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 = ACCESS_ONCE(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
 413        tun = rtnl_dereference(tfile->tun);
 414
 415        if (tun && !tfile->detached) {
 416                u16 index = tfile->queue_index;
 417                BUG_ON(index >= tun->numqueues);
 418
 419                rcu_assign_pointer(tun->tfiles[index],
 420                                   tun->tfiles[tun->numqueues - 1]);
 421                ntfile = rtnl_dereference(tun->tfiles[index]);
 422                ntfile->queue_index = index;
 423
 424                --tun->numqueues;
 425                if (clean) {
 426                        rcu_assign_pointer(tfile->tun, NULL);
 427                        sock_put(&tfile->sk);
 428                } else
 429                        tun_disable_queue(tun, tfile);
 430
 431                synchronize_net();
 432                tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 433                /* Drop read queue */
 434                skb_queue_purge(&tfile->sk.sk_receive_queue);
 435                tun_set_real_num_queues(tun);
 436        } else if (tfile->detached && clean) {
 437                tun = tun_enable_queue(tfile);
 438                sock_put(&tfile->sk);
 439        }
 440
 441        if (clean) {
 442                if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 443                        netif_carrier_off(tun->dev);
 444
 445                        if (!(tun->flags & TUN_PERSIST) &&
 446                            tun->dev->reg_state == NETREG_REGISTERED)
 447                                unregister_netdevice(tun->dev);
 448                }
 449
 450                BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
 451                                 &tfile->socket.flags));
 452                sk_release_kernel(&tfile->sk);
 453        }
 454}
 455
 456static void tun_detach(struct tun_file *tfile, bool clean)
 457{
 458        rtnl_lock();
 459        __tun_detach(tfile, clean);
 460        rtnl_unlock();
 461}
 462
 463static void tun_detach_all(struct net_device *dev)
 464{
 465        struct tun_struct *tun = netdev_priv(dev);
 466        struct tun_file *tfile, *tmp;
 467        int i, n = tun->numqueues;
 468
 469        for (i = 0; i < n; i++) {
 470                tfile = rtnl_dereference(tun->tfiles[i]);
 471                BUG_ON(!tfile);
 472                wake_up_all(&tfile->wq.wait);
 473                rcu_assign_pointer(tfile->tun, NULL);
 474                --tun->numqueues;
 475        }
 476        list_for_each_entry(tfile, &tun->disabled, next) {
 477                wake_up_all(&tfile->wq.wait);
 478                rcu_assign_pointer(tfile->tun, NULL);
 479        }
 480        BUG_ON(tun->numqueues != 0);
 481
 482        synchronize_net();
 483        for (i = 0; i < n; i++) {
 484                tfile = rtnl_dereference(tun->tfiles[i]);
 485                /* Drop read queue */
 486                skb_queue_purge(&tfile->sk.sk_receive_queue);
 487                sock_put(&tfile->sk);
 488        }
 489        list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 490                tun_enable_queue(tfile);
 491                skb_queue_purge(&tfile->sk.sk_receive_queue);
 492                sock_put(&tfile->sk);
 493        }
 494        BUG_ON(tun->numdisabled != 0);
 495
 496        if (tun->flags & TUN_PERSIST)
 497                module_put(THIS_MODULE);
 498}
 499
 500static int tun_attach(struct tun_struct *tun, struct file *file)
 501{
 502        struct tun_file *tfile = file->private_data;
 503        int err;
 504
 505        err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 506        if (err < 0)
 507                goto out;
 508
 509        err = -EINVAL;
 510        if (rtnl_dereference(tfile->tun) && !tfile->detached)
 511                goto out;
 512
 513        err = -EBUSY;
 514        if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
 515                goto out;
 516
 517        err = -E2BIG;
 518        if (!tfile->detached &&
 519            tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 520                goto out;
 521
 522        err = 0;
 523
 524        /* Re-attach the filter to presist device */
 525        if (tun->filter_attached == true) {
 526                err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 527                if (!err)
 528                        goto out;
 529        }
 530        tfile->queue_index = tun->numqueues;
 531        rcu_assign_pointer(tfile->tun, tun);
 532        rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 533        tun->numqueues++;
 534
 535        if (tfile->detached)
 536                tun_enable_queue(tfile);
 537        else
 538                sock_hold(&tfile->sk);
 539
 540        tun_set_real_num_queues(tun);
 541
 542        /* device is allowed to go away first, so no need to hold extra
 543         * refcnt.
 544         */
 545
 546out:
 547        return err;
 548}
 549
 550static struct tun_struct *__tun_get(struct tun_file *tfile)
 551{
 552        struct tun_struct *tun;
 553
 554        rcu_read_lock();
 555        tun = rcu_dereference(tfile->tun);
 556        if (tun)
 557                dev_hold(tun->dev);
 558        rcu_read_unlock();
 559
 560        return tun;
 561}
 562
 563static struct tun_struct *tun_get(struct file *file)
 564{
 565        return __tun_get(file->private_data);
 566}
 567
 568static void tun_put(struct tun_struct *tun)
 569{
 570        dev_put(tun->dev);
 571}
 572
 573/* TAP filtering */
 574static void addr_hash_set(u32 *mask, const u8 *addr)
 575{
 576        int n = ether_crc(ETH_ALEN, addr) >> 26;
 577        mask[n >> 5] |= (1 << (n & 31));
 578}
 579
 580static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 581{
 582        int n = ether_crc(ETH_ALEN, addr) >> 26;
 583        return mask[n >> 5] & (1 << (n & 31));
 584}
 585
 586static int update_filter(struct tap_filter *filter, void __user *arg)
 587{
 588        struct { u8 u[ETH_ALEN]; } *addr;
 589        struct tun_filter uf;
 590        int err, alen, n, nexact;
 591
 592        if (copy_from_user(&uf, arg, sizeof(uf)))
 593                return -EFAULT;
 594
 595        if (!uf.count) {
 596                /* Disabled */
 597                filter->count = 0;
 598                return 0;
 599        }
 600
 601        alen = ETH_ALEN * uf.count;
 602        addr = kmalloc(alen, GFP_KERNEL);
 603        if (!addr)
 604                return -ENOMEM;
 605
 606        if (copy_from_user(addr, arg + sizeof(uf), alen)) {
 607                err = -EFAULT;
 608                goto done;
 609        }
 610
 611        /* The filter is updated without holding any locks. Which is
 612         * perfectly safe. We disable it first and in the worst
 613         * case we'll accept a few undesired packets. */
 614        filter->count = 0;
 615        wmb();
 616
 617        /* Use first set of addresses as an exact filter */
 618        for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 619                memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 620
 621        nexact = n;
 622
 623        /* Remaining multicast addresses are hashed,
 624         * unicast will leave the filter disabled. */
 625        memset(filter->mask, 0, sizeof(filter->mask));
 626        for (; n < uf.count; n++) {
 627                if (!is_multicast_ether_addr(addr[n].u)) {
 628                        err = 0; /* no filter */
 629                        goto done;
 630                }
 631                addr_hash_set(filter->mask, addr[n].u);
 632        }
 633
 634        /* For ALLMULTI just set the mask to all ones.
