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
  73 #include <asm/uaccess.h>
  74
  75 /* Uncomment to enable debugging */
  76 /* #define TUN_DEBUG 1 */
  77
  78 #ifdef TUN_DEBUG
  79 static int debug;
  80
  81 #define tun_debug(level, tun, fmt, args...)                     \
  82 do {                                                            \
  83         if (tun->debug)                                         \
  84                 netdev_printk(level, tun->dev, fmt, ##args);    \
  85 } while (0)
  86 #define DBG1(level, fmt, args...)                               \
  87 do {                                                            \
  88         if (debug == 2)                                         \
  89                 printk(level fmt, ##args);                      \
  90 } while (0)
  91 #else
  92 #define tun_debug(level, tun, fmt, args...)                     \
  93 do {                                                            \
  94         if (0)                                                  \
  95                 netdev_printk(level, tun->dev, fmt, ##args);    \
  96 } while (0)
  97 #define DBG1(level, fmt, args...)                               \
  98 do {                                                            \
  99         if (0)                                                  \
 100                 printk(level fmt, ##args);                      \
 101 } while (0)
 102 #endif
 103
 104 #define GOODCOPY_LEN 128
 105
 106 #define FLT_EXACT_COUNT 8
 107 struct tap_filter {
 108         unsigned int    count;    /* Number of addrs. Zero means disabled */
 109         u32             mask[2];  /* Mask of the hashed addrs */
 110         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 111 };
 112
 113 /* DEFAULT_MAX_NUM_RSS_QUEUES were chosen to let the rx/tx queues allocated for
 114  * the netdevice to be fit in one page. So we can make sure the success of
 115  * memory allocation. TODO: increase the limit. */
 116 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
 117 #define MAX_TAP_FLOWS  4096
 118
 119 #define TUN_FLOW_EXPIRE (3 * HZ)
 120
 121 /* A tun_file connects an open character device to a tuntap netdevice. It
 122  * also contains all socket related structures (except sock_fprog and tap_filter)
 123  * to serve as one transmit queue for tuntap device. The sock_fprog and
 124  * tap_filter were kept in tun_struct since they were used for filtering for the
 125  * netdevice not for a specific queue (at least I didn't see the requirement for
 126  * this).
 127  *
 128  * RCU usage:
 129  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 130  * other can only be read while rcu_read_lock or rtnl_lock is held.
 131  */
 132 struct tun_file {
 133         struct sock sk;
 134         struct socket socket;
 135         struct socket_wq wq;
 136         struct tun_struct __rcu *tun;
 137         struct net *net;
 138         struct fasync_struct *fasync;
 139         /* only used for fasnyc */
 140         unsigned int flags;
 141         union {
 142                 u16 queue_index;
 143                 unsigned int ifindex;
 144         };
 145         struct list_head next;
 146         struct tun_struct *detached;
 147 };
 148
 149 struct tun_flow_entry {
 150         struct hlist_node hash_link;
 151         struct rcu_head rcu;
 152         struct tun_struct *tun;
 153
 154         u32 rxhash;
 155         u32 rps_rxhash;
 156         int queue_index;
 157         unsigned long updated;
 158 };
 159
 160 #define TUN_NUM_FLOW_ENTRIES 1024
 161
 162 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 163  * device, socket filter, sndbuf and vnet header size were restore when the
 164  * file were attached to a persist device.
 165  */
 166 struct tun_struct {
 167         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 168         unsigned int            numqueues;
 169         unsigned int            flags;
 170         kuid_t                  owner;
 171         kgid_t                  group;
 172
 173         struct net_device       *dev;
 174         netdev_features_t       set_features;
 175 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 176                           NETIF_F_TSO6|NETIF_F_UFO)
 177
 178         int                     vnet_hdr_sz;
 179         int                     sndbuf;
 180         struct tap_filter       txflt;
 181         struct sock_fprog       fprog;
 182         /* protected by rtnl lock */
 183         bool                    filter_attached;
 184 #ifdef TUN_DEBUG
 185         int debug;
 186 #endif
 187         spinlock_t lock;
 188         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 189         struct timer_list flow_gc_timer;
 190         unsigned long ageing_time;
 191         unsigned int numdisabled;
 192         struct list_head disabled;
 193         void *security;
 194         u32 flow_count;
 195 };
 196
 197 static inline u32 tun_hashfn(u32 rxhash)
 198 {
 199         return rxhash & 0x3ff;
 200 }
 201
 202 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 203 {
 204         struct tun_flow_entry *e;
 205
 206         hlist_for_each_entry_rcu(e, head, hash_link) {
 207                 if (e->rxhash == rxhash)
 208                         return e;
 209         }
 210         return NULL;
 211 }
 212
 213 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 214                                               struct hlist_head *head,
 215                                               u32 rxhash, u16 queue_index)
 216 {
 217         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 218
 219         if (e) {
 220                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 221                           rxhash, queue_index);
 222                 e->updated = jiffies;
 223                 e->rxhash = rxhash;
 224                 e->rps_rxhash = 0;
 225                 e->queue_index = queue_index;
 226                 e->tun = tun;
 227                 hlist_add_head_rcu(&e->hash_link, head);
 228                 ++tun->flow_count;
 229         }
 230         return e;
 231 }
 232
 233 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 234 {
 235         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 236                   e->rxhash, e->queue_index);
 237         sock_rps_reset_flow_hash(e->rps_rxhash);
 238         hlist_del_rcu(&e->hash_link);
 239         kfree_rcu(e, rcu);
 240         --tun->flow_count;
 241 }
 242
 243 static void tun_flow_flush(struct tun_struct *tun)
 244 {
 245         int i;
 246
 247         spin_lock_bh(&tun->lock);
 248         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 249                 struct tun_flow_entry *e;
 250                 struct hlist_node *n;
 251
 252                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 253                         tun_flow_delete(tun, e);
 254         }
 255         spin_unlock_bh(&tun->lock);
 256 }
 257
 258 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 259 {
 260         int i;
 261
 262         spin_lock_bh(&tun->lock);
 263         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 264                 struct tun_flow_entry *e;
 265                 struct hlist_node *n;
 266
 267                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 268                         if (e->queue_index == queue_index)
 269                                 tun_flow_delete(tun, e);
 270                 }
 271         }
 272         spin_unlock_bh(&tun->lock);
 273 }
 274
