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/sched/signal.h>
  48 #include <linux/major.h>
  49 #include <linux/slab.h>
  50 #include <linux/poll.h>
  51 #include <linux/fcntl.h>
  52 #include <linux/init.h>
  53 #include <linux/skbuff.h>
  54 #include <linux/netdevice.h>
  55 #include <linux/etherdevice.h>
  56 #include <linux/miscdevice.h>
  57 #include <linux/ethtool.h>
  58 #include <linux/rtnetlink.h>
  59 #include <linux/compat.h>
  60 #include <linux/if.h>
  61 #include <linux/if_arp.h>
  62 #include <linux/if_ether.h>
  63 #include <linux/if_tun.h>
  64 #include <linux/if_vlan.h>
  65 #include <linux/crc32.h>
  66 #include <linux/nsproxy.h>
  67 #include <linux/virtio_net.h>
  68 #include <linux/rcupdate.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <linux/seq_file.h>
  74 #include <linux/uio.h>
  75 #include <linux/skb_array.h>
  76 #include <linux/bpf.h>
  77 #include <linux/bpf_trace.h>
  78
  79 #include <linux/uaccess.h>
  80
  81 /* Uncomment to enable debugging */
  82 /* #define TUN_DEBUG 1 */
  83
  84 #ifdef TUN_DEBUG
  85 static int debug;
  86
  87 #define tun_debug(level, tun, fmt, args...)                     \
  88 do {                                                            \
  89         if (tun->debug)                                         \
  90                 netdev_printk(level, tun->dev, fmt, ##args);    \
  91 } while (0)
  92 #define DBG1(level, fmt, args...)                               \
  93 do {                                                            \
  94         if (debug == 2)                                         \
  95                 printk(level fmt, ##args);                      \
  96 } while (0)
  97 #else
  98 #define tun_debug(level, tun, fmt, args...)                     \
  99 do {                                                            \
 100         if (0)                                                  \
 101                 netdev_printk(level, tun->dev, fmt, ##args);    \
 102 } while (0)
 103 #define DBG1(level, fmt, args...)                               \
 104 do {                                                            \
 105         if (0)                                                  \
 106                 printk(level fmt, ##args);                      \
 107 } while (0)
 108 #endif
 109
 110 #define TUN_HEADROOM 256
 111 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 112
 113 /* TUN device flags */
 114
 115 /* IFF_ATTACH_QUEUE is never stored in device flags,
 116  * overload it to mean fasync when stored there.
 117  */
 118 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 119 /* High bits in flags field are unused. */
 120 #define TUN_VNET_LE     0x80000000
 121 #define TUN_VNET_BE     0x40000000
 122
 123 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 124                       IFF_MULTI_QUEUE)
 125 #define GOODCOPY_LEN 128
 126
 127 #define FLT_EXACT_COUNT 8
 128 struct tap_filter {
 129         unsigned int    count;    /* Number of addrs. Zero means disabled */
 130         u32             mask[2];  /* Mask of the hashed addrs */
 131         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 132 };
 133
 134 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 135  * to max number of VCPUs in guest. */
 136 #define MAX_TAP_QUEUES 256
 137 #define MAX_TAP_FLOWS  4096
 138
 139 #define TUN_FLOW_EXPIRE (3 * HZ)
 140
 141 struct tun_pcpu_stats {
 142         u64 rx_packets;
 143         u64 rx_bytes;
 144         u64 tx_packets;
 145         u64 tx_bytes;
 146         struct u64_stats_sync syncp;
 147         u32 rx_dropped;
 148         u32 tx_dropped;
 149         u32 rx_frame_errors;
 150 };
 151
 152 /* A tun_file connects an open character device to a tuntap netdevice. It
 153  * also contains all socket related structures (except sock_fprog and tap_filter)
 154  * to serve as one transmit queue for tuntap device. The sock_fprog and
 155  * tap_filter were kept in tun_struct since they were used for filtering for the
 156  * netdevice not for a specific queue (at least I didn't see the requirement for
 157  * this).
 158  *
 159  * RCU usage:
 160  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 161  * other can only be read while rcu_read_lock or rtnl_lock is held.
 162  */
 163 struct tun_file {
 164         struct sock sk;
 165         struct socket socket;
 166         struct socket_wq wq;
 167         struct tun_struct __rcu *tun;
 168         struct fasync_struct *fasync;
 169         /* only used for fasnyc */
 170         unsigned int flags;
 171         union {
 172                 u16 queue_index;
 173                 unsigned int ifindex;
 174         };
 175         struct list_head next;
 176         struct tun_struct *detached;
 177         struct skb_array tx_array;
 178 };
 179
 180 struct tun_flow_entry {
 181         struct hlist_node hash_link;
 182         struct rcu_head rcu;
 183         struct tun_struct *tun;
 184
 185         u32 rxhash;
 186         u32 rps_rxhash;
 187         int queue_index;
 188         unsigned long updated;
 189 };
 190
 191 #define TUN_NUM_FLOW_ENTRIES 1024
 192
 193 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 194  * device, socket filter, sndbuf and vnet header size were restore when the
 195  * file were attached to a persist device.
 196  */
 197 struct tun_struct {
 198         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 199         unsigned int            numqueues;
 200         unsigned int            flags;
 201         kuid_t                  owner;
 202         kgid_t                  group;
 203
 204         struct net_device       *dev;
 205         netdev_features_t       set_features;
 206 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 207                           NETIF_F_TSO6)
 208
 209         int                     align;
 210         int                     vnet_hdr_sz;
 211         int                     sndbuf;
 212         struct tap_filter       txflt;
 213         struct sock_fprog       fprog;
 214         /* protected by rtnl lock */
 215         bool                    filter_attached;
 216 #ifdef TUN_DEBUG
 217         int debug;
 218 #endif
 219         spinlock_t lock;
 220         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 221         struct timer_list flow_gc_timer;
 222         unsigned long ageing_time;
 223         unsigned int numdisabled;
 224         struct list_head disabled;
 225         void *security;
 226         u32 flow_count;
 227         u32 rx_batched;
 228         struct tun_pcpu_stats __percpu *pcpu_stats;
 229         struct bpf_prog __rcu *xdp_prog;
 230 };
 231
 232 #ifdef CONFIG_TUN_VNET_CROSS_LE
 233 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 234 {
 235         return tun->flags & TUN_VNET_BE ? false :
 236                 virtio_legacy_is_little_endian();
 237 }
 238
 239 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 240 {
 241         int be = !!(tun->flags & TUN_VNET_BE);
 242
 243         if (put_user(be, argp))
 244                 return -EFAULT;
 245
 246         return 0;
 247 }
 248
 249 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 250 {
 251         int be;
 252
 253         if (get_user(be, argp))
 254                 return -EFAULT;
 255
 256         if (be)
 257                 tun->flags |= TUN_VNET_BE;
 258         else
 259                 tun->flags &= ~TUN_VNET_BE;
 260
 261         return 0;
 262 }
 263 #else
 264 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 265 {
 266         return virtio_legacy_is_little_endian();
 267 }
 268
 269 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 270 {
 271         return -EINVAL;
 272 }
 273
 274 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 275 {
 276         return -EINVAL;
 277 }
 278 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 279
 280 static inline bool tun_is_little_endian(struct tun_struct *tun)
 281 {
 282         return tun->flags & TUN_VNET_LE ||
 283                 tun_legacy_is_little_endian(tun);
 284 }
 285
 286 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 287 {
 288         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 289 }
 290
 291 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 292 {
 293         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 294 }
 295
 296 static inline u32 tun_hashfn(u32 rxhash)
 297 {
 298         return rxhash & 0x3ff;
 299 }
 300
 301 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 302 {
 303         struct tun_flow_entry *e;
 304
 305         hlist_for_each_entry_rcu(e, head, hash_link) {
 306                 if (e->rxhash == rxhash)
 307                         return e;
 308         }
 309         return NULL;
 310 }
 311
 312 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 313                                               struct hlist_head *head,
 314                                               u32 rxhash, u16 queue_index)
 315 {
 316         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 317
 318         if (e) {
 319                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 320                           rxhash, queue_index);
 321                 e->updated = jiffies;
 322                 e->rxhash = rxhash;
 323                 e->rps_rxhash = 0;
 324                 e->queue_index = queue_index;
 325                 e->tun = tun;
 326                 hlist_add_head_rcu(&e->hash_link, head);
 327                 ++tun->flow_count;
 328         }
 329         return e;
 330 }
 331
 332 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 333 {
 334         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 335                   e->rxhash, e->queue_index);
 336         hlist_del_rcu(&e->hash_link);
 337         kfree_rcu(e, rcu);
 338         --tun->flow_count;
 339 }
 340
 341 static void tun_flow_flush(struct tun_struct *tun)
 342 {
 343         int i;
 344
 345         spin_lock_bh(&tun->lock);
 346         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 347                 struct tun_flow_entry *e;
 348                 struct hlist_node *n;
 349
 350                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 351                         tun_flow_delete(tun, e);
 352         }
 353         spin_unlock_bh(&tun->lock);
 354 }
 355
 356 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 357 {
 358         int i;
 359
 360         spin_lock_bh(&tun->lock);
 361         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 362                 struct tun_flow_entry *e;
 363                 struct hlist_node *n;
 364
 365                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 366                         if (e->queue_index == queue_index)
 367                                 tun_flow_delete(tun, e);
 368                 }
 369         }
 370         spin_unlock_bh(&tun->lock);
 371 }
 372
 373 static void tun_flow_cleanup(unsigned long data)
 374 {
 375         struct tun_struct *tun = (struct tun_struct *)data;
 376         unsigned long delay = tun->ageing_time;
 377         unsigned long next_timer = jiffies + delay;
 378         unsigned long count = 0;
 379         int i;
 380
 381         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 382
 383         spin_lock_bh(&tun->lock);
 384         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 385                 struct tun_flow_entry *e;
 386                 struct hlist_node *n;
 387
 388                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 389                         unsigned long this_timer;
 390                         count++;
 391                         this_timer = e->updated + delay;
 392                         if (time_before_eq(this_timer, jiffies))
 393                                 tun_flow_delete(tun, e);
 394                         else if (time_before(this_timer, next_timer))
 395                                 next_timer = this_timer;
 396                 }
 397         }
 398
 399         if (count)
 400                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 401         spin_unlock_bh(&tun->lock);
 402 }
 403
 404 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 405                             struct tun_file *tfile)
 406 {
 407         struct hlist_head *head;
 408         struct tun_flow_entry *e;
 409         unsigned long delay = tun->ageing_time;
 410         u16 queue_index = tfile->queue_index;
 411
 412         if (!rxhash)
 413                 return;
 414         else
 415                 head = &tun->flows[tun_hashfn(rxhash)];
 416
 417         rcu_read_lock();
 418
 419         /* We may get a very small possibility of OOO during switching, not
 420          * worth to optimize.*/
 421         if (tun->numqueues == 1 || tfile->detached)
 422                 goto unlock;
 423
 424         e = tun_flow_find(head, rxhash);
 425         if (likely(e)) {
 426                 /* TODO: keep queueing to old queue until it's empty? */
 427                 e->queue_index = queue_index;
 428                 e->updated = jiffies;
 429                 sock_rps_record_flow_hash(e->rps_rxhash);
 430         } else {
 431                 spin_lock_bh(&tun->lock);
 432                 if (!tun_flow_find(head, rxhash) &&
 433                     tun->flow_count < MAX_TAP_FLOWS)
 434                         tun_flow_create(tun, head, rxhash, queue_index);
 435
 436                 if (!timer_pending(&tun->flow_gc_timer))
 437                         mod_timer(&tun->flow_gc_timer,
 438                                   round_jiffies_up(jiffies + delay));
 439                 spin_unlock_bh(&tun->lock);
 440         }
 441
 442 unlock:
 443         rcu_read_unlock();
 444 }
 445
 446 /**
 447  * Save the hash received in the stack receive path and update the
 448  * flow_hash table accordingly.
