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/ipv6.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <linux/seq_file.h>
  74 #include <linux/uio.h>
  75
  76 #include <asm/uaccess.h>
  77
  78 /* Uncomment to enable debugging */
  79 /* #define TUN_DEBUG 1 */
  80
  81 #ifdef TUN_DEBUG
  82 static int debug;
  83
  84 #define tun_debug(level, tun, fmt, args...)                     \
  85 do {                                                            \
  86         if (tun->debug)                                         \
  87                 netdev_printk(level, tun->dev, fmt, ##args);    \
  88 } while (0)
  89 #define DBG1(level, fmt, args...)                               \
  90 do {                                                            \
  91         if (debug == 2)                                         \
  92                 printk(level fmt, ##args);                      \
  93 } while (0)
  94 #else
  95 #define tun_debug(level, tun, fmt, args...)                     \
  96 do {                                                            \
  97         if (0)                                                  \
  98                 netdev_printk(level, tun->dev, fmt, ##args);    \
  99 } while (0)
 100 #define DBG1(level, fmt, args...)                               \
 101 do {                                                            \
 102         if (0)                                                  \
 103                 printk(level fmt, ##args);                      \
 104 } while (0)
 105 #endif
 106
 107 /* TUN device flags */
 108
 109 /* IFF_ATTACH_QUEUE is never stored in device flags,
 110  * overload it to mean fasync when stored there.
 111  */
 112 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 113 /* High bits in flags field are unused. */
 114 #define TUN_VNET_LE     0x80000000
 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 /* DEFAULT_MAX_NUM_RSS_QUEUES were chosen to let the rx/tx queues allocated for
 128  * the netdevice to be fit in one page. So we can make sure the success of
 129  * memory allocation. TODO: increase the limit. */
 130 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
 131 #define MAX_TAP_FLOWS  4096
 132
 133 #define TUN_FLOW_EXPIRE (3 * HZ)
 134
 135 /* A tun_file connects an open character device to a tuntap netdevice. It
 136  * also contains all socket related structures (except sock_fprog and tap_filter)
 137  * to serve as one transmit queue for tuntap device. The sock_fprog and
 138  * tap_filter were kept in tun_struct since they were used for filtering for the
 139  * netdevice not for a specific queue (at least I didn't see the requirement for
 140  * this).
 141  *
 142  * RCU usage:
 143  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 144  * other can only be read while rcu_read_lock or rtnl_lock is held.
 145  */
 146 struct tun_file {
 147         struct sock sk;
 148         struct socket socket;
 149         struct socket_wq wq;
 150         struct tun_struct __rcu *tun;
 151         struct net *net;
 152         struct fasync_struct *fasync;
 153         /* only used for fasnyc */
 154         unsigned int flags;
 155         union {
 156                 u16 queue_index;
 157                 unsigned int ifindex;
 158         };
 159         struct list_head next;
 160         struct tun_struct *detached;
 161 };
 162
 163 struct tun_flow_entry {
 164         struct hlist_node hash_link;
 165         struct rcu_head rcu;
 166         struct tun_struct *tun;
 167
 168         u32 rxhash;
 169         u32 rps_rxhash;
 170         int queue_index;
 171         unsigned long updated;
 172 };
 173
 174 #define TUN_NUM_FLOW_ENTRIES 1024
 175
 176 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 177  * device, socket filter, sndbuf and vnet header size were restore when the
 178  * file were attached to a persist device.
 179  */
 180 struct tun_struct {
 181         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 182         unsigned int            numqueues;
 183         unsigned int            flags;
 184         kuid_t                  owner;
 185         kgid_t                  group;
 186
 187         struct net_device       *dev;
 188         netdev_features_t       set_features;
 189 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 190                           NETIF_F_TSO6)
 191
 192         int                     vnet_hdr_sz;
 193         int                     sndbuf;
 194         struct tap_filter       txflt;
 195         struct sock_fprog       fprog;
 196         /* protected by rtnl lock */
 197         bool                    filter_attached;
 198 #ifdef TUN_DEBUG
 199         int debug;
 200 #endif
 201         spinlock_t lock;
 202         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 203         struct timer_list flow_gc_timer;
 204         unsigned long ageing_time;
 205         unsigned int numdisabled;
 206         struct list_head disabled;
 207         void *security;
 208         u32 flow_count;
 209 };
 210
 211 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 212 {
 213         return __virtio16_to_cpu(tun->flags & TUN_VNET_LE, val);
 214 }
 215
 216 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 217 {
 218         return __cpu_to_virtio16(tun->flags & TUN_VNET_LE, val);
 219 }
 220
 221 static inline u32 tun_hashfn(u32 rxhash)
 222 {
 223         return rxhash & 0x3ff;
 224 }
 225
 226 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 227 {
 228         struct tun_flow_entry *e;
 229
 230         hlist_for_each_entry_rcu(e, head, hash_link) {
 231                 if (e->rxhash == rxhash)
 232                         return e;
 233         }
 234         return NULL;
 235 }
 236
 237 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 238                                               struct hlist_head *head,
 239                                               u32 rxhash, u16 queue_index)
 240 {
 241         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 242
 243         if (e) {
 244                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 245                           rxhash, queue_index);
 246                 e->updated = jiffies;
 247                 e->rxhash = rxhash;
 248                 e->rps_rxhash = 0;
 249                 e->queue_index = queue_index;
 250                 e->tun = tun;
 251                 hlist_add_head_rcu(&e->hash_link, head);
 252                 ++tun->flow_count;
 253         }
 254         return e;
 255 }
 256
 257 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 258 {
 259         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 260                   e->rxhash, e->queue_index);
 261         sock_rps_reset_flow_hash(e->rps_rxhash);
 262         hlist_del_rcu(&e->hash_link);
 263         kfree_rcu(e, rcu);
 264         --tun->flow_count;
 265 }
 266
 267 static void tun_flow_flush(struct tun_struct *tun)
 268 {
 269         int i;
 270
 271         spin_lock_bh(&tun->lock);
 272         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 273                 struct tun_flow_entry *e;
 274                 struct hlist_node *n;
 275
 276                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 277                         tun_flow_delete(tun, e);
 278         }
 279         spin_unlock_bh(&tun->lock);
 280 }
 281
 282 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 283 {
 284         int i;
 285
 286         spin_lock_bh(&tun->lock);
 287         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 288                 struct tun_flow_entry *e;
 289                 struct hlist_node *n;
 290
 291                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 292                         if (e->queue_index == queue_index)
 293                                 tun_flow_delete(tun, e);
 294                 }
 295         }
 296         spin_unlock_bh(&tun->lock);
 297 }
 298
 299 static void tun_flow_cleanup(unsigned long data)
 300 {
 301         struct tun_struct *tun = (struct tun_struct *)data;
 302         unsigned long delay = tun->ageing_time;
 303         unsigned long next_timer = jiffies + delay;
 304         unsigned long count = 0;
 305         int i;
 306
 307         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 308
 309         spin_lock_bh(&tun->lock);
 310         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 311                 struct tun_flow_entry *e;
 312                 struct hlist_node *n;
 313
 314                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 315                         unsigned long this_timer;
 316                         count++;
 317                         this_timer = e->updated + delay;
 318                         if (time_before_eq(this_timer, jiffies))
 319                                 tun_flow_delete(tun, e);
 320                         else if (time_before(this_timer, next_timer))
 321                                 next_timer = this_timer;
 322                 }
 323         }
 324
 325         if (count)
 326                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 327         spin_unlock_bh(&tun->lock);
 328 }
 329
 330 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 331                             struct tun_file *tfile)
 332 {
 333         struct hlist_head *head;
 334         struct tun_flow_entry *e;
 335         unsigned long delay = tun->ageing_time;
 336         u16 queue_index = tfile->queue_index;
 337
 338         if (!rxhash)
 339                 return;
 340         else
 341                 head = &tun->flows[tun_hashfn(rxhash)];
 342
 343         rcu_read_lock();
 344
 345         /* We may get a very small possibility of OOO during switching, not
 346          * worth to optimize.*/
 347         if (tun->numqueues == 1 || tfile->detached)
 348                 goto unlock;
 349
 350         e = tun_flow_find(head, rxhash);
 351         if (likely(e)) {
 352                 /* TODO: keep queueing to old queue until it's empty? */
 353                 e->queue_index = queue_index;
 354                 e->updated = jiffies;
 355                 sock_rps_record_flow_hash(e->rps_rxhash);
 356         } else {
 357                 spin_lock_bh(&tun->lock);
 358                 if (!tun_flow_find(head, rxhash) &&
 359                     tun->flow_count < MAX_TAP_FLOWS)
 360                         tun_flow_create(tun, head, rxhash, queue_index);
 361
 362                 if (!timer_pending(&tun->flow_gc_timer))
 363                         mod_timer(&tun->flow_gc_timer,
 364                                   round_jiffies_up(jiffies + delay));
 365                 spin_unlock_bh(&tun->lock);
 366         }
 367
 368 unlock:
 369         rcu_read_unlock();
 370 }
 371
 372 /**
 373  * Save the hash received in the stack receive path and update the
 374  * flow_hash table accordingly.
