drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <[email protected]>
   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 <[email protected]> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <[email protected]>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <[email protected]>  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 <[email protected]>
  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 <[email protected]>"
  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/ipv6.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <linux/seq_file.h>
  74 #include <linux/uio.h>
  75
  76 #include <asm/uaccess.h>
  77
  78 /* Uncomment to enable debugging */
  79 /* #define TUN_DEBUG 1 */
  80
  81 #ifdef TUN_DEBUG
  82 static int debug;
  83
  84 #define tun_debug(level, tun, fmt, args...)                     \
  85 do {                                                            \
  86         if (tun->debug)                                         \
  87                 netdev_printk(level, tun->dev, fmt, ##args);    \
  88 } while (0)
  89 #define DBG1(level, fmt, args...)                               \
  90 do {                                                            \
  91         if (debug == 2)                                         \
  92                 printk(level fmt, ##args);                      \
  93 } while (0)
  94 #else
  95 #define tun_debug(level, tun, fmt, args...)                     \
  96 do {                                                            \
  97         if (0)                                                  \
  98                 netdev_printk(level, tun->dev, fmt, ##args);    \
  99 } while (0)
 100 #define DBG1(level, fmt, args...)                               \
 101 do {                                                            \
 102         if (0)                                                  \
 103                 printk(level fmt, ##args);                      \
 104 } while (0)
 105 #endif
 106
 107 #define GOODCOPY_LEN 128
 108
 109 #define FLT_EXACT_COUNT 8
 110 struct tap_filter {
 111         unsigned int    count;    /* Number of addrs. Zero means disabled */
 112         u32             mask[2];  /* Mask of the hashed addrs */
 113         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 114 };
 115
 116 /* DEFAULT_MAX_NUM_RSS_QUEUES were chosen to let the rx/tx queues allocated for
 117  * the netdevice to be fit in one page. So we can make sure the success of
 118  * memory allocation. TODO: increase the limit. */
 119 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
 120 #define MAX_TAP_FLOWS  4096
 121
 122 #define TUN_FLOW_EXPIRE (3 * HZ)
 123
 124 /* A tun_file connects an open character device to a tuntap netdevice. It
 125  * also contains all socket related structures (except sock_fprog and tap_filter)
 126  * to serve as one transmit queue for tuntap device. The sock_fprog and
 127  * tap_filter were kept in tun_struct since they were used for filtering for the
 128  * netdevice not for a specific queue (at least I didn't see the requirement for
 129  * this).
 130  *
 131  * RCU usage:
 132  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 133  * other can only be read while rcu_read_lock or rtnl_lock is held.
 134  */
 135 struct tun_file {
 136         struct sock sk;
 137         struct socket socket;
 138         struct socket_wq wq;
 139         struct tun_struct __rcu *tun;
 140         struct net *net;
 141         struct fasync_struct *fasync;
 142         /* only used for fasnyc */
 143         unsigned int flags;
 144         union {
 145                 u16 queue_index;
 146                 unsigned int ifindex;
 147         };
 148         struct list_head next;
 149         struct tun_struct *detached;
 150 };
 151
 152 struct tun_flow_entry {
 153         struct hlist_node hash_link;
 154         struct rcu_head rcu;
 155         struct tun_struct *tun;
 156
 157         u32 rxhash;
 158         u32 rps_rxhash;
 159         int queue_index;
 160         unsigned long updated;
 161 };
 162
 163 #define TUN_NUM_FLOW_ENTRIES 1024
 164
 165 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 166  * device, socket filter, sndbuf and vnet header size were restore when the
 167  * file were attached to a persist device.
 168  */
 169 struct tun_struct {
 170         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 171         unsigned int            numqueues;
 172         unsigned int            flags;
 173         kuid_t                  owner;
 174         kgid_t                  group;
 175
 176         struct net_device       *dev;
 177         netdev_features_t       set_features;
 178 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 179                           NETIF_F_TSO6)
 180
 181         int                     vnet_hdr_sz;
 182         int                     sndbuf;
 183         struct tap_filter       txflt;
 184         struct sock_fprog       fprog;
 185         /* protected by rtnl lock */
 186         bool                    filter_attached;
 187 #ifdef TUN_DEBUG
 188         int debug;
 189 #endif
 190         spinlock_t lock;
 191         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 192         struct timer_list flow_gc_timer;
 193         unsigned long ageing_time;
 194         unsigned int numdisabled;
 195         struct list_head disabled;
 196         void *security;
 197         u32 flow_count;
 198 };
 199
 200 static inline u32 tun_hashfn(u32 rxhash)
 201 {
 202         return rxhash & 0x3ff;
 203 }
 204
 205 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 206 {
 207         struct tun_flow_entry *e;
 208
 209         hlist_for_each_entry_rcu(e, head, hash_link) {
 210                 if (e->rxhash == rxhash)
 211                         return e;
 212         }
 213         return NULL;
 214 }
 215
 216 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 217                                               struct hlist_head *head,
 218                                               u32 rxhash, u16 queue_index)
 219 {
 220         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 221
 222         if (e) {
 223                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 224                           rxhash, queue_index);
 225                 e->updated = jiffies;
 226                 e->rxhash = rxhash;
 227                 e->rps_rxhash = 0;
 228                 e->queue_index = queue_index;
 229                 e->tun = tun;
 230                 hlist_add_head_rcu(&e->hash_link, head);
 231                 ++tun->flow_count;
 232         }
 233         return e;
 234 }
 235
 236 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 237 {
 238         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 239                   e->rxhash, e->queue_index);
 240         sock_rps_reset_flow_hash(e->rps_rxhash);
 241         hlist_del_rcu(&e->hash_link);
 242         kfree_rcu(e, rcu);
 243         --tun->flow_count;
 244 }
 245
 246 static void tun_flow_flush(struct tun_struct *tun)
 247 {
 248         int i;
 249
 250         spin_lock_bh(&tun->lock);
 251         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 252                 struct tun_flow_entry *e;
 253                 struct hlist_node *n;
 254
 255                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 256                         tun_flow_delete(tun, e);
 257         }
 258         spin_unlock_bh(&tun->lock);
 259 }
 260
 261 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 262 {
 263         int i;
 264
 265         spin_lock_bh(&tun->lock);
 266         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 267                 struct tun_flow_entry *e;
 268                 struct hlist_node *n;
 269
 270                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 271                         if (e->queue_index == queue_index)
 272                                 tun_flow_delete(tun, e);
 273                 }
 274         }
 275         spin_unlock_bh(&tun->lock);
 276 }
 277
 278 static void tun_flow_cleanup(unsigned long data)
 279 {
 280         struct tun_struct *tun = (struct tun_struct *)data;
 281         unsigned long delay = tun->ageing_time;
 282         unsigned long next_timer = jiffies + delay;
 283         unsigned long count = 0;
 284         int i;
 285
 286         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 287
 288         spin_lock_bh(&tun->lock);
 289         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 290                 struct tun_flow_entry *e;
 291                 struct hlist_node *n;
 292
 293                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 294                         unsigned long this_timer;
 295                         count++;
 296                         this_timer = e->updated + delay;
 297                         if (time_before_eq(this_timer, jiffies))
 298                                 tun_flow_delete(tun, e);
 299                         else if (time_before(this_timer, next_timer))
 300                                 next_timer = this_timer;
 301                 }
 302         }
 303
 304         if (count)
 305                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 306         spin_unlock_bh(&tun->lock);
 307 }
 308
 309 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 310                             struct tun_file *tfile)
 311 {
 312         struct hlist_head *head;
 313         struct tun_flow_entry *e;
 314         unsigned long delay = tun->ageing_time;
 315         u16 queue_index = tfile->queue_index;
 316
 317         if (!rxhash)
 318                 return;
 319         else
 320                 head = &tun->flows[tun_hashfn(rxhash)];
 321
 322         rcu_read_lock();
 323
 324         /* We may get a very small possibility of OOO during switching, not
 325          * worth to optimize.*/
 326         if (tun->numqueues == 1 || tfile->detached)
 327                 goto unlock;
 328
 329         e = tun_flow_find(head, rxhash);
 330         if (likely(e)) {
 331                 /* TODO: keep queueing to old queue until it's empty? */
 332                 e->queue_index = queue_index;
 333                 e->updated = jiffies;
 334                 sock_rps_record_flow_hash(e->rps_rxhash);
 335         } else {
 336                 spin_lock_bh(&tun->lock);
 337                 if (!tun_flow_find(head, rxhash) &&
 338                     tun->flow_count < MAX_TAP_FLOWS)
 339                         tun_flow_create(tun, head, rxhash, queue_index);
 340
 341                 if (!timer_pending(&tun->flow_gc_timer))
 342                         mod_timer(&tun->flow_gc_timer,
 343                                   round_jiffies_up(jiffies + delay));
 344                 spin_unlock_bh(&tun->lock);
 345         }
 346
 347 unlock:
 348         rcu_read_unlock();
 349 }
 350
 351 /**
 352  * Save the hash received in the stack receive path and update the
 353  * flow_hash table accordingly.
