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 struct tun_pcpu_stats {
 135         u64 rx_packets;
 136         u64 rx_bytes;
 137         u64 tx_packets;
 138         u64 tx_bytes;
 139         struct u64_stats_sync syncp;
 140         u32 rx_dropped;
 141         u32 tx_dropped;
 142         u32 rx_frame_errors;
 143 };
 144
 145 /* A tun_file connects an open character device to a tuntap netdevice. It
 146  * also contains all socket related structures (except sock_fprog and tap_filter)
 147  * to serve as one transmit queue for tuntap device. The sock_fprog and
 148  * tap_filter were kept in tun_struct since they were used for filtering for the
 149  * netdevice not for a specific queue (at least I didn't see the requirement for
 150  * this).
 151  *
 152  * RCU usage:
 153  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 154  * other can only be read while rcu_read_lock or rtnl_lock is held.
 155  */
 156 struct tun_file {
 157         struct sock sk;
 158         struct socket socket;
 159         struct socket_wq wq;
 160         struct tun_struct __rcu *tun;
 161         struct fasync_struct *fasync;
 162         /* only used for fasnyc */
 163         unsigned int flags;
 164         union {
 165                 u16 queue_index;
 166                 unsigned int ifindex;
 167         };
 168         struct list_head next;
 169         struct tun_struct *detached;
 170 };
 171
 172 struct tun_flow_entry {
 173         struct hlist_node hash_link;
 174         struct rcu_head rcu;
 175         struct tun_struct *tun;
 176
 177         u32 rxhash;
 178         u32 rps_rxhash;
 179         int queue_index;
 180         unsigned long updated;
 181 };
 182
 183 #define TUN_NUM_FLOW_ENTRIES 1024
 184
 185 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 186  * device, socket filter, sndbuf and vnet header size were restore when the
 187  * file were attached to a persist device.
 188  */
 189 struct tun_struct {
 190         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 191         unsigned int            numqueues;
 192         unsigned int            flags;
 193         kuid_t                  owner;
 194         kgid_t                  group;
 195
 196         struct net_device       *dev;
 197         netdev_features_t       set_features;
 198 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 199                           NETIF_F_TSO6|NETIF_F_UFO)
 200
 201         int                     align;
 202         int                     vnet_hdr_sz;
 203         int                     sndbuf;
 204         struct tap_filter       txflt;
 205         struct sock_fprog       fprog;
 206         /* protected by rtnl lock */
 207         bool                    filter_attached;
 208 #ifdef TUN_DEBUG
 209         int debug;
 210 #endif
 211         spinlock_t lock;
 212         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 213         struct timer_list flow_gc_timer;
 214         unsigned long ageing_time;
 215         unsigned int numdisabled;
 216         struct list_head disabled;
 217         void *security;
 218         u32 flow_count;
 219         struct tun_pcpu_stats __percpu *pcpu_stats;
 220 };
 221
 222 #ifdef CONFIG_TUN_VNET_CROSS_LE
 223 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 224 {
 225         return tun->flags & TUN_VNET_BE ? false :
 226                 virtio_legacy_is_little_endian();
 227 }
 228
 229 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 230 {
 231         int be = !!(tun->flags & TUN_VNET_BE);
 232
 233         if (put_user(be, argp))
 234                 return -EFAULT;
 235
 236         return 0;
 237 }
 238
 239 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 240 {
 241         int be;
 242
 243         if (get_user(be, argp))
 244                 return -EFAULT;
 245
 246         if (be)
 247                 tun->flags |= TUN_VNET_BE;
 248         else
 249                 tun->flags &= ~TUN_VNET_BE;
 250
 251         return 0;
 252 }
 253 #else
 254 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 255 {
 256         return virtio_legacy_is_little_endian();
 257 }
 258
 259 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 260 {
 261         return -EINVAL;
 262 }
 263
 264 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 265 {
 266         return -EINVAL;
 267 }
 268 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 269
 270 static inline bool tun_is_little_endian(struct tun_struct *tun)
 271 {
 272         return tun->flags & TUN_VNET_LE ||
 273                 tun_legacy_is_little_endian(tun);
 274 }
 275
 276 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 277 {
 278         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 279 }
 280
 281 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 282 {
 283         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 284 }
 285
 286 static inline u32 tun_hashfn(u32 rxhash)
 287 {
 288         return rxhash & 0x3ff;
 289 }
 290
 291 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 292 {
 293         struct tun_flow_entry *e;
 294
 295         hlist_for_each_entry_rcu(e, head, hash_link) {
 296                 if (e->rxhash == rxhash)
 297                         return e;
 298         }
 299         return NULL;
 300 }
 301
 302 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 303                                               struct hlist_head *head,
 304                                               u32 rxhash, u16 queue_index)
 305 {
 306         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 307
 308         if (e) {
 309                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 310                           rxhash, queue_index);
 311                 e->updated = jiffies;
 312                 e->rxhash = rxhash;
 313                 e->rps_rxhash = 0;
 314                 e->queue_index = queue_index;
 315                 e->tun = tun;
 316                 hlist_add_head_rcu(&e->hash_link, head);
 317                 ++tun->flow_count;
 318         }
 319         return e;
 320 }
 321
 322 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 323 {
 324         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 325                   e->rxhash, e->queue_index);
 326         hlist_del_rcu(&e->hash_link);
 327         kfree_rcu(e, rcu);
 328         --tun->flow_count;
 329 }
 330
 331 static void tun_flow_flush(struct tun_struct *tun)
 332 {
 333         int i;
 334
 335         spin_lock_bh(&tun->lock);
 336         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 337                 struct tun_flow_entry *e;
 338                 struct hlist_node *n;
 339
 340                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 341                         tun_flow_delete(tun, e);
 342         }
 343         spin_unlock_bh(&tun->lock);
 344 }
 345
 346 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 347 {
 348         int i;
 349
 350         spin_lock_bh(&tun->lock);
 351         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 352                 struct tun_flow_entry *e;
 353                 struct hlist_node *n;
 354
 355                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 356                         if (e->queue_index == queue_index)
 357                                 tun_flow_delete(tun, e);
 358                 }
 359         }
 360         spin_unlock_bh(&tun->lock);
 361 }
 362
 363 static void tun_flow_cleanup(unsigned long data)
 364 {
 365         struct tun_struct *tun = (struct tun_struct *)data;
 366         unsigned long delay = tun->ageing_time;
 367         unsigned long next_timer = jiffies + delay;
 368         unsigned long count = 0;
 369         int i;
 370
 371         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 372
 373         spin_lock_bh(&tun->lock);
 374         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 375                 struct tun_flow_entry *e;
 376                 struct hlist_node *n;
 377
 378                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 379                         unsigned long this_timer;
 380                         count++;
 381                         this_timer = e->updated + delay;
 382                         if (time_before_eq(this_timer, jiffies))
 383                                 tun_flow_delete(tun, e);
 384                         else if (time_before(this_timer, next_timer))
 385                                 next_timer = this_timer;
 386                 }
 387         }
 388
 389         if (count)
 390                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 391         spin_unlock_bh(&tun->lock);
 392 }
 393
 394 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 395                             struct tun_file *tfile)
 396 {
 397         struct hlist_head *head;
 398         struct tun_flow_entry *e;
 399         unsigned long delay = tun->ageing_time;
 400         u16 queue_index = tfile->queue_index;
 401
 402         if (!rxhash)
 403                 return;
 404         else
 405                 head = &tun->flows[tun_hashfn(rxhash)];
 406
 407         rcu_read_lock();
 408
 409         /* We may get a very small possibility of OOO during switching, not
 410          * worth to optimize.*/
 411         if (tun->numqueues == 1 || tfile->detached)
 412                 goto unlock;
 413
 414         e = tun_flow_find(head, rxhash);
 415         if (likely(e)) {
 416                 /* TODO: keep queueing to old queue until it's empty? */
 417                 e->queue_index = queue_index;
 418                 e->updated = jiffies;
 419                 sock_rps_record_flow_hash(e->rps_rxhash);
 420         } else {
 421                 spin_lock_bh(&tun->lock);
 422                 if (!tun_flow_find(head, rxhash) &&
 423                     tun->flow_count < MAX_TAP_FLOWS)
 424                         tun_flow_create(tun, head, rxhash, queue_index);
 425
 426                 if (!timer_pending(&tun->flow_gc_timer))
 427                         mod_timer(&tun->flow_gc_timer,
 428                                   round_jiffies_up(jiffies + delay));
 429                 spin_unlock_bh(&tun->lock);
 430         }
 431
 432 unlock:
 433         rcu_read_unlock();
 434 }
 435
 436 /**
 437  * Save the hash received in the stack receive path and update the
 438  * flow_hash table accordingly.
