drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/sched/signal.h>
  48 #include <linux/major.h>
  49 #include <linux/slab.h>
  50 #include <linux/poll.h>
  51 #include <linux/fcntl.h>
  52 #include <linux/init.h>
  53 #include <linux/skbuff.h>
  54 #include <linux/netdevice.h>
  55 #include <linux/etherdevice.h>
  56 #include <linux/miscdevice.h>
  57 #include <linux/ethtool.h>
  58 #include <linux/rtnetlink.h>
  59 #include <linux/compat.h>
  60 #include <linux/if.h>
  61 #include <linux/if_arp.h>
  62 #include <linux/if_ether.h>
  63 #include <linux/if_tun.h>
  64 #include <linux/if_vlan.h>
  65 #include <linux/crc32.h>
  66 #include <linux/nsproxy.h>
  67 #include <linux/virtio_net.h>
  68 #include <linux/rcupdate.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <linux/seq_file.h>
  74 #include <linux/uio.h>
  75 #include <linux/skb_array.h>
  76 #include <linux/bpf.h>
  77 #include <linux/bpf_trace.h>
  78 #include <linux/ieee802154.h>
  79 #include <linux/if_ltalk.h>
  80 #include <uapi/linux/if_fddi.h>
  81 #include <uapi/linux/if_hippi.h>
  82 #include <uapi/linux/if_fc.h>
  83 #include <net/ax25.h>
  84 #include <net/rose.h>
  85 #include <net/6lowpan.h>
  86
  87 #include <linux/uaccess.h>
  88
  89 /* Uncomment to enable debugging */
  90 /* #define TUN_DEBUG 1 */
  91
  92 #ifdef TUN_DEBUG
  93 static int debug;
  94
  95 #define tun_debug(level, tun, fmt, args...)                     \
  96 do {                                                            \
  97         if (tun->debug)                                         \
  98                 netdev_printk(level, tun->dev, fmt, ##args);    \
  99 } while (0)
 100 #define DBG1(level, fmt, args...)                               \
 101 do {                                                            \
 102         if (debug == 2)                                         \
 103                 printk(level fmt, ##args);                      \
 104 } while (0)
 105 #else
 106 #define tun_debug(level, tun, fmt, args...)                     \
 107 do {                                                            \
 108         if (0)                                                  \
 109                 netdev_printk(level, tun->dev, fmt, ##args);    \
 110 } while (0)
 111 #define DBG1(level, fmt, args...)                               \
 112 do {                                                            \
 113         if (0)                                                  \
 114                 printk(level fmt, ##args);                      \
 115 } while (0)
 116 #endif
 117
 118 #define TUN_HEADROOM 256
 119 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 120
 121 /* TUN device flags */
 122
 123 /* IFF_ATTACH_QUEUE is never stored in device flags,
 124  * overload it to mean fasync when stored there.
 125  */
 126 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 127 /* High bits in flags field are unused. */
 128 #define TUN_VNET_LE     0x80000000
 129 #define TUN_VNET_BE     0x40000000
 130
 131 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 132                       IFF_MULTI_QUEUE)
 133 #define GOODCOPY_LEN 128
 134
 135 #define FLT_EXACT_COUNT 8
 136 struct tap_filter {
 137         unsigned int    count;    /* Number of addrs. Zero means disabled */
 138         u32             mask[2];  /* Mask of the hashed addrs */
 139         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 140 };
 141
 142 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 143  * to max number of VCPUs in guest. */
 144 #define MAX_TAP_QUEUES 256
 145 #define MAX_TAP_FLOWS  4096
 146
 147 #define TUN_FLOW_EXPIRE (3 * HZ)
 148
 149 struct tun_pcpu_stats {
 150         u64 rx_packets;
 151         u64 rx_bytes;
 152         u64 tx_packets;
 153         u64 tx_bytes;
 154         struct u64_stats_sync syncp;
 155         u32 rx_dropped;
 156         u32 tx_dropped;
 157         u32 rx_frame_errors;
 158 };
 159
 160 /* A tun_file connects an open character device to a tuntap netdevice. It
 161  * also contains all socket related structures (except sock_fprog and tap_filter)
 162  * to serve as one transmit queue for tuntap device. The sock_fprog and
 163  * tap_filter were kept in tun_struct since they were used for filtering for the
 164  * netdevice not for a specific queue (at least I didn't see the requirement for
 165  * this).
 166  *
 167  * RCU usage:
 168  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 169  * other can only be read while rcu_read_lock or rtnl_lock is held.
 170  */
 171 struct tun_file {
 172         struct sock sk;
 173         struct socket socket;
 174         struct socket_wq wq;
 175         struct tun_struct __rcu *tun;
 176         struct fasync_struct *fasync;
 177         /* only used for fasnyc */
 178         unsigned int flags;
 179         union {
 180                 u16 queue_index;
 181                 unsigned int ifindex;
 182         };
 183         struct list_head next;
 184         struct tun_struct *detached;
 185         struct skb_array tx_array;
 186 };
 187
 188 struct tun_flow_entry {
 189         struct hlist_node hash_link;
 190         struct rcu_head rcu;
 191         struct tun_struct *tun;
 192
 193         u32 rxhash;
 194         u32 rps_rxhash;
 195         int queue_index;
 196         unsigned long updated;
 197 };
 198
 199 #define TUN_NUM_FLOW_ENTRIES 1024
 200
 201 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 202  * device, socket filter, sndbuf and vnet header size were restore when the
 203  * file were attached to a persist device.
 204  */
 205 struct tun_struct {
 206         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 207         unsigned int            numqueues;
 208         unsigned int            flags;
 209         kuid_t                  owner;
 210         kgid_t                  group;
 211
 212         struct net_device       *dev;
 213         netdev_features_t       set_features;
 214 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 215                           NETIF_F_TSO6)
 216
 217         int                     align;
 218         int                     vnet_hdr_sz;
 219         int                     sndbuf;
 220         struct tap_filter       txflt;
 221         struct sock_fprog       fprog;
 222         /* protected by rtnl lock */
 223         bool                    filter_attached;
 224 #ifdef TUN_DEBUG
 225         int debug;
 226 #endif
 227         spinlock_t lock;
 228         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 229         struct timer_list flow_gc_timer;
 230         unsigned long ageing_time;
 231         unsigned int numdisabled;
 232         struct list_head disabled;
 233         void *security;
 234         u32 flow_count;
 235         u32 rx_batched;
 236         struct tun_pcpu_stats __percpu *pcpu_stats;
 237         struct bpf_prog __rcu *xdp_prog;
 238 };
 239
 240 #ifdef CONFIG_TUN_VNET_CROSS_LE
 241 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 242 {
 243         return tun->flags & TUN_VNET_BE ? false :
 244                 virtio_legacy_is_little_endian();
 245 }
 246
 247 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 248 {
 249         int be = !!(tun->flags & TUN_VNET_BE);
 250
 251         if (put_user(be, argp))
 252                 return -EFAULT;
 253
 254         return 0;
 255 }
 256
 257 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 258 {
 259         int be;
 260
 261         if (get_user(be, argp))
 262                 return -EFAULT;
 263
 264         if (be)
 265                 tun->flags |= TUN_VNET_BE;
 266         else
 267                 tun->flags &= ~TUN_VNET_BE;
 268
 269         return 0;
 270 }
 271 #else
 272 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 273 {
 274         return virtio_legacy_is_little_endian();
 275 }
 276
 277 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 278 {
 279         return -EINVAL;
 280 }
 281
 282 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 283 {
 284         return -EINVAL;
 285 }
 286 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 287
 288 static inline bool tun_is_little_endian(struct tun_struct *tun)
 289 {
 290         return tun->flags & TUN_VNET_LE ||
 291                 tun_legacy_is_little_endian(tun);
 292 }
 293
 294 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 295 {
 296         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 297 }
 298
 299 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 300 {
 301         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 302 }
 303
 304 static inline u32 tun_hashfn(u32 rxhash)
 305 {
 306         return rxhash & 0x3ff;
 307 }
 308
 309 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 310 {
 311         struct tun_flow_entry *e;
 312
 313         hlist_for_each_entry_rcu(e, head, hash_link) {
 314                 if (e->rxhash == rxhash)
 315                         return e;
 316         }
 317         return NULL;
 318 }
 319
 320 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 321                                               struct hlist_head *head,
 322                                               u32 rxhash, u16 queue_index)
 323 {
 324         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 325
 326         if (e) {
 327                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 328                           rxhash, queue_index);
 329                 e->updated = jiffies;
 330                 e->rxhash = rxhash;
 331                 e->rps_rxhash = 0;
 332                 e->queue_index = queue_index;
 333                 e->tun = tun;
 334                 hlist_add_head_rcu(&e->hash_link, head);
 335                 ++tun->flow_count;
 336         }
 337         return e;
 338 }
 339
 340 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 341 {
 342         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 343                   e->rxhash, e->queue_index);
 344         hlist_del_rcu(&e->hash_link);
 345         kfree_rcu(e, rcu);
 346         --tun->flow_count;
 347 }
 348
 349 static void tun_flow_flush(struct tun_struct *tun)
 350 {
 351         int i;
 352
 353         spin_lock_bh(&tun->lock);
 354         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 355                 struct tun_flow_entry *e;
 356                 struct hlist_node *n;
 357
 358                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 359                         tun_flow_delete(tun, e);
 360         }
 361         spin_unlock_bh(&tun->lock);
 362 }
 363
 364 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 365 {
 366         int i;
 367
 368         spin_lock_bh(&tun->lock);
 369         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 370                 struct tun_flow_entry *e;
 371                 struct hlist_node *n;
 372
 373                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 374                         if (e->queue_index == queue_index)
 375                                 tun_flow_delete(tun, e);
 376                 }
 377         }
 378         spin_unlock_bh(&tun->lock);
 379 }
 380
 381 static void tun_flow_cleanup(unsigned long data)
 382 {
 383         struct tun_struct *tun = (struct tun_struct *)data;
 384         unsigned long delay = tun->ageing_time;
 385         unsigned long next_timer = jiffies + delay;
 386         unsigned long count = 0;
 387         int i;
 388
 389         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 390
 391         spin_lock_bh(&tun->lock);
 392         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 393                 struct tun_flow_entry *e;
 394                 struct hlist_node *n;
 395
 396                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 397                         unsigned long this_timer;
 398                         count++;
 399                         this_timer = e->updated + delay;
 400                         if (time_before_eq(this_timer, jiffies))
 401                                 tun_flow_delete(tun, e);
 402                         else if (time_before(this_timer, next_timer))
 403                                 next_timer = this_timer;
 404                 }
 405         }
 406
 407         if (count)
 408                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 409         spin_unlock_bh(&tun->lock);
 410 }
 411
 412 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 413                             struct tun_file *tfile)
 414 {
 415         struct hlist_head *head;
 416         struct tun_flow_entry *e;
 417         unsigned long delay = tun->ageing_time;
 418         u16 queue_index = tfile->queue_index;
 419
 420         if (!rxhash)
 421                 return;
 422         else
 423                 head = &tun->flows[tun_hashfn(rxhash)];
 424
 425         rcu_read_lock();
 426
 427         /* We may get a very small possibility of OOO during switching, not
 428          * worth to optimize.*/
 429         if (tun->numqueues == 1 || tfile->detached)
 430                 goto unlock;
 431
 432         e = tun_flow_find(head, rxhash);
 433         if (likely(e)) {
 434                 /* TODO: keep queueing to old queue until it's empty? */
 435                 e->queue_index = queue_index;
 436                 e->updated = jiffies;
 437                 sock_rps_record_flow_hash(e->rps_rxhash);
 438         } else {
 439                 spin_lock_bh(&tun->lock);
 440                 if (!tun_flow_find(head, rxhash) &&
 441                     tun->flow_count < MAX_TAP_FLOWS)
 442                         tun_flow_create(tun, head, rxhash, queue_index);
 443
 444                 if (!timer_pending(&tun->flow_gc_timer))
 445                         mod_timer(&tun->flow_gc_timer,
 446                                   round_jiffies_up(jiffies + delay));
 447                 spin_unlock_bh(&tun->lock);
 448         }
 449
 450 unlock:
 451         rcu_read_unlock();
 452 }
 453
 454 /**
 455  * Save the hash received in the stack receive path and update the
 456  * flow_hash table accordingly.
