Pileus Git - ~andy/linux/blob - 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
 742         rcu_read_lock();
 743         tfile = rcu_dereference(tun->tfiles[txq]);
 744
 745         /* Drop packet if interface is not attached */
 746         if (txq >= tun->numqueues)
 747                 goto drop;
 748
 749         if (tun->numqueues == 1) {
 750                 /* Select queue was not called for the skbuff, so we extract the
 751                  * RPS hash and save it into the flow_table here.
 752                  */
 753                 __u32 rxhash;
 754
 755                 rxhash = skb_get_hash(skb);
 756                 if (rxhash) {
 757                         struct tun_flow_entry *e;
 758                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 759                                         rxhash);
 760                         if (e)
 761                                 tun_flow_save_rps_rxhash(e, rxhash);
 762                 }
 763         }
 764
 765         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 766
 767         BUG_ON(!tfile);
 768
 769         /* Drop if the filter does not like it.
 770          * This is a noop if the filter is disabled.
 771          * Filter can be enabled only for the TAP devices. */
 772         if (!check_filter(&tun->txflt, skb))
 773                 goto drop;
 774
 775         if (tfile->socket.sk->sk_filter &&
 776             sk_filter(tfile->socket.sk, skb))
 777                 goto drop;
 778
 779         /* Limit the number of packets queued by dividing txq length with the
 780          * number of queues.
 781          */
 782         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
 783                           >= dev->tx_queue_len / tun->numqueues)
 784                 goto drop;
 785
 786         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 787                 goto drop;
 788
 789         if (skb->sk) {
 790                 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
 791                 sw_tx_timestamp(skb);
 792         }
 793
 794         /* Orphan the skb - required as we might hang on to it
 795          * for indefinite time.
 796          */
 797         skb_orphan(skb);
 798
 799         nf_reset(skb);
 800
 801         /* Enqueue packet */
 802         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 803
 804         /* Notify and wake up reader process */
 805         if (tfile->flags & TUN_FASYNC)
 806                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 807         wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
 808                                    POLLRDNORM | POLLRDBAND);
 809
 810         rcu_read_unlock();
 811         return NETDEV_TX_OK;
 812
 813 drop:
 814         dev->stats.tx_dropped++;
 815         skb_tx_error(skb);
 816         kfree_skb(skb);
 817         rcu_read_unlock();
 818         return NETDEV_TX_OK;
 819 }
 820
 821 static void tun_net_mclist(struct net_device *dev)
 822 {
 823         /*
 824          * This callback is supposed to deal with mc filter in
 825          * _rx_ path and has nothing to do with the _tx_ path.
 826          * In rx path we always accept everything userspace gives us.
 827          */
 828 }
 829
 830 #define MIN_MTU 68
 831 #define MAX_MTU 65535
 832
 833 static int
 834 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 835 {
 836         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 837                 return -EINVAL;
 838         dev->mtu = new_mtu;
 839         return 0;
 840 }
 841
 842 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 843         netdev_features_t features)
 844 {
 845         struct tun_struct *tun = netdev_priv(dev);
 846
 847         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 848 }
 849 #ifdef CONFIG_NET_POLL_CONTROLLER
 850 static void tun_poll_controller(struct net_device *dev)
 851 {
 852         /*
 853          * Tun only receives frames when:
 854          * 1) the char device endpoint gets data from user space
 855          * 2) the tun socket gets a sendmsg call from user space
 856          * Since both of those are synchronous operations, we are guaranteed
 857          * never to have pending data when we poll for it
 858          * so there is nothing to do here but return.
 859          * We need this though so netpoll recognizes us as an interface that
 860          * supports polling, which enables bridge devices in virt setups to
 861          * still use netconsole
 862          */
 863         return;
 864 }
 865 #endif
 866 static const struct net_device_ops tun_netdev_ops = {
 867         .ndo_uninit             = tun_net_uninit,
 868         .ndo_open               = tun_net_open,
 869         .ndo_stop               = tun_net_close,
 870         .ndo_start_xmit         = tun_net_xmit,
 871         .ndo_change_mtu         = tun_net_change_mtu,
 872         .ndo_fix_features       = tun_net_fix_features,
 873         .ndo_select_queue       = tun_select_queue,
 874 #ifdef CONFIG_NET_POLL_CONTROLLER
 875         .ndo_poll_controller    = tun_poll_controller,
 876 #endif
 877 };
 878
 879 static const struct net_device_ops tap_netdev_ops = {
 880         .ndo_uninit             = tun_net_uninit,
 881         .ndo_open               = tun_net_open,
 882         .ndo_stop               = tun_net_close,
 883         .ndo_start_xmit         = tun_net_xmit,
 884         .ndo_change_mtu         = tun_net_change_mtu,
 885         .ndo_fix_features       = tun_net_fix_features,
 886         .ndo_set_rx_mode        = tun_net_mclist,
 887         .ndo_set_mac_address    = eth_mac_addr,
 888         .ndo_validate_addr      = eth_validate_addr,
 889         .ndo_select_queue       = tun_select_queue,
 890 #ifdef CONFIG_NET_POLL_CONTROLLER
 891         .ndo_poll_controller    = tun_poll_controller,
 892 #endif
 893 };
 894
 895 static void tun_flow_init(struct tun_struct *tun)
 896 {
 897         int i;
 898
 899         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 900                 INIT_HLIST_HEAD(&tun->flows[i]);
 901
 902         tun->ageing_time = TUN_FLOW_EXPIRE;
 903         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 904         mod_timer(&tun->flow_gc_timer,
 905                   round_jiffies_up(jiffies + tun->ageing_time));
 906 }
 907
 908 static void tun_flow_uninit(struct tun_struct *tun)
 909 {
 910         del_timer_sync(&tun->flow_gc_timer);
 911         tun_flow_flush(tun);
 912 }
 913
 914 /* Initialize net device. */
 915 static void tun_net_init(struct net_device *dev)
 916 {
 917         struct tun_struct *tun = netdev_priv(dev);
 918
 919         switch (tun->flags & TUN_TYPE_MASK) {
 920         case TUN_TUN_DEV:
 921                 dev->netdev_ops = &tun_netdev_ops;
 922
 923                 /* Point-to-Point TUN Device */
 924                 dev->hard_header_len = 0;
 925                 dev->addr_len = 0;
 926                 dev->mtu = 1500;
 927
 928                 /* Zero header length */
 929                 dev->type = ARPHRD_NONE;
 930                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 931                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 932                 break;
 933
 934         case TUN_TAP_DEV:
 935                 dev->netdev_ops = &tap_netdev_ops;
 936                 /* Ethernet TAP Device */
 937                 ether_setup(dev);
 938                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 939                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 940
 941                 eth_hw_addr_random(dev);
 942
 943                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 944                 break;
 945         }
 946 }
 947
 948 /* Character device part */
 949
 950 /* Poll */
 951 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 952 {
 953         struct tun_file *tfile = file->private_data;
 954         struct tun_struct *tun = __tun_get(tfile);
 955         struct sock *sk;
 956         unsigned int mask = 0;
 957
 958         if (!tun)
 959                 return POLLERR;
 960
 961         sk = tfile->socket.sk;
 962
 963         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 964
 965         poll_wait(file, &tfile->wq.wait, wait);
 966
 967         if (!skb_queue_empty(&sk->sk_receive_queue))
 968                 mask |= POLLIN | POLLRDNORM;
 969
 970         if (sock_writeable(sk) ||
 971             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 972              sock_writeable(sk)))
 973                 mask |= POLLOUT | POLLWRNORM;
 974
 975         if (tun->dev->reg_state != NETREG_REGISTERED)
 976                 mask = POLLERR;
 977
 978         tun_put(tun);
 979         return mask;
 980 }
 981
 982 /* prepad is the amount to reserve at front.  len is length after that.