 439  */
 440 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 441 {
 442         if (unlikely(e->rps_rxhash != hash))
 443                 e->rps_rxhash = hash;
 444 }
 445
 446 /* We try to identify a flow through its rxhash first. The reason that
 447  * we do not check rxq no. is because some cards(e.g 82599), chooses
 448  * the rxq based on the txq where the last packet of the flow comes. As
 449  * the userspace application move between processors, we may get a
 450  * different rxq no. here. If we could not get rxhash, then we would
 451  * hope the rxq no. may help here.
 452  */
 453 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 454                             void *accel_priv, select_queue_fallback_t fallback)
 455 {
 456         struct tun_struct *tun = netdev_priv(dev);
 457         struct tun_flow_entry *e;
 458         u32 txq = 0;
 459         u32 numqueues = 0;
 460
 461         rcu_read_lock();
 462         numqueues = ACCESS_ONCE(tun->numqueues);
 463
 464         txq = skb_get_hash(skb);
 465         if (txq) {
 466                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 467                 if (e) {
 468                         tun_flow_save_rps_rxhash(e, txq);
 469                         txq = e->queue_index;
 470                 } else
 471                         /* use multiply and shift instead of expensive divide */
 472                         txq = ((u64)txq * numqueues) >> 32;
 473         } else if (likely(skb_rx_queue_recorded(skb))) {
 474                 txq = skb_get_rx_queue(skb);
 475                 while (unlikely(txq >= numqueues))
 476                         txq -= numqueues;
 477         }
 478
 479         rcu_read_unlock();
 480         return txq;
 481 }
 482
 483 static inline bool tun_not_capable(struct tun_struct *tun)
 484 {
 485         const struct cred *cred = current_cred();
 486         struct net *net = dev_net(tun->dev);
 487
 488         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 489                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 490                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 491 }
 492
 493 static void tun_set_real_num_queues(struct tun_struct *tun)
 494 {
 495         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 496         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 497 }
 498
 499 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 500 {
 501         tfile->detached = tun;
 502         list_add_tail(&tfile->next, &tun->disabled);
 503         ++tun->numdisabled;
 504 }
 505
 506 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 507 {
 508         struct tun_struct *tun = tfile->detached;
 509
 510         tfile->detached = NULL;
 511         list_del_init(&tfile->next);
 512         --tun->numdisabled;
 513         return tun;
 514 }
 515
 516 static void tun_queue_purge(struct tun_file *tfile)
 517 {
 518         skb_queue_purge(&tfile->sk.sk_receive_queue);
 519         skb_queue_purge(&tfile->sk.sk_error_queue);
 520 }
 521
 522 static void __tun_detach(struct tun_file *tfile, bool clean)
 523 {
 524         struct tun_file *ntfile;
 525         struct tun_struct *tun;
 526
 527         tun = rtnl_dereference(tfile->tun);
 528
 529         if (tun && !tfile->detached) {
 530                 u16 index = tfile->queue_index;
 531                 BUG_ON(index >= tun->numqueues);
 532
 533                 rcu_assign_pointer(tun->tfiles[index],
 534                                    tun->tfiles[tun->numqueues - 1]);
 535                 ntfile = rtnl_dereference(tun->tfiles[index]);
 536                 ntfile->queue_index = index;
 537
 538                 --tun->numqueues;
 539                 if (clean) {
 540                         RCU_INIT_POINTER(tfile->tun, NULL);
 541                         sock_put(&tfile->sk);
 542                 } else
 543                         tun_disable_queue(tun, tfile);
 544
 545                 synchronize_net();
 546                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 547                 /* Drop read queue */
 548                 tun_queue_purge(tfile);
 549                 tun_set_real_num_queues(tun);
 550         } else if (tfile->detached && clean) {
 551                 tun = tun_enable_queue(tfile);
 552                 sock_put(&tfile->sk);
 553         }
 554
 555         if (clean) {
 556                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 557                         netif_carrier_off(tun->dev);
 558
 559                         if (!(tun->flags & IFF_PERSIST) &&
 560                             tun->dev->reg_state == NETREG_REGISTERED)
 561                                 unregister_netdevice(tun->dev);
 562                 }
 563                 sock_put(&tfile->sk);
 564         }
 565 }
 566
 567 static void tun_detach(struct tun_file *tfile, bool clean)
 568 {
 569         rtnl_lock();
 570         __tun_detach(tfile, clean);
 571         rtnl_unlock();
 572 }
 573
 574 static void tun_detach_all(struct net_device *dev)
 575 {
 576         struct tun_struct *tun = netdev_priv(dev);
 577         struct tun_file *tfile, *tmp;
 578         int i, n = tun->numqueues;
 579
 580         for (i = 0; i < n; i++) {
 581                 tfile = rtnl_dereference(tun->tfiles[i]);
 582                 BUG_ON(!tfile);
 583                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 584                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 585                 RCU_INIT_POINTER(tfile->tun, NULL);
 586                 --tun->numqueues;
 587         }
 588         list_for_each_entry(tfile, &tun->disabled, next) {
 589                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 590                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 591                 RCU_INIT_POINTER(tfile->tun, NULL);
 592         }
 593         BUG_ON(tun->numqueues != 0);
 594
 595         synchronize_net();
 596         for (i = 0; i < n; i++) {
 597                 tfile = rtnl_dereference(tun->tfiles[i]);
 598                 /* Drop read queue */
 599                 tun_queue_purge(tfile);
 600                 sock_put(&tfile->sk);
 601         }
 602         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 603                 tun_enable_queue(tfile);
 604                 tun_queue_purge(tfile);
 605                 sock_put(&tfile->sk);
 606         }
 607         BUG_ON(tun->numdisabled != 0);
 608
 609         if (tun->flags & IFF_PERSIST)
 610                 module_put(THIS_MODULE);
 611 }
 612
 613 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 614 {
 615         struct tun_file *tfile = file->private_data;
 616         int err;
 617
 618         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 619         if (err < 0)
 620                 goto out;
 621
 622         err = -EINVAL;
 623         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 624                 goto out;
 625
 626         err = -EBUSY;
 627         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 628                 goto out;
 629
 630         err = -E2BIG;
 631         if (!tfile->detached &&
 632             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 633                 goto out;
 634
 635         err = 0;
 636
 637         /* Re-attach the filter to persist device */
 638         if (!skip_filter && (tun->filter_attached == true)) {
 639                 lock_sock(tfile->socket.sk);
 640                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 641                 release_sock(tfile->socket.sk);
 642                 if (!err)
 643                         goto out;
 644         }
 645         tfile->queue_index = tun->numqueues;
 646         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 647         rcu_assign_pointer(tfile->tun, tun);
 648         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 649         tun->numqueues++;
 650
 651         if (tfile->detached)
 652                 tun_enable_queue(tfile);
 653         else
 654                 sock_hold(&tfile->sk);
 655
 656         tun_set_real_num_queues(tun);
 657
 658         /* device is allowed to go away first, so no need to hold extra
 659          * refcnt.
 660          */
 661
 662 out:
 663         return err;
 664 }
 665
 666 static struct tun_struct *__tun_get(struct tun_file *tfile)
 667 {
 668         struct tun_struct *tun;
 669
 670         rcu_read_lock();
 671         tun = rcu_dereference(tfile->tun);
 672         if (tun)
 673                 dev_hold(tun->dev);
 674         rcu_read_unlock();
 675
 676         return tun;
 677 }
 678
 679 static struct tun_struct *tun_get(struct file *file)
 680 {
 681         return __tun_get(file->private_data);
 682 }
 683
 684 static void tun_put(struct tun_struct *tun)
 685 {
 686         dev_put(tun->dev);
 687 }
 688
 689 /* TAP filtering */
 690 static void addr_hash_set(u32 *mask, const u8 *addr)
 691 {
 692         int n = ether_crc(ETH_ALEN, addr) >> 26;
 693         mask[n >> 5] |= (1 << (n & 31));
 694 }
 695
 696 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 697 {
 698         int n = ether_crc(ETH_ALEN, addr) >> 26;
 699         return mask[n >> 5] & (1 << (n & 31));
 700 }
 701
 702 static int update_filter(struct tap_filter *filter, void __user *arg)
 703 {
 704         struct { u8 u[ETH_ALEN]; } *addr;
 705         struct tun_filter uf;
 706         int err, alen, n, nexact;
 707
 708         if (copy_from_user(&uf, arg, sizeof(uf)))
 709                 return -EFAULT;
 710
 711         if (!uf.count) {
 712                 /* Disabled */
 713                 filter->count = 0;
 714                 return 0;
 715         }
 716
 717         alen = ETH_ALEN * uf.count;
 718         addr = kmalloc(alen, GFP_KERNEL);
 719         if (!addr)
 720                 return -ENOMEM;
 721
 722         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 723                 err = -EFAULT;
 724                 goto done;
 725         }
 726
 727         /* The filter is updated without holding any locks. Which is
 728          * perfectly safe. We disable it first and in the worst
 729          * case we'll accept a few undesired packets. */
 730         filter->count = 0;
 731         wmb();
 732
 733         /* Use first set of addresses as an exact filter */
 734         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 735                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 736
 737         nexact = n;
 738
 739         /* Remaining multicast addresses are hashed,
 740          * unicast will leave the filter disabled. */
 741         memset(filter->mask, 0, sizeof(filter->mask));
 742         for (; n < uf.count; n++) {
 743                 if (!is_multicast_ether_addr(addr[n].u)) {
 744                         err = 0; /* no filter */
 745                         goto done;
 746                 }
 747                 addr_hash_set(filter->mask, addr[n].u);
 748         }
 749
 750         /* For ALLMULTI just set the mask to all ones.
