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

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/major.h>
  48 #include <linux/slab.h>
  49 #include <linux/poll.h>
  50 #include <linux/fcntl.h>
  51 #include <linux/init.h>
  52 #include <linux/skbuff.h>
  53 #include <linux/netdevice.h>
  54 #include <linux/etherdevice.h>
  55 #include <linux/miscdevice.h>
  56 #include <linux/ethtool.h>
  57 #include <linux/rtnetlink.h>
  58 #include <linux/compat.h>
  59 #include <linux/if.h>
  60 #include <linux/if_arp.h>
  61 #include <linux/if_ether.h>
  62 #include <linux/if_tun.h>
  63 #include <linux/if_vlan.h>
  64 #include <linux/crc32.h>
  65 #include <linux/nsproxy.h>
  66 #include <linux/virtio_net.h>
  67 #include <linux/rcupdate.h>
  68 #include <net/net_namespace.h>
  69 #include <net/netns/generic.h>
  70 #include <net/rtnetlink.h>
  71 #include <net/sock.h>
  72
  73 #include <asm/uaccess.h>
  74
  75 /* Uncomment to enable debugging */
  76 /* #define TUN_DEBUG 1 */
  77
  78 #ifdef TUN_DEBUG
  79 static int debug;
  80
  81 #define tun_debug(level, tun, fmt, args...)                     \
  82 do {                                                            \
  83         if (tun->debug)                                         \
  84                 netdev_printk(level, tun->dev, fmt, ##args);    \
  85 } while (0)
  86 #define DBG1(level, fmt, args...)                               \
  87 do {                                                            \
  88         if (debug == 2)                                         \
  89                 printk(level fmt, ##args);                      \
  90 } while (0)
  91 #else
  92 #define tun_debug(level, tun, fmt, args...)                     \
  93 do {                                                            \
  94         if (0)                                                  \
  95                 netdev_printk(level, tun->dev, fmt, ##args);    \
  96 } while (0)
  97 #define DBG1(level, fmt, args...)                               \
  98 do {                                                            \
  99         if (0)                                                  \
 100                 printk(level fmt, ##args);                      \
 101 } while (0)
 102 #endif
 103
 104 #define GOODCOPY_LEN 128
 105
 106 #define FLT_EXACT_COUNT 8
 107 struct tap_filter {
 108         unsigned int    count;    /* Number of addrs. Zero means disabled */
 109         u32             mask[2];  /* Mask of the hashed addrs */
 110         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 111 };
 112
 113 /* DEFAULT_MAX_NUM_RSS_QUEUES were choosed to let the rx/tx queues allocated for
 114  * the netdevice to be fit in one page. So we can make sure the success of
 115  * memory allocation. TODO: increase the limit. */
 116 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
 117 #define MAX_TAP_FLOWS  4096
 118
 119 #define TUN_FLOW_EXPIRE (3 * HZ)
 120
 121 /* A tun_file connects an open character device to a tuntap netdevice. It
 122  * also contains all socket related strctures (except sock_fprog and tap_filter)
 123  * to serve as one transmit queue for tuntap device. The sock_fprog and
 124  * tap_filter were kept in tun_struct since they were used for filtering for the
 125  * netdevice not for a specific queue (at least I didn't see the requirement for
 126  * this).
 127  *
 128  * RCU usage:
 129  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 130  * other can only be read while rcu_read_lock or rtnl_lock is held.
 131  */
 132 struct tun_file {
 133         struct sock sk;
 134         struct socket socket;
 135         struct socket_wq wq;
 136         struct tun_struct __rcu *tun;
 137         struct net *net;
 138         struct fasync_struct *fasync;
 139         /* only used for fasnyc */
 140         unsigned int flags;
 141         union {
 142                 u16 queue_index;
 143                 unsigned int ifindex;
 144         };
 145         struct list_head next;
 146         struct tun_struct *detached;
 147 };
 148
 149 struct tun_flow_entry {
 150         struct hlist_node hash_link;
 151         struct rcu_head rcu;
 152         struct tun_struct *tun;
 153
 154         u32 rxhash;
 155         int queue_index;
 156         unsigned long updated;
 157 };
 158
 159 #define TUN_NUM_FLOW_ENTRIES 1024
 160
 161 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 162  * device, socket filter, sndbuf and vnet header size were restore when the
 163  * file were attached to a persist device.
 164  */
 165 struct tun_struct {
 166         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 167         unsigned int            numqueues;
 168         unsigned int            flags;
 169         kuid_t                  owner;
 170         kgid_t                  group;
 171
 172         struct net_device       *dev;
 173         netdev_features_t       set_features;
 174 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 175                           NETIF_F_TSO6|NETIF_F_UFO)
 176
 177         int                     vnet_hdr_sz;
 178         int                     sndbuf;
 179         struct tap_filter       txflt;
 180         struct sock_fprog       fprog;
 181         /* protected by rtnl lock */
 182         bool                    filter_attached;
 183 #ifdef TUN_DEBUG
 184         int debug;
 185 #endif
 186         spinlock_t lock;
 187         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 188         struct timer_list flow_gc_timer;
 189         unsigned long ageing_time;
 190         unsigned int numdisabled;
 191         struct list_head disabled;
 192         void *security;
 193         u32 flow_count;
 194 };
 195
 196 static inline u32 tun_hashfn(u32 rxhash)
 197 {
 198         return rxhash & 0x3ff;
 199 }
 200
 201 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 202 {
 203         struct tun_flow_entry *e;
 204
 205         hlist_for_each_entry_rcu(e, head, hash_link) {
 206                 if (e->rxhash == rxhash)
 207                         return e;
 208         }
 209         return NULL;
 210 }
 211
 212 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 213                                               struct hlist_head *head,
 214                                               u32 rxhash, u16 queue_index)
 215 {
 216         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 217
 218         if (e) {
 219                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 220                           rxhash, queue_index);
 221                 e->updated = jiffies;
 222                 e->rxhash = rxhash;
 223                 e->queue_index = queue_index;
 224                 e->tun = tun;
 225                 hlist_add_head_rcu(&e->hash_link, head);
 226                 ++tun->flow_count;
 227         }
 228         return e;
 229 }
 230
 231 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 232 {
 233         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 234                   e->rxhash, e->queue_index);
 235         hlist_del_rcu(&e->hash_link);
 236         kfree_rcu(e, rcu);
 237         --tun->flow_count;
 238 }
 239
 240 static void tun_flow_flush(struct tun_struct *tun)
 241 {
 242         int i;
 243
 244         spin_lock_bh(&tun->lock);
 245         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 246                 struct tun_flow_entry *e;
 247                 struct hlist_node *n;
 248
 249                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 250                         tun_flow_delete(tun, e);
 251         }
 252         spin_unlock_bh(&tun->lock);
 253 }
 254
 255 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 256 {
 257         int i;
 258
 259         spin_lock_bh(&tun->lock);
 260         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 261                 struct tun_flow_entry *e;
 262                 struct hlist_node *n;
 263
 264                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 265                         if (e->queue_index == queue_index)
 266                                 tun_flow_delete(tun, e);
 267                 }
 268         }
 269         spin_unlock_bh(&tun->lock);
 270 }
 271
 272 static void tun_flow_cleanup(unsigned long data)
 273 {
 274         struct tun_struct *tun = (struct tun_struct *)data;
 275         unsigned long delay = tun->ageing_time;
 276         unsigned long next_timer = jiffies + delay;
 277         unsigned long count = 0;
 278         int i;
 279
 280         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 281
 282         spin_lock_bh(&tun->lock);
 283         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 284                 struct tun_flow_entry *e;
 285                 struct hlist_node *n;
 286
 287                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 288                         unsigned long this_timer;
 289                         count++;
 290                         this_timer = e->updated + delay;
 291                         if (time_before_eq(this_timer, jiffies))
 292                                 tun_flow_delete(tun, e);
 293                         else if (time_before(this_timer, next_timer))
 294                                 next_timer = this_timer;
 295                 }
 296         }
 297
 298         if (count)
 299                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 300         spin_unlock_bh(&tun->lock);
 301 }
 302
 303 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 304                             struct tun_file *tfile)
 305 {
 306         struct hlist_head *head;
 307         struct tun_flow_entry *e;
 308         unsigned long delay = tun->ageing_time;
 309         u16 queue_index = tfile->queue_index;
 310
 311         if (!rxhash)
 312                 return;
 313         else
 314                 head = &tun->flows[tun_hashfn(rxhash)];
 315
 316         rcu_read_lock();
 317
 318         /* We may get a very small possibility of OOO during switching, not
 319          * worth to optimize.*/
 320         if (tun->numqueues == 1 || tfile->detached)
 321                 goto unlock;
 322
 323         e = tun_flow_find(head, rxhash);
 324         if (likely(e)) {
 325                 /* TODO: keep queueing to old queue until it's empty? */
 326                 e->queue_index = queue_index;
 327                 e->updated = jiffies;
 328         } else {
 329                 spin_lock_bh(&tun->lock);
 330                 if (!tun_flow_find(head, rxhash) &&
 331                     tun->flow_count < MAX_TAP_FLOWS)
 332                         tun_flow_create(tun, head, rxhash, queue_index);
 333
 334                 if (!timer_pending(&tun->flow_gc_timer))
 335                         mod_timer(&tun->flow_gc_timer,
 336                                   round_jiffies_up(jiffies + delay));
 337                 spin_unlock_bh(&tun->lock);
 338         }
 339
 340 unlock:
 341         rcu_read_unlock();
 342 }
 343
 344 /* We try to identify a flow through its rxhash first. The reason that
 345  * we do not check rxq no. is becuase some cards(e.g 82599), chooses
 346  * the rxq based on the txq where the last packet of the flow comes. As
 347  * the userspace application move between processors, we may get a
 348  * different rxq no. here. If we could not get rxhash, then we would
 349  * hope the rxq no. may help here.