 449  */
 450 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 451 {
 452         if (unlikely(e->rps_rxhash != hash))
 453                 e->rps_rxhash = hash;
 454 }
 455
 456 /* We try to identify a flow through its rxhash first. The reason that
 457  * we do not check rxq no. is because some cards(e.g 82599), chooses
 458  * the rxq based on the txq where the last packet of the flow comes. As
 459  * the userspace application move between processors, we may get a
 460  * different rxq no. here. If we could not get rxhash, then we would
 461  * hope the rxq no. may help here.
 462  */
 463 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 464                             void *accel_priv, select_queue_fallback_t fallback)
 465 {
 466         struct tun_struct *tun = netdev_priv(dev);
 467         struct tun_flow_entry *e;
 468         u32 txq = 0;
 469         u32 numqueues = 0;
 470
 471         rcu_read_lock();
 472         numqueues = ACCESS_ONCE(tun->numqueues);
 473
 474         txq = __skb_get_hash_symmetric(skb);
 475         if (txq) {
 476                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 477                 if (e) {
 478                         tun_flow_save_rps_rxhash(e, txq);
 479                         txq = e->queue_index;
 480                 } else
 481                         /* use multiply and shift instead of expensive divide */
 482                         txq = ((u64)txq * numqueues) >> 32;
 483         } else if (likely(skb_rx_queue_recorded(skb))) {
 484                 txq = skb_get_rx_queue(skb);
 485                 while (unlikely(txq >= numqueues))
 486                         txq -= numqueues;
 487         }
 488
 489         rcu_read_unlock();
 490         return txq;
 491 }
 492
 493 static inline bool tun_not_capable(struct tun_struct *tun)
 494 {
 495         const struct cred *cred = current_cred();
 496         struct net *net = dev_net(tun->dev);
 497
 498         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 499                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 500                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 501 }
 502
 503 static void tun_set_real_num_queues(struct tun_struct *tun)
 504 {
 505         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 506         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 507 }
 508
 509 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 510 {
 511         tfile->detached = tun;
 512         list_add_tail(&tfile->next, &tun->disabled);
 513         ++tun->numdisabled;
 514 }
 515
 516 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 517 {
 518         struct tun_struct *tun = tfile->detached;
 519
 520         tfile->detached = NULL;
 521         list_del_init(&tfile->next);
 522         --tun->numdisabled;
 523         return tun;
 524 }
 525
 526 static void tun_queue_purge(struct tun_file *tfile)
 527 {
 528         struct sk_buff *skb;
 529
 530         while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
 531                 kfree_skb(skb);
 532
 533         skb_queue_purge(&tfile->sk.sk_write_queue);
 534         skb_queue_purge(&tfile->sk.sk_error_queue);
 535 }
 536
 537 static void __tun_detach(struct tun_file *tfile, bool clean)
 538 {
 539         struct tun_file *ntfile;
 540         struct tun_struct *tun;
 541
 542         tun = rtnl_dereference(tfile->tun);
 543
 544         if (tun && !tfile->detached) {
 545                 u16 index = tfile->queue_index;
 546                 BUG_ON(index >= tun->numqueues);
 547
 548                 rcu_assign_pointer(tun->tfiles[index],
 549                                    tun->tfiles[tun->numqueues - 1]);
 550                 ntfile = rtnl_dereference(tun->tfiles[index]);
 551                 ntfile->queue_index = index;
 552
 553                 --tun->numqueues;
 554                 if (clean) {
 555                         RCU_INIT_POINTER(tfile->tun, NULL);
 556                         sock_put(&tfile->sk);
 557                 } else
 558                         tun_disable_queue(tun, tfile);
 559
 560                 synchronize_net();
 561                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 562                 /* Drop read queue */
 563                 tun_queue_purge(tfile);
 564                 tun_set_real_num_queues(tun);
 565         } else if (tfile->detached && clean) {
 566                 tun = tun_enable_queue(tfile);
 567                 sock_put(&tfile->sk);
 568         }
 569
 570         if (clean) {
 571                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 572                         netif_carrier_off(tun->dev);
 573
 574                         if (!(tun->flags & IFF_PERSIST) &&
 575                             tun->dev->reg_state == NETREG_REGISTERED)
 576                                 unregister_netdevice(tun->dev);
 577                 }
 578                 if (tun)
 579                         skb_array_cleanup(&tfile->tx_array);
 580                 sock_put(&tfile->sk);
 581         }
 582 }
 583
 584 static void tun_detach(struct tun_file *tfile, bool clean)
 585 {
 586         rtnl_lock();
 587         __tun_detach(tfile, clean);
 588         rtnl_unlock();
 589 }
 590
 591 static void tun_detach_all(struct net_device *dev)
 592 {
 593         struct tun_struct *tun = netdev_priv(dev);
 594         struct bpf_prog *xdp_prog = rtnl_dereference(tun->xdp_prog);
 595         struct tun_file *tfile, *tmp;
 596         int i, n = tun->numqueues;
 597
 598         for (i = 0; i < n; i++) {
 599                 tfile = rtnl_dereference(tun->tfiles[i]);
 600                 BUG_ON(!tfile);
 601                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 602                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 603                 RCU_INIT_POINTER(tfile->tun, NULL);
 604                 --tun->numqueues;
 605         }
 606         list_for_each_entry(tfile, &tun->disabled, next) {
 607                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 608                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 609                 RCU_INIT_POINTER(tfile->tun, NULL);
 610         }
 611         BUG_ON(tun->numqueues != 0);
 612
 613         synchronize_net();
 614         for (i = 0; i < n; i++) {
 615                 tfile = rtnl_dereference(tun->tfiles[i]);
 616                 /* Drop read queue */
 617                 tun_queue_purge(tfile);
 618                 sock_put(&tfile->sk);
 619         }
 620         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 621                 tun_enable_queue(tfile);
 622                 tun_queue_purge(tfile);
 623                 sock_put(&tfile->sk);
 624         }
 625         BUG_ON(tun->numdisabled != 0);
 626
 627         if (xdp_prog)
 628                 bpf_prog_put(xdp_prog);
 629
 630         if (tun->flags & IFF_PERSIST)
 631                 module_put(THIS_MODULE);
 632 }
 633
 634 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 635 {
 636         struct tun_file *tfile = file->private_data;
 637         struct net_device *dev = tun->dev;
 638         int err;
 639
 640         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 641         if (err < 0)
 642                 goto out;
 643
 644         err = -EINVAL;
 645         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 646                 goto out;
 647
 648         err = -EBUSY;
 649         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 650                 goto out;
 651
 652         err = -E2BIG;
 653         if (!tfile->detached &&
 654             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 655                 goto out;
 656
 657         err = 0;
 658
 659         /* Re-attach the filter to persist device */
 660         if (!skip_filter && (tun->filter_attached == true)) {
 661                 lock_sock(tfile->socket.sk);
 662                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 663                 release_sock(tfile->socket.sk);
 664                 if (!err)
 665                         goto out;
 666         }
 667
 668         if (!tfile->detached &&
 669             skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
 670                 err = -ENOMEM;
 671                 goto out;
 672         }
 673
 674         tfile->queue_index = tun->numqueues;
 675         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 676         rcu_assign_pointer(tfile->tun, tun);
 677         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 678         tun->numqueues++;
 679
 680         if (tfile->detached)
 681                 tun_enable_queue(tfile);
 682         else
 683                 sock_hold(&tfile->sk);
 684
 685         tun_set_real_num_queues(tun);
 686
 687         /* device is allowed to go away first, so no need to hold extra
 688          * refcnt.
 689          */
 690
 691 out:
 692         return err;
 693 }
 694
 695 static struct tun_struct *__tun_get(struct tun_file *tfile)
 696 {
 697         struct tun_struct *tun;
 698
 699         rcu_read_lock();
 700         tun = rcu_dereference(tfile->tun);
 701         if (tun)
 702                 dev_hold(tun->dev);
 703         rcu_read_unlock();
 704
 705         return tun;
 706 }
 707
 708 static struct tun_struct *tun_get(struct file *file)
 709 {
 710         return __tun_get(file->private_data);
 711 }
 712
 713 static void tun_put(struct tun_struct *tun)
 714 {
 715         dev_put(tun->dev);
 716 }
 717
 718 /* TAP filtering */
 719 static void addr_hash_set(u32 *mask, const u8 *addr)
 720 {
 721         int n = ether_crc(ETH_ALEN, addr) >> 26;
 722         mask[n >> 5] |= (1 << (n & 31));
 723 }
 724
 725 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 726 {
 727         int n = ether_crc(ETH_ALEN, addr) >> 26;
 728         return mask[n >> 5] & (1 << (n & 31));
 729 }
 730
 731 static int update_filter(struct tap_filter *filter, void __user *arg)
 732 {
 733         struct { u8 u[ETH_ALEN]; } *addr;
 734         struct tun_filter uf;
 735         int err, alen, n, nexact;
 736
 737         if (copy_from_user(&uf, arg, sizeof(uf)))
 738                 return -EFAULT;
 739
 740         if (!uf.count) {
 741                 /* Disabled */
 742                 filter->count = 0;
 743                 return 0;
 744         }
 745
 746         alen = ETH_ALEN * uf.count;
 747         addr = memdup_user(arg + sizeof(uf), alen);
 748         if (IS_ERR(addr))
 749                 return PTR_ERR(addr);
 750
 751         /* The filter is updated without holding any locks. Which is
 752          * perfectly safe. We disable it first and in the worst
 753          * case we'll accept a few undesired packets. */
 754         filter->count = 0;
 755         wmb();
 756
 757         /* Use first set of addresses as an exact filter */
 758         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 759                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 760
 761         nexact = n;
 762
 763         /* Remaining multicast addresses are hashed,
 764          * unicast will leave the filter disabled. */
 765         memset(filter->mask, 0, sizeof(filter->mask));
 766         for (; n < uf.count; n++) {
 767                 if (!is_multicast_ether_addr(addr[n].u)) {
 768                         err = 0; /* no filter */
 769                         goto free_addr;
 770                 }
 771                 addr_hash_set(filter->mask, addr[n].u);
 772         }
 773
 774         /* For ALLMULTI just set the mask to all ones.