 354  */
 355 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 356 {
 357         if (unlikely(e->rps_rxhash != hash)) {
 358                 sock_rps_reset_flow_hash(e->rps_rxhash);
 359                 e->rps_rxhash = hash;
 360         }
 361 }
 362
 363 /* We try to identify a flow through its rxhash first. The reason that
 364  * we do not check rxq no. is because some cards(e.g 82599), chooses
 365  * the rxq based on the txq where the last packet of the flow comes. As
 366  * the userspace application move between processors, we may get a
 367  * different rxq no. here. If we could not get rxhash, then we would
 368  * hope the rxq no. may help here.
 369  */
 370 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 371                             void *accel_priv, select_queue_fallback_t fallback)
 372 {
 373         struct tun_struct *tun = netdev_priv(dev);
 374         struct tun_flow_entry *e;
 375         u32 txq = 0;
 376         u32 numqueues = 0;
 377
 378         rcu_read_lock();
 379         numqueues = ACCESS_ONCE(tun->numqueues);
 380
 381         txq = skb_get_hash(skb);
 382         if (txq) {
 383                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 384                 if (e) {
 385                         tun_flow_save_rps_rxhash(e, txq);
 386                         txq = e->queue_index;
 387                 } else
 388                         /* use multiply and shift instead of expensive divide */
 389                         txq = ((u64)txq * numqueues) >> 32;
 390         } else if (likely(skb_rx_queue_recorded(skb))) {
 391                 txq = skb_get_rx_queue(skb);
 392                 while (unlikely(txq >= numqueues))
 393                         txq -= numqueues;
 394         }
 395
 396         rcu_read_unlock();
 397         return txq;
 398 }
 399
 400 static inline bool tun_not_capable(struct tun_struct *tun)
 401 {
 402         const struct cred *cred = current_cred();
 403         struct net *net = dev_net(tun->dev);
 404
 405         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 406                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 407                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 408 }
 409
 410 static void tun_set_real_num_queues(struct tun_struct *tun)
 411 {
 412         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 413         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 414 }
 415
 416 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 417 {
 418         tfile->detached = tun;
 419         list_add_tail(&tfile->next, &tun->disabled);
 420         ++tun->numdisabled;
 421 }
 422
 423 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 424 {
 425         struct tun_struct *tun = tfile->detached;
 426
 427         tfile->detached = NULL;
 428         list_del_init(&tfile->next);
 429         --tun->numdisabled;
 430         return tun;
 431 }
 432
 433 static void tun_queue_purge(struct tun_file *tfile)
 434 {
 435         skb_queue_purge(&tfile->sk.sk_receive_queue);
 436         skb_queue_purge(&tfile->sk.sk_error_queue);
 437 }
 438
 439 static void __tun_detach(struct tun_file *tfile, bool clean)
 440 {
 441         struct tun_file *ntfile;
 442         struct tun_struct *tun;
 443
 444         tun = rtnl_dereference(tfile->tun);
 445
 446         if (tun && !tfile->detached) {
 447                 u16 index = tfile->queue_index;
 448                 BUG_ON(index >= tun->numqueues);
 449
 450                 rcu_assign_pointer(tun->tfiles[index],
 451                                    tun->tfiles[tun->numqueues - 1]);
 452                 ntfile = rtnl_dereference(tun->tfiles[index]);
 453                 ntfile->queue_index = index;
 454
 455                 --tun->numqueues;
 456                 if (clean) {
 457                         RCU_INIT_POINTER(tfile->tun, NULL);
 458                         sock_put(&tfile->sk);
 459                 } else
 460                         tun_disable_queue(tun, tfile);
 461
 462                 synchronize_net();
 463                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 464                 /* Drop read queue */
 465                 tun_queue_purge(tfile);
 466                 tun_set_real_num_queues(tun);
 467         } else if (tfile->detached && clean) {
 468                 tun = tun_enable_queue(tfile);
 469                 sock_put(&tfile->sk);
 470         }
 471
 472         if (clean) {
 473                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 474                         netif_carrier_off(tun->dev);
 475
 476                         if (!(tun->flags & TUN_PERSIST) &&
 477                             tun->dev->reg_state == NETREG_REGISTERED)
 478                                 unregister_netdevice(tun->dev);
 479                 }
 480
 481                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 482                                  &tfile->socket.flags));
 483                 sk_release_kernel(&tfile->sk);
 484         }
 485 }
 486
 487 static void tun_detach(struct tun_file *tfile, bool clean)
 488 {
 489         rtnl_lock();
 490         __tun_detach(tfile, clean);
 491         rtnl_unlock();
 492 }
 493
 494 static void tun_detach_all(struct net_device *dev)
 495 {
 496         struct tun_struct *tun = netdev_priv(dev);
 497         struct tun_file *tfile, *tmp;
 498         int i, n = tun->numqueues;
 499
 500         for (i = 0; i < n; i++) {
 501                 tfile = rtnl_dereference(tun->tfiles[i]);
 502                 BUG_ON(!tfile);
 503                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 504                 RCU_INIT_POINTER(tfile->tun, NULL);
 505                 --tun->numqueues;
 506         }
 507         list_for_each_entry(tfile, &tun->disabled, next) {
 508                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 509                 RCU_INIT_POINTER(tfile->tun, NULL);
 510         }
 511         BUG_ON(tun->numqueues != 0);
 512
 513         synchronize_net();
 514         for (i = 0; i < n; i++) {
 515                 tfile = rtnl_dereference(tun->tfiles[i]);
 516                 /* Drop read queue */
 517                 tun_queue_purge(tfile);
 518                 sock_put(&tfile->sk);
 519         }
 520         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 521                 tun_enable_queue(tfile);
 522                 tun_queue_purge(tfile);
 523                 sock_put(&tfile->sk);
 524         }
 525         BUG_ON(tun->numdisabled != 0);
 526
 527         if (tun->flags & TUN_PERSIST)
 528                 module_put(THIS_MODULE);
 529 }
 530
 531 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 532 {
 533         struct tun_file *tfile = file->private_data;
 534         int err;
 535
 536         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 537         if (err < 0)
 538                 goto out;
 539
 540         err = -EINVAL;
 541         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 542                 goto out;
 543
 544         err = -EBUSY;
 545         if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 546                 goto out;
 547
 548         err = -E2BIG;
 549         if (!tfile->detached &&
 550             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 551                 goto out;
 552
 553         err = 0;
 554
 555         /* Re-attach the filter to persist device */
 556         if (!skip_filter && (tun->filter_attached == true)) {
 557                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 558                 if (!err)
 559                         goto out;
 560         }
 561         tfile->queue_index = tun->numqueues;
 562         rcu_assign_pointer(tfile->tun, tun);
 563         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 564         tun->numqueues++;
 565
 566         if (tfile->detached)
 567                 tun_enable_queue(tfile);
 568         else
 569                 sock_hold(&tfile->sk);
 570
 571         tun_set_real_num_queues(tun);
 572
 573         /* device is allowed to go away first, so no need to hold extra
 574          * refcnt.
 575          */
 576
 577 out:
 578         return err;
 579 }
 580
 581 static struct tun_struct *__tun_get(struct tun_file *tfile)
 582 {
 583         struct tun_struct *tun;
 584
 585         rcu_read_lock();
 586         tun = rcu_dereference(tfile->tun);
 587         if (tun)
 588                 dev_hold(tun->dev);
 589         rcu_read_unlock();
 590
 591         return tun;
 592 }
 593
 594 static struct tun_struct *tun_get(struct file *file)
 595 {
 596         return __tun_get(file->private_data);
 597 }
 598
 599 static void tun_put(struct tun_struct *tun)
 600 {
 601         dev_put(tun->dev);
 602 }
 603
 604 /* TAP filtering */
 605 static void addr_hash_set(u32 *mask, const u8 *addr)
 606 {
 607         int n = ether_crc(ETH_ALEN, addr) >> 26;
 608         mask[n >> 5] |= (1 << (n & 31));
 609 }
 610
 611 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 612 {
 613         int n = ether_crc(ETH_ALEN, addr) >> 26;
 614         return mask[n >> 5] & (1 << (n & 31));
 615 }
 616
 617 static int update_filter(struct tap_filter *filter, void __user *arg)
 618 {
 619         struct { u8 u[ETH_ALEN]; } *addr;
 620         struct tun_filter uf;
 621         int err, alen, n, nexact;
 622
 623         if (copy_from_user(&uf, arg, sizeof(uf)))
 624                 return -EFAULT;
 625
 626         if (!uf.count) {
 627                 /* Disabled */
 628                 filter->count = 0;
 629                 return 0;
 630         }
 631
 632         alen = ETH_ALEN * uf.count;
 633         addr = kmalloc(alen, GFP_KERNEL);
 634         if (!addr)
 635                 return -ENOMEM;
 636
 637         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 638                 err = -EFAULT;
 639                 goto done;
 640         }
 641
 642         /* The filter is updated without holding any locks. Which is
 643          * perfectly safe. We disable it first and in the worst
 644          * case we'll accept a few undesired packets. */
 645         filter->count = 0;
 646         wmb();
 647
 648         /* Use first set of addresses as an exact filter */
 649         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 650                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 651
 652         nexact = n;
 653
 654         /* Remaining multicast addresses are hashed,
 655          * unicast will leave the filter disabled. */
 656         memset(filter->mask, 0, sizeof(filter->mask));
 657         for (; n < uf.count; n++) {
 658                 if (!is_multicast_ether_addr(addr[n].u)) {
 659                         err = 0; /* no filter */
 660                         goto done;
 661                 }
 662                 addr_hash_set(filter->mask, addr[n].u);
 663         }
 664
 665         /* For ALLMULTI just set the mask to all ones.