 751          * This overrides the mask populated above. */
 752         if ((uf.flags & TUN_FLT_ALLMULTI))
 753                 memset(filter->mask, ~0, sizeof(filter->mask));
 754
 755         /* Now enable the filter */
 756         wmb();
 757         filter->count = nexact;
 758
 759         /* Return the number of exact filters */
 760         err = nexact;
 761
 762 done:
 763         kfree(addr);
 764         return err;
 765 }
 766
 767 /* Returns: 0 - drop, !=0 - accept */
 768 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 769 {
 770         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 771          * at this point. */
 772         struct ethhdr *eh = (struct ethhdr *) skb->data;
 773         int i;
 774
 775         /* Exact match */
 776         for (i = 0; i < filter->count; i++)
 777                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 778                         return 1;
 779
 780         /* Inexact match (multicast only) */
 781         if (is_multicast_ether_addr(eh->h_dest))
 782                 return addr_hash_test(filter->mask, eh->h_dest);
 783
 784         return 0;
 785 }
 786
 787 /*
 788  * Checks whether the packet is accepted or not.
 789  * Returns: 0 - drop, !=0 - accept
 790  */
 791 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 792 {
 793         if (!filter->count)
 794                 return 1;
 795
 796         return run_filter(filter, skb);
 797 }
 798
 799 /* Network device part of the driver */
 800
 801 static const struct ethtool_ops tun_ethtool_ops;
 802
 803 /* Net device detach from fd. */
 804 static void tun_net_uninit(struct net_device *dev)
 805 {
 806         tun_detach_all(dev);
 807 }
 808
 809 /* Net device open. */
 810 static int tun_net_open(struct net_device *dev)
 811 {
 812         netif_tx_start_all_queues(dev);
 813         return 0;
 814 }
 815
 816 /* Net device close. */
 817 static int tun_net_close(struct net_device *dev)
 818 {
 819         netif_tx_stop_all_queues(dev);
 820         return 0;
 821 }
 822
 823 /* Net device start xmit */
 824 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 825 {
 826         struct tun_struct *tun = netdev_priv(dev);
 827         int txq = skb->queue_mapping;
 828         struct tun_file *tfile;
 829         u32 numqueues = 0;
 830
 831         rcu_read_lock();
 832         tfile = rcu_dereference(tun->tfiles[txq]);
 833         numqueues = ACCESS_ONCE(tun->numqueues);
 834
 835         /* Drop packet if interface is not attached */
 836         if (txq >= numqueues)
 837                 goto drop;
 838
 839 #ifdef CONFIG_RPS
 840         if (numqueues == 1 && static_key_false(&rps_needed)) {
 841                 /* Select queue was not called for the skbuff, so we extract the
 842                  * RPS hash and save it into the flow_table here.
 843                  */
 844                 __u32 rxhash;
 845
 846                 rxhash = skb_get_hash(skb);
 847                 if (rxhash) {
 848                         struct tun_flow_entry *e;
 849                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 850                                         rxhash);
 851                         if (e)
 852                                 tun_flow_save_rps_rxhash(e, rxhash);
 853                 }
 854         }
 855 #endif
 856
 857         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 858
 859         BUG_ON(!tfile);
 860
 861         /* Drop if the filter does not like it.
 862          * This is a noop if the filter is disabled.
 863          * Filter can be enabled only for the TAP devices. */
 864         if (!check_filter(&tun->txflt, skb))
 865                 goto drop;
 866
 867         if (tfile->socket.sk->sk_filter &&
 868             sk_filter(tfile->socket.sk, skb))
 869                 goto drop;
 870
 871         /* Limit the number of packets queued by dividing txq length with the
 872          * number of queues.
 873          */
 874         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
 875                           >= dev->tx_queue_len)
 876                 goto drop;
 877
 878         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 879                 goto drop;
 880
 881         if (skb->sk && sk_fullsock(skb->sk)) {
 882                 sock_tx_timestamp(skb->sk, skb->sk->sk_tsflags,
 883                                   &skb_shinfo(skb)->tx_flags);
 884                 sw_tx_timestamp(skb);
 885         }
 886
 887         /* Orphan the skb - required as we might hang on to it
 888          * for indefinite time.
 889          */
 890         skb_orphan(skb);
 891
 892         nf_reset(skb);
 893
 894         /* Enqueue packet */
 895         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 896
 897         /* Notify and wake up reader process */
 898         if (tfile->flags & TUN_FASYNC)
 899                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 900         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 901
 902         rcu_read_unlock();
 903         return NETDEV_TX_OK;
 904
 905 drop:
 906         this_cpu_inc(tun->pcpu_stats->tx_dropped);
 907         skb_tx_error(skb);
 908         kfree_skb(skb);
 909         rcu_read_unlock();
 910         return NET_XMIT_DROP;
 911 }
 912
 913 static void tun_net_mclist(struct net_device *dev)
 914 {
 915         /*
 916          * This callback is supposed to deal with mc filter in
 917          * _rx_ path and has nothing to do with the _tx_ path.
 918          * In rx path we always accept everything userspace gives us.
 919          */
 920 }
 921
 922 #define MIN_MTU 68
 923 #define MAX_MTU 65535
 924
 925 static int
 926 tun_net_change_mtu(struct net_device *dev, int new_mtu)
 927 {
 928         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 929                 return -EINVAL;
 930         dev->mtu = new_mtu;
 931         return 0;
 932 }
 933
 934 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 935         netdev_features_t features)
 936 {
 937         struct tun_struct *tun = netdev_priv(dev);
 938
 939         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 940 }
 941 #ifdef CONFIG_NET_POLL_CONTROLLER
 942 static void tun_poll_controller(struct net_device *dev)
 943 {
 944         /*
 945          * Tun only receives frames when:
 946          * 1) the char device endpoint gets data from user space
 947          * 2) the tun socket gets a sendmsg call from user space
 948          * Since both of those are synchronous operations, we are guaranteed
 949          * never to have pending data when we poll for it
 950          * so there is nothing to do here but return.
