linux/drivers/net/tun.c
<<
>>
Prefs
   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/crc32.h>
  64#include <linux/nsproxy.h>
  65#include <linux/virtio_net.h>
  66#include <linux/rcupdate.h>
  67#include <net/net_namespace.h>
  68#include <net/netns/generic.h>
  69#include <net/rtnetlink.h>
  70#include <net/sock.h>
  71
  72#include <asm/uaccess.h>
  73
  74/* Uncomment to enable debugging */
  75/* #define TUN_DEBUG 1 */
  76
  77#ifdef TUN_DEBUG
  78static int debug;
  79
  80#define tun_debug(level, tun, fmt, args...)                     \
  81do {                                                            \
  82        if (tun->debug)                                         \
  83                netdev_printk(level, tun->dev, fmt, ##args);    \
  84} while (0)
  85#define DBG1(level, fmt, args...)                               \
  86do {                                                            \
  87        if (debug == 2)                                         \
  88                printk(level fmt, ##args);                      \
  89} while (0)
  90#else
  91#define tun_debug(level, tun, fmt, args...)                     \
  92do {                                                            \
  93        if (0)                                                  \
  94                netdev_printk(level, tun->dev, fmt, ##args);    \
  95} while (0)
  96#define DBG1(level, fmt, args...)                               \
  97do {                                                            \
  98        if (0)                                                  \
  99                printk(level fmt, ##args);                      \
 100} while (0)
 101#endif
 102
 103#define GOODCOPY_LEN 128
 104
 105#define FLT_EXACT_COUNT 8
 106struct tap_filter {
 107        unsigned int    count;    /* Number of addrs. Zero means disabled */
 108        u32             mask[2];  /* Mask of the hashed addrs */
 109        unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 110};
 111
 112/* DEFAULT_MAX_NUM_RSS_QUEUES were choosed to let the rx/tx queues allocated for
 113 * the netdevice to be fit in one page. So we can make sure the success of
 114 * memory allocation. TODO: increase the limit. */
 115#define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
 116#define MAX_TAP_FLOWS  4096
 117
 118#define TUN_FLOW_EXPIRE (3 * HZ)
 119
 120/* A tun_file connects an open character device to a tuntap netdevice. It
 121 * also contains all socket related strctures (except sock_fprog and tap_filter)
 122 * to serve as one transmit queue for tuntap device. The sock_fprog and
 123 * tap_filter were kept in tun_struct since they were used for filtering for the
 124 * netdevice not for a specific queue (at least I didn't see the requirement for
 125 * this).
 126 *
 127 * RCU usage:
 128 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 129 * other can only be read while rcu_read_lock or rtnl_lock is held.
 130 */
 131struct tun_file {
 132        struct sock sk;
 133        struct socket socket;
 134        struct socket_wq wq;
 135        struct tun_struct __rcu *tun;
 136        struct net *net;
 137        struct fasync_struct *fasync;
 138        /* only used for fasnyc */
 139        unsigned int flags;
 140        u16 queue_index;
 141        struct list_head next;
 142        struct tun_struct *detached;
 143};
 144
 145struct tun_flow_entry {
 146        struct hlist_node hash_link;
 147        struct rcu_head rcu;
 148        struct tun_struct *tun;
 149
 150        u32 rxhash;
 151        int queue_index;
 152        unsigned long updated;
 153};
 154
 155#define TUN_NUM_FLOW_ENTRIES 1024
 156
 157/* Since the socket were moved to tun_file, to preserve the behavior of persist
 158 * device, socket filter, sndbuf and vnet header size were restore when the
 159 * file were attached to a persist device.
 160 */
 161struct tun_struct {
 162        struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 163        unsigned int            numqueues;
 164        unsigned int            flags;
 165        kuid_t                  owner;
 166        kgid_t                  group;
 167
 168        struct net_device       *dev;
 169        netdev_features_t       set_features;
 170#define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 171                          NETIF_F_TSO6|NETIF_F_UFO)
 172
 173        int                     vnet_hdr_sz;
 174        int                     sndbuf;
 175        struct tap_filter       txflt;
 176        struct sock_fprog       fprog;
 177        /* protected by rtnl lock */
 178        bool                    filter_attached;
 179#ifdef TUN_DEBUG
 180        int debug;
 181#endif
 182        spinlock_t lock;
 183        struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 184        struct timer_list flow_gc_timer;
 185        unsigned long ageing_time;
 186        unsigned int numdisabled;
 187        struct list_head disabled;
 188        void *security;
 189        u32 flow_count;
 190};
 191
 192static inline u32 tun_hashfn(u32 rxhash)
 193{
 194        return rxhash & 0x3ff;
 195}
 196
 197static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 198{
 199        struct tun_flow_entry *e;
 200
 201        hlist_for_each_entry_rcu(e, head, hash_link) {
 202                if (e->rxhash == rxhash)
 203                        return e;
 204        }
 205        return NULL;
 206}
 207
 208static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 209                                              struct hlist_head *head,
 210                                              u32 rxhash, u16 queue_index)
 211{
 212        struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 213
 214        if (e) {
 215                tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 216                          rxhash, queue_index);
 217                e->updated = jiffies;
 218                e->rxhash = rxhash;
 219                e->queue_index = queue_index;
 220                e->tun = tun;
 221                hlist_add_head_rcu(&e->hash_link, head);
 222                ++tun->flow_count;
 223        }
 224        return e;
 225}
 226
 227static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 228{
 229        tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 230                  e->rxhash, e->queue_index);
 231        hlist_del_rcu(&e->hash_link);
 232        kfree_rcu(e, rcu);
 233        --tun->flow_count;
 234}
 235
 236static void tun_flow_flush(struct tun_struct *tun)
 237{
 238        int i;
 239
 240        spin_lock_bh(&tun->lock);
 241        for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 242                struct tun_flow_entry *e;
 243                struct hlist_node *n;
 244
 245                hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 246                        tun_flow_delete(tun, e);
 247        }
 248        spin_unlock_bh(&tun->lock);
 249}
 250
 251static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 252{
 253        int i;
 254
 255        spin_lock_bh(&tun->lock);
 256        for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 257                struct tun_flow_entry *e;
 258                struct hlist_node *n;
 259
 260                hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 261                        if (e->queue_index == queue_index)
 262                                tun_flow_delete(tun, e);
 263                }
 264        }
 265        spin_unlock_bh(&tun->lock);
 266}
 267
 268static void tun_flow_cleanup(unsigned long data)
 269{
 270        struct tun_struct *tun = (struct tun_struct *)data;
 271        unsigned long delay = tun->ageing_time;
 272        unsigned long next_timer = jiffies + delay;
 273        unsigned long count = 0;
 274        int i;
 275
 276        tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 277
 278        spin_lock_bh(&tun->lock);
 279        for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 280                struct tun_flow_entry *e;
 281                struct hlist_node *n;
 282
 283                hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 284                        unsigned long this_timer;
 285                        count++;
 286                        this_timer = e->updated + delay;
 287                        if (time_before_eq(this_timer, jiffies))
 288                                tun_flow_delete(tun, e);
 289                        else if (time_before(this_timer, next_timer))
 290                                next_timer = this_timer;
 291                }
 292        }
 293
 294        if (count)
 295                mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 296        spin_unlock_bh(&tun->lock);
 297}
 298
 299static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 300                            struct tun_file *tfile)
 301{
 302        struct hlist_head *head;
 303        struct tun_flow_entry *e;
 304        unsigned long delay = tun->ageing_time;
 305        u16 queue_index = tfile->queue_index;
 306
 307        if (!rxhash)
 308                return;
 309        else
 310                head = &tun->flows[tun_hashfn(rxhash)];
 311
 312        rcu_read_lock();
 313
 314        /* We may get a very small possibility of OOO during switching, not
 315         * worth to optimize.*/
 316        if (tun->numqueues == 1 || tfile->detached)
 317                goto unlock;
 318
 319        e = tun_flow_find(head, rxhash);
 320        if (likely(e)) {
 321                /* TODO: keep queueing to old queue until it's empty? */
 322                e->queue_index = queue_index;
 323                e->updated = jiffies;
 324        } else {
 325                spin_lock_bh(&tun->lock);
 326                if (!tun_flow_find(head, rxhash) &&
 327                    tun->flow_count < MAX_TAP_FLOWS)
 328                        tun_flow_create(tun, head, rxhash, queue_index);
 329
 330                if (!timer_pending(&tun->flow_gc_timer))
 331                        mod_timer(&tun->flow_gc_timer,
 332                                  round_jiffies_up(jiffies + delay));
 333                spin_unlock_bh(&tun->lock);
 334        }
 335
 336unlock:
 337        rcu_read_unlock();
 338}
 339
 340/* We try to identify a flow through its rxhash first. The reason that
 341 * we do not check rxq no. is becuase some cards(e.g 82599), chooses
 342 * the rxq based on the txq where the last packet of the flow comes. As
 343 * the userspace application move between processors, we may get a
 344 * different rxq no. here. If we could not get rxhash, then we would
 345 * hope the rxq no. may help here.