 666          * This overrides the mask populated above. */
 667         if ((uf.flags & TUN_FLT_ALLMULTI))
 668                 memset(filter->mask, ~0, sizeof(filter->mask));
 669
 670         /* Now enable the filter */
 671         wmb();
 672         filter->count = nexact;
 673
 674         /* Return the number of exact filters */
 675         err = nexact;
 676
 677 done:
 678         kfree(addr);
 679         return err;
 680 }
 681
 682 /* Returns: 0 - drop, !=0 - accept */
 683 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 684 {
 685         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 686          * at this point. */
 687         struct ethhdr *eh = (struct ethhdr *) skb->data;
 688         int i;
 689
 690         /* Exact match */
 691         for (i = 0; i < filter->count; i++)
 692                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 693                         return 1;
 694
 695         /* Inexact match (multicast only) */
 696         if (is_multicast_ether_addr(eh->h_dest))
 697                 return addr_hash_test(filter->mask, eh->h_dest);
 698
 699         return 0;
 700 }
 701
 702 /*
 703  * Checks whether the packet is accepted or not.
 704  * Returns: 0 - drop, !=0 - accept
 705  */
 706 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 707 {
 708         if (!filter->count)
 709                 return 1;
 710
 711         return run_filter(filter, skb);
 712 }
 713
 714 /* Network device part of the driver */
 715
 716 static const struct ethtool_ops tun_ethtool_ops;
 717
 718 /* Net device detach from fd. */
 719 static void tun_net_uninit(struct net_device *dev)
 720 {
 721         tun_detach_all(dev);
 722 }
 723
 724 /* Net device open. */
 725 static int tun_net_open(struct net_device *dev)
 726 {
 727         netif_tx_start_all_queues(dev);
 728         return 0;
 729 }
 730
 731 /* Net device close. */
 732 static int tun_net_close(struct net_device *dev)
 733 {
 734         netif_tx_stop_all_queues(dev);
 735         return 0;
 736 }
 737
 738 /* Net device start xmit */
 739 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 740 {
 741         struct tun_struct *tun = netdev_priv(dev);
 742         int txq = skb->queue_mapping;
 743         struct tun_file *tfile;
 744         u32 numqueues = 0;
 745
 746         rcu_read_lock();
 747         tfile = rcu_dereference(tun->tfiles[txq]);
 748         numqueues = ACCESS_ONCE(tun->numqueues);
 749
 750         /* Drop packet if interface is not attached */
 751         if (txq >= numqueues)
 752                 goto drop;
 753
 754         if (numqueues == 1) {
 755                 /* Select queue was not called for the skbuff, so we extract the
 756                  * RPS hash and save it into the flow_table here.
 757                  */
 758                 __u32 rxhash;
 759
 760                 rxhash = skb_get_hash(skb);
 761                 if (rxhash) {
 762                         struct tun_flow_entry *e;
 763                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 764                                         rxhash);
 765                         if (e)
 766                                 tun_flow_save_rps_rxhash(e, rxhash);
 767                 }
 768         }
 769
 770         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 771
 772         BUG_ON(!tfile);
 773
 774         /* Drop if the filter does not like it.
 775          * This is a noop if the filter is disabled.
 776          * Filter can be enabled only for the TAP devices. */
 777         if (!check_filter(&tun->txflt, skb))
 778                 goto drop;
 779
 780         if (tfile->socket.sk->sk_filter &&
 781             sk_filter(tfile->socket.sk, skb))
 782                 goto drop;
 783
 784         /* Limit the number of packets queued by dividing txq length with the
 785          * number of queues.
 786          */
 787         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
 788                           >= dev->tx_queue_len)
 789                 goto drop;
 790
 791         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 792                 goto drop;
 793
 794         if (skb->sk) {
 795                 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
 796                 sw_tx_timestamp(skb);
 797         }
 798
 799         /* Orphan the skb - required as we might hang on to it
 800          * for indefinite time.
 801          */
 802         skb_orphan(skb);
 803
 804         nf_reset(skb);
 805
 806         /* Enqueue packet */
 807         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 808
 809         /* Notify and wake up reader process */
 810         if (tfile->flags & TUN_FASYNC)
 811                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 812         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 813
 814         rcu_read_unlock();
 815         return NETDEV_TX_OK;
 816
 817 drop:
 818         dev->stats.tx_dropped++;
 819         skb_tx_error(skb);
 820         kfree_skb(skb);
 821         rcu_read_unlock();
 822         return NET_XMIT_DROP;
 823 }
 824
 825 static void tun_net_mclist(struct net_device *dev)
 826 {
 827         /*
 828          * This callback is supposed to deal with mc filter in
 829          * _rx_ path and has nothing to do with the _tx_ path.
 830          * In rx path we always accept everything userspace gives us.
 831          */
 832 }
 833
 834 #define MIN_MTU 68
 835 #define MAX_MTU 65535
 836
 837 static int
 838 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 839 {
 840         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 841                 return -EINVAL;
 842         dev->mtu = new_mtu;
 843         return 0;
 844 }
 845
 846 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 847         netdev_features_t features)
 848 {
 849         struct tun_struct *tun = netdev_priv(dev);
 850
 851         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 852 }
 853 #ifdef CONFIG_NET_POLL_CONTROLLER
 854 static void tun_poll_controller(struct net_device *dev)
 855 {
 856         /*
 857          * Tun only receives frames when:
 858          * 1) the char device endpoint gets data from user space
 859          * 2) the tun socket gets a sendmsg call from user space
 860          * Since both of those are synchronous operations, we are guaranteed
 861          * never to have pending data when we poll for it
 862          * so there is nothing to do here but return.