 951          * We need this though so netpoll recognizes us as an interface that
 952          * supports polling, which enables bridge devices in virt setups to
 953          * still use netconsole
 954          */
 955         return;
 956 }
 957 #endif
 958
 959 static void tun_set_headroom(struct net_device *dev, int new_hr)
 960 {
 961         struct tun_struct *tun = netdev_priv(dev);
 962
 963         if (new_hr < NET_SKB_PAD)
 964                 new_hr = NET_SKB_PAD;
 965
 966         tun->align = new_hr;
 967 }
 968
 969 static struct rtnl_link_stats64 *
 970 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
 971 {
 972         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
 973         struct tun_struct *tun = netdev_priv(dev);
 974         struct tun_pcpu_stats *p;
 975         int i;
 976
 977         for_each_possible_cpu(i) {
 978                 u64 rxpackets, rxbytes, txpackets, txbytes;
 979                 unsigned int start;
 980
 981                 p = per_cpu_ptr(tun->pcpu_stats, i);
 982                 do {
 983                         start = u64_stats_fetch_begin(&p->syncp);
 984                         rxpackets       = p->rx_packets;
 985                         rxbytes         = p->rx_bytes;
 986                         txpackets       = p->tx_packets;
 987                         txbytes         = p->tx_bytes;
 988                 } while (u64_stats_fetch_retry(&p->syncp, start));
 989
 990                 stats->rx_packets       += rxpackets;
 991                 stats->rx_bytes         += rxbytes;
 992                 stats->tx_packets       += txpackets;
 993                 stats->tx_bytes         += txbytes;
 994
 995                 /* u32 counters */
 996                 rx_dropped      += p->rx_dropped;
 997                 rx_frame_errors += p->rx_frame_errors;
 998                 tx_dropped      += p->tx_dropped;
 999         }
1000         stats->rx_dropped  = rx_dropped;
1001         stats->rx_frame_errors = rx_frame_errors;
1002         stats->tx_dropped = tx_dropped;
1003         return stats;
1004 }
1005
1006 static const struct net_device_ops tun_netdev_ops = {
1007         .ndo_uninit             = tun_net_uninit,
1008         .ndo_open               = tun_net_open,
1009         .ndo_stop               = tun_net_close,
1010         .ndo_start_xmit         = tun_net_xmit,
1011         .ndo_change_mtu         = tun_net_change_mtu,
1012         .ndo_fix_features       = tun_net_fix_features,
1013         .ndo_select_queue       = tun_select_queue,
1014 #ifdef CONFIG_NET_POLL_CONTROLLER
1015         .ndo_poll_controller    = tun_poll_controller,
1016 #endif
1017         .ndo_set_rx_headroom    = tun_set_headroom,
1018         .ndo_get_stats64        = tun_net_get_stats64,
1019 };
1020
1021 static const struct net_device_ops tap_netdev_ops = {
1022         .ndo_uninit             = tun_net_uninit,
1023         .ndo_open               = tun_net_open,
1024         .ndo_stop               = tun_net_close,
1025         .ndo_start_xmit         = tun_net_xmit,
1026         .ndo_change_mtu         = tun_net_change_mtu,
1027         .ndo_fix_features       = tun_net_fix_features,
1028         .ndo_set_rx_mode        = tun_net_mclist,
1029         .ndo_set_mac_address    = eth_mac_addr,
1030         .ndo_validate_addr      = eth_validate_addr,
1031         .ndo_select_queue       = tun_select_queue,
1032 #ifdef CONFIG_NET_POLL_CONTROLLER
1033         .ndo_poll_controller    = tun_poll_controller,
1034 #endif
1035         .ndo_features_check     = passthru_features_check,
1036         .ndo_set_rx_headroom    = tun_set_headroom,
1037         .ndo_get_stats64        = tun_net_get_stats64,
1038 };
1039
1040 static void tun_flow_init(struct tun_struct *tun)
1041 {
1042         int i;
1043
1044         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1045                 INIT_HLIST_HEAD(&tun->flows[i]);
1046
1047         tun->ageing_time = TUN_FLOW_EXPIRE;
1048         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1049         mod_timer(&tun->flow_gc_timer,
1050                   round_jiffies_up(jiffies + tun->ageing_time));
1051 }
1052
1053 static void tun_flow_uninit(struct tun_struct *tun)
1054 {
1055         del_timer_sync(&tun->flow_gc_timer);
1056         tun_flow_flush(tun);
1057 }
1058
1059 /* Initialize net device. */
1060 static void tun_net_init(struct net_device *dev)
1061 {
1062         struct tun_struct *tun = netdev_priv(dev);
1063
1064         switch (tun->flags & TUN_TYPE_MASK) {
1065         case IFF_TUN:
1066                 dev->netdev_ops = &tun_netdev_ops;
1067
1068                 /* Point-to-Point TUN Device */
1069                 dev->hard_header_len = 0;
1070                 dev->addr_len = 0;
1071                 dev->mtu = 1500;
1072
1073                 /* Zero header length */
1074                 dev->type = ARPHRD_NONE;
1075                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1076                 break;
1077
1078         case IFF_TAP:
1079                 dev->netdev_ops = &tap_netdev_ops;
1080                 /* Ethernet TAP Device */
1081                 ether_setup(dev);
1082                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1083                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1084
1085                 eth_hw_addr_random(dev);
1086
1087                 break;
1088         }
1089 }
1090
1091 /* Character device part */
1092
1093 /* Poll */
1094 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1095 {
1096         struct tun_file *tfile = file->private_data;
1097         struct tun_struct *tun = __tun_get(tfile);
1098         struct sock *sk;
1099         unsigned int mask = 0;
1100
1101         if (!tun)
1102                 return POLLERR;
1103
1104         sk = tfile->socket.sk;
1105
1106         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1107
1108         poll_wait(file, sk_sleep(sk), wait);
1109
1110         if (!skb_queue_empty(&sk->sk_receive_queue))
1111                 mask |= POLLIN | POLLRDNORM;
1112
1113         if (sock_writeable(sk) ||
1114             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1115              sock_writeable(sk)))
1116                 mask |= POLLOUT | POLLWRNORM;
1117
1118         if (tun->dev->reg_state != NETREG_REGISTERED)
1119                 mask = POLLERR;
1120
1121         tun_put(tun);
1122         return mask;
1123 }
1124
1125 /* prepad is the amount to reserve at front.  len is length after that.
1126  * linear is a hint as to how much to copy (usually headers). */
1127 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1128                                      size_t prepad, size_t len,
1129                                      size_t linear, int noblock)
1130 {
1131         struct sock *sk = tfile->socket.sk;
1132         struct sk_buff *skb;
1133         int err;
1134
1135         /* Under a page?  Don't bother with paged skb. */
1136         if (prepad + len < PAGE_SIZE || !linear)
1137                 linear = len;
1138
1139         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1140                                    &err, 0);
1141         if (!skb)
1142                 return ERR_PTR(err);
1143
1144         skb_reserve(skb, prepad);
1145         skb_put(skb, linear);
1146         skb->data_len = len - linear;
1147         skb->len += len - linear;
1148
1149         return skb;
1150 }
1151
1152 /* Get packet from user space buffer */
1153 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1154                             void *msg_control, struct iov_iter *from,
1155                             int noblock)
1156 {
1157         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1158         struct sk_buff *skb;
1159         size_t total_len = iov_iter_count(from);
1160         size_t len = total_len, align = tun->align, linear;
1161         struct virtio_net_hdr gso = { 0 };
1162         struct tun_pcpu_stats *stats;
1163         int good_linear;
1164         int copylen;
1165         bool zerocopy = false;
1166         int err;
1167         u32 rxhash;
1168         ssize_t n;
1169
1170         if (!(tun->dev->flags & IFF_UP))
1171                 return -EIO;
1172
1173         if (!(tun->flags & IFF_NO_PI)) {
1174                 if (len < sizeof(pi))
1175                         return -EINVAL;
1176                 len -= sizeof(pi);
1177
1178                 n = copy_from_iter(&pi, sizeof(pi), from);
1179                 if (n != sizeof(pi))
1180                         return -EFAULT;
1181         }
1182
1183         if (tun->flags & IFF_VNET_HDR) {
1184                 if (len < tun->vnet_hdr_sz)
1185                         return -EINVAL;
1186                 len -= tun->vnet_hdr_sz;
1187
1188                 n = copy_from_iter(&gso, sizeof(gso), from);
1189                 if (n != sizeof(gso))
1190                         return -EFAULT;
1191
1192                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1193                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1194                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1195
1196                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1197                         return -EINVAL;
1198                 iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1199         }
1200
1201         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1202                 align += NET_IP_ALIGN;
1203                 if (unlikely(len < ETH_HLEN ||
1204                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1205                         return -EINVAL;
1206         }
1207
1208         good_linear = SKB_MAX_HEAD(align);
1209
1210         if (msg_control) {
1211                 struct iov_iter i = *from;
1212
1213                 /* There are 256 bytes to be copied in skb, so there is
1214                  * enough room for skb expand head in case it is used.
1215                  * The rest of the buffer is mapped from userspace.