 346 */
 347static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
 348{
 349        struct tun_struct *tun = netdev_priv(dev);
 350        struct tun_flow_entry *e;
 351        u32 txq = 0;
 352        u32 numqueues = 0;
 353
 354        rcu_read_lock();
 355        numqueues = tun->numqueues;
 356
 357        txq = skb_get_rxhash(skb);
 358        if (txq) {
 359                e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 360                if (e)
 361                        txq = e->queue_index;
 362                else
 363                        /* use multiply and shift instead of expensive divide */
 364                        txq = ((u64)txq * numqueues) >> 32;
 365        } else if (likely(skb_rx_queue_recorded(skb))) {
 366                txq = skb_get_rx_queue(skb);
 367                while (unlikely(txq >= numqueues))
 368                        txq -= numqueues;
 369        }
 370
 371        rcu_read_unlock();
 372        return txq;
 373}
 374
 375static inline bool tun_not_capable(struct tun_struct *tun)
 376{
 377        const struct cred *cred = current_cred();
 378        struct net *net = dev_net(tun->dev);
 379
 380        return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 381                  (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 382                !ns_capable(net->user_ns, CAP_NET_ADMIN);
 383}
 384
 385static void tun_set_real_num_queues(struct tun_struct *tun)
 386{
 387        netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 388        netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 389}
 390
 391static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 392{
 393        tfile->detached = tun;
 394        list_add_tail(&tfile->next, &tun->disabled);
 395        ++tun->numdisabled;
 396}
 397
 398static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 399{
 400        struct tun_struct *tun = tfile->detached;
 401
 402        tfile->detached = NULL;
 403        list_del_init(&tfile->next);
 404        --tun->numdisabled;
 405        return tun;
 406}
 407
 408static void __tun_detach(struct tun_file *tfile, bool clean)
 409{
 410        struct tun_file *ntfile;
 411        struct tun_struct *tun;
 412        struct net_device *dev;
 413
 414        tun = rtnl_dereference(tfile->tun);
 415
 416        if (tun && !tfile->detached) {
 417                u16 index = tfile->queue_index;
 418                BUG_ON(index >= tun->numqueues);
 419                dev = tun->dev;
 420
 421                rcu_assign_pointer(tun->tfiles[index],
 422                                   tun->tfiles[tun->numqueues - 1]);
 423                ntfile = rtnl_dereference(tun->tfiles[index]);
 424                ntfile->queue_index = index;
 425
 426                --tun->numqueues;
 427                if (clean) {
 428                        rcu_assign_pointer(tfile->tun, NULL);
 429                        sock_put(&tfile->sk);
 430                } else
 431                        tun_disable_queue(tun, tfile);
 432
 433                synchronize_net();
 434                tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 435                /* Drop read queue */
 436                skb_queue_purge(&tfile->sk.sk_receive_queue);
 437                tun_set_real_num_queues(tun);
 438        } else if (tfile->detached && clean) {
 439                tun = tun_enable_queue(tfile);
 440                sock_put(&tfile->sk);
 441        }
 442
 443        if (clean) {
 444                if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 445                        netif_carrier_off(tun->dev);
 446
 447                        if (!(tun->flags & TUN_PERSIST) &&
 448                            tun->dev->reg_state == NETREG_REGISTERED)
 449                                unregister_netdevice(tun->dev);
 450                }
 451
 452                BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 453                                 &tfile->socket.flags));
 454                sk_release_kernel(&tfile->sk);
 455        }
 456}
 457
 458static void tun_detach(struct tun_file *tfile, bool clean)
 459{
 460        rtnl_lock();
 461        __tun_detach(tfile, clean);
 462        rtnl_unlock();
 463}
 464
 465static void tun_detach_all(struct net_device *dev)
 466{
 467        struct tun_struct *tun = netdev_priv(dev);
 468        struct tun_file *tfile, *tmp;
 469        int i, n = tun->numqueues;
 470
 471        for (i = 0; i < n; i++) {
 472                tfile = rtnl_dereference(tun->tfiles[i]);
 473                BUG_ON(!tfile);
 474                wake_up_all(&tfile->wq.wait);
 475                rcu_assign_pointer(tfile->tun, NULL);
 476                --tun->numqueues;
 477        }
 478        list_for_each_entry(tfile, &tun->disabled, next) {
 479                wake_up_all(&tfile->wq.wait);
 480                rcu_assign_pointer(tfile->tun, NULL);
 481        }
 482        BUG_ON(tun->numqueues != 0);
 483
 484        synchronize_net();
 485        for (i = 0; i < n; i++) {
 486                tfile = rtnl_dereference(tun->tfiles[i]);
 487                /* Drop read queue */
 488                skb_queue_purge(&tfile->sk.sk_receive_queue);
 489                sock_put(&tfile->sk);
 490        }
 491        list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 492                tun_enable_queue(tfile);
 493                skb_queue_purge(&tfile->sk.sk_receive_queue);
 494                sock_put(&tfile->sk);
 495        }
 496        BUG_ON(tun->numdisabled != 0);
 497
 498        if (tun->flags & TUN_PERSIST)
 499                module_put(THIS_MODULE);
 500}
 501
 502static int tun_attach(struct tun_struct *tun, struct file *file)
 503{
 504        struct tun_file *tfile = file->private_data;
 505        int err;
 506
 507        err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 508        if (err < 0)
 509                goto out;
 510
 511        err = -EINVAL;
 512        if (rtnl_dereference(tfile->tun) && !tfile->detached)
 513                goto out;
 514
 515        err = -EBUSY;
 516        if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 517                goto out;
 518
 519        err = -E2BIG;
 520        if (!tfile->detached &&
 521            tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 522                goto out;
 523
 524        err = 0;
 525
 526        /* Re-attach the filter to presist device */
 527        if (tun->filter_attached == true) {
 528                err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 529                if (!err)
 530                        goto out;
 531        }
 532        tfile->queue_index = tun->numqueues;
 533        rcu_assign_pointer(tfile->tun, tun);
 534        rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 535        tun->numqueues++;
 536
 537        if (tfile->detached)
 538                tun_enable_queue(tfile);
 539        else
 540                sock_hold(&tfile->sk);
 541
 542        tun_set_real_num_queues(tun);
 543
 544        /* device is allowed to go away first, so no need to hold extra
 545         * refcnt.
 546         */
 547
 548out:
 549        return err;
 550}
 551
 552static struct tun_struct *__tun_get(struct tun_file *tfile)
 553{
 554        struct tun_struct *tun;
 555
 556        rcu_read_lock();
 557        tun = rcu_dereference(tfile->tun);
 558        if (tun)
 559                dev_hold(tun->dev);
 560        rcu_read_unlock();
 561
 562        return tun;
 563}
 564
 565static struct tun_struct *tun_get(struct file *file)
 566{
 567        return __tun_get(file->private_data);
 568}
 569
 570static void tun_put(struct tun_struct *tun)
 571{
 572        dev_put(tun->dev);
 573}
 574
 575/* TAP filtering */
 576static void addr_hash_set(u32 *mask, const u8 *addr)
 577{
 578        int n = ether_crc(ETH_ALEN, addr) >> 26;
 579        mask[n >> 5] |= (1 << (n & 31));
 580}
 581
 582static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 583{
 584        int n = ether_crc(ETH_ALEN, addr) >> 26;
 585        return mask[n >> 5] & (1 << (n & 31));
 586}
 587
 588static int update_filter(struct tap_filter *filter, void __user *arg)
 589{
 590        struct { u8 u[ETH_ALEN]; } *addr;
 591        struct tun_filter uf;
 592        int err, alen, n, nexact;
 593
 594        if (copy_from_user(&uf, arg, sizeof(uf)))
 595                return -EFAULT;
 596
 597        if (!uf.count) {
 598                /* Disabled */
 599                filter->count = 0;
 600                return 0;
 601        }
 602
 603        alen = ETH_ALEN * uf.count;
 604        addr = kmalloc(alen, GFP_KERNEL);
 605        if (!addr)
 606                return -ENOMEM;
 607
 608        if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 609                err = -EFAULT;
 610                goto done;
 611        }
 612
 613        /* The filter is updated without holding any locks. Which is
 614         * perfectly safe. We disable it first and in the worst
 615         * case we'll accept a few undesired packets. */
 616        filter->count = 0;
 617        wmb();
 618
 619        /* Use first set of addresses as an exact filter */
 620        for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 621                memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 622
 623        nexact = n;
 624
 625        /* Remaining multicast addresses are hashed,
 626         * unicast will leave the filter disabled. */
 627        memset(filter->mask, 0, sizeof(filter->mask));
 628        for (; n < uf.count; n++) {
 629                if (!is_multicast_ether_addr(addr[n].u)) {
 630                        err = 0; /* no filter */
 631                        goto done;
 632                }
 633                addr_hash_set(filter->mask, addr[n].u);
 634        }
 635
 636        /* For ALLMULTI just set the mask to all ones.
