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 #include <linux/skb_array.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 #define TUN_VNET_BE     0x40000000
 116
 117 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 118                       IFF_MULTI_QUEUE)
 119 #define GOODCOPY_LEN 128
 120
 121 #define FLT_EXACT_COUNT 8
 122 struct tap_filter {
 123         unsigned int    count;    /* Number of addrs. Zero means disabled */
 124         u32             mask[2];  /* Mask of the hashed addrs */
 125         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 126 };
 127
 128 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 129  * to max number of VCPUs in guest. */
 130 #define MAX_TAP_QUEUES 256
 131 #define MAX_TAP_FLOWS  4096
 132
 133 #define TUN_FLOW_EXPIRE (3 * HZ)
 134
 135 struct tun_pcpu_stats {
 136         u64 rx_packets;
 137         u64 rx_bytes;
 138         u64 tx_packets;
 139         u64 tx_bytes;
 140         struct u64_stats_sync syncp;
 141         u32 rx_dropped;
 142         u32 tx_dropped;
 143         u32 rx_frame_errors;
 144 };
 145
 146 /* A tun_file connects an open character device to a tuntap netdevice. It
 147  * also contains all socket related structures (except sock_fprog and tap_filter)
 148  * to serve as one transmit queue for tuntap device. The sock_fprog and
 149  * tap_filter were kept in tun_struct since they were used for filtering for the
 150  * netdevice not for a specific queue (at least I didn't see the requirement for
 151  * this).
 152  *
 153  * RCU usage:
 154  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 155  * other can only be read while rcu_read_lock or rtnl_lock is held.
 156  */
 157 struct tun_file {
 158         struct sock sk;
 159         struct socket socket;
 160         struct socket_wq wq;
 161         struct tun_struct __rcu *tun;
 162         struct fasync_struct *fasync;
 163         /* only used for fasnyc */
 164         unsigned int flags;
 165         union {
 166                 u16 queue_index;
 167                 unsigned int ifindex;
 168         };
 169         struct list_head next;
 170         struct tun_struct *detached;
 171         struct skb_array tx_array;
 172 };
 173
 174 struct tun_flow_entry {
 175         struct hlist_node hash_link;
 176         struct rcu_head rcu;
 177         struct tun_struct *tun;
 178
 179         u32 rxhash;
 180         u32 rps_rxhash;
 181         int queue_index;
 182         unsigned long updated;
 183 };
 184
 185 #define TUN_NUM_FLOW_ENTRIES 1024
 186
 187 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 188  * device, socket filter, sndbuf and vnet header size were restore when the
 189  * file were attached to a persist device.
 190  */
 191 struct tun_struct {
 192         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 193         unsigned int            numqueues;
 194         unsigned int            flags;
 195         kuid_t                  owner;
 196         kgid_t                  group;
 197
 198         struct net_device       *dev;
 199         netdev_features_t       set_features;
 200 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 201                           NETIF_F_TSO6|NETIF_F_UFO)
 202
 203         int                     align;
 204         int                     vnet_hdr_sz;
 205         int                     sndbuf;
 206         struct tap_filter       txflt;
 207         struct sock_fprog       fprog;
 208         /* protected by rtnl lock */
 209         bool                    filter_attached;
 210 #ifdef TUN_DEBUG
 211         int debug;
 212 #endif
 213         spinlock_t lock;
 214         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 215         struct timer_list flow_gc_timer;
 216         unsigned long ageing_time;
 217         unsigned int numdisabled;
 218         struct list_head disabled;
 219         void *security;
 220         u32 flow_count;
 221         struct tun_pcpu_stats __percpu *pcpu_stats;
 222 };
 223
 224 #ifdef CONFIG_TUN_VNET_CROSS_LE
 225 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 226 {
 227         return tun->flags & TUN_VNET_BE ? false :
 228                 virtio_legacy_is_little_endian();
 229 }
 230
 231 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 232 {
 233         int be = !!(tun->flags & TUN_VNET_BE);
 234
 235         if (put_user(be, argp))
 236                 return -EFAULT;
 237
 238         return 0;
 239 }
 240
 241 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 242 {
 243         int be;
 244
 245         if (get_user(be, argp))
 246                 return -EFAULT;
 247
 248         if (be)
 249                 tun->flags |= TUN_VNET_BE;
 250         else
 251                 tun->flags &= ~TUN_VNET_BE;
 252
 253         return 0;
 254 }
 255 #else
 256 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 257 {
 258         return virtio_legacy_is_little_endian();
 259 }
 260
 261 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 262 {
 263         return -EINVAL;
 264 }
 265
 266 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 267 {
 268         return -EINVAL;
 269 }
 270 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 271
 272 static inline bool tun_is_little_endian(struct tun_struct *tun)
 273 {
 274         return tun->flags & TUN_VNET_LE ||
 275                 tun_legacy_is_little_endian(tun);
 276 }
 277
 278 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 279 {
 280         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 281 }
 282
 283 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 284 {
 285         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 286 }
 287
 288 static inline u32 tun_hashfn(u32 rxhash)
 289 {
 290         return rxhash & 0x3ff;
 291 }
 292
 293 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 294 {
 295         struct tun_flow_entry *e;
 296
 297         hlist_for_each_entry_rcu(e, head, hash_link) {
 298                 if (e->rxhash == rxhash)
 299                         return e;
 300         }
 301         return NULL;
 302 }
 303
 304 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 305                                               struct hlist_head *head,
 306                                               u32 rxhash, u16 queue_index)
 307 {
 308         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 309
 310         if (e) {
 311                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 312                           rxhash, queue_index);
 313                 e->updated = jiffies;
 314                 e->rxhash = rxhash;
 315                 e->rps_rxhash = 0;
 316                 e->queue_index = queue_index;
 317                 e->tun = tun;
 318                 hlist_add_head_rcu(&e->hash_link, head);
 319                 ++tun->flow_count;
 320         }
 321         return e;
 322 }
 323
 324 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 325 {
 326         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 327                   e->rxhash, e->queue_index);
 328         hlist_del_rcu(&e->hash_link);
 329         kfree_rcu(e, rcu);
 330         --tun->flow_count;
 331 }
 332
 333 static void tun_flow_flush(struct tun_struct *tun)
 334 {
 335         int i;
 336
 337         spin_lock_bh(&tun->lock);
 338         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 339                 struct tun_flow_entry *e;
 340                 struct hlist_node *n;
 341
 342                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 343                         tun_flow_delete(tun, e);
 344         }
 345         spin_unlock_bh(&tun->lock);
 346 }
 347
 348 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 349 {
 350         int i;
 351
 352         spin_lock_bh(&tun->lock);
 353         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 354                 struct tun_flow_entry *e;
 355                 struct hlist_node *n;
 356
 357                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 358                         if (e->queue_index == queue_index)
 359                                 tun_flow_delete(tun, e);
 360                 }
 361         }
 362         spin_unlock_bh(&tun->lock);
 363 }
 364
 365 static void tun_flow_cleanup(unsigned long data)
 366 {
 367         struct tun_struct *tun = (struct tun_struct *)data;
 368         unsigned long delay = tun->ageing_time;
 369         unsigned long next_timer = jiffies + delay;
 370         unsigned long count = 0;
 371         int i;
 372
 373         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 374
 375         spin_lock_bh(&tun->lock);
 376         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 377                 struct tun_flow_entry *e;
 378                 struct hlist_node *n;
 379
 380                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 381                         unsigned long this_timer;
 382                         count++;
 383                         this_timer = e->updated + delay;
 384                         if (time_before_eq(this_timer, jiffies))
 385                                 tun_flow_delete(tun, e);
 386                         else if (time_before(this_timer, next_timer))
 387                                 next_timer = this_timer;
 388                 }
 389         }
 390
 391         if (count)
 392                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 393         spin_unlock_bh(&tun->lock);
 394 }
 395
 396 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 397                             struct tun_file *tfile)
 398 {
 399         struct hlist_head *head;
 400         struct tun_flow_entry *e;
 401         unsigned long delay = tun->ageing_time;
 402         u16 queue_index = tfile->queue_index;
 403
 404         if (!rxhash)
 405                 return;
 406         else
 407                 head = &tun->flows[tun_hashfn(rxhash)];
 408
 409         rcu_read_lock();
 410
 411         /* We may get a very small possibility of OOO during switching, not
 412          * worth to optimize.*/
 413         if (tun->numqueues == 1 || tfile->detached)
 414                 goto unlock;
 415
 416         e = tun_flow_find(head, rxhash);
 417         if (likely(e)) {
 418                 /* TODO: keep queueing to old queue until it's empty? */
 419                 e->queue_index = queue_index;
 420                 e->updated = jiffies;
 421                 sock_rps_record_flow_hash(e->rps_rxhash);
 422         } else {
 423                 spin_lock_bh(&tun->lock);
 424                 if (!tun_flow_find(head, rxhash) &&
 425                     tun->flow_count < MAX_TAP_FLOWS)
 426                         tun_flow_create(tun, head, rxhash, queue_index);
 427
 428                 if (!timer_pending(&tun->flow_gc_timer))
 429                         mod_timer(&tun->flow_gc_timer,
 430                                   round_jiffies_up(jiffies + delay));
 431                 spin_unlock_bh(&tun->lock);
 432         }
 433
 434 unlock:
 435         rcu_read_unlock();
 436 }
 437
 438 /**
 439  * Save the hash received in the stack receive path and update the
 440  * flow_hash table accordingly.