 775          * This overrides the mask populated above. */
 776         if ((uf.flags & TUN_FLT_ALLMULTI))
 777                 memset(filter->mask, ~0, sizeof(filter->mask));
 778
 779         /* Now enable the filter */
 780         wmb();
 781         filter->count = nexact;
 782
 783         /* Return the number of exact filters */
 784         err = nexact;
 785 free_addr:
 786         kfree(addr);
 787         return err;
 788 }
 789
 790 /* Returns: 0 - drop, !=0 - accept */
 791 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 792 {
 793         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 794          * at this point. */
 795         struct ethhdr *eh = (struct ethhdr *) skb->data;
 796         int i;
 797
 798         /* Exact match */
 799         for (i = 0; i < filter->count; i++)
 800                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 801                         return 1;
 802
 803         /* Inexact match (multicast only) */
 804         if (is_multicast_ether_addr(eh->h_dest))
 805                 return addr_hash_test(filter->mask, eh->h_dest);
 806
 807         return 0;
 808 }
 809
 810 /*
 811  * Checks whether the packet is accepted or not.
 812  * Returns: 0 - drop, !=0 - accept
 813  */
 814 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 815 {
 816         if (!filter->count)
 817                 return 1;
 818
 819         return run_filter(filter, skb);
 820 }
 821
 822 /* Network device part of the driver */
 823
 824 static const struct ethtool_ops tun_ethtool_ops;
 825
 826 /* Net device detach from fd. */
 827 static void tun_net_uninit(struct net_device *dev)
 828 {
 829         tun_detach_all(dev);
 830 }
 831
 832 /* Net device open. */
 833 static int tun_net_open(struct net_device *dev)
 834 {
 835         struct tun_struct *tun = netdev_priv(dev);
 836         int i;
 837
 838         netif_tx_start_all_queues(dev);
 839
 840         for (i = 0; i < tun->numqueues; i++) {
 841                 struct tun_file *tfile;
 842
 843                 tfile = rtnl_dereference(tun->tfiles[i]);
 844                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
 845         }
 846
 847         return 0;
 848 }
 849
 850 /* Net device close. */
 851 static int tun_net_close(struct net_device *dev)
 852 {
 853         netif_tx_stop_all_queues(dev);
 854         return 0;
 855 }
 856
 857 /* Net device start xmit */
 858 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 859 {
 860         struct tun_struct *tun = netdev_priv(dev);
 861         int txq = skb->queue_mapping;
 862         struct tun_file *tfile;
 863         u32 numqueues = 0;
 864
 865         rcu_read_lock();
 866         tfile = rcu_dereference(tun->tfiles[txq]);
 867         numqueues = ACCESS_ONCE(tun->numqueues);
 868
 869         /* Drop packet if interface is not attached */
 870         if (txq >= numqueues)
 871                 goto drop;
 872
 873 #ifdef CONFIG_RPS
 874         if (numqueues == 1 && static_key_false(&rps_needed)) {
 875                 /* Select queue was not called for the skbuff, so we extract the
 876                  * RPS hash and save it into the flow_table here.
 877                  */
 878                 __u32 rxhash;
 879
 880                 rxhash = __skb_get_hash_symmetric(skb);
 881                 if (rxhash) {
 882                         struct tun_flow_entry *e;
 883                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 884                                         rxhash);
 885                         if (e)
 886                                 tun_flow_save_rps_rxhash(e, rxhash);
 887                 }
 888         }
 889 #endif
 890
 891         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 892
 893         BUG_ON(!tfile);
 894
 895         /* Drop if the filter does not like it.
 896          * This is a noop if the filter is disabled.
 897          * Filter can be enabled only for the TAP devices. */
 898         if (!check_filter(&tun->txflt, skb))
 899                 goto drop;
 900
 901         if (tfile->socket.sk->sk_filter &&
 902             sk_filter(tfile->socket.sk, skb))
 903                 goto drop;
 904
 905         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
 906                 goto drop;
 907
 908         skb_tx_timestamp(skb);
 909
 910         /* Orphan the skb - required as we might hang on to it
 911          * for indefinite time.
 912          */
 913         skb_orphan(skb);
 914
 915         nf_reset(skb);
 916
 917         if (skb_array_produce(&tfile->tx_array, skb))
 918                 goto drop;
 919
 920         /* Notify and wake up reader process */
 921         if (tfile->flags & TUN_FASYNC)
 922                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 923         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 924
 925         rcu_read_unlock();
 926         return NETDEV_TX_OK;
 927
 928 drop:
 929         this_cpu_inc(tun->pcpu_stats->tx_dropped);
 930         skb_tx_error(skb);
 931         kfree_skb(skb);
 932         rcu_read_unlock();
 933         return NET_XMIT_DROP;
 934 }
 935
 936 static void tun_net_mclist(struct net_device *dev)
 937 {
 938         /*
 939          * This callback is supposed to deal with mc filter in
 940          * _rx_ path and has nothing to do with the _tx_ path.
 941          * In rx path we always accept everything userspace gives us.
 942          */
 943 }
 944
 945 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 946         netdev_features_t features)
 947 {
 948         struct tun_struct *tun = netdev_priv(dev);
 949
 950         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 951 }
 952 #ifdef CONFIG_NET_POLL_CONTROLLER
 953 static void tun_poll_controller(struct net_device *dev)
 954 {
 955         /*
 956          * Tun only receives frames when:
 957          * 1) the char device endpoint gets data from user space
 958          * 2) the tun socket gets a sendmsg call from user space
 959          * Since both of those are synchronous operations, we are guaranteed
 960          * never to have pending data when we poll for it
 961          * so there is nothing to do here but return.
 962          * We need this though so netpoll recognizes us as an interface that
 963          * supports polling, which enables bridge devices in virt setups to
 964          * still use netconsole
 965          */
 966         return;
 967 }
 968 #endif
 969
 970 static void tun_set_headroom(struct net_device *dev, int new_hr)
 971 {
 972         struct tun_struct *tun = netdev_priv(dev);
 973
 974         if (new_hr < NET_SKB_PAD)
 975                 new_hr = NET_SKB_PAD;
 976
 977         tun->align = new_hr;
 978 }
 979
 980 static void
 981 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
 982 {
 983         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
 984         struct tun_struct *tun = netdev_priv(dev);
 985         struct tun_pcpu_stats *p;
 986         int i;
 987
 988         for_each_possible_cpu(i) {
 989                 u64 rxpackets, rxbytes, txpackets, txbytes;
 990                 unsigned int start;
 991
 992                 p = per_cpu_ptr(tun->pcpu_stats, i);
 993                 do {
 994                         start = u64_stats_fetch_begin(&p->syncp);
 995                         rxpackets       = p->rx_packets;
 996                         rxbytes         = p->rx_bytes;
 997                         txpackets       = p->tx_packets;
 998                         txbytes         = p->tx_bytes;
 999                 } while (u64_stats_fetch_retry(&p->syncp, start));
1000
1001                 stats->rx_packets       += rxpackets;
1002                 stats->rx_bytes         += rxbytes;
1003                 stats->tx_packets       += txpackets;
1004                 stats->tx_bytes         += txbytes;
1005
1006                 /* u32 counters */
1007                 rx_dropped      += p->rx_dropped;
1008                 rx_frame_errors += p->rx_frame_errors;
1009                 tx_dropped      += p->tx_dropped;
1010         }
1011         stats->rx_dropped  = rx_dropped;
1012         stats->rx_frame_errors = rx_frame_errors;
1013         stats->tx_dropped = tx_dropped;
1014 }
1015
1016 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1017                        struct netlink_ext_ack *extack)
1018 {
1019         struct tun_struct *tun = netdev_priv(dev);
1020         struct bpf_prog *old_prog;
1021
1022         old_prog = rtnl_dereference(tun->xdp_prog);
1023         rcu_assign_pointer(tun->xdp_prog, prog);
1024         if (old_prog)
1025                 bpf_prog_put(old_prog);
1026
1027         return 0;
1028 }
1029
1030 static u32 tun_xdp_query(struct net_device *dev)
1031 {
1032         struct tun_struct *tun = netdev_priv(dev);
1033         const struct bpf_prog *xdp_prog;
1034
1035         xdp_prog = rtnl_dereference(tun->xdp_prog);
1036         if (xdp_prog)
1037                 return xdp_prog->aux->id;
1038
1039         return 0;
1040 }
1041
1042 static int tun_xdp(struct net_device *dev, struct netdev_xdp *xdp)
1043 {
1044         switch (xdp->command) {
1045         case XDP_SETUP_PROG:
1046                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1047         case XDP_QUERY_PROG:
1048                 xdp->prog_id = tun_xdp_query(dev);
1049                 xdp->prog_attached = !!xdp->prog_id;
1050                 return 0;
1051         default:
1052                 return -EINVAL;
1053         }
1054 }
1055
1056 static const struct net_device_ops tun_netdev_ops = {
1057         .ndo_uninit             = tun_net_uninit,
1058         .ndo_open               = tun_net_open,
1059         .ndo_stop               = tun_net_close,
1060         .ndo_start_xmit         = tun_net_xmit,
1061         .ndo_fix_features       = tun_net_fix_features,
1062         .ndo_select_queue       = tun_select_queue,
1063 #ifdef CONFIG_NET_POLL_CONTROLLER
1064         .ndo_poll_controller    = tun_poll_controller,
1065 #endif
1066         .ndo_set_rx_headroom    = tun_set_headroom,
1067         .ndo_get_stats64        = tun_net_get_stats64,
1068 };
1069
1070 static const struct net_device_ops tap_netdev_ops = {
1071         .ndo_uninit             = tun_net_uninit,
1072         .ndo_open               = tun_net_open,
1073         .ndo_stop               = tun_net_close,
1074         .ndo_start_xmit         = tun_net_xmit,
1075         .ndo_fix_features       = tun_net_fix_features,
1076         .ndo_set_rx_mode        = tun_net_mclist,
1077         .ndo_set_mac_address    = eth_mac_addr,
1078         .ndo_validate_addr      = eth_validate_addr,
1079         .ndo_select_queue       = tun_select_queue,
1080 #ifdef CONFIG_NET_POLL_CONTROLLER
1081         .ndo_poll_controller    = tun_poll_controller,
1082 #endif
1083         .ndo_features_check     = passthru_features_check,
1084         .ndo_set_rx_headroom    = tun_set_headroom,
1085         .ndo_get_stats64        = tun_net_get_stats64,
1086         .ndo_xdp                = tun_xdp,
1087 };
1088
1089 static void tun_flow_init(struct tun_struct *tun)
1090 {
1091         int i;
1092
1093         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1094                 INIT_HLIST_HEAD(&tun->flows[i]);
1095
1096         tun->ageing_time = TUN_FLOW_EXPIRE;
1097         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1098         mod_timer(&tun->flow_gc_timer,
1099                   round_jiffies_up(jiffies + tun->ageing_time));
1100 }
1101
1102 static void tun_flow_uninit(struct tun_struct *tun)
1103 {
1104         del_timer_sync(&tun->flow_gc_timer);
1105         tun_flow_flush(tun);
1106 }
1107
1108 #define MIN_MTU 68
1109 #define MAX_MTU 65535
1110
1111 /* Initialize net device. */
1112 static void tun_net_init(struct net_device *dev)
1113 {
1114         struct tun_struct *tun = netdev_priv(dev);
