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/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 #include <linux/uio.h>
  74
  75 #include <asm/uaccess.h>
  76
  77 /* Uncomment to enable debugging */
  78 /* #define TUN_DEBUG 1 */
  79
  80 #ifdef TUN_DEBUG
  81 static int debug;
  82
  83 #define tun_debug(level, tun, fmt, args...)                     \
  84 do {                                                            \
  85         if (tun->debug)                                         \
  86                 netdev_printk(level, tun->dev, fmt, ##args);    \
  87 } while (0)
  88 #define DBG1(level, fmt, args...)                               \
  89 do {                                                            \
  90         if (debug == 2)                                         \
  91                 printk(level fmt, ##args);                      \
  92 } while (0)
  93 #else
  94 #define tun_debug(level, tun, fmt, args...)                     \
  95 do {                                                            \
  96         if (0)                                                  \
  97                 netdev_printk(level, tun->dev, fmt, ##args);    \
  98 } while (0)
  99 #define DBG1(level, fmt, args...)                               \
 100 do {                                                            \
 101         if (0)                                                  \
 102                 printk(level fmt, ##args);                      \
 103 } while (0)
 104 #endif
 105
 106 /* TUN device flags */
 107
 108 /* IFF_ATTACH_QUEUE is never stored in device flags,
 109  * overload it to mean fasync when stored there.
 110  */
 111 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 112 /* High bits in flags field are unused. */
 113 #define TUN_VNET_LE     0x80000000
 114 #define TUN_VNET_BE     0x40000000
 115
 116 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 117                       IFF_MULTI_QUEUE)
 118 #define GOODCOPY_LEN 128
 119
 120 #define FLT_EXACT_COUNT 8
 121 struct tap_filter {
 122         unsigned int    count;    /* Number of addrs. Zero means disabled */
 123         u32             mask[2];  /* Mask of the hashed addrs */
 124         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 125 };
 126
 127 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 128  * to max number of VCPUs in guest. */
 129 #define MAX_TAP_QUEUES 256
 130 #define MAX_TAP_FLOWS  4096
 131
 132 #define TUN_FLOW_EXPIRE (3 * HZ)
 133
 134 /* A tun_file connects an open character device to a tuntap netdevice. It
 135  * also contains all socket related structures (except sock_fprog and tap_filter)
 136  * to serve as one transmit queue for tuntap device. The sock_fprog and
 137  * tap_filter were kept in tun_struct since they were used for filtering for the
 138  * netdevice not for a specific queue (at least I didn't see the requirement for
 139  * this).
 140  *
 141  * RCU usage:
 142  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 143  * other can only be read while rcu_read_lock or rtnl_lock is held.
 144  */
 145 struct tun_file {
 146         struct sock sk;
 147         struct socket socket;
 148         struct socket_wq wq;
 149         struct tun_struct __rcu *tun;
 150         struct fasync_struct *fasync;
 151         /* only used for fasnyc */
 152         unsigned int flags;
 153         union {
 154                 u16 queue_index;
 155                 unsigned int ifindex;
 156         };
 157         struct list_head next;
 158         struct tun_struct *detached;
 159 };
 160
 161 struct tun_flow_entry {
 162         struct hlist_node hash_link;
 163         struct rcu_head rcu;
 164         struct tun_struct *tun;
 165
 166         u32 rxhash;
 167         u32 rps_rxhash;
 168         int queue_index;
 169         unsigned long updated;
 170 };
 171
 172 #define TUN_NUM_FLOW_ENTRIES 1024
 173
 174 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 175  * device, socket filter, sndbuf and vnet header size were restore when the
 176  * file were attached to a persist device.
 177  */
 178 struct tun_struct {
 179         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 180         unsigned int            numqueues;
 181         unsigned int            flags;
 182         kuid_t                  owner;
 183         kgid_t                  group;
 184
 185         struct net_device       *dev;
 186         netdev_features_t       set_features;
 187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 188                           NETIF_F_TSO6|NETIF_F_UFO)
 189
 190         int                     vnet_hdr_sz;
 191         int                     sndbuf;
 192         struct tap_filter       txflt;
 193         struct sock_fprog       fprog;
 194         /* protected by rtnl lock */
 195         bool                    filter_attached;
 196 #ifdef TUN_DEBUG
 197         int debug;
 198 #endif
 199         spinlock_t lock;
 200         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 201         struct timer_list flow_gc_timer;
 202         unsigned long ageing_time;
 203         unsigned int numdisabled;
 204         struct list_head disabled;
 205         void *security;
 206         u32 flow_count;
 207 };
 208
 209 #ifdef CONFIG_TUN_VNET_CROSS_LE
 210 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 211 {
 212         return tun->flags & TUN_VNET_BE ? false :
 213                 virtio_legacy_is_little_endian();
 214 }
 215
 216 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 217 {
 218         int be = !!(tun->flags & TUN_VNET_BE);
 219
 220         if (put_user(be, argp))
 221                 return -EFAULT;
 222
 223         return 0;
 224 }
 225
 226 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 227 {
 228         int be;
 229
 230         if (get_user(be, argp))
 231                 return -EFAULT;
 232
 233         if (be)
 234                 tun->flags |= TUN_VNET_BE;
 235         else
 236                 tun->flags &= ~TUN_VNET_BE;
 237
 238         return 0;
 239 }
 240 #else
 241 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 242 {
 243         return virtio_legacy_is_little_endian();
 244 }
 245
 246 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 247 {
 248         return -EINVAL;
 249 }
 250
 251 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 252 {
 253         return -EINVAL;
 254 }
 255 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 256
 257 static inline bool tun_is_little_endian(struct tun_struct *tun)
 258 {
 259         return tun->flags & TUN_VNET_LE ||
 260                 tun_legacy_is_little_endian(tun);
 261 }
 262
 263 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 264 {
 265         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 266 }
 267
 268 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 269 {
 270         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 271 }
 272
 273 static inline u32 tun_hashfn(u32 rxhash)
 274 {
 275         return rxhash & 0x3ff;
 276 }
 277
 278 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 279 {
 280         struct tun_flow_entry *e;
 281
 282         hlist_for_each_entry_rcu(e, head, hash_link) {
 283                 if (e->rxhash == rxhash)
 284                         return e;
 285         }
 286         return NULL;
 287 }
 288
 289 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 290                                               struct hlist_head *head,
 291                                               u32 rxhash, u16 queue_index)
 292 {
 293         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 294
 295         if (e) {
 296                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 297                           rxhash, queue_index);
 298                 e->updated = jiffies;
 299                 e->rxhash = rxhash;
 300                 e->rps_rxhash = 0;
 301                 e->queue_index = queue_index;
 302                 e->tun = tun;
 303                 hlist_add_head_rcu(&e->hash_link, head);
 304                 ++tun->flow_count;
 305         }
 306         return e;
 307 }
 308
 309 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 310 {
 311         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 312                   e->rxhash, e->queue_index);
 313         hlist_del_rcu(&e->hash_link);
 314         kfree_rcu(e, rcu);
 315         --tun->flow_count;
 316 }
 317
 318 static void tun_flow_flush(struct tun_struct *tun)
 319 {
 320         int i;
 321
 322         spin_lock_bh(&tun->lock);
 323         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 324                 struct tun_flow_entry *e;
 325                 struct hlist_node *n;
 326
 327                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 328                         tun_flow_delete(tun, e);
 329         }
 330         spin_unlock_bh(&tun->lock);
 331 }
 332
 333 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 334 {
 335         int i;
 336
 337         spin_lock_bh(&tun->lock);
 338         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 339                 struct tun_flow_entry *e;
 340                 struct hlist_node *n;
 341
 342                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 343                         if (e->queue_index == queue_index)
 344                                 tun_flow_delete(tun, e);
 345                 }
 346         }
 347         spin_unlock_bh(&tun->lock);
 348 }
 349
 350 static void tun_flow_cleanup(unsigned long data)
 351 {
 352         struct tun_struct *tun = (struct tun_struct *)data;
 353         unsigned long delay = tun->ageing_time;
 354         unsigned long next_timer = jiffies + delay;
 355         unsigned long count = 0;
 356         int i;
 357
 358         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 359
 360         spin_lock_bh(&tun->lock);
 361         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 362                 struct tun_flow_entry *e;
 363                 struct hlist_node *n;
 364
 365                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 366                         unsigned long this_timer;
 367                         count++;
 368                         this_timer = e->updated + delay;
 369                         if (time_before_eq(this_timer, jiffies))
 370                                 tun_flow_delete(tun, e);
 371                         else if (time_before(this_timer, next_timer))
 372                                 next_timer = this_timer;
 373                 }
 374         }
 375
 376         if (count)
 377                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 378         spin_unlock_bh(&tun->lock);
 379 }
 380
 381 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 382                             struct tun_file *tfile)
 383 {
 384         struct hlist_head *head;
 385         struct tun_flow_entry *e;
 386         unsigned long delay = tun->ageing_time;
 387         u16 queue_index = tfile->queue_index;
 388
 389         if (!rxhash)
 390                 return;
 391         else
 392                 head = &tun->flows[tun_hashfn(rxhash)];
 393
 394         rcu_read_lock();
 395
 396         /* We may get a very small possibility of OOO during switching, not
 397          * worth to optimize.*/
 398         if (tun->numqueues == 1 || tfile->detached)
 399                 goto unlock;
 400
 401         e = tun_flow_find(head, rxhash);
 402         if (likely(e)) {
 403                 /* TODO: keep queueing to old queue until it's empty? */
 404                 e->queue_index = queue_index;
 405                 e->updated = jiffies;
 406                 sock_rps_record_flow_hash(e->rps_rxhash);
 407         } else {
 408                 spin_lock_bh(&tun->lock);
 409                 if (!tun_flow_find(head, rxhash) &&
 410                     tun->flow_count < MAX_TAP_FLOWS)
 411                         tun_flow_create(tun, head, rxhash, queue_index);
 412
 413                 if (!timer_pending(&tun->flow_gc_timer))
 414                         mod_timer(&tun->flow_gc_timer,
 415                                   round_jiffies_up(jiffies + delay));
 416                 spin_unlock_bh(&tun->lock);
 417         }
 418
 419 unlock:
 420         rcu_read_unlock();
 421 }
 422
 423 /**
 424  * Save the hash received in the stack receive path and update the
 425  * flow_hash table accordingly.
