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
  74 #include <asm/uaccess.h>
  75
  76 /* Uncomment to enable debugging */
  77 /* #define TUN_DEBUG 1 */
  78
  79 #ifdef TUN_DEBUG
  80 static int debug;
  81
  82 #define tun_debug(level, tun, fmt, args...)                     \
  83 do {                                                            \
  84         if (tun->debug)                                         \
  85                 netdev_printk(level, tun->dev, fmt, ##args);    \
  86 } while (0)
  87 #define DBG1(level, fmt, args...)                               \
  88 do {                                                            \
  89         if (debug == 2)                                         \
  90                 printk(level fmt, ##args);                      \
  91 } while (0)
  92 #else
  93 #define tun_debug(level, tun, fmt, args...)                     \
  94 do {                                                            \
  95         if (0)                                                  \
  96                 netdev_printk(level, tun->dev, fmt, ##args);    \
  97 } while (0)
  98 #define DBG1(level, fmt, args...)                               \
  99 do {                                                            \
 100         if (0)                                                  \
 101                 printk(level fmt, ##args);                      \
 102 } while (0)
 103 #endif
 104
 105 #define GOODCOPY_LEN 128
 106
 107 #define FLT_EXACT_COUNT 8
 108 struct tap_filter {
 109         unsigned int    count;    /* Number of addrs. Zero means disabled */
 110         u32             mask[2];  /* Mask of the hashed addrs */
 111         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 112 };
 113
 114 /* DEFAULT_MAX_NUM_RSS_QUEUES were chosen to let the rx/tx queues allocated for
 115  * the netdevice to be fit in one page. So we can make sure the success of
 116  * memory allocation. TODO: increase the limit. */
 117 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
 118 #define MAX_TAP_FLOWS  4096
 119
 120 #define TUN_FLOW_EXPIRE (3 * HZ)
 121
 122 /* A tun_file connects an open character device to a tuntap netdevice. It
 123  * also contains all socket related structures (except sock_fprog and tap_filter)
 124  * to serve as one transmit queue for tuntap device. The sock_fprog and
 125  * tap_filter were kept in tun_struct since they were used for filtering for the
 126  * netdevice not for a specific queue (at least I didn't see the requirement for
 127  * this).
 128  *
 129  * RCU usage:
 130  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 131  * other can only be read while rcu_read_lock or rtnl_lock is held.
 132  */
 133 struct tun_file {
 134         struct sock sk;
 135         struct socket socket;
 136         struct socket_wq wq;
 137         struct tun_struct __rcu *tun;
 138         struct net *net;
 139         struct fasync_struct *fasync;
 140         /* only used for fasnyc */
 141         unsigned int flags;
 142         union {
 143                 u16 queue_index;
 144                 unsigned int ifindex;
 145         };
 146         struct list_head next;
 147         struct tun_struct *detached;
 148 };
 149
 150 struct tun_flow_entry {
 151         struct hlist_node hash_link;
 152         struct rcu_head rcu;
 153         struct tun_struct *tun;
 154
 155         u32 rxhash;
 156         u32 rps_rxhash;
 157         int queue_index;
 158         unsigned long updated;
 159 };
 160
 161 #define TUN_NUM_FLOW_ENTRIES 1024
 162
 163 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 164  * device, socket filter, sndbuf and vnet header size were restore when the
 165  * file were attached to a persist device.
 166  */
 167 struct tun_struct {
 168         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 169         unsigned int            numqueues;
 170         unsigned int            flags;
 171         kuid_t                  owner;
 172         kgid_t                  group;
 173
 174         struct net_device       *dev;
 175         netdev_features_t       set_features;
 176 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 177                           NETIF_F_TSO6|NETIF_F_UFO)
 178
 179         int                     vnet_hdr_sz;
 180         int                     sndbuf;
 181         struct tap_filter       txflt;
 182         struct sock_fprog       fprog;
 183         /* protected by rtnl lock */
 184         bool                    filter_attached;
 185 #ifdef TUN_DEBUG
 186         int debug;
 187 #endif
 188         spinlock_t lock;
 189         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 190         struct timer_list flow_gc_timer;
 191         unsigned long ageing_time;
 192         unsigned int numdisabled;
 193         struct list_head disabled;
 194         void *security;
 195         u32 flow_count;
 196 };
 197
 198 static inline u32 tun_hashfn(u32 rxhash)
 199 {
 200         return rxhash & 0x3ff;
 201 }
 202
 203 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 204 {
 205         struct tun_flow_entry *e;
 206
 207         hlist_for_each_entry_rcu(e, head, hash_link) {
 208                 if (e->rxhash == rxhash)
 209                         return e;
 210         }
 211         return NULL;
 212 }
 213
 214 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 215                                               struct hlist_head *head,
 216                                               u32 rxhash, u16 queue_index)
 217 {
 218         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 219
 220         if (e) {
 221                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 222                           rxhash, queue_index);
 223                 e->updated = jiffies;
 224                 e->rxhash = rxhash;
 225                 e->rps_rxhash = 0;
 226                 e->queue_index = queue_index;
 227                 e->tun = tun;
 228                 hlist_add_head_rcu(&e->hash_link, head);
 229                 ++tun->flow_count;
 230         }
 231         return e;
 232 }
 233
 234 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 235 {
 236         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 237                   e->rxhash, e->queue_index);
 238         sock_rps_reset_flow_hash(e->rps_rxhash);
 239         hlist_del_rcu(&e->hash_link);
 240         kfree_rcu(e, rcu);
 241         --tun->flow_count;
 242 }
 243
 244 static void tun_flow_flush(struct tun_struct *tun)
 245 {
 246         int i;
 247
 248         spin_lock_bh(&tun->lock);
 249         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 250                 struct tun_flow_entry *e;
 251                 struct hlist_node *n;
 252
 253                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 254                         tun_flow_delete(tun, e);
 255         }
 256         spin_unlock_bh(&tun->lock);
 257 }
 258
 259 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 260 {
 261         int i;
 262
 263         spin_lock_bh(&tun->lock);
 264         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 265                 struct tun_flow_entry *e;
 266                 struct hlist_node *n;
 267
 268                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 269                         if (e->queue_index == queue_index)
 270                                 tun_flow_delete(tun, e);
 271                 }
 272         }
 273         spin_unlock_bh(&tun->lock);
 274 }
 275
 276 static void tun_flow_cleanup(unsigned long data)
 277 {
 278         struct tun_struct *tun = (struct tun_struct *)data;
 279         unsigned long delay = tun->ageing_time;
 280         unsigned long next_timer = jiffies + delay;
 281         unsigned long count = 0;
 282         int i;
 283
 284         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 285
 286         spin_lock_bh(&tun->lock);
 287         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 288                 struct tun_flow_entry *e;
 289                 struct hlist_node *n;
 290
 291                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 292                         unsigned long this_timer;
 293                         count++;
 294                         this_timer = e->updated + delay;
 295                         if (time_before_eq(this_timer, jiffies))
 296                                 tun_flow_delete(tun, e);
 297                         else if (time_before(this_timer, next_timer))
 298                                 next_timer = this_timer;
 299                 }
 300         }
 301
 302         if (count)
 303                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 304         spin_unlock_bh(&tun->lock);
 305 }
 306
 307 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 308                             struct tun_file *tfile)
 309 {
 310         struct hlist_head *head;
 311         struct tun_flow_entry *e;
 312         unsigned long delay = tun->ageing_time;
 313         u16 queue_index = tfile->queue_index;
 314
 315         if (!rxhash)
 316                 return;
 317         else
 318                 head = &tun->flows[tun_hashfn(rxhash)];
 319
 320         rcu_read_lock();
 321
 322         /* We may get a very small possibility of OOO during switching, not
 323          * worth to optimize.*/
 324         if (tun->numqueues == 1 || tfile->detached)
 325                 goto unlock;
 326
 327         e = tun_flow_find(head, rxhash);
 328         if (likely(e)) {
 329                 /* TODO: keep queueing to old queue until it's empty? */
 330                 e->queue_index = queue_index;
 331                 e->updated = jiffies;
 332                 sock_rps_record_flow_hash(e->rps_rxhash);
 333         } else {
 334                 spin_lock_bh(&tun->lock);
 335                 if (!tun_flow_find(head, rxhash) &&
 336                     tun->flow_count < MAX_TAP_FLOWS)
 337                         tun_flow_create(tun, head, rxhash, queue_index);
 338
 339                 if (!timer_pending(&tun->flow_gc_timer))
 340                         mod_timer(&tun->flow_gc_timer,
 341                                   round_jiffies_up(jiffies + delay));
 342                 spin_unlock_bh(&tun->lock);
 343         }
 344
 345 unlock:
 346         rcu_read_unlock();
 347 }
 348
 349 /**
 350  * Save the hash received in the stack receive path and update the
 351  * flow_hash table accordingly.
