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 #include <linux/seq_file.h>
  73
  74 #include <asm/uaccess.h>
  75
  76 /* Uncomment to enable debugging */
  77 /* #define TUN_DEBUG 1 */
  78
  79 #ifdef TUN_DEBUG
  80 static int debug;
  81
  82 #define tun_debug(level, tun, fmt, args...)                     \
  83 do {                                                            \
  84         if (tun->debug)                                         \
  85                 netdev_printk(level, tun->dev, fmt, ##args);    \
  86 } while (0)
  87 #define DBG1(level, fmt, args...)                               \
  88 do {                                                            \
  89         if (debug == 2)                                         \
  90                 printk(level fmt, ##args);                      \
  91 } while (0)
  92 #else
  93 #define tun_debug(level, tun, fmt, args...)                     \
  94 do {                                                            \
  95         if (0)                                                  \
  96                 netdev_printk(level, tun->dev, fmt, ##args);    \
  97 } while (0)
  98 #define DBG1(level, fmt, args...)                               \
  99 do {                                                            \
 100         if (0)                                                  \
 101                 printk(level fmt, ##args);                      \
 102 } while (0)
 103 #endif
 104
 105 #define GOODCOPY_LEN 128
 106
 107 #define FLT_EXACT_COUNT 8
 108 struct tap_filter {
 109         unsigned int    count;    /* Number of addrs. Zero means disabled */
 110         u32             mask[2];  /* Mask of the hashed addrs */
 111         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 112 };
 113
 114 /* DEFAULT_MAX_NUM_RSS_QUEUES were chosen to let the rx/tx queues allocated for
 115  * the netdevice to be fit in one page. So we can make sure the success of
 116  * memory allocation. TODO: increase the limit. */
 117 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
 118 #define MAX_TAP_FLOWS  4096
 119
 120 #define TUN_FLOW_EXPIRE (3 * HZ)
 121
 122 /* A tun_file connects an open character device to a tuntap netdevice. It
 123  * also contains all socket related structures (except sock_fprog and tap_filter)
 124  * to serve as one transmit queue for tuntap device. The sock_fprog and
 125  * tap_filter were kept in tun_struct since they were used for filtering for the
 126  * netdevice not for a specific queue (at least I didn't see the requirement for
 127  * this).
 128  *
 129  * RCU usage:
 130  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 131  * other can only be read while rcu_read_lock or rtnl_lock is held.
 132  */
 133 struct tun_file {
 134         struct sock sk;
 135         struct socket socket;
 136         struct socket_wq wq;
 137         struct tun_struct __rcu *tun;
 138         struct net *net;
 139         struct fasync_struct *fasync;
 140         /* only used for fasnyc */
 141         unsigned int flags;
 142         union {
 143                 u16 queue_index;
 144                 unsigned int ifindex;
 145         };
 146         struct list_head next;
 147         struct tun_struct *detached;
 148 };
 149
 150 struct tun_flow_entry {
 151         struct hlist_node hash_link;
 152         struct rcu_head rcu;
 153         struct tun_struct *tun;
 154
 155         u32 rxhash;
 156         u32 rps_rxhash;
 157         int queue_index;
 158         unsigned long updated;
 159 };
 160
 161 #define TUN_NUM_FLOW_ENTRIES 1024
 162
 163 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 164  * device, socket filter, sndbuf and vnet header size were restore when the
 165  * file were attached to a persist device.
 166  */
 167 struct tun_struct {
 168         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 169         unsigned int            numqueues;
 170         unsigned int            flags;
 171         kuid_t                  owner;
 172         kgid_t                  group;
 173
 174         struct net_device       *dev;
 175         netdev_features_t       set_features;
 176 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 177                           NETIF_F_TSO6)
 178
 179         int                     vnet_hdr_sz;
 180         int                     sndbuf;
 181         struct tap_filter       txflt;
 182         struct sock_fprog       fprog;
 183         /* protected by rtnl lock */
 184         bool                    filter_attached;
 185 #ifdef TUN_DEBUG
 186         int debug;
 187 #endif
 188         spinlock_t lock;
 189         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 190         struct timer_list flow_gc_timer;
 191         unsigned long ageing_time;
 192         unsigned int numdisabled;
 193         struct list_head disabled;
 194         void *security;
 195         u32 flow_count;
 196 };
 197
 198 static inline u32 tun_hashfn(u32 rxhash)
 199 {
 200         return rxhash & 0x3ff;
 201 }
 202
 203 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 204 {
 205         struct tun_flow_entry *e;
 206
 207         hlist_for_each_entry_rcu(e, head, hash_link) {
 208                 if (e->rxhash == rxhash)
 209                         return e;
 210         }
 211         return NULL;
 212 }
 213
 214 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 215                                               struct hlist_head *head,
 216                                               u32 rxhash, u16 queue_index)
 217 {
 218         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 219
 220         if (e) {
 221                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 222                           rxhash, queue_index);
 223                 e->updated = jiffies;
 224                 e->rxhash = rxhash;
 225                 e->rps_rxhash = 0;
 226                 e->queue_index = queue_index;
 227                 e->tun = tun;
 228                 hlist_add_head_rcu(&e->hash_link, head);
 229                 ++tun->flow_count;
 230         }
 231         return e;
 232 }
 233
 234 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 235 {
 236         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 237                   e->rxhash, e->queue_index);
 238         sock_rps_reset_flow_hash(e->rps_rxhash);
 239         hlist_del_rcu(&e->hash_link);
 240         kfree_rcu(e, rcu);
 241         --tun->flow_count;
 242 }
 243
 244 static void tun_flow_flush(struct tun_struct *tun)
 245 {
 246         int i;
 247
 248         spin_lock_bh(&tun->lock);
 249         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 250                 struct tun_flow_entry *e;
 251                 struct hlist_node *n;
 252
 253                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 254                         tun_flow_delete(tun, e);
 255         }
 256         spin_unlock_bh(&tun->lock);
 257 }
 258
 259 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 260 {
 261         int i;
 262
 263         spin_lock_bh(&tun->lock);
 264         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 265                 struct tun_flow_entry *e;
 266                 struct hlist_node *n;
 267
 268                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 269                         if (e->queue_index == queue_index)
 270                                 tun_flow_delete(tun, e);
 271                 }
 272         }
 273         spin_unlock_bh(&tun->lock);
 274 }
 275
 276 static void tun_flow_cleanup(unsigned long data)
 277 {
 278         struct tun_struct *tun = (struct tun_struct *)data;
 279         unsigned long delay = tun->ageing_time;
 280         unsigned long next_timer = jiffies + delay;
 281         unsigned long count = 0;
 282         int i;
 283
 284         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 285
 286         spin_lock_bh(&tun->lock);
 287         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 288                 struct tun_flow_entry *e;
 289                 struct hlist_node *n;
 290
 291                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 292                         unsigned long this_timer;
 293                         count++;
 294                         this_timer = e->updated + delay;
 295                         if (time_before_eq(this_timer, jiffies))
 296                                 tun_flow_delete(tun, e);
 297                         else if (time_before(this_timer, next_timer))
 298                                 next_timer = this_timer;
 299                 }
 300         }
 301
 302         if (count)
 303                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 304         spin_unlock_bh(&tun->lock);
 305 }
 306
 307 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 308                             struct tun_file *tfile)
 309 {
 310         struct hlist_head *head;
 311         struct tun_flow_entry *e;
 312         unsigned long delay = tun->ageing_time;
 313         u16 queue_index = tfile->queue_index;
 314
 315         if (!rxhash)
 316                 return;
 317         else
 318                 head = &tun->flows[tun_hashfn(rxhash)];
 319
 320         rcu_read_lock();
 321
 322         /* We may get a very small possibility of OOO during switching, not
 323          * worth to optimize.*/
 324         if (tun->numqueues == 1 || tfile->detached)
 325                 goto unlock;
 326
 327         e = tun_flow_find(head, rxhash);
 328         if (likely(e)) {
 329                 /* TODO: keep queueing to old queue until it's empty? */
 330                 e->queue_index = queue_index;
 331                 e->updated = jiffies;
 332                 sock_rps_record_flow_hash(e->rps_rxhash);
 333         } else {
 334                 spin_lock_bh(&tun->lock);
 335                 if (!tun_flow_find(head, rxhash) &&
 336                     tun->flow_count < MAX_TAP_FLOWS)
 337                         tun_flow_create(tun, head, rxhash, queue_index);
 338
 339                 if (!timer_pending(&tun->flow_gc_timer))
 340                         mod_timer(&tun->flow_gc_timer,
 341                                   round_jiffies_up(jiffies + delay));
 342                 spin_unlock_bh(&tun->lock);
 343         }
 344
 345 unlock:
 346         rcu_read_unlock();
 347 }
 348
 349 /**
 350  * Save the hash received in the stack receive path and update the
 351  * flow_hash table accordingly.