 441  */
 442 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 443 {
 444         if (unlikely(e->rps_rxhash != hash))
 445                 e->rps_rxhash = hash;
 446 }
 447
 448 /* We try to identify a flow through its rxhash first. The reason that
 449  * we do not check rxq no. is because some cards(e.g 82599), chooses
 450  * the rxq based on the txq where the last packet of the flow comes. As
 451  * the userspace application move between processors, we may get a
 452  * different rxq no. here. If we could not get rxhash, then we would
 453  * hope the rxq no. may help here.
 454  */
 455 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 456                             void *accel_priv, select_queue_fallback_t fallback)
 457 {
 458         struct tun_struct *tun = netdev_priv(dev);
 459         struct tun_flow_entry *e;
 460         u32 txq = 0;
 461         u32 numqueues = 0;
 462
 463         rcu_read_lock();
 464         numqueues = ACCESS_ONCE(tun->numqueues);
 465
 466         txq = skb_get_hash(skb);
 467         if (txq) {
 468                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 469                 if (e) {
 470                         tun_flow_save_rps_rxhash(e, txq);
 471                         txq = e->queue_index;
 472                 } else
 473                         /* use multiply and shift instead of expensive divide */
 474                         txq = ((u64)txq * numqueues) >> 32;
 475         } else if (likely(skb_rx_queue_recorded(skb))) {
 476                 txq = skb_get_rx_queue(skb);
 477                 while (unlikely(txq >= numqueues))
 478                         txq -= numqueues;
 479         }
 480
 481         rcu_read_unlock();
 482         return txq;
 483 }
 484
 485 static inline bool tun_not_capable(struct tun_struct *tun)
 486 {
 487         const struct cred *cred = current_cred();
 488         struct net *net = dev_net(tun->dev);
 489
 490         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 491                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 492                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 493 }
 494
 495 static void tun_set_real_num_queues(struct tun_struct *tun)
 496 {
 497         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 498         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 499 }
 500
 501 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 502 {
 503         tfile->detached = tun;
 504         list_add_tail(&tfile->next, &tun->disabled);
 505         ++tun->numdisabled;
 506 }
 507
 508 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 509 {
 510         struct tun_struct *tun = tfile->detached;
 511
 512         tfile->detached = NULL;
 513         list_del_init(&tfile->next);
 514         --tun->numdisabled;
 515         return tun;
 516 }
 517
 518 static void tun_queue_purge(struct tun_file *tfile)
 519 {
 520         struct sk_buff *skb;
 521
 522         while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
 523                 kfree_skb(skb);
 524
 525         skb_queue_purge(&tfile->sk.sk_error_queue);
 526 }
 527
 528 static void tun_cleanup_tx_array(struct tun_file *tfile)
 529 {
 530         if (tfile->tx_array.ring.queue) {
 531                 skb_array_cleanup(&tfile->tx_array);
 532                 memset(&tfile->tx_array, 0, sizeof(tfile->tx_array));
 533         }
 534 }
 535
 536 static void __tun_detach(struct tun_file *tfile, bool clean)
 537 {
 538         struct tun_file *ntfile;
 539         struct tun_struct *tun;
 540
 541         tun = rtnl_dereference(tfile->tun);
 542
 543         if (tun && !tfile->detached) {
 544                 u16 index = tfile->queue_index;
 545                 BUG_ON(index >= tun->numqueues);
 546
 547                 rcu_assign_pointer(tun->tfiles[index],
 548                                    tun->tfiles[tun->numqueues - 1]);
 549                 ntfile = rtnl_dereference(tun->tfiles[index]);
 550                 ntfile->queue_index = index;
 551
 552                 --tun->numqueues;
 553                 if (clean) {
 554                         RCU_INIT_POINTER(tfile->tun, NULL);
 555                         sock_put(&tfile->sk);
 556                 } else
 557                         tun_disable_queue(tun, tfile);
 558
 559                 synchronize_net();
 560                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 561                 /* Drop read queue */
 562                 tun_queue_purge(tfile);
 563                 tun_set_real_num_queues(tun);
 564         } else if (tfile->detached && clean) {
 565                 tun = tun_enable_queue(tfile);
 566                 sock_put(&tfile->sk);
 567         }
 568
 569         if (clean) {
 570                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 571                         netif_carrier_off(tun->dev);
 572
 573                         if (!(tun->flags & IFF_PERSIST) &&
 574                             tun->dev->reg_state == NETREG_REGISTERED)
 575                                 unregister_netdevice(tun->dev);
 576                 }
 577                 tun_cleanup_tx_array(tfile);
 578                 sock_put(&tfile->sk);
 579         }
 580 }
 581
 582 static void tun_detach(struct tun_file *tfile, bool clean)
 583 {
 584         rtnl_lock();
 585         __tun_detach(tfile, clean);
 586         rtnl_unlock();
 587 }
 588
 589 static void tun_detach_all(struct net_device *dev)
 590 {
 591         struct tun_struct *tun = netdev_priv(dev);
 592         struct tun_file *tfile, *tmp;
 593         int i, n = tun->numqueues;
 594
 595         for (i = 0; i < n; i++) {
 596                 tfile = rtnl_dereference(tun->tfiles[i]);
 597                 BUG_ON(!tfile);
 598                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 599                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 600                 RCU_INIT_POINTER(tfile->tun, NULL);
 601                 --tun->numqueues;
 602         }
 603         list_for_each_entry(tfile, &tun->disabled, next) {
 604                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 605                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 606                 RCU_INIT_POINTER(tfile->tun, NULL);
 607         }
 608         BUG_ON(tun->numqueues != 0);
 609
 610         synchronize_net();
 611         for (i = 0; i < n; i++) {
 612                 tfile = rtnl_dereference(tun->tfiles[i]);
 613                 /* Drop read queue */
 614                 tun_queue_purge(tfile);
 615                 sock_put(&tfile->sk);
 616                 tun_cleanup_tx_array(tfile);
 617         }
 618         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 619                 tun_enable_queue(tfile);
 620                 tun_queue_purge(tfile);
 621                 sock_put(&tfile->sk);
 622                 tun_cleanup_tx_array(tfile);
 623         }
 624         BUG_ON(tun->numdisabled != 0);
 625
 626         if (tun->flags & IFF_PERSIST)
 627                 module_put(THIS_MODULE);
 628 }
 629
 630 static int tun_attach(struct tun_struct *tun, struct file *file,
 631                       bool skip_filter, bool publish_tun)
 632 {
 633         struct tun_file *tfile = file->private_data;
 634         struct net_device *dev = tun->dev;
 635         int err;
 636
 637         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 638         if (err < 0)
 639                 goto out;
 640
 641         err = -EINVAL;
 642         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 643                 goto out;
 644
 645         err = -EBUSY;
 646         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 647                 goto out;
 648
 649         err = -E2BIG;
 650         if (!tfile->detached &&
 651             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 652                 goto out;
 653
 654         err = 0;
 655
 656         /* Re-attach the filter to persist device */
 657         if (!skip_filter && (tun->filter_attached == true)) {
 658                 lock_sock(tfile->socket.sk);
 659                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 660                 release_sock(tfile->socket.sk);
 661                 if (!err)
 662                         goto out;
 663         }
 664
 665         if (!tfile->detached &&
 666             skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
 667                 err = -ENOMEM;
 668                 goto out;
 669         }
 670
 671         tfile->queue_index = tun->numqueues;
 672         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 673         if (publish_tun)
 674                 rcu_assign_pointer(tfile->tun, tun);
 675         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 676         tun->numqueues++;
 677
 678         if (tfile->detached)
 679                 tun_enable_queue(tfile);
 680         else
 681                 sock_hold(&tfile->sk);
 682
 683         tun_set_real_num_queues(tun);
 684
 685         /* device is allowed to go away first, so no need to hold extra
 686          * refcnt.
 687          */
 688
 689 out:
 690         return err;
 691 }
 692
 693 static struct tun_struct *__tun_get(struct tun_file *tfile)
 694 {
 695         struct tun_struct *tun;
 696
 697         rcu_read_lock();
 698         tun = rcu_dereference(tfile->tun);
 699         if (tun)
 700                 dev_hold(tun->dev);
 701         rcu_read_unlock();
 702
 703         return tun;
 704 }
 705
 706 static struct tun_struct *tun_get(struct file *file)
 707 {
 708         return __tun_get(file->private_data);
 709 }
 710
 711 static void tun_put(struct tun_struct *tun)
 712 {
 713         dev_put(tun->dev);
 714 }
 715
 716 /* TAP filtering */
 717 static void addr_hash_set(u32 *mask, const u8 *addr)
 718 {
 719         int n = ether_crc(ETH_ALEN, addr) >> 26;
 720         mask[n >> 5] |= (1 << (n & 31));
 721 }
 722
 723 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 724 {
 725         int n = ether_crc(ETH_ALEN, addr) >> 26;
 726         return mask[n >> 5] & (1 << (n & 31));
 727 }
 728
 729 static int update_filter(struct tap_filter *filter, void __user *arg)
 730 {
 731         struct { u8 u[ETH_ALEN]; } *addr;
 732         struct tun_filter uf;
 733         int err, alen, n, nexact;
 734
 735         if (copy_from_user(&uf, arg, sizeof(uf)))
 736                 return -EFAULT;
 737
 738         if (!uf.count) {
 739                 /* Disabled */
 740                 filter->count = 0;
 741                 return 0;
 742         }
 743
 744         alen = ETH_ALEN * uf.count;
 745         addr = memdup_user(arg + sizeof(uf), alen);
 746         if (IS_ERR(addr))
 747                 return PTR_ERR(addr);
 748
 749         /* The filter is updated without holding any locks. Which is
 750          * perfectly safe. We disable it first and in the worst
 751          * case we'll accept a few undesired packets. */
 752         filter->count = 0;
 753         wmb();
 754
 755         /* Use first set of addresses as an exact filter */
 756         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 757                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 758
 759         nexact = n;
 760
 761         /* Remaining multicast addresses are hashed,
 762          * unicast will leave the filter disabled. */
 763         memset(filter->mask, 0, sizeof(filter->mask));
 764         for (; n < uf.count; n++) {
 765                 if (!is_multicast_ether_addr(addr[n].u)) {
 766                         err = 0; /* no filter */
 767                         goto free_addr;
 768                 }
 769                 addr_hash_set(filter->mask, addr[n].u);
 770         }
 771
 772         /* For ALLMULTI just set the mask to all ones.