 635         * This overrides the mask populated above. */
 636        if ((uf.flags & TUN_FLT_ALLMULTI))
 637                memset(filter->mask, ~0, sizeof(filter->mask));
 638
 639        /* Now enable the filter */
 640        wmb();
 641        filter->count = nexact;
 642
 643        /* Return the number of exact filters */
 644        err = nexact;
 645
 646done:
 647        kfree(addr);
 648        return err;
 649}
 650
 651/* Returns: 0 - drop, !=0 - accept */
 652static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 653{
 654        /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 655         * at this point. */
 656        struct ethhdr *eh = (struct ethhdr *) skb->data;
 657        int i;
 658
 659        /* Exact match */
 660        for (i = 0; i < filter->count; i++)
 661                if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 662                        return 1;
 663
 664        /* Inexact match (multicast only) */
 665        if (is_multicast_ether_addr(eh->h_dest))
 666                return addr_hash_test(filter->mask, eh->h_dest);
 667
 668        return 0;
 669}
 670
 671/*
 672 * Checks whether the packet is accepted or not.
 673 * Returns: 0 - drop, !=0 - accept
 674 */
 675static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 676{
 677        if (!filter->count)
 678                return 1;
 679
 680        return run_filter(filter, skb);
 681}
 682
 683/* Network device part of the driver */
 684
 685static const struct ethtool_ops tun_ethtool_ops;
 686
 687/* Net device detach from fd. */
 688static void tun_net_uninit(struct net_device *dev)
 689{
 690        tun_detach_all(dev);
 691}
 692
 693/* Net device open. */
 694static int tun_net_open(struct net_device *dev)
 695{
 696        netif_tx_start_all_queues(dev);
 697        return 0;
 698}
 699
 700/* Net device close. */
 701static int tun_net_close(struct net_device *dev)
 702{
 703        netif_tx_stop_all_queues(dev);
 704        return 0;
 705}
 706
 707/* Net device start xmit */
 708static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 709{
 710        struct tun_struct *tun = netdev_priv(dev);
 711        int txq = skb->queue_mapping;
 712        struct tun_file *tfile;
 713
 714        rcu_read_lock();
 715        tfile = rcu_dereference(tun->tfiles[txq]);
 716
 717        /* Drop packet if interface is not attached */
 718        if (txq >= tun->numqueues)
 719                goto drop;
 720
 721        tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 722
 723        BUG_ON(!tfile);
 724
 725        /* Drop if the filter does not like it.
 726         * This is a noop if the filter is disabled.
 727         * Filter can be enabled only for the TAP devices. */
 728        if (!check_filter(&tun->txflt, skb))
 729                goto drop;
 730
 731        if (tfile->socket.sk->sk_filter &&
 732            sk_filter(tfile->socket.sk, skb))
 733                goto drop;
 734
 735        /* Limit the number of packets queued by dividing txq length with the
 736         * number of queues.
 737         */
 738        if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
 739                          >= dev->tx_queue_len / tun->numqueues)
 740                goto drop;
 741
 742        /* Orphan the skb - required as we might hang on to it
 743         * for indefinite time. */
 744        if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
 745                goto drop;
 746        skb_orphan(skb);
 747
 748        nf_reset(skb);
 749
 750        /* Enqueue packet */
 751        skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
 752
 753        /* Notify and wake up reader process */
 754        if (tfile->flags & TUN_FASYNC)
 755                kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 756        wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
 757                                   POLLRDNORM | POLLRDBAND);
 758
 759        rcu_read_unlock();
 760        return NETDEV_TX_OK;
 761
 762drop:
 763        dev->stats.tx_dropped++;
 764        skb_tx_error(skb);
 765        kfree_skb(skb);
 766        rcu_read_unlock();
 767        return NETDEV_TX_OK;
 768}
 769
 770static void tun_net_mclist(struct net_device *dev)
 771{
 772        /*
 773         * This callback is supposed to deal with mc filter in
 774         * _rx_ path and has nothing to do with the _tx_ path.
 775         * In rx path we always accept everything userspace gives us.
 776         */
 777}
 778
 779#define MIN_MTU 68
 780#define MAX_MTU 65535
 781
 782static int
 783tun_net_change_mtu(struct net_device *dev, int new_mtu)
 784{
 785        if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
 786                return -EINVAL;
 787        dev->mtu = new_mtu;
 788        return 0;
 789}
 790
 791static netdev_features_t tun_net_fix_features(struct net_device *dev,
 792        netdev_features_t features)
 793{
 794        struct tun_struct *tun = netdev_priv(dev);
 795
 796        return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 797}
 798#ifdef CONFIG_NET_POLL_CONTROLLER
 799static void tun_poll_controller(struct net_device *dev)
 800{
 801        /*
 802         * Tun only receives frames when:
 803         * 1) the char device endpoint gets data from user space
 804         * 2) the tun socket gets a sendmsg call from user space
 805         * Since both of those are syncronous operations, we are guaranteed
 806         * never to have pending data when we poll for it
 807         * so theres nothing to do here but return.
 808         * We need this though so netpoll recognizes us as an interface that
 809         * supports polling, which enables bridge devices in virt setups to
 810         * still use netconsole
 811         */
 812        return;
 813}
 814#endif
 815static const struct net_device_ops tun_netdev_ops = {
 816        .ndo_uninit             = tun_net_uninit,
 817        .ndo_open               = tun_net_open,
 818        .ndo_stop               = tun_net_close,
 819        .ndo_start_xmit         = tun_net_xmit,
 820        .ndo_change_mtu         = tun_net_change_mtu,
 821        .ndo_fix_features       = tun_net_fix_features,
 822        .ndo_select_queue       = tun_select_queue,
 823#ifdef CONFIG_NET_POLL_CONTROLLER
 824        .ndo_poll_controller    = tun_poll_controller,
 825#endif
 826};
 827
 828static const struct net_device_ops tap_netdev_ops = {
 829        .ndo_uninit             = tun_net_uninit,
 830        .ndo_open               = tun_net_open,
 831        .ndo_stop               = tun_net_close,
 832        .ndo_start_xmit         = tun_net_xmit,
 833        .ndo_change_mtu         = tun_net_change_mtu,
 834        .ndo_fix_features       = tun_net_fix_features,
 835        .ndo_set_rx_mode        = tun_net_mclist,
 836        .ndo_set_mac_address    = eth_mac_addr,
 837        .ndo_validate_addr      = eth_validate_addr,
 838        .ndo_select_queue       = tun_select_queue,
 839#ifdef CONFIG_NET_POLL_CONTROLLER
 840        .ndo_poll_controller    = tun_poll_controller,
 841#endif
 842};
 843
 844static void tun_flow_init(struct tun_struct *tun)
 845{
 846        int i;
 847
 848        for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
 849                INIT_HLIST_HEAD(&tun->flows[i]);
 850
 851        tun->ageing_time = TUN_FLOW_EXPIRE;
 852        setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
 853        mod_timer(&tun->flow_gc_timer,
 854                  round_jiffies_up(jiffies + tun->ageing_time));
 855}
 856
 857static void tun_flow_uninit(struct tun_struct *tun)
 858{
 859        del_timer_sync(&tun->flow_gc_timer);
 860        tun_flow_flush(tun);
 861}
 862
 863/* Initialize net device. */
 864static void tun_net_init(struct net_device *dev)
 865{
 866        struct tun_struct *tun = netdev_priv(dev);
 867
 868        switch (tun->flags & TUN_TYPE_MASK) {
 869        case TUN_TUN_DEV:
 870                dev->netdev_ops = &tun_netdev_ops;
 871
 872                /* Point-to-Point TUN Device */
 873                dev->hard_header_len = 0;
 874                dev->addr_len = 0;
 875                dev->mtu = 1500;
 876
 877                /* Zero header length */
 878                dev->type = ARPHRD_NONE;
 879                dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
 880                dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 881                break;
 882
 883        case TUN_TAP_DEV:
 884                dev->netdev_ops = &tap_netdev_ops;
 885                /* Ethernet TAP Device */
 886                ether_setup(dev);
 887                dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 888                dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 889
 890                eth_hw_addr_random(dev);
 891
 892                dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
 893                break;
 894        }
 895}
 896
 897/* Character device part */
 898
 899/* Poll */
 900static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 901{
 902        struct tun_file *tfile = file->private_data;
 903        struct tun_struct *tun = __tun_get(tfile);
 904        struct sock *sk;
 905        unsigned int mask = 0;
 906
 907        if (!tun)
 908                return POLLERR;
 909
 910        sk = tfile->socket.sk;
 911
 912        tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
 913
 914        poll_wait(file, &tfile->wq.wait, wait);
 915
 916        if (!skb_queue_empty(&sk->sk_receive_queue))
 917                mask |= POLLIN | POLLRDNORM;
 918
 919        if (sock_writeable(sk) ||
 920            (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
 921             sock_writeable(sk)))
 922                mask |= POLLOUT | POLLWRNORM;
 923
 924        if (tun->dev->reg_state != NETREG_REGISTERED)
 925                mask = POLLERR;
 926
 927        tun_put(tun);
 928        return mask;
 929}
 930
 931/* prepad is the amount to reserve at front.  len is length after that.