 375  */
 376 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 377 {
 378         if (unlikely(e->rps_rxhash != hash)) {
 379                 sock_rps_reset_flow_hash(e->rps_rxhash);
 380                 e->rps_rxhash = hash;
 381         }
 382 }
 383
 384 /* We try to identify a flow through its rxhash first. The reason that
 385  * we do not check rxq no. is because some cards(e.g 82599), chooses
 386  * the rxq based on the txq where the last packet of the flow comes. As
 387  * the userspace application move between processors, we may get a
 388  * different rxq no. here. If we could not get rxhash, then we would
 389  * hope the rxq no. may help here.
 390  */
 391 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 392                             void *accel_priv, select_queue_fallback_t fallback)
 393 {
 394         struct tun_struct *tun = netdev_priv(dev);
 395         struct tun_flow_entry *e;
 396         u32 txq = 0;
 397         u32 numqueues = 0;
 398
 399         rcu_read_lock();
 400         numqueues = ACCESS_ONCE(tun->numqueues);
 401
 402         txq = skb_get_hash(skb);
 403         if (txq) {
 404                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 405                 if (e) {
 406                         tun_flow_save_rps_rxhash(e, txq);
 407                         txq = e->queue_index;
 408                 } else
 409                         /* use multiply and shift instead of expensive divide */
 410                         txq = ((u64)txq * numqueues) >> 32;
 411         } else if (likely(skb_rx_queue_recorded(skb))) {
 412                 txq = skb_get_rx_queue(skb);
 413                 while (unlikely(txq >= numqueues))
 414                         txq -= numqueues;
 415         }
 416
 417         rcu_read_unlock();
 418         return txq;
 419 }
 420
 421 static inline bool tun_not_capable(struct tun_struct *tun)
 422 {
 423         const struct cred *cred = current_cred();
 424         struct net *net = dev_net(tun->dev);
 425
 426         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 427                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 428                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 429 }
 430
 431 static void tun_set_real_num_queues(struct tun_struct *tun)
 432 {
 433         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 434         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 435 }
 436
 437 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 438 {
 439         tfile->detached = tun;
 440         list_add_tail(&tfile->next, &tun->disabled);
 441         ++tun->numdisabled;
 442 }
 443
 444 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 445 {
 446         struct tun_struct *tun = tfile->detached;
 447
 448         tfile->detached = NULL;
 449         list_del_init(&tfile->next);
 450         --tun->numdisabled;
 451         return tun;
 452 }
 453
 454 static void tun_queue_purge(struct tun_file *tfile)
 455 {
 456         skb_queue_purge(&tfile->sk.sk_receive_queue);
 457         skb_queue_purge(&tfile->sk.sk_error_queue);
 458 }
 459
 460 static void __tun_detach(struct tun_file *tfile, bool clean)
 461 {
 462         struct tun_file *ntfile;
 463         struct tun_struct *tun;
 464
 465         tun = rtnl_dereference(tfile->tun);
 466
 467         if (tun && !tfile->detached) {
 468                 u16 index = tfile->queue_index;
 469                 BUG_ON(index >= tun->numqueues);
 470
 471                 rcu_assign_pointer(tun->tfiles[index],
 472                                    tun->tfiles[tun->numqueues - 1]);
 473                 ntfile = rtnl_dereference(tun->tfiles[index]);
 474                 ntfile->queue_index = index;
 475
 476                 --tun->numqueues;
 477                 if (clean) {
 478                         RCU_INIT_POINTER(tfile->tun, NULL);
 479                         sock_put(&tfile->sk);
 480                 } else
 481                         tun_disable_queue(tun, tfile);
 482
 483                 synchronize_net();
 484                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 485                 /* Drop read queue */
 486                 tun_queue_purge(tfile);
 487                 tun_set_real_num_queues(tun);
 488         } else if (tfile->detached && clean) {
 489                 tun = tun_enable_queue(tfile);
 490                 sock_put(&tfile->sk);
 491         }
 492
 493         if (clean) {
 494                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 495                         netif_carrier_off(tun->dev);
 496
 497                         if (!(tun->flags & IFF_PERSIST) &&
 498                             tun->dev->reg_state == NETREG_REGISTERED)
 499                                 unregister_netdevice(tun->dev);
 500                 }
 501
 502                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 503                                  &tfile->socket.flags));
 504                 sk_release_kernel(&tfile->sk);
 505         }
 506 }
 507
 508 static void tun_detach(struct tun_file *tfile, bool clean)
 509 {
 510         rtnl_lock();
 511         __tun_detach(tfile, clean);
 512         rtnl_unlock();
 513 }
 514
 515 static void tun_detach_all(struct net_device *dev)
 516 {
 517         struct tun_struct *tun = netdev_priv(dev);
 518         struct tun_file *tfile, *tmp;
 519         int i, n = tun->numqueues;
 520
 521         for (i = 0; i < n; i++) {
 522                 tfile = rtnl_dereference(tun->tfiles[i]);
 523                 BUG_ON(!tfile);
 524                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 525                 RCU_INIT_POINTER(tfile->tun, NULL);
 526                 --tun->numqueues;
 527         }
 528         list_for_each_entry(tfile, &tun->disabled, next) {
 529                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 530                 RCU_INIT_POINTER(tfile->tun, NULL);
 531         }
 532         BUG_ON(tun->numqueues != 0);
 533
 534         synchronize_net();
 535         for (i = 0; i < n; i++) {
 536                 tfile = rtnl_dereference(tun->tfiles[i]);
 537                 /* Drop read queue */
 538                 tun_queue_purge(tfile);
 539                 sock_put(&tfile->sk);
 540         }
 541         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 542                 tun_enable_queue(tfile);
 543                 tun_queue_purge(tfile);
 544                 sock_put(&tfile->sk);
 545         }
 546         BUG_ON(tun->numdisabled != 0);
 547
 548         if (tun->flags & IFF_PERSIST)
 549                 module_put(THIS_MODULE);
 550 }
 551
 552 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 553 {
 554         struct tun_file *tfile = file->private_data;
 555         int err;
 556
 557         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 558         if (err < 0)
 559                 goto out;
 560
 561         err = -EINVAL;
 562         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 563                 goto out;
 564
 565         err = -EBUSY;
 566         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 567                 goto out;
 568
 569         err = -E2BIG;
 570         if (!tfile->detached &&
 571             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 572                 goto out;
 573
 574         err = 0;
 575
 576         /* Re-attach the filter to persist device */
 577         if (!skip_filter && (tun->filter_attached == true)) {
 578                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 579                 if (!err)
 580                         goto out;
 581         }
 582         tfile->queue_index = tun->numqueues;
 583         rcu_assign_pointer(tfile->tun, tun);
 584         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 585         tun->numqueues++;
 586
 587         if (tfile->detached)
 588                 tun_enable_queue(tfile);
 589         else
 590                 sock_hold(&tfile->sk);
 591
 592         tun_set_real_num_queues(tun);
 593
 594         /* device is allowed to go away first, so no need to hold extra
 595          * refcnt.
