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 offset = 0;
 985         int copylen;
 986         bool zerocopy = false;
 987         int err;
 988         u32 rxhash;
 989
 990         if (!(tun->flags & TUN_NO_PI)) {
 991                 if (len < sizeof(pi))
 992                         return -EINVAL;
 993                 len -= sizeof(pi);
 994
 995                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
 996                         return -EFAULT;
 997                 offset += sizeof(pi);
 998         }
 999
1000         if (tun->flags & TUN_VNET_HDR) {
1001                 if (len < tun->vnet_hdr_sz)
1002                         return -EINVAL;
1003                 len -= tun->vnet_hdr_sz;
1004
1005                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1006                         return -EFAULT;
1007
1008                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1009                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1010                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1011
1012                 if (gso.hdr_len > len)
1013                         return -EINVAL;
1014                 offset += tun->vnet_hdr_sz;
1015         }
1016
1017         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1018                 align += NET_IP_ALIGN;
1019                 if (unlikely(len < ETH_HLEN ||
1020                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1021                         return -EINVAL;
1022         }
1023
1024         if (msg_control) {
1025                 /* There are 256 bytes to be copied in skb, so there is
1026                  * enough room for skb expand head in case it is used.
1027                  * The rest of the buffer is mapped from userspace.
1028                  */
1029                 copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1030                 linear = copylen;
1031                 if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1032                         zerocopy = true;
1033         }
1034
1035         if (!zerocopy) {
1036                 copylen = len;
1037                 linear = gso.hdr_len;
1038         }
1039
1040         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1041         if (IS_ERR(skb)) {
1042                 if (PTR_ERR(skb) != -EAGAIN)
1043                         tun->dev->stats.rx_dropped++;
1044                 return PTR_ERR(skb);
1045         }
1046
1047         if (zerocopy)
1048                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1049         else {
1050                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1051                 if (!err && msg_control) {
1052                         struct ubuf_info *uarg = msg_control;
1053                         uarg->callback(uarg, false);
1054                 }
1055         }
1056
1057         if (err) {
1058                 tun->dev->stats.rx_dropped++;
1059                 kfree_skb(skb);
1060                 return -EFAULT;
1061         }
1062
1063         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1064                 if (!skb_partial_csum_set(skb, gso.csum_start,
1065                                           gso.csum_offset)) {
1066                         tun->dev->stats.rx_frame_errors++;
1067                         kfree_skb(skb);
1068                         return -EINVAL;
1069                 }
1070         }
1071
1072         switch (tun->flags & TUN_TYPE_MASK) {
1073         case TUN_TUN_DEV:
1074                 if (tun->flags & TUN_NO_PI) {
1075                         switch (skb->data[0] & 0xf0) {
1076                         case 0x40:
1077                                 pi.proto = htons(ETH_P_IP);
1078                                 break;
1079                         case 0x60:
1080                                 pi.proto = htons(ETH_P_IPV6);
1081                                 break;
1082                         default:
1083                                 tun->dev->stats.rx_dropped++;
1084                                 kfree_skb(skb);
1085                                 return -EINVAL;
1086                         }
1087                 }
1088
1089                 skb_reset_mac_header(skb);
1090                 skb->protocol = pi.proto;
1091                 skb->dev = tun->dev;
1092                 break;
1093         case TUN_TAP_DEV:
1094                 skb->protocol = eth_type_trans(skb, tun->dev);
1095                 break;
1096         }
1097
1098         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1099                 pr_debug("GSO!\n");
1100                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1101                 case VIRTIO_NET_HDR_GSO_TCPV4:
1102                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1103                         break;
1104                 case VIRTIO_NET_HDR_GSO_TCPV6:
1105                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1106                         break;
1107                 case VIRTIO_NET_HDR_GSO_UDP:
1108                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1109                         break;
1110                 default:
1111                         tun->dev->stats.rx_frame_errors++;
1112                         kfree_skb(skb);
1113                         return -EINVAL;
1114                 }
1115
1116                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1117                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1118
1119                 skb_shinfo(skb)->gso_size = gso.gso_size;
1120                 if (skb_shinfo(skb)->gso_size == 0) {
1121                         tun->dev->stats.rx_frame_errors++;
1122                         kfree_skb(skb);
1123                         return -EINVAL;
1124                 }
1125
1126                 /* Header must be checked, and gso_segs computed. */
1127                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1128                 skb_shinfo(skb)->gso_segs = 0;
1129         }
1130
1131         /* copy skb_ubuf_info for callback when skb has no error */
1132         if (zerocopy) {
1133                 skb_shinfo(skb)->destructor_arg = msg_control;
1134                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1135                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1136         }
1137
1138         skb_reset_network_header(skb);
1139         skb_probe_transport_header(skb, 0);
1140
