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 #include <linux/seq_file.h>
  73
  74 #include <asm/uaccess.h>
  75
  76 /* Uncomment to enable debugging */
  77 /* #define TUN_DEBUG 1 */
  78
  79 #ifdef TUN_DEBUG
  80 static int debug;
  81
  82 #define tun_debug(level, tun, fmt, args...)                     \
  83 do {                                                            \
  84         if (tun->debug)                                         \
  85                 netdev_printk(level, tun->dev, fmt, ##args);    \
  86 } while (0)
  87 #define DBG1(level, fmt, args...)                               \
  88 do {                                                            \
  89         if (debug == 2)                                         \
  90                 printk(level fmt, ##args);                      \
  91 } while (0)
  92 #else
  93 #define tun_debug(level, tun, fmt, args...)                     \
  94 do {                                                            \
  95         if (0)                                                  \
  96                 netdev_printk(level, tun->dev, fmt, ##args);    \
  97 } while (0)
  98 #define DBG1(level, fmt, args...)                               \
  99 do {                                                            \
 100         if (0)                                                  \
 101                 printk(level fmt, ##args);                      \
 102 } while (0)
 103 #endif
 104
 105 #define GOODCOPY_LEN 128
 106
 107 #define FLT_EXACT_COUNT 8
 108 struct tap_filter {
 109         unsigned int    count;    /* Number of addrs. Zero means disabled */
 110         u32             mask[2];  /* Mask of the hashed addrs */
 111         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 112 };
 113
 114 /* DEFAULT_MAX_NUM_RSS_QUEUES were chosen to let the rx/tx queues allocated for
 115  * the netdevice to be fit in one page. So we can make sure the success of
 116  * memory allocation. TODO: increase the limit. */
 117 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
 118 #define MAX_TAP_FLOWS  4096
 119
 120 #define TUN_FLOW_EXPIRE (3 * HZ)
 121
 122 /* A tun_file connects an open character device to a tuntap netdevice. It
 123  * also contains all socket related structures (except sock_fprog and tap_filter)
 124  * to serve as one transmit queue for tuntap device. The sock_fprog and
 125  * tap_filter were kept in tun_struct since they were used for filtering for the
 126  * netdevice not for a specific queue (at least I didn't see the requirement for
 127  * this).
 128  *
 129  * RCU usage:
 130  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 131  * other can only be read while rcu_read_lock or rtnl_lock is held.
 132  */
 133 struct tun_file {
 134         struct sock sk;
 135         struct socket socket;
 136         struct socket_wq wq;
 137         struct tun_struct __rcu *tun;
 138         struct net *net;
 139         struct fasync_struct *fasync;
 140         /* only used for fasnyc */
 141         unsigned int flags;
 142         union {
 143                 u16 queue_index;
 144                 unsigned int ifindex;
 145         };
 146         struct list_head next;
 147         struct tun_struct *detached;
 148 };
 149
 150 struct tun_flow_entry {
 151         struct hlist_node hash_link;
 152         struct rcu_head rcu;
 153         struct tun_struct *tun;
 154
 155         u32 rxhash;
 156         u32 rps_rxhash;
 157         int queue_index;
 158         unsigned long updated;
 159 };
 160
 161 #define TUN_NUM_FLOW_ENTRIES 1024
 162
 163 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 164  * device, socket filter, sndbuf and vnet header size were restore when the
 165  * file were attached to a persist device.
 166  */
 167 struct tun_struct {
 168         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 169         unsigned int            numqueues;
 170         unsigned int            flags;
 171         kuid_t                  owner;
 172         kgid_t                  group;
 173
 174         struct net_device       *dev;
 175         netdev_features_t       set_features;
 176 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 177                           NETIF_F_TSO6|NETIF_F_UFO)
 178
 179         int                     vnet_hdr_sz;
 180         int                     sndbuf;
 181         struct tap_filter       txflt;
 182         struct sock_fprog       fprog;
 183         /* protected by rtnl lock */
 184         bool                    filter_attached;
 185 #ifdef TUN_DEBUG
 186         int debug;
 187 #endif
 188         spinlock_t lock;
 189         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 190         struct timer_list flow_gc_timer;
 191         unsigned long ageing_time;
 192         unsigned int numdisabled;
 193         struct list_head disabled;
 194         void *security;
 195         u32 flow_count;
 196 };
 197
 198 static inline u32 tun_hashfn(u32 rxhash)
 199 {
 200         return rxhash & 0x3ff;
 201 }
 202
 203 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 204 {
 205         struct tun_flow_entry *e;
 206
 207         hlist_for_each_entry_rcu(e, head, hash_link) {
 208                 if (e->rxhash == rxhash)
 209                         return e;
 210         }
 211         return NULL;
 212 }
 213
 214 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 215                                               struct hlist_head *head,
 216                                               u32 rxhash, u16 queue_index)
 217 {
 218         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 219
 220         if (e) {
 221                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 222                           rxhash, queue_index);
 223                 e->updated = jiffies;
 224                 e->rxhash = rxhash;
 225                 e->rps_rxhash = 0;
 226                 e->queue_index = queue_index;
 227                 e->tun = tun;
 228                 hlist_add_head_rcu(&e->hash_link, head);
 229                 ++tun->flow_count;
 230         }
 231         return e;
 232 }
 233
 234 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 235 {
 236         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 237                   e->rxhash, e->queue_index);
 238         sock_rps_reset_flow_hash(e->rps_rxhash);
 239         hlist_del_rcu(&e->hash_link);
 240         kfree_rcu(e, rcu);
 241         --tun->flow_count;
 242 }
 243
 244 static void tun_flow_flush(struct tun_struct *tun)
 245 {
 246         int i;
 247
 248         spin_lock_bh(&tun->lock);
 249         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 250                 struct tun_flow_entry *e;
 251                 struct hlist_node *n;
 252
 253                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 254                         tun_flow_delete(tun, e);
 255         }
 256         spin_unlock_bh(&tun->lock);
 257 }
 258
 259 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 260 {
 261         int i;
 262
 263         spin_lock_bh(&tun->lock);
 264         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 265                 struct tun_flow_entry *e;
 266                 struct hlist_node *n;
 267
 268                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 269                         if (e->queue_index == queue_index)
 270                                 tun_flow_delete(tun, e);
 271                 }
 272         }
 273         spin_unlock_bh(&tun->lock);
 274 }
 275
 276 static void tun_flow_cleanup(unsigned long data)
 277 {
 278         struct tun_struct *tun = (struct tun_struct *)data;
 279         unsigned long delay = tun->ageing_time;
 280         unsigned long next_timer = jiffies + delay;
 281         unsigned long count = 0;
 282         int i;
 283
 284         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 285
 286         spin_lock_bh(&tun->lock);
 287         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 288                 struct tun_flow_entry *e;
 289                 struct hlist_node *n;
 290
 291                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 292                         unsigned long this_timer;
 293                         count++;
 294                         this_timer = e->updated + delay;
 295                         if (time_before_eq(this_timer, jiffies))
 296                                 tun_flow_delete(tun, e);
 297                         else if (time_before(this_timer, next_timer))
 298                                 next_timer = this_timer;
 299                 }
 300         }
 301
 302         if (count)
 303                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 304         spin_unlock_bh(&tun->lock);
 305 }
 306
 307 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 308                             struct tun_file *tfile)
 309 {
 310         struct hlist_head *head;
 311         struct tun_flow_entry *e;
 312         unsigned long delay = tun->ageing_time;
 313         u16 queue_index = tfile->queue_index;
 314
 315         if (!rxhash)
 316                 return;
 317         else
 318                 head = &tun->flows[tun_hashfn(rxhash)];
 319
 320         rcu_read_lock();
 321
 322         /* We may get a very small possibility of OOO during switching, not
 323          * worth to optimize.*/
 324         if (tun->numqueues == 1 || tfile->detached)
 325                 goto unlock;
 326
 327         e = tun_flow_find(head, rxhash);
 328         if (likely(e)) {
 329                 /* TODO: keep queueing to old queue until it's empty? */
 330                 e->queue_index = queue_index;
 331                 e->updated = jiffies;
 332                 sock_rps_record_flow_hash(e->rps_rxhash);
 333         } else {
 334                 spin_lock_bh(&tun->lock);
 335                 if (!tun_flow_find(head, rxhash) &&
 336                     tun->flow_count < MAX_TAP_FLOWS)
 337                         tun_flow_create(tun, head, rxhash, queue_index);
 338
 339                 if (!timer_pending(&tun->flow_gc_timer))
 340                         mod_timer(&tun->flow_gc_timer,
 341                                   round_jiffies_up(jiffies + delay));
 342                 spin_unlock_bh(&tun->lock);
 343         }
 344
 345 unlock:
 346         rcu_read_unlock();
 347 }
 348
 349 /**
 350  * Save the hash received in the stack receive path and update the
 351  * flow_hash table accordingly.
