drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/major.h>
  48 #include <linux/slab.h>
  49 #include <linux/poll.h>
  50 #include <linux/fcntl.h>
  51 #include <linux/init.h>
  52 #include <linux/skbuff.h>
  53 #include <linux/netdevice.h>
  54 #include <linux/etherdevice.h>
  55 #include <linux/miscdevice.h>
  56 #include <linux/ethtool.h>
  57 #include <linux/rtnetlink.h>
  58 #include <linux/compat.h>
  59 #include <linux/if.h>
  60 #include <linux/if_arp.h>
  61 #include <linux/if_ether.h>
  62 #include <linux/if_tun.h>
  63 #include <linux/if_vlan.h>
  64 #include <linux/crc32.h>
  65 #include <linux/nsproxy.h>
  66 #include <linux/virtio_net.h>
  67 #include <linux/rcupdate.h>
  68 #include <net/net_namespace.h>
  69 #include <net/netns/generic.h>
  70 #include <net/rtnetlink.h>
  71 #include <net/sock.h>
  72 #include <linux/seq_file.h>
  73 #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
 115 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 116                       IFF_MULTI_QUEUE)
 117 #define GOODCOPY_LEN 128
 118
 119 #define FLT_EXACT_COUNT 8
 120 struct tap_filter {
 121         unsigned int    count;    /* Number of addrs. Zero means disabled */
 122         u32             mask[2];  /* Mask of the hashed addrs */
 123         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 124 };
 125
 126 /* DEFAULT_MAX_NUM_RSS_QUEUES were chosen to let the rx/tx queues allocated for
 127  * the netdevice to be fit in one page. So we can make sure the success of
 128  * memory allocation. TODO: increase the limit. */
 129 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
 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 net *net;
 151         struct fasync_struct *fasync;
 152         /* only used for fasnyc */
 153         unsigned int flags;
 154         union {
 155                 u16 queue_index;
 156                 unsigned int ifindex;
 157         };
 158         struct list_head next;
 159         struct tun_struct *detached;
 160 };
 161
 162 struct tun_flow_entry {
 163         struct hlist_node hash_link;
 164         struct rcu_head rcu;
 165         struct tun_struct *tun;
 166
 167         u32 rxhash;
 168         u32 rps_rxhash;
 169         int queue_index;
 170         unsigned long updated;
 171 };
 172
 173 #define TUN_NUM_FLOW_ENTRIES 1024
 174
 175 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 176  * device, socket filter, sndbuf and vnet header size were restore when the
 177  * file were attached to a persist device.
 178  */
 179 struct tun_struct {
 180         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 181         unsigned int            numqueues;
 182         unsigned int            flags;
 183         kuid_t                  owner;
 184         kgid_t                  group;
 185
 186         struct net_device       *dev;
 187         netdev_features_t       set_features;
 188 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 189                           NETIF_F_TSO6|NETIF_F_UFO)
 190
 191         int                     vnet_hdr_sz;
 192         int                     sndbuf;
 193         struct tap_filter       txflt;
 194         struct sock_fprog       fprog;
 195         /* protected by rtnl lock */
 196         bool                    filter_attached;
 197 #ifdef TUN_DEBUG
 198         int debug;
 199 #endif
 200         spinlock_t lock;
 201         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 202         struct timer_list flow_gc_timer;
 203         unsigned long ageing_time;
 204         unsigned int numdisabled;
 205         struct list_head disabled;
 206         void *security;
 207         u32 flow_count;
 208 };
 209
 210 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 211 {
 212         return __virtio16_to_cpu(tun->flags & TUN_VNET_LE, val);
 213 }
 214
 215 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 216 {
 217         return __cpu_to_virtio16(tun->flags & TUN_VNET_LE, val);
 218 }
 219
 220 static inline u32 tun_hashfn(u32 rxhash)
 221 {
 222         return rxhash & 0x3ff;
 223 }
 224
 225 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 226 {
 227         struct tun_flow_entry *e;
 228
 229         hlist_for_each_entry_rcu(e, head, hash_link) {
 230                 if (e->rxhash == rxhash)
 231                         return e;
 232         }
 233         return NULL;
 234 }
 235
 236 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 237                                               struct hlist_head *head,
 238                                               u32 rxhash, u16 queue_index)
 239 {
 240         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 241
 242         if (e) {
 243                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 244                           rxhash, queue_index);
 245                 e->updated = jiffies;
 246                 e->rxhash = rxhash;
 247                 e->rps_rxhash = 0;
 248                 e->queue_index = queue_index;
 249                 e->tun = tun;
 250                 hlist_add_head_rcu(&e->hash_link, head);
 251                 ++tun->flow_count;
 252         }
 253         return e;
 254 }
 255
 256 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 257 {
 258         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 259                   e->rxhash, e->queue_index);
 260         sock_rps_reset_flow_hash(e->rps_rxhash);
 261         hlist_del_rcu(&e->hash_link);
 262         kfree_rcu(e, rcu);
 263         --tun->flow_count;
 264 }
 265
 266 static void tun_flow_flush(struct tun_struct *tun)
 267 {
 268         int i;
 269
 270         spin_lock_bh(&tun->lock);
 271         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 272                 struct tun_flow_entry *e;
 273                 struct hlist_node *n;
 274
 275                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 276                         tun_flow_delete(tun, e);
 277         }
 278         spin_unlock_bh(&tun->lock);
 279 }
 280
 281 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 282 {
 283         int i;
 284
 285         spin_lock_bh(&tun->lock);
 286         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 287                 struct tun_flow_entry *e;
 288                 struct hlist_node *n;
 289
 290                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 291                         if (e->queue_index == queue_index)
 292                                 tun_flow_delete(tun, e);
 293                 }
 294         }
 295         spin_unlock_bh(&tun->lock);
 296 }
 297
 298 static void tun_flow_cleanup(unsigned long data)
 299 {
 300         struct tun_struct *tun = (struct tun_struct *)data;
 301         unsigned long delay = tun->ageing_time;
 302         unsigned long next_timer = jiffies + delay;
 303         unsigned long count = 0;
 304         int i;
 305
 306         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 307
 308         spin_lock_bh(&tun->lock);
 309         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 310                 struct tun_flow_entry *e;
 311                 struct hlist_node *n;
 312
 313                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 314                         unsigned long this_timer;
 315                         count++;
 316                         this_timer = e->updated + delay;
 317                         if (time_before_eq(this_timer, jiffies))
 318                                 tun_flow_delete(tun, e);
 319                         else if (time_before(this_timer, next_timer))
 320                                 next_timer = this_timer;
 321                 }
 322         }
 323
 324         if (count)
 325                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 326         spin_unlock_bh(&tun->lock);
 327 }
 328
 329 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 330                             struct tun_file *tfile)
 331 {
 332         struct hlist_head *head;
 333         struct tun_flow_entry *e;
 334         unsigned long delay = tun->ageing_time;
 335         u16 queue_index = tfile->queue_index;
 336
 337         if (!rxhash)
 338                 return;
 339         else
 340                 head = &tun->flows[tun_hashfn(rxhash)];
 341
 342         rcu_read_lock();
 343
 344         /* We may get a very small possibility of OOO during switching, not
 345          * worth to optimize.*/
 346         if (tun->numqueues == 1 || tfile->detached)
 347                 goto unlock;
 348
 349         e = tun_flow_find(head, rxhash);
 350         if (likely(e)) {
 351                 /* TODO: keep queueing to old queue until it's empty? */
 352                 e->queue_index = queue_index;
 353                 e->updated = jiffies;
 354                 sock_rps_record_flow_hash(e->rps_rxhash);
 355         } else {
 356                 spin_lock_bh(&tun->lock);
 357                 if (!tun_flow_find(head, rxhash) &&
 358                     tun->flow_count < MAX_TAP_FLOWS)
 359                         tun_flow_create(tun, head, rxhash, queue_index);
 360
 361                 if (!timer_pending(&tun->flow_gc_timer))
 362                         mod_timer(&tun->flow_gc_timer,
 363                                   round_jiffies_up(jiffies + delay));
 364                 spin_unlock_bh(&tun->lock);
 365         }
 366
 367 unlock:
 368         rcu_read_unlock();
 369 }
 370
 371 /**
 372  * Save the hash received in the stack receive path and update the
 373  * flow_hash table accordingly.