1141         rxhash = skb_get_rxhash(skb);
1142         netif_rx_ni(skb);
1143
1144         tun->dev->stats.rx_packets++;
1145         tun->dev->stats.rx_bytes += len;
1146
1147         tun_flow_update(tun, rxhash, tfile);
1148         return total_len;
1149 }
1150
1151 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1152                               unsigned long count, loff_t pos)
1153 {
1154         struct file *file = iocb->ki_filp;
1155         struct tun_struct *tun = tun_get(file);
1156         struct tun_file *tfile = file->private_data;
1157         ssize_t result;
1158
1159         if (!tun)
1160                 return -EBADFD;
1161
1162         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1163
1164         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1165                               count, file->f_flags & O_NONBLOCK);
1166
1167         tun_put(tun);
1168         return result;
1169 }
1170
1171 /* Put packet to the user space buffer */
1172 static ssize_t tun_put_user(struct tun_struct *tun,
1173                             struct tun_file *tfile,
1174                             struct sk_buff *skb,
1175                             const struct iovec *iv, int len)
1176 {
1177         struct tun_pi pi = { 0, skb->protocol };
1178         ssize_t total = 0;
1179         int vlan_offset = 0;
1180
1181         if (!(tun->flags & TUN_NO_PI)) {
1182                 if ((len -= sizeof(pi)) < 0)
1183                         return -EINVAL;
1184
1185                 if (len < skb->len) {
1186                         /* Packet will be striped */
1187                         pi.flags |= TUN_PKT_STRIP;
1188                 }
1189
1190                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1191                         return -EFAULT;
1192                 total += sizeof(pi);
1193         }
1194
1195         if (tun->flags & TUN_VNET_HDR) {
1196                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1197                 if ((len -= tun->vnet_hdr_sz) < 0)
1198                         return -EINVAL;
1199
1200                 if (skb_is_gso(skb)) {
1201                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1202
1203                         /* This is a hint as to how much should be linear. */
1204                         gso.hdr_len = skb_headlen(skb);
1205                         gso.gso_size = sinfo->gso_size;
1206                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1207                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1208                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1209                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1210                         else if (sinfo->gso_type & SKB_GSO_UDP)
1211                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1212                         else {
1213                                 pr_err("unexpected GSO type: "
1214                                        "0x%x, gso_size %d, hdr_len %d\n",
1215                                        sinfo->gso_type, gso.gso_size,
1216                                        gso.hdr_len);
1217                                 print_hex_dump(KERN_ERR, "tun: ",
1218                                                DUMP_PREFIX_NONE,
1219                                                16, 1, skb->head,
1220                                                min((int)gso.hdr_len, 64), true);
1221                                 WARN_ON_ONCE(1);
1222                                 return -EINVAL;
1223                         }
1224                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1225                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1226                 } else
1227                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1228
1229                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1230                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1231                         gso.csum_start = skb_checksum_start_offset(skb);
1232                         gso.csum_offset = skb->csum_offset;
1233                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1234                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1235                 } /* else everything is zero */
1236
1237                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1238                                                sizeof(gso))))
1239                         return -EFAULT;
1240                 total += tun->vnet_hdr_sz;
1241         }
1242
1243         if (!vlan_tx_tag_present(skb)) {
1244                 len = min_t(int, skb->len, len);
1245         } else {
1246                 int copy, ret;
1247                 struct {
1248                         __be16 h_vlan_proto;
1249                         __be16 h_vlan_TCI;
1250                 } veth;
1251
1252                 veth.h_vlan_proto = skb->vlan_proto;
1253                 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1254
1255                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1256                 len = min_t(int, skb->len + VLAN_HLEN, len);
1257
1258                 copy = min_t(int, vlan_offset, len);
1259                 ret = skb_copy_datagram_const_iovec(skb, 0, iv, total, copy);
1260                 len -= copy;
1261                 total += copy;
1262                 if (ret || !len)
1263                         goto done;
1264
1265                 copy = min_t(int, sizeof(veth), len);
1266                 ret = memcpy_toiovecend(iv, (void *)&veth, total, copy);
1267                 len -= copy;
1268                 total += copy;
1269                 if (ret || !len)
1270                         goto done;
1271         }
1272
1273         skb_copy_datagram_const_iovec(skb, vlan_offset, iv, total, len);
1274         total += len;
1275
1276 done:
1277         tun->dev->stats.tx_packets++;
1278         tun->dev->stats.tx_bytes += len;
1279
1280         return total;
1281 }
1282
1283 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1284                            struct kiocb *iocb, const struct iovec *iv,
1285                            ssize_t len, int noblock)
1286 {
1287         DECLARE_WAITQUEUE(wait, current);
1288         struct sk_buff *skb;
1289         ssize_t ret = 0;
1290
1291         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1292
1293         if (unlikely(!noblock))
1294                 add_wait_queue(&tfile->wq.wait, &wait);
1295         while (len) {