 352  */
 353 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 354 {
 355         if (unlikely(e->rps_rxhash != hash)) {
 356                 sock_rps_reset_flow_hash(e->rps_rxhash);
 357                 e->rps_rxhash = hash;
 358         }
 359 }
 360
 361 /* We try to identify a flow through its rxhash first. The reason that
 362  * we do not check rxq no. is because some cards(e.g 82599), chooses
 363  * the rxq based on the txq where the last packet of the flow comes. As
 364  * the userspace application move between processors, we may get a
 365  * different rxq no. here. If we could not get rxhash, then we would
 366  * hope the rxq no. may help here.
 367  */
 368 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 369                             void *accel_priv)
 370 {
 371         struct tun_struct *tun = netdev_priv(dev);
 372         struct tun_flow_entry *e;
 373         u32 txq = 0;
 374         u32 numqueues = 0;
 375
 376         rcu_read_lock();
 377         numqueues = ACCESS_ONCE(tun->numqueues);
 378
 379         txq = skb_get_hash(skb);
 380         if (txq) {
 381                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 382                 if (e) {
 383                         tun_flow_save_rps_rxhash(e, txq);
 384                         txq = e->queue_index;
 385                 } else
 386                         /* use multiply and shift instead of expensive divide */
 387                         txq = ((u64)txq * numqueues) >> 32;
 388         } else if (likely(skb_rx_queue_recorded(skb))) {
 389                 txq = skb_get_rx_queue(skb);
 390                 while (unlikely(txq >= numqueues))
 391                         txq -= numqueues;
 392         }
 393
 394         rcu_read_unlock();
 395         return txq;
 396 }
 397
 398 static inline bool tun_not_capable(struct tun_struct *tun)
 399 {
 400         const struct cred *cred = current_cred();
 401         struct net *net = dev_net(tun->dev);
 402
 403         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 404                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 405                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 406 }
 407
 408 static void tun_set_real_num_queues(struct tun_struct *tun)
 409 {
 410         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 411         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 412 }
 413
 414 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 415 {
 416         tfile->detached = tun;
 417         list_add_tail(&tfile->next, &tun->disabled);
 418         ++tun->numdisabled;
 419 }
 420
 421 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 422 {
 423         struct tun_struct *tun = tfile->detached;
 424
 425         tfile->detached = NULL;
 426         list_del_init(&tfile->next);
 427         --tun->numdisabled;
 428         return tun;
 429 }
 430
 431 static void tun_queue_purge(struct tun_file *tfile)
 432 {
 433         skb_queue_purge(&tfile->sk.sk_receive_queue);
 434         skb_queue_purge(&tfile->sk.sk_error_queue);
 435 }
 436
 437 static void __tun_detach(struct tun_file *tfile, bool clean)
 438 {
 439         struct tun_file *ntfile;
 440         struct tun_struct *tun;
 441
 442         tun = rtnl_dereference(tfile->tun);
 443
 444         if (tun && !tfile->detached) {
 445                 u16 index = tfile->queue_index;
 446                 BUG_ON(index >= tun->numqueues);
 447
 448                 rcu_assign_pointer(tun->tfiles[index],
 449                                    tun->tfiles[tun->numqueues - 1]);
 450                 ntfile = rtnl_dereference(tun->tfiles[index]);
 451                 ntfile->queue_index = index;
 452
 453                 --tun->numqueues;
 454                 if (clean) {
 455                         rcu_assign_pointer(tfile->tun, NULL);
 456                         sock_put(&tfile->sk);
 457                 } else
 458                         tun_disable_queue(tun, tfile);
 459
 460                 synchronize_net();
 461                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 462                 /* Drop read queue */
 463                 tun_queue_purge(tfile);
 464                 tun_set_real_num_queues(tun);
 465         } else if (tfile->detached && clean) {
 466                 tun = tun_enable_queue(tfile);
 467                 sock_put(&tfile->sk);
 468         }
 469
 470         if (clean) {
 471                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 472                         netif_carrier_off(tun->dev);
 473
 474                         if (!(tun->flags & TUN_PERSIST) &&
 475                             tun->dev->reg_state == NETREG_REGISTERED)
 476                                 unregister_netdevice(tun->dev);
 477                 }
 478
 479                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 480                                  &tfile->socket.flags));
 481                 sk_release_kernel(&tfile->sk);
 482         }
 483 }
 484
 485 static void tun_detach(struct tun_file *tfile, bool clean)
 486 {
 487         rtnl_lock();
 488         __tun_detach(tfile, clean);
 489         rtnl_unlock();
 490 }
 491
 492 static void tun_detach_all(struct net_device *dev)
 493 {
 494         struct tun_struct *tun = netdev_priv(dev);
 495         struct tun_file *tfile, *tmp;
 496         int i, n = tun->numqueues;
 497
 498         for (i = 0; i < n; i++) {
 499                 tfile = rtnl_dereference(tun->tfiles[i]);
 500                 BUG_ON(!tfile);
 501                 wake_up_all(&tfile->wq.wait);
 502                 rcu_assign_pointer(tfile->tun, NULL);
 503                 --tun->numqueues;
 504         }
 505         list_for_each_entry(tfile, &tun->disabled, next) {
 506                 wake_up_all(&tfile->wq.wait);
 507                 rcu_assign_pointer(tfile->tun, NULL);
 508         }
 509         BUG_ON(tun->numqueues != 0);
 510
 511         synchronize_net();
 512         for (i = 0; i < n; i++) {
 513                 tfile = rtnl_dereference(tun->tfiles[i]);
 514                 /* Drop read queue */
 515                 tun_queue_purge(tfile);
 516                 sock_put(&tfile->sk);
 517         }
 518         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 519                 tun_enable_queue(tfile);
 520                 tun_queue_purge(tfile);
 521                 sock_put(&tfile->sk);
 522         }
 523         BUG_ON(tun->numdisabled != 0);
 524
 525         if (tun->flags & TUN_PERSIST)
 526                 module_put(THIS_MODULE);
 527 }
 528
 529 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 530 {
 531         struct tun_file *tfile = file->private_data;
 532         int err;
 533
 534         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 535         if (err < 0)
 536                 goto out;
 537
 538         err = -EINVAL;
 539         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 540                 goto out;
 541
 542         err = -EBUSY;
 543         if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 544                 goto out;
 545
 546         err = -E2BIG;
 547         if (!tfile->detached &&
 548             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 549                 goto out;
 550
 551         err = 0;
 552
 553         /* Re-attach the filter to persist device */
 554         if (!skip_filter && (tun->filter_attached == true)) {
 555                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 556                 if (!err)
 557                         goto out;
 558         }
 559         tfile->queue_index = tun->numqueues;
 560         rcu_assign_pointer(tfile->tun, tun);
 561         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 562         tun->numqueues++;
 563
 564         if (tfile->detached)
 565                 tun_enable_queue(tfile);
 566         else
 567                 sock_hold(&tfile->sk);
 568
 569         tun_set_real_num_queues(tun);
 570
 571         /* device is allowed to go away first, so no need to hold extra
 572          * refcnt.
 573          */
 574
 575 out:
 576         return err;
 577 }
 578
 579 static struct tun_struct *__tun_get(struct tun_file *tfile)
 580 {
 581         struct tun_struct *tun;
 582
 583         rcu_read_lock();
 584         tun = rcu_dereference(tfile->tun);
 585         if (tun)
 586                 dev_hold(tun->dev);
 587         rcu_read_unlock();
 588
 589         return tun;
 590 }
 591
 592 static struct tun_struct *tun_get(struct file *file)
 593 {
 594         return __tun_get(file->private_data);
 595 }
 596
 597 static void tun_put(struct tun_struct *tun)
 598 {
 599         dev_put(tun->dev);
 600 }
 601
 602 /* TAP filtering */
 603 static void addr_hash_set(u32 *mask, const u8 *addr)
 604 {
 605         int n = ether_crc(ETH_ALEN, addr) >> 26;
 606         mask[n >> 5] |= (1 << (n & 31));
 607 }
 608
 609 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 610 {
 611         int n = ether_crc(ETH_ALEN, addr) >> 26;
 612         return mask[n >> 5] & (1 << (n & 31));
 613 }
 614
 615 static int update_filter(struct tap_filter *filter, void __user *arg)
 616 {
 617         struct { u8 u[ETH_ALEN]; } *addr;
 618         struct tun_filter uf;
 619         int err, alen, n, nexact;
 620
 621         if (copy_from_user(&uf, arg, sizeof(uf)))
 622                 return -EFAULT;
 623
 624         if (!uf.count) {
 625                 /* Disabled */
 626                 filter->count = 0;
 627                 return 0;
 628         }
 629
 630         alen = ETH_ALEN * uf.count;
 631         addr = kmalloc(alen, GFP_KERNEL);
 632         if (!addr)
 633                 return -ENOMEM;
 634
 635         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 636                 err = -EFAULT;
 637                 goto done;
 638         }
 639
 640         /* The filter is updated without holding any locks. Which is
 641          * perfectly safe. We disable it first and in the worst
 642          * case we'll accept a few undesired packets. */
 643         filter->count = 0;
 644         wmb();
 645
 646         /* Use first set of addresses as an exact filter */
 647         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 648                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 649
 650         nexact = n;
 651
 652         /* Remaining multicast addresses are hashed,
 653          * unicast will leave the filter disabled. */
 654         memset(filter->mask, 0, sizeof(filter->mask));
 655         for (; n < uf.count; n++) {
 656                 if (!is_multicast_ether_addr(addr[n].u)) {
 657                         err = 0; /* no filter */
 658                         goto done;
 659                 }
 660                 addr_hash_set(filter->mask, addr[n].u);
 661         }
 662
 663         /* For ALLMULTI just set the mask to all ones.
