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/crc32.h>
  64 #include <linux/nsproxy.h>
  65 #include <linux/virtio_net.h>
  66 #include <linux/rcupdate.h>
  67 #include <net/net_namespace.h>
  68 #include <net/netns/generic.h>
  69 #include <net/rtnetlink.h>
  70 #include <net/sock.h>
  71
  72 #include <asm/uaccess.h>
  73
  74 /* Uncomment to enable debugging */
  75 /* #define TUN_DEBUG 1 */
  76
  77 #ifdef TUN_DEBUG
  78 static int debug;
  79
  80 #define tun_debug(level, tun, fmt, args...)                     \
  81 do {                                                            \
  82         if (tun->debug)                                         \
  83                 netdev_printk(level, tun->dev, fmt, ##args);    \
  84 } while (0)
  85 #define DBG1(level, fmt, args...)                               \
  86 do {                                                            \
  87         if (debug == 2)                                         \
  88                 printk(level fmt, ##args);                      \
  89 } while (0)
  90 #else
  91 #define tun_debug(level, tun, fmt, args...)                     \
  92 do {                                                            \
  93         if (0)                                                  \
  94                 netdev_printk(level, tun->dev, fmt, ##args);    \
  95 } while (0)
  96 #define DBG1(level, fmt, args...)                               \
  97 do {                                                            \
  98         if (0)                                                  \
  99                 printk(level fmt, ##args);                      \
 100 } while (0)
 101 #endif
 102
 103 #define GOODCOPY_LEN 128
 104
 105 #define FLT_EXACT_COUNT 8
 106 struct tap_filter {
 107         unsigned int    count;    /* Number of addrs. Zero means disabled */
 108         u32             mask[2];  /* Mask of the hashed addrs */
 109         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 110 };
 111
 112 /* DEFAULT_MAX_NUM_RSS_QUEUES were choosed to let the rx/tx queues allocated for
 113  * the netdevice to be fit in one page. So we can make sure the success of
 114  * memory allocation. TODO: increase the limit. */
 115 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
 116
 117 #define TUN_FLOW_EXPIRE (3 * HZ)
 118
 119 /* A tun_file connects an open character device to a tuntap netdevice. It
 120  * also contains all socket related strctures (except sock_fprog and tap_filter)
 121  * to serve as one transmit queue for tuntap device. The sock_fprog and
 122  * tap_filter were kept in tun_struct since they were used for filtering for the
 123  * netdevice not for a specific queue (at least I didn't see the requirement for
 124  * this).
 125  *
 126  * RCU usage:
 127  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 128  * other can only be read while rcu_read_lock or rtnl_lock is held.
 129  */
 130 struct tun_file {
 131         struct sock sk;
 132         struct socket socket;
 133         struct socket_wq wq;
 134         struct tun_struct __rcu *tun;
 135         struct net *net;
 136         struct fasync_struct *fasync;
 137         /* only used for fasnyc */
 138         unsigned int flags;
 139         u16 queue_index;
 140         struct list_head next;
 141         struct tun_struct *detached;
 142 };
 143
 144 struct tun_flow_entry {
 145         struct hlist_node hash_link;
 146         struct rcu_head rcu;
 147         struct tun_struct *tun;
 148
 149         u32 rxhash;
 150         int queue_index;
 151         unsigned long updated;
 152 };
 153
 154 #define TUN_NUM_FLOW_ENTRIES 1024
 155
 156 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 157  * device, socket filter, sndbuf and vnet header size were restore when the
 158  * file were attached to a persist device.
 159  */
 160 struct tun_struct {
 161         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 162         unsigned int            numqueues;
 163         unsigned int            flags;
 164         kuid_t                  owner;
 165         kgid_t                  group;
 166
 167         struct net_device       *dev;
 168         netdev_features_t       set_features;
 169 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 170                           NETIF_F_TSO6|NETIF_F_UFO)
 171
 172         int                     vnet_hdr_sz;
 173         int                     sndbuf;
 174         struct tap_filter       txflt;
 175         struct sock_fprog       fprog;
 176         /* protected by rtnl lock */
 177         bool                    filter_attached;
 178 #ifdef TUN_DEBUG
 179         int debug;
 180 #endif
 181         spinlock_t lock;
 182         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 183         struct timer_list flow_gc_timer;
 184         unsigned long ageing_time;
 185         unsigned int numdisabled;
 186         struct list_head disabled;
 187         void *security;
 188 };
 189
 190 static inline u32 tun_hashfn(u32 rxhash)
 191 {
 192         return rxhash & 0x3ff;
 193 }
 194
 195 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 196 {
 197         struct tun_flow_entry *e;
 198         struct hlist_node *n;
 199
 200         hlist_for_each_entry_rcu(e, n, head, hash_link) {
 201                 if (e->rxhash == rxhash)
 202                         return e;
 203         }
 204         return NULL;
 205 }
 206
 207 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 208                                               struct hlist_head *head,
 209                                               u32 rxhash, u16 queue_index)
 210 {
 211         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 212
 213         if (e) {
 214                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 215                           rxhash, queue_index);
 216                 e->updated = jiffies;
 217                 e->rxhash = rxhash;
 218                 e->queue_index = queue_index;
 219                 e->tun = tun;
 220                 hlist_add_head_rcu(&e->hash_link, head);
 221         }
 222         return e;
 223 }
 224
 225 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 226 {
 227         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 228                   e->rxhash, e->queue_index);
 229         hlist_del_rcu(&e->hash_link);
 230         kfree_rcu(e, rcu);
 231 }
 232
 233 static void tun_flow_flush(struct tun_struct *tun)
 234 {
 235         int i;
 236
 237         spin_lock_bh(&tun->lock);
 238         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 239                 struct tun_flow_entry *e;
 240                 struct hlist_node *h, *n;
 241
 242                 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link)
 243                         tun_flow_delete(tun, e);
 244         }
 245         spin_unlock_bh(&tun->lock);
 246 }
 247
 248 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 249 {
 250         int i;
 251
 252         spin_lock_bh(&tun->lock);
 253         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 254                 struct tun_flow_entry *e;
 255                 struct hlist_node *h, *n;
 256
 257                 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
 258                         if (e->queue_index == queue_index)
 259                                 tun_flow_delete(tun, e);
 260                 }
 261         }
 262         spin_unlock_bh(&tun->lock);
 263 }
 264
 265 static void tun_flow_cleanup(unsigned long data)
 266 {
 267         struct tun_struct *tun = (struct tun_struct *)data;
 268         unsigned long delay = tun->ageing_time;
 269         unsigned long next_timer = jiffies + delay;
 270         unsigned long count = 0;
 271         int i;
 272
 273         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 274
 275         spin_lock_bh(&tun->lock);
 276         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 277                 struct tun_flow_entry *e;
 278                 struct hlist_node *h, *n;
 279
 280                 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
 281                         unsigned long this_timer;
 282                         count++;
 283                         this_timer = e->updated + delay;
 284                         if (time_before_eq(this_timer, jiffies))
 285                                 tun_flow_delete(tun, e);
 286                         else if (time_before(this_timer, next_timer))
 287                                 next_timer = this_timer;
 288                 }
 289         }
 290
 291         if (count)
 292                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 293         spin_unlock_bh(&tun->lock);
 294 }
 295
 296 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 297                             u16 queue_index)
 298 {
 299         struct hlist_head *head;
 300         struct tun_flow_entry *e;
 301         unsigned long delay = tun->ageing_time;
 302
 303         if (!rxhash)
 304                 return;
 305         else
 306                 head = &tun->flows[tun_hashfn(rxhash)];
 307
 308         rcu_read_lock();
 309
 310         if (tun->numqueues == 1)
 311                 goto unlock;
 312
 313         e = tun_flow_find(head, rxhash);
 314         if (likely(e)) {
 315                 /* TODO: keep queueing to old queue until it's empty? */
 316                 e->queue_index = queue_index;
 317                 e->updated = jiffies;
 318         } else {
 319                 spin_lock_bh(&tun->lock);
 320                 if (!tun_flow_find(head, rxhash))
 321                         tun_flow_create(tun, head, rxhash, queue_index);
 322
 323                 if (!timer_pending(&tun->flow_gc_timer))
 324                         mod_timer(&tun->flow_gc_timer,
 325                                   round_jiffies_up(jiffies + delay));
 326                 spin_unlock_bh(&tun->lock);
 327         }
 328
 329 unlock:
 330         rcu_read_unlock();
 331 }
 332
 333 /* We try to identify a flow through its rxhash first. The reason that
 334  * we do not check rxq no. is becuase some cards(e.g 82599), chooses
 335  * the rxq based on the txq where the last packet of the flow comes. As
 336  * the userspace application move between processors, we may get a
 337  * different rxq no. here. If we could not get rxhash, then we would
 338  * hope the rxq no. may help here.
