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