drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/major.h>
  48 #include <linux/slab.h>
  49 #include <linux/poll.h>
  50 #include <linux/fcntl.h>
  51 #include <linux/init.h>
  52 #include <linux/skbuff.h>
  53 #include <linux/netdevice.h>
  54 #include <linux/etherdevice.h>
  55 #include <linux/miscdevice.h>
  56 #include <linux/ethtool.h>
  57 #include <linux/rtnetlink.h>
  58 #include <linux/compat.h>
  59 #include <linux/if.h>
  60 #include <linux/if_arp.h>
  61 #include <linux/if_ether.h>
  62 #include <linux/if_tun.h>
  63 #include <linux/if_vlan.h>
  64 #include <linux/crc32.h>
  65 #include <linux/nsproxy.h>
  66 #include <linux/virtio_net.h>
  67 #include <linux/rcupdate.h>
  68 #include <net/net_namespace.h>
  69 #include <net/netns/generic.h>
  70 #include <net/rtnetlink.h>
  71 #include <net/sock.h>
  72 #include <linux/seq_file.h>
  73 #include <linux/uio.h>
  74
  75 #include <asm/uaccess.h>
  76
  77 /* Uncomment to enable debugging */
  78 /* #define TUN_DEBUG 1 */
  79
  80 #ifdef TUN_DEBUG
  81 static int debug;
  82
  83 #define tun_debug(level, tun, fmt, args...)                     \
  84 do {                                                            \
  85         if (tun->debug)                                         \
  86                 netdev_printk(level, tun->dev, fmt, ##args);    \
  87 } while (0)
  88 #define DBG1(level, fmt, args...)                               \
  89 do {                                                            \
  90         if (debug == 2)                                         \
  91                 printk(level fmt, ##args);                      \
  92 } while (0)
  93 #else
  94 #define tun_debug(level, tun, fmt, args...)                     \
  95 do {                                                            \
  96         if (0)                                                  \
  97                 netdev_printk(level, tun->dev, fmt, ##args);    \
  98 } while (0)
  99 #define DBG1(level, fmt, args...)                               \
 100 do {                                                            \
 101         if (0)                                                  \
 102                 printk(level fmt, ##args);                      \
 103 } while (0)
 104 #endif
 105
 106 /* TUN device flags */
 107
 108 /* IFF_ATTACH_QUEUE is never stored in device flags,
 109  * overload it to mean fasync when stored there.
 110  */
 111 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 112 /* High bits in flags field are unused. */
 113 #define TUN_VNET_LE     0x80000000
 114 #define TUN_VNET_BE     0x40000000
 115
 116 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 117                       IFF_MULTI_QUEUE)
 118 #define GOODCOPY_LEN 128
 119
 120 #define FLT_EXACT_COUNT 8
 121 struct tap_filter {
 122         unsigned int    count;    /* Number of addrs. Zero means disabled */
 123         u32             mask[2];  /* Mask of the hashed addrs */
 124         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 125 };
 126
 127 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 128  * to max number of VCPUs in guest. */
 129 #define MAX_TAP_QUEUES 256
 130 #define MAX_TAP_FLOWS  4096
 131
 132 #define TUN_FLOW_EXPIRE (3 * HZ)
 133
 134 /* A tun_file connects an open character device to a tuntap netdevice. It
 135  * also contains all socket related structures (except sock_fprog and tap_filter)
 136  * to serve as one transmit queue for tuntap device. The sock_fprog and
 137  * tap_filter were kept in tun_struct since they were used for filtering for the
 138  * netdevice not for a specific queue (at least I didn't see the requirement for
 139  * this).
 140  *
 141  * RCU usage:
 142  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 143  * other can only be read while rcu_read_lock or rtnl_lock is held.
 144  */
 145 struct tun_file {
 146         struct sock sk;
 147         struct socket socket;
 148         struct socket_wq wq;
 149         struct tun_struct __rcu *tun;
 150         struct fasync_struct *fasync;
 151         /* only used for fasnyc */
 152         unsigned int flags;
 153         union {
 154                 u16 queue_index;
 155                 unsigned int ifindex;
 156         };
 157         struct list_head next;
 158         struct tun_struct *detached;
 159 };
 160
 161 struct tun_flow_entry {
 162         struct hlist_node hash_link;
 163         struct rcu_head rcu;
 164         struct tun_struct *tun;
 165
 166         u32 rxhash;
 167         u32 rps_rxhash;
 168         int queue_index;
 169         unsigned long updated;
 170 };
 171
 172 #define TUN_NUM_FLOW_ENTRIES 1024
 173
 174 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 175  * device, socket filter, sndbuf and vnet header size were restore when the
 176  * file were attached to a persist device.
 177  */
 178 struct tun_struct {
 179         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 180         unsigned int            numqueues;
 181         unsigned int            flags;
 182         kuid_t                  owner;
 183         kgid_t                  group;
 184
 185         struct net_device       *dev;
 186         netdev_features_t       set_features;
 187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 188                           NETIF_F_TSO6|NETIF_F_UFO)
 189
 190         int                     vnet_hdr_sz;
 191         int                     sndbuf;
 192         struct tap_filter       txflt;
 193         struct sock_fprog       fprog;
 194         /* protected by rtnl lock */
 195         bool                    filter_attached;
 196 #ifdef TUN_DEBUG
 197         int debug;
 198 #endif
 199         spinlock_t lock;
 200         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 201         struct timer_list flow_gc_timer;
 202         unsigned long ageing_time;
 203         unsigned int numdisabled;
 204         struct list_head disabled;
 205         void *security;
 206         u32 flow_count;
 207 };
 208
 209 #ifdef CONFIG_TUN_VNET_CROSS_LE
 210 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 211 {
 212         return tun->flags & TUN_VNET_BE ? false :
 213                 virtio_legacy_is_little_endian();
 214 }
 215
 216 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 217 {
 218         int be = !!(tun->flags & TUN_VNET_BE);
 219
 220         if (put_user(be, argp))
 221                 return -EFAULT;
 222
 223         return 0;
 224 }
 225
 226 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 227 {
 228         int be;
 229
 230         if (get_user(be, argp))
 231                 return -EFAULT;
 232
 233         if (be)
 234                 tun->flags |= TUN_VNET_BE;
 235         else
 236                 tun->flags &= ~TUN_VNET_BE;
 237
 238         return 0;
 239 }
 240 #else
 241 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 242 {
 243         return virtio_legacy_is_little_endian();
 244 }
 245
 246 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 247 {
 248         return -EINVAL;
 249 }
 250
 251 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 252 {
 253         return -EINVAL;
 254 }
 255 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 256
 257 static inline bool tun_is_little_endian(struct tun_struct *tun)
 258 {
 259         return tun->flags & TUN_VNET_LE ||
 260                 tun_legacy_is_little_endian(tun);
 261 }
 262
 263 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 264 {
 265         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 266 }
 267
 268 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 269 {
 270         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 271 }
 272
 273 static inline u32 tun_hashfn(u32 rxhash)
 274 {
 275         return rxhash & 0x3ff;
 276 }
 277
 278 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 279 {
 280         struct tun_flow_entry *e;
 281
 282         hlist_for_each_entry_rcu(e, head, hash_link) {
 283                 if (e->rxhash == rxhash)
 284                         return e;
 285         }
 286         return NULL;
 287 }
 288
 289 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 290                                               struct hlist_head *head,
 291                                               u32 rxhash, u16 queue_index)
 292 {
 293         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 294
 295         if (e) {
 296                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 297                           rxhash, queue_index);
 298                 e->updated = jiffies;
 299                 e->rxhash = rxhash;
 300                 e->rps_rxhash = 0;
 301                 e->queue_index = queue_index;
 302                 e->tun = tun;
 303                 hlist_add_head_rcu(&e->hash_link, head);
 304                 ++tun->flow_count;
 305         }
 306         return e;
 307 }
 308
 309 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 310 {
 311         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 312                   e->rxhash, e->queue_index);
 313         hlist_del_rcu(&e->hash_link);
 314         kfree_rcu(e, rcu);
 315         --tun->flow_count;
 316 }
 317
 318 static void tun_flow_flush(struct tun_struct *tun)
 319 {
 320         int i;
 321
 322         spin_lock_bh(&tun->lock);
 323         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 324                 struct tun_flow_entry *e;
 325                 struct hlist_node *n;
 326
 327                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 328                         tun_flow_delete(tun, e);
 329         }
 330         spin_unlock_bh(&tun->lock);
 331 }
 332
 333 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 334 {
 335         int i;
 336
 337         spin_lock_bh(&tun->lock);
 338         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 339                 struct tun_flow_entry *e;
 340                 struct hlist_node *n;
 341
 342                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 343                         if (e->queue_index == queue_index)
 344                                 tun_flow_delete(tun, e);
 345                 }
 346         }
 347         spin_unlock_bh(&tun->lock);
 348 }
 349
 350 static void tun_flow_cleanup(unsigned long data)
 351 {
 352         struct tun_struct *tun = (struct tun_struct *)data;
 353         unsigned long delay = tun->ageing_time;
 354         unsigned long next_timer = jiffies + delay;
 355         unsigned long count = 0;
 356         int i;
 357
 358         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 359
 360         spin_lock_bh(&tun->lock);
 361         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 362                 struct tun_flow_entry *e;
 363                 struct hlist_node *n;
 364
 365                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 366                         unsigned long this_timer;
 367                         count++;
 368                         this_timer = e->updated + delay;
 369                         if (time_before_eq(this_timer, jiffies))
 370                                 tun_flow_delete(tun, e);
 371                         else if (time_before(this_timer, next_timer))
 372                                 next_timer = this_timer;
 373                 }
 374         }
 375
 376         if (count)
 377                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 378         spin_unlock_bh(&tun->lock);
 379 }
 380
 381 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 382                             struct tun_file *tfile)
 383 {
 384         struct hlist_head *head;
 385         struct tun_flow_entry *e;
 386         unsigned long delay = tun->ageing_time;
 387         u16 queue_index = tfile->queue_index;
 388
 389         if (!rxhash)
 390                 return;
 391         else
 392                 head = &tun->flows[tun_hashfn(rxhash)];
 393
 394         rcu_read_lock();
 395
 396         /* We may get a very small possibility of OOO during switching, not
 397          * worth to optimize.*/
 398         if (tun->numqueues == 1 || tfile->detached)
 399                 goto unlock;
 400
 401         e = tun_flow_find(head, rxhash);
 402         if (likely(e)) {
 403                 /* TODO: keep queueing to old queue until it's empty? */
 404                 e->queue_index = queue_index;
 405                 e->updated = jiffies;
 406                 sock_rps_record_flow_hash(e->rps_rxhash);
 407         } else {
 408                 spin_lock_bh(&tun->lock);
 409                 if (!tun_flow_find(head, rxhash) &&
 410                     tun->flow_count < MAX_TAP_FLOWS)
 411                         tun_flow_create(tun, head, rxhash, queue_index);
 412
 413                 if (!timer_pending(&tun->flow_gc_timer))
 414                         mod_timer(&tun->flow_gc_timer,
 415                                   round_jiffies_up(jiffies + delay));
 416                 spin_unlock_bh(&tun->lock);
 417         }
 418
 419 unlock:
 420         rcu_read_unlock();
 421 }
 422
 423 /**
 424  * Save the hash received in the stack receive path and update the
 425  * flow_hash table accordingly.