1115
1116         switch (tun->flags & TUN_TYPE_MASK) {
1117         case IFF_TUN:
1118                 dev->netdev_ops = &tun_netdev_ops;
1119
1120                 /* Point-to-Point TUN Device */
1121                 dev->hard_header_len = 0;
1122                 dev->addr_len = 0;
1123                 dev->mtu = 1500;
1124
1125                 /* Zero header length */
1126                 dev->type = ARPHRD_NONE;
1127                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1128                 break;
1129
1130         case IFF_TAP:
1131                 dev->netdev_ops = &tap_netdev_ops;
1132                 /* Ethernet TAP Device */
1133                 ether_setup(dev);
1134                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1135                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1136
1137                 eth_hw_addr_random(dev);
1138
1139                 break;
1140         }
1141
1142         dev->min_mtu = MIN_MTU;
1143         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1144 }
1145
1146 /* Character device part */
1147
1148 /* Poll */
1149 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1150 {
1151         struct tun_file *tfile = file->private_data;
1152         struct tun_struct *tun = __tun_get(tfile);
1153         struct sock *sk;
1154         unsigned int mask = 0;
1155
1156         if (!tun)
1157                 return POLLERR;
1158
1159         sk = tfile->socket.sk;
1160
1161         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1162
1163         poll_wait(file, sk_sleep(sk), wait);
1164
1165         if (!skb_array_empty(&tfile->tx_array))
1166                 mask |= POLLIN | POLLRDNORM;
1167
1168         if (tun->dev->flags & IFF_UP &&
1169             (sock_writeable(sk) ||
1170              (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1171               sock_writeable(sk))))
1172                 mask |= POLLOUT | POLLWRNORM;
1173
1174         if (tun->dev->reg_state != NETREG_REGISTERED)
1175                 mask = POLLERR;
1176
1177         tun_put(tun);
1178         return mask;
1179 }
1180
1181 /* prepad is the amount to reserve at front.  len is length after that.
1182  * linear is a hint as to how much to copy (usually headers). */
1183 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1184                                      size_t prepad, size_t len,
1185                                      size_t linear, int noblock)
1186 {
1187         struct sock *sk = tfile->socket.sk;
1188         struct sk_buff *skb;
1189         int err;
1190
1191         /* Under a page?  Don't bother with paged skb. */
1192         if (prepad + len < PAGE_SIZE || !linear)
1193                 linear = len;
1194
1195         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1196                                    &err, 0);
1197         if (!skb)
1198                 return ERR_PTR(err);
1199
1200         skb_reserve(skb, prepad);
1201         skb_put(skb, linear);
1202         skb->data_len = len - linear;
1203         skb->len += len - linear;
1204
1205         return skb;
1206 }
1207
1208 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1209                            struct sk_buff *skb, int more)
1210 {
1211         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1212         struct sk_buff_head process_queue;
1213         u32 rx_batched = tun->rx_batched;
1214         bool rcv = false;
1215
1216         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1217                 local_bh_disable();
1218                 netif_receive_skb(skb);
1219                 local_bh_enable();
1220                 return;
1221         }
1222
1223         spin_lock(&queue->lock);
1224         if (!more || skb_queue_len(queue) == rx_batched) {
1225                 __skb_queue_head_init(&process_queue);
1226                 skb_queue_splice_tail_init(queue, &process_queue);
1227                 rcv = true;
1228         } else {
1229                 __skb_queue_tail(queue, skb);
1230         }
1231         spin_unlock(&queue->lock);
1232
1233         if (rcv) {
1234                 struct sk_buff *nskb;
1235
1236                 local_bh_disable();
1237                 while ((nskb = __skb_dequeue(&process_queue)))
1238                         netif_receive_skb(nskb);
1239                 netif_receive_skb(skb);
1240                 local_bh_enable();
1241         }
1242 }
1243
1244 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1245                               int len, int noblock, bool zerocopy)
1246 {
1247         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1248                 return false;
1249
1250         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1251                 return false;
1252
1253         if (!noblock)
1254                 return false;
1255
1256         if (zerocopy)
1257                 return false;
1258
1259         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1260             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1261                 return false;
1262
1263         return true;
1264 }
1265
1266 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1267                                      struct tun_file *tfile,
1268                                      struct iov_iter *from,
1269                                      struct virtio_net_hdr *hdr,
1270                                      int len, int *skb_xdp)
1271 {
1272         struct page_frag *alloc_frag = &current->task_frag;
1273         struct sk_buff *skb;
1274         struct bpf_prog *xdp_prog;
1275         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1276         unsigned int delta = 0;
1277         char *buf;
1278         size_t copied;
1279         bool xdp_xmit = false;
1280         int err, pad = TUN_RX_PAD;
1281
1282         rcu_read_lock();
1283         xdp_prog = rcu_dereference(tun->xdp_prog);
1284         if (xdp_prog)
1285                 pad += TUN_HEADROOM;
1286         buflen += SKB_DATA_ALIGN(len + pad);
1287         rcu_read_unlock();
1288
1289         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1290                 return ERR_PTR(-ENOMEM);
1291
1292         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1293         copied = copy_page_from_iter(alloc_frag->page,
1294                                      alloc_frag->offset + pad,
1295                                      len, from);
1296         if (copied != len)
1297                 return ERR_PTR(-EFAULT);
1298
1299         /* There's a small window that XDP may be set after the check
1300          * of xdp_prog above, this should be rare and for simplicity
1301          * we do XDP on skb in case the headroom is not enough.
1302          */
1303         if (hdr->gso_type || !xdp_prog)
1304                 *skb_xdp = 1;
1305         else
1306                 *skb_xdp = 0;
1307
1308         rcu_read_lock();
1309         xdp_prog = rcu_dereference(tun->xdp_prog);
1310         if (xdp_prog && !*skb_xdp) {
1311                 struct xdp_buff xdp;
1312                 void *orig_data;
1313                 u32 act;
1314
1315                 xdp.data_hard_start = buf;
1316                 xdp.data = buf + pad;
1317                 xdp.data_end = xdp.data + len;
1318                 orig_data = xdp.data;
1319                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1320
1321                 switch (act) {
1322                 case XDP_REDIRECT:
1323                         get_page(alloc_frag->page);
1324                         alloc_frag->offset += buflen;
1325                         err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1326                         if (err)
1327                                 goto err_redirect;
1328                         return NULL;
1329                 case XDP_TX:
1330                         xdp_xmit = true;
1331                         /* fall through */
1332                 case XDP_PASS:
1333                         delta = orig_data - xdp.data;
1334                         break;
1335                 default:
1336                         bpf_warn_invalid_xdp_action(act);
1337                         /* fall through */
1338                 case XDP_ABORTED:
1339                         trace_xdp_exception(tun->dev, xdp_prog, act);
1340                         /* fall through */
1341                 case XDP_DROP:
1342                         goto err_xdp;
1343                 }
1344         }
1345
1346         skb = build_skb(buf, buflen);
1347         if (!skb) {
1348                 rcu_read_unlock();
1349                 return ERR_PTR(-ENOMEM);
1350         }
1351
1352         skb_reserve(skb, pad - delta);
1353         skb_put(skb, len + delta);
1354         get_page(alloc_frag->page);
1355         alloc_frag->offset += buflen;
1356
1357         if (xdp_xmit) {
1358                 skb->dev = tun->dev;
1359                 generic_xdp_tx(skb, xdp_prog);
1360                 rcu_read_lock();
1361                 return NULL;
1362         }
1363
1364         rcu_read_unlock();
1365
1366         return skb;
1367
1368 err_redirect:
1369         put_page(alloc_frag->page);
1370 err_xdp:
1371         rcu_read_unlock();
1372         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1373         return NULL;
1374 }
1375
1376 /* Get packet from user space buffer */
1377 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1378                             void *msg_control, struct iov_iter *from,
1379                             int noblock, bool more)
1380 {
1381         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1382         struct sk_buff *skb;
1383         size_t total_len = iov_iter_count(from);
1384         size_t len = total_len, align = tun->align, linear;
1385         struct virtio_net_hdr gso = { 0 };
1386         struct tun_pcpu_stats *stats;
1387         int good_linear;
1388         int copylen;
1389         bool zerocopy = false;
1390         int err;
1391         u32 rxhash;
1392         int skb_xdp = 1;
1393
1394         if (!(tun->dev->flags & IFF_UP))
1395                 return -EIO;
1396
1397         if (!(tun->flags & IFF_NO_PI)) {
1398                 if (len < sizeof(pi))
1399                         return -EINVAL;
1400                 len -= sizeof(pi);
1401
1402                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1403                         return -EFAULT;
1404         }
1405
1406         if (tun->flags & IFF_VNET_HDR) {
1407                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1408
1409                 if (len < vnet_hdr_sz)
1410                         return -EINVAL;
1411                 len -= vnet_hdr_sz;
1412
1413                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1414                         return -EFAULT;
1415
1416                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1417                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1418                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1419
1420                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1421                         return -EINVAL;
1422                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1423         }
1424
1425         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1426                 align += NET_IP_ALIGN;
1427                 if (unlikely(len < ETH_HLEN ||
1428                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1429                         return -EINVAL;
1430         }
1431
1432         good_linear = SKB_MAX_HEAD(align);
1433
1434         if (msg_control) {
1435                 struct iov_iter i = *from;
1436
1437                 /* There are 256 bytes to be copied in skb, so there is
1438                  * enough room for skb expand head in case it is used.