1296                 current->state = TASK_INTERRUPTIBLE;
1297
1298                 /* Read frames from the queue */
1299                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1300                         if (noblock) {
1301                                 ret = -EAGAIN;
1302                                 break;
1303                         }
1304                         if (signal_pending(current)) {
1305                                 ret = -ERESTARTSYS;
1306                                 break;
1307                         }
1308                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1309                                 ret = -EIO;
1310                                 break;
1311                         }
1312
1313                         /* Nothing to read, let's sleep */
1314                         schedule();
1315                         continue;
1316                 }
1317
1318                 ret = tun_put_user(tun, tfile, skb, iv, len);
1319                 kfree_skb(skb);
1320                 break;
1321         }
1322
1323         current->state = TASK_RUNNING;
1324         if (unlikely(!noblock))
1325                 remove_wait_queue(&tfile->wq.wait, &wait);
1326
1327         return ret;
1328 }
1329
1330 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1331                             unsigned long count, loff_t pos)
1332 {
1333         struct file *file = iocb->ki_filp;
1334         struct tun_file *tfile = file->private_data;
1335         struct tun_struct *tun = __tun_get(tfile);
1336         ssize_t len, ret;
1337
1338         if (!tun)
1339                 return -EBADFD;
1340         len = iov_length(iv, count);
1341         if (len < 0) {
1342                 ret = -EINVAL;
1343                 goto out;
1344         }
1345
1346         ret = tun_do_read(tun, tfile, iocb, iv, len,
1347                           file->f_flags & O_NONBLOCK);
1348         ret = min_t(ssize_t, ret, len);
1349 out:
1350         tun_put(tun);
1351         return ret;
1352 }
1353
1354 static void tun_free_netdev(struct net_device *dev)
1355 {
1356         struct tun_struct *tun = netdev_priv(dev);
1357
1358         BUG_ON(!(list_empty(&tun->disabled)));
1359         tun_flow_uninit(tun);
1360         security_tun_dev_free_security(tun->security);
1361         free_netdev(dev);
1362 }
1363
1364 static void tun_setup(struct net_device *dev)
1365 {
1366         struct tun_struct *tun = netdev_priv(dev);
1367
1368         tun->owner = INVALID_UID;
1369         tun->group = INVALID_GID;
1370
1371         dev->ethtool_ops = &tun_ethtool_ops;
1372         dev->destructor = tun_free_netdev;
1373 }
1374
1375 /* Trivial set of netlink ops to allow deleting tun or tap
1376  * device with netlink.
1377  */
1378 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1379 {
1380         return -EINVAL;
1381 }
1382
1383 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1384         .kind           = DRV_NAME,
1385         .priv_size      = sizeof(struct tun_struct),
1386         .setup          = tun_setup,
1387         .validate       = tun_validate,
1388 };
1389
1390 static void tun_sock_write_space(struct sock *sk)
1391 {
1392         struct tun_file *tfile;
1393         wait_queue_head_t *wqueue;
1394
1395         if (!sock_writeable(sk))
1396                 return;
1397
1398         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1399                 return;
1400
1401         wqueue = sk_sleep(sk);
1402         if (wqueue && waitqueue_active(wqueue))
1403                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1404                                                 POLLWRNORM | POLLWRBAND);
1405
1406         tfile = container_of(sk, struct tun_file, sk);
1407         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1408 }
1409
1410 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1411                        struct msghdr *m, size_t total_len)
1412 {
1413         int ret;
1414         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1415         struct tun_struct *tun = __tun_get(tfile);
1416
1417         if (!tun)
1418                 return -EBADFD;
1419         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1420                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1421         tun_put(tun);
1422         return ret;
1423 }
1424
1425 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1426                        struct msghdr *m, size_t total_len,
1427                        int flags)
1428 {
1429         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1430         struct tun_struct *tun = __tun_get(tfile);
1431         int ret;
1432
1433         if (!tun)
1434                 return -EBADFD;
1435
1436         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1437                 ret = -EINVAL;
1438                 goto out;
1439         }
1440         if (flags & MSG_ERRQUEUE) {
1441                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1442                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1443                 goto out;
1444         }
1445         ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1446                           flags & MSG_DONTWAIT);
1447         if (ret > total_len) {
1448                 m->msg_flags |= MSG_TRUNC;
1449                 ret = flags & MSG_TRUNC ? ret : total_len;
1450         }
1451 out:
1452         tun_put(tun);
1453         return ret;
1454 }
1455
1456 static int tun_release(struct socket *sock)
1457 {
1458         if (sock->sk)
1459                 sock_put(sock->sk);
1460         return 0;
1461 }
1462
1463 /* Ops structure to mimic raw sockets with tun */
1464 static const struct proto_ops tun_socket_ops = {
1465         .sendmsg = tun_sendmsg,
1466         .recvmsg = tun_recvmsg,
1467         .release = tun_release,
1468 };
1469
1470 static struct proto tun_proto = {
1471         .name           = "tun",
1472         .owner          = THIS_MODULE,
1473         .obj_size       = sizeof(struct tun_file),
1474 };
1475
1476 static int tun_flags(struct tun_struct *tun)
1477 {
1478         int flags = 0;
1479
1480         if (tun->flags & TUN_TUN_DEV)
1481                 flags |= IFF_TUN;
1482         else
1483                 flags |= IFF_TAP;
1484
1485         if (tun->flags & TUN_NO_PI)
1486                 flags |= IFF_NO_PI;
1487
1488         /* This flag has no real effect.  We track the value for backwards
1489          * compatibility.