 352  */
 353 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 354 {
 355         if (unlikely(e->rps_rxhash != hash)) {
 356                 sock_rps_reset_flow_hash(e->rps_rxhash);
 357                 e->rps_rxhash = hash;
 358         }
 359 }
 360
 361 /* We try to identify a flow through its rxhash first. The reason that
 362  * we do not check rxq no. is because some cards(e.g 82599), chooses
 363  * the rxq based on the txq where the last packet of the flow comes. As
 364  * the userspace application move between processors, we may get a
 365  * different rxq no. here. If we could not get rxhash, then we would
 366  * hope the rxq no. may help here.
 367  */
 368 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 369                             void *accel_priv, select_queue_fallback_t fallback)
 370 {
 371         struct tun_struct *tun = netdev_priv(dev);
 372         struct tun_flow_entry *e;
 373         u32 txq = 0;
 374         u32 numqueues = 0;
 375
 376         rcu_read_lock();
 377         numqueues = ACCESS_ONCE(tun->numqueues);
 378
 379         txq = skb_get_hash(skb);
 380         if (txq) {
 381                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 382                 if (e) {
 383                         tun_flow_save_rps_rxhash(e, txq);
 384                         txq = e->queue_index;
 385                 } else
 386                         /* use multiply and shift instead of expensive divide */
 387                         txq = ((u64)txq * numqueues) >> 32;
 388         } else if (likely(skb_rx_queue_recorded(skb))) {
 389                 txq = skb_get_rx_queue(skb);
 390                 while (unlikely(txq >= numqueues))
 391                         txq -= numqueues;
 392         }
 393
 394         rcu_read_unlock();
 395         return txq;
 396 }
 397
 398 static inline bool tun_not_capable(struct tun_struct *tun)
 399 {
 400         const struct cred *cred = current_cred();
 401         struct net *net = dev_net(tun->dev);
 402
 403         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 404                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 405                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 406 }
 407
 408 static void tun_set_real_num_queues(struct tun_struct *tun)
 409 {
 410         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 411         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 412 }
 413
 414 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 415 {
 416         tfile->detached = tun;
 417         list_add_tail(&tfile->next, &tun->disabled);
 418         ++tun->numdisabled;
 419 }
 420
 421 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 422 {
 423         struct tun_struct *tun = tfile->detached;
 424
 425         tfile->detached = NULL;
 426         list_del_init(&tfile->next);
 427         --tun->numdisabled;
 428         return tun;
 429 }
 430
 431 static void tun_queue_purge(struct tun_file *tfile)
 432 {
 433         skb_queue_purge(&tfile->sk.sk_receive_queue);
 434         skb_queue_purge(&tfile->sk.sk_error_queue);
 435 }
 436
 437 static void __tun_detach(struct tun_file *tfile, bool clean)
 438 {
 439         struct tun_file *ntfile;
 440         struct tun_struct *tun;
 441
 442         tun = rtnl_dereference(tfile->tun);
 443
 444         if (tun && !tfile->detached) {
 445                 u16 index = tfile->queue_index;
 446                 BUG_ON(index >= tun->numqueues);
 447
 448                 rcu_assign_pointer(tun->tfiles[index],
 449                                    tun->tfiles[tun->numqueues - 1]);
 450                 ntfile = rtnl_dereference(tun->tfiles[index]);
 451                 ntfile->queue_index = index;
 452
 453                 --tun->numqueues;
 454                 if (clean) {
 455                         RCU_INIT_POINTER(tfile->tun, NULL);
 456                         sock_put(&tfile->sk);
 457                 } else
 458                         tun_disable_queue(tun, tfile);
 459
 460                 synchronize_net();
 461                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 462                 /* Drop read queue */
 463                 tun_queue_purge(tfile);
 464                 tun_set_real_num_queues(tun);
 465         } else if (tfile->detached && clean) {
 466                 tun = tun_enable_queue(tfile);
 467                 sock_put(&tfile->sk);
 468         }
 469
 470         if (clean) {
 471                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 472                         netif_carrier_off(tun->dev);
 473
 474                         if (!(tun->flags & TUN_PERSIST) &&
 475                             tun->dev->reg_state == NETREG_REGISTERED)
 476                                 unregister_netdevice(tun->dev);
 477                 }
 478
 479                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 480                                  &tfile->socket.flags));
 481                 sk_release_kernel(&tfile->sk);
 482         }
 483 }
 484
 485 static void tun_detach(struct tun_file *tfile, bool clean)
 486 {
 487         rtnl_lock();
 488         __tun_detach(tfile, clean);
 489         rtnl_unlock();
 490 }
 491
 492 static void tun_detach_all(struct net_device *dev)
 493 {
 494         struct tun_struct *tun = netdev_priv(dev);
 495         struct tun_file *tfile, *tmp;
 496         int i, n = tun->numqueues;
 497
 498         for (i = 0; i < n; i++) {
 499                 tfile = rtnl_dereference(tun->tfiles[i]);
 500                 BUG_ON(!tfile);
 501                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 502                 RCU_INIT_POINTER(tfile->tun, NULL);
 503                 --tun->numqueues;
 504         }
 505         list_for_each_entry(tfile, &tun->disabled, next) {
 506                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 507                 RCU_INIT_POINTER(tfile->tun, NULL);
 508         }
 509         BUG_ON(tun->numqueues != 0);
 510
 511         synchronize_net();
 512         for (i = 0; i < n; i++) {
 513                 tfile = rtnl_dereference(tun->tfiles[i]);
 514                 /* Drop read queue */
 515                 tun_queue_purge(tfile);
 516                 sock_put(&tfile->sk);
 517         }
 518         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 519                 tun_enable_queue(tfile);
 520                 tun_queue_purge(tfile);
 521                 sock_put(&tfile->sk);
 522         }
 523         BUG_ON(tun->numdisabled != 0);
 524
 525         if (tun->flags & TUN_PERSIST)
 526                 module_put(THIS_MODULE);
 527 }
 528
 529 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 530 {
 531         struct tun_file *tfile = file->private_data;
 532         int err;
 533
 534         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 535         if (err < 0)
 536                 goto out;
 537
 538         err = -EINVAL;
 539         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 540                 goto out;
 541
 542         err = -EBUSY;
 543         if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 544                 goto out;
 545
 546         err = -E2BIG;
 547         if (!tfile->detached &&
 548             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 549                 goto out;
 550
 551         err = 0;
 552
 553         /* Re-attach the filter to persist device */
 554         if (!skip_filter && (tun->filter_attached == true)) {
 555                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 556                 if (!err)
 557                         goto out;
 558         }
 559         tfile->queue_index = tun->numqueues;
 560         rcu_assign_pointer(tfile->tun, tun);
 561         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 562         tun->numqueues++;
 563
 564         if (tfile->detached)
 565                 tun_enable_queue(tfile);
 566         else
 567                 sock_hold(&tfile->sk);
 568
 569         tun_set_real_num_queues(tun);
 570
 571         /* device is allowed to go away first, so no need to hold extra
 572          * refcnt.
 573          */
 574
 575 out:
 576         return err;
 577 }
 578
 579 static struct tun_struct *__tun_get(struct tun_file *tfile)
 580 {
 581         struct tun_struct *tun;
 582
 583         rcu_read_lock();
 584         tun = rcu_dereference(tfile->tun);
 585         if (tun)
 586                 dev_hold(tun->dev);
 587         rcu_read_unlock();
 588
 589         return tun;
 590 }
 591
 592 static struct tun_struct *tun_get(struct file *file)
 593 {
 594         return __tun_get(file->private_data);
 595 }
 596
 597 static void tun_put(struct tun_struct *tun)
 598 {
 599         dev_put(tun->dev);
 600 }
 601
 602 /* TAP filtering */
 603 static void addr_hash_set(u32 *mask, const u8 *addr)
 604 {
 605         int n = ether_crc(ETH_ALEN, addr) >> 26;
 606         mask[n >> 5] |= (1 << (n & 31));
 607 }
 608
 609 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 610 {
 611         int n = ether_crc(ETH_ALEN, addr) >> 26;
 612         return mask[n >> 5] & (1 << (n & 31));
 613 }
 614
 615 static int update_filter(struct tap_filter *filter, void __user *arg)
 616 {
 617         struct { u8 u[ETH_ALEN]; } *addr;
 618         struct tun_filter uf;
 619         int err, alen, n, nexact;
 620
 621         if (copy_from_user(&uf, arg, sizeof(uf)))
 622                 return -EFAULT;
 623
 624         if (!uf.count) {
 625                 /* Disabled */
 626                 filter->count = 0;
 627                 return 0;
 628         }
 629
 630         alen = ETH_ALEN * uf.count;
 631         addr = kmalloc(alen, GFP_KERNEL);
 632         if (!addr)
 633                 return -ENOMEM;
 634
 635         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 636                 err = -EFAULT;
 637                 goto done;
 638         }
 639
 640         /* The filter is updated without holding any locks. Which is
 641          * perfectly safe. We disable it first and in the worst
 642          * case we'll accept a few undesired packets. */
 643         filter->count = 0;
 644         wmb();
 645
 646         /* Use first set of addresses as an exact filter */
 647         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 648                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 649
 650         nexact = n;
 651
 652         /* Remaining multicast addresses are hashed,
 653          * unicast will leave the filter disabled. */
 654         memset(filter->mask, 0, sizeof(filter->mask));
 655         for (; n < uf.count; n++) {
 656                 if (!is_multicast_ether_addr(addr[n].u)) {
 657                         err = 0; /* no filter */
 658                         goto done;
 659                 }
 660                 addr_hash_set(filter->mask, addr[n].u);
 661         }
 662
 663         /* For ALLMULTI just set the mask to all ones.