 350  */
 351 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
 352 {
 353         struct tun_struct *tun = netdev_priv(dev);
 354         struct tun_flow_entry *e;
 355         u32 txq = 0;
 356         u32 numqueues = 0;
 357
 358         rcu_read_lock();
 359         numqueues = ACCESS_ONCE(tun->numqueues);
 360
 361         txq = skb_get_rxhash(skb);
 362         if (txq) {
 363                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 364                 if (e)
 365                         txq = e->queue_index;
 366                 else
 367                         /* use multiply and shift instead of expensive divide */
 368                         txq = ((u64)txq * numqueues) >> 32;
 369         } else if (likely(skb_rx_queue_recorded(skb))) {
 370                 txq = skb_get_rx_queue(skb);
 371                 while (unlikely(txq >= numqueues))
 372                         txq -= numqueues;
 373         }
 374
 375         rcu_read_unlock();
 376         return txq;
 377 }
 378
 379 static inline bool tun_not_capable(struct tun_struct *tun)
 380 {
 381         const struct cred *cred = current_cred();
 382         struct net *net = dev_net(tun->dev);
 383
 384         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 385                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 386                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 387 }
 388
 389 static void tun_set_real_num_queues(struct tun_struct *tun)
 390 {
 391         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 392         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 393 }
 394
 395 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 396 {
 397         tfile->detached = tun;
 398         list_add_tail(&tfile->next, &tun->disabled);
 399         ++tun->numdisabled;
 400 }
 401
 402 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 403 {
 404         struct tun_struct *tun = tfile->detached;
 405
 406         tfile->detached = NULL;
 407         list_del_init(&tfile->next);
 408         --tun->numdisabled;
 409         return tun;
 410 }
 411
 412 static void __tun_detach(struct tun_file *tfile, bool clean)
 413 {
 414         struct tun_file *ntfile;
 415         struct tun_struct *tun;
 416
 417         tun = rtnl_dereference(tfile->tun);
 418
 419         if (tun && !tfile->detached) {
 420                 u16 index = tfile->queue_index;
 421                 BUG_ON(index >= tun->numqueues);
 422
 423                 rcu_assign_pointer(tun->tfiles[index],
 424                                    tun->tfiles[tun->numqueues - 1]);
 425                 ntfile = rtnl_dereference(tun->tfiles[index]);
 426                 ntfile->queue_index = index;
 427
 428                 --tun->numqueues;
 429                 if (clean) {
 430                         rcu_assign_pointer(tfile->tun, NULL);
 431                         sock_put(&tfile->sk);
 432                 } else
 433                         tun_disable_queue(tun, tfile);
 434
 435                 synchronize_net();
 436                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 437                 /* Drop read queue */
 438                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 439                 tun_set_real_num_queues(tun);
 440         } else if (tfile->detached && clean) {
 441                 tun = tun_enable_queue(tfile);
 442                 sock_put(&tfile->sk);
 443         }
 444
 445         if (clean) {
 446                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 447                         netif_carrier_off(tun->dev);
 448
 449                         if (!(tun->flags & TUN_PERSIST) &&
 450                             tun->dev->reg_state == NETREG_REGISTERED)
 451                                 unregister_netdevice(tun->dev);
 452                 }
 453
 454                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 455                                  &tfile->socket.flags));
 456                 sk_release_kernel(&tfile->sk);
 457         }
 458 }
 459
 460 static void tun_detach(struct tun_file *tfile, bool clean)
 461 {
 462         rtnl_lock();
 463         __tun_detach(tfile, clean);
 464         rtnl_unlock();
 465 }
 466
 467 static void tun_detach_all(struct net_device *dev)
 468 {
 469         struct tun_struct *tun = netdev_priv(dev);
 470         struct tun_file *tfile, *tmp;
 471         int i, n = tun->numqueues;
 472
 473         for (i = 0; i < n; i++) {
 474                 tfile = rtnl_dereference(tun->tfiles[i]);
 475                 BUG_ON(!tfile);
 476                 wake_up_all(&tfile->wq.wait);
 477                 rcu_assign_pointer(tfile->tun, NULL);
 478                 --tun->numqueues;
 479         }
 480         list_for_each_entry(tfile, &tun->disabled, next) {
 481                 wake_up_all(&tfile->wq.wait);
 482                 rcu_assign_pointer(tfile->tun, NULL);
 483         }
 484         BUG_ON(tun->numqueues != 0);
 485
 486         synchronize_net();
 487         for (i = 0; i < n; i++) {
 488                 tfile = rtnl_dereference(tun->tfiles[i]);
 489                 /* Drop read queue */
 490                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 491                 sock_put(&tfile->sk);
 492         }
 493         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 494                 tun_enable_queue(tfile);
 495                 skb_queue_purge(&tfile->sk.sk_receive_queue);
 496                 sock_put(&tfile->sk);
 497         }
 498         BUG_ON(tun->numdisabled != 0);
 499
 500         if (tun->flags & TUN_PERSIST)
 501                 module_put(THIS_MODULE);
 502 }
 503
 504 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 505 {
 506         struct tun_file *tfile = file->private_data;
 507         int err;
 508
 509         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 510         if (err < 0)
 511                 goto out;
 512
 513         err = -EINVAL;
 514         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 515                 goto out;
 516
 517         err = -EBUSY;
 518         if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 519                 goto out;
 520
 521         err = -E2BIG;
 522         if (!tfile->detached &&
 523             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 524                 goto out;
 525
 526         err = 0;
 527
 528         /* Re-attach the filter to presist device */
 529         if (!skip_filter && (tun->filter_attached == true)) {
 530                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 531                 if (!err)
 532                         goto out;
 533         }
 534         tfile->queue_index = tun->numqueues;
 535         rcu_assign_pointer(tfile->tun, tun);
 536         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 537         tun->numqueues++;
 538
 539         if (tfile->detached)
 540                 tun_enable_queue(tfile);
 541         else
 542                 sock_hold(&tfile->sk);
 543
 544         tun_set_real_num_queues(tun);
 545
 546         /* device is allowed to go away first, so no need to hold extra
 547          * refcnt.