 637         * This overrides the mask populated above. */
 638        if ((uf.flags & TUN_FLT_ALLMULTI))
 639                memset(filter->mask, ~0, sizeof(filter->mask));
 640
 641        /* Now enable the filter */
 642        wmb();
 643        filter->count = nexact;
 644
 645        /* Return the number of exact filters */
 646        err = nexact;
 647
 648done:
 649        kfree(addr);
 650        return err;
 651}
 652
 653/* Returns: 0 - drop, !=0 - accept */
 654static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 655{
 656        /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 657         * at this point. */
 658        struct ethhdr *eh = (struct ethhdr *) skb->data;
 659        int i;
 660
 661        /* Exact match */
 662        for (i = 0; i < filter->count; i++)
 663                if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 664                        return 1;
 665
 666        /* Inexact match (multicast only) */
 667        if (is_multicast_ether_addr(eh->h_dest))
 668                return addr_hash_test(filter->mask, eh->h_dest);
 669
 670        return 0;
 671}
 672
 673/*
 674 * Checks whether the packet is accepted or not.
 675 * Returns: 0 - drop, !=0 - accept
 676 */
 677static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 678{
 679        if (!filter->count)
 680                return 1;
 681
 682        return run_filter(filter, skb);
 683}
 684
 685/* Network device part of the driver */
 686
 687static const struct ethtool_ops tun_ethtool_ops;
 688
 689/* Net device detach from fd. */
 690static void tun_net_uninit(struct net_device *dev)
 691{
 692        tun_detach_all(dev);
 693}
 694
 695/* Net device open. */
 696static int tun_net_open(struct net_device *dev)
 697{
 698        netif_tx_start_all_queues(dev);
 699        return 0;
 700}
 701
 702/* Net device close. */
 703static int tun_net_close(struct net_device *dev)
 704{
 705        netif_tx_stop_all_queues(dev);
 706        return 0;
 707}
 708
 709/* Net device start xmit */
 710static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 711{
 712        struct tun_struct *tun = netdev_priv(dev);
 713        int txq = skb->queue_mapping;
 714        struct tun_file *tfile;
 715
 716        rcu_read_lock();
 717        tfile = rcu_dereference(tun->tfiles[txq]);
 718
 719        /* Drop packet if interface is not attached */
 720        if (txq >= tun->numqueues)
 721                goto drop;
 722
 723        tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 724
 725        BUG_ON(!tfile);
 726
 727        /* Drop if the filter does not like it.
 728         * This is a noop if the filter is disabled.
 729         * Filter can be enabled only for the TAP devices. */
 730        if (!check_filter(&tun->txflt, skb))
 731                goto drop;
 732
 733        if (tfile->socket.sk->sk_filter &&
 734            sk_filter(tfile->socket.sk, skb))
 735                goto drop;
 736
 737        /* Limit the number of packets queued by dividing txq length with the
 738         * number of queues.
 739         */
 740        if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
 741                          >= dev->tx_queue_len / tun->numqueues)
 742                goto drop;
 743
 744        /* Orphan the skb - required as we might hang on to it
 745         * for indefinite time. */
 746        if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 747                goto drop;
 748        skb_orphan(skb);
 749
 750        nf_reset(skb);
 751
 752        /* Enqueue packet */
 753        skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 754
 755        /* Notify and wake up reader process */
 756        if (tfile->flags & TUN_FASYNC)
 757                kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 758        wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
 759                                   POLLRDNORM | POLLRDBAND);
 760
 761        rcu_read_unlock();
 762        return NETDEV_TX_OK;
 763
 764drop:
 765        dev->stats.tx_dropped++;
 766        skb_tx_error(skb);
 767        kfree_skb(skb);
 768        rcu_read_unlock();
 769        return NETDEV_TX_OK;
 770}
 771
 772static void tun_net_mclist(struct net_device *dev)
 773{
 774        /*
 775         * This callback is supposed to deal with mc filter in
 776         * _rx_ path and has nothing to do with the _tx_ path.
 777         * In rx path we always accept everything userspace gives us.
 778         */
 779}
 780
 781#define MIN_MTU 68
 782#define MAX_MTU 65535
 783
 784static int
 785tun_net_change_mtu(struct net_device *dev, int new_mtu)
 786{
 787        if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 788                return -EINVAL;
 789        dev->mtu = new_mtu;
 790        return 0;
 791}
 792
 793static netdev_features_t tun_net_fix_features(struct net_device *dev,
 794        netdev_features_t features)
 795{
 796        struct tun_struct *tun = netdev_priv(dev);
 797
 798        return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 799}
 800#ifdef CONFIG_NET_POLL_CONTROLLER
 801static void tun_poll_controller(struct net_device *dev)
 802{
 803        /*
 804         * Tun only receives frames when:
 805         * 1) the char device endpoint gets data from user space
 806         * 2) the tun socket gets a sendmsg call from user space
 807         * Since both of those are syncronous operations, we are guaranteed
 808         * never to have pending data when we poll for it
 809         * so theres nothing to do here but return.
 810         * We need this though so netpoll recognizes us as an interface that
 811         * supports polling, which enables bridge devices in virt setups to
 812         * still use netconsole
 813         */
 814        return;
 815}
 816#endif
 817static const struct net_device_ops tun_netdev_ops = {
 818        .ndo_uninit             = tun_net_uninit,
 819        .ndo_open               = tun_net_open,
 820        .ndo_stop               = tun_net_close,
 821        .ndo_start_xmit         = tun_net_xmit,
 822        .ndo_change_mtu         = tun_net_change_mtu,
 823        .ndo_fix_features       = tun_net_fix_features,
 824        .ndo_select_queue       = tun_select_queue,
 825#ifdef CONFIG_NET_POLL_CONTROLLER
 826        .ndo_poll_controller    = tun_poll_controller,
 827#endif
 828};
 829
 830static const struct net_device_ops tap_netdev_ops = {
 831        .ndo_uninit             = tun_net_uninit,
 832        .ndo_open               = tun_net_open,
 833        .ndo_stop               = tun_net_close,
 834        .ndo_start_xmit         = tun_net_xmit,
 835        .ndo_change_mtu         = tun_net_change_mtu,
 836        .ndo_fix_features       = tun_net_fix_features,
 837        .ndo_set_rx_mode        = tun_net_mclist,
 838        .ndo_set_mac_address    = eth_mac_addr,
 839        .ndo_validate_addr      = eth_validate_addr,
 840        .ndo_select_queue       = tun_select_queue,
 841#ifdef CONFIG_NET_POLL_CONTROLLER
 842        .ndo_poll_controller    = tun_poll_controller,
 843#endif
 844};
 845
 846static int tun_flow_init(struct tun_struct *tun)
 847{
 848        int i;
 849
 850        for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 851                INIT_HLIST_HEAD(&tun->flows[i]);
 852
 853        tun->ageing_time = TUN_FLOW_EXPIRE;
 854        setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 855        mod_timer(&tun->flow_gc_timer,
 856                  round_jiffies_up(jiffies + tun->ageing_time));
 857
 858        return 0;
 859}
 860
 861static void tun_flow_uninit(struct tun_struct *tun)
 862{
 863        del_timer_sync(&tun->flow_gc_timer);
 864        tun_flow_flush(tun);
 865}
 866
 867/* Initialize net device. */
 868static void tun_net_init(struct net_device *dev)
 869{
 870        struct tun_struct *tun = netdev_priv(dev);
 871
 872        switch (tun->flags & TUN_TYPE_MASK) {
 873        case TUN_TUN_DEV:
 874                dev->netdev_ops = &tun_netdev_ops;
 875
 876                /* Point-to-Point TUN Device */
 877                dev->hard_header_len = 0;
 878                dev->addr_len = 0;
 879                dev->mtu = 1500;
 880
 881                /* Zero header length */
 882                dev->type = ARPHRD_NONE;
 883                dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 884                dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 885                break;
 886
 887        case TUN_TAP_DEV:
 888                dev->netdev_ops = &tap_netdev_ops;
 889                /* Ethernet TAP Device */
 890                ether_setup(dev);
 891                dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 892                dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 893
 894                eth_hw_addr_random(dev);
 895
 896                dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 897                break;
 898        }
 899}
 900
 901/* Character device part */
 902
 903/* Poll */
 904static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 905{
 906        struct tun_file *tfile = file->private_data;
 907        struct tun_struct *tun = __tun_get(tfile);
 908        struct sock *sk;
 909        unsigned int mask = 0;
 910
 911        if (!tun)
 912                return POLLERR;
 913
 914        sk = tfile->socket.sk;
 915
 916        tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 917
 918        poll_wait(file, &tfile->wq.wait, wait);
 919
 920        if (!skb_queue_empty(&sk->sk_receive_queue))
 921                mask |= POLLIN | POLLRDNORM;
 922
 923        if (sock_writeable(sk) ||
 924            (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 925             sock_writeable(sk)))
 926                mask |= POLLOUT | POLLWRNORM;
 927
 928        if (tun->dev->reg_state != NETREG_REGISTERED)
 929                mask = POLLERR;
 930
 931        tun_put(tun);
 932        return mask;
 933}
 934
 935/* prepad is the amount to reserve at front.  len is length after that.