 374  */
 375 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 376 {
 377         if (unlikely(e->rps_rxhash != hash)) {
 378                 sock_rps_reset_flow_hash(e->rps_rxhash);
 379                 e->rps_rxhash = hash;
 380         }
 381 }
 382
 383 /* We try to identify a flow through its rxhash first. The reason that
 384  * we do not check rxq no. is because some cards(e.g 82599), chooses
 385  * the rxq based on the txq where the last packet of the flow comes. As
 386  * the userspace application move between processors, we may get a
 387  * different rxq no. here. If we could not get rxhash, then we would
 388  * hope the rxq no. may help here.
 389  */
 390 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 391                             void *accel_priv, select_queue_fallback_t fallback)
 392 {
 393         struct tun_struct *tun = netdev_priv(dev);
 394         struct tun_flow_entry *e;
 395         u32 txq = 0;
 396         u32 numqueues = 0;
 397
 398         rcu_read_lock();
 399         numqueues = ACCESS_ONCE(tun->numqueues);
 400
 401         txq = skb_get_hash(skb);
 402         if (txq) {
 403                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 404                 if (e) {
 405                         tun_flow_save_rps_rxhash(e, txq);
 406                         txq = e->queue_index;
 407                 } else
 408                         /* use multiply and shift instead of expensive divide */
 409                         txq = ((u64)txq * numqueues) >> 32;
 410         } else if (likely(skb_rx_queue_recorded(skb))) {
 411                 txq = skb_get_rx_queue(skb);
 412                 while (unlikely(txq >= numqueues))
 413                         txq -= numqueues;
 414         }
 415
 416         rcu_read_unlock();
 417         return txq;
 418 }
 419
 420 static inline bool tun_not_capable(struct tun_struct *tun)
 421 {
 422         const struct cred *cred = current_cred();
 423         struct net *net = dev_net(tun->dev);
 424
 425         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 426                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 427                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 428 }
 429
 430 static void tun_set_real_num_queues(struct tun_struct *tun)
 431 {
 432         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 433         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 434 }
 435
 436 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 437 {
 438         tfile->detached = tun;
 439         list_add_tail(&tfile->next, &tun->disabled);
 440         ++tun->numdisabled;
 441 }
 442
 443 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 444 {
 445         struct tun_struct *tun = tfile->detached;
 446
 447         tfile->detached = NULL;
 448         list_del_init(&tfile->next);
 449         --tun->numdisabled;
 450         return tun;
 451 }
 452
 453 static void tun_queue_purge(struct tun_file *tfile)
 454 {
 455         skb_queue_purge(&tfile->sk.sk_receive_queue);
 456         skb_queue_purge(&tfile->sk.sk_error_queue);
 457 }
 458
 459 static void __tun_detach(struct tun_file *tfile, bool clean)
 460 {
 461         struct tun_file *ntfile;
 462         struct tun_struct *tun;
 463
 464         tun = rtnl_dereference(tfile->tun);
 465
 466         if (tun && !tfile->detached) {
 467                 u16 index = tfile->queue_index;
 468                 BUG_ON(index >= tun->numqueues);
 469
 470                 rcu_assign_pointer(tun->tfiles[index],
 471                                    tun->tfiles[tun->numqueues - 1]);
 472                 ntfile = rtnl_dereference(tun->tfiles[index]);
 473                 ntfile->queue_index = index;
 474
 475                 --tun->numqueues;
 476                 if (clean) {
 477                         RCU_INIT_POINTER(tfile->tun, NULL);
 478                         sock_put(&tfile->sk);
 479                 } else
 480                         tun_disable_queue(tun, tfile);
 481
 482                 synchronize_net();
 483                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 484                 /* Drop read queue */
 485                 tun_queue_purge(tfile);
 486                 tun_set_real_num_queues(tun);
 487         } else if (tfile->detached && clean) {
 488                 tun = tun_enable_queue(tfile);
 489                 sock_put(&tfile->sk);
 490         }
 491
 492         if (clean) {
 493                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 494                         netif_carrier_off(tun->dev);
 495
 496                         if (!(tun->flags & IFF_PERSIST) &&
 497                             tun->dev->reg_state == NETREG_REGISTERED)
 498                                 unregister_netdevice(tun->dev);
 499                 }
 500
 501                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 502                                  &tfile->socket.flags));
 503                 sk_release_kernel(&tfile->sk);
 504         }
 505 }
 506
 507 static void tun_detach(struct tun_file *tfile, bool clean)
 508 {
 509         rtnl_lock();
 510         __tun_detach(tfile, clean);
 511         rtnl_unlock();
 512 }
 513
 514 static void tun_detach_all(struct net_device *dev)
 515 {
 516         struct tun_struct *tun = netdev_priv(dev);
 517         struct tun_file *tfile, *tmp;
 518         int i, n = tun->numqueues;
 519
 520         for (i = 0; i < n; i++) {
 521                 tfile = rtnl_dereference(tun->tfiles[i]);
 522                 BUG_ON(!tfile);
 523                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 524                 RCU_INIT_POINTER(tfile->tun, NULL);
 525                 --tun->numqueues;
 526         }
 527         list_for_each_entry(tfile, &tun->disabled, next) {
 528                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 529                 RCU_INIT_POINTER(tfile->tun, NULL);
 530         }
 531         BUG_ON(tun->numqueues != 0);
 532
 533         synchronize_net();
 534         for (i = 0; i < n; i++) {
 535                 tfile = rtnl_dereference(tun->tfiles[i]);
 536                 /* Drop read queue */
 537                 tun_queue_purge(tfile);
 538                 sock_put(&tfile->sk);
 539         }
 540         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 541                 tun_enable_queue(tfile);
 542                 tun_queue_purge(tfile);
 543                 sock_put(&tfile->sk);
 544         }
 545         BUG_ON(tun->numdisabled != 0);
 546
 547         if (tun->flags & IFF_PERSIST)
 548                 module_put(THIS_MODULE);
 549 }
 550
 551 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 552 {
 553         struct tun_file *tfile = file->private_data;
 554         int err;
 555
 556         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 557         if (err < 0)
 558                 goto out;
 559
 560         err = -EINVAL;
 561         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 562                 goto out;
 563
 564         err = -EBUSY;
 565         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 566                 goto out;
 567
 568         err = -E2BIG;
 569         if (!tfile->detached &&
 570             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 571                 goto out;
 572
 573         err = 0;
 574
 575         /* Re-attach the filter to persist device */
 576         if (!skip_filter && (tun->filter_attached == true)) {
 577                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 578                 if (!err)
 579                         goto out;
 580         }
 581         tfile->queue_index = tun->numqueues;
 582         rcu_assign_pointer(tfile->tun, tun);
 583         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 584         tun->numqueues++;
 585
 586         if (tfile->detached)
 587                 tun_enable_queue(tfile);
 588         else
 589                 sock_hold(&tfile->sk);
 590
 591         tun_set_real_num_queues(tun);
 592
 593         /* device is allowed to go away first, so no need to hold extra
 594          * refcnt.
 595          */
 596
 597 out:
 598         return err;
 599 }
 600
 601 static struct tun_struct *__tun_get(struct tun_file *tfile)
 602 {
 603         struct tun_struct *tun;
 604
 605         rcu_read_lock();
 606         tun = rcu_dereference(tfile->tun);
 607         if (tun)
 608                 dev_hold(tun->dev);
 609         rcu_read_unlock();
 610
 611         return tun;
 612 }
 613
 614 static struct tun_struct *tun_get(struct file *file)
 615 {
 616         return __tun_get(file->private_data);
 617 }
 618
 619 static void tun_put(struct tun_struct *tun)
 620 {
 621         dev_put(tun->dev);
 622 }
 623
 624 /* TAP filtering */
 625 static void addr_hash_set(u32 *mask, const u8 *addr)
 626 {
 627         int n = ether_crc(ETH_ALEN, addr) >> 26;
 628         mask[n >> 5] |= (1 << (n & 31));
 629 }
 630
 631 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 632 {
 633         int n = ether_crc(ETH_ALEN, addr) >> 26;
 634         return mask[n >> 5] & (1 << (n & 31));
 635 }
 636
 637 static int update_filter(struct tap_filter *filter, void __user *arg)
 638 {
 639         struct { u8 u[ETH_ALEN]; } *addr;
 640         struct tun_filter uf;
 641         int err, alen, n, nexact;
 642
 643         if (copy_from_user(&uf, arg, sizeof(uf)))
 644                 return -EFAULT;
 645
 646         if (!uf.count) {
 647                 /* Disabled */
 648                 filter->count = 0;
 649                 return 0;
 650         }
 651
 652         alen = ETH_ALEN * uf.count;
 653         addr = kmalloc(alen, GFP_KERNEL);
 654         if (!addr)
 655                 return -ENOMEM;
 656
 657         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 658                 err = -EFAULT;
 659                 goto done;
 660         }
 661
 662         /* The filter is updated without holding any locks. Which is
 663          * perfectly safe. We disable it first and in the worst
 664          * case we'll accept a few undesired packets. */
 665         filter->count = 0;
 666         wmb();
 667
 668         /* Use first set of addresses as an exact filter */
 669         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 670                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 671
 672         nexact = n;
 673
 674         /* Remaining multicast addresses are hashed,
 675          * unicast will leave the filter disabled. */
 676         memset(filter->mask, 0, sizeof(filter->mask));
 677         for (; n < uf.count; n++) {
 678                 if (!is_multicast_ether_addr(addr[n].u)) {
 679                         err = 0; /* no filter */
 680                         goto done;
 681                 }
 682                 addr_hash_set(filter->mask, addr[n].u);
 683         }
 684
 685         /* For ALLMULTI just set the mask to all ones.