 339  */
 340 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
 341 {
 342         struct tun_struct *tun = netdev_priv(dev);
 343         struct tun_flow_entry *e;
 344         u32 txq = 0;
 345         u32 numqueues = 0;
 346
 347         rcu_read_lock();
 348         numqueues = tun->numqueues;
 349
 350         txq = skb_get_rxhash(skb);
 351         if (txq) {
 352                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 353                 if (e)
 354                         txq = e->queue_index;
 355                 else
 356                         /* use multiply and shift instead of expensive divide */
 357                         txq = ((u64)txq * numqueues) >> 32;
 358         } else if (likely(skb_rx_queue_recorded(skb))) {
 359                 txq = skb_get_rx_queue(skb);
 360                 while (unlikely(txq >= numqueues))
 361                         txq -= numqueues;
 362         }
 363
 364         rcu_read_unlock();
 365         return txq;
 366 }
 367
 368 static inline bool tun_not_capable(struct tun_struct *tun)
 369 {
 370         const struct cred *cred = current_cred();
 371         struct net *net = dev_net(tun->dev);
 372
 373         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 374                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 375                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 376 }
 377
 378 static void tun_set_real_num_queues(struct tun_struct *tun)
 379 {
 380         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 381         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 382 }
 383
 384 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 385 {
 386         tfile->detached = tun;
 387         list_add_tail(&tfile->next, &tun->disabled);
 388         ++tun->numdisabled;
 389 }
 390
 391 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 392 {
 393         struct tun_struct *tun = tfile->detached;
 394
 395         tfile->detached = NULL;
 396         list_del_init(&tfile->next);
 397         --tun->numdisabled;
 398         return tun;
 399 }
 400
 401 static void __tun_detach(struct tun_file *tfile, bool clean)
 402 {
 403         struct tun_file *ntfile;
 404         struct tun_struct *tun;
 405         struct net_device *dev;
 406
 407         tun = rtnl_dereference(tfile->tun);
 408
 409         if (tun) {
 410                 u16 index = tfile->queue_index;
 411                 BUG_ON(index >= tun->numqueues);
 412                 dev = tun->dev;
 413
 414                 rcu_assign_pointer(tun->tfiles[index],
 415                                    tun->tfiles[tun->numqueues - 1]);
 416                 rcu_assign_pointer(tfile->tun, NULL);
 417                 ntfile = rtnl_dereference(tun->tfiles[index]);
 418                 ntfile->queue_index = index;
 419
 420                 --tun->numqueues;
 421                 if (clean)
 422                         sock_put(&tfile->sk);
 423                 else
 424                         tun_disable_queue(tun, tfile);
 425
 426                 synchronize_net();
 427                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 428                 /* Drop read queue */
 429                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 430                 tun_set_real_num_queues(tun);
 431         } else if (tfile->detached && clean) {
 432                 tun = tun_enable_queue(tfile);
 433                 sock_put(&tfile->sk);
 434         }
 435
 436         if (clean) {
 437                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0 &&
 438                     !(tun->flags & TUN_PERSIST))
 439                         if (tun->dev->reg_state == NETREG_REGISTERED)
 440                                 unregister_netdevice(tun->dev);
 441
 442                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 443                                  &tfile->socket.flags));
 444                 sk_release_kernel(&tfile->sk);
 445         }
 446 }
 447
 448 static void tun_detach(struct tun_file *tfile, bool clean)
 449 {
 450         rtnl_lock();
 451         __tun_detach(tfile, clean);
 452         rtnl_unlock();
 453 }
 454
 455 static void tun_detach_all(struct net_device *dev)
 456 {
 457         struct tun_struct *tun = netdev_priv(dev);
 458         struct tun_file *tfile, *tmp;
 459         int i, n = tun->numqueues;
 460
 461         for (i = 0; i < n; i++) {
 462                 tfile = rtnl_dereference(tun->tfiles[i]);
 463                 BUG_ON(!tfile);
 464                 wake_up_all(&tfile->wq.wait);
 465                 rcu_assign_pointer(tfile->tun, NULL);
 466                 --tun->numqueues;
 467         }
 468         BUG_ON(tun->numqueues != 0);
 469
 470         synchronize_net();
 471         for (i = 0; i < n; i++) {
 472                 tfile = rtnl_dereference(tun->tfiles[i]);
 473                 /* Drop read queue */
 474                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 475                 sock_put(&tfile->sk);
 476         }
 477         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 478                 tun_enable_queue(tfile);
 479                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 480                 sock_put(&tfile->sk);
 481         }
 482         BUG_ON(tun->numdisabled != 0);
 483
 484         if (tun->flags & TUN_PERSIST)
 485                 module_put(THIS_MODULE);
 486 }
 487
 488 static int tun_attach(struct tun_struct *tun, struct file *file)
 489 {
 490         struct tun_file *tfile = file->private_data;
 491         int err;
 492
 493         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 494         if (err < 0)
 495                 goto out;
 496
 497         err = -EINVAL;
 498         if (rtnl_dereference(tfile->tun))
 499                 goto out;
 500
 501         err = -EBUSY;
 502         if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 503                 goto out;
 504
 505         err = -E2BIG;
 506         if (!tfile->detached &&
 507             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 508                 goto out;
 509
 510         err = 0;
 511
 512         /* Re-attach the filter to presist device */
 513         if (tun->filter_attached == true) {
 514                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 515                 if (!err)
 516                         goto out;
 517         }
 518         tfile->queue_index = tun->numqueues;
 519         rcu_assign_pointer(tfile->tun, tun);
 520         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 521         tun->numqueues++;
 522
 523         if (tfile->detached)
 524                 tun_enable_queue(tfile);
 525         else
 526                 sock_hold(&tfile->sk);
 527
 528         tun_set_real_num_queues(tun);
 529
 530         /* device is allowed to go away first, so no need to hold extra
 531          * refcnt.
 532          */
 533
 534 out:
 535         return err;
 536 }
 537
 538 static struct tun_struct *__tun_get(struct tun_file *tfile)
 539 {
 540         struct tun_struct *tun;
 541
 542         rcu_read_lock();
 543         tun = rcu_dereference(tfile->tun);
 544         if (tun)
 545                 dev_hold(tun->dev);
 546         rcu_read_unlock();
 547
 548         return tun;
 549 }
 550
 551 static struct tun_struct *tun_get(struct file *file)
 552 {
 553         return __tun_get(file->private_data);
 554 }
 555
 556 static void tun_put(struct tun_struct *tun)
 557 {
 558         dev_put(tun->dev);
 559 }
 560
 561 /* TAP filtering */
 562 static void addr_hash_set(u32 *mask, const u8 *addr)
 563 {
 564         int n = ether_crc(ETH_ALEN, addr) >> 26;
 565         mask[n >> 5] |= (1 << (n & 31));
 566 }
 567
 568 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 569 {
 570         int n = ether_crc(ETH_ALEN, addr) >> 26;
 571         return mask[n >> 5] & (1 << (n & 31));
 572 }
 573
 574 static int update_filter(struct tap_filter *filter, void __user *arg)
 575 {
 576         struct { u8 u[ETH_ALEN]; } *addr;
 577         struct tun_filter uf;
 578         int err, alen, n, nexact;
 579
 580         if (copy_from_user(&uf, arg, sizeof(uf)))
 581                 return -EFAULT;
 582
 583         if (!uf.count) {
 584                 /* Disabled */
 585                 filter->count = 0;
 586                 return 0;
 587         }
 588
 589         alen = ETH_ALEN * uf.count;
 590         addr = kmalloc(alen, GFP_KERNEL);
 591         if (!addr)
 592                 return -ENOMEM;
 593
 594         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 595                 err = -EFAULT;
 596                 goto done;
 597         }
 598
 599         /* The filter is updated without holding any locks. Which is
 600          * perfectly safe. We disable it first and in the worst
 601          * case we'll accept a few undesired packets. */
 602         filter->count = 0;
 603         wmb();
 604
 605         /* Use first set of addresses as an exact filter */
 606         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 607                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 608
 609         nexact = n;
 610
 611         /* Remaining multicast addresses are hashed,
 612          * unicast will leave the filter disabled. */
 613         memset(filter->mask, 0, sizeof(filter->mask));
 614         for (; n < uf.count; n++) {
 615                 if (!is_multicast_ether_addr(addr[n].u)) {
 616                         err = 0; /* no filter */
 617                         goto done;
 618                 }
 619                 addr_hash_set(filter->mask, addr[n].u);
 620         }
 621
 622         /* For ALLMULTI just set the mask to all ones.