 773          * This overrides the mask populated above. */
 774         if ((uf.flags & TUN_FLT_ALLMULTI))
 775                 memset(filter->mask, ~0, sizeof(filter->mask));
 776
 777         /* Now enable the filter */
 778         wmb();
 779         filter->count = nexact;
 780
 781         /* Return the number of exact filters */
 782         err = nexact;
 783 free_addr:
 784         kfree(addr);
 785         return err;
 786 }
 787
 788 /* Returns: 0 - drop, !=0 - accept */
 789 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 790 {
 791         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 792          * at this point. */
 793         struct ethhdr *eh = (struct ethhdr *) skb->data;
 794         int i;
 795
 796         /* Exact match */
 797         for (i = 0; i < filter->count; i++)
 798                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 799                         return 1;
 800
 801         /* Inexact match (multicast only) */
 802         if (is_multicast_ether_addr(eh->h_dest))
 803                 return addr_hash_test(filter->mask, eh->h_dest);
 804
 805         return 0;
 806 }
 807
 808 /*
 809  * Checks whether the packet is accepted or not.
 810  * Returns: 0 - drop, !=0 - accept
 811  */
 812 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 813 {
 814         if (!filter->count)
 815                 return 1;
 816
 817         return run_filter(filter, skb);
 818 }
 819
 820 /* Network device part of the driver */
 821
 822 static const struct ethtool_ops tun_ethtool_ops;
 823
 824 /* Net device detach from fd. */
 825 static void tun_net_uninit(struct net_device *dev)
 826 {
 827         tun_detach_all(dev);
 828 }
 829
 830 /* Net device open. */
 831 static int tun_net_open(struct net_device *dev)
 832 {
 833         netif_tx_start_all_queues(dev);
 834
 835         return 0;
 836 }
 837
 838 /* Net device close. */
 839 static int tun_net_close(struct net_device *dev)
 840 {
 841         netif_tx_stop_all_queues(dev);
 842         return 0;
 843 }
 844
 845 /* Net device start xmit */
 846 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 847 {
 848         struct tun_struct *tun = netdev_priv(dev);
 849         int txq = skb->queue_mapping;
 850         struct tun_file *tfile;
 851         u32 numqueues = 0;
 852
 853         rcu_read_lock();
 854         tfile = rcu_dereference(tun->tfiles[txq]);
 855         numqueues = ACCESS_ONCE(tun->numqueues);
 856
 857         /* Drop packet if interface is not attached */
 858         if (txq >= numqueues)
 859                 goto drop;
 860
 861 #ifdef CONFIG_RPS
 862         if (numqueues == 1 && static_key_false(&rps_needed)) {
 863                 /* Select queue was not called for the skbuff, so we extract the
 864                  * RPS hash and save it into the flow_table here.
 865                  */
 866                 __u32 rxhash;
 867
 868                 rxhash = skb_get_hash(skb);
 869                 if (rxhash) {
 870                         struct tun_flow_entry *e;
 871                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 872                                         rxhash);
 873                         if (e)
 874                                 tun_flow_save_rps_rxhash(e, rxhash);
 875                 }
 876         }
 877 #endif
 878
 879         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 880
 881         BUG_ON(!tfile);
 882
 883         /* Drop if the filter does not like it.
 884          * This is a noop if the filter is disabled.
 885          * Filter can be enabled only for the TAP devices. */
 886         if (!check_filter(&tun->txflt, skb))
 887                 goto drop;
 888
 889         if (tfile->socket.sk->sk_filter &&
 890             sk_filter(tfile->socket.sk, skb))
 891                 goto drop;
 892
 893         /* Limit the number of packets queued by dividing txq length with the
 894          * number of queues.
 895          */
 896         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
 897                           >= dev->tx_queue_len)
 898                 goto drop;
 899
 900         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 901                 goto drop;
 902
 903         skb_tx_timestamp(skb);
 904
 905         /* Orphan the skb - required as we might hang on to it
 906          * for indefinite time.
 907          */
 908         skb_orphan(skb);
 909
 910         nf_reset(skb);
 911
 912         if (skb_array_produce(&tfile->tx_array, skb))
 913                 goto drop;
 914
 915         /* Notify and wake up reader process */
 916         if (tfile->flags & TUN_FASYNC)
 917                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 918         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 919
 920         rcu_read_unlock();
 921         return NETDEV_TX_OK;
 922
 923 drop:
 924         this_cpu_inc(tun->pcpu_stats->tx_dropped);
 925         skb_tx_error(skb);
 926         kfree_skb(skb);
 927         rcu_read_unlock();
 928         return NET_XMIT_DROP;
 929 }
 930
 931 static void tun_net_mclist(struct net_device *dev)
 932 {
 933         /*
 934          * This callback is supposed to deal with mc filter in
 935          * _rx_ path and has nothing to do with the _tx_ path.
 936          * In rx path we always accept everything userspace gives us.
 937          */
 938 }
 939
 940 #define MIN_MTU 68
 941 #define MAX_MTU 65535
 942
 943 static int
 944 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 945 {
 946         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 947                 return -EINVAL;
 948         dev->mtu = new_mtu;
 949         return 0;
 950 }
 951
 952 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 953         netdev_features_t features)
 954 {
 955         struct tun_struct *tun = netdev_priv(dev);
 956
 957         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 958 }
 959 #ifdef CONFIG_NET_POLL_CONTROLLER
 960 static void tun_poll_controller(struct net_device *dev)
 961 {
 962         /*
 963          * Tun only receives frames when:
 964          * 1) the char device endpoint gets data from user space
 965          * 2) the tun socket gets a sendmsg call from user space
 966          * Since both of those are synchronous operations, we are guaranteed
 967          * never to have pending data when we poll for it
 968          * so there is nothing to do here but return.