 936 * linear is a hint as to how much to copy (usually headers). */
 937static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 938                                     size_t prepad, size_t len,
 939                                     size_t linear, int noblock)
 940{
 941        struct sock *sk = tfile->socket.sk;
 942        struct sk_buff *skb;
 943        int err;
 944
 945        /* Under a page?  Don't bother with paged skb. */
 946        if (prepad + len < PAGE_SIZE || !linear)
 947                linear = len;
 948
 949        skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 950                                   &err);
 951        if (!skb)
 952                return ERR_PTR(err);
 953
 954        skb_reserve(skb, prepad);
 955        skb_put(skb, linear);
 956        skb->data_len = len - linear;
 957        skb->len += len - linear;
 958
 959        return skb;
 960}
 961
 962/* set skb frags from iovec, this can move to core network code for reuse */
 963static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 964                                  int offset, size_t count)
 965{
 966        int len = iov_length(from, count) - offset;
 967        int copy = skb_headlen(skb);
 968        int size, offset1 = 0;
 969        int i = 0;
 970
 971        /* Skip over from offset */
 972        while (count && (offset >= from->iov_len)) {
 973                offset -= from->iov_len;
 974                ++from;
 975                --count;
 976        }
 977
 978        /* copy up to skb headlen */
 979        while (count && (copy > 0)) {
 980                size = min_t(unsigned int, copy, from->iov_len - offset);
 981                if (copy_from_user(skb->data + offset1, from->iov_base + offset,
 982                                   size))
 983                        return -EFAULT;
 984                if (copy > size) {
 985                        ++from;
 986                        --count;
 987                        offset = 0;
 988                } else
 989                        offset += size;
 990                copy -= size;
 991                offset1 += size;
 992        }
 993
 994        if (len == offset1)
 995                return 0;
 996
 997        while (count--) {
 998                struct page *page[MAX_SKB_FRAGS];
 999                int num_pages;
1000                unsigned long base;
1001                unsigned long truesize;
1002
1003                len = from->iov_len - offset;
1004                if (!len) {
1005                        offset = 0;
1006                        ++from;
1007                        continue;
1008                }
1009                base = (unsigned long)from->iov_base + offset;
1010                size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
1011                if (i + size > MAX_SKB_FRAGS)
1012                        return -EMSGSIZE;
1013                num_pages = get_user_pages_fast(base, size, 0, &page[i]);
1014                if (num_pages != size) {
1015                        for (i = 0; i < num_pages; i++)
1016                                put_page(page[i]);
1017                        return -EFAULT;
1018                }
1019                truesize = size * PAGE_SIZE;
1020                skb->data_len += len;
1021                skb->len += len;
1022                skb->truesize += truesize;
1023                atomic_add(truesize, &skb->sk->sk_wmem_alloc);
1024                while (len) {
1025                        int off = base & ~PAGE_MASK;
1026                        int size = min_t(int, len, PAGE_SIZE - off);
1027                        __skb_fill_page_desc(skb, i, page[i], off, size);
1028                        skb_shinfo(skb)->nr_frags++;
1029                        /* increase sk_wmem_alloc */
1030                        base += size;
1031                        len -= size;
1032                        i++;
1033                }
1034                offset = 0;
1035                ++from;
1036        }
1037        return 0;
1038}
1039
1040/* Get packet from user space buffer */
1041static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1042                            void *msg_control, const struct iovec *iv,
1043                            size_t total_len, size_t count, int noblock)
1044{
1045        struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1046        struct sk_buff *skb;
1047        size_t len = total_len, align = NET_SKB_PAD;
1048        struct virtio_net_hdr gso = { 0 };
1049        int offset = 0;
1050        int copylen;
1051        bool zerocopy = false;
1052        int err;
1053        u32 rxhash;
1054
1055        if (!(tun->flags & TUN_NO_PI)) {
1056                if ((len -= sizeof(pi)) > total_len)
1057                        return -EINVAL;
1058
1059                if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1060                        return -EFAULT;
1061                offset += sizeof(pi);
1062        }
1063
1064        if (tun->flags & TUN_VNET_HDR) {
1065                if ((len -= tun->vnet_hdr_sz) > total_len)
1066                        return -EINVAL;
1067
1068                if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1069                        return -EFAULT;
1070
1071                if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1072                    gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1073                        gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1074
1075                if (gso.hdr_len > len)
1076                        return -EINVAL;
1077                offset += tun->vnet_hdr_sz;
1078        }
1079
1080        if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1081                align += NET_IP_ALIGN;
1082                if (unlikely(len < ETH_HLEN ||
1083                             (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1084                        return -EINVAL;
1085        }
1086
1087        if (msg_control)
1088                zerocopy = true;
1089
1090        if (zerocopy) {
1091                /* Userspace may produce vectors with count greater than
1092                 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1093                 * to let the rest of data to be fit in the frags.
1094                 */
1095                if (count > MAX_SKB_FRAGS) {
1096                        copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1097                        if (copylen < offset)
1098                                copylen = 0;
1099                        else
1100                                copylen -= offset;
1101                } else
1102                                copylen = 0;
1103                /* There are 256 bytes to be copied in skb, so there is enough
1104                 * room for skb expand head in case it is used.
1105                 * The rest of the buffer is mapped from userspace.