 664          * This overrides the mask populated above. */
 665         if ((uf.flags & TUN_FLT_ALLMULTI))
 666                 memset(filter->mask, ~0, sizeof(filter->mask));
 667
 668         /* Now enable the filter */
 669         wmb();
 670         filter->count = nexact;
 671
 672         /* Return the number of exact filters */
 673         err = nexact;
 674
 675 done:
 676         kfree(addr);
 677         return err;
 678 }
 679
 680 /* Returns: 0 - drop, !=0 - accept */
 681 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 682 {
 683         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 684          * at this point. */
 685         struct ethhdr *eh = (struct ethhdr *) skb->data;
 686         int i;
 687
 688         /* Exact match */
 689         for (i = 0; i < filter->count; i++)
 690                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 691                         return 1;
 692
 693         /* Inexact match (multicast only) */
 694         if (is_multicast_ether_addr(eh->h_dest))
 695                 return addr_hash_test(filter->mask, eh->h_dest);
 696
 697         return 0;
 698 }
 699
 700 /*
 701  * Checks whether the packet is accepted or not.
 702  * Returns: 0 - drop, !=0 - accept
 703  */
 704 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 705 {
 706         if (!filter->count)
 707                 return 1;
 708
 709         return run_filter(filter, skb);
 710 }
 711
 712 /* Network device part of the driver */
 713
 714 static const struct ethtool_ops tun_ethtool_ops;
 715
 716 /* Net device detach from fd. */
 717 static void tun_net_uninit(struct net_device *dev)
 718 {
 719         tun_detach_all(dev);
 720 }
 721
 722 /* Net device open. */
 723 static int tun_net_open(struct net_device *dev)
 724 {
 725         netif_tx_start_all_queues(dev);
 726         return 0;
 727 }
 728
 729 /* Net device close. */
 730 static int tun_net_close(struct net_device *dev)
 731 {
 732         netif_tx_stop_all_queues(dev);
 733         return 0;
 734 }
 735
 736 /* Net device start xmit */
 737 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 738 {
 739         struct tun_struct *tun = netdev_priv(dev);
 740         int txq = skb->queue_mapping;
 741         struct tun_file *tfile;
 742         u32 numqueues = 0;
 743
 744         rcu_read_lock();
 745         tfile = rcu_dereference(tun->tfiles[txq]);
 746         numqueues = ACCESS_ONCE(tun->numqueues);
 747
 748         /* Drop packet if interface is not attached */
 749         if (txq >= numqueues)
 750                 goto drop;
 751
 752         if (numqueues == 1) {
 753                 /* Select queue was not called for the skbuff, so we extract the
 754                  * RPS hash and save it into the flow_table here.
 755                  */
 756                 __u32 rxhash;
 757
 758                 rxhash = skb_get_hash(skb);
 759                 if (rxhash) {
 760                         struct tun_flow_entry *e;
 761                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 762                                         rxhash);
 763                         if (e)
 764                                 tun_flow_save_rps_rxhash(e, rxhash);
 765                 }
 766         }
 767
 768         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 769
 770         BUG_ON(!tfile);
 771
 772         /* Drop if the filter does not like it.
 773          * This is a noop if the filter is disabled.
 774          * Filter can be enabled only for the TAP devices. */
 775         if (!check_filter(&tun->txflt, skb))
 776                 goto drop;
 777
 778         if (tfile->socket.sk->sk_filter &&
 779             sk_filter(tfile->socket.sk, skb))
 780                 goto drop;
 781
 782         /* Limit the number of packets queued by dividing txq length with the
 783          * number of queues.
 784          */
 785         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
 786                           >= dev->tx_queue_len)
 787                 goto drop;
 788
 789         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 790                 goto drop;
 791
 792         if (skb->sk) {
 793                 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
 794                 sw_tx_timestamp(skb);
 795         }
 796
 797         /* Orphan the skb - required as we might hang on to it
 798          * for indefinite time.
 799          */
 800         skb_orphan(skb);
 801
 802         nf_reset(skb);
 803
 804         /* Enqueue packet */
 805         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 806
 807         /* Notify and wake up reader process */
 808         if (tfile->flags & TUN_FASYNC)
 809                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 810         wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
 811                                    POLLRDNORM | POLLRDBAND);
 812
 813         rcu_read_unlock();
 814         return NETDEV_TX_OK;
 815
 816 drop:
 817         dev->stats.tx_dropped++;
 818         skb_tx_error(skb);
 819         kfree_skb(skb);
 820         rcu_read_unlock();
 821         return NETDEV_TX_OK;
 822 }
 823
 824 static void tun_net_mclist(struct net_device *dev)
 825 {
 826         /*
 827          * This callback is supposed to deal with mc filter in
 828          * _rx_ path and has nothing to do with the _tx_ path.
 829          * In rx path we always accept everything userspace gives us.
 830          */
 831 }
 832
 833 #define MIN_MTU 68
 834 #define MAX_MTU 65535
 835
 836 static int
 837 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 838 {
 839         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 840                 return -EINVAL;
 841         dev->mtu = new_mtu;
 842         return 0;
 843 }
 844
 845 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 846         netdev_features_t features)
 847 {
 848         struct tun_struct *tun = netdev_priv(dev);
 849
 850         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 851 }
 852 #ifdef CONFIG_NET_POLL_CONTROLLER
 853 static void tun_poll_controller(struct net_device *dev)
 854 {
 855         /*
 856          * Tun only receives frames when:
 857          * 1) the char device endpoint gets data from user space
 858          * 2) the tun socket gets a sendmsg call from user space
 859          * Since both of those are synchronous operations, we are guaranteed
 860          * never to have pending data when we poll for it
 861          * so there is nothing to do here but return.
 862          * We need this though so netpoll recognizes us as an interface that
 863          * supports polling, which enables bridge devices in virt setups to
 864          * still use netconsole
 865          */
 866         return;
 867 }
 868 #endif
 869 static const struct net_device_ops tun_netdev_ops = {
 870         .ndo_uninit             = tun_net_uninit,
 871         .ndo_open               = tun_net_open,
 872         .ndo_stop               = tun_net_close,
 873         .ndo_start_xmit         = tun_net_xmit,
 874         .ndo_change_mtu         = tun_net_change_mtu,
 875         .ndo_fix_features       = tun_net_fix_features,
 876         .ndo_select_queue       = tun_select_queue,
 877 #ifdef CONFIG_NET_POLL_CONTROLLER
 878         .ndo_poll_controller    = tun_poll_controller,
 879 #endif
 880 };
 881
 882 static const struct net_device_ops tap_netdev_ops = {
 883         .ndo_uninit             = tun_net_uninit,
 884         .ndo_open               = tun_net_open,
 885         .ndo_stop               = tun_net_close,
 886         .ndo_start_xmit         = tun_net_xmit,
 887         .ndo_change_mtu         = tun_net_change_mtu,
 888         .ndo_fix_features       = tun_net_fix_features,
 889         .ndo_set_rx_mode        = tun_net_mclist,
 890         .ndo_set_mac_address    = eth_mac_addr,
 891         .ndo_validate_addr      = eth_validate_addr,
 892         .ndo_select_queue       = tun_select_queue,
 893 #ifdef CONFIG_NET_POLL_CONTROLLER
 894         .ndo_poll_controller    = tun_poll_controller,
 895 #endif
 896 };
 897
 898 static void tun_flow_init(struct tun_struct *tun)
 899 {
 900         int i;
 901
 902         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 903                 INIT_HLIST_HEAD(&tun->flows[i]);
 904
 905         tun->ageing_time = TUN_FLOW_EXPIRE;
 906         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 907         mod_timer(&tun->flow_gc_timer,
 908                   round_jiffies_up(jiffies + tun->ageing_time));
 909 }
 910
 911 static void tun_flow_uninit(struct tun_struct *tun)
 912 {
 913         del_timer_sync(&tun->flow_gc_timer);
 914         tun_flow_flush(tun);
 915 }
 916
 917 /* Initialize net device. */
 918 static void tun_net_init(struct net_device *dev)
 919 {
 920         struct tun_struct *tun = netdev_priv(dev);
 921
 922         switch (tun->flags & TUN_TYPE_MASK) {
 923         case TUN_TUN_DEV:
 924                 dev->netdev_ops = &tun_netdev_ops;
 925
 926                 /* Point-to-Point TUN Device */
 927                 dev->hard_header_len = 0;
 928                 dev->addr_len = 0;
 929                 dev->mtu = 1500;
 930
 931                 /* Zero header length */
 932                 dev->type = ARPHRD_NONE;
 933                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 934                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 935                 break;
 936
 937         case TUN_TAP_DEV:
 938                 dev->netdev_ops = &tap_netdev_ops;
 939                 /* Ethernet TAP Device */
 940                 ether_setup(dev);
 941                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 942                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 943
 944                 eth_hw_addr_random(dev);
 945
 946                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 947                 break;
 948         }
 949 }
 950
 951 /* Character device part */
 952
 953 /* Poll */
 954 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 955 {
 956         struct tun_file *tfile = file->private_data;
 957         struct tun_struct *tun = __tun_get(tfile);
 958         struct sock *sk;
 959         unsigned int mask = 0;
 960
 961         if (!tun)
 962                 return POLLERR;
 963
 964         sk = tfile->socket.sk;
 965
 966         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 967
 968         poll_wait(file, &tfile->wq.wait, wait);
 969
 970         if (!skb_queue_empty(&sk->sk_receive_queue))
 971                 mask |= POLLIN | POLLRDNORM;
 972
 973         if (sock_writeable(sk) ||
 974             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 975              sock_writeable(sk)))
 976                 mask |= POLLOUT | POLLWRNORM;
 977
 978         if (tun->dev->reg_state != NETREG_REGISTERED)
 979                 mask = POLLERR;
 980
 981         tun_put(tun);
 982         return mask;
 983 }
 984
 985 /* prepad is the amount to reserve at front.  len is length after that.