 664          * This overrides the mask populated above. */
 665         if ((uf.flags & TUN_FLT_ALLMULTI))
 666                 memset(filter->mask, ~0, sizeof(filter->mask));
 667
 668         /* Now enable the filter */
 669         wmb();
 670         filter->count = nexact;
 671
 672         /* Return the number of exact filters */
 673         err = nexact;
 674
 675 done:
 676         kfree(addr);
 677         return err;
 678 }
 679
 680 /* Returns: 0 - drop, !=0 - accept */
 681 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 682 {
 683         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 684          * at this point. */
 685         struct ethhdr *eh = (struct ethhdr *) skb->data;
 686         int i;
 687
 688         /* Exact match */
 689         for (i = 0; i < filter->count; i++)
 690                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 691                         return 1;
 692
 693         /* Inexact match (multicast only) */
 694         if (is_multicast_ether_addr(eh->h_dest))
 695                 return addr_hash_test(filter->mask, eh->h_dest);
 696
 697         return 0;
 698 }
 699
 700 /*
 701  * Checks whether the packet is accepted or not.
 702  * Returns: 0 - drop, !=0 - accept
 703  */
 704 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 705 {
 706         if (!filter->count)
 707                 return 1;
 708
 709         return run_filter(filter, skb);
 710 }
 711
 712 /* Network device part of the driver */
 713
 714 static const struct ethtool_ops tun_ethtool_ops;
 715
 716 /* Net device detach from fd. */
 717 static void tun_net_uninit(struct net_device *dev)
 718 {
 719         tun_detach_all(dev);
 720 }
 721
 722 /* Net device open. */
 723 static int tun_net_open(struct net_device *dev)
 724 {
 725         netif_tx_start_all_queues(dev);
 726         return 0;
 727 }
 728
 729 /* Net device close. */
 730 static int tun_net_close(struct net_device *dev)
 731 {
 732         netif_tx_stop_all_queues(dev);
 733         return 0;
 734 }
 735
 736 /* Net device start xmit */
 737 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 738 {
 739         struct tun_struct *tun = netdev_priv(dev);
 740         int txq = skb->queue_mapping;
 741         struct tun_file *tfile;
 742         u32 numqueues = 0;
 743
 744         rcu_read_lock();
 745         tfile = rcu_dereference(tun->tfiles[txq]);
 746         numqueues = ACCESS_ONCE(tun->numqueues);
 747
 748         /* Drop packet if interface is not attached */
 749         if (txq >= numqueues)
 750                 goto drop;
 751
 752         if (numqueues == 1) {
 753                 /* Select queue was not called for the skbuff, so we extract the
 754                  * RPS hash and save it into the flow_table here.
 755                  */
 756                 __u32 rxhash;
 757
 758                 rxhash = skb_get_hash(skb);
 759                 if (rxhash) {
 760                         struct tun_flow_entry *e;
 761                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 762                                         rxhash);
 763                         if (e)
 764                                 tun_flow_save_rps_rxhash(e, rxhash);
 765                 }
 766         }
 767
 768         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 769
 770         BUG_ON(!tfile);
 771
 772         /* Drop if the filter does not like it.
 773          * This is a noop if the filter is disabled.
 774          * Filter can be enabled only for the TAP devices. */
 775         if (!check_filter(&tun->txflt, skb))
 776                 goto drop;
 777
 778         if (tfile->socket.sk->sk_filter &&
 779             sk_filter(tfile->socket.sk, skb))
 780                 goto drop;
 781
 782         /* Limit the number of packets queued by dividing txq length with the
 783          * number of queues.
 784          */
 785         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
 786                           >= dev->tx_queue_len)
 787                 goto drop;
 788
 789         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 790                 goto drop;
 791
 792         if (skb->sk) {
 793                 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
 794                 sw_tx_timestamp(skb);
 795         }
 796
 797         /* Orphan the skb - required as we might hang on to it
 798          * for indefinite time.
 799          */
 800         skb_orphan(skb);
 801
 802         nf_reset(skb);
 803
 804         /* Enqueue packet */
 805         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 806
 807         /* Notify and wake up reader process */
 808         if (tfile->flags & TUN_FASYNC)
 809                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 810         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 811
 812         rcu_read_unlock();
 813         return NETDEV_TX_OK;
 814
 815 drop:
 816         dev->stats.tx_dropped++;
 817         skb_tx_error(skb);
 818         kfree_skb(skb);
 819         rcu_read_unlock();
 820         return NETDEV_TX_OK;
 821 }
 822
 823 static void tun_net_mclist(struct net_device *dev)
 824 {
 825         /*
 826          * This callback is supposed to deal with mc filter in
 827          * _rx_ path and has nothing to do with the _tx_ path.
 828          * In rx path we always accept everything userspace gives us.
 829          */
 830 }
 831
 832 #define MIN_MTU 68
 833 #define MAX_MTU 65535
 834
 835 static int
 836 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 837 {
 838         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 839                 return -EINVAL;
 840         dev->mtu = new_mtu;
 841         return 0;
 842 }
 843
 844 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 845         netdev_features_t features)
 846 {
 847         struct tun_struct *tun = netdev_priv(dev);
 848
 849         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 850 }
 851 #ifdef CONFIG_NET_POLL_CONTROLLER
 852 static void tun_poll_controller(struct net_device *dev)
 853 {
 854         /*
 855          * Tun only receives frames when:
 856          * 1) the char device endpoint gets data from user space
 857          * 2) the tun socket gets a sendmsg call from user space
 858          * Since both of those are synchronous operations, we are guaranteed
 859          * never to have pending data when we poll for it
 860          * so there is nothing to do here but return.