 275 static void tun_flow_cleanup(unsigned long data)
 276 {
 277         struct tun_struct *tun = (struct tun_struct *)data;
 278         unsigned long delay = tun->ageing_time;
 279         unsigned long next_timer = jiffies + delay;
 280         unsigned long count = 0;
 281         int i;
 282
 283         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 284
 285         spin_lock_bh(&tun->lock);
 286         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 287                 struct tun_flow_entry *e;
 288                 struct hlist_node *n;
 289
 290                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 291                         unsigned long this_timer;
 292                         count++;
 293                         this_timer = e->updated + delay;
 294                         if (time_before_eq(this_timer, jiffies))
 295                                 tun_flow_delete(tun, e);
 296                         else if (time_before(this_timer, next_timer))
 297                                 next_timer = this_timer;
 298                 }
 299         }
 300
 301         if (count)
 302                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 303         spin_unlock_bh(&tun->lock);
 304 }
 305
 306 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 307                             struct tun_file *tfile)
 308 {
 309         struct hlist_head *head;
 310         struct tun_flow_entry *e;
 311         unsigned long delay = tun->ageing_time;
 312         u16 queue_index = tfile->queue_index;
 313
 314         if (!rxhash)
 315                 return;
 316         else
 317                 head = &tun->flows[tun_hashfn(rxhash)];
 318
 319         rcu_read_lock();
 320
 321         /* We may get a very small possibility of OOO during switching, not
 322          * worth to optimize.*/
 323         if (tun->numqueues == 1 || tfile->detached)
 324                 goto unlock;
 325
 326         e = tun_flow_find(head, rxhash);
 327         if (likely(e)) {
 328                 /* TODO: keep queueing to old queue until it's empty? */
 329                 e->queue_index = queue_index;
 330                 e->updated = jiffies;
 331                 sock_rps_record_flow_hash(e->rps_rxhash);
 332         } else {
 333                 spin_lock_bh(&tun->lock);
 334                 if (!tun_flow_find(head, rxhash) &&
 335                     tun->flow_count < MAX_TAP_FLOWS)
 336                         tun_flow_create(tun, head, rxhash, queue_index);
 337
 338                 if (!timer_pending(&tun->flow_gc_timer))
 339                         mod_timer(&tun->flow_gc_timer,
 340                                   round_jiffies_up(jiffies + delay));
 341                 spin_unlock_bh(&tun->lock);
 342         }
 343
 344 unlock:
 345         rcu_read_unlock();
 346 }
 347
 348 /**
 349  * Save the hash received in the stack receive path and update the
 350  * flow_hash table accordingly.
 351  */
 352 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 353 {
 354         if (unlikely(e->rps_rxhash != hash)) {
 355                 sock_rps_reset_flow_hash(e->rps_rxhash);
 356                 e->rps_rxhash = hash;
 357         }
 358 }
 359
 360 /* We try to identify a flow through its rxhash first. The reason that
 361  * we do not check rxq no. is because some cards(e.g 82599), chooses
 362  * the rxq based on the txq where the last packet of the flow comes. As
 363  * the userspace application move between processors, we may get a
 364  * different rxq no. here. If we could not get rxhash, then we would
 365  * hope the rxq no. may help here.
 366  */
 367 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 368                             void *accel_priv)
 369 {
 370         struct tun_struct *tun = netdev_priv(dev);
 371         struct tun_flow_entry *e;
 372         u32 txq = 0;
 373         u32 numqueues = 0;
 374
 375         rcu_read_lock();
 376         numqueues = ACCESS_ONCE(tun->numqueues);
 377
 378         txq = skb_get_hash(skb);
 379         if (txq) {
 380                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 381                 if (e) {
 382                         tun_flow_save_rps_rxhash(e, txq);
 383                         txq = e->queue_index;
 384                 } else
 385                         /* use multiply and shift instead of expensive divide */
 386                         txq = ((u64)txq * numqueues) >> 32;
 387         } else if (likely(skb_rx_queue_recorded(skb))) {
 388                 txq = skb_get_rx_queue(skb);
 389                 while (unlikely(txq >= numqueues))
 390                         txq -= numqueues;
 391         }
 392
 393         rcu_read_unlock();
 394         return txq;
 395 }
 396
 397 static inline bool tun_not_capable(struct tun_struct *tun)
 398 {
 399         const struct cred *cred = current_cred();
 400         struct net *net = dev_net(tun->dev);
 401
 402         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 403                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 404                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 405 }
 406
 407 static void tun_set_real_num_queues(struct tun_struct *tun)
 408 {
 409         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 410         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 411 }
 412
 413 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 414 {
 415         tfile->detached = tun;
 416         list_add_tail(&tfile->next, &tun->disabled);
 417         ++tun->numdisabled;
 418 }
 419
 420 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 421 {
 422         struct tun_struct *tun = tfile->detached;
 423
 424         tfile->detached = NULL;
 425         list_del_init(&tfile->next);
 426         --tun->numdisabled;
 427         return tun;
 428 }
 429
 430 static void tun_queue_purge(struct tun_file *tfile)
 431 {
 432         skb_queue_purge(&tfile->sk.sk_receive_queue);
 433         skb_queue_purge(&tfile->sk.sk_error_queue);
 434 }
 435
 436 static void __tun_detach(struct tun_file *tfile, bool clean)
 437 {
 438         struct tun_file *ntfile;
 439         struct tun_struct *tun;
 440
 441         tun = rtnl_dereference(tfile->tun);
 442
 443         if (tun && !tfile->detached) {
 444                 u16 index = tfile->queue_index;
 445                 BUG_ON(index >= tun->numqueues);
 446
 447                 rcu_assign_pointer(tun->tfiles[index],
 448                                    tun->tfiles[tun->numqueues - 1]);
 449                 ntfile = rtnl_dereference(tun->tfiles[index]);
 450                 ntfile->queue_index = index;
 451
 452                 --tun->numqueues;
 453                 if (clean) {
 454                         rcu_assign_pointer(tfile->tun, NULL);
 455                         sock_put(&tfile->sk);
 456                 } else
 457                         tun_disable_queue(tun, tfile);
 458
 459                 synchronize_net();
 460                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 461                 /* Drop read queue */
 462                 tun_queue_purge(tfile);
 463                 tun_set_real_num_queues(tun);
 464         } else if (tfile->detached && clean) {
 465                 tun = tun_enable_queue(tfile);
 466                 sock_put(&tfile->sk);
 467         }
 468
 469         if (clean) {
 470                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 471                         netif_carrier_off(tun->dev);
 472