 983  * linear is a hint as to how much to copy (usually headers). */
 984 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 985                                      size_t prepad, size_t len,
 986                                      size_t linear, int noblock)
 987 {
 988         struct sock *sk = tfile->socket.sk;
 989         struct sk_buff *skb;
 990         int err;
 991
 992         /* Under a page?  Don't bother with paged skb. */
 993         if (prepad + len < PAGE_SIZE || !linear)
 994                 linear = len;
 995
 996         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 997                                    &err, 0);
 998         if (!skb)
 999                 return ERR_PTR(err);
1000
1001         skb_reserve(skb, prepad);
1002         skb_put(skb, linear);
1003         skb->data_len = len - linear;
1004         skb->len += len - linear;
1005
1006         return skb;
1007 }
1008
1009 /* Get packet from user space buffer */
1010 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1011                             void *msg_control, const struct iovec *iv,
1012                             size_t total_len, size_t count, int noblock)
1013 {
1014         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1015         struct sk_buff *skb;
1016         size_t len = total_len, align = NET_SKB_PAD, linear;
1017         struct virtio_net_hdr gso = { 0 };
1018         int good_linear;
1019         int offset = 0;
1020         int copylen;
1021         bool zerocopy = false;
1022         int err;
1023         u32 rxhash;
1024
1025         if (!(tun->flags & TUN_NO_PI)) {
1026                 if (len < sizeof(pi))
1027                         return -EINVAL;
1028                 len -= sizeof(pi);
1029
1030                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1031                         return -EFAULT;
1032                 offset += sizeof(pi);
1033         }
1034
1035         if (tun->flags & TUN_VNET_HDR) {
1036                 if (len < tun->vnet_hdr_sz)
1037                         return -EINVAL;
1038                 len -= tun->vnet_hdr_sz;
1039
1040                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1041                         return -EFAULT;
1042
1043                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1044                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1045                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1046
1047                 if (gso.hdr_len > len)
1048                         return -EINVAL;
1049                 offset += tun->vnet_hdr_sz;
1050         }
1051
1052         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1053                 align += NET_IP_ALIGN;
1054                 if (unlikely(len < ETH_HLEN ||
1055                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1056                         return -EINVAL;
1057         }
1058
1059         good_linear = SKB_MAX_HEAD(align);
1060
1061         if (msg_control) {
1062                 /* There are 256 bytes to be copied in skb, so there is
1063                  * enough room for skb expand head in case it is used.
1064                  * The rest of the buffer is mapped from userspace.
1065                  */
1066                 copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1067                 if (copylen > good_linear)
1068                         copylen = good_linear;
1069                 linear = copylen;
1070                 if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1071                         zerocopy = true;
1072         }
1073
1074         if (!zerocopy) {
1075                 copylen = len;
1076                 if (gso.hdr_len > good_linear)
1077                         linear = good_linear;
1078                 else
1079                         linear = gso.hdr_len;
1080         }
1081
1082         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1083         if (IS_ERR(skb)) {
1084                 if (PTR_ERR(skb) != -EAGAIN)
1085                         tun->dev->stats.rx_dropped++;
1086                 return PTR_ERR(skb);
1087         }
1088
1089         if (zerocopy)
1090                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1091         else {
1092                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1093                 if (!err && msg_control) {
1094                         struct ubuf_info *uarg = msg_control;
1095                         uarg->callback(uarg, false);
1096                 }
1097         }
1098
1099         if (err) {
1100                 tun->dev->stats.rx_dropped++;
1101                 kfree_skb(skb);
1102                 return -EFAULT;
1103         }
1104
1105         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1106                 if (!skb_partial_csum_set(skb, gso.csum_start,
1107                                           gso.csum_offset)) {
1108                         tun->dev->stats.rx_frame_errors++;
1109                         kfree_skb(skb);
1110                         return -EINVAL;
1111                 }
1112         }
1113
1114         switch (tun->flags & TUN_TYPE_MASK) {
1115         case TUN_TUN_DEV:
1116                 if (tun->flags & TUN_NO_PI) {
1117                         switch (skb->data[0] & 0xf0) {
1118                         case 0x40:
1119                                 pi.proto = htons(ETH_P_IP);
1120                                 break;
1121                         case 0x60:
1122                                 pi.proto = htons(ETH_P_IPV6);
1123                                 break;
1124                         default:
1125                                 tun->dev->stats.rx_dropped++;
1126                                 kfree_skb(skb);
1127                                 return -EINVAL;
1128                         }
1129                 }
1130
1131                 skb_reset_mac_header(skb);
1132                 skb->protocol = pi.proto;
1133                 skb->dev = tun->dev;
1134                 break;
1135         case TUN_TAP_DEV:
1136                 skb->protocol = eth_type_trans(skb, tun->dev);
1137                 break;
1138         }
1139
1140         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1141                 pr_debug("GSO!\n");
1142                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1143                 case VIRTIO_NET_HDR_GSO_TCPV4:
1144                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1145                         break;
1146                 case VIRTIO_NET_HDR_GSO_TCPV6:
1147                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1148                         break;
1149                 case VIRTIO_NET_HDR_GSO_UDP:
1150                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1151                         break;
1152                 default:
1153                         tun->dev->stats.rx_frame_errors++;
1154                         kfree_skb(skb);
1155                         return -EINVAL;
1156                 }
1157
1158                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1159                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1160
1161                 skb_shinfo(skb)->gso_size = gso.gso_size;
1162                 if (skb_shinfo(skb)->gso_size == 0) {
1163                         tun->dev->stats.rx_frame_errors++;