 548          */
 549
 550 out:
 551         return err;
 552 }
 553
 554 static struct tun_struct *__tun_get(struct tun_file *tfile)
 555 {
 556         struct tun_struct *tun;
 557
 558         rcu_read_lock();
 559         tun = rcu_dereference(tfile->tun);
 560         if (tun)
 561                 dev_hold(tun->dev);
 562         rcu_read_unlock();
 563
 564         return tun;
 565 }
 566
 567 static struct tun_struct *tun_get(struct file *file)
 568 {
 569         return __tun_get(file->private_data);
 570 }
 571
 572 static void tun_put(struct tun_struct *tun)
 573 {
 574         dev_put(tun->dev);
 575 }
 576
 577 /* TAP filtering */
 578 static void addr_hash_set(u32 *mask, const u8 *addr)
 579 {
 580         int n = ether_crc(ETH_ALEN, addr) >> 26;
 581         mask[n >> 5] |= (1 << (n & 31));
 582 }
 583
 584 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 585 {
 586         int n = ether_crc(ETH_ALEN, addr) >> 26;
 587         return mask[n >> 5] & (1 << (n & 31));
 588 }
 589
 590 static int update_filter(struct tap_filter *filter, void __user *arg)
 591 {
 592         struct { u8 u[ETH_ALEN]; } *addr;
 593         struct tun_filter uf;
 594         int err, alen, n, nexact;
 595
 596         if (copy_from_user(&uf, arg, sizeof(uf)))
 597                 return -EFAULT;
 598
 599         if (!uf.count) {
 600                 /* Disabled */
 601                 filter->count = 0;
 602                 return 0;
 603         }
 604
 605         alen = ETH_ALEN * uf.count;
 606         addr = kmalloc(alen, GFP_KERNEL);
 607         if (!addr)
 608                 return -ENOMEM;
 609
 610         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 611                 err = -EFAULT;
 612                 goto done;
 613         }
 614
 615         /* The filter is updated without holding any locks. Which is
 616          * perfectly safe. We disable it first and in the worst
 617          * case we'll accept a few undesired packets. */
 618         filter->count = 0;
 619         wmb();
 620
 621         /* Use first set of addresses as an exact filter */
 622         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 623                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 624
 625         nexact = n;
 626
 627         /* Remaining multicast addresses are hashed,
 628          * unicast will leave the filter disabled. */
 629         memset(filter->mask, 0, sizeof(filter->mask));
 630         for (; n < uf.count; n++) {
 631                 if (!is_multicast_ether_addr(addr[n].u)) {
 632                         err = 0; /* no filter */
 633                         goto done;
 634                 }
 635                 addr_hash_set(filter->mask, addr[n].u);
 636         }
 637
 638         /* For ALLMULTI just set the mask to all ones.
 639          * This overrides the mask populated above. */
 640         if ((uf.flags & TUN_FLT_ALLMULTI))
 641                 memset(filter->mask, ~0, sizeof(filter->mask));
 642
 643         /* Now enable the filter */
 644         wmb();
 645         filter->count = nexact;
 646
 647         /* Return the number of exact filters */
 648         err = nexact;
 649
 650 done:
 651         kfree(addr);
 652         return err;
 653 }
 654
 655 /* Returns: 0 - drop, !=0 - accept */
 656 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 657 {
 658         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 659          * at this point. */
 660         struct ethhdr *eh = (struct ethhdr *) skb->data;
 661         int i;
 662
 663         /* Exact match */
 664         for (i = 0; i < filter->count; i++)
 665                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 666                         return 1;
 667
 668         /* Inexact match (multicast only) */
 669         if (is_multicast_ether_addr(eh->h_dest))
 670                 return addr_hash_test(filter->mask, eh->h_dest);
 671
 672         return 0;
 673 }
 674
 675 /*
 676  * Checks whether the packet is accepted or not.
 677  * Returns: 0 - drop, !=0 - accept
 678  */
 679 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 680 {
 681         if (!filter->count)
 682                 return 1;
 683
 684         return run_filter(filter, skb);
 685 }
 686
 687 /* Network device part of the driver */
 688
 689 static const struct ethtool_ops tun_ethtool_ops;
 690
 691 /* Net device detach from fd. */
 692 static void tun_net_uninit(struct net_device *dev)
 693 {
 694         tun_detach_all(dev);
 695 }
 696
 697 /* Net device open. */
 698 static int tun_net_open(struct net_device *dev)
 699 {
 700         netif_tx_start_all_queues(dev);
 701         return 0;
 702 }
 703
 704 /* Net device close. */
 705 static int tun_net_close(struct net_device *dev)
 706 {
 707         netif_tx_stop_all_queues(dev);
 708         return 0;
 709 }
 710
 711 /* Net device start xmit */
 712 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 713 {
 714         struct tun_struct *tun = netdev_priv(dev);
 715         int txq = skb->queue_mapping;
 716         struct tun_file *tfile;
 717
 718         rcu_read_lock();
 719         tfile = rcu_dereference(tun->tfiles[txq]);
 720
 721         /* Drop packet if interface is not attached */
 722         if (txq >= tun->numqueues)
 723                 goto drop;
 724
 725         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 726
 727         BUG_ON(!tfile);
 728
 729         /* Drop if the filter does not like it.
 730          * This is a noop if the filter is disabled.
 731          * Filter can be enabled only for the TAP devices. */
 732         if (!check_filter(&tun->txflt, skb))
 733                 goto drop;
 734
 735         if (tfile->socket.sk->sk_filter &&
 736             sk_filter(tfile->socket.sk, skb))
 737                 goto drop;
 738
 739         /* Limit the number of packets queued by dividing txq length with the
 740          * number of queues.
 741          */
 742         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
 743                           >= dev->tx_queue_len / tun->numqueues)
 744                 goto drop;
 745
 746         if (skb->sk) {
 747                 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
 748                 sw_tx_timestamp(skb);
 749         }
 750
 751         /* Orphan the skb - required as we might hang on to it
 752          * for indefinite time. */
 753         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 754                 goto drop;
 755         skb_orphan(skb);
 756
 757         nf_reset(skb);
 758
 759         /* Enqueue packet */
 760         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 761
 762         /* Notify and wake up reader process */
 763         if (tfile->flags & TUN_FASYNC)
 764                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 765         wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
 766                                    POLLRDNORM | POLLRDBAND);
 767
 768         rcu_read_unlock();
 769         return NETDEV_TX_OK;
 770
 771 drop:
 772         dev->stats.tx_dropped++;
 773         skb_tx_error(skb);
 774         kfree_skb(skb);
 775         rcu_read_unlock();
 776         return NETDEV_TX_OK;
 777 }
 778
 779 static void tun_net_mclist(struct net_device *dev)
 780 {
 781         /*
 782          * This callback is supposed to deal with mc filter in
 783          * _rx_ path and has nothing to do with the _tx_ path.
 784          * In rx path we always accept everything userspace gives us.
 785          */
 786 }
 787
 788 #define MIN_MTU 68
 789 #define MAX_MTU 65535
 790
 791 static int
 792 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 793 {
 794         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 795                 return -EINVAL;
 796         dev->mtu = new_mtu;
 797         return 0;
 798 }
 799
 800 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 801         netdev_features_t features)
 802 {
 803         struct tun_struct *tun = netdev_priv(dev);
 804
 805         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 806 }
 807 #ifdef CONFIG_NET_POLL_CONTROLLER
 808 static void tun_poll_controller(struct net_device *dev)
 809 {
 810         /*
 811          * Tun only receives frames when:
 812          * 1) the char device endpoint gets data from user space
 813          * 2) the tun socket gets a sendmsg call from user space
 814          * Since both of those are syncronous operations, we are guaranteed
 815          * never to have pending data when we poll for it
 816          * so theres nothing to do here but return.
 817          * We need this though so netpoll recognizes us as an interface that
 818          * supports polling, which enables bridge devices in virt setups to
 819          * still use netconsole
 820          */
 821         return;
 822 }
 823 #endif
 824 static const struct net_device_ops tun_netdev_ops = {
 825         .ndo_uninit             = tun_net_uninit,
 826         .ndo_open               = tun_net_open,
 827         .ndo_stop               = tun_net_close,
 828         .ndo_start_xmit         = tun_net_xmit,
 829         .ndo_change_mtu         = tun_net_change_mtu,
 830         .ndo_fix_features       = tun_net_fix_features,
 831         .ndo_select_queue       = tun_select_queue,
 832 #ifdef CONFIG_NET_POLL_CONTROLLER
 833         .ndo_poll_controller    = tun_poll_controller,
 834 #endif
 835 };
 836
 837 static const struct net_device_ops tap_netdev_ops = {
 838         .ndo_uninit             = tun_net_uninit,
 839         .ndo_open               = tun_net_open,
 840         .ndo_stop               = tun_net_close,
 841         .ndo_start_xmit         = tun_net_xmit,
 842         .ndo_change_mtu         = tun_net_change_mtu,
 843         .ndo_fix_features       = tun_net_fix_features,
 844         .ndo_set_rx_mode        = tun_net_mclist,
 845         .ndo_set_mac_address    = eth_mac_addr,
 846         .ndo_validate_addr      = eth_validate_addr,
 847         .ndo_select_queue       = tun_select_queue,
 848 #ifdef CONFIG_NET_POLL_CONTROLLER
 849         .ndo_poll_controller    = tun_poll_controller,
 850 #endif
 851 };
 852
 853 static void tun_flow_init(struct tun_struct *tun)
 854 {
 855         int i;
 856
 857         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 858                 INIT_HLIST_HEAD(&tun->flows[i]);
 859
 860         tun->ageing_time = TUN_FLOW_EXPIRE;
 861         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 862         mod_timer(&tun->flow_gc_timer,
 863                   round_jiffies_up(jiffies + tun->ageing_time));
 864 }
 865
 866 static void tun_flow_uninit(struct tun_struct *tun)
 867 {
 868         del_timer_sync(&tun->flow_gc_timer);
 869         tun_flow_flush(tun);
 870 }
 871
 872 /* Initialize net device. */
 873 static void tun_net_init(struct net_device *dev)
 874 {
 875         struct tun_struct *tun = netdev_priv(dev);
 876
 877         switch (tun->flags & TUN_TYPE_MASK) {
 878         case TUN_TUN_DEV:
 879                 dev->netdev_ops = &tun_netdev_ops;
 880
 881                 /* Point-to-Point TUN Device */
 882                 dev->hard_header_len = 0;
 883                 dev->addr_len = 0;
 884                 dev->mtu = 1500;
 885
 886                 /* Zero header length */
 887                 dev->type = ARPHRD_NONE;
 888                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 889                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 890                 break;
 891
 892         case TUN_TAP_DEV:
 893                 dev->netdev_ops = &tap_netdev_ops;
 894                 /* Ethernet TAP Device */
 895                 ether_setup(dev);
 896                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 897                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 898
 899                 eth_hw_addr_random(dev);
 900
 901                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 902                 break;
 903         }
 904 }
 905
 906 /* Character device part */
 907
 908 /* Poll */
 909 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 910 {
 911         struct tun_file *tfile = file->private_data;
 912         struct tun_struct *tun = __tun_get(tfile);
 913         struct sock *sk;
 914         unsigned int mask = 0;
 915
 916         if (!tun)
 917                 return POLLERR;
 918
 919         sk = tfile->socket.sk;
 920
 921         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 922
 923         poll_wait(file, &tfile->wq.wait, wait);
 924
 925         if (!skb_queue_empty(&sk->sk_receive_queue))
 926                 mask |= POLLIN | POLLRDNORM;
 927
 928         if (sock_writeable(sk) ||
 929             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 930              sock_writeable(sk)))
 931                 mask |= POLLOUT | POLLWRNORM;
 932
 933         if (tun->dev->reg_state != NETREG_REGISTERED)
 934                 mask = POLLERR;
 935
 936         tun_put(tun);
 937         return mask;
 938 }
 939
 940 /* prepad is the amount to reserve at front.  len is length after that.