 969          * We need this though so netpoll recognizes us as an interface that
 970          * supports polling, which enables bridge devices in virt setups to
 971          * still use netconsole
 972          */
 973         return;
 974 }
 975 #endif
 976
 977 static void tun_set_headroom(struct net_device *dev, int new_hr)
 978 {
 979         struct tun_struct *tun = netdev_priv(dev);
 980
 981         if (new_hr < NET_SKB_PAD)
 982                 new_hr = NET_SKB_PAD;
 983
 984         tun->align = new_hr;
 985 }
 986
 987 static struct rtnl_link_stats64 *
 988 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
 989 {
 990         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
 991         struct tun_struct *tun = netdev_priv(dev);
 992         struct tun_pcpu_stats *p;
 993         int i;
 994
 995         for_each_possible_cpu(i) {
 996                 u64 rxpackets, rxbytes, txpackets, txbytes;
 997                 unsigned int start;
 998
 999                 p = per_cpu_ptr(tun->pcpu_stats, i);
1000                 do {
1001                         start = u64_stats_fetch_begin(&p->syncp);
1002                         rxpackets       = p->rx_packets;
1003                         rxbytes         = p->rx_bytes;
1004                         txpackets       = p->tx_packets;
1005                         txbytes         = p->tx_bytes;
1006                 } while (u64_stats_fetch_retry(&p->syncp, start));
1007
1008                 stats->rx_packets       += rxpackets;
1009                 stats->rx_bytes         += rxbytes;
1010                 stats->tx_packets       += txpackets;
1011                 stats->tx_bytes         += txbytes;
1012
1013                 /* u32 counters */
1014                 rx_dropped      += p->rx_dropped;
1015                 rx_frame_errors += p->rx_frame_errors;
1016                 tx_dropped      += p->tx_dropped;
1017         }
1018         stats->rx_dropped  = rx_dropped;
1019         stats->rx_frame_errors = rx_frame_errors;
1020         stats->tx_dropped = tx_dropped;
1021         return stats;
1022 }
1023
1024 static const struct net_device_ops tun_netdev_ops = {
1025         .ndo_uninit             = tun_net_uninit,
1026         .ndo_open               = tun_net_open,
1027         .ndo_stop               = tun_net_close,
1028         .ndo_start_xmit         = tun_net_xmit,
1029         .ndo_change_mtu         = tun_net_change_mtu,
1030         .ndo_fix_features       = tun_net_fix_features,
1031         .ndo_select_queue       = tun_select_queue,
1032 #ifdef CONFIG_NET_POLL_CONTROLLER
1033         .ndo_poll_controller    = tun_poll_controller,
1034 #endif
1035         .ndo_set_rx_headroom    = tun_set_headroom,
1036         .ndo_get_stats64        = tun_net_get_stats64,
1037 };
1038
1039 static const struct net_device_ops tap_netdev_ops = {
1040         .ndo_uninit             = tun_net_uninit,
1041         .ndo_open               = tun_net_open,
1042         .ndo_stop               = tun_net_close,
1043         .ndo_start_xmit         = tun_net_xmit,
1044         .ndo_change_mtu         = tun_net_change_mtu,
1045         .ndo_fix_features       = tun_net_fix_features,
1046         .ndo_set_rx_mode        = tun_net_mclist,
1047         .ndo_set_mac_address    = eth_mac_addr,
1048         .ndo_validate_addr      = eth_validate_addr,
1049         .ndo_select_queue       = tun_select_queue,
1050 #ifdef CONFIG_NET_POLL_CONTROLLER
1051         .ndo_poll_controller    = tun_poll_controller,
1052 #endif
1053         .ndo_features_check     = passthru_features_check,
1054         .ndo_set_rx_headroom    = tun_set_headroom,
1055         .ndo_get_stats64        = tun_net_get_stats64,
1056 };
1057
1058 static void tun_flow_init(struct tun_struct *tun)
1059 {
1060         int i;
1061
1062         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1063                 INIT_HLIST_HEAD(&tun->flows[i]);
1064
1065         tun->ageing_time = TUN_FLOW_EXPIRE;
1066         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1067         mod_timer(&tun->flow_gc_timer,
1068                   round_jiffies_up(jiffies + tun->ageing_time));
1069 }
1070
1071 static void tun_flow_uninit(struct tun_struct *tun)
1072 {
1073         del_timer_sync(&tun->flow_gc_timer);
1074         tun_flow_flush(tun);
1075 }
1076
1077 /* Initialize net device. */
1078 static void tun_net_init(struct net_device *dev)
1079 {
1080         struct tun_struct *tun = netdev_priv(dev);
1081
1082         switch (tun->flags & TUN_TYPE_MASK) {
1083         case IFF_TUN:
1084                 dev->netdev_ops = &tun_netdev_ops;
1085
1086                 /* Point-to-Point TUN Device */
1087                 dev->hard_header_len = 0;
1088                 dev->addr_len = 0;
1089                 dev->mtu = 1500;
1090
1091                 /* Zero header length */
1092                 dev->type = ARPHRD_NONE;
1093                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1094                 break;
1095
1096         case IFF_TAP:
1097                 dev->netdev_ops = &tap_netdev_ops;
1098                 /* Ethernet TAP Device */
1099                 ether_setup(dev);
1100                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1101                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1102
1103                 eth_hw_addr_random(dev);
1104
1105                 break;
1106         }
1107 }
1108
1109 /* Character device part */
1110
1111 /* Poll */
1112 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1113 {
1114         struct tun_file *tfile = file->private_data;
1115         struct tun_struct *tun = __tun_get(tfile);
1116         struct sock *sk;
1117         unsigned int mask = 0;
1118
1119         if (!tun)
1120                 return POLLERR;
1121
1122         sk = tfile->socket.sk;
1123
1124         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1125
1126         poll_wait(file, sk_sleep(sk), wait);
1127
1128         if (!skb_array_empty(&tfile->tx_array))
1129                 mask |= POLLIN | POLLRDNORM;
1130
1131         if (tun->dev->flags & IFF_UP &&
1132             (sock_writeable(sk) ||
1133              (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1134               sock_writeable(sk))))
1135                 mask |= POLLOUT | POLLWRNORM;
1136
1137         if (tun->dev->reg_state != NETREG_REGISTERED)
1138                 mask = POLLERR;
1139
1140         tun_put(tun);
1141         return mask;
1142 }
1143
1144 /* prepad is the amount to reserve at front.  len is length after that.
1145  * linear is a hint as to how much to copy (usually headers). */
1146 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1147                                      size_t prepad, size_t len,
1148                                      size_t linear, int noblock)
1149 {
1150         struct sock *sk = tfile->socket.sk;
1151         struct sk_buff *skb;
1152         int err;
1153
1154         /* Under a page?  Don't bother with paged skb. */
1155         if (prepad + len < PAGE_SIZE || !linear)
1156                 linear = len;
1157
1158         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1159                                    &err, 0);
1160         if (!skb)
1161                 return ERR_PTR(err);
1162
1163         skb_reserve(skb, prepad);
1164         skb_put(skb, linear);
1165         skb->data_len = len - linear;
1166         skb->len += len - linear;
1167
1168         return skb;
1169 }
1170
1171 /* Get packet from user space buffer */
1172 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1173                             void *msg_control, struct iov_iter *from,
1174                             int noblock)
1175 {
1176         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1177         struct sk_buff *skb;
1178         size_t total_len = iov_iter_count(from);
1179         size_t len = total_len, align = tun->align, linear;
1180         struct virtio_net_hdr gso = { 0 };
1181         struct tun_pcpu_stats *stats;
1182         int good_linear;
1183         int copylen;
1184         bool zerocopy = false;
1185         int err;
1186         u32 rxhash;
1187         ssize_t n;
1188
1189         if (!(tun->flags & IFF_NO_PI)) {
1190                 if (len < sizeof(pi))
1191                         return -EINVAL;
1192                 len -= sizeof(pi);
1193
1194                 n = copy_from_iter(&pi, sizeof(pi), from);
1195                 if (n != sizeof(pi))
1196                         return -EFAULT;
1197         }
1198
1199         if (tun->flags & IFF_VNET_HDR) {
1200                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1201
1202                 if (len < vnet_hdr_sz)
1203                         return -EINVAL;
1204                 len -= vnet_hdr_sz;
1205
1206                 n = copy_from_iter(&gso, sizeof(gso), from);
1207                 if (n != sizeof(gso))
1208                         return -EFAULT;
1209
1210                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1211                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1212                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1213
1214                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1215                         return -EINVAL;
1216                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1217         }
1218
1219         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1220                 align += NET_IP_ALIGN;
1221                 if (unlikely(len < ETH_HLEN ||
1222                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1223                         return -EINVAL;
1224         }
1225
1226         good_linear = SKB_MAX_HEAD(align);
1227
1228         if (msg_control) {
1229                 struct iov_iter i = *from;
1230
1231                 /* There are 256 bytes to be copied in skb, so there is
1232                  * enough room for skb expand head in case it is used.
1233                  * The rest of the buffer is mapped from userspace.
1234                  */
1235                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1236                 if (copylen > good_linear)
1237                         copylen = good_linear;
1238                 linear = copylen;
1239                 iov_iter_advance(&i, copylen);
1240                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1241                         zerocopy = true;
1242         }
1243
1244         if (!zerocopy) {
1245                 copylen = len;
1246                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1247                         linear = good_linear;
1248                 else
1249                         linear = tun16_to_cpu(tun, gso.hdr_len);
1250         }
1251
1252         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1253         if (IS_ERR(skb)) {
1254                 if (PTR_ERR(skb) != -EAGAIN)
1255                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1256                 return PTR_ERR(skb);
1257         }
1258
1259         if (zerocopy)
1260                 err = zerocopy_sg_from_iter(skb, from);
1261         else
1262                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1263
1264         if (err) {
1265                 err = -EFAULT;
1266 drop:
1267                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1268                 kfree_skb(skb);
1269                 return err;
1270         }
1271
1272         err = virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun));
1273         if (err) {
1274                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1275                 kfree_skb(skb);
1276                 return -EINVAL;
1277         }
1278
1279         switch (tun->flags & TUN_TYPE_MASK) {
1280         case IFF_TUN:
1281                 if (tun->flags & IFF_NO_PI) {
1282                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1283
1284                         switch (ip_version) {
1285                         case 4:
1286                                 pi.proto = htons(ETH_P_IP);
1287                                 break;
1288                         case 6:
1289                                 pi.proto = htons(ETH_P_IPV6);
1290                                 break;
1291                         default:
1292                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1293                                 kfree_skb(skb);
1294                                 return -EINVAL;
1295                         }
1296                 }
1297
1298                 skb_reset_mac_header(skb);
1299                 skb->protocol = pi.proto;
1300                 skb->dev = tun->dev;
1301                 break;
1302         case IFF_TAP:
1303                 skb->protocol = eth_type_trans(skb, tun->dev);
1304                 break;
1305         }
1306
1307         /* copy skb_ubuf_info for callback when skb has no error */
1308         if (zerocopy) {
1309                 skb_shinfo(skb)->destructor_arg = msg_control;