1490          */
1491         if (tun->flags & TUN_ONE_QUEUE)
1492                 flags |= IFF_ONE_QUEUE;
1493
1494         if (tun->flags & TUN_VNET_HDR)
1495                 flags |= IFF_VNET_HDR;
1496
1497         if (tun->flags & TUN_TAP_MQ)
1498                 flags |= IFF_MULTI_QUEUE;
1499
1500         if (tun->flags & TUN_PERSIST)
1501                 flags |= IFF_PERSIST;
1502
1503         return flags;
1504 }
1505
1506 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1507                               char *buf)
1508 {
1509         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1510         return sprintf(buf, "0x%x\n", tun_flags(tun));
1511 }
1512
1513 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1514                               char *buf)
1515 {
1516         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1517         return uid_valid(tun->owner)?
1518                 sprintf(buf, "%u\n",
1519                         from_kuid_munged(current_user_ns(), tun->owner)):
1520                 sprintf(buf, "-1\n");
1521 }
1522
1523 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1524                               char *buf)
1525 {
1526         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1527         return gid_valid(tun->group) ?
1528                 sprintf(buf, "%u\n",
1529                         from_kgid_munged(current_user_ns(), tun->group)):
1530                 sprintf(buf, "-1\n");
1531 }
1532
1533 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1534 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1535 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1536
1537 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1538 {
1539         struct tun_struct *tun;
1540         struct tun_file *tfile = file->private_data;
1541         struct net_device *dev;
1542         int err;
1543
1544         if (tfile->detached)
1545                 return -EINVAL;
1546
1547         dev = __dev_get_by_name(net, ifr->ifr_name);
1548         if (dev) {
1549                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1550                         return -EBUSY;
1551                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1552                         tun = netdev_priv(dev);
1553                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1554                         tun = netdev_priv(dev);
1555                 else
1556                         return -EINVAL;
1557
1558                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1559                     !!(tun->flags & TUN_TAP_MQ))
1560                         return -EINVAL;
1561
1562                 if (tun_not_capable(tun))
1563                         return -EPERM;
1564                 err = security_tun_dev_open(tun->security);
1565                 if (err < 0)
1566                         return err;
1567
1568                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1569                 if (err < 0)
1570                         return err;
1571
1572                 if (tun->flags & TUN_TAP_MQ &&
1573                     (tun->numqueues + tun->numdisabled > 1)) {
1574                         /* One or more queue has already been attached, no need
1575                          * to initialize the device again.
1576                          */
1577                         return 0;
1578                 }
1579         }
1580         else {
1581                 char *name;
1582                 unsigned long flags = 0;
1583                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1584                              MAX_TAP_QUEUES : 1;
1585
1586                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1587                         return -EPERM;
1588                 err = security_tun_dev_create();
1589                 if (err < 0)
1590                         return err;
1591
1592                 /* Set dev type */
1593                 if (ifr->ifr_flags & IFF_TUN) {
1594                         /* TUN device */
1595                         flags |= TUN_TUN_DEV;
1596                         name = "tun%d";
1597                 } else if (ifr->ifr_flags & IFF_TAP) {
1598                         /* TAP device */
1599                         flags |= TUN_TAP_DEV;
1600                         name = "tap%d";
1601                 } else
1602                         return -EINVAL;
1603
1604                 if (*ifr->ifr_name)
1605                         name = ifr->ifr_name;
1606
1607                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1608                                        tun_setup, queues, queues);
1609
1610                 if (!dev)
1611                         return -ENOMEM;
1612
1613                 dev_net_set(dev, net);
1614                 dev->rtnl_link_ops = &tun_link_ops;
1615                 dev->ifindex = tfile->ifindex;
1616
1617                 tun = netdev_priv(dev);
1618                 tun->dev = dev;
1619                 tun->flags = flags;
1620                 tun->txflt.count = 0;
1621                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1622
1623                 tun->filter_attached = false;
1624                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1625
1626                 spin_lock_init(&tun->lock);
1627
1628                 err = security_tun_dev_alloc_security(&tun->security);
1629                 if (err < 0)
1630                         goto err_free_dev;
1631
1632                 tun_net_init(dev);
1633                 tun_flow_init(tun);
1634
1635                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1636                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1637                                    NETIF_F_HW_VLAN_STAG_TX;
1638                 dev->features = dev->hw_features;
1639                 dev->vlan_features = dev->features;
1640
1641                 INIT_LIST_HEAD(&tun->disabled);
1642                 err = tun_attach(tun, file, false);
1643                 if (err < 0)
1644                         goto err_free_dev;
1645
1646                 err = register_netdevice(tun->dev);
1647                 if (err < 0)
1648                         goto err_free_dev;
1649
1650                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1651                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1652                     device_create_file(&tun->dev->dev, &dev_attr_group))
1653                         pr_err("Failed to create tun sysfs files\n");
1654         }
1655
1656         netif_carrier_on(tun->dev);
1657
1658         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1659
1660         if (ifr->ifr_flags & IFF_NO_PI)
1661                 tun->flags |= TUN_NO_PI;
1662         else
1663                 tun->flags &= ~TUN_NO_PI;
1664
1665         /* This flag has no real effect.  We track the value for backwards
1666          * compatibility.