 623          * This overrides the mask populated above. */
 624         if ((uf.flags & TUN_FLT_ALLMULTI))
 625                 memset(filter->mask, ~0, sizeof(filter->mask));
 626
 627         /* Now enable the filter */
 628         wmb();
 629         filter->count = nexact;
 630
 631         /* Return the number of exact filters */
 632         err = nexact;
 633
 634 done:
 635         kfree(addr);
 636         return err;
 637 }
 638
 639 /* Returns: 0 - drop, !=0 - accept */
 640 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 641 {
 642         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 643          * at this point. */
 644         struct ethhdr *eh = (struct ethhdr *) skb->data;
 645         int i;
 646
 647         /* Exact match */
 648         for (i = 0; i < filter->count; i++)
 649                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 650                         return 1;
 651
 652         /* Inexact match (multicast only) */
 653         if (is_multicast_ether_addr(eh->h_dest))
 654                 return addr_hash_test(filter->mask, eh->h_dest);
 655
 656         return 0;
 657 }
 658
 659 /*
 660  * Checks whether the packet is accepted or not.
 661  * Returns: 0 - drop, !=0 - accept
 662  */
 663 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 664 {
 665         if (!filter->count)
 666                 return 1;
 667
 668         return run_filter(filter, skb);
 669 }
 670
 671 /* Network device part of the driver */
 672
 673 static const struct ethtool_ops tun_ethtool_ops;
 674
 675 /* Net device detach from fd. */
 676 static void tun_net_uninit(struct net_device *dev)
 677 {
 678         tun_detach_all(dev);
 679 }
 680
 681 /* Net device open. */
 682 static int tun_net_open(struct net_device *dev)
 683 {
 684         netif_tx_start_all_queues(dev);
 685         return 0;
 686 }
 687
 688 /* Net device close. */
 689 static int tun_net_close(struct net_device *dev)
 690 {
 691         netif_tx_stop_all_queues(dev);
 692         return 0;
 693 }
 694
 695 /* Net device start xmit */
 696 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 697 {
 698         struct tun_struct *tun = netdev_priv(dev);
 699         int txq = skb->queue_mapping;
 700         struct tun_file *tfile;
 701
 702         rcu_read_lock();
 703         tfile = rcu_dereference(tun->tfiles[txq]);
 704
 705         /* Drop packet if interface is not attached */
 706         if (txq >= tun->numqueues)
 707                 goto drop;
 708
 709         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 710
 711         BUG_ON(!tfile);
 712
 713         /* Drop if the filter does not like it.
 714          * This is a noop if the filter is disabled.
 715          * Filter can be enabled only for the TAP devices. */
 716         if (!check_filter(&tun->txflt, skb))
 717                 goto drop;
 718
 719         if (tfile->socket.sk->sk_filter &&
 720             sk_filter(tfile->socket.sk, skb))
 721                 goto drop;
 722
 723         /* Limit the number of packets queued by dividing txq length with the
 724          * number of queues.
 725          */
 726         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
 727                           >= dev->tx_queue_len / tun->numqueues)
 728                 goto drop;
 729
 730         /* Orphan the skb - required as we might hang on to it
 731          * for indefinite time. */
 732         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 733                 goto drop;
 734         skb_orphan(skb);
 735
 736         /* Enqueue packet */
 737         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 738
 739         /* Notify and wake up reader process */
 740         if (tfile->flags & TUN_FASYNC)
 741                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 742         wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
 743                                    POLLRDNORM | POLLRDBAND);
 744
 745         rcu_read_unlock();
 746         return NETDEV_TX_OK;
 747
 748 drop:
 749         dev->stats.tx_dropped++;
 750         skb_tx_error(skb);
 751         kfree_skb(skb);
 752         rcu_read_unlock();
 753         return NETDEV_TX_OK;
 754 }
 755
 756 static void tun_net_mclist(struct net_device *dev)
 757 {
 758         /*
 759          * This callback is supposed to deal with mc filter in
 760          * _rx_ path and has nothing to do with the _tx_ path.
 761          * In rx path we always accept everything userspace gives us.
 762          */
 763 }
 764
 765 #define MIN_MTU 68
 766 #define MAX_MTU 65535
 767
 768 static int
 769 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 770 {
 771         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 772                 return -EINVAL;
 773         dev->mtu = new_mtu;
 774         return 0;
 775 }
 776
 777 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 778         netdev_features_t features)
 779 {
 780         struct tun_struct *tun = netdev_priv(dev);
 781
 782         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 783 }
 784 #ifdef CONFIG_NET_POLL_CONTROLLER
 785 static void tun_poll_controller(struct net_device *dev)
 786 {
 787         /*
 788          * Tun only receives frames when:
 789          * 1) the char device endpoint gets data from user space
 790          * 2) the tun socket gets a sendmsg call from user space
 791          * Since both of those are syncronous operations, we are guaranteed
 792          * never to have pending data when we poll for it
 793          * so theres nothing to do here but return.
 794          * We need this though so netpoll recognizes us as an interface that
 795          * supports polling, which enables bridge devices in virt setups to
 796          * still use netconsole
 797          */
 798         return;
 799 }
 800 #endif
 801 static const struct net_device_ops tun_netdev_ops = {
 802         .ndo_uninit             = tun_net_uninit,
 803         .ndo_open               = tun_net_open,
 804         .ndo_stop               = tun_net_close,
 805         .ndo_start_xmit         = tun_net_xmit,
 806         .ndo_change_mtu         = tun_net_change_mtu,
 807         .ndo_fix_features       = tun_net_fix_features,
 808         .ndo_select_queue       = tun_select_queue,
 809 #ifdef CONFIG_NET_POLL_CONTROLLER
 810         .ndo_poll_controller    = tun_poll_controller,
 811 #endif
 812 };
 813
 814 static const struct net_device_ops tap_netdev_ops = {
 815         .ndo_uninit             = tun_net_uninit,
 816         .ndo_open               = tun_net_open,
 817         .ndo_stop               = tun_net_close,
 818         .ndo_start_xmit         = tun_net_xmit,
 819         .ndo_change_mtu         = tun_net_change_mtu,
 820         .ndo_fix_features       = tun_net_fix_features,
 821         .ndo_set_rx_mode        = tun_net_mclist,
 822         .ndo_set_mac_address    = eth_mac_addr,
 823         .ndo_validate_addr      = eth_validate_addr,
 824         .ndo_select_queue       = tun_select_queue,
 825 #ifdef CONFIG_NET_POLL_CONTROLLER
 826         .ndo_poll_controller    = tun_poll_controller,
 827 #endif
 828 };
 829
 830 static int tun_flow_init(struct tun_struct *tun)
 831 {
 832         int i;
 833
 834         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 835                 INIT_HLIST_HEAD(&tun->flows[i]);
 836
 837         tun->ageing_time = TUN_FLOW_EXPIRE;
 838         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 839         mod_timer(&tun->flow_gc_timer,
 840                   round_jiffies_up(jiffies + tun->ageing_time));
 841
 842         return 0;
 843 }
 844
 845 static void tun_flow_uninit(struct tun_struct *tun)
 846 {
 847         del_timer_sync(&tun->flow_gc_timer);
 848         tun_flow_flush(tun);
 849 }
 850
 851 /* Initialize net device. */
 852 static void tun_net_init(struct net_device *dev)
 853 {
 854         struct tun_struct *tun = netdev_priv(dev);
 855
 856         switch (tun->flags & TUN_TYPE_MASK) {
 857         case TUN_TUN_DEV:
 858                 dev->netdev_ops = &tun_netdev_ops;
 859
 860                 /* Point-to-Point TUN Device */
 861                 dev->hard_header_len = 0;
 862                 dev->addr_len = 0;
 863                 dev->mtu = 1500;
 864
 865                 /* Zero header length */
 866                 dev->type = ARPHRD_NONE;
 867                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 868                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 869                 break;
 870
 871         case TUN_TAP_DEV:
 872                 dev->netdev_ops = &tap_netdev_ops;
 873                 /* Ethernet TAP Device */
 874                 ether_setup(dev);
 875                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 876                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 877
 878                 eth_hw_addr_random(dev);
 879
 880                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 881                 break;
 882         }
 883 }
 884
 885 /* Character device part */
 886
 887 /* Poll */
 888 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 889 {
 890         struct tun_file *tfile = file->private_data;
 891         struct tun_struct *tun = __tun_get(tfile);
 892         struct sock *sk;
 893         unsigned int mask = 0;
 894
 895         if (!tun)
 896                 return POLLERR;
 897
 898         sk = tfile->socket.sk;
 899
 900         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 901
 902         poll_wait(file, &tfile->wq.wait, wait);
 903
 904         if (!skb_queue_empty(&sk->sk_receive_queue))
 905                 mask |= POLLIN | POLLRDNORM;
 906
 907         if (sock_writeable(sk) ||
 908             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 909              sock_writeable(sk)))
 910                 mask |= POLLOUT | POLLWRNORM;
 911
 912         if (tun->dev->reg_state != NETREG_REGISTERED)
 913                 mask = POLLERR;
 914
 915         tun_put(tun);
 916         return mask;
 917 }
 918
 919 /* prepad is the amount to reserve at front.  len is length after that.