 473                         if (!(tun->flags & TUN_PERSIST) &&
 474                             tun->dev->reg_state == NETREG_REGISTERED)
 475                                 unregister_netdevice(tun->dev);
 476                 }
 477
 478                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 479                                  &tfile->socket.flags));
 480                 sk_release_kernel(&tfile->sk);
 481         }
 482 }
 483
 484 static void tun_detach(struct tun_file *tfile, bool clean)
 485 {
 486         rtnl_lock();
 487         __tun_detach(tfile, clean);
 488         rtnl_unlock();
 489 }
 490
 491 static void tun_detach_all(struct net_device *dev)
 492 {
 493         struct tun_struct *tun = netdev_priv(dev);
 494         struct tun_file *tfile, *tmp;
 495         int i, n = tun->numqueues;
 496
 497         for (i = 0; i < n; i++) {
 498                 tfile = rtnl_dereference(tun->tfiles[i]);
 499                 BUG_ON(!tfile);
 500                 wake_up_all(&tfile->wq.wait);
 501                 rcu_assign_pointer(tfile->tun, NULL);
 502                 --tun->numqueues;
 503         }
 504         list_for_each_entry(tfile, &tun->disabled, next) {
 505                 wake_up_all(&tfile->wq.wait);
 506                 rcu_assign_pointer(tfile->tun, NULL);
 507         }
 508         BUG_ON(tun->numqueues != 0);
 509
 510         synchronize_net();
 511         for (i = 0; i < n; i++) {
 512                 tfile = rtnl_dereference(tun->tfiles[i]);
 513                 /* Drop read queue */
 514                 tun_queue_purge(tfile);
 515                 sock_put(&tfile->sk);
 516         }
 517         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 518                 tun_enable_queue(tfile);
 519                 tun_queue_purge(tfile);
 520                 sock_put(&tfile->sk);
 521         }
 522         BUG_ON(tun->numdisabled != 0);
 523
 524         if (tun->flags & TUN_PERSIST)
 525                 module_put(THIS_MODULE);
 526 }
 527
 528 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 529 {
 530         struct tun_file *tfile = file->private_data;
 531         int err;
 532
 533         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 534         if (err < 0)
 535                 goto out;
 536
 537         err = -EINVAL;
 538         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 539                 goto out;
 540
 541         err = -EBUSY;
 542         if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 543                 goto out;
 544
 545         err = -E2BIG;
 546         if (!tfile->detached &&
 547             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 548                 goto out;
 549
 550         err = 0;
 551
 552         /* Re-attach the filter to persist device */
 553         if (!skip_filter && (tun->filter_attached == true)) {
 554                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 555                 if (!err)
 556                         goto out;
 557         }
 558         tfile->queue_index = tun->numqueues;
 559         rcu_assign_pointer(tfile->tun, tun);
 560         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 561         tun->numqueues++;
 562
 563         if (tfile->detached)
 564                 tun_enable_queue(tfile);
 565         else
 566                 sock_hold(&tfile->sk);
 567
 568         tun_set_real_num_queues(tun);
 569
 570         /* device is allowed to go away first, so no need to hold extra
 571          * refcnt.
 572          */
 573
 574 out:
 575         return err;
 576 }
 577
 578 static struct tun_struct *__tun_get(struct tun_file *tfile)
 579 {
 580         struct tun_struct *tun;
 581
 582         rcu_read_lock();
 583         tun = rcu_dereference(tfile->tun);
 584         if (tun)
 585                 dev_hold(tun->dev);
 586         rcu_read_unlock();
 587
 588         return tun;
 589 }
 590
 591 static struct tun_struct *tun_get(struct file *file)
 592 {
 593         return __tun_get(file->private_data);
 594 }
 595
 596 static void tun_put(struct tun_struct *tun)
 597 {
 598         dev_put(tun->dev);
 599 }
 600
 601 /* TAP filtering */
 602 static void addr_hash_set(u32 *mask, const u8 *addr)
 603 {
 604         int n = ether_crc(ETH_ALEN, addr) >> 26;
 605         mask[n >> 5] |= (1 << (n & 31));
 606 }
 607
 608 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 609 {
 610         int n = ether_crc(ETH_ALEN, addr) >> 26;
 611         return mask[n >> 5] & (1 << (n & 31));
 612 }
 613
 614 static int update_filter(struct tap_filter *filter, void __user *arg)
 615 {
 616         struct { u8 u[ETH_ALEN]; } *addr;
 617         struct tun_filter uf;
 618         int err, alen, n, nexact;
 619
 620         if (copy_from_user(&uf, arg, sizeof(uf)))
 621                 return -EFAULT;
 622
 623         if (!uf.count) {
 624                 /* Disabled */
 625                 filter->count = 0;
 626                 return 0;
 627         }
 628
 629         alen = ETH_ALEN * uf.count;
 630         addr = kmalloc(alen, GFP_KERNEL);
 631         if (!addr)
 632                 return -ENOMEM;
 633
 634         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 635                 err = -EFAULT;
 636                 goto done;
 637         }
 638
 639         /* The filter is updated without holding any locks. Which is
 640          * perfectly safe. We disable it first and in the worst
 641          * case we'll accept a few undesired packets. */
 642         filter->count = 0;
 643         wmb();
 644
 645         /* Use first set of addresses as an exact filter */
 646         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 647                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 648
 649         nexact = n;
 650
 651         /* Remaining multicast addresses are hashed,
 652          * unicast will leave the filter disabled. */
 653         memset(filter->mask, 0, sizeof(filter->mask));
 654         for (; n < uf.count; n++) {
 655                 if (!is_multicast_ether_addr(addr[n].u)) {
 656                         err = 0; /* no filter */
 657                         goto done;
 658                 }
 659                 addr_hash_set(filter->mask, addr[n].u);
 660         }
 661
 662         /* For ALLMULTI just set the mask to all ones.
 663          * This overrides the mask populated above. */
 664         if ((uf.flags & TUN_FLT_ALLMULTI))
 665                 memset(filter->mask, ~0, sizeof(filter->mask));
 666
 667         /* Now enable the filter */
 668         wmb();
 669         filter->count = nexact;
 670
 671         /* Return the number of exact filters */
 672         err = nexact;
 673
 674 done:
 675         kfree(addr);
 676         return err;
 677 }
 678
 679 /* Returns: 0 - drop, !=0 - accept */
 680 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 681 {
 682         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 683          * at this point. */
 684         struct ethhdr *eh = (struct ethhdr *) skb->data;
 685         int i;
 686
 687         /* Exact match */
 688         for (i = 0; i < filter->count; i++)
 689                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 690                         return 1;
 691
 692         /* Inexact match (multicast only) */
 693         if (is_multicast_ether_addr(eh->h_dest))
 694                 return addr_hash_test(filter->mask, eh->h_dest);
 695
 696         return 0;
 697 }
 698
 699 /*
 700  * Checks whether the packet is accepted or not.