 932 * linear is a hint as to how much to copy (usually headers). */
 933static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
 934                                     size_t prepad, size_t len,
 935                                     size_t linear, int noblock)
 936{
 937        struct sock *sk = tfile->socket.sk;
 938        struct sk_buff *skb;
 939        int err;
 940
 941        /* Under a page?  Don't bother with paged skb. */
 942        if (prepad + len < PAGE_SIZE || !linear)
 943                linear = len;
 944
 945        skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 946                                   &err);
 947        if (!skb)
 948                return ERR_PTR(err);
 949
 950        skb_reserve(skb, prepad);
 951        skb_put(skb, linear);
 952        skb->data_len = len - linear;
 953        skb->len += len - linear;
 954
 955        return skb;
 956}
 957
 958/* set skb frags from iovec, this can move to core network code for reuse */
 959static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 960                                  int offset, size_t count)
 961{
 962        int len = iov_length(from, count) - offset;
 963        int copy = skb_headlen(skb);
 964        int size, offset1 = 0;
 965        int i = 0;
 966
 967        /* Skip over from offset */
 968        while (count && (offset >= from->iov_len)) {
 969                offset -= from->iov_len;
 970                ++from;
 971                --count;
 972        }
 973
 974        /* copy up to skb headlen */
 975        while (count && (copy > 0)) {
 976                size = min_t(unsigned int, copy, from->iov_len - offset);
 977                if (copy_from_user(skb->data + offset1, from->iov_base + offset,
 978                                   size))
 979                        return -EFAULT;
 980                if (copy > size) {
 981                        ++from;
 982                        --count;
 983                        offset = 0;
 984                } else
 985                        offset += size;
 986                copy -= size;
 987                offset1 += size;
 988        }
 989
 990        if (len == offset1)
 991                return 0;
 992
 993        while (count--) {
 994                struct page *page[MAX_SKB_FRAGS];
 995                int num_pages;
 996                unsigned long base;
 997                unsigned long truesize;
 998
 999                len = from->iov_len - offset;
1000                if (!len) {
1001                        offset = 0;
1002                        ++from;
1003                        continue;
1004                }
1005                base = (unsigned long)from->iov_base + offset;
1006                size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
1007                if (i + size > MAX_SKB_FRAGS)
1008                        return -EMSGSIZE;
1009                num_pages = get_user_pages_fast(base, size, 0, &page[i]);
1010                if (num_pages != size) {
1011                        int j;
1012
1013                        for (j = 0; j < num_pages; j++)
1014                                put_page(page[i + j]);
1015                        return -EFAULT;
1016                }
1017                truesize = size * PAGE_SIZE;
1018                skb->data_len += len;
1019                skb->len += len;
1020                skb->truesize += truesize;
1021                atomic_add(truesize, &skb->sk->sk_wmem_alloc);
1022                while (len) {
1023                        int off = base & ~PAGE_MASK;
1024                        int size = min_t(int, len, PAGE_SIZE - off);
1025                        __skb_fill_page_desc(skb, i, page[i], off, size);
1026                        skb_shinfo(skb)->nr_frags++;
1027                        /* increase sk_wmem_alloc */
1028                        base += size;
1029                        len -= size;
1030                        i++;
1031                }
1032                offset = 0;
1033                ++from;
1034        }
1035        return 0;
1036}
1037
1038static unsigned long iov_pages(const struct iovec *iv, int offset,
1039                               unsigned long nr_segs)
1040{
1041        unsigned long seg, base;
1042        int pages = 0, len, size;
1043
1044        while (nr_segs && (offset >= iv->iov_len)) {
1045                offset -= iv->iov_len;
1046                ++iv;
1047                --nr_segs;
1048        }
1049
1050        for (seg = 0; seg < nr_segs; seg++) {
1051                base = (unsigned long)iv[seg].iov_base + offset;
1052                len = iv[seg].iov_len - offset;
1053                size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
1054                pages += size;
1055                offset = 0;
1056        }
1057
1058        return pages;
1059}
1060
1061/* Get packet from user space buffer */
1062static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1063                            void *msg_control, const struct iovec *iv,
1064                            size_t total_len, size_t count, int noblock)
1065{
1066        struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1067        struct sk_buff *skb;
1068        size_t len = total_len, align = NET_SKB_PAD, linear;
1069        struct virtio_net_hdr gso = { 0 };
1070        int offset = 0;
1071        int copylen;
1072        bool zerocopy = false;
1073        int err;
1074        u32 rxhash;
1075
1076        if (!(tun->flags & TUN_NO_PI)) {
1077                if (len < sizeof(pi))
1078                        return -EINVAL;
1079                len -= sizeof(pi);
1080
1081                if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1082                        return -EFAULT;
1083                offset += sizeof(pi);
1084        }
1085
1086        if (tun->flags & TUN_VNET_HDR) {
1087                if (len < tun->vnet_hdr_sz)
1088                        return -EINVAL;
1089                len -= tun->vnet_hdr_sz;
1090
1091                if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1092                        return -EFAULT;
1093
1094                if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1095                    gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1096                        gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1097
1098                if (gso.hdr_len > len)
1099                        return -EINVAL;
1100                offset += tun->vnet_hdr_sz;
1101        }
1102
1103        if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1104                align += NET_IP_ALIGN;
1105                if (unlikely(len < ETH_HLEN ||
1106                             (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1107                        return -EINVAL;
1108        }
1109
1110        if (msg_control) {
1111                /* There are 256 bytes to be copied in skb, so there is
1112                 * enough room for skb expand head in case it is used.
1113                 * The rest of the buffer is mapped from userspace.