1164                         kfree_skb(skb);
1165                         return -EINVAL;
1166                 }
1167
1168                 /* Header must be checked, and gso_segs computed. */
1169                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1170                 skb_shinfo(skb)->gso_segs = 0;
1171         }
1172
1173         /* copy skb_ubuf_info for callback when skb has no error */
1174         if (zerocopy) {
1175                 skb_shinfo(skb)->destructor_arg = msg_control;
1176                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1177                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1178         }
1179
1180         skb_reset_network_header(skb);
1181         skb_probe_transport_header(skb, 0);
1182
1183         rxhash = skb_get_hash(skb);
1184         netif_rx_ni(skb);
1185
1186         tun->dev->stats.rx_packets++;
1187         tun->dev->stats.rx_bytes += len;
1188
1189         tun_flow_update(tun, rxhash, tfile);
1190         return total_len;
1191 }
1192
1193 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1194                               unsigned long count, loff_t pos)
1195 {
1196         struct file *file = iocb->ki_filp;
1197         struct tun_struct *tun = tun_get(file);
1198         struct tun_file *tfile = file->private_data;
1199         ssize_t result;
1200
1201         if (!tun)
1202                 return -EBADFD;
1203
1204         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1205
1206         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1207                               count, file->f_flags & O_NONBLOCK);
1208
1209         tun_put(tun);
1210         return result;
1211 }
1212
1213 /* Put packet to the user space buffer */
1214 static ssize_t tun_put_user(struct tun_struct *tun,
1215                             struct tun_file *tfile,
1216                             struct sk_buff *skb,
1217                             const struct iovec *iv, int len)
1218 {
1219         struct tun_pi pi = { 0, skb->protocol };
1220         ssize_t total = 0;
1221         int vlan_offset = 0, copied;
1222
1223         if (!(tun->flags & TUN_NO_PI)) {
1224                 if ((len -= sizeof(pi)) < 0)
1225                         return -EINVAL;
1226
1227                 if (len < skb->len) {
1228                         /* Packet will be striped */
1229                         pi.flags |= TUN_PKT_STRIP;
1230                 }
1231
1232                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1233                         return -EFAULT;
1234                 total += sizeof(pi);
1235         }
1236
1237         if (tun->flags & TUN_VNET_HDR) {
1238                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1239                 if ((len -= tun->vnet_hdr_sz) < 0)
1240                         return -EINVAL;
1241
1242                 if (skb_is_gso(skb)) {
1243                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1244
1245                         /* This is a hint as to how much should be linear. */
1246                         gso.hdr_len = skb_headlen(skb);
1247                         gso.gso_size = sinfo->gso_size;
1248                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1249                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1250                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1251                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1252                         else if (sinfo->gso_type & SKB_GSO_UDP)
1253                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1254                         else {
1255                                 pr_err("unexpected GSO type: "
1256                                        "0x%x, gso_size %d, hdr_len %d\n",
1257                                        sinfo->gso_type, gso.gso_size,
1258                                        gso.hdr_len);
1259                                 print_hex_dump(KERN_ERR, "tun: ",
1260                                                DUMP_PREFIX_NONE,
1261                                                16, 1, skb->head,
1262                                                min((int)gso.hdr_len, 64), true);
1263                                 WARN_ON_ONCE(1);
1264                                 return -EINVAL;
1265                         }
1266                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1267                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1268                 } else
1269                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1270
1271                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1272                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1273                         gso.csum_start = skb_checksum_start_offset(skb);
1274                         gso.csum_offset = skb->csum_offset;
1275                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1276                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1277                 } /* else everything is zero */
1278
1279                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1280                                                sizeof(gso))))
1281                         return -EFAULT;
1282                 total += tun->vnet_hdr_sz;
1283         }
1284
1285         copied = total;
1286         total += skb->len;
1287         if (!vlan_tx_tag_present(skb)) {
1288                 len = min_t(int, skb->len, len);
1289         } else {
1290                 int copy, ret;
1291                 struct {
1292                         __be16 h_vlan_proto;
1293                         __be16 h_vlan_TCI;
1294                 } veth;
1295
1296                 veth.h_vlan_proto = skb->vlan_proto;
1297                 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1298
1299                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1300                 len = min_t(int, skb->len + VLAN_HLEN, len);
1301                 total += VLAN_HLEN;
1302
1303                 copy = min_t(int, vlan_offset, len);
1304                 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
1305                 len -= copy;
1306                 copied += copy;
1307                 if (ret || !len)
1308                         goto done;
1309
1310                 copy = min_t(int, sizeof(veth), len);
1311                 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
1312                 len -= copy;
1313                 copied += copy;
1314                 if (ret || !len)
1315                         goto done;
1316         }
1317
1318         skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
1319
1320 done:
1321         tun->dev->stats.tx_packets++;
1322         tun->dev->stats.tx_bytes += len;
1323
1324         return total;
1325 }
1326
1327 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1328                            const struct iovec *iv, ssize_t len, int noblock)
1329 {
1330         DECLARE_WAITQUEUE(wait, current);
1331         struct sk_buff *skb;
1332         ssize_t ret = 0;
1333
1334         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1335
1336         if (unlikely(!noblock))
1337                 add_wait_queue(&tfile->wq.wait, &wait);
1338         while (len) {
1339                 if (unlikely(!noblock))
1340                         current->state = TASK_INTERRUPTIBLE;
1341