 426  */
 427 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 428 {
 429         if (unlikely(e->rps_rxhash != hash))
 430                 e->rps_rxhash = hash;
 431 }
 432
 433 /* We try to identify a flow through its rxhash first. The reason that
 434  * we do not check rxq no. is because some cards(e.g 82599), chooses
 435  * the rxq based on the txq where the last packet of the flow comes. As
 436  * the userspace application move between processors, we may get a
 437  * different rxq no. here. If we could not get rxhash, then we would
 438  * hope the rxq no. may help here.
 439  */
 440 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 441                             void *accel_priv, select_queue_fallback_t fallback)
 442 {
 443         struct tun_struct *tun = netdev_priv(dev);
 444         struct tun_flow_entry *e;
 445         u32 txq = 0;
 446         u32 numqueues = 0;
 447
 448         rcu_read_lock();
 449         numqueues = ACCESS_ONCE(tun->numqueues);
 450
 451         txq = skb_get_hash(skb);
 452         if (txq) {
 453                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 454                 if (e) {
 455                         tun_flow_save_rps_rxhash(e, txq);
 456                         txq = e->queue_index;
 457                 } else
 458                         /* use multiply and shift instead of expensive divide */
 459                         txq = ((u64)txq * numqueues) >> 32;
 460         } else if (likely(skb_rx_queue_recorded(skb))) {
 461                 txq = skb_get_rx_queue(skb);
 462                 while (unlikely(txq >= numqueues))
 463                         txq -= numqueues;
 464         }
 465
 466         rcu_read_unlock();
 467         return txq;
 468 }
 469
 470 static inline bool tun_not_capable(struct tun_struct *tun)
 471 {
 472         const struct cred *cred = current_cred();
 473         struct net *net = dev_net(tun->dev);
 474
 475         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 476                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 477                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 478 }
 479
 480 static void tun_set_real_num_queues(struct tun_struct *tun)
 481 {
 482         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 483         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 484 }
 485
 486 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 487 {
 488         tfile->detached = tun;
 489         list_add_tail(&tfile->next, &tun->disabled);
 490         ++tun->numdisabled;
 491 }
 492
 493 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 494 {
 495         struct tun_struct *tun = tfile->detached;
 496
 497         tfile->detached = NULL;
 498         list_del_init(&tfile->next);
 499         --tun->numdisabled;
 500         return tun;
 501 }
 502
 503 static void tun_queue_purge(struct tun_file *tfile)
 504 {
 505         skb_queue_purge(&tfile->sk.sk_receive_queue);
 506         skb_queue_purge(&tfile->sk.sk_error_queue);
 507 }
 508
 509 static void __tun_detach(struct tun_file *tfile, bool clean)
 510 {
 511         struct tun_file *ntfile;
 512         struct tun_struct *tun;
 513
 514         tun = rtnl_dereference(tfile->tun);
 515
 516         if (tun && !tfile->detached) {
 517                 u16 index = tfile->queue_index;
 518                 BUG_ON(index >= tun->numqueues);
 519
 520                 rcu_assign_pointer(tun->tfiles[index],
 521                                    tun->tfiles[tun->numqueues - 1]);
 522                 ntfile = rtnl_dereference(tun->tfiles[index]);
 523                 ntfile->queue_index = index;
 524
 525                 --tun->numqueues;
 526                 if (clean) {
 527                         RCU_INIT_POINTER(tfile->tun, NULL);
 528                         sock_put(&tfile->sk);
 529                 } else
 530                         tun_disable_queue(tun, tfile);
 531
 532                 synchronize_net();
 533                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 534                 /* Drop read queue */
 535                 tun_queue_purge(tfile);
 536                 tun_set_real_num_queues(tun);
 537         } else if (tfile->detached && clean) {
 538                 tun = tun_enable_queue(tfile);
 539                 sock_put(&tfile->sk);
 540         }
 541
 542         if (clean) {
 543                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 544                         netif_carrier_off(tun->dev);
 545
 546                         if (!(tun->flags & IFF_PERSIST) &&
 547                             tun->dev->reg_state == NETREG_REGISTERED)
 548                                 unregister_netdevice(tun->dev);
 549                 }
 550                 sock_put(&tfile->sk);
 551         }
 552 }
 553
 554 static void tun_detach(struct tun_file *tfile, bool clean)
 555 {
 556         rtnl_lock();
 557         __tun_detach(tfile, clean);
 558         rtnl_unlock();
 559 }
 560
 561 static void tun_detach_all(struct net_device *dev)
 562 {
 563         struct tun_struct *tun = netdev_priv(dev);
 564         struct tun_file *tfile, *tmp;
 565         int i, n = tun->numqueues;
 566
 567         for (i = 0; i < n; i++) {
 568                 tfile = rtnl_dereference(tun->tfiles[i]);
 569                 BUG_ON(!tfile);
 570                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 571                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 572                 RCU_INIT_POINTER(tfile->tun, NULL);
 573                 --tun->numqueues;
 574         }
 575         list_for_each_entry(tfile, &tun->disabled, next) {
 576                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 577                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 578                 RCU_INIT_POINTER(tfile->tun, NULL);
 579         }
 580         BUG_ON(tun->numqueues != 0);
 581
 582         synchronize_net();
 583         for (i = 0; i < n; i++) {
 584                 tfile = rtnl_dereference(tun->tfiles[i]);
 585                 /* Drop read queue */
 586                 tun_queue_purge(tfile);
 587                 sock_put(&tfile->sk);
 588         }
 589         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 590                 tun_enable_queue(tfile);
 591                 tun_queue_purge(tfile);
 592                 sock_put(&tfile->sk);
 593         }
 594         BUG_ON(tun->numdisabled != 0);
 595
 596         if (tun->flags & IFF_PERSIST)
 597                 module_put(THIS_MODULE);
 598 }
 599
 600 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 601 {
 602         struct tun_file *tfile = file->private_data;
 603         int err;
 604
 605         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 606         if (err < 0)
 607                 goto out;
 608
 609         err = -EINVAL;
 610         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 611                 goto out;
 612
 613         err = -EBUSY;
 614         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 615                 goto out;
 616
 617         err = -E2BIG;
 618         if (!tfile->detached &&
 619             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 620                 goto out;
 621
 622         err = 0;
 623
 624         /* Re-attach the filter to persist device */
 625         if (!skip_filter && (tun->filter_attached == true)) {
 626                 err = __sk_attach_filter(&tun->fprog, tfile->socket.sk,
 627                                          lockdep_rtnl_is_held());
 628                 if (!err)
 629                         goto out;
 630         }
 631         tfile->queue_index = tun->numqueues;
 632         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 633         rcu_assign_pointer(tfile->tun, tun);
 634         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 635         tun->numqueues++;
 636
 637         if (tfile->detached)
 638                 tun_enable_queue(tfile);
 639         else
 640                 sock_hold(&tfile->sk);
 641
 642         tun_set_real_num_queues(tun);
 643
 644         /* device is allowed to go away first, so no need to hold extra
 645          * refcnt.