1216                  */
1217                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1218                 if (copylen > good_linear)
1219                         copylen = good_linear;
1220                 linear = copylen;
1221                 iov_iter_advance(&i, copylen);
1222                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1223                         zerocopy = true;
1224         }
1225
1226         if (!zerocopy) {
1227                 copylen = len;
1228                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1229                         linear = good_linear;
1230                 else
1231                         linear = tun16_to_cpu(tun, gso.hdr_len);
1232         }
1233
1234         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1235         if (IS_ERR(skb)) {
1236                 if (PTR_ERR(skb) != -EAGAIN)
1237                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1238                 return PTR_ERR(skb);
1239         }
1240
1241         if (zerocopy)
1242                 err = zerocopy_sg_from_iter(skb, from);
1243         else {
1244                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1245                 if (!err && msg_control) {
1246                         struct ubuf_info *uarg = msg_control;
1247                         uarg->callback(uarg, false);
1248                 }
1249         }
1250
1251         if (err) {
1252                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1253                 kfree_skb(skb);
1254                 return -EFAULT;
1255         }
1256
1257         switch (tun->flags & TUN_TYPE_MASK) {
1258         case IFF_TUN:
1259                 if (tun->flags & IFF_NO_PI) {
1260                         switch (skb->data[0] & 0xf0) {
1261                         case 0x40:
1262                                 pi.proto = htons(ETH_P_IP);
1263                                 break;
1264                         case 0x60:
1265                                 pi.proto = htons(ETH_P_IPV6);
1266                                 break;
1267                         default:
1268                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1269                                 kfree_skb(skb);
1270                                 return -EINVAL;
1271                         }
1272                 }
1273
1274                 skb_reset_mac_header(skb);
1275                 skb->protocol = pi.proto;
1276                 skb->dev = tun->dev;
1277                 break;
1278         case IFF_TAP:
1279                 skb->protocol = eth_type_trans(skb, tun->dev);
1280                 break;
1281         }
1282
1283         err = virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun));
1284         if (err) {
1285                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1286                 kfree_skb(skb);
1287                 return -EINVAL;
1288         }
1289
1290         /* copy skb_ubuf_info for callback when skb has no error */
1291         if (zerocopy) {
1292                 skb_shinfo(skb)->destructor_arg = msg_control;
1293                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1294                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1295         }
1296
1297         skb_reset_network_header(skb);
1298         skb_probe_transport_header(skb, 0);
1299
1300         rxhash = skb_get_hash(skb);
1301         netif_rx_ni(skb);
1302
1303         stats = get_cpu_ptr(tun->pcpu_stats);
1304         u64_stats_update_begin(&stats->syncp);
1305         stats->rx_packets++;
1306         stats->rx_bytes += len;
1307         u64_stats_update_end(&stats->syncp);
1308         put_cpu_ptr(stats);
1309
1310         tun_flow_update(tun, rxhash, tfile);
1311         return total_len;
1312 }
1313
1314 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1315 {
1316         struct file *file = iocb->ki_filp;
1317         struct tun_struct *tun = tun_get(file);
1318         struct tun_file *tfile = file->private_data;
1319         ssize_t result;
1320
1321         if (!tun)
1322                 return -EBADFD;
1323
1324         result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1325
1326         tun_put(tun);
1327         return result;
1328 }
1329
1330 /* Put packet to the user space buffer */
1331 static ssize_t tun_put_user(struct tun_struct *tun,
1332                             struct tun_file *tfile,
1333                             struct sk_buff *skb,
1334                             struct iov_iter *iter)
1335 {
1336         struct tun_pi pi = { 0, skb->protocol };
1337         struct tun_pcpu_stats *stats;
1338         ssize_t total;
1339         int vlan_offset = 0;
1340         int vlan_hlen = 0;
1341         int vnet_hdr_sz = 0;
1342
1343         if (skb_vlan_tag_present(skb))
1344                 vlan_hlen = VLAN_HLEN;
1345
1346         if (tun->flags & IFF_VNET_HDR)
1347                 vnet_hdr_sz = tun->vnet_hdr_sz;
1348
1349         total = skb->len + vlan_hlen + vnet_hdr_sz;
1350
1351         if (!(tun->flags & IFF_NO_PI)) {
1352                 if (iov_iter_count(iter) < sizeof(pi))
1353                         return -EINVAL;
1354
1355                 total += sizeof(pi);
1356                 if (iov_iter_count(iter) < total) {
1357                         /* Packet will be striped */
1358                         pi.flags |= TUN_PKT_STRIP;
1359                 }
1360
1361                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1362                         return -EFAULT;
1363         }
1364
1365         if (vnet_hdr_sz) {
1366                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1367                 int ret;
1368
1369                 if (iov_iter_count(iter) < vnet_hdr_sz)
1370                         return -EINVAL;
1371
1372                 ret = virtio_net_hdr_from_skb(skb, &gso,
1373                                               tun_is_little_endian(tun));
1374                 if (ret) {
1375                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1376                         pr_err("unexpected GSO type: "
1377                                "0x%x, gso_size %d, hdr_len %d\n",
1378                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1379                                tun16_to_cpu(tun, gso.hdr_len));
1380                         print_hex_dump(KERN_ERR, "tun: ",
1381                                        DUMP_PREFIX_NONE,
1382                                        16, 1, skb->head,
1383                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1384                         WARN_ON_ONCE(1);
1385                         return -EINVAL;
1386                 }
1387
1388                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1389                         return -EFAULT;
1390
1391                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1392         }
1393
1394         if (vlan_hlen) {
1395                 int ret;
1396                 struct {
1397                         __be16 h_vlan_proto;
1398                         __be16 h_vlan_TCI;
1399                 } veth;
1400
1401                 veth.h_vlan_proto = skb->vlan_proto;
1402                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1403
1404                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1405
1406                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1407                 if (ret || !iov_iter_count(iter))
1408                         goto done;
1409
1410                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1411                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1412                         goto done;
1413         }
1414
1415         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1416
1417 done:
1418         /* caller is in process context, */
1419         stats = get_cpu_ptr(tun->pcpu_stats);
1420         u64_stats_update_begin(&stats->syncp);
1421         stats->tx_packets++;
1422         stats->tx_bytes += skb->len + vlan_hlen;
1423         u64_stats_update_end(&stats->syncp);
1424         put_cpu_ptr(tun->pcpu_stats);
1425
1426         return total;
1427 }
1428
1429 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1430                            struct iov_iter *to,
1431                            int noblock)
1432 {
1433         struct sk_buff *skb;
1434         ssize_t ret;
1435         int peeked, err, off = 0;
1436
1437         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1438
1439         if (!iov_iter_count(to))
1440                 return 0;
1441
1442         /* Read frames from queue */
1443         skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1444                                   &peeked, &off, &err);
1445         if (!skb)
1446                 return err;
1447
1448         ret = tun_put_user(tun, tfile, skb, to);
1449         if (unlikely(ret < 0))
1450                 kfree_skb(skb);
1451         else
1452                 consume_skb(skb);
1453
1454         return ret;
1455 }
1456
1457 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1458 {
1459         struct file *file = iocb->ki_filp;
1460         struct tun_file *tfile = file->private_data;
1461         struct tun_struct *tun = __tun_get(tfile);
1462         ssize_t len = iov_iter_count(to), ret;
1463
1464         if (!tun)
1465                 return -EBADFD;
1466         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1467         ret = min_t(ssize_t, ret, len);
1468         if (ret > 0)
1469                 iocb->ki_pos = ret;
1470         tun_put(tun);
1471         return ret;
1472 }
1473
1474 static void tun_free_netdev(struct net_device *dev)
1475 {
1476         struct tun_struct *tun = netdev_priv(dev);
1477
1478         BUG_ON(!(list_empty(&tun->disabled)));
1479         free_percpu(tun->pcpu_stats);
1480         tun_flow_uninit(tun);
1481         security_tun_dev_free_security(tun->security);
1482         free_netdev(dev);
1483 }
1484
1485 static void tun_setup(struct net_device *dev)
1486 {
1487         struct tun_struct *tun = netdev_priv(dev);
1488
1489         tun->owner = INVALID_UID;
1490         tun->group = INVALID_GID;
1491
1492         dev->ethtool_ops = &tun_ethtool_ops;
1493         dev->destructor = tun_free_netdev;
1494         /* We prefer our own queue length */
1495         dev->tx_queue_len = TUN_READQ_SIZE;
1496 }
1497
1498 /* Trivial set of netlink ops to allow deleting tun or tap
1499  * device with netlink.
1500  */
1501 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1502 {
1503         return -EINVAL;
1504 }
1505
1506 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1507         .kind           = DRV_NAME,
1508         .priv_size      = sizeof(struct tun_struct),
1509         .setup          = tun_setup,
1510         .validate       = tun_validate,
1511 };
1512
1513 static void tun_sock_write_space(struct sock *sk)
1514 {
1515         struct tun_file *tfile;
1516         wait_queue_head_t *wqueue;
1517
1518         if (!sock_writeable(sk))
1519                 return;
1520
1521         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1522                 return;
1523
1524         wqueue = sk_sleep(sk);
1525         if (wqueue && waitqueue_active(wqueue))
1526                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1527                                                 POLLWRNORM | POLLWRBAND);
1528
1529         tfile = container_of(sk, struct tun_file, sk);
1530         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1531 }
1532
1533 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1534 {
1535         int ret;
1536         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1537         struct tun_struct *tun = __tun_get(tfile);
1538
1539         if (!tun)
1540                 return -EBADFD;
1541
1542         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1543                            m->msg_flags & MSG_DONTWAIT);
1544         tun_put(tun);
1545         return ret;
1546 }
1547
1548 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1549                        int flags)
1550 {
1551         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1552         struct tun_struct *tun = __tun_get(tfile);
1553         int ret;
1554
1555         if (!tun)
1556                 return -EBADFD;
1557
1558         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1559                 ret = -EINVAL;
1560                 goto out;
1561         }
1562         if (flags & MSG_ERRQUEUE) {
1563                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1564                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1565                 goto out;
1566         }
1567         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1568         if (ret > (ssize_t)total_len) {
1569                 m->msg_flags |= MSG_TRUNC;
1570                 ret = flags & MSG_TRUNC ? ret : total_len;
1571         }
1572 out:
1573         tun_put(tun);
1574         return ret;
1575 }
1576
1577 /* Ops structure to mimic raw sockets with tun */
1578 static const struct proto_ops tun_socket_ops = {
1579         .sendmsg = tun_sendmsg,
1580         .recvmsg = tun_recvmsg,
1581 };
1582
1583 static struct proto tun_proto = {
1584         .name           = "tun",
1585         .owner          = THIS_MODULE,
1586         .obj_size       = sizeof(struct tun_file),
1587 };
1588
1589 static int tun_flags(struct tun_struct *tun)
1590 {
1591         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1592 }
1593
1594 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1595                               char *buf)
1596 {
1597         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1598         return sprintf(buf, "0x%x\n", tun_flags(tun));
1599 }
1600
1601 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1602                               char *buf)
1603 {
1604         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1605         return uid_valid(tun->owner)?