 986  * linear is a hint as to how much to copy (usually headers). */
 987 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 988                                      size_t prepad, size_t len,
 989                                      size_t linear, int noblock)
 990 {
 991         struct sock *sk = tfile->socket.sk;
 992         struct sk_buff *skb;
 993         int err;
 994
 995         /* Under a page?  Don't bother with paged skb. */
 996         if (prepad + len < PAGE_SIZE || !linear)
 997                 linear = len;
 998
 999         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1000                                    &err, 0);
1001         if (!skb)
1002                 return ERR_PTR(err);
1003
1004         skb_reserve(skb, prepad);
1005         skb_put(skb, linear);
1006         skb->data_len = len - linear;
1007         skb->len += len - linear;
1008
1009         return skb;
1010 }
1011
1012 /* Get packet from user space buffer */
1013 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1014                             void *msg_control, const struct iovec *iv,
1015                             size_t total_len, size_t count, int noblock)
1016 {
1017         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1018         struct sk_buff *skb;
1019         size_t len = total_len, align = NET_SKB_PAD, linear;
1020         struct virtio_net_hdr gso = { 0 };
1021         int good_linear;
1022         int offset = 0;
1023         int copylen;
1024         bool zerocopy = false;
1025         int err;
1026         u32 rxhash;
1027
1028         if (!(tun->flags & TUN_NO_PI)) {
1029                 if (len < sizeof(pi))
1030                         return -EINVAL;
1031                 len -= sizeof(pi);
1032
1033                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1034                         return -EFAULT;
1035                 offset += sizeof(pi);
1036         }
1037
1038         if (tun->flags & TUN_VNET_HDR) {
1039                 if (len < tun->vnet_hdr_sz)
1040                         return -EINVAL;
1041                 len -= tun->vnet_hdr_sz;
1042
1043                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1044                         return -EFAULT;
1045
1046                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1047                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1048                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1049
1050                 if (gso.hdr_len > len)
1051                         return -EINVAL;
1052                 offset += tun->vnet_hdr_sz;
1053         }
1054
1055         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1056                 align += NET_IP_ALIGN;
1057                 if (unlikely(len < ETH_HLEN ||
1058                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1059                         return -EINVAL;
1060         }
1061
1062         good_linear = SKB_MAX_HEAD(align);
1063
1064         if (msg_control) {
1065                 /* There are 256 bytes to be copied in skb, so there is
1066                  * enough room for skb expand head in case it is used.
1067                  * The rest of the buffer is mapped from userspace.
1068                  */
1069                 copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1070                 if (copylen > good_linear)
1071                         copylen = good_linear;
1072                 linear = copylen;
1073                 if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1074                         zerocopy = true;
1075         }
1076
1077         if (!zerocopy) {
1078                 copylen = len;
1079                 if (gso.hdr_len > good_linear)
1080                         linear = good_linear;
1081                 else
1082                         linear = gso.hdr_len;
1083         }
1084
1085         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1086         if (IS_ERR(skb)) {
1087                 if (PTR_ERR(skb) != -EAGAIN)
1088                         tun->dev->stats.rx_dropped++;
1089                 return PTR_ERR(skb);
1090         }
1091
1092         if (zerocopy)
1093                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1094         else {
1095                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1096                 if (!err && msg_control) {
1097                         struct ubuf_info *uarg = msg_control;
1098                         uarg->callback(uarg, false);
1099                 }
1100         }
1101
1102         if (err) {
1103                 tun->dev->stats.rx_dropped++;
1104                 kfree_skb(skb);
1105                 return -EFAULT;
1106         }
1107
1108         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1109                 if (!skb_partial_csum_set(skb, gso.csum_start,
1110                                           gso.csum_offset)) {
1111                         tun->dev->stats.rx_frame_errors++;
1112                         kfree_skb(skb);
1113                         return -EINVAL;
1114                 }
1115         }
1116
1117         switch (tun->flags & TUN_TYPE_MASK) {
1118         case TUN_TUN_DEV:
1119                 if (tun->flags & TUN_NO_PI) {
1120                         switch (skb->data[0] & 0xf0) {
1121                         case 0x40:
1122                                 pi.proto = htons(ETH_P_IP);
1123                                 break;
1124                         case 0x60:
1125                                 pi.proto = htons(ETH_P_IPV6);
1126                                 break;
1127                         default:
1128                                 tun->dev->stats.rx_dropped++;
1129                                 kfree_skb(skb);
1130                                 return -EINVAL;
1131                         }
1132                 }
1133
1134                 skb_reset_mac_header(skb);
1135                 skb->protocol = pi.proto;
1136                 skb->dev = tun->dev;
1137                 break;
1138         case TUN_TAP_DEV:
1139                 skb->protocol = eth_type_trans(skb, tun->dev);
1140                 break;
1141         }
1142
1143         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1144                 pr_debug("GSO!\n");
1145                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1146                 case VIRTIO_NET_HDR_GSO_TCPV4:
1147                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1148                         break;
1149                 case VIRTIO_NET_HDR_GSO_TCPV6:
1150                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1151                         break;
1152                 case VIRTIO_NET_HDR_GSO_UDP:
1153                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1154                         break;
1155                 default:
1156                         tun->dev->stats.rx_frame_errors++;
1157                         kfree_skb(skb);
1158                         return -EINVAL;
1159                 }
1160
1161                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1162                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1163
1164                 skb_shinfo(skb)->gso_size = gso.gso_size;
1165                 if (skb_shinfo(skb)->gso_size == 0) {
1166                         tun->dev->stats.rx_frame_errors++;
1167                         kfree_skb(skb);
1168                         return -EINVAL;
1169                 }
1170
1171                 /* Header must be checked, and gso_segs computed. */
1172                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1173                 skb_shinfo(skb)->gso_segs = 0;
1174         }
1175
1176         /* copy skb_ubuf_info for callback when skb has no error */
1177         if (zerocopy) {
1178                 skb_shinfo(skb)->destructor_arg = msg_control;
1179                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1180                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1181         }
1182
1183         skb_reset_network_header(skb);
1184         skb_probe_transport_header(skb, 0);
1185
1186         rxhash = skb_get_hash(skb);
1187         netif_rx_ni(skb);
1188
1189         tun->dev->stats.rx_packets++;
1190         tun->dev->stats.rx_bytes += len;
1191
1192         tun_flow_update(tun, rxhash, tfile);
1193         return total_len;
1194 }
1195
1196 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1197                               unsigned long count, loff_t pos)
1198 {
1199         struct file *file = iocb->ki_filp;
1200         struct tun_struct *tun = tun_get(file);
1201         struct tun_file *tfile = file->private_data;
1202         ssize_t result;
1203
1204         if (!tun)
1205                 return -EBADFD;
1206
1207         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1208
1209         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1210                               count, file->f_flags & O_NONBLOCK);
1211
1212         tun_put(tun);
1213         return result;
1214 }
1215
1216 /* Put packet to the user space buffer */
1217 static ssize_t tun_put_user(struct tun_struct *tun,
1218                             struct tun_file *tfile,
1219                             struct sk_buff *skb,
1220                             const struct iovec *iv, int len)
1221 {
1222         struct tun_pi pi = { 0, skb->protocol };
1223         ssize_t total = 0;
1224         int vlan_offset = 0, copied;
1225
1226         if (!(tun->flags & TUN_NO_PI)) {
1227                 if ((len -= sizeof(pi)) < 0)
1228                         return -EINVAL;
1229
1230                 if (len < skb->len) {
1231                         /* Packet will be striped */
1232                         pi.flags |= TUN_PKT_STRIP;
1233                 }
1234
1235                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1236                         return -EFAULT;
1237                 total += sizeof(pi);
1238         }
1239
1240         if (tun->flags & TUN_VNET_HDR) {
1241                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1242                 if ((len -= tun->vnet_hdr_sz) < 0)
1243                         return -EINVAL;
1244
1245                 if (skb_is_gso(skb)) {
1246                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1247
1248                         /* This is a hint as to how much should be linear. */
1249                         gso.hdr_len = skb_headlen(skb);
1250                         gso.gso_size = sinfo->gso_size;