 646          */
 647
 648 out:
 649         return err;
 650 }
 651
 652 static struct tun_struct *__tun_get(struct tun_file *tfile)
 653 {
 654         struct tun_struct *tun;
 655
 656         rcu_read_lock();
 657         tun = rcu_dereference(tfile->tun);
 658         if (tun)
 659                 dev_hold(tun->dev);
 660         rcu_read_unlock();
 661
 662         return tun;
 663 }
 664
 665 static struct tun_struct *tun_get(struct file *file)
 666 {
 667         return __tun_get(file->private_data);
 668 }
 669
 670 static void tun_put(struct tun_struct *tun)
 671 {
 672         dev_put(tun->dev);
 673 }
 674
 675 /* TAP filtering */
 676 static void addr_hash_set(u32 *mask, const u8 *addr)
 677 {
 678         int n = ether_crc(ETH_ALEN, addr) >> 26;
 679         mask[n >> 5] |= (1 << (n & 31));
 680 }
 681
 682 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 683 {
 684         int n = ether_crc(ETH_ALEN, addr) >> 26;
 685         return mask[n >> 5] & (1 << (n & 31));
 686 }
 687
 688 static int update_filter(struct tap_filter *filter, void __user *arg)
 689 {
 690         struct { u8 u[ETH_ALEN]; } *addr;
 691         struct tun_filter uf;
 692         int err, alen, n, nexact;
 693
 694         if (copy_from_user(&uf, arg, sizeof(uf)))
 695                 return -EFAULT;
 696
 697         if (!uf.count) {
 698                 /* Disabled */
 699                 filter->count = 0;
 700                 return 0;
 701         }
 702
 703         alen = ETH_ALEN * uf.count;
 704         addr = kmalloc(alen, GFP_KERNEL);
 705         if (!addr)
 706                 return -ENOMEM;
 707
 708         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 709                 err = -EFAULT;
 710                 goto done;
 711         }
 712
 713         /* The filter is updated without holding any locks. Which is
 714          * perfectly safe. We disable it first and in the worst
 715          * case we'll accept a few undesired packets. */
 716         filter->count = 0;
 717         wmb();
 718
 719         /* Use first set of addresses as an exact filter */
 720         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 721                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 722
 723         nexact = n;
 724
 725         /* Remaining multicast addresses are hashed,
 726          * unicast will leave the filter disabled. */
 727         memset(filter->mask, 0, sizeof(filter->mask));
 728         for (; n < uf.count; n++) {
 729                 if (!is_multicast_ether_addr(addr[n].u)) {
 730                         err = 0; /* no filter */
 731                         goto done;
 732                 }
 733                 addr_hash_set(filter->mask, addr[n].u);
 734         }
 735
 736         /* For ALLMULTI just set the mask to all ones.
 737          * This overrides the mask populated above. */
 738         if ((uf.flags & TUN_FLT_ALLMULTI))
 739                 memset(filter->mask, ~0, sizeof(filter->mask));
 740
 741         /* Now enable the filter */
 742         wmb();
 743         filter->count = nexact;
 744
 745         /* Return the number of exact filters */
 746         err = nexact;
 747
 748 done:
 749         kfree(addr);
 750         return err;
 751 }
 752
 753 /* Returns: 0 - drop, !=0 - accept */
 754 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 755 {
 756         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 757          * at this point. */
 758         struct ethhdr *eh = (struct ethhdr *) skb->data;
 759         int i;
 760
 761         /* Exact match */
 762         for (i = 0; i < filter->count; i++)
 763                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 764                         return 1;
 765
 766         /* Inexact match (multicast only) */
 767         if (is_multicast_ether_addr(eh->h_dest))
 768                 return addr_hash_test(filter->mask, eh->h_dest);
 769
 770         return 0;
 771 }
 772
 773 /*
 774  * Checks whether the packet is accepted or not.
 775  * Returns: 0 - drop, !=0 - accept
 776  */
 777 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 778 {
 779         if (!filter->count)
 780                 return 1;
 781
 782         return run_filter(filter, skb);
 783 }
 784
 785 /* Network device part of the driver */
 786
 787 static const struct ethtool_ops tun_ethtool_ops;
 788
 789 /* Net device detach from fd. */
 790 static void tun_net_uninit(struct net_device *dev)
 791 {
 792         tun_detach_all(dev);
 793 }
 794
 795 /* Net device open. */
 796 static int tun_net_open(struct net_device *dev)
 797 {
 798         netif_tx_start_all_queues(dev);
 799         return 0;
 800 }
 801
 802 /* Net device close. */
 803 static int tun_net_close(struct net_device *dev)
 804 {
 805         netif_tx_stop_all_queues(dev);
 806         return 0;
 807 }
 808
 809 /* Net device start xmit */
 810 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 811 {
 812         struct tun_struct *tun = netdev_priv(dev);
 813         int txq = skb->queue_mapping;
 814         struct tun_file *tfile;
 815         u32 numqueues = 0;
 816
 817         rcu_read_lock();
 818         tfile = rcu_dereference(tun->tfiles[txq]);
 819         numqueues = ACCESS_ONCE(tun->numqueues);
 820
 821         /* Drop packet if interface is not attached */
 822         if (txq >= numqueues)
 823                 goto drop;
 824
 825         if (numqueues == 1) {
 826                 /* Select queue was not called for the skbuff, so we extract the
 827                  * RPS hash and save it into the flow_table here.
 828                  */
 829                 __u32 rxhash;
 830
 831                 rxhash = skb_get_hash(skb);
 832                 if (rxhash) {
 833                         struct tun_flow_entry *e;
 834                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 835                                         rxhash);
 836                         if (e)
 837                                 tun_flow_save_rps_rxhash(e, rxhash);
 838                 }
 839         }
 840
 841         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 842
 843         BUG_ON(!tfile);
 844
 845         /* Drop if the filter does not like it.
 846          * This is a noop if the filter is disabled.
 847          * Filter can be enabled only for the TAP devices. */
 848         if (!check_filter(&tun->txflt, skb))
 849                 goto drop;
 850
 851         if (tfile->socket.sk->sk_filter &&
 852             sk_filter(tfile->socket.sk, skb))
 853                 goto drop;
 854
 855         /* Limit the number of packets queued by dividing txq length with the
 856          * number of queues.
 857          */
 858         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
 859                           >= dev->tx_queue_len)
 860                 goto drop;
 861
 862         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 863                 goto drop;
 864
 865         if (skb->sk && sk_fullsock(skb->sk)) {
 866                 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
 867                 sw_tx_timestamp(skb);
 868         }
 869
 870         /* Orphan the skb - required as we might hang on to it
 871          * for indefinite time.
 872          */
 873         skb_orphan(skb);
 874
 875         nf_reset(skb);
 876
 877         /* Enqueue packet */
 878         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 879
 880         /* Notify and wake up reader process */
 881         if (tfile->flags & TUN_FASYNC)
 882                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 883         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 884
 885         rcu_read_unlock();
 886         return NETDEV_TX_OK;
 887
 888 drop:
 889         dev->stats.tx_dropped++;
 890         skb_tx_error(skb);
 891         kfree_skb(skb);
 892         rcu_read_unlock();
 893         return NET_XMIT_DROP;
 894 }
 895
 896 static void tun_net_mclist(struct net_device *dev)
 897 {
 898         /*
 899          * This callback is supposed to deal with mc filter in
 900          * _rx_ path and has nothing to do with the _tx_ path.
 901          * In rx path we always accept everything userspace gives us.
 902          */
 903 }
 904
 905 #define MIN_MTU 68
 906 #define MAX_MTU 65535
 907
 908 static int
 909 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 910 {
 911         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 912                 return -EINVAL;
 913         dev->mtu = new_mtu;
 914         return 0;
 915 }
 916
 917 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 918         netdev_features_t features)
 919 {
 920         struct tun_struct *tun = netdev_priv(dev);
 921
 922         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 923 }
 924 #ifdef CONFIG_NET_POLL_CONTROLLER
 925 static void tun_poll_controller(struct net_device *dev)
 926 {
 927         /*
 928          * Tun only receives frames when:
 929          * 1) the char device endpoint gets data from user space
 930          * 2) the tun socket gets a sendmsg call from user space
 931          * Since both of those are synchronous operations, we are guaranteed
 932          * never to have pending data when we poll for it
 933          * so there is nothing to do here but return.