1439                  * The rest of the buffer is mapped from userspace.
1440                  */
1441                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1442                 if (copylen > good_linear)
1443                         copylen = good_linear;
1444                 linear = copylen;
1445                 iov_iter_advance(&i, copylen);
1446                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1447                         zerocopy = true;
1448         }
1449
1450         if (tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1451                 /* For the packet that is not easy to be processed
1452                  * (e.g gso or jumbo packet), we will do it at after
1453                  * skb was created with generic XDP routine.
1454                  */
1455                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1456                 if (IS_ERR(skb)) {
1457                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1458                         return PTR_ERR(skb);
1459                 }
1460                 if (!skb)
1461                         return total_len;
1462         } else {
1463                 if (!zerocopy) {
1464                         copylen = len;
1465                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1466                                 linear = good_linear;
1467                         else
1468                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1469                 }
1470
1471                 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1472                 if (IS_ERR(skb)) {
1473                         if (PTR_ERR(skb) != -EAGAIN)
1474                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1475                         return PTR_ERR(skb);
1476                 }
1477
1478                 if (zerocopy)
1479                         err = zerocopy_sg_from_iter(skb, from);
1480                 else
1481                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1482
1483                 if (err) {
1484                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1485                         kfree_skb(skb);
1486                         return -EFAULT;
1487                 }
1488         }
1489
1490         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1491                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1492                 kfree_skb(skb);
1493                 return -EINVAL;
1494         }
1495
1496         switch (tun->flags & TUN_TYPE_MASK) {
1497         case IFF_TUN:
1498                 if (tun->flags & IFF_NO_PI) {
1499                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1500
1501                         switch (ip_version) {
1502                         case 4:
1503                                 pi.proto = htons(ETH_P_IP);
1504                                 break;
1505                         case 6:
1506                                 pi.proto = htons(ETH_P_IPV6);
1507                                 break;
1508                         default:
1509                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1510                                 kfree_skb(skb);
1511                                 return -EINVAL;
1512                         }
1513                 }
1514
1515                 skb_reset_mac_header(skb);
1516                 skb->protocol = pi.proto;
1517                 skb->dev = tun->dev;
1518                 break;
1519         case IFF_TAP:
1520                 skb->protocol = eth_type_trans(skb, tun->dev);
1521                 break;
1522         }
1523
1524         /* copy skb_ubuf_info for callback when skb has no error */
1525         if (zerocopy) {
1526                 skb_shinfo(skb)->destructor_arg = msg_control;
1527                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1528                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1529         } else if (msg_control) {
1530                 struct ubuf_info *uarg = msg_control;
1531                 uarg->callback(uarg, false);
1532         }
1533
1534         skb_reset_network_header(skb);
1535         skb_probe_transport_header(skb, 0);
1536
1537         if (skb_xdp) {
1538                 struct bpf_prog *xdp_prog;
1539                 int ret;
1540
1541                 rcu_read_lock();
1542                 xdp_prog = rcu_dereference(tun->xdp_prog);
1543                 if (xdp_prog) {
1544                         ret = do_xdp_generic(xdp_prog, skb);
1545                         if (ret != XDP_PASS) {
1546                                 rcu_read_unlock();
1547                                 return total_len;
1548                         }
1549                 }
1550                 rcu_read_unlock();
1551         }
1552
1553         rxhash = __skb_get_hash_symmetric(skb);
1554 #ifndef CONFIG_4KSTACKS
1555         tun_rx_batched(tun, tfile, skb, more);
1556 #else
1557         netif_rx_ni(skb);
1558 #endif
1559
1560         stats = get_cpu_ptr(tun->pcpu_stats);
1561         u64_stats_update_begin(&stats->syncp);
1562         stats->rx_packets++;
1563         stats->rx_bytes += len;
1564         u64_stats_update_end(&stats->syncp);
1565         put_cpu_ptr(stats);
1566
1567         tun_flow_update(tun, rxhash, tfile);
1568         return total_len;
1569 }
1570
1571 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1572 {
1573         struct file *file = iocb->ki_filp;
1574         struct tun_struct *tun = tun_get(file);
1575         struct tun_file *tfile = file->private_data;
1576         ssize_t result;
1577
1578         if (!tun)
1579                 return -EBADFD;
1580
1581         result = tun_get_user(tun, tfile, NULL, from,
1582                               file->f_flags & O_NONBLOCK, false);
1583
1584         tun_put(tun);
1585         return result;
1586 }
1587
1588 /* Put packet to the user space buffer */
1589 static ssize_t tun_put_user(struct tun_struct *tun,
1590                             struct tun_file *tfile,
1591                             struct sk_buff *skb,
1592                             struct iov_iter *iter)
1593 {
1594         struct tun_pi pi = { 0, skb->protocol };
1595         struct tun_pcpu_stats *stats;
1596         ssize_t total;
1597         int vlan_offset = 0;
1598         int vlan_hlen = 0;
1599         int vnet_hdr_sz = 0;
1600
1601         if (skb_vlan_tag_present(skb))
1602                 vlan_hlen = VLAN_HLEN;
1603
1604         if (tun->flags & IFF_VNET_HDR)
1605                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1606
1607         total = skb->len + vlan_hlen + vnet_hdr_sz;
1608
1609         if (!(tun->flags & IFF_NO_PI)) {
1610                 if (iov_iter_count(iter) < sizeof(pi))
1611                         return -EINVAL;
1612
1613                 total += sizeof(pi);
1614                 if (iov_iter_count(iter) < total) {
1615                         /* Packet will be striped */
1616                         pi.flags |= TUN_PKT_STRIP;
1617                 }
1618
1619                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1620                         return -EFAULT;
1621         }
1622
1623         if (vnet_hdr_sz) {
1624                 struct virtio_net_hdr gso;
1625
1626                 if (iov_iter_count(iter) < vnet_hdr_sz)
1627                         return -EINVAL;
1628
1629                 if (virtio_net_hdr_from_skb(skb, &gso,
1630                                             tun_is_little_endian(tun), true)) {
1631                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1632                         pr_err("unexpected GSO type: "
1633                                "0x%x, gso_size %d, hdr_len %d\n",
1634                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1635                                tun16_to_cpu(tun, gso.hdr_len));
1636                         print_hex_dump(KERN_ERR, "tun: ",
1637                                        DUMP_PREFIX_NONE,
1638                                        16, 1, skb->head,
1639                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1640                         WARN_ON_ONCE(1);
1641                         return -EINVAL;
1642                 }
1643
1644                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1645                         return -EFAULT;
1646
1647                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1648         }
1649
1650         if (vlan_hlen) {
1651                 int ret;
1652                 struct {
1653                         __be16 h_vlan_proto;
1654                         __be16 h_vlan_TCI;
1655                 } veth;
1656
1657                 veth.h_vlan_proto = skb->vlan_proto;
1658                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1659
1660                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1661
1662                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1663                 if (ret || !iov_iter_count(iter))
1664                         goto done;
1665
1666                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1667                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1668                         goto done;
1669         }
1670
1671         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1672
1673 done:
1674         /* caller is in process context, */
1675         stats = get_cpu_ptr(tun->pcpu_stats);
1676         u64_stats_update_begin(&stats->syncp);
1677         stats->tx_packets++;
1678         stats->tx_bytes += skb->len + vlan_hlen;
1679         u64_stats_update_end(&stats->syncp);
1680         put_cpu_ptr(tun->pcpu_stats);
1681
1682         return total;
1683 }
1684
1685 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1686                                      int *err)
1687 {
1688         DECLARE_WAITQUEUE(wait, current);
1689         struct sk_buff *skb = NULL;
1690         int error = 0;
1691
1692         skb = skb_array_consume(&tfile->tx_array);
1693         if (skb)
1694                 goto out;
1695         if (noblock) {
1696                 error = -EAGAIN;
1697                 goto out;
1698         }
1699
1700         add_wait_queue(&tfile->wq.wait, &wait);
1701         current->state = TASK_INTERRUPTIBLE;
1702
1703         while (1) {
1704                 skb = skb_array_consume(&tfile->tx_array);
1705                 if (skb)
1706                         break;
1707                 if (signal_pending(current)) {
1708                         error = -ERESTARTSYS;
1709                         break;
1710                 }
1711                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1712                         error = -EFAULT;
1713                         break;
1714                 }
1715
1716                 schedule();
1717         }
1718
1719         current->state = TASK_RUNNING;
1720         remove_wait_queue(&tfile->wq.wait, &wait);
1721
1722 out:
1723         *err = error;
1724         return skb;
1725 }
1726
1727 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1728                            struct iov_iter *to,
1729                            int noblock, struct sk_buff *skb)
1730 {
1731         ssize_t ret;
1732         int err;
1733
1734         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1735
1736         if (!iov_iter_count(to))
1737                 return 0;
1738
1739         if (!skb) {
1740                 /* Read frames from ring */
1741                 skb = tun_ring_recv(tfile, noblock, &err);
1742                 if (!skb)
1743                         return err;
1744         }
1745
1746         ret = tun_put_user(tun, tfile, skb, to);
1747         if (unlikely(ret < 0))
1748                 kfree_skb(skb);
1749         else
1750                 consume_skb(skb);
1751
1752         return ret;
1753 }
1754
1755 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1756 {
1757         struct file *file = iocb->ki_filp;
1758         struct tun_file *tfile = file->private_data;
1759         struct tun_struct *tun = __tun_get(tfile);
1760         ssize_t len = iov_iter_count(to), ret;
1761
1762         if (!tun)
1763                 return -EBADFD;
1764         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
1765         ret = min_t(ssize_t, ret, len);
1766         if (ret > 0)
1767                 iocb->ki_pos = ret;
1768         tun_put(tun);
1769         return ret;
1770 }
1771
1772 static void tun_free_netdev(struct net_device *dev)
1773 {
1774         struct tun_struct *tun = netdev_priv(dev);
1775
1776         BUG_ON(!(list_empty(&tun->disabled)));
1777         free_percpu(tun->pcpu_stats);
1778         tun_flow_uninit(tun);
1779         security_tun_dev_free_security(tun->security);
1780 }
1781
1782 static void tun_setup(struct net_device *dev)
1783 {
1784         struct tun_struct *tun = netdev_priv(dev);
1785
1786         tun->owner = INVALID_UID;
1787         tun->group = INVALID_GID;
1788
1789         dev->ethtool_ops = &tun_ethtool_ops;
1790         dev->needs_free_netdev = true;
1791         dev->priv_destructor = tun_free_netdev;
1792         /* We prefer our own queue length */
1793         dev->tx_queue_len = TUN_READQ_SIZE;
1794 }
1795
1796 /* Trivial set of netlink ops to allow deleting tun or tap
1797  * device with netlink.