 863          * We need this though so netpoll recognizes us as an interface that
 864          * supports polling, which enables bridge devices in virt setups to
 865          * still use netconsole
 866          */
 867         return;
 868 }
 869 #endif
 870 static const struct net_device_ops tun_netdev_ops = {
 871         .ndo_uninit             = tun_net_uninit,
 872         .ndo_open               = tun_net_open,
 873         .ndo_stop               = tun_net_close,
 874         .ndo_start_xmit         = tun_net_xmit,
 875         .ndo_change_mtu         = tun_net_change_mtu,
 876         .ndo_fix_features       = tun_net_fix_features,
 877         .ndo_select_queue       = tun_select_queue,
 878 #ifdef CONFIG_NET_POLL_CONTROLLER
 879         .ndo_poll_controller    = tun_poll_controller,
 880 #endif
 881 };
 882
 883 static const struct net_device_ops tap_netdev_ops = {
 884         .ndo_uninit             = tun_net_uninit,
 885         .ndo_open               = tun_net_open,
 886         .ndo_stop               = tun_net_close,
 887         .ndo_start_xmit         = tun_net_xmit,
 888         .ndo_change_mtu         = tun_net_change_mtu,
 889         .ndo_fix_features       = tun_net_fix_features,
 890         .ndo_set_rx_mode        = tun_net_mclist,
 891         .ndo_set_mac_address    = eth_mac_addr,
 892         .ndo_validate_addr      = eth_validate_addr,
 893         .ndo_select_queue       = tun_select_queue,
 894 #ifdef CONFIG_NET_POLL_CONTROLLER
 895         .ndo_poll_controller    = tun_poll_controller,
 896 #endif
 897 };
 898
 899 static void tun_flow_init(struct tun_struct *tun)
 900 {
 901         int i;
 902
 903         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 904                 INIT_HLIST_HEAD(&tun->flows[i]);
 905
 906         tun->ageing_time = TUN_FLOW_EXPIRE;
 907         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 908         mod_timer(&tun->flow_gc_timer,
 909                   round_jiffies_up(jiffies + tun->ageing_time));
 910 }
 911
 912 static void tun_flow_uninit(struct tun_struct *tun)
 913 {
 914         del_timer_sync(&tun->flow_gc_timer);
 915         tun_flow_flush(tun);
 916 }
 917
 918 /* Initialize net device. */
 919 static void tun_net_init(struct net_device *dev)
 920 {
 921         struct tun_struct *tun = netdev_priv(dev);
 922
 923         switch (tun->flags & TUN_TYPE_MASK) {
 924         case TUN_TUN_DEV:
 925                 dev->netdev_ops = &tun_netdev_ops;
 926
 927                 /* Point-to-Point TUN Device */
 928                 dev->hard_header_len = 0;
 929                 dev->addr_len = 0;
 930                 dev->mtu = 1500;
 931
 932                 /* Zero header length */
 933                 dev->type = ARPHRD_NONE;
 934                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 935                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 936                 break;
 937
 938         case TUN_TAP_DEV:
 939                 dev->netdev_ops = &tap_netdev_ops;
 940                 /* Ethernet TAP Device */
 941                 ether_setup(dev);
 942                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 943                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 944
 945                 eth_hw_addr_random(dev);
 946
 947                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 948                 break;
 949         }
 950 }
 951
 952 /* Character device part */
 953
 954 /* Poll */
 955 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 956 {
 957         struct tun_file *tfile = file->private_data;
 958         struct tun_struct *tun = __tun_get(tfile);
 959         struct sock *sk;
 960         unsigned int mask = 0;
 961
 962         if (!tun)
 963                 return POLLERR;
 964
 965         sk = tfile->socket.sk;
 966
 967         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 968
 969         poll_wait(file, sk_sleep(sk), wait);
 970
 971         if (!skb_queue_empty(&sk->sk_receive_queue))
 972                 mask |= POLLIN | POLLRDNORM;
 973
 974         if (sock_writeable(sk) ||
 975             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 976              sock_writeable(sk)))
 977                 mask |= POLLOUT | POLLWRNORM;
 978
 979         if (tun->dev->reg_state != NETREG_REGISTERED)
 980                 mask = POLLERR;
 981
 982         tun_put(tun);
 983         return mask;
 984 }
 985
 986 /* prepad is the amount to reserve at front.  len is length after that.
 987  * linear is a hint as to how much to copy (usually headers). */
 988 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 989                                      size_t prepad, size_t len,
 990                                      size_t linear, int noblock)
 991 {
 992         struct sock *sk = tfile->socket.sk;
 993         struct sk_buff *skb;
 994         int err;
 995
 996         /* Under a page?  Don't bother with paged skb. */
 997         if (prepad + len < PAGE_SIZE || !linear)
 998                 linear = len;
 999
1000         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1001                                    &err, 0);
1002         if (!skb)
1003                 return ERR_PTR(err);
1004
1005         skb_reserve(skb, prepad);
1006         skb_put(skb, linear);
1007         skb->data_len = len - linear;
1008         skb->len += len - linear;
1009
1010         return skb;
1011 }
1012
1013 /* Get packet from user space buffer */
1014 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1015                             void *msg_control, struct iov_iter *from,
1016                             int noblock)
1017 {
1018         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1019         struct sk_buff *skb;
1020         size_t total_len = iov_iter_count(from);
1021         size_t len = total_len, align = NET_SKB_PAD, linear;
1022         struct virtio_net_hdr gso = { 0 };
1023         int good_linear;
1024         int copylen;
1025         bool zerocopy = false;
1026         int err;
1027         u32 rxhash;
1028         ssize_t n;
1029
1030         if (!(tun->flags & TUN_NO_PI)) {
1031                 if (len < sizeof(pi))
1032                         return -EINVAL;
1033                 len -= sizeof(pi);
1034
1035                 n = copy_from_iter(&pi, sizeof(pi), from);
1036                 if (n != sizeof(pi))
1037                         return -EFAULT;
1038         }
1039
1040         if (tun->flags & TUN_VNET_HDR) {
1041                 if (len < tun->vnet_hdr_sz)
1042                         return -EINVAL;
1043                 len -= tun->vnet_hdr_sz;
1044
1045                 n = copy_from_iter(&gso, sizeof(gso), from);
1046                 if (n != sizeof(gso))
1047                         return -EFAULT;
1048
1049                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1050                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1051                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1052
1053                 if (gso.hdr_len > len)
1054                         return -EINVAL;
1055                 iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1056         }
1057
1058         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1059                 align += NET_IP_ALIGN;
1060                 if (unlikely(len < ETH_HLEN ||
1061                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1062                         return -EINVAL;
1063         }
1064
1065         good_linear = SKB_MAX_HEAD(align);
1066
1067         if (msg_control) {
1068                 struct iov_iter i = *from;
1069
1070                 /* There are 256 bytes to be copied in skb, so there is
1071                  * enough room for skb expand head in case it is used.
1072                  * The rest of the buffer is mapped from userspace.
1073                  */
1074                 copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1075                 if (copylen > good_linear)
1076                         copylen = good_linear;
1077                 linear = copylen;
1078                 iov_iter_advance(&i, copylen);
1079                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1080                         zerocopy = true;
1081         }
1082
1083         if (!zerocopy) {
1084                 copylen = len;
1085                 if (gso.hdr_len > good_linear)
1086                         linear = good_linear;
1087                 else
1088                         linear = gso.hdr_len;
1089         }
1090
1091         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1092         if (IS_ERR(skb)) {
1093                 if (PTR_ERR(skb) != -EAGAIN)
1094                         tun->dev->stats.rx_dropped++;
1095                 return PTR_ERR(skb);
1096         }
1097
1098         if (zerocopy)
1099                 err = zerocopy_sg_from_iter(skb, from);
1100         else {
1101                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1102                 if (!err && msg_control) {
1103                         struct ubuf_info *uarg = msg_control;
1104                         uarg->callback(uarg, false);
1105                 }
1106         }
1107
1108         if (err) {
1109                 tun->dev->stats.rx_dropped++;
1110                 kfree_skb(skb);
1111                 return -EFAULT;
1112         }
1113
1114         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1115                 if (!skb_partial_csum_set(skb, gso.csum_start,
1116                                           gso.csum_offset)) {
1117                         tun->dev->stats.rx_frame_errors++;
1118                         kfree_skb(skb);
1119                         return -EINVAL;
1120                 }
1121         }
1122
1123         switch (tun->flags & TUN_TYPE_MASK) {
1124         case TUN_TUN_DEV:
1125                 if (tun->flags & TUN_NO_PI) {
1126                         switch (skb->data[0] & 0xf0) {
1127                         case 0x40:
1128                                 pi.proto = htons(ETH_P_IP);
1129                                 break;
1130                         case 0x60:
1131                                 pi.proto = htons(ETH_P_IPV6);
1132                                 break;
1133                         default:
1134                                 tun->dev->stats.rx_dropped++;
1135                                 kfree_skb(skb);
1136                                 return -EINVAL;
1137                         }
1138                 }
1139
1140                 skb_reset_mac_header(skb);
1141                 skb->protocol = pi.proto;
1142                 skb->dev = tun->dev;
1143                 break;
1144         case TUN_TAP_DEV:
1145                 skb->protocol = eth_type_trans(skb, tun->dev);
1146                 break;
1147         }
1148
1149         skb_reset_network_header(skb);
1150
1151         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1152                 pr_debug("GSO!\n");
1153                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1154                 case VIRTIO_NET_HDR_GSO_TCPV4:
1155                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1156                         break;
1157                 case VIRTIO_NET_HDR_GSO_TCPV6:
1158                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1159                         break;
1160                 case VIRTIO_NET_HDR_GSO_UDP:
1161                 {
1162                         static bool warned;
1163
1164                         if (!warned) {
1165                                 warned = true;
1166                                 netdev_warn(tun->dev,
1167                                             "%s: using disabled UFO feature; please fix this program\n",
1168                                             current->comm);
1169                         }
1170                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1171                         if (skb->protocol == htons(ETH_P_IPV6))
1172                                 ipv6_proxy_select_ident(skb);
1173                         break;
1174                 }
1175                 default:
1176                         tun->dev->stats.rx_frame_errors++;
1177                         kfree_skb(skb);
1178                         return -EINVAL;
1179                 }
1180