1251                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1252                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1253                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1254                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1255                         else if (sinfo->gso_type & SKB_GSO_UDP)
1256                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1257                         else {
1258                                 pr_err("unexpected GSO type: "
1259                                        "0x%x, gso_size %d, hdr_len %d\n",
1260                                        sinfo->gso_type, gso.gso_size,
1261                                        gso.hdr_len);
1262                                 print_hex_dump(KERN_ERR, "tun: ",
1263                                                DUMP_PREFIX_NONE,
1264                                                16, 1, skb->head,
1265                                                min((int)gso.hdr_len, 64), true);
1266                                 WARN_ON_ONCE(1);
1267                                 return -EINVAL;
1268                         }
1269                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1270                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1271                 } else
1272                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1273
1274                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1275                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1276                         gso.csum_start = skb_checksum_start_offset(skb);
1277                         gso.csum_offset = skb->csum_offset;
1278                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1279                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1280                 } /* else everything is zero */
1281
1282                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1283                                                sizeof(gso))))
1284                         return -EFAULT;
1285                 total += tun->vnet_hdr_sz;
1286         }
1287
1288         copied = total;
1289         total += skb->len;
1290         if (!vlan_tx_tag_present(skb)) {
1291                 len = min_t(int, skb->len, len);
1292         } else {
1293                 int copy, ret;
1294                 struct {
1295                         __be16 h_vlan_proto;
1296                         __be16 h_vlan_TCI;
1297                 } veth;
1298
1299                 veth.h_vlan_proto = skb->vlan_proto;
1300                 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1301
1302                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1303                 len = min_t(int, skb->len + VLAN_HLEN, len);
1304                 total += VLAN_HLEN;
1305
1306                 copy = min_t(int, vlan_offset, len);
1307                 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
1308                 len -= copy;
1309                 copied += copy;
1310                 if (ret || !len)
1311                         goto done;
1312
1313                 copy = min_t(int, sizeof(veth), len);
1314                 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
1315                 len -= copy;
1316                 copied += copy;
1317                 if (ret || !len)
1318                         goto done;
1319         }
1320
1321         skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
1322
1323 done:
1324         tun->dev->stats.tx_packets++;
1325         tun->dev->stats.tx_bytes += len;
1326
1327         return total;
1328 }
1329
1330 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1331                            const struct iovec *iv, ssize_t len, int noblock)
1332 {
1333         DECLARE_WAITQUEUE(wait, current);
1334         struct sk_buff *skb;
1335         ssize_t ret = 0;
1336
1337         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1338
1339         if (unlikely(!noblock))
1340                 add_wait_queue(&tfile->wq.wait, &wait);
1341         while (len) {
1342                 if (unlikely(!noblock))
1343                         current->state = TASK_INTERRUPTIBLE;
1344
1345                 /* Read frames from the queue */
1346                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1347                         if (noblock) {
1348                                 ret = -EAGAIN;
1349                                 break;
1350                         }
1351                         if (signal_pending(current)) {
1352                                 ret = -ERESTARTSYS;
1353                                 break;
1354                         }
1355                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1356                                 ret = -EIO;
1357                                 break;
1358                         }
1359
1360                         /* Nothing to read, let's sleep */
1361                         schedule();
1362                         continue;
1363                 }
1364
1365                 ret = tun_put_user(tun, tfile, skb, iv, len);
1366                 kfree_skb(skb);
1367                 break;
1368         }
1369
1370         if (unlikely(!noblock)) {
1371                 current->state = TASK_RUNNING;
1372                 remove_wait_queue(&tfile->wq.wait, &wait);
1373         }
1374
1375         return ret;
1376 }
1377
1378 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1379                             unsigned long count, loff_t pos)
1380 {
1381         struct file *file = iocb->ki_filp;
1382         struct tun_file *tfile = file->private_data;
1383         struct tun_struct *tun = __tun_get(tfile);
1384         ssize_t len, ret;
1385
1386         if (!tun)
1387                 return -EBADFD;
1388         len = iov_length(iv, count);
1389         if (len < 0) {
1390                 ret = -EINVAL;
1391                 goto out;
1392         }
1393
1394         ret = tun_do_read(tun, tfile, iv, len,
1395                           file->f_flags & O_NONBLOCK);
1396         ret = min_t(ssize_t, ret, len);
1397         if (ret > 0)
1398                 iocb->ki_pos = ret;
1399 out:
1400         tun_put(tun);
1401         return ret;
1402 }
1403
1404 static void tun_free_netdev(struct net_device *dev)
1405 {
1406         struct tun_struct *tun = netdev_priv(dev);
1407
1408         BUG_ON(!(list_empty(&tun->disabled)));
1409         tun_flow_uninit(tun);
1410         security_tun_dev_free_security(tun->security);
1411         free_netdev(dev);
1412 }
1413
1414 static void tun_setup(struct net_device *dev)
1415 {
1416         struct tun_struct *tun = netdev_priv(dev);
1417
1418         tun->owner = INVALID_UID;
1419         tun->group = INVALID_GID;
1420
1421         dev->ethtool_ops = &tun_ethtool_ops;
1422         dev->destructor = tun_free_netdev;
1423 }
1424
1425 /* Trivial set of netlink ops to allow deleting tun or tap
1426  * device with netlink.
1427  */
1428 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1429 {
1430         return -EINVAL;
1431 }
1432
1433 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1434         .kind           = DRV_NAME,
1435         .priv_size      = sizeof(struct tun_struct),
1436         .setup          = tun_setup,
1437         .validate       = tun_validate,
1438 };
1439
1440 static void tun_sock_write_space(struct sock *sk)
1441 {
1442         struct tun_file *tfile;
1443         wait_queue_head_t *wqueue;
1444
1445         if (!sock_writeable(sk))
1446                 return;
1447
1448         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1449                 return;
1450
1451         wqueue = sk_sleep(sk);
1452         if (wqueue && waitqueue_active(wqueue))
1453                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1454                                                 POLLWRNORM | POLLWRBAND);
1455
1456         tfile = container_of(sk, struct tun_file, sk);
1457         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1458 }
1459
1460 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1461                        struct msghdr *m, size_t total_len)
1462 {
1463         int ret;
1464         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1465         struct tun_struct *tun = __tun_get(tfile);
1466
1467         if (!tun)
1468                 return -EBADFD;
1469         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1470                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1471         tun_put(tun);
1472         return ret;
1473 }
1474
1475 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1476                        struct msghdr *m, size_t total_len,
1477                        int flags)
1478 {
1479         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1480         struct tun_struct *tun = __tun_get(tfile);
1481         int ret;
1482
1483         if (!tun)
1484                 return -EBADFD;
1485
1486         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1487                 ret = -EINVAL;
1488                 goto out;
1489         }
1490         if (flags & MSG_ERRQUEUE) {
1491                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1492                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1493                 goto out;
1494         }
1495         ret = tun_do_read(tun, tfile, m->msg_iov, total_len,
1496                           flags & MSG_DONTWAIT);
1497         if (ret > total_len) {
1498                 m->msg_flags |= MSG_TRUNC;
1499                 ret = flags & MSG_TRUNC ? ret : total_len;
1500         }
1501 out:
1502         tun_put(tun);
1503         return ret;
1504 }
1505
1506 static int tun_release(struct socket *sock)
1507 {
1508         if (sock->sk)
1509                 sock_put(sock->sk);
1510         return 0;
1511 }
1512
1513 /* Ops structure to mimic raw sockets with tun */
1514 static const struct proto_ops tun_socket_ops = {
1515         .sendmsg = tun_sendmsg,
1516         .recvmsg = tun_recvmsg,
1517         .release = tun_release,
1518 };
1519
1520 static struct proto tun_proto = {
1521         .name           = "tun",
1522         .owner          = THIS_MODULE,
1523         .obj_size       = sizeof(struct tun_file),
1524 };
1525
1526 static int tun_flags(struct tun_struct *tun)
1527 {
1528         int flags = 0;
1529
1530         if (tun->flags & TUN_TUN_DEV)
1531                 flags |= IFF_TUN;
1532         else
1533                 flags |= IFF_TAP;
1534
1535         if (tun->flags & TUN_NO_PI)
1536                 flags |= IFF_NO_PI;
1537
1538         /* This flag has no real effect.  We track the value for backwards
1539          * compatibility.