 920  * linear is a hint as to how much to copy (usually headers). */
 921 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 922                                      size_t prepad, size_t len,
 923                                      size_t linear, int noblock)
 924 {
 925         struct sock *sk = tfile->socket.sk;
 926         struct sk_buff *skb;
 927         int err;
 928
 929         /* Under a page?  Don't bother with paged skb. */
 930         if (prepad + len < PAGE_SIZE || !linear)
 931                 linear = len;
 932
 933         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 934                                    &err);
 935         if (!skb)
 936                 return ERR_PTR(err);
 937
 938         skb_reserve(skb, prepad);
 939         skb_put(skb, linear);
 940         skb->data_len = len - linear;
 941         skb->len += len - linear;
 942
 943         return skb;
 944 }
 945
 946 /* set skb frags from iovec, this can move to core network code for reuse */
 947 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 948                                   int offset, size_t count)
 949 {
 950         int len = iov_length(from, count) - offset;
 951         int copy = skb_headlen(skb);
 952         int size, offset1 = 0;
 953         int i = 0;
 954
 955         /* Skip over from offset */
 956         while (count && (offset >= from->iov_len)) {
 957                 offset -= from->iov_len;
 958                 ++from;
 959                 --count;
 960         }
 961
 962         /* copy up to skb headlen */
 963         while (count && (copy > 0)) {
 964                 size = min_t(unsigned int, copy, from->iov_len - offset);
 965                 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
 966                                    size))
 967                         return -EFAULT;
 968                 if (copy > size) {
 969                         ++from;
 970                         --count;
 971                         offset = 0;
 972                 } else
 973                         offset += size;
 974                 copy -= size;
 975                 offset1 += size;
 976         }
 977
 978         if (len == offset1)
 979                 return 0;
 980
 981         while (count--) {
 982                 struct page *page[MAX_SKB_FRAGS];
 983                 int num_pages;
 984                 unsigned long base;
 985                 unsigned long truesize;
 986
 987                 len = from->iov_len - offset;
 988                 if (!len) {
 989                         offset = 0;
 990                         ++from;
 991                         continue;
 992                 }
 993                 base = (unsigned long)from->iov_base + offset;
 994                 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
 995                 if (i + size > MAX_SKB_FRAGS)
 996                         return -EMSGSIZE;
 997                 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
 998                 if (num_pages != size) {
 999                         for (i = 0; i < num_pages; i++)
1000                                 put_page(page[i]);
1001                         return -EFAULT;
1002                 }
1003                 truesize = size * PAGE_SIZE;
1004                 skb->data_len += len;
1005                 skb->len += len;
1006                 skb->truesize += truesize;
1007                 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
1008                 while (len) {
1009                         int off = base & ~PAGE_MASK;
1010                         int size = min_t(int, len, PAGE_SIZE - off);
1011                         __skb_fill_page_desc(skb, i, page[i], off, size);
1012                         skb_shinfo(skb)->nr_frags++;
1013                         /* increase sk_wmem_alloc */
1014                         base += size;
1015                         len -= size;
1016                         i++;
1017                 }
1018                 offset = 0;
1019                 ++from;
1020         }
1021         return 0;
1022 }
1023
1024 /* Get packet from user space buffer */
1025 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1026                             void *msg_control, const struct iovec *iv,
1027                             size_t total_len, size_t count, int noblock)
1028 {
1029         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1030         struct sk_buff *skb;
1031         size_t len = total_len, align = NET_SKB_PAD;
1032         struct virtio_net_hdr gso = { 0 };
1033         int offset = 0;
1034         int copylen;
1035         bool zerocopy = false;
1036         int err;
1037         u32 rxhash;
1038
1039         if (!(tun->flags & TUN_NO_PI)) {
1040                 if ((len -= sizeof(pi)) > total_len)
1041                         return -EINVAL;
1042
1043                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1044                         return -EFAULT;
1045                 offset += sizeof(pi);
1046         }
1047
1048         if (tun->flags & TUN_VNET_HDR) {
1049                 if ((len -= tun->vnet_hdr_sz) > total_len)
1050                         return -EINVAL;
1051
1052                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1053                         return -EFAULT;
1054
1055                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1056                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1057                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1058
1059                 if (gso.hdr_len > len)
1060                         return -EINVAL;
1061                 offset += tun->vnet_hdr_sz;
1062         }
1063
1064         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1065                 align += NET_IP_ALIGN;
1066                 if (unlikely(len < ETH_HLEN ||
1067                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1068                         return -EINVAL;
1069         }
1070
1071         if (msg_control)
1072                 zerocopy = true;
1073
1074         if (zerocopy) {
1075                 /* Userspace may produce vectors with count greater than
1076                  * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1077                  * to let the rest of data to be fit in the frags.
1078                  */
1079                 if (count > MAX_SKB_FRAGS) {
1080                         copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1081                         if (copylen < offset)
1082                                 copylen = 0;
1083                         else
1084                                 copylen -= offset;
1085                 } else
1086                                 copylen = 0;
1087                 /* There are 256 bytes to be copied in skb, so there is enough
1088                  * room for skb expand head in case it is used.
1089                  * The rest of the buffer is mapped from userspace.