1114                 */
1115                copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1116                linear = copylen;
1117                if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1118                        zerocopy = true;
1119        }
1120
1121        if (!zerocopy) {
1122                copylen = len;
1123                linear = gso.hdr_len;
1124        }
1125
1126        skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1127        if (IS_ERR(skb)) {
1128                if (PTR_ERR(skb) != -EAGAIN)
1129                        tun->dev->stats.rx_dropped++;
1130                return PTR_ERR(skb);
1131        }
1132
1133        if (zerocopy)
1134                err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1135        else {
1136                err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1137                if (!err && msg_control) {
1138                        struct ubuf_info *uarg = msg_control;
1139                        uarg->callback(uarg, false);
1140                }
1141        }
1142
1143        if (err) {
1144                tun->dev->stats.rx_dropped++;
1145                kfree_skb(skb);
1146                return -EFAULT;
1147        }
1148
1149        if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1150                if (!skb_partial_csum_set(skb, gso.csum_start,
1151                                          gso.csum_offset)) {
1152                        tun->dev->stats.rx_frame_errors++;
1153                        kfree_skb(skb);
1154                        return -EINVAL;
1155                }
1156        }
1157
1158        switch (tun->flags & TUN_TYPE_MASK) {
1159        case TUN_TUN_DEV:
1160                if (tun->flags & TUN_NO_PI) {
1161                        switch (skb->data[0] & 0xf0) {
1162                        case 0x40:
1163                                pi.proto = htons(ETH_P_IP);
1164                                break;
1165                        case 0x60:
1166                                pi.proto = htons(ETH_P_IPV6);
1167                                break;
1168                        default:
1169                                tun->dev->stats.rx_dropped++;
1170                                kfree_skb(skb);
1171                                return -EINVAL;
1172                        }
1173                }
1174
1175                skb_reset_mac_header(skb);
1176                skb->protocol = pi.proto;
1177                skb->dev = tun->dev;
1178                break;
1179        case TUN_TAP_DEV:
1180                skb->protocol = eth_type_trans(skb, tun->dev);
1181                break;
1182        }
1183
1184        if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1185                pr_debug("GSO!\n");
1186                switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1187                case VIRTIO_NET_HDR_GSO_TCPV4:
1188                        skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1189                        break;
1190                case VIRTIO_NET_HDR_GSO_TCPV6:
1191                        skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1192                        break;
1193                case VIRTIO_NET_HDR_GSO_UDP:
1194                        skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1195                        break;
1196                default:
1197                        tun->dev->stats.rx_frame_errors++;
1198                        kfree_skb(skb);
1199                        return -EINVAL;
1200                }
1201
1202                if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1203                        skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1204
1205                skb_shinfo(skb)->gso_size = gso.gso_size;
1206                if (skb_shinfo(skb)->gso_size == 0) {
1207                        tun->dev->stats.rx_frame_errors++;
1208                        kfree_skb(skb);
1209                        return -EINVAL;
1210                }
1211
1212                /* Header must be checked, and gso_segs computed. */
1213                skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1214                skb_shinfo(skb)->gso_segs = 0;
1215        }
1216
1217        /* copy skb_ubuf_info for callback when skb has no error */
1218        if (zerocopy) {
1219                skb_shinfo(skb)->destructor_arg = msg_control;
1220                skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1221                skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1222        }
1223
1224        skb_reset_network_header(skb);
1225        skb_probe_transport_header(skb, 0);
1226
1227        rxhash = skb_get_rxhash(skb);
1228        netif_rx_ni(skb);
1229
1230        tun->dev->stats.rx_packets++;
1231        tun->dev->stats.rx_bytes += len;
1232
1233        tun_flow_update(tun, rxhash, tfile);
1234        return total_len;
1235}
1236
1237static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1238                              unsigned long count, loff_t pos)
1239{
1240        struct file *file = iocb->ki_filp;
1241        struct tun_struct *tun = tun_get(file);
1242        struct tun_file *tfile = file->private_data;
1243        ssize_t result;
1244
1245        if (!tun)
1246                return -EBADFD;
1247
1248        tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1249
1250        result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1251                              count, file->f_flags & O_NONBLOCK);
1252
1253        tun_put(tun);
1254        return result;
1255}
1256
1257/* Put packet to the user space buffer */
1258static ssize_t tun_put_user(struct tun_struct *tun,
1259                            struct tun_file *tfile,
1260                            struct sk_buff *skb,
1261                            const struct iovec *iv, int len)
1262{
1263        struct tun_pi pi = { 0, skb->protocol };
1264        ssize_t total = 0;
1265
1266        if (!(tun->flags & TUN_NO_PI)) {
1267                if ((len -= sizeof(pi)) < 0)
1268                        return -EINVAL;
1269
1270                if (len < skb->len) {
1271                        /* Packet will be striped */
1272                        pi.flags |= TUN_PKT_STRIP;
1273                }
1274
1275                if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1276                        return -EFAULT;
1277                total += sizeof(pi);
1278        }
1279
1280        if (tun->flags & TUN_VNET_HDR) {
1281                struct virtio_net_hdr gso = { 0 }; /* no info leak */
1282                if ((len -= tun->vnet_hdr_sz) < 0)
1283                        return -EINVAL;
1284
1285                if (skb_is_gso(skb)) {
1286                        struct skb_shared_info *sinfo = skb_shinfo(skb);
1287
1288                        /* This is a hint as to how much should be linear. */
1289                        gso.hdr_len = skb_headlen(skb);
1290                        gso.gso_size = sinfo->gso_size;
1291                        if (sinfo->gso_type & SKB_GSO_TCPV4)
1292                                gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1293                        else if (sinfo->gso_type & SKB_GSO_TCPV6)
1294                                gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1295                        else if (sinfo->gso_type & SKB_GSO_UDP)
1296                                gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1297                        else {
1298                                pr_err("unexpected GSO type: "
1299                                       "0x%x, gso_size %d, hdr_len %d\n",
1300                                       sinfo->gso_type, gso.gso_size,
1301                                       gso.hdr_len);
1302                                print_hex_dump(KERN_ERR, "tun: ",
1303                                               DUMP_PREFIX_NONE,
1304                                               16, 1, skb->head,
1305                                               min((int)gso.hdr_len, 64), true);
1306                                WARN_ON_ONCE(1);
1307                                return -EINVAL;
1308                        }
1309                        if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1310                                gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1311                } else
1312                        gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1313
1314                if (skb->ip_summed == CHECKSUM_PARTIAL) {
1315                        gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1316                        gso.csum_start = skb_checksum_start_offset(skb);
1317                        gso.csum_offset = skb->csum_offset;
1318                } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1319                        gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1320                } /* else everything is zero */
1321
1322                if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1323                                               sizeof(gso))))
1324                        return -EFAULT;
1325                total += tun->vnet_hdr_sz;
1326        }
1327
1328        len = min_t(int, skb->len, len);
1329
1330        skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1331        total += skb->len;
1332
1333        tun->dev->stats.tx_packets++;
1334        tun->dev->stats.tx_bytes += len;
1335
1336        return total;
1337}
1338
1339static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1340                           struct kiocb *iocb, const struct iovec *iv,
1341                           ssize_t len, int noblock)