1540          */
1541         if (tun->flags & TUN_ONE_QUEUE)
1542                 flags |= IFF_ONE_QUEUE;
1543
1544         if (tun->flags & TUN_VNET_HDR)
1545                 flags |= IFF_VNET_HDR;
1546
1547         if (tun->flags & TUN_TAP_MQ)
1548                 flags |= IFF_MULTI_QUEUE;
1549
1550         if (tun->flags & TUN_PERSIST)
1551                 flags |= IFF_PERSIST;
1552
1553         return flags;
1554 }
1555
1556 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1557                               char *buf)
1558 {
1559         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1560         return sprintf(buf, "0x%x\n", tun_flags(tun));
1561 }
1562
1563 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1564                               char *buf)
1565 {
1566         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1567         return uid_valid(tun->owner)?
1568                 sprintf(buf, "%u\n",
1569                         from_kuid_munged(current_user_ns(), tun->owner)):
1570                 sprintf(buf, "-1\n");
1571 }
1572
1573 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1574                               char *buf)
1575 {
1576         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1577         return gid_valid(tun->group) ?
1578                 sprintf(buf, "%u\n",
1579                         from_kgid_munged(current_user_ns(), tun->group)):
1580                 sprintf(buf, "-1\n");
1581 }
1582
1583 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1584 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1585 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1586
1587 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1588 {
1589         struct tun_struct *tun;
1590         struct tun_file *tfile = file->private_data;
1591         struct net_device *dev;
1592         int err;
1593
1594         if (tfile->detached)
1595                 return -EINVAL;
1596
1597         dev = __dev_get_by_name(net, ifr->ifr_name);
1598         if (dev) {
1599                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1600                         return -EBUSY;
1601                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1602                         tun = netdev_priv(dev);
1603                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1604                         tun = netdev_priv(dev);
1605                 else
1606                         return -EINVAL;
1607
1608                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1609                     !!(tun->flags & TUN_TAP_MQ))
1610                         return -EINVAL;
1611
1612                 if (tun_not_capable(tun))
1613                         return -EPERM;
1614                 err = security_tun_dev_open(tun->security);
1615                 if (err < 0)
1616                         return err;
1617
1618                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1619                 if (err < 0)
1620                         return err;
1621
1622                 if (tun->flags & TUN_TAP_MQ &&
1623                     (tun->numqueues + tun->numdisabled > 1)) {
1624                         /* One or more queue has already been attached, no need
1625                          * to initialize the device again.
1626                          */
1627                         return 0;
1628                 }
1629         }
1630         else {
1631                 char *name;
1632                 unsigned long flags = 0;
1633                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1634                              MAX_TAP_QUEUES : 1;
1635
1636                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1637                         return -EPERM;
1638                 err = security_tun_dev_create();
1639                 if (err < 0)
1640                         return err;
1641
1642                 /* Set dev type */
1643                 if (ifr->ifr_flags & IFF_TUN) {
1644                         /* TUN device */
1645                         flags |= TUN_TUN_DEV;
1646                         name = "tun%d";
1647                 } else if (ifr->ifr_flags & IFF_TAP) {
1648                         /* TAP device */
1649                         flags |= TUN_TAP_DEV;
1650                         name = "tap%d";
1651                 } else
1652                         return -EINVAL;
1653
1654                 if (*ifr->ifr_name)
1655                         name = ifr->ifr_name;
1656
1657                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1658                                        tun_setup, queues, queues);
1659
1660                 if (!dev)
1661                         return -ENOMEM;
1662
1663                 dev_net_set(dev, net);
1664                 dev->rtnl_link_ops = &tun_link_ops;
1665                 dev->ifindex = tfile->ifindex;
1666
1667                 tun = netdev_priv(dev);
1668                 tun->dev = dev;
1669                 tun->flags = flags;
1670                 tun->txflt.count = 0;
1671                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1672
1673                 tun->filter_attached = false;
1674                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1675
1676                 spin_lock_init(&tun->lock);
1677
1678                 err = security_tun_dev_alloc_security(&tun->security);
1679                 if (err < 0)
1680                         goto err_free_dev;
1681
1682                 tun_net_init(dev);
1683                 tun_flow_init(tun);
1684
1685                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1686                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1687                                    NETIF_F_HW_VLAN_STAG_TX;
1688                 dev->features = dev->hw_features;
1689                 dev->vlan_features = dev->features;
1690
1691                 INIT_LIST_HEAD(&tun->disabled);
1692                 err = tun_attach(tun, file, false);
1693                 if (err < 0)
1694                         goto err_free_flow;
1695
1696                 err = register_netdevice(tun->dev);
1697                 if (err < 0)
1698                         goto err_detach;
1699
1700                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1701                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1702                     device_create_file(&tun->dev->dev, &dev_attr_group))
1703                         pr_err("Failed to create tun sysfs files\n");
1704         }
1705
1706         netif_carrier_on(tun->dev);
1707
1708         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1709
1710         if (ifr->ifr_flags & IFF_NO_PI)
1711                 tun->flags |= TUN_NO_PI;
1712         else
1713                 tun->flags &= ~TUN_NO_PI;
1714
1715         /* This flag has no real effect.  We track the value for backwards
1716          * compatibility.
1717          */
1718         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1719                 tun->flags |= TUN_ONE_QUEUE;
1720         else
1721                 tun->flags &= ~TUN_ONE_QUEUE;
1722
1723         if (ifr->ifr_flags & IFF_VNET_HDR)
1724                 tun->flags |= TUN_VNET_HDR;
1725         else
1726                 tun->flags &= ~TUN_VNET_HDR;
1727
1728         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1729                 tun->flags |= TUN_TAP_MQ;
1730         else
1731                 tun->flags &= ~TUN_TAP_MQ;
1732
1733         /* Make sure persistent devices do not get stuck in
1734          * xoff state.