1090                  */
1091                 if (copylen < gso.hdr_len)
1092                         copylen = gso.hdr_len;
1093                 if (!copylen)
1094                         copylen = GOODCOPY_LEN;
1095         } else
1096                 copylen = len;
1097
1098         skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
1099         if (IS_ERR(skb)) {
1100                 if (PTR_ERR(skb) != -EAGAIN)
1101                         tun->dev->stats.rx_dropped++;
1102                 return PTR_ERR(skb);
1103         }
1104
1105         if (zerocopy)
1106                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1107         else
1108                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1109
1110         if (err) {
1111                 tun->dev->stats.rx_dropped++;
1112                 kfree_skb(skb);
1113                 return -EFAULT;
1114         }
1115
1116         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1117                 if (!skb_partial_csum_set(skb, gso.csum_start,
1118                                           gso.csum_offset)) {
1119                         tun->dev->stats.rx_frame_errors++;
1120                         kfree_skb(skb);
1121                         return -EINVAL;
1122                 }
1123         }
1124
1125         switch (tun->flags & TUN_TYPE_MASK) {
1126         case TUN_TUN_DEV:
1127                 if (tun->flags & TUN_NO_PI) {
1128                         switch (skb->data[0] & 0xf0) {
1129                         case 0x40:
1130                                 pi.proto = htons(ETH_P_IP);
1131                                 break;
1132                         case 0x60:
1133                                 pi.proto = htons(ETH_P_IPV6);
1134                                 break;
1135                         default:
1136                                 tun->dev->stats.rx_dropped++;
1137                                 kfree_skb(skb);
1138                                 return -EINVAL;
1139                         }
1140                 }
1141
1142                 skb_reset_mac_header(skb);
1143                 skb->protocol = pi.proto;
1144                 skb->dev = tun->dev;
1145                 break;
1146         case TUN_TAP_DEV:
1147                 skb->protocol = eth_type_trans(skb, tun->dev);
1148                 break;
1149         }
1150
1151         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1152                 pr_debug("GSO!\n");
1153                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1154                 case VIRTIO_NET_HDR_GSO_TCPV4:
1155                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1156                         break;
1157                 case VIRTIO_NET_HDR_GSO_TCPV6:
1158                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1159                         break;
1160                 case VIRTIO_NET_HDR_GSO_UDP:
1161                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1162                         break;
1163                 default:
1164                         tun->dev->stats.rx_frame_errors++;
1165                         kfree_skb(skb);
1166                         return -EINVAL;
1167                 }
1168
1169                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1170                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1171
1172                 skb_shinfo(skb)->gso_size = gso.gso_size;
1173                 if (skb_shinfo(skb)->gso_size == 0) {
1174                         tun->dev->stats.rx_frame_errors++;
1175                         kfree_skb(skb);
1176                         return -EINVAL;
1177                 }
1178
1179                 /* Header must be checked, and gso_segs computed. */
1180                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1181                 skb_shinfo(skb)->gso_segs = 0;
1182         }
1183
1184         /* copy skb_ubuf_info for callback when skb has no error */
1185         if (zerocopy) {
1186                 skb_shinfo(skb)->destructor_arg = msg_control;
1187                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1188         }
1189
1190         skb_reset_network_header(skb);
1191         rxhash = skb_get_rxhash(skb);
1192         netif_rx_ni(skb);
1193
1194         tun->dev->stats.rx_packets++;
1195         tun->dev->stats.rx_bytes += len;
1196
1197         tun_flow_update(tun, rxhash, tfile->queue_index);
1198         return total_len;
1199 }
1200
1201 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1202                               unsigned long count, loff_t pos)
1203 {
1204         struct file *file = iocb->ki_filp;
1205         struct tun_struct *tun = tun_get(file);
1206         struct tun_file *tfile = file->private_data;
1207         ssize_t result;
1208
1209         if (!tun)
1210                 return -EBADFD;
1211
1212         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1213
1214         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1215                               count, file->f_flags & O_NONBLOCK);
1216
1217         tun_put(tun);
1218         return result;
1219 }
1220
1221 /* Put packet to the user space buffer */
1222 static ssize_t tun_put_user(struct tun_struct *tun,
1223                             struct tun_file *tfile,
1224                             struct sk_buff *skb,
1225                             const struct iovec *iv, int len)
1226 {
1227         struct tun_pi pi = { 0, skb->protocol };
1228         ssize_t total = 0;
1229
1230         if (!(tun->flags & TUN_NO_PI)) {
1231                 if ((len -= sizeof(pi)) < 0)
1232                         return -EINVAL;
1233
1234                 if (len < skb->len) {
1235                         /* Packet will be striped */
1236                         pi.flags |= TUN_PKT_STRIP;
1237                 }
1238
1239                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1240                         return -EFAULT;
1241                 total += sizeof(pi);
1242         }
1243
1244         if (tun->flags & TUN_VNET_HDR) {
1245                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1246                 if ((len -= tun->vnet_hdr_sz) < 0)
1247                         return -EINVAL;
1248
1249                 if (skb_is_gso(skb)) {
1250                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1251
1252                         /* This is a hint as to how much should be linear. */
1253                         gso.hdr_len = skb_headlen(skb);
1254                         gso.gso_size = sinfo->gso_size;
1255                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1256                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1257                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1258                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1259                         else if (sinfo->gso_type & SKB_GSO_UDP)
1260                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1261                         else {
1262                                 pr_err("unexpected GSO type: "
1263                                        "0x%x, gso_size %d, hdr_len %d\n",
1264                                        sinfo->gso_type, gso.gso_size,
1265                                        gso.hdr_len);
1266                                 print_hex_dump(KERN_ERR, "tun: ",
1267                                                DUMP_PREFIX_NONE,
1268                                                16, 1, skb->head,
1269                                                min((int)gso.hdr_len, 64), true);
1270                                 WARN_ON_ONCE(1);
1271                                 return -EINVAL;
1272                         }
1273                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1274                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1275                 } else
1276                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1277
1278                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1279                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1280                         gso.csum_start = skb_checksum_start_offset(skb);
1281                         gso.csum_offset = skb->csum_offset;
1282                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1283                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1284                 } /* else everything is zero */
1285
1286                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1287                                                sizeof(gso))))
1288                         return -EFAULT;
1289                 total += tun->vnet_hdr_sz;
1290         }
1291
1292         len = min_t(int, skb->len, len);
1293
1294         skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1295         total += skb->len;
1296
1297         tun->dev->stats.tx_packets++;
1298         tun->dev->stats.tx_bytes += len;
1299
1300         return total;
1301 }
1302
1303 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1304                            struct kiocb *iocb, const struct iovec *iv,
1305                            ssize_t len, int noblock)
1306 {
1307         DECLARE_WAITQUEUE(wait, current);
1308         struct sk_buff *skb;
1309         ssize_t ret = 0;
1310
1311         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1312
1313         if (unlikely(!noblock))
1314                 add_wait_queue(&tfile->wq.wait, &wait);
1315         while (len) {
1316                 current->state = TASK_INTERRUPTIBLE;
1317
1318                 /* Read frames from the queue */
1319                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1320                         if (noblock) {
1321                                 ret = -EAGAIN;
1322                                 break;
1323                         }
1324                         if (signal_pending(current)) {
1325                                 ret = -ERESTARTSYS;
1326                                 break;
1327                         }
1328                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1329                                 ret = -EIO;
1330                                 break;
1331                         }
1332
1333                         /* Nothing to read, let's sleep */
1334                         schedule();
1335                         continue;
1336                 }
1337
1338                 ret = tun_put_user(tun, tfile, skb, iv, len);
1339                 kfree_skb(skb);
1340                 break;
1341         }
1342
1343         current->state = TASK_RUNNING;
1344         if (unlikely(!noblock))
1345                 remove_wait_queue(&tfile->wq.wait, &wait);
1346
1347         return ret;
1348 }
1349
1350 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1351                             unsigned long count, loff_t pos)
1352 {
1353         struct file *file = iocb->ki_filp;
1354         struct tun_file *tfile = file->private_data;
1355         struct tun_struct *tun = __tun_get(tfile);
1356         ssize_t len, ret;
1357
1358         if (!tun)
1359                 return -EBADFD;
1360         len = iov_length(iv, count);
1361         if (len < 0) {
1362                 ret = -EINVAL;
1363                 goto out;
1364         }
1365
1366         ret = tun_do_read(tun, tfile, iocb, iv, len,
1367                           file->f_flags & O_NONBLOCK);
1368         ret = min_t(ssize_t, ret, len);
1369 out:
1370         tun_put(tun);
1371         return ret;
1372 }
1373
1374 static void tun_free_netdev(struct net_device *dev)
1375 {
1376         struct tun_struct *tun = netdev_priv(dev);
1377
1378         BUG_ON(!(list_empty(&tun->disabled)));
1379         tun_flow_uninit(tun);
1380         security_tun_dev_free_security(tun->security);
1381         free_netdev(dev);
1382 }
1383
1384 static void tun_setup(struct net_device *dev)
1385 {
1386         struct tun_struct *tun = netdev_priv(dev);
1387
1388         tun->owner = INVALID_UID;
1389         tun->group = INVALID_GID;
1390
1391         dev->ethtool_ops = &tun_ethtool_ops;
1392         dev->destructor = tun_free_netdev;
1393 }
1394
1395 /* Trivial set of netlink ops to allow deleting tun or tap
1396  * device with netlink.