 686          * This overrides the mask populated above. */
 687         if ((uf.flags & TUN_FLT_ALLMULTI))
 688                 memset(filter->mask, ~0, sizeof(filter->mask));
 689
 690         /* Now enable the filter */
 691         wmb();
 692         filter->count = nexact;
 693
 694         /* Return the number of exact filters */
 695         err = nexact;
 696
 697 done:
 698         kfree(addr);
 699         return err;
 700 }
 701
 702 /* Returns: 0 - drop, !=0 - accept */
 703 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 704 {
 705         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 706          * at this point. */
 707         struct ethhdr *eh = (struct ethhdr *) skb->data;
 708         int i;
 709
 710         /* Exact match */
 711         for (i = 0; i < filter->count; i++)
 712                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 713                         return 1;
 714
 715         /* Inexact match (multicast only) */
 716         if (is_multicast_ether_addr(eh->h_dest))
 717                 return addr_hash_test(filter->mask, eh->h_dest);
 718
 719         return 0;
 720 }
 721
 722 /*
 723  * Checks whether the packet is accepted or not.
 724  * Returns: 0 - drop, !=0 - accept
 725  */
 726 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 727 {
 728         if (!filter->count)
 729                 return 1;
 730
 731         return run_filter(filter, skb);
 732 }
 733
 734 /* Network device part of the driver */
 735
 736 static const struct ethtool_ops tun_ethtool_ops;
 737
 738 /* Net device detach from fd. */
 739 static void tun_net_uninit(struct net_device *dev)
 740 {
 741         tun_detach_all(dev);
 742 }
 743
 744 /* Net device open. */
 745 static int tun_net_open(struct net_device *dev)
 746 {
 747         netif_tx_start_all_queues(dev);
 748         return 0;
 749 }
 750
 751 /* Net device close. */
 752 static int tun_net_close(struct net_device *dev)
 753 {
 754         netif_tx_stop_all_queues(dev);
 755         return 0;
 756 }
 757
 758 /* Net device start xmit */
 759 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 760 {
 761         struct tun_struct *tun = netdev_priv(dev);
 762         int txq = skb->queue_mapping;
 763         struct tun_file *tfile;
 764         u32 numqueues = 0;
 765
 766         rcu_read_lock();
 767         tfile = rcu_dereference(tun->tfiles[txq]);
 768         numqueues = ACCESS_ONCE(tun->numqueues);
 769
 770         /* Drop packet if interface is not attached */
 771         if (txq >= numqueues)
 772                 goto drop;
 773
 774         if (numqueues == 1) {
 775                 /* Select queue was not called for the skbuff, so we extract the
 776                  * RPS hash and save it into the flow_table here.
 777                  */
 778                 __u32 rxhash;
 779
 780                 rxhash = skb_get_hash(skb);
 781                 if (rxhash) {
 782                         struct tun_flow_entry *e;
 783                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 784                                         rxhash);
 785                         if (e)
 786                                 tun_flow_save_rps_rxhash(e, rxhash);
 787                 }
 788         }
 789
 790         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 791
 792         BUG_ON(!tfile);
 793
 794         /* Drop if the filter does not like it.
 795          * This is a noop if the filter is disabled.
 796          * Filter can be enabled only for the TAP devices. */
 797         if (!check_filter(&tun->txflt, skb))
 798                 goto drop;
 799
 800         if (tfile->socket.sk->sk_filter &&
 801             sk_filter(tfile->socket.sk, skb))
 802                 goto drop;
 803
 804         /* Limit the number of packets queued by dividing txq length with the
 805          * number of queues.
 806          */
 807         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
 808                           >= dev->tx_queue_len)
 809                 goto drop;
 810
 811         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 812                 goto drop;
 813
 814         if (skb->sk) {
 815                 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
 816                 sw_tx_timestamp(skb);
 817         }
 818
 819         /* Orphan the skb - required as we might hang on to it
 820          * for indefinite time.
 821          */
 822         skb_orphan(skb);
 823
 824         nf_reset(skb);
 825
 826         /* Enqueue packet */
 827         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 828
 829         /* Notify and wake up reader process */
 830         if (tfile->flags & TUN_FASYNC)
 831                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 832         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 833
 834         rcu_read_unlock();
 835         return NETDEV_TX_OK;
 836
 837 drop:
 838         dev->stats.tx_dropped++;
 839         skb_tx_error(skb);
 840         kfree_skb(skb);
 841         rcu_read_unlock();
 842         return NET_XMIT_DROP;
 843 }
 844
 845 static void tun_net_mclist(struct net_device *dev)
 846 {
 847         /*
 848          * This callback is supposed to deal with mc filter in
 849          * _rx_ path and has nothing to do with the _tx_ path.
 850          * In rx path we always accept everything userspace gives us.
 851          */
 852 }
 853
 854 #define MIN_MTU 68
 855 #define MAX_MTU 65535
 856
 857 static int
 858 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 859 {
 860         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 861                 return -EINVAL;
 862         dev->mtu = new_mtu;
 863         return 0;
 864 }
 865
 866 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 867         netdev_features_t features)
 868 {
 869         struct tun_struct *tun = netdev_priv(dev);
 870
 871         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 872 }
 873 #ifdef CONFIG_NET_POLL_CONTROLLER
 874 static void tun_poll_controller(struct net_device *dev)
 875 {
 876         /*
 877          * Tun only receives frames when:
 878          * 1) the char device endpoint gets data from user space
 879          * 2) the tun socket gets a sendmsg call from user space
 880          * Since both of those are synchronous operations, we are guaranteed
 881          * never to have pending data when we poll for it
 882          * so there is nothing to do here but return.
