Pileus Git - ~andy/linux/blob - drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/major.h>
  48 #include <linux/slab.h>
  49 #include <linux/poll.h>
  50 #include <linux/fcntl.h>
  51 #include <linux/init.h>
  52 #include <linux/skbuff.h>
  53 #include <linux/netdevice.h>
  54 #include <linux/etherdevice.h>
  55 #include <linux/miscdevice.h>
  56 #include <linux/ethtool.h>
  57 #include <linux/rtnetlink.h>
  58 #include <linux/compat.h>
  59 #include <linux/if.h>
  60 #include <linux/if_arp.h>
  61 #include <linux/if_ether.h>
  62 #include <linux/if_tun.h>
  63 #include <linux/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 /* 1024 is probably a high enough limit: modern hypervisors seem to support on
 113  * the order of 100-200 CPUs so this leaves us some breathing space if we want
 114  * to match a queue per guest CPU.
 115  */
 116 #define MAX_TAP_QUEUES 1024
 117
 118 #define TUN_FLOW_EXPIRE (3 * HZ)
 119
 120 /* A tun_file connects an open character device to a tuntap netdevice. It
 121  * also contains all socket related strctures (except sock_fprog and tap_filter)
 122  * to serve as one transmit queue for tuntap device. The sock_fprog and
 123  * tap_filter were kept in tun_struct since they were used for filtering for the
 124  * netdevice not for a specific queue (at least I didn't see the reqirement for
 125  * this).
 126  *
 127  * RCU usage:
 128  * The tun_file and tun_struct are loosely coupled, the pointer from on to the
 129  * other can only be read while rcu_read_lock or rtnl_lock is held.
 130  */
 131 struct tun_file {
 132         struct sock sk;
 133         struct socket socket;
 134         struct socket_wq wq;
 135         struct tun_struct __rcu *tun;
 136         struct net *net;
 137         struct fasync_struct *fasync;
 138         /* only used for fasnyc */
 139         unsigned int flags;
 140         u16 queue_index;
 141 };
 142
 143 struct tun_flow_entry {
 144         struct hlist_node hash_link;
 145         struct rcu_head rcu;
 146         struct tun_struct *tun;
 147
 148         u32 rxhash;
 149         int queue_index;
 150         unsigned long updated;
 151 };
 152
 153 #define TUN_NUM_FLOW_ENTRIES 1024
 154
 155 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 156  * device, socket fileter, sndbuf and vnet header size were restore when the
 157  * file were attached to a persist device.
 158  */
 159 struct tun_struct {
 160         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 161         unsigned int            numqueues;
 162         unsigned int            flags;
 163         kuid_t                  owner;
 164         kgid_t                  group;
 165
 166         struct net_device       *dev;
 167         netdev_features_t       set_features;
 168 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 169                           NETIF_F_TSO6|NETIF_F_UFO)
 170
 171         int                     vnet_hdr_sz;
 172         int                     sndbuf;
 173         struct tap_filter       txflt;
 174         struct sock_fprog       fprog;
 175         /* protected by rtnl lock */
 176         bool                    filter_attached;
 177 #ifdef TUN_DEBUG
 178         int debug;
 179 #endif
 180         spinlock_t lock;
 181         struct kmem_cache *flow_cache;
 182         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 183         struct timer_list flow_gc_timer;
 184         unsigned long ageing_time;
 185 };
 186
 187 static inline u32 tun_hashfn(u32 rxhash)
 188 {
 189         return rxhash & 0x3ff;
 190 }
 191
 192 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 193 {
 194         struct tun_flow_entry *e;
 195         struct hlist_node *n;
 196
 197         hlist_for_each_entry_rcu(e, n, head, hash_link) {
 198                 if (e->rxhash == rxhash)
 199                         return e;
 200         }
 201         return NULL;
 202 }
 203
 204 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 205                                               struct hlist_head *head,
 206                                               u32 rxhash, u16 queue_index)
 207 {
 208         struct tun_flow_entry *e = kmem_cache_alloc(tun->flow_cache,
 209                                                     GFP_ATOMIC);
 210         if (e) {
 211                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 212                           rxhash, queue_index);
 213                 e->updated = jiffies;
 214                 e->rxhash = rxhash;
 215                 e->queue_index = queue_index;
 216                 e->tun = tun;
 217                 hlist_add_head_rcu(&e->hash_link, head);
 218         }
 219         return e;
 220 }
 221
 222 static void tun_flow_free(struct rcu_head *head)
 223 {
 224         struct tun_flow_entry *e
 225                 = container_of(head, struct tun_flow_entry, rcu);
 226         kmem_cache_free(e->tun->flow_cache, e);
 227 }
 228
 229 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 230 {
 231         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 232                   e->rxhash, e->queue_index);
 233         hlist_del_rcu(&e->hash_link);
 234         call_rcu(&e->rcu, tun_flow_free);
 235 }
 236
 237 static void tun_flow_flush(struct tun_struct *tun)
 238 {
 239         int i;
 240
 241         spin_lock_bh(&tun->lock);
 242         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 243                 struct tun_flow_entry *e;
 244                 struct hlist_node *h, *n;
 245
 246                 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link)
 247                         tun_flow_delete(tun, e);
 248         }
 249         spin_unlock_bh(&tun->lock);
 250 }
 251
 252 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 253 {
 254         int i;
 255
 256         spin_lock_bh(&tun->lock);
 257         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 258                 struct tun_flow_entry *e;
 259                 struct hlist_node *h, *n;
 260
 261                 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
 262                         if (e->queue_index == queue_index)
 263                                 tun_flow_delete(tun, e);
 264                 }
 265         }
 266         spin_unlock_bh(&tun->lock);
 267 }
 268
 269 static void tun_flow_cleanup(unsigned long data)
 270 {
 271         struct tun_struct *tun = (struct tun_struct *)data;
 272         unsigned long delay = tun->ageing_time;
 273         unsigned long next_timer = jiffies + delay;
 274         unsigned long count = 0;
 275         int i;
 276
 277         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 278
 279         spin_lock_bh(&tun->lock);
 280         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 281                 struct tun_flow_entry *e;
 282                 struct hlist_node *h, *n;
 283
 284                 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
 285                         unsigned long this_timer;
 286                         count++;
 287                         this_timer = e->updated + delay;
 288                         if (time_before_eq(this_timer, jiffies))
 289                                 tun_flow_delete(tun, e);
 290                         else if (time_before(this_timer, next_timer))
 291                                 next_timer = this_timer;
 292                 }
 293         }
 294
 295         if (count)
 296                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 297         spin_unlock_bh(&tun->lock);
 298 }
 299
 300 static void tun_flow_update(struct tun_struct *tun, struct sk_buff *skb,
 301                             u16 queue_index)
 302 {
 303         struct hlist_head *head;
 304         struct tun_flow_entry *e;
 305         unsigned long delay = tun->ageing_time;
 306         u32 rxhash = skb_get_rxhash(skb);
 307
 308         if (!rxhash)
 309                 return;
 310         else
 311                 head = &tun->flows[tun_hashfn(rxhash)];
 312
 313         rcu_read_lock();
 314
 315         if (tun->numqueues == 1)
 316                 goto unlock;
 317
 318         e = tun_flow_find(head, rxhash);
 319         if (likely(e)) {
 320                 /* TODO: keep queueing to old queue until it's empty? */
 321                 e->queue_index = queue_index;
 322                 e->updated = jiffies;
 323         } else {
 324                 spin_lock_bh(&tun->lock);
 325                 if (!tun_flow_find(head, rxhash))
 326                         tun_flow_create(tun, head, rxhash, queue_index);
 327
 328                 if (!timer_pending(&tun->flow_gc_timer))
 329                         mod_timer(&tun->flow_gc_timer,
 330                                   round_jiffies_up(jiffies + delay));
 331                 spin_unlock_bh(&tun->lock);
 332         }
 333
 334 unlock:
 335         rcu_read_unlock();
 336 }
 337
 338 /* We try to identify a flow through its rxhash first. The reason that
 339  * we do not check rxq no. is becuase some cards(e.g 82599), chooses
 340  * the rxq based on the txq where the last packet of the flow comes. As
 341  * the userspace application move between processors, we may get a
 342  * different rxq no. here. If we could not get rxhash, then we would
 343  * hope the rxq no. may help here.