1397  */
1398 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1399 {
1400         return -EINVAL;
1401 }
1402
1403 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1404         .kind           = DRV_NAME,
1405         .priv_size      = sizeof(struct tun_struct),
1406         .setup          = tun_setup,
1407         .validate       = tun_validate,
1408 };
1409
1410 static void tun_sock_write_space(struct sock *sk)
1411 {
1412         struct tun_file *tfile;
1413         wait_queue_head_t *wqueue;
1414
1415         if (!sock_writeable(sk))
1416                 return;
1417
1418         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1419                 return;
1420
1421         wqueue = sk_sleep(sk);
1422         if (wqueue && waitqueue_active(wqueue))
1423                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1424                                                 POLLWRNORM | POLLWRBAND);
1425
1426         tfile = container_of(sk, struct tun_file, sk);
1427         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1428 }
1429
1430 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1431                        struct msghdr *m, size_t total_len)
1432 {
1433         int ret;
1434         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1435         struct tun_struct *tun = __tun_get(tfile);
1436
1437         if (!tun)
1438                 return -EBADFD;
1439         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1440                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1441         tun_put(tun);
1442         return ret;
1443 }
1444
1445
1446 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1447                        struct msghdr *m, size_t total_len,
1448                        int flags)
1449 {
1450         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1451         struct tun_struct *tun = __tun_get(tfile);
1452         int ret;
1453
1454         if (!tun)
1455                 return -EBADFD;
1456
1457         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1458                 return -EINVAL;
1459         ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1460                           flags & MSG_DONTWAIT);
1461         if (ret > total_len) {
1462                 m->msg_flags |= MSG_TRUNC;
1463                 ret = flags & MSG_TRUNC ? ret : total_len;
1464         }
1465         tun_put(tun);
1466         return ret;
1467 }
1468
1469 static int tun_release(struct socket *sock)
1470 {
1471         if (sock->sk)
1472                 sock_put(sock->sk);
1473         return 0;
1474 }
1475
1476 /* Ops structure to mimic raw sockets with tun */
1477 static const struct proto_ops tun_socket_ops = {
1478         .sendmsg = tun_sendmsg,
1479         .recvmsg = tun_recvmsg,
1480         .release = tun_release,
1481 };
1482
1483 static struct proto tun_proto = {
1484         .name           = "tun",
1485         .owner          = THIS_MODULE,
1486         .obj_size       = sizeof(struct tun_file),
1487 };
1488
1489 static int tun_flags(struct tun_struct *tun)
1490 {
1491         int flags = 0;
1492
1493         if (tun->flags & TUN_TUN_DEV)
1494                 flags |= IFF_TUN;
1495         else
1496                 flags |= IFF_TAP;
1497
1498         if (tun->flags & TUN_NO_PI)
1499                 flags |= IFF_NO_PI;
1500
1501         /* This flag has no real effect.  We track the value for backwards
1502          * compatibility.
1503          */
1504         if (tun->flags & TUN_ONE_QUEUE)
1505                 flags |= IFF_ONE_QUEUE;
1506
1507         if (tun->flags & TUN_VNET_HDR)
1508                 flags |= IFF_VNET_HDR;
1509
1510         if (tun->flags & TUN_TAP_MQ)
1511                 flags |= IFF_MULTI_QUEUE;
1512
1513         return flags;
1514 }
1515
1516 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1517                               char *buf)
1518 {
1519         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1520         return sprintf(buf, "0x%x\n", tun_flags(tun));
1521 }
1522
1523 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1524                               char *buf)
1525 {
1526         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1527         return uid_valid(tun->owner)?
1528                 sprintf(buf, "%u\n",
1529                         from_kuid_munged(current_user_ns(), tun->owner)):
1530                 sprintf(buf, "-1\n");
1531 }
1532
1533 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1534                               char *buf)
1535 {
1536         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1537         return gid_valid(tun->group) ?
1538                 sprintf(buf, "%u\n",
1539                         from_kgid_munged(current_user_ns(), tun->group)):
1540                 sprintf(buf, "-1\n");
1541 }
1542
1543 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1544 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1545 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1546
1547 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1548 {
1549         struct tun_struct *tun;
1550         struct tun_file *tfile = file->private_data;
1551         struct net_device *dev;
1552         int err;
1553
1554         if (tfile->detached)
1555                 return -EINVAL;
1556
1557         dev = __dev_get_by_name(net, ifr->ifr_name);
1558         if (dev) {
1559                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1560                         return -EBUSY;
1561                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1562                         tun = netdev_priv(dev);
1563                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1564                         tun = netdev_priv(dev);
1565                 else
1566                         return -EINVAL;
1567
1568                 if (tun_not_capable(tun))
1569                         return -EPERM;
1570                 err = security_tun_dev_open(tun->security);
1571                 if (err < 0)
1572                         return err;
1573
1574                 err = tun_attach(tun, file);
1575                 if (err < 0)
1576                         return err;
1577
1578                 if (tun->flags & TUN_TAP_MQ &&
1579                     (tun->numqueues + tun->numdisabled > 1))
1580                         return err;
1581         }
1582         else {
1583                 char *name;
1584                 unsigned long flags = 0;
1585                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1586                              MAX_TAP_QUEUES : 1;
1587
1588                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1589                         return -EPERM;
1590                 err = security_tun_dev_create();
1591                 if (err < 0)
1592                         return err;
1593
1594                 /* Set dev type */
1595                 if (ifr->ifr_flags & IFF_TUN) {
1596                         /* TUN device */
1597                         flags |= TUN_TUN_DEV;
1598                         name = "tun%d";
1599                 } else if (ifr->ifr_flags & IFF_TAP) {
1600                         /* TAP device */
1601                         flags |= TUN_TAP_DEV;
1602                         name = "tap%d";
1603                 } else
1604                         return -EINVAL;
1605
1606                 if (*ifr->ifr_name)
1607                         name = ifr->ifr_name;
1608
1609                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1610                                        tun_setup, queues, queues);
1611
1612                 if (!dev)
1613                         return -ENOMEM;
1614
1615                 dev_net_set(dev, net);
1616                 dev->rtnl_link_ops = &tun_link_ops;
1617
1618                 tun = netdev_priv(dev);
1619                 tun->dev = dev;
1620                 tun->flags = flags;
1621                 tun->txflt.count = 0;
1622                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1623
1624                 tun->filter_attached = false;
1625                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1626
1627                 spin_lock_init(&tun->lock);
1628
1629                 err = security_tun_dev_alloc_security(&tun->security);
1630                 if (err < 0)
1631                         goto err_free_dev;
1632
1633                 tun_net_init(dev);
1634
1635                 err = tun_flow_init(tun);
1636                 if (err < 0)
1637                         goto err_free_dev;
1638
1639                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1640                         TUN_USER_FEATURES;
1641                 dev->features = dev->hw_features;
1642
1643                 INIT_LIST_HEAD(&tun->disabled);
1644                 err = tun_attach(tun, file);
1645                 if (err < 0)
1646                         goto err_free_dev;
1647
1648                 err = register_netdevice(tun->dev);
1649                 if (err < 0)
1650                         goto err_free_dev;
1651
1652                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1653                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1654                     device_create_file(&tun->dev->dev, &dev_attr_group))
1655                         pr_err("Failed to create tun sysfs files\n");
1656
1657                 netif_carrier_on(tun->dev);
1658         }
1659
1660         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1661
1662         if (ifr->ifr_flags & IFF_NO_PI)
1663                 tun->flags |= TUN_NO_PI;
1664         else
1665                 tun->flags &= ~TUN_NO_PI;
1666
1667         /* This flag has no real effect.  We track the value for backwards
1668          * compatibility.
1669          */
1670         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1671                 tun->flags |= TUN_ONE_QUEUE;
1672         else
1673                 tun->flags &= ~TUN_ONE_QUEUE;
1674
1675         if (ifr->ifr_flags & IFF_VNET_HDR)
1676                 tun->flags |= TUN_VNET_HDR;
1677         else
1678                 tun->flags &= ~TUN_VNET_HDR;
1679
1680         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1681                 tun->flags |= TUN_TAP_MQ;
1682         else
1683                 tun->flags &= ~TUN_TAP_MQ;
1684
1685         /* Make sure persistent devices do not get stuck in
1686          * xoff state.