 596          */
 597
 598 out:
 599         return err;
 600 }
 601
 602 static struct tun_struct *__tun_get(struct tun_file *tfile)
 603 {
 604         struct tun_struct *tun;
 605
 606         rcu_read_lock();
 607         tun = rcu_dereference(tfile->tun);
 608         if (tun)
 609                 dev_hold(tun->dev);
 610         rcu_read_unlock();
 611
 612         return tun;
 613 }
 614
 615 static struct tun_struct *tun_get(struct file *file)
 616 {
 617         return __tun_get(file->private_data);
 618 }
 619
 620 static void tun_put(struct tun_struct *tun)
 621 {
 622         dev_put(tun->dev);
 623 }
 624
 625 /* TAP filtering */
 626 static void addr_hash_set(u32 *mask, const u8 *addr)
 627 {
 628         int n = ether_crc(ETH_ALEN, addr) >> 26;
 629         mask[n >> 5] |= (1 << (n & 31));
 630 }
 631
 632 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 633 {
 634         int n = ether_crc(ETH_ALEN, addr) >> 26;
 635         return mask[n >> 5] & (1 << (n & 31));
 636 }
 637
 638 static int update_filter(struct tap_filter *filter, void __user *arg)
 639 {
 640         struct { u8 u[ETH_ALEN]; } *addr;
 641         struct tun_filter uf;
 642         int err, alen, n, nexact;
 643
 644         if (copy_from_user(&uf, arg, sizeof(uf)))
 645                 return -EFAULT;
 646
 647         if (!uf.count) {
 648                 /* Disabled */
 649                 filter->count = 0;
 650                 return 0;
 651         }
 652
 653         alen = ETH_ALEN * uf.count;
 654         addr = kmalloc(alen, GFP_KERNEL);
 655         if (!addr)
 656                 return -ENOMEM;
 657
 658         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 659                 err = -EFAULT;
 660                 goto done;
 661         }
 662
 663         /* The filter is updated without holding any locks. Which is
 664          * perfectly safe. We disable it first and in the worst
 665          * case we'll accept a few undesired packets. */
 666         filter->count = 0;
 667         wmb();
 668
 669         /* Use first set of addresses as an exact filter */
 670         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 671                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 672
 673         nexact = n;
 674
 675         /* Remaining multicast addresses are hashed,
 676          * unicast will leave the filter disabled. */
 677         memset(filter->mask, 0, sizeof(filter->mask));
 678         for (; n < uf.count; n++) {
 679                 if (!is_multicast_ether_addr(addr[n].u)) {
 680                         err = 0; /* no filter */
 681                         goto done;
 682                 }
 683                 addr_hash_set(filter->mask, addr[n].u);
 684         }
 685
 686         /* For ALLMULTI just set the mask to all ones.
 687          * This overrides the mask populated above. */
 688         if ((uf.flags & TUN_FLT_ALLMULTI))
 689                 memset(filter->mask, ~0, sizeof(filter->mask));
 690
 691         /* Now enable the filter */
 692         wmb();
 693         filter->count = nexact;
 694
 695         /* Return the number of exact filters */
 696         err = nexact;
 697
 698 done:
 699         kfree(addr);
 700         return err;
 701 }
 702
 703 /* Returns: 0 - drop, !=0 - accept */
 704 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 705 {
 706         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 707          * at this point. */
 708         struct ethhdr *eh = (struct ethhdr *) skb->data;
 709         int i;
 710
 711         /* Exact match */
 712         for (i = 0; i < filter->count; i++)
 713                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 714                         return 1;
 715
 716         /* Inexact match (multicast only) */
 717         if (is_multicast_ether_addr(eh->h_dest))
 718                 return addr_hash_test(filter->mask, eh->h_dest);
 719
 720         return 0;
 721 }
 722
 723 /*
 724  * Checks whether the packet is accepted or not.
 725  * Returns: 0 - drop, !=0 - accept
 726  */
 727 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 728 {
 729         if (!filter->count)
 730                 return 1;
 731
 732         return run_filter(filter, skb);
 733 }
 734
 735 /* Network device part of the driver */
 736
 737 static const struct ethtool_ops tun_ethtool_ops;
 738
 739 /* Net device detach from fd. */
 740 static void tun_net_uninit(struct net_device *dev)
 741 {
 742         tun_detach_all(dev);
 743 }
 744
 745 /* Net device open. */
 746 static int tun_net_open(struct net_device *dev)
 747 {
 748         netif_tx_start_all_queues(dev);
 749         return 0;
 750 }
 751
 752 /* Net device close. */
 753 static int tun_net_close(struct net_device *dev)
 754 {
 755         netif_tx_stop_all_queues(dev);
 756         return 0;
 757 }
 758
 759 /* Net device start xmit */
 760 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 761 {
 762         struct tun_struct *tun = netdev_priv(dev);
 763         int txq = skb->queue_mapping;
 764         struct tun_file *tfile;
 765         u32 numqueues = 0;
 766
 767         rcu_read_lock();
 768         tfile = rcu_dereference(tun->tfiles[txq]);
 769         numqueues = ACCESS_ONCE(tun->numqueues);
 770
 771         /* Drop packet if interface is not attached */
 772         if (txq >= numqueues)
 773                 goto drop;
 774
 775         if (numqueues == 1) {
 776                 /* Select queue was not called for the skbuff, so we extract the
 777                  * RPS hash and save it into the flow_table here.
 778                  */
 779                 __u32 rxhash;
 780
 781                 rxhash = skb_get_hash(skb);
 782                 if (rxhash) {
 783                         struct tun_flow_entry *e;
 784                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 785                                         rxhash);
 786                         if (e)
 787                                 tun_flow_save_rps_rxhash(e, rxhash);
 788                 }
 789         }
 790
 791         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 792
 793         BUG_ON(!tfile);
 794
 795         /* Drop if the filter does not like it.
 796          * This is a noop if the filter is disabled.
 797          * Filter can be enabled only for the TAP devices. */
 798         if (!check_filter(&tun->txflt, skb))
 799                 goto drop;
 800
 801         if (tfile->socket.sk->sk_filter &&
 802             sk_filter(tfile->socket.sk, skb))
 803                 goto drop;
 804
 805         /* Limit the number of packets queued by dividing txq length with the
 806          * number of queues.
 807          */
 808         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
 809                           >= dev->tx_queue_len)
 810                 goto drop;
 811
 812         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 813                 goto drop;
 814
 815         if (skb->sk) {
 816                 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
 817                 sw_tx_timestamp(skb);
 818         }
 819
 820         /* Orphan the skb - required as we might hang on to it
 821          * for indefinite time.
 822          */
 823         skb_orphan(skb);
 824
 825         nf_reset(skb);
 826
 827         /* Enqueue packet */
 828         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 829
 830         /* Notify and wake up reader process */
 831         if (tfile->flags & TUN_FASYNC)
 832                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 833         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 834
 835         rcu_read_unlock();
 836         return NETDEV_TX_OK;
 837
 838 drop:
 839         dev->stats.tx_dropped++;
 840         skb_tx_error(skb);
 841         kfree_skb(skb);
 842         rcu_read_unlock();
 843         return NET_XMIT_DROP;
 844 }
 845
 846 static void tun_net_mclist(struct net_device *dev)
 847 {
 848         /*
 849          * This callback is supposed to deal with mc filter in
 850          * _rx_ path and has nothing to do with the _tx_ path.
 851          * In rx path we always accept everything userspace gives us.
 852          */
 853 }
 854
 855 #define MIN_MTU 68
 856 #define MAX_MTU 65535
 857
 858 static int
 859 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 860 {
 861         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 862                 return -EINVAL;
 863         dev->mtu = new_mtu;
 864         return 0;
 865 }
 866
 867 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 868         netdev_features_t features)
 869 {
 870         struct tun_struct *tun = netdev_priv(dev);
 871
 872         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 873 }
 874 #ifdef CONFIG_NET_POLL_CONTROLLER
 875 static void tun_poll_controller(struct net_device *dev)
 876 {
 877         /*
 878          * Tun only receives frames when:
 879          * 1) the char device endpoint gets data from user space
 880          * 2) the tun socket gets a sendmsg call from user space
 881          * Since both of those are synchronous operations, we are guaranteed
 882          * never to have pending data when we poll for it
 883          * so there is nothing to do here but return.