 861          * We need this though so netpoll recognizes us as an interface that
 862          * supports polling, which enables bridge devices in virt setups to
 863          * still use netconsole
 864          */
 865         return;
 866 }
 867 #endif
 868 static const struct net_device_ops tun_netdev_ops = {
 869         .ndo_uninit             = tun_net_uninit,
 870         .ndo_open               = tun_net_open,
 871         .ndo_stop               = tun_net_close,
 872         .ndo_start_xmit         = tun_net_xmit,
 873         .ndo_change_mtu         = tun_net_change_mtu,
 874         .ndo_fix_features       = tun_net_fix_features,
 875         .ndo_select_queue       = tun_select_queue,
 876 #ifdef CONFIG_NET_POLL_CONTROLLER
 877         .ndo_poll_controller    = tun_poll_controller,
 878 #endif
 879 };
 880
 881 static const struct net_device_ops tap_netdev_ops = {
 882         .ndo_uninit             = tun_net_uninit,
 883         .ndo_open               = tun_net_open,
 884         .ndo_stop               = tun_net_close,
 885         .ndo_start_xmit         = tun_net_xmit,
 886         .ndo_change_mtu         = tun_net_change_mtu,
 887         .ndo_fix_features       = tun_net_fix_features,
 888         .ndo_set_rx_mode        = tun_net_mclist,
 889         .ndo_set_mac_address    = eth_mac_addr,
 890         .ndo_validate_addr      = eth_validate_addr,
 891         .ndo_select_queue       = tun_select_queue,
 892 #ifdef CONFIG_NET_POLL_CONTROLLER
 893         .ndo_poll_controller    = tun_poll_controller,
 894 #endif
 895 };
 896
 897 static void tun_flow_init(struct tun_struct *tun)
 898 {
 899         int i;
 900
 901         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 902                 INIT_HLIST_HEAD(&tun->flows[i]);
 903
 904         tun->ageing_time = TUN_FLOW_EXPIRE;
 905         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 906         mod_timer(&tun->flow_gc_timer,
 907                   round_jiffies_up(jiffies + tun->ageing_time));
 908 }
 909
 910 static void tun_flow_uninit(struct tun_struct *tun)
 911 {
 912         del_timer_sync(&tun->flow_gc_timer);
 913         tun_flow_flush(tun);
 914 }
 915
 916 /* Initialize net device. */
 917 static void tun_net_init(struct net_device *dev)
 918 {
 919         struct tun_struct *tun = netdev_priv(dev);
 920
 921         switch (tun->flags & TUN_TYPE_MASK) {
 922         case TUN_TUN_DEV:
 923                 dev->netdev_ops = &tun_netdev_ops;
 924
 925                 /* Point-to-Point TUN Device */
 926                 dev->hard_header_len = 0;
 927                 dev->addr_len = 0;
 928                 dev->mtu = 1500;
 929
 930                 /* Zero header length */
 931                 dev->type = ARPHRD_NONE;
 932                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 933                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 934                 break;
 935
 936         case TUN_TAP_DEV:
 937                 dev->netdev_ops = &tap_netdev_ops;
 938                 /* Ethernet TAP Device */
 939                 ether_setup(dev);
 940                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 941                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 942
 943                 eth_hw_addr_random(dev);
 944
 945                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 946                 break;
 947         }
 948 }
 949
 950 /* Character device part */
 951
 952 /* Poll */
 953 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 954 {
 955         struct tun_file *tfile = file->private_data;
 956         struct tun_struct *tun = __tun_get(tfile);
 957         struct sock *sk;
 958         unsigned int mask = 0;
 959
 960         if (!tun)
 961                 return POLLERR;
 962
 963         sk = tfile->socket.sk;
 964
 965         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 966
 967         poll_wait(file, sk_sleep(sk), wait);
 968
 969         if (!skb_queue_empty(&sk->sk_receive_queue))
 970                 mask |= POLLIN | POLLRDNORM;
 971
 972         if (sock_writeable(sk) ||
 973             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 974              sock_writeable(sk)))
 975                 mask |= POLLOUT | POLLWRNORM;
 976
 977         if (tun->dev->reg_state != NETREG_REGISTERED)
 978                 mask = POLLERR;
 979
 980         tun_put(tun);
 981         return mask;
 982 }
 983
 984 /* prepad is the amount to reserve at front.  len is length after that.
 985  * linear is a hint as to how much to copy (usually headers). */
 986 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 987                                      size_t prepad, size_t len,
 988                                      size_t linear, int noblock)
 989 {
 990         struct sock *sk = tfile->socket.sk;
 991         struct sk_buff *skb;
 992         int err;
 993
 994         /* Under a page?  Don't bother with paged skb. */
 995         if (prepad + len < PAGE_SIZE || !linear)
 996                 linear = len;
 997
 998         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 999                                    &err, 0);
1000         if (!skb)
1001                 return ERR_PTR(err);
1002
1003         skb_reserve(skb, prepad);
1004         skb_put(skb, linear);
1005         skb->data_len = len - linear;
1006         skb->len += len - linear;
1007
1008         return skb;
1009 }
1010
1011 /* Get packet from user space buffer */
1012 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1013                             void *msg_control, const struct iovec *iv,
1014                             size_t total_len, size_t count, int noblock)
1015 {
1016         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1017         struct sk_buff *skb;
1018         size_t len = total_len, align = NET_SKB_PAD, linear;
1019         struct virtio_net_hdr gso = { 0 };
1020         int good_linear;
1021         int offset = 0;
1022         int copylen;
1023         bool zerocopy = false;
1024         int err;
1025         u32 rxhash;
1026
1027         if (!(tun->flags & TUN_NO_PI)) {
1028                 if (len < sizeof(pi))
1029                         return -EINVAL;
1030                 len -= sizeof(pi);
1031
1032                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1033                         return -EFAULT;
1034                 offset += sizeof(pi);
1035         }
1036
1037         if (tun->flags & TUN_VNET_HDR) {
1038                 if (len < tun->vnet_hdr_sz)
1039                         return -EINVAL;
1040                 len -= tun->vnet_hdr_sz;
1041
1042                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1043                         return -EFAULT;
1044
1045                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1046                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1047                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1048
1049                 if (gso.hdr_len > len)
1050                         return -EINVAL;
1051                 offset += tun->vnet_hdr_sz;
1052         }
1053
1054         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1055                 align += NET_IP_ALIGN;
1056                 if (unlikely(len < ETH_HLEN ||
1057                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1058                         return -EINVAL;
1059         }
1060
1061         good_linear = SKB_MAX_HEAD(align);
1062
1063         if (msg_control) {
1064                 /* There are 256 bytes to be copied in skb, so there is
1065                  * enough room for skb expand head in case it is used.
1066                  * The rest of the buffer is mapped from userspace.
1067                  */
1068                 copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1069                 if (copylen > good_linear)
1070                         copylen = good_linear;
1071                 linear = copylen;
1072                 if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1073                         zerocopy = true;
1074         }
1075
1076         if (!zerocopy) {
1077                 copylen = len;
1078                 if (gso.hdr_len > good_linear)
1079                         linear = good_linear;
1080                 else
1081                         linear = gso.hdr_len;
1082         }
1083
1084         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1085         if (IS_ERR(skb)) {
1086                 if (PTR_ERR(skb) != -EAGAIN)
1087                         tun->dev->stats.rx_dropped++;
1088                 return PTR_ERR(skb);
1089         }
1090
1091         if (zerocopy)
1092                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1093         else {
1094                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1095                 if (!err && msg_control) {
1096                         struct ubuf_info *uarg = msg_control;
1097                         uarg->callback(uarg, false);
1098                 }
1099         }
1100
1101         if (err) {
1102                 tun->dev->stats.rx_dropped++;
1103                 kfree_skb(skb);
1104                 return -EFAULT;
1105         }
1106
1107         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1108                 if (!skb_partial_csum_set(skb, gso.csum_start,
1109                                           gso.csum_offset)) {
1110                         tun->dev->stats.rx_frame_errors++;
1111                         kfree_skb(skb);
1112                         return -EINVAL;
1113                 }
1114         }
1115
1116         switch (tun->flags & TUN_TYPE_MASK) {
1117         case TUN_TUN_DEV:
1118                 if (tun->flags & TUN_NO_PI) {
1119                         switch (skb->data[0] & 0xf0) {
1120                         case 0x40:
1121                                 pi.proto = htons(ETH_P_IP);
1122                                 break;
1123                         case 0x60:
1124                                 pi.proto = htons(ETH_P_IPV6);
1125                                 break;
1126                         default:
1127                                 tun->dev->stats.rx_dropped++;
1128                                 kfree_skb(skb);
1129                                 return -EINVAL;
1130                         }
1131                 }
1132
1133                 skb_reset_mac_header(skb);
1134                 skb->protocol = pi.proto;
1135                 skb->dev = tun->dev;
1136                 break;
1137         case TUN_TAP_DEV:
1138                 skb->protocol = eth_type_trans(skb, tun->dev);
1139                 break;
1140         }
1141
1142         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1143                 pr_debug("GSO!\n");
1144                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1145                 case VIRTIO_NET_HDR_GSO_TCPV4:
1146                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1147                         break;
1148                 case VIRTIO_NET_HDR_GSO_TCPV6:
1149                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1150                         break;
1151                 case VIRTIO_NET_HDR_GSO_UDP:
1152                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1153                         break;
1154                 default:
1155                         tun->dev->stats.rx_frame_errors++;
1156                         kfree_skb(skb);
1157                         return -EINVAL;
1158                 }
1159
1160                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1161                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1162
1163                 skb_shinfo(skb)->gso_size = gso.gso_size;
1164                 if (skb_shinfo(skb)->gso_size == 0) {
1165                         tun->dev->stats.rx_frame_errors++;
1166                         kfree_skb(skb);
1167                         return -EINVAL;
1168                 }
1169
1170                 /* Header must be checked, and gso_segs computed. */
1171                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1172                 skb_shinfo(skb)->gso_segs = 0;
1173         }
1174
1175         /* copy skb_ubuf_info for callback when skb has no error */
1176         if (zerocopy) {
1177                 skb_shinfo(skb)->destructor_arg = msg_control;