1687          */
1688         if (netif_running(tun->dev))
1689                 netif_tx_wake_all_queues(tun->dev);
1690
1691         strcpy(ifr->ifr_name, tun->dev->name);
1692         return 0;
1693
1694  err_free_dev:
1695         free_netdev(dev);
1696         return err;
1697 }
1698
1699 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1700                        struct ifreq *ifr)
1701 {
1702         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1703
1704         strcpy(ifr->ifr_name, tun->dev->name);
1705
1706         ifr->ifr_flags = tun_flags(tun);
1707
1708 }
1709
1710 /* This is like a cut-down ethtool ops, except done via tun fd so no
1711  * privs required. */
1712 static int set_offload(struct tun_struct *tun, unsigned long arg)
1713 {
1714         netdev_features_t features = 0;
1715
1716         if (arg & TUN_F_CSUM) {
1717                 features |= NETIF_F_HW_CSUM;
1718                 arg &= ~TUN_F_CSUM;
1719
1720                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1721                         if (arg & TUN_F_TSO_ECN) {
1722                                 features |= NETIF_F_TSO_ECN;
1723                                 arg &= ~TUN_F_TSO_ECN;
1724                         }
1725                         if (arg & TUN_F_TSO4)
1726                                 features |= NETIF_F_TSO;
1727                         if (arg & TUN_F_TSO6)
1728                                 features |= NETIF_F_TSO6;
1729                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1730                 }
1731
1732                 if (arg & TUN_F_UFO) {
1733                         features |= NETIF_F_UFO;
1734                         arg &= ~TUN_F_UFO;
1735                 }
1736         }
1737
1738         /* This gives the user a way to test for new features in future by
1739          * trying to set them. */
1740         if (arg)
1741                 return -EINVAL;
1742
1743         tun->set_features = features;
1744         netdev_update_features(tun->dev);
1745
1746         return 0;
1747 }
1748
1749 static void tun_detach_filter(struct tun_struct *tun, int n)
1750 {
1751         int i;
1752         struct tun_file *tfile;
1753
1754         for (i = 0; i < n; i++) {
1755                 tfile = rtnl_dereference(tun->tfiles[i]);
1756                 sk_detach_filter(tfile->socket.sk);
1757         }
1758
1759         tun->filter_attached = false;
1760 }
1761
1762 static int tun_attach_filter(struct tun_struct *tun)
1763 {
1764         int i, ret = 0;
1765         struct tun_file *tfile;
1766
1767         for (i = 0; i < tun->numqueues; i++) {
1768                 tfile = rtnl_dereference(tun->tfiles[i]);
1769                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1770                 if (ret) {
1771                         tun_detach_filter(tun, i);
1772                         return ret;
1773                 }
1774         }
1775
1776         tun->filter_attached = true;
1777         return ret;
1778 }
1779
1780 static void tun_set_sndbuf(struct tun_struct *tun)
1781 {
1782         struct tun_file *tfile;
1783         int i;
1784
1785         for (i = 0; i < tun->numqueues; i++) {
1786                 tfile = rtnl_dereference(tun->tfiles[i]);
1787                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1788         }
1789 }
1790
1791 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1792 {
1793         struct tun_file *tfile = file->private_data;
1794         struct tun_struct *tun;
1795         int ret = 0;
1796
1797         rtnl_lock();
1798
1799         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1800                 tun = tfile->detached;
1801                 if (!tun) {
1802                         ret = -EINVAL;
1803                         goto unlock;
1804                 }
1805                 ret = security_tun_dev_attach_queue(tun->security);
1806                 if (ret < 0)
1807                         goto unlock;
1808                 ret = tun_attach(tun, file);
1809         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1810                 tun = rtnl_dereference(tfile->tun);
1811                 if (!tun || !(tun->flags & TUN_TAP_MQ))
1812                         ret = -EINVAL;
1813                 else
1814                         __tun_detach(tfile, false);
1815         } else
1816                 ret = -EINVAL;
1817
1818 unlock:
1819         rtnl_unlock();
1820         return ret;
1821 }
1822
1823 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1824                             unsigned long arg, int ifreq_len)
1825 {
1826         struct tun_file *tfile = file->private_data;
1827         struct tun_struct *tun;
1828         void __user* argp = (void __user*)arg;
1829         struct ifreq ifr;
1830         kuid_t owner;
1831         kgid_t group;
1832         int sndbuf;
1833         int vnet_hdr_sz;
1834         int ret;
1835
1836         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1837                 if (copy_from_user(&ifr, argp, ifreq_len))
1838                         return -EFAULT;
1839         } else {
1840                 memset(&ifr, 0, sizeof(ifr));
1841         }
1842         if (cmd == TUNGETFEATURES) {
1843                 /* Currently this just means: "what IFF flags are valid?".
1844                  * This is needed because we never checked for invalid flags on
1845                  * TUNSETIFF. */
1846                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1847                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1848                                 (unsigned int __user*)argp);
1849         } else if (cmd == TUNSETQUEUE)
1850                 return tun_set_queue(file, &ifr);
1851
1852         ret = 0;
1853         rtnl_lock();
1854
1855         tun = __tun_get(tfile);
1856         if (cmd == TUNSETIFF && !tun) {
1857                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1858
1859                 ret = tun_set_iff(tfile->net, file, &ifr);
1860
1861                 if (ret)
1862                         goto unlock;
1863
1864                 if (copy_to_user(argp, &ifr, ifreq_len))
1865                         ret = -EFAULT;
1866                 goto unlock;
1867         }
1868
1869         ret = -EBADFD;
1870         if (!tun)
1871                 goto unlock;
1872
1873         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1874
1875         ret = 0;
1876         switch (cmd) {
1877         case TUNGETIFF:
1878                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1879
1880                 if (copy_to_user(argp, &ifr, ifreq_len))
1881                         ret = -EFAULT;
1882                 break;
1883
1884         case TUNSETNOCSUM:
1885                 /* Disable/Enable checksum */
1886
1887                 /* [unimplemented] */
1888                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1889                           arg ? "disabled" : "enabled");
1890                 break;
1891
1892         case TUNSETPERSIST:
1893                 /* Disable/Enable persist mode. Keep an extra reference to the
1894                  * module to prevent the module being unprobed.
1895                  */
1896                 if (arg && !(tun->flags & TUN_PERSIST)) {
1897                         tun->flags |= TUN_PERSIST;
1898                         __module_get(THIS_MODULE);
1899                 }
1900                 if (!arg && (tun->flags & TUN_PERSIST)) {
1901                         tun->flags &= ~TUN_PERSIST;
1902                         module_put(THIS_MODULE);
1903                 }
1904
1905                 tun_debug(KERN_INFO, tun, "persist %s\n",
1906                           arg ? "enabled" : "disabled");
1907                 break;
1908
1909         case TUNSETOWNER:
1910                 /* Set owner of the device */
1911                 owner = make_kuid(current_user_ns(), arg);
1912                 if (!uid_valid(owner)) {
1913                         ret = -EINVAL;
1914                         break;
1915                 }
1916                 tun->owner = owner;
1917                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1918                           from_kuid(&init_user_ns, tun->owner));
1919                 break;
1920
1921         case TUNSETGROUP:
1922                 /* Set group of the device */
1923                 group = make_kgid(current_user_ns(), arg);
1924                 if (!gid_valid(group)) {
1925                         ret = -EINVAL;
1926                         break;
1927                 }
1928                 tun->group = group;
1929                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1930                           from_kgid(&init_user_ns, tun->group));
1931                 break;
1932
1933         case TUNSETLINK:
1934                 /* Only allow setting the type when the interface is down */
1935                 if (tun->dev->flags & IFF_UP) {
1936                         tun_debug(KERN_INFO, tun,
1937                                   "Linktype set failed because interface is up\n");
1938                         ret = -EBUSY;
1939                 } else {
1940                         tun->dev->type = (int) arg;
1941                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1942                                   tun->dev->type);
1943                         ret = 0;
1944                 }
1945                 break;
1946
1947 #ifdef TUN_DEBUG
1948         case TUNSETDEBUG:
1949                 tun->debug = arg;
1950                 break;
1951 #endif
1952         case TUNSETOFFLOAD:
1953                 ret = set_offload(tun, arg);
1954                 break;
1955
1956         case TUNSETTXFILTER:
1957                 /* Can be set only for TAPs */
1958                 ret = -EINVAL;
1959                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1960                         break;
1961                 ret = update_filter(&tun->txflt, (void __user *)arg);
1962                 break;
1963
1964         case SIOCGIFHWADDR:
1965                 /* Get hw address */
1966                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1967                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1968                 if (copy_to_user(argp, &ifr, ifreq_len))
1969                         ret = -EFAULT;
1970                 break;
1971
1972         case SIOCSIFHWADDR:
1973                 /* Set hw address */
1974                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1975                           ifr.ifr_hwaddr.sa_data);
1976
1977                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1978                 break;
1979
1980         case TUNGETSNDBUF:
1981                 sndbuf = tfile->socket.sk->sk_sndbuf;
1982                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1983                         ret = -EFAULT;
1984                 break;
1985
1986         case TUNSETSNDBUF:
1987                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1988                         ret = -EFAULT;
1989                         break;
1990                 }
1991
1992                 tun->sndbuf = sndbuf;
1993                 tun_set_sndbuf(tun);
1994                 break;
1995
1996         case TUNGETVNETHDRSZ:
1997                 vnet_hdr_sz = tun->vnet_hdr_sz;
1998                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1999                         ret = -EFAULT;
2000                 break;
2001
2002         case TUNSETVNETHDRSZ:
2003                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2004                         ret = -EFAULT;
2005                         break;
2006                 }
2007                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2008                         ret = -EINVAL;
2009                         break;
2010                 }
2011
2012                 tun->vnet_hdr_sz = vnet_hdr_sz;
2013                 break;
2014
2015         case TUNATTACHFILTER:
2016                 /* Can be set only for TAPs */
2017                 ret = -EINVAL;
2018                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2019                         break;
2020                 ret = -EFAULT;
2021                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2022                         break;
2023
2024                 ret = tun_attach_filter(tun);
2025                 break;
2026
2027         case TUNDETACHFILTER:
2028                 /* Can be set only for TAPs */
2029                 ret = -EINVAL;
2030                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2031                         break;
2032                 ret = 0;
2033                 tun_detach_filter(tun, tun->numqueues);
2034                 break;
2035
2036         default:
2037                 ret = -EINVAL;
2038                 break;
2039         }
2040
2041 unlock:
2042         rtnl_unlock();
2043         if (tun)
2044                 tun_put(tun);
2045         return ret;
2046 }
2047
2048 static long tun_chr_ioctl(struct file *file,
2049                           unsigned int cmd, unsigned long arg)
2050 {
2051         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2052 }
2053
2054 #ifdef CONFIG_COMPAT
2055 static long tun_chr_compat_ioctl(struct file *file,
2056                          unsigned int cmd, unsigned long arg)
2057 {
2058         switch (cmd) {
2059         case TUNSETIFF:
2060         case TUNGETIFF:
2061         case TUNSETTXFILTER:
2062         case TUNGETSNDBUF:
2063         case TUNSETSNDBUF:
2064         case SIOCGIFHWADDR:
2065         case SIOCSIFHWADDR:
2066                 arg = (unsigned long)compat_ptr(arg);
2067                 break;
2068         default:
2069                 arg = (compat_ulong_t)arg;
2070                 break;
2071         }
2072
2073         /*
2074          * compat_ifreq is shorter than ifreq, so we must not access beyond
2075          * the end of that structure. All fields that are used in this
2076          * driver are compatible though, we don't need to convert the
2077          * contents.