 884          * We need this though so netpoll recognizes us as an interface that
 885          * supports polling, which enables bridge devices in virt setups to
 886          * still use netconsole
 887          */
 888         return;
 889 }
 890 #endif
 891 static const struct net_device_ops tun_netdev_ops = {
 892         .ndo_uninit             = tun_net_uninit,
 893         .ndo_open               = tun_net_open,
 894         .ndo_stop               = tun_net_close,
 895         .ndo_start_xmit         = tun_net_xmit,
 896         .ndo_change_mtu         = tun_net_change_mtu,
 897         .ndo_fix_features       = tun_net_fix_features,
 898         .ndo_select_queue       = tun_select_queue,
 899 #ifdef CONFIG_NET_POLL_CONTROLLER
 900         .ndo_poll_controller    = tun_poll_controller,
 901 #endif
 902 };
 903
 904 static const struct net_device_ops tap_netdev_ops = {
 905         .ndo_uninit             = tun_net_uninit,
 906         .ndo_open               = tun_net_open,
 907         .ndo_stop               = tun_net_close,
 908         .ndo_start_xmit         = tun_net_xmit,
 909         .ndo_change_mtu         = tun_net_change_mtu,
 910         .ndo_fix_features       = tun_net_fix_features,
 911         .ndo_set_rx_mode        = tun_net_mclist,
 912         .ndo_set_mac_address    = eth_mac_addr,
 913         .ndo_validate_addr      = eth_validate_addr,
 914         .ndo_select_queue       = tun_select_queue,
 915 #ifdef CONFIG_NET_POLL_CONTROLLER
 916         .ndo_poll_controller    = tun_poll_controller,
 917 #endif
 918 };
 919
 920 static void tun_flow_init(struct tun_struct *tun)
 921 {
 922         int i;
 923
 924         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 925                 INIT_HLIST_HEAD(&tun->flows[i]);
 926
 927         tun->ageing_time = TUN_FLOW_EXPIRE;
 928         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 929         mod_timer(&tun->flow_gc_timer,
 930                   round_jiffies_up(jiffies + tun->ageing_time));
 931 }
 932
 933 static void tun_flow_uninit(struct tun_struct *tun)
 934 {
 935         del_timer_sync(&tun->flow_gc_timer);
 936         tun_flow_flush(tun);
 937 }
 938
 939 /* Initialize net device. */
 940 static void tun_net_init(struct net_device *dev)
 941 {
 942         struct tun_struct *tun = netdev_priv(dev);
 943
 944         switch (tun->flags & TUN_TYPE_MASK) {
 945         case IFF_TUN:
 946                 dev->netdev_ops = &tun_netdev_ops;
 947
 948                 /* Point-to-Point TUN Device */
 949                 dev->hard_header_len = 0;
 950                 dev->addr_len = 0;
 951                 dev->mtu = 1500;
 952
 953                 /* Zero header length */
 954                 dev->type = ARPHRD_NONE;
 955                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 956                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 957                 break;
 958
 959         case IFF_TAP:
 960                 dev->netdev_ops = &tap_netdev_ops;
 961                 /* Ethernet TAP Device */
 962                 ether_setup(dev);
 963                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 964                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 965
 966                 eth_hw_addr_random(dev);
 967
 968                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 969                 break;
 970         }
 971 }
 972
 973 /* Character device part */
 974
 975 /* Poll */
 976 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 977 {
 978         struct tun_file *tfile = file->private_data;
 979         struct tun_struct *tun = __tun_get(tfile);
 980         struct sock *sk;
 981         unsigned int mask = 0;
 982
 983         if (!tun)
 984                 return POLLERR;
 985
 986         sk = tfile->socket.sk;
 987
 988         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 989
 990         poll_wait(file, sk_sleep(sk), wait);
 991
 992         if (!skb_queue_empty(&sk->sk_receive_queue))
 993                 mask |= POLLIN | POLLRDNORM;
 994
 995         if (sock_writeable(sk) ||
 996             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 997              sock_writeable(sk)))
 998                 mask |= POLLOUT | POLLWRNORM;
 999
1000         if (tun->dev->reg_state != NETREG_REGISTERED)
1001                 mask = POLLERR;
1002
1003         tun_put(tun);
1004         return mask;
1005 }
1006
1007 /* prepad is the amount to reserve at front.  len is length after that.
1008  * linear is a hint as to how much to copy (usually headers). */
1009 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1010                                      size_t prepad, size_t len,
1011                                      size_t linear, int noblock)
1012 {
1013         struct sock *sk = tfile->socket.sk;
1014         struct sk_buff *skb;
1015         int err;
1016
1017         /* Under a page?  Don't bother with paged skb. */
1018         if (prepad + len < PAGE_SIZE || !linear)
1019                 linear = len;
1020
1021         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1022                                    &err, 0);
1023         if (!skb)
1024                 return ERR_PTR(err);
1025
1026         skb_reserve(skb, prepad);
1027         skb_put(skb, linear);
1028         skb->data_len = len - linear;
1029         skb->len += len - linear;
1030
1031         return skb;
1032 }
1033
1034 /* Get packet from user space buffer */
1035 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1036                             void *msg_control, struct iov_iter *from,
1037                             int noblock)
1038 {
1039         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1040         struct sk_buff *skb;
1041         size_t total_len = iov_iter_count(from);
1042         size_t len = total_len, align = NET_SKB_PAD, linear;
1043         struct virtio_net_hdr gso = { 0 };
1044         int good_linear;
1045         int copylen;
1046         bool zerocopy = false;
1047         int err;
1048         u32 rxhash;
1049         ssize_t n;
1050
1051         if (!(tun->flags & IFF_NO_PI)) {
1052                 if (len < sizeof(pi))
1053                         return -EINVAL;
1054                 len -= sizeof(pi);
1055
1056                 n = copy_from_iter(&pi, sizeof(pi), from);
1057                 if (n != sizeof(pi))
1058                         return -EFAULT;
1059         }
1060
1061         if (tun->flags & IFF_VNET_HDR) {
1062                 if (len < tun->vnet_hdr_sz)
1063                         return -EINVAL;
1064                 len -= tun->vnet_hdr_sz;
1065
1066                 n = copy_from_iter(&gso, sizeof(gso), from);
1067                 if (n != sizeof(gso))
1068                         return -EFAULT;
1069
1070                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1071                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1072                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1073
1074                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1075                         return -EINVAL;
1076                 iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1077         }
1078
1079         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1080                 align += NET_IP_ALIGN;
1081                 if (unlikely(len < ETH_HLEN ||
1082                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1083                         return -EINVAL;
1084         }
1085
1086         good_linear = SKB_MAX_HEAD(align);
1087
1088         if (msg_control) {
1089                 struct iov_iter i = *from;
1090
1091                 /* There are 256 bytes to be copied in skb, so there is
1092                  * enough room for skb expand head in case it is used.
1093                  * The rest of the buffer is mapped from userspace.
1094                  */
1095                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1096                 if (copylen > good_linear)
1097                         copylen = good_linear;
1098                 linear = copylen;
1099                 iov_iter_advance(&i, copylen);
1100                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1101                         zerocopy = true;
1102         }
1103
1104         if (!zerocopy) {
1105                 copylen = len;
1106                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1107                         linear = good_linear;
1108                 else
1109                         linear = tun16_to_cpu(tun, gso.hdr_len);
1110         }
1111
1112         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1113         if (IS_ERR(skb)) {
1114                 if (PTR_ERR(skb) != -EAGAIN)
1115                         tun->dev->stats.rx_dropped++;
1116                 return PTR_ERR(skb);
1117         }
1118
1119         if (zerocopy)
1120                 err = zerocopy_sg_from_iter(skb, from);
1121         else {
1122                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1123                 if (!err && msg_control) {
1124                         struct ubuf_info *uarg = msg_control;
1125                         uarg->callback(uarg, false);
1126                 }
1127         }
1128
1129         if (err) {
1130                 tun->dev->stats.rx_dropped++;
1131                 kfree_skb(skb);
1132                 return -EFAULT;
1133         }
1134
1135         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1136                 if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start),
1137                                           tun16_to_cpu(tun, gso.csum_offset))) {
1138                         tun->dev->stats.rx_frame_errors++;
1139                         kfree_skb(skb);
1140                         return -EINVAL;
1141                 }
1142         }
1143
1144         switch (tun->flags & TUN_TYPE_MASK) {
1145         case IFF_TUN:
1146                 if (tun->flags & IFF_NO_PI) {
1147                         switch (skb->data[0] & 0xf0) {
1148                         case 0x40:
1149                                 pi.proto = htons(ETH_P_IP);
1150                                 break;
1151                         case 0x60:
1152                                 pi.proto = htons(ETH_P_IPV6);
1153                                 break;
1154                         default:
1155                                 tun->dev->stats.rx_dropped++;
1156                                 kfree_skb(skb);
1157                                 return -EINVAL;
1158                         }
1159                 }
1160
1161                 skb_reset_mac_header(skb);
1162                 skb->protocol = pi.proto;
1163                 skb->dev = tun->dev;
1164                 break;
1165         case IFF_TAP:
1166                 skb->protocol = eth_type_trans(skb, tun->dev);
1167                 break;
1168         }
1169
1170         skb_reset_network_header(skb);
1171
1172         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1173                 pr_debug("GSO!\n");
1174                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1175                 case VIRTIO_NET_HDR_GSO_TCPV4:
1176                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1177                         break;
1178                 case VIRTIO_NET_HDR_GSO_TCPV6:
1179                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1180                         break;
1181                 case VIRTIO_NET_HDR_GSO_UDP:
1182                 {
1183                         static bool warned;
1184
1185                         if (!warned) {
1186                                 warned = true;
1187                                 netdev_warn(tun->dev,
1188                                             "%s: using disabled UFO feature; please fix this program\n",
1189                                             current->comm);
1190                         }
1191                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1192                         if (skb->protocol == htons(ETH_P_IPV6))
1193                                 ipv6_proxy_select_ident(skb);
1194                         break;
1195                 }
1196                 default:
1197                         tun->dev->stats.rx_frame_errors++;
1198                         kfree_skb(skb);
1199                         return -EINVAL;
1200                 }
1201
1202                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1203                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1204
1205                 skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size);