1667          */
1668         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1669                 tun->flags |= TUN_ONE_QUEUE;
1670         else
1671                 tun->flags &= ~TUN_ONE_QUEUE;
1672
1673         if (ifr->ifr_flags & IFF_VNET_HDR)
1674                 tun->flags |= TUN_VNET_HDR;
1675         else
1676                 tun->flags &= ~TUN_VNET_HDR;
1677
1678         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1679                 tun->flags |= TUN_TAP_MQ;
1680         else
1681                 tun->flags &= ~TUN_TAP_MQ;
1682
1683         /* Make sure persistent devices do not get stuck in
1684          * xoff state.
1685          */
1686         if (netif_running(tun->dev))
1687                 netif_tx_wake_all_queues(tun->dev);
1688
1689         strcpy(ifr->ifr_name, tun->dev->name);
1690         return 0;
1691
1692  err_free_dev:
1693         free_netdev(dev);
1694         return err;
1695 }
1696
1697 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1698                        struct ifreq *ifr)
1699 {
1700         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1701
1702         strcpy(ifr->ifr_name, tun->dev->name);
1703
1704         ifr->ifr_flags = tun_flags(tun);
1705
1706 }
1707
1708 /* This is like a cut-down ethtool ops, except done via tun fd so no
1709  * privs required. */
1710 static int set_offload(struct tun_struct *tun, unsigned long arg)
1711 {
1712         netdev_features_t features = 0;
1713
1714         if (arg & TUN_F_CSUM) {
1715                 features |= NETIF_F_HW_CSUM;
1716                 arg &= ~TUN_F_CSUM;
1717
1718                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1719                         if (arg & TUN_F_TSO_ECN) {
1720                                 features |= NETIF_F_TSO_ECN;
1721                                 arg &= ~TUN_F_TSO_ECN;
1722                         }
1723                         if (arg & TUN_F_TSO4)
1724                                 features |= NETIF_F_TSO;
1725                         if (arg & TUN_F_TSO6)
1726                                 features |= NETIF_F_TSO6;
1727                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1728                 }
1729
1730                 if (arg & TUN_F_UFO) {
1731                         features |= NETIF_F_UFO;
1732                         arg &= ~TUN_F_UFO;
1733                 }
1734         }
1735
1736         /* This gives the user a way to test for new features in future by
1737          * trying to set them. */
1738         if (arg)
1739                 return -EINVAL;
1740
1741         tun->set_features = features;
1742         netdev_update_features(tun->dev);
1743
1744         return 0;
1745 }
1746
1747 static void tun_detach_filter(struct tun_struct *tun, int n)
1748 {
1749         int i;
1750         struct tun_file *tfile;
1751
1752         for (i = 0; i < n; i++) {
1753                 tfile = rtnl_dereference(tun->tfiles[i]);
1754                 sk_detach_filter(tfile->socket.sk);
1755         }
1756
1757         tun->filter_attached = false;
1758 }
1759
1760 static int tun_attach_filter(struct tun_struct *tun)
1761 {
1762         int i, ret = 0;
1763         struct tun_file *tfile;
1764
1765         for (i = 0; i < tun->numqueues; i++) {
1766                 tfile = rtnl_dereference(tun->tfiles[i]);
1767                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1768                 if (ret) {
1769                         tun_detach_filter(tun, i);
1770                         return ret;
1771                 }
1772         }
1773
1774         tun->filter_attached = true;
1775         return ret;
1776 }
1777
1778 static void tun_set_sndbuf(struct tun_struct *tun)
1779 {
1780         struct tun_file *tfile;
1781         int i;
1782
1783         for (i = 0; i < tun->numqueues; i++) {
1784                 tfile = rtnl_dereference(tun->tfiles[i]);
1785                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1786         }
1787 }
1788
1789 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1790 {
1791         struct tun_file *tfile = file->private_data;
1792         struct tun_struct *tun;
1793         int ret = 0;
1794
1795         rtnl_lock();
1796
1797         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1798                 tun = tfile->detached;
1799                 if (!tun) {
1800                         ret = -EINVAL;
1801                         goto unlock;
1802                 }
1803                 ret = security_tun_dev_attach_queue(tun->security);
1804                 if (ret < 0)
1805                         goto unlock;
1806                 ret = tun_attach(tun, file, false);
1807         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1808                 tun = rtnl_dereference(tfile->tun);
1809                 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1810                         ret = -EINVAL;
1811                 else
1812                         __tun_detach(tfile, false);
1813         } else
1814                 ret = -EINVAL;
1815
1816 unlock:
1817         rtnl_unlock();
1818         return ret;
1819 }
1820
1821 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1822                             unsigned long arg, int ifreq_len)
1823 {
1824         struct tun_file *tfile = file->private_data;
1825         struct tun_struct *tun;
1826         void __user* argp = (void __user*)arg;
1827         struct ifreq ifr;
1828         kuid_t owner;
1829         kgid_t group;
1830         int sndbuf;
1831         int vnet_hdr_sz;
1832         unsigned int ifindex;
1833         int ret;
1834
1835         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1836                 if (copy_from_user(&ifr, argp, ifreq_len))
1837                         return -EFAULT;
1838         } else {
1839                 memset(&ifr, 0, sizeof(ifr));
1840         }
1841         if (cmd == TUNGETFEATURES) {
1842                 /* Currently this just means: "what IFF flags are valid?".