1606                 sprintf(buf, "%u\n",
1607                         from_kuid_munged(current_user_ns(), tun->owner)):
1608                 sprintf(buf, "-1\n");
1609 }
1610
1611 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1612                               char *buf)
1613 {
1614         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1615         return gid_valid(tun->group) ?
1616                 sprintf(buf, "%u\n",
1617                         from_kgid_munged(current_user_ns(), tun->group)):
1618                 sprintf(buf, "-1\n");
1619 }
1620
1621 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1622 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1623 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1624
1625 static struct attribute *tun_dev_attrs[] = {
1626         &dev_attr_tun_flags.attr,
1627         &dev_attr_owner.attr,
1628         &dev_attr_group.attr,
1629         NULL
1630 };
1631
1632 static const struct attribute_group tun_attr_group = {
1633         .attrs = tun_dev_attrs
1634 };
1635
1636 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1637 {
1638         struct tun_struct *tun;
1639         struct tun_file *tfile = file->private_data;
1640         struct net_device *dev;
1641         int err;
1642
1643         if (tfile->detached)
1644                 return -EINVAL;
1645
1646         dev = __dev_get_by_name(net, ifr->ifr_name);
1647         if (dev) {
1648                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1649                         return -EBUSY;
1650                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1651                         tun = netdev_priv(dev);
1652                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1653                         tun = netdev_priv(dev);
1654                 else
1655                         return -EINVAL;
1656
1657                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1658                     !!(tun->flags & IFF_MULTI_QUEUE))
1659                         return -EINVAL;
1660
1661                 if (tun_not_capable(tun))
1662                         return -EPERM;
1663                 err = security_tun_dev_open(tun->security);
1664                 if (err < 0)
1665                         return err;
1666
1667                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1668                 if (err < 0)
1669                         return err;
1670
1671                 if (tun->flags & IFF_MULTI_QUEUE &&
1672                     (tun->numqueues + tun->numdisabled > 1)) {
1673                         /* One or more queue has already been attached, no need
1674                          * to initialize the device again.
1675                          */
1676                         return 0;
1677                 }
1678         }
1679         else {
1680                 char *name;
1681                 unsigned long flags = 0;
1682                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1683                              MAX_TAP_QUEUES : 1;
1684
1685                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1686                         return -EPERM;
1687                 err = security_tun_dev_create();
1688                 if (err < 0)
1689                         return err;
1690
1691                 /* Set dev type */
1692                 if (ifr->ifr_flags & IFF_TUN) {
1693                         /* TUN device */
1694                         flags |= IFF_TUN;
1695                         name = "tun%d";
1696                 } else if (ifr->ifr_flags & IFF_TAP) {
1697                         /* TAP device */
1698                         flags |= IFF_TAP;
1699                         name = "tap%d";
1700                 } else
1701                         return -EINVAL;
1702
1703                 if (*ifr->ifr_name)
1704                         name = ifr->ifr_name;
1705
1706                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1707                                        NET_NAME_UNKNOWN, tun_setup, queues,
1708                                        queues);
1709
1710                 if (!dev)
1711                         return -ENOMEM;
1712
1713                 dev_net_set(dev, net);
1714                 dev->rtnl_link_ops = &tun_link_ops;
1715                 dev->ifindex = tfile->ifindex;
1716                 dev->sysfs_groups[0] = &tun_attr_group;
1717
1718                 tun = netdev_priv(dev);
1719                 tun->dev = dev;
1720                 tun->flags = flags;
1721                 tun->txflt.count = 0;
1722                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1723
1724                 tun->align = NET_SKB_PAD;
1725                 tun->filter_attached = false;
1726                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1727
1728                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
1729                 if (!tun->pcpu_stats) {
1730                         err = -ENOMEM;
1731                         goto err_free_dev;
1732                 }
1733
1734                 spin_lock_init(&tun->lock);
1735
1736                 err = security_tun_dev_alloc_security(&tun->security);
1737                 if (err < 0)
1738                         goto err_free_stat;
1739
1740                 tun_net_init(dev);
1741                 tun_flow_init(tun);
1742
1743                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1744                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1745                                    NETIF_F_HW_VLAN_STAG_TX;
1746                 dev->features = dev->hw_features | NETIF_F_LLTX;
1747                 dev->vlan_features = dev->features &
1748                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1749                                        NETIF_F_HW_VLAN_STAG_TX);
1750
1751                 INIT_LIST_HEAD(&tun->disabled);
1752                 err = tun_attach(tun, file, false);
1753                 if (err < 0)
1754                         goto err_free_flow;
1755
1756                 err = register_netdevice(tun->dev);
1757                 if (err < 0)
1758                         goto err_detach;
1759         }
1760
1761         netif_carrier_on(tun->dev);
1762
1763         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1764
1765         tun->flags = (tun->flags & ~TUN_FEATURES) |
1766                 (ifr->ifr_flags & TUN_FEATURES);
1767
1768         /* Make sure persistent devices do not get stuck in
1769          * xoff state.
1770          */
1771         if (netif_running(tun->dev))
1772                 netif_tx_wake_all_queues(tun->dev);
1773
1774         strcpy(ifr->ifr_name, tun->dev->name);
1775         return 0;
1776
1777 err_detach:
1778         tun_detach_all(dev);
1779 err_free_flow:
1780         tun_flow_uninit(tun);
1781         security_tun_dev_free_security(tun->security);
1782 err_free_stat:
1783         free_percpu(tun->pcpu_stats);
1784 err_free_dev:
1785         free_netdev(dev);
1786         return err;
1787 }
1788
1789 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1790                        struct ifreq *ifr)
1791 {
1792         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1793
1794         strcpy(ifr->ifr_name, tun->dev->name);
1795
1796         ifr->ifr_flags = tun_flags(tun);
1797
1798 }
1799
1800 /* This is like a cut-down ethtool ops, except done via tun fd so no
1801  * privs required. */
1802 static int set_offload(struct tun_struct *tun, unsigned long arg)
1803 {
1804         netdev_features_t features = 0;
1805
1806         if (arg & TUN_F_CSUM) {
1807                 features |= NETIF_F_HW_CSUM;
1808                 arg &= ~TUN_F_CSUM;
1809
1810                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1811                         if (arg & TUN_F_TSO_ECN) {
1812                                 features |= NETIF_F_TSO_ECN;
1813                                 arg &= ~TUN_F_TSO_ECN;
1814                         }
1815                         if (arg & TUN_F_TSO4)
1816                                 features |= NETIF_F_TSO;
1817                         if (arg & TUN_F_TSO6)
1818                                 features |= NETIF_F_TSO6;
1819                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1820                 }
1821
1822                 if (arg & TUN_F_UFO) {
1823                         features |= NETIF_F_UFO;
1824                         arg &= ~TUN_F_UFO;
1825                 }
1826         }
1827
1828         /* This gives the user a way to test for new features in future by
1829          * trying to set them. */
1830         if (arg)
1831                 return -EINVAL;
1832
1833         tun->set_features = features;
1834         netdev_update_features(tun->dev);
1835
1836         return 0;
1837 }
1838
1839 static void tun_detach_filter(struct tun_struct *tun, int n)
1840 {
1841         int i;
1842         struct tun_file *tfile;
1843
1844         for (i = 0; i < n; i++) {
1845                 tfile = rtnl_dereference(tun->tfiles[i]);
1846                 lock_sock(tfile->socket.sk);
1847                 sk_detach_filter(tfile->socket.sk);
1848                 release_sock(tfile->socket.sk);
1849         }
1850
1851         tun->filter_attached = false;
1852 }
1853
1854 static int tun_attach_filter(struct tun_struct *tun)
1855 {
1856         int i, ret = 0;
1857         struct tun_file *tfile;
1858
1859         for (i = 0; i < tun->numqueues; i++) {
1860                 tfile = rtnl_dereference(tun->tfiles[i]);
1861                 lock_sock(tfile->socket.sk);
1862                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1863                 release_sock(tfile->socket.sk);
1864                 if (ret) {
1865                         tun_detach_filter(tun, i);
1866                         return ret;
1867                 }
1868         }
1869
1870         tun->filter_attached = true;
1871         return ret;
1872 }
1873
1874 static void tun_set_sndbuf(struct tun_struct *tun)
1875 {
1876         struct tun_file *tfile;
1877         int i;
1878
1879         for (i = 0; i < tun->numqueues; i++) {
1880                 tfile = rtnl_dereference(tun->tfiles[i]);
1881                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1882         }
1883 }
1884
1885 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1886 {
1887         struct tun_file *tfile = file->private_data;
1888         struct tun_struct *tun;
1889         int ret = 0;
1890
1891         rtnl_lock();
1892
1893         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1894                 tun = tfile->detached;
1895                 if (!tun) {
1896                         ret = -EINVAL;
1897                         goto unlock;
1898                 }
1899                 ret = security_tun_dev_attach_queue(tun->security);
1900                 if (ret < 0)
1901                         goto unlock;
1902                 ret = tun_attach(tun, file, false);
1903         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1904                 tun = rtnl_dereference(tfile->tun);
1905                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1906                         ret = -EINVAL;
1907                 else
1908                         __tun_detach(tfile, false);
1909         } else
1910                 ret = -EINVAL;
1911
1912 unlock:
1913         rtnl_unlock();
1914         return ret;
1915 }
1916
1917 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1918                             unsigned long arg, int ifreq_len)
1919 {
1920         struct tun_file *tfile = file->private_data;
1921         struct tun_struct *tun;
1922         void __user* argp = (void __user*)arg;
1923         struct ifreq ifr;
1924         kuid_t owner;
1925         kgid_t group;
1926         int sndbuf;
1927         int vnet_hdr_sz;
1928         unsigned int ifindex;
1929         int le;
1930         int ret;
1931
1932         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1933                 if (copy_from_user(&ifr, argp, ifreq_len))
1934                         return -EFAULT;
1935         } else {
1936                 memset(&ifr, 0, sizeof(ifr));
1937         }
1938         if (cmd == TUNGETFEATURES) {
1939                 /* Currently this just means: "what IFF flags are valid?".