1206                 if (skb_shinfo(skb)->gso_size == 0) {
1207                         tun->dev->stats.rx_frame_errors++;
1208                         kfree_skb(skb);
1209                         return -EINVAL;
1210                 }
1211
1212                 /* Header must be checked, and gso_segs computed. */
1213                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1214                 skb_shinfo(skb)->gso_segs = 0;
1215         }
1216
1217         /* copy skb_ubuf_info for callback when skb has no error */
1218         if (zerocopy) {
1219                 skb_shinfo(skb)->destructor_arg = msg_control;
1220                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1221                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1222         }
1223
1224         skb_probe_transport_header(skb, 0);
1225
1226         rxhash = skb_get_hash(skb);
1227         netif_rx_ni(skb);
1228
1229         tun->dev->stats.rx_packets++;
1230         tun->dev->stats.rx_bytes += len;
1231
1232         tun_flow_update(tun, rxhash, tfile);
1233         return total_len;
1234 }
1235
1236 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1237 {
1238         struct file *file = iocb->ki_filp;
1239         struct tun_struct *tun = tun_get(file);
1240         struct tun_file *tfile = file->private_data;
1241         ssize_t result;
1242
1243         if (!tun)
1244                 return -EBADFD;
1245
1246         result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1247
1248         tun_put(tun);
1249         return result;
1250 }
1251
1252 /* Put packet to the user space buffer */
1253 static ssize_t tun_put_user(struct tun_struct *tun,
1254                             struct tun_file *tfile,
1255                             struct sk_buff *skb,
1256                             struct iov_iter *iter)
1257 {
1258         struct tun_pi pi = { 0, skb->protocol };
1259         ssize_t total;
1260         int vlan_offset = 0;
1261         int vlan_hlen = 0;
1262         int vnet_hdr_sz = 0;
1263
1264         if (vlan_tx_tag_present(skb))
1265                 vlan_hlen = VLAN_HLEN;
1266
1267         if (tun->flags & IFF_VNET_HDR)
1268                 vnet_hdr_sz = tun->vnet_hdr_sz;
1269
1270         total = skb->len + vlan_hlen + vnet_hdr_sz;
1271
1272         if (!(tun->flags & IFF_NO_PI)) {
1273                 if (iov_iter_count(iter) < sizeof(pi))
1274                         return -EINVAL;
1275
1276                 total += sizeof(pi);
1277                 if (iov_iter_count(iter) < total) {
1278                         /* Packet will be striped */
1279                         pi.flags |= TUN_PKT_STRIP;
1280                 }
1281
1282                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1283                         return -EFAULT;
1284         }
1285
1286         if (vnet_hdr_sz) {
1287                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1288                 if (iov_iter_count(iter) < vnet_hdr_sz)
1289                         return -EINVAL;
1290
1291                 if (skb_is_gso(skb)) {
1292                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1293
1294                         /* This is a hint as to how much should be linear. */
1295                         gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb));
1296                         gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size);
1297                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1298                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1299                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1300                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1301                         else {
1302                                 pr_err("unexpected GSO type: "
1303                                        "0x%x, gso_size %d, hdr_len %d\n",
1304                                        sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1305                                        tun16_to_cpu(tun, gso.hdr_len));
1306                                 print_hex_dump(KERN_ERR, "tun: ",
1307                                                DUMP_PREFIX_NONE,
1308                                                16, 1, skb->head,
1309                                                min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1310                                 WARN_ON_ONCE(1);
1311                                 return -EINVAL;
1312                         }
1313                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1314                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1315                 } else
1316                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1317
1318                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1319                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1320                         gso.csum_start = cpu_to_tun16(tun, skb_checksum_start_offset(skb) +
1321                                                       vlan_hlen);
1322                         gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset);
1323                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1324                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1325                 } /* else everything is zero */
1326
1327                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1328                         return -EFAULT;
1329
1330                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1331         }
1332
1333         if (vlan_hlen) {
1334                 int ret;
1335                 struct {
1336                         __be16 h_vlan_proto;
1337                         __be16 h_vlan_TCI;
1338                 } veth;
1339
1340                 veth.h_vlan_proto = skb->vlan_proto;
1341                 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1342
1343                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1344
1345                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1346                 if (ret || !iov_iter_count(iter))
1347                         goto done;
1348
1349                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1350                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1351                         goto done;
1352         }
1353
1354         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1355
1356 done:
1357         tun->dev->stats.tx_packets++;
1358         tun->dev->stats.tx_bytes += skb->len + vlan_hlen;
1359
1360         return total;
1361 }
1362
1363 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1364                            struct iov_iter *to,
1365                            int noblock)
1366 {
1367         struct sk_buff *skb;
1368         ssize_t ret;
1369         int peeked, err, off = 0;
1370
1371         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1372
1373         if (!iov_iter_count(to))
1374                 return 0;
1375
1376         if (tun->dev->reg_state != NETREG_REGISTERED)
1377                 return -EIO;
1378
1379         /* Read frames from queue */
1380         skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1381                                   &peeked, &off, &err);
1382         if (!skb)
1383                 return 0;
1384
1385         ret = tun_put_user(tun, tfile, skb, to);
1386         if (unlikely(ret < 0))
1387                 kfree_skb(skb);
1388         else
1389                 consume_skb(skb);
1390
1391         return ret;
1392 }
1393
1394 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1395 {
1396         struct file *file = iocb->ki_filp;
1397         struct tun_file *tfile = file->private_data;
1398         struct tun_struct *tun = __tun_get(tfile);
1399         ssize_t len = iov_iter_count(to), ret;
1400
1401         if (!tun)
1402                 return -EBADFD;
1403         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1404         ret = min_t(ssize_t, ret, len);
1405         if (ret > 0)
1406                 iocb->ki_pos = ret;
1407         tun_put(tun);
1408         return ret;
1409 }
1410
1411 static void tun_free_netdev(struct net_device *dev)
1412 {
1413         struct tun_struct *tun = netdev_priv(dev);
1414
1415         BUG_ON(!(list_empty(&tun->disabled)));
1416         tun_flow_uninit(tun);
1417         security_tun_dev_free_security(tun->security);
1418         free_netdev(dev);
1419 }
1420
1421 static void tun_setup(struct net_device *dev)
1422 {
1423         struct tun_struct *tun = netdev_priv(dev);
1424
1425         tun->owner = INVALID_UID;
1426         tun->group = INVALID_GID;
1427
1428         dev->ethtool_ops = &tun_ethtool_ops;
1429         dev->destructor = tun_free_netdev;
1430 }
1431
1432 /* Trivial set of netlink ops to allow deleting tun or tap
1433  * device with netlink.
1434  */
1435 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1436 {
1437         return -EINVAL;
1438 }
1439
1440 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1441         .kind           = DRV_NAME,
1442         .priv_size      = sizeof(struct tun_struct),
1443         .setup          = tun_setup,
1444         .validate       = tun_validate,
1445 };
1446
1447 static void tun_sock_write_space(struct sock *sk)
1448 {
1449         struct tun_file *tfile;
1450         wait_queue_head_t *wqueue;
1451
1452         if (!sock_writeable(sk))
1453                 return;
1454
1455         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1456                 return;
1457
1458         wqueue = sk_sleep(sk);
1459         if (wqueue && waitqueue_active(wqueue))
1460                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1461                                                 POLLWRNORM | POLLWRBAND);
1462
1463         tfile = container_of(sk, struct tun_file, sk);
1464         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1465 }
1466
1467 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1468                        struct msghdr *m, size_t total_len)
1469 {
1470         int ret;
1471         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1472         struct tun_struct *tun = __tun_get(tfile);
1473
1474         if (!tun)
1475                 return -EBADFD;
1476
1477         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1478                            m->msg_flags & MSG_DONTWAIT);
1479         tun_put(tun);
1480         return ret;
1481 }
1482
1483 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1484                        struct msghdr *m, size_t total_len,
1485                        int flags)
1486 {
1487         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1488         struct tun_struct *tun = __tun_get(tfile);
1489         int ret;
1490
1491         if (!tun)
1492                 return -EBADFD;
1493
1494         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1495                 ret = -EINVAL;
1496                 goto out;
1497         }
1498         if (flags & MSG_ERRQUEUE) {
1499                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1500                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1501                 goto out;
1502         }
1503         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1504         if (ret > total_len) {
1505                 m->msg_flags |= MSG_TRUNC;
1506                 ret = flags & MSG_TRUNC ? ret : total_len;
1507         }
1508 out:
1509         tun_put(tun);
1510         return ret;
1511 }
1512
1513 static int tun_release(struct socket *sock)
1514 {
1515         if (sock->sk)
1516                 sock_put(sock->sk);
1517         return 0;
1518 }
1519
1520 /* Ops structure to mimic raw sockets with tun */
1521 static const struct proto_ops tun_socket_ops = {
1522         .sendmsg = tun_sendmsg,
1523         .recvmsg = tun_recvmsg,
1524         .release = tun_release,
1525 };
1526
1527 static struct proto tun_proto = {
1528         .name           = "tun",
1529         .owner          = THIS_MODULE,
1530         .obj_size       = sizeof(struct tun_file),
1531 };
1532
1533 static int tun_flags(struct tun_struct *tun)
1534 {
1535         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1536 }
1537
1538 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1539                               char *buf)
1540 {
1541         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1542         return sprintf(buf, "0x%x\n", tun_flags(tun));
1543 }
1544
1545 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1546                               char *buf)
1547 {
1548         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1549         return uid_valid(tun->owner)?