1843                  * This is needed because we never checked for invalid flags on
1844                  * TUNSETIFF. */
1845                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1846                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1847                                 (unsigned int __user*)argp);
1848         } else if (cmd == TUNSETQUEUE)
1849                 return tun_set_queue(file, &ifr);
1850
1851         ret = 0;
1852         rtnl_lock();
1853
1854         tun = __tun_get(tfile);
1855         if (cmd == TUNSETIFF && !tun) {
1856                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1857
1858                 ret = tun_set_iff(tfile->net, file, &ifr);
1859
1860                 if (ret)
1861                         goto unlock;
1862
1863                 if (copy_to_user(argp, &ifr, ifreq_len))
1864                         ret = -EFAULT;
1865                 goto unlock;
1866         }
1867         if (cmd == TUNSETIFINDEX) {
1868                 ret = -EPERM;
1869                 if (tun)
1870                         goto unlock;
1871
1872                 ret = -EFAULT;
1873                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1874                         goto unlock;
1875
1876                 ret = 0;
1877                 tfile->ifindex = ifindex;
1878                 goto unlock;
1879         }
1880
1881         ret = -EBADFD;
1882         if (!tun)
1883                 goto unlock;
1884
1885         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1886
1887         ret = 0;
1888         switch (cmd) {
1889         case TUNGETIFF:
1890                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1891
1892                 if (tfile->detached)
1893                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
1894                 if (!tfile->socket.sk->sk_filter)
1895                         ifr.ifr_flags |= IFF_NOFILTER;
1896
1897                 if (copy_to_user(argp, &ifr, ifreq_len))
1898                         ret = -EFAULT;
1899                 break;
1900
1901         case TUNSETNOCSUM:
1902                 /* Disable/Enable checksum */
1903
1904                 /* [unimplemented] */
1905                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1906                           arg ? "disabled" : "enabled");
1907                 break;
1908
1909         case TUNSETPERSIST:
1910                 /* Disable/Enable persist mode. Keep an extra reference to the
1911                  * module to prevent the module being unprobed.
1912                  */
1913                 if (arg && !(tun->flags & TUN_PERSIST)) {
1914                         tun->flags |= TUN_PERSIST;
1915                         __module_get(THIS_MODULE);
1916                 }
1917                 if (!arg && (tun->flags & TUN_PERSIST)) {
1918                         tun->flags &= ~TUN_PERSIST;
1919                         module_put(THIS_MODULE);
1920                 }
1921
1922                 tun_debug(KERN_INFO, tun, "persist %s\n",
1923                           arg ? "enabled" : "disabled");
1924                 break;
1925
1926         case TUNSETOWNER:
1927                 /* Set owner of the device */
1928                 owner = make_kuid(current_user_ns(), arg);
1929                 if (!uid_valid(owner)) {
1930                         ret = -EINVAL;
1931                         break;
1932                 }
1933                 tun->owner = owner;
1934                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1935                           from_kuid(&init_user_ns, tun->owner));
1936                 break;
1937
1938         case TUNSETGROUP:
1939                 /* Set group of the device */
1940                 group = make_kgid(current_user_ns(), arg);
1941                 if (!gid_valid(group)) {
1942                         ret = -EINVAL;
1943                         break;
1944                 }
1945                 tun->group = group;
1946                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1947                           from_kgid(&init_user_ns, tun->group));
1948                 break;
1949
1950         case TUNSETLINK:
1951                 /* Only allow setting the type when the interface is down */
1952                 if (tun->dev->flags & IFF_UP) {
1953                         tun_debug(KERN_INFO, tun,
1954                                   "Linktype set failed because interface is up\n");
1955                         ret = -EBUSY;
1956                 } else {
1957                         tun->dev->type = (int) arg;
1958                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1959                                   tun->dev->type);
1960                         ret = 0;
1961                 }
1962                 break;
1963
1964 #ifdef TUN_DEBUG
1965         case TUNSETDEBUG:
1966                 tun->debug = arg;
1967                 break;
1968 #endif
1969         case TUNSETOFFLOAD:
1970                 ret = set_offload(tun, arg);
1971                 break;
1972
1973         case TUNSETTXFILTER:
1974                 /* Can be set only for TAPs */
1975                 ret = -EINVAL;
1976                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1977                         break;
1978                 ret = update_filter(&tun->txflt, (void __user *)arg);
1979                 break;
1980
1981         case SIOCGIFHWADDR:
1982                 /* Get hw address */
1983                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1984                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1985                 if (copy_to_user(argp, &ifr, ifreq_len))
