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