 457  */
 458 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 459 {
 460         if (unlikely(e->rps_rxhash != hash))
 461                 e->rps_rxhash = hash;
 462 }
 463
 464 /* We try to identify a flow through its rxhash first. The reason that
 465  * we do not check rxq no. is because some cards(e.g 82599), chooses
 466  * the rxq based on the txq where the last packet of the flow comes. As
 467  * the userspace application move between processors, we may get a
 468  * different rxq no. here. If we could not get rxhash, then we would
 469  * hope the rxq no. may help here.
 470  */
 471 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 472                             void *accel_priv, select_queue_fallback_t fallback)
 473 {
 474         struct tun_struct *tun = netdev_priv(dev);
 475         struct tun_flow_entry *e;
 476         u32 txq = 0;
 477         u32 numqueues = 0;
 478
 479         rcu_read_lock();
 480         numqueues = ACCESS_ONCE(tun->numqueues);
 481
 482         txq = __skb_get_hash_symmetric(skb);
 483         if (txq) {
 484                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 485                 if (e) {
 486                         tun_flow_save_rps_rxhash(e, txq);
 487                         txq = e->queue_index;
 488                 } else
 489                         /* use multiply and shift instead of expensive divide */
 490                         txq = ((u64)txq * numqueues) >> 32;
 491         } else if (likely(skb_rx_queue_recorded(skb))) {
 492                 txq = skb_get_rx_queue(skb);
 493                 while (unlikely(txq >= numqueues))
 494                         txq -= numqueues;
 495         }
 496
 497         rcu_read_unlock();
 498         return txq;
 499 }
 500
 501 static inline bool tun_not_capable(struct tun_struct *tun)
 502 {
 503         const struct cred *cred = current_cred();
 504         struct net *net = dev_net(tun->dev);
 505
 506         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 507                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 508                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 509 }
 510
 511 static void tun_set_real_num_queues(struct tun_struct *tun)
 512 {
 513         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 514         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 515 }
 516
 517 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 518 {
 519         tfile->detached = tun;
 520         list_add_tail(&tfile->next, &tun->disabled);
 521         ++tun->numdisabled;
 522 }
 523
 524 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 525 {
 526         struct tun_struct *tun = tfile->detached;
 527
 528         tfile->detached = NULL;
 529         list_del_init(&tfile->next);
 530         --tun->numdisabled;
 531         return tun;
 532 }
 533
 534 static void tun_queue_purge(struct tun_file *tfile)
 535 {
 536         struct sk_buff *skb;
 537
 538         while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
 539                 kfree_skb(skb);
 540
 541         skb_queue_purge(&tfile->sk.sk_write_queue);
 542         skb_queue_purge(&tfile->sk.sk_error_queue);
 543 }
 544
 545 static void __tun_detach(struct tun_file *tfile, bool clean)
 546 {
 547         struct tun_file *ntfile;
 548         struct tun_struct *tun;
 549
 550         tun = rtnl_dereference(tfile->tun);
 551
 552         if (tun && !tfile->detached) {
 553                 u16 index = tfile->queue_index;
 554                 BUG_ON(index >= tun->numqueues);
 555
 556                 rcu_assign_pointer(tun->tfiles[index],
 557                                    tun->tfiles[tun->numqueues - 1]);
 558                 ntfile = rtnl_dereference(tun->tfiles[index]);
 559                 ntfile->queue_index = index;
 560
 561                 --tun->numqueues;
 562                 if (clean) {
 563                         RCU_INIT_POINTER(tfile->tun, NULL);
 564                         sock_put(&tfile->sk);
 565                 } else
 566                         tun_disable_queue(tun, tfile);
 567
 568                 synchronize_net();
 569                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 570                 /* Drop read queue */
 571                 tun_queue_purge(tfile);
 572                 tun_set_real_num_queues(tun);
 573         } else if (tfile->detached && clean) {
 574                 tun = tun_enable_queue(tfile);
 575                 sock_put(&tfile->sk);
 576         }
 577
 578         if (clean) {
 579                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 580                         netif_carrier_off(tun->dev);
 581
 582                         if (!(tun->flags & IFF_PERSIST) &&
 583                             tun->dev->reg_state == NETREG_REGISTERED)
 584                                 unregister_netdevice(tun->dev);
 585                 }
 586                 skb_array_cleanup(&tfile->tx_array);
 587                 sock_put(&tfile->sk);
 588         }
 589 }
 590
 591 static void tun_detach(struct tun_file *tfile, bool clean)
 592 {
 593         rtnl_lock();
 594         __tun_detach(tfile, clean);
 595         rtnl_unlock();
 596 }
 597
 598 static void tun_detach_all(struct net_device *dev)
 599 {
 600         struct tun_struct *tun = netdev_priv(dev);
 601         struct bpf_prog *xdp_prog = rtnl_dereference(tun->xdp_prog);
 602         struct tun_file *tfile, *tmp;
 603         int i, n = tun->numqueues;
 604
 605         for (i = 0; i < n; i++) {
 606                 tfile = rtnl_dereference(tun->tfiles[i]);
 607                 BUG_ON(!tfile);
 608                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 609                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 610                 RCU_INIT_POINTER(tfile->tun, NULL);
 611                 --tun->numqueues;
 612         }
 613         list_for_each_entry(tfile, &tun->disabled, next) {
 614                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 615                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 616                 RCU_INIT_POINTER(tfile->tun, NULL);
 617         }
 618         BUG_ON(tun->numqueues != 0);
 619
 620         synchronize_net();
 621         for (i = 0; i < n; i++) {
 622                 tfile = rtnl_dereference(tun->tfiles[i]);
 623                 /* Drop read queue */
 624                 tun_queue_purge(tfile);
 625                 sock_put(&tfile->sk);
 626         }
 627         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 628                 tun_enable_queue(tfile);
 629                 tun_queue_purge(tfile);
 630                 sock_put(&tfile->sk);
 631         }
 632         BUG_ON(tun->numdisabled != 0);
 633
 634         if (xdp_prog)
 635                 bpf_prog_put(xdp_prog);
 636
 637         if (tun->flags & IFF_PERSIST)
 638                 module_put(THIS_MODULE);
 639 }
 640
 641 static int tun_attach(struct tun_struct *tun, struct file *file,
 642                       bool skip_filter, bool publish_tun)
 643 {
 644         struct tun_file *tfile = file->private_data;
 645         struct net_device *dev = tun->dev;
 646         int err;
 647
 648         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 649         if (err < 0)
 650                 goto out;
 651
 652         err = -EINVAL;
 653         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 654                 goto out;
 655
 656         err = -EBUSY;
 657         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 658                 goto out;
 659
 660         err = -E2BIG;
 661         if (!tfile->detached &&
 662             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 663                 goto out;
 664
 665         err = 0;
 666
 667         /* Re-attach the filter to persist device */
 668         if (!skip_filter && (tun->filter_attached == true)) {
 669                 lock_sock(tfile->socket.sk);
 670                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 671                 release_sock(tfile->socket.sk);
 672                 if (!err)
 673                         goto out;
 674         }
 675
 676         if (!tfile->detached &&
 677             skb_array_resize(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
 678                 err = -ENOMEM;
 679                 goto out;
 680         }
 681
 682         tfile->queue_index = tun->numqueues;
 683         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 684         if (publish_tun)
 685                 rcu_assign_pointer(tfile->tun, tun);
 686         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 687         tun->numqueues++;
 688
 689         if (tfile->detached)
 690                 tun_enable_queue(tfile);
 691         else
 692                 sock_hold(&tfile->sk);
 693
 694         tun_set_real_num_queues(tun);
 695
 696         /* device is allowed to go away first, so no need to hold extra
 697          * refcnt.
 698          */
 699
 700 out:
 701         return err;
 702 }
 703
 704 static struct tun_struct *__tun_get(struct tun_file *tfile)
 705 {
 706         struct tun_struct *tun;
 707
 708         rcu_read_lock();
 709         tun = rcu_dereference(tfile->tun);
 710         if (tun)
 711                 dev_hold(tun->dev);
 712         rcu_read_unlock();
 713
 714         return tun;
 715 }
 716
 717 static struct tun_struct *tun_get(struct file *file)
 718 {
 719         return __tun_get(file->private_data);
 720 }
 721
 722 static void tun_put(struct tun_struct *tun)
 723 {
 724         dev_put(tun->dev);
 725 }
 726
 727 /* TAP filtering */
 728 static void addr_hash_set(u32 *mask, const u8 *addr)
 729 {
 730         int n = ether_crc(ETH_ALEN, addr) >> 26;
 731         mask[n >> 5] |= (1 << (n & 31));
 732 }
 733
 734 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 735 {
 736         int n = ether_crc(ETH_ALEN, addr) >> 26;
 737         return mask[n >> 5] & (1 << (n & 31));
 738 }
 739
 740 static int update_filter(struct tap_filter *filter, void __user *arg)
 741 {
 742         struct { u8 u[ETH_ALEN]; } *addr;
 743         struct tun_filter uf;
 744         int err, alen, n, nexact;
 745
 746         if (copy_from_user(&uf, arg, sizeof(uf)))
 747                 return -EFAULT;
 748
 749         if (!uf.count) {
 750                 /* Disabled */
 751                 filter->count = 0;
 752                 return 0;
 753         }
 754
 755         alen = ETH_ALEN * uf.count;
 756         addr = memdup_user(arg + sizeof(uf), alen);
 757         if (IS_ERR(addr))
 758                 return PTR_ERR(addr);
 759
 760         /* The filter is updated without holding any locks. Which is
 761          * perfectly safe. We disable it first and in the worst
 762          * case we'll accept a few undesired packets. */
 763         filter->count = 0;
 764         wmb();
 765
 766         /* Use first set of addresses as an exact filter */
 767         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 768                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 769
 770         nexact = n;
 771
 772         /* Remaining multicast addresses are hashed,
 773          * unicast will leave the filter disabled. */
 774         memset(filter->mask, 0, sizeof(filter->mask));
 775         for (; n < uf.count; n++) {
 776                 if (!is_multicast_ether_addr(addr[n].u)) {
 777                         err = 0; /* no filter */
 778                         goto free_addr;
 779                 }
 780                 addr_hash_set(filter->mask, addr[n].u);
 781         }
 782
 783         /* For ALLMULTI just set the mask to all ones.