1986                         ret = -EFAULT;
1987                 break;
1988
1989         case SIOCSIFHWADDR:
1990                 /* Set hw address */
1991                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1992                           ifr.ifr_hwaddr.sa_data);
1993
1994                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1995                 break;
1996
1997         case TUNGETSNDBUF:
1998                 sndbuf = tfile->socket.sk->sk_sndbuf;
1999                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2000                         ret = -EFAULT;
2001                 break;
2002
2003         case TUNSETSNDBUF:
2004                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2005                         ret = -EFAULT;
2006                         break;
2007                 }
2008
2009                 tun->sndbuf = sndbuf;
2010                 tun_set_sndbuf(tun);
2011                 break;
2012
2013         case TUNGETVNETHDRSZ:
2014                 vnet_hdr_sz = tun->vnet_hdr_sz;
2015                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2016                         ret = -EFAULT;
2017                 break;
2018
2019         case TUNSETVNETHDRSZ:
2020                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2021                         ret = -EFAULT;
2022                         break;
2023                 }
2024                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2025                         ret = -EINVAL;
2026                         break;
2027                 }
2028
2029                 tun->vnet_hdr_sz = vnet_hdr_sz;
2030                 break;
2031
2032         case TUNATTACHFILTER:
2033                 /* Can be set only for TAPs */
2034                 ret = -EINVAL;
2035                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2036                         break;
2037                 ret = -EFAULT;
2038                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2039                         break;
2040
2041                 ret = tun_attach_filter(tun);
2042                 break;
2043
2044         case TUNDETACHFILTER:
2045                 /* Can be set only for TAPs */
2046                 ret = -EINVAL;
2047                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2048                         break;
2049                 ret = 0;
2050                 tun_detach_filter(tun, tun->numqueues);
2051                 break;
2052
2053         case TUNGETFILTER:
2054                 ret = -EINVAL;
2055                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2056                         break;
2057                 ret = -EFAULT;
2058                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2059                         break;
2060                 ret = 0;
2061                 break;
2062
2063         default:
2064                 ret = -EINVAL;
2065                 break;
2066         }
2067
2068 unlock:
2069         rtnl_unlock();
2070         if (tun)
2071                 tun_put(tun);
2072         return ret;
2073 }
2074
2075 static long tun_chr_ioctl(struct file *file,
2076                           unsigned int cmd, unsigned long arg)
2077 {
2078         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2079 }
2080
2081 #ifdef CONFIG_COMPAT
2082 static long tun_chr_compat_ioctl(struct file *file,
2083                          unsigned int cmd, unsigned long arg)
2084 {
2085         switch (cmd) {
2086         case TUNSETIFF:
2087         case TUNGETIFF:
2088         case TUNSETTXFILTER:
2089         case TUNGETSNDBUF:
2090         case TUNSETSNDBUF:
2091         case SIOCGIFHWADDR:
2092         case SIOCSIFHWADDR:
2093                 arg = (unsigned long)compat_ptr(arg);
2094                 break;
2095         default:
2096                 arg = (compat_ulong_t)arg;
2097                 break;
2098         }
2099
2100         /*
2101          * compat_ifreq is shorter than ifreq, so we must not access beyond
2102          * the end of that structure. All fields that are used in this
2103          * driver are compatible though, we don't need to convert the
2104          * contents.
2105          */
2106         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2107 }
2108 #endif /* CONFIG_COMPAT */
2109
2110 static int tun_chr_fasync(int fd, struct file *file, int on)
2111 {
2112         struct tun_file *tfile = file->private_data;
2113         int ret;
2114
2115         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2116                 goto out;
2117
2118         if (on) {
2119                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2120                 if (ret)
2121                         goto out;
2122                 tfile->flags |= TUN_FASYNC;
2123         } else
2124                 tfile->flags &= ~TUN_FASYNC;
2125         ret = 0;
2126 out:
2127         return ret;
2128 }
2129
2130 static int tun_chr_open(struct inode *inode, struct file * file)
2131 {
2132         struct tun_file *tfile;
2133
2134         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2135
2136         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2137                                             &tun_proto);
2138         if (!tfile)
2139                 return -ENOMEM;
2140         rcu_assign_pointer(tfile->tun, NULL);
2141         tfile->net = get_net(current->nsproxy->net_ns);
2142         tfile->flags = 0;
2143         tfile->ifindex = 0;
2144