1342                 /* Read frames from the queue */
1343                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1344                         if (noblock) {
1345                                 ret = -EAGAIN;
1346                                 break;
1347                         }
1348                         if (signal_pending(current)) {
1349                                 ret = -ERESTARTSYS;
1350                                 break;
1351                         }
1352                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1353                                 ret = -EIO;
1354                                 break;
1355                         }
1356
1357                         /* Nothing to read, let's sleep */
1358                         schedule();
1359                         continue;
1360                 }
1361
1362                 ret = tun_put_user(tun, tfile, skb, iv, len);
1363                 kfree_skb(skb);
1364                 break;
1365         }
1366
1367         if (unlikely(!noblock)) {
1368                 current->state = TASK_RUNNING;
1369                 remove_wait_queue(&tfile->wq.wait, &wait);
1370         }
1371
1372         return ret;
1373 }
1374
1375 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1376                             unsigned long count, loff_t pos)
1377 {
1378         struct file *file = iocb->ki_filp;
1379         struct tun_file *tfile = file->private_data;
1380         struct tun_struct *tun = __tun_get(tfile);
1381         ssize_t len, ret;
1382
1383         if (!tun)
1384                 return -EBADFD;
1385         len = iov_length(iv, count);
1386         if (len < 0) {
1387                 ret = -EINVAL;
1388                 goto out;
1389         }
1390
1391         ret = tun_do_read(tun, tfile, iv, len,
1392                           file->f_flags & O_NONBLOCK);
1393         ret = min_t(ssize_t, ret, len);
1394         if (ret > 0)
1395                 iocb->ki_pos = ret;
1396 out:
1397         tun_put(tun);
1398         return ret;
1399 }
1400
1401 static void tun_free_netdev(struct net_device *dev)
1402 {
1403         struct tun_struct *tun = netdev_priv(dev);
1404
1405         BUG_ON(!(list_empty(&tun->disabled)));
1406         tun_flow_uninit(tun);
1407         security_tun_dev_free_security(tun->security);
1408         free_netdev(dev);
1409 }
1410
1411 static void tun_setup(struct net_device *dev)
1412 {
1413         struct tun_struct *tun = netdev_priv(dev);
1414
1415         tun->owner = INVALID_UID;
1416         tun->group = INVALID_GID;
1417
1418         dev->ethtool_ops = &tun_ethtool_ops;
1419         dev->destructor = tun_free_netdev;
1420 }
1421
1422 /* Trivial set of netlink ops to allow deleting tun or tap
1423  * device with netlink.
1424  */
1425 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1426 {
1427         return -EINVAL;
1428 }
1429
1430 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1431         .kind           = DRV_NAME,
1432         .priv_size      = sizeof(struct tun_struct),
1433         .setup          = tun_setup,
1434         .validate       = tun_validate,
1435 };
1436
1437 static void tun_sock_write_space(struct sock *sk)
1438 {
1439         struct tun_file *tfile;
1440         wait_queue_head_t *wqueue;
1441
1442         if (!sock_writeable(sk))
1443                 return;
1444
1445         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1446                 return;
1447
1448         wqueue = sk_sleep(sk);
1449         if (wqueue && waitqueue_active(wqueue))
1450                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1451                                                 POLLWRNORM | POLLWRBAND);
1452
1453         tfile = container_of(sk, struct tun_file, sk);
1454         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1455 }
1456
1457 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1458                        struct msghdr *m, size_t total_len)
1459 {
1460         int ret;
1461         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1462         struct tun_struct *tun = __tun_get(tfile);
1463
1464         if (!tun)
1465                 return -EBADFD;
1466         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1467                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1468         tun_put(tun);
1469         return ret;
1470 }
1471
1472 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1473                        struct msghdr *m, size_t total_len,
1474                        int flags)
1475 {
1476         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1477         struct tun_struct *tun = __tun_get(tfile);
1478         int ret;
1479
1480         if (!tun)
1481                 return -EBADFD;
1482
1483         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1484                 ret = -EINVAL;
1485                 goto out;
1486         }
1487         if (flags & MSG_ERRQUEUE) {
1488                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1489                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1490                 goto out;
1491         }
1492         ret = tun_do_read(tun, tfile, m->msg_iov, total_len,
1493                           flags & MSG_DONTWAIT);
1494         if (ret > total_len) {
1495                 m->msg_flags |= MSG_TRUNC;
1496                 ret = flags & MSG_TRUNC ? ret : total_len;
1497         }
1498 out:
1499         tun_put(tun);
1500         return ret;
1501 }
1502
1503 static int tun_release(struct socket *sock)
1504 {
1505         if (sock->sk)
1506                 sock_put(sock->sk);
1507         return 0;
1508 }
1509
1510 /* Ops structure to mimic raw sockets with tun */
1511 static const struct proto_ops tun_socket_ops = {
1512         .sendmsg = tun_sendmsg,
1513         .recvmsg = tun_recvmsg,
1514         .release = tun_release,
1515 };
1516
1517 static struct proto tun_proto = {
1518         .name           = "tun",
1519         .owner          = THIS_MODULE,
1520         .obj_size       = sizeof(struct tun_file),
1521 };
1522
1523 static int tun_flags(struct tun_struct *tun)
1524 {
1525         int flags = 0;
1526
1527         if (tun->flags & TUN_TUN_DEV)
1528                 flags |= IFF_TUN;
1529         else
1530                 flags |= IFF_TAP;
1531
1532         if (tun->flags & TUN_NO_PI)
1533                 flags |= IFF_NO_PI;
1534
1535         /* This flag has no real effect.  We track the value for backwards
1536          * compatibility.
1537          */
1538         if (tun->flags & TUN_ONE_QUEUE)
1539                 flags |= IFF_ONE_QUEUE;
1540
1541         if (tun->flags & TUN_VNET_HDR)
1542                 flags |= IFF_VNET_HDR;
1543
1544         if (tun->flags & TUN_TAP_MQ)
1545                 flags |= IFF_MULTI_QUEUE;
1546
1547         if (tun->flags & TUN_PERSIST)
1548                 flags |= IFF_PERSIST;
1549
1550         return flags;
1551 }
1552
1553 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1554                               char *buf)
1555 {
1556         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1557         return sprintf(buf, "0x%x\n", tun_flags(tun));
1558 }
1559
1560 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1561                               char *buf)
1562 {
1563         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1564         return uid_valid(tun->owner)?