 784          * This overrides the mask populated above. */
 785         if ((uf.flags & TUN_FLT_ALLMULTI))
 786                 memset(filter->mask, ~0, sizeof(filter->mask));
 787
 788         /* Now enable the filter */
 789         wmb();
 790         filter->count = nexact;
 791
 792         /* Return the number of exact filters */
 793         err = nexact;
 794 free_addr:
 795         kfree(addr);
 796         return err;
 797 }
 798
 799 /* Returns: 0 - drop, !=0 - accept */
 800 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 801 {
 802         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 803          * at this point. */
 804         struct ethhdr *eh = (struct ethhdr *) skb->data;
 805         int i;
 806
 807         /* Exact match */
 808         for (i = 0; i < filter->count; i++)
 809                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 810                         return 1;
 811
 812         /* Inexact match (multicast only) */
 813         if (is_multicast_ether_addr(eh->h_dest))
 814                 return addr_hash_test(filter->mask, eh->h_dest);
 815
 816         return 0;
 817 }
 818
 819 /*
 820  * Checks whether the packet is accepted or not.
 821  * Returns: 0 - drop, !=0 - accept
 822  */
 823 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 824 {
 825         if (!filter->count)
 826                 return 1;
 827
 828         return run_filter(filter, skb);
 829 }
 830
 831 /* Network device part of the driver */
 832
 833 static const struct ethtool_ops tun_ethtool_ops;
 834
 835 /* Net device detach from fd. */
 836 static void tun_net_uninit(struct net_device *dev)
 837 {
 838         tun_detach_all(dev);
 839 }
 840
 841 /* Net device open. */
 842 static int tun_net_open(struct net_device *dev)
 843 {
 844         netif_tx_start_all_queues(dev);
 845
 846         return 0;
 847 }
 848
 849 /* Net device close. */
 850 static int tun_net_close(struct net_device *dev)
 851 {
 852         netif_tx_stop_all_queues(dev);
 853         return 0;
 854 }
 855
 856 /* Net device start xmit */
 857 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 858 {
 859         struct tun_struct *tun = netdev_priv(dev);
 860         int txq = skb->queue_mapping;
 861         struct tun_file *tfile;
 862         u32 numqueues = 0;
 863
 864         rcu_read_lock();
 865         tfile = rcu_dereference(tun->tfiles[txq]);
 866         numqueues = ACCESS_ONCE(tun->numqueues);
 867
 868         /* Drop packet if interface is not attached */
 869         if (txq >= numqueues)
 870                 goto drop;
 871
 872 #ifdef CONFIG_RPS
 873         if (numqueues == 1 && static_key_false(&rps_needed)) {
 874                 /* Select queue was not called for the skbuff, so we extract the
 875                  * RPS hash and save it into the flow_table here.
 876                  */
 877                 __u32 rxhash;
 878
 879                 rxhash = __skb_get_hash_symmetric(skb);
 880                 if (rxhash) {
 881                         struct tun_flow_entry *e;
 882                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 883                                         rxhash);
 884                         if (e)
 885                                 tun_flow_save_rps_rxhash(e, rxhash);
 886                 }
 887         }
 888 #endif
 889
 890         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 891
 892         BUG_ON(!tfile);
 893
 894         /* Drop if the filter does not like it.
 895          * This is a noop if the filter is disabled.
 896          * Filter can be enabled only for the TAP devices. */
 897         if (!check_filter(&tun->txflt, skb))
 898                 goto drop;
 899
 900         if (tfile->socket.sk->sk_filter &&
 901             sk_filter(tfile->socket.sk, skb))
 902                 goto drop;
 903
 904         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
 905                 goto drop;
 906
 907         skb_tx_timestamp(skb);
 908
 909         /* Orphan the skb - required as we might hang on to it
 910          * for indefinite time.
 911          */
 912         skb_orphan(skb);
 913
 914         nf_reset(skb);
 915
 916         if (skb_array_produce(&tfile->tx_array, skb))
 917                 goto drop;
 918
 919         /* Notify and wake up reader process */
 920         if (tfile->flags & TUN_FASYNC)
 921                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 922         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 923
 924         rcu_read_unlock();
 925         return NETDEV_TX_OK;
 926
 927 drop:
 928         this_cpu_inc(tun->pcpu_stats->tx_dropped);
 929         skb_tx_error(skb);
 930         kfree_skb(skb);
 931         rcu_read_unlock();
 932         return NET_XMIT_DROP;
 933 }
 934
 935 static void tun_net_mclist(struct net_device *dev)
 936 {
 937         /*
 938          * This callback is supposed to deal with mc filter in
 939          * _rx_ path and has nothing to do with the _tx_ path.
 940          * In rx path we always accept everything userspace gives us.
 941          */
 942 }
 943
 944 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 945         netdev_features_t features)
 946 {
 947         struct tun_struct *tun = netdev_priv(dev);
 948
 949         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 950 }
 951 #ifdef CONFIG_NET_POLL_CONTROLLER
 952 static void tun_poll_controller(struct net_device *dev)
 953 {
 954         /*
 955          * Tun only receives frames when:
 956          * 1) the char device endpoint gets data from user space
 957          * 2) the tun socket gets a sendmsg call from user space
 958          * Since both of those are synchronous operations, we are guaranteed
 959          * never to have pending data when we poll for it
 960          * so there is nothing to do here but return.
 961          * We need this though so netpoll recognizes us as an interface that
 962          * supports polling, which enables bridge devices in virt setups to
 963          * still use netconsole
 964          */
 965         return;
 966 }
 967 #endif
 968
 969 static void tun_set_headroom(struct net_device *dev, int new_hr)
 970 {
 971         struct tun_struct *tun = netdev_priv(dev);
 972
 973         if (new_hr < NET_SKB_PAD)
 974                 new_hr = NET_SKB_PAD;
 975
 976         tun->align = new_hr;
 977 }
 978
 979 static void
 980 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
 981 {
 982         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
 983         struct tun_struct *tun = netdev_priv(dev);
 984         struct tun_pcpu_stats *p;
 985         int i;
 986
 987         for_each_possible_cpu(i) {
 988                 u64 rxpackets, rxbytes, txpackets, txbytes;
 989                 unsigned int start;
 990
 991                 p = per_cpu_ptr(tun->pcpu_stats, i);
 992                 do {
 993                         start = u64_stats_fetch_begin(&p->syncp);
 994                         rxpackets       = p->rx_packets;
 995                         rxbytes         = p->rx_bytes;
 996                         txpackets       = p->tx_packets;
 997                         txbytes         = p->tx_bytes;
 998                 } while (u64_stats_fetch_retry(&p->syncp, start));
 999
1000                 stats->rx_packets       += rxpackets;
1001                 stats->rx_bytes         += rxbytes;
1002                 stats->tx_packets       += txpackets;
1003                 stats->tx_bytes         += txbytes;
1004
1005                 /* u32 counters */
1006                 rx_dropped      += p->rx_dropped;
1007                 rx_frame_errors += p->rx_frame_errors;
1008                 tx_dropped      += p->tx_dropped;
1009         }
1010         stats->rx_dropped  = rx_dropped;
1011         stats->rx_frame_errors = rx_frame_errors;
1012         stats->tx_dropped = tx_dropped;
1013 }
1014
1015 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1016                        struct netlink_ext_ack *extack)
1017 {
1018         struct tun_struct *tun = netdev_priv(dev);
1019         struct bpf_prog *old_prog;
1020
1021         old_prog = rtnl_dereference(tun->xdp_prog);
1022         rcu_assign_pointer(tun->xdp_prog, prog);
1023         if (old_prog)
1024                 bpf_prog_put(old_prog);
1025
1026         return 0;
1027 }
1028
1029 static u32 tun_xdp_query(struct net_device *dev)
1030 {
1031         struct tun_struct *tun = netdev_priv(dev);
1032         const struct bpf_prog *xdp_prog;
1033
1034         xdp_prog = rtnl_dereference(tun->xdp_prog);
1035         if (xdp_prog)
1036                 return xdp_prog->aux->id;
1037
1038         return 0;
1039 }
1040
1041 static int tun_xdp(struct net_device *dev, struct netdev_xdp *xdp)
1042 {
1043         switch (xdp->command) {
1044         case XDP_SETUP_PROG:
1045                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1046         case XDP_QUERY_PROG:
1047                 xdp->prog_id = tun_xdp_query(dev);
1048                 xdp->prog_attached = !!xdp->prog_id;
1049                 return 0;
1050         default:
1051                 return -EINVAL;
1052         }
1053 }
1054
1055 static const struct net_device_ops tun_netdev_ops = {
1056         .ndo_uninit             = tun_net_uninit,
1057         .ndo_open               = tun_net_open,
1058         .ndo_stop               = tun_net_close,
1059         .ndo_start_xmit         = tun_net_xmit,
1060         .ndo_fix_features       = tun_net_fix_features,
1061         .ndo_select_queue       = tun_select_queue,
1062 #ifdef CONFIG_NET_POLL_CONTROLLER
1063         .ndo_poll_controller    = tun_poll_controller,
1064 #endif
1065         .ndo_set_rx_headroom    = tun_set_headroom,
1066         .ndo_get_stats64        = tun_net_get_stats64,
1067 };
1068
1069 static const struct net_device_ops tap_netdev_ops = {
1070         .ndo_uninit             = tun_net_uninit,
1071         .ndo_open               = tun_net_open,
1072         .ndo_stop               = tun_net_close,
1073         .ndo_start_xmit         = tun_net_xmit,
1074         .ndo_fix_features       = tun_net_fix_features,
1075         .ndo_set_rx_mode        = tun_net_mclist,
1076         .ndo_set_mac_address    = eth_mac_addr,
1077         .ndo_validate_addr      = eth_validate_addr,
1078         .ndo_select_queue       = tun_select_queue,
1079 #ifdef CONFIG_NET_POLL_CONTROLLER
1080         .ndo_poll_controller    = tun_poll_controller,
1081 #endif
1082         .ndo_features_check     = passthru_features_check,
1083         .ndo_set_rx_headroom    = tun_set_headroom,
1084         .ndo_get_stats64        = tun_net_get_stats64,
1085         .ndo_xdp                = tun_xdp,
1086 };
1087
1088 static void tun_flow_init(struct tun_struct *tun)
1089 {
1090         int i;
1091
1092         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1093                 INIT_HLIST_HEAD(&tun->flows[i]);
1094
1095         tun->ageing_time = TUN_FLOW_EXPIRE;
1096         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1097         mod_timer(&tun->flow_gc_timer,
1098                   round_jiffies_up(jiffies + tun->ageing_time));
1099 }
1100
1101 static void tun_flow_uninit(struct tun_struct *tun)
1102 {
1103         del_timer_sync(&tun->flow_gc_timer);
1104         tun_flow_flush(tun);
1105 }
1106
1107 #define MIN_MTU 68
1108 #define MAX_MTU 65535
1109
1110 /* Initialize net device. */
1111 static void tun_net_init(struct net_device *dev)
1112 {
1113         struct tun_struct *tun = netdev_priv(dev);
1114
1115         switch (tun->flags & TUN_TYPE_MASK) {
1116         case IFF_TUN:
1117                 dev->netdev_ops = &tun_netdev_ops;
1118
1119                 /* Point-to-Point TUN Device */
1120                 dev->hard_header_len = 0;
1121                 dev->addr_len = 0;
1122                 dev->mtu = 1500;
1123
1124                 /* Zero header length */
1125                 dev->type = ARPHRD_NONE;
1126                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1127                 break;
1128
1129         case IFF_TAP:
1130                 dev->netdev_ops = &tap_netdev_ops;
1131                 /* Ethernet TAP Device */
1132                 ether_setup(dev);
1133                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1134                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1135
1136                 eth_hw_addr_random(dev);
1137
1138                 break;
1139         }
1140
1141         dev->min_mtu = MIN_MTU;
1142         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1143 }
1144
1145 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1146 {
1147         struct sock *sk = tfile->socket.sk;
1148
1149         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1150 }
1151
1152 /* Character device part */
1153
1154 /* Poll */
1155 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1156 {
1157         struct tun_file *tfile = file->private_data;
1158         struct tun_struct *tun = __tun_get(tfile);
1159         struct sock *sk;
1160         unsigned int mask = 0;
1161
1162         if (!tun)
1163                 return POLLERR;
1164
1165         sk = tfile->socket.sk;
1166
1167         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1168
1169         poll_wait(file, sk_sleep(sk), wait);
1170
1171         if (!skb_array_empty(&tfile->tx_array))
1172                 mask |= POLLIN | POLLRDNORM;
1173
1174         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1175          * guarantee EPOLLOUT to be raised by either here or
1176          * tun_sock_write_space(). Then process could get notification
1177          * after it writes to a down device and meets -EIO.