 934          * We need this though so netpoll recognizes us as an interface that
 935          * supports polling, which enables bridge devices in virt setups to
 936          * still use netconsole
 937          */
 938         return;
 939 }
 940 #endif
 941 static const struct net_device_ops tun_netdev_ops = {
 942         .ndo_uninit             = tun_net_uninit,
 943         .ndo_open               = tun_net_open,
 944         .ndo_stop               = tun_net_close,
 945         .ndo_start_xmit         = tun_net_xmit,
 946         .ndo_change_mtu         = tun_net_change_mtu,
 947         .ndo_fix_features       = tun_net_fix_features,
 948         .ndo_select_queue       = tun_select_queue,
 949 #ifdef CONFIG_NET_POLL_CONTROLLER
 950         .ndo_poll_controller    = tun_poll_controller,
 951 #endif
 952 };
 953
 954 static const struct net_device_ops tap_netdev_ops = {
 955         .ndo_uninit             = tun_net_uninit,
 956         .ndo_open               = tun_net_open,
 957         .ndo_stop               = tun_net_close,
 958         .ndo_start_xmit         = tun_net_xmit,
 959         .ndo_change_mtu         = tun_net_change_mtu,
 960         .ndo_fix_features       = tun_net_fix_features,
 961         .ndo_set_rx_mode        = tun_net_mclist,
 962         .ndo_set_mac_address    = eth_mac_addr,
 963         .ndo_validate_addr      = eth_validate_addr,
 964         .ndo_select_queue       = tun_select_queue,
 965 #ifdef CONFIG_NET_POLL_CONTROLLER
 966         .ndo_poll_controller    = tun_poll_controller,
 967 #endif
 968         .ndo_features_check     = passthru_features_check,
 969 };
 970
 971 static void tun_flow_init(struct tun_struct *tun)
 972 {
 973         int i;
 974
 975         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 976                 INIT_HLIST_HEAD(&tun->flows[i]);
 977
 978         tun->ageing_time = TUN_FLOW_EXPIRE;
 979         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 980         mod_timer(&tun->flow_gc_timer,
 981                   round_jiffies_up(jiffies + tun->ageing_time));
 982 }
 983
 984 static void tun_flow_uninit(struct tun_struct *tun)
 985 {
 986         del_timer_sync(&tun->flow_gc_timer);
 987         tun_flow_flush(tun);
 988 }
 989
 990 /* Initialize net device. */
 991 static void tun_net_init(struct net_device *dev)
 992 {
 993         struct tun_struct *tun = netdev_priv(dev);
 994
 995         switch (tun->flags & TUN_TYPE_MASK) {
 996         case IFF_TUN:
 997                 dev->netdev_ops = &tun_netdev_ops;
 998
 999                 /* Point-to-Point TUN Device */
1000                 dev->hard_header_len = 0;
1001                 dev->addr_len = 0;
1002                 dev->mtu = 1500;
1003
1004                 /* Zero header length */
1005                 dev->type = ARPHRD_NONE;
1006                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1007                 break;
1008
1009         case IFF_TAP:
1010                 dev->netdev_ops = &tap_netdev_ops;
1011                 /* Ethernet TAP Device */
1012                 ether_setup(dev);
1013                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1014                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1015
1016                 eth_hw_addr_random(dev);
1017
1018                 break;
1019         }
1020 }
1021
1022 /* Character device part */
1023
1024 /* Poll */
1025 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1026 {
1027         struct tun_file *tfile = file->private_data;
1028         struct tun_struct *tun = __tun_get(tfile);
1029         struct sock *sk;
1030         unsigned int mask = 0;
1031
1032         if (!tun)
1033                 return POLLERR;
1034
1035         sk = tfile->socket.sk;
1036
1037         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1038
1039         poll_wait(file, sk_sleep(sk), wait);
1040
1041         if (!skb_queue_empty(&sk->sk_receive_queue))
1042                 mask |= POLLIN | POLLRDNORM;
1043
1044         if (sock_writeable(sk) ||
1045             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1046              sock_writeable(sk)))
1047                 mask |= POLLOUT | POLLWRNORM;
1048
1049         if (tun->dev->reg_state != NETREG_REGISTERED)
1050                 mask = POLLERR;
1051
1052         tun_put(tun);
1053         return mask;
1054 }
1055
1056 /* prepad is the amount to reserve at front.  len is length after that.
1057  * linear is a hint as to how much to copy (usually headers). */
1058 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1059                                      size_t prepad, size_t len,
1060                                      size_t linear, int noblock)
1061 {
1062         struct sock *sk = tfile->socket.sk;
1063         struct sk_buff *skb;
1064         int err;
1065
1066         /* Under a page?  Don't bother with paged skb. */
1067         if (prepad + len < PAGE_SIZE || !linear)
1068                 linear = len;
1069
1070         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1071                                    &err, 0);
1072         if (!skb)
1073                 return ERR_PTR(err);
1074
1075         skb_reserve(skb, prepad);
1076         skb_put(skb, linear);
1077         skb->data_len = len - linear;
1078         skb->len += len - linear;
1079
1080         return skb;
1081 }
1082
1083 /* Get packet from user space buffer */
1084 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1085                             void *msg_control, struct iov_iter *from,
1086                             int noblock)
1087 {
1088         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1089         struct sk_buff *skb;
1090         size_t total_len = iov_iter_count(from);
1091         size_t len = total_len, align = NET_SKB_PAD, linear;
1092         struct virtio_net_hdr gso = { 0 };
1093         int good_linear;
1094         int copylen;
1095         bool zerocopy = false;
1096         int err;
1097         u32 rxhash;
1098         ssize_t n;
1099
1100         if (!(tun->flags & IFF_NO_PI)) {
1101                 if (len < sizeof(pi))
1102                         return -EINVAL;
1103                 len -= sizeof(pi);
1104
1105                 n = copy_from_iter(&pi, sizeof(pi), from);
1106                 if (n != sizeof(pi))
1107                         return -EFAULT;
1108         }
1109
1110         if (tun->flags & IFF_VNET_HDR) {
1111                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1112
1113                 if (len < vnet_hdr_sz)
1114                         return -EINVAL;
1115                 len -= vnet_hdr_sz;
1116
1117                 n = copy_from_iter(&gso, sizeof(gso), from);
1118                 if (n != sizeof(gso))
1119                         return -EFAULT;
1120
1121                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1122                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1123                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1124
1125                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1126                         return -EINVAL;
1127                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1128         }
1129
1130         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1131                 align += NET_IP_ALIGN;
1132                 if (unlikely(len < ETH_HLEN ||
1133                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1134                         return -EINVAL;
1135         }
1136
1137         good_linear = SKB_MAX_HEAD(align);
1138
1139         if (msg_control) {
1140                 struct iov_iter i = *from;
1141
1142                 /* There are 256 bytes to be copied in skb, so there is
1143                  * enough room for skb expand head in case it is used.
1144                  * The rest of the buffer is mapped from userspace.