 883          * We need this though so netpoll recognizes us as an interface that
 884          * supports polling, which enables bridge devices in virt setups to
 885          * still use netconsole
 886          */
 887         return;
 888 }
 889 #endif
 890 static const struct net_device_ops tun_netdev_ops = {
 891         .ndo_uninit             = tun_net_uninit,
 892         .ndo_open               = tun_net_open,
 893         .ndo_stop               = tun_net_close,
 894         .ndo_start_xmit         = tun_net_xmit,
 895         .ndo_change_mtu         = tun_net_change_mtu,
 896         .ndo_fix_features       = tun_net_fix_features,
 897         .ndo_select_queue       = tun_select_queue,
 898 #ifdef CONFIG_NET_POLL_CONTROLLER
 899         .ndo_poll_controller    = tun_poll_controller,
 900 #endif
 901 };
 902
 903 static const struct net_device_ops tap_netdev_ops = {
 904         .ndo_uninit             = tun_net_uninit,
 905         .ndo_open               = tun_net_open,
 906         .ndo_stop               = tun_net_close,
 907         .ndo_start_xmit         = tun_net_xmit,
 908         .ndo_change_mtu         = tun_net_change_mtu,
 909         .ndo_fix_features       = tun_net_fix_features,
 910         .ndo_set_rx_mode        = tun_net_mclist,
 911         .ndo_set_mac_address    = eth_mac_addr,
 912         .ndo_validate_addr      = eth_validate_addr,
 913         .ndo_select_queue       = tun_select_queue,
 914 #ifdef CONFIG_NET_POLL_CONTROLLER
 915         .ndo_poll_controller    = tun_poll_controller,
 916 #endif
 917 };
 918
 919 static void tun_flow_init(struct tun_struct *tun)
 920 {
 921         int i;
 922
 923         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 924                 INIT_HLIST_HEAD(&tun->flows[i]);
 925
 926         tun->ageing_time = TUN_FLOW_EXPIRE;
 927         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 928         mod_timer(&tun->flow_gc_timer,
 929                   round_jiffies_up(jiffies + tun->ageing_time));
 930 }
 931
 932 static void tun_flow_uninit(struct tun_struct *tun)
 933 {
 934         del_timer_sync(&tun->flow_gc_timer);
 935         tun_flow_flush(tun);
 936 }
 937
 938 /* Initialize net device. */
 939 static void tun_net_init(struct net_device *dev)
 940 {
 941         struct tun_struct *tun = netdev_priv(dev);
 942
 943         switch (tun->flags & TUN_TYPE_MASK) {
 944         case IFF_TUN:
 945                 dev->netdev_ops = &tun_netdev_ops;
 946
 947                 /* Point-to-Point TUN Device */
 948                 dev->hard_header_len = 0;
 949                 dev->addr_len = 0;
 950                 dev->mtu = 1500;
 951
 952                 /* Zero header length */
 953                 dev->type = ARPHRD_NONE;
 954                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 955                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 956                 break;
 957
 958         case IFF_TAP:
 959                 dev->netdev_ops = &tap_netdev_ops;
 960                 /* Ethernet TAP Device */
 961                 ether_setup(dev);
 962                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 963                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 964
 965                 eth_hw_addr_random(dev);
 966
 967                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 968                 break;
 969         }
 970 }
 971
 972 /* Character device part */
 973
 974 /* Poll */
 975 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 976 {
 977         struct tun_file *tfile = file->private_data;
 978         struct tun_struct *tun = __tun_get(tfile);
 979         struct sock *sk;
 980         unsigned int mask = 0;
 981
 982         if (!tun)
 983                 return POLLERR;
 984
 985         sk = tfile->socket.sk;
 986
 987         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 988
 989         poll_wait(file, sk_sleep(sk), wait);
 990
 991         if (!skb_queue_empty(&sk->sk_receive_queue))
 992                 mask |= POLLIN | POLLRDNORM;
 993
 994         if (sock_writeable(sk) ||
 995             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 996              sock_writeable(sk)))
 997                 mask |= POLLOUT | POLLWRNORM;
 998
 999         if (tun->dev->reg_state != NETREG_REGISTERED)
1000                 mask = POLLERR;
1001
1002         tun_put(tun);
1003         return mask;
1004 }
1005
1006 /* prepad is the amount to reserve at front.  len is length after that.
1007  * linear is a hint as to how much to copy (usually headers). */
1008 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1009                                      size_t prepad, size_t len,
1010                                      size_t linear, int noblock)
1011 {
1012         struct sock *sk = tfile->socket.sk;
1013         struct sk_buff *skb;
1014         int err;
1015
1016         /* Under a page?  Don't bother with paged skb. */
1017         if (prepad + len < PAGE_SIZE || !linear)
1018                 linear = len;
1019
1020         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1021                                    &err, 0);
1022         if (!skb)
1023                 return ERR_PTR(err);
1024
1025         skb_reserve(skb, prepad);
1026         skb_put(skb, linear);
1027         skb->data_len = len - linear;
1028         skb->len += len - linear;
1029
1030         return skb;
1031 }
1032
1033 /* Get packet from user space buffer */
1034 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1035                             void *msg_control, struct iov_iter *from,
1036                             int noblock)
1037 {
1038         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1039         struct sk_buff *skb;
1040         size_t total_len = iov_iter_count(from);
1041         size_t len = total_len, align = NET_SKB_PAD, linear;
1042         struct virtio_net_hdr gso = { 0 };
1043         int good_linear;
1044         int copylen;
1045         bool zerocopy = false;
1046         int err;
1047         u32 rxhash;
1048         ssize_t n;
1049
1050         if (!(tun->flags & IFF_NO_PI)) {
1051                 if (len < sizeof(pi))
1052                         return -EINVAL;
1053                 len -= sizeof(pi);
1054
1055                 n = copy_from_iter(&pi, sizeof(pi), from);
1056                 if (n != sizeof(pi))
1057                         return -EFAULT;
1058         }
1059
1060         if (tun->flags & IFF_VNET_HDR) {
1061                 if (len < tun->vnet_hdr_sz)
1062                         return -EINVAL;
1063                 len -= tun->vnet_hdr_sz;
1064
1065                 n = copy_from_iter(&gso, sizeof(gso), from);
1066                 if (n != sizeof(gso))
1067                         return -EFAULT;
1068
1069                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1070                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1071                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1072
1073                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1074                         return -EINVAL;
1075                 iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1076         }
1077
1078         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1079                 align += NET_IP_ALIGN;
1080                 if (unlikely(len < ETH_HLEN ||
1081                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1082                         return -EINVAL;
1083         }
1084
1085         good_linear = SKB_MAX_HEAD(align);
1086
1087         if (msg_control) {
1088                 struct iov_iter i = *from;
1089
1090                 /* There are 256 bytes to be copied in skb, so there is
1091                  * enough room for skb expand head in case it is used.
1092                  * The rest of the buffer is mapped from userspace.
1093                  */
1094                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1095                 if (copylen > good_linear)
1096                         copylen = good_linear;
1097                 linear = copylen;
1098                 iov_iter_advance(&i, copylen);
1099                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1100                         zerocopy = true;
1101         }
1102
1103         if (!zerocopy) {
1104                 copylen = len;
1105                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1106                         linear = good_linear;
1107                 else
1108                         linear = tun16_to_cpu(tun, gso.hdr_len);
1109         }
1110
1111         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1112         if (IS_ERR(skb)) {
1113                 if (PTR_ERR(skb) != -EAGAIN)
1114                         tun->dev->stats.rx_dropped++;
1115                 return PTR_ERR(skb);
1116         }
1117
1118         if (zerocopy)
1119                 err = zerocopy_sg_from_iter(skb, from);
1120         else {
1121                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1122                 if (!err && msg_control) {
1123                         struct ubuf_info *uarg = msg_control;
1124                         uarg->callback(uarg, false);
1125                 }
1126         }
1127
1128         if (err) {
1129                 tun->dev->stats.rx_dropped++;
1130                 kfree_skb(skb);
1131                 return -EFAULT;
1132         }
1133
1134         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1135                 if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start),
1136                                           tun16_to_cpu(tun, gso.csum_offset))) {
1137                         tun->dev->stats.rx_frame_errors++;
1138                         kfree_skb(skb);
1139                         return -EINVAL;
1140                 }
1141         }
1142
1143         switch (tun->flags & TUN_TYPE_MASK) {
1144         case IFF_TUN:
1145                 if (tun->flags & IFF_NO_PI) {
1146                         switch (skb->data[0] & 0xf0) {
1147                         case 0x40:
1148                                 pi.proto = htons(ETH_P_IP);
1149                                 break;
1150                         case 0x60:
1151                                 pi.proto = htons(ETH_P_IPV6);
1152                                 break;
1153                         default:
1154                                 tun->dev->stats.rx_dropped++;
1155                                 kfree_skb(skb);
1156                                 return -EINVAL;
1157                         }
1158                 }
1159
1160                 skb_reset_mac_header(skb);
1161                 skb->protocol = pi.proto;
1162                 skb->dev = tun->dev;
1163                 break;
1164         case IFF_TAP:
1165                 skb->protocol = eth_type_trans(skb, tun->dev);
1166                 break;
1167         }
1168
1169         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1170                 pr_debug("GSO!\n");
1171                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1172                 case VIRTIO_NET_HDR_GSO_TCPV4:
1173                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1174                         break;
1175                 case VIRTIO_NET_HDR_GSO_TCPV6:
1176                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1177                         break;
1178                 case VIRTIO_NET_HDR_GSO_UDP:
1179                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1180                         break;
1181                 default:
1182                         tun->dev->stats.rx_frame_errors++;
1183                         kfree_skb(skb);
1184                         return -EINVAL;
1185                 }
1186
1187                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1188                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1189
1190                 skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size);
1191                 if (skb_shinfo(skb)->gso_size == 0) {
1192                         tun->dev->stats.rx_frame_errors++;
1193                         kfree_skb(skb);
1194                         return -EINVAL;
1195                 }
1196
1197                 /* Header must be checked, and gso_segs computed. */
1198                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1199                 skb_shinfo(skb)->gso_segs = 0;