1178          */
1179         if (tun_sock_writeable(tun, tfile) ||
1180             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1181              tun_sock_writeable(tun, tfile)))
1182                 mask |= POLLOUT | POLLWRNORM;
1183
1184         if (tun->dev->reg_state != NETREG_REGISTERED)
1185                 mask = POLLERR;
1186
1187         tun_put(tun);
1188         return mask;
1189 }
1190
1191 /* prepad is the amount to reserve at front.  len is length after that.
1192  * linear is a hint as to how much to copy (usually headers). */
1193 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1194                                      size_t prepad, size_t len,
1195                                      size_t linear, int noblock)
1196 {
1197         struct sock *sk = tfile->socket.sk;
1198         struct sk_buff *skb;
1199         int err;
1200
1201         /* Under a page?  Don't bother with paged skb. */
1202         if (prepad + len < PAGE_SIZE || !linear)
1203                 linear = len;
1204
1205         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1206                                    &err, 0);
1207         if (!skb)
1208                 return ERR_PTR(err);
1209
1210         skb_reserve(skb, prepad);
1211         skb_put(skb, linear);
1212         skb->data_len = len - linear;
1213         skb->len += len - linear;
1214
1215         return skb;
1216 }
1217
1218 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1219                            struct sk_buff *skb, int more)
1220 {
1221         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1222         struct sk_buff_head process_queue;
1223         u32 rx_batched = tun->rx_batched;
1224         bool rcv = false;
1225
1226         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1227                 local_bh_disable();
1228                 skb_record_rx_queue(skb, tfile->queue_index);
1229                 netif_receive_skb(skb);
1230                 local_bh_enable();
1231                 return;
1232         }
1233
1234         spin_lock(&queue->lock);
1235         if (!more || skb_queue_len(queue) == rx_batched) {
1236                 __skb_queue_head_init(&process_queue);
1237                 skb_queue_splice_tail_init(queue, &process_queue);
1238                 rcv = true;
1239         } else {
1240                 __skb_queue_tail(queue, skb);
1241         }
1242         spin_unlock(&queue->lock);
1243
1244         if (rcv) {
1245                 struct sk_buff *nskb;
1246
1247                 local_bh_disable();
1248                 while ((nskb = __skb_dequeue(&process_queue))) {
1249                         skb_record_rx_queue(nskb, tfile->queue_index);
1250                         netif_receive_skb(nskb);
1251                 }
1252                 skb_record_rx_queue(skb, tfile->queue_index);
1253                 netif_receive_skb(skb);
1254                 local_bh_enable();
1255         }
1256 }
1257
1258 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1259                               int len, int noblock, bool zerocopy)
1260 {
1261         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1262                 return false;
1263
1264         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1265                 return false;
1266
1267         if (!noblock)
1268                 return false;
1269
1270         if (zerocopy)
1271                 return false;
1272
1273         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1274             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1275                 return false;
1276
1277         return true;
1278 }
1279
1280 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1281                                      struct tun_file *tfile,
1282                                      struct iov_iter *from,
1283                                      struct virtio_net_hdr *hdr,
1284                                      int len, int *skb_xdp)
1285 {
1286         struct page_frag *alloc_frag = &current->task_frag;
1287         struct sk_buff *skb;
1288         struct bpf_prog *xdp_prog;
1289         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1290         unsigned int delta = 0;
1291         char *buf;
1292         size_t copied;
1293         bool xdp_xmit = false;
1294         int err, pad = TUN_RX_PAD;
1295
1296         rcu_read_lock();
1297         xdp_prog = rcu_dereference(tun->xdp_prog);
1298         if (xdp_prog)
1299                 pad += TUN_HEADROOM;
1300         buflen += SKB_DATA_ALIGN(len + pad);
1301         rcu_read_unlock();
1302
1303         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1304         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1305                 return ERR_PTR(-ENOMEM);
1306
1307         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1308         copied = copy_page_from_iter(alloc_frag->page,
1309                                      alloc_frag->offset + pad,
1310                                      len, from);
1311         if (copied != len)
1312                 return ERR_PTR(-EFAULT);
1313
1314         /* There's a small window that XDP may be set after the check
1315          * of xdp_prog above, this should be rare and for simplicity
1316          * we do XDP on skb in case the headroom is not enough.
1317          */
1318         if (hdr->gso_type || !xdp_prog)
1319                 *skb_xdp = 1;
1320         else
1321                 *skb_xdp = 0;
1322
1323         local_bh_disable();
1324         rcu_read_lock();
1325         xdp_prog = rcu_dereference(tun->xdp_prog);
1326         if (xdp_prog && !*skb_xdp) {
1327                 struct xdp_buff xdp;
1328                 void *orig_data;
1329                 u32 act;
1330
1331                 xdp.data_hard_start = buf;
1332                 xdp.data = buf + pad;
1333                 xdp.data_end = xdp.data + len;
1334                 orig_data = xdp.data;
1335                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1336
1337                 switch (act) {
1338                 case XDP_REDIRECT:
1339                         get_page(alloc_frag->page);
1340                         alloc_frag->offset += buflen;
1341                         err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1342                         xdp_do_flush_map();
1343                         if (err)
1344                                 goto err_redirect;
1345                         rcu_read_unlock();
1346                         local_bh_enable();
1347                         return NULL;
1348                 case XDP_TX:
1349                         xdp_xmit = true;
1350                         /* fall through */
1351                 case XDP_PASS:
1352                         delta = orig_data - xdp.data;
1353                         break;
1354                 default:
1355                         bpf_warn_invalid_xdp_action(act);
1356                         /* fall through */
1357                 case XDP_ABORTED:
1358                         trace_xdp_exception(tun->dev, xdp_prog, act);
1359                         /* fall through */
1360                 case XDP_DROP:
1361                         goto err_xdp;
1362                 }
1363         }
1364
1365         skb = build_skb(buf, buflen);
1366         if (!skb) {
1367                 rcu_read_unlock();
1368                 local_bh_enable();
1369                 return ERR_PTR(-ENOMEM);
1370         }
1371
1372         skb_reserve(skb, pad - delta);
1373         skb_put(skb, len + delta);
1374         skb_set_owner_w(skb, tfile->socket.sk);
1375         get_page(alloc_frag->page);
1376         alloc_frag->offset += buflen;
1377
1378         if (xdp_xmit) {
1379                 skb->dev = tun->dev;
1380                 generic_xdp_tx(skb, xdp_prog);
1381                 rcu_read_unlock();
1382                 local_bh_enable();
1383                 return NULL;
1384         }
1385
1386         rcu_read_unlock();
1387         local_bh_enable();
1388
1389         return skb;
1390
1391 err_redirect:
1392         put_page(alloc_frag->page);
1393 err_xdp:
1394         rcu_read_unlock();
1395         local_bh_enable();
1396         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1397         return NULL;
1398 }
1399
1400 /* Get packet from user space buffer */
1401 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1402                             void *msg_control, struct iov_iter *from,
1403                             int noblock, bool more)
1404 {
1405         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1406         struct sk_buff *skb;
1407         size_t total_len = iov_iter_count(from);
1408         size_t len = total_len, align = tun->align, linear;
1409         struct virtio_net_hdr gso = { 0 };
1410         struct tun_pcpu_stats *stats;
1411         int good_linear;
1412         int copylen;
1413         bool zerocopy = false;
1414         int err;
1415         u32 rxhash;
1416         int skb_xdp = 1;
1417
1418         if (!(tun->flags & IFF_NO_PI)) {
1419                 if (len < sizeof(pi))
1420                         return -EINVAL;
1421                 len -= sizeof(pi);
1422
1423                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1424                         return -EFAULT;
1425         }
1426
1427         if (tun->flags & IFF_VNET_HDR) {
1428                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1429
1430                 if (len < vnet_hdr_sz)
1431                         return -EINVAL;
1432                 len -= vnet_hdr_sz;
1433
1434                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1435                         return -EFAULT;
1436
1437                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1438                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1439                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1440
1441                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1442                         return -EINVAL;
1443                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1444         }
1445
1446         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1447                 align += NET_IP_ALIGN;
1448                 if (unlikely(len < ETH_HLEN ||
1449                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1450                         return -EINVAL;
1451         }
1452
1453         good_linear = SKB_MAX_HEAD(align);
1454
1455         if (msg_control) {
1456                 struct iov_iter i = *from;
1457
1458                 /* There are 256 bytes to be copied in skb, so there is
1459                  * enough room for skb expand head in case it is used.
1460                  * The rest of the buffer is mapped from userspace.
1461                  */
1462                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1463                 if (copylen > good_linear)
1464                         copylen = good_linear;
1465                 linear = copylen;
1466                 iov_iter_advance(&i, copylen);
1467                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1468                         zerocopy = true;
1469         }
1470
1471         if (tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1472                 /* For the packet that is not easy to be processed
1473                  * (e.g gso or jumbo packet), we will do it at after
1474                  * skb was created with generic XDP routine.