1145                  */
1146                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1147                 if (copylen > good_linear)
1148                         copylen = good_linear;
1149                 linear = copylen;
1150                 iov_iter_advance(&i, copylen);
1151                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1152                         zerocopy = true;
1153         }
1154
1155         if (!zerocopy) {
1156                 copylen = len;
1157                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1158                         linear = good_linear;
1159                 else
1160                         linear = tun16_to_cpu(tun, gso.hdr_len);
1161         }
1162
1163         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1164         if (IS_ERR(skb)) {
1165                 if (PTR_ERR(skb) != -EAGAIN)
1166                         tun->dev->stats.rx_dropped++;
1167                 return PTR_ERR(skb);
1168         }
1169
1170         if (zerocopy)
1171                 err = zerocopy_sg_from_iter(skb, from);
1172         else {
1173                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1174                 if (!err && msg_control) {
1175                         struct ubuf_info *uarg = msg_control;
1176                         uarg->callback(uarg, false);
1177                 }
1178         }
1179
1180         if (err) {
1181                 tun->dev->stats.rx_dropped++;
1182                 kfree_skb(skb);
1183                 return -EFAULT;
1184         }
1185
1186         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1187                 if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start),
1188                                           tun16_to_cpu(tun, gso.csum_offset))) {
1189                         tun->dev->stats.rx_frame_errors++;
1190                         kfree_skb(skb);
1191                         return -EINVAL;
1192                 }
1193         }
1194
1195         switch (tun->flags & TUN_TYPE_MASK) {
1196         case IFF_TUN:
1197                 if (tun->flags & IFF_NO_PI) {
1198                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1199
1200                         switch (ip_version) {
1201                         case 4:
1202                                 pi.proto = htons(ETH_P_IP);
1203                                 break;
1204                         case 6:
1205                                 pi.proto = htons(ETH_P_IPV6);
1206                                 break;
1207                         default:
1208                                 tun->dev->stats.rx_dropped++;
1209                                 kfree_skb(skb);
1210                                 return -EINVAL;
1211                         }
1212                 }
1213
1214                 skb_reset_mac_header(skb);
1215                 skb->protocol = pi.proto;
1216                 skb->dev = tun->dev;
1217                 break;
1218         case IFF_TAP:
1219                 skb->protocol = eth_type_trans(skb, tun->dev);
1220                 break;
1221         }
1222
1223         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1224                 pr_debug("GSO!\n");
1225                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1226                 case VIRTIO_NET_HDR_GSO_TCPV4:
1227                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1228                         break;
1229                 case VIRTIO_NET_HDR_GSO_TCPV6:
1230                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1231                         break;
1232                 case VIRTIO_NET_HDR_GSO_UDP:
1233                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1234                         break;
1235                 default:
1236                         tun->dev->stats.rx_frame_errors++;
1237                         kfree_skb(skb);
1238                         return -EINVAL;
1239                 }
1240
1241                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1242                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1243
1244                 skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size);
1245                 if (skb_shinfo(skb)->gso_size == 0) {
1246                         tun->dev->stats.rx_frame_errors++;
1247                         kfree_skb(skb);
1248                         return -EINVAL;
1249                 }
1250
1251                 /* Header must be checked, and gso_segs computed. */
1252                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1253                 skb_shinfo(skb)->gso_segs = 0;
1254         }
1255
1256         /* copy skb_ubuf_info for callback when skb has no error */
1257         if (zerocopy) {
1258                 skb_shinfo(skb)->destructor_arg = msg_control;
1259                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1260                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1261         }
1262
1263         skb_reset_network_header(skb);
1264         skb_probe_transport_header(skb, 0);
1265
1266         rxhash = skb_get_hash(skb);
1267         netif_rx_ni(skb);
1268
1269         tun->dev->stats.rx_packets++;
1270         tun->dev->stats.rx_bytes += len;
1271
1272         tun_flow_update(tun, rxhash, tfile);
1273         return total_len;
1274 }
1275
1276 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1277 {
1278         struct file *file = iocb->ki_filp;
1279         struct tun_struct *tun = tun_get(file);
1280         struct tun_file *tfile = file->private_data;
1281         ssize_t result;
1282
1283         if (!tun)
1284                 return -EBADFD;
1285
1286         result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1287
1288         tun_put(tun);
1289         return result;
1290 }
1291
1292 /* Put packet to the user space buffer */
1293 static ssize_t tun_put_user(struct tun_struct *tun,
1294                             struct tun_file *tfile,
1295                             struct sk_buff *skb,
1296                             struct iov_iter *iter)
1297 {
1298         struct tun_pi pi = { 0, skb->protocol };
1299         ssize_t total;
1300         int vlan_offset = 0;
1301         int vlan_hlen = 0;
1302         int vnet_hdr_sz = 0;
1303
1304         if (skb_vlan_tag_present(skb))
1305                 vlan_hlen = VLAN_HLEN;
1306
1307         if (tun->flags & IFF_VNET_HDR)
1308                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1309
1310         total = skb->len + vlan_hlen + vnet_hdr_sz;
1311
1312         if (!(tun->flags & IFF_NO_PI)) {
1313                 if (iov_iter_count(iter) < sizeof(pi))
1314                         return -EINVAL;
1315
1316                 total += sizeof(pi);
1317                 if (iov_iter_count(iter) < total) {
1318                         /* Packet will be striped */
1319                         pi.flags |= TUN_PKT_STRIP;
1320                 }
1321
1322                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1323                         return -EFAULT;
1324         }
1325
1326         if (vnet_hdr_sz) {
1327                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1328                 if (iov_iter_count(iter) < vnet_hdr_sz)
1329                         return -EINVAL;
1330
1331                 if (skb_is_gso(skb)) {
1332                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1333
1334                         /* This is a hint as to how much should be linear. */
1335                         gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb));
1336                         gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size);
1337                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1338                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1339                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1340                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1341                         else if (sinfo->gso_type & SKB_GSO_UDP)
1342                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1343                         else {
1344                                 pr_err("unexpected GSO type: "
1345                                        "0x%x, gso_size %d, hdr_len %d\n",
1346                                        sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1347                                        tun16_to_cpu(tun, gso.hdr_len));
1348                                 print_hex_dump(KERN_ERR, "tun: ",
1349                                                DUMP_PREFIX_NONE,
1350                                                16, 1, skb->head,
1351                                                min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1352                                 WARN_ON_ONCE(1);
1353                                 return -EINVAL;
1354                         }
1355                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1356                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1357                 } else
1358                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1359
1360                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1361                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1362                         gso.csum_start = cpu_to_tun16(tun, skb_checksum_start_offset(skb) +
1363                                                       vlan_hlen);
1364                         gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset);
1365                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1366                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1367                 } /* else everything is zero */
1368
1369                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1370                         return -EFAULT;
1371
1372                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1373         }
1374
1375         if (vlan_hlen) {
1376                 int ret;
1377                 struct {
1378                         __be16 h_vlan_proto;
1379                         __be16 h_vlan_TCI;
1380                 } veth;
1381
1382                 veth.h_vlan_proto = skb->vlan_proto;
1383                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1384
1385                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1386
1387                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1388                 if (ret || !iov_iter_count(iter))
1389                         goto done;
1390
1391                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1392                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1393                         goto done;
1394         }
1395
1396         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1397
1398 done:
1399         tun->dev->stats.tx_packets++;
1400         tun->dev->stats.tx_bytes += skb->len + vlan_hlen;
1401
1402         return total;
1403 }
1404
1405 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1406                            struct iov_iter *to,
1407                            int noblock)
1408 {
1409         struct sk_buff *skb;
1410         ssize_t ret;
1411         int peeked, err, off = 0;
1412
1413         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1414
1415         if (!iov_iter_count(to))
1416                 return 0;
1417
1418         /* Read frames from queue */
1419         skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1420                                   &peeked, &off, &err);
1421         if (!skb)
1422                 return err;
1423
1424         ret = tun_put_user(tun, tfile, skb, to);
1425         if (unlikely(ret < 0))
1426                 kfree_skb(skb);
1427         else
1428                 consume_skb(skb);
1429
1430         return ret;
1431 }
1432
1433 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1434 {
1435         struct file *file = iocb->ki_filp;
1436         struct tun_file *tfile = file->private_data;
1437         struct tun_struct *tun = __tun_get(tfile);
1438         ssize_t len = iov_iter_count(to), ret;
1439
1440         if (!tun)
1441                 return -EBADFD;
1442         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1443         ret = min_t(ssize_t, ret, len);
1444         if (ret > 0)
1445                 iocb->ki_pos = ret;
1446         tun_put(tun);
1447         return ret;
1448 }
1449
1450 static void tun_free_netdev(struct net_device *dev)
1451 {
1452         struct tun_struct *tun = netdev_priv(dev);
1453
1454         BUG_ON(!(list_empty(&tun->disabled)));
1455         tun_flow_uninit(tun);
1456         security_tun_dev_free_security(tun->security);
1457         free_netdev(dev);
1458 }
1459
1460 static void tun_setup(struct net_device *dev)
1461 {
1462         struct tun_struct *tun = netdev_priv(dev);
1463
1464         tun->owner = INVALID_UID;
1465         tun->group = INVALID_GID;
1466
1467         dev->ethtool_ops = &tun_ethtool_ops;
1468         dev->destructor = tun_free_netdev;
1469         /* We prefer our own queue length */
1470         dev->tx_queue_len = TUN_READQ_SIZE;
1471 }
1472
1473 /* Trivial set of netlink ops to allow deleting tun or tap
1474  * device with netlink.
1475  */
1476 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1477 {
1478         if (!data)
1479                 return 0;
1480         return -EINVAL;
1481 }
1482
1483 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1484         .kind           = DRV_NAME,
1485         .priv_size      = sizeof(struct tun_struct),
1486         .setup          = tun_setup,
1487         .validate       = tun_validate,
1488 };
1489
1490 static void tun_sock_write_space(struct sock *sk)
1491 {
1492         struct tun_file *tfile;
1493         wait_queue_head_t *wqueue;
1494
1495         if (!sock_writeable(sk))
1496                 return;
1497
1498         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1499                 return;
1500
1501         wqueue = sk_sleep(sk);
1502         if (wqueue && waitqueue_active(wqueue))
1503                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1504                                                 POLLWRNORM | POLLWRBAND);
1505
1506         tfile = container_of(sk, struct tun_file, sk);
1507         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1508 }
1509
1510 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1511 {
1512         int ret;
1513         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1514         struct tun_struct *tun = __tun_get(tfile);
1515
1516         if (!tun)
1517                 return -EBADFD;
1518
1519         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1520                            m->msg_flags & MSG_DONTWAIT);
1521         tun_put(tun);
1522         return ret;
1523 }
1524
1525 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1526                        int flags)
1527 {
1528         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1529         struct tun_struct *tun = __tun_get(tfile);
1530         int ret;
1531
1532         if (!tun)
1533                 return -EBADFD;
1534
1535         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1536                 ret = -EINVAL;
1537                 goto out;
1538         }
1539         if (flags & MSG_ERRQUEUE) {
1540                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1541                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1542                 goto out;
1543         }
1544         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1545         if (ret > (ssize_t)total_len) {
1546                 m->msg_flags |= MSG_TRUNC;
1547                 ret = flags & MSG_TRUNC ? ret : total_len;
1548         }
1549 out:
1550         tun_put(tun);
1551         return ret;
1552 }
1553
1554 /* Ops structure to mimic raw sockets with tun */
1555 static const struct proto_ops tun_socket_ops = {
1556         .sendmsg = tun_sendmsg,
1557         .recvmsg = tun_recvmsg,
1558 };
1559
1560 static struct proto tun_proto = {
1561         .name           = "tun",
1562         .owner          = THIS_MODULE,
1563         .obj_size       = sizeof(struct tun_file),
1564 };
1565
1566 static int tun_flags(struct tun_struct *tun)
1567 {
1568         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1569 }
1570
1571 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1572                               char *buf)
1573 {
1574         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1575         return sprintf(buf, "0x%x\n", tun_flags(tun));
1576 }
1577
1578 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1579                               char *buf)
1580 {
1581         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1582         return uid_valid(tun->owner)?