 426  */
 427 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 428 {
 429         if (unlikely(e->rps_rxhash != hash))
 430                 e->rps_rxhash = hash;
 431 }
 432
 433 /* We try to identify a flow through its rxhash first. The reason that
 434  * we do not check rxq no. is because some cards(e.g 82599), chooses
 435  * the rxq based on the txq where the last packet of the flow comes. As
 436  * the userspace application move between processors, we may get a
 437  * different rxq no. here. If we could not get rxhash, then we would
 438  * hope the rxq no. may help here.
 439  */
 440 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 441                             void *accel_priv, select_queue_fallback_t fallback)
 442 {
 443         struct tun_struct *tun = netdev_priv(dev);
 444         struct tun_flow_entry *e;
 445         u32 txq = 0;
 446         u32 numqueues = 0;
 447
 448         rcu_read_lock();
 449         numqueues = ACCESS_ONCE(tun->numqueues);
 450
 451         txq = skb_get_hash(skb);
 452         if (txq) {
 453                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 454                 if (e) {
 455                         tun_flow_save_rps_rxhash(e, txq);
 456                         txq = e->queue_index;
 457                 } else
 458                         /* use multiply and shift instead of expensive divide */
 459                         txq = ((u64)txq * numqueues) >> 32;
 460         } else if (likely(skb_rx_queue_recorded(skb))) {
 461                 txq = skb_get_rx_queue(skb);
 462                 while (unlikely(txq >= numqueues))
 463                         txq -= numqueues;
 464         }
 465
 466         rcu_read_unlock();
 467         return txq;
 468 }
 469
 470 static inline bool tun_not_capable(struct tun_struct *tun)
 471 {
 472         const struct cred *cred = current_cred();
 473         struct net *net = dev_net(tun->dev);
 474
 475         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 476                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 477                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 478 }
 479
 480 static void tun_set_real_num_queues(struct tun_struct *tun)
 481 {
 482         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 483         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 484 }
 485
 486 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 487 {
 488         tfile->detached = tun;
 489         list_add_tail(&tfile->next, &tun->disabled);
 490         ++tun->numdisabled;
 491 }
 492
 493 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 494 {
 495         struct tun_struct *tun = tfile->detached;
 496
 497         tfile->detached = NULL;
 498         list_del_init(&tfile->next);
 499         --tun->numdisabled;
 500         return tun;
 501 }
 502
 503 static void tun_queue_purge(struct tun_file *tfile)
 504 {
 505         skb_queue_purge(&tfile->sk.sk_receive_queue);
 506         skb_queue_purge(&tfile->sk.sk_error_queue);
 507 }
 508
 509 static void __tun_detach(struct tun_file *tfile, bool clean)
 510 {
 511         struct tun_file *ntfile;
 512         struct tun_struct *tun;
 513
 514         tun = rtnl_dereference(tfile->tun);
 515
 516         if (tun && !tfile->detached) {
 517                 u16 index = tfile->queue_index;
 518                 BUG_ON(index >= tun->numqueues);
 519
 520                 rcu_assign_pointer(tun->tfiles[index],
 521                                    tun->tfiles[tun->numqueues - 1]);
 522                 ntfile = rtnl_dereference(tun->tfiles[index]);
 523                 ntfile->queue_index = index;
 524
 525                 --tun->numqueues;
 526                 if (clean) {
 527                         RCU_INIT_POINTER(tfile->tun, NULL);
 528                         sock_put(&tfile->sk);
 529                 } else
 530                         tun_disable_queue(tun, tfile);
 531
 532                 synchronize_net();
 533                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 534                 /* Drop read queue */
 535                 tun_queue_purge(tfile);
 536                 tun_set_real_num_queues(tun);
 537         } else if (tfile->detached && clean) {
 538                 tun = tun_enable_queue(tfile);
 539                 sock_put(&tfile->sk);
 540         }
 541
 542         if (clean) {
 543                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 544                         netif_carrier_off(tun->dev);
 545
 546                         if (!(tun->flags & IFF_PERSIST) &&
 547                             tun->dev->reg_state == NETREG_REGISTERED)
 548                                 unregister_netdevice(tun->dev);
 549                 }
 550                 sock_put(&tfile->sk);
 551         }
 552 }
 553
 554 static void tun_detach(struct tun_file *tfile, bool clean)
 555 {
 556         rtnl_lock();
 557         __tun_detach(tfile, clean);
 558         rtnl_unlock();
 559 }
 560
 561 static void tun_detach_all(struct net_device *dev)
 562 {
 563         struct tun_struct *tun = netdev_priv(dev);
 564         struct tun_file *tfile, *tmp;
 565         int i, n = tun->numqueues;
 566
 567         for (i = 0; i < n; i++) {
 568                 tfile = rtnl_dereference(tun->tfiles[i]);
 569                 BUG_ON(!tfile);
 570                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 571                 RCU_INIT_POINTER(tfile->tun, NULL);
 572                 --tun->numqueues;
 573         }
 574         list_for_each_entry(tfile, &tun->disabled, next) {
 575                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 576                 RCU_INIT_POINTER(tfile->tun, NULL);
 577         }
 578         BUG_ON(tun->numqueues != 0);
 579
 580         synchronize_net();
 581         for (i = 0; i < n; i++) {
 582                 tfile = rtnl_dereference(tun->tfiles[i]);
 583                 /* Drop read queue */
 584                 tun_queue_purge(tfile);
 585                 sock_put(&tfile->sk);
 586         }
 587         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 588                 tun_enable_queue(tfile);
 589                 tun_queue_purge(tfile);
 590                 sock_put(&tfile->sk);
 591         }
 592         BUG_ON(tun->numdisabled != 0);
 593
 594         if (tun->flags & IFF_PERSIST)
 595                 module_put(THIS_MODULE);
 596 }
 597
 598 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 599 {
 600         struct tun_file *tfile = file->private_data;
 601         int err;
 602
 603         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 604         if (err < 0)
 605                 goto out;
 606
 607         err = -EINVAL;
 608         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 609                 goto out;
 610
 611         err = -EBUSY;
 612         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 613                 goto out;
 614
 615         err = -E2BIG;
 616         if (!tfile->detached &&
 617             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 618                 goto out;
 619
 620         err = 0;
 621
 622         /* Re-attach the filter to persist device */
 623         if (!skip_filter && (tun->filter_attached == true)) {
 624                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 625                 if (!err)
 626                         goto out;
 627         }
 628         tfile->queue_index = tun->numqueues;
 629         rcu_assign_pointer(tfile->tun, tun);
 630         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 631         tun->numqueues++;
 632
 633         if (tfile->detached)
 634                 tun_enable_queue(tfile);
 635         else
 636                 sock_hold(&tfile->sk);
 637
 638         tun_set_real_num_queues(tun);
 639
 640         /* device is allowed to go away first, so no need to hold extra
 641          * refcnt.