1475                  */
1476                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1477                 if (IS_ERR(skb)) {
1478                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1479                         return PTR_ERR(skb);
1480                 }
1481                 if (!skb)
1482                         return total_len;
1483         } else {
1484                 if (!zerocopy) {
1485                         copylen = len;
1486                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1487                                 linear = good_linear;
1488                         else
1489                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1490                 }
1491
1492                 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1493                 if (IS_ERR(skb)) {
1494                         if (PTR_ERR(skb) != -EAGAIN)
1495                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1496                         return PTR_ERR(skb);
1497                 }
1498
1499                 if (zerocopy)
1500                         err = zerocopy_sg_from_iter(skb, from);
1501                 else
1502                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1503
1504                 if (err) {
1505                         err = -EFAULT;
1506 drop:
1507                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1508                         kfree_skb(skb);
1509                         return err;
1510                 }
1511         }
1512
1513         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1514                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1515                 kfree_skb(skb);
1516                 return -EINVAL;
1517         }
1518
1519         switch (tun->flags & TUN_TYPE_MASK) {
1520         case IFF_TUN:
1521                 if (tun->flags & IFF_NO_PI) {
1522                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1523
1524                         switch (ip_version) {
1525                         case 4:
1526                                 pi.proto = htons(ETH_P_IP);
1527                                 break;
1528                         case 6:
1529                                 pi.proto = htons(ETH_P_IPV6);
1530                                 break;
1531                         default:
1532                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1533                                 kfree_skb(skb);
1534                                 return -EINVAL;
1535                         }
1536                 }
1537
1538                 skb_reset_mac_header(skb);
1539                 skb->protocol = pi.proto;
1540                 skb->dev = tun->dev;
1541                 break;
1542         case IFF_TAP:
1543                 skb->protocol = eth_type_trans(skb, tun->dev);
1544                 break;
1545         }
1546
1547         /* copy skb_ubuf_info for callback when skb has no error */
1548         if (zerocopy) {
1549                 skb_shinfo(skb)->destructor_arg = msg_control;
1550                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1551                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1552         } else if (msg_control) {
1553                 struct ubuf_info *uarg = msg_control;
1554                 uarg->callback(uarg, false);
1555         }
1556
1557         skb_reset_network_header(skb);
1558         skb_probe_transport_header(skb, 0);
1559
1560         if (skb_xdp) {
1561                 struct bpf_prog *xdp_prog;
1562                 int ret;
1563
1564                 local_bh_disable();
1565                 rcu_read_lock();
1566                 xdp_prog = rcu_dereference(tun->xdp_prog);
1567                 if (xdp_prog) {
1568                         ret = do_xdp_generic(xdp_prog, skb);
1569                         if (ret != XDP_PASS) {
1570                                 rcu_read_unlock();
1571                                 local_bh_enable();
1572                                 return total_len;
1573                         }
1574                 }
1575                 rcu_read_unlock();
1576                 local_bh_enable();
1577         }
1578
1579         rxhash = __skb_get_hash_symmetric(skb);
1580
1581         rcu_read_lock();
1582         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1583                 err = -EIO;
1584                 rcu_read_unlock();
1585                 goto drop;
1586         }
1587
1588 #ifndef CONFIG_4KSTACKS
1589         tun_rx_batched(tun, tfile, skb, more);
1590 #else
1591         netif_rx_ni(skb);
1592 #endif
1593         rcu_read_unlock();
1594
1595         stats = get_cpu_ptr(tun->pcpu_stats);
1596         u64_stats_update_begin(&stats->syncp);
1597         stats->rx_packets++;
1598         stats->rx_bytes += len;
1599         u64_stats_update_end(&stats->syncp);
1600         put_cpu_ptr(stats);
1601
1602         tun_flow_update(tun, rxhash, tfile);
1603         return total_len;
1604 }
1605
1606 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1607 {
1608         struct file *file = iocb->ki_filp;
1609         struct tun_struct *tun = tun_get(file);
1610         struct tun_file *tfile = file->private_data;
1611         ssize_t result;
1612         int noblock = 0;
1613
1614         if (!tun)
1615                 return -EBADFD;
1616
1617         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
1618                 noblock = 1;
1619
1620         result = tun_get_user(tun, tfile, NULL, from, noblock, false);
1621
1622         tun_put(tun);
1623         return result;
1624 }
1625
1626 /* Put packet to the user space buffer */
1627 static ssize_t tun_put_user(struct tun_struct *tun,
1628                             struct tun_file *tfile,
1629                             struct sk_buff *skb,
1630                             struct iov_iter *iter)
1631 {
1632         struct tun_pi pi = { 0, skb->protocol };
1633         struct tun_pcpu_stats *stats;
1634         ssize_t total;
1635         int vlan_offset = 0;
1636         int vlan_hlen = 0;
1637         int vnet_hdr_sz = 0;
1638
1639         if (skb_vlan_tag_present(skb))
1640                 vlan_hlen = VLAN_HLEN;
1641
1642         if (tun->flags & IFF_VNET_HDR)
1643                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1644
1645         total = skb->len + vlan_hlen + vnet_hdr_sz;
1646
1647         if (!(tun->flags & IFF_NO_PI)) {
1648                 if (iov_iter_count(iter) < sizeof(pi))
1649                         return -EINVAL;
1650
1651                 total += sizeof(pi);
1652                 if (iov_iter_count(iter) < total) {
1653                         /* Packet will be striped */
1654                         pi.flags |= TUN_PKT_STRIP;
1655                 }
1656
1657                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1658                         return -EFAULT;
1659         }
1660
1661         if (vnet_hdr_sz) {
1662                 struct virtio_net_hdr gso;
1663
1664                 if (iov_iter_count(iter) < vnet_hdr_sz)
1665                         return -EINVAL;
1666
1667                 if (virtio_net_hdr_from_skb(skb, &gso,
1668                                             tun_is_little_endian(tun), true,
1669                                             vlan_hlen)) {
1670                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1671                         pr_err("unexpected GSO type: "
1672                                "0x%x, gso_size %d, hdr_len %d\n",
1673                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1674                                tun16_to_cpu(tun, gso.hdr_len));
1675                         print_hex_dump(KERN_ERR, "tun: ",
1676                                        DUMP_PREFIX_NONE,
1677                                        16, 1, skb->head,
1678                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1679                         WARN_ON_ONCE(1);
1680                         return -EINVAL;
1681                 }
1682
1683                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1684                         return -EFAULT;
1685
1686                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1687         }
1688
1689         if (vlan_hlen) {
1690                 int ret;
1691                 struct {
1692                         __be16 h_vlan_proto;
1693                         __be16 h_vlan_TCI;
1694                 } veth;
1695
1696                 veth.h_vlan_proto = skb->vlan_proto;
1697                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1698
1699                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1700
1701                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1702                 if (ret || !iov_iter_count(iter))
1703                         goto done;
1704
1705                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1706                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1707                         goto done;
1708         }
1709
1710         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1711
1712 done:
1713         /* caller is in process context, */
1714         stats = get_cpu_ptr(tun->pcpu_stats);
1715         u64_stats_update_begin(&stats->syncp);
1716         stats->tx_packets++;
1717         stats->tx_bytes += skb->len + vlan_hlen;
1718         u64_stats_update_end(&stats->syncp);
1719         put_cpu_ptr(tun->pcpu_stats);
1720
1721         return total;
1722 }
1723
1724 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1725                                      int *err)
1726 {
1727         DECLARE_WAITQUEUE(wait, current);
1728         struct sk_buff *skb = NULL;
1729         int error = 0;
1730
1731         skb = skb_array_consume(&tfile->tx_array);
1732         if (skb)
1733                 goto out;
1734         if (noblock) {
1735                 error = -EAGAIN;
1736                 goto out;
1737         }
1738
1739         add_wait_queue(&tfile->wq.wait, &wait);
1740
1741         while (1) {
1742                 set_current_state(TASK_INTERRUPTIBLE);
1743                 skb = skb_array_consume(&tfile->tx_array);
1744                 if (skb)
1745                         break;
1746                 if (signal_pending(current)) {
1747                         error = -ERESTARTSYS;
1748                         break;
1749                 }
1750                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1751                         error = -EFAULT;
1752                         break;
1753                 }
1754
1755                 schedule();
1756         }
1757
1758         __set_current_state(TASK_RUNNING);
1759         remove_wait_queue(&tfile->wq.wait, &wait);
1760
1761 out:
1762         *err = error;
1763         return skb;
1764 }
1765
1766 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1767                            struct iov_iter *to,
1768                            int noblock, struct sk_buff *skb)
1769 {
1770         ssize_t ret;
1771         int err;
1772
1773         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1774
1775         if (!iov_iter_count(to)) {
1776                 if (skb)
1777                         kfree_skb(skb);
1778                 return 0;
1779         }
1780
1781         if (!skb) {
1782                 /* Read frames from ring */
1783                 skb = tun_ring_recv(tfile, noblock, &err);
1784                 if (!skb)
1785                         return err;
1786         }
1787
1788         ret = tun_put_user(tun, tfile, skb, to);
1789         if (unlikely(ret < 0))
1790                 kfree_skb(skb);
1791         else
1792                 consume_skb(skb);
1793
1794         return ret;
1795 }
1796
1797 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1798 {
1799         struct file *file = iocb->ki_filp;
1800         struct tun_file *tfile = file->private_data;
1801         struct tun_struct *tun = __tun_get(tfile);
1802         ssize_t len = iov_iter_count(to), ret;
1803         int noblock = 0;
1804
1805         if (!tun)
1806                 return -EBADFD;
1807
1808         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
1809                 noblock = 1;
1810
1811         ret = tun_do_read(tun, tfile, to, noblock, NULL);
1812         ret = min_t(ssize_t, ret, len);
1813         if (ret > 0)
1814                 iocb->ki_pos = ret;
1815         tun_put(tun);
1816         return ret;
1817 }
1818
1819 static void tun_free_netdev(struct net_device *dev)
1820 {
1821         struct tun_struct *tun = netdev_priv(dev);
1822
1823         BUG_ON(!(list_empty(&tun->disabled)));
1824         free_percpu(tun->pcpu_stats);
1825         tun_flow_uninit(tun);
1826         security_tun_dev_free_security(tun->security);
1827 }
1828
1829 static void tun_setup(struct net_device *dev)
1830 {
1831         struct tun_struct *tun = netdev_priv(dev);
1832
1833         tun->owner = INVALID_UID;
1834         tun->group = INVALID_GID;
1835
1836         dev->ethtool_ops = &tun_ethtool_ops;
1837         dev->needs_free_netdev = true;
1838         dev->priv_destructor = tun_free_netdev;
1839         /* We prefer our own queue length */
1840         dev->tx_queue_len = TUN_READQ_SIZE;
1841 }
1842
1843 /* Trivial set of netlink ops to allow deleting tun or tap
1844  * device with netlink.