1583                 sprintf(buf, "%u\n",
1584                         from_kuid_munged(current_user_ns(), tun->owner)):
1585                 sprintf(buf, "-1\n");
1586 }
1587
1588 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1589                               char *buf)
1590 {
1591         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1592         return gid_valid(tun->group) ?
1593                 sprintf(buf, "%u\n",
1594                         from_kgid_munged(current_user_ns(), tun->group)):
1595                 sprintf(buf, "-1\n");
1596 }
1597
1598 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1599 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1600 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1601
1602 static struct attribute *tun_dev_attrs[] = {
1603         &dev_attr_tun_flags.attr,
1604         &dev_attr_owner.attr,
1605         &dev_attr_group.attr,
1606         NULL
1607 };
1608
1609 static const struct attribute_group tun_attr_group = {
1610         .attrs = tun_dev_attrs
1611 };
1612
1613 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1614 {
1615         struct tun_struct *tun;
1616         struct tun_file *tfile = file->private_data;
1617         struct net_device *dev;
1618         int err;
1619
1620         if (tfile->detached)
1621                 return -EINVAL;
1622
1623         dev = __dev_get_by_name(net, ifr->ifr_name);
1624         if (dev) {
1625                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1626                         return -EBUSY;
1627                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1628                         tun = netdev_priv(dev);
1629                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1630                         tun = netdev_priv(dev);
1631                 else
1632                         return -EINVAL;
1633
1634                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1635                     !!(tun->flags & IFF_MULTI_QUEUE))
1636                         return -EINVAL;
1637
1638                 if (tun_not_capable(tun))
1639                         return -EPERM;
1640                 err = security_tun_dev_open(tun->security);
1641                 if (err < 0)
1642                         return err;
1643
1644                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1645                 if (err < 0)
1646                         return err;
1647
1648                 if (tun->flags & IFF_MULTI_QUEUE &&
1649                     (tun->numqueues + tun->numdisabled > 1)) {
1650                         /* One or more queue has already been attached, no need
1651                          * to initialize the device again.
1652                          */
1653                         return 0;
1654                 }
1655         }
1656         else {
1657                 char *name;
1658                 unsigned long flags = 0;
1659                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1660                              MAX_TAP_QUEUES : 1;
1661
1662                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1663                         return -EPERM;
1664                 err = security_tun_dev_create();
1665                 if (err < 0)
1666                         return err;
1667
1668                 /* Set dev type */
1669                 if (ifr->ifr_flags & IFF_TUN) {
1670                         /* TUN device */
1671                         flags |= IFF_TUN;
1672                         name = "tun%d";
1673                 } else if (ifr->ifr_flags & IFF_TAP) {
1674                         /* TAP device */
1675                         flags |= IFF_TAP;
1676                         name = "tap%d";
1677                 } else
1678                         return -EINVAL;
1679
1680                 if (*ifr->ifr_name)
1681                         name = ifr->ifr_name;
1682
1683                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1684                                        NET_NAME_UNKNOWN, tun_setup, queues,
1685                                        queues);
1686
1687                 if (!dev)
1688                         return -ENOMEM;
1689                 err = dev_get_valid_name(net, dev, name);
1690                 if (err < 0)
1691                         goto err_free_dev;
1692
1693                 dev_net_set(dev, net);
1694                 dev->rtnl_link_ops = &tun_link_ops;
1695                 dev->ifindex = tfile->ifindex;
1696                 dev->sysfs_groups[0] = &tun_attr_group;
1697
1698                 tun = netdev_priv(dev);
1699                 tun->dev = dev;
1700                 tun->flags = flags;
1701                 tun->txflt.count = 0;
1702                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1703
1704                 tun->filter_attached = false;
1705                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1706
1707                 spin_lock_init(&tun->lock);
1708
1709                 err = security_tun_dev_alloc_security(&tun->security);
1710                 if (err < 0)
1711                         goto err_free_dev;
1712
1713                 tun_net_init(dev);
1714                 tun_flow_init(tun);
1715
1716                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1717                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1718                                    NETIF_F_HW_VLAN_STAG_TX;
1719                 dev->features = dev->hw_features;
1720                 dev->vlan_features = dev->features &
1721                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1722                                        NETIF_F_HW_VLAN_STAG_TX);
1723
1724                 INIT_LIST_HEAD(&tun->disabled);
1725                 err = tun_attach(tun, file, false);
1726                 if (err < 0)
1727                         goto err_free_flow;
1728
1729                 err = register_netdevice(tun->dev);
1730                 if (err < 0)
1731                         goto err_detach;
1732         }
1733
1734         netif_carrier_on(tun->dev);
1735
1736         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1737
1738         tun->flags = (tun->flags & ~TUN_FEATURES) |
1739                 (ifr->ifr_flags & TUN_FEATURES);
1740
1741         /* Make sure persistent devices do not get stuck in
1742          * xoff state.
1743          */
1744         if (netif_running(tun->dev))
1745                 netif_tx_wake_all_queues(tun->dev);
1746
1747         strcpy(ifr->ifr_name, tun->dev->name);
1748         return 0;
1749
1750 err_detach:
1751         tun_detach_all(dev);
1752 err_free_flow:
1753         tun_flow_uninit(tun);
1754         security_tun_dev_free_security(tun->security);
1755 err_free_dev:
1756         free_netdev(dev);
1757         return err;
1758 }
1759
1760 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1761                        struct ifreq *ifr)
1762 {
1763         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1764
1765         strcpy(ifr->ifr_name, tun->dev->name);
1766
1767         ifr->ifr_flags = tun_flags(tun);
1768
1769 }
1770
1771 /* This is like a cut-down ethtool ops, except done via tun fd so no
1772  * privs required. */
1773 static int set_offload(struct tun_struct *tun, unsigned long arg)
1774 {
1775         netdev_features_t features = 0;
1776
1777         if (arg & TUN_F_CSUM) {
1778                 features |= NETIF_F_HW_CSUM;
1779                 arg &= ~TUN_F_CSUM;
1780
1781                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1782                         if (arg & TUN_F_TSO_ECN) {
1783                                 features |= NETIF_F_TSO_ECN;
1784                                 arg &= ~TUN_F_TSO_ECN;
1785                         }
1786                         if (arg & TUN_F_TSO4)
1787                                 features |= NETIF_F_TSO;
1788                         if (arg & TUN_F_TSO6)
1789                                 features |= NETIF_F_TSO6;
1790                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1791                 }
1792
1793                 if (arg & TUN_F_UFO) {
1794                         features |= NETIF_F_UFO;
1795                         arg &= ~TUN_F_UFO;
1796                 }
1797         }
1798
1799         /* This gives the user a way to test for new features in future by
1800          * trying to set them. */
1801         if (arg)
1802                 return -EINVAL;
1803
1804         tun->set_features = features;
1805         netdev_update_features(tun->dev);
1806
1807         return 0;
1808 }
1809
1810 static void tun_detach_filter(struct tun_struct *tun, int n)
1811 {
1812         int i;
1813         struct tun_file *tfile;
1814
1815         for (i = 0; i < n; i++) {
1816                 tfile = rtnl_dereference(tun->tfiles[i]);
1817                 __sk_detach_filter(tfile->socket.sk, lockdep_rtnl_is_held());
1818         }
1819
1820         tun->filter_attached = false;
1821 }
1822
1823 static int tun_attach_filter(struct tun_struct *tun)
1824 {
1825         int i, ret = 0;
1826         struct tun_file *tfile;
1827
1828         for (i = 0; i < tun->numqueues; i++) {
1829                 tfile = rtnl_dereference(tun->tfiles[i]);
1830                 ret = __sk_attach_filter(&tun->fprog, tfile->socket.sk,
1831                                          lockdep_rtnl_is_held());
1832                 if (ret) {
1833                         tun_detach_filter(tun, i);
1834                         return ret;
1835                 }
1836         }
1837
1838         tun->filter_attached = true;
1839         return ret;
1840 }
1841
1842 static void tun_set_sndbuf(struct tun_struct *tun)
1843 {
1844         struct tun_file *tfile;
1845         int i;
1846
1847         for (i = 0; i < tun->numqueues; i++) {
1848                 tfile = rtnl_dereference(tun->tfiles[i]);
1849                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1850         }
1851 }
1852
1853 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1854 {
1855         struct tun_file *tfile = file->private_data;
1856         struct tun_struct *tun;
1857         int ret = 0;
1858
1859         rtnl_lock();
1860
1861         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1862                 tun = tfile->detached;
1863                 if (!tun) {
1864                         ret = -EINVAL;
1865                         goto unlock;
1866                 }
1867                 ret = security_tun_dev_attach_queue(tun->security);
1868                 if (ret < 0)
1869                         goto unlock;
1870                 ret = tun_attach(tun, file, false);
1871         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1872                 tun = rtnl_dereference(tfile->tun);
1873                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1874                         ret = -EINVAL;
1875                 else
1876                         __tun_detach(tfile, false);
1877         } else
1878                 ret = -EINVAL;
1879
1880 unlock:
1881         rtnl_unlock();
1882         return ret;
1883 }
1884
1885 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1886                             unsigned long arg, int ifreq_len)
1887 {
1888         struct tun_file *tfile = file->private_data;
1889         struct tun_struct *tun;
1890         void __user* argp = (void __user*)arg;
1891         struct ifreq ifr;
1892         kuid_t owner;
1893         kgid_t group;
1894         int sndbuf;
1895         int vnet_hdr_sz;
1896         unsigned int ifindex;
1897         int le;
1898         int ret;
1899
1900         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1901                 if (copy_from_user(&ifr, argp, ifreq_len))
1902                         return -EFAULT;
1903         } else {
1904                 memset(&ifr, 0, sizeof(ifr));
1905         }
1906         if (cmd == TUNGETFEATURES) {
1907                 /* Currently this just means: "what IFF flags are valid?".