 344  */
 345 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
 346 {
 347         struct tun_struct *tun = netdev_priv(dev);
 348         struct tun_flow_entry *e;
 349         u32 txq = 0;
 350         u32 numqueues = 0;
 351
 352         rcu_read_lock();
 353         numqueues = tun->numqueues;
 354
 355         txq = skb_get_rxhash(skb);
 356         if (txq) {
 357                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 358                 if (e)
 359                         txq = e->queue_index;
 360                 else
 361                         /* use multiply and shift instead of expensive divide */
 362                         txq = ((u64)txq * numqueues) >> 32;
 363         } else if (likely(skb_rx_queue_recorded(skb))) {
 364                 txq = skb_get_rx_queue(skb);
 365                 while (unlikely(txq >= numqueues))
 366                         txq -= numqueues;
 367         }
 368
 369         rcu_read_unlock();
 370         return txq;
 371 }
 372
 373 static inline bool tun_not_capable(struct tun_struct *tun)
 374 {
 375         const struct cred *cred = current_cred();
 376         struct net *net = dev_net(tun->dev);
 377
 378         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 379                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 380                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 381 }
 382
 383 static void tun_set_real_num_queues(struct tun_struct *tun)
 384 {
 385         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 386         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 387 }
 388
 389 static void __tun_detach(struct tun_file *tfile, bool clean)
 390 {
 391         struct tun_file *ntfile;
 392         struct tun_struct *tun;
 393         struct net_device *dev;
 394
 395         tun = rcu_dereference_protected(tfile->tun,
 396                                         lockdep_rtnl_is_held());
 397         if (tun) {
 398                 u16 index = tfile->queue_index;
 399                 BUG_ON(index >= tun->numqueues);
 400                 dev = tun->dev;
 401
 402                 rcu_assign_pointer(tun->tfiles[index],
 403                                    tun->tfiles[tun->numqueues - 1]);
 404                 rcu_assign_pointer(tfile->tun, NULL);
 405                 ntfile = rcu_dereference_protected(tun->tfiles[index],
 406                                                    lockdep_rtnl_is_held());
 407                 ntfile->queue_index = index;
 408
 409                 --tun->numqueues;
 410                 sock_put(&tfile->sk);
 411
 412                 synchronize_net();
 413                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 414                 /* Drop read queue */
 415                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 416                 tun_set_real_num_queues(tun);
 417
 418                 if (tun->numqueues == 0 && !(tun->flags & TUN_PERSIST))
 419                         if (dev->reg_state == NETREG_REGISTERED)
 420                                 unregister_netdevice(dev);
 421         }
 422
 423         if (clean) {
 424                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 425                                  &tfile->socket.flags));
 426                 sk_release_kernel(&tfile->sk);
 427         }
 428 }
 429
 430 static void tun_detach(struct tun_file *tfile, bool clean)
 431 {
 432         rtnl_lock();
 433         __tun_detach(tfile, clean);
 434         rtnl_unlock();
 435 }
 436
 437 static void tun_detach_all(struct net_device *dev)
 438 {
 439         struct tun_struct *tun = netdev_priv(dev);
 440         struct tun_file *tfile;
 441         int i, n = tun->numqueues;
 442
 443         for (i = 0; i < n; i++) {
 444                 tfile = rcu_dereference_protected(tun->tfiles[i],
 445                                                   lockdep_rtnl_is_held());
 446                 BUG_ON(!tfile);
 447                 wake_up_all(&tfile->wq.wait);
 448                 rcu_assign_pointer(tfile->tun, NULL);
 449                 --tun->numqueues;
 450         }
 451         BUG_ON(tun->numqueues != 0);
 452
 453         synchronize_net();
 454         for (i = 0; i < n; i++) {
 455                 tfile = rcu_dereference_protected(tun->tfiles[i],
 456                                                   lockdep_rtnl_is_held());
 457                 /* Drop read queue */
 458                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 459                 sock_put(&tfile->sk);
 460         }
 461 }
 462
 463 static int tun_attach(struct tun_struct *tun, struct file *file)
 464 {
 465         struct tun_file *tfile = file->private_data;
 466         int err;
 467
 468         err = -EINVAL;
 469         if (rcu_dereference_protected(tfile->tun, lockdep_rtnl_is_held()))
 470                 goto out;
 471
 472         err = -EBUSY;
 473         if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 474                 goto out;
 475
 476         err = -E2BIG;
 477         if (tun->numqueues == MAX_TAP_QUEUES)
 478                 goto out;
 479
 480         err = 0;
 481
 482         /* Re-attach the filter to presist device */
 483         if (tun->filter_attached == true) {
 484                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 485                 if (!err)
 486                         goto out;
 487         }
 488         tfile->queue_index = tun->numqueues;
 489         rcu_assign_pointer(tfile->tun, tun);
 490         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 491         sock_hold(&tfile->sk);
 492         tun->numqueues++;
 493
 494         tun_set_real_num_queues(tun);
 495
 496         if (tun->numqueues == 1)
 497                 netif_carrier_on(tun->dev);
 498
 499         /* device is allowed to go away first, so no need to hold extra
 500          * refcnt.
 501          */
 502
 503 out:
 504         return err;
 505 }
 506
 507 static struct tun_struct *__tun_get(struct tun_file *tfile)
 508 {
 509         struct tun_struct *tun;
 510
 511         rcu_read_lock();
 512         tun = rcu_dereference(tfile->tun);
 513         if (tun)
 514                 dev_hold(tun->dev);
 515         rcu_read_unlock();
 516
 517         return tun;
 518 }
 519
 520 static struct tun_struct *tun_get(struct file *file)
 521 {
 522         return __tun_get(file->private_data);
 523 }
 524
 525 static void tun_put(struct tun_struct *tun)
 526 {
 527         dev_put(tun->dev);
 528 }
 529
 530 /* TAP filtering */
 531 static void addr_hash_set(u32 *mask, const u8 *addr)
 532 {
 533         int n = ether_crc(ETH_ALEN, addr) >> 26;
 534         mask[n >> 5] |= (1 << (n & 31));
 535 }
 536
 537 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 538 {
 539         int n = ether_crc(ETH_ALEN, addr) >> 26;
 540         return mask[n >> 5] & (1 << (n & 31));
 541 }
 542
 543 static int update_filter(struct tap_filter *filter, void __user *arg)
 544 {
 545         struct { u8 u[ETH_ALEN]; } *addr;
 546         struct tun_filter uf;
 547         int err, alen, n, nexact;
 548
 549         if (copy_from_user(&uf, arg, sizeof(uf)))
 550                 return -EFAULT;
 551
 552         if (!uf.count) {
 553                 /* Disabled */
 554                 filter->count = 0;
 555                 return 0;
 556         }
 557
 558         alen = ETH_ALEN * uf.count;
 559         addr = kmalloc(alen, GFP_KERNEL);
 560         if (!addr)
 561                 return -ENOMEM;
 562
 563         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 564                 err = -EFAULT;
 565                 goto done;
 566         }
 567
 568         /* The filter is updated without holding any locks. Which is
 569          * perfectly safe. We disable it first and in the worst
 570          * case we'll accept a few undesired packets. */
 571         filter->count = 0;
 572         wmb();
 573
 574         /* Use first set of addresses as an exact filter */
 575         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 576                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 577
 578         nexact = n;
 579
 580         /* Remaining multicast addresses are hashed,
 581          * unicast will leave the filter disabled. */
 582         memset(filter->mask, 0, sizeof(filter->mask));
 583         for (; n < uf.count; n++) {
 584                 if (!is_multicast_ether_addr(addr[n].u)) {
 585                         err = 0; /* no filter */
 586                         goto done;
 587                 }
 588                 addr_hash_set(filter->mask, addr[n].u);
 589         }
 590
 591         /* For ALLMULTI just set the mask to all ones.
 592          * This overrides the mask populated above. */
 593         if ((uf.flags & TUN_FLT_ALLMULTI))
 594                 memset(filter->mask, ~0, sizeof(filter->mask));
 595
 596         /* Now enable the filter */
 597         wmb();
 598         filter->count = nexact;
 599
 600         /* Return the number of exact filters */
 601         err = nexact;
 602
 603 done:
 604         kfree(addr);
 605         return err;
 606 }
 607
 608 /* Returns: 0 - drop, !=0 - accept */
 609 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 610 {
 611         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 612          * at this point. */
 613         struct ethhdr *eh = (struct ethhdr *) skb->data;
 614         int i;
 615
 616         /* Exact match */
 617         for (i = 0; i < filter->count; i++)
 618                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 619                         return 1;
 620
 621         /* Inexact match (multicast only) */
 622         if (is_multicast_ether_addr(eh->h_dest))
 623                 return addr_hash_test(filter->mask, eh->h_dest);
 624
 625         return 0;
 626 }
 627
 628 /*
 629  * Checks whether the packet is accepted or not.