 941  * linear is a hint as to how much to copy (usually headers). */
 942 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 943                                      size_t prepad, size_t len,
 944                                      size_t linear, int noblock)
 945 {
 946         struct sock *sk = tfile->socket.sk;
 947         struct sk_buff *skb;
 948         int err;
 949
 950         /* Under a page?  Don't bother with paged skb. */
 951         if (prepad + len < PAGE_SIZE || !linear)
 952                 linear = len;
 953
 954         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 955                                    &err, 0);
 956         if (!skb)
 957                 return ERR_PTR(err);
 958
 959         skb_reserve(skb, prepad);
 960         skb_put(skb, linear);
 961         skb->data_len = len - linear;
 962         skb->len += len - linear;
 963
 964         return skb;
 965 }
 966
 967 /* Get packet from user space buffer */
 968 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
 969                             void *msg_control, const struct iovec *iv,
 970                             size_t total_len, size_t count, int noblock)
 971 {
 972         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
 973         struct sk_buff *skb;
 974         size_t len = total_len, align = NET_SKB_PAD, linear;
 975         struct virtio_net_hdr gso = { 0 };
 976         int offset = 0;
 977         int copylen;
 978         bool zerocopy = false;
 979         int err;
 980         u32 rxhash;
 981
 982         if (!(tun->flags & TUN_NO_PI)) {
 983                 if (len < sizeof(pi))
 984                         return -EINVAL;
 985                 len -= sizeof(pi);
 986
 987                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
 988                         return -EFAULT;
 989                 offset += sizeof(pi);
 990         }
 991
 992         if (tun->flags & TUN_VNET_HDR) {
 993                 if (len < tun->vnet_hdr_sz)
 994                         return -EINVAL;
 995                 len -= tun->vnet_hdr_sz;
 996
 997                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
 998                         return -EFAULT;
 999
1000                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1001                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1002                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1003
1004                 if (gso.hdr_len > len)
1005                         return -EINVAL;
1006                 offset += tun->vnet_hdr_sz;
1007         }
1008
1009         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1010                 align += NET_IP_ALIGN;
1011                 if (unlikely(len < ETH_HLEN ||
1012                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1013                         return -EINVAL;
1014         }
1015
1016         if (msg_control) {
1017                 /* There are 256 bytes to be copied in skb, so there is
1018                  * enough room for skb expand head in case it is used.
1019                  * The rest of the buffer is mapped from userspace.
1020                  */
1021                 copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1022                 linear = copylen;
1023                 if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1024                         zerocopy = true;
1025         }
1026
1027         if (!zerocopy) {
1028                 copylen = len;
1029                 linear = gso.hdr_len;
1030         }
1031
1032         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1033         if (IS_ERR(skb)) {
1034                 if (PTR_ERR(skb) != -EAGAIN)
1035                         tun->dev->stats.rx_dropped++;
1036                 return PTR_ERR(skb);
1037         }
1038
1039         if (zerocopy)
1040                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1041         else {
1042                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1043                 if (!err && msg_control) {
1044                         struct ubuf_info *uarg = msg_control;
1045                         uarg->callback(uarg, false);
1046                 }
1047         }
1048
1049         if (err) {
1050                 tun->dev->stats.rx_dropped++;
1051                 kfree_skb(skb);
1052                 return -EFAULT;
1053         }
1054
1055         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1056                 if (!skb_partial_csum_set(skb, gso.csum_start,
1057                                           gso.csum_offset)) {
1058                         tun->dev->stats.rx_frame_errors++;
1059                         kfree_skb(skb);
1060                         return -EINVAL;
1061                 }
1062         }
1063
1064         switch (tun->flags & TUN_TYPE_MASK) {
1065         case TUN_TUN_DEV:
1066                 if (tun->flags & TUN_NO_PI) {
1067                         switch (skb->data[0] & 0xf0) {
1068                         case 0x40:
1069                                 pi.proto = htons(ETH_P_IP);
1070                                 break;
1071                         case 0x60:
1072                                 pi.proto = htons(ETH_P_IPV6);
1073                                 break;
1074                         default:
1075                                 tun->dev->stats.rx_dropped++;
1076                                 kfree_skb(skb);
1077                                 return -EINVAL;
1078                         }
1079                 }
1080
1081                 skb_reset_mac_header(skb);
1082                 skb->protocol = pi.proto;
1083                 skb->dev = tun->dev;
1084                 break;
1085         case TUN_TAP_DEV:
1086                 skb->protocol = eth_type_trans(skb, tun->dev);
1087                 break;
1088         }
1089
1090         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1091                 pr_debug("GSO!\n");
1092                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1093                 case VIRTIO_NET_HDR_GSO_TCPV4:
1094                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1095                         break;
1096                 case VIRTIO_NET_HDR_GSO_TCPV6:
1097                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1098                         break;
1099                 case VIRTIO_NET_HDR_GSO_UDP:
1100                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1101                         break;
1102                 default:
1103                         tun->dev->stats.rx_frame_errors++;
1104                         kfree_skb(skb);
1105                         return -EINVAL;
1106                 }
1107
1108                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1109                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1110
1111                 skb_shinfo(skb)->gso_size = gso.gso_size;
1112                 if (skb_shinfo(skb)->gso_size == 0) {
1113                         tun->dev->stats.rx_frame_errors++;
1114                         kfree_skb(skb);
1115                         return -EINVAL;
1116                 }
1117
1118                 /* Header must be checked, and gso_segs computed. */
1119                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1120                 skb_shinfo(skb)->gso_segs = 0;
1121         }
1122
1123         /* copy skb_ubuf_info for callback when skb has no error */
1124         if (zerocopy) {
1125                 skb_shinfo(skb)->destructor_arg = msg_control;
1126                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1127                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1128         }
1129
1130         skb_reset_network_header(skb);
1131         skb_probe_transport_header(skb, 0);
1132
1133         rxhash = skb_get_rxhash(skb);
1134         netif_rx_ni(skb);
1135
1136         tun->dev->stats.rx_packets++;
1137         tun->dev->stats.rx_bytes += len;
1138
1139         tun_flow_update(tun, rxhash, tfile);
1140         return total_len;
1141 }
1142
1143 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1144                               unsigned long count, loff_t pos)
1145 {
1146         struct file *file = iocb->ki_filp;
1147         struct tun_struct *tun = tun_get(file);
1148         struct tun_file *tfile = file->private_data;
1149         ssize_t result;
1150
1151         if (!tun)
1152                 return -EBADFD;
1153
1154         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1155
1156         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1157                               count, file->f_flags & O_NONBLOCK);
1158
1159         tun_put(tun);
1160         return result;
1161 }
1162
1163 /* Put packet to the user space buffer */
1164 static ssize_t tun_put_user(struct tun_struct *tun,
1165                             struct tun_file *tfile,
1166                             struct sk_buff *skb,
1167                             const struct iovec *iv, int len)
1168 {
1169         struct tun_pi pi = { 0, skb->protocol };
1170         ssize_t total = 0;
1171         int vlan_offset = 0;
1172
1173         if (!(tun->flags & TUN_NO_PI)) {
1174                 if ((len -= sizeof(pi)) < 0)
1175                         return -EINVAL;
1176
1177                 if (len < skb->len) {
1178                         /* Packet will be striped */
1179                         pi.flags |= TUN_PKT_STRIP;
1180                 }
1181
1182                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1183                         return -EFAULT;
1184                 total += sizeof(pi);
1185         }
1186
1187         if (tun->flags & TUN_VNET_HDR) {
1188                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1189                 if ((len -= tun->vnet_hdr_sz) < 0)
1190                         return -EINVAL;
1191
1192                 if (skb_is_gso(skb)) {
1193                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1194
1195                         /* This is a hint as to how much should be linear. */
1196                         gso.hdr_len = skb_headlen(skb);
1197                         gso.gso_size = sinfo->gso_size;
1198                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1199                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1200                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1201                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1202                         else if (sinfo->gso_type & SKB_GSO_UDP)