 352  */
 353 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 354 {
 355         if (unlikely(e->rps_rxhash != hash)) {
 356                 sock_rps_reset_flow_hash(e->rps_rxhash);
 357                 e->rps_rxhash = hash;
 358         }
 359 }
 360
 361 /* We try to identify a flow through its rxhash first. The reason that
 362  * we do not check rxq no. is because some cards(e.g 82599), chooses
 363  * the rxq based on the txq where the last packet of the flow comes. As
 364  * the userspace application move between processors, we may get a
 365  * different rxq no. here. If we could not get rxhash, then we would
 366  * hope the rxq no. may help here.
 367  */
 368 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 369                             void *accel_priv, select_queue_fallback_t fallback)
 370 {
 371         struct tun_struct *tun = netdev_priv(dev);
 372         struct tun_flow_entry *e;
 373         u32 txq = 0;
 374         u32 numqueues = 0;
 375
 376         rcu_read_lock();
 377         numqueues = ACCESS_ONCE(tun->numqueues);
 378
 379         txq = skb_get_hash(skb);
 380         if (txq) {
 381                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 382                 if (e) {
 383                         tun_flow_save_rps_rxhash(e, txq);
 384                         txq = e->queue_index;
 385                 } else
 386                         /* use multiply and shift instead of expensive divide */
 387                         txq = ((u64)txq * numqueues) >> 32;
 388         } else if (likely(skb_rx_queue_recorded(skb))) {
 389                 txq = skb_get_rx_queue(skb);
 390                 while (unlikely(txq >= numqueues))
 391                         txq -= numqueues;
 392         }
 393
 394         rcu_read_unlock();
 395         return txq;
 396 }
 397
 398 static inline bool tun_not_capable(struct tun_struct *tun)
 399 {
 400         const struct cred *cred = current_cred();
 401         struct net *net = dev_net(tun->dev);
 402
 403         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 404                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 405                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 406 }
 407
 408 static void tun_set_real_num_queues(struct tun_struct *tun)
 409 {
 410         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 411         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 412 }
 413
 414 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 415 {
 416         tfile->detached = tun;
 417         list_add_tail(&tfile->next, &tun->disabled);
 418         ++tun->numdisabled;
 419 }
 420
 421 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 422 {
 423         struct tun_struct *tun = tfile->detached;
 424
 425         tfile->detached = NULL;
 426         list_del_init(&tfile->next);
 427         --tun->numdisabled;
 428         return tun;
 429 }
 430
 431 static void tun_queue_purge(struct tun_file *tfile)
 432 {
 433         skb_queue_purge(&tfile->sk.sk_receive_queue);
 434         skb_queue_purge(&tfile->sk.sk_error_queue);
 435 }
 436
 437 static void __tun_detach(struct tun_file *tfile, bool clean)
 438 {
 439         struct tun_file *ntfile;
 440         struct tun_struct *tun;
 441
 442         tun = rtnl_dereference(tfile->tun);
 443
 444         if (tun && !tfile->detached) {
 445                 u16 index = tfile->queue_index;
 446                 BUG_ON(index >= tun->numqueues);
 447
 448                 rcu_assign_pointer(tun->tfiles[index],
 449                                    tun->tfiles[tun->numqueues - 1]);
 450                 ntfile = rtnl_dereference(tun->tfiles[index]);
 451                 ntfile->queue_index = index;
 452
 453                 --tun->numqueues;
 454                 if (clean) {
 455                         RCU_INIT_POINTER(tfile->tun, NULL);
 456                         sock_put(&tfile->sk);
 457                 } else
 458                         tun_disable_queue(tun, tfile);
 459
 460                 synchronize_net();
 461                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 462                 /* Drop read queue */
 463                 tun_queue_purge(tfile);
 464                 tun_set_real_num_queues(tun);
 465         } else if (tfile->detached && clean) {
 466                 tun = tun_enable_queue(tfile);
 467                 sock_put(&tfile->sk);
 468         }
 469
 470         if (clean) {
 471                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 472                         netif_carrier_off(tun->dev);
 473
 474                         if (!(tun->flags & TUN_PERSIST) &&
 475                             tun->dev->reg_state == NETREG_REGISTERED)
 476                                 unregister_netdevice(tun->dev);
 477                 }
 478
 479                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 480                                  &tfile->socket.flags));
 481                 sk_release_kernel(&tfile->sk);
 482         }
 483 }
 484
 485 static void tun_detach(struct tun_file *tfile, bool clean)
 486 {
 487         rtnl_lock();
 488         __tun_detach(tfile, clean);
 489         rtnl_unlock();
 490 }
 491
 492 static void tun_detach_all(struct net_device *dev)
 493 {
 494         struct tun_struct *tun = netdev_priv(dev);
 495         struct tun_file *tfile, *tmp;
 496         int i, n = tun->numqueues;
 497
 498         for (i = 0; i < n; i++) {
 499                 tfile = rtnl_dereference(tun->tfiles[i]);
 500                 BUG_ON(!tfile);
 501                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 502                 RCU_INIT_POINTER(tfile->tun, NULL);
 503                 --tun->numqueues;
 504         }
 505         list_for_each_entry(tfile, &tun->disabled, next) {
 506                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 507                 RCU_INIT_POINTER(tfile->tun, NULL);
 508         }
 509         BUG_ON(tun->numqueues != 0);
 510
 511         synchronize_net();
 512         for (i = 0; i < n; i++) {
 513                 tfile = rtnl_dereference(tun->tfiles[i]);
 514                 /* Drop read queue */
 515                 tun_queue_purge(tfile);
 516                 sock_put(&tfile->sk);
 517         }
 518         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 519                 tun_enable_queue(tfile);
 520                 tun_queue_purge(tfile);
 521                 sock_put(&tfile->sk);
 522         }
 523         BUG_ON(tun->numdisabled != 0);
 524
 525         if (tun->flags & TUN_PERSIST)
 526                 module_put(THIS_MODULE);
 527 }
 528
 529 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 530 {
 531         struct tun_file *tfile = file->private_data;
 532         int err;
 533
 534         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 535         if (err < 0)
 536                 goto out;
 537
 538         err = -EINVAL;
 539         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 540                 goto out;
 541
 542         err = -EBUSY;
 543         if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 544                 goto out;
 545
 546         err = -E2BIG;
 547         if (!tfile->detached &&
 548             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 549                 goto out;
 550
 551         err = 0;
 552
 553         /* Re-attach the filter to persist device */
 554         if (!skip_filter && (tun->filter_attached == true)) {
 555                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 556                 if (!err)
 557                         goto out;
 558         }
 559         tfile->queue_index = tun->numqueues;
 560         rcu_assign_pointer(tfile->tun, tun);
 561         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 562         tun->numqueues++;
 563
 564         if (tfile->detached)
 565                 tun_enable_queue(tfile);
 566         else
 567                 sock_hold(&tfile->sk);
 568
 569         tun_set_real_num_queues(tun);
 570
 571         /* device is allowed to go away first, so no need to hold extra
 572          * refcnt.
 573          */
 574
 575 out:
 576         return err;
 577 }
 578
 579 static struct tun_struct *__tun_get(struct tun_file *tfile)
 580 {
 581         struct tun_struct *tun;
 582
 583         rcu_read_lock();
 584         tun = rcu_dereference(tfile->tun);
 585         if (tun)
 586                 dev_hold(tun->dev);
 587         rcu_read_unlock();
 588
 589         return tun;
 590 }
 591
 592 static struct tun_struct *tun_get(struct file *file)
 593 {
 594         return __tun_get(file->private_data);
 595 }
 596
 597 static void tun_put(struct tun_struct *tun)
 598 {
 599         dev_put(tun->dev);
 600 }
 601
 602 /* TAP filtering */
 603 static void addr_hash_set(u32 *mask, const u8 *addr)
 604 {
 605         int n = ether_crc(ETH_ALEN, addr) >> 26;
 606         mask[n >> 5] |= (1 << (n & 31));
 607 }
 608
 609 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 610 {
 611         int n = ether_crc(ETH_ALEN, addr) >> 26;
 612         return mask[n >> 5] & (1 << (n & 31));
 613 }
 614
 615 static int update_filter(struct tap_filter *filter, void __user *arg)
 616 {
 617         struct { u8 u[ETH_ALEN]; } *addr;
 618         struct tun_filter uf;
 619         int err, alen, n, nexact;
 620
 621         if (copy_from_user(&uf, arg, sizeof(uf)))
 622                 return -EFAULT;
 623
 624         if (!uf.count) {
 625                 /* Disabled */
 626                 filter->count = 0;
 627                 return 0;
 628         }
 629
 630         alen = ETH_ALEN * uf.count;
 631         addr = kmalloc(alen, GFP_KERNEL);
 632         if (!addr)
 633                 return -ENOMEM;
 634
 635         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 636                 err = -EFAULT;
 637                 goto done;
 638         }
 639
 640         /* The filter is updated without holding any locks. Which is
 641          * perfectly safe. We disable it first and in the worst
 642          * case we'll accept a few undesired packets. */
 643         filter->count = 0;
 644         wmb();
 645
 646         /* Use first set of addresses as an exact filter */
 647         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 648                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 649
 650         nexact = n;
 651
 652         /* Remaining multicast addresses are hashed,
 653          * unicast will leave the filter disabled. */
 654         memset(filter->mask, 0, sizeof(filter->mask));
 655         for (; n < uf.count; n++) {
 656                 if (!is_multicast_ether_addr(addr[n].u)) {
 657                         err = 0; /* no filter */
 658                         goto done;
 659                 }
 660                 addr_hash_set(filter->mask, addr[n].u);
 661         }
 662
 663         /* For ALLMULTI just set the mask to all ones.