 630  * Returns: 0 - drop, !=0 - accept
 631  */
 632 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 633 {
 634         if (!filter->count)
 635                 return 1;
 636
 637         return run_filter(filter, skb);
 638 }
 639
 640 /* Network device part of the driver */
 641
 642 static const struct ethtool_ops tun_ethtool_ops;
 643
 644 /* Net device detach from fd. */
 645 static void tun_net_uninit(struct net_device *dev)
 646 {
 647         tun_detach_all(dev);
 648 }
 649
 650 /* Net device open. */
 651 static int tun_net_open(struct net_device *dev)
 652 {
 653         netif_tx_start_all_queues(dev);
 654         return 0;
 655 }
 656
 657 /* Net device close. */
 658 static int tun_net_close(struct net_device *dev)
 659 {
 660         netif_tx_stop_all_queues(dev);
 661         return 0;
 662 }
 663
 664 /* Net device start xmit */
 665 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 666 {
 667         struct tun_struct *tun = netdev_priv(dev);
 668         int txq = skb->queue_mapping;
 669         struct tun_file *tfile;
 670
 671         rcu_read_lock();
 672         tfile = rcu_dereference(tun->tfiles[txq]);
 673
 674         /* Drop packet if interface is not attached */
 675         if (txq >= tun->numqueues)
 676                 goto drop;
 677
 678         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 679
 680         BUG_ON(!tfile);
 681
 682         /* Drop if the filter does not like it.
 683          * This is a noop if the filter is disabled.
 684          * Filter can be enabled only for the TAP devices. */
 685         if (!check_filter(&tun->txflt, skb))
 686                 goto drop;
 687
 688         if (tfile->socket.sk->sk_filter &&
 689             sk_filter(tfile->socket.sk, skb))
 690                 goto drop;
 691
 692         /* Limit the number of packets queued by divining txq length with the
 693          * number of queues.
 694          */
 695         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
 696                           >= dev->tx_queue_len / tun->numqueues){
 697                 if (!(tun->flags & TUN_ONE_QUEUE)) {
 698                         /* Normal queueing mode. */
 699                         /* Packet scheduler handles dropping of further packets. */
 700                         netif_stop_subqueue(dev, txq);
 701
 702                         /* We won't see all dropped packets individually, so overrun
 703                          * error is more appropriate. */
 704                         dev->stats.tx_fifo_errors++;
 705                 } else {
 706                         /* Single queue mode.
 707                          * Driver handles dropping of all packets itself. */
 708                         goto drop;
 709                 }
 710         }
 711
 712         /* Orphan the skb - required as we might hang on to it
 713          * for indefinite time. */
 714         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 715                 goto drop;
 716         skb_orphan(skb);
 717
 718         /* Enqueue packet */
 719         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 720
 721         /* Notify and wake up reader process */
 722         if (tfile->flags & TUN_FASYNC)
 723                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 724         wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
 725                                    POLLRDNORM | POLLRDBAND);
 726
 727         rcu_read_unlock();
 728         return NETDEV_TX_OK;
 729
 730 drop:
 731         dev->stats.tx_dropped++;
 732         skb_tx_error(skb);
 733         kfree_skb(skb);
 734         rcu_read_unlock();
 735         return NETDEV_TX_OK;
 736 }
 737
 738 static void tun_net_mclist(struct net_device *dev)
 739 {
 740         /*
 741          * This callback is supposed to deal with mc filter in
 742          * _rx_ path and has nothing to do with the _tx_ path.
 743          * In rx path we always accept everything userspace gives us.
 744          */
 745 }
 746
 747 #define MIN_MTU 68
 748 #define MAX_MTU 65535
 749
 750 static int
 751 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 752 {
 753         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 754                 return -EINVAL;
 755         dev->mtu = new_mtu;
 756         return 0;
 757 }
 758
 759 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 760         netdev_features_t features)
 761 {
 762         struct tun_struct *tun = netdev_priv(dev);
 763
 764         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 765 }
 766 #ifdef CONFIG_NET_POLL_CONTROLLER
 767 static void tun_poll_controller(struct net_device *dev)
 768 {
 769         /*
 770          * Tun only receives frames when:
 771          * 1) the char device endpoint gets data from user space
 772          * 2) the tun socket gets a sendmsg call from user space
 773          * Since both of those are syncronous operations, we are guaranteed
 774          * never to have pending data when we poll for it
 775          * so theres nothing to do here but return.
 776          * We need this though so netpoll recognizes us as an interface that
 777          * supports polling, which enables bridge devices in virt setups to
 778          * still use netconsole
 779          */
 780         return;
 781 }
 782 #endif
 783 static const struct net_device_ops tun_netdev_ops = {
 784         .ndo_uninit             = tun_net_uninit,
 785         .ndo_open               = tun_net_open,
 786         .ndo_stop               = tun_net_close,
 787         .ndo_start_xmit         = tun_net_xmit,
 788         .ndo_change_mtu         = tun_net_change_mtu,
 789         .ndo_fix_features       = tun_net_fix_features,
 790         .ndo_select_queue       = tun_select_queue,
 791 #ifdef CONFIG_NET_POLL_CONTROLLER
 792         .ndo_poll_controller    = tun_poll_controller,
 793 #endif
 794 };
 795
 796 static const struct net_device_ops tap_netdev_ops = {
 797         .ndo_uninit             = tun_net_uninit,
 798         .ndo_open               = tun_net_open,
 799         .ndo_stop               = tun_net_close,
 800         .ndo_start_xmit         = tun_net_xmit,
 801         .ndo_change_mtu         = tun_net_change_mtu,
 802         .ndo_fix_features       = tun_net_fix_features,
 803         .ndo_set_rx_mode        = tun_net_mclist,
 804         .ndo_set_mac_address    = eth_mac_addr,
 805         .ndo_validate_addr      = eth_validate_addr,
 806         .ndo_select_queue       = tun_select_queue,
 807 #ifdef CONFIG_NET_POLL_CONTROLLER
 808         .ndo_poll_controller    = tun_poll_controller,
 809 #endif
 810 };
 811
 812 static int tun_flow_init(struct tun_struct *tun)
 813 {
 814         int i;
 815
 816         tun->flow_cache = kmem_cache_create("tun_flow_cache",
 817                                             sizeof(struct tun_flow_entry), 0, 0,
 818                                             NULL);
 819         if (!tun->flow_cache)
 820                 return -ENOMEM;
 821
 822         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 823                 INIT_HLIST_HEAD(&tun->flows[i]);
 824
 825         tun->ageing_time = TUN_FLOW_EXPIRE;
 826         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 827         mod_timer(&tun->flow_gc_timer,
 828                   round_jiffies_up(jiffies + tun->ageing_time));
 829
 830         return 0;
 831 }
 832
 833 static void tun_flow_uninit(struct tun_struct *tun)
 834 {
 835         del_timer_sync(&tun->flow_gc_timer);
 836         tun_flow_flush(tun);
 837
 838         /* Wait for completion of call_rcu()'s */
 839         rcu_barrier();
 840         kmem_cache_destroy(tun->flow_cache);
 841 }
 842
 843 /* Initialize net device. */
 844 static void tun_net_init(struct net_device *dev)
 845 {
 846         struct tun_struct *tun = netdev_priv(dev);
 847
 848         switch (tun->flags & TUN_TYPE_MASK) {
 849         case TUN_TUN_DEV:
 850                 dev->netdev_ops = &tun_netdev_ops;
 851
 852                 /* Point-to-Point TUN Device */
 853                 dev->hard_header_len = 0;
 854                 dev->addr_len = 0;
 855                 dev->mtu = 1500;
 856
 857                 /* Zero header length */
 858                 dev->type = ARPHRD_NONE;
 859                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 860                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 861                 break;
 862
 863         case TUN_TAP_DEV:
 864                 dev->netdev_ops = &tap_netdev_ops;
 865                 /* Ethernet TAP Device */
 866                 ether_setup(dev);
 867                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 868
 869                 eth_hw_addr_random(dev);
 870
 871                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 872                 break;
 873         }
 874 }
 875
 876 /* Character device part */
 877
 878 /* Poll */
 879 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 880 {
 881         struct tun_file *tfile = file->private_data;
 882         struct tun_struct *tun = __tun_get(tfile);
 883         struct sock *sk;
 884         unsigned int mask = 0;
 885
 886         if (!tun)
 887                 return POLLERR;
 888
 889         sk = tfile->socket.sk;
 890
 891         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 892
 893         poll_wait(file, &tfile->wq.wait, wait);
 894
 895         if (!skb_queue_empty(&sk->sk_receive_queue))
 896                 mask |= POLLIN | POLLRDNORM;
 897
 898         if (sock_writeable(sk) ||
 899             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 900              sock_writeable(sk)))
 901                 mask |= POLLOUT | POLLWRNORM;
 902
 903         if (tun->dev->reg_state != NETREG_REGISTERED)
 904                 mask = POLLERR;
 905
 906         tun_put(tun);
 907         return mask;
 908 }
 909
 910 /* prepad is the amount to reserve at front.  len is length after that.