1200         }
1201
1202         /* copy skb_ubuf_info for callback when skb has no error */
1203         if (zerocopy) {
1204                 skb_shinfo(skb)->destructor_arg = msg_control;
1205                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1206                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1207         }
1208
1209         skb_reset_network_header(skb);
1210         skb_probe_transport_header(skb, 0);
1211
1212         rxhash = skb_get_hash(skb);
1213         netif_rx_ni(skb);
1214
1215         tun->dev->stats.rx_packets++;
1216         tun->dev->stats.rx_bytes += len;
1217
1218         tun_flow_update(tun, rxhash, tfile);
1219         return total_len;
1220 }
1221
1222 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1223 {
1224         struct file *file = iocb->ki_filp;
1225         struct tun_struct *tun = tun_get(file);
1226         struct tun_file *tfile = file->private_data;
1227         ssize_t result;
1228
1229         if (!tun)
1230                 return -EBADFD;
1231
1232         result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1233
1234         tun_put(tun);
1235         return result;
1236 }
1237
1238 /* Put packet to the user space buffer */
1239 static ssize_t tun_put_user(struct tun_struct *tun,
1240                             struct tun_file *tfile,
1241                             struct sk_buff *skb,
1242                             struct iov_iter *iter)
1243 {
1244         struct tun_pi pi = { 0, skb->protocol };
1245         ssize_t total;
1246         int vlan_offset = 0;
1247         int vlan_hlen = 0;
1248         int vnet_hdr_sz = 0;
1249
1250         if (vlan_tx_tag_present(skb))
1251                 vlan_hlen = VLAN_HLEN;
1252
1253         if (tun->flags & IFF_VNET_HDR)
1254                 vnet_hdr_sz = tun->vnet_hdr_sz;
1255
1256         total = skb->len + vlan_hlen + vnet_hdr_sz;
1257
1258         if (!(tun->flags & IFF_NO_PI)) {
1259                 if (iov_iter_count(iter) < sizeof(pi))
1260                         return -EINVAL;
1261
1262                 total += sizeof(pi);
1263                 if (iov_iter_count(iter) < total) {
1264                         /* Packet will be striped */
1265                         pi.flags |= TUN_PKT_STRIP;
1266                 }
1267
1268                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1269                         return -EFAULT;
1270         }
1271
1272         if (vnet_hdr_sz) {
1273                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1274                 if (iov_iter_count(iter) < vnet_hdr_sz)
1275                         return -EINVAL;
1276
1277                 if (skb_is_gso(skb)) {
1278                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1279
1280                         /* This is a hint as to how much should be linear. */
1281                         gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb));
1282                         gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size);
1283                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1284                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1285                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1286                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1287                         else if (sinfo->gso_type & SKB_GSO_UDP)
1288                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1289                         else {
1290                                 pr_err("unexpected GSO type: "
1291                                        "0x%x, gso_size %d, hdr_len %d\n",
1292                                        sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1293                                        tun16_to_cpu(tun, gso.hdr_len));
1294                                 print_hex_dump(KERN_ERR, "tun: ",
1295                                                DUMP_PREFIX_NONE,
1296                                                16, 1, skb->head,
1297                                                min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1298                                 WARN_ON_ONCE(1);
1299                                 return -EINVAL;
1300                         }
1301                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1302                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1303                 } else
1304                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1305
1306                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1307                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1308                         gso.csum_start = cpu_to_tun16(tun, skb_checksum_start_offset(skb) +
1309                                                       vlan_hlen);
1310                         gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset);
1311                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1312                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1313                 } /* else everything is zero */
1314
1315                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1316                         return -EFAULT;
1317
1318                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1319         }
1320
1321         if (vlan_hlen) {
1322                 int ret;
1323                 struct {
1324                         __be16 h_vlan_proto;
1325                         __be16 h_vlan_TCI;
1326                 } veth;
1327
1328                 veth.h_vlan_proto = skb->vlan_proto;
1329                 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1330
1331                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1332
1333                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1334                 if (ret || !iov_iter_count(iter))
1335                         goto done;
1336
1337                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1338                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1339                         goto done;
1340         }
1341
1342         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1343
1344 done:
1345         tun->dev->stats.tx_packets++;
1346         tun->dev->stats.tx_bytes += skb->len + vlan_hlen;
1347
1348         return total;
1349 }
1350
1351 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1352                            struct iov_iter *to,
1353                            int noblock)
1354 {
1355         struct sk_buff *skb;
1356         ssize_t ret;
1357         int peeked, err, off = 0;
1358
1359         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1360
1361         if (!iov_iter_count(to))
1362                 return 0;
1363
1364         if (tun->dev->reg_state != NETREG_REGISTERED)
1365                 return -EIO;
1366
1367         /* Read frames from queue */
1368         skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1369                                   &peeked, &off, &err);
1370         if (!skb)
1371                 return 0;
1372
1373         ret = tun_put_user(tun, tfile, skb, to);
1374         if (unlikely(ret < 0))
1375                 kfree_skb(skb);
1376         else
1377                 consume_skb(skb);
1378
1379         return ret;
1380 }
1381
1382 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1383 {
1384         struct file *file = iocb->ki_filp;
1385         struct tun_file *tfile = file->private_data;
1386         struct tun_struct *tun = __tun_get(tfile);
1387         ssize_t len = iov_iter_count(to), ret;
1388
1389         if (!tun)
1390                 return -EBADFD;
1391         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1392         ret = min_t(ssize_t, ret, len);
1393         if (ret > 0)
1394                 iocb->ki_pos = ret;
1395         tun_put(tun);
1396         return ret;
1397 }
1398
1399 static void tun_free_netdev(struct net_device *dev)
1400 {
1401         struct tun_struct *tun = netdev_priv(dev);
1402
1403         BUG_ON(!(list_empty(&tun->disabled)));
1404         tun_flow_uninit(tun);
1405         security_tun_dev_free_security(tun->security);
1406         free_netdev(dev);
1407 }
1408
1409 static void tun_setup(struct net_device *dev)
1410 {
1411         struct tun_struct *tun = netdev_priv(dev);
1412
1413         tun->owner = INVALID_UID;
1414         tun->group = INVALID_GID;
1415
1416         dev->ethtool_ops = &tun_ethtool_ops;
1417         dev->destructor = tun_free_netdev;
1418 }
1419
1420 /* Trivial set of netlink ops to allow deleting tun or tap
1421  * device with netlink.
1422  */
1423 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1424 {
1425         return -EINVAL;
1426 }
1427
1428 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1429         .kind           = DRV_NAME,
1430         .priv_size      = sizeof(struct tun_struct),
1431         .setup          = tun_setup,
1432         .validate       = tun_validate,
1433 };
1434
1435 static void tun_sock_write_space(struct sock *sk)
1436 {
1437         struct tun_file *tfile;
1438         wait_queue_head_t *wqueue;
1439
1440         if (!sock_writeable(sk))
1441                 return;
1442
1443         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1444                 return;
1445
1446         wqueue = sk_sleep(sk);
1447         if (wqueue && waitqueue_active(wqueue))
1448                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1449                                                 POLLWRNORM | POLLWRBAND);
1450
1451         tfile = container_of(sk, struct tun_file, sk);
1452         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1453 }
1454
1455 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1456                        struct msghdr *m, size_t total_len)
1457 {
1458         int ret;
1459         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1460         struct tun_struct *tun = __tun_get(tfile);
1461
1462         if (!tun)
1463                 return -EBADFD;
1464
1465         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1466                            m->msg_flags & MSG_DONTWAIT);
1467         tun_put(tun);
1468         return ret;
1469 }
1470
1471 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1472                        struct msghdr *m, size_t total_len,
1473                        int flags)
1474 {
1475         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1476         struct tun_struct *tun = __tun_get(tfile);
1477         int ret;
1478
1479         if (!tun)
1480                 return -EBADFD;
1481
1482         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1483                 ret = -EINVAL;
1484                 goto out;
1485         }
1486         if (flags & MSG_ERRQUEUE) {
1487                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1488                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1489                 goto out;
1490         }
1491         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1492         if (ret > total_len) {
1493                 m->msg_flags |= MSG_TRUNC;
1494                 ret = flags & MSG_TRUNC ? ret : total_len;
1495         }
1496 out:
1497         tun_put(tun);
1498         return ret;
1499 }
1500
1501 static int tun_release(struct socket *sock)
1502 {
1503         if (sock->sk)
1504                 sock_put(sock->sk);
1505         return 0;
1506 }
1507
1508 /* Ops structure to mimic raw sockets with tun */
1509 static const struct proto_ops tun_socket_ops = {
1510         .sendmsg = tun_sendmsg,
1511         .recvmsg = tun_recvmsg,
1512         .release = tun_release,
1513 };
1514
1515 static struct proto tun_proto = {
1516         .name           = "tun",
1517         .owner          = THIS_MODULE,
1518         .obj_size       = sizeof(struct tun_file),
1519 };
1520
1521 static int tun_flags(struct tun_struct *tun)
1522 {
1523         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1524 }
1525
1526 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1527                               char *buf)
1528 {
1529         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1530         return sprintf(buf, "0x%x\n", tun_flags(tun));
1531 }
1532
1533 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1534                               char *buf)
1535 {
1536         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1537         return uid_valid(tun->owner)?