1203                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1204                         else {
1205                                 pr_err("unexpected GSO type: "
1206                                        "0x%x, gso_size %d, hdr_len %d\n",
1207                                        sinfo->gso_type, gso.gso_size,
1208                                        gso.hdr_len);
1209                                 print_hex_dump(KERN_ERR, "tun: ",
1210                                                DUMP_PREFIX_NONE,
1211                                                16, 1, skb->head,
1212                                                min((int)gso.hdr_len, 64), true);
1213                                 WARN_ON_ONCE(1);
1214                                 return -EINVAL;
1215                         }
1216                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1217                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1218                 } else
1219                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1220
1221                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1222                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1223                         gso.csum_start = skb_checksum_start_offset(skb);
1224                         gso.csum_offset = skb->csum_offset;
1225                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1226                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1227                 } /* else everything is zero */
1228
1229                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1230                                                sizeof(gso))))
1231                         return -EFAULT;
1232                 total += tun->vnet_hdr_sz;
1233         }
1234
1235         if (!vlan_tx_tag_present(skb)) {
1236                 len = min_t(int, skb->len, len);
1237         } else {
1238                 int copy, ret;
1239                 struct {
1240                         __be16 h_vlan_proto;
1241                         __be16 h_vlan_TCI;
1242                 } veth;
1243
1244                 veth.h_vlan_proto = skb->vlan_proto;
1245                 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1246
1247                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1248                 len = min_t(int, skb->len + VLAN_HLEN, len);
1249
1250                 copy = min_t(int, vlan_offset, len);
1251                 ret = skb_copy_datagram_const_iovec(skb, 0, iv, total, copy);
1252                 len -= copy;
1253                 total += copy;
1254                 if (ret || !len)
1255                         goto done;
1256
1257                 copy = min_t(int, sizeof(veth), len);
1258                 ret = memcpy_toiovecend(iv, (void *)&veth, total, copy);
1259                 len -= copy;
1260                 total += copy;
1261                 if (ret || !len)
1262                         goto done;
1263         }
1264
1265         skb_copy_datagram_const_iovec(skb, vlan_offset, iv, total, len);
1266         total += len;
1267
1268 done:
1269         tun->dev->stats.tx_packets++;
1270         tun->dev->stats.tx_bytes += len;
1271
1272         return total;
1273 }
1274
1275 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1276                            struct kiocb *iocb, const struct iovec *iv,
1277                            ssize_t len, int noblock)
1278 {
1279         DECLARE_WAITQUEUE(wait, current);
1280         struct sk_buff *skb;
1281         ssize_t ret = 0;
1282
1283         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1284
1285         if (unlikely(!noblock))
1286                 add_wait_queue(&tfile->wq.wait, &wait);
1287         while (len) {
1288                 current->state = TASK_INTERRUPTIBLE;
1289
1290                 /* Read frames from the queue */
1291                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1292                         if (noblock) {
1293                                 ret = -EAGAIN;
1294                                 break;
1295                         }
1296                         if (signal_pending(current)) {
1297                                 ret = -ERESTARTSYS;
1298                                 break;
1299                         }
1300                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1301                                 ret = -EIO;
1302                                 break;
1303                         }
1304
1305                         /* Nothing to read, let's sleep */
1306                         schedule();
1307                         continue;
1308                 }
1309
1310                 ret = tun_put_user(tun, tfile, skb, iv, len);
1311                 kfree_skb(skb);
1312                 break;
1313         }
1314
1315         current->state = TASK_RUNNING;
1316         if (unlikely(!noblock))
1317                 remove_wait_queue(&tfile->wq.wait, &wait);
1318
1319         return ret;
1320 }
1321
1322 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1323                             unsigned long count, loff_t pos)
1324 {
1325         struct file *file = iocb->ki_filp;
1326         struct tun_file *tfile = file->private_data;
1327         struct tun_struct *tun = __tun_get(tfile);
1328         ssize_t len, ret;
1329
1330         if (!tun)
1331                 return -EBADFD;
1332         len = iov_length(iv, count);
1333         if (len < 0) {
1334                 ret = -EINVAL;
1335                 goto out;
1336         }
1337
1338         ret = tun_do_read(tun, tfile, iocb, iv, len,
1339                           file->f_flags & O_NONBLOCK);
1340         ret = min_t(ssize_t, ret, len);
1341 out:
1342         tun_put(tun);
1343         return ret;
1344 }
1345
1346 static void tun_free_netdev(struct net_device *dev)
1347 {
1348         struct tun_struct *tun = netdev_priv(dev);
1349
1350         BUG_ON(!(list_empty(&tun->disabled)));
1351         tun_flow_uninit(tun);
1352         security_tun_dev_free_security(tun->security);
1353         free_netdev(dev);
1354 }
1355
1356 static void tun_setup(struct net_device *dev)
1357 {
1358         struct tun_struct *tun = netdev_priv(dev);
1359
1360         tun->owner = INVALID_UID;
1361         tun->group = INVALID_GID;
1362
1363         dev->ethtool_ops = &tun_ethtool_ops;
1364         dev->destructor = tun_free_netdev;
1365 }
1366
1367 /* Trivial set of netlink ops to allow deleting tun or tap
1368  * device with netlink.
1369  */
1370 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1371 {
1372         return -EINVAL;
1373 }
1374
1375 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1376         .kind           = DRV_NAME,
1377         .priv_size      = sizeof(struct tun_struct),
1378         .setup          = tun_setup,
1379         .validate       = tun_validate,
1380 };
1381
1382 static void tun_sock_write_space(struct sock *sk)
1383 {
1384         struct tun_file *tfile;
1385         wait_queue_head_t *wqueue;
1386
1387         if (!sock_writeable(sk))
1388                 return;
1389
1390         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1391                 return;
1392
1393         wqueue = sk_sleep(sk);
1394         if (wqueue && waitqueue_active(wqueue))
1395                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1396                                                 POLLWRNORM | POLLWRBAND);
1397
1398         tfile = container_of(sk, struct tun_file, sk);
1399         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1400 }
1401
1402 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1403                        struct msghdr *m, size_t total_len)
1404 {
1405         int ret;
1406         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1407         struct tun_struct *tun = __tun_get(tfile);
1408
1409         if (!tun)
1410                 return -EBADFD;
1411         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1412                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1413         tun_put(tun);
1414         return ret;
1415 }
1416
1417 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1418                        struct msghdr *m, size_t total_len,
1419                        int flags)
1420 {
1421         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1422         struct tun_struct *tun = __tun_get(tfile);
1423         int ret;
1424
1425         if (!tun)
1426                 return -EBADFD;
1427
1428         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1429                 ret = -EINVAL;
1430                 goto out;
1431         }
1432         if (flags & MSG_ERRQUEUE) {
1433                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1434                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1435                 goto out;
1436         }
1437         ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1438                           flags & MSG_DONTWAIT);
1439         if (ret > total_len) {
1440                 m->msg_flags |= MSG_TRUNC;
1441                 ret = flags & MSG_TRUNC ? ret : total_len;
1442         }
1443 out:
1444         tun_put(tun);
1445         return ret;
1446 }
1447
1448 static int tun_release(struct socket *sock)
1449 {
1450         if (sock->sk)
1451                 sock_put(sock->sk);
1452         return 0;
1453 }
1454
1455 /* Ops structure to mimic raw sockets with tun */
1456 static const struct proto_ops tun_socket_ops = {
1457         .sendmsg = tun_sendmsg,
1458         .recvmsg = tun_recvmsg,
1459         .release = tun_release,
1460 };
1461
1462 static struct proto tun_proto = {
1463         .name           = "tun",
1464         .owner          = THIS_MODULE,
1465         .obj_size       = sizeof(struct tun_file),
1466 };
1467
1468 static int tun_flags(struct tun_struct *tun)
1469 {
1470         int flags = 0;
1471
1472         if (tun->flags & TUN_TUN_DEV)
1473                 flags |= IFF_TUN;
1474         else
1475                 flags |= IFF_TAP;
1476
1477         if (tun->flags & TUN_NO_PI)
1478                 flags |= IFF_NO_PI;
1479
1480         /* This flag has no real effect.  We track the value for backwards
1481          * compatibility.