 911  * linear is a hint as to how much to copy (usually headers). */
 912 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 913                                      size_t prepad, size_t len,
 914                                      size_t linear, int noblock)
 915 {
 916         struct sock *sk = tfile->socket.sk;
 917         struct sk_buff *skb;
 918         int err;
 919
 920         /* Under a page?  Don't bother with paged skb. */
 921         if (prepad + len < PAGE_SIZE || !linear)
 922                 linear = len;
 923
 924         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 925                                    &err);
 926         if (!skb)
 927                 return ERR_PTR(err);
 928
 929         skb_reserve(skb, prepad);
 930         skb_put(skb, linear);
 931         skb->data_len = len - linear;
 932         skb->len += len - linear;
 933
 934         return skb;
 935 }
 936
 937 /* set skb frags from iovec, this can move to core network code for reuse */
 938 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 939                                   int offset, size_t count)
 940 {
 941         int len = iov_length(from, count) - offset;
 942         int copy = skb_headlen(skb);
 943         int size, offset1 = 0;
 944         int i = 0;
 945
 946         /* Skip over from offset */
 947         while (count && (offset >= from->iov_len)) {
 948                 offset -= from->iov_len;
 949                 ++from;
 950                 --count;
 951         }
 952
 953         /* copy up to skb headlen */
 954         while (count && (copy > 0)) {
 955                 size = min_t(unsigned int, copy, from->iov_len - offset);
 956                 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
 957                                    size))
 958                         return -EFAULT;
 959                 if (copy > size) {
 960                         ++from;
 961                         --count;
 962                         offset = 0;
 963                 } else
 964                         offset += size;
 965                 copy -= size;
 966                 offset1 += size;
 967         }
 968
 969         if (len == offset1)
 970                 return 0;
 971
 972         while (count--) {
 973                 struct page *page[MAX_SKB_FRAGS];
 974                 int num_pages;
 975                 unsigned long base;
 976                 unsigned long truesize;
 977
 978                 len = from->iov_len - offset;
 979                 if (!len) {
 980                         offset = 0;
 981                         ++from;
 982                         continue;
 983                 }
 984                 base = (unsigned long)from->iov_base + offset;
 985                 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
 986                 if (i + size > MAX_SKB_FRAGS)
 987                         return -EMSGSIZE;
 988                 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
 989                 if (num_pages != size) {
 990                         for (i = 0; i < num_pages; i++)
 991                                 put_page(page[i]);
 992                         return -EFAULT;
 993                 }
 994                 truesize = size * PAGE_SIZE;
 995                 skb->data_len += len;
 996                 skb->len += len;
 997                 skb->truesize += truesize;
 998                 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
 999                 while (len) {
1000                         int off = base & ~PAGE_MASK;
1001                         int size = min_t(int, len, PAGE_SIZE - off);
1002                         __skb_fill_page_desc(skb, i, page[i], off, size);
1003                         skb_shinfo(skb)->nr_frags++;
1004                         /* increase sk_wmem_alloc */
1005                         base += size;
1006                         len -= size;
1007                         i++;
1008                 }
1009                 offset = 0;
1010                 ++from;
1011         }
1012         return 0;
1013 }
1014
1015 /* Get packet from user space buffer */
1016 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1017                             void *msg_control, const struct iovec *iv,
1018                             size_t total_len, size_t count, int noblock)
1019 {
1020         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1021         struct sk_buff *skb;
1022         size_t len = total_len, align = NET_SKB_PAD;
1023         struct virtio_net_hdr gso = { 0 };
1024         int offset = 0;
1025         int copylen;
1026         bool zerocopy = false;
1027         int err;
1028
1029         if (!(tun->flags & TUN_NO_PI)) {
1030                 if ((len -= sizeof(pi)) > total_len)
1031                         return -EINVAL;
1032
1033                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1034                         return -EFAULT;
1035                 offset += sizeof(pi);
1036         }
1037
1038         if (tun->flags & TUN_VNET_HDR) {
1039                 if ((len -= tun->vnet_hdr_sz) > total_len)
1040                         return -EINVAL;
1041
1042                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1043                         return -EFAULT;
1044
1045                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1046                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1047                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1048
1049                 if (gso.hdr_len > len)
1050                         return -EINVAL;
1051                 offset += tun->vnet_hdr_sz;
1052         }
1053
1054         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1055                 align += NET_IP_ALIGN;
1056                 if (unlikely(len < ETH_HLEN ||
1057                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1058                         return -EINVAL;
1059         }
1060
1061         if (msg_control)
1062                 zerocopy = true;
1063
1064         if (zerocopy) {
1065                 /* Userspace may produce vectors with count greater than
1066                  * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1067                  * to let the rest of data to be fit in the frags.
1068                  */
1069                 if (count > MAX_SKB_FRAGS) {
1070                         copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1071                         if (copylen < offset)
1072                                 copylen = 0;
1073                         else
1074                                 copylen -= offset;
1075                 } else
1076                                 copylen = 0;
1077                 /* There are 256 bytes to be copied in skb, so there is enough
1078                  * room for skb expand head in case it is used.
1079                  * The rest of the buffer is mapped from userspace.