1310                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1311                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1312         } else if (msg_control) {
1313                 struct ubuf_info *uarg = msg_control;
1314                 uarg->callback(uarg, false);
1315         }
1316
1317         skb_reset_network_header(skb);
1318         skb_probe_transport_header(skb, 0);
1319
1320         rxhash = skb_get_hash(skb);
1321
1322         rcu_read_lock();
1323         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1324                 err = -EIO;
1325                 rcu_read_unlock();
1326                 goto drop;
1327         }
1328
1329         netif_rx_ni(skb);
1330         rcu_read_unlock();
1331
1332         stats = get_cpu_ptr(tun->pcpu_stats);
1333         u64_stats_update_begin(&stats->syncp);
1334         stats->rx_packets++;
1335         stats->rx_bytes += len;
1336         u64_stats_update_end(&stats->syncp);
1337         put_cpu_ptr(stats);
1338
1339         tun_flow_update(tun, rxhash, tfile);
1340         return total_len;
1341 }
1342
1343 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1344 {
1345         struct file *file = iocb->ki_filp;
1346         struct tun_struct *tun = tun_get(file);
1347         struct tun_file *tfile = file->private_data;
1348         ssize_t result;
1349
1350         if (!tun)
1351                 return -EBADFD;
1352
1353         result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1354
1355         tun_put(tun);
1356         return result;
1357 }
1358
1359 /* Put packet to the user space buffer */
1360 static ssize_t tun_put_user(struct tun_struct *tun,
1361                             struct tun_file *tfile,
1362                             struct sk_buff *skb,
1363                             struct iov_iter *iter)
1364 {
1365         struct tun_pi pi = { 0, skb->protocol };
1366         struct tun_pcpu_stats *stats;
1367         ssize_t total;
1368         int vlan_offset = 0;
1369         int vlan_hlen = 0;
1370         int vnet_hdr_sz = 0;
1371
1372         if (skb_vlan_tag_present(skb))
1373                 vlan_hlen = VLAN_HLEN;
1374
1375         if (tun->flags & IFF_VNET_HDR)
1376                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1377
1378         total = skb->len + vlan_hlen + vnet_hdr_sz;
1379
1380         if (!(tun->flags & IFF_NO_PI)) {
1381                 if (iov_iter_count(iter) < sizeof(pi))
1382                         return -EINVAL;
1383
1384                 total += sizeof(pi);
1385                 if (iov_iter_count(iter) < total) {
1386                         /* Packet will be striped */
1387                         pi.flags |= TUN_PKT_STRIP;
1388                 }
1389
1390                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1391                         return -EFAULT;
1392         }
1393
1394         if (vnet_hdr_sz) {
1395                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1396                 int ret;
1397
1398                 if (iov_iter_count(iter) < vnet_hdr_sz)
1399                         return -EINVAL;
1400
1401                 ret = virtio_net_hdr_from_skb(skb, &gso,
1402                                               tun_is_little_endian(tun), true);
1403                 if (ret) {
1404                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1405                         pr_err("unexpected GSO type: "
1406                                "0x%x, gso_size %d, hdr_len %d\n",
1407                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1408                                tun16_to_cpu(tun, gso.hdr_len));
1409                         print_hex_dump(KERN_ERR, "tun: ",
1410                                        DUMP_PREFIX_NONE,
1411                                        16, 1, skb->head,
1412                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1413                         WARN_ON_ONCE(1);
1414                         return -EINVAL;
1415                 }
1416
1417                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1418                         return -EFAULT;
1419
1420                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1421         }
1422
1423         if (vlan_hlen) {
1424                 int ret;
1425                 struct {
1426                         __be16 h_vlan_proto;
1427                         __be16 h_vlan_TCI;
1428                 } veth;
1429
1430                 veth.h_vlan_proto = skb->vlan_proto;
1431                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1432
1433                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1434
1435                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1436                 if (ret || !iov_iter_count(iter))
1437                         goto done;
1438
1439                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1440                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1441                         goto done;
1442         }
1443
1444         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1445
1446 done:
1447         /* caller is in process context, */
1448         stats = get_cpu_ptr(tun->pcpu_stats);
1449         u64_stats_update_begin(&stats->syncp);
1450         stats->tx_packets++;
1451         stats->tx_bytes += skb->len + vlan_hlen;
1452         u64_stats_update_end(&stats->syncp);
1453         put_cpu_ptr(tun->pcpu_stats);
1454
1455         return total;
1456 }
1457
1458 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1459                                      int *err)
1460 {
1461         DECLARE_WAITQUEUE(wait, current);
1462         struct sk_buff *skb = NULL;
1463         int error = 0;
1464
1465         skb = skb_array_consume(&tfile->tx_array);
1466         if (skb)
1467                 goto out;
1468         if (noblock) {
1469                 error = -EAGAIN;
1470                 goto out;
1471         }
1472
1473         add_wait_queue(&tfile->wq.wait, &wait);
1474
1475         while (1) {
1476                 set_current_state(TASK_INTERRUPTIBLE);
1477                 skb = skb_array_consume(&tfile->tx_array);
1478                 if (skb)
1479                         break;
1480                 if (signal_pending(current)) {
1481                         error = -ERESTARTSYS;
1482                         break;
1483                 }
1484                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1485                         error = -EFAULT;
1486                         break;
1487                 }
1488
1489                 schedule();
1490         }
1491
1492         __set_current_state(TASK_RUNNING);
1493         remove_wait_queue(&tfile->wq.wait, &wait);
1494
1495 out:
1496         *err = error;
1497         return skb;
1498 }
1499
1500 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1501                            struct iov_iter *to,
1502                            int noblock)
1503 {
1504         struct sk_buff *skb;
1505         ssize_t ret;
1506         int err;
1507
1508         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1509
1510         if (!iov_iter_count(to))
1511                 return 0;
1512
1513         /* Read frames from ring */
1514         skb = tun_ring_recv(tfile, noblock, &err);
1515         if (!skb)
1516                 return err;
1517
1518         ret = tun_put_user(tun, tfile, skb, to);
1519         if (unlikely(ret < 0))
1520                 kfree_skb(skb);
1521         else
1522                 consume_skb(skb);
1523
1524         return ret;
1525 }
1526
1527 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1528 {
1529         struct file *file = iocb->ki_filp;
1530         struct tun_file *tfile = file->private_data;
1531         struct tun_struct *tun = __tun_get(tfile);
1532         ssize_t len = iov_iter_count(to), ret;
1533
1534         if (!tun)
1535                 return -EBADFD;
1536         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1537         ret = min_t(ssize_t, ret, len);
1538         if (ret > 0)
1539                 iocb->ki_pos = ret;
1540         tun_put(tun);
1541         return ret;
1542 }
1543
1544 static void tun_free_netdev(struct net_device *dev)
1545 {
1546         struct tun_struct *tun = netdev_priv(dev);
1547
1548         BUG_ON(!(list_empty(&tun->disabled)));
1549         free_percpu(tun->pcpu_stats);
1550         tun_flow_uninit(tun);
1551         security_tun_dev_free_security(tun->security);
1552         free_netdev(dev);
1553 }
1554
1555 static void tun_setup(struct net_device *dev)
1556 {
1557         struct tun_struct *tun = netdev_priv(dev);
1558
1559         tun->owner = INVALID_UID;
1560         tun->group = INVALID_GID;
1561
1562         dev->ethtool_ops = &tun_ethtool_ops;
1563         dev->destructor = tun_free_netdev;
1564         /* We prefer our own queue length */
1565         dev->tx_queue_len = TUN_READQ_SIZE;
1566 }
1567
1568 /* Trivial set of netlink ops to allow deleting tun or tap
1569  * device with netlink.
1570  */
1571 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1572 {
1573         /* NL_SET_ERR_MSG(extack,
1574                        "tun/tap creation via rtnetlink is not supported."); */
1575         return -EOPNOTSUPP;
1576 }
1577
1578 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1579         .kind           = DRV_NAME,
1580         .priv_size      = sizeof(struct tun_struct),
1581         .setup          = tun_setup,
1582         .validate       = tun_validate,
1583 };
1584
1585 static void tun_sock_write_space(struct sock *sk)
1586 {
1587         struct tun_file *tfile;
1588         wait_queue_head_t *wqueue;
1589
1590         if (!sock_writeable(sk))
1591                 return;
1592
1593         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1594                 return;
1595
1596         wqueue = sk_sleep(sk);
1597         if (wqueue && waitqueue_active(wqueue))
1598                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1599                                                 POLLWRNORM | POLLWRBAND);
1600
1601         tfile = container_of(sk, struct tun_file, sk);
1602         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1603 }
1604
1605 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1606 {
1607         int ret;
1608         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1609         struct tun_struct *tun = __tun_get(tfile);
1610
1611         if (!tun)
1612                 return -EBADFD;
1613
1614         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1615                            m->msg_flags & MSG_DONTWAIT);
1616         tun_put(tun);
1617         return ret;
1618 }
1619
1620 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1621                        int flags)
1622 {
1623         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1624         struct tun_struct *tun = __tun_get(tfile);
1625         int ret;
1626
1627         if (!tun)
1628                 return -EBADFD;
1629
1630         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1631                 ret = -EINVAL;
1632                 goto out;
1633         }
1634         if (flags & MSG_ERRQUEUE) {
1635                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1636                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1637                 goto out;
1638         }
1639         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1640         if (ret > (ssize_t)total_len) {
1641                 m->msg_flags |= MSG_TRUNC;
1642                 ret = flags & MSG_TRUNC ? ret : total_len;
1643         }
1644 out:
1645         tun_put(tun);
1646         return ret;
1647 }
1648
1649 static int tun_peek_len(struct socket *sock)
1650 {
1651         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1652         struct tun_struct *tun;
1653         int ret = 0;
1654
1655         tun = __tun_get(tfile);
1656         if (!tun)
1657                 return 0;
1658
1659         ret = skb_array_peek_len(&tfile->tx_array);
1660         tun_put(tun);
1661
1662         return ret;
1663 }
1664
1665 /* Ops structure to mimic raw sockets with tun */
1666 static const struct proto_ops tun_socket_ops = {
1667         .peek_len = tun_peek_len,
1668         .sendmsg = tun_sendmsg,
1669         .recvmsg = tun_recvmsg,
1670 };
1671
1672 static struct proto tun_proto = {
1673         .name           = "tun",
1674         .owner          = THIS_MODULE,
1675         .obj_size       = sizeof(struct tun_file),
1676 };
1677
1678 static int tun_flags(struct tun_struct *tun)
1679 {
1680         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1681 }
1682
1683 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1684                               char *buf)
1685 {
1686         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1687         return sprintf(buf, "0x%x\n", tun_flags(tun));
1688 }
1689
1690 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1691                               char *buf)
1692 {
1693         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1694         return uid_valid(tun->owner)?