1845  */
1846 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
1847                         struct netlink_ext_ack *extack)
1848 {
1849         NL_SET_ERR_MSG(extack,
1850                        "tun/tap creation via rtnetlink is not supported.");
1851         return -EOPNOTSUPP;
1852 }
1853
1854 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1855         .kind           = DRV_NAME,
1856         .priv_size      = sizeof(struct tun_struct),
1857         .setup          = tun_setup,
1858         .validate       = tun_validate,
1859 };
1860
1861 static void tun_sock_write_space(struct sock *sk)
1862 {
1863         struct tun_file *tfile;
1864         wait_queue_head_t *wqueue;
1865
1866         if (!sock_writeable(sk))
1867                 return;
1868
1869         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1870                 return;
1871
1872         wqueue = sk_sleep(sk);
1873         if (wqueue && waitqueue_active(wqueue))
1874                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1875                                                 POLLWRNORM | POLLWRBAND);
1876
1877         tfile = container_of(sk, struct tun_file, sk);
1878         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1879 }
1880
1881 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1882 {
1883         int ret;
1884         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1885         struct tun_struct *tun = __tun_get(tfile);
1886
1887         if (!tun)
1888                 return -EBADFD;
1889
1890         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1891                            m->msg_flags & MSG_DONTWAIT,
1892                            m->msg_flags & MSG_MORE);
1893         tun_put(tun);
1894         return ret;
1895 }
1896
1897 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1898                        int flags)
1899 {
1900         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1901         struct tun_struct *tun = __tun_get(tfile);
1902         struct sk_buff *skb = m->msg_control;
1903         int ret;
1904
1905         if (!tun) {
1906                 ret = -EBADFD;
1907                 goto out_free_skb;
1908         }
1909
1910         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1911                 ret = -EINVAL;
1912                 goto out_put_tun;
1913         }
1914         if (flags & MSG_ERRQUEUE) {
1915                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1916                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1917                 goto out;
1918         }
1919         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, skb);
1920         if (ret > (ssize_t)total_len) {
1921                 m->msg_flags |= MSG_TRUNC;
1922                 ret = flags & MSG_TRUNC ? ret : total_len;
1923         }
1924 out:
1925         tun_put(tun);
1926         return ret;
1927
1928 out_put_tun:
1929         tun_put(tun);
1930 out_free_skb:
1931         if (skb)
1932                 kfree_skb(skb);
1933         return ret;
1934 }
1935
1936 static int tun_peek_len(struct socket *sock)
1937 {
1938         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1939         struct tun_struct *tun;
1940         int ret = 0;
1941
1942         tun = __tun_get(tfile);
1943         if (!tun)
1944                 return 0;
1945
1946         ret = skb_array_peek_len(&tfile->tx_array);
1947         tun_put(tun);
1948
1949         return ret;
1950 }
1951
1952 /* Ops structure to mimic raw sockets with tun */
1953 static const struct proto_ops tun_socket_ops = {
1954         .peek_len = tun_peek_len,
1955         .sendmsg = tun_sendmsg,
1956         .recvmsg = tun_recvmsg,
1957 };
1958
1959 static struct proto tun_proto = {
1960         .name           = "tun",
1961         .owner          = THIS_MODULE,
1962         .obj_size       = sizeof(struct tun_file),
1963 };
1964
1965 static int tun_flags(struct tun_struct *tun)
1966 {
1967         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1968 }
1969
1970 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1971                               char *buf)
1972 {
1973         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1974         return sprintf(buf, "0x%x\n", tun_flags(tun));
1975 }
1976
1977 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1978                               char *buf)
1979 {
1980         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1981         return uid_valid(tun->owner)?
1982                 sprintf(buf, "%u\n",
1983                         from_kuid_munged(current_user_ns(), tun->owner)):
1984                 sprintf(buf, "-1\n");
1985 }
1986
1987 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1988                               char *buf)
1989 {
1990         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1991         return gid_valid(tun->group) ?
1992                 sprintf(buf, "%u\n",
1993                         from_kgid_munged(current_user_ns(), tun->group)):
1994                 sprintf(buf, "-1\n");
1995 }
1996
1997 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1998 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1999 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2000
2001 static struct attribute *tun_dev_attrs[] = {
2002         &dev_attr_tun_flags.attr,
2003         &dev_attr_owner.attr,
2004         &dev_attr_group.attr,
2005         NULL
2006 };
2007
2008 static const struct attribute_group tun_attr_group = {
2009         .attrs = tun_dev_attrs
2010 };
2011
2012 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2013 {
2014         struct tun_struct *tun;
2015         struct tun_file *tfile = file->private_data;
2016         struct net_device *dev;
2017         int err;
2018
2019         if (tfile->detached)
2020                 return -EINVAL;
2021
2022         dev = __dev_get_by_name(net, ifr->ifr_name);
2023         if (dev) {
2024                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2025                         return -EBUSY;
2026                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2027                         tun = netdev_priv(dev);
2028                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2029                         tun = netdev_priv(dev);
2030                 else
2031                         return -EINVAL;
2032
2033                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2034                     !!(tun->flags & IFF_MULTI_QUEUE))
2035                         return -EINVAL;
2036
2037                 if (tun_not_capable(tun))
2038                         return -EPERM;
2039                 err = security_tun_dev_open(tun->security);
2040                 if (err < 0)
2041                         return err;
2042
2043                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER, true);
2044                 if (err < 0)
2045                         return err;
2046
2047                 if (tun->flags & IFF_MULTI_QUEUE &&
2048                     (tun->numqueues + tun->numdisabled > 1)) {
2049                         /* One or more queue has already been attached, no need
2050                          * to initialize the device again.
2051                          */
2052                         return 0;
2053                 }
2054         }
2055         else {
2056                 char *name;
2057                 unsigned long flags = 0;
2058                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2059                              MAX_TAP_QUEUES : 1;
2060
2061                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2062                         return -EPERM;
2063                 err = security_tun_dev_create();
2064                 if (err < 0)
2065                         return err;
2066
2067                 /* Set dev type */
2068                 if (ifr->ifr_flags & IFF_TUN) {
2069                         /* TUN device */
2070                         flags |= IFF_TUN;
2071                         name = "tun%d";
2072                 } else if (ifr->ifr_flags & IFF_TAP) {
2073                         /* TAP device */
2074                         flags |= IFF_TAP;
2075                         name = "tap%d";
2076                 } else
2077                         return -EINVAL;
2078
2079                 if (*ifr->ifr_name)
2080                         name = ifr->ifr_name;
2081
2082                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2083                                        NET_NAME_UNKNOWN, tun_setup, queues,
2084                                        queues);
2085
2086                 if (!dev)
2087                         return -ENOMEM;
2088                 err = dev_get_valid_name(net, dev, name);
2089                 if (err < 0)
2090                         goto err_free_dev;
2091
2092                 dev_net_set(dev, net);
2093                 dev->rtnl_link_ops = &tun_link_ops;
2094                 dev->ifindex = tfile->ifindex;
2095                 dev->sysfs_groups[0] = &tun_attr_group;
2096
2097                 tun = netdev_priv(dev);
2098                 tun->dev = dev;
2099                 tun->flags = flags;
2100                 tun->txflt.count = 0;
2101                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2102
2103                 tun->align = NET_SKB_PAD;
2104                 tun->filter_attached = false;
2105                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2106                 tun->rx_batched = 0;
2107
2108                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2109                 if (!tun->pcpu_stats) {
2110                         err = -ENOMEM;
2111                         goto err_free_dev;
2112                 }
2113
2114                 spin_lock_init(&tun->lock);
2115
2116                 err = security_tun_dev_alloc_security(&tun->security);
2117                 if (err < 0)
2118                         goto err_free_stat;
2119
2120                 tun_net_init(dev);
2121                 tun_flow_init(tun);
2122
2123                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2124                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2125                                    NETIF_F_HW_VLAN_STAG_TX;
2126                 dev->features = dev->hw_features | NETIF_F_LLTX;
2127                 dev->vlan_features = dev->features &
2128                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2129                                        NETIF_F_HW_VLAN_STAG_TX);
2130
2131                 INIT_LIST_HEAD(&tun->disabled);
2132                 err = tun_attach(tun, file, false, false);
2133                 if (err < 0)
2134                         goto err_free_flow;
2135
2136                 err = register_netdevice(tun->dev);
2137                 if (err < 0)
2138                         goto err_detach;
2139                 /* free_netdev() won't check refcnt, to aovid race
2140                  * with dev_put() we need publish tun after registration.
2141                  */
2142                 rcu_assign_pointer(tfile->tun, tun);
2143         }
2144
2145         netif_carrier_on(tun->dev);
2146
2147         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2148
2149         tun->flags = (tun->flags & ~TUN_FEATURES) |
2150                 (ifr->ifr_flags & TUN_FEATURES);
2151
2152         /* Make sure persistent devices do not get stuck in
2153          * xoff state.
2154          */
2155         if (netif_running(tun->dev))
2156                 netif_tx_wake_all_queues(tun->dev);
2157
2158         strcpy(ifr->ifr_name, tun->dev->name);
2159         return 0;
2160
2161 err_detach:
2162         tun_detach_all(dev);
2163         /* register_netdevice() already called tun_free_netdev() */
2164         goto err_free_dev;
2165
2166 err_free_flow:
2167         tun_flow_uninit(tun);
2168         security_tun_dev_free_security(tun->security);
2169 err_free_stat:
2170         free_percpu(tun->pcpu_stats);
2171 err_free_dev:
2172         free_netdev(dev);
2173         return err;
2174 }
2175
2176 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2177                        struct ifreq *ifr)
2178 {
2179         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2180
2181         strcpy(ifr->ifr_name, tun->dev->name);
2182
2183         ifr->ifr_flags = tun_flags(tun);
2184
2185 }
2186
2187 /* This is like a cut-down ethtool ops, except done via tun fd so no
2188  * privs required. */
2189 static int set_offload(struct tun_struct *tun, unsigned long arg)
2190 {
2191         netdev_features_t features = 0;
2192
2193         if (arg & TUN_F_CSUM) {
2194                 features |= NETIF_F_HW_CSUM;
2195                 arg &= ~TUN_F_CSUM;
2196
2197                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2198                         if (arg & TUN_F_TSO_ECN) {
2199                                 features |= NETIF_F_TSO_ECN;
2200                                 arg &= ~TUN_F_TSO_ECN;
2201                         }
2202                         if (arg & TUN_F_TSO4)
2203                                 features |= NETIF_F_TSO;
2204                         if (arg & TUN_F_TSO6)
2205                                 features |= NETIF_F_TSO6;
2206                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2207                 }
2208
2209                 arg &= ~TUN_F_UFO;
2210         }
2211
2212         /* This gives the user a way to test for new features in future by
2213          * trying to set them. */
2214         if (arg)
2215                 return -EINVAL;
2216
2217         tun->set_features = features;
2218         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2219         tun->dev->wanted_features |= features;
2220         netdev_update_features(tun->dev);
2221
2222         return 0;
2223 }
2224
2225 static void tun_detach_filter(struct tun_struct *tun, int n)
2226 {
2227         int i;
2228         struct tun_file *tfile;
2229
2230         for (i = 0; i < n; i++) {
2231                 tfile = rtnl_dereference(tun->tfiles[i]);
2232                 lock_sock(tfile->socket.sk);
2233                 sk_detach_filter(tfile->socket.sk);
2234                 release_sock(tfile->socket.sk);
2235         }
2236
2237         tun->filter_attached = false;
2238 }
2239
2240 static int tun_attach_filter(struct tun_struct *tun)
2241 {
2242         int i, ret = 0;
2243         struct tun_file *tfile;
2244
2245         for (i = 0; i < tun->numqueues; i++) {
2246                 tfile = rtnl_dereference(tun->tfiles[i]);
2247                 lock_sock(tfile->socket.sk);
2248                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2249                 release_sock(tfile->socket.sk);
2250                 if (ret) {
2251                         tun_detach_filter(tun, i);
2252                         return ret;
2253                 }
2254         }
2255
2256         tun->filter_attached = true;
2257         return ret;
2258 }
2259
2260 static void tun_set_sndbuf(struct tun_struct *tun)
2261 {
2262         struct tun_file *tfile;
2263         int i;
2264
2265         for (i = 0; i < tun->numqueues; i++) {
2266                 tfile = rtnl_dereference(tun->tfiles[i]);
2267                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2268         }
2269 }
2270
2271 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2272 {
2273         struct tun_file *tfile = file->private_data;
2274         struct tun_struct *tun;
2275         int ret = 0;
2276
2277         rtnl_lock();
2278
2279         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2280                 tun = tfile->detached;
2281                 if (!tun) {
2282                         ret = -EINVAL;
2283                         goto unlock;
2284                 }
2285                 ret = security_tun_dev_attach_queue(tun->security);
2286                 if (ret < 0)
2287                         goto unlock;
2288                 ret = tun_attach(tun, file, false, true);
2289         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2290                 tun = rtnl_dereference(tfile->tun);
2291                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2292                         ret = -EINVAL;
2293                 else
2294                         __tun_detach(tfile, false);
2295         } else
2296                 ret = -EINVAL;
2297
2298 unlock:
2299         rtnl_unlock();
2300         return ret;
2301 }
2302
2303 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
2304 static unsigned char tun_get_addr_len(unsigned short type)
2305 {
2306         switch (type) {
2307         case ARPHRD_IP6GRE:
2308         case ARPHRD_TUNNEL6:
2309                 return sizeof(struct in6_addr);
2310         case ARPHRD_IPGRE:
2311         case ARPHRD_TUNNEL:
2312         case ARPHRD_SIT:
2313                 return 4;
2314         case ARPHRD_ETHER:
2315                 return ETH_ALEN;
2316         case ARPHRD_IEEE802154:
2317         case ARPHRD_IEEE802154_MONITOR:
2318                 return IEEE802154_EXTENDED_ADDR_LEN;
2319         case ARPHRD_PHONET_PIPE:
2320         case ARPHRD_PPP:
2321         case ARPHRD_NONE:
2322                 return 0;
2323         case ARPHRD_6LOWPAN:
2324                 return EUI64_ADDR_LEN;
2325         case ARPHRD_FDDI:
2326                 return FDDI_K_ALEN;
2327         case ARPHRD_HIPPI:
2328                 return HIPPI_ALEN;
2329         case ARPHRD_IEEE802:
2330                 return FC_ALEN;
2331         case ARPHRD_ROSE:
2332                 return ROSE_ADDR_LEN;
2333         case ARPHRD_NETROM:
2334                 return AX25_ADDR_LEN;
2335         case ARPHRD_LOCALTLK:
2336                 return LTALK_ALEN;
2337         default:
2338                 return 0;
2339         }
2340 }
2341
2342 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2343                             unsigned long arg, int ifreq_len)
2344 {
2345         struct tun_file *tfile = file->private_data;
2346         struct tun_struct *tun;
2347         void __user* argp = (void __user*)arg;
2348         struct ifreq ifr;
2349         kuid_t owner;
2350         kgid_t group;
2351         int sndbuf;
2352         int vnet_hdr_sz;
2353         unsigned int ifindex;
2354         int le;
2355         int ret;
2356
2357         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2358                 if (copy_from_user(&ifr, argp, ifreq_len))
2359                         return -EFAULT;
2360         } else {
2361                 memset(&ifr, 0, sizeof(ifr));
2362         }
2363         if (cmd == TUNGETFEATURES) {
2364                 /* Currently this just means: "what IFF flags are valid?".