1080                  */
1081                 if (copylen < gso.hdr_len)
1082                         copylen = gso.hdr_len;
1083                 if (!copylen)
1084                         copylen = GOODCOPY_LEN;
1085         } else
1086                 copylen = len;
1087
1088         skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
1089         if (IS_ERR(skb)) {
1090                 if (PTR_ERR(skb) != -EAGAIN)
1091                         tun->dev->stats.rx_dropped++;
1092                 return PTR_ERR(skb);
1093         }
1094
1095         if (zerocopy)
1096                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1097         else
1098                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1099
1100         if (err) {
1101                 tun->dev->stats.rx_dropped++;
1102                 kfree_skb(skb);
1103                 return -EFAULT;
1104         }
1105
1106         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1107                 if (!skb_partial_csum_set(skb, gso.csum_start,
1108                                           gso.csum_offset)) {
1109                         tun->dev->stats.rx_frame_errors++;
1110                         kfree_skb(skb);
1111                         return -EINVAL;
1112                 }
1113         }
1114
1115         switch (tun->flags & TUN_TYPE_MASK) {
1116         case TUN_TUN_DEV:
1117                 if (tun->flags & TUN_NO_PI) {
1118                         switch (skb->data[0] & 0xf0) {
1119                         case 0x40:
1120                                 pi.proto = htons(ETH_P_IP);
1121                                 break;
1122                         case 0x60:
1123                                 pi.proto = htons(ETH_P_IPV6);
1124                                 break;
1125                         default:
1126                                 tun->dev->stats.rx_dropped++;
1127                                 kfree_skb(skb);
1128                                 return -EINVAL;
1129                         }
1130                 }
1131
1132                 skb_reset_mac_header(skb);
1133                 skb->protocol = pi.proto;
1134                 skb->dev = tun->dev;
1135                 break;
1136         case TUN_TAP_DEV:
1137                 skb->protocol = eth_type_trans(skb, tun->dev);
1138                 break;
1139         }
1140
1141         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1142                 pr_debug("GSO!\n");
1143                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1144                 case VIRTIO_NET_HDR_GSO_TCPV4:
1145                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1146                         break;
1147                 case VIRTIO_NET_HDR_GSO_TCPV6:
1148                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1149                         break;
1150                 case VIRTIO_NET_HDR_GSO_UDP:
1151                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1152                         break;
1153                 default:
1154                         tun->dev->stats.rx_frame_errors++;
1155                         kfree_skb(skb);
1156                         return -EINVAL;
1157                 }
1158
1159                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1160                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1161
1162                 skb_shinfo(skb)->gso_size = gso.gso_size;
1163                 if (skb_shinfo(skb)->gso_size == 0) {
1164                         tun->dev->stats.rx_frame_errors++;
1165                         kfree_skb(skb);
1166                         return -EINVAL;
1167                 }
1168
1169                 /* Header must be checked, and gso_segs computed. */
1170                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1171                 skb_shinfo(skb)->gso_segs = 0;
1172         }
1173
1174         /* copy skb_ubuf_info for callback when skb has no error */
1175         if (zerocopy) {
1176                 skb_shinfo(skb)->destructor_arg = msg_control;
1177                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1178         }
1179
1180         netif_rx_ni(skb);
1181
1182         tun->dev->stats.rx_packets++;
1183         tun->dev->stats.rx_bytes += len;
1184
1185         tun_flow_update(tun, skb, tfile->queue_index);
1186         return total_len;
1187 }
1188
1189 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1190                               unsigned long count, loff_t pos)
1191 {
1192         struct file *file = iocb->ki_filp;
1193         struct tun_struct *tun = tun_get(file);
1194         struct tun_file *tfile = file->private_data;
1195         ssize_t result;
1196
1197         if (!tun)
1198                 return -EBADFD;
1199
1200         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1201
1202         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1203                               count, file->f_flags & O_NONBLOCK);
1204
1205         tun_put(tun);
1206         return result;
1207 }
1208
1209 /* Put packet to the user space buffer */
1210 static ssize_t tun_put_user(struct tun_struct *tun,
1211                             struct tun_file *tfile,
1212                             struct sk_buff *skb,
1213                             const struct iovec *iv, int len)
1214 {
1215         struct tun_pi pi = { 0, skb->protocol };
1216         ssize_t total = 0;
1217
1218         if (!(tun->flags & TUN_NO_PI)) {
1219                 if ((len -= sizeof(pi)) < 0)
1220                         return -EINVAL;
1221
1222                 if (len < skb->len) {
1223                         /* Packet will be striped */
1224                         pi.flags |= TUN_PKT_STRIP;
1225                 }
1226
1227                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1228                         return -EFAULT;
1229                 total += sizeof(pi);
1230         }
1231
1232         if (tun->flags & TUN_VNET_HDR) {
1233                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1234                 if ((len -= tun->vnet_hdr_sz) < 0)
1235                         return -EINVAL;
1236
1237                 if (skb_is_gso(skb)) {
1238                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1239
1240                         /* This is a hint as to how much should be linear. */
1241                         gso.hdr_len = skb_headlen(skb);
1242                         gso.gso_size = sinfo->gso_size;
1243                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1244                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1245                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1246                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1247                         else if (sinfo->gso_type & SKB_GSO_UDP)
1248                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1249                         else {
1250                                 pr_err("unexpected GSO type: "
1251                                        "0x%x, gso_size %d, hdr_len %d\n",
1252                                        sinfo->gso_type, gso.gso_size,
1253                                        gso.hdr_len);
1254                                 print_hex_dump(KERN_ERR, "tun: ",
1255                                                DUMP_PREFIX_NONE,
1256                                                16, 1, skb->head,
1257                                                min((int)gso.hdr_len, 64), true);
1258                                 WARN_ON_ONCE(1);
1259                                 return -EINVAL;
1260                         }
1261                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1262                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1263                 } else
1264                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1265
1266                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1267                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1268                         gso.csum_start = skb_checksum_start_offset(skb);
1269                         gso.csum_offset = skb->csum_offset;
1270                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1271                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1272                 } /* else everything is zero */
1273
1274                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1275                                                sizeof(gso))))
1276                         return -EFAULT;
1277                 total += tun->vnet_hdr_sz;
1278         }
1279
1280         len = min_t(int, skb->len, len);
1281
1282         skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1283         total += skb->len;
1284
1285         tun->dev->stats.tx_packets++;
1286         tun->dev->stats.tx_bytes += len;
1287
1288         return total;
1289 }
1290
1291 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1292                            struct kiocb *iocb, const struct iovec *iv,
1293                            ssize_t len, int noblock)
1294 {
1295         DECLARE_WAITQUEUE(wait, current);
1296         struct sk_buff *skb;
1297         ssize_t ret = 0;
1298
1299         tun_debug(KERN_INFO, tun, "tun_chr_read\n");
1300
1301         if (unlikely(!noblock))
1302                 add_wait_queue(&tfile->wq.wait, &wait);
1303         while (len) {
1304                 current->state = TASK_INTERRUPTIBLE;
1305
1306                 /* Read frames from the queue */
1307                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1308                         if (noblock) {
1309                                 ret = -EAGAIN;
1310                                 break;
1311                         }
1312                         if (signal_pending(current)) {
1313                                 ret = -ERESTARTSYS;
1314                                 break;
1315                         }
1316                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1317                                 ret = -EIO;
1318                                 break;
1319                         }
1320
1321                         /* Nothing to read, let's sleep */
1322                         schedule();
1323                         continue;
1324                 }
1325                 netif_wake_subqueue(tun->dev, tfile->queue_index);
1326
1327                 ret = tun_put_user(tun, tfile, skb, iv, len);
1328                 kfree_skb(skb);
1329                 break;
1330         }
1331
1332         current->state = TASK_RUNNING;
1333         if (unlikely(!noblock))
1334                 remove_wait_queue(&tfile->wq.wait, &wait);
1335
1336         return ret;
1337 }
1338
1339 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1340                             unsigned long count, loff_t pos)
1341 {
1342         struct file *file = iocb->ki_filp;
1343         struct tun_file *tfile = file->private_data;
1344         struct tun_struct *tun = __tun_get(tfile);
1345         ssize_t len, ret;
1346
1347         if (!tun)
1348                 return -EBADFD;
1349         len = iov_length(iv, count);
1350         if (len < 0) {
1351                 ret = -EINVAL;
1352                 goto out;
1353         }
1354
1355         ret = tun_do_read(tun, tfile, iocb, iv, len,
1356                           file->f_flags & O_NONBLOCK);
1357         ret = min_t(ssize_t, ret, len);
1358 out:
1359         tun_put(tun);
1360         return ret;
1361 }
1362
1363 static void tun_free_netdev(struct net_device *dev)
1364 {
1365         struct tun_struct *tun = netdev_priv(dev);
1366
1367         tun_flow_uninit(tun);
1368         free_netdev(dev);
1369 }
1370
1371 static void tun_setup(struct net_device *dev)
1372 {
1373         struct tun_struct *tun = netdev_priv(dev);
1374
1375         tun->owner = INVALID_UID;
1376         tun->group = INVALID_GID;
1377
1378         dev->ethtool_ops = &tun_ethtool_ops;
1379         dev->destructor = tun_free_netdev;
1380 }
1381
1382 /* Trivial set of netlink ops to allow deleting tun or tap
1383  * device with netlink.