1565                 sprintf(buf, "%u\n",
1566                         from_kuid_munged(current_user_ns(), tun->owner)):
1567                 sprintf(buf, "-1\n");
1568 }
1569
1570 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1571                               char *buf)
1572 {
1573         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1574         return gid_valid(tun->group) ?
1575                 sprintf(buf, "%u\n",
1576                         from_kgid_munged(current_user_ns(), tun->group)):
1577                 sprintf(buf, "-1\n");
1578 }
1579
1580 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1581 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1582 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1583
1584 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1585 {
1586         struct tun_struct *tun;
1587         struct tun_file *tfile = file->private_data;
1588         struct net_device *dev;
1589         int err;
1590
1591         if (tfile->detached)
1592                 return -EINVAL;
1593
1594         dev = __dev_get_by_name(net, ifr->ifr_name);
1595         if (dev) {
1596                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1597                         return -EBUSY;
1598                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1599                         tun = netdev_priv(dev);
1600                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1601                         tun = netdev_priv(dev);
1602                 else
1603                         return -EINVAL;
1604
1605                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1606                     !!(tun->flags & TUN_TAP_MQ))
1607                         return -EINVAL;
1608
1609                 if (tun_not_capable(tun))
1610                         return -EPERM;
1611                 err = security_tun_dev_open(tun->security);
1612                 if (err < 0)
1613                         return err;
1614
1615                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1616                 if (err < 0)
1617                         return err;
1618
1619                 if (tun->flags & TUN_TAP_MQ &&
1620                     (tun->numqueues + tun->numdisabled > 1)) {
1621                         /* One or more queue has already been attached, no need
1622                          * to initialize the device again.
1623                          */
1624                         return 0;
1625                 }
1626         }
1627         else {
1628                 char *name;
1629                 unsigned long flags = 0;
1630                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1631                              MAX_TAP_QUEUES : 1;
1632
1633                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1634                         return -EPERM;
1635                 err = security_tun_dev_create();
1636                 if (err < 0)
1637                         return err;
1638
1639                 /* Set dev type */
1640                 if (ifr->ifr_flags & IFF_TUN) {
1641                         /* TUN device */
1642                         flags |= TUN_TUN_DEV;
1643                         name = "tun%d";
1644                 } else if (ifr->ifr_flags & IFF_TAP) {
1645                         /* TAP device */
1646                         flags |= TUN_TAP_DEV;
1647                         name = "tap%d";
1648                 } else
1649                         return -EINVAL;
1650
1651                 if (*ifr->ifr_name)
1652                         name = ifr->ifr_name;
1653
1654                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1655                                        tun_setup, queues, queues);
1656
1657                 if (!dev)
1658                         return -ENOMEM;
1659
1660                 dev_net_set(dev, net);
1661                 dev->rtnl_link_ops = &tun_link_ops;
1662                 dev->ifindex = tfile->ifindex;
1663
1664                 tun = netdev_priv(dev);
1665                 tun->dev = dev;
1666                 tun->flags = flags;
1667                 tun->txflt.count = 0;
1668                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1669
1670                 tun->filter_attached = false;
1671                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1672
1673                 spin_lock_init(&tun->lock);
1674
1675                 err = security_tun_dev_alloc_security(&tun->security);
1676                 if (err < 0)
1677                         goto err_free_dev;
1678
1679                 tun_net_init(dev);
1680                 tun_flow_init(tun);
1681
1682                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1683                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1684                                    NETIF_F_HW_VLAN_STAG_TX;
1685                 dev->features = dev->hw_features;
1686                 dev->vlan_features = dev->features;
1687
1688                 INIT_LIST_HEAD(&tun->disabled);
1689                 err = tun_attach(tun, file, false);
1690                 if (err < 0)
1691                         goto err_free_flow;
1692
1693                 err = register_netdevice(tun->dev);
1694                 if (err < 0)
1695                         goto err_detach;
1696
1697                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1698                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1699                     device_create_file(&tun->dev->dev, &dev_attr_group))
1700                         pr_err("Failed to create tun sysfs files\n");
1701         }
1702
1703         netif_carrier_on(tun->dev);
1704
1705         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1706
1707         if (ifr->ifr_flags & IFF_NO_PI)
1708                 tun->flags |= TUN_NO_PI;
1709         else
1710                 tun->flags &= ~TUN_NO_PI;
1711
1712         /* This flag has no real effect.  We track the value for backwards
1713          * compatibility.
1714          */
1715         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1716                 tun->flags |= TUN_ONE_QUEUE;
1717         else
1718                 tun->flags &= ~TUN_ONE_QUEUE;
1719
1720         if (ifr->ifr_flags & IFF_VNET_HDR)
1721                 tun->flags |= TUN_VNET_HDR;
1722         else
1723                 tun->flags &= ~TUN_VNET_HDR;
1724
1725         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1726                 tun->flags |= TUN_TAP_MQ;
1727         else
1728                 tun->flags &= ~TUN_TAP_MQ;
1729
1730         /* Make sure persistent devices do not get stuck in
1731          * xoff state.