1798  */
1799 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
1800                         struct netlink_ext_ack *extack)
1801 {
1802         return -EINVAL;
1803 }
1804
1805 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1806         .kind           = DRV_NAME,
1807         .priv_size      = sizeof(struct tun_struct),
1808         .setup          = tun_setup,
1809         .validate       = tun_validate,
1810 };
1811
1812 static void tun_sock_write_space(struct sock *sk)
1813 {
1814         struct tun_file *tfile;
1815         wait_queue_head_t *wqueue;
1816
1817         if (!sock_writeable(sk))
1818                 return;
1819
1820         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1821                 return;
1822
1823         wqueue = sk_sleep(sk);
1824         if (wqueue && waitqueue_active(wqueue))
1825                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1826                                                 POLLWRNORM | POLLWRBAND);
1827
1828         tfile = container_of(sk, struct tun_file, sk);
1829         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1830 }
1831
1832 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1833 {
1834         int ret;
1835         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1836         struct tun_struct *tun = __tun_get(tfile);
1837
1838         if (!tun)
1839                 return -EBADFD;
1840
1841         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1842                            m->msg_flags & MSG_DONTWAIT,
1843                            m->msg_flags & MSG_MORE);
1844         tun_put(tun);
1845         return ret;
1846 }
1847
1848 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1849                        int flags)
1850 {
1851         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1852         struct tun_struct *tun = __tun_get(tfile);
1853         int ret;
1854
1855         if (!tun)
1856                 return -EBADFD;
1857
1858         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1859                 ret = -EINVAL;
1860                 goto out;
1861         }
1862         if (flags & MSG_ERRQUEUE) {
1863                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1864                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1865                 goto out;
1866         }
1867         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT,
1868                           m->msg_control);
1869         if (ret > (ssize_t)total_len) {
1870                 m->msg_flags |= MSG_TRUNC;
1871                 ret = flags & MSG_TRUNC ? ret : total_len;
1872         }
1873 out:
1874         tun_put(tun);
1875         return ret;
1876 }
1877
1878 static int tun_peek_len(struct socket *sock)
1879 {
1880         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1881         struct tun_struct *tun;
1882         int ret = 0;
1883
1884         tun = __tun_get(tfile);
1885         if (!tun)
1886                 return 0;
1887
1888         ret = skb_array_peek_len(&tfile->tx_array);
1889         tun_put(tun);
1890
1891         return ret;
1892 }
1893
1894 /* Ops structure to mimic raw sockets with tun */
1895 static const struct proto_ops tun_socket_ops = {
1896         .peek_len = tun_peek_len,
1897         .sendmsg = tun_sendmsg,
1898         .recvmsg = tun_recvmsg,
1899 };
1900
1901 static struct proto tun_proto = {
1902         .name           = "tun",
1903         .owner          = THIS_MODULE,
1904         .obj_size       = sizeof(struct tun_file),
1905 };
1906
1907 static int tun_flags(struct tun_struct *tun)
1908 {
1909         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1910 }
1911
1912 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1913                               char *buf)
1914 {
1915         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1916         return sprintf(buf, "0x%x\n", tun_flags(tun));
1917 }
1918
1919 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1920                               char *buf)
1921 {
1922         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1923         return uid_valid(tun->owner)?
1924                 sprintf(buf, "%u\n",
1925                         from_kuid_munged(current_user_ns(), tun->owner)):
1926                 sprintf(buf, "-1\n");
1927 }
1928
1929 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1930                               char *buf)
1931 {
1932         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1933         return gid_valid(tun->group) ?
1934                 sprintf(buf, "%u\n",
1935                         from_kgid_munged(current_user_ns(), tun->group)):
1936                 sprintf(buf, "-1\n");
1937 }
1938
1939 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1940 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1941 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1942
1943 static struct attribute *tun_dev_attrs[] = {
1944         &dev_attr_tun_flags.attr,
1945         &dev_attr_owner.attr,
1946         &dev_attr_group.attr,
1947         NULL
1948 };
1949
1950 static const struct attribute_group tun_attr_group = {
1951         .attrs = tun_dev_attrs
1952 };
1953
1954 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1955 {
1956         struct tun_struct *tun;
1957         struct tun_file *tfile = file->private_data;
1958         struct net_device *dev;
1959         int err;
1960
1961         if (tfile->detached)
1962                 return -EINVAL;
1963
1964         dev = __dev_get_by_name(net, ifr->ifr_name);
1965         if (dev) {
1966                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1967                         return -EBUSY;
1968                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1969                         tun = netdev_priv(dev);
1970                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1971                         tun = netdev_priv(dev);
1972                 else
1973                         return -EINVAL;
1974
1975                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1976                     !!(tun->flags & IFF_MULTI_QUEUE))
1977                         return -EINVAL;
1978
1979                 if (tun_not_capable(tun))
1980                         return -EPERM;
1981                 err = security_tun_dev_open(tun->security);
1982                 if (err < 0)
1983                         return err;
1984
1985                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1986                 if (err < 0)
1987                         return err;
1988
1989                 if (tun->flags & IFF_MULTI_QUEUE &&
1990                     (tun->numqueues + tun->numdisabled > 1)) {
1991                         /* One or more queue has already been attached, no need
1992                          * to initialize the device again.
1993                          */
1994                         return 0;
1995                 }
1996         }
1997         else {
1998                 char *name;
1999                 unsigned long flags = 0;
2000                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2001                              MAX_TAP_QUEUES : 1;
2002
2003                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2004                         return -EPERM;
2005                 err = security_tun_dev_create();
2006                 if (err < 0)
2007                         return err;
2008
2009                 /* Set dev type */
2010                 if (ifr->ifr_flags & IFF_TUN) {
2011                         /* TUN device */
2012                         flags |= IFF_TUN;
2013                         name = "tun%d";
2014                 } else if (ifr->ifr_flags & IFF_TAP) {
2015                         /* TAP device */
2016                         flags |= IFF_TAP;
2017                         name = "tap%d";
2018                 } else
2019                         return -EINVAL;
2020
2021                 if (*ifr->ifr_name)
2022                         name = ifr->ifr_name;
2023
2024                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2025                                        NET_NAME_UNKNOWN, tun_setup, queues,
2026                                        queues);
2027
2028                 if (!dev)
2029                         return -ENOMEM;
2030                 err = dev_get_valid_name(net, dev, name);
2031                 if (err)
2032                         goto err_free_dev;
2033
2034                 dev_net_set(dev, net);
2035                 dev->rtnl_link_ops = &tun_link_ops;
2036                 dev->ifindex = tfile->ifindex;
2037                 dev->sysfs_groups[0] = &tun_attr_group;
2038
2039                 tun = netdev_priv(dev);
2040                 tun->dev = dev;
2041                 tun->flags = flags;
2042                 tun->txflt.count = 0;
2043                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2044
2045                 tun->align = NET_SKB_PAD;
2046                 tun->filter_attached = false;
2047                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2048                 tun->rx_batched = 0;
2049
2050                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2051                 if (!tun->pcpu_stats) {
2052                         err = -ENOMEM;
2053                         goto err_free_dev;
2054                 }
2055
2056                 spin_lock_init(&tun->lock);
2057
2058                 err = security_tun_dev_alloc_security(&tun->security);
2059                 if (err < 0)
2060                         goto err_free_stat;
2061
2062                 tun_net_init(dev);
2063                 tun_flow_init(tun);
2064
2065                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2066                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2067                                    NETIF_F_HW_VLAN_STAG_TX;
2068                 dev->features = dev->hw_features | NETIF_F_LLTX;
2069                 dev->vlan_features = dev->features &
2070                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2071                                        NETIF_F_HW_VLAN_STAG_TX);
2072
2073                 INIT_LIST_HEAD(&tun->disabled);
2074                 err = tun_attach(tun, file, false);
2075                 if (err < 0)
2076                         goto err_free_flow;
2077
2078                 err = register_netdevice(tun->dev);
2079                 if (err < 0)
2080                         goto err_detach;
2081         }
2082
2083         netif_carrier_on(tun->dev);
2084
2085         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2086
2087         tun->flags = (tun->flags & ~TUN_FEATURES) |
2088                 (ifr->ifr_flags & TUN_FEATURES);
2089
2090         /* Make sure persistent devices do not get stuck in
2091          * xoff state.