1384  */
1385 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1386 {
1387         return -EINVAL;
1388 }
1389
1390 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1391         .kind           = DRV_NAME,
1392         .priv_size      = sizeof(struct tun_struct),
1393         .setup          = tun_setup,
1394         .validate       = tun_validate,
1395 };
1396
1397 static void tun_sock_write_space(struct sock *sk)
1398 {
1399         struct tun_file *tfile;
1400         wait_queue_head_t *wqueue;
1401
1402         if (!sock_writeable(sk))
1403                 return;
1404
1405         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1406                 return;
1407
1408         wqueue = sk_sleep(sk);
1409         if (wqueue && waitqueue_active(wqueue))
1410                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1411                                                 POLLWRNORM | POLLWRBAND);
1412
1413         tfile = container_of(sk, struct tun_file, sk);
1414         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1415 }
1416
1417 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1418                        struct msghdr *m, size_t total_len)
1419 {
1420         int ret;
1421         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1422         struct tun_struct *tun = __tun_get(tfile);
1423
1424         if (!tun)
1425                 return -EBADFD;
1426         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1427                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1428         tun_put(tun);
1429         return ret;
1430 }
1431
1432
1433 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1434                        struct msghdr *m, size_t total_len,
1435                        int flags)
1436 {
1437         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1438         struct tun_struct *tun = __tun_get(tfile);
1439         int ret;
1440
1441         if (!tun)
1442                 return -EBADFD;
1443
1444         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1445                 return -EINVAL;
1446         ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1447                           flags & MSG_DONTWAIT);
1448         if (ret > total_len) {
1449                 m->msg_flags |= MSG_TRUNC;
1450                 ret = flags & MSG_TRUNC ? ret : total_len;
1451         }
1452         tun_put(tun);
1453         return ret;
1454 }
1455
1456 static int tun_release(struct socket *sock)
1457 {
1458         if (sock->sk)
1459                 sock_put(sock->sk);
1460         return 0;
1461 }
1462
1463 /* Ops structure to mimic raw sockets with tun */
1464 static const struct proto_ops tun_socket_ops = {
1465         .sendmsg = tun_sendmsg,
1466         .recvmsg = tun_recvmsg,
1467         .release = tun_release,
1468 };
1469
1470 static struct proto tun_proto = {
1471         .name           = "tun",
1472         .owner          = THIS_MODULE,
1473         .obj_size       = sizeof(struct tun_file),
1474 };
1475
1476 static int tun_flags(struct tun_struct *tun)
1477 {
1478         int flags = 0;
1479
1480         if (tun->flags & TUN_TUN_DEV)
1481                 flags |= IFF_TUN;
1482         else
1483                 flags |= IFF_TAP;
1484
1485         if (tun->flags & TUN_NO_PI)
1486                 flags |= IFF_NO_PI;
1487
1488         if (tun->flags & TUN_ONE_QUEUE)
1489                 flags |= IFF_ONE_QUEUE;
1490
1491         if (tun->flags & TUN_VNET_HDR)
1492                 flags |= IFF_VNET_HDR;
1493
1494         if (tun->flags & TUN_TAP_MQ)
1495                 flags |= IFF_MULTI_QUEUE;
1496
1497         return flags;
1498 }
1499
1500 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1501                               char *buf)
1502 {
1503         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1504         return sprintf(buf, "0x%x\n", tun_flags(tun));
1505 }
1506
1507 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1508                               char *buf)
1509 {
1510         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1511         return uid_valid(tun->owner)?
1512                 sprintf(buf, "%u\n",
1513                         from_kuid_munged(current_user_ns(), tun->owner)):
1514                 sprintf(buf, "-1\n");
1515 }
1516
1517 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1518                               char *buf)
1519 {
1520         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1521         return gid_valid(tun->group) ?
1522                 sprintf(buf, "%u\n",
1523                         from_kgid_munged(current_user_ns(), tun->group)):
1524                 sprintf(buf, "-1\n");
1525 }
1526
1527 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1528 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1529 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1530
1531 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1532 {
1533         struct tun_struct *tun;
1534         struct tun_file *tfile = file->private_data;
1535         struct net_device *dev;
1536         int err;
1537
1538         dev = __dev_get_by_name(net, ifr->ifr_name);
1539         if (dev) {
1540                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1541                         return -EBUSY;
1542                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1543                         tun = netdev_priv(dev);
1544                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1545                         tun = netdev_priv(dev);
1546                 else
1547                         return -EINVAL;
1548
1549                 if (tun_not_capable(tun))
1550                         return -EPERM;
1551                 err = security_tun_dev_attach(tfile->socket.sk);
1552                 if (err < 0)
1553                         return err;
1554
1555                 err = tun_attach(tun, file);
1556                 if (err < 0)
1557                         return err;
1558         }
1559         else {
1560                 char *name;
1561                 unsigned long flags = 0;
1562
1563                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1564                         return -EPERM;
1565                 err = security_tun_dev_create();
1566                 if (err < 0)
1567                         return err;
1568
1569                 /* Set dev type */
1570                 if (ifr->ifr_flags & IFF_TUN) {
1571                         /* TUN device */
1572                         flags |= TUN_TUN_DEV;
1573                         name = "tun%d";
1574                 } else if (ifr->ifr_flags & IFF_TAP) {
1575                         /* TAP device */
1576                         flags |= TUN_TAP_DEV;
1577                         name = "tap%d";
1578                 } else
1579                         return -EINVAL;
1580
1581                 if (*ifr->ifr_name)
1582                         name = ifr->ifr_name;
1583
1584                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1585                                        tun_setup,
1586                                        MAX_TAP_QUEUES, MAX_TAP_QUEUES);
1587                 if (!dev)
1588                         return -ENOMEM;
1589
1590                 dev_net_set(dev, net);
1591                 dev->rtnl_link_ops = &tun_link_ops;
1592
1593                 tun = netdev_priv(dev);
1594                 tun->dev = dev;
1595                 tun->flags = flags;
1596                 tun->txflt.count = 0;
1597                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1598
1599                 tun->filter_attached = false;
1600                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1601
1602                 spin_lock_init(&tun->lock);
1603
1604                 security_tun_dev_post_create(&tfile->sk);
1605
1606                 tun_net_init(dev);
1607
1608                 if (tun_flow_init(tun))
1609                         goto err_free_dev;
1610
1611                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1612                         TUN_USER_FEATURES;
1613                 dev->features = dev->hw_features;
1614
1615                 err = register_netdevice(tun->dev);
1616                 if (err < 0)
1617                         goto err_free_dev;
1618
1619                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1620                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1621                     device_create_file(&tun->dev->dev, &dev_attr_group))
1622                         pr_err("Failed to create tun sysfs files\n");
1623
1624                 err = tun_attach(tun, file);
1625                 if (err < 0)
1626                         goto err_free_dev;
1627         }
1628
1629         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1630
1631         if (ifr->ifr_flags & IFF_NO_PI)
1632                 tun->flags |= TUN_NO_PI;
1633         else
1634                 tun->flags &= ~TUN_NO_PI;
1635
1636         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1637                 tun->flags |= TUN_ONE_QUEUE;
1638         else
1639                 tun->flags &= ~TUN_ONE_QUEUE;
1640
1641         if (ifr->ifr_flags & IFF_VNET_HDR)
1642                 tun->flags |= TUN_VNET_HDR;
1643         else
1644                 tun->flags &= ~TUN_VNET_HDR;
1645
1646         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1647                 tun->flags |= TUN_TAP_MQ;
1648         else
1649                 tun->flags &= ~TUN_TAP_MQ;
1650
1651         /* Make sure persistent devices do not get stuck in
1652          * xoff state.