1106                 */
1107                if (copylen < gso.hdr_len)
1108                        copylen = gso.hdr_len;
1109                if (!copylen)
1110                        copylen = GOODCOPY_LEN;
1111        } else
1112                copylen = len;
1113
1114        skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
1115        if (IS_ERR(skb)) {
1116                if (PTR_ERR(skb) != -EAGAIN)
1117                        tun->dev->stats.rx_dropped++;
1118                return PTR_ERR(skb);
1119        }
1120
1121        if (zerocopy)
1122                err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1123        else
1124                err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1125
1126        if (err) {
1127                tun->dev->stats.rx_dropped++;
1128                kfree_skb(skb);
1129                return -EFAULT;
1130        }
1131
1132        if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1133                if (!skb_partial_csum_set(skb, gso.csum_start,
1134                                          gso.csum_offset)) {
1135                        tun->dev->stats.rx_frame_errors++;
1136                        kfree_skb(skb);
1137                        return -EINVAL;
1138                }
1139        }
1140
1141        switch (tun->flags & TUN_TYPE_MASK) {
1142        case TUN_TUN_DEV:
1143                if (tun->flags & TUN_NO_PI) {
1144                        switch (skb->data[0] & 0xf0) {
1145                        case 0x40:
1146                                pi.proto = htons(ETH_P_IP);
1147                                break;
1148                        case 0x60:
1149                                pi.proto = htons(ETH_P_IPV6);
1150                                break;
1151                        default:
1152                                tun->dev->stats.rx_dropped++;
1153                                kfree_skb(skb);
1154                                return -EINVAL;
1155                        }
1156                }
1157
1158                skb_reset_mac_header(skb);
1159                skb->protocol = pi.proto;
1160                skb->dev = tun->dev;
1161                break;
1162        case TUN_TAP_DEV:
1163                skb->protocol = eth_type_trans(skb, tun->dev);
1164                break;
1165        }
1166
1167        if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1168                pr_debug("GSO!\n");
1169                switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1170                case VIRTIO_NET_HDR_GSO_TCPV4:
1171                        skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1172                        break;
1173                case VIRTIO_NET_HDR_GSO_TCPV6:
1174                        skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1175                        break;
1176                case VIRTIO_NET_HDR_GSO_UDP:
1177                        skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1178                        break;
1179                default:
1180                        tun->dev->stats.rx_frame_errors++;
1181                        kfree_skb(skb);
1182                        return -EINVAL;
1183                }
1184
1185                if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1186                        skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1187
1188                skb_shinfo(skb)->gso_size = gso.gso_size;
1189                if (skb_shinfo(skb)->gso_size == 0) {
1190                        tun->dev->stats.rx_frame_errors++;
1191                        kfree_skb(skb);
1192                        return -EINVAL;
1193                }
1194
1195                /* Header must be checked, and gso_segs computed. */
1196                skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1197                skb_shinfo(skb)->gso_segs = 0;
1198        }
1199
1200        /* copy skb_ubuf_info for callback when skb has no error */
1201        if (zerocopy) {
1202                skb_shinfo(skb)->destructor_arg = msg_control;
1203                skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1204                skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1205        }
1206
1207        skb_reset_network_header(skb);
1208        rxhash = skb_get_rxhash(skb);
1209        netif_rx_ni(skb);
1210
1211        tun->dev->stats.rx_packets++;
1212        tun->dev->stats.rx_bytes += len;
1213
1214        tun_flow_update(tun, rxhash, tfile);
1215        return total_len;
1216}
1217
1218static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1219                              unsigned long count, loff_t pos)
1220{
1221        struct file *file = iocb->ki_filp;
1222        struct tun_struct *tun = tun_get(file);
1223        struct tun_file *tfile = file->private_data;
1224        ssize_t result;
1225
1226        if (!tun)
1227                return -EBADFD;
1228
1229        tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1230
1231        result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1232                              count, file->f_flags & O_NONBLOCK);
1233
1234        tun_put(tun);
1235        return result;
1236}
1237
1238/* Put packet to the user space buffer */
1239static ssize_t tun_put_user(struct tun_struct *tun,
1240                            struct tun_file *tfile,
1241                            struct sk_buff *skb,
1242                            const struct iovec *iv, int len)
1243{
1244        struct tun_pi pi = { 0, skb->protocol };
1245        ssize_t total = 0;
1246
1247        if (!(tun->flags & TUN_NO_PI)) {
1248                if ((len -= sizeof(pi)) < 0)
1249                        return -EINVAL;
1250
1251                if (len < skb->len) {
1252                        /* Packet will be striped */
1253                        pi.flags |= TUN_PKT_STRIP;
1254                }
1255
1256                if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1257                        return -EFAULT;
1258                total += sizeof(pi);
1259        }
1260
1261        if (tun->flags & TUN_VNET_HDR) {
1262                struct virtio_net_hdr gso = { 0 }; /* no info leak */
1263                if ((len -= tun->vnet_hdr_sz) < 0)
1264                        return -EINVAL;
1265
1266                if (skb_is_gso(skb)) {
1267                        struct skb_shared_info *sinfo = skb_shinfo(skb);
1268
1269                        /* This is a hint as to how much should be linear. */
1270                        gso.hdr_len = skb_headlen(skb);
1271                        gso.gso_size = sinfo->gso_size;
1272                        if (sinfo->gso_type & SKB_GSO_TCPV4)
1273                                gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1274                        else if (sinfo->gso_type & SKB_GSO_TCPV6)
1275                                gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1276                        else if (sinfo->gso_type & SKB_GSO_UDP)
1277                                gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1278                        else {
1279                                pr_err("unexpected GSO type: "
1280                                       "0x%x, gso_size %d, hdr_len %d\n",
1281                                       sinfo->gso_type, gso.gso_size,
1282                                       gso.hdr_len);
1283                                print_hex_dump(KERN_ERR, "tun: ",
1284                                               DUMP_PREFIX_NONE,
1285                                               16, 1, skb->head,
1286                                               min((int)gso.hdr_len, 64), true);
1287                                WARN_ON_ONCE(1);
1288                                return -EINVAL;
1289                        }
1290                        if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1291                                gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1292                } else
1293                        gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1294
1295                if (skb->ip_summed == CHECKSUM_PARTIAL) {
1296                        gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1297                        gso.csum_start = skb_checksum_start_offset(skb);
1298                        gso.csum_offset = skb->csum_offset;
1299                } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1300                        gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1301                } /* else everything is zero */
1302
1303                if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1304                                               sizeof(gso))))
1305                        return -EFAULT;
1306                total += tun->vnet_hdr_sz;
1307        }
1308
1309        len = min_t(int, skb->len, len);
1310
1311        skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1312        total += skb->len;
1313
1314        tun->dev->stats.tx_packets++;
1315        tun->dev->stats.tx_bytes += len;
1316
1317        return total;
1318}
1319
1320static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1321                           struct kiocb *iocb, const struct iovec *iv,
1322                           ssize_t len, int noblock)
1323{
1324        DECLARE_WAITQUEUE(wait, current);
1325        struct sk_buff *skb;
1326        ssize_t ret = 0;
1327
1328        tun_debug(KERN_INFO, tun, "tun_do_read\n");
1329
1330        if (unlikely(!noblock))
1331                add_wait_queue(&tfile->wq.wait, &wait);
1332        while (len) {
1333                current->state = TASK_INTERRUPTIBLE;
1334
1335                /* Read frames from the queue */
1336                if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1337                        if (noblock) {
1338                                ret = -EAGAIN;
1339                                break;
1340                        }
1341                        if (signal_pending(current)) {
1342                                ret = -ERESTARTSYS;
1343                                break;
1344                        }
1345                        if (tun->dev->reg_state != NETREG_REGISTERED) {
1346                                ret = -EIO;
1347                                break;
1348                        }
1349
1350                        /* Nothing to read, let's sleep */
1351                        schedule();
1352                        continue;
1353                }
1354
1355                ret = tun_put_user(tun, tfile, skb, iv, len);
1356                kfree_skb(skb);
1357                break;
1358        }
1359
1360        current->state = TASK_RUNNING;
1361        if (unlikely(!noblock))
1362                remove_wait_queue(&tfile->wq.wait, &wait);
1363
1364        return ret;
1365}
1366
1367static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1368                            unsigned long count, loff_t pos)
1369{
1370        struct file *file = iocb->ki_filp;
1371        struct tun_file *tfile = file->private_data;
1372        struct tun_struct *tun = __tun_get(tfile);
1373        ssize_t len, ret;
1374
1375        if (!tun)
1376                return -EBADFD;
1377        len = iov_length(iv, count);
1378        if (len < 0) {
1379                ret = -EINVAL;
1380                goto out;
1381        }
1382
1383        ret = tun_do_read(tun, tfile, iocb, iv, len,
1384                          file->f_flags & O_NONBLOCK);
1385        ret = min_t(ssize_t, ret, len);
1386out:
1387        tun_put(tun);
1388        return ret;
1389}
1390
1391static void tun_free_netdev(struct net_device *dev)
1392{
1393        struct tun_struct *tun = netdev_priv(dev);
1394
1395        BUG_ON(!(list_empty(&tun->disabled)));
1396        tun_flow_uninit(tun);
1397        security_tun_dev_free_security(tun->security);
1398        free_netdev(dev);
1399}
1400
1401static void tun_setup(struct net_device *dev)
1402{
1403        struct tun_struct *tun = netdev_priv(dev);
1404
1405        tun->owner = INVALID_UID;
1406        tun->group = INVALID_GID;
1407
1408        dev->ethtool_ops = &tun_ethtool_ops;
1409        dev->destructor = tun_free_netdev;
1410}
1411
1412/* Trivial set of netlink ops to allow deleting tun or tap
1413 * device with netlink.