2365                  * This is needed because we never checked for invalid flags on
2366                  * TUNSETIFF.
2367                  */
2368                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2369                                 (unsigned int __user*)argp);
2370         } else if (cmd == TUNSETQUEUE)
2371                 return tun_set_queue(file, &ifr);
2372
2373         ret = 0;
2374         rtnl_lock();
2375
2376         tun = __tun_get(tfile);
2377         if (cmd == TUNSETIFF) {
2378                 ret = -EEXIST;
2379                 if (tun)
2380                         goto unlock;
2381
2382                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2383
2384                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2385
2386                 if (ret)
2387                         goto unlock;
2388
2389                 if (copy_to_user(argp, &ifr, ifreq_len))
2390                         ret = -EFAULT;
2391                 goto unlock;
2392         }
2393         if (cmd == TUNSETIFINDEX) {
2394                 ret = -EPERM;
2395                 if (tun)
2396                         goto unlock;
2397
2398                 ret = -EFAULT;
2399                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2400                         goto unlock;
2401
2402                 ret = 0;
2403                 tfile->ifindex = ifindex;
2404                 goto unlock;
2405         }
2406
2407         ret = -EBADFD;
2408         if (!tun)
2409                 goto unlock;
2410
2411         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2412
2413         ret = 0;
2414         switch (cmd) {
2415         case TUNGETIFF:
2416                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2417
2418                 if (tfile->detached)
2419                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2420                 if (!tfile->socket.sk->sk_filter)
2421                         ifr.ifr_flags |= IFF_NOFILTER;
2422
2423                 if (copy_to_user(argp, &ifr, ifreq_len))
2424                         ret = -EFAULT;
2425                 break;
2426
2427         case TUNSETNOCSUM:
2428                 /* Disable/Enable checksum */
2429
2430                 /* [unimplemented] */
2431                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2432                           arg ? "disabled" : "enabled");
2433                 break;
2434
2435         case TUNSETPERSIST:
2436                 /* Disable/Enable persist mode. Keep an extra reference to the
2437                  * module to prevent the module being unprobed.
2438                  */
2439                 if (arg && !(tun->flags & IFF_PERSIST)) {
2440                         tun->flags |= IFF_PERSIST;
2441                         __module_get(THIS_MODULE);
2442                 }
2443                 if (!arg && (tun->flags & IFF_PERSIST)) {
2444                         tun->flags &= ~IFF_PERSIST;
2445                         module_put(THIS_MODULE);
2446                 }
2447
2448                 tun_debug(KERN_INFO, tun, "persist %s\n",
2449                           arg ? "enabled" : "disabled");
2450                 break;
2451
2452         case TUNSETOWNER:
2453                 /* Set owner of the device */
2454                 owner = make_kuid(current_user_ns(), arg);
2455                 if (!uid_valid(owner)) {
2456                         ret = -EINVAL;
2457                         break;
2458                 }
2459                 tun->owner = owner;
2460                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2461                           from_kuid(&init_user_ns, tun->owner));
2462                 break;
2463
2464         case TUNSETGROUP:
2465                 /* Set group of the device */
2466                 group = make_kgid(current_user_ns(), arg);
2467                 if (!gid_valid(group)) {
2468                         ret = -EINVAL;
2469                         break;
2470                 }
2471                 tun->group = group;
2472                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2473                           from_kgid(&init_user_ns, tun->group));
2474                 break;
2475
2476         case TUNSETLINK:
2477                 /* Only allow setting the type when the interface is down */
2478                 if (tun->dev->flags & IFF_UP) {
2479                         tun_debug(KERN_INFO, tun,
2480                                   "Linktype set failed because interface is up\n");
2481                         ret = -EBUSY;
2482                 } else {
2483                         tun->dev->type = (int) arg;
2484                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
2485                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2486                                   tun->dev->type);
2487                         ret = 0;
2488                 }
2489                 break;
2490
2491 #ifdef TUN_DEBUG
2492         case TUNSETDEBUG:
2493                 tun->debug = arg;
2494                 break;
2495 #endif
2496         case TUNSETOFFLOAD:
2497                 ret = set_offload(tun, arg);
2498                 break;
2499
2500         case TUNSETTXFILTER:
2501                 /* Can be set only for TAPs */
2502                 ret = -EINVAL;
2503                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2504                         break;
2505                 ret = update_filter(&tun->txflt, (void __user *)arg);
2506                 break;
2507
2508         case SIOCGIFHWADDR:
2509                 /* Get hw address */
2510                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2511                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2512                 if (copy_to_user(argp, &ifr, ifreq_len))
2513                         ret = -EFAULT;
2514                 break;
2515
2516         case SIOCSIFHWADDR:
2517                 /* Set hw address */
2518                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2519                           ifr.ifr_hwaddr.sa_data);
2520
2521                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2522                 break;
2523
2524         case TUNGETSNDBUF:
2525                 sndbuf = tfile->socket.sk->sk_sndbuf;
2526                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2527                         ret = -EFAULT;
2528                 break;
2529
2530         case TUNSETSNDBUF:
2531                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2532                         ret = -EFAULT;
2533                         break;
2534                 }
2535                 if (sndbuf <= 0) {
2536                         ret = -EINVAL;
2537                         break;
2538                 }
2539
2540                 tun->sndbuf = sndbuf;
2541                 tun_set_sndbuf(tun);
2542                 break;
2543
2544         case TUNGETVNETHDRSZ:
2545                 vnet_hdr_sz = tun->vnet_hdr_sz;
2546                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2547                         ret = -EFAULT;
2548                 break;
2549
2550         case TUNSETVNETHDRSZ:
2551                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2552                         ret = -EFAULT;
2553                         break;
2554                 }
2555                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2556                         ret = -EINVAL;
2557                         break;
2558                 }
2559
2560                 tun->vnet_hdr_sz = vnet_hdr_sz;
2561                 break;
2562
2563         case TUNGETVNETLE:
2564                 le = !!(tun->flags & TUN_VNET_LE);
2565                 if (put_user(le, (int __user *)argp))
2566                         ret = -EFAULT;
2567                 break;
2568
2569         case TUNSETVNETLE:
2570                 if (get_user(le, (int __user *)argp)) {
2571                         ret = -EFAULT;
2572                         break;
2573                 }
2574                 if (le)
2575                         tun->flags |= TUN_VNET_LE;
2576                 else
2577                         tun->flags &= ~TUN_VNET_LE;
2578                 break;
2579
2580         case TUNGETVNETBE:
2581                 ret = tun_get_vnet_be(tun, argp);
2582                 break;
2583
2584         case TUNSETVNETBE:
2585                 ret = tun_set_vnet_be(tun, argp);
2586                 break;
2587
2588         case TUNATTACHFILTER:
2589                 /* Can be set only for TAPs */
2590                 ret = -EINVAL;
2591                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2592                         break;
2593                 ret = -EFAULT;
2594                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2595                         break;
2596
2597                 ret = tun_attach_filter(tun);
2598                 break;
2599
2600         case TUNDETACHFILTER:
2601                 /* Can be set only for TAPs */
2602                 ret = -EINVAL;
2603                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2604                         break;
2605                 ret = 0;
2606                 tun_detach_filter(tun, tun->numqueues);
2607                 break;
2608
2609         case TUNGETFILTER:
2610                 ret = -EINVAL;
2611                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2612                         break;
2613                 ret = -EFAULT;
2614                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2615                         break;
2616                 ret = 0;
2617                 break;
2618
2619         default:
2620                 ret = -EINVAL;
2621                 break;
2622         }
2623
2624 unlock:
2625         rtnl_unlock();
2626         if (tun)
2627                 tun_put(tun);
2628         return ret;
2629 }
2630
2631 static long tun_chr_ioctl(struct file *file,
2632                           unsigned int cmd, unsigned long arg)
2633 {
2634         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2635 }
2636
2637 #ifdef CONFIG_COMPAT
2638 static long tun_chr_compat_ioctl(struct file *file,
2639                          unsigned int cmd, unsigned long arg)
2640 {
2641         switch (cmd) {
2642         case TUNSETIFF:
2643         case TUNGETIFF:
2644         case TUNSETTXFILTER:
2645         case TUNGETSNDBUF:
2646         case TUNSETSNDBUF:
2647         case SIOCGIFHWADDR:
2648         case SIOCSIFHWADDR:
2649                 arg = (unsigned long)compat_ptr(arg);
2650                 break;
2651         default:
2652                 arg = (compat_ulong_t)arg;
2653                 break;
2654         }
2655
2656         /*
2657          * compat_ifreq is shorter than ifreq, so we must not access beyond
2658          * the end of that structure. All fields that are used in this
2659          * driver are compatible though, we don't need to convert the
2660          * contents.