1181                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1182                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1183
1184                 skb_shinfo(skb)->gso_size = gso.gso_size;
1185                 if (skb_shinfo(skb)->gso_size == 0) {
1186                         tun->dev->stats.rx_frame_errors++;
1187                         kfree_skb(skb);
1188                         return -EINVAL;
1189                 }
1190
1191                 /* Header must be checked, and gso_segs computed. */
1192                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1193                 skb_shinfo(skb)->gso_segs = 0;
1194         }
1195
1196         /* copy skb_ubuf_info for callback when skb has no error */
1197         if (zerocopy) {
1198                 skb_shinfo(skb)->destructor_arg = msg_control;
1199                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1200                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1201         }
1202
1203         skb_probe_transport_header(skb, 0);
1204
1205         rxhash = skb_get_hash(skb);
1206         netif_rx_ni(skb);
1207
1208         tun->dev->stats.rx_packets++;
1209         tun->dev->stats.rx_bytes += len;
1210
1211         tun_flow_update(tun, rxhash, tfile);
1212         return total_len;
1213 }
1214
1215 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1216 {
1217         struct file *file = iocb->ki_filp;
1218         struct tun_struct *tun = tun_get(file);
1219         struct tun_file *tfile = file->private_data;
1220         ssize_t result;
1221
1222         if (!tun)
1223                 return -EBADFD;
1224
1225         result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1226
1227         tun_put(tun);
1228         return result;
1229 }
1230
1231 /* Put packet to the user space buffer */
1232 static ssize_t tun_put_user(struct tun_struct *tun,
1233                             struct tun_file *tfile,
1234                             struct sk_buff *skb,
1235                             struct iov_iter *iter)
1236 {
1237         struct tun_pi pi = { 0, skb->protocol };
1238         ssize_t total;
1239         int vlan_offset = 0;
1240         int vlan_hlen = 0;
1241         int vnet_hdr_sz = 0;
1242
1243         if (vlan_tx_tag_present(skb))
1244                 vlan_hlen = VLAN_HLEN;
1245
1246         if (tun->flags & TUN_VNET_HDR)
1247                 vnet_hdr_sz = tun->vnet_hdr_sz;
1248
1249         total = skb->len + vlan_hlen + vnet_hdr_sz;
1250
1251         if (!(tun->flags & TUN_NO_PI)) {
1252                 if (iov_iter_count(iter) < sizeof(pi))
1253                         return -EINVAL;
1254
1255                 total += sizeof(pi);
1256                 if (iov_iter_count(iter) < total) {
1257                         /* Packet will be striped */
1258                         pi.flags |= TUN_PKT_STRIP;
1259                 }
1260
1261                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1262                         return -EFAULT;
1263         }
1264
1265         if (vnet_hdr_sz) {
1266                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1267                 if (iov_iter_count(iter) < vnet_hdr_sz)
1268                         return -EINVAL;
1269
1270                 if (skb_is_gso(skb)) {
1271                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1272
1273                         /* This is a hint as to how much should be linear. */
1274                         gso.hdr_len = skb_headlen(skb);
1275                         gso.gso_size = sinfo->gso_size;
1276                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1277                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1278                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1279                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1280                         else {
1281                                 pr_err("unexpected GSO type: "
1282                                        "0x%x, gso_size %d, hdr_len %d\n",
1283                                        sinfo->gso_type, gso.gso_size,
1284                                        gso.hdr_len);
1285                                 print_hex_dump(KERN_ERR, "tun: ",
1286                                                DUMP_PREFIX_NONE,
1287                                                16, 1, skb->head,
1288                                                min((int)gso.hdr_len, 64), true);
1289                                 WARN_ON_ONCE(1);
1290                                 return -EINVAL;
1291                         }
1292                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1293                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1294                 } else
1295                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1296
1297                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1298                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1299                         gso.csum_start = skb_checksum_start_offset(skb) +
1300                                          vlan_hlen;
1301                         gso.csum_offset = skb->csum_offset;
1302                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1303                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1304                 } /* else everything is zero */
1305
1306                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1307                         return -EFAULT;
1308
1309                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1310         }
1311
1312         if (vlan_hlen) {
1313                 int ret;
1314                 struct {
1315                         __be16 h_vlan_proto;
1316                         __be16 h_vlan_TCI;
1317                 } veth;
1318
1319                 veth.h_vlan_proto = skb->vlan_proto;
1320                 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1321
1322                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1323
1324                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1325                 if (ret || !iov_iter_count(iter))
1326                         goto done;
1327
1328                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1329                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1330                         goto done;
1331         }
1332
1333         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1334
1335 done:
1336         tun->dev->stats.tx_packets++;
1337         tun->dev->stats.tx_bytes += skb->len + vlan_hlen;
1338
1339         return total;
1340 }
1341
1342 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1343                            struct iov_iter *to,
1344                            int noblock)
1345 {
1346         struct sk_buff *skb;
1347         ssize_t ret;
1348         int peeked, err, off = 0;
1349
1350         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1351
1352         if (!iov_iter_count(to))
1353                 return 0;
1354
1355         if (tun->dev->reg_state != NETREG_REGISTERED)
1356                 return -EIO;
1357
1358         /* Read frames from queue */
1359         skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1360                                   &peeked, &off, &err);
1361         if (!skb)
1362                 return 0;
1363
1364         ret = tun_put_user(tun, tfile, skb, to);
1365         kfree_skb(skb);
1366
1367         return ret;
1368 }
1369
1370 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1371 {
1372         struct file *file = iocb->ki_filp;
1373         struct tun_file *tfile = file->private_data;
1374         struct tun_struct *tun = __tun_get(tfile);
1375         ssize_t len = iov_iter_count(to), ret;
1376
1377         if (!tun)
1378                 return -EBADFD;
1379         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1380         ret = min_t(ssize_t, ret, len);
1381         if (ret > 0)
1382                 iocb->ki_pos = ret;
1383         tun_put(tun);
1384         return ret;
1385 }
1386
1387 static void tun_free_netdev(struct net_device *dev)
1388 {
1389         struct tun_struct *tun = netdev_priv(dev);
1390
1391         BUG_ON(!(list_empty(&tun->disabled)));
1392         tun_flow_uninit(tun);
1393         security_tun_dev_free_security(tun->security);
1394         free_netdev(dev);
1395 }
1396
1397 static void tun_setup(struct net_device *dev)
1398 {
1399         struct tun_struct *tun = netdev_priv(dev);
1400
1401         tun->owner = INVALID_UID;
1402         tun->group = INVALID_GID;
1403
1404         dev->ethtool_ops = &tun_ethtool_ops;
1405         dev->destructor = tun_free_netdev;
1406 }
1407
1408 /* Trivial set of netlink ops to allow deleting tun or tap
1409  * device with netlink.
1410  */
1411 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1412 {
1413         return -EINVAL;
1414 }
1415
1416 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1417         .kind           = DRV_NAME,
1418         .priv_size      = sizeof(struct tun_struct),
1419         .setup          = tun_setup,
1420         .validate       = tun_validate,
1421 };
1422
1423 static void tun_sock_write_space(struct sock *sk)
1424 {
1425         struct tun_file *tfile;
1426         wait_queue_head_t *wqueue;
1427
1428         if (!sock_writeable(sk))
1429                 return;
1430
1431         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1432                 return;
1433
1434         wqueue = sk_sleep(sk);
1435         if (wqueue && waitqueue_active(wqueue))
1436                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1437                                                 POLLWRNORM | POLLWRBAND);
1438
1439         tfile = container_of(sk, struct tun_file, sk);
1440         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1441 }
1442
1443 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1444                        struct msghdr *m, size_t total_len)
1445 {
1446         int ret;
1447         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1448         struct tun_struct *tun = __tun_get(tfile);
1449         struct iov_iter from;
1450
1451         if (!tun)
1452                 return -EBADFD;
1453
1454         iov_iter_init(&from, WRITE, m->msg_iov, m->msg_iovlen, total_len);
1455         ret = tun_get_user(tun, tfile, m->msg_control, &from,
1456                            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         struct iov_iter to;
1468         int ret;
1469
1470         if (!tun)
1471                 return -EBADFD;
1472
1473         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1474                 ret = -EINVAL;
1475                 goto out;
1476         }
1477         if (flags & MSG_ERRQUEUE) {
1478                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1479                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1480                 goto out;
1481         }
1482         iov_iter_init(&to, READ, m->msg_iov, m->msg_iovlen, total_len);
1483         ret = tun_do_read(tun, tfile, &to, flags & MSG_DONTWAIT);
1484         if (ret > total_len) {
1485                 m->msg_flags |= MSG_TRUNC;
1486                 ret = flags & MSG_TRUNC ? ret : total_len;
1487         }
1488 out:
1489         tun_put(tun);
1490         return ret;
1491 }
1492
1493 static int tun_release(struct socket *sock)
1494 {
1495         if (sock->sk)
1496                 sock_put(sock->sk);
1497         return 0;
1498 }
1499
1500 /* Ops structure to mimic raw sockets with tun */
1501 static const struct proto_ops tun_socket_ops = {
1502         .sendmsg = tun_sendmsg,
1503         .recvmsg = tun_recvmsg,
1504         .release = tun_release,
1505 };
1506
1507 static struct proto tun_proto = {
1508         .name           = "tun",
1509         .owner          = THIS_MODULE,
1510         .obj_size       = sizeof(struct tun_file),
1511 };
1512
1513 static int tun_flags(struct tun_struct *tun)
1514 {
1515         int flags = 0;
1516
1517         if (tun->flags & TUN_TUN_DEV)
1518                 flags |= IFF_TUN;
1519         else
1520                 flags |= IFF_TAP;
1521
1522         if (tun->flags & TUN_NO_PI)
1523                 flags |= IFF_NO_PI;
1524
1525         /* This flag has no real effect.  We track the value for backwards
1526          * compatibility.