1414 */
1415static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1416{
1417        return -EINVAL;
1418}
1419
1420static struct rtnl_link_ops tun_link_ops __read_mostly = {
1421        .kind           = DRV_NAME,
1422        .priv_size      = sizeof(struct tun_struct),
1423        .setup          = tun_setup,
1424        .validate       = tun_validate,
1425};
1426
1427static void tun_sock_write_space(struct sock *sk)
1428{
1429        struct tun_file *tfile;
1430        wait_queue_head_t *wqueue;
1431
1432        if (!sock_writeable(sk))
1433                return;
1434
1435        if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1436                return;
1437
1438        wqueue = sk_sleep(sk);
1439        if (wqueue && waitqueue_active(wqueue))
1440                wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1441                                                POLLWRNORM | POLLWRBAND);
1442
1443        tfile = container_of(sk, struct tun_file, sk);
1444        kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1445}
1446
1447static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1448                       struct msghdr *m, size_t total_len)
1449{
1450        int ret;
1451        struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1452        struct tun_struct *tun = __tun_get(tfile);
1453
1454        if (!tun)
1455                return -EBADFD;
1456        ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1457                           m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1458        tun_put(tun);
1459        return ret;
1460}
1461
1462
1463static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1464                       struct msghdr *m, size_t total_len,
1465                       int flags)
1466{
1467        struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1468        struct tun_struct *tun = __tun_get(tfile);
1469        int ret;
1470
1471        if (!tun)
1472                return -EBADFD;
1473
1474        if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) {
1475                ret = -EINVAL;
1476                goto out;
1477        }
1478        ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1479                          flags & MSG_DONTWAIT);
1480        if (ret > total_len) {
1481                m->msg_flags |= MSG_TRUNC;
1482                ret = flags & MSG_TRUNC ? ret : total_len;
1483        }
1484out:
1485        tun_put(tun);
1486        return ret;
1487}
1488
1489static int tun_release(struct socket *sock)
1490{
1491        if (sock->sk)
1492                sock_put(sock->sk);
1493        return 0;
1494}
1495
1496/* Ops structure to mimic raw sockets with tun */
1497static const struct proto_ops tun_socket_ops = {
1498        .sendmsg = tun_sendmsg,
1499        .recvmsg = tun_recvmsg,
1500        .release = tun_release,
1501};
1502
1503static struct proto tun_proto = {
1504        .name           = "tun",
1505        .owner          = THIS_MODULE,
1506        .obj_size       = sizeof(struct tun_file),
1507};
1508
1509static int tun_flags(struct tun_struct *tun)
1510{
1511        int flags = 0;
1512
1513        if (tun->flags & TUN_TUN_DEV)
1514                flags |= IFF_TUN;
1515        else
1516                flags |= IFF_TAP;
1517
1518        if (tun->flags & TUN_NO_PI)
1519                flags |= IFF_NO_PI;
1520
1521        /* This flag has no real effect.  We track the value for backwards
1522         * compatibility.
1523         */
1524        if (tun->flags & TUN_ONE_QUEUE)
1525                flags |= IFF_ONE_QUEUE;
1526
1527        if (tun->flags & TUN_VNET_HDR)
1528                flags |= IFF_VNET_HDR;
1529
1530        if (tun->flags & TUN_TAP_MQ)
1531                flags |= IFF_MULTI_QUEUE;
1532
1533        return flags;
1534}
1535
1536static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1537                              char *buf)
1538{
1539        struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1540        return sprintf(buf, "0x%x\n", tun_flags(tun));
1541}
1542
1543static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1544                              char *buf)
1545{
1546        struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1547        return uid_valid(tun->owner)?
1548                sprintf(buf, "%u\n",
1549                        from_kuid_munged(current_user_ns(), tun->owner)):
1550                sprintf(buf, "-1\n");
1551}
1552
1553static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1554                              char *buf)
1555{
1556        struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1557        return gid_valid(tun->group) ?
1558                sprintf(buf, "%u\n",
1559                        from_kgid_munged(current_user_ns(), tun->group)):
1560                sprintf(buf, "-1\n");
1561}
1562
1563static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1564static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1565static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1566
1567static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1568{
1569        struct tun_struct *tun;
1570        struct tun_file *tfile = file->private_data;
1571        struct net_device *dev;
1572        int err;
1573
1574        if (tfile->detached)
1575                return -EINVAL;
1576
1577        dev = __dev_get_by_name(net, ifr->ifr_name);
1578        if (dev) {
1579                if (ifr->ifr_flags & IFF_TUN_EXCL)
1580                        return -EBUSY;
1581                if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1582                        tun = netdev_priv(dev);
1583                else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1584                        tun = netdev_priv(dev);
1585                else
1586                        return -EINVAL;
1587
1588                if (tun_not_capable(tun))
1589                        return -EPERM;
1590                err = security_tun_dev_open(tun->security);
1591                if (err < 0)
1592                        return err;
1593
1594                err = tun_attach(tun, file);
1595                if (err < 0)
1596                        return err;
1597
1598                if (tun->flags & TUN_TAP_MQ &&
1599                    (tun->numqueues + tun->numdisabled > 1))
1600                        return -EBUSY;
1601        }
1602        else {
1603                char *name;
1604                unsigned long flags = 0;
1605                int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1606                             MAX_TAP_QUEUES : 1;
1607
1608                if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1609                        return -EPERM;
1610                err = security_tun_dev_create();
1611                if (err < 0)
1612                        return err;
1613
1614                /* Set dev type */
1615                if (ifr->ifr_flags & IFF_TUN) {
1616                        /* TUN device */
1617                        flags |= TUN_TUN_DEV;
1618                        name = "tun%d";
1619                } else if (ifr->ifr_flags & IFF_TAP) {
1620                        /* TAP device */
1621                        flags |= TUN_TAP_DEV;
1622                        name = "tap%d";
1623                } else
1624                        return -EINVAL;
1625
1626                if (*ifr->ifr_name)
1627                        name = ifr->ifr_name;
1628
1629                dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1630                                       tun_setup, queues, queues);
1631
1632                if (!dev)
1633                        return -ENOMEM;
1634
1635                dev_net_set(dev, net);
1636                dev->rtnl_link_ops = &tun_link_ops;
1637
1638                tun = netdev_priv(dev);
1639                tun->dev = dev;
1640                tun->flags = flags;
1641                tun->txflt.count = 0;
1642                tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1643
1644                tun->filter_attached = false;
1645                tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1646
1647                spin_lock_init(&tun->lock);
1648
1649                err = security_tun_dev_alloc_security(&tun->security);
1650                if (err < 0)
1651                        goto err_free_dev;
1652
1653                tun_net_init(dev);
1654
1655                err = tun_flow_init(tun);
1656                if (err < 0)
1657                        goto err_free_dev;
1658
1659                dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1660                        TUN_USER_FEATURES;
1661                dev->features = dev->hw_features;
1662
1663                INIT_LIST_HEAD(&tun->disabled);
1664                err = tun_attach(tun, file);
1665                if (err < 0)
1666                        goto err_free_dev;
1667
1668                err = register_netdevice(tun->dev);
1669                if (err < 0)
1670                        goto err_free_dev;
1671
1672                if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1673                    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1674                    device_create_file(&tun->dev->dev, &dev_attr_group))
1675                        pr_err("Failed to create tun sysfs files\n");
1676        }
1677
1678        netif_carrier_on(tun->dev);
1679
1680        tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1681
1682        if (ifr->ifr_flags & IFF_NO_PI)
1683                tun->flags |= TUN_NO_PI;
1684        else
1685                tun->flags &= ~TUN_NO_PI;
1686
1687        /* This flag has no real effect.  We track the value for backwards
1688         * compatibility.
1689         */
1690        if (ifr->ifr_flags & IFF_ONE_QUEUE)
1691                tun->flags |= TUN_ONE_QUEUE;
1692        else
1693                tun->flags &= ~TUN_ONE_QUEUE;
1694
1695        if (ifr->ifr_flags & IFF_VNET_HDR)
1696                tun->flags |= TUN_VNET_HDR;
1697        else
1698                tun->flags &= ~TUN_VNET_HDR;
1699
1700        if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1701                tun->flags |= TUN_TAP_MQ;
1702        else
1703                tun->flags &= ~TUN_TAP_MQ;
1704
1705        /* Make sure persistent devices do not get stuck in
1706         * xoff state.