 642          */
 643
 644 out:
 645         return err;
 646 }
 647
 648 static struct tun_struct *__tun_get(struct tun_file *tfile)
 649 {
 650         struct tun_struct *tun;
 651
 652         rcu_read_lock();
 653         tun = rcu_dereference(tfile->tun);
 654         if (tun)
 655                 dev_hold(tun->dev);
 656         rcu_read_unlock();
 657
 658         return tun;
 659 }
 660
 661 static struct tun_struct *tun_get(struct file *file)
 662 {
 663         return __tun_get(file->private_data);
 664 }
 665
 666 static void tun_put(struct tun_struct *tun)
 667 {
 668         dev_put(tun->dev);
 669 }
 670
 671 /* TAP filtering */
 672 static void addr_hash_set(u32 *mask, const u8 *addr)
 673 {
 674         int n = ether_crc(ETH_ALEN, addr) >> 26;
 675         mask[n >> 5] |= (1 << (n & 31));
 676 }
 677
 678 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 679 {
 680         int n = ether_crc(ETH_ALEN, addr) >> 26;
 681         return mask[n >> 5] & (1 << (n & 31));
 682 }
 683
 684 static int update_filter(struct tap_filter *filter, void __user *arg)
 685 {
 686         struct { u8 u[ETH_ALEN]; } *addr;
 687         struct tun_filter uf;
 688         int err, alen, n, nexact;
 689
 690         if (copy_from_user(&uf, arg, sizeof(uf)))
 691                 return -EFAULT;
 692
 693         if (!uf.count) {
 694                 /* Disabled */
 695                 filter->count = 0;
 696                 return 0;
 697         }
 698
 699         alen = ETH_ALEN * uf.count;
 700         addr = kmalloc(alen, GFP_KERNEL);
 701         if (!addr)
 702                 return -ENOMEM;
 703
 704         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 705                 err = -EFAULT;
 706                 goto done;
 707         }
 708
 709         /* The filter is updated without holding any locks. Which is
 710          * perfectly safe. We disable it first and in the worst
 711          * case we'll accept a few undesired packets. */
 712         filter->count = 0;
 713         wmb();
 714
 715         /* Use first set of addresses as an exact filter */
 716         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 717                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 718
 719         nexact = n;
 720
 721         /* Remaining multicast addresses are hashed,
 722          * unicast will leave the filter disabled. */
 723         memset(filter->mask, 0, sizeof(filter->mask));
 724         for (; n < uf.count; n++) {
 725                 if (!is_multicast_ether_addr(addr[n].u)) {
 726                         err = 0; /* no filter */
 727                         goto done;
 728                 }
 729                 addr_hash_set(filter->mask, addr[n].u);
 730         }
 731
 732         /* For ALLMULTI just set the mask to all ones.
 733          * This overrides the mask populated above. */
 734         if ((uf.flags & TUN_FLT_ALLMULTI))
 735                 memset(filter->mask, ~0, sizeof(filter->mask));
 736
 737         /* Now enable the filter */
 738         wmb();
 739         filter->count = nexact;
 740
 741         /* Return the number of exact filters */
 742         err = nexact;
 743
 744 done:
 745         kfree(addr);
 746         return err;
 747 }
 748
 749 /* Returns: 0 - drop, !=0 - accept */
 750 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 751 {
 752         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 753          * at this point. */
 754         struct ethhdr *eh = (struct ethhdr *) skb->data;
 755         int i;
 756
 757         /* Exact match */
 758         for (i = 0; i < filter->count; i++)
 759                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 760                         return 1;
 761
 762         /* Inexact match (multicast only) */
 763         if (is_multicast_ether_addr(eh->h_dest))
 764                 return addr_hash_test(filter->mask, eh->h_dest);
 765
 766         return 0;
 767 }
 768
 769 /*
 770  * Checks whether the packet is accepted or not.
 771  * Returns: 0 - drop, !=0 - accept
 772  */
 773 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 774 {
 775         if (!filter->count)
 776                 return 1;
 777
 778         return run_filter(filter, skb);
 779 }
 780
 781 /* Network device part of the driver */
 782
 783 static const struct ethtool_ops tun_ethtool_ops;
 784
 785 /* Net device detach from fd. */
 786 static void tun_net_uninit(struct net_device *dev)
 787 {
 788         tun_detach_all(dev);
 789 }
 790
 791 /* Net device open. */
 792 static int tun_net_open(struct net_device *dev)
 793 {
 794         netif_tx_start_all_queues(dev);
 795         return 0;
 796 }
 797
 798 /* Net device close. */
 799 static int tun_net_close(struct net_device *dev)
 800 {
 801         netif_tx_stop_all_queues(dev);
 802         return 0;
 803 }
 804
 805 /* Net device start xmit */
 806 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 807 {
 808         struct tun_struct *tun = netdev_priv(dev);
 809         int txq = skb->queue_mapping;
 810         struct tun_file *tfile;
 811         u32 numqueues = 0;
 812
 813         rcu_read_lock();
 814         tfile = rcu_dereference(tun->tfiles[txq]);
 815         numqueues = ACCESS_ONCE(tun->numqueues);
 816
 817         /* Drop packet if interface is not attached */
 818         if (txq >= numqueues)
 819                 goto drop;
 820
 821         if (numqueues == 1) {
 822                 /* Select queue was not called for the skbuff, so we extract the
 823                  * RPS hash and save it into the flow_table here.
 824                  */
 825                 __u32 rxhash;
 826
 827                 rxhash = skb_get_hash(skb);
 828                 if (rxhash) {
 829                         struct tun_flow_entry *e;
 830                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 831                                         rxhash);
 832                         if (e)
 833                                 tun_flow_save_rps_rxhash(e, rxhash);
 834                 }
 835         }
 836
 837         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 838
 839         BUG_ON(!tfile);
 840
 841         /* Drop if the filter does not like it.
 842          * This is a noop if the filter is disabled.
 843          * Filter can be enabled only for the TAP devices. */
 844         if (!check_filter(&tun->txflt, skb))
 845                 goto drop;
 846
 847         if (tfile->socket.sk->sk_filter &&
 848             sk_filter(tfile->socket.sk, skb))
 849                 goto drop;
 850
 851         /* Limit the number of packets queued by dividing txq length with the
 852          * number of queues.
 853          */
 854         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
 855                           >= dev->tx_queue_len)
 856                 goto drop;
 857
 858         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 859                 goto drop;
 860
 861         if (skb->sk && sk_fullsock(skb->sk)) {
 862                 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
 863                 sw_tx_timestamp(skb);
 864         }
 865
 866         /* Orphan the skb - required as we might hang on to it
 867          * for indefinite time.
 868          */
 869         skb_orphan(skb);
 870
 871         nf_reset(skb);
 872
 873         /* Enqueue packet */
 874         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 875
 876         /* Notify and wake up reader process */
 877         if (tfile->flags & TUN_FASYNC)
 878                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 879         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 880
 881         rcu_read_unlock();
 882         return NETDEV_TX_OK;
 883
 884 drop:
 885         dev->stats.tx_dropped++;
 886         skb_tx_error(skb);
 887         kfree_skb(skb);
 888         rcu_read_unlock();
 889         return NET_XMIT_DROP;
 890 }
 891
 892 static void tun_net_mclist(struct net_device *dev)
 893 {
 894         /*
 895          * This callback is supposed to deal with mc filter in
 896          * _rx_ path and has nothing to do with the _tx_ path.
 897          * In rx path we always accept everything userspace gives us.
 898          */
 899 }
 900
 901 #define MIN_MTU 68
 902 #define MAX_MTU 65535
 903
 904 static int
 905 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 906 {
 907         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 908                 return -EINVAL;
 909         dev->mtu = new_mtu;
 910         return 0;
 911 }
 912
 913 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 914         netdev_features_t features)
 915 {
 916         struct tun_struct *tun = netdev_priv(dev);
 917
 918         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 919 }
 920 #ifdef CONFIG_NET_POLL_CONTROLLER
 921 static void tun_poll_controller(struct net_device *dev)
 922 {
 923         /*
 924          * Tun only receives frames when:
 925          * 1) the char device endpoint gets data from user space
 926          * 2) the tun socket gets a sendmsg call from user space
 927          * Since both of those are synchronous operations, we are guaranteed
 928          * never to have pending data when we poll for it
 929          * so there is nothing to do here but return.