2661          */
2662         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2663 }
2664 #endif /* CONFIG_COMPAT */
2665
2666 static int tun_chr_fasync(int fd, struct file *file, int on)
2667 {
2668         struct tun_file *tfile = file->private_data;
2669         int ret;
2670
2671         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2672                 goto out;
2673
2674         if (on) {
2675                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2676                 tfile->flags |= TUN_FASYNC;
2677         } else
2678                 tfile->flags &= ~TUN_FASYNC;
2679         ret = 0;
2680 out:
2681         return ret;
2682 }
2683
2684 static int tun_chr_open(struct inode *inode, struct file * file)
2685 {
2686         struct net *net = current->nsproxy->net_ns;
2687         struct tun_file *tfile;
2688
2689         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2690
2691         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2692                                             &tun_proto, 0);
2693         if (!tfile)
2694                 return -ENOMEM;
2695         if (skb_array_init(&tfile->tx_array, 0, GFP_KERNEL)) {
2696                 sk_free(&tfile->sk);
2697                 return -ENOMEM;
2698         }
2699
2700         RCU_INIT_POINTER(tfile->tun, NULL);
2701         tfile->flags = 0;
2702         tfile->ifindex = 0;
2703
2704         init_waitqueue_head(&tfile->wq.wait);
2705         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2706
2707         tfile->socket.file = file;
2708         tfile->socket.ops = &tun_socket_ops;
2709
2710         sock_init_data(&tfile->socket, &tfile->sk);
2711
2712         tfile->sk.sk_write_space = tun_sock_write_space;
2713         tfile->sk.sk_sndbuf = INT_MAX;
2714
2715         file->private_data = tfile;
2716         INIT_LIST_HEAD(&tfile->next);
2717
2718         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2719
2720         return 0;
2721 }
2722
2723 static int tun_chr_close(struct inode *inode, struct file *file)
2724 {
2725         struct tun_file *tfile = file->private_data;
2726
2727         tun_detach(tfile, true);
2728
2729         return 0;
2730 }
2731
2732 #ifdef CONFIG_PROC_FS
2733 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2734 {
2735         struct tun_struct *tun;
2736         struct ifreq ifr;
2737
2738         memset(&ifr, 0, sizeof(ifr));
2739
2740         rtnl_lock();
2741         tun = tun_get(f);
2742         if (tun)
2743                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2744         rtnl_unlock();
2745
2746         if (tun)
2747                 tun_put(tun);
2748
2749         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2750 }
2751 #endif
2752
2753 static const struct file_operations tun_fops = {
2754         .owner  = THIS_MODULE,
2755         .llseek = no_llseek,
2756         .read_iter  = tun_chr_read_iter,
2757         .write_iter = tun_chr_write_iter,
2758         .poll   = tun_chr_poll,
2759         .unlocked_ioctl = tun_chr_ioctl,
2760 #ifdef CONFIG_COMPAT
2761         .compat_ioctl = tun_chr_compat_ioctl,
2762 #endif
2763         .open   = tun_chr_open,
2764         .release = tun_chr_close,
2765         .fasync = tun_chr_fasync,
2766 #ifdef CONFIG_PROC_FS
2767         .show_fdinfo = tun_chr_show_fdinfo,
2768 #endif
2769 };
2770
2771 static struct miscdevice tun_miscdev = {
2772         .minor = TUN_MINOR,
2773         .name = "tun",
2774         .nodename = "net/tun",
2775         .fops = &tun_fops,
2776 };
2777
2778 /* ethtool interface */
2779
2780 static int tun_get_link_ksettings(struct net_device *dev,
2781                                   struct ethtool_link_ksettings *cmd)
2782 {
2783         ethtool_link_ksettings_zero_link_mode(cmd, supported);
2784         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
2785         cmd->base.speed         = SPEED_10;
2786         cmd->base.duplex        = DUPLEX_FULL;
2787         cmd->base.port          = PORT_TP;
2788         cmd->base.phy_address   = 0;
2789         cmd->base.autoneg       = AUTONEG_DISABLE;
2790         return 0;
2791 }
2792
2793 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2794 {
2795         struct tun_struct *tun = netdev_priv(dev);
2796
2797         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2798         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2799
2800         switch (tun->flags & TUN_TYPE_MASK) {
2801         case IFF_TUN:
2802                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2803                 break;
2804         case IFF_TAP:
2805                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2806                 break;
2807         }
2808 }
2809
2810 static u32 tun_get_msglevel(struct net_device *dev)
2811 {
2812 #ifdef TUN_DEBUG
2813         struct tun_struct *tun = netdev_priv(dev);
2814         return tun->debug;
2815 #else
2816         return -EOPNOTSUPP;
2817 #endif
2818 }
2819
2820 static void tun_set_msglevel(struct net_device *dev, u32 value)
2821 {
2822 #ifdef TUN_DEBUG
2823         struct tun_struct *tun = netdev_priv(dev);
2824         tun->debug = value;
2825 #endif
2826 }
2827
2828 static int tun_get_coalesce(struct net_device *dev,
2829                             struct ethtool_coalesce *ec)
2830 {
2831         struct tun_struct *tun = netdev_priv(dev);
2832
2833         ec->rx_max_coalesced_frames = tun->rx_batched;
2834
2835         return 0;
2836 }
2837
2838 static int tun_set_coalesce(struct net_device *dev,
2839                             struct ethtool_coalesce *ec)
2840 {
2841         struct tun_struct *tun = netdev_priv(dev);
2842
2843         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
2844                 tun->rx_batched = NAPI_POLL_WEIGHT;
2845         else
2846                 tun->rx_batched = ec->rx_max_coalesced_frames;
2847
2848         return 0;
2849 }
2850
2851 static const struct ethtool_ops tun_ethtool_ops = {
2852         .get_drvinfo    = tun_get_drvinfo,
2853         .get_msglevel   = tun_get_msglevel,
2854         .set_msglevel   = tun_set_msglevel,
2855         .get_link       = ethtool_op_get_link,
2856         .get_ts_info    = ethtool_op_get_ts_info,
2857         .get_coalesce   = tun_get_coalesce,
2858         .set_coalesce   = tun_set_coalesce,
2859         .get_link_ksettings = tun_get_link_ksettings,
2860 };
2861
2862 static int tun_queue_resize(struct tun_struct *tun)
2863 {
2864         struct net_device *dev = tun->dev;
2865         struct tun_file *tfile;
2866         struct skb_array **arrays;
2867         int n = tun->numqueues + tun->numdisabled;
2868         int ret, i;
2869
2870         arrays = kmalloc_array(n, sizeof(*arrays), GFP_KERNEL);
2871         if (!arrays)
2872                 return -ENOMEM;
2873
2874         for (i = 0; i < tun->numqueues; i++) {
2875                 tfile = rtnl_dereference(tun->tfiles[i]);
2876                 arrays[i] = &tfile->tx_array;
2877         }
2878         list_for_each_entry(tfile, &tun->disabled, next)
2879                 arrays[i++] = &tfile->tx_array;
2880
2881         ret = skb_array_resize_multiple(arrays, n,
2882                                         dev->tx_queue_len, GFP_KERNEL);
2883
2884         kfree(arrays);
2885         return ret;
2886 }
2887
2888 static int tun_device_event(struct notifier_block *unused,
2889                             unsigned long event, void *ptr)
2890 {
2891         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2892         struct tun_struct *tun = netdev_priv(dev);
2893         int i;
2894
2895         if (dev->rtnl_link_ops != &tun_link_ops)
2896                 return NOTIFY_DONE;
2897
2898         switch (event) {
2899         case NETDEV_CHANGE_TX_QUEUE_LEN:
2900                 if (tun_queue_resize(tun))
2901                         return NOTIFY_BAD;
2902                 break;
2903         case NETDEV_UP:
2904                 for (i = 0; i < tun->numqueues; i++) {
2905                         struct tun_file *tfile;
2906
2907                         tfile = rtnl_dereference(tun->tfiles[i]);
2908                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
2909                 }
2910                 break;
2911         default:
2912                 break;
2913         }
2914
2915         return NOTIFY_DONE;
2916 }
2917
2918 static struct notifier_block tun_notifier_block __read_mostly = {
2919         .notifier_call  = tun_device_event,
2920 };
2921
2922 static int __init tun_init(void)
2923 {
2924         int ret = 0;
2925
2926         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2927
2928         ret = rtnl_link_register(&tun_link_ops);
2929         if (ret) {
2930                 pr_err("Can't register link_ops\n");
2931                 goto err_linkops;
2932         }
2933
2934         ret = misc_register(&tun_miscdev);
2935         if (ret) {
2936                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2937                 goto err_misc;
2938         }
2939
2940         ret = register_netdevice_notifier(&tun_notifier_block);
2941         if (ret) {
2942                 pr_err("Can't register netdevice notifier\n");
2943                 goto err_notifier;
2944         }
2945
2946         return  0;
2947
2948 err_notifier:
2949         misc_deregister(&tun_miscdev);
2950 err_misc:
2951         rtnl_link_unregister(&tun_link_ops);
2952 err_linkops:
2953         return ret;
2954 }
2955
2956 static void tun_cleanup(void)
2957 {
2958         misc_deregister(&tun_miscdev);
2959         rtnl_link_unregister(&tun_link_ops);
2960         unregister_netdevice_notifier(&tun_notifier_block);
2961 }
2962
2963 /* Get an underlying socket object from tun file.  Returns error unless file is
2964  * attached to a device.  The returned object works like a packet socket, it
2965  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2966  * holding a reference to the file for as long as the socket is in use. */
2967 struct socket *tun_get_socket(struct file *file)
2968 {
2969         struct tun_file *tfile;
2970         if (file->f_op != &tun_fops)
2971                 return ERR_PTR(-EINVAL);
2972         tfile = file->private_data;
2973         if (!tfile)
2974                 return ERR_PTR(-EBADFD);
2975         return &tfile->socket;
2976 }
2977 EXPORT_SYMBOL_GPL(tun_get_socket);
2978
2979 struct skb_array *tun_get_skb_array(struct file *file)
2980 {
2981         struct tun_file *tfile;
2982
2983         if (file->f_op != &tun_fops)
2984                 return ERR_PTR(-EINVAL);
2985         tfile = file->private_data;
2986         if (!tfile)
2987                 return ERR_PTR(-EBADFD);
2988         return &tfile->tx_array;
2989 }
2990 EXPORT_SYMBOL_GPL(tun_get_skb_array);
2991
2992 module_init(tun_init);
2993 module_exit(tun_cleanup);
2994 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2995 MODULE_AUTHOR(DRV_COPYRIGHT);
2996 MODULE_LICENSE("GPL");
2997 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2998 MODULE_ALIAS("devname:net/tun");