1908                  * This is needed because we never checked for invalid flags on
1909                  * TUNSETIFF.
1910                  */
1911                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1912                                 (unsigned int __user*)argp);
1913         } else if (cmd == TUNSETQUEUE)
1914                 return tun_set_queue(file, &ifr);
1915
1916         ret = 0;
1917         rtnl_lock();
1918
1919         tun = __tun_get(tfile);
1920         if (cmd == TUNSETIFF && !tun) {
1921                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1922
1923                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
1924
1925                 if (ret)
1926                         goto unlock;
1927
1928                 if (copy_to_user(argp, &ifr, ifreq_len))
1929                         ret = -EFAULT;
1930                 goto unlock;
1931         }
1932         if (cmd == TUNSETIFINDEX) {
1933                 ret = -EPERM;
1934                 if (tun)
1935                         goto unlock;
1936
1937                 ret = -EFAULT;
1938                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1939                         goto unlock;
1940
1941                 ret = 0;
1942                 tfile->ifindex = ifindex;
1943                 goto unlock;
1944         }
1945
1946         ret = -EBADFD;
1947         if (!tun)
1948                 goto unlock;
1949
1950         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1951
1952         ret = 0;
1953         switch (cmd) {
1954         case TUNGETIFF:
1955                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1956
1957                 if (tfile->detached)
1958                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
1959                 if (!tfile->socket.sk->sk_filter)
1960                         ifr.ifr_flags |= IFF_NOFILTER;
1961
1962                 if (copy_to_user(argp, &ifr, ifreq_len))
1963                         ret = -EFAULT;
1964                 break;
1965
1966         case TUNSETNOCSUM:
1967                 /* Disable/Enable checksum */
1968
1969                 /* [unimplemented] */
1970                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1971                           arg ? "disabled" : "enabled");
1972                 break;
1973
1974         case TUNSETPERSIST:
1975                 /* Disable/Enable persist mode. Keep an extra reference to the
1976                  * module to prevent the module being unprobed.
1977                  */
1978                 if (arg && !(tun->flags & IFF_PERSIST)) {
1979                         tun->flags |= IFF_PERSIST;
1980                         __module_get(THIS_MODULE);
1981                 }
1982                 if (!arg && (tun->flags & IFF_PERSIST)) {
1983                         tun->flags &= ~IFF_PERSIST;
1984                         module_put(THIS_MODULE);
1985                 }
1986
1987                 tun_debug(KERN_INFO, tun, "persist %s\n",
1988                           arg ? "enabled" : "disabled");
1989                 break;
1990
1991         case TUNSETOWNER:
1992                 /* Set owner of the device */
1993                 owner = make_kuid(current_user_ns(), arg);
1994                 if (!uid_valid(owner)) {
1995                         ret = -EINVAL;
1996                         break;
1997                 }
1998                 tun->owner = owner;
1999                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2000                           from_kuid(&init_user_ns, tun->owner));
2001                 break;
2002
2003         case TUNSETGROUP:
2004                 /* Set group of the device */
2005                 group = make_kgid(current_user_ns(), arg);
2006                 if (!gid_valid(group)) {
2007                         ret = -EINVAL;
2008                         break;
2009                 }
2010                 tun->group = group;
2011                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2012                           from_kgid(&init_user_ns, tun->group));
2013                 break;
2014
2015         case TUNSETLINK:
2016                 /* Only allow setting the type when the interface is down */
2017                 if (tun->dev->flags & IFF_UP) {
2018                         tun_debug(KERN_INFO, tun,
2019                                   "Linktype set failed because interface is up\n");
2020                         ret = -EBUSY;
2021                 } else {
2022                         tun->dev->type = (int) arg;
2023                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2024                                   tun->dev->type);
2025                         ret = 0;
2026                 }
2027                 break;
2028
2029 #ifdef TUN_DEBUG
2030         case TUNSETDEBUG:
2031                 tun->debug = arg;
2032                 break;
2033 #endif
2034         case TUNSETOFFLOAD:
2035                 ret = set_offload(tun, arg);
2036                 break;
2037
2038         case TUNSETTXFILTER:
2039                 /* Can be set only for TAPs */
2040                 ret = -EINVAL;
2041                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2042                         break;
2043                 ret = update_filter(&tun->txflt, (void __user *)arg);
2044                 break;
2045
2046         case SIOCGIFHWADDR:
2047                 /* Get hw address */
2048                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2049                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2050                 if (copy_to_user(argp, &ifr, ifreq_len))
2051                         ret = -EFAULT;
2052                 break;
2053
2054         case SIOCSIFHWADDR:
2055                 /* Set hw address */
2056                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2057                           ifr.ifr_hwaddr.sa_data);
2058
2059                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2060                 break;
2061
2062         case TUNGETSNDBUF:
2063                 sndbuf = tfile->socket.sk->sk_sndbuf;
2064                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2065                         ret = -EFAULT;
2066                 break;
2067
2068         case TUNSETSNDBUF:
2069                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2070                         ret = -EFAULT;
2071                         break;
2072                 }
2073                 if (sndbuf <= 0) {
2074                         ret = -EINVAL;
2075                         break;
2076                 }
2077
2078                 tun->sndbuf = sndbuf;
2079                 tun_set_sndbuf(tun);
2080                 break;
2081
2082         case TUNGETVNETHDRSZ:
2083                 vnet_hdr_sz = tun->vnet_hdr_sz;
2084                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2085                         ret = -EFAULT;
2086                 break;
2087
2088         case TUNSETVNETHDRSZ:
2089                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2090                         ret = -EFAULT;
2091                         break;
2092                 }
2093                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2094                         ret = -EINVAL;
2095                         break;
2096                 }
2097
2098                 tun->vnet_hdr_sz = vnet_hdr_sz;
2099                 break;
2100
2101         case TUNGETVNETLE:
2102                 le = !!(tun->flags & TUN_VNET_LE);
2103                 if (put_user(le, (int __user *)argp))
2104                         ret = -EFAULT;
2105                 break;
2106
2107         case TUNSETVNETLE:
2108                 if (get_user(le, (int __user *)argp)) {
2109                         ret = -EFAULT;
2110                         break;
2111                 }
2112                 if (le)
2113                         tun->flags |= TUN_VNET_LE;
2114                 else
2115                         tun->flags &= ~TUN_VNET_LE;
2116                 break;
2117
2118         case TUNGETVNETBE:
2119                 ret = tun_get_vnet_be(tun, argp);
2120                 break;
2121
2122         case TUNSETVNETBE:
2123                 ret = tun_set_vnet_be(tun, argp);
2124                 break;
2125
2126         case TUNATTACHFILTER:
2127                 /* Can be set only for TAPs */
2128                 ret = -EINVAL;
2129                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2130                         break;
2131                 ret = -EFAULT;
2132                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2133                         break;
2134
2135                 ret = tun_attach_filter(tun);
2136                 break;
2137
2138         case TUNDETACHFILTER:
2139                 /* Can be set only for TAPs */
2140                 ret = -EINVAL;
2141                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2142                         break;
2143                 ret = 0;
2144                 tun_detach_filter(tun, tun->numqueues);
2145                 break;
2146
2147         case TUNGETFILTER:
2148                 ret = -EINVAL;
2149                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2150                         break;
2151                 ret = -EFAULT;
2152                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2153                         break;
2154                 ret = 0;
2155                 break;
2156
2157         default:
2158                 ret = -EINVAL;
2159                 break;
2160         }
2161
2162 unlock:
2163         rtnl_unlock();
2164         if (tun)
2165                 tun_put(tun);
2166         return ret;
2167 }
2168
2169 static long tun_chr_ioctl(struct file *file,
2170                           unsigned int cmd, unsigned long arg)
2171 {
2172         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2173 }
2174
2175 #ifdef CONFIG_COMPAT
2176 static long tun_chr_compat_ioctl(struct file *file,
2177                          unsigned int cmd, unsigned long arg)
2178 {
2179         switch (cmd) {
2180         case TUNSETIFF:
2181         case TUNGETIFF:
2182         case TUNSETTXFILTER:
2183         case TUNGETSNDBUF:
2184         case TUNSETSNDBUF:
2185         case SIOCGIFHWADDR:
2186         case SIOCSIFHWADDR:
2187                 arg = (unsigned long)compat_ptr(arg);
2188                 break;
2189         default:
2190                 arg = (compat_ulong_t)arg;
2191                 break;
2192         }
2193
2194         /*
2195          * compat_ifreq is shorter than ifreq, so we must not access beyond
2196          * the end of that structure. All fields that are used in this
2197          * driver are compatible though, we don't need to convert the
2198          * contents.
