct tun_struct *tun = __tun_get(tfile);
1543         int ret;
1544
1545         if (!tun)
1546                 return -EBADFD;
1547
1548         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1549                 ret = -EINVAL;
1550                 goto out;
1551         }
1552         if (flags & MSG_ERRQUEUE) {
1553                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1554                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1555                 goto out;
1556         }
1557         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1558         if (ret > (ssize_t)total_len) {
1559                 m->msg_flags |= MSG_TRUNC;
1560                 ret = flags & MSG_TRUNC ? ret : total_len;
1561         }
1562 out:
1563         tun_put(tun);
1564         return ret;
1565 }
1566
1567 /* Ops structure to mimic raw sockets with tun */
1568 static const struct proto_ops tun_socket_ops = {
1569         .sendmsg = tun_sendmsg,
1570         .recvmsg = tun_recvmsg,
1571 };
1572
1573 static struct proto tun_proto = {
1574         .name           = "tun",
1575         .owner          = THIS_MODULE,
1576         .obj_size       = sizeof(struct tun_file),
1577 };
1578
1579 static int tun_flags(struct tun_struct *tun)
1580 {
1581         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1582 }
1583
1584 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1585                               char *buf)
1586 {
1587         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1588         return sprintf(buf, "0x%x\n", tun_flags(tun));
1589 }
1590
1591 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1592                               char *buf)
1593 {
1594         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1595         return uid_valid(tun->owner)?
1596                 sprintf(buf, "%u\n",
1597                         from_kuid_munged(current_user_ns(), tun->owner)):
1598                 sprintf(buf, "-1\n");
1599 }
1600
1601 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1602                               char *buf)
1603 {
1604         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1605         return gid_valid(tun->group) ?
1606                 sprintf(buf, "%u\n",
1607                         from_kgid_munged(current_user_ns(), tun->group)):
1608                 sprintf(buf, "-1\n");
1609 }
1610
1611 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1612 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1613 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1614
1615 static struct attribute *tun_dev_attrs[] = {
1616         &dev_attr_tun_flags.attr,
1617         &dev_attr_owner.attr,
1618         &dev_attr_group.attr,
1619         NULL
1620 };
1621
1622 static const struct attribute_group tun_attr_group = {
1623         .attrs = tun_dev_attrs
1624 };
1625
1626 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1627 {
1628         struct tun_struct *tun;
1629         struct tun_file *tfile = file->private_data;
1630         struct net_device *dev;
1631         int err;
1632
1633         if (tfile->detached)
1634                 return -EINVAL;
1635
1636         dev = __dev_get_by_name(net, ifr->ifr_name);
1637         if (dev) {
1638                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1639                         return -EBUSY;
1640                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1641                         tun = netdev_priv(dev);
1642                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1643                         tun = netdev_priv(dev);
1644                 else
1645                         return -EINVAL;
1646
1647                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1648                     !!(tun->flags & IFF_MULTI_QUEUE))
1649                         return -EINVAL;
1650
1651                 if (tun_not_capable(tun))
1652                         return -EPERM;
1653                 err = security_tun_dev_open(tun->security);
1654                 if (err < 0)
1655                         return err;
1656
1657                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1658                 if (err < 0)
1659                         return err;
1660
1661                 if (tun->flags & IFF_MULTI_QUEUE &&
1662                     (tun->numqueues + tun->numdisabled > 1)) {
1663                         /* One or more queue has already been attached, no need
1664                          * to initialize the device again.