1538                 sprintf(buf, "%u\n",
1539                         from_kuid_munged(current_user_ns(), tun->owner)):
1540                 sprintf(buf, "-1\n");
1541 }
1542
1543 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1544                               char *buf)
1545 {
1546         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1547         return gid_valid(tun->group) ?
1548                 sprintf(buf, "%u\n",
1549                         from_kgid_munged(current_user_ns(), tun->group)):
1550                 sprintf(buf, "-1\n");
1551 }
1552
1553 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1554 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1555 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1556
1557 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1558 {
1559         struct tun_struct *tun;
1560         struct tun_file *tfile = file->private_data;
1561         struct net_device *dev;
1562         int err;
1563
1564         if (tfile->detached)
1565                 return -EINVAL;
1566
1567         dev = __dev_get_by_name(net, ifr->ifr_name);
1568         if (dev) {
1569                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1570                         return -EBUSY;
1571                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1572                         tun = netdev_priv(dev);
1573                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1574                         tun = netdev_priv(dev);
1575                 else
1576                         return -EINVAL;
1577
1578                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1579                     !!(tun->flags & IFF_MULTI_QUEUE))
1580                         return -EINVAL;
1581
1582                 if (tun_not_capable(tun))
1583                         return -EPERM;
1584                 err = security_tun_dev_open(tun->security);
1585                 if (err < 0)
1586                         return err;
1587
1588                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1589                 if (err < 0)
1590                         return err;
1591
1592                 if (tun->flags & IFF_MULTI_QUEUE &&
1593                     (tun->numqueues + tun->numdisabled > 1)) {
1594                         /* One or more queue has already been attached, no need
1595                          * to initialize the device again.
1596                          */
1597                         return 0;
1598                 }
1599         }
1600         else {
1601                 char *name;
1602                 unsigned long flags = 0;
1603                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1604                              MAX_TAP_QUEUES : 1;
1605
1606                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1607                         return -EPERM;
1608                 err = security_tun_dev_create();
1609                 if (err < 0)
1610                         return err;
1611
1612                 /* Set dev type */
1613                 if (ifr->ifr_flags & IFF_TUN) {
1614                         /* TUN device */
1615                         flags |= IFF_TUN;
1616                         name = "tun%d";
1617                 } else if (ifr->ifr_flags & IFF_TAP) {
1618                         /* TAP device */
1619                         flags |= IFF_TAP;
1620                         name = "tap%d";
1621                 } else
1622                         return -EINVAL;
1623
1624                 if (*ifr->ifr_name)
1625                         name = ifr->ifr_name;
1626
1627                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1628                                        NET_NAME_UNKNOWN, tun_setup, queues,
1629                                        queues);
1630
1631                 if (!dev)
1632                         return -ENOMEM;
1633
1634                 dev_net_set(dev, net);
1635                 dev->rtnl_link_ops = &tun_link_ops;
1636                 dev->ifindex = tfile->ifindex;
1637
1638                 tun = netdev_priv(dev);
1639                 tun->dev = dev;
1640                 tun->flags = flags;
1641                 tun->txflt.count = 0;
1642                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1643
1644                 tun->filter_attached = false;
1645                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1646
1647                 spin_lock_init(&tun->lock);
1648
1649                 err = security_tun_dev_alloc_security(&tun->security);
1650                 if (err < 0)
1651                         goto err_free_dev;
1652
1653                 tun_net_init(dev);
1654                 tun_flow_init(tun);
1655
1656                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1657                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1658                                    NETIF_F_HW_VLAN_STAG_TX;
1659                 dev->features = dev->hw_features;
1660                 dev->vlan_features = dev->features &
1661                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1662                                        NETIF_F_HW_VLAN_STAG_TX);
1663
1664                 INIT_LIST_HEAD(&tun->disabled);
1665                 err = tun_attach(tun, file, false);
1666                 if (err < 0)
1667                         goto err_free_flow;
1668
1669                 err = register_netdevice(tun->dev);
1670                 if (err < 0)
1671                         goto err_detach;
1672
1673                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1674                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1675                     device_create_file(&tun->dev->dev, &dev_attr_group))
1676                         pr_err("Failed to create tun sysfs files\n");
1677         }
1678
1679         netif_carrier_on(tun->dev);
1680
1681         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1682
1683         tun->flags = (tun->flags & ~TUN_FEATURES) |
1684                 (ifr->ifr_flags & TUN_FEATURES);
1685
1686         /* Make sure persistent devices do not get stuck in
1687          * xoff state.
1688          */
1689         if (netif_running(tun->dev))
1690                 netif_tx_wake_all_queues(tun->dev);
1691
1692         strcpy(ifr->ifr_name, tun->dev->name);
1693         return 0;
1694
1695 err_detach:
1696         tun_detach_all(dev);
1697 err_free_flow:
1698         tun_flow_uninit(tun);
1699         security_tun_dev_free_security(tun->security);
1700 err_free_dev:
1701         free_netdev(dev);
1702         return err;
1703 }
1704
1705 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1706                        struct ifreq *ifr)
1707 {
1708         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1709
1710         strcpy(ifr->ifr_name, tun->dev->name);
1711
1712         ifr->ifr_flags = tun_flags(tun);
1713
1714 }
1715
1716 /* This is like a cut-down ethtool ops, except done via tun fd so no
1717  * privs required. */
1718 static int set_offload(struct tun_struct *tun, unsigned long arg)
1719 {
1720         netdev_features_t features = 0;
1721
1722         if (arg & TUN_F_CSUM) {
1723                 features |= NETIF_F_HW_CSUM;
1724                 arg &= ~TUN_F_CSUM;
1725
1726                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1727                         if (arg & TUN_F_TSO_ECN) {
1728                                 features |= NETIF_F_TSO_ECN;
1729                                 arg &= ~TUN_F_TSO_ECN;
1730                         }
1731                         if (arg & TUN_F_TSO4)
1732                                 features |= NETIF_F_TSO;
1733                         if (arg & TUN_F_TSO6)
1734                                 features |= NETIF_F_TSO6;
1735                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1736                 }
1737
1738                 if (arg & TUN_F_UFO) {
1739                         features |= NETIF_F_UFO;
1740                         arg &= ~TUN_F_UFO;
1741                 }
1742         }
1743
1744         /* This gives the user a way to test for new features in future by
1745          * trying to set them. */
1746         if (arg)
1747                 return -EINVAL;
1748
1749         tun->set_features = features;
1750         netdev_update_features(tun->dev);
1751
1752         return 0;
1753 }
1754
1755 static void tun_detach_filter(struct tun_struct *tun, int n)
1756 {
1757         int i;
1758         struct tun_file *tfile;
1759
1760         for (i = 0; i < n; i++) {
1761                 tfile = rtnl_dereference(tun->tfiles[i]);
1762                 sk_detach_filter(tfile->socket.sk);
1763         }
1764
1765         tun->filter_attached = false;
1766 }
1767
1768 static int tun_attach_filter(struct tun_struct *tun)
1769 {
1770         int i, ret = 0;
1771         struct tun_file *tfile;
1772
1773         for (i = 0; i < tun->numqueues; i++) {
1774                 tfile = rtnl_dereference(tun->tfiles[i]);
1775                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1776                 if (ret) {
1777                         tun_detach_filter(tun, i);
1778                         return ret;
1779                 }
1780         }
1781
1782         tun->filter_attached = true;
1783         return ret;
1784 }
1785
1786 static void tun_set_sndbuf(struct tun_struct *tun)
1787 {
1788         struct tun_file *tfile;
1789         int i;
1790
1791         for (i = 0; i < tun->numqueues; i++) {
1792                 tfile = rtnl_dereference(tun->tfiles[i]);
1793                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1794         }
1795 }
1796
1797 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1798 {
1799         struct tun_file *tfile = file->private_data;
1800         struct tun_struct *tun;
1801         int ret = 0;
1802
1803         rtnl_lock();
1804
1805         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1806                 tun = tfile->detached;
1807                 if (!tun) {
1808                         ret = -EINVAL;
1809                         goto unlock;
1810                 }
1811                 ret = security_tun_dev_attach_queue(tun->security);
1812                 if (ret < 0)
1813                         goto unlock;
1814                 ret = tun_attach(tun, file, false);
1815         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1816                 tun = rtnl_dereference(tfile->tun);
1817                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1818                         ret = -EINVAL;
1819                 else
1820                         __tun_detach(tfile, false);
1821         } else
1822                 ret = -EINVAL;
1823
1824 unlock:
1825         rtnl_unlock();
1826         return ret;
1827 }
1828
1829 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1830                             unsigned long arg, int ifreq_len)
1831 {
1832         struct tun_file *tfile = file->private_data;
1833         struct tun_struct *tun;
1834         void __user* argp = (void __user*)arg;
1835         struct ifreq ifr;
1836         kuid_t owner;
1837         kgid_t group;
1838         int sndbuf;
1839         int vnet_hdr_sz;
1840         unsigned int ifindex;
1841         int le;
1842         int ret;
1843
1844         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1845                 if (copy_from_user(&ifr, argp, ifreq_len))
1846                         return -EFAULT;
1847         } else {
1848                 memset(&ifr, 0, sizeof(ifr));
1849         }
1850         if (cmd == TUNGETFEATURES) {
1851                 /* Currently this just means: "what IFF flags are valid?".