 701  * Returns: 0 - drop, !=0 - accept
 702  */
 703 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 704 {
 705         if (!filter->count)
 706                 return 1;
 707
 708         return run_filter(filter, skb);
 709 }
 710
 711 /* Network device part of the driver */
 712
 713 static const struct ethtool_ops tun_ethtool_ops;
 714
 715 /* Net device detach from fd. */
 716 static void tun_net_uninit(struct net_device *dev)
 717 {
 718         tun_detach_all(dev);
 719 }
 720
 721 /* Net device open. */
 722 static int tun_net_open(struct net_device *dev)
 723 {
 724         netif_tx_start_all_queues(dev);
 725         return 0;
 726 }
 727
 728 /* Net device close. */
 729 static int tun_net_close(struct net_device *dev)
 730 {
 731         netif_tx_stop_all_queues(dev);
 732         return 0;
 733 }
 734
 735 /* Net device start xmit */
 736 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 737 {
 738         struct tun_struct *tun = netdev_priv(dev);
 739         int txq = skb->queue_mapping;
 740         struct tun_file *tfile;
 741         u32 numqueues = 0;
 742
 743         rcu_read_lock();
 744         tfile = rcu_dereference(tun->tfiles[txq]);
 745         numqueues = ACCESS_ONCE(tun->numqueues);
 746
 747         /* Drop packet if interface is not attached */
 748         if (txq >= numqueues)
 749                 goto drop;
 750
 751         if (numqueues == 1) {
 752                 /* Select queue was not called for the skbuff, so we extract the
 753                  * RPS hash and save it into the flow_table here.
 754                  */
 755                 __u32 rxhash;
 756
 757                 rxhash = skb_get_hash(skb);
 758                 if (rxhash) {
 759                         struct tun_flow_entry *e;
 760                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 761                                         rxhash);
 762                         if (e)
 763                                 tun_flow_save_rps_rxhash(e, rxhash);
 764                 }
 765         }
 766
 767         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 768
 769         BUG_ON(!tfile);
 770
 771         /* Drop if the filter does not like it.
 772          * This is a noop if the filter is disabled.
 773          * Filter can be enabled only for the TAP devices. */
 774         if (!check_filter(&tun->txflt, skb))
 775                 goto drop;
 776
 777         if (tfile->socket.sk->sk_filter &&
 778             sk_filter(tfile->socket.sk, skb))
 779                 goto drop;
 780
 781         /* Limit the number of packets queued by dividing txq length with the
 782          * number of queues.
 783          */
 784         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
 785                           >= dev->tx_queue_len)
 786                 goto drop;
 787
 788         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 789                 goto drop;
 790
 791         if (skb->sk) {
 792                 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
 793                 sw_tx_timestamp(skb);
 794         }
 795
 796         /* Orphan the skb - required as we might hang on to it
 797          * for indefinite time.
 798          */
 799         skb_orphan(skb);
 800
 801         nf_reset(skb);
 802
 803         /* Enqueue packet */
 804         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 805
 806         /* Notify and wake up reader process */
 807         if (tfile->flags & TUN_FASYNC)
 808                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 809         wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
 810                                    POLLRDNORM | POLLRDBAND);
 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, &tfile->wq.wait, 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         DECLARE_WAITQUEUE(wait, current);
1333         struct sk_buff *skb;
1334         ssize_t ret = 0;
1335
1336         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1337
1338         if (unlikely(!noblock))
1339                 add_wait_queue(&tfile->wq.wait, &wait);
1340         while (len) {
1341                 if (unlikely(!noblock))
1342                         current->state = TASK_INTERRUPTIBLE;
1343
1344                 /* Read frames from the queue */
1345                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1346                         if (noblock) {
1347                                 ret = -EAGAIN;
1348                                 break;
1349                         }
1350                         if (signal_pending(current)) {
1351                                 ret = -ERESTARTSYS;
1352                                 break;
1353                         }
1354                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1355                                 ret = -EIO;
1356                                 break;
1357                         }
1358
1359                         /* Nothing to read, let's sleep */
1360                         schedule();
1361                         continue;
1362                 }
1363
1364                 ret = tun_put_user(tun, tfile, skb, iv, len);
1365                 kfree_skb(skb);
1366                 break;
1367         }
1368
1369         if (unlikely(!noblock)) {
1370                 current->state = TASK_RUNNING;
1371                 remove_wait_queue(&tfile->wq.wait, &wait);
1372         }
1373
1374         return ret;
1375 }
1376
1377 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1378                             unsigned long count, loff_t pos)
1379 {
1380         struct file *file = iocb->ki_filp;
1381         struct tun_file *tfile = file->private_data;
1382         struct tun_struct *tun = __tun_get(tfile);
1383         ssize_t len, ret;
1384
1385         if (!tun)
1386                 return -EBADFD;
1387         len = iov_length(iv, count);
1388         if (len < 0) {
1389                 ret = -EINVAL;
1390                 goto out;
1391         }
1392
1393         ret = tun_do_read(tun, tfile, iv, len,
1394                           file->f_flags & O_NONBLOCK);
1395         ret = min_t(ssize_t, ret, len);
1396         if (ret > 0)
1397                 iocb->ki_pos = ret;
1398 out:
1399         tun_put(tun);
1400         return ret;
1401 }
1402
1403 static void tun_free_netdev(struct net_device *dev)
1404 {
1405         struct tun_struct *tun = netdev_priv(dev);
1406
1407         BUG_ON(!(list_empty(&tun->disabled)));
1408         tun_flow_uninit(tun);
1409         security_tun_dev_free_security(tun->security);
1410         free_netdev(dev);
1411 }
1412
1413 static void tun_setup(struct net_device *dev)
1414 {
1415         struct tun_struct *tun = netdev_priv(dev);
1416
1417         tun->owner = INVALID_UID;
1418         tun->group = INVALID_GID;
1419
1420         dev->ethtool_ops = &tun_ethtool_ops;
1421         dev->destructor = tun_free_netdev;
1422 }
1423
1424 /* Trivial set of netlink ops to allow deleting tun or tap
1425  * device with netlink.