1695                 sprintf(buf, "%u\n",
1696                         from_kuid_munged(current_user_ns(), tun->owner)):
1697                 sprintf(buf, "-1\n");
1698 }
1699
1700 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1701                               char *buf)
1702 {
1703         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1704         return gid_valid(tun->group) ?
1705                 sprintf(buf, "%u\n",
1706                         from_kgid_munged(current_user_ns(), tun->group)):
1707                 sprintf(buf, "-1\n");
1708 }
1709
1710 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1711 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1712 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1713
1714 static struct attribute *tun_dev_attrs[] = {
1715         &dev_attr_tun_flags.attr,
1716         &dev_attr_owner.attr,
1717         &dev_attr_group.attr,
1718         NULL
1719 };
1720
1721 static const struct attribute_group tun_attr_group = {
1722         .attrs = tun_dev_attrs
1723 };
1724
1725 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1726 {
1727         struct tun_struct *tun;
1728         struct tun_file *tfile = file->private_data;
1729         struct net_device *dev;
1730         int err;
1731
1732         if (tfile->detached)
1733                 return -EINVAL;
1734
1735         dev = __dev_get_by_name(net, ifr->ifr_name);
1736         if (dev) {
1737                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1738                         return -EBUSY;
1739                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1740                         tun = netdev_priv(dev);
1741                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1742                         tun = netdev_priv(dev);
1743                 else
1744                         return -EINVAL;
1745
1746                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1747                     !!(tun->flags & IFF_MULTI_QUEUE))
1748                         return -EINVAL;
1749
1750                 if (tun_not_capable(tun))
1751                         return -EPERM;
1752                 err = security_tun_dev_open(tun->security);
1753                 if (err < 0)
1754                         return err;
1755
1756                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER, true);
1757                 if (err < 0)
1758                         return err;
1759
1760                 if (tun->flags & IFF_MULTI_QUEUE &&
1761                     (tun->numqueues + tun->numdisabled > 1)) {
1762                         /* One or more queue has already been attached, no need
1763                          * to initialize the device again.
1764                          */
1765                         return 0;
1766                 }
1767         }
1768         else {
1769                 char *name;
1770                 unsigned long flags = 0;
1771                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1772                              MAX_TAP_QUEUES : 1;
1773
1774                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1775                         return -EPERM;
1776                 err = security_tun_dev_create();
1777                 if (err < 0)
1778                         return err;
1779
1780                 /* Set dev type */
1781                 if (ifr->ifr_flags & IFF_TUN) {
1782                         /* TUN device */
1783                         flags |= IFF_TUN;
1784                         name = "tun%d";
1785                 } else if (ifr->ifr_flags & IFF_TAP) {
1786                         /* TAP device */
1787                         flags |= IFF_TAP;
1788                         name = "tap%d";
1789                 } else
1790                         return -EINVAL;
1791
1792                 if (*ifr->ifr_name)
1793                         name = ifr->ifr_name;
1794
1795                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1796                                        NET_NAME_UNKNOWN, tun_setup, queues,
1797                                        queues);
1798
1799                 if (!dev)
1800                         return -ENOMEM;
1801                 err = dev_get_valid_name(net, dev, name);
1802                 if (err < 0)
1803                         goto err_free_dev;
1804
1805                 dev_net_set(dev, net);
1806                 dev->rtnl_link_ops = &tun_link_ops;
1807                 dev->ifindex = tfile->ifindex;
1808                 dev->sysfs_groups[0] = &tun_attr_group;
1809
1810                 tun = netdev_priv(dev);
1811                 tun->dev = dev;
1812                 tun->flags = flags;
1813                 tun->txflt.count = 0;
1814                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1815
1816                 tun->align = NET_SKB_PAD;
1817                 tun->filter_attached = false;
1818                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1819
1820                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
1821                 if (!tun->pcpu_stats) {
1822                         err = -ENOMEM;
1823                         goto err_free_dev;
1824                 }
1825
1826                 spin_lock_init(&tun->lock);
1827
1828                 err = security_tun_dev_alloc_security(&tun->security);
1829                 if (err < 0)
1830                         goto err_free_stat;
1831
1832                 tun_net_init(dev);
1833                 tun_flow_init(tun);
1834
1835                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1836                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1837                                    NETIF_F_HW_VLAN_STAG_TX;
1838                 dev->features = dev->hw_features | NETIF_F_LLTX;
1839                 dev->vlan_features = dev->features &
1840                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1841                                        NETIF_F_HW_VLAN_STAG_TX);
1842
1843                 INIT_LIST_HEAD(&tun->disabled);
1844                 err = tun_attach(tun, file, false, false);
1845                 if (err < 0)
1846                         goto err_free_flow;
1847
1848                 err = register_netdevice(tun->dev);
1849                 if (err < 0)
1850                         goto err_detach;
1851                 /* free_netdev() won't check refcnt, to aovid race
1852                  * with dev_put() we need publish tun after registration.
1853                  */
1854                 rcu_assign_pointer(tfile->tun, tun);
1855         }
1856
1857         netif_carrier_on(tun->dev);
1858
1859         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1860
1861         tun->flags = (tun->flags & ~TUN_FEATURES) |
1862                 (ifr->ifr_flags & TUN_FEATURES);
1863
1864         /* Make sure persistent devices do not get stuck in
1865          * xoff state.
1866          */
1867         if (netif_running(tun->dev))
1868                 netif_tx_wake_all_queues(tun->dev);
1869
1870         strcpy(ifr->ifr_name, tun->dev->name);
1871         return 0;
1872
1873 err_detach:
1874         tun_detach_all(dev);
1875 err_free_flow:
1876         tun_flow_uninit(tun);
1877         security_tun_dev_free_security(tun->security);
1878 err_free_stat:
1879         free_percpu(tun->pcpu_stats);
1880 err_free_dev:
1881         free_netdev(dev);
1882         return err;
1883 }
1884
1885 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1886                        struct ifreq *ifr)
1887 {
1888         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1889
1890         strcpy(ifr->ifr_name, tun->dev->name);
1891
1892         ifr->ifr_flags = tun_flags(tun);
1893
1894 }
1895
1896 /* This is like a cut-down ethtool ops, except done via tun fd so no
1897  * privs required. */
1898 static int set_offload(struct tun_struct *tun, unsigned long arg)
1899 {
1900         netdev_features_t features = 0;
1901
1902         if (arg & TUN_F_CSUM) {
1903                 features |= NETIF_F_HW_CSUM;
1904                 arg &= ~TUN_F_CSUM;
1905
1906                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1907                         if (arg & TUN_F_TSO_ECN) {
1908                                 features |= NETIF_F_TSO_ECN;
1909                                 arg &= ~TUN_F_TSO_ECN;
1910                         }
1911                         if (arg & TUN_F_TSO4)
1912                                 features |= NETIF_F_TSO;
1913                         if (arg & TUN_F_TSO6)
1914                                 features |= NETIF_F_TSO6;
1915                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1916                 }
1917
1918                 if (arg & TUN_F_UFO) {
1919                         features |= NETIF_F_UFO;
1920                         arg &= ~TUN_F_UFO;
1921                 }
1922         }
1923
1924         /* This gives the user a way to test for new features in future by
1925          * trying to set them. */
1926         if (arg)
1927                 return -EINVAL;
1928
1929         tun->set_features = features;
1930         netdev_update_features(tun->dev);
1931
1932         return 0;
1933 }
1934
1935 static void tun_detach_filter(struct tun_struct *tun, int n)
1936 {
1937         int i;
1938         struct tun_file *tfile;
1939
1940         for (i = 0; i < n; i++) {
1941                 tfile = rtnl_dereference(tun->tfiles[i]);
1942                 lock_sock(tfile->socket.sk);
1943                 sk_detach_filter(tfile->socket.sk);
1944                 release_sock(tfile->socket.sk);
1945         }
1946
1947         tun->filter_attached = false;
1948 }
1949
1950 static int tun_attach_filter(struct tun_struct *tun)
1951 {
1952         int i, ret = 0;
1953         struct tun_file *tfile;
1954
1955         for (i = 0; i < tun->numqueues; i++) {
1956                 tfile = rtnl_dereference(tun->tfiles[i]);
1957                 lock_sock(tfile->socket.sk);
1958                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1959                 release_sock(tfile->socket.sk);
1960                 if (ret) {
1961                         tun_detach_filter(tun, i);
1962                         return ret;
1963                 }
1964         }
1965
1966         tun->filter_attached = true;
1967         return ret;
1968 }
1969
1970 static void tun_set_sndbuf(struct tun_struct *tun)
1971 {
1972         struct tun_file *tfile;
1973         int i;
1974
1975         for (i = 0; i < tun->numqueues; i++) {
1976                 tfile = rtnl_dereference(tun->tfiles[i]);
1977                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1978         }
1979 }
1980
1981 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1982 {
1983         struct tun_file *tfile = file->private_data;
1984         struct tun_struct *tun;
1985         int ret = 0;
1986
1987         rtnl_lock();
1988
1989         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1990                 tun = tfile->detached;
1991                 if (!tun) {
1992                         ret = -EINVAL;
1993                         goto unlock;
1994                 }
1995                 ret = security_tun_dev_attach_queue(tun->security);
1996                 if (ret < 0)
1997                         goto unlock;
1998                 ret = tun_attach(tun, file, false, true);
1999         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2000                 tun = rtnl_dereference(tfile->tun);
2001                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2002                         ret = -EINVAL;
2003                 else
2004                         __tun_detach(tfile, false);
2005         } else
2006                 ret = -EINVAL;
2007
2008 unlock:
2009         rtnl_unlock();
2010         return ret;
2011 }
2012
2013 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2014                             unsigned long arg, int ifreq_len)
2015 {
2016         struct tun_file *tfile = file->private_data;
2017         struct tun_struct *tun;
2018         void __user* argp = (void __user*)arg;
2019         struct ifreq ifr;
2020         kuid_t owner;
2021         kgid_t group;
2022         int sndbuf;
2023         int vnet_hdr_sz;
2024         unsigned int ifindex;
2025         int le;
2026         int ret;
2027
2028         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
2029                 if (copy_from_user(&ifr, argp, ifreq_len))
2030                         return -EFAULT;
2031         } else {
2032                 memset(&ifr, 0, sizeof(ifr));
2033         }
2034         if (cmd == TUNGETFEATURES) {
2035                 /* Currently this just means: "what IFF flags are valid?".