1665                          */
1666                         return 0;
1667                 }
1668         }
1669         else {
1670                 char *name;
1671                 unsigned long flags = 0;
1672                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1673                              MAX_TAP_QUEUES : 1;
1674
1675                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1676                         return -EPERM;
1677                 err = security_tun_dev_create();
1678                 if (err < 0)
1679                         return err;
1680
1681                 /* Set dev type */
1682                 if (ifr->ifr_flags & IFF_TUN) {
1683                         /* TUN device */
1684                         flags |= IFF_TUN;
1685                         name = "tun%d";
1686                 } else if (ifr->ifr_flags & IFF_TAP) {
1687                         /* TAP device */
1688                         flags |= IFF_TAP;
1689                         name = "tap%d";
1690                 } else
1691                         return -EINVAL;
1692
1693                 if (*ifr->ifr_name)
1694                         name = ifr->ifr_name;
1695
1696                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1697                                        NET_NAME_UNKNOWN, tun_setup, queues,
1698                                        queues);
1699
1700                 if (!dev)
1701                         return -ENOMEM;
1702
1703                 dev_net_set(dev, net);
1704                 dev->rtnl_link_ops = &tun_link_ops;
1705                 dev->ifindex = tfile->ifindex;
1706                 dev->sysfs_groups[0] = &tun_attr_group;
1707
1708                 tun = netdev_priv(dev);
1709                 tun->dev = dev;
1710                 tun->flags = flags;
1711                 tun->txflt.count = 0;
1712                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1713
1714                 tun->align = NET_SKB_PAD;
1715                 tun->filter_attached = false;
1716                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1717
1718                 spin_lock_init(&tun->lock);
1719
1720                 err = security_tun_dev_alloc_security(&tun->security);
1721                 if (err < 0)
1722                         goto err_free_dev;
1723
1724                 tun_net_init(dev);
1725                 tun_flow_init(tun);
1726
1727                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1728                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1729                                    NETIF_F_HW_VLAN_STAG_TX;
1730                 dev->features = dev->hw_features;
1731                 dev->vlan_features = dev->features &
1732                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1733                                        NETIF_F_HW_VLAN_STAG_TX);
1734
1735                 INIT_LIST_HEAD(&tun->disabled);
1736                 err = tun_attach(tun, file, false);
1737                 if (err < 0)
1738                         goto err_free_flow;
1739
1740                 err = register_netdevice(tun->dev);
1741                 if (err < 0)
1742                         goto err_detach;
1743         }
1744
1745         netif_carrier_on(tun->dev);
1746
1747         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1748
1749         tun->flags = (tun->flags & ~TUN_FEATURES) |
1750                 (ifr->ifr_flags & TUN_FEATURES);
1751
1752         /* Make sure persistent devices do not get stuck in
1753          * xoff state.
1754          */
1755         if (netif_running(tun->dev))
1756                 netif_tx_wake_all_queues(tun->dev);
1757
1758         strcpy(ifr->ifr_name, tun->dev->name);
1759         return 0;
1760
1761 err_detach:
1762         tun_detach_all(dev);
1763 err_free_flow:
1764         tun_flow_uninit(tun);
1765         security_tun_dev_free_security(tun->security);
1766 err_free_dev:
1767         free_netdev(dev);
1768         return err;
1769 }
1770
1771 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1772                        struct ifreq *ifr)
1773 {
1774         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1775
1776         strcpy(ifr->ifr_name, tun->dev->name);
1777
1778         ifr->ifr_flags = tun_flags(tun);
1779
1780 }
1781
1782 /* This is like a cut-down ethtool ops, except done via tun fd so no
1783  * privs required. */
1784 static int set_offload(struct tun_struct *tun, unsigned long arg)
1785 {
1786         netdev_features_t features = 0;
1787
1788         if (arg & TUN_F_CSUM) {
1789                 features |= NETIF_F_HW_CSUM;
1790                 arg &= ~TUN_F_CSUM;
1791