1852                  * This is needed because we never checked for invalid flags on
1853                  * TUNSETIFF.
1854                  */
1855                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1856                                 (unsigned int __user*)argp);
1857         } else if (cmd == TUNSETQUEUE)
1858                 return tun_set_queue(file, &ifr);
1859
1860         ret = 0;
1861         rtnl_lock();
1862
1863         tun = __tun_get(tfile);
1864         if (cmd == TUNSETIFF && !tun) {
1865                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1866
1867                 ret = tun_set_iff(tfile->net, file, &ifr);
1868
1869                 if (ret)
1870                         goto unlock;
1871
1872                 if (copy_to_user(argp, &ifr, ifreq_len))
1873                         ret = -EFAULT;
1874                 goto unlock;
1875         }
1876         if (cmd == TUNSETIFINDEX) {
1877                 ret = -EPERM;
1878                 if (tun)
1879                         goto unlock;
1880
1881                 ret = -EFAULT;
1882                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1883                         goto unlock;
1884
1885                 ret = 0;
1886                 tfile->ifindex = ifindex;
1887                 goto unlock;
1888         }
1889
1890         ret = -EBADFD;
1891         if (!tun)
1892                 goto unlock;
1893
1894         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1895
1896         ret = 0;
1897         switch (cmd) {
1898         case TUNGETIFF:
1899                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1900
1901                 if (tfile->detached)
1902                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
1903                 if (!tfile->socket.sk->sk_filter)
1904                         ifr.ifr_flags |= IFF_NOFILTER;
1905
1906                 if (copy_to_user(argp, &ifr, ifreq_len))
1907                         ret = -EFAULT;
1908                 break;
1909
1910         case TUNSETNOCSUM:
1911                 /* Disable/Enable checksum */
1912
1913                 /* [unimplemented] */
1914                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1915                           arg ? "disabled" : "enabled");
1916                 break;
1917
1918         case TUNSETPERSIST:
1919                 /* Disable/Enable persist mode. Keep an extra reference to the
1920                  * module to prevent the module being unprobed.
1921                  */
1922                 if (arg && !(tun->flags & IFF_PERSIST)) {
1923                         tun->flags |= IFF_PERSIST;
1924                         __module_get(THIS_MODULE);
1925                 }
1926                 if (!arg && (tun->flags & IFF_PERSIST)) {
1927                         tun->flags &= ~IFF_PERSIST;
1928                         module_put(THIS_MODULE);
1929                 }
1930
1931                 tun_debug(KERN_INFO, tun, "persist %s\n",
1932                           arg ? "enabled" : "disabled");
1933                 break;
1934
1935         case TUNSETOWNER:
1936                 /* Set owner of the device */
1937                 owner = make_kuid(current_user_ns(), arg);
1938                 if (!uid_valid(owner)) {
1939                         ret = -EINVAL;
1940                         break;
1941                 }
1942                 tun->owner = owner;
1943                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1944                           from_kuid(&init_user_ns, tun->owner));
1945                 break;
1946
1947         case TUNSETGROUP:
1948                 /* Set group of the device */
1949                 group = make_kgid(current_user_ns(), arg);
1950                 if (!gid_valid(group)) {
1951                         ret = -EINVAL;
1952                         break;
1953                 }
1954                 tun->group = group;
1955                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1956                           from_kgid(&init_user_ns, tun->group));
1957                 break;
1958
1959         case TUNSETLINK:
1960                 /* Only allow setting the type when the interface is down */
1961                 if (tun->dev->flags & IFF_UP) {
1962                         tun_debug(KERN_INFO, tun,
1963                                   "Linktype set failed because interface is up\n");
1964                         ret = -EBUSY;
1965                 } else {
1966                         tun->dev->type = (int) arg;
1967                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1968                                   tun->dev->type);
1969                         ret = 0;
1970                 }
1971                 break;
1972
1973 #ifdef TUN_DEBUG
1974         case TUNSETDEBUG:
1975                 tun->debug = arg;
1976                 break;
1977 #endif
1978         case TUNSETOFFLOAD:
1979                 ret = set_offload(tun, arg);
1980                 break;
1981
1982         case TUNSETTXFILTER:
1983                 /* Can be set only for TAPs */
1984                 ret = -EINVAL;
1985                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1986                         break;
1987                 ret = update_filter(&tun->txflt, (void __user *)arg);
1988                 break;
1989
1990         case SIOCGIFHWADDR:
1991                 /* Get hw address */
1992                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1993                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1994                 if (copy_to_user(argp, &ifr, ifreq_len))
1995                         ret = -EFAULT;
1996                 break;
1997
1998         case SIOCSIFHWADDR:
1999                 /* Set hw address */
2000                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2001                           ifr.ifr_hwaddr.sa_data);
2002
2003                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2004                 break;
2005
2006         case TUNGETSNDBUF:
2007                 sndbuf = tfile->socket.sk->sk_sndbuf;
2008                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2009                         ret = -EFAULT;
2010                 break;
2011
2012         case TUNSETSNDBUF:
2013                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2014                         ret = -EFAULT;
2015                         break;
2016                 }
2017
2018                 tun->sndbuf = sndbuf;
2019                 tun_set_sndbuf(tun);
2020                 break;
2021
2022         case TUNGETVNETHDRSZ:
2023                 vnet_hdr_sz = tun->vnet_hdr_sz;
2024                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2025                         ret = -EFAULT;
2026                 break;
2027
2028         case TUNSETVNETHDRSZ:
2029                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2030                         ret = -EFAULT;
2031                         break;
2032                 }
2033                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2034                         ret = -EINVAL;
2035                         break;
2036                 }
2037
2038                 tun->vnet_hdr_sz = vnet_hdr_sz;
2039                 break;
2040
2041         case TUNGETVNETLE:
2042                 le = !!(tun->flags & TUN_VNET_LE);
2043                 if (put_user(le, (int __user *)argp))
2044                         ret = -EFAULT;
2045                 break;
2046
2047         case TUNSETVNETLE:
2048                 if (get_user(le, (int __user *)argp)) {
2049                         ret = -EFAULT;
2050                         break;
2051                 }
2052                 if (le)
2053                         tun->flags |= TUN_VNET_LE;
2054                 else
2055                         tun->flags &= ~TUN_VNET_LE;
2056                 break;
2057
2058         case TUNATTACHFILTER:
2059                 /* Can be set only for TAPs */
2060                 ret = -EINVAL;
2061                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2062                         break;
2063                 ret = -EFAULT;
2064                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2065                         break;
2066
2067                 ret = tun_attach_filter(tun);
2068                 break;
2069
2070         case TUNDETACHFILTER:
2071                 /* Can be set only for TAPs */
2072                 ret = -EINVAL;
2073                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2074                         break;
2075                 ret = 0;
2076                 tun_detach_filter(tun, tun->numqueues);
2077                 break;
2078
2079         case TUNGETFILTER:
2080                 ret = -EINVAL;
2081                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2082                         break;
2083                 ret = -EFAULT;
2084                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2085                         break;
2086                 ret = 0;
2087                 break;
2088
2089         default:
2090                 ret = -EINVAL;
2091                 break;
2092         }
2093
2094 unlock:
2095         rtnl_unlock();
2096         if (tun)
2097                 tun_put(tun);
2098         return ret;
2099 }
2100
2101 static long tun_chr_ioctl(struct file *file,
2102                           unsigned int cmd, unsigned long arg)
2103 {
2104         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2105 }
2106
2107 #ifdef CONFIG_COMPAT
2108 static long tun_chr_compat_ioctl(struct file *file,
2109                          unsigned int cmd, unsigned long arg)
2110 {
2111         switch (cmd) {
2112         case TUNSETIFF:
2113         case TUNGETIFF:
2114         case TUNSETTXFILTER:
2115         case TUNGETSNDBUF:
2116         case TUNSETSNDBUF:
2117         case SIOCGIFHWADDR:
2118         case SIOCSIFHWADDR:
2119                 arg = (unsigned long)compat_ptr(arg);
2120                 break;
2121         default:
2122                 arg = (compat_ulong_t)arg;
2123                 break;
2124         }
2125
2126         /*
2127          * compat_ifreq is shorter than ifreq, so we must not access beyond
2128          * the end of that structure. All fields that are used in this
2129          * driver are compatible though, we don't need to convert the
2130          * contents.