1426  */
1427 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1428 {
1429         return -EINVAL;
1430 }
1431
1432 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1433         .kind           = DRV_NAME,
1434         .priv_size      = sizeof(struct tun_struct),
1435         .setup          = tun_setup,
1436         .validate       = tun_validate,
1437 };
1438
1439 static void tun_sock_write_space(struct sock *sk)
1440 {
1441         struct tun_file *tfile;
1442         wait_queue_head_t *wqueue;
1443
1444         if (!sock_writeable(sk))
1445                 return;
1446
1447         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1448                 return;
1449
1450         wqueue = sk_sleep(sk);
1451         if (wqueue && waitqueue_active(wqueue))
1452                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1453                                                 POLLWRNORM | POLLWRBAND);
1454
1455         tfile = container_of(sk, struct tun_file, sk);
1456         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1457 }
1458
1459 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1460                        struct msghdr *m, size_t total_len)
1461 {
1462         int ret;
1463         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1464         struct tun_struct *tun = __tun_get(tfile);
1465
1466         if (!tun)
1467                 return -EBADFD;
1468         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1469                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1470         tun_put(tun);
1471         return ret;
1472 }
1473
1474 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1475                        struct msghdr *m, size_t total_len,
1476                        int flags)
1477 {
1478         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1479         struct tun_struct *tun = __tun_get(tfile);
1480         int ret;
1481
1482         if (!tun)
1483                 return -EBADFD;
1484
1485         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1486                 ret = -EINVAL;
1487                 goto out;
1488         }
1489         if (flags & MSG_ERRQUEUE) {
1490                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1491                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1492                 goto out;
1493         }
1494         ret = tun_do_read(tun, tfile, m->msg_iov, total_len,
1495                           flags & MSG_DONTWAIT);
1496         if (ret > total_len) {
1497                 m->msg_flags |= MSG_TRUNC;
1498                 ret = flags & MSG_TRUNC ? ret : total_len;
1499         }
1500 out:
1501         tun_put(tun);
1502         return ret;
1503 }
1504
1505 static int tun_release(struct socket *sock)
1506 {
1507         if (sock->sk)
1508                 sock_put(sock->sk);
1509         return 0;
1510 }
1511
1512 /* Ops structure to mimic raw sockets with tun */
1513 static const struct proto_ops tun_socket_ops = {
1514         .sendmsg = tun_sendmsg,
1515         .recvmsg = tun_recvmsg,
1516         .release = tun_release,
1517 };
1518
1519 static struct proto tun_proto = {
1520         .name           = "tun",
1521         .owner          = THIS_MODULE,
1522         .obj_size       = sizeof(struct tun_file),
1523 };
1524
1525 static int tun_flags(struct tun_struct *tun)
1526 {
1527         int flags = 0;
1528
1529         if (tun->flags & TUN_TUN_DEV)
1530                 flags |= IFF_TUN;
1531         else
1532                 flags |= IFF_TAP;
1533
1534         if (tun->flags & TUN_NO_PI)
1535                 flags |= IFF_NO_PI;
1536
1537         /* This flag has no real effect.  We track the value for backwards
1538          * compatibility.
1539          */
1540         if (tun->flags & TUN_ONE_QUEUE)
1541                 flags |= IFF_ONE_QUEUE;
1542
1543         if (tun->flags & TUN_VNET_HDR)
1544                 flags |= IFF_VNET_HDR;
1545
1546         if (tun->flags & TUN_TAP_MQ)
1547                 flags |= IFF_MULTI_QUEUE;
1548
1549         if (tun->flags & TUN_PERSIST)
1550                 flags |= IFF_PERSIST;
1551
1552         return flags;
1553 }
1554
1555 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1556                               char *buf)
1557 {
1558         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1559         return sprintf(buf, "0x%x\n", tun_flags(tun));
1560 }
1561
1562 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1563                               char *buf)
1564 {
1565         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1566         return uid_valid(tun->owner)?
1567                 sprintf(buf, "%u\n",
1568                         from_kuid_munged(current_user_ns(), tun->owner)):
1569                 sprintf(buf, "-1\n");
1570 }
1571
1572 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1573                               char *buf)
1574 {
1575         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1576         return gid_valid(tun->group) ?
1577                 sprintf(buf, "%u\n",
1578                         from_kgid_munged(current_user_ns(), tun->group)):
1579                 sprintf(buf, "-1\n");
1580 }
1581
1582 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1583 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1584 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1585
1586 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1587 {
1588         struct tun_struct *tun;
1589         struct tun_file *tfile = file->private_data;
1590         struct net_device *dev;
1591         int err;
1592
1593         if (tfile->detached)
1594                 return -EINVAL;
1595
1596         dev = __dev_get_by_name(net, ifr->ifr_name);
1597         if (dev) {
1598                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1599                         return -EBUSY;
1600                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1601                         tun = netdev_priv(dev);
1602                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1603                         tun = netdev_priv(dev);
1604                 else
1605                         return -EINVAL;
1606
1607                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1608                     !!(tun->flags & TUN_TAP_MQ))
1609                         return -EINVAL;
1610
1611                 if (tun_not_capable(tun))
1612                         return -EPERM;
1613                 err = security_tun_dev_open(tun->security);
1614                 if (err < 0)
1615                         return err;
1616
1617                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1618                 if (err < 0)
1619                         return err;
1620
1621                 if (tun->flags & TUN_TAP_MQ &&
1622                     (tun->numqueues + tun->numdisabled > 1)) {
1623                         /* One or more queue has already been attached, no need
1624                          * to initialize the device again.
1625                          */
1626                         return 0;
1627                 }
1628         }
1629         else {
1630                 char *name;
1631                 unsigned long flags = 0;
1632                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1633                              MAX_TAP_QUEUES : 1;
1634
1635                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1636                         return -EPERM;
1637                 err = security_tun_dev_create();
1638                 if (err < 0)
1639                         return err;
1640
1641                 /* Set dev type */
1642                 if (ifr->ifr_flags & IFF_TUN) {
1643                         /* TUN device */
1644                         flags |= TUN_TUN_DEV;
1645                         name = "tun%d";
1646                 } else if (ifr->ifr_flags & IFF_TAP) {
1647                         /* TAP device */
1648                         flags |= TUN_TAP_DEV;
1649                         name = "tap%d";
1650                 } else
1651                         return -EINVAL;
1652
1653                 if (*ifr->ifr_name)
1654                         name = ifr->ifr_name;
1655
1656                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1657                                        tun_setup, queues, queues);
1658
1659                 if (!dev)
1660                         return -ENOMEM;
1661
1662                 dev_net_set(dev, net);
1663                 dev->rtnl_link_ops = &tun_link_ops;
1664                 dev->ifindex = tfile->ifindex;
1665
1666                 tun = netdev_priv(dev);
1667                 tun->dev = dev;
1668                 tun->flags = flags;
1669                 tun->txflt.count = 0;
1670                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1671
1672                 tun->filter_attached = false;
1673                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1674
1675                 spin_lock_init(&tun->lock);
1676
1677                 err = security_tun_dev_alloc_security(&tun->security);
1678                 if (err < 0)
1679                         goto err_free_dev;
1680
1681                 tun_net_init(dev);
1682                 tun_flow_init(tun);
1683
1684                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1685                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1686                                    NETIF_F_HW_VLAN_STAG_TX;
1687                 dev->features = dev->hw_features;
1688                 dev->vlan_features = dev->features;
1689
1690                 INIT_LIST_HEAD(&tun->disabled);
1691                 err = tun_attach(tun, file, false);
1692                 if (err < 0)
1693                         goto err_free_flow;
1694
1695                 err = register_netdevice(tun->dev);
1696                 if (err < 0)
1697                         goto err_detach;
1698
1699                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1700                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1701                     device_create_file(&tun->dev->dev, &dev_attr_group))
1702                         pr_err("Failed to create tun sysfs files\n");
1703         }
1704
1705         netif_carrier_on(tun->dev);
1706
1707         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1708
1709         if (ifr->ifr_flags & IFF_NO_PI)
1710                 tun->flags |= TUN_NO_PI;
1711         else
1712                 tun->flags &= ~TUN_NO_PI;
1713
1714         /* This flag has no real effect.  We track the value for backwards
1715          * compatibility.