1178                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1179                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1180         }
1181
1182         skb_reset_network_header(skb);
1183         skb_probe_transport_header(skb, 0);
1184
1185         rxhash = skb_get_hash(skb);
1186         netif_rx_ni(skb);
1187
1188         tun->dev->stats.rx_packets++;
1189         tun->dev->stats.rx_bytes += len;
1190
1191         tun_flow_update(tun, rxhash, tfile);
1192         return total_len;
1193 }
1194
1195 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1196                               unsigned long count, loff_t pos)
1197 {
1198         struct file *file = iocb->ki_filp;
1199         struct tun_struct *tun = tun_get(file);
1200         struct tun_file *tfile = file->private_data;
1201         ssize_t result;
1202
1203         if (!tun)
1204                 return -EBADFD;
1205
1206         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1207
1208         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1209                               count, file->f_flags & O_NONBLOCK);
1210
1211         tun_put(tun);
1212         return result;
1213 }
1214
1215 /* Put packet to the user space buffer */
1216 static ssize_t tun_put_user(struct tun_struct *tun,
1217                             struct tun_file *tfile,
1218                             struct sk_buff *skb,
1219                             const struct iovec *iv, int len)
1220 {
1221         struct tun_pi pi = { 0, skb->protocol };
1222         ssize_t total = 0;
1223         int vlan_offset = 0, copied;
1224
1225         if (!(tun->flags & TUN_NO_PI)) {
1226                 if ((len -= sizeof(pi)) < 0)
1227                         return -EINVAL;
1228
1229                 if (len < skb->len) {
1230                         /* Packet will be striped */
1231                         pi.flags |= TUN_PKT_STRIP;
1232                 }
1233
1234                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1235                         return -EFAULT;
1236                 total += sizeof(pi);
1237         }
1238
1239         if (tun->flags & TUN_VNET_HDR) {
1240                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1241                 if ((len -= tun->vnet_hdr_sz) < 0)
1242                         return -EINVAL;
1243
1244                 if (skb_is_gso(skb)) {
1245                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1246
1247                         /* This is a hint as to how much should be linear. */
1248                         gso.hdr_len = skb_headlen(skb);
1249                         gso.gso_size = sinfo->gso_size;
1250                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1251                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1252                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1253                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1254                         else if (sinfo->gso_type & SKB_GSO_UDP)
1255                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1256                         else {
1257                                 pr_err("unexpected GSO type: "
1258                                        "0x%x, gso_size %d, hdr_len %d\n",
1259                                        sinfo->gso_type, gso.gso_size,
1260                                        gso.hdr_len);
1261                                 print_hex_dump(KERN_ERR, "tun: ",
1262                                                DUMP_PREFIX_NONE,
1263                                                16, 1, skb->head,
1264                                                min((int)gso.hdr_len, 64), true);
1265                                 WARN_ON_ONCE(1);
1266                                 return -EINVAL;
1267                         }
1268                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1269                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1270                 } else
1271                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1272
1273                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1274                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1275                         gso.csum_start = skb_checksum_start_offset(skb);
1276                         gso.csum_offset = skb->csum_offset;
1277                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1278                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1279                 } /* else everything is zero */
1280
1281                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1282                                                sizeof(gso))))
1283                         return -EFAULT;
1284                 total += tun->vnet_hdr_sz;
1285         }
1286
1287         copied = total;
1288         total += skb->len;
1289         if (!vlan_tx_tag_present(skb)) {
1290                 len = min_t(int, skb->len, len);
1291         } else {
1292                 int copy, ret;
1293                 struct {
1294                         __be16 h_vlan_proto;
1295                         __be16 h_vlan_TCI;
1296                 } veth;
1297
1298                 veth.h_vlan_proto = skb->vlan_proto;
1299                 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1300
1301                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1302                 len = min_t(int, skb->len + VLAN_HLEN, len);
1303                 total += VLAN_HLEN;
1304
1305                 copy = min_t(int, vlan_offset, len);
1306                 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
1307                 len -= copy;
1308                 copied += copy;
1309                 if (ret || !len)
1310                         goto done;
1311
1312                 copy = min_t(int, sizeof(veth), len);
1313                 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
1314                 len -= copy;
1315                 copied += copy;
1316                 if (ret || !len)
1317                         goto done;
1318         }
1319
1320         skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
1321
1322 done:
1323         tun->dev->stats.tx_packets++;
1324         tun->dev->stats.tx_bytes += len;
1325
1326         return total;
1327 }
1328
1329 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1330                            const struct iovec *iv, ssize_t len, int noblock)
1331 {
1332         struct sk_buff *skb;
1333         ssize_t ret = 0;
1334         int peeked, err, off = 0;
1335
1336         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1337
1338         if (!len)
1339                 return ret;
1340
1341         if (tun->dev->reg_state != NETREG_REGISTERED)
1342                 return -EIO;
1343
1344         /* Read frames from queue */
1345         skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1346                                   &peeked, &off, &err);
1347         if (skb) {
1348                 ret = tun_put_user(tun, tfile, skb, iv, len);
1349                 kfree_skb(skb);
1350         } else
1351                 ret = err;
1352
1353         return ret;
1354 }
1355
1356 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1357                             unsigned long count, loff_t pos)
1358 {
1359         struct file *file = iocb->ki_filp;
1360         struct tun_file *tfile = file->private_data;
1361         struct tun_struct *tun = __tun_get(tfile);
1362         ssize_t len, ret;
1363
1364         if (!tun)
1365                 return -EBADFD;
1366         len = iov_length(iv, count);
1367         if (len < 0) {
1368                 ret = -EINVAL;
1369                 goto out;
1370         }
1371
1372         ret = tun_do_read(tun, tfile, iv, len,
1373                           file->f_flags & O_NONBLOCK);
1374         ret = min_t(ssize_t, ret, len);
1375         if (ret > 0)
1376                 iocb->ki_pos = ret;
1377 out:
1378         tun_put(tun);
1379         return ret;
1380 }
1381
1382 static void tun_free_netdev(struct net_device *dev)
1383 {
1384         struct tun_struct *tun = netdev_priv(dev);
1385
1386         BUG_ON(!(list_empty(&tun->disabled)));
1387         tun_flow_uninit(tun);
1388         security_tun_dev_free_security(tun->security);
1389         free_netdev(dev);
1390 }
1391
1392 static void tun_setup(struct net_device *dev)
1393 {
1394         struct tun_struct *tun = netdev_priv(dev);
1395
1396         tun->owner = INVALID_UID;
1397         tun->group = INVALID_GID;
1398
1399         dev->ethtool_ops = &tun_ethtool_ops;
1400         dev->destructor = tun_free_netdev;
1401 }
1402
1403 /* Trivial set of netlink ops to allow deleting tun or tap
1404  * device with netlink.
1405  */
1406 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1407 {
1408         return -EINVAL;
1409 }
1410
1411 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1412         .kind           = DRV_NAME,
1413         .priv_size      = sizeof(struct tun_struct),
1414         .setup          = tun_setup,
1415         .validate       = tun_validate,
1416 };
1417
1418 static void tun_sock_write_space(struct sock *sk)
1419 {
1420         struct tun_file *tfile;
1421         wait_queue_head_t *wqueue;
1422
1423         if (!sock_writeable(sk))
1424                 return;
1425
1426         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1427                 return;
1428
1429         wqueue = sk_sleep(sk);
1430         if (wqueue && waitqueue_active(wqueue))
1431                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1432                                                 POLLWRNORM | POLLWRBAND);
1433
1434         tfile = container_of(sk, struct tun_file, sk);
1435         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1436 }
1437
1438 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1439                        struct msghdr *m, size_t total_len)
1440 {
1441         int ret;
1442         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1443         struct tun_struct *tun = __tun_get(tfile);
1444
1445         if (!tun)
1446                 return -EBADFD;
1447         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1448                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1449         tun_put(tun);
1450         return ret;
1451 }
1452
1453 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1454                        struct msghdr *m, size_t total_len,
1455                        int flags)
1456 {
1457         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1458         struct tun_struct *tun = __tun_get(tfile);
1459         int ret;
1460
1461         if (!tun)
1462                 return -EBADFD;
1463
1464         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1465                 ret = -EINVAL;
1466                 goto out;
1467         }
1468         if (flags & MSG_ERRQUEUE) {
1469                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1470                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1471                 goto out;
1472         }
1473         ret = tun_do_read(tun, tfile, m->msg_iov, total_len,
1474                           flags & MSG_DONTWAIT);
1475         if (ret > total_len) {
1476                 m->msg_flags |= MSG_TRUNC;
1477                 ret = flags & MSG_TRUNC ? ret : total_len;
1478         }
1479 out:
1480         tun_put(tun);
1481         return ret;
1482 }
1483
1484 static int tun_release(struct socket *sock)
1485 {
1486         if (sock->sk)
1487                 sock_put(sock->sk);
1488         return 0;
1489 }
1490
1491 /* Ops structure to mimic raw sockets with tun */
1492 static const struct proto_ops tun_socket_ops = {
1493         .sendmsg = tun_sendmsg,
1494         .recvmsg = tun_recvmsg,
1495         .release = tun_release,
1496 };
1497
1498 static struct proto tun_proto = {
1499         .name           = "tun",
1500         .owner          = THIS_MODULE,
1501         .obj_size       = sizeof(struct tun_file),
1502 };
1503
1504 static int tun_flags(struct tun_struct *tun)
1505 {
1506         int flags = 0;
1507
1508         if (tun->flags & TUN_TUN_DEV)
1509                 flags |= IFF_TUN;
1510         else
1511                 flags |= IFF_TAP;
1512
1513         if (tun->flags & TUN_NO_PI)
1514                 flags |= IFF_NO_PI;
1515
1516         /* This flag has no real effect.  We track the value for backwards
1517          * compatibility.