1653          */
1654         if (netif_running(tun->dev))
1655                 netif_tx_wake_all_queues(tun->dev);
1656
1657         strcpy(ifr->ifr_name, tun->dev->name);
1658         return 0;
1659
1660  err_free_dev:
1661         free_netdev(dev);
1662         return err;
1663 }
1664
1665 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1666                        struct ifreq *ifr)
1667 {
1668         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1669
1670         strcpy(ifr->ifr_name, tun->dev->name);
1671
1672         ifr->ifr_flags = tun_flags(tun);
1673
1674 }
1675
1676 /* This is like a cut-down ethtool ops, except done via tun fd so no
1677  * privs required. */
1678 static int set_offload(struct tun_struct *tun, unsigned long arg)
1679 {
1680         netdev_features_t features = 0;
1681
1682         if (arg & TUN_F_CSUM) {
1683                 features |= NETIF_F_HW_CSUM;
1684                 arg &= ~TUN_F_CSUM;
1685
1686                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1687                         if (arg & TUN_F_TSO_ECN) {
1688                                 features |= NETIF_F_TSO_ECN;
1689                                 arg &= ~TUN_F_TSO_ECN;
1690                         }
1691                         if (arg & TUN_F_TSO4)
1692                                 features |= NETIF_F_TSO;
1693                         if (arg & TUN_F_TSO6)
1694                                 features |= NETIF_F_TSO6;
1695                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1696                 }
1697
1698                 if (arg & TUN_F_UFO) {
1699                         features |= NETIF_F_UFO;
1700                         arg &= ~TUN_F_UFO;
1701                 }
1702         }
1703
1704         /* This gives the user a way to test for new features in future by
1705          * trying to set them. */
1706         if (arg)
1707                 return -EINVAL;
1708
1709         tun->set_features = features;
1710         netdev_update_features(tun->dev);
1711
1712         return 0;
1713 }
1714
1715 static void tun_detach_filter(struct tun_struct *tun, int n)
1716 {
1717         int i;
1718         struct tun_file *tfile;
1719
1720         for (i = 0; i < n; i++) {
1721                 tfile = rcu_dereference_protected(tun->tfiles[i],
1722                                                   lockdep_rtnl_is_held());
1723                 sk_detach_filter(tfile->socket.sk);
1724         }
1725
1726         tun->filter_attached = false;
1727 }
1728
1729 static int tun_attach_filter(struct tun_struct *tun)
1730 {
1731         int i, ret = 0;
1732         struct tun_file *tfile;
1733
1734         for (i = 0; i < tun->numqueues; i++) {
1735                 tfile = rcu_dereference_protected(tun->tfiles[i],
1736                                                   lockdep_rtnl_is_held());
1737                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1738                 if (ret) {
1739                         tun_detach_filter(tun, i);
1740                         return ret;
1741                 }
1742         }
1743
1744         tun->filter_attached = true;
1745         return ret;
1746 }
1747
1748 static void tun_set_sndbuf(struct tun_struct *tun)
1749 {
1750         struct tun_file *tfile;
1751         int i;
1752
1753         for (i = 0; i < tun->numqueues; i++) {
1754                 tfile = rcu_dereference_protected(tun->tfiles[i],
1755                                                 lockdep_rtnl_is_held());
1756                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1757         }
1758 }
1759
1760 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1761 {
1762         struct tun_file *tfile = file->private_data;
1763         struct tun_struct *tun;
1764         struct net_device *dev;
1765         int ret = 0;
1766
1767         rtnl_lock();
1768
1769         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1770                 dev = __dev_get_by_name(tfile->net, ifr->ifr_name);
1771                 if (!dev) {
1772                         ret = -EINVAL;
1773                         goto unlock;
1774                 }
1775
1776                 tun = netdev_priv(dev);
1777                 if (dev->netdev_ops != &tap_netdev_ops &&
1778                         dev->netdev_ops != &tun_netdev_ops)
1779                         ret = -EINVAL;
1780                 else if (tun_not_capable(tun))
1781                         ret = -EPERM;
1782                 else
1783                         ret = tun_attach(tun, file);
1784         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE)
1785                 __tun_detach(tfile, false);
1786         else
1787                 ret = -EINVAL;
1788
1789 unlock:
1790         rtnl_unlock();
1791         return ret;
1792 }
1793
1794 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1795                             unsigned long arg, int ifreq_len)
1796 {
1797         struct tun_file *tfile = file->private_data;
1798         struct tun_struct *tun;
1799         void __user* argp = (void __user*)arg;
1800         struct ifreq ifr;
1801         kuid_t owner;
1802         kgid_t group;
1803         int sndbuf;
1804         int vnet_hdr_sz;
1805         int ret;
1806
1807         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1808                 if (copy_from_user(&ifr, argp, ifreq_len))
1809                         return -EFAULT;
1810         } else {
1811                 memset(&ifr, 0, sizeof(ifr));
1812         }
1813         if (cmd == TUNGETFEATURES) {
1814                 /* Currently this just means: "what IFF flags are valid?".
1815                  * This is needed because we never checked for invalid flags on
1816                  * TUNSETIFF. */
1817                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1818                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1819                                 (unsigned int __user*)argp);
1820         } else if (cmd == TUNSETQUEUE)
1821                 return tun_set_queue(file, &ifr);
1822
1823         ret = 0;
1824         rtnl_lock();
1825
1826         tun = __tun_get(tfile);
1827         if (cmd == TUNSETIFF && !tun) {
1828                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1829
1830                 ret = tun_set_iff(tfile->net, file, &ifr);
1831
1832                 if (ret)
1833                         goto unlock;
1834
1835                 if (copy_to_user(argp, &ifr, ifreq_len))
1836                         ret = -EFAULT;
1837                 goto unlock;
1838         }
1839
1840         ret = -EBADFD;
1841         if (!tun)
1842                 goto unlock;
1843
1844         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1845
1846         ret = 0;
1847         switch (cmd) {
1848         case TUNGETIFF:
1849                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1850
1851                 if (copy_to_user(argp, &ifr, ifreq_len))
1852                         ret = -EFAULT;
1853                 break;
1854
1855         case TUNSETNOCSUM:
1856                 /* Disable/Enable checksum */
1857
1858                 /* [unimplemented] */
1859                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1860                           arg ? "disabled" : "enabled");
1861                 break;
1862
1863         case TUNSETPERSIST:
1864                 /* Disable/Enable persist mode. Keep an extra reference to the
1865                  * module to prevent the module being unprobed.
1866                  */
1867                 if (arg) {
1868                         tun->flags |= TUN_PERSIST;
1869                         __module_get(THIS_MODULE);
1870                 } else {
1871                         tun->flags &= ~TUN_PERSIST;
1872                         module_put(THIS_MODULE);
1873                 }
1874
1875                 tun_debug(KERN_INFO, tun, "persist %s\n",
1876                           arg ? "enabled" : "disabled");
1877                 break;
1878
1879         case TUNSETOWNER:
1880                 /* Set owner of the device */
1881                 owner = make_kuid(current_user_ns(), arg);
1882                 if (!uid_valid(owner)) {
1883                         ret = -EINVAL;
1884                         break;
1885                 }
1886                 tun->owner = owner;
1887                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1888                           from_kuid(&init_user_ns, tun->owner));
1889                 break;
1890
1891         case TUNSETGROUP:
1892                 /* Set group of the device */
1893                 group = make_kgid(current_user_ns(), arg);
1894                 if (!gid_valid(group)) {
1895                         ret = -EINVAL;
1896                         break;
1897                 }
1898                 tun->group = group;
1899                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1900                           from_kgid(&init_user_ns, tun->group));
1901                 break;
1902
1903         case TUNSETLINK:
1904                 /* Only allow setting the type when the interface is down */
1905                 if (tun->dev->flags & IFF_UP) {
1906                         tun_debug(KERN_INFO, tun,
1907                                   "Linktype set failed because interface is up\n");
1908                         ret = -EBUSY;
1909                 } else {
1910                         tun->dev->type = (int) arg;
1911                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1912                                   tun->dev->type);
1913                         ret = 0;
1914                 }
1915                 break;
1916
1917 #ifdef TUN_DEBUG
1918         case TUNSETDEBUG:
1919                 tun->debug = arg;
1920                 break;
1921 #endif
1922         case TUNSETOFFLOAD:
1923                 ret = set_offload(tun, arg);
1924                 break;
1925
1926         case TUNSETTXFILTER:
1927                 /* Can be set only for TAPs */
1928                 ret = -EINVAL;
1929                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1930                         break;
1931                 ret = update_filter(&tun->txflt, (void __user *)arg);
1932                 break;
1933
1934         case SIOCGIFHWADDR:
1935                 /* Get hw address */
1936                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1937                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1938                 if (copy_to_user(argp, &ifr, ifreq_len))
1939                         ret = -EFAULT;
1940                 break;
1941
1942         case SIOCSIFHWADDR:
1943                 /* Set hw address */
1944                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1945                           ifr.ifr_hwaddr.sa_data);
1946
1947                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1948                 break;
1949
1950         case TUNGETSNDBUF:
1951                 sndbuf = tfile->socket.sk->sk_sndbuf;
1952                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1953                         ret = -EFAULT;
1954                 break;
1955
1956         case TUNSETSNDBUF:
1957                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1958                         ret = -EFAULT;
1959                         break;
1960                 }
1961
1962                 tun->sndbuf = sndbuf;
1963                 tun_set_sndbuf(tun);
1964                 break;
1965
1966         case TUNGETVNETHDRSZ:
1967                 vnet_hdr_sz = tun->vnet_hdr_sz;
1968                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1969                         ret = -EFAULT;
1970                 break;
1971
1972         case TUNSETVNETHDRSZ:
1973                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1974                         ret = -EFAULT;
1975                         break;
1976                 }
1977                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1978                         ret = -EINVAL;
1979                         break;
1980                 }
1981
1982                 tun->vnet_hdr_sz = vnet_hdr_sz;
1983                 break;
1984
1985         case TUNATTACHFILTER:
1986                 /* Can be set only for TAPs */
1987                 ret = -EINVAL;
1988                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1989                         break;
1990                 ret = -EFAULT;
1991                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
1992                         break;
1993
1994                 ret = tun_attach_filter(tun);
1995                 break;
1996
1997         case TUNDETACHFILTER:
1998                 /* Can be set only for TAPs */
1999                 ret = -EINVAL;
2000                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2001                         break;
2002                 ret = 0;
2003                 tun_detach_filter(tun, tun->numqueues);
2004                 break;
2005
2006         default:
2007                 ret = -EINVAL;
2008                 break;
2009         }
2010
2011 unlock:
2012         rtnl_unlock();
2013         if (tun)
2014                 tun_put(tun);
2015         return ret;
2016 }
2017
2018 static long tun_chr_ioctl(struct file *file,
2019                           unsigned int cmd, unsigned long arg)
2020 {
2021         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2022 }
2023
2024 #ifdef CONFIG_COMPAT
2025 static long tun_chr_compat_ioctl(struct file *file,
2026                          unsigned int cmd, unsigned long arg)
2027 {
2028         switch (cmd) {
2029         case TUNSETIFF:
2030         case TUNGETIFF:
2031         case TUNSETTXFILTER:
2032         case TUNGETSNDBUF:
2033         case TUNSETSNDBUF:
2034         case SIOCGIFHWADDR:
2035         case SIOCSIFHWADDR:
2036                 arg = (unsigned long)compat_ptr(arg);
2037                 break;
2038         default:
2039                 arg = (compat_ulong_t)arg;
2040                 break;
2041         }
2042
2043         /*
2044          * compat_ifreq is shorter than ifreq, so we must not access beyond
2045          * the end of that structure. All fields that are used in this
2046          * driver are compatible though, we don't need to convert the
2047          * contents.