1792                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1793                         if (arg & TUN_F_TSO_ECN) {
1794                                 features |= NETIF_F_TSO_ECN;
1795                                 arg &= ~TUN_F_TSO_ECN;
1796                         }
1797                         if (arg & TUN_F_TSO4)
1798                                 features |= NETIF_F_TSO;
1799                         if (arg & TUN_F_TSO6)
1800                                 features |= NETIF_F_TSO6;
1801                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1802                 }
1803
1804                 if (arg & TUN_F_UFO) {
1805                         features |= NETIF_F_UFO;
1806                         arg &= ~TUN_F_UFO;
1807                 }
1808         }
1809
1810         /* This gives the user a way to test for new features in future by
1811          * trying to set them. */
1812         if (arg)
1813                 return -EINVAL;
1814
1815         tun->set_features = features;
1816         netdev_update_features(tun->dev);
1817
1818         return 0;
1819 }
1820
1821 static void tun_detach_filter(struct tun_struct *tun, int n)
1822 {
1823         int i;
1824         struct tun_file *tfile;
1825
1826         for (i = 0; i < n; i++) {
1827                 tfile = rtnl_dereference(tun->tfiles[i]);
1828                 lock_sock(tfile->socket.sk);
1829                 sk_detach_filter(tfile->socket.sk);
1830                 release_sock(tfile->socket.sk);
1831         }
1832
1833         tun->filter_attached = false;
1834 }
1835
1836 static int tun_attach_filter(struct tun_struct *tun)
1837 {
1838         int i, ret = 0;
1839         struct tun_file *tfile;
1840
1841         for (i = 0; i < tun->numqueues; i++) {
1842                 tfile = rtnl_dereference(tun->tfiles[i]);
1843                 lock_sock(tfile->socket.sk);
1844                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1845                 release_sock(tfile->socket.sk);
1846                 if (ret) {
1847                         tun_detach_filter(tun, i);
1848                         return ret;
1849                 }
1850         }
1851
1852         tun->filter_attached = true;
1853         return ret;
1854 }
1855
1856 static void tun_set_sndbuf(struct tun_struct *tun)
1857 {
1858         struct tun_file *tfile;
1859         int i;
1860
1861         for (i = 0; i < tun->numqueues; i++) {
1862                 tfile = rtnl_dereference(tun->tfiles[i]);
1863                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1864         }
1865 }
1866
1867 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1868 {
1869         struct tun_file *tfile = file->private_data;
1870         struct tun_struct *tun;
1871         int ret = 0;
1872
1873         rtnl_lock();
1874
1875         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1876                 tun = tfile->detached;
1877                 if (!tun) {
1878                         ret = -EINVAL;
1879                         goto unlock;
1880                 }
1881                 ret = security_tun_dev_attach_queue(tun->security);
1882                 if (ret < 0)
1883                         goto unlock;
1884                 ret = tun_attach(tun, file, false);
1885         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1886                 tun = rtnl_dereference(tfile->tun);
1887                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1888                         ret = -EINVAL;
1889                 else
1890                         __tun_detach(tfile, false);
1891         } else
1892                 ret = -EINVAL;
1893
1894 unlock:
1895         rtnl_unlock();
1896         return ret;
1897 }
1898
1899 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1900                             unsigned long arg, int ifreq_len)
1901 {
1902         struct tun_file *tfile = file->private_data;
1903         struct tun_struct *tun;
1904         void __user* argp = (void __user*)arg;
1905         struct ifreq ifr;
1906         kuid_t owner;
1907         kgid_t group;
1908         int sndbuf;
1909         int vnet_hdr_sz;
1910         unsigned int ifindex;
1911         int le;
1912         int ret;
1913
1914         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1915                 if (copy_from_user(&ifr, argp, ifreq_len))
1916                         return -EFAULT;
1917         } else {
1918                 memset(&ifr, 0, sizeof(ifr));
1919         }
1920         if (cmd == TUNGETFEATURES) {
1921                 /* Currently this just means: "what IFF flags are valid?".
1922                  * This is needed because we never checked for invalid flags on
1923                  * TUNSETIFF.