1342{
1343        DECLARE_WAITQUEUE(wait, current);
1344        struct sk_buff *skb;
1345        ssize_t ret = 0;
1346
1347        tun_debug(KERN_INFO, tun, "tun_do_read\n");
1348
1349        if (unlikely(!noblock))
1350                add_wait_queue(&tfile->wq.wait, &wait);
1351        while (len) {
1352                current->state = TASK_INTERRUPTIBLE;
1353
1354                /* Read frames from the queue */
1355                if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1356                        if (noblock) {
1357                                ret = -EAGAIN;
1358                                break;
1359                        }
1360                        if (signal_pending(current)) {
1361                                ret = -ERESTARTSYS;
1362                                break;
1363                        }
1364                        if (tun->dev->reg_state != NETREG_REGISTERED) {
1365                                ret = -EIO;
1366                                break;
1367                        }
1368
1369                        /* Nothing to read, let's sleep */
1370                        schedule();
1371                        continue;
1372                }
1373
1374                ret = tun_put_user(tun, tfile, skb, iv, len);
1375                kfree_skb(skb);
1376                break;
1377        }
1378
1379        current->state = TASK_RUNNING;
1380        if (unlikely(!noblock))
1381                remove_wait_queue(&tfile->wq.wait, &wait);
1382
1383        return ret;
1384}
1385
1386static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1387                            unsigned long count, loff_t pos)
1388{
1389        struct file *file = iocb->ki_filp;
1390        struct tun_file *tfile = file->private_data;
1391        struct tun_struct *tun = __tun_get(tfile);
1392        ssize_t len, ret;
1393
1394        if (!tun)
1395                return -EBADFD;
1396        len = iov_length(iv, count);
1397        if (len < 0) {
1398                ret = -EINVAL;
1399                goto out;
1400        }
1401
1402        ret = tun_do_read(tun, tfile, iocb, iv, len,
1403                          file->f_flags & O_NONBLOCK);
1404        ret = min_t(ssize_t, ret, len);
1405out:
1406        tun_put(tun);
1407        return ret;
1408}
1409
1410static void tun_free_netdev(struct net_device *dev)
1411{
1412        struct tun_struct *tun = netdev_priv(dev);
1413
1414        BUG_ON(!(list_empty(&tun->disabled)));
1415        tun_flow_uninit(tun);
1416        security_tun_dev_free_security(tun->security);
1417        free_netdev(dev);
1418}
1419
1420static void tun_setup(struct net_device *dev)
1421{
1422        struct tun_struct *tun = netdev_priv(dev);
1423
1424        tun->owner = INVALID_UID;
1425        tun->group = INVALID_GID;
1426
1427        dev->ethtool_ops = &tun_ethtool_ops;
1428        dev->destructor = tun_free_netdev;
1429}
1430
1431/* Trivial set of netlink ops to allow deleting tun or tap
1432 * device with netlink.
1433 */
1434static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1435{
1436        return -EINVAL;
1437}
1438
1439static struct rtnl_link_ops tun_link_ops __read_mostly = {
1440        .kind           = DRV_NAME,
1441        .priv_size      = sizeof(struct tun_struct),
1442        .setup          = tun_setup,
1443        .validate       = tun_validate,
1444};
1445
1446static void tun_sock_write_space(struct sock *sk)
1447{
1448        struct tun_file *tfile;
1449        wait_queue_head_t *wqueue;
1450
1451        if (!sock_writeable(sk))
1452                return;
1453
1454        if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1455                return;
1456
1457        wqueue = sk_sleep(sk);
1458        if (wqueue && waitqueue_active(wqueue))
1459                wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1460                                                POLLWRNORM | POLLWRBAND);
1461
1462        tfile = container_of(sk, struct tun_file, sk);
1463        kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1464}
1465
1466static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1467                       struct msghdr *m, size_t total_len)
1468{
1469        int ret;
1470        struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1471        struct tun_struct *tun = __tun_get(tfile);
1472
1473        if (!tun)
1474                return -EBADFD;
1475        ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1476                           m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1477        tun_put(tun);
1478        return ret;
1479}
1480
1481
1482static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1483                       struct msghdr *m, size_t total_len,
1484                       int flags)
1485{
1486        struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1487        struct tun_struct *tun = __tun_get(tfile);
1488        int ret;
1489
1490        if (!tun)
1491                return -EBADFD;
1492
1493        if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) {
1494                ret = -EINVAL;
1495                goto out;
1496        }
1497        ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1498                          flags & MSG_DONTWAIT);
1499        if (ret > total_len) {
1500                m->msg_flags |= MSG_TRUNC;
1501                ret = flags & MSG_TRUNC ? ret : total_len;
1502        }
1503out:
1504        tun_put(tun);
1505        return ret;
1506}
1507
1508static int tun_release(struct socket *sock)
1509{
1510        if (sock->sk)
1511                sock_put(sock->sk);
1512        return 0;
1513}
1514
1515/* Ops structure to mimic raw sockets with tun */
1516static const struct proto_ops tun_socket_ops = {
1517        .sendmsg = tun_sendmsg,
1518        .recvmsg = tun_recvmsg,
1519        .release = tun_release,
1520};
1521
1522static struct proto tun_proto = {
1523        .name           = "tun",
1524        .owner          = THIS_MODULE,
1525        .obj_size       = sizeof(struct tun_file),
1526};
1527
1528static int tun_flags(struct tun_struct *tun)
1529{
1530        int flags = 0;
1531
1532        if (tun->flags & TUN_TUN_DEV)
1533                flags |= IFF_TUN;
1534        else
1535                flags |= IFF_TAP;
1536
1537        if (tun->flags & TUN_NO_PI)
1538                flags |= IFF_NO_PI;
1539
1540        /* This flag has no real effect.  We track the value for backwards
1541         * compatibility.
1542         */
1543        if (tun->flags & TUN_ONE_QUEUE)
1544                flags |= IFF_ONE_QUEUE;
1545
1546        if (tun->flags & TUN_VNET_HDR)
1547                flags |= IFF_VNET_HDR;
1548
1549        if (tun->flags & TUN_TAP_MQ)
1550                flags |= IFF_MULTI_QUEUE;
1551
1552        if (tun->flags & TUN_PERSIST)
1553                flags |= IFF_PERSIST;
1554
1555        return flags;
1556}
1557
1558static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1559                              char *buf)
1560{
1561        struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1562        return sprintf(buf, "0x%x\n", tun_flags(tun));
1563}
1564
1565static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1566                              char *buf)
1567{
1568        struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1569        return uid_valid(tun->owner)?
1570                sprintf(buf, "%u\n",
1571                        from_kuid_munged(current_user_ns(), tun->owner)):
1572                sprintf(buf, "-1\n");
1573}
1574
1575static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1576                              char *buf)
1577{
1578        struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1579        return gid_valid(tun->group) ?
1580                sprintf(buf, "%u\n",
1581                        from_kgid_munged(current_user_ns(), tun->group)):
1582                sprintf(buf, "-1\n");
1583}
1584
1585static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1586static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1587static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1588
1589static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1590{
1591        struct tun_struct *tun;
1592        struct tun_file *tfile = file->private_data;
1593        struct net_device *dev;
1594        int err;
1595
1596        if (tfile->detached)
1597                return -EINVAL;
1598
1599        dev = __dev_get_by_name(net, ifr->ifr_name);
1600        if (dev) {
1601                if (ifr->ifr_flags & IFF_TUN_EXCL)
1602                        return -EBUSY;
1603                if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1604                        tun = netdev_priv(dev);
1605                else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1606                        tun = netdev_priv(dev);
1607                else
1608                        return -EINVAL;
1609
1610                if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1611                    !!(tun->flags & TUN_TAP_MQ))
1612                        return -EINVAL;
1613
1614                if (tun_not_capable(tun))
1615                        return -EPERM;
1616                err = security_tun_dev_open(tun->security);
1617                if (err < 0)
1618                        return err;
1619
1620                err = tun_attach(tun, file);
1621                if (err < 0)
1622                        return err;
1623
1624                if (tun->flags & TUN_TAP_MQ &&
1625                    (tun->numqueues + tun->numdisabled > 1)) {
1626                        /* One or more queue has already been attached, no need
1627                         * to initialize the device again.
