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