1924                  */
1925                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1926                                 (unsigned int __user*)argp);
1927         } else if (cmd == TUNSETQUEUE)
1928                 return tun_set_queue(file, &ifr);
1929
1930         ret = 0;
1931         rtnl_lock();
1932
1933         tun = __tun_get(tfile);
1934         if (cmd == TUNSETIFF && !tun) {
1935                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1936
1937                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
1938
1939                 if (ret)
1940                         goto unlock;
1941
1942                 if (copy_to_user(argp, &ifr, ifreq_len))
1943                         ret = -EFAULT;
1944                 goto unlock;
1945         }
1946         if (cmd == TUNSETIFINDEX) {
1947                 ret = -EPERM;
1948                 if (tun)
1949                         goto unlock;
1950
1951                 ret = -EFAULT;
1952                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1953                         goto unlock;
1954
1955                 ret = 0;
1956                 tfile->ifindex = ifindex;
1957                 goto unlock;
1958         }
1959
1960         ret = -EBADFD;
1961         if (!tun)
1962                 goto unlock;
1963
1964         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1965
1966         ret = 0;
1967         switch (cmd) {
1968         case TUNGETIFF:
1969                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1970
1971                 if (tfile->detached)
1972                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
1973                 if (!tfile->socket.sk->sk_filter)
1974                         ifr.ifr_flags |= IFF_NOFILTER;
1975
1976                 if (copy_to_user(argp, &ifr, ifreq_len))
1977                         ret = -EFAULT;
1978                 break;
1979
1980         case TUNSETNOCSUM:
1981                 /* Disable/Enable checksum */
1982
1983                 /* [unimplemented] */
1984                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1985                           arg ? "disabled" : "enabled");
1986                 break;
1987
1988         case TUNSETPERSIST:
1989                 /* Disable/Enable persist mode. Keep an extra reference to the
1990                  * module to prevent the module being unprobed.
1991                  */
1992                 if (arg && !(tun->flags & IFF_PERSIST)) {
1993                         tun->flags |= IFF_PERSIST;
1994                         __module_get(THIS_MODULE);
1995                 }
1996                 if (!arg && (tun->flags & IFF_PERSIST)) {
1997                         tun->flags &= ~IFF_PERSIST;
1998                         module_put(THIS_MODULE);
1999                 }
2000
2001                 tun_debug(KERN_INFO, tun, "persist %s\n",
2002                           arg ? "enabled" : "disabled");
2003                 break;
2004
2005         case TUNSETOWNER:
2006                 /* Set owner of the device */
2007                 owner = make_kuid(current_user_ns(), arg);
2008                 if (!uid_valid(owner)) {
2009                         ret = -EINVAL;
2010                         break;
2011                 }
2012                 tun->owner = owner;
2013                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2014                           from_kuid(&init_user_ns, tun->owner));
2015                 break;
2016
2017         case TUNSETGROUP:
2018                 /* Set group of the device */
2019                 group = make_kgid(current_user_ns(), arg);
2020                 if (!gid_valid(group)) {
2021                         ret = -EINVAL;
2022                         break;
2023                 }
2024                 tun->group = group;
2025                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2026                           from_kgid(&init_user_ns, tun->group));
2027                 break;
2028
2029         case TUNSETLINK:
2030                 /* Only allow setting the type when the interface is down */
2031                 if (tun->dev->flags & IFF_UP) {
2032                         tun_debug(KERN_INFO, tun,
2033                                   "Linktype set failed because interface is up\n");
2034                         ret = -EBUSY;
2035                 } else {
2036                         tun->dev->type = (int) arg;
2037                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2038                                   tun->dev->type);
2039                         ret = 0;
2040                 }
2041                 break;
2042
2043 #ifdef TUN_DEBUG
2044         case TUNSETDEBUG:
2045                 tun->debug = arg;
2046                 break;
2047 #endif
2048         case TUNSETOFFLOAD:
2049                 ret = set_offload(tun, arg);
2050                 break;
2051
2052         case TUNSETTXFILTER:
2053                 /* Can be set only for TAPs */
2054                 ret = -EINVAL;
2055                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2056                         break;
2057                 ret = update_filter(&tun->txflt, (void __user *)arg);
2058                 break;
2059
2060         case SIOCGIFHWADDR:
2061                 /* Get hw address */
2062                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2063                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2064                 if (copy_to_user(argp, &ifr, ifreq_len))
2065                         ret = -EFAULT;
2066                 break;
2067
2068         case SIOCSIFHWADDR:
2069                 /* Set hw address */
2070                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2071                           ifr.ifr_hwaddr.sa_data);
2072
2073                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2074                 break;
2075
2076         case TUNGETSNDBUF:
2077                 sndbuf = tfile->socket.sk->sk_sndbuf;
2078                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2079                         ret = -EFAULT;
2080                 break;
2081
2082         case TUNSETSNDBUF:
2083                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2084                         ret = -EFAULT;
2085                         break;
2086                 }
2087
2088                 tun->sndbuf = sndbuf;
2089                 tun_set_sndbuf(tun);
2090                 break;
2091
2092         case TUNGETVNETHDRSZ:
2093                 vnet_hdr_sz = tun->vnet_hdr_sz;
2094                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2095                         ret = -EFAULT;
2096                 break;
2097
2098         case TUNSETVNETHDRSZ:
2099                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2100                         ret = -EFAULT;
2101                         break;
2102                 }
2103                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2104                         ret = -EINVAL;
2105                         break;
2106                 }
2107
2108                 tun->vnet_hdr_sz = vnet_hdr_sz;
2109                 break;
2110
2111         case TUNGETVNETLE:
2112                 le = !!(tun->flags & TUN_VNET_LE);
2113                 if (put_user(le, (int __user *)argp))
2114                         ret = -EFAULT;
2115                 break;
2116
2117         case TUNSETVNETLE:
2118                 if (get_user(le, (int __user *)argp)) {
2119                         ret = -EFAULT;
2120                         break;
2121                 }
2122                 if (le)
2123                         tun->flags |= TUN_VNET_LE;
2124                 else
2125                         tun->flags &= ~TUN_VNET_LE;
2126                 break;
2127
2128         case TUNGETVNETBE:
2129                 ret = tun_get_vnet_be(tun, argp);
2130                 break;
2131
2132         case TUNSETVNETBE:
2133                 ret = tun_set_vnet_be(tun, argp);
2134                 break;
2135
2136         case TUNATTACHFILTER:
2137                 /* Can be set only for TAPs */
2138                 ret = -EINVAL;
2139                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2140                         break;
2141                 ret = -EFAULT;
2142                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2143                         break;
2144
2145                 ret = tun_attach_filter(tun);
2146                 break;
2147
2148         case TUNDETACHFILTER:
2149                 /* Can be set only for TAPs */
2150                 ret = -EINVAL;
2151                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2152                         break;
2153                 ret = 0;
2154                 tun_detach_filter(tun, tun->numqueues);
2155                 break;
2156
2157         case TUNGETFILTER:
2158                 ret = -EINVAL;
2159                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2160                         break;
2161                 ret = -EFAULT;
2162                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2163                         break;
2164                 ret = 0;
2165                 break;
2166
2167         default:
2168                 ret = -EINVAL;
2169                 break;
2170         }
2171
2172 unlock:
2173         rtnl_unlock();
2174         if (tun)
2175                 tun_put(tun);
2176         return ret;
2177 }
2178
2179 static long tun_chr_ioctl(struct file *file,
2180                           unsigned int cmd, unsigned long arg)
2181 {
2182         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2183 }
2184
2185 #ifdef CONFIG_COMPAT
2186 static long tun_chr_compat_ioctl(struct file *file,
2187                          unsigned int cmd, unsigned long arg)
2188 {
2189         switch (cmd) {
2190         case TUNSETIFF:
2191         case TUNGETIFF:
2192         case TUNSETTXFILTER:
2193         case TUNGETSNDBUF:
2194         case TUNSETSNDBUF:
2195         case SIOCGIFHWADDR:
2196         case SIOCSIFHWADDR:
2197                 arg = (unsigned long)compat_ptr(arg);
2198                 break;
2199         default:
2200                 arg = (compat_ulong_t)arg;
2201                 break;
2202         }
2203
2204         /*
2205          * compat_ifreq is shorter than ifreq, so we must not access beyond
2206          * the end of that structure. All fields that are used in this
2207          * driver are compatible though, we don't need to convert the
2208          * contents.