1628                         */
1629                        return 0;
1630                }
1631        }
1632        else {
1633                char *name;
1634                unsigned long flags = 0;
1635                int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1636                             MAX_TAP_QUEUES : 1;
1637
1638                if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1639                        return -EPERM;
1640                err = security_tun_dev_create();
1641                if (err < 0)
1642                        return err;
1643
1644                /* Set dev type */
1645                if (ifr->ifr_flags & IFF_TUN) {
1646                        /* TUN device */
1647                        flags |= TUN_TUN_DEV;
1648                        name = "tun%d";
1649                } else if (ifr->ifr_flags & IFF_TAP) {
1650                        /* TAP device */
1651                        flags |= TUN_TAP_DEV;
1652                        name = "tap%d";
1653                } else
1654                        return -EINVAL;
1655
1656                if (*ifr->ifr_name)
1657                        name = ifr->ifr_name;
1658
1659                dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1660                                       tun_setup, queues, queues);
1661
1662                if (!dev)
1663                        return -ENOMEM;
1664
1665                dev_net_set(dev, net);
1666                dev->rtnl_link_ops = &tun_link_ops;
1667
1668                tun = netdev_priv(dev);
1669                tun->dev = dev;
1670                tun->flags = flags;
1671                tun->txflt.count = 0;
1672                tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1673
1674                tun->filter_attached = false;
1675                tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1676
1677                spin_lock_init(&tun->lock);
1678
1679                err = security_tun_dev_alloc_security(&tun->security);
1680                if (err < 0)
1681                        goto err_free_dev;
1682
1683                tun_net_init(dev);
1684                tun_flow_init(tun);
1685
1686                dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1687                        TUN_USER_FEATURES;
1688                dev->features = dev->hw_features;
1689                dev->vlan_features = dev->features;
1690
1691                INIT_LIST_HEAD(&tun->disabled);
1692                err = tun_attach(tun, file);
1693                if (err < 0)
1694                        goto err_free_dev;
1695
1696                err = register_netdevice(tun->dev);
1697                if (err < 0)
1698                        goto err_free_dev;
1699
1700                if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1701                    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1702                    device_create_file(&tun->dev->dev, &dev_attr_group))
1703                        pr_err("Failed to create tun sysfs files\n");
1704        }
1705
1706        netif_carrier_on(tun->dev);
1707
1708        tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1709
1710        if (ifr->ifr_flags & IFF_NO_PI)
1711                tun->flags |= TUN_NO_PI;
1712        else
1713                tun->flags &= ~TUN_NO_PI;
1714
1715        /* This flag has no real effect.  We track the value for backwards
1716         * compatibility.
1717         */
1718        if (ifr->ifr_flags & IFF_ONE_QUEUE)
1719                tun->flags |= TUN_ONE_QUEUE;
1720        else
1721                tun->flags &= ~TUN_ONE_QUEUE;
1722
1723        if (ifr->ifr_flags & IFF_VNET_HDR)
1724                tun->flags |= TUN_VNET_HDR;
1725        else
1726                tun->flags &= ~TUN_VNET_HDR;
1727
1728        if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1729                tun->flags |= TUN_TAP_MQ;
1730        else
1731                tun->flags &= ~TUN_TAP_MQ;
1732
1733        /* Make sure persistent devices do not get stuck in
1734         * xoff state.
1735         */
1736        if (netif_running(tun->dev))
1737                netif_tx_wake_all_queues(tun->dev);
1738
1739        strcpy(ifr->ifr_name, tun->dev->name);
1740        return 0;
1741
1742 err_free_dev:
1743        free_netdev(dev);
1744        return err;
1745}
1746
1747static void tun_get_iff(struct net *net, struct tun_struct *tun,
1748                       struct ifreq *ifr)
1749{
1750        tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1751
1752        strcpy(ifr->ifr_name, tun->dev->name);
1753
1754        ifr->ifr_flags = tun_flags(tun);
1755
1756}
1757
1758/* This is like a cut-down ethtool ops, except done via tun fd so no
1759 * privs required. */
1760static int set_offload(struct tun_struct *tun, unsigned long arg)
1761{
1762        netdev_features_t features = 0;
1763
1764        if (arg & TUN_F_CSUM) {
1765                features |= NETIF_F_HW_CSUM;
1766                arg &= ~TUN_F_CSUM;
1767
1768                if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1769                        if (arg & TUN_F_TSO_ECN) {
1770                                features |= NETIF_F_TSO_ECN;
1771                                arg &= ~TUN_F_TSO_ECN;
1772                        }
1773                        if (arg & TUN_F_TSO4)
1774                                features |= NETIF_F_TSO;
1775                        if (arg & TUN_F_TSO6)
1776                                features |= NETIF_F_TSO6;
1777                        arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1778                }
1779
1780                if (arg & TUN_F_UFO) {
1781                        features |= NETIF_F_UFO;
1782                        arg &= ~TUN_F_UFO;
1783                }
1784        }
1785
1786        /* This gives the user a way to test for new features in future by
1787         * trying to set them. */
1788        if (arg)
1789                return -EINVAL;
1790
1791        tun->set_features = features;
1792        netdev_update_features(tun->dev);
1793
1794        return 0;
1795}
1796
1797static void tun_detach_filter(struct tun_struct *tun, int n)
1798{
1799        int i;
1800        struct tun_file *tfile;
1801
1802        for (i = 0; i < n; i++) {
1803                tfile = rtnl_dereference(tun->tfiles[i]);
1804                sk_detach_filter(tfile->socket.sk);
1805        }
1806
1807        tun->filter_attached = false;
1808}
1809
1810static int tun_attach_filter(struct tun_struct *tun)
1811{
1812        int i, ret = 0;
1813        struct tun_file *tfile;
1814
1815        for (i = 0; i < tun->numqueues; i++) {
1816                tfile = rtnl_dereference(tun->tfiles[i]);
1817                ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1818                if (ret) {
1819                        tun_detach_filter(tun, i);
1820                        return ret;
1821                }
1822        }
1823
1824        tun->filter_attached = true;
1825        return ret;
1826}
1827
1828static void tun_set_sndbuf(struct tun_struct *tun)
1829{
1830        struct tun_file *tfile;
1831        int i;
1832
1833        for (i = 0; i < tun->numqueues; i++) {
1834                tfile = rtnl_dereference(tun->tfiles[i]);
1835                tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1836        }
1837}
1838
1839static int tun_set_queue(struct file *file, struct ifreq *ifr)
1840{
1841        struct tun_file *tfile = file->private_data;
1842        struct tun_struct *tun;
1843        int ret = 0;
1844
1845        rtnl_lock();
1846
1847        if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1848                tun = tfile->detached;
1849                if (!tun) {
1850                        ret = -EINVAL;
1851                        goto unlock;
1852                }
1853                ret = security_tun_dev_attach_queue(tun->security);
1854                if (ret < 0)
1855                        goto unlock;
1856                ret = tun_attach(tun, file);
1857        } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1858                tun = rtnl_dereference(tfile->tun);
1859                if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1860                        ret = -EINVAL;
1861                else
1862                        __tun_detach(tfile, false);
1863        } else
1864                ret = -EINVAL;
1865
1866unlock:
1867        rtnl_unlock();
1868        return ret;
1869}
1870
1871static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1872                            unsigned long arg, int ifreq_len)
1873{
1874        struct tun_file *tfile = file->private_data;
1875        struct tun_struct *tun;
1876        void __user* argp = (void __user*)arg;
1877        struct ifreq ifr;
1878        kuid_t owner;
1879        kgid_t group;
1880        int sndbuf;
1881        int vnet_hdr_sz;
1882        int ret;
1883
1884        if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1885                if (copy_from_user(&ifr, argp, ifreq_len))
1886                        return -EFAULT;
1887        } else {
1888                memset(&ifr, 0, sizeof(ifr));
1889        }
1890        if (cmd == TUNGETFEATURES) {
1891                /* Currently this just means: "what IFF flags are valid?".