 930          * We need this though so netpoll recognizes us as an interface that
 931          * supports polling, which enables bridge devices in virt setups to
 932          * still use netconsole
 933          */
 934         return;
 935 }
 936 #endif
 937 static const struct net_device_ops tun_netdev_ops = {
 938         .ndo_uninit             = tun_net_uninit,
 939         .ndo_open               = tun_net_open,
 940         .ndo_stop               = tun_net_close,
 941         .ndo_start_xmit         = tun_net_xmit,
 942         .ndo_change_mtu         = tun_net_change_mtu,
 943         .ndo_fix_features       = tun_net_fix_features,
 944         .ndo_select_queue       = tun_select_queue,
 945 #ifdef CONFIG_NET_POLL_CONTROLLER
 946         .ndo_poll_controller    = tun_poll_controller,
 947 #endif
 948 };
 949
 950 static const struct net_device_ops tap_netdev_ops = {
 951         .ndo_uninit             = tun_net_uninit,
 952         .ndo_open               = tun_net_open,
 953         .ndo_stop               = tun_net_close,
 954         .ndo_start_xmit         = tun_net_xmit,
 955         .ndo_change_mtu         = tun_net_change_mtu,
 956         .ndo_fix_features       = tun_net_fix_features,
 957         .ndo_set_rx_mode        = tun_net_mclist,
 958         .ndo_set_mac_address    = eth_mac_addr,
 959         .ndo_validate_addr      = eth_validate_addr,
 960         .ndo_select_queue       = tun_select_queue,
 961 #ifdef CONFIG_NET_POLL_CONTROLLER
 962         .ndo_poll_controller    = tun_poll_controller,
 963 #endif
 964         .ndo_features_check     = passthru_features_check,
 965 };
 966
 967 static void tun_flow_init(struct tun_struct *tun)
 968 {
 969         int i;
 970
 971         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 972                 INIT_HLIST_HEAD(&tun->flows[i]);
 973
 974         tun->ageing_time = TUN_FLOW_EXPIRE;
 975         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 976         mod_timer(&tun->flow_gc_timer,
 977                   round_jiffies_up(jiffies + tun->ageing_time));
 978 }
 979
 980 static void tun_flow_uninit(struct tun_struct *tun)
 981 {
 982         del_timer_sync(&tun->flow_gc_timer);
 983         tun_flow_flush(tun);
 984 }
 985
 986 /* Initialize net device. */
 987 static void tun_net_init(struct net_device *dev)
 988 {
 989         struct tun_struct *tun = netdev_priv(dev);
 990
 991         switch (tun->flags & TUN_TYPE_MASK) {
 992         case IFF_TUN:
 993                 dev->netdev_ops = &tun_netdev_ops;
 994
 995                 /* Point-to-Point TUN Device */
 996                 dev->hard_header_len = 0;
 997                 dev->addr_len = 0;
 998                 dev->mtu = 1500;
 999
1000                 /* Zero header length */
1001                 dev->type = ARPHRD_NONE;
1002                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1003                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
1004                 break;
1005
1006         case IFF_TAP:
1007                 dev->netdev_ops = &tap_netdev_ops;
1008                 /* Ethernet TAP Device */
1009                 ether_setup(dev);
1010                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1011                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1012
1013                 eth_hw_addr_random(dev);
1014
1015                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
1016                 break;
1017         }
1018 }
1019
1020 /* Character device part */
1021
1022 /* Poll */
1023 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1024 {
1025         struct tun_file *tfile = file->private_data;
1026         struct tun_struct *tun = __tun_get(tfile);
1027         struct sock *sk;
1028         unsigned int mask = 0;
1029
1030         if (!tun)
1031                 return POLLERR;
1032
1033         sk = tfile->socket.sk;
1034
1035         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1036
1037         poll_wait(file, sk_sleep(sk), wait);
1038
1039         if (!skb_queue_empty(&sk->sk_receive_queue))
1040                 mask |= POLLIN | POLLRDNORM;
1041
1042         if (sock_writeable(sk) ||
1043             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1044              sock_writeable(sk)))
1045                 mask |= POLLOUT | POLLWRNORM;
1046
1047         if (tun->dev->reg_state != NETREG_REGISTERED)
1048                 mask = POLLERR;
1049
1050         tun_put(tun);
1051         return mask;
1052 }
1053
1054 /* prepad is the amount to reserve at front.  len is length after that.
1055  * linear is a hint as to how much to copy (usually headers). */
1056 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1057                                      size_t prepad, size_t len,
1058                                      size_t linear, int noblock)
1059 {
1060         struct sock *sk = tfile->socket.sk;
1061         struct sk_buff *skb;
1062         int err;
1063
1064         /* Under a page?  Don't bother with paged skb. */
1065         if (prepad + len < PAGE_SIZE || !linear)
1066                 linear = len;
1067
1068         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1069                                    &err, 0);
1070         if (!skb)
1071                 return ERR_PTR(err);
1072
1073         skb_reserve(skb, prepad);
1074         skb_put(skb, linear);
1075         skb->data_len = len - linear;
1076         skb->len += len - linear;
1077
1078         return skb;
1079 }
1080
1081 /* Get packet from user space buffer */
1082 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1083                             void *msg_control, struct iov_iter *from,
1084                             int noblock)
1085 {
1086         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1087         struct sk_buff *skb;
1088         size_t total_len = iov_iter_count(from);
1089         size_t len = total_len, align = NET_SKB_PAD, linear;
1090         struct virtio_net_hdr gso = { 0 };
1091         int good_linear;
1092         int copylen;
1093         bool zerocopy = false;
1094         int err;
1095         u32 rxhash;
1096         ssize_t n;
1097
1098         if (!(tun->flags & IFF_NO_PI)) {
1099                 if (len < sizeof(pi))
1100                         return -EINVAL;
1101                 len -= sizeof(pi);
1102
1103                 n = copy_from_iter(&pi, sizeof(pi), from);
1104                 if (n != sizeof(pi))
1105                         return -EFAULT;
1106         }
1107
1108         if (tun->flags & IFF_VNET_HDR) {
1109                 if (len < tun->vnet_hdr_sz)
1110                         return -EINVAL;
1111                 len -= tun->vnet_hdr_sz;
1112
1113                 n = copy_from_iter(&gso, sizeof(gso), from);
1114                 if (n != sizeof(gso))
1115                         return -EFAULT;
1116
1117                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1118                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1119                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1120
1121                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1122                         return -EINVAL;
1123                 iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1124         }
1125
1126         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1127                 align += NET_IP_ALIGN;
1128                 if (unlikely(len < ETH_HLEN ||
1129                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1130                         return -EINVAL;
1131         }
1132
1133         good_linear = SKB_MAX_HEAD(align);
1134
1135         if (msg_control) {
1136                 struct iov_iter i = *from;
1137
1138                 /* There are 256 bytes to be copied in skb, so there is
1139                  * enough room for skb expand head in case it is used.
1140                  * The rest of the buffer is mapped from userspace.