1527          */
1528         if (tun->flags & TUN_ONE_QUEUE)
1529                 flags |= IFF_ONE_QUEUE;
1530
1531         if (tun->flags & TUN_VNET_HDR)
1532                 flags |= IFF_VNET_HDR;
1533
1534         if (tun->flags & TUN_TAP_MQ)
1535                 flags |= IFF_MULTI_QUEUE;
1536
1537         if (tun->flags & TUN_PERSIST)
1538                 flags |= IFF_PERSIST;
1539
1540         return flags;
1541 }
1542
1543 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1544                               char *buf)
1545 {
1546         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1547         return sprintf(buf, "0x%x\n", tun_flags(tun));
1548 }
1549
1550 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1551                               char *buf)
1552 {
1553         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1554         return uid_valid(tun->owner)?
1555                 sprintf(buf, "%u\n",
1556                         from_kuid_munged(current_user_ns(), tun->owner)):
1557                 sprintf(buf, "-1\n");
1558 }
1559
1560 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1561                               char *buf)
1562 {
1563         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1564         return gid_valid(tun->group) ?
1565                 sprintf(buf, "%u\n",
1566                         from_kgid_munged(current_user_ns(), tun->group)):
1567                 sprintf(buf, "-1\n");
1568 }
1569
1570 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1571 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1572 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1573
1574 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1575 {
1576         struct tun_struct *tun;
1577         struct tun_file *tfile = file->private_data;
1578         struct net_device *dev;
1579         int err;
1580
1581         if (tfile->detached)
1582                 return -EINVAL;
1583
1584         dev = __dev_get_by_name(net, ifr->ifr_name);
1585         if (dev) {
1586                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1587                         return -EBUSY;
1588                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1589                         tun = netdev_priv(dev);
1590                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1591                         tun = netdev_priv(dev);
1592                 else
1593                         return -EINVAL;
1594
1595                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1596                     !!(tun->flags & TUN_TAP_MQ))
1597                         return -EINVAL;
1598
1599                 if (tun_not_capable(tun))
1600                         return -EPERM;
1601                 err = security_tun_dev_open(tun->security);
1602                 if (err < 0)
1603                         return err;
1604
1605                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1606                 if (err < 0)
1607                         return err;
1608
1609                 if (tun->flags & TUN_TAP_MQ &&
1610                     (tun->numqueues + tun->numdisabled > 1)) {
1611                         /* One or more queue has already been attached, no need
1612                          * to initialize the device again.
1613                          */
1614                         return 0;
1615                 }
1616         }
1617         else {
1618                 char *name;
1619                 unsigned long flags = 0;
1620                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1621                              MAX_TAP_QUEUES : 1;
1622
1623                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1624                         return -EPERM;
1625                 err = security_tun_dev_create();
1626                 if (err < 0)
1627                         return err;
1628
1629                 /* Set dev type */
1630                 if (ifr->ifr_flags & IFF_TUN) {
1631                         /* TUN device */
1632                         flags |= TUN_TUN_DEV;
1633                         name = "tun%d";
1634                 } else if (ifr->ifr_flags & IFF_TAP) {
1635                         /* TAP device */
1636                         flags |= TUN_TAP_DEV;
1637                         name = "tap%d";
1638                 } else
1639                         return -EINVAL;
1640
1641                 if (*ifr->ifr_name)
1642                         name = ifr->ifr_name;
1643
1644                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1645                                        NET_NAME_UNKNOWN, tun_setup, queues,
1646                                        queues);
1647
1648                 if (!dev)
1649                         return -ENOMEM;
1650
1651                 dev_net_set(dev, net);
1652                 dev->rtnl_link_ops = &tun_link_ops;
1653                 dev->ifindex = tfile->ifindex;
1654
1655                 tun = netdev_priv(dev);
1656                 tun->dev = dev;
1657                 tun->flags = flags;
1658                 tun->txflt.count = 0;
1659                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1660
1661                 tun->filter_attached = false;
1662                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1663
1664                 spin_lock_init(&tun->lock);
1665
1666                 err = security_tun_dev_alloc_security(&tun->security);
1667                 if (err < 0)
1668                         goto err_free_dev;
1669
1670                 tun_net_init(dev);
1671                 tun_flow_init(tun);
1672
1673                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1674                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1675                                    NETIF_F_HW_VLAN_STAG_TX;
1676                 dev->features = dev->hw_features;
1677                 dev->vlan_features = dev->features &
1678                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1679                                        NETIF_F_HW_VLAN_STAG_TX);
1680
1681                 INIT_LIST_HEAD(&tun->disabled);
1682                 err = tun_attach(tun, file, false);
1683                 if (err < 0)
1684                         goto err_free_flow;
1685
1686                 err = register_netdevice(tun->dev);
1687                 if (err < 0)
1688                         goto err_detach;
1689
1690                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1691                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1692                     device_create_file(&tun->dev->dev, &dev_attr_group))
1693                         pr_err("Failed to create tun sysfs files\n");
1694         }
1695
1696         netif_carrier_on(tun->dev);
1697
1698         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1699
1700         if (ifr->ifr_flags & IFF_NO_PI)
1701                 tun->flags |= TUN_NO_PI;
1702         else
1703                 tun->flags &= ~TUN_NO_PI;
1704
1705         /* This flag has no real effect.  We track the value for backwards
1706          * compatibility.
1707          */
1708         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1709                 tun->flags |= TUN_ONE_QUEUE;
1710         else
1711                 tun->flags &= ~TUN_ONE_QUEUE;
1712
1713         if (ifr->ifr_flags & IFF_VNET_HDR)
1714                 tun->flags |= TUN_VNET_HDR;
1715         else
1716                 tun->flags &= ~TUN_VNET_HDR;
1717
1718         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1719                 tun->flags |= TUN_TAP_MQ;
1720         else
1721                 tun->flags &= ~TUN_TAP_MQ;
1722
1723         /* Make sure persistent devices do not get stuck in
1724          * xoff state.