1940                  * This is needed because we never checked for invalid flags on
1941                  * TUNSETIFF.
1942                  */
1943                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1944                                 (unsigned int __user*)argp);
1945         } else if (cmd == TUNSETQUEUE)
1946                 return tun_set_queue(file, &ifr);
1947
1948         ret = 0;
1949         rtnl_lock();
1950
1951         tun = __tun_get(tfile);
1952         if (cmd == TUNSETIFF && !tun) {
1953                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1954
1955                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
1956
1957                 if (ret)
1958                         goto unlock;
1959
1960                 if (copy_to_user(argp, &ifr, ifreq_len))
1961                         ret = -EFAULT;
1962                 goto unlock;
1963         }
1964         if (cmd == TUNSETIFINDEX) {
1965                 ret = -EPERM;
1966                 if (tun)
1967                         goto unlock;
1968
1969                 ret = -EFAULT;
1970                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1971                         goto unlock;
1972
1973                 ret = 0;
1974                 tfile->ifindex = ifindex;
1975                 goto unlock;
1976         }
1977
1978         ret = -EBADFD;
1979         if (!tun)
1980                 goto unlock;
1981
1982         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1983
1984         ret = 0;
1985         switch (cmd) {
1986         case TUNGETIFF:
1987                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1988
1989                 if (tfile->detached)
1990                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
1991                 if (!tfile->socket.sk->sk_filter)
1992                         ifr.ifr_flags |= IFF_NOFILTER;
1993
1994                 if (copy_to_user(argp, &ifr, ifreq_len))
1995                         ret = -EFAULT;
1996                 break;
1997
1998         case TUNSETNOCSUM:
1999                 /* Disable/Enable checksum */
2000
2001                 /* [unimplemented] */
2002                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2003                           arg ? "disabled" : "enabled");
2004                 break;
2005
2006         case TUNSETPERSIST:
2007                 /* Disable/Enable persist mode. Keep an extra reference to the
2008                  * module to prevent the module being unprobed.
2009                  */
2010                 if (arg && !(tun->flags & IFF_PERSIST)) {
2011                         tun->flags |= IFF_PERSIST;
2012                         __module_get(THIS_MODULE);
2013                 }
2014                 if (!arg && (tun->flags & IFF_PERSIST)) {
2015                         tun->flags &= ~IFF_PERSIST;
2016                         module_put(THIS_MODULE);
2017                 }
2018
2019                 tun_debug(KERN_INFO, tun, "persist %s\n",
2020                           arg ? "enabled" : "disabled");
2021                 break;
2022
2023         case TUNSETOWNER:
2024                 /* Set owner of the device */
2025                 owner = make_kuid(current_user_ns(), arg);
2026                 if (!uid_valid(owner)) {
2027                         ret = -EINVAL;
2028                         break;
2029                 }
2030                 tun->owner = owner;
2031                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2032                           from_kuid(&init_user_ns, tun->owner));
2033                 break;
2034
2035         case TUNSETGROUP:
2036                 /* Set group of the device */
2037                 group = make_kgid(current_user_ns(), arg);
2038                 if (!gid_valid(group)) {
2039                         ret = -EINVAL;
2040                         break;
2041                 }
2042                 tun->group = group;
2043                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2044                           from_kgid(&init_user_ns, tun->group));
2045                 break;
2046
2047         case TUNSETLINK:
2048                 /* Only allow setting the type when the interface is down */
2049                 if (tun->dev->flags & IFF_UP) {
2050                         tun_debug(KERN_INFO, tun,
2051                                   "Linktype set failed because interface is up\n");
2052                         ret = -EBUSY;
2053                 } else {
2054                         tun->dev->type = (int) arg;
2055                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2056                                   tun->dev->type);
2057                         ret = 0;
2058                 }
2059                 break;
2060
2061 #ifdef TUN_DEBUG
2062         case TUNSETDEBUG:
2063                 tun->debug = arg;
2064                 break;
2065 #endif
2066         case TUNSETOFFLOAD:
2067                 ret = set_offload(tun, arg);
2068                 break;
2069
2070         case TUNSETTXFILTER:
2071                 /* Can be set only for TAPs */
2072                 ret = -EINVAL;
2073                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2074                         break;
2075                 ret = update_filter(&tun->txflt, (void __user *)arg);
2076                 break;
2077
2078         case SIOCGIFHWADDR:
2079                 /* Get hw address */
2080                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2081                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2082                 if (copy_to_user(argp, &ifr, ifreq_len))
2083                         ret = -EFAULT;
2084                 break;
2085
2086         case SIOCSIFHWADDR:
2087                 /* Set hw address */
2088                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2089                           ifr.ifr_hwaddr.sa_data);
2090
2091                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2092                 break;
2093
2094         case TUNGETSNDBUF:
2095                 sndbuf = tfile->socket.sk->sk_sndbuf;
2096                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2097                         ret = -EFAULT;
2098                 break;
2099
2100         case TUNSETSNDBUF:
2101                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2102                         ret = -EFAULT;
2103                         break;
2104                 }
2105
2106                 tun->sndbuf = sndbuf;
2107                 tun_set_sndbuf(tun);
2108                 break;
2109
2110         case TUNGETVNETHDRSZ:
2111                 vnet_hdr_sz = tun->vnet_hdr_sz;
2112                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2113                         ret = -EFAULT;
2114                 break;
2115
2116         case TUNSETVNETHDRSZ:
2117                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2118                         ret = -EFAULT;
2119                         break;
2120                 }
2121                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2122                         ret = -EINVAL;
2123                         break;
2124                 }
2125
2126                 tun->vnet_hdr_sz = vnet_hdr_sz;
2127                 break;
2128
2129         case TUNGETVNETLE:
2130                 le = !!(tun->flags & TUN_VNET_LE);
2131                 if (put_user(le, (int __user *)argp))
2132                         ret = -EFAULT;
2133                 break;
2134
2135         case TUNSETVNETLE:
2136                 if (get_user(le, (int __user *)argp)) {
2137                         ret = -EFAULT;
2138                         break;
2139                 }
2140                 if (le)
2141                         tun->flags |= TUN_VNET_LE;
2142                 else
2143                         tun->flags &= ~TUN_VNET_LE;
2144                 break;
2145
2146         case TUNGETVNETBE:
2147                 ret = tun_get_vnet_be(tun, argp);
2148                 break;
2149
2150         case TUNSETVNETBE:
2151                 ret = tun_set_vnet_be(tun, argp);
2152                 break;
2153
2154         case TUNATTACHFILTER:
2155                 /* Can be set only for TAPs */
2156                 ret = -EINVAL;
2157                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2158                         break;
2159                 ret = -EFAULT;
2160                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2161                         break;
2162
2163                 ret = tun_attach_filter(tun);
2164                 break;
2165
2166         case TUNDETACHFILTER:
2167                 /* Can be set only for TAPs */
2168                 ret = -EINVAL;
2169                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2170                         break;
2171                 ret = 0;
2172                 tun_detach_filter(tun, tun->numqueues);
2173                 break;
2174
2175         case TUNGETFILTER:
2176                 ret = -EINVAL;
2177                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2178                         break;
2179                 ret = -EFAULT;
2180                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2181                         break;
2182                 ret = 0;
2183                 break;
2184
2185         default:
2186                 ret = -EINVAL;
2187                 break;
2188         }
2189
2190 unlock:
2191         rtnl_unlock();
2192         if (tun)
2193                 tun_put(tun);
2194         return ret;
2195 }
2196
2197 static long tun_chr_ioctl(struct file *file,
2198                           unsigned int cmd, unsigned long arg)
2199 {
2200         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2201 }
2202
2203 #ifdef CONFIG_COMPAT
2204 static long tun_chr_compat_ioctl(struct file *file,
2205                          unsigned int cmd, unsigned long arg)
2206 {
2207         switch (cmd) {
2208         case TUNSETIFF:
2209         case TUNGETIFF:
2210         case TUNSETTXFILTER:
2211         case TUNGETSNDBUF:
2212         case TUNSETSNDBUF:
2213         case SIOCGIFHWADDR:
2214         case SIOCSIFHWADDR:
2215                 arg = (unsigned long)compat_ptr(arg);
2216                 break;
2217         default:
2218                 arg = (compat_ulong_t)arg;
2219                 break;
2220         }
2221
2222         /*
2223          * compat_ifreq is shorter than ifreq, so we must not access beyond
2224          * the end of that structure. All fields that are used in this
2225          * driver are compatible though, we don't need to convert the
2226          * contents.