1141                  */
1142                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1143                 if (copylen > good_linear)
1144                         copylen = good_linear;
1145                 linear = copylen;
1146                 iov_iter_advance(&i, copylen);
1147                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1148                         zerocopy = true;
1149         }
1150
1151         if (!zerocopy) {
1152                 copylen = len;
1153                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1154                         linear = good_linear;
1155                 else
1156                         linear = tun16_to_cpu(tun, gso.hdr_len);
1157         }
1158
1159         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1160         if (IS_ERR(skb)) {
1161                 if (PTR_ERR(skb) != -EAGAIN)
1162                         tun->dev->stats.rx_dropped++;
1163                 return PTR_ERR(skb);
1164         }
1165
1166         if (zerocopy)
1167                 err = zerocopy_sg_from_iter(skb, from);
1168         else {
1169                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1170                 if (!err && msg_control) {
1171                         struct ubuf_info *uarg = msg_control;
1172                         uarg->callback(uarg, false);
1173                 }
1174         }
1175
1176         if (err) {
1177                 tun->dev->stats.rx_dropped++;
1178                 kfree_skb(skb);
1179                 return -EFAULT;
1180         }
1181
1182         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1183                 if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start),
1184                                           tun16_to_cpu(tun, gso.csum_offset))) {
1185                         tun->dev->stats.rx_frame_errors++;
1186                         kfree_skb(skb);
1187                         return -EINVAL;
1188                 }
1189         }
1190
1191         switch (tun->flags & TUN_TYPE_MASK) {
1192         case IFF_TUN:
1193                 if (tun->flags & IFF_NO_PI) {
1194                         switch (skb->data[0] & 0xf0) {
1195                         case 0x40:
1196                                 pi.proto = htons(ETH_P_IP);
1197                                 break;
1198                         case 0x60:
1199                                 pi.proto = htons(ETH_P_IPV6);
1200                                 break;
1201                         default:
1202                                 tun->dev->stats.rx_dropped++;
1203                                 kfree_skb(skb);
1204                                 return -EINVAL;
1205                         }
1206                 }
1207
1208                 skb_reset_mac_header(skb);
1209                 skb->protocol = pi.proto;
1210                 skb->dev = tun->dev;
1211                 break;
1212         case IFF_TAP:
1213                 skb->protocol = eth_type_trans(skb, tun->dev);
1214                 break;
1215         }
1216
1217         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1218                 pr_debug("GSO!\n");
1219                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1220                 case VIRTIO_NET_HDR_GSO_TCPV4:
1221                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1222                         break;
1223                 case VIRTIO_NET_HDR_GSO_TCPV6:
1224                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1225                         break;
1226                 case VIRTIO_NET_HDR_GSO_UDP:
1227                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1228                         break;
1229                 default:
1230                         tun->dev->stats.rx_frame_errors++;
1231                         kfree_skb(skb);
1232                         return -EINVAL;
1233                 }
1234
1235                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1236                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1237
1238                 skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size);
1239                 if (skb_shinfo(skb)->gso_size == 0) {
1240                         tun->dev->stats.rx_frame_errors++;
1241                         kfree_skb(skb);
1242                         return -EINVAL;
1243                 }
1244
1245                 /* Header must be checked, and gso_segs computed. */
1246                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1247                 skb_shinfo(skb)->gso_segs = 0;
1248         }
1249
1250         /* copy skb_ubuf_info for callback when skb has no error */
1251         if (zerocopy) {
1252                 skb_shinfo(skb)->destructor_arg = msg_control;
1253                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1254                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1255         }
1256
1257         skb_reset_network_header(skb);
1258         skb_probe_transport_header(skb, 0);
1259
1260         rxhash = skb_get_hash(skb);
1261         netif_rx_ni(skb);
1262
1263         tun->dev->stats.rx_packets++;
1264         tun->dev->stats.rx_bytes += len;
1265
1266         tun_flow_update(tun, rxhash, tfile);
1267         return total_len;
1268 }
1269
1270 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1271 {
1272         struct file *file = iocb->ki_filp;
1273         struct tun_struct *tun = tun_get(file);
1274         struct tun_file *tfile = file->private_data;
1275         ssize_t result;
1276
1277         if (!tun)
1278                 return -EBADFD;
1279
1280         result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1281
1282         tun_put(tun);
1283         return result;
1284 }
1285
1286 /* Put packet to the user space buffer */
1287 static ssize_t tun_put_user(struct tun_struct *tun,
1288                             struct tun_file *tfile,
1289                             struct sk_buff *skb,
1290                             struct iov_iter *iter)
1291 {
1292         struct tun_pi pi = { 0, skb->protocol };
1293         ssize_t total;
1294         int vlan_offset = 0;
1295         int vlan_hlen = 0;
1296         int vnet_hdr_sz = 0;
1297
1298         if (skb_vlan_tag_present(skb))
1299                 vlan_hlen = VLAN_HLEN;
1300
1301         if (tun->flags & IFF_VNET_HDR)
1302                 vnet_hdr_sz = tun->vnet_hdr_sz;
1303
1304         total = skb->len + vlan_hlen + vnet_hdr_sz;
1305
1306         if (!(tun->flags & IFF_NO_PI)) {
1307                 if (iov_iter_count(iter) < sizeof(pi))
1308                         return -EINVAL;
1309
1310                 total += sizeof(pi);
1311                 if (iov_iter_count(iter) < total) {
1312                         /* Packet will be striped */
1313                         pi.flags |= TUN_PKT_STRIP;
1314                 }
1315
1316                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1317                         return -EFAULT;
1318         }
1319
1320         if (vnet_hdr_sz) {
1321                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1322                 if (iov_iter_count(iter) < vnet_hdr_sz)
1323                         return -EINVAL;
1324
1325                 if (skb_is_gso(skb)) {
1326                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1327
1328                         /* This is a hint as to how much should be linear. */
1329                         gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb));
1330                         gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size);
1331                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1332                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1333                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1334                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1335                         else if (sinfo->gso_type & SKB_GSO_UDP)
1336                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1337                         else {
1338                                 pr_err("unexpected GSO type: "
1339                                        "0x%x, gso_size %d, hdr_len %d\n",
1340                                        sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1341                                        tun16_to_cpu(tun, gso.hdr_len));
1342                                 print_hex_dump(KERN_ERR, "tun: ",
1343                                                DUMP_PREFIX_NONE,
1344                                                16, 1, skb->head,
1345                                                min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1346                                 WARN_ON_ONCE(1);
1347                                 return -EINVAL;
1348                         }
1349                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1350                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1351                 } else
1352                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1353
1354                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1355                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1356                         gso.csum_start = cpu_to_tun16(tun, skb_checksum_start_offset(skb) +
1357                                                       vlan_hlen);
1358                         gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset);
1359                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1360                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1361                 } /* else everything is zero */
1362
1363                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1364                         return -EFAULT;
1365
1366                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1367         }
1368
1369         if (vlan_hlen) {
1370                 int ret;
1371                 struct {
1372                         __be16 h_vlan_proto;
1373                         __be16 h_vlan_TCI;
1374                 } veth;
1375
1376                 veth.h_vlan_proto = skb->vlan_proto;
1377                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1378
1379                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1380
1381                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1382                 if (ret || !iov_iter_count(iter))
1383                         goto done;
1384
1385                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1386                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1387                         goto done;
1388         }
1389
1390         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1391
1392 done:
1393         tun->dev->stats.tx_packets++;
1394         tun->dev->stats.tx_bytes += skb->len + vlan_hlen;
1395
1396         return total;
1397 }
1398
1399 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1400                            struct iov_iter *to,
1401                            int noblock)
1402 {
1403         struct sk_buff *skb;
1404         ssize_t ret;
1405         int peeked, err, off = 0;
1406
1407         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1408
1409         if (!iov_iter_count(to))
1410                 return 0;
1411
1412         if (tun->dev->reg_state != NETREG_REGISTERED)
1413                 return -EIO;
1414
1415         /* Read frames from queue */
1416         skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1417                                   &peeked, &off, &err);
1418         if (!skb)
1419                 return err;
1420
1421         ret = tun_put_user(tun, tfile, skb, to);
1422         if (unlikely(ret < 0))
1423                 kfree_skb(skb);
1424         else
1425                 consume_skb(skb);
1426
1427         return ret;
1428 }
1429
1430 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1431 {
1432         struct file *file = iocb->ki_filp;
1433         struct tun_file *tfile = file->private_data;
1434         struct tun_struct *tun = __tun_get(tfile);
1435         ssize_t len = iov_iter_count(to), ret;
1436
1437         if (!tun)
1438                 return -EBADFD;
1439         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1440         ret = min_t(ssize_t, ret, len);
1441         if (ret > 0)
1442                 iocb->ki_pos = ret;
1443         tun_put(tun);
1444         return ret;
1445 }
1446
1447 static void tun_free_netdev(struct net_device *dev)
1448 {
1449         struct tun_struct *tun = netdev_priv(dev);
1450
1451         BUG_ON(!(list_empty(&tun->disabled)));
1452         tun_flow_uninit(tun);
1453         security_tun_dev_free_security(tun->security);
1454         free_netdev(dev);
1455 }
1456
1457 static void tun_setup(struct net_device *dev)
1458 {
1459         struct tun_struct *tun = netdev_priv(dev);
1460
1461         tun->owner = INVALID_UID;
1462         tun->group = INVALID_GID;
1463
1464         dev->ethtool_ops = &tun_ethtool_ops;
1465         dev->destructor = tun_free_netdev;
1466 }
1467
1468 /* Trivial set of netlink ops to allow deleting tun or tap
1469  * device with netlink.
1470  */
1471 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1472 {
1473         return -EINVAL;
1474 }
1475
1476 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1477         .kind           = DRV_NAME,
1478         .priv_size      = sizeof(struct tun_struct),
1479         .setup          = tun_setup,
1480         .validate       = tun_validate,
1481 };
1482
1483 static void tun_sock_write_space(struct sock *sk)
1484 {
1485         struct tun_file *tfile;
1486         wait_queue_head_t *wqueue;
1487
1488         if (!sock_writeable(sk))
1489                 return;
1490
1491         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1492                 return;
1493
1494         wqueue = sk_sleep(sk);
1495         if (wqueue && waitqueue_active(wqueue))
1496                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1497                                                 POLLWRNORM | POLLWRBAND);
1498
1499         tfile = container_of(sk, struct tun_file, sk);
1500         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1501 }
1502
1503 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1504 {
1505         int ret;
1506         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1507         struct tun_struct *tun = __tun_get(tfile);
1508
1509         if (!tun)
1510                 return -EBADFD;
1511
1512         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1513                            m->msg_flags & MSG_DONTWAIT);
1514         tun_put(tun);
1515         return ret;
1516 }
1517
1518 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1519                        int flags)
1520 {
1521         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1522         struct tun_struct *tun = __tun_get(tfile);
1523         int ret;
1524
1525         if (!tun)
1526                 return -EBADFD;
1527
1528         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1529                 ret = -EINVAL;
1530                 goto out;
1531         }
1532         if (flags & MSG_ERRQUEUE) {
1533                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1534                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1535                 goto out;
1536         }
1537         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1538         if (ret > (ssize_t)total_len) {
1539                 m->msg_flags |= MSG_TRUNC;
1540                 ret = flags & MSG_TRUNC ? ret : total_len;
1541         }
1542 out:
1543         tun_put(tun);
1544         return ret;
1545 }
1546
1547 /* Ops structure to mimic raw sockets with tun */
1548 static const struct proto_ops tun_socket_ops = {
1549         .sendmsg = tun_sendmsg,
1550         .recvmsg = tun_recvmsg,
1551 };
1552
1553 static struct proto tun_proto = {
1554         .name           = "tun",
1555         .owner          = THIS_MODULE,
1556         .obj_size       = sizeof(struct tun_file),
1557 };
1558
1559 static int tun_flags(struct tun_struct *tun)
1560 {
1561         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1562 }
1563
1564 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1565                               char *buf)
1566 {
1567         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1568         return sprintf(buf, "0x%x\n", tun_flags(tun));
1569 }
1570
1571 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1572                               char *buf)
1573 {
1574         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1575         return uid_valid(tun->owner)?