1482          */
1483         if (tun->flags & TUN_ONE_QUEUE)
1484                 flags |= IFF_ONE_QUEUE;
1485
1486         if (tun->flags & TUN_VNET_HDR)
1487                 flags |= IFF_VNET_HDR;
1488
1489         if (tun->flags & TUN_TAP_MQ)
1490                 flags |= IFF_MULTI_QUEUE;
1491
1492         if (tun->flags & TUN_PERSIST)
1493                 flags |= IFF_PERSIST;
1494
1495         return flags;
1496 }
1497
1498 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1499                               char *buf)
1500 {
1501         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1502         return sprintf(buf, "0x%x\n", tun_flags(tun));
1503 }
1504
1505 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1506                               char *buf)
1507 {
1508         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1509         return uid_valid(tun->owner)?
1510                 sprintf(buf, "%u\n",
1511                         from_kuid_munged(current_user_ns(), tun->owner)):
1512                 sprintf(buf, "-1\n");
1513 }
1514
1515 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1516                               char *buf)
1517 {
1518         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1519         return gid_valid(tun->group) ?
1520                 sprintf(buf, "%u\n",
1521                         from_kgid_munged(current_user_ns(), tun->group)):
1522                 sprintf(buf, "-1\n");
1523 }
1524
1525 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1526 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1527 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1528
1529 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1530 {
1531         struct tun_struct *tun;
1532         struct tun_file *tfile = file->private_data;
1533         struct net_device *dev;
1534         int err;
1535
1536         if (tfile->detached)
1537                 return -EINVAL;
1538
1539         dev = __dev_get_by_name(net, ifr->ifr_name);
1540         if (dev) {
1541                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1542                         return -EBUSY;
1543                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1544                         tun = netdev_priv(dev);
1545                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1546                         tun = netdev_priv(dev);
1547                 else
1548                         return -EINVAL;
1549
1550                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1551                     !!(tun->flags & TUN_TAP_MQ))
1552                         return -EINVAL;
1553
1554                 if (tun_not_capable(tun))
1555                         return -EPERM;
1556                 err = security_tun_dev_open(tun->security);
1557                 if (err < 0)
1558                         return err;
1559
1560                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1561                 if (err < 0)
1562                         return err;
1563
1564                 if (tun->flags & TUN_TAP_MQ &&
1565                     (tun->numqueues + tun->numdisabled > 1)) {
1566                         /* One or more queue has already been attached, no need
1567                          * to initialize the device again.
1568                          */
1569                         return 0;
1570                 }
1571         }
1572         else {
1573                 char *name;
1574                 unsigned long flags = 0;
1575                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1576                              MAX_TAP_QUEUES : 1;
1577
1578                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1579                         return -EPERM;
1580                 err = security_tun_dev_create();
1581                 if (err < 0)
1582                         return err;
1583
1584                 /* Set dev type */
1585                 if (ifr->ifr_flags & IFF_TUN) {
1586                         /* TUN device */
1587                         flags |= TUN_TUN_DEV;
1588                         name = "tun%d";
1589                 } else if (ifr->ifr_flags & IFF_TAP) {
1590                         /* TAP device */
1591                         flags |= TUN_TAP_DEV;
1592                         name = "tap%d";
1593                 } else
1594                         return -EINVAL;
1595
1596                 if (*ifr->ifr_name)
1597                         name = ifr->ifr_name;
1598
1599                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1600                                        tun_setup, queues, queues);
1601
1602                 if (!dev)
1603                         return -ENOMEM;
1604
1605                 dev_net_set(dev, net);
1606                 dev->rtnl_link_ops = &tun_link_ops;
1607                 dev->ifindex = tfile->ifindex;
1608
1609                 tun = netdev_priv(dev);
1610                 tun->dev = dev;
1611                 tun->flags = flags;
1612                 tun->txflt.count = 0;
1613                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1614
1615                 tun->filter_attached = false;
1616                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1617
1618                 spin_lock_init(&tun->lock);
1619
1620                 err = security_tun_dev_alloc_security(&tun->security);
1621                 if (err < 0)
1622                         goto err_free_dev;
1623
1624                 tun_net_init(dev);
1625                 tun_flow_init(tun);
1626
1627                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1628                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1629                                    NETIF_F_HW_VLAN_STAG_TX;
1630                 dev->features = dev->hw_features;
1631                 dev->vlan_features = dev->features;
1632
1633                 INIT_LIST_HEAD(&tun->disabled);
1634                 err = tun_attach(tun, file, false);
1635                 if (err < 0)
1636                         goto err_free_dev;
1637
1638                 err = register_netdevice(tun->dev);
1639                 if (err < 0)
1640                         goto err_free_dev;
1641
1642                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1643                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1644                     device_create_file(&tun->dev->dev, &dev_attr_group))
1645                         pr_err("Failed to create tun sysfs files\n");
1646         }
1647
1648         netif_carrier_on(tun->dev);
1649
1650         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1651
1652         if (ifr->ifr_flags & IFF_NO_PI)
1653                 tun->flags |= TUN_NO_PI;
1654         else
1655                 tun->flags &= ~TUN_NO_PI;
1656
1657         /* This flag has no real effect.  We track the value for backwards
1658          * compatibility.
1659          */
1660         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1661                 tun->flags |= TUN_ONE_QUEUE;
1662         else
1663                 tun->flags &= ~TUN_ONE_QUEUE;
1664
1665         if (ifr->ifr_flags & IFF_VNET_HDR)
1666                 tun->flags |= TUN_VNET_HDR;
1667         else
1668                 tun->flags &= ~TUN_VNET_HDR;
1669
1670         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1671                 tun->flags |= TUN_TAP_MQ;
1672         else
1673                 tun->flags &= ~TUN_TAP_MQ;
1674
1675         /* Make sure persistent devices do not get stuck in
1676          * xoff state.