 664          * This overrides the mask populated above. */
 665         if ((uf.flags & TUN_FLT_ALLMULTI))
 666                 memset(filter->mask, ~0, sizeof(filter->mask));
 667
 668         /* Now enable the filter */
 669         wmb();
 670         filter->count = nexact;
 671
 672         /* Return the number of exact filters */
 673         err = nexact;
 674
 675 done:
 676         kfree(addr);
 677         return err;
 678 }
 679
 680 /* Returns: 0 - drop, !=0 - accept */
 681 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 682 {
 683         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 684          * at this point. */
 685         struct ethhdr *eh = (struct ethhdr *) skb->data;
 686         int i;
 687
 688         /* Exact match */
 689         for (i = 0; i < filter->count; i++)
 690                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 691                         return 1;
 692
 693         /* Inexact match (multicast only) */
 694         if (is_multicast_ether_addr(eh->h_dest))
 695                 return addr_hash_test(filter->mask, eh->h_dest);
 696
 697         return 0;
 698 }
 699
 700 /*
 701  * Checks whether the packet is accepted or not.
 702  * Returns: 0 - drop, !=0 - accept
 703  */
 704 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 705 {
 706         if (!filter->count)
 707                 return 1;
 708
 709         return run_filter(filter, skb);
 710 }
 711
 712 /* Network device part of the driver */
 713
 714 static const struct ethtool_ops tun_ethtool_ops;
 715
 716 /* Net device detach from fd. */
 717 static void tun_net_uninit(struct net_device *dev)
 718 {
 719         tun_detach_all(dev);
 720 }
 721
 722 /* Net device open. */
 723 static int tun_net_open(struct net_device *dev)
 724 {
 725         netif_tx_start_all_queues(dev);
 726         return 0;
 727 }
 728
 729 /* Net device close. */
 730 static int tun_net_close(struct net_device *dev)
 731 {
 732         netif_tx_stop_all_queues(dev);
 733         return 0;
 734 }
 735
 736 /* Net device start xmit */
 737 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 738 {
 739         struct tun_struct *tun = netdev_priv(dev);
 740         int txq = skb->queue_mapping;
 741         struct tun_file *tfile;
 742         u32 numqueues = 0;
 743
 744         rcu_read_lock();
 745         tfile = rcu_dereference(tun->tfiles[txq]);
 746         numqueues = ACCESS_ONCE(tun->numqueues);
 747
 748         /* Drop packet if interface is not attached */
 749         if (txq >= numqueues)
 750                 goto drop;
 751
 752         if (numqueues == 1) {
 753                 /* Select queue was not called for the skbuff, so we extract the
 754                  * RPS hash and save it into the flow_table here.
 755                  */
 756                 __u32 rxhash;
 757
 758                 rxhash = skb_get_hash(skb);
 759                 if (rxhash) {
 760                         struct tun_flow_entry *e;
 761                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 762                                         rxhash);
 763                         if (e)
 764                                 tun_flow_save_rps_rxhash(e, rxhash);
 765                 }
 766         }
 767
 768         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 769
 770         BUG_ON(!tfile);
 771
 772         /* Drop if the filter does not like it.
 773          * This is a noop if the filter is disabled.
 774          * Filter can be enabled only for the TAP devices. */
 775         if (!check_filter(&tun->txflt, skb))
 776                 goto drop;
 777
 778         if (tfile->socket.sk->sk_filter &&
 779             sk_filter(tfile->socket.sk, skb))
 780                 goto drop;
 781
 782         /* Limit the number of packets queued by dividing txq length with the
 783          * number of queues.
 784          */
 785         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
 786                           >= dev->tx_queue_len)
 787                 goto drop;
 788
 789         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 790                 goto drop;
 791
 792         if (skb->sk) {
 793                 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
 794                 sw_tx_timestamp(skb);
 795         }
 796
 797         /* Orphan the skb - required as we might hang on to it
 798          * for indefinite time.
 799          */
 800         skb_orphan(skb);
 801
 802         nf_reset(skb);
 803
 804         /* Enqueue packet */
 805         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 806
 807         /* Notify and wake up reader process */
 808         if (tfile->flags & TUN_FASYNC)
 809                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 810         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 811
 812         rcu_read_unlock();
 813         return NETDEV_TX_OK;
 814
 815 drop:
 816         dev->stats.tx_dropped++;
 817         skb_tx_error(skb);
 818         kfree_skb(skb);
 819         rcu_read_unlock();
 820         return NETDEV_TX_OK;
 821 }
 822
 823 static void tun_net_mclist(struct net_device *dev)
 824 {
 825         /*
 826          * This callback is supposed to deal with mc filter in
 827          * _rx_ path and has nothing to do with the _tx_ path.
 828          * In rx path we always accept everything userspace gives us.
 829          */
 830 }
 831
 832 #define MIN_MTU 68
 833 #define MAX_MTU 65535
 834
 835 static int
 836 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 837 {
 838         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 839                 return -EINVAL;
 840         dev->mtu = new_mtu;
 841         return 0;
 842 }
 843
 844 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 845         netdev_features_t features)
 846 {
 847         struct tun_struct *tun = netdev_priv(dev);
 848
 849         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 850 }
 851 #ifdef CONFIG_NET_POLL_CONTROLLER
 852 static void tun_poll_controller(struct net_device *dev)
 853 {
 854         /*
 855          * Tun only receives frames when:
 856          * 1) the char device endpoint gets data from user space
 857          * 2) the tun socket gets a sendmsg call from user space
 858          * Since both of those are synchronous operations, we are guaranteed
 859          * never to have pending data when we poll for it
 860          * so there is nothing to do here but return.