2227          */
2228         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2229 }
2230 #endif /* CONFIG_COMPAT */
2231
2232 static int tun_chr_fasync(int fd, struct file *file, int on)
2233 {
2234         struct tun_file *tfile = file->private_data;
2235         int ret;
2236
2237         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2238                 goto out;
2239
2240         if (on) {
2241                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2242                 tfile->flags |= TUN_FASYNC;
2243         } else
2244                 tfile->flags &= ~TUN_FASYNC;
2245         ret = 0;
2246 out:
2247         return ret;
2248 }
2249
2250 static int tun_chr_open(struct inode *inode, struct file * file)
2251 {
2252         struct net *net = current->nsproxy->net_ns;
2253         struct tun_file *tfile;
2254
2255         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2256
2257         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2258                                             &tun_proto, 0);
2259         if (!tfile)
2260                 return -ENOMEM;
2261         RCU_INIT_POINTER(tfile->tun, NULL);
2262         tfile->flags = 0;
2263         tfile->ifindex = 0;
2264
2265         init_waitqueue_head(&tfile->wq.wait);
2266         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2267
2268         tfile->socket.file = file;
2269         tfile->socket.ops = &tun_socket_ops;
2270
2271         sock_init_data(&tfile->socket, &tfile->sk);
2272
2273         tfile->sk.sk_write_space = tun_sock_write_space;
2274         tfile->sk.sk_sndbuf = INT_MAX;
2275
2276         file->private_data = tfile;
2277         INIT_LIST_HEAD(&tfile->next);
2278
2279         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2280
2281         return 0;
2282 }
2283
2284 static int tun_chr_close(struct inode *inode, struct file *file)
2285 {
2286         struct tun_file *tfile = file->private_data;
2287
2288         tun_detach(tfile, true);
2289
2290         return 0;
2291 }
2292
2293 #ifdef CONFIG_PROC_FS
2294 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2295 {
2296         struct tun_struct *tun;
2297         struct ifreq ifr;
2298
2299         memset(&ifr, 0, sizeof(ifr));
2300
2301         rtnl_lock();
2302         tun = tun_get(f);
2303         if (tun)
2304                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2305         rtnl_unlock();
2306
2307         if (tun)
2308                 tun_put(tun);
2309
2310         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2311 }
2312 #endif
2313
2314 static const struct file_operations tun_fops = {
2315         .owner  = THIS_MODULE,
2316         .llseek = no_llseek,
2317         .read_iter  = tun_chr_read_iter,
2318         .write_iter = tun_chr_write_iter,
2319         .poll   = tun_chr_poll,
2320         .unlocked_ioctl = tun_chr_ioctl,
2321 #ifdef CONFIG_COMPAT
2322         .compat_ioctl = tun_chr_compat_ioctl,
2323 #endif
2324         .open   = tun_chr_open,
2325         .release = tun_chr_close,
2326         .fasync = tun_chr_fasync,
2327 #ifdef CONFIG_PROC_FS
2328         .show_fdinfo = tun_chr_show_fdinfo,
2329 #endif
2330 };
2331
2332 static struct miscdevice tun_miscdev = {
2333         .minor = TUN_MINOR,
2334         .name = "tun",
2335         .nodename = "net/tun",
2336         .fops = &tun_fops,
2337 };
2338
2339 /* ethtool interface */
2340
2341 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2342 {
2343         cmd->supported          = 0;
2344         cmd->advertising        = 0;
2345         ethtool_cmd_speed_set(cmd, SPEED_10);
2346         cmd->duplex             = DUPLEX_FULL;
2347         cmd->port               = PORT_TP;
2348         cmd->phy_address        = 0;
2349         cmd->transceiver        = XCVR_INTERNAL;
2350         cmd->autoneg            = AUTONEG_DISABLE;
2351         cmd->maxtxpkt           = 0;
2352         cmd->maxrxpkt           = 0;
2353         return 0;
2354 }
2355
2356 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2357 {
2358         struct tun_struct *tun = netdev_priv(dev);
2359
2360         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2361         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2362
2363         switch (tun->flags & TUN_TYPE_MASK) {
2364         case IFF_TUN:
2365                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2366                 break;
2367         case IFF_TAP:
2368                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2369                 break;
2370         }
2371 }
2372
2373 static u32 tun_get_msglevel(struct net_device *dev)
2374 {
2375 #ifdef TUN_DEBUG
2376         struct tun_struct *tun = netdev_priv(dev);
2377         return tun->debug;
2378 #else
2379         return -EOPNOTSUPP;
2380 #endif
2381 }
2382
2383 static void tun_set_msglevel(struct net_device *dev, u32 value)
2384 {
2385 #ifdef TUN_DEBUG
2386         struct tun_struct *tun = netdev_priv(dev);
2387         tun->debug = value;
2388 #endif
2389 }
2390
2391 static const struct ethtool_ops tun_ethtool_ops = {
2392         .get_settings   = tun_get_settings,
2393         .get_drvinfo    = tun_get_drvinfo,
2394         .get_msglevel   = tun_get_msglevel,
2395         .set_msglevel   = tun_set_msglevel,
2396         .get_link       = ethtool_op_get_link,
2397         .get_ts_info    = ethtool_op_get_ts_info,
2398 };
2399
2400
2401 static int __init tun_init(void)
2402 {
2403         int ret = 0;
2404
2405         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2406         pr_info("%s\n", DRV_COPYRIGHT);
2407
2408         ret = rtnl_link_register(&tun_link_ops);
2409         if (ret) {
2410                 pr_err("Can't register link_ops\n");
2411                 goto err_linkops;
2412         }
2413
2414         ret = misc_register(&tun_miscdev);
2415         if (ret) {
2416                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2417                 goto err_misc;
2418         }
2419         return  0;
2420 err_misc:
2421         rtnl_link_unregister(&tun_link_ops);
2422 err_linkops:
2423         return ret;
2424 }
2425
2426 static void tun_cleanup(void)
2427 {
2428         misc_deregister(&tun_miscdev);
2429         rtnl_link_unregister(&tun_link_ops);
2430 }
2431
2432 /* Get an underlying socket object from tun file.  Returns error unless file is
2433  * attached to a device.  The returned object works like a packet socket, it
2434  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2435  * holding a reference to the file for as long as the socket is in use. */
2436 struct socket *tun_get_socket(struct file *file)
2437 {
2438         struct tun_file *tfile;
2439         if (file->f_op != &tun_fops)
2440                 return ERR_PTR(-EINVAL);
2441         tfile = file->private_data;
2442         if (!tfile)
2443                 return ERR_PTR(-EBADFD);
2444         return &tfile->socket;
2445 }
2446 EXPORT_SYMBOL_GPL(tun_get_socket);
2447
2448 module_init(tun_init);
2449 module_exit(tun_cleanup);
2450 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2451 MODULE_AUTHOR(DRV_COPYRIGHT);
2452 MODULE_LICENSE("GPL");
2453 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2454 MODULE_ALIAS("devname:net/tun");