1576                 sprintf(buf, "%u\n",
1577                         from_kuid_munged(current_user_ns(), tun->owner)):
1578                 sprintf(buf, "-1\n");
1579 }
1580
1581 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1582                               char *buf)
1583 {
1584         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1585         return gid_valid(tun->group) ?
1586                 sprintf(buf, "%u\n",
1587                         from_kgid_munged(current_user_ns(), tun->group)):
1588                 sprintf(buf, "-1\n");
1589 }
1590
1591 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1592 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1593 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1594
1595 static struct attribute *tun_dev_attrs[] = {
1596         &dev_attr_tun_flags.attr,
1597         &dev_attr_owner.attr,
1598         &dev_attr_group.attr,
1599         NULL
1600 };
1601
1602 static const struct attribute_group tun_attr_group = {
1603         .attrs = tun_dev_attrs
1604 };
1605
1606 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1607 {
1608         struct tun_struct *tun;
1609         struct tun_file *tfile = file->private_data;
1610         struct net_device *dev;
1611         int err;
1612
1613         if (tfile->detached)
1614                 return -EINVAL;
1615
1616         dev = __dev_get_by_name(net, ifr->ifr_name);
1617         if (dev) {
1618                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1619                         return -EBUSY;
1620                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1621                         tun = netdev_priv(dev);
1622                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1623                         tun = netdev_priv(dev);
1624                 else
1625                         return -EINVAL;
1626
1627                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1628                     !!(tun->flags & IFF_MULTI_QUEUE))
1629                         return -EINVAL;
1630
1631                 if (tun_not_capable(tun))
1632                         return -EPERM;
1633                 err = security_tun_dev_open(tun->security);
1634                 if (err < 0)
1635                         return err;
1636
1637                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1638                 if (err < 0)
1639                         return err;
1640
1641                 if (tun->flags & IFF_MULTI_QUEUE &&
1642                     (tun->numqueues + tun->numdisabled > 1)) {
1643                         /* One or more queue has already been attached, no need
1644                          * to initialize the device again.
1645                          */
1646                         return 0;
1647                 }
1648         }
1649         else {
1650                 char *name;
1651                 unsigned long flags = 0;
1652                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1653                              MAX_TAP_QUEUES : 1;
1654
1655                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1656                         return -EPERM;
1657                 err = security_tun_dev_create();
1658                 if (err < 0)
1659                         return err;
1660
1661                 /* Set dev type */
1662                 if (ifr->ifr_flags & IFF_TUN) {
1663                         /* TUN device */
1664                         flags |= IFF_TUN;
1665                         name = "tun%d";
1666                 } else if (ifr->ifr_flags & IFF_TAP) {
1667                         /* TAP device */
1668                         flags |= IFF_TAP;
1669                         name = "tap%d";
1670                 } else
1671                         return -EINVAL;
1672
1673                 if (*ifr->ifr_name)
1674                         name = ifr->ifr_name;
1675
1676                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1677                                        NET_NAME_UNKNOWN, tun_setup, queues,
1678                                        queues);
1679
1680                 if (!dev)
1681                         return -ENOMEM;
1682
1683                 dev_net_set(dev, net);
1684                 dev->rtnl_link_ops = &tun_link_ops;
1685                 dev->ifindex = tfile->ifindex;
1686                 dev->sysfs_groups[0] = &tun_attr_group;
1687
1688                 tun = netdev_priv(dev);
1689                 tun->dev = dev;
1690                 tun->flags = flags;
1691                 tun->txflt.count = 0;
1692                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1693
1694                 tun->filter_attached = false;
1695                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1696
1697                 spin_lock_init(&tun->lock);
1698
1699                 err = security_tun_dev_alloc_security(&tun->security);
1700                 if (err < 0)
1701                         goto err_free_dev;
1702
1703                 tun_net_init(dev);
1704                 tun_flow_init(tun);
1705
1706                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1707                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1708                                    NETIF_F_HW_VLAN_STAG_TX;
1709                 dev->features = dev->hw_features;
1710                 dev->vlan_features = dev->features &
1711                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1712                                        NETIF_F_HW_VLAN_STAG_TX);
1713
1714                 INIT_LIST_HEAD(&tun->disabled);
1715                 err = tun_attach(tun, file, false);
1716                 if (err < 0)
1717                         goto err_free_flow;
1718
1719                 err = register_netdevice(tun->dev);
1720                 if (err < 0)
1721                         goto err_detach;
1722         }
1723
1724         netif_carrier_on(tun->dev);
1725
1726         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1727
1728         tun->flags = (tun->flags & ~TUN_FEATURES) |
1729                 (ifr->ifr_flags & TUN_FEATURES);
1730
1731         /* Make sure persistent devices do not get stuck in
1732          * xoff state.
1733          */
1734         if (netif_running(tun->dev))
1735                 netif_tx_wake_all_queues(tun->dev);
1736
1737         strcpy(ifr->ifr_name, tun->dev->name);
1738         return 0;
1739
1740 err_detach:
1741         tun_detach_all(dev);
1742 err_free_flow:
1743         tun_flow_uninit(tun);
1744         security_tun_dev_free_security(tun->security);
1745 err_free_dev:
1746         free_netdev(dev);
1747         return err;
1748 }
1749
1750 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1751                        struct ifreq *ifr)
1752 {
1753         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1754
1755         strcpy(ifr->ifr_name, tun->dev->name);
1756
1757         ifr->ifr_flags = tun_flags(tun);
1758
1759 }
1760
1761 /* This is like a cut-down ethtool ops, except done via tun fd so no
1762  * privs required. */
1763 static int set_offload(struct tun_struct *tun, unsigned long arg)
1764 {
1765         netdev_features_t features = 0;
1766
1767         if (arg & TUN_F_CSUM) {
1768                 features |= NETIF_F_HW_CSUM;
1769                 arg &= ~TUN_F_CSUM;
1770
1771                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1772                         if (arg & TUN_F_TSO_ECN) {
1773                                 features |= NETIF_F_TSO_ECN;
1774                                 arg &= ~TUN_F_TSO_ECN;
1775                         }
1776                         if (arg & TUN_F_TSO4)
1777                                 features |= NETIF_F_TSO;
1778                         if (arg & TUN_F_TSO6)
1779                                 features |= NETIF_F_TSO6;
1780                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1781                 }
1782
1783                 if (arg & TUN_F_UFO) {
1784                         features |= NETIF_F_UFO;
1785                         arg &= ~TUN_F_UFO;
1786                 }
1787         }
1788
1789         /* This gives the user a way to test for new features in future by
1790          * trying to set them. */
1791         if (arg)
1792                 return -EINVAL;
1793
1794         tun->set_features = features;
1795         netdev_update_features(tun->dev);
1796
1797         return 0;
1798 }
1799
1800 static void tun_detach_filter(struct tun_struct *tun, int n)
1801 {
1802         int i;
1803         struct tun_file *tfile;
1804
1805         for (i = 0; i < n; i++) {
1806                 tfile = rtnl_dereference(tun->tfiles[i]);
1807                 sk_detach_filter(tfile->socket.sk);
1808         }
1809
1810         tun->filter_attached = false;
1811 }
1812
1813 static int tun_attach_filter(struct tun_struct *tun)
1814 {
1815         int i, ret = 0;
1816         struct tun_file *tfile;
1817
1818         for (i = 0; i < tun->numqueues; i++) {
1819                 tfile = rtnl_dereference(tun->tfiles[i]);
1820                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1821                 if (ret) {
1822                         tun_detach_filter(tun, i);
1823                         return ret;
1824                 }
1825         }
1826
1827         tun->filter_attached = true;
1828         return ret;
1829 }
1830
1831 static void tun_set_sndbuf(struct tun_struct *tun)
1832 {
1833         struct tun_file *tfile;
1834         int i;
1835
1836         for (i = 0; i < tun->numqueues; i++) {
1837                 tfile = rtnl_dereference(tun->tfiles[i]);
1838                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1839         }
1840 }
1841
1842 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1843 {
1844         struct tun_file *tfile = file->private_data;
1845         struct tun_struct *tun;
1846         int ret = 0;
1847
1848         rtnl_lock();
1849
1850         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1851                 tun = tfile->detached;
1852                 if (!tun) {
1853                         ret = -EINVAL;
1854                         goto unlock;
1855                 }
1856                 ret = security_tun_dev_attach_queue(tun->security);
1857                 if (ret < 0)
1858                         goto unlock;
1859                 ret = tun_attach(tun, file, false);
1860         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1861                 tun = rtnl_dereference(tfile->tun);
1862                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1863                         ret = -EINVAL;
1864                 else
1865                         __tun_detach(tfile, false);
1866         } else
1867                 ret = -EINVAL;
1868
1869 unlock:
1870         rtnl_unlock();
1871         return ret;
1872 }
1873
1874 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1875                             unsigned long arg, int ifreq_len)
1876 {
1877         struct tun_file *tfile = file->private_data;
1878         struct tun_struct *tun;
1879         void __user* argp = (void __user*)arg;
1880         struct ifreq ifr;
1881         kuid_t owner;
1882         kgid_t group;
1883         int sndbuf;
1884         int vnet_hdr_sz;
1885         unsigned int ifindex;
1886         int le;
1887         int ret;
1888
1889         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1890                 if (copy_from_user(&ifr, argp, ifreq_len))
1891                         return -EFAULT;
1892         } else {
1893                 memset(&ifr, 0, sizeof(ifr));
1894         }
1895         if (cmd == TUNGETFEATURES) {
1896                 /* Currently this just means: "what IFF flags are valid?".