 861          * We need this though so netpoll recognizes us as an interface that
 862          * supports polling, which enables bridge devices in virt setups to
 863          * still use netconsole
 864          */
 865         return;
 866 }
 867 #endif
 868 static const struct net_device_ops tun_netdev_ops = {
 869         .ndo_uninit             = tun_net_uninit,
 870         .ndo_open               = tun_net_open,
 871         .ndo_stop               = tun_net_close,
 872         .ndo_start_xmit         = tun_net_xmit,
 873         .ndo_change_mtu         = tun_net_change_mtu,
 874         .ndo_fix_features       = tun_net_fix_features,
 875         .ndo_select_queue       = tun_select_queue,
 876 #ifdef CONFIG_NET_POLL_CONTROLLER
 877         .ndo_poll_controller    = tun_poll_controller,
 878 #endif
 879 };
 880
 881 static const struct net_device_ops tap_netdev_ops = {
 882         .ndo_uninit             = tun_net_uninit,
 883         .ndo_open               = tun_net_open,
 884         .ndo_stop               = tun_net_close,
 885         .ndo_start_xmit         = tun_net_xmit,
 886         .ndo_change_mtu         = tun_net_change_mtu,
 887         .ndo_fix_features       = tun_net_fix_features,
 888         .ndo_set_rx_mode        = tun_net_mclist,
 889         .ndo_set_mac_address    = eth_mac_addr,
 890         .ndo_validate_addr      = eth_validate_addr,
 891         .ndo_select_queue       = tun_select_queue,
 892 #ifdef CONFIG_NET_POLL_CONTROLLER
 893         .ndo_poll_controller    = tun_poll_controller,
 894 #endif
 895 };
 896
 897 static void tun_flow_init(struct tun_struct *tun)
 898 {
 899         int i;
 900
 901         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 902                 INIT_HLIST_HEAD(&tun->flows[i]);
 903
 904         tun->ageing_time = TUN_FLOW_EXPIRE;
 905         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 906         mod_timer(&tun->flow_gc_timer,
 907                   round_jiffies_up(jiffies + tun->ageing_time));
 908 }
 909
 910 static void tun_flow_uninit(struct tun_struct *tun)
 911 {
 912         del_timer_sync(&tun->flow_gc_timer);
 913         tun_flow_flush(tun);
 914 }
 915
 916 /* Initialize net device. */
 917 static void tun_net_init(struct net_device *dev)
 918 {
 919         struct tun_struct *tun = netdev_priv(dev);
 920
 921         switch (tun->flags & TUN_TYPE_MASK) {
 922         case TUN_TUN_DEV:
 923                 dev->netdev_ops = &tun_netdev_ops;
 924
 925                 /* Point-to-Point TUN Device */
 926                 dev->hard_header_len = 0;
 927                 dev->addr_len = 0;
 928                 dev->mtu = 1500;
 929
 930                 /* Zero header length */
 931                 dev->type = ARPHRD_NONE;
 932                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 933                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 934                 break;
 935
 936         case TUN_TAP_DEV:
 937                 dev->netdev_ops = &tap_netdev_ops;
 938                 /* Ethernet TAP Device */
 939                 ether_setup(dev);
 940                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 941                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 942
 943                 eth_hw_addr_random(dev);
 944
 945                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 946                 break;
 947         }
 948 }
 949
 950 /* Character device part */
 951
 952 /* Poll */
 953 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 954 {
 955         struct tun_file *tfile = file->private_data;
 956         struct tun_struct *tun = __tun_get(tfile);
 957         struct sock *sk;
 958         unsigned int mask = 0;
 959
 960         if (!tun)
 961                 return POLLERR;
 962
 963         sk = tfile->socket.sk;
 964
 965         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 966
 967         poll_wait(file, sk_sleep(sk), wait);
 968
 969         if (!skb_queue_empty(&sk->sk_receive_queue))
 970                 mask |= POLLIN | POLLRDNORM;
 971
 972         if (sock_writeable(sk) ||
 973             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 974              sock_writeable(sk)))
 975                 mask |= POLLOUT | POLLWRNORM;
 976
 977         if (tun->dev->reg_state != NETREG_REGISTERED)
 978                 mask = POLLERR;
 979
 980         tun_put(tun);
 981         return mask;
 982 }
 983
 984 /* prepad is the amount to reserve at front.  len is length after that.
 985  * linear is a hint as to how much to copy (usually headers). */
 986 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 987                                      size_t prepad, size_t len,
 988                                      size_t linear, int noblock)
 989 {
 990         struct sock *sk = tfile->socket.sk;
 991         struct sk_buff *skb;
 992         int err;
 993
 994         /* Under a page?  Don't bother with paged skb. */
 995         if (prepad + len < PAGE_SIZE || !linear)
 996                 linear = len;
 997
 998         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 999                                    &err, 0);
1000         if (!skb)
1001                 return ERR_PTR(err);
1002
1003         skb_reserve(skb, prepad);
1004         skb_put(skb, linear);
1005         skb->data_len = len - linear;
1006         skb->len += len - linear;
1007
1008         return skb;
1009 }
1010
1011 /* Get packet from user space buffer */
1012 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1013                             void *msg_control, const struct iovec *iv,
1014                             size_t total_len, size_t count, int noblock)
1015 {
1016         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1017         struct sk_buff *skb;
1018         size_t len = total_len, align = NET_SKB_PAD, linear;
1019         struct virtio_net_hdr gso = { 0 };
1020         int good_linear;
1021         int offset = 0;
1022         int copylen;
1023         bool zerocopy = false;
1024         int err;
1025         u32 rxhash;
1026
1027         if (!(tun->flags & TUN_NO_PI)) {
1028                 if (len < sizeof(pi))
1029                         return -EINVAL;
1030                 len -= sizeof(pi);
1031
1032                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1033                         return -EFAULT;
1034                 offset += sizeof(pi);
1035         }
1036
1037         if (tun->flags & TUN_VNET_HDR) {
1038                 if (len < tun->vnet_hdr_sz)
1039                         return -EINVAL;
1040                 len -= tun->vnet_hdr_sz;
1041
1042                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1043                         return -EFAULT;
1044
1045                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1046                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1047                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1048
1049                 if (gso.hdr_len > len)
1050                         return -EINVAL;
1051                 offset += tun->vnet_hdr_sz;
1052         }
1053
1054         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1055                 align += NET_IP_ALIGN;
1056                 if (unlikely(len < ETH_HLEN ||
1057                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1058                         return -EINVAL;
1059         }
1060
1061         good_linear = SKB_MAX_HEAD(align);
1062
1063         if (msg_control) {
1064                 /* There are 256 bytes to be copied in skb, so there is
1065                  * enough room for skb expand head in case it is used.
1066                  * The rest of the buffer is mapped from userspace.
1067                  */
1068                 copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1069                 if (copylen > good_linear)
1070                         copylen = good_linear;
1071                 linear = copylen;
1072                 if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1073                         zerocopy = true;
1074         }
1075
1076         if (!zerocopy) {
1077                 copylen = len;
1078                 if (gso.hdr_len > good_linear)
1079                         linear = good_linear;
1080                 else
1081                         linear = gso.hdr_len;
1082         }
1083
1084         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1085         if (IS_ERR(skb)) {
1086                 if (PTR_ERR(skb) != -EAGAIN)
1087                         tun->dev->stats.rx_dropped++;
1088                 return PTR_ERR(skb);
1089         }
1090
1091         if (zerocopy)
1092                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1093         else {
1094                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1095                 if (!err && msg_control) {
1096                         struct ubuf_info *uarg = msg_control;
1097                         uarg->callback(uarg, false);
1098                 }
1099         }
1100
1101         if (err) {
1102                 tun->dev->stats.rx_dropped++;
1103                 kfree_skb(skb);
1104                 return -EFAULT;
1105         }
1106
1107         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1108                 if (!skb_partial_csum_set(skb, gso.csum_start,
1109                                           gso.csum_offset)) {
1110                         tun->dev->stats.rx_frame_errors++;
1111                         kfree_skb(skb);
1112                         return -EINVAL;
1113                 }
1114         }
1115
1116         switch (tun->flags & TUN_TYPE_MASK) {
1117         case TUN_TUN_DEV:
1118                 if (tun->flags & TUN_NO_PI) {
1119                         switch (skb->data[0] & 0xf0) {
1120                         case 0x40:
1121                                 pi.proto = htons(ETH_P_IP);
1122                                 break;
1123                         case 0x60:
1124                                 pi.proto = htons(ETH_P_IPV6);
1125                                 break;
1126                         default:
1127                                 tun->dev->stats.rx_dropped++;
1128                                 kfree_skb(skb);
1129                                 return -EINVAL;
1130                         }
1131                 }
1132
1133                 skb_reset_mac_header(skb);
1134                 skb->protocol = pi.proto;
1135                 skb->dev = tun->dev;
1136                 break;
1137         case TUN_TAP_DEV:
1138                 skb->protocol = eth_type_trans(skb, tun->dev);
1139                 break;
1140         }
1141
1142         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1143                 pr_debug("GSO!\n");
1144                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1145                 case VIRTIO_NET_HDR_GSO_TCPV4:
1146                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1147                         break;
1148                 case VIRTIO_NET_HDR_GSO_TCPV6:
1149                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1150                         break;
1151                 case VIRTIO_NET_HDR_GSO_UDP:
1152                 {
1153                         static bool warned;
1154
1155                         if (!warned) {
1156                                 warned = true;
1157                                 netdev_warn(tun->dev,
1158                                             "%s: using disabled UFO feature; please fix this program\n",
1159                                             current->comm);
1160                         }
1161                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1162                         break;
1163                 }
1164                 default:
1165                         tun->dev->stats.rx_frame_errors++;
1166                         kfree_skb(skb);
1167                         return -EINVAL;
1168                 }
1169
1170                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1171                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1172
1173                 skb_shinfo(skb)->gso_size = gso.gso_size;
1174                 if (skb_shinfo(skb)->gso_size == 0) {
1175                         tun->dev->stats.rx_frame_errors++;
1176                         kfree_skb(skb);
1177                         return -EINVAL;
1178                 }
1179
1180                 /* Header must be checked, and gso_segs computed. */
1181                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1182                 skb_shinfo(skb)->gso_segs = 0;