2131          */
2132         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2133 }
2134 #endif /* CONFIG_COMPAT */
2135
2136 static int tun_chr_fasync(int fd, struct file *file, int on)
2137 {
2138         struct tun_file *tfile = file->private_data;
2139         int ret;
2140
2141         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2142                 goto out;
2143
2144         if (on) {
2145                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2146                 tfile->flags |= TUN_FASYNC;
2147         } else
2148                 tfile->flags &= ~TUN_FASYNC;
2149         ret = 0;
2150 out:
2151         return ret;
2152 }
2153
2154 static int tun_chr_open(struct inode *inode, struct file * file)
2155 {
2156         struct tun_file *tfile;
2157
2158         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2159
2160         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2161                                             &tun_proto);
2162         if (!tfile)
2163                 return -ENOMEM;
2164         RCU_INIT_POINTER(tfile->tun, NULL);
2165         tfile->net = get_net(current->nsproxy->net_ns);
2166         tfile->flags = 0;
2167         tfile->ifindex = 0;
2168
2169         init_waitqueue_head(&tfile->wq.wait);
2170         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2171
2172         tfile->socket.file = file;
2173         tfile->socket.ops = &tun_socket_ops;
2174
2175         sock_init_data(&tfile->socket, &tfile->sk);
2176         sk_change_net(&tfile->sk, tfile->net);
2177
2178         tfile->sk.sk_write_space = tun_sock_write_space;
2179         tfile->sk.sk_sndbuf = INT_MAX;
2180
2181         file->private_data = tfile;
2182         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2183         INIT_LIST_HEAD(&tfile->next);
2184
2185         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2186
2187         return 0;
2188 }
2189
2190 static int tun_chr_close(struct inode *inode, struct file *file)
2191 {
2192         struct tun_file *tfile = file->private_data;
2193         struct net *net = tfile->net;
2194
2195         tun_detach(tfile, true);
2196         put_net(net);
2197
2198         return 0;
2199 }
2200
2201 #ifdef CONFIG_PROC_FS
2202 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2203 {
2204         struct tun_struct *tun;
2205         struct ifreq ifr;
2206
2207         memset(&ifr, 0, sizeof(ifr));
2208
2209         rtnl_lock();
2210         tun = tun_get(f);
2211         if (tun)
2212                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2213         rtnl_unlock();
2214
2215         if (tun)
2216                 tun_put(tun);
2217
2218         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2219 }
2220 #endif
2221
2222 static const struct file_operations tun_fops = {
2223         .owner  = THIS_MODULE,
2224         .llseek = no_llseek,
2225         .read  = new_sync_read,
2226         .write = new_sync_write,
2227         .read_iter  = tun_chr_read_iter,
2228         .write_iter = tun_chr_write_iter,
2229         .poll   = tun_chr_poll,
2230         .unlocked_ioctl = tun_chr_ioctl,
2231 #ifdef CONFIG_COMPAT
2232         .compat_ioctl = tun_chr_compat_ioctl,
2233 #endif
2234         .open   = tun_chr_open,
2235         .release = tun_chr_close,
2236         .fasync = tun_chr_fasync,
2237 #ifdef CONFIG_PROC_FS
2238         .show_fdinfo = tun_chr_show_fdinfo,
2239 #endif
2240 };
2241
2242 static struct miscdevice tun_miscdev = {
2243         .minor = TUN_MINOR,
2244         .name = "tun",
2245         .nodename = "net/tun",
2246         .fops = &tun_fops,
2247 };
2248
2249 /* ethtool interface */
2250
2251 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2252 {
2253         cmd->supported          = 0;
2254         cmd->advertising        = 0;
2255         ethtool_cmd_speed_set(cmd, SPEED_10);
2256         cmd->duplex             = DUPLEX_FULL;
2257         cmd->port               = PORT_TP;
2258         cmd->phy_address        = 0;
2259         cmd->transceiver        = XCVR_INTERNAL;
2260         cmd->autoneg            = AUTONEG_DISABLE;
2261         cmd->maxtxpkt           = 0;
2262         cmd->maxrxpkt           = 0;
2263         return 0;
2264 }
2265
2266 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2267 {
2268         struct tun_struct *tun = netdev_priv(dev);
2269
2270         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2271         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2272
2273         switch (tun->flags & TUN_TYPE_MASK) {
2274         case IFF_TUN:
2275                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2276                 break;
2277         case IFF_TAP:
2278                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2279                 break;
2280         }
2281 }
2282
2283 static u32 tun_get_msglevel(struct net_device *dev)
2284 {
2285 #ifdef TUN_DEBUG
2286         struct tun_struct *tun = netdev_priv(dev);
2287         return tun->debug;
2288 #else
2289         return -EOPNOTSUPP;
2290 #endif
2291 }
2292
2293 static void tun_set_msglevel(struct net_device *dev, u32 value)
2294 {
2295 #ifdef TUN_DEBUG
2296         struct tun_struct *tun = netdev_priv(dev);
2297         tun->debug = value;
2298 #endif
2299 }
2300
2301 static const struct ethtool_ops tun_ethtool_ops = {
2302         .get_settings   = tun_get_settings,
2303         .get_drvinfo    = tun_get_drvinfo,
2304         .get_msglevel   = tun_get_msglevel,
2305         .set_msglevel   = tun_set_msglevel,
2306         .get_link       = ethtool_op_get_link,
2307         .get_ts_info    = ethtool_op_get_ts_info,
2308 };
2309
2310
2311 static int __init tun_init(void)
2312 {
2313         int ret = 0;
2314
2315         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2316         pr_info("%s\n", DRV_COPYRIGHT);
2317
2318         ret = rtnl_link_register(&tun_link_ops);
2319         if (ret) {
2320                 pr_err("Can't register link_ops\n");
2321                 goto err_linkops;
2322         }
2323
2324         ret = misc_register(&tun_miscdev);
2325         if (ret) {
2326                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2327                 goto err_misc;
2328         }
2329         return  0;
2330 err_misc:
2331         rtnl_link_unregister(&tun_link_ops);
2332 err_linkops:
2333         return ret;
2334 }
2335
2336 static void tun_cleanup(void)
2337 {
2338         misc_deregister(&tun_miscdev);
2339         rtnl_link_unregister(&tun_link_ops);
2340 }
2341
2342 /* Get an underlying socket object from tun file.  Returns error unless file is
2343  * attached to a device.  The returned object works like a packet socket, it
2344  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2345  * holding a reference to the file for as long as the socket is in use. */
2346 struct socket *tun_get_socket(struct file *file)
2347 {
2348         struct tun_file *tfile;
2349         if (file->f_op != &tun_fops)
2350                 return ERR_PTR(-EINVAL);
2351         tfile = file->private_data;
2352         if (!tfile)
2353                 return ERR_PTR(-EBADFD);
2354         return &tfile->socket;
2355 }
2356 EXPORT_SYMBOL_GPL(tun_get_socket);
2357
2358 module_init(tun_init);
2359 module_exit(tun_cleanup);
2360 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2361 MODULE_AUTHOR(DRV_COPYRIGHT);
2362 MODULE_LICENSE("GPL");
2363 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2364 MODULE_ALIAS("devname:net/tun");