1725          */
1726         if (netif_running(tun->dev))
1727                 netif_tx_wake_all_queues(tun->dev);
1728
1729         strcpy(ifr->ifr_name, tun->dev->name);
1730         return 0;
1731
1732 err_detach:
1733         tun_detach_all(dev);
1734 err_free_flow:
1735         tun_flow_uninit(tun);
1736         security_tun_dev_free_security(tun->security);
1737 err_free_dev:
1738         free_netdev(dev);
1739         return err;
1740 }
1741
1742 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1743                        struct ifreq *ifr)
1744 {
1745         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1746
1747         strcpy(ifr->ifr_name, tun->dev->name);
1748
1749         ifr->ifr_flags = tun_flags(tun);
1750
1751 }
1752
1753 /* This is like a cut-down ethtool ops, except done via tun fd so no
1754  * privs required. */
1755 static int set_offload(struct tun_struct *tun, unsigned long arg)
1756 {
1757         netdev_features_t features = 0;
1758
1759         if (arg & TUN_F_CSUM) {
1760                 features |= NETIF_F_HW_CSUM;
1761                 arg &= ~TUN_F_CSUM;
1762
1763                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1764                         if (arg & TUN_F_TSO_ECN) {
1765                                 features |= NETIF_F_TSO_ECN;
1766                                 arg &= ~TUN_F_TSO_ECN;
1767                         }
1768                         if (arg & TUN_F_TSO4)
1769                                 features |= NETIF_F_TSO;
1770                         if (arg & TUN_F_TSO6)
1771                                 features |= NETIF_F_TSO6;
1772                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1773                 }
1774         }
1775
1776         /* This gives the user a way to test for new features in future by
1777          * trying to set them. */
1778         if (arg)
1779                 return -EINVAL;
1780
1781         tun->set_features = features;
1782         netdev_update_features(tun->dev);
1783
1784         return 0;
1785 }
1786
1787 static void tun_detach_filter(struct tun_struct *tun, int n)
1788 {
1789         int i;
1790         struct tun_file *tfile;
1791
1792         for (i = 0; i < n; i++) {
1793                 tfile = rtnl_dereference(tun->tfiles[i]);
1794                 sk_detach_filter(tfile->socket.sk);
1795         }
1796
1797         tun->filter_attached = false;
1798 }
1799
1800 static int tun_attach_filter(struct tun_struct *tun)
1801 {
1802         int i, ret = 0;
1803         struct tun_file *tfile;
1804
1805         for (i = 0; i < tun->numqueues; i++) {
1806                 tfile = rtnl_dereference(tun->tfiles[i]);
1807                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1808                 if (ret) {
1809                         tun_detach_filter(tun, i);
1810                         return ret;
1811                 }
1812         }
1813
1814         tun->filter_attached = true;
1815         return ret;
1816 }
1817
1818 static void tun_set_sndbuf(struct tun_struct *tun)
1819 {
1820         struct tun_file *tfile;
1821         int i;
1822
1823         for (i = 0; i < tun->numqueues; i++) {
1824                 tfile = rtnl_dereference(tun->tfiles[i]);
1825                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1826         }
1827 }
1828
1829 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1830 {
1831         struct tun_file *tfile = file->private_data;
1832         struct tun_struct *tun;
1833         int ret = 0;
1834
1835         rtnl_lock();
1836
1837         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1838                 tun = tfile->detached;
1839                 if (!tun) {
1840                         ret = -EINVAL;
1841                         goto unlock;
1842                 }
1843                 ret = security_tun_dev_attach_queue(tun->security);
1844                 if (ret < 0)
1845                         goto unlock;
1846                 ret = tun_attach(tun, file, false);
1847         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1848                 tun = rtnl_dereference(tfile->tun);
1849                 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1850                         ret = -EINVAL;
1851                 else
1852                         __tun_detach(tfile, false);
1853         } else
1854                 ret = -EINVAL;
1855
1856 unlock:
1857         rtnl_unlock();
1858         return ret;
1859 }
1860
1861 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1862                             unsigned long arg, int ifreq_len)
1863 {
1864         struct tun_file *tfile = file->private_data;
1865         struct tun_struct *tun;
1866         void __user* argp = (void __user*)arg;
1867         struct ifreq ifr;
1868         kuid_t owner;
1869         kgid_t group;
1870         int sndbuf;
1871         int vnet_hdr_sz;
1872         unsigned int ifindex;
1873         int ret;
1874
1875         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1876                 if (copy_from_user(&ifr, argp, ifreq_len))
1877                         return -EFAULT;
1878         } else {
1879                 memset(&ifr, 0, sizeof(ifr));
1880         }
1881         if (cmd == TUNGETFEATURES) {
1882                 /* Currently this just means: "what IFF flags are valid?".
1883                  * This is needed because we never checked for invalid flags on
1884                  * TUNSETIFF. */
1885                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1886                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1887                                 (unsigned int __user*)argp);
1888         } else if (cmd == TUNSETQUEUE)
1889                 return tun_set_queue(file, &ifr);
1890
1891         ret = 0;
1892         rtnl_lock();
1893
1894         tun = __tun_get(tfile);
1895         if (cmd == TUNSETIFF && !tun) {
1896                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1897
1898                 ret = tun_set_iff(tfile->net, file, &ifr);
1899
1900                 if (ret)
1901                         goto unlock;
1902
1903                 if (copy_to_user(argp, &ifr, ifreq_len))
1904                         ret = -EFAULT;
1905                 goto unlock;
1906         }
1907         if (cmd == TUNSETIFINDEX) {
1908                 ret = -EPERM;
1909                 if (tun)
1910                         goto unlock;
1911
1912                 ret = -EFAULT;
1913                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1914                         goto unlock;
1915
1916                 ret = 0;
1917                 tfile->ifindex = ifindex;
1918                 goto unlock;
1919         }
1920
1921         ret = -EBADFD;
1922         if (!tun)
1923                 goto unlock;
1924
1925         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1926
1927         ret = 0;
1928         switch (cmd) {
1929         case TUNGETIFF:
1930                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1931
1932                 if (tfile->detached)
1933                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
1934                 if (!tfile->socket.sk->sk_filter)
1935                         ifr.ifr_flags |= IFF_NOFILTER;
1936
1937                 if (copy_to_user(argp, &ifr, ifreq_len))
1938                         ret = -EFAULT;
1939                 break;
1940
1941         case TUNSETNOCSUM:
1942                 /* Disable/Enable checksum */
1943
1944                 /* [unimplemented] */
1945                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1946                           arg ? "disabled" : "enabled");
1947                 break;
1948
1949         case TUNSETPERSIST:
1950                 /* Disable/Enable persist mode. Keep an extra reference to the
1951                  * module to prevent the module being unprobed.
1952                  */
1953                 if (arg && !(tun->flags & TUN_PERSIST)) {
1954                         tun->flags |= TUN_PERSIST;
1955                         __module_get(THIS_MODULE);
1956                 }
1957                 if (!arg && (tun->flags & TUN_PERSIST)) {
1958                         tun->flags &= ~TUN_PERSIST;
1959                         module_put(THIS_MODULE);
1960                 }
1961
1962                 tun_debug(KERN_INFO, tun, "persist %s\n",
1963                           arg ? "enabled" : "disabled");
1964                 break;
1965
1966         case TUNSETOWNER:
1967                 /* Set owner of the device */
1968                 owner = make_kuid(current_user_ns(), arg);
1969                 if (!uid_valid(owner)) {
1970                         ret = -EINVAL;
1971                         break;
1972                 }
1973                 tun->owner = owner;
1974                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1975                           from_kuid(&init_user_ns, tun->owner));
1976                 break;
1977
1978         case TUNSETGROUP:
1979                 /* Set group of the device */
1980                 group = make_kgid(current_user_ns(), arg);
1981                 if (!gid_valid(group)) {
1982                         ret = -EINVAL;
1983                         break;
1984                 }
1985                 tun->group = group;
1986                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1987                           from_kgid(&init_user_ns, tun->group));
1988                 break;
1989
1990         case TUNSETLINK:
1991                 /* Only allow setting the type when the interface is down */
1992                 if (tun->dev->flags & IFF_UP) {
1993                         tun_debug(KERN_INFO, tun,
1994                                   "Linktype set failed because interface is up\n");
1995                         ret = -EBUSY;
1996                 } else {
1997                         tun->dev->type = (int) arg;
1998                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1999                                   tun->dev->type);
2000                         ret = 0;
2001                 }
2002                 break;
2003
2004 #ifdef TUN_DEBUG
2005         case TUNSETDEBUG:
2006                 tun->debug = arg;
2007                 break;
2008 #endif
2009         case TUNSETOFFLOAD:
2010                 ret = set_offload(tun, arg);
2011                 break;
2012
2013         case TUNSETTXFILTER:
2014                 /* Can be set only for TAPs */
2015                 ret = -EINVAL;
2016                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2017                         break;
2018                 ret = update_filter(&tun->txflt, (void __user *)arg);
2019                 break;
2020
2021         case SIOCGIFHWADDR:
2022                 /* Get hw address */
2023                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2024                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2025                 if (copy_to_user(argp, &ifr, ifreq_len))
2026                         ret = -EFAULT;
2027                 break;
2028
2029         case SIOCSIFHWADDR:
2030                 /* Set hw address */
2031                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2032                           ifr.ifr_hwaddr.sa_data);
2033
2034                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2035                 break;
2036
2037         case TUNGETSNDBUF:
2038                 sndbuf = tfile->socket.sk->sk_sndbuf;
2039                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2040                         ret = -EFAULT;
2041                 break;
2042
2043         case TUNSETSNDBUF:
2044                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2045                         ret = -EFAULT;
2046                         break;
2047                 }
2048
2049                 tun->sndbuf = sndbuf;
2050                 tun_set_sndbuf(tun);
2051                 break;
2052
2053         case TUNGETVNETHDRSZ:
2054                 vnet_hdr_sz = tun->vnet_hdr_sz;
2055                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2056                         ret = -EFAULT;
2057                 break;
2058
2059         case TUNSETVNETHDRSZ:
2060                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2061                         ret = -EFAULT;
2062                         break;
2063                 }
2064                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2065                         ret = -EINVAL;
2066                         break;
2067                 }
2068
2069                 tun->vnet_hdr_sz = vnet_hdr_sz;
2070                 break;
2071
2072         case TUNATTACHFILTER:
2073                 /* Can be set only for TAPs */
2074                 ret = -EINVAL;
2075                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2076                         break;
2077                 ret = -EFAULT;
2078                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2079                         break;
2080
2081                 ret = tun_attach_filter(tun);
2082                 break;
2083
2084         case TUNDETACHFILTER:
2085                 /* Can be set only for TAPs */
2086                 ret = -EINVAL;
2087                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2088                         break;
2089                 ret = 0;
2090                 tun_detach_filter(tun, tun->numqueues);
2091                 break;
2092
2093         case TUNGETFILTER:
2094                 ret = -EINVAL;
2095                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2096                         break;
2097                 ret = -EFAULT;
2098                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2099                         break;
2100                 ret = 0;
2101                 break;
2102
2103         default:
2104                 ret = -EINVAL;
2105                 break;
2106         }
2107
2108 unlock:
2109         rtnl_unlock();
2110         if (tun)
2111                 tun_put(tun);
2112         return ret;
2113 }
2114
2115 static long tun_chr_ioctl(struct file *file,
2116                           unsigned int cmd, unsigned long arg)
2117 {
2118         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2119 }
2120
2121 #ifdef CONFIG_COMPAT
2122 static long tun_chr_compat_ioctl(struct file *file,
2123                          unsigned int cmd, unsigned long arg)
2124 {
2125         switch (cmd) {
2126         case TUNSETIFF:
2127         case TUNGETIFF:
2128         case TUNSETTXFILTER:
2129         case TUNGETSNDBUF:
2130         case TUNSETSNDBUF:
2131         case SIOCGIFHWADDR:
2132         case SIOCSIFHWADDR:
2133                 arg = (unsigned long)compat_ptr(arg);
2134                 break;
2135         default:
2136                 arg = (compat_ulong_t)arg;
2137                 break;
2138         }
2139
2140         /*
2141          * compat_ifreq is shorter than ifreq, so we must not access beyond
2142          * the end of that structure. All fields that are used in this
2143          * driver are compatible though, we don't need to convert the
2144          * contents.