1183         }
1184
1185         /* copy skb_ubuf_info for callback when skb has no error */
1186         if (zerocopy) {
1187                 skb_shinfo(skb)->destructor_arg = msg_control;
1188                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1189                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1190         }
1191
1192         skb_reset_network_header(skb);
1193         skb_probe_transport_header(skb, 0);
1194
1195         rxhash = skb_get_hash(skb);
1196         netif_rx_ni(skb);
1197
1198         tun->dev->stats.rx_packets++;
1199         tun->dev->stats.rx_bytes += len;
1200
1201         tun_flow_update(tun, rxhash, tfile);
1202         return total_len;
1203 }
1204
1205 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1206                               unsigned long count, loff_t pos)
1207 {
1208         struct file *file = iocb->ki_filp;
1209         struct tun_struct *tun = tun_get(file);
1210         struct tun_file *tfile = file->private_data;
1211         ssize_t result;
1212
1213         if (!tun)
1214                 return -EBADFD;
1215
1216         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1217
1218         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1219                               count, file->f_flags & O_NONBLOCK);
1220
1221         tun_put(tun);
1222         return result;
1223 }
1224
1225 /* Put packet to the user space buffer */
1226 static ssize_t tun_put_user(struct tun_struct *tun,
1227                             struct tun_file *tfile,
1228                             struct sk_buff *skb,
1229                             const struct iovec *iv, int len)
1230 {
1231         struct tun_pi pi = { 0, skb->protocol };
1232         ssize_t total = 0;
1233         int vlan_offset = 0, copied;
1234
1235         if (!(tun->flags & TUN_NO_PI)) {
1236                 if ((len -= sizeof(pi)) < 0)
1237                         return -EINVAL;
1238
1239                 if (len < skb->len) {
1240                         /* Packet will be striped */
1241                         pi.flags |= TUN_PKT_STRIP;
1242                 }
1243
1244                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1245                         return -EFAULT;
1246                 total += sizeof(pi);
1247         }
1248
1249         if (tun->flags & TUN_VNET_HDR) {
1250                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1251                 if ((len -= tun->vnet_hdr_sz) < 0)
1252                         return -EINVAL;
1253
1254                 if (skb_is_gso(skb)) {
1255                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1256
1257                         /* This is a hint as to how much should be linear. */
1258                         gso.hdr_len = skb_headlen(skb);
1259                         gso.gso_size = sinfo->gso_size;
1260                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1261                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1262                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1263                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1264                         else {
1265                                 pr_err("unexpected GSO type: "
1266                                        "0x%x, gso_size %d, hdr_len %d\n",
1267                                        sinfo->gso_type, gso.gso_size,
1268                                        gso.hdr_len);
1269                                 print_hex_dump(KERN_ERR, "tun: ",
1270                                                DUMP_PREFIX_NONE,
1271                                                16, 1, skb->head,
1272                                                min((int)gso.hdr_len, 64), true);
1273                                 WARN_ON_ONCE(1);
1274                                 return -EINVAL;
1275                         }
1276                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1277                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1278                 } else
1279                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1280
1281                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1282                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1283                         gso.csum_start = skb_checksum_start_offset(skb);
1284                         gso.csum_offset = skb->csum_offset;
1285                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1286                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1287                 } /* else everything is zero */
1288
1289                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1290                                                sizeof(gso))))
1291                         return -EFAULT;
1292                 total += tun->vnet_hdr_sz;
1293         }
1294
1295         copied = total;
1296         total += skb->len;
1297         if (!vlan_tx_tag_present(skb)) {
1298                 len = min_t(int, skb->len, len);
1299         } else {
1300                 int copy, ret;
1301                 struct {
1302                         __be16 h_vlan_proto;
1303                         __be16 h_vlan_TCI;
1304                 } veth;
1305
1306                 veth.h_vlan_proto = skb->vlan_proto;
1307                 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1308
1309                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1310                 len = min_t(int, skb->len + VLAN_HLEN, len);
1311                 total += VLAN_HLEN;
1312
1313                 copy = min_t(int, vlan_offset, len);
1314                 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
1315                 len -= copy;
1316                 copied += copy;
1317                 if (ret || !len)
1318                         goto done;
1319
1320                 copy = min_t(int, sizeof(veth), len);
1321                 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
1322                 len -= copy;
1323                 copied += copy;
1324                 if (ret || !len)
1325                         goto done;
1326         }
1327
1328         skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
1329
1330 done:
1331         tun->dev->stats.tx_packets++;
1332         tun->dev->stats.tx_bytes += len;
1333
1334         return total;
1335 }
1336
1337 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1338                            const struct iovec *iv, ssize_t len, int noblock)
1339 {
1340         struct sk_buff *skb;
1341         ssize_t ret = 0;
1342         int peeked, err, off = 0;
1343
1344         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1345
1346         if (!len)
1347                 return ret;
1348
1349         if (tun->dev->reg_state != NETREG_REGISTERED)
1350                 return -EIO;
1351
1352         /* Read frames from queue */
1353         skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1354                                   &peeked, &off, &err);
1355         if (skb) {
1356                 ret = tun_put_user(tun, tfile, skb, iv, len);
1357                 kfree_skb(skb);
1358         } else
1359                 ret = err;
1360
1361         return ret;
1362 }
1363
1364 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1365                             unsigned long count, loff_t pos)
1366 {
1367         struct file *file = iocb->ki_filp;
1368         struct tun_file *tfile = file->private_data;
1369         struct tun_struct *tun = __tun_get(tfile);
1370         ssize_t len, ret;
1371
1372         if (!tun)
1373                 return -EBADFD;
1374         len = iov_length(iv, count);
1375         if (len < 0) {
1376                 ret = -EINVAL;
1377                 goto out;
1378         }
1379
1380         ret = tun_do_read(tun, tfile, iv, len,
1381                           file->f_flags & O_NONBLOCK);
1382         ret = min_t(ssize_t, ret, len);
1383         if (ret > 0)
1384                 iocb->ki_pos = ret;
1385 out:
1386         tun_put(tun);
1387         return ret;
1388 }
1389
1390 static void tun_free_netdev(struct net_device *dev)
1391 {
1392         struct tun_struct *tun = netdev_priv(dev);
1393
1394         BUG_ON(!(list_empty(&tun->disabled)));
1395         tun_flow_uninit(tun);
1396         security_tun_dev_free_security(tun->security);
1397         free_netdev(dev);
1398 }
1399
1400 static void tun_setup(struct net_device *dev)
1401 {
1402         struct tun_struct *tun = netdev_priv(dev);
1403
1404         tun->owner = INVALID_UID;
1405         tun->group = INVALID_GID;
1406
1407         dev->ethtool_ops = &tun_ethtool_ops;
1408         dev->destructor = tun_free_netdev;
1409 }
1410
1411 /* Trivial set of netlink ops to allow deleting tun or tap
1412  * device with netlink.
1413  */
1414 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1415 {
1416         return -EINVAL;
1417 }
1418
1419 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1420         .kind           = DRV_NAME,
1421         .priv_size      = sizeof(struct tun_struct),
1422         .setup          = tun_setup,
1423         .validate       = tun_validate,
1424 };
1425
1426 static void tun_sock_write_space(struct sock *sk)
1427 {
1428         struct tun_file *tfile;
1429         wait_queue_head_t *wqueue;
1430
1431         if (!sock_writeable(sk))
1432                 return;
1433
1434         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1435                 return;
1436
1437         wqueue = sk_sleep(sk);
1438         if (wqueue && waitqueue_active(wqueue))
1439                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1440                                                 POLLWRNORM | POLLWRBAND);
1441
1442         tfile = container_of(sk, struct tun_file, sk);
1443         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1444 }
1445
1446 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1447                        struct msghdr *m, size_t total_len)
1448 {
1449         int ret;
1450         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1451         struct tun_struct *tun = __tun_get(tfile);
1452
1453         if (!tun)
1454                 return -EBADFD;
1455         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1456                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1457         tun_put(tun);
1458         return ret;
1459 }
1460
1461 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1462                        struct msghdr *m, size_t total_len,
1463                        int flags)
1464 {
1465         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1466         struct tun_struct *tun = __tun_get(tfile);
1467         int ret;
1468
1469         if (!tun)
1470                 return -EBADFD;
1471
1472         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1473                 ret = -EINVAL;
1474                 goto out;
1475         }
1476         if (flags & MSG_ERRQUEUE) {
1477                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1478                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1479                 goto out;
1480         }
1481         ret = tun_do_read(tun, tfile, m->msg_iov, total_len,
1482                           flags & MSG_DONTWAIT);
1483         if (ret > total_len) {
1484                 m->msg_flags |= MSG_TRUNC;
1485                 ret = flags & MSG_TRUNC ? ret : total_len;
1486         }
1487 out:
1488         tun_put(tun);
1489         return ret;
1490 }
1491
1492 static int tun_release(struct socket *sock)
1493 {
1494         if (sock->sk)
1495                 sock_put(sock->sk);
1496         return 0;
1497 }
1498
1499 /* Ops structure to mimic raw sockets with tun */
1500 static const struct proto_ops tun_socket_ops = {
1501         .sendmsg = tun_sendmsg,
1502         .recvmsg = tun_recvmsg,
1503         .release = tun_release,
1504 };
1505
1506 static struct proto tun_proto = {
1507         .name           = "tun",
1508         .owner          = THIS_MODULE,
1509         .obj_size       = sizeof(struct tun_file),
1510 };
1511
1512 static int tun_flags(struct tun_struct *tun)
1513 {
1514         int flags = 0;
1515
1516         if (tun->flags & TUN_TUN_DEV)
1517                 flags |= IFF_TUN;
1518         else
1519                 flags |= IFF_TAP;
1520
1521         if (tun->flags & TUN_NO_PI)
1522                 flags |= IFF_NO_PI;
1523
1524         /* This flag has no real effect.  We track the value for backwards
1525          * compatibility.