1897                  * This is needed because we never checked for invalid flags on
1898                  * TUNSETIFF.
1899                  */
1900                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1901                                 (unsigned int __user*)argp);
1902         } else if (cmd == TUNSETQUEUE)
1903                 return tun_set_queue(file, &ifr);
1904
1905         ret = 0;
1906         rtnl_lock();
1907
1908         tun = __tun_get(tfile);
1909         if (cmd == TUNSETIFF && !tun) {
1910                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1911
1912                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
1913
1914                 if (ret)
1915                         goto unlock;
1916
1917                 if (copy_to_user(argp, &ifr, ifreq_len))
1918                         ret = -EFAULT;
1919                 goto unlock;
1920         }
1921         if (cmd == TUNSETIFINDEX) {
1922                 ret = -EPERM;
1923                 if (tun)
1924                         goto unlock;
1925
1926                 ret = -EFAULT;
1927                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1928                         goto unlock;
1929
1930                 ret = 0;
1931                 tfile->ifindex = ifindex;
1932                 goto unlock;
1933         }
1934
1935         ret = -EBADFD;
1936         if (!tun)
1937                 goto unlock;
1938
1939         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1940
1941         ret = 0;
1942         switch (cmd) {
1943         case TUNGETIFF:
1944                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1945
1946                 if (tfile->detached)
1947                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
1948                 if (!tfile->socket.sk->sk_filter)
1949                         ifr.ifr_flags |= IFF_NOFILTER;
1950
1951                 if (copy_to_user(argp, &ifr, ifreq_len))
1952                         ret = -EFAULT;
1953                 break;
1954
1955         case TUNSETNOCSUM:
1956                 /* Disable/Enable checksum */
1957
1958                 /* [unimplemented] */
1959                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1960                           arg ? "disabled" : "enabled");
1961                 break;
1962
1963         case TUNSETPERSIST:
1964                 /* Disable/Enable persist mode. Keep an extra reference to the
1965                  * module to prevent the module being unprobed.
1966                  */
1967                 if (arg && !(tun->flags & IFF_PERSIST)) {
1968                         tun->flags |= IFF_PERSIST;
1969                         __module_get(THIS_MODULE);
1970                 }
1971                 if (!arg && (tun->flags & IFF_PERSIST)) {
1972                         tun->flags &= ~IFF_PERSIST;
1973                         module_put(THIS_MODULE);
1974                 }
1975
1976                 tun_debug(KERN_INFO, tun, "persist %s\n",
1977                           arg ? "enabled" : "disabled");
1978                 break;
1979
1980         case TUNSETOWNER:
1981                 /* Set owner of the device */
1982                 owner = make_kuid(current_user_ns(), arg);
1983                 if (!uid_valid(owner)) {
1984                         ret = -EINVAL;
1985                         break;
1986                 }
1987                 tun->owner = owner;
1988                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1989                           from_kuid(&init_user_ns, tun->owner));
1990                 break;
1991
1992         case TUNSETGROUP:
1993                 /* Set group of the device */
1994                 group = make_kgid(current_user_ns(), arg);
1995                 if (!gid_valid(group)) {
1996                         ret = -EINVAL;
1997                         break;
1998                 }
1999                 tun->group = group;
2000                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2001                           from_kgid(&init_user_ns, tun->group));
2002                 break;
2003
2004         case TUNSETLINK:
2005                 /* Only allow setting the type when the interface is down */
2006                 if (tun->dev->flags & IFF_UP) {
2007                         tun_debug(KERN_INFO, tun,
2008                                   "Linktype set failed because interface is up\n");
2009                         ret = -EBUSY;
2010                 } else {
2011                         tun->dev->type = (int) arg;
2012                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2013                                   tun->dev->type);
2014                         ret = 0;
2015                 }
2016                 break;
2017
2018 #ifdef TUN_DEBUG
2019         case TUNSETDEBUG:
2020                 tun->debug = arg;
2021                 break;
2022 #endif
2023         case TUNSETOFFLOAD:
2024                 ret = set_offload(tun, arg);
2025                 break;
2026
2027         case TUNSETTXFILTER:
2028                 /* Can be set only for TAPs */
2029                 ret = -EINVAL;
2030                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2031                         break;
2032                 ret = update_filter(&tun->txflt, (void __user *)arg);
2033                 break;
2034
2035         case SIOCGIFHWADDR:
2036                 /* Get hw address */
2037                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2038                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2039                 if (copy_to_user(argp, &ifr, ifreq_len))
2040                         ret = -EFAULT;
2041                 break;
2042
2043         case SIOCSIFHWADDR:
2044                 /* Set hw address */
2045                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2046                           ifr.ifr_hwaddr.sa_data);
2047
2048                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2049                 break;
2050
2051         case TUNGETSNDBUF:
2052                 sndbuf = tfile->socket.sk->sk_sndbuf;
2053                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2054                         ret = -EFAULT;
2055                 break;
2056
2057         case TUNSETSNDBUF:
2058                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2059                         ret = -EFAULT;
2060                         break;
2061                 }
2062
2063                 tun->sndbuf = sndbuf;
2064                 tun_set_sndbuf(tun);
2065                 break;
2066
2067         case TUNGETVNETHDRSZ:
2068                 vnet_hdr_sz = tun->vnet_hdr_sz;
2069                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2070                         ret = -EFAULT;
2071                 break;
2072
2073         case TUNSETVNETHDRSZ:
2074                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2075                         ret = -EFAULT;
2076                         break;
2077                 }
2078                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2079                         ret = -EINVAL;
2080                         break;
2081                 }
2082
2083                 tun->vnet_hdr_sz = vnet_hdr_sz;
2084                 break;
2085
2086         case TUNGETVNETLE:
2087                 le = !!(tun->flags & TUN_VNET_LE);
2088                 if (put_user(le, (int __user *)argp))
2089                         ret = -EFAULT;
2090                 break;
2091
2092         case TUNSETVNETLE:
2093                 if (get_user(le, (int __user *)argp)) {
2094                         ret = -EFAULT;
2095                         break;
2096                 }
2097                 if (le)
2098                         tun->flags |= TUN_VNET_LE;
2099                 else
2100                         tun->flags &= ~TUN_VNET_LE;
2101                 break;
2102
2103         case TUNGETVNETBE:
2104                 ret = tun_get_vnet_be(tun, argp);
2105                 break;
2106
2107         case TUNSETVNETBE:
2108                 ret = tun_set_vnet_be(tun, argp);
2109                 break;
2110
2111         case TUNATTACHFILTER:
2112                 /* Can be set only for TAPs */
2113                 ret = -EINVAL;
2114                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2115                         break;
2116                 ret = -EFAULT;
2117                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2118                         break;
2119
2120                 ret = tun_attach_filter(tun);
2121                 break;
2122
2123         case TUNDETACHFILTER:
2124                 /* Can be set only for TAPs */
2125                 ret = -EINVAL;
2126                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2127                         break;
2128                 ret = 0;
2129                 tun_detach_filter(tun, tun->numqueues);
2130                 break;
2131
2132         case TUNGETFILTER:
2133                 ret = -EINVAL;
2134                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2135                         break;
2136                 ret = -EFAULT;
2137                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2138                         break;
2139                 ret = 0;
2140                 break;
2141
2142         default:
2143                 ret = -EINVAL;
2144                 break;
2145         }
2146
2147 unlock:
2148         rtnl_unlock();
2149         if (tun)
2150                 tun_put(tun);
2151         return ret;
2152 }
2153
2154 static long tun_chr_ioctl(struct file *file,
2155                           unsigned int cmd, unsigned long arg)
2156 {
2157         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2158 }
2159
2160 #ifdef CONFIG_COMPAT
2161 static long tun_chr_compat_ioctl(struct file *file,
2162                          unsigned int cmd, unsigned long arg)
2163 {
2164         switch (cmd) {
2165         case TUNSETIFF:
2166         case TUNGETIFF:
2167         case TUNSETTXFILTER:
2168         case TUNGETSNDBUF:
2169         case TUNSETSNDBUF:
2170         case SIOCGIFHWADDR:
2171         case SIOCSIFHWADDR:
2172                 arg = (unsigned long)compat_ptr(arg);
2173                 break;
2174         default:
2175                 arg = (compat_ulong_t)arg;
2176                 break;
2177         }
2178
2179         /*
2180          * compat_ifreq is shorter than ifreq, so we must not access beyond
2181          * the end of that structure. All fields that are used in this
2182          * driver are compatible though, we don't need to convert the
2183          * contents.