1518          */
1519         if (tun->flags & TUN_ONE_QUEUE)
1520                 flags |= IFF_ONE_QUEUE;
1521
1522         if (tun->flags & TUN_VNET_HDR)
1523                 flags |= IFF_VNET_HDR;
1524
1525         if (tun->flags & TUN_TAP_MQ)
1526                 flags |= IFF_MULTI_QUEUE;
1527
1528         if (tun->flags & TUN_PERSIST)
1529                 flags |= IFF_PERSIST;
1530
1531         return flags;
1532 }
1533
1534 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1535                               char *buf)
1536 {
1537         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1538         return sprintf(buf, "0x%x\n", tun_flags(tun));
1539 }
1540
1541 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1542                               char *buf)
1543 {
1544         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1545         return uid_valid(tun->owner)?
1546                 sprintf(buf, "%u\n",
1547                         from_kuid_munged(current_user_ns(), tun->owner)):
1548                 sprintf(buf, "-1\n");
1549 }
1550
1551 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1552                               char *buf)
1553 {
1554         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1555         return gid_valid(tun->group) ?
1556                 sprintf(buf, "%u\n",
1557                         from_kgid_munged(current_user_ns(), tun->group)):
1558                 sprintf(buf, "-1\n");
1559 }
1560
1561 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1562 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1563 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1564
1565 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1566 {
1567         struct tun_struct *tun;
1568         struct tun_file *tfile = file->private_data;
1569         struct net_device *dev;
1570         int err;
1571
1572         if (tfile->detached)
1573                 return -EINVAL;
1574
1575         dev = __dev_get_by_name(net, ifr->ifr_name);
1576         if (dev) {
1577                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1578                         return -EBUSY;
1579                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1580                         tun = netdev_priv(dev);
1581                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1582                         tun = netdev_priv(dev);
1583                 else
1584                         return -EINVAL;
1585
1586                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1587                     !!(tun->flags & TUN_TAP_MQ))
1588                         return -EINVAL;
1589
1590                 if (tun_not_capable(tun))
1591                         return -EPERM;
1592                 err = security_tun_dev_open(tun->security);
1593                 if (err < 0)
1594                         return err;
1595
1596                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1597                 if (err < 0)
1598                         return err;
1599
1600                 if (tun->flags & TUN_TAP_MQ &&
1601                     (tun->numqueues + tun->numdisabled > 1)) {
1602                         /* One or more queue has already been attached, no need
1603                          * to initialize the device again.
1604                          */
1605                         return 0;
1606                 }
1607         }
1608         else {
1609                 char *name;
1610                 unsigned long flags = 0;
1611                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1612                              MAX_TAP_QUEUES : 1;
1613
1614                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1615                         return -EPERM;
1616                 err = security_tun_dev_create();
1617                 if (err < 0)
1618                         return err;
1619
1620                 /* Set dev type */
1621                 if (ifr->ifr_flags & IFF_TUN) {
1622                         /* TUN device */
1623                         flags |= TUN_TUN_DEV;
1624                         name = "tun%d";
1625                 } else if (ifr->ifr_flags & IFF_TAP) {
1626                         /* TAP device */
1627                         flags |= TUN_TAP_DEV;
1628                         name = "tap%d";
1629                 } else
1630                         return -EINVAL;
1631
1632                 if (*ifr->ifr_name)
1633                         name = ifr->ifr_name;
1634
1635                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1636                                        NET_NAME_UNKNOWN, tun_setup, queues,
1637                                        queues);
1638
1639                 if (!dev)
1640                         return -ENOMEM;
1641
1642                 dev_net_set(dev, net);
1643                 dev->rtnl_link_ops = &tun_link_ops;
1644                 dev->ifindex = tfile->ifindex;
1645
1646                 tun = netdev_priv(dev);
1647                 tun->dev = dev;
1648                 tun->flags = flags;
1649                 tun->txflt.count = 0;
1650                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1651
1652                 tun->filter_attached = false;
1653                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1654
1655                 spin_lock_init(&tun->lock);
1656
1657                 err = security_tun_dev_alloc_security(&tun->security);
1658                 if (err < 0)
1659                         goto err_free_dev;
1660
1661                 tun_net_init(dev);
1662                 tun_flow_init(tun);
1663
1664                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1665                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1666                                    NETIF_F_HW_VLAN_STAG_TX;
1667                 dev->features = dev->hw_features;
1668                 dev->vlan_features = dev->features &
1669                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1670                                        NETIF_F_HW_VLAN_STAG_TX);
1671
1672                 INIT_LIST_HEAD(&tun->disabled);
1673                 err = tun_attach(tun, file, false);
1674                 if (err < 0)
1675                         goto err_free_flow;
1676
1677                 err = register_netdevice(tun->dev);
1678                 if (err < 0)
1679                         goto err_detach;
1680
1681                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1682                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1683                     device_create_file(&tun->dev->dev, &dev_attr_group))
1684                         pr_err("Failed to create tun sysfs files\n");
1685         }
1686
1687         netif_carrier_on(tun->dev);
1688
1689         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1690
1691         if (ifr->ifr_flags & IFF_NO_PI)
1692                 tun->flags |= TUN_NO_PI;
1693         else
1694                 tun->flags &= ~TUN_NO_PI;
1695
1696         /* This flag has no real effect.  We track the value for backwards
1697          * compatibility.
1698          */
1699         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1700                 tun->flags |= TUN_ONE_QUEUE;
1701         else
1702                 tun->flags &= ~TUN_ONE_QUEUE;
1703
1704         if (ifr->ifr_flags & IFF_VNET_HDR)
1705                 tun->flags |= TUN_VNET_HDR;
1706         else
1707                 tun->flags &= ~TUN_VNET_HDR;
1708
1709         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1710                 tun->flags |= TUN_TAP_MQ;
1711         else
1712                 tun->flags &= ~TUN_TAP_MQ;
1713
1714         /* Make sure persistent devices do not get stuck in
1715          * xoff state.