1707         */
1708        if (netif_running(tun->dev))
1709                netif_tx_wake_all_queues(tun->dev);
1710
1711        strcpy(ifr->ifr_name, tun->dev->name);
1712        return 0;
1713
1714 err_free_dev:
1715        free_netdev(dev);
1716        return err;
1717}
1718
1719static void tun_get_iff(struct net *net, struct tun_struct *tun,
1720                       struct ifreq *ifr)
1721{
1722        tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1723
1724        strcpy(ifr->ifr_name, tun->dev->name);
1725
1726        ifr->ifr_flags = tun_flags(tun);
1727
1728}
1729
1730/* This is like a cut-down ethtool ops, except done via tun fd so no
1731 * privs required. */
1732static int set_offload(struct tun_struct *tun, unsigned long arg)
1733{
1734        netdev_features_t features = 0;
1735
1736        if (arg & TUN_F_CSUM) {
1737                features |= NETIF_F_HW_CSUM;
1738                arg &= ~TUN_F_CSUM;
1739
1740                if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1741                        if (arg & TUN_F_TSO_ECN) {
1742                                features |= NETIF_F_TSO_ECN;
1743                                arg &= ~TUN_F_TSO_ECN;
1744                        }
1745                        if (arg & TUN_F_TSO4)
1746                                features |= NETIF_F_TSO;
1747                        if (arg & TUN_F_TSO6)
1748                                features |= NETIF_F_TSO6;
1749                        arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1750                }
1751
1752                if (arg & TUN_F_UFO) {
1753                        features |= NETIF_F_UFO;
1754                        arg &= ~TUN_F_UFO;
1755                }
1756        }
1757
1758        /* This gives the user a way to test for new features in future by
1759         * trying to set them. */
1760        if (arg)
1761                return -EINVAL;
1762
1763        tun->set_features = features;
1764        netdev_update_features(tun->dev);
1765
1766        return 0;
1767}
1768
1769static void tun_detach_filter(struct tun_struct *tun, int n)
1770{
1771        int i;
1772        struct tun_file *tfile;
1773
1774        for (i = 0; i < n; i++) {
1775                tfile = rtnl_dereference(tun->tfiles[i]);
1776                sk_detach_filter(tfile->socket.sk);
1777        }
1778
1779        tun->filter_attached = false;
1780}
1781
1782static int tun_attach_filter(struct tun_struct *tun)
1783{
1784        int i, ret = 0;
1785        struct tun_file *tfile;
1786
1787        for (i = 0; i < tun->numqueues; i++) {
1788                tfile = rtnl_dereference(tun->tfiles[i]);
1789                ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1790                if (ret) {
1791                        tun_detach_filter(tun, i);
1792                        return ret;
1793                }
1794        }
1795
1796        tun->filter_attached = true;
1797        return ret;
1798}
1799
1800static void tun_set_sndbuf(struct tun_struct *tun)
1801{
1802        struct tun_file *tfile;
1803        int i;
1804
1805        for (i = 0; i < tun->numqueues; i++) {
1806                tfile = rtnl_dereference(tun->tfiles[i]);
1807                tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1808        }
1809}
1810
1811static int tun_set_queue(struct file *file, struct ifreq *ifr)
1812{
1813        struct tun_file *tfile = file->private_data;
1814        struct tun_struct *tun;
1815        int ret = 0;
1816
1817        rtnl_lock();
1818
1819        if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1820                tun = tfile->detached;
1821                if (!tun) {
1822                        ret = -EINVAL;
1823                        goto unlock;
1824                }
1825                ret = security_tun_dev_attach_queue(tun->security);
1826                if (ret < 0)
1827                        goto unlock;
1828                ret = tun_attach(tun, file);
1829        } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1830                tun = rtnl_dereference(tfile->tun);
1831                if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1832                        ret = -EINVAL;
1833                else
1834                        __tun_detach(tfile, false);
1835        } else
1836                ret = -EINVAL;
1837
1838unlock:
1839        rtnl_unlock();
1840        return ret;
1841}
1842
1843static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1844                            unsigned long arg, int ifreq_len)
1845{
1846        struct tun_file *tfile = file->private_data;
1847        struct tun_struct *tun;
1848        void __user* argp = (void __user*)arg;
1849        struct ifreq ifr;
1850        kuid_t owner;
1851        kgid_t group;
1852        int sndbuf;
1853        int vnet_hdr_sz;
1854        int ret;
1855
1856        if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1857                if (copy_from_user(&ifr, argp, ifreq_len))
1858                        return -EFAULT;
1859        } else {
1860                memset(&ifr, 0, sizeof(ifr));
1861        }
1862        if (cmd == TUNGETFEATURES) {
1863                /* Currently this just means: "what IFF flags are valid?".
1864                 * This is needed because we never checked for invalid flags on
1865                 * TUNSETIFF. */
1866                return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1867                                IFF_VNET_HDR | IFF_MULTI_QUEUE,
1868                                (unsigned int __user*)argp);
1869        } else if (cmd == TUNSETQUEUE)
1870                return tun_set_queue(file, &ifr);
1871
1872        ret = 0;
1873        rtnl_lock();
1874
1875        tun = __tun_get(tfile);
1876        if (cmd == TUNSETIFF && !tun) {
1877                ifr.ifr_name[IFNAMSIZ-1] = '\0';
1878
1879                ret = tun_set_iff(tfile->net, file, &ifr);
1880
1881                if (ret)
1882                        goto unlock;
1883
1884                if (copy_to_user(argp, &ifr, ifreq_len))
1885                        ret = -EFAULT;
1886                goto unlock;
1887        }
1888
1889        ret = -EBADFD;
1890        if (!tun)
1891                goto unlock;
1892
1893        tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1894
1895        ret = 0;
1896        switch (cmd) {
1897        case TUNGETIFF:
1898                tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1899
1900                if (copy_to_user(argp, &ifr, ifreq_len))
1901                        ret = -EFAULT;
1902                break;
1903
1904        case TUNSETNOCSUM:
1905                /* Disable/Enable checksum */
1906
1907                /* [unimplemented] */
1908                tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1909                          arg ? "disabled" : "enabled");
1910                break;
1911
1912        case TUNSETPERSIST:
1913                /* Disable/Enable persist mode. Keep an extra reference to the
1914                 * module to prevent the module being unprobed.
1915                 */
1916                if (arg && !(tun->flags & TUN_PERSIST)) {
1917                        tun->flags |= TUN_PERSIST;
1918                        __module_get(THIS_MODULE);
1919                }
1920                if (!arg && (tun->flags & TUN_PERSIST)) {
1921                        tun->flags &= ~TUN_PERSIST;
1922                        module_put(THIS_MODULE);
1923                }
1924
1925                tun_debug(KERN_INFO, tun, "persist %s\n",
1926                          arg ? "enabled" : "disabled");
1927                break;
1928
1929        case TUNSETOWNER:
1930                /* Set owner of the device */
1931                owner = make_kuid(current_user_ns(), arg);
1932                if (!uid_valid(owner)) {
1933                        ret = -EINVAL;
1934                        break;
1935                }
1936                tun->owner = owner;
1937                tun_debug(KERN_INFO, tun, "owner set to %u\n",
1938                          from_kuid(&init_user_ns, tun->owner));
1939                break;
1940
1941        case TUNSETGROUP:
1942                /* Set group of the device */
1943                group = make_kgid(current_user_ns(), arg);
1944                if (!gid_valid(group)) {
1945                        ret = -EINVAL;
1946                        break;
1947                }
1948                tun->group = group;
1949                tun_debug(KERN_INFO, tun, "group set to %u\n",
1950                          from_kgid(&init_user_ns, tun->group));
1951                break;
1952
1953        case TUNSETLINK:
1954                /* Only allow setting the type when the interface is down */
1955                if (tun->dev->flags & IFF_UP) {
1956                        tun_debug(KERN_INFO, tun,
1957                                  "Linktype set failed because interface is up\n");
1958                        ret = -EBUSY;
1959                } else {
1960                        tun->dev->type = (int) arg;
1961                        tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1962                                  tun->dev->type);
1963                        ret = 0;
1964                }
1965                break;
1966
1967#ifdef TUN_DEBUG
1968        case TUNSETDEBUG:
1969                tun->debug = arg;
1970                break;
1971#endif
1972        case TUNSETOFFLOAD:
1973                ret = set_offload(tun, arg);
1974                break;
1975
1976        case TUNSETTXFILTER:
1977                /* Can be set only for TAPs */
1978                ret = -EINVAL;
1979                if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1980                        break;
1981                ret = update_filter(&tun->txflt, (void __user *)arg);
1982                break;
1983
1984        case SIOCGIFHWADDR:
1985                /* Get hw address */
1986                memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1987                ifr.ifr_hwaddr.sa_family = tun->dev->type;
1988                if (copy_to_user(argp, &ifr, ifreq_len))
1989                        ret = -EFAULT;
1990                break;
1991
1992        case SIOCSIFHWADDR:
1993                /* Set hw address */
1994                tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1995                          ifr.ifr_hwaddr.sa_data);
1996
1997                ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1998                break;
1999
2000        case TUNGETSNDBUF:
2001                sndbuf = tfile->socket.sk->sk_sndbuf;
2002                if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2003                        ret = -EFAULT;
2004                break;
2005
2006        case TUNSETSNDBUF:
2007                if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2008                        ret = -EFAULT;
2009                        break;
2010                }
2011
2012                tun->sndbuf = sndbuf;
2013                tun_set_sndbuf(tun);
2014                break;
2015
2016        case TUNGETVNETHDRSZ:
2017                vnet_hdr_sz = tun->vnet_hdr_sz;
2018                if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2019                        ret = -EFAULT;
2020                break;
2021
2022        case TUNSETVNETHDRSZ:
2023                if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2024                        ret = -EFAULT;
2025                        break;
2026                }
2027                if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2028                        ret = -EINVAL;
2029                        break;
2030                }
2031
2032                tun->vnet_hdr_sz = vnet_hdr_sz;
2033                break;
2034
2035        case TUNATTACHFILTER:
2036                /* Can be set only for TAPs */
2037                ret = -EINVAL;
2038                if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2039                        break;
2040                ret = -EFAULT;
2041                if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2042                        break;
2043
2044                ret = tun_attach_filter(tun);
2045                break;
2046
2047        case TUNDETACHFILTER:
2048                /* Can be set only for TAPs */
2049                ret = -EINVAL;
2050                if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2051                        break;
2052                ret = 0;
2053                tun_detach_filter(tun, tun->numqueues);
2054                break;
2055
2056        default:
2057                ret = -EINVAL;
2058                break;
2059        }
2060
2061unlock:
2062        rtnl_unlock();
2063        if (tun)
2064                tun_put(tun);
2065        return ret;
2066}
2067
2068static long tun_chr_ioctl(struct file *file,
2069                          unsigned int cmd, unsigned long arg)
2070{
2071        return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2072}
2073
2074#ifdef CONFIG_COMPAT
2075static long tun_chr_compat_ioctl(struct file *file,
2076                         unsigned int cmd, unsigned long arg)
2077{
2078        switch (cmd) {
2079        case TUNSETIFF:
2080        case TUNGETIFF:
2081        case TUNSETTXFILTER:
2082        case TUNGETSNDBUF:
2083        case TUNSETSNDBUF:
2084        case SIOCGIFHWADDR:
2085        case SIOCSIFHWADDR:
2086                arg = (unsigned long)compat_ptr(arg);
2087                break;
2088        default:
2089                arg = (compat_ulong_t)arg;
2090                break;
2091        }
2092
2093        /*
2094         * compat_ifreq is shorter than ifreq, so we must not access beyond
2095         * the end of that structure. All fields that are used in this
2096         * driver are compatible though, we don't need to convert the
2097         * contents.