1550                 sprintf(buf, "%u\n",
1551                         from_kuid_munged(current_user_ns(), tun->owner)):
1552                 sprintf(buf, "-1\n");
1553 }
1554
1555 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1556                               char *buf)
1557 {
1558         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1559         return gid_valid(tun->group) ?
1560                 sprintf(buf, "%u\n",
1561                         from_kgid_munged(current_user_ns(), tun->group)):
1562                 sprintf(buf, "-1\n");
1563 }
1564
1565 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1566 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1567 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1568
1569 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1570 {
1571         struct tun_struct *tun;
1572         struct tun_file *tfile = file->private_data;
1573         struct net_device *dev;
1574         int err;
1575
1576         if (tfile->detached)
1577                 return -EINVAL;
1578
1579         dev = __dev_get_by_name(net, ifr->ifr_name);
1580         if (dev) {
1581                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1582                         return -EBUSY;
1583                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1584                         tun = netdev_priv(dev);
1585                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1586                         tun = netdev_priv(dev);
1587                 else
1588                         return -EINVAL;
1589
1590                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1591                     !!(tun->flags & IFF_MULTI_QUEUE))
1592                         return -EINVAL;
1593
1594                 if (tun_not_capable(tun))
1595                         return -EPERM;
1596                 err = security_tun_dev_open(tun->security);
1597                 if (err < 0)
1598                         return err;
1599
1600                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1601                 if (err < 0)
1602                         return err;
1603
1604                 if (tun->flags & IFF_MULTI_QUEUE &&
1605                     (tun->numqueues + tun->numdisabled > 1)) {
1606                         /* One or more queue has already been attached, no need
1607                          * to initialize the device again.
1608                          */
1609                         return 0;
1610                 }
1611         }
1612         else {
1613                 char *name;
1614                 unsigned long flags = 0;
1615                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1616                              MAX_TAP_QUEUES : 1;
1617
1618                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1619                         return -EPERM;
1620                 err = security_tun_dev_create();
1621                 if (err < 0)
1622                         return err;
1623
1624                 /* Set dev type */
1625                 if (ifr->ifr_flags & IFF_TUN) {
1626                         /* TUN device */
1627                         flags |= IFF_TUN;
1628                         name = "tun%d";
1629                 } else if (ifr->ifr_flags & IFF_TAP) {
1630                         /* TAP device */
1631                         flags |= IFF_TAP;
1632                         name = "tap%d";
1633                 } else
1634                         return -EINVAL;
1635
1636                 if (*ifr->ifr_name)
1637                         name = ifr->ifr_name;
1638
1639                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1640                                        NET_NAME_UNKNOWN, tun_setup, queues,
1641                                        queues);
1642
1643                 if (!dev)
1644                         return -ENOMEM;
1645
1646                 dev_net_set(dev, net);
1647                 dev->rtnl_link_ops = &tun_link_ops;
1648                 dev->ifindex = tfile->ifindex;
1649
1650                 tun = netdev_priv(dev);
1651                 tun->dev = dev;
1652                 tun->flags = flags;
1653                 tun->txflt.count = 0;
1654                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1655
1656                 tun->filter_attached = false;
1657                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1658
1659                 spin_lock_init(&tun->lock);
1660
1661                 err = security_tun_dev_alloc_security(&tun->security);
1662                 if (err < 0)
1663                         goto err_free_dev;
1664
1665                 tun_net_init(dev);
1666                 tun_flow_init(tun);
1667
1668                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1669                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1670                                    NETIF_F_HW_VLAN_STAG_TX;
1671                 dev->features = dev->hw_features;
1672                 dev->vlan_features = dev->features &
1673                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1674                                        NETIF_F_HW_VLAN_STAG_TX);
1675
1676                 INIT_LIST_HEAD(&tun->disabled);
1677                 err = tun_attach(tun, file, false);
1678                 if (err < 0)
1679                         goto err_free_flow;
1680
1681                 err = register_netdevice(tun->dev);
1682                 if (err < 0)
1683                         goto err_detach;
1684
1685                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1686                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1687                     device_create_file(&tun->dev->dev, &dev_attr_group))
1688                         pr_err("Failed to create tun sysfs files\n");
1689         }
1690
1691         netif_carrier_on(tun->dev);
1692
1693         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1694
1695         tun->flags = (tun->flags & ~TUN_FEATURES) |
1696                 (ifr->ifr_flags & TUN_FEATURES);
1697
1698         /* Make sure persistent devices do not get stuck in
1699          * xoff state.
1700          */
1701         if (netif_running(tun->dev))
1702                 netif_tx_wake_all_queues(tun->dev);
1703
1704         strcpy(ifr->ifr_name, tun->dev->name);
1705         return 0;
1706
1707 err_detach:
1708         tun_detach_all(dev);
1709 err_free_flow:
1710         tun_flow_uninit(tun);
1711         security_tun_dev_free_security(tun->security);
1712 err_free_dev:
1713         free_netdev(dev);
1714         return err;
1715 }
1716
1717 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1718                        struct ifreq *ifr)
1719 {
1720         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1721
1722         strcpy(ifr->ifr_name, tun->dev->name);
1723
1724         ifr->ifr_flags = tun_flags(tun);
1725
1726 }
1727
1728 /* This is like a cut-down ethtool ops, except done via tun fd so no
1729  * privs required. */
1730 static int set_offload(struct tun_struct *tun, unsigned long arg)
1731 {
1732         netdev_features_t features = 0;
1733
1734         if (arg & TUN_F_CSUM) {
1735                 features |= NETIF_F_HW_CSUM;
1736                 arg &= ~TUN_F_CSUM;
1737
1738                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1739                         if (arg & TUN_F_TSO_ECN) {
1740                                 features |= NETIF_F_TSO_ECN;
1741                                 arg &= ~TUN_F_TSO_ECN;
1742                         }
1743                         if (arg & TUN_F_TSO4)
1744                                 features |= NETIF_F_TSO;
1745                         if (arg & TUN_F_TSO6)
1746                                 features |= NETIF_F_TSO6;
1747                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1748                 }
1749         }
1750
1751         /* This gives the user a way to test for new features in future by
1752          * trying to set them. */
1753         if (arg)
1754                 return -EINVAL;
1755
1756         tun->set_features = features;
1757         netdev_update_features(tun->dev);
1758
1759         return 0;
1760 }
1761
1762 static void tun_detach_filter(struct tun_struct *tun, int n)
1763 {
1764         int i;
1765         struct tun_file *tfile;
1766
1767         for (i = 0; i < n; i++) {
1768                 tfile = rtnl_dereference(tun->tfiles[i]);
1769                 sk_detach_filter(tfile->socket.sk);
1770         }
1771
1772         tun->filter_attached = false;
1773 }
1774
1775 static int tun_attach_filter(struct tun_struct *tun)
1776 {
1777         int i, ret = 0;
1778         struct tun_file *tfile;
1779
1780         for (i = 0; i < tun->numqueues; i++) {
1781                 tfile = rtnl_dereference(tun->tfiles[i]);
1782                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1783                 if (ret) {
1784                         tun_detach_filter(tun, i);
1785                         return ret;
1786                 }
1787         }
1788
1789         tun->filter_attached = true;
1790         return ret;
1791 }
1792
1793 static void tun_set_sndbuf(struct tun_struct *tun)
1794 {
1795         struct tun_file *tfile;
1796         int i;
1797
1798         for (i = 0; i < tun->numqueues; i++) {
1799                 tfile = rtnl_dereference(tun->tfiles[i]);
1800                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1801         }
1802 }
1803
1804 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1805 {
1806         struct tun_file *tfile = file->private_data;
1807         struct tun_struct *tun;
1808         int ret = 0;
1809
1810         rtnl_lock();
1811
1812         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1813                 tun = tfile->detached;
1814                 if (!tun) {
1815                         ret = -EINVAL;
1816                         goto unlock;
1817                 }
1818                 ret = security_tun_dev_attach_queue(tun->security);
1819                 if (ret < 0)
1820                         goto unlock;
1821                 ret = tun_attach(tun, file, false);
1822         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1823                 tun = rtnl_dereference(tfile->tun);
1824                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1825                         ret = -EINVAL;
1826                 else
1827                         __tun_detach(tfile, false);
1828         } else
1829                 ret = -EINVAL;
1830
1831 unlock:
1832         rtnl_unlock();
1833         return ret;
1834 }
1835
1836 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1837                             unsigned long arg, int ifreq_len)
1838 {
1839         struct tun_file *tfile = file->private_data;
1840         struct tun_struct *tun;
1841         void __user* argp = (void __user*)arg;
1842         struct ifreq ifr;
1843         kuid_t owner;
1844         kgid_t group;
1845         int sndbuf;
1846         int vnet_hdr_sz;
1847         unsigned int ifindex;
1848         int le;
1849         int ret;
1850
1851         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1852                 if (copy_from_user(&ifr, argp, ifreq_len))
1853                         return -EFAULT;
1854         } else {
1855                 memset(&ifr, 0, sizeof(ifr));
1856         }
1857         if (cmd == TUNGETFEATURES) {
1858                 /* Currently this just means: "what IFF flags are valid?".