1716          */
1717         if (netif_running(tun->dev))
1718                 netif_tx_wake_all_queues(tun->dev);
1719
1720         strcpy(ifr->ifr_name, tun->dev->name);
1721         return 0;
1722
1723 err_detach:
1724         tun_detach_all(dev);
1725 err_free_flow:
1726         tun_flow_uninit(tun);
1727         security_tun_dev_free_security(tun->security);
1728 err_free_dev:
1729         free_netdev(dev);
1730         return err;
1731 }
1732
1733 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1734                        struct ifreq *ifr)
1735 {
1736         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1737
1738         strcpy(ifr->ifr_name, tun->dev->name);
1739
1740         ifr->ifr_flags = tun_flags(tun);
1741
1742 }
1743
1744 /* This is like a cut-down ethtool ops, except done via tun fd so no
1745  * privs required. */
1746 static int set_offload(struct tun_struct *tun, unsigned long arg)
1747 {
1748         netdev_features_t features = 0;
1749
1750         if (arg & TUN_F_CSUM) {
1751                 features |= NETIF_F_HW_CSUM;
1752                 arg &= ~TUN_F_CSUM;
1753
1754                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1755                         if (arg & TUN_F_TSO_ECN) {
1756                                 features |= NETIF_F_TSO_ECN;
1757                                 arg &= ~TUN_F_TSO_ECN;
1758                         }
1759                         if (arg & TUN_F_TSO4)
1760                                 features |= NETIF_F_TSO;
1761                         if (arg & TUN_F_TSO6)
1762                                 features |= NETIF_F_TSO6;
1763                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1764                 }
1765
1766                 if (arg & TUN_F_UFO) {
1767                         features |= NETIF_F_UFO;
1768                         arg &= ~TUN_F_UFO;
1769                 }
1770         }
1771
1772         /* This gives the user a way to test for new features in future by
1773          * trying to set them. */
1774         if (arg)
1775                 return -EINVAL;
1776
1777         tun->set_features = features;
1778         netdev_update_features(tun->dev);
1779
1780         return 0;
1781 }
1782
1783 static void tun_detach_filter(struct tun_struct *tun, int n)
1784 {
1785         int i;
1786         struct tun_file *tfile;
1787
1788         for (i = 0; i < n; i++) {
1789                 tfile = rtnl_dereference(tun->tfiles[i]);
1790                 sk_detach_filter(tfile->socket.sk);
1791         }
1792
1793         tun->filter_attached = false;
1794 }
1795
1796 static int tun_attach_filter(struct tun_struct *tun)
1797 {
1798         int i, ret = 0;
1799         struct tun_file *tfile;
1800
1801         for (i = 0; i < tun->numqueues; i++) {
1802                 tfile = rtnl_dereference(tun->tfiles[i]);
1803                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1804                 if (ret) {
1805                         tun_detach_filter(tun, i);
1806                         return ret;
1807                 }
1808         }
1809
1810         tun->filter_attached = true;
1811         return ret;
1812 }
1813
1814 static void tun_set_sndbuf(struct tun_struct *tun)
1815 {
1816         struct tun_file *tfile;
1817         int i;
1818
1819         for (i = 0; i < tun->numqueues; i++) {
1820                 tfile = rtnl_dereference(tun->tfiles[i]);
1821                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1822         }
1823 }
1824
1825 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1826 {
1827         struct tun_file *tfile = file->private_data;
1828         struct tun_struct *tun;
1829         int ret = 0;
1830
1831         rtnl_lock();
1832
1833         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1834                 tun = tfile->detached;
1835                 if (!tun) {
1836                         ret = -EINVAL;
1837                         goto unlock;
1838                 }
1839                 ret = security_tun_dev_attach_queue(tun->security);
1840                 if (ret < 0)
1841                         goto unlock;
1842                 ret = tun_attach(tun, file, false);
1843         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1844                 tun = rtnl_dereference(tfile->tun);
1845                 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1846                         ret = -EINVAL;
1847                 else
1848                         __tun_detach(tfile, false);
1849         } else
1850                 ret = -EINVAL;
1851
1852 unlock:
1853         rtnl_unlock();
1854         return ret;
1855 }
1856
1857 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1858                             unsigned long arg, int ifreq_len)
1859 {
1860         struct tun_file *tfile = file->private_data;
1861         struct tun_struct *tun;
1862         void __user* argp = (void __user*)arg;
1863         struct ifreq ifr;
1864         kuid_t owner;
1865         kgid_t group;
1866         int sndbuf;
1867         int vnet_hdr_sz;
1868         unsigned int ifindex;
1869         int ret;
1870
1871         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1872                 if (copy_from_user(&ifr, argp, ifreq_len))
1873                         return -EFAULT;
1874         } else {
1875                 memset(&ifr, 0, sizeof(ifr));
1876         }
1877         if (cmd == TUNGETFEATURES) {
1878                 /* Currently this just means: "what IFF flags are valid?".
1879                  * This is needed because we never checked for invalid flags on
1880                  * TUNSETIFF. */
1881                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1882                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1883                                 (unsigned int __user*)argp);
1884         } else if (cmd == TUNSETQUEUE)
1885                 return tun_set_queue(file, &ifr);
1886
1887         ret = 0;
1888         rtnl_lock();
1889
1890         tun = __tun_get(tfile);
1891         if (cmd == TUNSETIFF && !tun) {
1892                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1893
1894                 ret = tun_set_iff(tfile->net, file, &ifr);
1895
1896                 if (ret)
1897                         goto unlock;
1898
1899                 if (copy_to_user(argp, &ifr, ifreq_len))
1900                         ret = -EFAULT;
1901                 goto unlock;
1902         }
1903         if (cmd == TUNSETIFINDEX) {
1904                 ret = -EPERM;
1905                 if (tun)
1906                         goto unlock;
1907
1908                 ret = -EFAULT;
1909                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1910                         goto unlock;
1911
1912                 ret = 0;
1913                 tfile->ifindex = ifindex;
1914                 goto unlock;
1915         }
1916
1917         ret = -EBADFD;
1918         if (!tun)
1919                 goto unlock;
1920
1921         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1922
1923         ret = 0;
1924         switch (cmd) {
1925         case TUNGETIFF:
1926                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1927
1928                 if (tfile->detached)
1929                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
1930                 if (!tfile->socket.sk->sk_filter)
1931                         ifr.ifr_flags |= IFF_NOFILTER;
1932
1933                 if (copy_to_user(argp, &ifr, ifreq_len))
1934                         ret = -EFAULT;
1935                 break;
1936
1937         case TUNSETNOCSUM:
1938                 /* Disable/Enable checksum */
1939
1940                 /* [unimplemented] */
1941                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1942                           arg ? "disabled" : "enabled");
1943                 break;
1944
1945         case TUNSETPERSIST:
1946                 /* Disable/Enable persist mode. Keep an extra reference to the
1947                  * module to prevent the module being unprobed.
1948                  */
1949                 if (arg && !(tun->flags & TUN_PERSIST)) {
1950                         tun->flags |= TUN_PERSIST;
1951                         __module_get(THIS_MODULE);
1952                 }
1953                 if (!arg && (tun->flags & TUN_PERSIST)) {
1954                         tun->flags &= ~TUN_PERSIST;
1955                         module_put(THIS_MODULE);
1956                 }
1957
1958                 tun_debug(KERN_INFO, tun, "persist %s\n",
1959                           arg ? "enabled" : "disabled");
1960                 break;
1961
1962         case TUNSETOWNER:
1963                 /* Set owner of the device */
1964                 owner = make_kuid(current_user_ns(), arg);
1965                 if (!uid_valid(owner)) {
1966                         ret = -EINVAL;
1967                         break;
1968                 }
1969                 tun->owner = owner;
1970                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1971                           from_kuid(&init_user_ns, tun->owner));
1972                 break;
1973
1974         case TUNSETGROUP:
1975                 /* Set group of the device */
1976                 group = make_kgid(current_user_ns(), arg);
1977                 if (!gid_valid(group)) {
1978                         ret = -EINVAL;
1979                         break;
1980                 }
1981                 tun->group = group;
1982                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1983                           from_kgid(&init_user_ns, tun->group));
1984                 break;
1985
1986         case TUNSETLINK:
1987                 /* Only allow setting the type when the interface is down */
1988                 if (tun->dev->flags & IFF_UP) {
1989                         tun_debug(KERN_INFO, tun,
1990                                   "Linktype set failed because interface is up\n");
1991                         ret = -EBUSY;
1992                 } else {
1993                         tun->dev->type = (int) arg;
1994                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1995                                   tun->dev->type);
1996                         ret = 0;
1997                 }
1998                 break;
1999
2000 #ifdef TUN_DEBUG
2001         case TUNSETDEBUG:
2002                 tun->debug = arg;
2003                 break;
2004 #endif
2005         case TUNSETOFFLOAD:
2006                 ret = set_offload(tun, arg);
2007                 break;
2008
2009         case TUNSETTXFILTER:
2010                 /* Can be set only for TAPs */
2011                 ret = -EINVAL;
2012                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2013                         break;
2014                 ret = update_filter(&tun->txflt, (void __user *)arg);
2015                 break;
2016
2017         case SIOCGIFHWADDR:
2018                 /* Get hw address */
2019                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2020                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2021                 if (copy_to_user(argp, &ifr, ifreq_len))
2022                         ret = -EFAULT;
2023                 break;
2024
2025         case SIOCSIFHWADDR:
2026                 /* Set hw address */
2027                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2028                           ifr.ifr_hwaddr.sa_data);
2029
2030                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2031                 break;
2032
2033         case TUNGETSNDBUF:
2034                 sndbuf = tfile->socket.sk->sk_sndbuf;
2035                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2036                         ret = -EFAULT;
2037                 break;
2038
2039         case TUNSETSNDBUF:
2040                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2041                         ret = -EFAULT;
2042                         break;
2043                 }
2044
2045                 tun->sndbuf = sndbuf;
2046                 tun_set_sndbuf(tun);
2047                 break;
2048
2049         case TUNGETVNETHDRSZ:
2050                 vnet_hdr_sz = tun->vnet_hdr_sz;
2051                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2052                         ret = -EFAULT;
2053                 break;
2054
2055         case TUNSETVNETHDRSZ:
2056                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2057                         ret = -EFAULT;
2058                         break;
2059                 }
2060                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2061                         ret = -EINVAL;
2062                         break;
2063                 }
2064
2065                 tun->vnet_hdr_sz = vnet_hdr_sz;
2066                 break;
2067
2068         case TUNATTACHFILTER:
2069                 /* Can be set only for TAPs */
2070                 ret = -EINVAL;
2071                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2072                         break;
2073                 ret = -EFAULT;
2074                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2075                         break;
2076
2077                 ret = tun_attach_filter(tun);
2078                 break;
2079
2080         case TUNDETACHFILTER:
2081                 /* Can be set only for TAPs */
2082                 ret = -EINVAL;
2083                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2084                         break;
2085                 ret = 0;
2086                 tun_detach_filter(tun, tun->numqueues);
2087                 break;
2088
2089         case TUNGETFILTER:
2090                 ret = -EINVAL;
2091                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2092                         break;
2093                 ret = -EFAULT;
2094                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2095                         break;
2096                 ret = 0;
2097                 break;
2098
2099         default:
2100                 ret = -EINVAL;
2101                 break;
2102         }
2103
2104 unlock:
2105         rtnl_unlock();
2106         if (tun)
2107                 tun_put(tun);
2108         return ret;
2109 }
2110
2111 static long tun_chr_ioctl(struct file *file,
2112                           unsigned int cmd, unsigned long arg)
2113 {
2114         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2115 }
2116
2117 #ifdef CONFIG_COMPAT
2118 static long tun_chr_compat_ioctl(struct file *file,
2119                          unsigned int cmd, unsigned long arg)
2120 {
2121         switch (cmd) {
2122         case TUNSETIFF:
2123         case TUNGETIFF:
2124         case TUNSETTXFILTER:
2125         case TUNGETSNDBUF:
2126         case TUNSETSNDBUF:
2127         case SIOCGIFHWADDR:
2128         case SIOCSIFHWADDR:
2129                 arg = (unsigned long)compat_ptr(arg);
2130                 break;
2131         default:
2132                 arg = (compat_ulong_t)arg;
2133                 break;
2134         }
2135
2136         /*
2137          * compat_ifreq is shorter than ifreq, so we must not access beyond
2138          * the end of that structure. All fields that are used in this
2139          * driver are compatible though, we don't need to convert the
2140          * contents.