1526          */
1527         if (tun->flags & TUN_ONE_QUEUE)
1528                 flags |= IFF_ONE_QUEUE;
1529
1530         if (tun->flags & TUN_VNET_HDR)
1531                 flags |= IFF_VNET_HDR;
1532
1533         if (tun->flags & TUN_TAP_MQ)
1534                 flags |= IFF_MULTI_QUEUE;
1535
1536         if (tun->flags & TUN_PERSIST)
1537                 flags |= IFF_PERSIST;
1538
1539         return flags;
1540 }
1541
1542 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1543                               char *buf)
1544 {
1545         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1546         return sprintf(buf, "0x%x\n", tun_flags(tun));
1547 }
1548
1549 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1550                               char *buf)
1551 {
1552         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1553         return uid_valid(tun->owner)?
1554                 sprintf(buf, "%u\n",
1555                         from_kuid_munged(current_user_ns(), tun->owner)):
1556                 sprintf(buf, "-1\n");
1557 }
1558
1559 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1560                               char *buf)
1561 {
1562         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1563         return gid_valid(tun->group) ?
1564                 sprintf(buf, "%u\n",
1565                         from_kgid_munged(current_user_ns(), tun->group)):
1566                 sprintf(buf, "-1\n");
1567 }
1568
1569 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1570 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1571 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1572
1573 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1574 {
1575         struct tun_struct *tun;
1576         struct tun_file *tfile = file->private_data;
1577         struct net_device *dev;
1578         int err;
1579
1580         if (tfile->detached)
1581                 return -EINVAL;
1582
1583         dev = __dev_get_by_name(net, ifr->ifr_name);
1584         if (dev) {
1585                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1586                         return -EBUSY;
1587                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1588                         tun = netdev_priv(dev);
1589                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1590                         tun = netdev_priv(dev);
1591                 else
1592                         return -EINVAL;
1593
1594                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1595                     !!(tun->flags & TUN_TAP_MQ))
1596                         return -EINVAL;
1597
1598                 if (tun_not_capable(tun))
1599                         return -EPERM;
1600                 err = security_tun_dev_open(tun->security);
1601                 if (err < 0)
1602                         return err;
1603
1604                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1605                 if (err < 0)
1606                         return err;
1607
1608                 if (tun->flags & TUN_TAP_MQ &&
1609                     (tun->numqueues + tun->numdisabled > 1)) {
1610                         /* One or more queue has already been attached, no need
1611                          * to initialize the device again.
1612                          */
1613                         return 0;
1614                 }
1615         }
1616         else {
1617                 char *name;
1618                 unsigned long flags = 0;
1619                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1620                              MAX_TAP_QUEUES : 1;
1621
1622                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1623                         return -EPERM;
1624                 err = security_tun_dev_create();
1625                 if (err < 0)
1626                         return err;
1627
1628                 /* Set dev type */
1629                 if (ifr->ifr_flags & IFF_TUN) {
1630                         /* TUN device */
1631                         flags |= TUN_TUN_DEV;
1632                         name = "tun%d";
1633                 } else if (ifr->ifr_flags & IFF_TAP) {
1634                         /* TAP device */
1635                         flags |= TUN_TAP_DEV;
1636                         name = "tap%d";
1637                 } else
1638                         return -EINVAL;
1639
1640                 if (*ifr->ifr_name)
1641                         name = ifr->ifr_name;
1642
1643                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1644                                        NET_NAME_UNKNOWN, tun_setup, queues,
1645                                        queues);
1646
1647                 if (!dev)
1648                         return -ENOMEM;
1649
1650                 dev_net_set(dev, net);
1651                 dev->rtnl_link_ops = &tun_link_ops;
1652                 dev->ifindex = tfile->ifindex;
1653
1654                 tun = netdev_priv(dev);
1655                 tun->dev = dev;
1656                 tun->flags = flags;
1657                 tun->txflt.count = 0;
1658                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1659
1660                 tun->filter_attached = false;
1661                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1662
1663                 spin_lock_init(&tun->lock);
1664
1665                 err = security_tun_dev_alloc_security(&tun->security);
1666                 if (err < 0)
1667                         goto err_free_dev;
1668
1669                 tun_net_init(dev);
1670                 tun_flow_init(tun);
1671
1672                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1673                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1674                                    NETIF_F_HW_VLAN_STAG_TX;
1675                 dev->features = dev->hw_features;
1676                 dev->vlan_features = dev->features &
1677                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1678                                        NETIF_F_HW_VLAN_STAG_TX);
1679
1680                 INIT_LIST_HEAD(&tun->disabled);
1681                 err = tun_attach(tun, file, false);
1682                 if (err < 0)
1683                         goto err_free_flow;
1684
1685                 err = register_netdevice(tun->dev);
1686                 if (err < 0)
1687                         goto err_detach;
1688
1689                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1690                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1691                     device_create_file(&tun->dev->dev, &dev_attr_group))
1692                         pr_err("Failed to create tun sysfs files\n");
1693         }
1694
1695         netif_carrier_on(tun->dev);
1696
1697         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1698
1699         if (ifr->ifr_flags & IFF_NO_PI)
1700                 tun->flags |= TUN_NO_PI;
1701         else
1702                 tun->flags &= ~TUN_NO_PI;
1703
1704         /* This flag has no real effect.  We track the value for backwards
1705          * compatibility.
1706          */
1707         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1708                 tun->flags |= TUN_ONE_QUEUE;
1709         else
1710                 tun->flags &= ~TUN_ONE_QUEUE;
1711
1712         if (ifr->ifr_flags & IFF_VNET_HDR)
1713                 tun->flags |= TUN_VNET_HDR;
1714         else
1715                 tun->flags &= ~TUN_VNET_HDR;
1716
1717         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1718                 tun->flags |= TUN_TAP_MQ;
1719         else
1720                 tun->flags &= ~TUN_TAP_MQ;
1721
1722         /* Make sure persistent devices do not get stuck in
1723          * xoff state.