1732          */
1733         if (netif_running(tun->dev))
1734                 netif_tx_wake_all_queues(tun->dev);
1735
1736         strcpy(ifr->ifr_name, tun->dev->name);
1737         return 0;
1738
1739 err_detach:
1740         tun_detach_all(dev);
1741 err_free_flow:
1742         tun_flow_uninit(tun);
1743         security_tun_dev_free_security(tun->security);
1744 err_free_dev:
1745         free_netdev(dev);
1746         return err;
1747 }
1748
1749 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1750                        struct ifreq *ifr)
1751 {
1752         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1753
1754         strcpy(ifr->ifr_name, tun->dev->name);
1755
1756         ifr->ifr_flags = tun_flags(tun);
1757
1758 }
1759
1760 /* This is like a cut-down ethtool ops, except done via tun fd so no
1761  * privs required. */
1762 static int set_offload(struct tun_struct *tun, unsigned long arg)
1763 {
1764         netdev_features_t features = 0;
1765
1766         if (arg & TUN_F_CSUM) {
1767                 features |= NETIF_F_HW_CSUM;
1768                 arg &= ~TUN_F_CSUM;
1769
1770                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1771                         if (arg & TUN_F_TSO_ECN) {
1772                                 features |= NETIF_F_TSO_ECN;
1773                                 arg &= ~TUN_F_TSO_ECN;
1774                         }
1775                         if (arg & TUN_F_TSO4)
1776                                 features |= NETIF_F_TSO;
1777                         if (arg & TUN_F_TSO6)
1778                                 features |= NETIF_F_TSO6;
1779                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1780                 }
1781
1782                 if (arg & TUN_F_UFO) {
1783                         features |= NETIF_F_UFO;
1784                         arg &= ~TUN_F_UFO;
1785                 }
1786         }
1787
1788         /* This gives the user a way to test for new features in future by
1789          * trying to set them. */
1790         if (arg)
1791                 return -EINVAL;
1792
1793         tun->set_features = features;
1794         netdev_update_features(tun->dev);
1795
1796         return 0;
1797 }
1798
1799 static void tun_detach_filter(struct tun_struct *tun, int n)
1800 {
1801         int i;
1802         struct tun_file *tfile;
1803
1804         for (i = 0; i < n; i++) {
1805                 tfile = rtnl_dereference(tun->tfiles[i]);
1806                 sk_detach_filter(tfile->socket.sk);
1807         }
1808
1809         tun->filter_attached = false;
1810 }
1811
1812 static int tun_attach_filter(struct tun_struct *tun)
1813 {
1814         int i, ret = 0;
1815         struct tun_file *tfile;
1816
1817         for (i = 0; i < tun->numqueues; i++) {
1818                 tfile = rtnl_dereference(tun->tfiles[i]);
1819                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1820                 if (ret) {
1821                         tun_detach_filter(tun, i);
1822                         return ret;
1823                 }
1824         }
1825
1826         tun->filter_attached = true;
1827         return ret;
1828 }
1829
1830 static void tun_set_sndbuf(struct tun_struct *tun)
1831 {
1832         struct tun_file *tfile;
1833         int i;
1834
1835         for (i = 0; i < tun->numqueues; i++) {
1836                 tfile = rtnl_dereference(tun->tfiles[i]);
1837                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1838         }
1839 }
1840
1841 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1842 {
1843         struct tun_file *tfile = file->private_data;
1844         struct tun_struct *tun;
1845         int ret = 0;
1846
1847         rtnl_lock();
1848
1849         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1850                 tun = tfile->detached;
1851                 if (!tun) {
1852                         ret = -EINVAL;
1853                         goto unlock;
1854                 }
1855                 ret = security_tun_dev_attach_queue(tun->security);
1856                 if (ret < 0)
1857                         goto unlock;
1858                 ret = tun_attach(tun, file, false);
1859         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1860                 tun = rtnl_dereference(tfile->tun);
1861                 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1862                         ret = -EINVAL;
1863                 else
1864                         __tun_detach(tfile, false);
1865         } else
1866                 ret = -EINVAL;
1867
1868 unlock:
1869         rtnl_unlock();
1870         return ret;
1871 }
1872
1873 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1874                             unsigned long arg, int ifreq_len)
1875 {
1876         struct tun_file *tfile = file->private_data;
1877         struct tun_struct *tun;
1878         void __user* argp = (void __user*)arg;
1879         struct ifreq ifr;
1880         kuid_t owner;
1881         kgid_t group;
1882         int sndbuf;
1883         int vnet_hdr_sz;
1884         unsigned int ifindex;
1885         int ret;
1886
1887         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1888                 if (copy_from_user(&ifr, argp, ifreq_len))
1889                         return -EFAULT;
1890         } else {
1891                 memset(&ifr, 0, sizeof(ifr));
1892         }
1893         if (cmd == TUNGETFEATURES) {
1894                 /* Currently this just means: "what IFF flags are valid?".
1895                  * This is needed because we never checked for invalid flags on
1896                  * TUNSETIFF. */
1897                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1898                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1899                                 (unsigned int __user*)argp);
1900         } else if (cmd == TUNSETQUEUE)
1901                 return tun_set_queue(file, &ifr);
1902
1903         ret = 0;
1904         rtnl_lock();
1905
1906         tun = __tun_get(tfile);
1907         if (cmd == TUNSETIFF && !tun) {
1908                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1909
1910                 ret = tun_set_iff(tfile->net, file, &ifr);
1911
1912                 if (ret)
1913                         goto unlock;
1914
1915                 if (copy_to_user(argp, &ifr, ifreq_len))
1916                         ret = -EFAULT;
1917                 goto unlock;
1918         }
1919         if (cmd == TUNSETIFINDEX) {
1920                 ret = -EPERM;
1921                 if (tun)
1922                         goto unlock;
1923
1924                 ret = -EFAULT;
1925                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1926                         goto unlock;
1927
1928                 ret = 0;
1929                 tfile->ifindex = ifindex;
1930                 goto unlock;
1931         }
1932
1933         ret = -EBADFD;
1934         if (!tun)
1935                 goto unlock;
1936
1937         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1938
1939         ret = 0;
1940         switch (cmd) {
1941         case TUNGETIFF:
1942                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1943
1944                 if (tfile->detached)
1945                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
1946                 if (!tfile->socket.sk->sk_filter)
1947                         ifr.ifr_flags |= IFF_NOFILTER;
1948
1949                 if (copy_to_user(argp, &ifr, ifreq_len))
1950                         ret = -EFAULT;
1951                 break;
1952
1953         case TUNSETNOCSUM:
1954                 /* Disable/Enable checksum */
1955
1956                 /* [unimplemented] */
1957                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1958                           arg ? "disabled" : "enabled");
1959                 break;
1960
1961         case TUNSETPERSIST:
1962                 /* Disable/Enable persist mode. Keep an extra reference to the
1963                  * module to prevent the module being unprobed.