2145          */
2146         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2147 }
2148 #endif /* CONFIG_COMPAT */
2149
2150 static int tun_chr_fasync(int fd, struct file *file, int on)
2151 {
2152         struct tun_file *tfile = file->private_data;
2153         int ret;
2154
2155         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2156                 goto out;
2157
2158         if (on) {
2159                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2160                 tfile->flags |= TUN_FASYNC;
2161         } else
2162                 tfile->flags &= ~TUN_FASYNC;
2163         ret = 0;
2164 out:
2165         return ret;
2166 }
2167
2168 static int tun_chr_open(struct inode *inode, struct file * file)
2169 {
2170         struct tun_file *tfile;
2171
2172         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2173
2174         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2175                                             &tun_proto);
2176         if (!tfile)
2177                 return -ENOMEM;
2178         RCU_INIT_POINTER(tfile->tun, NULL);
2179         tfile->net = get_net(current->nsproxy->net_ns);
2180         tfile->flags = 0;
2181         tfile->ifindex = 0;
2182
2183         init_waitqueue_head(&tfile->wq.wait);
2184         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2185
2186         tfile->socket.file = file;
2187         tfile->socket.ops = &tun_socket_ops;
2188
2189         sock_init_data(&tfile->socket, &tfile->sk);
2190         sk_change_net(&tfile->sk, tfile->net);
2191
2192         tfile->sk.sk_write_space = tun_sock_write_space;
2193         tfile->sk.sk_sndbuf = INT_MAX;
2194
2195         file->private_data = tfile;
2196         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2197         INIT_LIST_HEAD(&tfile->next);
2198
2199         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2200
2201         return 0;
2202 }
2203
2204 static int tun_chr_close(struct inode *inode, struct file *file)
2205 {
2206         struct tun_file *tfile = file->private_data;
2207         struct net *net = tfile->net;
2208
2209         tun_detach(tfile, true);
2210         put_net(net);
2211
2212         return 0;
2213 }
2214
2215 #ifdef CONFIG_PROC_FS
2216 static int tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2217 {
2218         struct tun_struct *tun;
2219         struct ifreq ifr;
2220
2221         memset(&ifr, 0, sizeof(ifr));
2222
2223         rtnl_lock();
2224         tun = tun_get(f);
2225         if (tun)
2226                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2227         rtnl_unlock();
2228
2229         if (tun)
2230                 tun_put(tun);
2231
2232         return seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2233 }
2234 #endif
2235
2236 static const struct file_operations tun_fops = {
2237         .owner  = THIS_MODULE,
2238         .llseek = no_llseek,
2239         .read  = new_sync_read,
2240         .write = new_sync_write,
2241         .read_iter  = tun_chr_read_iter,
2242         .write_iter = tun_chr_write_iter,
2243         .poll   = tun_chr_poll,
2244         .unlocked_ioctl = tun_chr_ioctl,
2245 #ifdef CONFIG_COMPAT
2246         .compat_ioctl = tun_chr_compat_ioctl,
2247 #endif
2248         .open   = tun_chr_open,
2249         .release = tun_chr_close,
2250         .fasync = tun_chr_fasync,
2251 #ifdef CONFIG_PROC_FS
2252         .show_fdinfo = tun_chr_show_fdinfo,
2253 #endif
2254 };
2255
2256 static struct miscdevice tun_miscdev = {
2257         .minor = TUN_MINOR,
2258         .name = "tun",
2259         .nodename = "net/tun",
2260         .fops = &tun_fops,
2261 };
2262
2263 /* ethtool interface */
2264
2265 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2266 {
2267         cmd->supported          = 0;
2268         cmd->advertising        = 0;
2269         ethtool_cmd_speed_set(cmd, SPEED_10);
2270         cmd->duplex             = DUPLEX_FULL;
2271         cmd->port               = PORT_TP;
2272         cmd->phy_address        = 0;
2273         cmd->transceiver        = XCVR_INTERNAL;
2274         cmd->autoneg            = AUTONEG_DISABLE;
2275         cmd->maxtxpkt           = 0;
2276         cmd->maxrxpkt           = 0;
2277         return 0;
2278 }
2279
2280 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2281 {
2282         struct tun_struct *tun = netdev_priv(dev);
2283
2284         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2285         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2286
2287         switch (tun->flags & TUN_TYPE_MASK) {
2288         case TUN_TUN_DEV:
2289                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2290                 break;
2291         case TUN_TAP_DEV:
2292                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2293                 break;
2294         }
2295 }
2296
2297 static u32 tun_get_msglevel(struct net_device *dev)
2298 {
2299 #ifdef TUN_DEBUG
2300         struct tun_struct *tun = netdev_priv(dev);
2301         return tun->debug;
2302 #else
2303         return -EOPNOTSUPP;
2304 #endif
2305 }
2306
2307 static void tun_set_msglevel(struct net_device *dev, u32 value)
2308 {
2309 #ifdef TUN_DEBUG
2310         struct tun_struct *tun = netdev_priv(dev);
2311         tun->debug = value;
2312 #endif
2313 }
2314
2315 static const struct ethtool_ops tun_ethtool_ops = {
2316         .get_settings   = tun_get_settings,
2317         .get_drvinfo    = tun_get_drvinfo,
2318         .get_msglevel   = tun_get_msglevel,
2319         .set_msglevel   = tun_set_msglevel,
2320         .get_link       = ethtool_op_get_link,
2321         .get_ts_info    = ethtool_op_get_ts_info,
2322 };
2323
2324
2325 static int __init tun_init(void)
2326 {
2327         int ret = 0;
2328
2329         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2330         pr_info("%s\n", DRV_COPYRIGHT);
2331
2332         ret = rtnl_link_register(&tun_link_ops);
2333         if (ret) {
2334                 pr_err("Can't register link_ops\n");
2335                 goto err_linkops;
2336         }
2337
2338         ret = misc_register(&tun_miscdev);
2339         if (ret) {
2340                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2341                 goto err_misc;
2342         }
2343         return  0;
2344 err_misc:
2345         rtnl_link_unregister(&tun_link_ops);
2346 err_linkops:
2347         return ret;
2348 }
2349
2350 static void tun_cleanup(void)
2351 {
2352         misc_deregister(&tun_miscdev);
2353         rtnl_link_unregister(&tun_link_ops);
2354 }
2355
2356 /* Get an underlying socket object from tun file.  Returns error unless file is
2357  * attached to a device.  The returned object works like a packet socket, it
2358  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2359  * holding a reference to the file for as long as the socket is in use. */
2360 struct socket *tun_get_socket(struct file *file)
2361 {
2362         struct tun_file *tfile;
2363         if (file->f_op != &tun_fops)
2364                 return ERR_PTR(-EINVAL);
2365         tfile = file->private_data;
2366         if (!tfile)
2367                 return ERR_PTR(-EBADFD);
2368         return &tfile->socket;
2369 }
2370 EXPORT_SYMBOL_GPL(tun_get_socket);
2371
2372 module_init(tun_init);
2373 module_exit(tun_cleanup);
2374 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2375 MODULE_AUTHOR(DRV_COPYRIGHT);
2376 MODULE_LICENSE("GPL");
2377 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2378 MODULE_ALIAS("devname:net/tun");