2036                  * This is needed because we never checked for invalid flags on
2037                  * TUNSETIFF.
2038                  */
2039                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2040                                 (unsigned int __user*)argp);
2041         } else if (cmd == TUNSETQUEUE)
2042                 return tun_set_queue(file, &ifr);
2043
2044         ret = 0;
2045         rtnl_lock();
2046
2047         tun = __tun_get(tfile);
2048         if (cmd == TUNSETIFF && !tun) {
2049                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2050
2051                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2052
2053                 if (ret)
2054                         goto unlock;
2055
2056                 if (copy_to_user(argp, &ifr, ifreq_len))
2057                         ret = -EFAULT;
2058                 goto unlock;
2059         }
2060         if (cmd == TUNSETIFINDEX) {
2061                 ret = -EPERM;
2062                 if (tun)
2063                         goto unlock;
2064
2065                 ret = -EFAULT;
2066                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2067                         goto unlock;
2068
2069                 ret = 0;
2070                 tfile->ifindex = ifindex;
2071                 goto unlock;
2072         }
2073
2074         ret = -EBADFD;
2075         if (!tun)
2076                 goto unlock;
2077
2078         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2079
2080         ret = 0;
2081         switch (cmd) {
2082         case TUNGETIFF:
2083                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2084
2085                 if (tfile->detached)
2086                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2087                 if (!tfile->socket.sk->sk_filter)
2088                         ifr.ifr_flags |= IFF_NOFILTER;
2089
2090                 if (copy_to_user(argp, &ifr, ifreq_len))
2091                         ret = -EFAULT;
2092                 break;
2093
2094         case TUNSETNOCSUM:
2095                 /* Disable/Enable checksum */
2096
2097                 /* [unimplemented] */
2098                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2099                           arg ? "disabled" : "enabled");
2100                 break;
2101
2102         case TUNSETPERSIST:
2103                 /* Disable/Enable persist mode. Keep an extra reference to the
2104                  * module to prevent the module being unprobed.
2105                  */
2106                 if (arg && !(tun->flags & IFF_PERSIST)) {
2107                         tun->flags |= IFF_PERSIST;
2108                         __module_get(THIS_MODULE);
2109                 }
2110                 if (!arg && (tun->flags & IFF_PERSIST)) {
2111                         tun->flags &= ~IFF_PERSIST;
2112                         module_put(THIS_MODULE);
2113                 }
2114
2115                 tun_debug(KERN_INFO, tun, "persist %s\n",
2116                           arg ? "enabled" : "disabled");
2117                 break;
2118
2119         case TUNSETOWNER:
2120                 /* Set owner of the device */
2121                 owner = make_kuid(current_user_ns(), arg);
2122                 if (!uid_valid(owner)) {
2123                         ret = -EINVAL;
2124                         break;
2125                 }
2126                 tun->owner = owner;
2127                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2128                           from_kuid(&init_user_ns, tun->owner));
2129                 break;
2130
2131         case TUNSETGROUP:
2132                 /* Set group of the device */
2133                 group = make_kgid(current_user_ns(), arg);
2134                 if (!gid_valid(group)) {
2135                         ret = -EINVAL;
2136                         break;
2137                 }
2138                 tun->group = group;
2139                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2140                           from_kgid(&init_user_ns, tun->group));
2141                 break;
2142
2143         case TUNSETLINK:
2144                 /* Only allow setting the type when the interface is down */
2145                 if (tun->dev->flags & IFF_UP) {
2146                         tun_debug(KERN_INFO, tun,
2147                                   "Linktype set failed because interface is up\n");
2148                         ret = -EBUSY;
2149                 } else {
2150                         tun->dev->type = (int) arg;
2151                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2152                                   tun->dev->type);
2153                         ret = 0;
2154                 }
2155                 break;
2156
2157 #ifdef TUN_DEBUG
2158         case TUNSETDEBUG:
2159                 tun->debug = arg;
2160                 break;
2161 #endif
2162         case TUNSETOFFLOAD:
2163                 ret = set_offload(tun, arg);
2164                 break;
2165
2166         case TUNSETTXFILTER:
2167                 /* Can be set only for TAPs */
2168                 ret = -EINVAL;
2169                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2170                         break;
2171                 ret = update_filter(&tun->txflt, (void __user *)arg);
2172                 break;
2173
2174         case SIOCGIFHWADDR:
2175                 /* Get hw address */
2176                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2177                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2178                 if (copy_to_user(argp, &ifr, ifreq_len))
2179                         ret = -EFAULT;
2180                 break;
2181
2182         case SIOCSIFHWADDR:
2183                 /* Set hw address */
2184                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2185                           ifr.ifr_hwaddr.sa_data);
2186
2187                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2188                 break;
2189
2190         case TUNGETSNDBUF:
2191                 sndbuf = tfile->socket.sk->sk_sndbuf;
2192                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2193                         ret = -EFAULT;
2194                 break;
2195
2196         case TUNSETSNDBUF:
2197                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2198                         ret = -EFAULT;
2199                         break;
2200                 }
2201                 if (sndbuf <= 0) {
2202                         ret = -EINVAL;
2203                         break;
2204                 }
2205
2206                 tun->sndbuf = sndbuf;
2207                 tun_set_sndbuf(tun);
2208                 break;
2209
2210         case TUNGETVNETHDRSZ:
2211                 vnet_hdr_sz = tun->vnet_hdr_sz;
2212                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2213                         ret = -EFAULT;
2214                 break;
2215
2216         case TUNSETVNETHDRSZ:
2217                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2218                         ret = -EFAULT;
2219                         break;
2220                 }
2221                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2222                         ret = -EINVAL;
2223                         break;
2224                 }
2225
2226                 tun->vnet_hdr_sz = vnet_hdr_sz;
2227                 break;
2228
2229         case TUNGETVNETLE:
2230                 le = !!(tun->flags & TUN_VNET_LE);
2231                 if (put_user(le, (int __user *)argp))
2232                         ret = -EFAULT;
2233                 break;
2234
2235         case TUNSETVNETLE:
2236                 if (get_user(le, (int __user *)argp)) {
2237                         ret = -EFAULT;
2238                         break;
2239                 }
2240                 if (le)
2241                         tun->flags |= TUN_VNET_LE;
2242                 else
2243                         tun->flags &= ~TUN_VNET_LE;
2244                 break;
2245
2246         case TUNGETVNETBE:
2247                 ret = tun_get_vnet_be(tun, argp);
2248                 break;
2249
2250         case TUNSETVNETBE:
2251                 ret = tun_set_vnet_be(tun, argp);
2252                 break;
2253
2254         case TUNATTACHFILTER:
2255                 /* Can be set only for TAPs */
2256                 ret = -EINVAL;
2257                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2258                         break;
2259                 ret = -EFAULT;
2260                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2261                         break;
2262
2263                 ret = tun_attach_filter(tun);
2264                 break;
2265
2266         case TUNDETACHFILTER:
2267                 /* Can be set only for TAPs */
2268                 ret = -EINVAL;
2269                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2270                         break;
2271                 ret = 0;
2272                 tun_detach_filter(tun, tun->numqueues);
2273                 break;
2274
2275         case TUNGETFILTER:
2276                 ret = -EINVAL;
2277                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2278                         break;
2279                 ret = -EFAULT;
2280                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2281                         break;
2282                 ret = 0;
2283                 break;
2284
2285         default:
2286                 ret = -EINVAL;
2287                 break;
2288         }
2289
2290 unlock:
2291         rtnl_unlock();
2292         if (tun)
2293                 tun_put(tun);
2294         return ret;
2295 }
2296
2297 static long tun_chr_ioctl(struct file *file,
2298                           unsigned int cmd, unsigned long arg)
2299 {
2300         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2301 }
2302
2303 #ifdef CONFIG_COMPAT
2304 static long tun_chr_compat_ioctl(struct file *file,
2305                          unsigned int cmd, unsigned long arg)
2306 {
2307         switch (cmd) {
2308         case TUNSETIFF:
2309         case TUNGETIFF:
2310         case TUNSETTXFILTER:
2311         case TUNGETSNDBUF:
2312         case TUNSETSNDBUF:
2313         case SIOCGIFHWADDR:
2314         case SIOCSIFHWADDR:
2315                 arg = (unsigned long)compat_ptr(arg);
2316                 break;
2317         default:
2318                 arg = (compat_ulong_t)arg;
2319                 break;
2320         }
2321
2322         /*
2323          * compat_ifreq is shorter than ifreq, so we must not access beyond
2324          * the end of that structure. All fields that are used in this
2325          * driver are compatible though, we don't need to convert the
2326          * contents.