2078          */
2079         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2080 }
2081 #endif /* CONFIG_COMPAT */
2082
2083 static int tun_chr_fasync(int fd, struct file *file, int on)
2084 {
2085         struct tun_file *tfile = file->private_data;
2086         int ret;
2087
2088         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2089                 goto out;
2090
2091         if (on) {
2092                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2093                 if (ret)
2094                         goto out;
2095                 tfile->flags |= TUN_FASYNC;
2096         } else
2097                 tfile->flags &= ~TUN_FASYNC;
2098         ret = 0;
2099 out:
2100         return ret;
2101 }
2102
2103 static int tun_chr_open(struct inode *inode, struct file * file)
2104 {
2105         struct tun_file *tfile;
2106
2107         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2108
2109         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2110                                             &tun_proto);
2111         if (!tfile)
2112                 return -ENOMEM;
2113         rcu_assign_pointer(tfile->tun, NULL);
2114         tfile->net = get_net(current->nsproxy->net_ns);
2115         tfile->flags = 0;
2116
2117         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2118         init_waitqueue_head(&tfile->wq.wait);
2119
2120         tfile->socket.file = file;
2121         tfile->socket.ops = &tun_socket_ops;
2122
2123         sock_init_data(&tfile->socket, &tfile->sk);
2124         sk_change_net(&tfile->sk, tfile->net);
2125
2126         tfile->sk.sk_write_space = tun_sock_write_space;
2127         tfile->sk.sk_sndbuf = INT_MAX;
2128
2129         file->private_data = tfile;
2130         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2131         INIT_LIST_HEAD(&tfile->next);
2132
2133         return 0;
2134 }
2135
2136 static int tun_chr_close(struct inode *inode, struct file *file)
2137 {
2138         struct tun_file *tfile = file->private_data;
2139         struct net *net = tfile->net;
2140
2141         tun_detach(tfile, true);
2142         put_net(net);
2143
2144         return 0;
2145 }
2146
2147 static const struct file_operations tun_fops = {
2148         .owner  = THIS_MODULE,
2149         .llseek = no_llseek,
2150         .read  = do_sync_read,
2151         .aio_read  = tun_chr_aio_read,
2152         .write = do_sync_write,
2153         .aio_write = tun_chr_aio_write,
2154         .poll   = tun_chr_poll,
2155         .unlocked_ioctl = tun_chr_ioctl,
2156 #ifdef CONFIG_COMPAT
2157         .compat_ioctl = tun_chr_compat_ioctl,
2158 #endif
2159         .open   = tun_chr_open,
2160         .release = tun_chr_close,
2161         .fasync = tun_chr_fasync
2162 };
2163
2164 static struct miscdevice tun_miscdev = {
2165         .minor = TUN_MINOR,
2166         .name = "tun",
2167         .nodename = "net/tun",
2168         .fops = &tun_fops,
2169 };
2170
2171 /* ethtool interface */
2172
2173 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2174 {
2175         cmd->supported          = 0;
2176         cmd->advertising        = 0;
2177         ethtool_cmd_speed_set(cmd, SPEED_10);
2178         cmd->duplex             = DUPLEX_FULL;
2179         cmd->port               = PORT_TP;
2180         cmd->phy_address        = 0;
2181         cmd->transceiver        = XCVR_INTERNAL;
2182         cmd->autoneg            = AUTONEG_DISABLE;
2183         cmd->maxtxpkt           = 0;
2184         cmd->maxrxpkt           = 0;
2185         return 0;
2186 }
2187
2188 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2189 {
2190         struct tun_struct *tun = netdev_priv(dev);
2191
2192         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2193         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2194
2195         switch (tun->flags & TUN_TYPE_MASK) {
2196         case TUN_TUN_DEV:
2197                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2198                 break;
2199         case TUN_TAP_DEV:
2200                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2201                 break;
2202         }
2203 }
2204
2205 static u32 tun_get_msglevel(struct net_device *dev)
2206 {
2207 #ifdef TUN_DEBUG
2208         struct tun_struct *tun = netdev_priv(dev);
2209         return tun->debug;
2210 #else
2211         return -EOPNOTSUPP;
2212 #endif
2213 }
2214
2215 static void tun_set_msglevel(struct net_device *dev, u32 value)
2216 {
2217 #ifdef TUN_DEBUG
2218         struct tun_struct *tun = netdev_priv(dev);
2219         tun->debug = value;
2220 #endif
2221 }
2222
2223 static const struct ethtool_ops tun_ethtool_ops = {
2224         .get_settings   = tun_get_settings,
2225         .get_drvinfo    = tun_get_drvinfo,
2226         .get_msglevel   = tun_get_msglevel,
2227         .set_msglevel   = tun_set_msglevel,
2228         .get_link       = ethtool_op_get_link,
2229 };
2230
2231
2232 static int __init tun_init(void)
2233 {
2234         int ret = 0;
2235
2236         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2237         pr_info("%s\n", DRV_COPYRIGHT);
2238
2239         ret = rtnl_link_register(&tun_link_ops);
2240         if (ret) {
2241                 pr_err("Can't register link_ops\n");
2242                 goto err_linkops;
2243         }
2244
2245         ret = misc_register(&tun_miscdev);
2246         if (ret) {
2247                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2248                 goto err_misc;
2249         }
2250         return  0;
2251 err_misc:
2252         rtnl_link_unregister(&tun_link_ops);
2253 err_linkops:
2254         return ret;
2255 }
2256
2257 static void tun_cleanup(void)
2258 {
2259         misc_deregister(&tun_miscdev);
2260         rtnl_link_unregister(&tun_link_ops);
2261 }
2262
2263 /* Get an underlying socket object from tun file.  Returns error unless file is
2264  * attached to a device.  The returned object works like a packet socket, it
2265  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2266  * holding a reference to the file for as long as the socket is in use. */
2267 struct socket *tun_get_socket(struct file *file)
2268 {
2269         struct tun_file *tfile;
2270         if (file->f_op != &tun_fops)
2271                 return ERR_PTR(-EINVAL);
2272         tfile = file->private_data;
2273         if (!tfile)
2274                 return ERR_PTR(-EBADFD);
2275         return &tfile->socket;
2276 }
2277 EXPORT_SYMBOL_GPL(tun_get_socket);
2278
2279 module_init(tun_init);
2280 module_exit(tun_cleanup);
2281 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2282 MODULE_AUTHOR(DRV_COPYRIGHT);
2283 MODULE_LICENSE("GPL");
2284 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2285 MODULE_ALIAS("devname:net/tun");