1724          */
1725         if (netif_running(tun->dev))
1726                 netif_tx_wake_all_queues(tun->dev);
1727
1728         strcpy(ifr->ifr_name, tun->dev->name);
1729         return 0;
1730
1731 err_detach:
1732         tun_detach_all(dev);
1733 err_free_flow:
1734         tun_flow_uninit(tun);
1735         security_tun_dev_free_security(tun->security);
1736 err_free_dev:
1737         free_netdev(dev);
1738         return err;
1739 }
1740
1741 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1742                        struct ifreq *ifr)
1743 {
1744         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1745
1746         strcpy(ifr->ifr_name, tun->dev->name);
1747
1748         ifr->ifr_flags = tun_flags(tun);
1749
1750 }
1751
1752 /* This is like a cut-down ethtool ops, except done via tun fd so no
1753  * privs required. */
1754 static int set_offload(struct tun_struct *tun, unsigned long arg)
1755 {
1756         netdev_features_t features = 0;
1757
1758         if (arg & TUN_F_CSUM) {
1759                 features |= NETIF_F_HW_CSUM;
1760                 arg &= ~TUN_F_CSUM;
1761
1762                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1763                         if (arg & TUN_F_TSO_ECN) {
1764                                 features |= NETIF_F_TSO_ECN;
1765                                 arg &= ~TUN_F_TSO_ECN;
1766                         }
1767                         if (arg & TUN_F_TSO4)
1768                                 features |= NETIF_F_TSO;
1769                         if (arg & TUN_F_TSO6)
1770                                 features |= NETIF_F_TSO6;
1771                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1772                 }
1773         }
1774
1775         /* This gives the user a way to test for new features in future by
1776          * trying to set them. */
1777         if (arg)
1778                 return -EINVAL;
1779
1780         tun->set_features = features;
1781         netdev_update_features(tun->dev);
1782
1783         return 0;
1784 }
1785
1786 static void tun_detach_filter(struct tun_struct *tun, int n)
1787 {
1788         int i;
1789         struct tun_file *tfile;
1790
1791         for (i = 0; i < n; i++) {
1792                 tfile = rtnl_dereference(tun->tfiles[i]);
1793                 sk_detach_filter(tfile->socket.sk);
1794         }
1795
1796         tun->filter_attached = false;
1797 }
1798
1799 static int tun_attach_filter(struct tun_struct *tun)
1800 {
1801         int i, ret = 0;
1802         struct tun_file *tfile;
1803
1804         for (i = 0; i < tun->numqueues; i++) {
1805                 tfile = rtnl_dereference(tun->tfiles[i]);
1806                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1807                 if (ret) {
1808                         tun_detach_filter(tun, i);
1809                         return ret;
1810                 }
1811         }
1812
1813         tun->filter_attached = true;
1814         return ret;
1815 }
1816
1817 static void tun_set_sndbuf(struct tun_struct *tun)
1818 {
1819         struct tun_file *tfile;
1820         int i;
1821
1822         for (i = 0; i < tun->numqueues; i++) {
1823                 tfile = rtnl_dereference(tun->tfiles[i]);
1824                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1825         }
1826 }
1827
1828 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1829 {
1830         struct tun_file *tfile = file->private_data;
1831         struct tun_struct *tun;
1832         int ret = 0;
1833
1834         rtnl_lock();
1835
1836         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1837                 tun = tfile->detached;
1838                 if (!tun) {
1839                         ret = -EINVAL;
1840                         goto unlock;
1841                 }
1842                 ret = security_tun_dev_attach_queue(tun->security);
1843                 if (ret < 0)
1844                         goto unlock;
1845                 ret = tun_attach(tun, file, false);
1846         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1847                 tun = rtnl_dereference(tfile->tun);
1848                 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1849                         ret = -EINVAL;
1850                 else
1851                         __tun_detach(tfile, false);
1852         } else
1853                 ret = -EINVAL;
1854
1855 unlock:
1856         rtnl_unlock();
1857         return ret;
1858 }
1859
1860 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1861                             unsigned long arg, int ifreq_len)
1862 {
1863         struct tun_file *tfile = file->private_data;
1864         struct tun_struct *tun;
1865         void __user* argp = (void __user*)arg;
1866         struct ifreq ifr;
1867         kuid_t owner;
1868         kgid_t group;
1869         int sndbuf;
1870         int vnet_hdr_sz;
1871         unsigned int ifindex;
1872         int ret;
1873
1874         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1875                 if (copy_from_user(&ifr, argp, ifreq_len))
1876                         return -EFAULT;
1877         } else {
1878                 memset(&ifr, 0, sizeof(ifr));
1879         }
1880         if (cmd == TUNGETFEATURES) {
1881                 /* Currently this just means: "what IFF flags are valid?".
1882                  * This is needed because we never checked for invalid flags on
1883                  * TUNSETIFF. */
1884                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1885                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1886                                 (unsigned int __user*)argp);
1887         } else if (cmd == TUNSETQUEUE)
1888                 return tun_set_queue(file, &ifr);
1889
1890         ret = 0;
1891         rtnl_lock();
1892
1893         tun = __tun_get(tfile);
1894         if (cmd == TUNSETIFF && !tun) {
1895                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1896
1897                 ret = tun_set_iff(tfile->net, file, &ifr);
1898
1899                 if (ret)
1900                         goto unlock;
1901
1902                 if (copy_to_user(argp, &ifr, ifreq_len))
1903                         ret = -EFAULT;
1904                 goto unlock;
1905         }
1906         if (cmd == TUNSETIFINDEX) {
1907                 ret = -EPERM;
1908                 if (tun)
1909                         goto unlock;
1910
1911                 ret = -EFAULT;
1912                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1913                         goto unlock;
1914
1915                 ret = 0;
1916                 tfile->ifindex = ifindex;
1917                 goto unlock;
1918         }
1919
1920         ret = -EBADFD;
1921         if (!tun)
1922                 goto unlock;
1923
1924         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1925
1926         ret = 0;
1927         switch (cmd) {
1928         case TUNGETIFF:
1929                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1930
1931                 if (tfile->detached)
1932                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
1933                 if (!tfile->socket.sk->sk_filter)
1934                         ifr.ifr_flags |= IFF_NOFILTER;
1935
1936                 if (copy_to_user(argp, &ifr, ifreq_len))
1937                         ret = -EFAULT;
1938                 break;
1939
1940         case TUNSETNOCSUM:
1941                 /* Disable/Enable checksum */
1942
1943                 /* [unimplemented] */
1944                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1945                           arg ? "disabled" : "enabled");
1946                 break;
1947
1948         case TUNSETPERSIST:
1949                 /* Disable/Enable persist mode. Keep an extra reference to the
1950                  * module to prevent the module being unprobed.
1951                  */
1952                 if (arg && !(tun->flags & TUN_PERSIST)) {
1953                         tun->flags |= TUN_PERSIST;
1954                         __module_get(THIS_MODULE);
1955                 }
1956                 if (!arg && (tun->flags & TUN_PERSIST)) {
1957                         tun->flags &= ~TUN_PERSIST;
1958                         module_put(THIS_MODULE);
1959                 }
1960
1961                 tun_debug(KERN_INFO, tun, "persist %s\n",
1962                           arg ? "enabled" : "disabled");
1963                 break;
1964
1965         case TUNSETOWNER:
1966                 /* Set owner of the device */
1967                 owner = make_kuid(current_user_ns(), arg);
1968                 if (!uid_valid(owner)) {
1969                         ret = -EINVAL;
1970                         break;
1971                 }
1972                 tun->owner = owner;
1973                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1974                           from_kuid(&init_user_ns, tun->owner));
1975                 break;
1976
1977         case TUNSETGROUP:
1978                 /* Set group of the device */
1979                 group = make_kgid(current_user_ns(), arg);
1980                 if (!gid_valid(group)) {
1981                         ret = -EINVAL;
1982                         break;
1983                 }
1984                 tun->group = group;
1985                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1986                           from_kgid(&init_user_ns, tun->group));
1987                 break;
1988
1989         case TUNSETLINK:
1990                 /* Only allow setting the type when the interface is down */
1991                 if (tun->dev->flags & IFF_UP) {
1992                         tun_debug(KERN_INFO, tun,
1993                                   "Linktype set failed because interface is up\n");
1994                         ret = -EBUSY;
1995                 } else {
1996                         tun->dev->type = (int) arg;
1997                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1998                                   tun->dev->type);
1999                         ret = 0;
2000                 }
2001                 break;
2002
2003 #ifdef TUN_DEBUG
2004         case TUNSETDEBUG:
2005                 tun->debug = arg;
2006                 break;
2007 #endif
2008         case TUNSETOFFLOAD:
2009                 ret = set_offload(tun, arg);
2010                 break;
2011
2012         case TUNSETTXFILTER:
2013                 /* Can be set only for TAPs */
2014                 ret = -EINVAL;
2015                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2016                         break;
2017                 ret = update_filter(&tun->txflt, (void __user *)arg);
2018                 break;
2019
2020         case SIOCGIFHWADDR:
2021                 /* Get hw address */
2022                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2023                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2024                 if (copy_to_user(argp, &ifr, ifreq_len))
2025                         ret = -EFAULT;
2026                 break;
2027
2028         case SIOCSIFHWADDR:
2029                 /* Set hw address */
2030                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2031                           ifr.ifr_hwaddr.sa_data);
2032
2033                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2034                 break;
2035
2036         case TUNGETSNDBUF:
2037                 sndbuf = tfile->socket.sk->sk_sndbuf;
2038                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2039                         ret = -EFAULT;
2040                 break;
2041
2042         case TUNSETSNDBUF:
2043                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2044                         ret = -EFAULT;
2045                         break;
2046                 }
2047
2048                 tun->sndbuf = sndbuf;
2049                 tun_set_sndbuf(tun);
2050                 break;
2051
2052         case TUNGETVNETHDRSZ:
2053                 vnet_hdr_sz = tun->vnet_hdr_sz;
2054                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2055                         ret = -EFAULT;
2056                 break;
2057
2058         case TUNSETVNETHDRSZ:
2059                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2060                         ret = -EFAULT;
2061                         break;
2062                 }
2063                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2064                         ret = -EINVAL;
2065                         break;
2066                 }
2067
2068                 tun->vnet_hdr_sz = vnet_hdr_sz;
2069                 break;
2070
2071         case TUNATTACHFILTER:
2072                 /* Can be set only for TAPs */
2073                 ret = -EINVAL;
2074                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2075                         break;
2076                 ret = -EFAULT;
2077                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2078                         break;
2079
2080                 ret = tun_attach_filter(tun);
2081                 break;
2082
2083         case TUNDETACHFILTER:
2084                 /* Can be set only for TAPs */
2085                 ret = -EINVAL;
2086                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2087                         break;
2088                 ret = 0;
2089                 tun_detach_filter(tun, tun->numqueues);
2090                 break;
2091
2092         case TUNGETFILTER:
2093                 ret = -EINVAL;
2094                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2095                         break;
2096                 ret = -EFAULT;
2097                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2098                         break;
2099                 ret = 0;
2100                 break;
2101
2102         default:
2103                 ret = -EINVAL;
2104                 break;
2105         }
2106
2107 unlock:
2108         rtnl_unlock();
2109         if (tun)
2110                 tun_put(tun);
2111         return ret;
2112 }
2113
2114 static long tun_chr_ioctl(struct file *file,
2115                           unsigned int cmd, unsigned long arg)
2116 {
2117         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2118 }
2119
2120 #ifdef CONFIG_COMPAT
2121 static long tun_chr_compat_ioctl(struct file *file,
2122                          unsigned int cmd, unsigned long arg)
2123 {
2124         switch (cmd) {
2125         case TUNSETIFF:
2126         case TUNGETIFF:
2127         case TUNSETTXFILTER:
2128         case TUNGETSNDBUF:
2129         case TUNSETSNDBUF:
2130         case SIOCGIFHWADDR:
2131         case SIOCSIFHWADDR:
2132                 arg = (unsigned long)compat_ptr(arg);
2133                 break;
2134         default:
2135                 arg = (compat_ulong_t)arg;
2136                 break;
2137         }
2138
2139         /*
2140          * compat_ifreq is shorter than ifreq, so we must not access beyond
2141          * the end of that structure. All fields that are used in this
2142          * driver are compatible though, we don't need to convert the
2143          * contents.