1964                  */
1965                 if (arg && !(tun->flags & TUN_PERSIST)) {
1966                         tun->flags |= TUN_PERSIST;
1967                         __module_get(THIS_MODULE);
1968                 }
1969                 if (!arg && (tun->flags & TUN_PERSIST)) {
1970                         tun->flags &= ~TUN_PERSIST;
1971                         module_put(THIS_MODULE);
1972                 }
1973
1974                 tun_debug(KERN_INFO, tun, "persist %s\n",
1975                           arg ? "enabled" : "disabled");
1976                 break;
1977
1978         case TUNSETOWNER:
1979                 /* Set owner of the device */
1980                 owner = make_kuid(current_user_ns(), arg);
1981                 if (!uid_valid(owner)) {
1982                         ret = -EINVAL;
1983                         break;
1984                 }
1985                 tun->owner = owner;
1986                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1987                           from_kuid(&init_user_ns, tun->owner));
1988                 break;
1989
1990         case TUNSETGROUP:
1991                 /* Set group of the device */
1992                 group = make_kgid(current_user_ns(), arg);
1993                 if (!gid_valid(group)) {
1994                         ret = -EINVAL;
1995                         break;
1996                 }
1997                 tun->group = group;
1998                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1999                           from_kgid(&init_user_ns, tun->group));
2000                 break;
2001
2002         case TUNSETLINK:
2003                 /* Only allow setting the type when the interface is down */
2004                 if (tun->dev->flags & IFF_UP) {
2005                         tun_debug(KERN_INFO, tun,
2006                                   "Linktype set failed because interface is up\n");
2007                         ret = -EBUSY;
2008                 } else {
2009                         tun->dev->type = (int) arg;
2010                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2011                                   tun->dev->type);
2012                         ret = 0;
2013                 }
2014                 break;
2015
2016 #ifdef TUN_DEBUG
2017         case TUNSETDEBUG:
2018                 tun->debug = arg;
2019                 break;
2020 #endif
2021         case TUNSETOFFLOAD:
2022                 ret = set_offload(tun, arg);
2023                 break;
2024
2025         case TUNSETTXFILTER:
2026                 /* Can be set only for TAPs */
2027                 ret = -EINVAL;
2028                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2029                         break;
2030                 ret = update_filter(&tun->txflt, (void __user *)arg);
2031                 break;
2032
2033         case SIOCGIFHWADDR:
2034                 /* Get hw address */
2035                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2036                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2037                 if (copy_to_user(argp, &ifr, ifreq_len))
2038                         ret = -EFAULT;
2039                 break;
2040
2041         case SIOCSIFHWADDR:
2042                 /* Set hw address */
2043                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2044                           ifr.ifr_hwaddr.sa_data);
2045
2046                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2047                 break;
2048
2049         case TUNGETSNDBUF:
2050                 sndbuf = tfile->socket.sk->sk_sndbuf;
2051                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2052                         ret = -EFAULT;
2053                 break;
2054
2055         case TUNSETSNDBUF:
2056                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2057                         ret = -EFAULT;
2058                         break;
2059                 }
2060
2061                 tun->sndbuf = sndbuf;
2062                 tun_set_sndbuf(tun);
2063                 break;
2064
2065         case TUNGETVNETHDRSZ:
2066                 vnet_hdr_sz = tun->vnet_hdr_sz;
2067                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2068                         ret = -EFAULT;
2069                 break;
2070
2071         case TUNSETVNETHDRSZ:
2072                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2073                         ret = -EFAULT;
2074                         break;
2075                 }
2076                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2077                         ret = -EINVAL;
2078                         break;
2079                 }
2080
2081                 tun->vnet_hdr_sz = vnet_hdr_sz;
2082                 break;
2083
2084         case TUNATTACHFILTER:
2085                 /* Can be set only for TAPs */
2086                 ret = -EINVAL;
2087                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2088                         break;
2089                 ret = -EFAULT;
2090                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2091                         break;
2092
2093                 ret = tun_attach_filter(tun);
2094                 break;
2095
2096         case TUNDETACHFILTER:
2097                 /* Can be set only for TAPs */
2098                 ret = -EINVAL;
2099                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2100                         break;
2101                 ret = 0;
2102                 tun_detach_filter(tun, tun->numqueues);
2103                 break;
2104
2105         case TUNGETFILTER:
2106                 ret = -EINVAL;
2107                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2108                         break;
2109                 ret = -EFAULT;
2110                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2111                         break;
2112                 ret = 0;
2113                 break;
2114
2115         default:
2116                 ret = -EINVAL;
2117                 break;
2118         }
2119
2120 unlock:
2121         rtnl_unlock();
2122         if (tun)
2123                 tun_put(tun);
2124         return ret;
2125 }
2126
2127 static long tun_chr_ioctl(struct file *file,
2128                           unsigned int cmd, unsigned long arg)
2129 {
2130         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2131 }
2132
2133 #ifdef CONFIG_COMPAT
2134 static long tun_chr_compat_ioctl(struct file *file,
2135                          unsigned int cmd, unsigned long arg)
2136 {
2137         switch (cmd) {
2138         case TUNSETIFF:
2139         case TUNGETIFF:
2140         case TUNSETTXFILTER:
2141         case TUNGETSNDBUF:
2142         case TUNSETSNDBUF:
2143         case SIOCGIFHWADDR:
2144         case SIOCSIFHWADDR:
2145                 arg = (unsigned long)compat_ptr(arg);
2146                 break;
2147         default:
2148                 arg = (compat_ulong_t)arg;
2149                 break;
2150         }
2151
2152         /*
2153          * compat_ifreq is shorter than ifreq, so we must not access beyond
2154          * the end of that structure. All fields that are used in this
2155          * driver are compatible though, we don't need to convert the
2156          * contents.