2327          */
2328         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2329 }
2330 #endif /* CONFIG_COMPAT */
2331
2332 static int tun_chr_fasync(int fd, struct file *file, int on)
2333 {
2334         struct tun_file *tfile = file->private_data;
2335         int ret;
2336
2337         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2338                 goto out;
2339
2340         if (on) {
2341                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2342                 tfile->flags |= TUN_FASYNC;
2343         } else
2344                 tfile->flags &= ~TUN_FASYNC;
2345         ret = 0;
2346 out:
2347         return ret;
2348 }
2349
2350 static int tun_chr_open(struct inode *inode, struct file * file)
2351 {
2352         struct net *net = current->nsproxy->net_ns;
2353         struct tun_file *tfile;
2354
2355         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2356
2357         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2358                                             &tun_proto, 0);
2359         if (!tfile)
2360                 return -ENOMEM;
2361         RCU_INIT_POINTER(tfile->tun, NULL);
2362         tfile->flags = 0;
2363         tfile->ifindex = 0;
2364
2365         init_waitqueue_head(&tfile->wq.wait);
2366         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2367
2368         tfile->socket.file = file;
2369         tfile->socket.ops = &tun_socket_ops;
2370
2371         sock_init_data(&tfile->socket, &tfile->sk);
2372
2373         tfile->sk.sk_write_space = tun_sock_write_space;
2374         tfile->sk.sk_sndbuf = INT_MAX;
2375
2376         file->private_data = tfile;
2377         INIT_LIST_HEAD(&tfile->next);
2378
2379         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2380
2381         memset(&tfile->tx_array, 0, sizeof(tfile->tx_array));
2382
2383         return 0;
2384 }
2385
2386 static int tun_chr_close(struct inode *inode, struct file *file)
2387 {
2388         struct tun_file *tfile = file->private_data;
2389
2390         tun_detach(tfile, true);
2391
2392         return 0;
2393 }
2394
2395 #ifdef CONFIG_PROC_FS
2396 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2397 {
2398         struct tun_struct *tun;
2399         struct ifreq ifr;
2400
2401         memset(&ifr, 0, sizeof(ifr));
2402
2403         rtnl_lock();
2404         tun = tun_get(f);
2405         if (tun)
2406                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2407         rtnl_unlock();
2408
2409         if (tun)
2410                 tun_put(tun);
2411
2412         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2413 }
2414 #endif
2415
2416 static const struct file_operations tun_fops = {
2417         .owner  = THIS_MODULE,
2418         .llseek = no_llseek,
2419         .read_iter  = tun_chr_read_iter,
2420         .write_iter = tun_chr_write_iter,
2421         .poll   = tun_chr_poll,
2422         .unlocked_ioctl = tun_chr_ioctl,
2423 #ifdef CONFIG_COMPAT
2424         .compat_ioctl = tun_chr_compat_ioctl,
2425 #endif
2426         .open   = tun_chr_open,
2427         .release = tun_chr_close,
2428         .fasync = tun_chr_fasync,
2429 #ifdef CONFIG_PROC_FS
2430         .show_fdinfo = tun_chr_show_fdinfo,
2431 #endif
2432 };
2433
2434 static struct miscdevice tun_miscdev = {
2435         .minor = TUN_MINOR,
2436         .name = "tun",
2437         .nodename = "net/tun",
2438         .fops = &tun_fops,
2439 };
2440
2441 /* ethtool interface */
2442
2443 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2444 {
2445         cmd->supported          = 0;
2446         cmd->advertising        = 0;
2447         ethtool_cmd_speed_set(cmd, SPEED_10);
2448         cmd->duplex             = DUPLEX_FULL;
2449         cmd->port               = PORT_TP;
2450         cmd->phy_address        = 0;
2451         cmd->transceiver        = XCVR_INTERNAL;
2452         cmd->autoneg            = AUTONEG_DISABLE;
2453         cmd->maxtxpkt           = 0;
2454         cmd->maxrxpkt           = 0;
2455         return 0;
2456 }
2457
2458 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2459 {
2460         struct tun_struct *tun = netdev_priv(dev);
2461
2462         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2463         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2464
2465         switch (tun->flags & TUN_TYPE_MASK) {
2466         case IFF_TUN:
2467                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2468                 break;
2469         case IFF_TAP:
2470                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2471                 break;
2472         }
2473 }
2474
2475 static u32 tun_get_msglevel(struct net_device *dev)
2476 {
2477 #ifdef TUN_DEBUG
2478         struct tun_struct *tun = netdev_priv(dev);
2479         return tun->debug;
2480 #else
2481         return -EOPNOTSUPP;
2482 #endif
2483 }
2484
2485 static void tun_set_msglevel(struct net_device *dev, u32 value)
2486 {
2487 #ifdef TUN_DEBUG
2488         struct tun_struct *tun = netdev_priv(dev);
2489         tun->debug = value;
2490 #endif
2491 }
2492
2493 static const struct ethtool_ops tun_ethtool_ops = {
2494         .get_settings   = tun_get_settings,
2495         .get_drvinfo    = tun_get_drvinfo,
2496         .get_msglevel   = tun_get_msglevel,
2497         .set_msglevel   = tun_set_msglevel,
2498         .get_link       = ethtool_op_get_link,
2499         .get_ts_info    = ethtool_op_get_ts_info,
2500 };
2501
2502 static int tun_queue_resize(struct tun_struct *tun)
2503 {
2504         struct net_device *dev = tun->dev;
2505         struct tun_file *tfile;
2506         struct skb_array **arrays;
2507         int n = tun->numqueues + tun->numdisabled;
2508         int ret, i;
2509
2510         arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL);
2511         if (!arrays)
2512                 return -ENOMEM;
2513
2514         for (i = 0; i < tun->numqueues; i++) {
2515                 tfile = rtnl_dereference(tun->tfiles[i]);
2516                 arrays[i] = &tfile->tx_array;
2517         }
2518         list_for_each_entry(tfile, &tun->disabled, next)
2519                 arrays[i++] = &tfile->tx_array;
2520
2521         ret = skb_array_resize_multiple(arrays, n,
2522                                         dev->tx_queue_len, GFP_KERNEL);
2523
2524         kfree(arrays);
2525         return ret;
2526 }
2527
2528 static int tun_device_event(struct notifier_block *unused,
2529                             unsigned long event, void *ptr)
2530 {
2531         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2532         struct tun_struct *tun = netdev_priv(dev);
2533         int i;
2534
2535         if (dev->rtnl_link_ops != &tun_link_ops)
2536                 return NOTIFY_DONE;
2537
2538         switch (event) {
2539         case NETDEV_CHANGE_TX_QUEUE_LEN:
2540                 if (tun_queue_resize(tun))
2541                         return NOTIFY_BAD;
2542                 break;
2543         case NETDEV_UP:
2544                 for (i = 0; i < tun->numqueues; i++) {
2545                         struct tun_file *tfile;
2546
2547                         tfile = rtnl_dereference(tun->tfiles[i]);
2548                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
2549                 }
2550                 break;
2551         default:
2552                 break;
2553         }
2554
2555         return NOTIFY_DONE;
2556 }
2557
2558 static struct notifier_block tun_notifier_block __read_mostly = {
2559         .notifier_call  = tun_device_event,
2560 };
2561
2562 static int __init tun_init(void)
2563 {
2564         int ret = 0;
2565
2566         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2567         pr_info("%s\n", DRV_COPYRIGHT);
2568
2569         ret = rtnl_link_register(&tun_link_ops);
2570         if (ret) {
2571                 pr_err("Can't register link_ops\n");
2572                 goto err_linkops;
2573         }
2574
2575         ret = misc_register(&tun_miscdev);
2576         if (ret) {
2577                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2578                 goto err_misc;
2579         }
2580
2581         register_netdevice_notifier(&tun_notifier_block);
2582         return  0;
2583 err_misc:
2584         rtnl_link_unregister(&tun_link_ops);
2585 err_linkops:
2586         return ret;
2587 }
2588
2589 static void tun_cleanup(void)
2590 {
2591         misc_deregister(&tun_miscdev);
2592         rtnl_link_unregister(&tun_link_ops);
2593         unregister_netdevice_notifier(&tun_notifier_block);
2594 }
2595
2596 /* Get an underlying socket object from tun file.  Returns error unless file is
2597  * attached to a device.  The returned object works like a packet socket, it
2598  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2599  * holding a reference to the file for as long as the socket is in use. */
2600 struct socket *tun_get_socket(struct file *file)
2601 {
2602         struct tun_file *tfile;
2603         if (file->f_op != &tun_fops)
2604                 return ERR_PTR(-EINVAL);
2605         tfile = file->private_data;
2606         if (!tfile)
2607                 return ERR_PTR(-EBADFD);
2608         return &tfile->socket;
2609 }
2610 EXPORT_SYMBOL_GPL(tun_get_socket);
2611
2612 module_init(tun_init);
2613 module_exit(tun_cleanup);
2614 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2615 MODULE_AUTHOR(DRV_COPYRIGHT);
2616 MODULE_LICENSE("GPL");
2617 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2618 MODULE_ALIAS("devname:net/tun");