2199          */
2200         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2201 }
2202 #endif /* CONFIG_COMPAT */
2203
2204 static int tun_chr_fasync(int fd, struct file *file, int on)
2205 {
2206         struct tun_file *tfile = file->private_data;
2207         int ret;
2208
2209         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2210                 goto out;
2211
2212         if (on) {
2213                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2214                 tfile->flags |= TUN_FASYNC;
2215         } else
2216                 tfile->flags &= ~TUN_FASYNC;
2217         ret = 0;
2218 out:
2219         return ret;
2220 }
2221
2222 static int tun_chr_open(struct inode *inode, struct file * file)
2223 {
2224         struct net *net = current->nsproxy->net_ns;
2225         struct tun_file *tfile;
2226
2227         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2228
2229         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2230                                             &tun_proto, 0);
2231         if (!tfile)
2232                 return -ENOMEM;
2233         RCU_INIT_POINTER(tfile->tun, NULL);
2234         tfile->flags = 0;
2235         tfile->ifindex = 0;
2236
2237         init_waitqueue_head(&tfile->wq.wait);
2238         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2239
2240         tfile->socket.file = file;
2241         tfile->socket.ops = &tun_socket_ops;
2242
2243         sock_init_data(&tfile->socket, &tfile->sk);
2244
2245         tfile->sk.sk_write_space = tun_sock_write_space;
2246         tfile->sk.sk_sndbuf = INT_MAX;
2247
2248         file->private_data = tfile;
2249         INIT_LIST_HEAD(&tfile->next);
2250
2251         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2252
2253         return 0;
2254 }
2255
2256 static int tun_chr_close(struct inode *inode, struct file *file)
2257 {
2258         struct tun_file *tfile = file->private_data;
2259
2260         tun_detach(tfile, true);
2261
2262         return 0;
2263 }
2264
2265 #ifdef CONFIG_PROC_FS
2266 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2267 {
2268         struct tun_struct *tun;
2269         struct ifreq ifr;
2270
2271         memset(&ifr, 0, sizeof(ifr));
2272
2273         rtnl_lock();
2274         tun = tun_get(f);
2275         if (tun)
2276                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2277         rtnl_unlock();
2278
2279         if (tun)
2280                 tun_put(tun);
2281
2282         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2283 }
2284 #endif
2285
2286 static const struct file_operations tun_fops = {
2287         .owner  = THIS_MODULE,
2288         .llseek = no_llseek,
2289         .read_iter  = tun_chr_read_iter,
2290         .write_iter = tun_chr_write_iter,
2291         .poll   = tun_chr_poll,
2292         .unlocked_ioctl = tun_chr_ioctl,
2293 #ifdef CONFIG_COMPAT
2294         .compat_ioctl = tun_chr_compat_ioctl,
2295 #endif
2296         .open   = tun_chr_open,
2297         .release = tun_chr_close,
2298         .fasync = tun_chr_fasync,
2299 #ifdef CONFIG_PROC_FS
2300         .show_fdinfo = tun_chr_show_fdinfo,
2301 #endif
2302 };
2303
2304 static struct miscdevice tun_miscdev = {
2305         .minor = TUN_MINOR,
2306         .name = "tun",
2307         .nodename = "net/tun",
2308         .fops = &tun_fops,
2309 };
2310
2311 /* ethtool interface */
2312
2313 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2314 {
2315         cmd->supported          = 0;
2316         cmd->advertising        = 0;
2317         ethtool_cmd_speed_set(cmd, SPEED_10);
2318         cmd->duplex             = DUPLEX_FULL;
2319         cmd->port               = PORT_TP;
2320         cmd->phy_address        = 0;
2321         cmd->transceiver        = XCVR_INTERNAL;
2322         cmd->autoneg            = AUTONEG_DISABLE;
2323         cmd->maxtxpkt           = 0;
2324         cmd->maxrxpkt           = 0;
2325         return 0;
2326 }
2327
2328 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2329 {
2330         struct tun_struct *tun = netdev_priv(dev);
2331
2332         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2333         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2334
2335         switch (tun->flags & TUN_TYPE_MASK) {
2336         case IFF_TUN:
2337                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2338                 break;
2339         case IFF_TAP:
2340                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2341                 break;
2342         }
2343 }
2344
2345 static u32 tun_get_msglevel(struct net_device *dev)
2346 {
2347 #ifdef TUN_DEBUG
2348         struct tun_struct *tun = netdev_priv(dev);
2349         return tun->debug;
2350 #else
2351         return -EOPNOTSUPP;
2352 #endif
2353 }
2354
2355 static void tun_set_msglevel(struct net_device *dev, u32 value)
2356 {
2357 #ifdef TUN_DEBUG
2358         struct tun_struct *tun = netdev_priv(dev);
2359         tun->debug = value;
2360 #endif
2361 }
2362
2363 static const struct ethtool_ops tun_ethtool_ops = {
2364         .get_settings   = tun_get_settings,
2365         .get_drvinfo    = tun_get_drvinfo,
2366         .get_msglevel   = tun_get_msglevel,
2367         .set_msglevel   = tun_set_msglevel,
2368         .get_link       = ethtool_op_get_link,
2369         .get_ts_info    = ethtool_op_get_ts_info,
2370 };
2371
2372
2373 static int __init tun_init(void)
2374 {
2375         int ret = 0;
2376
2377         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2378         pr_info("%s\n", DRV_COPYRIGHT);
2379
2380         ret = rtnl_link_register(&tun_link_ops);
2381         if (ret) {
2382                 pr_err("Can't register link_ops\n");
2383                 goto err_linkops;
2384         }
2385
2386         ret = misc_register(&tun_miscdev);
2387         if (ret) {
2388                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2389                 goto err_misc;
2390         }
2391         return  0;
2392 err_misc:
2393         rtnl_link_unregister(&tun_link_ops);
2394 err_linkops:
2395         return ret;
2396 }
2397
2398 static void tun_cleanup(void)
2399 {
2400         misc_deregister(&tun_miscdev);
2401         rtnl_link_unregister(&tun_link_ops);
2402 }
2403
2404 /* Get an underlying socket object from tun file.  Returns error unless file is
2405  * attached to a device.  The returned object works like a packet socket, it
2406  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2407  * holding a reference to the file for as long as the socket is in use. */
2408 struct socket *tun_get_socket(struct file *file)
2409 {
2410         struct tun_file *tfile;
2411         if (file->f_op != &tun_fops)
2412                 return ERR_PTR(-EINVAL);
2413         tfile = file->private_data;
2414         if (!tfile)
2415                 return ERR_PTR(-EBADFD);
2416         return &tfile->socket;
2417 }
2418 EXPORT_SYMBOL_GPL(tun_get_socket);
2419
2420 module_init(tun_init);
2421 module_exit(tun_cleanup);
2422 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2423 MODULE_AUTHOR(DRV_COPYRIGHT);
2424 MODULE_LICENSE("GPL");
2425 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2426 MODULE_ALIAS("devname:net/tun");