1892                 * This is needed because we never checked for invalid flags on
1893                 * TUNSETIFF. */
1894                return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1895                                IFF_VNET_HDR | IFF_MULTI_QUEUE,
1896                                (unsigned int __user*)argp);
1897        } else if (cmd == TUNSETQUEUE)
1898                return tun_set_queue(file, &ifr);
1899
1900        ret = 0;
1901        rtnl_lock();
1902
1903        tun = __tun_get(tfile);
1904        if (cmd == TUNSETIFF && !tun) {
1905                ifr.ifr_name[IFNAMSIZ-1] = '\0';
1906
1907                ret = tun_set_iff(tfile->net, file, &ifr);
1908
1909                if (ret)
1910                        goto unlock;
1911
1912                if (copy_to_user(argp, &ifr, ifreq_len))
1913                        ret = -EFAULT;
1914                goto unlock;
1915        }
1916
1917        ret = -EBADFD;
1918        if (!tun)
1919                goto unlock;
1920
1921        tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1922
1923        ret = 0;
1924        switch (cmd) {
1925        case TUNGETIFF:
1926                tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1927
1928                if (copy_to_user(argp, &ifr, ifreq_len))
1929                        ret = -EFAULT;
1930                break;
1931
1932        case TUNSETNOCSUM:
1933                /* Disable/Enable checksum */
1934
1935                /* [unimplemented] */
1936                tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1937                          arg ? "disabled" : "enabled");
1938                break;
1939
1940        case TUNSETPERSIST:
1941                /* Disable/Enable persist mode. Keep an extra reference to the
1942                 * module to prevent the module being unprobed.
1943                 */
1944                if (arg && !(tun->flags & TUN_PERSIST)) {
1945                        tun->flags |= TUN_PERSIST;
1946                        __module_get(THIS_MODULE);
1947                }
1948                if (!arg && (tun->flags & TUN_PERSIST)) {
1949                        tun->flags &= ~TUN_PERSIST;
1950                        module_put(THIS_MODULE);
1951                }
1952
1953                tun_debug(KERN_INFO, tun, "persist %s\n",
1954                          arg ? "enabled" : "disabled");
1955                break;
1956
1957        case TUNSETOWNER:
1958                /* Set owner of the device */
1959                owner = make_kuid(current_user_ns(), arg);
1960                if (!uid_valid(owner)) {
1961                        ret = -EINVAL;
1962                        break;
1963                }
1964                tun->owner = owner;
1965                tun_debug(KERN_INFO, tun, "owner set to %u\n",
1966                          from_kuid(&init_user_ns, tun->owner));
1967                break;
1968
1969        case TUNSETGROUP:
1970                /* Set group of the device */
1971                group = make_kgid(current_user_ns(), arg);
1972                if (!gid_valid(group)) {
1973                        ret = -EINVAL;
1974                        break;
1975                }
1976                tun->group = group;
1977                tun_debug(KERN_INFO, tun, "group set to %u\n",
1978                          from_kgid(&init_user_ns, tun->group));
1979                break;
1980
1981        case TUNSETLINK:
1982                /* Only allow setting the type when the interface is down */
1983                if (tun->dev->flags & IFF_UP) {
1984                        tun_debug(KERN_INFO, tun,
1985                                  "Linktype set failed because interface is up\n");
1986                        ret = -EBUSY;
1987                } else {
1988                        tun->dev->type = (int) arg;
1989                        tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1990                                  tun->dev->type);
1991                        ret = 0;
1992                }
1993                break;
1994
1995#ifdef TUN_DEBUG
1996        case TUNSETDEBUG:
1997                tun->debug = arg;
1998                break;
1999#endif
2000        case TUNSETOFFLOAD:
2001                ret = set_offload(tun, arg);
2002                break;
2003
2004        case TUNSETTXFILTER:
2005                /* Can be set only for TAPs */
2006                ret = -EINVAL;
2007                if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2008                        break;
2009                ret = update_filter(&tun->txflt, (void __user *)arg);
2010                break;
2011
2012        case SIOCGIFHWADDR:
2013                /* Get hw address */
2014                memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2015                ifr.ifr_hwaddr.sa_family = tun->dev->type;
2016                if (copy_to_user(argp, &ifr, ifreq_len))
2017                        ret = -EFAULT;
2018                break;
2019
2020        case SIOCSIFHWADDR:
2021                /* Set hw address */
2022                tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2023                          ifr.ifr_hwaddr.sa_data);
2024
2025                ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2026                break;
2027
2028        case TUNGETSNDBUF:
2029                sndbuf = tfile->socket.sk->sk_sndbuf;
2030                if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2031                        ret = -EFAULT;
2032                break;
2033
2034        case TUNSETSNDBUF:
2035                if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2036                        ret = -EFAULT;
2037                        break;
2038                }
2039
2040                tun->sndbuf = sndbuf;
2041                tun_set_sndbuf(tun);
2042                break;
2043
2044        case TUNGETVNETHDRSZ:
2045                vnet_hdr_sz = tun->vnet_hdr_sz;
2046                if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2047                        ret = -EFAULT;
2048                break;
2049
2050        case TUNSETVNETHDRSZ:
2051                if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2052                        ret = -EFAULT;
2053                        break;
2054                }
2055                if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2056                        ret = -EINVAL;
2057                        break;
2058                }
2059
2060                tun->vnet_hdr_sz = vnet_hdr_sz;
2061                break;
2062
2063        case TUNATTACHFILTER:
2064                /* Can be set only for TAPs */
2065                ret = -EINVAL;
2066                if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2067                        break;
2068                ret = -EFAULT;
2069                if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2070                        break;
2071
2072                ret = tun_attach_filter(tun);
2073                break;
2074
2075        case TUNDETACHFILTER:
2076                /* Can be set only for TAPs */
2077                ret = -EINVAL;
2078                if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2079                        break;
2080                ret = 0;
2081                tun_detach_filter(tun, tun->numqueues);
2082                break;
2083
2084        default:
2085                ret = -EINVAL;
2086                break;
2087        }
2088
2089unlock:
2090        rtnl_unlock();
2091        if (tun)
2092                tun_put(tun);
2093        return ret;
2094}
2095
2096static long tun_chr_ioctl(struct file *file,
2097                          unsigned int cmd, unsigned long arg)
2098{
2099        return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2100}
2101
2102#ifdef CONFIG_COMPAT
2103static long tun_chr_compat_ioctl(struct file *file,
2104                         unsigned int cmd, unsigned long arg)
2105{
2106        switch (cmd) {
2107        case TUNSETIFF:
2108        case TUNGETIFF:
2109        case TUNSETTXFILTER:
2110        case TUNGETSNDBUF:
2111        case TUNSETSNDBUF:
2112        case SIOCGIFHWADDR:
2113        case SIOCSIFHWADDR:
2114                arg = (unsigned long)compat_ptr(arg);
2115                break;
2116        default:
2117                arg = (compat_ulong_t)arg;
2118                break;
2119        }
2120
2121        /*
2122         * compat_ifreq is shorter than ifreq, so we must not access beyond
2123         * the end of that structure. All fields that are used in this
2124         * driver are compatible though, we don't need to convert the
2125         * contents.