1716          */
1717         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1718                 tun->flags |= TUN_ONE_QUEUE;
1719         else
1720                 tun->flags &= ~TUN_ONE_QUEUE;
1721
1722         if (ifr->ifr_flags & IFF_VNET_HDR)
1723                 tun->flags |= TUN_VNET_HDR;
1724         else
1725                 tun->flags &= ~TUN_VNET_HDR;
1726
1727         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1728                 tun->flags |= TUN_TAP_MQ;
1729         else
1730                 tun->flags &= ~TUN_TAP_MQ;
1731
1732         /* Make sure persistent devices do not get stuck in
1733          * xoff state.
1734          */
1735         if (netif_running(tun->dev))
1736                 netif_tx_wake_all_queues(tun->dev);
1737
1738         strcpy(ifr->ifr_name, tun->dev->name);
1739         return 0;
1740
1741 err_detach:
1742         tun_detach_all(dev);
1743 err_free_flow:
1744         tun_flow_uninit(tun);
1745         security_tun_dev_free_security(tun->security);
1746 err_free_dev:
1747         free_netdev(dev);
1748         return err;
1749 }
1750
1751 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1752                        struct ifreq *ifr)
1753 {
1754         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1755
1756         strcpy(ifr->ifr_name, tun->dev->name);
1757
1758         ifr->ifr_flags = tun_flags(tun);
1759
1760 }
1761
1762 /* This is like a cut-down ethtool ops, except done via tun fd so no
1763  * privs required. */
1764 static int set_offload(struct tun_struct *tun, unsigned long arg)
1765 {
1766         netdev_features_t features = 0;
1767
1768         if (arg & TUN_F_CSUM) {
1769                 features |= NETIF_F_HW_CSUM;
1770                 arg &= ~TUN_F_CSUM;
1771
1772                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1773                         if (arg & TUN_F_TSO_ECN) {
1774                                 features |= NETIF_F_TSO_ECN;
1775                                 arg &= ~TUN_F_TSO_ECN;
1776                         }
1777                         if (arg & TUN_F_TSO4)
1778                                 features |= NETIF_F_TSO;
1779                         if (arg & TUN_F_TSO6)
1780                                 features |= NETIF_F_TSO6;
1781                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1782                 }
1783
1784                 if (arg & TUN_F_UFO) {
1785                         features |= NETIF_F_UFO;
1786                         arg &= ~TUN_F_UFO;
1787                 }
1788         }
1789
1790         /* This gives the user a way to test for new features in future by
1791          * trying to set them. */
1792         if (arg)
1793                 return -EINVAL;
1794
1795         tun->set_features = features;
1796         netdev_update_features(tun->dev);
1797
1798         return 0;
1799 }
1800
1801 static void tun_detach_filter(struct tun_struct *tun, int n)
1802 {
1803         int i;
1804         struct tun_file *tfile;
1805
1806         for (i = 0; i < n; i++) {
1807                 tfile = rtnl_dereference(tun->tfiles[i]);
1808                 sk_detach_filter(tfile->socket.sk);
1809         }
1810
1811         tun->filter_attached = false;
1812 }
1813
1814 static int tun_attach_filter(struct tun_struct *tun)
1815 {
1816         int i, ret = 0;
1817         struct tun_file *tfile;
1818
1819         for (i = 0; i < tun->numqueues; i++) {
1820                 tfile = rtnl_dereference(tun->tfiles[i]);
1821                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1822                 if (ret) {
1823                         tun_detach_filter(tun, i);
1824                         return ret;
1825                 }
1826         }
1827
1828         tun->filter_attached = true;
1829         return ret;
1830 }
1831
1832 static void tun_set_sndbuf(struct tun_struct *tun)
1833 {
1834         struct tun_file *tfile;
1835         int i;
1836
1837         for (i = 0; i < tun->numqueues; i++) {
1838                 tfile = rtnl_dereference(tun->tfiles[i]);
1839                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1840         }
1841 }
1842
1843 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1844 {
1845         struct tun_file *tfile = file->private_data;
1846         struct tun_struct *tun;
1847         int ret = 0;
1848
1849         rtnl_lock();
1850
1851         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1852                 tun = tfile->detached;
1853                 if (!tun) {
1854                         ret = -EINVAL;
1855                         goto unlock;
1856                 }
1857                 ret = security_tun_dev_attach_queue(tun->security);
1858                 if (ret < 0)
1859                         goto unlock;
1860                 ret = tun_attach(tun, file, false);
1861         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1862                 tun = rtnl_dereference(tfile->tun);
1863                 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1864                         ret = -EINVAL;
1865                 else
1866                         __tun_detach(tfile, false);
1867         } else
1868                 ret = -EINVAL;
1869
1870 unlock:
1871         rtnl_unlock();
1872         return ret;
1873 }
1874
1875 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1876                             unsigned long arg, int ifreq_len)
1877 {
1878         struct tun_file *tfile = file->private_data;
1879         struct tun_struct *tun;
1880         void __user* argp = (void __user*)arg;
1881         struct ifreq ifr;
1882         kuid_t owner;
1883         kgid_t group;
1884         int sndbuf;
1885         int vnet_hdr_sz;
1886         unsigned int ifindex;
1887         int ret;
1888
1889         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1890                 if (copy_from_user(&ifr, argp, ifreq_len))
1891                         return -EFAULT;
1892         } else {
1893                 memset(&ifr, 0, sizeof(ifr));
1894         }
1895         if (cmd == TUNGETFEATURES) {
1896                 /* Currently this just means: "what IFF flags are valid?".