2157          */
2158         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2159 }
2160 #endif /* CONFIG_COMPAT */
2161
2162 static int tun_chr_fasync(int fd, struct file *file, int on)
2163 {
2164         struct tun_file *tfile = file->private_data;
2165         int ret;
2166
2167         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2168                 goto out;
2169
2170         if (on) {
2171                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2172                 if (ret)
2173                         goto out;
2174                 tfile->flags |= TUN_FASYNC;
2175         } else
2176                 tfile->flags &= ~TUN_FASYNC;
2177         ret = 0;
2178 out:
2179         return ret;
2180 }
2181
2182 static int tun_chr_open(struct inode *inode, struct file * file)
2183 {
2184         struct tun_file *tfile;
2185
2186         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2187
2188         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2189                                             &tun_proto);
2190         if (!tfile)
2191                 return -ENOMEM;
2192         rcu_assign_pointer(tfile->tun, NULL);
2193         tfile->net = get_net(current->nsproxy->net_ns);
2194         tfile->flags = 0;
2195         tfile->ifindex = 0;
2196
2197         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2198         init_waitqueue_head(&tfile->wq.wait);
2199
2200         tfile->socket.file = file;
2201         tfile->socket.ops = &tun_socket_ops;
2202
2203         sock_init_data(&tfile->socket, &tfile->sk);
2204         sk_change_net(&tfile->sk, tfile->net);
2205
2206         tfile->sk.sk_write_space = tun_sock_write_space;
2207         tfile->sk.sk_sndbuf = INT_MAX;
2208
2209         file->private_data = tfile;
2210         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2211         INIT_LIST_HEAD(&tfile->next);
2212
2213         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2214
2215         return 0;
2216 }
2217
2218 static int tun_chr_close(struct inode *inode, struct file *file)
2219 {
2220         struct tun_file *tfile = file->private_data;
2221         struct net *net = tfile->net;
2222
2223         tun_detach(tfile, true);
2224         put_net(net);
2225
2226         return 0;
2227 }
2228
2229 static const struct file_operations tun_fops = {
2230         .owner  = THIS_MODULE,
2231         .llseek = no_llseek,
2232         .read  = do_sync_read,
2233         .aio_read  = tun_chr_aio_read,
2234         .write = do_sync_write,
2235         .aio_write = tun_chr_aio_write,
2236         .poll   = tun_chr_poll,
2237         .unlocked_ioctl = tun_chr_ioctl,
2238 #ifdef CONFIG_COMPAT
2239         .compat_ioctl = tun_chr_compat_ioctl,
2240 #endif
2241         .open   = tun_chr_open,
2242         .release = tun_chr_close,
2243         .fasync = tun_chr_fasync
2244 };
2245
2246 static struct miscdevice tun_miscdev = {
2247         .minor = TUN_MINOR,
2248         .name = "tun",
2249         .nodename = "net/tun",
2250         .fops = &tun_fops,
2251 };
2252
2253 /* ethtool interface */
2254
2255 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2256 {
2257         cmd->supported          = 0;
2258         cmd->advertising        = 0;
2259         ethtool_cmd_speed_set(cmd, SPEED_10);
2260         cmd->duplex             = DUPLEX_FULL;
2261         cmd->port               = PORT_TP;
2262         cmd->phy_address        = 0;
2263         cmd->transceiver        = XCVR_INTERNAL;
2264         cmd->autoneg            = AUTONEG_DISABLE;
2265         cmd->maxtxpkt           = 0;
2266         cmd->maxrxpkt           = 0;
2267         return 0;
2268 }
2269
2270 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2271 {
2272         struct tun_struct *tun = netdev_priv(dev);
2273
2274         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2275         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2276
2277         switch (tun->flags & TUN_TYPE_MASK) {
2278         case TUN_TUN_DEV:
2279                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2280                 break;
2281         case TUN_TAP_DEV:
2282                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2283                 break;
2284         }
2285 }
2286
2287 static u32 tun_get_msglevel(struct net_device *dev)
2288 {
2289 #ifdef TUN_DEBUG
2290         struct tun_struct *tun = netdev_priv(dev);
2291         return tun->debug;
2292 #else
2293         return -EOPNOTSUPP;
2294 #endif
2295 }
2296
2297 static void tun_set_msglevel(struct net_device *dev, u32 value)
2298 {
2299 #ifdef TUN_DEBUG
2300         struct tun_struct *tun = netdev_priv(dev);
2301         tun->debug = value;
2302 #endif
2303 }
2304
2305 static const struct ethtool_ops tun_ethtool_ops = {
2306         .get_settings   = tun_get_settings,
2307         .get_drvinfo    = tun_get_drvinfo,
2308         .get_msglevel   = tun_get_msglevel,
2309         .set_msglevel   = tun_set_msglevel,
2310         .get_link       = ethtool_op_get_link,
2311         .get_ts_info    = ethtool_op_get_ts_info,
2312 };
2313
2314
2315 static int __init tun_init(void)
2316 {
2317         int ret = 0;
2318
2319         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2320         pr_info("%s\n", DRV_COPYRIGHT);
2321
2322         ret = rtnl_link_register(&tun_link_ops);
2323         if (ret) {
2324                 pr_err("Can't register link_ops\n");
2325                 goto err_linkops;
2326         }
2327
2328         ret = misc_register(&tun_miscdev);
2329         if (ret) {
2330                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2331                 goto err_misc;
2332         }
2333         return  0;
2334 err_misc:
2335         rtnl_link_unregister(&tun_link_ops);
2336 err_linkops:
2337         return ret;
2338 }
2339
2340 static void tun_cleanup(void)
2341 {
2342         misc_deregister(&tun_miscdev);
2343         rtnl_link_unregister(&tun_link_ops);
2344 }
2345
2346 /* Get an underlying socket object from tun file.  Returns error unless file is
2347  * attached to a device.  The returned object works like a packet socket, it
2348  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2349  * holding a reference to the file for as long as the socket is in use. */
2350 struct socket *tun_get_socket(struct file *file)
2351 {
2352         struct tun_file *tfile;
2353         if (file->f_op != &tun_fops)
2354                 return ERR_PTR(-EINVAL);
2355         tfile = file->private_data;
2356         if (!tfile)
2357                 return ERR_PTR(-EBADFD);
2358         return &tfile->socket;
2359 }
2360 EXPORT_SYMBOL_GPL(tun_get_socket);
2361
2362 module_init(tun_init);
2363 module_exit(tun_cleanup);
2364 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2365 MODULE_AUTHOR(DRV_COPYRIGHT);
2366 MODULE_LICENSE("GPL");
2367 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2368 MODULE_ALIAS("devname:net/tun");