2209          */
2210         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2211 }
2212 #endif /* CONFIG_COMPAT */
2213
2214 static int tun_chr_fasync(int fd, struct file *file, int on)
2215 {
2216         struct tun_file *tfile = file->private_data;
2217         int ret;
2218
2219         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2220                 goto out;
2221
2222         if (on) {
2223                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2224                 tfile->flags |= TUN_FASYNC;
2225         } else
2226                 tfile->flags &= ~TUN_FASYNC;
2227         ret = 0;
2228 out:
2229         return ret;
2230 }
2231
2232 static int tun_chr_open(struct inode *inode, struct file * file)
2233 {
2234         struct net *net = current->nsproxy->net_ns;
2235         struct tun_file *tfile;
2236
2237         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2238
2239         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2240                                             &tun_proto, 0);
2241         if (!tfile)
2242                 return -ENOMEM;
2243         RCU_INIT_POINTER(tfile->tun, NULL);
2244         tfile->flags = 0;
2245         tfile->ifindex = 0;
2246
2247         init_waitqueue_head(&tfile->wq.wait);
2248         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2249
2250         tfile->socket.file = file;
2251         tfile->socket.ops = &tun_socket_ops;
2252
2253         sock_init_data(&tfile->socket, &tfile->sk);
2254
2255         tfile->sk.sk_write_space = tun_sock_write_space;
2256         tfile->sk.sk_sndbuf = INT_MAX;
2257
2258         file->private_data = tfile;
2259         INIT_LIST_HEAD(&tfile->next);
2260
2261         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2262
2263         return 0;
2264 }
2265
2266 static int tun_chr_close(struct inode *inode, struct file *file)
2267 {
2268         struct tun_file *tfile = file->private_data;
2269
2270         tun_detach(tfile, true);
2271
2272         return 0;
2273 }
2274
2275 #ifdef CONFIG_PROC_FS
2276 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2277 {
2278         struct tun_struct *tun;
2279         struct ifreq ifr;
2280
2281         memset(&ifr, 0, sizeof(ifr));
2282
2283         rtnl_lock();
2284         tun = tun_get(f);
2285         if (tun)
2286                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2287         rtnl_unlock();
2288
2289         if (tun)
2290                 tun_put(tun);
2291
2292         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2293 }
2294 #endif
2295
2296 static const struct file_operations tun_fops = {
2297         .owner  = THIS_MODULE,
2298         .llseek = no_llseek,
2299         .read_iter  = tun_chr_read_iter,
2300         .write_iter = tun_chr_write_iter,
2301         .poll   = tun_chr_poll,
2302         .unlocked_ioctl = tun_chr_ioctl,
2303 #ifdef CONFIG_COMPAT
2304         .compat_ioctl = tun_chr_compat_ioctl,
2305 #endif
2306         .open   = tun_chr_open,
2307         .release = tun_chr_close,
2308         .fasync = tun_chr_fasync,
2309 #ifdef CONFIG_PROC_FS
2310         .show_fdinfo = tun_chr_show_fdinfo,
2311 #endif
2312 };
2313
2314 static struct miscdevice tun_miscdev = {
2315         .minor = TUN_MINOR,
2316         .name = "tun",
2317         .nodename = "net/tun",
2318         .fops = &tun_fops,
2319 };
2320
2321 /* ethtool interface */
2322
2323 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2324 {
2325         cmd->supported          = 0;
2326         cmd->advertising        = 0;
2327         ethtool_cmd_speed_set(cmd, SPEED_10);