1677          */
1678         if (netif_running(tun->dev))
1679                 netif_tx_wake_all_queues(tun->dev);
1680
1681         strcpy(ifr->ifr_name, tun->dev->name);
1682         return 0;
1683
1684  err_free_dev:
1685         free_netdev(dev);
1686         return err;
1687 }
1688
1689 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1690                        struct ifreq *ifr)
1691 {
1692         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1693
1694         strcpy(ifr->ifr_name, tun->dev->name);
1695
1696         ifr->ifr_flags = tun_flags(tun);
1697
1698 }
1699
1700 /* This is like a cut-down ethtool ops, except done via tun fd so no
1701  * privs required. */
1702 static int set_offload(struct tun_struct *tun, unsigned long arg)
1703 {
1704         netdev_features_t features = 0;
1705
1706         if (arg & TUN_F_CSUM) {
1707                 features |= NETIF_F_HW_CSUM;
1708                 arg &= ~TUN_F_CSUM;
1709
1710                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1711                         if (arg & TUN_F_TSO_ECN) {
1712                                 features |= NETIF_F_TSO_ECN;
1713                                 arg &= ~TUN_F_TSO_ECN;
1714                         }
1715                         if (arg & TUN_F_TSO4)
1716                                 features |= NETIF_F_TSO;
1717                         if (arg & TUN_F_TSO6)
1718                                 features |= NETIF_F_TSO6;
1719                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1720                 }
1721
1722                 if (arg & TUN_F_UFO) {
1723                         features |= NETIF_F_UFO;
1724                         arg &= ~TUN_F_UFO;
1725                 }
1726         }
1727
1728         /* This gives the user a way to test for new features in future by
1729          * trying to set them. */
1730         if (arg)
1731                 return -EINVAL;
1732
1733         tun->set_features = features;
1734         netdev_update_features(tun->dev);
1735
1736         return 0;
1737 }
1738
1739 static void tun_detach_filter(struct tun_struct *tun, int n)
1740 {
1741         int i;
1742         struct tun_file *tfile;
1743
1744         for (i = 0; i < n; i++) {
1745                 tfile = rtnl_dereference(tun->tfiles[i]);
1746                 sk_detach_filter(tfile->socket.sk);
1747         }
1748
1749         tun->filter_attached = false;
1750 }
1751
1752 static int tun_attach_filter(struct tun_struct *tun)
1753 {
1754         int i, ret = 0;
1755         struct tun_file *tfile;
1756
1757         for (i = 0; i < tun->numqueues; i++) {
1758                 tfile = rtnl_dereference(tun->tfiles[i]);
1759                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1760                 if (ret) {
1761                         tun_detach_filter(tun, i);
1762                         return ret;
1763                 }
1764         }
1765
1766         tun->filter_attached = true;
1767         return ret;
1768 }
1769
1770 static void tun_set_sndbuf(struct tun_struct *tun)
1771 {
1772         struct tun_file *tfile;
1773         int i;
1774
1775         for (i = 0; i < tun->numqueues; i++) {
1776                 tfile = rtnl_dereference(tun->tfiles[i]);
1777                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1778         }
1779 }
1780
1781 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1782 {
1783         struct tun_file *tfile = file->private_data;
1784         struct tun_struct *tun;
1785         int ret = 0;
1786
1787         rtnl_lock();
1788
1789         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1790                 tun = tfile->detached;
1791                 if (!tun) {
1792                         ret = -EINVAL;
1793                         goto unlock;
1794                 }
1795                 ret = security_tun_dev_attach_queue(tun->security);
1796                 if (ret < 0)
1797                         goto unlock;
1798                 ret = tun_attach(tun, file, false);
1799         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1800                 tun = rtnl_dereference(tfile->tun);
1801                 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1802                         ret = -EINVAL;
1803                 else
1804                         __tun_detach(tfile, false);
1805         } else
1806                 ret = -EINVAL;
1807
1808 unlock:
1809         rtnl_unlock();
1810         return ret;
1811 }
1812
1813 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1814                             unsigned long arg, int ifreq_len)
1815 {
1816         struct tun_file *tfile = file->private_data;
1817         struct tun_struct *tun;
1818         void __user* argp = (void __user*)arg;
1819         struct ifreq ifr;
1820         kuid_t owner;
1821         kgid_t group;
1822         int sndbuf;
1823         int vnet_hdr_sz;
1824         unsigned int ifindex;
1825         int ret;
1826
1827         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1828                 if (copy_from_user(&ifr, argp, ifreq_len))
1829                         return -EFAULT;
1830         } else {
1831                 memset(&ifr, 0, sizeof(ifr));
1832         }
1833         if (cmd == TUNGETFEATURES) {
1834                 /* Currently this just means: "what IFF flags are valid?".
1835                  * This is needed because we never checked for invalid flags on
1836                  * TUNSETIFF. */
1837                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1838                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1839                                 (unsigned int __user*)argp);
1840         } else if (cmd == TUNSETQUEUE)
1841                 return tun_set_queue(file, &ifr);
1842
1843         ret = 0;
1844         rtnl_lock();
1845
1846         tun = __tun_get(tfile);
1847         if (cmd == TUNSETIFF && !tun) {
1848                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1849
1850                 ret = tun_set_iff(tfile->net, file, &ifr);
1851
1852                 if (ret)
1853                         goto unlock;
1854
1855                 if (copy_to_user(argp, &ifr, ifreq_len))
1856                         ret = -EFAULT;
1857                 goto unlock;
1858         }
1859         if (cmd == TUNSETIFINDEX) {
1860                 ret = -EPERM;
1861                 if (tun)
1862                         goto unlock;
1863
1864                 ret = -EFAULT;
1865                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1866                         goto unlock;
1867
1868                 ret = 0;
1869                 tfile->ifindex = ifindex;
1870                 goto unlock;
1871         }
1872
1873         ret = -EBADFD;
1874         if (!tun)
1875                 goto unlock;
1876
1877         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1878
1879         ret = 0;
1880         switch (cmd) {
1881         case TUNGETIFF:
1882                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1883
1884                 if (tfile->detached)
1885                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
1886                 if (!tfile->socket.sk->sk_filter)
1887                         ifr.ifr_flags |= IFF_NOFILTER;
1888
1889                 if (copy_to_user(argp, &ifr, ifreq_len))
1890                         ret = -EFAULT;
1891                 break;
1892
1893         case TUNSETNOCSUM:
1894                 /* Disable/Enable checksum */
1895
1896                 /* [unimplemented] */
1897                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1898                           arg ? "disabled" : "enabled");
1899                 break;
1900
1901         case TUNSETPERSIST:
1902                 /* Disable/Enable persist mode. Keep an extra reference to the
1903                  * module to prevent the module being unprobed.
1904                  */
1905                 if (arg && !(tun->flags & TUN_PERSIST)) {
1906                         tun->flags |= TUN_PERSIST;
1907                         __module_get(THIS_MODULE);
1908                 }
1909                 if (!arg && (tun->flags & TUN_PERSIST)) {
1910                         tun->flags &= ~TUN_PERSIST;
1911                         module_put(THIS_MODULE);
1912                 }
1913
1914                 tun_debug(KERN_INFO, tun, "persist %s\n",
1915                           arg ? "enabled" : "disabled");
1916                 break;
1917
1918         case TUNSETOWNER:
1919                 /* Set owner of the device */
1920                 owner = make_kuid(current_user_ns(), arg);
1921                 if (!uid_valid(owner)) {
1922                         ret = -EINVAL;
1923                         break;
1924                 }
1925                 tun->owner = owner;
1926                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1927                           from_kuid(&init_user_ns, tun->owner));
1928                 break;
1929
1930         case TUNSETGROUP:
1931                 /* Set group of the device */
1932                 group = make_kgid(current_user_ns(), arg);
1933                 if (!gid_valid(group)) {
1934                         ret = -EINVAL;
1935                         break;
1936                 }
1937                 tun->group = group;
1938                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1939                           from_kgid(&init_user_ns, tun->group));
1940                 break;
1941
1942         case TUNSETLINK:
1943                 /* Only allow setting the type when the interface is down */
1944                 if (tun->dev->flags & IFF_UP) {
1945                         tun_debug(KERN_INFO, tun,
1946                                   "Linktype set failed because interface is up\n");
1947                         ret = -EBUSY;
1948                 } else {
1949                         tun->dev->type = (int) arg;
1950                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1951                                   tun->dev->type);
1952                         ret = 0;
1953                 }
1954                 break;
1955
1956 #ifdef TUN_DEBUG
1957         case TUNSETDEBUG:
1958                 tun->debug = arg;
1959                 break;
1960 #endif
1961         case TUNSETOFFLOAD:
1962                 ret = set_offload(tun, arg);
1963                 break;
1964
1965         case TUNSETTXFILTER:
1966                 /* Can be set only for TAPs */
1967                 ret = -EINVAL;
1968                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1969                         break;
1970                 ret = update_filter(&tun->txflt, (void __user *)arg);
1971                 break;
1972
1973         case SIOCGIFHWADDR:
1974                 /* Get hw address */
1975                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1976                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1977                 if (copy_to_user(argp, &ifr, ifreq_len))
1978                         ret = -EFAULT;
1979                 break;
1980
1981         case SIOCSIFHWADDR:
1982                 /* Set hw address */
1983                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1984                           ifr.ifr_hwaddr.sa_data);
1985
1986                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1987                 break;
1988
1989         case TUNGETSNDBUF:
1990                 sndbuf = tfile->socket.sk->sk_sndbuf;
1991                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1992                         ret = -EFAULT;
1993                 break;
1994
1995         case TUNSETSNDBUF:
1996                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1997                         ret = -EFAULT;
1998                         break;
1999                 }
2000
2001                 tun->sndbuf = sndbuf;
2002                 tun_set_sndbuf(tun);
2003                 break;
2004
2005         case TUNGETVNETHDRSZ:
2006                 vnet_hdr_sz = tun->vnet_hdr_sz;
2007                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2008                         ret = -EFAULT;
2009                 break;
2010
2011         case TUNSETVNETHDRSZ:
2012                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2013                         ret = -EFAULT;
2014                         break;
2015                 }
2016                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2017                         ret = -EINVAL;
2018                         break;
2019                 }
2020
2021                 tun->vnet_hdr_sz = vnet_hdr_sz;
2022                 break;
2023
2024         case TUNATTACHFILTER:
2025                 /* Can be set only for TAPs */
2026                 ret = -EINVAL;
2027                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2028                         break;
2029                 ret = -EFAULT;
2030                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2031                         break;
2032
2033                 ret = tun_attach_filter(tun);
2034                 break;
2035
2036         case TUNDETACHFILTER:
2037                 /* Can be set only for TAPs */
2038                 ret = -EINVAL;
2039                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2040                         break;
2041                 ret = 0;
2042                 tun_detach_filter(tun, tun->numqueues);
2043                 break;
2044
2045         case TUNGETFILTER:
2046                 ret = -EINVAL;
2047                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2048                         break;
2049                 ret = -EFAULT;
2050                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2051                         break;
2052                 ret = 0;
2053                 break;
2054
2055         default:
2056                 ret = -EINVAL;
2057                 break;
2058         }
2059
2060 unlock:
2061         rtnl_unlock();
2062         if (tun)
2063                 tun_put(tun);
2064         return ret;
2065 }
2066
2067 static long tun_chr_ioctl(struct file *file,
2068                           unsigned int cmd, unsigned long arg)
2069 {
2070         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2071 }
2072
2073 #ifdef CONFIG_COMPAT
2074 static long tun_chr_compat_ioctl(struct file *file,
2075                          unsigned int cmd, unsigned long arg)
2076 {
2077         switch (cmd) {
2078         case TUNSETIFF:
2079         case TUNGETIFF:
2080         case TUNSETTXFILTER:
2081         case TUNGETSNDBUF:
2082         case TUNSETSNDBUF:
2083         case SIOCGIFHWADDR:
2084         case SIOCSIFHWADDR:
2085                 arg = (unsigned long)compat_ptr(arg);
2086                 break;
2087         default:
2088                 arg = (compat_ulong_t)arg;
2089                 break;
2090         }
2091
2092         /*
2093          * compat_ifreq is shorter than ifreq, so we must not access beyond
2094          * the end of that structure. All fields that are used in this
2095          * driver are compatible though, we don't need to convert the
2096          * contents.