2098         */
2099        return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2100}
2101#endif /* CONFIG_COMPAT */
2102
2103static int tun_chr_fasync(int fd, struct file *file, int on)
2104{
2105        struct tun_file *tfile = file->private_data;
2106        int ret;
2107
2108        if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2109                goto out;
2110
2111        if (on) {
2112                ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2113                if (ret)
2114                        goto out;
2115                tfile->flags |= TUN_FASYNC;
2116        } else
2117                tfile->flags &= ~TUN_FASYNC;
2118        ret = 0;
2119out:
2120        return ret;
2121}
2122
2123static int tun_chr_open(struct inode *inode, struct file * file)
2124{
2125        struct tun_file *tfile;
2126
2127        DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2128
2129        tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2130                                            &tun_proto);
2131        if (!tfile)
2132                return -ENOMEM;
2133        rcu_assign_pointer(tfile->tun, NULL);
2134        tfile->net = get_net(current->nsproxy->net_ns);
2135        tfile->flags = 0;
2136
2137        rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2138        init_waitqueue_head(&tfile->wq.wait);
2139
2140        tfile->socket.file = file;
2141        tfile->socket.ops = &tun_socket_ops;
2142
2143        sock_init_data(&tfile->socket, &tfile->sk);
2144        sk_change_net(&tfile->sk, tfile->net);
2145
2146        tfile->sk.sk_write_space = tun_sock_write_space;
2147        tfile->sk.sk_sndbuf = INT_MAX;
2148
2149        file->private_data = tfile;
2150        set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2151        INIT_LIST_HEAD(&tfile->next);
2152
2153        return 0;
2154}
2155
2156static int tun_chr_close(struct inode *inode, struct file *file)
2157{
2158        struct tun_file *tfile = file->private_data;
2159        struct net *net = tfile->net;
2160
2161        tun_detach(tfile, true);
2162        put_net(net);
2163
2164        return 0;
2165}
2166
2167static const struct file_operations tun_fops = {
2168        .owner  = THIS_MODULE,
2169        .llseek = no_llseek,
2170        .read  = do_sync_read,
2171        .aio_read  = tun_chr_aio_read,
2172        .write = do_sync_write,
2173        .aio_write = tun_chr_aio_write,
2174        .poll   = tun_chr_poll,
2175        .unlocked_ioctl = tun_chr_ioctl,
2176#ifdef CONFIG_COMPAT
2177        .compat_ioctl = tun_chr_compat_ioctl,
2178#endif
2179        .open   = tun_chr_open,
2180        .release = tun_chr_close,
2181        .fasync = tun_chr_fasync
2182};
2183
2184static struct miscdevice tun_miscdev = {
2185        .minor = TUN_MINOR,
2186        .name = "tun",
2187        .nodename = "net/tun",
2188        .fops = &tun_fops,
2189};
2190
2191/* ethtool interface */
2192
2193static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2194{
2195        cmd->supported          = 0;
2196        cmd->advertising        = 0;
2197        ethtool_cmd_speed_set(cmd, SPEED_10);
2198        cmd->duplex             = DUPLEX_FULL;
2199        cmd->port               = PORT_TP;
2200        cmd->phy_address        = 0;
2201        cmd->transceiver        = XCVR_INTERNAL;
2202        cmd->autoneg            = AUTONEG_DISABLE;
2203        cmd->maxtxpkt           = 0;
2204        cmd->maxrxpkt           = 0;
2205        return 0;
2206}
2207
2208static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2209{
2210        struct tun_struct *tun = netdev_priv(dev);
2211
2212        strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2213        strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2214
2215        switch (tun->flags & TUN_TYPE_MASK) {
2216        case TUN_TUN_DEV:
2217                strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2218                break;
2219        case TUN_TAP_DEV:
2220                strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2221                break;
2222        }
2223}
2224
2225static u32 tun_get_msglevel(struct net_device *dev)
2226{
2227#ifdef TUN_DEBUG
2228        struct tun_struct *tun = netdev_priv(dev);
2229        return tun->debug;
2230#else
2231        return -EOPNOTSUPP;
2232#endif
2233}
2234
2235static void tun_set_msglevel(struct net_device *dev, u32 value)
2236{
2237#ifdef TUN_DEBUG
2238        struct tun_struct *tun = netdev_priv(dev);
2239        tun->debug = value;
2240#endif
2241}
2242
2243static const struct ethtool_ops tun_ethtool_ops = {
2244        .get_settings   = tun_get_settings,
2245        .get_drvinfo    = tun_get_drvinfo,
2246        .get_msglevel   = tun_get_msglevel,
2247        .set_msglevel   = tun_set_msglevel,
2248        .get_link       = ethtool_op_get_link,
2249};
2250
2251
2252static int __init tun_init(void)
2253{
2254        int ret = 0;
2255
2256        pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2257        pr_info("%s\n", DRV_COPYRIGHT);
2258
2259        ret = rtnl_link_register(&tun_link_ops);
2260        if (ret) {
2261                pr_err("Can't register link_ops\n");
2262                goto err_linkops;
2263        }
2264
2265        ret = misc_register(&tun_miscdev);
2266        if (ret) {
2267                pr_err("Can't register misc device %d\n", TUN_MINOR);
2268                goto err_misc;
2269        }
2270        return  0;
2271err_misc:
2272        rtnl_link_unregister(&tun_link_ops);
2273err_linkops:
2274        return ret;
2275}
2276
2277static void tun_cleanup(void)
2278{
2279        misc_deregister(&tun_miscdev);
2280        rtnl_link_unregister(&tun_link_ops);
2281}
2282
2283/* Get an underlying socket object from tun file.  Returns error unless file is
2284 * attached to a device.  The returned object works like a packet socket, it
2285 * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2286 * holding a reference to the file for as long as the socket is in use. */
2287struct socket *tun_get_socket(struct file *file)
2288{
2289        struct tun_file *tfile;
2290        if (file->f_op != &tun_fops)
2291                return ERR_PTR(-EINVAL);
2292        tfile = file->private_data;
2293        if (!tfile)
2294                return ERR_PTR(-EBADFD);
2295        return &tfile->socket;
2296}
2297EXPORT_SYMBOL_GPL(tun_get_socket);
2298
2299module_init(tun_init);
2300module_exit(tun_cleanup);
2301MODULE_DESCRIPTION(DRV_DESCRIPTION);
2302MODULE_AUTHOR(DRV_COPYRIGHT);
2303MODULE_LICENSE("GPL");
2304MODULE_ALIAS_MISCDEV(TUN_MINOR);
2305MODULE_ALIAS("devname:net/tun");
2306
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.