2092          */
2093         if (netif_running(tun->dev))
2094                 netif_tx_wake_all_queues(tun->dev);
2095
2096         strcpy(ifr->ifr_name, tun->dev->name);
2097         return 0;
2098
2099 err_detach:
2100         tun_detach_all(dev);
2101         /* register_netdevice() already called tun_free_netdev() */
2102         goto err_free_dev;
2103
2104 err_free_flow:
2105         tun_flow_uninit(tun);
2106         security_tun_dev_free_security(tun->security);
2107 err_free_stat:
2108         free_percpu(tun->pcpu_stats);
2109 err_free_dev:
2110         free_netdev(dev);
2111         return err;
2112 }
2113
2114 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2115                        struct ifreq *ifr)
2116 {
2117         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2118
2119         strcpy(ifr->ifr_name, tun->dev->name);
2120
2121         ifr->ifr_flags = tun_flags(tun);
2122
2123 }
2124
2125 /* This is like a cut-down ethtool ops, except done via tun fd so no
2126  * privs required. */
2127 static int set_offload(struct tun_struct *tun, unsigned long arg)
2128 {
2129         netdev_features_t features = 0;
2130
2131         if (arg & TUN_F_CSUM) {
2132                 features |= NETIF_F_HW_CSUM;
2133                 arg &= ~TUN_F_CSUM;
2134
2135                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2136                         if (arg & TUN_F_TSO_ECN) {
2137                                 features |= NETIF_F_TSO_ECN;
2138                                 arg &= ~TUN_F_TSO_ECN;
2139                         }
2140                         if (arg & TUN_F_TSO4)
2141                                 features |= NETIF_F_TSO;
2142                         if (arg & TUN_F_TSO6)
2143                                 features |= NETIF_F_TSO6;
2144                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2145                 }
2146         }
2147
2148         /* This gives the user a way to test for new features in future by
2149          * trying to set them. */
2150         if (arg)
2151                 return -EINVAL;
2152
2153         tun->set_features = features;
2154         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2155         tun->dev->wanted_features |= features;
2156         netdev_update_features(tun->dev);
2157
2158         return 0;
2159 }
2160
2161 static void tun_detach_filter(struct tun_struct *tun, int n)
2162 {
2163         int i;
2164         struct tun_file *tfile;
2165
2166         for (i = 0; i < n; i++) {
2167                 tfile = rtnl_dereference(tun->tfiles[i]);
2168                 lock_sock(tfile->socket.sk);
2169                 sk_detach_filter(tfile->socket.sk);
2170                 release_sock(tfile->socket.sk);
2171         }
2172
2173         tun->filter_attached = false;
2174 }
2175
2176 static int tun_attach_filter(struct tun_struct *tun)
2177 {
2178         int i, ret = 0;
2179         struct tun_file *tfile;
2180
2181         for (i = 0; i < tun->numqueues; i++) {
2182                 tfile = rtnl_dereference(tun->tfiles[i]);
2183                 lock_sock(tfile->socket.sk);
2184                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2185                 release_sock(tfile->socket.sk);
2186                 if (ret) {
2187                         tun_detach_filter(tun, i);
2188                         return ret;
2189                 }
2190         }
2191
2192         tun->filter_attached = true;
2193         return ret;
2194 }
2195
2196 static void tun_set_sndbuf(struct tun_struct *tun)
2197 {
2198         struct tun_file *tfile;
2199         int i;
2200
2201         for (i = 0; i < tun->numqueues; i++) {
2202                 tfile = rtnl_dereference(tun->tfiles[i]);
2203                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2204         }
2205 }
2206
2207 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2208 {
2209         struct tun_file *tfile = file->private_data;
2210         struct tun_struct *tun;
2211         int ret = 0;
2212
2213         rtnl_lock();
2214
2215         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2216                 tun = tfile->detached;
2217                 if (!tun) {
2218                         ret = -EINVAL;
2219                         goto unlock;
2220                 }
2221                 ret = security_tun_dev_attach_queue(tun->security);
2222                 if (ret < 0)
2223                         goto unlock;
2224                 ret = tun_attach(tun, file, false);
2225         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2226                 tun = rtnl_dereference(tfile->tun);
2227                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2228                         ret = -EINVAL;
2229                 else
2230                         __tun_detach(tfile, false);
2231         } else
2232                 ret = -EINVAL;
2233
2234 unlock:
2235         rtnl_unlock();
2236         return ret;
2237 }
2238
2239 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2240                             unsigned long arg, int ifreq_len)
2241 {
2242         struct tun_file *tfile = file->private_data;
2243         struct tun_struct *tun;
2244         void __user* argp = (void __user*)arg;
2245         struct ifreq ifr;
2246         kuid_t owner;
2247         kgid_t group;
2248         int sndbuf;
2249         int vnet_hdr_sz;
2250         unsigned int ifindex;
2251         int le;
2252         int ret;
2253
2254         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2255                 if (copy_from_user(&ifr, argp, ifreq_len))
2256                         return -EFAULT;
2257         } else {
2258                 memset(&ifr, 0, sizeof(ifr));
2259         }
2260         if (cmd == TUNGETFEATURES) {
2261                 /* Currently this just means: "what IFF flags are valid?".
2262                  * This is needed because we never checked for invalid flags on
2263                  * TUNSETIFF.
2264                  */
2265                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2266                                 (unsigned int __user*)argp);
2267         } else if (cmd == TUNSETQUEUE)
2268                 return tun_set_queue(file, &ifr);
2269
2270         ret = 0;
2271         rtnl_lock();
2272
2273         tun = __tun_get(tfile);
2274         if (cmd == TUNSETIFF) {
2275                 ret = -EEXIST;
2276                 if (tun)
2277                         goto unlock;
2278
2279                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2280
2281                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2282
2283                 if (ret)
2284                         goto unlock;
2285
2286                 if (copy_to_user(argp, &ifr, ifreq_len))
2287                         ret = -EFAULT;
2288                 goto unlock;
2289         }
2290         if (cmd == TUNSETIFINDEX) {
2291                 ret = -EPERM;
2292                 if (tun)
2293                         goto unlock;
2294
2295                 ret = -EFAULT;
2296                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2297                         goto unlock;
2298
2299                 ret = 0;
2300                 tfile->ifindex = ifindex;
2301                 goto unlock;
2302         }
2303
2304         ret = -EBADFD;
2305         if (!tun)
2306                 goto unlock;
2307
2308         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2309
2310         ret = 0;
2311         switch (cmd) {
2312         case TUNGETIFF:
2313                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2314
2315                 if (tfile->detached)
2316                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2317                 if (!tfile->socket.sk->sk_filter)
2318                         ifr.ifr_flags |= IFF_NOFILTER;
2319
2320                 if (copy_to_user(argp, &ifr, ifreq_len))
2321                         ret = -EFAULT;
2322                 break;
2323
2324         case TUNSETNOCSUM:
2325                 /* Disable/Enable checksum */
2326
2327                 /* [unimplemented] */
2328                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2329                           arg ? "disabled" : "enabled");
2330                 break;
2331
2332         case TUNSETPERSIST:
2333                 /* Disable/Enable persist mode. Keep an extra reference to the
2334                  * module to prevent the module being unprobed.
2335                  */
2336                 if (arg && !(tun->flags & IFF_PERSIST)) {
2337                         tun->flags |= IFF_PERSIST;
2338                         __module_get(THIS_MODULE);
2339                 }
2340                 if (!arg && (tun->flags & IFF_PERSIST)) {
2341                         tun->flags &= ~IFF_PERSIST;
2342                         module_put(THIS_MODULE);
2343                 }
2344
2345                 tun_debug(KERN_INFO, tun, "persist %s\n",
2346                           arg ? "enabled" : "disabled");
2347                 break;
2348
2349         case TUNSETOWNER:
2350                 /* Set owner of the device */
2351                 owner = make_kuid(current_user_ns(), arg);
2352                 if (!uid_valid(owner)) {
2353                         ret = -EINVAL;
2354                         break;
2355                 }
2356                 tun->owner = owner;
2357                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2358                           from_kuid(&init_user_ns, tun->owner));
2359                 break;
2360
2361         case TUNSETGROUP:
2362                 /* Set group of the device */
2363                 group = make_kgid(current_user_ns(), arg);
2364                 if (!gid_valid(group)) {
2365                         ret = -EINVAL;
2366                         break;
2367                 }
2368                 tun->group = group;
2369                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2370                           from_kgid(&init_user_ns, tun->group));
2371                 break;
2372
2373         case TUNSETLINK:
2374                 /* Only allow setting the type when the interface is down */
2375                 if (tun->dev->flags & IFF_UP) {
2376                         tun_debug(KERN_INFO, tun,
2377                                   "Linktype set failed because interface is up\n");
2378                         ret = -EBUSY;
2379                 } else {
2380                         tun->dev->type = (int) arg;
2381                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2382                                   tun->dev->type);
2383                         ret = 0;
2384                 }
2385                 break;
2386
2387 #ifdef TUN_DEBUG
2388         case TUNSETDEBUG:
2389                 tun->debug = arg;
2390                 break;
2391 #endif
2392         case TUNSETOFFLOAD:
2393                 ret = set_offload(tun, arg);
2394                 break;
2395
2396         case TUNSETTXFILTER:
2397                 /* Can be set only for TAPs */
2398                 ret = -EINVAL;
2399                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2400                         break;
2401                 ret = update_filter(&tun->txflt, (void __user *)arg);
2402                 break;
2403
2404         case SIOCGIFHWADDR:
2405                 /* Get hw address */
2406                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2407                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2408                 if (copy_to_user(argp, &ifr, ifreq_len))
2409                         ret = -EFAULT;
2410                 break;
2411
2412         case SIOCSIFHWADDR:
2413                 /* Set hw address */
2414                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2415                           ifr.ifr_hwaddr.sa_data);
2416
2417                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2418                 break;
2419
2420         case TUNGETSNDBUF:
2421                 sndbuf = tfile->socket.sk->sk_sndbuf;
2422                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2423                         ret = -EFAULT;
2424                 break;
2425
2426         case TUNSETSNDBUF:
2427                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2428                         ret = -EFAULT;
2429                         break;
2430                 }
2431
2432                 tun->sndbuf = sndbuf;
2433                 tun_set_sndbuf(tun);
2434                 break;
2435
2436         case TUNGETVNETHDRSZ:
2437                 vnet_hdr_sz = tun->vnet_hdr_sz;
2438                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2439                         ret = -EFAULT;
2440                 break;
2441
2442         case TUNSETVNETHDRSZ:
2443                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2444                         ret = -EFAULT;
2445                         break;
2446                 }
2447                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2448                         ret = -EINVAL;
2449                         break;
2450                 }
2451
2452                 tun->vnet_hdr_sz = vnet_hdr_sz;
2453                 break;
2454
2455         case TUNGETVNETLE:
2456                 le = !!(tun->flags & TUN_VNET_LE);
2457                 if (put_user(le, (int __user *)argp))
2458                         ret = -EFAULT;
2459                 break;
2460
2461         case TUNSETVNETLE:
2462                 if (get_user(le, (int __user *)argp)) {
2463                         ret = -EFAULT;
2464                         break;
2465                 }
2466                 if (le)
2467                         tun->flags |= TUN_VNET_LE;
2468                 else
2469                         tun->flags &= ~TUN_VNET_LE;
2470                 break;
2471
2472         case TUNGETVNETBE:
2473                 ret = tun_get_vnet_be(tun, argp);
2474                 break;
2475
2476         case TUNSETVNETBE:
2477                 ret = tun_set_vnet_be(tun, argp);
2478                 break;
2479
2480         case TUNATTACHFILTER:
2481                 /* Can be set only for TAPs */
2482                 ret = -EINVAL;
2483                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2484                         break;
2485                 ret = -EFAULT;
2486                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2487                         break;
2488
2489                 ret = tun_attach_filter(tun);
2490                 break;
2491
2492         case TUNDETACHFILTER:
2493                 /* Can be set only for TAPs */
2494                 ret = -EINVAL;
2495                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2496                         break;
2497                 ret = 0;
2498                 tun_detach_filter(tun, tun->numqueues);
2499                 break;
2500
2501         case TUNGETFILTER:
2502                 ret = -EINVAL;
2503                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2504                         break;
2505                 ret = -EFAULT;
2506                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2507                         break;
2508                 ret = 0;
2509                 break;
2510
2511         default:
2512                 ret = -EINVAL;
2513                 break;
2514         }
2515
2516 unlock:
2517         rtnl_unlock();
2518         if (tun)
2519                 tun_put(tun);
2520         return ret;
2521 }
2522
2523 static long tun_chr_ioctl(struct file *file,
2524                           unsigned int cmd, unsigned long arg)
2525 {
2526         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2527 }
2528
2529 #ifdef CONFIG_COMPAT
2530 static long tun_chr_compat_ioctl(struct file *file,
2531                          unsigned int cmd, unsigned long arg)
2532 {
2533         switch (cmd) {
2534         case TUNSETIFF:
2535         case TUNGETIFF:
2536         case TUNSETTXFILTER:
2537         case TUNGETSNDBUF:
2538         case TUNSETSNDBUF:
2539         case SIOCGIFHWADDR:
2540         case SIOCSIFHWADDR:
2541                 arg = (unsigned long)compat_ptr(arg);
2542                 break;
2543         default:
2544                 arg = (compat_ulong_t)arg;
2545                 break;
2546         }
2547
2548         /*
2549          * compat_ifreq is shorter than ifreq, so we must not access beyond
2550          * the end of that structure. All fields that are used in this
2551          * driver are compatible though, we don't need to convert the
2552          * contents.