2141          */
2142         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2143 }
2144 #endif /* CONFIG_COMPAT */
2145
2146 static int tun_chr_fasync(int fd, struct file *file, int on)
2147 {
2148         struct tun_file *tfile = file->private_data;
2149         int ret;
2150
2151         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2152                 goto out;
2153
2154         if (on) {
2155                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2156                 tfile->flags |= TUN_FASYNC;
2157         } else
2158                 tfile->flags &= ~TUN_FASYNC;
2159         ret = 0;
2160 out:
2161         return ret;
2162 }
2163
2164 static int tun_chr_open(struct inode *inode, struct file * file)
2165 {
2166         struct tun_file *tfile;
2167
2168         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2169
2170         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2171                                             &tun_proto);
2172         if (!tfile)
2173                 return -ENOMEM;
2174         RCU_INIT_POINTER(tfile->tun, NULL);
2175         tfile->net = get_net(current->nsproxy->net_ns);
2176         tfile->flags = 0;
2177         tfile->ifindex = 0;
2178
2179         init_waitqueue_head(&tfile->wq.wait);
2180         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2181
2182         tfile->socket.file = file;
2183         tfile->socket.ops = &tun_socket_ops;
2184
2185         sock_init_data(&tfile->socket, &tfile->sk);
2186         sk_change_net(&tfile->sk, tfile->net);
2187
2188         tfile->sk.sk_write_space = tun_sock_write_space;
2189         tfile->sk.sk_sndbuf = INT_MAX;
2190
2191         file->private_data = tfile;
2192         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2193         INIT_LIST_HEAD(&tfile->next);
2194
2195         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2196
2197         return 0;
2198 }
2199
2200 static int tun_chr_close(struct inode *inode, struct file *file)
2201 {
2202         struct tun_file *tfile = file->private_data;
2203         struct net *net = tfile->net;
2204
2205         tun_detach(tfile, true);
2206         put_net(net);
2207
2208         return 0;
2209 }
2210
2211 #ifdef CONFIG_PROC_FS
2212 static int tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2213 {
2214         struct tun_struct *tun;
2215         struct ifreq ifr;
2216
2217         memset(&ifr, 0, sizeof(ifr));
2218
2219         rtnl_lock();
2220         tun = tun_get(f);
2221         if (tun)
2222                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2223         rtnl_unlock();
2224
2225         if (tun)
2226                 tun_put(tun);
2227
2228         return seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2229 }
2230 #endif
2231
2232 static const struct file_operations tun_fops = {
2233         .owner  = THIS_MODULE,
2234         .llseek = no_llseek,
2235         .read  = do_sync_read,
2236         .aio_read  = tun_chr_aio_read,
2237         .write = do_sync_write,
2238         .aio_write = tun_chr_aio_write,
2239         .poll   = tun_chr_poll,
2240         .unlocked_ioctl = tun_chr_ioctl,
2241 #ifdef CONFIG_COMPAT
2242         .compat_ioctl = tun_chr_compat_ioctl,
2243 #endif
2244         .open   = tun_chr_open,
2245         .release = tun_chr_close,
2246         .fasync = tun_chr_fasync,
2247 #ifdef CONFIG_PROC_FS
2248         .show_fdinfo = tun_chr_show_fdinfo,
2249 #endif
2250 };
2251
2252 static struct miscdevice tun_miscdev = {
2253         .minor = TUN_MINOR,
2254         .name = "tun",
2255         .nodename = "net/tun",
2256         .fops = &tun_fops,
2257 };
2258
2259 /* ethtool interface */
2260
2261 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2262 {
2263         cmd->supported          = 0;
2264         cmd->advertising        = 0;
2265         ethtool_cmd_speed_set(cmd, SPEED_10);
2266         cmd->duplex             = DUPLEX_FULL;
2267         cmd->port               = PORT_TP;
2268         cmd->phy_address        = 0;
2269         cmd->transceiver        = XCVR_INTERNAL;
2270         cmd->autoneg            = AUTONEG_DISABLE;
2271         cmd->maxtxpkt           = 0;
2272         cmd->maxrxpkt           = 0;
2273         return 0;
2274 }
2275
2276 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2277 {
2278         struct tun_struct *tun = netdev_priv(dev);
2279
2280         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2281         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2282
2283         switch (tun->flags & TUN_TYPE_MASK) {
2284         case TUN_TUN_DEV:
2285                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2286                 break;
2287         case TUN_TAP_DEV:
2288                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2289                 break;
2290         }
2291 }
2292
2293 static u32 tun_get_msglevel(struct net_device *dev)
2294 {
2295 #ifdef TUN_DEBUG
2296         struct tun_struct *tun = netdev_priv(dev);
2297         return tun->debug;
2298 #else
2299         return -EOPNOTSUPP;
2300 #endif
2301 }
2302
2303 static void tun_set_msglevel(struct net_device *dev, u32 value)
2304 {
2305 #ifdef TUN_DEBUG
2306         struct tun_struct *tun = netdev_priv(dev);
2307         tun->debug = value;
2308 #endif
2309 }
2310
2311 static const struct ethtool_ops tun_ethtool_ops = {
2312         .get_settings   = tun_get_settings,
2313         .get_drvinfo    = tun_get_drvinfo,
2314         .get_msglevel   = tun_get_msglevel,
2315         .set_msglevel   = tun_set_msglevel,
2316         .get_link       = ethtool_op_get_link,
2317         .get_ts_info    = ethtool_op_get_ts_info,
2318 };
2319
2320
2321 static int __init tun_init(void)
2322 {
2323         int ret = 0;
2324
2325         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2326         pr_info("%s\n", DRV_COPYRIGHT);
2327
2328         ret = rtnl_link_register(&tun_link_ops);
2329         if (ret) {
2330                 pr_err("Can't register link_ops\n");
2331                 goto err_linkops;
2332         }
2333
2334         ret = misc_register(&tun_miscdev);
2335         if (ret) {
2336                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2337                 goto err_misc;
2338         }
2339         return  0;
2340 err_misc:
2341         rtnl_link_unregister(&tun_link_ops);
2342 err_linkops:
2343         return ret;
2344 }
2345
2346 static void tun_cleanup(void)
2347 {
2348         misc_deregister(&tun_miscdev);
2349         rtnl_link_unregister(&tun_link_ops);
2350 }
2351
2352 /* Get an underlying socket object from tun file.  Returns error unless file is
2353  * attached to a device.  The returned object works like a packet socket, it
2354  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2355  * holding a reference to the file for as long as the socket is in use. */
2356 struct socket *tun_get_socket(struct file *file)
2357 {
2358         struct tun_file *tfile;
2359         if (file->f_op != &tun_fops)
2360                 return ERR_PTR(-EINVAL);
2361         tfile = file->private_data;
2362         if (!tfile)
2363                 return ERR_PTR(-EBADFD);
2364         return &tfile->socket;
2365 }
2366 EXPORT_SYMBOL_GPL(tun_get_socket);
2367
2368 module_init(tun_init);
2369 module_exit(tun_cleanup);
2370 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2371 MODULE_AUTHOR(DRV_COPYRIGHT);
2372 MODULE_LICENSE("GPL");
2373 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2374 MODULE_ALIAS("devname:net/tun");