2184          */
2185         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2186 }
2187 #endif /* CONFIG_COMPAT */
2188
2189 static int tun_chr_fasync(int fd, struct file *file, int on)
2190 {
2191         struct tun_file *tfile = file->private_data;
2192         int ret;
2193
2194         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2195                 goto out;
2196
2197         if (on) {
2198                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2199                 tfile->flags |= TUN_FASYNC;
2200         } else
2201                 tfile->flags &= ~TUN_FASYNC;
2202         ret = 0;
2203 out:
2204         return ret;
2205 }
2206
2207 static int tun_chr_open(struct inode *inode, struct file * file)
2208 {
2209         struct net *net = current->nsproxy->net_ns;
2210         struct tun_file *tfile;
2211
2212         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2213
2214         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2215                                             &tun_proto, 0);
2216         if (!tfile)
2217                 return -ENOMEM;
2218         RCU_INIT_POINTER(tfile->tun, NULL);
2219         tfile->flags = 0;
2220         tfile->ifindex = 0;
2221
2222         init_waitqueue_head(&tfile->wq.wait);
2223         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2224
2225         tfile->socket.file = file;
2226         tfile->socket.ops = &tun_socket_ops;
2227
2228         sock_init_data(&tfile->socket, &tfile->sk);
2229
2230         tfile->sk.sk_write_space = tun_sock_write_space;
2231         tfile->sk.sk_sndbuf = INT_MAX;
2232
2233         file->private_data = tfile;
2234         INIT_LIST_HEAD(&tfile->next);
2235
2236         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2237
2238         return 0;
2239 }
2240
2241 static int tun_chr_close(struct inode *inode, struct file *file)
2242 {
2243         struct tun_file *tfile = file->private_data;
2244
2245         tun_detach(tfile, true);
2246
2247         return 0;
2248 }
2249
2250 #ifdef CONFIG_PROC_FS
2251 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2252 {
2253         struct tun_struct *tun;
2254         struct ifreq ifr;
2255
2256         memset(&ifr, 0, sizeof(ifr));
2257
2258         rtnl_lock();
2259         tun = tun_get(f);
2260         if (tun)
2261                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2262         rtnl_unlock();
2263
2264         if (tun)
2265                 tun_put(tun);
2266
2267         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2268 }
2269 #endif
2270
2271 static const struct file_operations tun_fops = {
2272         .owner  = THIS_MODULE,
2273         .llseek = no_llseek,
2274         .read_iter  = tun_chr_read_iter,
2275         .write_iter = tun_chr_write_iter,
2276         .poll   = tun_chr_poll,
2277         .unlocked_ioctl = tun_chr_ioctl,
2278 #ifdef CONFIG_COMPAT
2279         .compat_ioctl = tun_chr_compat_ioctl,
2280 #endif
2281         .open   = tun_chr_open,
2282         .release = tun_chr_close,
2283         .fasync = tun_chr_fasync,
2284 #ifdef CONFIG_PROC_FS
2285         .show_fdinfo = tun_chr_show_fdinfo,
2286 #endif
2287 };
2288
2289 static struct miscdevice tun_miscdev = {
2290         .minor = TUN_MINOR,
2291         .name = "tun",
2292         .nodename = "net/tun",
2293         .fops = &tun_fops,
2294 };
2295
2296 /* ethtool interface */
2297
2298 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2299 {
2300         cmd->supported          = 0;
2301         cmd->advertising        = 0;
2302         ethtool_cmd_speed_set(cmd, SPEED_10);
2303         cmd->duplex             = DUPLEX_FULL;
2304         cmd->port               = PORT_TP;
2305         cmd->phy_address        = 0;
2306         cmd->transceiver        = XCVR_INTERNAL;
2307         cmd->autoneg            = AUTONEG_DISABLE;
2308         cmd->maxtxpkt           = 0;
2309         cmd->maxrxpkt           = 0;
2310         return 0;
2311 }
2312
2313 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2314 {
2315         struct tun_struct *tun = netdev_priv(dev);
2316
2317         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2318         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2319
2320         switch (tun->flags & TUN_TYPE_MASK) {
2321         case IFF_TUN:
2322                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2323                 break;
2324         case IFF_TAP:
2325                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2326                 break;
2327         }
2328 }
2329
2330 static u32 tun_get_msglevel(struct net_device *dev)
2331 {
2332 #ifdef TUN_DEBUG
2333         struct tun_struct *tun = netdev_priv(dev);
2334         return tun->debug;
2335 #else
2336         return -EOPNOTSUPP;
2337 #endif
2338 }
2339
2340 static void tun_set_msglevel(struct net_device *dev, u32 value)
2341 {
2342 #ifdef TUN_DEBUG
2343         struct tun_struct *tun = netdev_priv(dev);
2344         tun->debug = value;
2345 #endif
2346 }
2347
2348 static const struct ethtool_ops tun_ethtool_ops = {
2349         .get_settings   = tun_get_settings,
2350         .get_drvinfo    = tun_get_drvinfo,
2351         .get_msglevel   = tun_get_msglevel,
2352         .set_msglevel   = tun_set_msglevel,
2353         .get_link       = ethtool_op_get_link,
2354         .get_ts_info    = ethtool_op_get_ts_info,
2355 };
2356
2357
2358 static int __init tun_init(void)
2359 {
2360         int ret = 0;
2361
2362         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2363         pr_info("%s\n", DRV_COPYRIGHT);
2364
2365         ret = rtnl_link_register(&tun_link_ops);
2366         if (ret) {
2367                 pr_err("Can't register link_ops\n");
2368                 goto err_linkops;
2369         }
2370
2371         ret = misc_register(&tun_miscdev);
2372         if (ret) {
2373                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2374                 goto err_misc;
2375         }
2376         return  0;
2377 err_misc:
2378         rtnl_link_unregister(&tun_link_ops);
2379 err_linkops:
2380         return ret;
2381 }
2382
2383 static void tun_cleanup(void)
2384 {
2385         misc_deregister(&tun_miscdev);
2386         rtnl_link_unregister(&tun_link_ops);
2387 }
2388
2389 /* Get an underlying socket object from tun file.  Returns error unless file is
2390  * attached to a device.  The returned object works like a packet socket, it
2391  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2392  * holding a reference to the file for as long as the socket is in use. */
2393 struct socket *tun_get_socket(struct file *file)
2394 {
2395         struct tun_file *tfile;
2396         if (file->f_op != &tun_fops)
2397                 return ERR_PTR(-EINVAL);
2398         tfile = file->private_data;
2399         if (!tfile)
2400                 return ERR_PTR(-EBADFD);
2401         return &tfile->socket;
2402 }
2403 EXPORT_SYMBOL_GPL(tun_get_socket);
2404
2405 module_init(tun_init);
2406 module_exit(tun_cleanup);
2407 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2408 MODULE_AUTHOR(DRV_COPYRIGHT);
2409 MODULE_LICENSE("GPL");
2410 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2411 MODULE_ALIAS("devname:net/tun");