1897                  * This is needed because we never checked for invalid flags on
1898                  * TUNSETIFF. */
1899                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1900                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1901                                 (unsigned int __user*)argp);
1902         } else if (cmd == TUNSETQUEUE)
1903                 return tun_set_queue(file, &ifr);
1904
1905         ret = 0;
1906         rtnl_lock();
1907
1908         tun = __tun_get(tfile);
1909         if (cmd == TUNSETIFF && !tun) {
1910                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1911
1912                 ret = tun_set_iff(tfile->net, file, &ifr);
1913
1914                 if (ret)
1915                         goto unlock;
1916
1917                 if (copy_to_user(argp, &ifr, ifreq_len))
1918                         ret = -EFAULT;
1919                 goto unlock;
1920         }
1921         if (cmd == TUNSETIFINDEX) {
1922                 ret = -EPERM;
1923                 if (tun)
1924                         goto unlock;
1925
1926                 ret = -EFAULT;
1927                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1928                         goto unlock;
1929
1930                 ret = 0;
1931                 tfile->ifindex = ifindex;
1932                 goto unlock;
1933         }
1934
1935         ret = -EBADFD;
1936         if (!tun)
1937                 goto unlock;
1938
1939         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1940
1941         ret = 0;
1942         switch (cmd) {
1943         case TUNGETIFF:
1944                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1945
1946                 if (tfile->detached)
1947                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
1948                 if (!tfile->socket.sk->sk_filter)
1949                         ifr.ifr_flags |= IFF_NOFILTER;
1950
1951                 if (copy_to_user(argp, &ifr, ifreq_len))
1952                         ret = -EFAULT;
1953                 break;
1954
1955         case TUNSETNOCSUM:
1956                 /* Disable/Enable checksum */
1957
1958                 /* [unimplemented] */
1959                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1960                           arg ? "disabled" : "enabled");
1961                 break;
1962
1963         case TUNSETPERSIST:
1964                 /* Disable/Enable persist mode. Keep an extra reference to the
1965                  * module to prevent the module being unprobed.
1966                  */
1967                 if (arg && !(tun->flags & TUN_PERSIST)) {
1968                         tun->flags |= TUN_PERSIST;
1969                         __module_get(THIS_MODULE);
1970                 }
1971                 if (!arg && (tun->flags & TUN_PERSIST)) {
1972                         tun->flags &= ~TUN_PERSIST;
1973                         module_put(THIS_MODULE);
1974                 }
1975
1976                 tun_debug(KERN_INFO, tun, "persist %s\n",
1977                           arg ? "enabled" : "disabled");
1978                 break;
1979
1980         case TUNSETOWNER:
1981                 /* Set owner of the device */
1982                 owner = make_kuid(current_user_ns(), arg);
1983                 if (!uid_valid(owner)) {
1984                         ret = -EINVAL;
1985                         break;
1986                 }
1987                 tun->owner = owner;
1988                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1989                           from_kuid(&init_user_ns, tun->owner));
1990                 break;
1991
1992         case TUNSETGROUP:
1993                 /* Set group of the device */
1994                 group = make_kgid(current_user_ns(), arg);
1995                 if (!gid_valid(group)) {
1996                         ret = -EINVAL;
1997                         break;
1998                 }
1999                 tun->group = group;
2000                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2001                           from_kgid(&init_user_ns, tun->group));
2002                 break;
2003
2004         case TUNSETLINK:
2005                 /* Only allow setting the type when the interface is down */
2006                 if (tun->dev->flags & IFF_UP) {
2007                         tun_debug(KERN_INFO, tun,
2008                                   "Linktype set failed because interface is up\n");
2009                         ret = -EBUSY;
2010                 } else {
2011                         tun->dev->type = (int) arg;
2012                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2013                                   tun->dev->type);
2014                         ret = 0;
2015                 }
2016                 break;
2017
2018 #ifdef TUN_DEBUG
2019         case TUNSETDEBUG:
2020                 tun->debug = arg;
2021                 break;
2022 #endif
2023         case TUNSETOFFLOAD:
2024                 ret = set_offload(tun, arg);
2025                 break;
2026
2027         case TUNSETTXFILTER:
2028                 /* Can be set only for TAPs */
2029                 ret = -EINVAL;
2030                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2031                         break;
2032                 ret = update_filter(&tun->txflt, (void __user *)arg);
2033                 break;
2034
2035         case SIOCGIFHWADDR:
2036                 /* Get hw address */
2037                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2038                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2039                 if (copy_to_user(argp, &ifr, ifreq_len))
2040                         ret = -EFAULT;
2041                 break;
2042
2043         case SIOCSIFHWADDR:
2044                 /* Set hw address */
2045                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2046                           ifr.ifr_hwaddr.sa_data);
2047
2048                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2049                 break;
2050
2051         case TUNGETSNDBUF:
2052                 sndbuf = tfile->socket.sk->sk_sndbuf;
2053                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2054                         ret = -EFAULT;
2055                 break;
2056
2057         case TUNSETSNDBUF:
2058                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2059                         ret = -EFAULT;
2060                         break;
2061                 }
2062
2063                 tun->sndbuf = sndbuf;
2064                 tun_set_sndbuf(tun);
2065                 break;
2066
2067         case TUNGETVNETHDRSZ:
2068                 vnet_hdr_sz = tun->vnet_hdr_sz;
2069                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2070                         ret = -EFAULT;
2071                 break;
2072
2073         case TUNSETVNETHDRSZ:
2074                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2075                         ret = -EFAULT;
2076                         break;
2077                 }
2078                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2079                         ret = -EINVAL;
2080                         break;
2081                 }
2082
2083                 tun->vnet_hdr_sz = vnet_hdr_sz;
2084                 break;
2085
2086         case TUNATTACHFILTER:
2087                 /* Can be set only for TAPs */
2088                 ret = -EINVAL;
2089                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2090                         break;
2091                 ret = -EFAULT;
2092                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2093                         break;
2094
2095                 ret = tun_attach_filter(tun);
2096                 break;
2097
2098         case TUNDETACHFILTER:
2099                 /* Can be set only for TAPs */
2100                 ret = -EINVAL;
2101                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2102                         break;
2103                 ret = 0;
2104                 tun_detach_filter(tun, tun->numqueues);
2105                 break;
2106
2107         case TUNGETFILTER:
2108                 ret = -EINVAL;
2109                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2110                         break;
2111                 ret = -EFAULT;
2112                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2113                         break;
2114                 ret = 0;
2115                 break;
2116
2117         default:
2118                 ret = -EINVAL;
2119                 break;
2120         }
2121
2122 unlock:
2123         rtnl_unlock();
2124         if (tun)
2125                 tun_put(tun);
2126         return ret;
2127 }
2128
2129 static long tun_chr_ioctl(struct file *file,
2130                           unsigned int cmd, unsigned long arg)
2131 {
2132         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2133 }
2134
2135 #ifdef CONFIG_COMPAT
2136 static long tun_chr_compat_ioctl(struct file *file,
2137                          unsigned int cmd, unsigned long arg)
2138 {
2139         switch (cmd) {
2140         case TUNSETIFF:
2141         case TUNGETIFF:
2142         case TUNSETTXFILTER:
2143         case TUNGETSNDBUF:
2144         case TUNSETSNDBUF:
2145         case SIOCGIFHWADDR:
2146         case SIOCSIFHWADDR:
2147                 arg = (unsigned long)compat_ptr(arg);
2148                 break;
2149         default:
2150                 arg = (compat_ulong_t)arg;
2151                 break;
2152         }
2153
2154         /*
2155          * compat_ifreq is shorter than ifreq, so we must not access beyond
2156          * the end of that structure. All fields that are used in this
2157          * driver are compatible though, we don't need to convert the
2158          * contents.