2553          */
2554         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2555 }
2556 #endif /* CONFIG_COMPAT */
2557
2558 static int tun_chr_fasync(int fd, struct file *file, int on)
2559 {
2560         struct tun_file *tfile = file->private_data;
2561         int ret;
2562
2563         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2564                 goto out;
2565
2566         if (on) {
2567                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2568                 tfile->flags |= TUN_FASYNC;
2569         } else
2570                 tfile->flags &= ~TUN_FASYNC;
2571         ret = 0;
2572 out:
2573         return ret;
2574 }
2575
2576 static int tun_chr_open(struct inode *inode, struct file * file)
2577 {
2578         struct net *net = current->nsproxy->net_ns;
2579         struct tun_file *tfile;
2580
2581         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2582
2583         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2584                                             &tun_proto, 0);
2585         if (!tfile)
2586                 return -ENOMEM;
2587         RCU_INIT_POINTER(tfile->tun, NULL);
2588         tfile->flags = 0;
2589         tfile->ifindex = 0;
2590
2591         init_waitqueue_head(&tfile->wq.wait);
2592         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2593
2594         tfile->socket.file = file;
2595         tfile->socket.ops = &tun_socket_ops;
2596
2597         sock_init_data(&tfile->socket, &tfile->sk);
2598
2599         tfile->sk.sk_write_space = tun_sock_write_space;
2600         tfile->sk.sk_sndbuf = INT_MAX;
2601
2602         file->private_data = tfile;
2603         INIT_LIST_HEAD(&tfile->next);
2604
2605         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2606
2607         return 0;
2608 }
2609
2610 static int tun_chr_close(struct inode *inode, struct file *file)
2611 {
2612         struct tun_file *tfile = file->private_data;
2613
2614         tun_detach(tfile, true);
2615
2616         return 0;
2617 }
2618
2619 #ifdef CONFIG_PROC_FS
2620 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2621 {
2622         struct tun_struct *tun;
2623         struct ifreq ifr;
2624
2625         memset(&ifr, 0, sizeof(ifr));
2626
2627         rtnl_lock();
2628         tun = tun_get(f);
2629         if (tun)
2630                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2631         rtnl_unlock();
2632
2633         if (tun)
2634                 tun_put(tun);
2635
2636         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2637 }
2638 #endif
2639
2640 static const struct file_operations tun_fops = {
2641         .owner  = THIS_MODULE,
2642         .llseek = no_llseek,
2643         .read_iter  = tun_chr_read_iter,
2644         .write_iter = tun_chr_write_iter,
2645         .poll   = tun_chr_poll,
2646         .unlocked_ioctl = tun_chr_ioctl,
2647 #ifdef CONFIG_COMPAT
2648         .compat_ioctl = tun_chr_compat_ioctl,
2649 #endif
2650         .open   = tun_chr_open,
2651         .release = tun_chr_close,
2652         .fasync = tun_chr_fasync,
2653 #ifdef CONFIG_PROC_FS
2654         .show_fdinfo = tun_chr_show_fdinfo,
2655 #endif
2656 };
2657
2658 static struct miscdevice tun_miscdev = {
2659         .minor = TUN_MINOR,
2660         .name = "tun",
2661         .nodename = "net/tun",
2662         .fops = &tun_fops,
2663 };
2664
2665 /* ethtool interface */
2666
2667 static int tun_get_link_ksettings(struct net_device *dev,
2668                                   struct ethtool_link_ksettings *cmd)
2669 {
2670         ethtool_link_ksettings_zero_link_mode(cmd, supported);
2671         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
2672         cmd->base.speed         = SPEED_10;
2673         cmd->base.duplex        = DUPLEX_FULL;
2674         cmd->base.port          = PORT_TP;
2675         cmd->base.phy_address   = 0;
2676         cmd->base.autoneg       = AUTONEG_DISABLE;
2677         return 0;
2678 }
2679
2680 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2681 {
2682         struct tun_struct *tun = netdev_priv(dev);
2683
2684         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2685         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2686
2687         switch (tun->flags & TUN_TYPE_MASK) {
2688         case IFF_TUN:
2689                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2690                 break;
2691         case IFF_TAP:
2692                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2693                 break;
2694         }
2695 }
2696
2697 static u32 tun_get_msglevel(struct net_device *dev)
2698 {
2699 #ifdef TUN_DEBUG
2700         struct tun_struct *tun = netdev_priv(dev);
2701         return tun->debug;
2702 #else
2703         return -EOPNOTSUPP;
2704 #endif
2705 }
2706
2707 static void tun_set_msglevel(struct net_device *dev, u32 value)
2708 {
2709 #ifdef TUN_DEBUG
2710         struct tun_struct *tun = netdev_priv(dev);
2711         tun->debug = value;
2712 #endif
2713 }
2714
2715 static int tun_get_coalesce(struct net_device *dev,
2716                             struct ethtool_coalesce *ec)
2717 {
2718         struct tun_struct *tun = netdev_priv(dev);
2719
2720         ec->rx_max_coalesced_frames = tun->rx_batched;
2721
2722         return 0;
2723 }
2724
2725 static int tun_set_coalesce(struct net_device *dev,
2726                             struct ethtool_coalesce *ec)
2727 {
2728         struct tun_struct *tun = netdev_priv(dev);
2729
2730         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
2731                 tun->rx_batched = NAPI_POLL_WEIGHT;
2732         else
2733                 tun->rx_batched = ec->rx_max_coalesced_frames;
2734
2735         return 0;
2736 }
2737
2738 static const struct ethtool_ops tun_ethtool_ops = {
2739         .get_drvinfo    = tun_get_drvinfo,
2740         .get_msglevel   = tun_get_msglevel,
2741         .set_msglevel   = tun_set_msglevel,
2742         .get_link       = ethtool_op_get_link,
2743         .get_ts_info    = ethtool_op_get_ts_info,
2744         .get_coalesce   = tun_get_coalesce,
2745         .set_coalesce   = tun_set_coalesce,
2746         .get_link_ksettings = tun_get_link_ksettings,
2747 };
2748
2749 static int tun_queue_resize(struct tun_struct *tun)
2750 {
2751         struct net_device *dev = tun->dev;
2752         struct tun_file *tfile;
2753         struct skb_array **arrays;
2754         int n = tun->numqueues + tun->numdisabled;
2755         int ret, i;
2756
2757         arrays = kmalloc_array(n, sizeof(*arrays), GFP_KERNEL);
2758         if (!arrays)
2759                 return -ENOMEM;
2760
2761         for (i = 0; i < tun->numqueues; i++) {
2762                 tfile = rtnl_dereference(tun->tfiles[i]);
2763                 arrays[i] = &tfile->tx_array;
2764         }
2765         list_for_each_entry(tfile, &tun->disabled, next)
2766                 arrays[i++] = &tfile->tx_array;
2767
2768         ret = skb_array_resize_multiple(arrays, n,
2769                                         dev->tx_queue_len, GFP_KERNEL);
2770
2771         kfree(arrays);
2772         return ret;
2773 }
2774
2775 static int tun_device_event(struct notifier_block *unused,
2776                             unsigned long event, void *ptr)
2777 {
2778         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2779         struct tun_struct *tun = netdev_priv(dev);
2780
2781         if (dev->rtnl_link_ops != &tun_link_ops)
2782                 return NOTIFY_DONE;
2783
2784         switch (event) {
2785         case NETDEV_CHANGE_TX_QUEUE_LEN:
2786                 if (tun_queue_resize(tun))
2787                         return NOTIFY_BAD;
2788                 break;
2789         default:
2790                 break;
2791         }
2792
2793         return NOTIFY_DONE;
2794 }
2795
2796 static struct notifier_block tun_notifier_block __read_mostly = {
2797         .notifier_call  = tun_device_event,
2798 };
2799
2800 static int __init tun_init(void)
2801 {
2802         int ret = 0;
2803
2804         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2805
2806         ret = rtnl_link_register(&tun_link_ops);
2807         if (ret) {
2808                 pr_err("Can't register link_ops\n");
2809                 goto err_linkops;
2810         }
2811
2812         ret = misc_register(&tun_miscdev);
2813         if (ret) {
2814                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2815                 goto err_misc;
2816         }
2817
2818         ret = register_netdevice_notifier(&tun_notifier_block);
2819         if (ret) {
2820                 pr_err("Can't register netdevice notifier\n");
2821                 goto err_notifier;
2822         }
2823
2824         return  0;
2825
2826 err_notifier:
2827         misc_deregister(&tun_miscdev);
2828 err_misc:
2829         rtnl_link_unregister(&tun_link_ops);
2830 err_linkops:
2831         return ret;
2832 }
2833
2834 static void tun_cleanup(void)
2835 {
2836         misc_deregister(&tun_miscdev);
2837         rtnl_link_unregister(&tun_link_ops);
2838         unregister_netdevice_notifier(&tun_notifier_block);
2839 }
2840
2841 /* Get an underlying socket object from tun file.  Returns error unless file is
2842  * attached to a device.  The returned object works like a packet socket, it
2843  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2844  * holding a reference to the file for as long as the socket is in use. */
2845 struct socket *tun_get_socket(struct file *file)
2846 {
2847         struct tun_file *tfile;
2848         if (file->f_op != &tun_fops)
2849                 return ERR_PTR(-EINVAL);
2850         tfile = file->private_data;
2851         if (!tfile)
2852                 return ERR_PTR(-EBADFD);
2853         return &tfile->socket;
2854 }
2855 EXPORT_SYMBOL_GPL(tun_get_socket);
2856
2857 struct skb_array *tun_get_skb_array(struct file *file)
2858 {
2859         struct tun_file *tfile;
2860
2861         if (file->f_op != &tun_fops)
2862                 return ERR_PTR(-EINVAL);
2863         tfile = file->private_data;
2864         if (!tfile)
2865                 return ERR_PTR(-EBADFD);
2866         return &tfile->tx_array;
2867 }
2868 EXPORT_SYMBOL_GPL(tun_get_skb_array);
2869
2870 module_init(tun_init);
2871 module_exit(tun_cleanup);
2872 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2873 MODULE_AUTHOR(DRV_COPYRIGHT);
2874 MODULE_LICENSE("GPL");
2875 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2876 MODULE_ALIAS("devname:net/tun");