2328         cmd->duplex             = DUPLEX_FULL;
2329         cmd->port               = PORT_TP;
2330         cmd->phy_address        = 0;
2331         cmd->transceiver        = XCVR_INTERNAL;
2332         cmd->autoneg            = AUTONEG_DISABLE;
2333         cmd->maxtxpkt           = 0;
2334         cmd->maxrxpkt           = 0;
2335         return 0;
2336 }
2337
2338 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2339 {
2340         struct tun_struct *tun = netdev_priv(dev);
2341
2342         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2343         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2344
2345         switch (tun->flags & TUN_TYPE_MASK) {
2346         case IFF_TUN:
2347                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2348                 break;
2349         case IFF_TAP:
2350                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2351                 break;
2352         }
2353 }
2354
2355 static u32 tun_get_msglevel(struct net_device *dev)
2356 {
2357 #ifdef TUN_DEBUG
2358         struct tun_struct *tun = netdev_priv(dev);
2359         return tun->debug;
2360 #else
2361         return -EOPNOTSUPP;
2362 #endif
2363 }
2364
2365 static void tun_set_msglevel(struct net_device *dev, u32 value)
2366 {
2367 #ifdef TUN_DEBUG
2368         struct tun_struct *tun = netdev_priv(dev);
2369         tun->debug = value;
2370 #endif
2371 }
2372
2373 static const struct ethtool_ops tun_ethtool_ops = {
2374         .get_settings   = tun_get_settings,
2375         .get_drvinfo    = tun_get_drvinfo,
2376         .get_msglevel   = tun_get_msglevel,
2377         .set_msglevel   = tun_set_msglevel,
2378         .get_link       = ethtool_op_get_link,
2379         .get_ts_info    = ethtool_op_get_ts_info,
2380 };
2381
2382
2383 static int __init tun_init(void)
2384 {
2385         int ret = 0;
2386
2387         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2388         pr_info("%s\n", DRV_COPYRIGHT);
2389
2390         ret = rtnl_link_register(&tun_link_ops);
2391         if (ret) {
2392                 pr_err("Can't register link_ops\n");
2393                 goto err_linkops;
2394         }
2395
2396         ret = misc_register(&tun_miscdev);
2397         if (ret) {
2398                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2399                 goto err_misc;
2400         }
2401         return  0;
2402 err_misc:
2403         rtnl_link_unregister(&tun_link_ops);
2404 err_linkops:
2405         return ret;
2406 }
2407
2408 static void tun_cleanup(void)
2409 {
2410         misc_deregister(&tun_miscdev);
2411         rtnl_link_unregister(&tun_link_ops);
2412 }
2413
2414 /* Get an underlying socket object from tun file.  Returns error unless file is
2415  * attached to a device.  The returned object works like a packet socket, it
2416  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2417  * holding a reference to the file for as long as the socket is in use. */
2418 struct socket *tun_get_socket(struct file *file)
2419 {
2420         struct tun_file *tfile;
2421         if (file->f_op != &tun_fops)
2422                 return ERR_PTR(-EINVAL);
2423         tfile = file->private_data;
2424         if (!tfile)
2425                 return ERR_PTR(-EBADFD);
2426         return &tfile->socket;
2427 }
2428 EXPORT_SYMBOL_GPL(tun_get_socket);
2429
2430 module_init(tun_init);
2431 module_exit(tun_cleanup);
2432 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2433 MODULE_AUTHOR(DRV_COPYRIGHT);
2434 MODULE_LICENSE("GPL");
2435 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2436 MODULE_ALIAS("devname:net/tun");