2159          */
2160         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2161 }
2162 #endif /* CONFIG_COMPAT */
2163
2164 static int tun_chr_fasync(int fd, struct file *file, int on)
2165 {
2166         struct tun_file *tfile = file->private_data;
2167         int ret;
2168
2169         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2170                 goto out;
2171
2172         if (on) {
2173                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2174                 if (ret)
2175                         goto out;
2176                 tfile->flags |= TUN_FASYNC;
2177         } else
2178                 tfile->flags &= ~TUN_FASYNC;
2179         ret = 0;
2180 out:
2181         return ret;
2182 }
2183
2184 static int tun_chr_open(struct inode *inode, struct file * file)
2185 {
2186         struct tun_file *tfile;
2187
2188         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2189
2190         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2191                                             &tun_proto);
2192         if (!tfile)
2193                 return -ENOMEM;
2194         rcu_assign_pointer(tfile->tun, NULL);
2195         tfile->net = get_net(current->nsproxy->net_ns);
2196         tfile->flags = 0;
2197         tfile->ifindex = 0;
2198
2199         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2200         init_waitqueue_head(&tfile->wq.wait);
2201
2202         tfile->socket.file = file;
2203         tfile->socket.ops = &tun_socket_ops;
2204
2205         sock_init_data(&tfile->socket, &tfile->sk);
2206         sk_change_net(&tfile->sk, tfile->net);
2207
2208         tfile->sk.sk_write_space = tun_sock_write_space;
2209         tfile->sk.sk_sndbuf = INT_MAX;
2210
2211         file->private_data = tfile;
2212         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2213         INIT_LIST_HEAD(&tfile->next);
2214
2215         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2216
2217         return 0;
2218 }
2219
2220 static int tun_chr_close(struct inode *inode, struct file *file)
2221 {
2222         struct tun_file *tfile = file->private_data;
2223         struct net *net = tfile->net;
2224
2225         tun_detach(tfile, true);
2226         put_net(net);
2227
2228         return 0;
2229 }
2230
2231 static const struct file_operations tun_fops = {
2232         .owner  = THIS_MODULE,
2233         .llseek = no_llseek,
2234         .read  = do_sync_read,
2235         .aio_read  = tun_chr_aio_read,
2236         .write = do_sync_write,
2237         .aio_write = tun_chr_aio_write,
2238         .poll   = tun_chr_poll,
2239         .unlocked_ioctl = tun_chr_ioctl,
2240 #ifdef CONFIG_COMPAT
2241         .compat_ioctl = tun_chr_compat_ioctl,
2242 #endif
2243         .open   = tun_chr_open,
2244         .release = tun_chr_close,
2245         .fasync = tun_chr_fasync
2246 };
2247
2248 static struct miscdevice tun_miscdev = {
2249         .minor = TUN_MINOR,
2250         .name = "tun",
2251         .nodename = "net/tun",
2252         .fops = &tun_fops,
2253 };
2254
2255 /* ethtool interface */
2256
2257 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2258 {
2259         cmd->supported          = 0;
2260         cmd->advertising        = 0;
2261         ethtool_cmd_speed_set(cmd, SPEED_10);
2262         cmd->duplex             = DUPLEX_FULL;
2263         cmd->port               = PORT_TP;
2264         cmd->phy_address        = 0;
2265         cmd->transceiver        = XCVR_INTERNAL;
2266         cmd->autoneg            = AUTONEG_DISABLE;
2267         cmd->maxtxpkt           = 0;
2268         cmd->maxrxpkt           = 0;
2269         return 0;
2270 }
2271
2272 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2273 {
2274         struct tun_struct *tun = netdev_priv(dev);
2275
2276         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2277         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2278
2279         switch (tun->flags & TUN_TYPE_MASK) {
2280         case TUN_TUN_DEV:
2281                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2282                 break;
2283         case TUN_TAP_DEV:
2284                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2285                 break;
2286         }
2287 }
2288
2289 static u32 tun_get_msglevel(struct net_device *dev)
2290 {
2291 #ifdef TUN_DEBUG
2292         struct tun_struct *tun = netdev_priv(dev);
2293         return tun->debug;
2294 #else
2295         return -EOPNOTSUPP;
2296 #endif
2297 }
2298
2299 static void tun_set_msglevel(struct net_device *dev, u32 value)
2300 {
2301 #ifdef TUN_DEBUG
2302         struct tun_struct *tun = netdev_priv(dev);
2303         tun->debug = value;
2304 #endif
2305 }
2306
2307 static const struct ethtool_ops tun_ethtool_ops = {
2308         .get_settings   = tun_get_settings,
2309         .get_drvinfo    = tun_get_drvinfo,
2310         .get_msglevel   = tun_get_msglevel,
2311         .set_msglevel   = tun_set_msglevel,
2312         .get_link       = ethtool_op_get_link,
2313         .get_ts_info    = ethtool_op_get_ts_info,
2314 };
2315
2316
2317 static int __init tun_init(void)
2318 {
2319         int ret = 0;
2320
2321         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2322         pr_info("%s\n", DRV_COPYRIGHT);
2323
2324         ret = rtnl_link_register(&tun_link_ops);
2325         if (ret) {
2326                 pr_err("Can't register link_ops\n");
2327                 goto err_linkops;
2328         }
2329
2330         ret = misc_register(&tun_miscdev);
2331         if (ret) {
2332                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2333                 goto err_misc;
2334         }
2335         return  0;
2336 err_misc:
2337         rtnl_link_unregister(&tun_link_ops);
2338 err_linkops:
2339         return ret;
2340 }
2341
2342 static void tun_cleanup(void)
2343 {
2344         misc_deregister(&tun_miscdev);
2345         rtnl_link_unregister(&tun_link_ops);
2346 }
2347
2348 /* Get an underlying socket object from tun file.  Returns error unless file is
2349  * attached to a device.  The returned object works like a packet socket, it
2350  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2351  * holding a reference to the file for as long as the socket is in use. */
2352 struct socket *tun_get_socket(struct file *file)
2353 {
2354         struct tun_file *tfile;
2355         if (file->f_op != &tun_fops)
2356                 return ERR_PTR(-EINVAL);
2357         tfile = file->private_data;
2358         if (!tfile)
2359                 return ERR_PTR(-EBADFD);
2360         return &tfile->socket;
2361 }
2362 EXPORT_SYMBOL_GPL(tun_get_socket);
2363
2364 module_init(tun_init);
2365 module_exit(tun_cleanup);
2366 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2367 MODULE_AUTHOR(DRV_COPYRIGHT);
2368 MODULE_LICENSE("GPL");
2369 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2370 MODULE_ALIAS("devname:net/tun");