1859                  * This is needed because we never checked for invalid flags on
1860                  * TUNSETIFF.
1861                  */
1862                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1863                                 (unsigned int __user*)argp);
1864         } else if (cmd == TUNSETQUEUE)
1865                 return tun_set_queue(file, &ifr);
1866
1867         ret = 0;
1868         rtnl_lock();
1869
1870         tun = __tun_get(tfile);
1871         if (cmd == TUNSETIFF && !tun) {
1872                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1873
1874                 ret = tun_set_iff(tfile->net, file, &ifr);
1875
1876                 if (ret)
1877                         goto unlock;
1878
1879                 if (copy_to_user(argp, &ifr, ifreq_len))
1880                         ret = -EFAULT;
1881                 goto unlock;
1882         }
1883         if (cmd == TUNSETIFINDEX) {
1884                 ret = -EPERM;
1885                 if (tun)
1886                         goto unlock;
1887
1888                 ret = -EFAULT;
1889                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1890                         goto unlock;
1891
1892                 ret = 0;
1893                 tfile->ifindex = ifindex;
1894                 goto unlock;
1895         }
1896
1897         ret = -EBADFD;
1898         if (!tun)
1899                 goto unlock;
1900
1901         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1902
1903         ret = 0;
1904         switch (cmd) {
1905         case TUNGETIFF:
1906                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1907
1908                 if (tfile->detached)
1909                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
1910                 if (!tfile->socket.sk->sk_filter)
1911                         ifr.ifr_flags |= IFF_NOFILTER;
1912
1913                 if (copy_to_user(argp, &ifr, ifreq_len))
1914                         ret = -EFAULT;
1915                 break;
1916
1917         case TUNSETNOCSUM:
1918                 /* Disable/Enable checksum */
1919
1920                 /* [unimplemented] */
1921                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1922                           arg ? "disabled" : "enabled");
1923                 break;
1924
1925         case TUNSETPERSIST:
1926                 /* Disable/Enable persist mode. Keep an extra reference to the
1927                  * module to prevent the module being unprobed.
1928                  */
1929                 if (arg && !(tun->flags & IFF_PERSIST)) {
1930                         tun->flags |= IFF_PERSIST;
1931                         __module_get(THIS_MODULE);
1932                 }
1933                 if (!arg && (tun->flags & IFF_PERSIST)) {
1934                         tun->flags &= ~IFF_PERSIST;
1935                         module_put(THIS_MODULE);
1936                 }
1937
1938                 tun_debug(KERN_INFO, tun, "persist %s\n",
1939                           arg ? "enabled" : "disabled");
1940                 break;
1941
1942         case TUNSETOWNER:
1943                 /* Set owner of the device */
1944                 owner = make_kuid(current_user_ns(), arg);
1945                 if (!uid_valid(owner)) {
1946                         ret = -EINVAL;
1947                         break;
1948                 }
1949                 tun->owner = owner;
1950                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1951                           from_kuid(&init_user_ns, tun->owner));
1952                 break;
1953
1954         case TUNSETGROUP:
1955                 /* Set group of the device */
1956                 group = make_kgid(current_user_ns(), arg);
1957                 if (!gid_valid(group)) {
1958                         ret = -EINVAL;
1959                         break;
1960                 }
1961                 tun->group = group;
1962                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1963                           from_kgid(&init_user_ns, tun->group));
1964                 break;
1965
1966         case TUNSETLINK:
1967                 /* Only allow setting the type when the interface is down */
1968                 if (tun->dev->flags & IFF_UP) {
1969                         tun_debug(KERN_INFO, tun,
1970                                   "Linktype set failed because interface is up\n");
1971                         ret = -EBUSY;
1972                 } else {
1973                         tun->dev->type = (int) arg;
1974                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1975                                   tun->dev->type);
1976                         ret = 0;
1977                 }
1978                 break;
1979
1980 #ifdef TUN_DEBUG
1981         case TUNSETDEBUG:
1982                 tun->debug = arg;
1983                 break;
1984 #endif
1985         case TUNSETOFFLOAD:
1986                 ret = set_offload(tun, arg);
1987                 break;
1988
1989         case TUNSETTXFILTER:
1990                 /* Can be set only for TAPs */
1991                 ret = -EINVAL;
1992                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1993                         break;
1994                 ret = update_filter(&tun->txflt, (void __user *)arg);
1995                 break;
1996
1997         case SIOCGIFHWADDR:
1998                 /* Get hw address */
1999                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2000                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2001                 if (copy_to_user(argp, &ifr, ifreq_len))
2002                         ret = -EFAULT;
2003                 break;
2004
2005         case SIOCSIFHWADDR:
2006                 /* Set hw address */
2007                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2008                           ifr.ifr_hwaddr.sa_data);
2009
2010                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2011                 break;
2012
2013         case TUNGETSNDBUF:
2014                 sndbuf = tfile->socket.sk->sk_sndbuf;
2015                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2016                         ret = -EFAULT;
2017                 break;
2018
2019         case TUNSETSNDBUF:
2020                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2021                         ret = -EFAULT;
2022                         break;
2023                 }
2024
2025                 tun->sndbuf = sndbuf;
2026                 tun_set_sndbuf(tun);
2027                 break;
2028
2029         case TUNGETVNETHDRSZ:
2030                 vnet_hdr_sz = tun->vnet_hdr_sz;
2031                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2032                         ret = -EFAULT;
2033                 break;
2034
2035         case TUNSETVNETHDRSZ:
2036                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2037                         ret = -EFAULT;
2038                         break;
2039                 }
2040                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2041                         ret = -EINVAL;
2042                         break;
2043                 }
2044
2045                 tun->vnet_hdr_sz = vnet_hdr_sz;
2046                 break;
2047
2048         case TUNGETVNETLE:
2049                 le = !!(tun->flags & TUN_VNET_LE);
2050                 if (put_user(le, (int __user *)argp))
2051                         ret = -EFAULT;
2052                 break;
2053
2054         case TUNSETVNETLE:
2055                 if (get_user(le, (int __user *)argp)) {
2056                         ret = -EFAULT;
2057                         break;
2058                 }
2059                 if (le)
2060                         tun->flags |= TUN_VNET_LE;
2061                 else
2062                         tun->flags &= ~TUN_VNET_LE;
2063                 break;
2064
2065         case TUNATTACHFILTER:
2066                 /* Can be set only for TAPs */
2067                 ret = -EINVAL;
2068                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2069                         break;
2070                 ret = -EFAULT;
2071                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2072                         break;
2073
2074                 ret = tun_attach_filter(tun);
2075                 break;
2076
2077         case TUNDETACHFILTER:
2078                 /* Can be set only for TAPs */
2079                 ret = -EINVAL;
2080                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2081                         break;
2082                 ret = 0;
2083                 tun_detach_filter(tun, tun->numqueues);
2084                 break;
2085
2086         case TUNGETFILTER:
2087                 ret = -EINVAL;
2088                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2089                         break;
2090                 ret = -EFAULT;
2091                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2092                         break;
2093                 ret = 0;
2094                 break;
2095
2096         default:
2097                 ret = -EINVAL;
2098                 break;
2099         }
2100
2101 unlock:
2102         rtnl_unlock();
2103         if (tun)
2104                 tun_put(tun);
2105         return ret;
2106 }
2107
2108 static long tun_chr_ioctl(struct file *file,
2109                           unsigned int cmd, unsigned long arg)
2110 {
2111         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2112 }
2113
2114 #ifdef CONFIG_COMPAT
2115 static long tun_chr_compat_ioctl(struct file *file,
2116                          unsigned int cmd, unsigned long arg)
2117 {
2118         switch (cmd) {
2119         case TUNSETIFF:
2120         case TUNGETIFF:
2121         case TUNSETTXFILTER:
2122         case TUNGETSNDBUF:
2123         case TUNSETSNDBUF:
2124         case SIOCGIFHWADDR:
2125         case SIOCSIFHWADDR:
2126                 arg = (unsigned long)compat_ptr(arg);
2127                 break;
2128         default:
2129                 arg = (compat_ulong_t)arg;
2130                 break;
2131         }
2132
2133         /*
2134          * compat_ifreq is shorter than ifreq, so we must not access beyond
2135          * the end of that structure. All fields that are used in this
2136          * driver are compatible though, we don't need to convert the
2137          * contents.