2097          */
2098         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2099 }
2100 #endif /* CONFIG_COMPAT */
2101
2102 static int tun_chr_fasync(int fd, struct file *file, int on)
2103 {
2104         struct tun_file *tfile = file->private_data;
2105         int ret;
2106
2107         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2108                 goto out;
2109
2110         if (on) {
2111                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2112                 if (ret)
2113                         goto out;
2114                 tfile->flags |= TUN_FASYNC;
2115         } else
2116                 tfile->flags &= ~TUN_FASYNC;
2117         ret = 0;
2118 out:
2119         return ret;
2120 }
2121
2122 static int tun_chr_open(struct inode *inode, struct file * file)
2123 {
2124         struct tun_file *tfile;
2125
2126         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2127
2128         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2129                                             &tun_proto);
2130         if (!tfile)
2131                 return -ENOMEM;
2132         rcu_assign_pointer(tfile->tun, NULL);
2133         tfile->net = get_net(current->nsproxy->net_ns);
2134         tfile->flags = 0;
2135         tfile->ifindex = 0;
2136
2137         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2138         init_waitqueue_head(&tfile->wq.wait);
2139
2140         tfile->socket.file = file;
2141         tfile->socket.ops = &tun_socket_ops;
2142
2143         sock_init_data(&tfile->socket, &tfile->sk);
2144         sk_change_net(&tfile->sk, tfile->net);
2145
2146         tfile->sk.sk_write_space = tun_sock_write_space;
2147         tfile->sk.sk_sndbuf = INT_MAX;
2148
2149         file->private_data = tfile;
2150         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2151         INIT_LIST_HEAD(&tfile->next);
2152
2153         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2154
2155         return 0;
2156 }
2157
2158 static int tun_chr_close(struct inode *inode, struct file *file)
2159 {
2160         struct tun_file *tfile = file->private_data;
2161         struct net *net = tfile->net;
2162
2163         tun_detach(tfile, true);
2164         put_net(net);
2165
2166         return 0;
2167 }
2168
2169 static const struct file_operations tun_fops = {
2170         .owner  = THIS_MODULE,
2171         .llseek = no_llseek,
2172         .read  = do_sync_read,
2173         .aio_read  = tun_chr_aio_read,
2174         .write = do_sync_write,
2175         .aio_write = tun_chr_aio_write,
2176         .poll   = tun_chr_poll,
2177         .unlocked_ioctl = tun_chr_ioctl,
2178 #ifdef CONFIG_COMPAT
2179         .compat_ioctl = tun_chr_compat_ioctl,
2180 #endif
2181         .open   = tun_chr_open,
2182         .release = tun_chr_close,
2183         .fasync = tun_chr_fasync
2184 };
2185
2186 static struct miscdevice tun_miscdev = {
2187         .minor = TUN_MINOR,
2188         .name = "tun",
2189         .nodename = "net/tun",
2190         .fops = &tun_fops,
2191 };
2192
2193 /* ethtool interface */
2194
2195 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2196 {
2197         cmd->supported          = 0;
2198         cmd->advertising        = 0;
2199         ethtool_cmd_speed_set(cmd, SPEED_10);
2200         cmd->duplex             = DUPLEX_FULL;
2201         cmd->port               = PORT_TP;
2202         cmd->phy_address        = 0;
2203         cmd->transceiver        = XCVR_INTERNAL;
2204         cmd->autoneg            = AUTONEG_DISABLE;
2205         cmd->maxtxpkt           = 0;
2206         cmd->maxrxpkt           = 0;
2207         return 0;
2208 }
2209
2210 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2211 {
2212         struct tun_struct *tun = netdev_priv(dev);
2213
2214         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2215         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2216
2217         switch (tun->flags & TUN_TYPE_MASK) {
2218         case TUN_TUN_DEV:
2219                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2220                 break;
2221         case TUN_TAP_DEV:
2222                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2223                 break;
2224         }
2225 }
2226
2227 static u32 tun_get_msglevel(struct net_device *dev)
2228 {
2229 #ifdef TUN_DEBUG
2230         struct tun_struct *tun = netdev_priv(dev);
2231         return tun->debug;
2232 #else
2233         return -EOPNOTSUPP;
2234 #endif
2235 }
2236
2237 static void tun_set_msglevel(struct net_device *dev, u32 value)
2238 {
2239 #ifdef TUN_DEBUG
2240         struct tun_struct *tun = netdev_priv(dev);
2241         tun->debug = value;
2242 #endif
2243 }
2244
2245 static const struct ethtool_ops tun_ethtool_ops = {
2246         .get_settings   = tun_get_settings,
2247         .get_drvinfo    = tun_get_drvinfo,
2248         .get_msglevel   = tun_get_msglevel,
2249         .set_msglevel   = tun_set_msglevel,
2250         .get_link       = ethtool_op_get_link,
2251         .get_ts_info    = ethtool_op_get_ts_info,
2252 };
2253
2254
2255 static int __init tun_init(void)
2256 {
2257         int ret = 0;
2258
2259         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2260         pr_info("%s\n", DRV_COPYRIGHT);
2261
2262         ret = rtnl_link_register(&tun_link_ops);
2263         if (ret) {
2264                 pr_err("Can't register link_ops\n");
2265                 goto err_linkops;
2266         }
2267
2268         ret = misc_register(&tun_miscdev);
2269         if (ret) {
2270                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2271                 goto err_misc;
2272         }
2273         return  0;
2274 err_misc:
2275         rtnl_link_unregister(&tun_link_ops);
2276 err_linkops:
2277         return ret;
2278 }
2279
2280 static void tun_cleanup(void)
2281 {
2282         misc_deregister(&tun_miscdev);
2283         rtnl_link_unregister(&tun_link_ops);
2284 }
2285
2286 /* Get an underlying socket object from tun file.  Returns error unless file is
2287  * attached to a device.  The returned object works like a packet socket, it
2288  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2289  * holding a reference to the file for as long as the socket is in use. */
2290 struct socket *tun_get_socket(struct file *file)
2291 {
2292         struct tun_file *tfile;
2293         if (file->f_op != &tun_fops)
2294                 return ERR_PTR(-EINVAL);
2295         tfile = file->private_data;
2296         if (!tfile)
2297                 return ERR_PTR(-EBADFD);
2298         return &tfile->socket;
2299 }
2300 EXPORT_SYMBOL_GPL(tun_get_socket);
2301
2302 module_init(tun_init);
2303 module_exit(tun_cleanup);
2304 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2305 MODULE_AUTHOR(DRV_COPYRIGHT);
2306 MODULE_LICENSE("GPL");
2307 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2308 MODULE_ALIAS("devname:net/tun");