2048          */
2049         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2050 }
2051 #endif /* CONFIG_COMPAT */
2052
2053 static int tun_chr_fasync(int fd, struct file *file, int on)
2054 {
2055         struct tun_file *tfile = file->private_data;
2056         int ret;
2057
2058         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2059                 goto out;
2060
2061         if (on) {
2062                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2063                 if (ret)
2064                         goto out;
2065                 tfile->flags |= TUN_FASYNC;
2066         } else
2067                 tfile->flags &= ~TUN_FASYNC;
2068         ret = 0;
2069 out:
2070         return ret;
2071 }
2072
2073 static int tun_chr_open(struct inode *inode, struct file * file)
2074 {
2075         struct tun_file *tfile;
2076
2077         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2078
2079         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2080                                             &tun_proto);
2081         if (!tfile)
2082                 return -ENOMEM;
2083         rcu_assign_pointer(tfile->tun, NULL);
2084         tfile->net = get_net(current->nsproxy->net_ns);
2085         tfile->flags = 0;
2086
2087         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2088         init_waitqueue_head(&tfile->wq.wait);
2089
2090         tfile->socket.file = file;
2091         tfile->socket.ops = &tun_socket_ops;
2092
2093         sock_init_data(&tfile->socket, &tfile->sk);
2094         sk_change_net(&tfile->sk, tfile->net);
2095
2096         tfile->sk.sk_write_space = tun_sock_write_space;
2097         tfile->sk.sk_sndbuf = INT_MAX;
2098
2099         file->private_data = tfile;
2100         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2101
2102         return 0;
2103 }
2104
2105 static int tun_chr_close(struct inode *inode, struct file *file)
2106 {
2107         struct tun_file *tfile = file->private_data;
2108         struct net *net = tfile->net;
2109
2110         tun_detach(tfile, true);
2111         put_net(net);
2112
2113         return 0;
2114 }
2115
2116 static const struct file_operations tun_fops = {
2117         .owner  = THIS_MODULE,
2118         .llseek = no_llseek,
2119         .read  = do_sync_read,
2120         .aio_read  = tun_chr_aio_read,
2121         .write = do_sync_write,
2122         .aio_write = tun_chr_aio_write,
2123         .poll   = tun_chr_poll,
2124         .unlocked_ioctl = tun_chr_ioctl,
2125 #ifdef CONFIG_COMPAT
2126         .compat_ioctl = tun_chr_compat_ioctl,
2127 #endif
2128         .open   = tun_chr_open,
2129         .release = tun_chr_close,
2130         .fasync = tun_chr_fasync
2131 };
2132
2133 static struct miscdevice tun_miscdev = {
2134         .minor = TUN_MINOR,
2135         .name = "tun",
2136         .nodename = "net/tun",
2137         .fops = &tun_fops,
2138 };
2139
2140 /* ethtool interface */
2141
2142 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2143 {
2144         cmd->supported          = 0;
2145         cmd->advertising        = 0;
2146         ethtool_cmd_speed_set(cmd, SPEED_10);
2147         cmd->duplex             = DUPLEX_FULL;
2148         cmd->port               = PORT_TP;
2149         cmd->phy_address        = 0;
2150         cmd->transceiver        = XCVR_INTERNAL;
2151         cmd->autoneg            = AUTONEG_DISABLE;
2152         cmd->maxtxpkt           = 0;
2153         cmd->maxrxpkt           = 0;
2154         return 0;
2155 }
2156
2157 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2158 {
2159         struct tun_struct *tun = netdev_priv(dev);
2160
2161         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2162         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2163
2164         switch (tun->flags & TUN_TYPE_MASK) {
2165         case TUN_TUN_DEV:
2166                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2167                 break;
2168         case TUN_TAP_DEV:
2169                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2170                 break;
2171         }
2172 }
2173
2174 static u32 tun_get_msglevel(struct net_device *dev)
2175 {
2176 #ifdef TUN_DEBUG
2177         struct tun_struct *tun = netdev_priv(dev);
2178         return tun->debug;
2179 #else
2180         return -EOPNOTSUPP;
2181 #endif
2182 }
2183
2184 static void tun_set_msglevel(struct net_device *dev, u32 value)
2185 {
2186 #ifdef TUN_DEBUG
2187         struct tun_struct *tun = netdev_priv(dev);
2188         tun->debug = value;
2189 #endif
2190 }
2191
2192 static const struct ethtool_ops tun_ethtool_ops = {
2193         .get_settings   = tun_get_settings,
2194         .get_drvinfo    = tun_get_drvinfo,
2195         .get_msglevel   = tun_get_msglevel,
2196         .set_msglevel   = tun_set_msglevel,
2197         .get_link       = ethtool_op_get_link,
2198 };
2199
2200
2201 static int __init tun_init(void)
2202 {
2203         int ret = 0;
2204
2205         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2206         pr_info("%s\n", DRV_COPYRIGHT);
2207
2208         ret = rtnl_link_register(&tun_link_ops);
2209         if (ret) {
2210                 pr_err("Can't register link_ops\n");
2211                 goto err_linkops;
2212         }
2213
2214         ret = misc_register(&tun_miscdev);
2215         if (ret) {
2216                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2217                 goto err_misc;
2218         }
2219         return  0;
2220 err_misc:
2221         rtnl_link_unregister(&tun_link_ops);
2222 err_linkops:
2223         return ret;
2224 }
2225
2226 static void tun_cleanup(void)
2227 {
2228         misc_deregister(&tun_miscdev);
2229         rtnl_link_unregister(&tun_link_ops);
2230 }
2231
2232 /* Get an underlying socket object from tun file.  Returns error unless file is
2233  * attached to a device.  The returned object works like a packet socket, it
2234  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2235  * holding a reference to the file for as long as the socket is in use. */
2236 struct socket *tun_get_socket(struct file *file)
2237 {
2238         struct tun_file *tfile;
2239         if (file->f_op != &tun_fops)
2240                 return ERR_PTR(-EINVAL);
2241         tfile = file->private_data;
2242         if (!tfile)
2243                 return ERR_PTR(-EBADFD);
2244         return &tfile->socket;
2245 }
2246 EXPORT_SYMBOL_GPL(tun_get_socket);
2247
2248 module_init(tun_init);
2249 module_exit(tun_cleanup);
2250 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2251 MODULE_AUTHOR(DRV_COPYRIGHT);
2252 MODULE_LICENSE("GPL");
2253 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2254 MODULE_ALIAS("devname:net/tun");