2145         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2146         init_waitqueue_head(&tfile->wq.wait);
2147
2148         tfile->socket.file = file;
2149         tfile->socket.ops = &tun_socket_ops;
2150
2151         sock_init_data(&tfile->socket, &tfile->sk);
2152         sk_change_net(&tfile->sk, tfile->net);
2153
2154         tfile->sk.sk_write_space = tun_sock_write_space;
2155         tfile->sk.sk_sndbuf = INT_MAX;
2156
2157         file->private_data = tfile;
2158         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2159         INIT_LIST_HEAD(&tfile->next);
2160
2161         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2162
2163         return 0;
2164 }
2165
2166 static int tun_chr_close(struct inode *inode, struct file *file)
2167 {
2168         struct tun_file *tfile = file->private_data;
2169         struct net *net = tfile->net;
2170
2171         tun_detach(tfile, true);
2172         put_net(net);
2173
2174         return 0;
2175 }
2176
2177 static const struct file_operations tun_fops = {
2178         .owner  = THIS_MODULE,
2179         .llseek = no_llseek,
2180         .read  = do_sync_read,
2181         .aio_read  = tun_chr_aio_read,
2182         .write = do_sync_write,
2183         .aio_write = tun_chr_aio_write,
2184         .poll   = tun_chr_poll,
2185         .unlocked_ioctl = tun_chr_ioctl,
2186 #ifdef CONFIG_COMPAT
2187         .compat_ioctl = tun_chr_compat_ioctl,
2188 #endif
2189         .open   = tun_chr_open,
2190         .release = tun_chr_close,
2191         .fasync = tun_chr_fasync
2192 };
2193
2194 static struct miscdevice tun_miscdev = {
2195         .minor = TUN_MINOR,
2196         .name = "tun",
2197         .nodename = "net/tun",
2198         .fops = &tun_fops,
2199 };
2200
2201 /* ethtool interface */
2202
2203 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2204 {
2205         cmd->supported          = 0;
2206         cmd->advertising        = 0;
2207         ethtool_cmd_speed_set(cmd, SPEED_10);
2208         cmd->duplex             = DUPLEX_FULL;
2209         cmd->port               = PORT_TP;
2210         cmd->phy_address        = 0;
2211         cmd->transceiver        = XCVR_INTERNAL;
2212         cmd->autoneg            = AUTONEG_DISABLE;
2213         cmd->maxtxpkt           = 0;
2214         cmd->maxrxpkt           = 0;
2215         return 0;
2216 }
2217
2218 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2219 {
2220         struct tun_struct *tun = netdev_priv(dev);
2221
2222         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2223         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2224
2225         switch (tun->flags & TUN_TYPE_MASK) {
2226         case TUN_TUN_DEV:
2227                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2228                 break;
2229         case TUN_TAP_DEV:
2230                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2231                 break;
2232         }
2233 }
2234
2235 static u32 tun_get_msglevel(struct net_device *dev)
2236 {
2237 #ifdef TUN_DEBUG
2238         struct tun_struct *tun = netdev_priv(dev);
2239         return tun->debug;
2240 #else
2241         return -EOPNOTSUPP;
2242 #endif
2243 }
2244
2245 static void tun_set_msglevel(struct net_device *dev, u32 value)
2246 {
2247 #ifdef TUN_DEBUG
2248         struct tun_struct *tun = netdev_priv(dev);
2249         tun->debug = value;
2250 #endif
2251 }
2252
2253 static const struct ethtool_ops tun_ethtool_ops = {
2254         .get_settings   = tun_get_settings,
2255         .get_drvinfo    = tun_get_drvinfo,
2256         .get_msglevel   = tun_get_msglevel,
2257         .set_msglevel   = tun_set_msglevel,
2258         .get_link       = ethtool_op_get_link,
2259         .get_ts_info    = ethtool_op_get_ts_info,
2260 };
2261
2262
2263 static int __init tun_init(void)
2264 {
2265         int ret = 0;
2266
2267         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2268         pr_info("%s\n", DRV_COPYRIGHT);
2269
2270         ret = rtnl_link_register(&tun_link_ops);
2271         if (ret) {
2272                 pr_err("Can't register link_ops\n");
2273                 goto err_linkops;
2274         }
2275
2276         ret = misc_register(&tun_miscdev);
2277         if (ret) {
2278                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2279                 goto err_misc;
2280         }
2281         return  0;
2282 err_misc:
2283         rtnl_link_unregister(&tun_link_ops);
2284 err_linkops:
2285         return ret;
2286 }
2287
2288 static void tun_cleanup(void)
2289 {
2290         misc_deregister(&tun_miscdev);
2291         rtnl_link_unregister(&tun_link_ops);
2292 }
2293
2294 /* Get an underlying socket object from tun file.  Returns error unless file is
2295  * attached to a device.  The returned object works like a packet socket, it
2296  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2297  * holding a reference to the file for as long as the socket is in use. */
2298 struct socket *tun_get_socket(struct file *file)
2299 {
2300         struct tun_file *tfile;
2301         if (file->f_op != &tun_fops)
2302                 return ERR_PTR(-EINVAL);
2303         tfile = file->private_data;
2304         if (!tfile)
2305                 return ERR_PTR(-EBADFD);
2306         return &tfile->socket;
2307 }
2308 EXPORT_SYMBOL_GPL(tun_get_socket);
2309
2310 module_init(tun_init);
2311 module_exit(tun_cleanup);
2312 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2313 MODULE_AUTHOR(DRV_COPYRIGHT);
2314 MODULE_LICENSE("GPL");
2315 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2316 MODULE_ALIAS("devname:net/tun");