1735          */
1736         if (netif_running(tun->dev))
1737                 netif_tx_wake_all_queues(tun->dev);
1738
1739         strcpy(ifr->ifr_name, tun->dev->name);
1740         return 0;
1741
1742 err_detach:
1743         tun_detach_all(dev);
1744 err_free_flow:
1745         tun_flow_uninit(tun);
1746         security_tun_dev_free_security(tun->security);
1747 err_free_dev:
1748         free_netdev(dev);
1749         return err;
1750 }
1751
1752 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1753                        struct ifreq *ifr)
1754 {
1755         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1756
1757         strcpy(ifr->ifr_name, tun->dev->name);
1758
1759         ifr->ifr_flags = tun_flags(tun);
1760
1761 }
1762
1763 /* This is like a cut-down ethtool ops, except done via tun fd so no
1764  * privs required. */
1765 static int set_offload(struct tun_struct *tun, unsigned long arg)
1766 {
1767         netdev_features_t features = 0;
1768
1769         if (arg & TUN_F_CSUM) {
1770                 features |= NETIF_F_HW_CSUM;
1771                 arg &= ~TUN_F_CSUM;
1772
1773                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1774                         if (arg & TUN_F_TSO_ECN) {
1775                                 features |= NETIF_F_TSO_ECN;
1776                                 arg &= ~TUN_F_TSO_ECN;
1777                         }
1778                         if (arg & TUN_F_TSO4)
1779                                 features |= NETIF_F_TSO;
1780                         if (arg & TUN_F_TSO6)
1781                                 features |= NETIF_F_TSO6;
1782                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1783                 }
1784
1785                 if (arg & TUN_F_UFO) {
1786                         features |= NETIF_F_UFO;
1787                         arg &= ~TUN_F_UFO;
1788                 }
1789         }
1790
1791         /* This gives the user a way to test for new features in future by
1792          * trying to set them. */
1793         if (arg)
1794                 return -EINVAL;
1795
1796         tun->set_features = features;
1797         netdev_update_features(tun->dev);
1798
1799         return 0;
1800 }
1801
1802 static void tun_detach_filter(struct tun_struct *tun, int n)
1803 {
1804         int i;
1805         struct tun_file *tfile;
1806
1807         for (i = 0; i < n; i++) {
1808                 tfile = rtnl_dereference(tun->tfiles[i]);
1809                 sk_detach_filter(tfile->socket.sk);
1810         }
1811
1812         tun->filter_attached = false;
1813 }
1814
1815 static int tun_attach_filter(struct tun_struct *tun)
1816 {
1817         int i, ret = 0;
1818         struct tun_file *tfile;
1819
1820         for (i = 0; i < tun->numqueues; i++) {
1821                 tfile = rtnl_dereference(tun->tfiles[i]);
1822                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1823                 if (ret) {
1824                         tun_detach_filter(tun, i);
1825                         return ret;
1826                 }
1827         }
1828
1829         tun->filter_attached = true;
1830         return ret;
1831 }
1832
1833 static void tun_set_sndbuf(struct tun_struct *tun)
1834 {
1835         struct tun_file *tfile;
1836         int i;
1837
1838         for (i = 0; i < tun->numqueues; i++) {
1839                 tfile = rtnl_dereference(tun->tfiles[i]);
1840                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1841         }
1842 }
1843
1844 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1845 {
1846         struct tun_file *tfile = file->private_data;
1847         struct tun_struct *tun;
1848         int ret = 0;
1849
1850         rtnl_lock();
1851
1852         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1853                 tun = tfile->detached;
1854                 if (!tun) {
1855                         ret = -EINVAL;
1856                         goto unlock;
1857                 }
1858                 ret = security_tun_dev_attach_queue(tun->security);
1859                 if (ret < 0)
1860                         goto unlock;
1861                 ret = tun_attach(tun, file, false);
1862         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1863                 tun = rtnl_dereference(tfile->tun);
1864                 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1865                         ret = -EINVAL;
1866                 else
1867                         __tun_detach(tfile, false);
1868         } else
1869                 ret = -EINVAL;
1870
1871 unlock:
1872         rtnl_unlock();
1873         return ret;
1874 }
1875
1876 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1877                             unsigned long arg, int ifreq_len)
1878 {
1879         struct tun_file *tfile = file->private_data;
1880         struct tun_struct *tun;
1881         void __user* argp = (void __user*)arg;
1882         struct ifreq ifr;
1883         kuid_t owner;
1884         kgid_t group;
1885         int sndbuf;
1886         int vnet_hdr_sz;
1887         unsigned int ifindex;
1888         int ret;
1889
1890         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1891                 if (copy_from_user(&ifr, argp, ifreq_len))
1892                         return -EFAULT;
1893         } else {
1894                 memset(&ifr, 0, sizeof(ifr));
1895         }
1896         if (cmd == TUNGETFEATURES) {
1897                 /* Currently this just means: "what IFF flags are valid?".
1898                  * This is needed because we never checked for invalid flags on
1899                  * TUNSETIFF. */
1900                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1901                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1902                                 (unsigned int __user*)argp);
1903         } else if (cmd == TUNSETQUEUE)
1904                 return tun_set_queue(file, &ifr);
1905
1906         ret = 0;
1907         rtnl_lock();
1908
1909         tun = __tun_get(tfile);
1910         if (cmd == TUNSETIFF && !tun) {
1911                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1912
1913                 ret = tun_set_iff(tfile->net, file, &ifr);
1914
1915                 if (ret)
1916                         goto unlock;
1917
1918                 if (copy_to_user(argp, &ifr, ifreq_len))
1919                         ret = -EFAULT;
1920                 goto unlock;
1921         }
1922         if (cmd == TUNSETIFINDEX) {
1923                 ret = -EPERM;
1924                 if (tun)
1925                         goto unlock;
1926
1927                 ret = -EFAULT;
1928                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1929                         goto unlock;
1930
1931                 ret = 0;
1932                 tfile->ifindex = ifindex;
1933                 goto unlock;
1934         }
1935
1936         ret = -EBADFD;
1937         if (!tun)
1938                 goto unlock;
1939
1940         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1941
1942         ret = 0;
1943         switch (cmd) {
1944         case TUNGETIFF:
1945                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1946
1947                 if (tfile->detached)
1948                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
1949                 if (!tfile->socket.sk->sk_filter)
1950                         ifr.ifr_flags |= IFF_NOFILTER;
1951
1952                 if (copy_to_user(argp, &ifr, ifreq_len))
1953                         ret = -EFAULT;
1954                 break;
1955
1956         case TUNSETNOCSUM:
1957                 /* Disable/Enable checksum */
1958
1959                 /* [unimplemented] */
1960                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1961                           arg ? "disabled" : "enabled");
1962                 break;
1963
1964         case TUNSETPERSIST:
1965                 /* Disable/Enable persist mode. Keep an extra reference to the
1966                  * module to prevent the module being unprobed.
1967                  */
1968                 if (arg && !(tun->flags & TUN_PERSIST)) {
1969                         tun->flags |= TUN_PERSIST;
1970                         __module_get(THIS_MODULE);
1971                 }
1972                 if (!arg && (tun->flags & TUN_PERSIST)) {
1973                         tun->flags &= ~TUN_PERSIST;
1974                         module_put(THIS_MODULE);
1975                 }
1976
1977                 tun_debug(KERN_INFO, tun, "persist %s\n",
1978                           arg ? "enabled" : "disabled");
1979                 break;
1980
1981         case TUNSETOWNER:
1982                 /* Set owner of the device */
1983                 owner = make_kuid(current_user_ns(), arg);
1984                 if (!uid_valid(owner)) {
1985                         ret = -EINVAL;
1986                         break;
1987                 }
1988                 tun->owner = owner;
1989                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1990                           from_kuid(&init_user_ns, tun->owner));
1991                 break;
1992
1993         case TUNSETGROUP:
1994                 /* Set group of the device */
1995                 group = make_kgid(current_user_ns(), arg);
1996                 if (!gid_valid(group)) {
1997                         ret = -EINVAL;
1998                         break;
1999                 }
2000                 tun->group = group;
2001                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2002                           from_kgid(&init_user_ns, tun->group));
2003                 break;
2004
2005         case TUNSETLINK:
2006                 /* Only allow setting the type when the interface is down */
2007                 if (tun->dev->flags & IFF_UP) {
2008                         tun_debug(KERN_INFO, tun,
2009                                   "Linktype set failed because interface is up\n");
2010                         ret = -EBUSY;
2011                 } else {
2012                         tun->dev->type = (int) arg;
2013                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2014                                   tun->dev->type);
2015                         ret = 0;
2016                 }
2017                 break;
2018
2019 #ifdef TUN_DEBUG
2020         case TUNSETDEBUG:
2021                 tun->debug = arg;
2022                 break;
2023 #endif
2024         case TUNSETOFFLOAD:
2025                 ret = set_offload(tun, arg);
2026                 break;
2027
2028         case TUNSETTXFILTER:
2029                 /* Can be set only for TAPs */
2030                 ret = -EINVAL;
2031                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2032                         break;
2033                 ret = update_filter(&tun->txflt, (void __user *)arg);
2034                 break;
2035
2036         case SIOCGIFHWADDR:
2037                 /* Get hw address */
2038                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2039                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2040                 if (copy_to_user(argp, &ifr, ifreq_len))
2041                         ret = -EFAULT;
2042                 break;
2043
2044         case SIOCSIFHWADDR:
2045                 /* Set hw address */
2046                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2047                           ifr.ifr_hwaddr.sa_data);
2048
2049                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2050                 break;
2051
2052         case TUNGETSNDBUF:
2053                 sndbuf = tfile->socket.sk->sk_sndbuf;
2054                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2055                         ret = -EFAULT;
2056                 break;
2057
2058         case TUNSETSNDBUF:
2059                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2060                         ret = -EFAULT;
2061                         break;
2062                 }
2063
2064                 tun->sndbuf = sndbuf;
2065                 tun_set_sndbuf(tun);
2066                 break;
2067
2068         case TUNGETVNETHDRSZ:
2069                 vnet_hdr_sz = tun->vnet_hdr_sz;
2070                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2071                         ret = -EFAULT;
2072                 break;
2073
2074         case TUNSETVNETHDRSZ:
2075                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2076                         ret = -EFAULT;
2077                         break;
2078                 }
2079                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2080                         ret = -EINVAL;
2081                         break;
2082                 }
2083
2084                 tun->vnet_hdr_sz = vnet_hdr_sz;
2085                 break;
2086
2087         case TUNATTACHFILTER:
2088                 /* Can be set only for TAPs */
2089                 ret = -EINVAL;
2090                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2091                         break;
2092                 ret = -EFAULT;
2093                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2094                         break;
2095
2096                 ret = tun_attach_filter(tun);
2097                 break;
2098
2099         case TUNDETACHFILTER:
2100                 /* Can be set only for TAPs */
2101                 ret = -EINVAL;
2102                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2103                         break;
2104                 ret = 0;
2105                 tun_detach_filter(tun, tun->numqueues);
2106                 break;
2107
2108         case TUNGETFILTER:
2109                 ret = -EINVAL;
2110                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2111                         break;
2112                 ret = -EFAULT;
2113                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2114                         break;
2115                 ret = 0;
2116                 break;
2117
2118         default:
2119                 ret = -EINVAL;
2120                 break;
2121         }
2122
2123 unlock:
2124         rtnl_unlock();
2125         if (tun)
2126                 tun_put(tun);
2127         return ret;
2128 }
2129
2130 static long tun_chr_ioctl(struct file *file,
2131                           unsigned int cmd, unsigned long arg)
2132 {
2133         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2134 }
2135
2136 #ifdef CONFIG_COMPAT
2137 static long tun_chr_compat_ioctl(struct file *file,
2138                          unsigned int cmd, unsigned long arg)
2139 {
2140         switch (cmd) {
2141         case TUNSETIFF:
2142         case TUNGETIFF:
2143         case TUNSETTXFILTER:
2144         case TUNGETSNDBUF:
2145         case TUNSETSNDBUF:
2146         case SIOCGIFHWADDR:
2147         case SIOCSIFHWADDR:
2148                 arg = (unsigned long)compat_ptr(arg);
2149                 break;
2150         default:
2151                 arg = (compat_ulong_t)arg;
2152                 break;
2153         }
2154
2155         /*
2156          * compat_ifreq is shorter than ifreq, so we must not access beyond
2157          * the end of that structure. All fields that are used in this
2158          * driver are compatible though, we don't need to convert the
2159          * contents.