2138          */
2139         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2140 }
2141 #endif /* CONFIG_COMPAT */
2142
2143 static int tun_chr_fasync(int fd, struct file *file, int on)
2144 {
2145         struct tun_file *tfile = file->private_data;
2146         int ret;
2147
2148         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2149                 goto out;
2150
2151         if (on) {
2152                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2153                 tfile->flags |= TUN_FASYNC;
2154         } else
2155                 tfile->flags &= ~TUN_FASYNC;
2156         ret = 0;
2157 out:
2158         return ret;
2159 }
2160
2161 static int tun_chr_open(struct inode *inode, struct file * file)
2162 {
2163         struct tun_file *tfile;
2164
2165         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2166
2167         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2168                                             &tun_proto);
2169         if (!tfile)
2170                 return -ENOMEM;
2171         RCU_INIT_POINTER(tfile->tun, NULL);
2172         tfile->net = get_net(current->nsproxy->net_ns);
2173         tfile->flags = 0;
2174         tfile->ifindex = 0;
2175
2176         init_waitqueue_head(&tfile->wq.wait);
2177         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2178
2179         tfile->socket.file = file;
2180         tfile->socket.ops = &tun_socket_ops;
2181
2182         sock_init_data(&tfile->socket, &tfile->sk);
2183         sk_change_net(&tfile->sk, tfile->net);
2184
2185         tfile->sk.sk_write_space = tun_sock_write_space;
2186         tfile->sk.sk_sndbuf = INT_MAX;
2187
2188         file->private_data = tfile;
2189         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2190         INIT_LIST_HEAD(&tfile->next);
2191
2192         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2193
2194         return 0;
2195 }
2196
2197 static int tun_chr_close(struct inode *inode, struct file *file)
2198 {
2199         struct tun_file *tfile = file->private_data;
2200         struct net *net = tfile->net;
2201
2202         tun_detach(tfile, true);
2203         put_net(net);
2204
2205         return 0;
2206 }
2207
2208 #ifdef CONFIG_PROC_FS
2209 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2210 {
2211         struct tun_struct *tun;
2212         struct ifreq ifr;
2213
2214         memset(&ifr, 0, sizeof(ifr));
2215
2216         rtnl_lock();
2217         tun = tun_get(f);
2218         if (tun)
2219                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2220         rtnl_unlock();
2221
2222         if (tun)
2223                 tun_put(tun);
2224
2225         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2226 }
2227 #endif
2228
2229 static const struct file_operations tun_fops = {
2230         .owner  = THIS_MODULE,
2231         .llseek = no_llseek,
2232         .read  = new_sync_read,
2233         .write = new_sync_write,
2234         .read_iter  = tun_chr_read_iter,
2235         .write_iter = tun_chr_write_iter,
2236         .poll   = tun_chr_poll,
2237         .unlocked_ioctl = tun_chr_ioctl,
2238 #ifdef CONFIG_COMPAT
2239         .compat_ioctl = tun_chr_compat_ioctl,
2240 #endif
2241         .open   = tun_chr_open,
2242         .release = tun_chr_close,
2243         .fasync = tun_chr_fasync,
2244 #ifdef CONFIG_PROC_FS
2245         .show_fdinfo = tun_chr_show_fdinfo,
2246 #endif
2247 };
2248
2249 static struct miscdevice tun_miscdev = {
2250         .minor = TUN_MINOR,
2251         .name = "tun",
2252         .nodename = "net/tun",
2253         .fops = &tun_fops,
2254 };
2255
2256 /* ethtool interface */
2257
2258 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2259 {
2260         cmd->supported          = 0;
2261         cmd->advertising        = 0;
2262         ethtool_cmd_speed_set(cmd, SPEED_10);
2263         cmd->duplex             = DUPLEX_FULL;
2264         cmd->port               = PORT_TP;
2265         cmd->phy_address        = 0;
2266         cmd->transceiver        = XCVR_INTERNAL;
2267         cmd->autoneg            = AUTONEG_DISABLE;
2268         cmd->maxtxpkt           = 0;
2269         cmd->maxrxpkt           = 0;
2270         return 0;
2271 }
2272
2273 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2274 {
2275         struct tun_struct *tun = netdev_priv(dev);
2276
2277         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2278         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2279
2280         switch (tun->flags & TUN_TYPE_MASK) {
2281         case IFF_TUN:
2282                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2283                 break;
2284         case IFF_TAP:
2285                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2286                 break;
2287         }
2288 }
2289
2290 static u32 tun_get_msglevel(struct net_device *dev)
2291 {
2292 #ifdef TUN_DEBUG
2293         struct tun_struct *tun = netdev_priv(dev);
2294         return tun->debug;
2295 #else
2296         return -EOPNOTSUPP;
2297 #endif
2298 }
2299
2300 static void tun_set_msglevel(struct net_device *dev, u32 value)
2301 {
2302 #ifdef TUN_DEBUG
2303         struct tun_struct *tun = netdev_priv(dev);
2304         tun->debug = value;
2305 #endif
2306 }
2307
2308 static const struct ethtool_ops tun_ethtool_ops = {
2309         .get_settings   = tun_get_settings,
2310         .get_drvinfo    = tun_get_drvinfo,
2311         .get_msglevel   = tun_get_msglevel,
2312         .set_msglevel   = tun_set_msglevel,
2313         .get_link       = ethtool_op_get_link,
2314         .get_ts_info    = ethtool_op_get_ts_info,
2315 };
2316
2317
2318 static int __init tun_init(void)
2319 {
2320         int ret = 0;
2321
2322         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2323         pr_info("%s\n", DRV_COPYRIGHT);
2324
2325         ret = rtnl_link_register(&tun_link_ops);
2326         if (ret) {
2327                 pr_err("Can't register link_ops\n");
2328                 goto err_linkops;
2329         }
2330
2331         ret = misc_register(&tun_miscdev);
2332         if (ret) {
2333                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2334                 goto err_misc;
2335         }
2336         return  0;
2337 err_misc:
2338         rtnl_link_unregister(&tun_link_ops);
2339 err_linkops:
2340         return ret;
2341 }
2342
2343 static void tun_cleanup(void)
2344 {
2345         misc_deregister(&tun_miscdev);
2346         rtnl_link_unregister(&tun_link_ops);
2347 }
2348
2349 /* Get an underlying socket object from tun file.  Returns error unless file is
2350  * attached to a device.  The returned object works like a packet socket, it
2351  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2352  * holding a reference to the file for as long as the socket is in use. */
2353 struct socket *tun_get_socket(struct file *file)
2354 {
2355         struct tun_file *tfile;
2356         if (file->f_op != &tun_fops)
2357                 return ERR_PTR(-EINVAL);
2358         tfile = file->private_data;
2359         if (!tfile)
2360                 return ERR_PTR(-EBADFD);
2361         return &tfile->socket;
2362 }
2363 EXPORT_SYMBOL_GPL(tun_get_socket);
2364
2365 module_init(tun_init);
2366 module_exit(tun_cleanup);
2367 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2368 MODULE_AUTHOR(DRV_COPYRIGHT);
2369 MODULE_LICENSE("GPL");
2370 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2371 MODULE_ALIAS("devname:net/tun");