2144          */
2145         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2146 }
2147 #endif /* CONFIG_COMPAT */
2148
2149 static int tun_chr_fasync(int fd, struct file *file, int on)
2150 {
2151         struct tun_file *tfile = file->private_data;
2152         int ret;
2153
2154         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2155                 goto out;
2156
2157         if (on) {
2158                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2159                 tfile->flags |= TUN_FASYNC;
2160         } else
2161                 tfile->flags &= ~TUN_FASYNC;
2162         ret = 0;
2163 out:
2164         return ret;
2165 }
2166
2167 static int tun_chr_open(struct inode *inode, struct file * file)
2168 {
2169         struct tun_file *tfile;
2170
2171         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2172
2173         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2174                                             &tun_proto);
2175         if (!tfile)
2176                 return -ENOMEM;
2177         RCU_INIT_POINTER(tfile->tun, NULL);
2178         tfile->net = get_net(current->nsproxy->net_ns);
2179         tfile->flags = 0;
2180         tfile->ifindex = 0;
2181
2182         init_waitqueue_head(&tfile->wq.wait);
2183         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2184
2185         tfile->socket.file = file;
2186         tfile->socket.ops = &tun_socket_ops;
2187
2188         sock_init_data(&tfile->socket, &tfile->sk);
2189         sk_change_net(&tfile->sk, tfile->net);
2190
2191         tfile->sk.sk_write_space = tun_sock_write_space;
2192         tfile->sk.sk_sndbuf = INT_MAX;
2193
2194         file->private_data = tfile;
2195         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2196         INIT_LIST_HEAD(&tfile->next);
2197
2198         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2199
2200         return 0;
2201 }
2202
2203 static int tun_chr_close(struct inode *inode, struct file *file)
2204 {
2205         struct tun_file *tfile = file->private_data;
2206         struct net *net = tfile->net;
2207
2208         tun_detach(tfile, true);
2209         put_net(net);
2210
2211         return 0;
2212 }
2213
2214 #ifdef CONFIG_PROC_FS
2215 static int tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2216 {
2217         struct tun_struct *tun;
2218         struct ifreq ifr;
2219
2220         memset(&ifr, 0, sizeof(ifr));
2221
2222         rtnl_lock();
2223         tun = tun_get(f);
2224         if (tun)
2225                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2226         rtnl_unlock();
2227
2228         if (tun)
2229                 tun_put(tun);
2230
2231         return seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2232 }
2233 #endif
2234
2235 static const struct file_operations tun_fops = {
2236         .owner  = THIS_MODULE,
2237         .llseek = no_llseek,
2238         .read  = do_sync_read,
2239         .aio_read  = tun_chr_aio_read,
2240         .write = do_sync_write,
2241         .aio_write = tun_chr_aio_write,
2242         .poll   = tun_chr_poll,
2243         .unlocked_ioctl = tun_chr_ioctl,
2244 #ifdef CONFIG_COMPAT
2245         .compat_ioctl = tun_chr_compat_ioctl,
2246 #endif
2247         .open   = tun_chr_open,
2248         .release = tun_chr_close,
2249         .fasync = tun_chr_fasync,
2250 #ifdef CONFIG_PROC_FS
2251         .show_fdinfo = tun_chr_show_fdinfo,
2252 #endif
2253 };
2254
2255 static struct miscdevice tun_miscdev = {
2256         .minor = TUN_MINOR,
2257         .name = "tun",
2258         .nodename = "net/tun",
2259         .fops = &tun_fops,
2260 };
2261
2262 /* ethtool interface */
2263
2264 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2265 {
2266         cmd->supported          = 0;
2267         cmd->advertising        = 0;
2268         ethtool_cmd_speed_set(cmd, SPEED_10);
2269         cmd->duplex             = DUPLEX_FULL;
2270         cmd->port               = PORT_TP;
2271         cmd->phy_address        = 0;
2272         cmd->transceiver        = XCVR_INTERNAL;
2273         cmd->autoneg            = AUTONEG_DISABLE;
2274         cmd->maxtxpkt           = 0;
2275         cmd->maxrxpkt           = 0;
2276         return 0;
2277 }
2278
2279 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2280 {
2281         struct tun_struct *tun = netdev_priv(dev);
2282
2283         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2284         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2285
2286         switch (tun->flags & TUN_TYPE_MASK) {
2287         case TUN_TUN_DEV:
2288                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2289                 break;
2290         case TUN_TAP_DEV:
2291                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2292                 break;
2293         }
2294 }
2295
2296 static u32 tun_get_msglevel(struct net_device *dev)
2297 {
2298 #ifdef TUN_DEBUG
2299         struct tun_struct *tun = netdev_priv(dev);
2300         return tun->debug;
2301 #else
2302         return -EOPNOTSUPP;
2303 #endif
2304 }
2305
2306 static void tun_set_msglevel(struct net_device *dev, u32 value)
2307 {
2308 #ifdef TUN_DEBUG
2309         struct tun_struct *tun = netdev_priv(dev);
2310         tun->debug = value;
2311 #endif
2312 }
2313
2314 static const struct ethtool_ops tun_ethtool_ops = {
2315         .get_settings   = tun_get_settings,
2316         .get_drvinfo    = tun_get_drvinfo,
2317         .get_msglevel   = tun_get_msglevel,
2318         .set_msglevel   = tun_set_msglevel,
2319         .get_link       = ethtool_op_get_link,
2320         .get_ts_info    = ethtool_op_get_ts_info,
2321 };
2322
2323
2324 static int __init tun_init(void)
2325 {
2326         int ret = 0;
2327
2328         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2329         pr_info("%s\n", DRV_COPYRIGHT);
2330
2331         ret = rtnl_link_register(&tun_link_ops);
2332         if (ret) {
2333                 pr_err("Can't register link_ops\n");
2334                 goto err_linkops;
2335         }
2336
2337         ret = misc_register(&tun_miscdev);
2338         if (ret) {
2339                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2340                 goto err_misc;
2341         }
2342         return  0;
2343 err_misc:
2344         rtnl_link_unregister(&tun_link_ops);
2345 err_linkops:
2346         return ret;
2347 }
2348
2349 static void tun_cleanup(void)
2350 {
2351         misc_deregister(&tun_miscdev);
2352         rtnl_link_unregister(&tun_link_ops);
2353 }
2354
2355 /* Get an underlying socket object from tun file.  Returns error unless file is
2356  * attached to a device.  The returned object works like a packet socket, it
2357  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2358  * holding a reference to the file for as long as the socket is in use. */
2359 struct socket *tun_get_socket(struct file *file)
2360 {
2361         struct tun_file *tfile;
2362         if (file->f_op != &tun_fops)
2363                 return ERR_PTR(-EINVAL);
2364         tfile = file->private_data;
2365         if (!tfile)
2366                 return ERR_PTR(-EBADFD);
2367         return &tfile->socket;
2368 }
2369 EXPORT_SYMBOL_GPL(tun_get_socket);
2370
2371 module_init(tun_init);
2372 module_exit(tun_cleanup);
2373 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2374 MODULE_AUTHOR(DRV_COPYRIGHT);
2375 MODULE_LICENSE("GPL");
2376 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2377 MODULE_ALIAS("devname:net/tun");