2160          */
2161         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2162 }
2163 #endif /* CONFIG_COMPAT */
2164
2165 static int tun_chr_fasync(int fd, struct file *file, int on)
2166 {
2167         struct tun_file *tfile = file->private_data;
2168         int ret;
2169
2170         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2171                 goto out;
2172
2173         if (on) {
2174                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2175                 if (ret)
2176                         goto out;
2177                 tfile->flags |= TUN_FASYNC;
2178         } else
2179                 tfile->flags &= ~TUN_FASYNC;
2180         ret = 0;
2181 out:
2182         return ret;
2183 }
2184
2185 static int tun_chr_open(struct inode *inode, struct file * file)
2186 {
2187         struct tun_file *tfile;
2188
2189         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2190
2191         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2192                                             &tun_proto);
2193         if (!tfile)
2194                 return -ENOMEM;
2195         rcu_assign_pointer(tfile->tun, NULL);
2196         tfile->net = get_net(current->nsproxy->net_ns);
2197         tfile->flags = 0;
2198         tfile->ifindex = 0;
2199
2200         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2201         init_waitqueue_head(&tfile->wq.wait);
2202
2203         tfile->socket.file = file;
2204         tfile->socket.ops = &tun_socket_ops;
2205
2206         sock_init_data(&tfile->socket, &tfile->sk);
2207         sk_change_net(&tfile->sk, tfile->net);
2208
2209         tfile->sk.sk_write_space = tun_sock_write_space;
2210         tfile->sk.sk_sndbuf = INT_MAX;
2211
2212         file->private_data = tfile;
2213         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2214         INIT_LIST_HEAD(&tfile->next);
2215
2216         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2217
2218         return 0;
2219 }
2220
2221 static int tun_chr_close(struct inode *inode, struct file *file)
2222 {
2223         struct tun_file *tfile = file->private_data;
2224         struct net *net = tfile->net;
2225
2226         tun_detach(tfile, true);
2227         put_net(net);
2228
2229         return 0;
2230 }
2231
2232 #ifdef CONFIG_PROC_FS
2233 static int tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2234 {
2235         struct tun_struct *tun;
2236         struct ifreq ifr;
2237
2238         memset(&ifr, 0, sizeof(ifr));
2239
2240         rtnl_lock();
2241         tun = tun_get(f);
2242         if (tun)
2243                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2244         rtnl_unlock();
2245
2246         if (tun)
2247                 tun_put(tun);
2248
2249         return seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2250 }
2251 #endif
2252
2253 static const struct file_operations tun_fops = {
2254         .owner  = THIS_MODULE,
2255         .llseek = no_llseek,
2256         .read  = do_sync_read,
2257         .aio_read  = tun_chr_aio_read,
2258         .write = do_sync_write,
2259         .aio_write = tun_chr_aio_write,
2260         .poll   = tun_chr_poll,
2261         .unlocked_ioctl = tun_chr_ioctl,
2262 #ifdef CONFIG_COMPAT
2263         .compat_ioctl = tun_chr_compat_ioctl,
2264 #endif
2265         .open   = tun_chr_open,
2266         .release = tun_chr_close,
2267         .fasync = tun_chr_fasync,
2268 #ifdef CONFIG_PROC_FS
2269         .show_fdinfo = tun_chr_show_fdinfo,
2270 #endif
2271 };
2272
2273 static struct miscdevice tun_miscdev = {
2274         .minor = TUN_MINOR,
2275         .name = "tun",
2276         .nodename = "net/tun",
2277         .fops = &tun_fops,
2278 };
2279
2280 /* ethtool interface */
2281
2282 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2283 {
2284         cmd->supported          = 0;
2285         cmd->advertising        = 0;
2286         ethtool_cmd_speed_set(cmd, SPEED_10);
2287         cmd->duplex             = DUPLEX_FULL;
2288         cmd->port               = PORT_TP;
2289         cmd->phy_address        = 0;
2290         cmd->transceiver        = XCVR_INTERNAL;
2291         cmd->autoneg            = AUTONEG_DISABLE;
2292         cmd->maxtxpkt           = 0;
2293         cmd->maxrxpkt           = 0;
2294         return 0;
2295 }
2296
2297 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2298 {
2299         struct tun_struct *tun = netdev_priv(dev);
2300
2301         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2302         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2303
2304         switch (tun->flags & TUN_TYPE_MASK) {
2305         case TUN_TUN_DEV:
2306                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2307                 break;
2308         case TUN_TAP_DEV:
2309                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2310                 break;
2311         }
2312 }
2313
2314 static u32 tun_get_msglevel(struct net_device *dev)
2315 {
2316 #ifdef TUN_DEBUG
2317         struct tun_struct *tun = netdev_priv(dev);
2318         return tun->debug;
2319 #else
2320         return -EOPNOTSUPP;
2321 #endif
2322 }
2323
2324 static void tun_set_msglevel(struct net_device *dev, u32 value)
2325 {
2326 #ifdef TUN_DEBUG
2327         struct tun_struct *tun = netdev_priv(dev);
2328         tun->debug = value;
2329 #endif
2330 }
2331
2332 static const struct ethtool_ops tun_ethtool_ops = {
2333         .get_settings   = tun_get_settings,
2334         .get_drvinfo    = tun_get_drvinfo,
2335         .get_msglevel   = tun_get_msglevel,
2336         .set_msglevel   = tun_set_msglevel,
2337         .get_link       = ethtool_op_get_link,
2338         .get_ts_info    = ethtool_op_get_ts_info,
2339 };
2340
2341
2342 static int __init tun_init(void)
2343 {
2344         int ret = 0;
2345
2346         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2347         pr_info("%s\n", DRV_COPYRIGHT);
2348
2349         ret = rtnl_link_register(&tun_link_ops);
2350         if (ret) {
2351                 pr_err("Can't register link_ops\n");
2352                 goto err_linkops;
2353         }
2354
2355         ret = misc_register(&tun_miscdev);
2356         if (ret) {
2357                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2358                 goto err_misc;
2359         }
2360         return  0;
2361 err_misc:
2362         rtnl_link_unregister(&tun_link_ops);
2363 err_linkops:
2364         return ret;
2365 }
2366
2367 static void tun_cleanup(void)
2368 {
2369         misc_deregister(&tun_miscdev);
2370         rtnl_link_unregister(&tun_link_ops);
2371 }
2372
2373 /* Get an underlying socket object from tun file.  Returns error unless file is
2374  * attached to a device.  The returned object works like a packet socket, it
2375  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2376  * holding a reference to the file for as long as the socket is in use. */
2377 struct socket *tun_get_socket(struct file *file)
2378 {
2379         struct tun_file *tfile;
2380         if (file->f_op != &tun_fops)
2381                 return ERR_PTR(-EINVAL);
2382         tfile = file->private_data;
2383         if (!tfile)
2384                 return ERR_PTR(-EBADFD);
2385         return &tfile->socket;
2386 }
2387 EXPORT_SYMBOL_GPL(tun_get_socket);
2388
2389 module_init(tun_init);
2390 module_exit(tun_cleanup);
2391 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2392 MODULE_AUTHOR(DRV_COPYRIGHT);
2393 MODULE_LICENSE("GPL");
2394 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2395 MODULE_ALIAS("devname:net/tun");