2126         */
2127        return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2128}
2129#endif /* CONFIG_COMPAT */
2130
2131static int tun_chr_fasync(int fd, struct file *file, int on)
2132{
2133        struct tun_file *tfile = file->private_data;
2134        int ret;
2135
2136        if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2137                goto out;
2138
2139        if (on) {
2140                ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2141                if (ret)
2142                        goto out;
2143                tfile->flags |= TUN_FASYNC;
2144        } else
2145                tfile->flags &= ~TUN_FASYNC;
2146        ret = 0;
2147out:
2148        return ret;
2149}
2150
2151static int tun_chr_open(struct inode *inode, struct file * file)
2152{
2153        struct tun_file *tfile;
2154
2155        DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2156
2157        tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2158                                            &tun_proto);
2159        if (!tfile)
2160                return -ENOMEM;
2161        rcu_assign_pointer(tfile->tun, NULL);
2162        tfile->net = get_net(current->nsproxy->net_ns);
2163        tfile->flags = 0;
2164
2165        rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2166        init_waitqueue_head(&tfile->wq.wait);
2167
2168        tfile->socket.file = file;
2169        tfile->socket.ops = &tun_socket_ops;
2170
2171        sock_init_data(&tfile->socket, &tfile->sk);
2172        sk_change_net(&tfile->sk, tfile->net);
2173
2174        tfile->sk.sk_write_space = tun_sock_write_space;
2175        tfile->sk.sk_sndbuf = INT_MAX;
2176
2177        file->private_data = tfile;
2178        set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2179        INIT_LIST_HEAD(&tfile->next);
2180
2181        sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2182
2183        return 0;
2184}
2185
2186static int tun_chr_close(struct inode *inode, struct file *file)
2187{
2188        struct tun_file *tfile = file->private_data;
2189        struct net *net = tfile->net;
2190
2191        tun_detach(tfile, true);
2192        put_net(net);
2193
2194        return 0;
2195}
2196
2197static const struct file_operations tun_fops = {
2198        .owner  = THIS_MODULE,
2199        .llseek = no_llseek,
2200        .read  = do_sync_read,
2201        .aio_read  = tun_chr_aio_read,
2202        .write = do_sync_write,
2203        .aio_write = tun_chr_aio_write,
2204        .poll   = tun_chr_poll,
2205        .unlocked_ioctl = tun_chr_ioctl,
2206#ifdef CONFIG_COMPAT
2207        .compat_ioctl = tun_chr_compat_ioctl,
2208#endif
2209        .open   = tun_chr_open,
2210        .release = tun_chr_close,
2211        .fasync = tun_chr_fasync
2212};
2213
2214static struct miscdevice tun_miscdev = {
2215        .minor = TUN_MINOR,
2216        .name = "tun",
2217        .nodename = "net/tun",
2218        .fops = &tun_fops,
2219};
2220
2221/* ethtool interface */
2222
2223static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2224{
2225        cmd->supported          = 0;
2226        cmd->advertising        = 0;
2227        ethtool_cmd_speed_set(cmd, SPEED_10);
2228        cmd->duplex             = DUPLEX_FULL;
2229        cmd->port               = PORT_TP;
2230        cmd->phy_address        = 0;
2231        cmd->transceiver        = XCVR_INTERNAL;
2232        cmd->autoneg            = AUTONEG_DISABLE;
2233        cmd->maxtxpkt           = 0;
2234        cmd->maxrxpkt           = 0;
2235        return 0;
2236}
2237
2238static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2239{
2240        struct tun_struct *tun = netdev_priv(dev);
2241
2242        strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2243        strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2244
2245        switch (tun->flags & TUN_TYPE_MASK) {
2246        case TUN_TUN_DEV:
2247                strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2248                break;
2249        case TUN_TAP_DEV:
2250                strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2251                break;
2252        }
2253}
2254
2255static u32 tun_get_msglevel(struct net_device *dev)
2256{
2257#ifdef TUN_DEBUG
2258        struct tun_struct *tun = netdev_priv(dev);
2259        return tun->debug;
2260#else
2261        return -EOPNOTSUPP;
2262#endif
2263}
2264
2265static void tun_set_msglevel(struct net_device *dev, u32 value)
2266{
2267#ifdef TUN_DEBUG
2268        struct tun_struct *tun = netdev_priv(dev);
2269        tun->debug = value;
2270#endif
2271}
2272
2273static const struct ethtool_ops tun_ethtool_ops = {
2274        .get_settings   = tun_get_settings,
2275        .get_drvinfo    = tun_get_drvinfo,
2276        .get_msglevel   = tun_get_msglevel,
2277        .set_msglevel   = tun_set_msglevel,
2278        .get_link       = ethtool_op_get_link,
2279};
2280
2281
2282static int __init tun_init(void)
2283{
2284        int ret = 0;
2285
2286        pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2287        pr_info("%s\n", DRV_COPYRIGHT);
2288
2289        ret = rtnl_link_register(&tun_link_ops);
2290        if (ret) {
2291                pr_err("Can't register link_ops\n");
2292                goto err_linkops;
2293        }
2294
2295        ret = misc_register(&tun_miscdev);
2296        if (ret) {
2297                pr_err("Can't register misc device %d\n", TUN_MINOR);
2298                goto err_misc;
2299        }
2300        return  0;
2301err_misc:
2302        rtnl_link_unregister(&tun_link_ops);
2303err_linkops:
2304        return ret;
2305}
2306
2307static void tun_cleanup(void)
2308{
2309        misc_deregister(&tun_miscdev);
2310        rtnl_link_unregister(&tun_link_ops);
2311}
2312
2313/* Get an underlying socket object from tun file.  Returns error unless file is
2314 * attached to a device.  The returned object works like a packet socket, it
2315 * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2316 * holding a reference to the file for as long as the socket is in use. */
2317struct socket *tun_get_socket(struct file *file)
2318{
2319        struct tun_file *tfile;
2320        if (file->f_op != &tun_fops)
2321                return ERR_PTR(-EINVAL);
2322        tfile = file->private_data;
2323        if (!tfile)
2324                return ERR_PTR(-EBADFD);
2325        return &tfile->socket;
2326}
2327EXPORT_SYMBOL_GPL(tun_get_socket);
2328
2329module_init(tun_init);
2330module_exit(tun_cleanup);
2331MODULE_DESCRIPTION(DRV_DESCRIPTION);
2332MODULE_AUTHOR(DRV_COPYRIGHT);
2333MODULE_LICENSE("GPL");
2334MODULE_ALIAS_MISCDEV(TUN_MINOR);
2335MODULE_ALIAS("devname:net/tun");
2336
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.