drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <[email protected]>
   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 <[email protected]> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <[email protected]>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <[email protected]>  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 <[email protected]>
  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 <[email protected]>"
  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/mutex.h>
  79
  80 #include <linux/uaccess.h>
  81
  82 /* Uncomment to enable debugging */
  83 /* #define TUN_DEBUG 1 */
  84
  85 #ifdef TUN_DEBUG
  86 static int debug;
  87
  88 #define tun_debug(level, tun, fmt, args...)                     \
  89 do {                                                            \
  90         if (tun->debug)                                         \
  91                 netdev_printk(level, tun->dev, fmt, ##args);    \
  92 } while (0)
  93 #define DBG1(level, fmt, args...)                               \
  94 do {                                                            \
  95         if (debug == 2)                                         \
  96                 printk(level fmt, ##args);                      \
  97 } while (0)
  98 #else
  99 #define tun_debug(level, tun, fmt, args...)                     \
 100 do {                                                            \
 101         if (0)                                                  \
 102                 netdev_printk(level, tun->dev, fmt, ##args);    \
 103 } while (0)
 104 #define DBG1(level, fmt, args...)                               \
 105 do {                                                            \
 106         if (0)                                                  \
 107                 printk(level fmt, ##args);                      \
 108 } while (0)
 109 #endif
 110
 111 #define TUN_HEADROOM 256
 112 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 113
 114 /* TUN device flags */
 115
 116 /* IFF_ATTACH_QUEUE is never stored in device flags,
 117  * overload it to mean fasync when stored there.
 118  */
 119 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 120 /* High bits in flags field are unused. */
 121 #define TUN_VNET_LE     0x80000000
 122 #define TUN_VNET_BE     0x40000000
 123
 124 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 125                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 126
 127 #define GOODCOPY_LEN 128
 128
 129 #define FLT_EXACT_COUNT 8
 130 struct tap_filter {
 131         unsigned int    count;    /* Number of addrs. Zero means disabled */
 132         u32             mask[2];  /* Mask of the hashed addrs */
 133         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 134 };
 135
 136 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 137  * to max number of VCPUs in guest. */
 138 #define MAX_TAP_QUEUES 256
 139 #define MAX_TAP_FLOWS  4096
 140
 141 #define TUN_FLOW_EXPIRE (3 * HZ)
 142
 143 struct tun_pcpu_stats {
 144         u64 rx_packets;
 145         u64 rx_bytes;
 146         u64 tx_packets;
 147         u64 tx_bytes;
 148         struct u64_stats_sync syncp;
 149         u32 rx_dropped;
 150         u32 tx_dropped;
 151         u32 rx_frame_errors;
 152 };
 153
 154 /* A tun_file connects an open character device to a tuntap netdevice. It
 155  * also contains all socket related structures (except sock_fprog and tap_filter)
 156  * to serve as one transmit queue for tuntap device. The sock_fprog and
 157  * tap_filter were kept in tun_struct since they were used for filtering for the
 158  * netdevice not for a specific queue (at least I didn't see the requirement for
 159  * this).
 160  *
 161  * RCU usage:
 162  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 163  * other can only be read while rcu_read_lock or rtnl_lock is held.
 164  */
 165 struct tun_file {
 166         struct sock sk;
 167         struct socket socket;
 168         struct socket_wq wq;
 169         struct tun_struct __rcu *tun;
 170         struct fasync_struct *fasync;
 171         /* only used for fasnyc */
 172         unsigned int flags;
 173         union {
 174                 u16 queue_index;
 175                 unsigned int ifindex;
 176         };
 177         struct napi_struct napi;
 178         bool napi_enabled;
 179         struct mutex napi_mutex;        /* Protects access to the above napi */
 180         struct list_head next;
 181         struct tun_struct *detached;
 182         struct ptr_ring tx_ring;
 183         struct xdp_rxq_info xdp_rxq;
 184 };
 185
 186 struct tun_flow_entry {
 187         struct hlist_node hash_link;
 188         struct rcu_head rcu;
 189         struct tun_struct *tun;
 190
 191         u32 rxhash;
 192         u32 rps_rxhash;
 193         int queue_index;
 194         unsigned long updated;
 195 };
 196
 197 #define TUN_NUM_FLOW_ENTRIES 1024
 198
 199 struct tun_prog {
 200         struct rcu_head rcu;
 201         struct bpf_prog *prog;
 202 };
 203
 204 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 205  * device, socket filter, sndbuf and vnet header size were restore when the
 206  * file were attached to a persist device.
 207  */
 208 struct tun_struct {
 209         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 210         unsigned int            numqueues;
 211         unsigned int            flags;
 212         kuid_t                  owner;
 213         kgid_t                  group;
 214
 215         struct net_device       *dev;
 216         netdev_features_t       set_features;
 217 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 218                           NETIF_F_TSO6)
 219
 220         int                     align;
 221         int                     vnet_hdr_sz;
 222         int                     sndbuf;
 223         struct tap_filter       txflt;
 224         struct sock_fprog       fprog;
 225         /* protected by rtnl lock */
 226         bool                    filter_attached;
 227 #ifdef TUN_DEBUG
 228         int debug;
 229 #endif
 230         spinlock_t lock;
 231         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 232         struct timer_list flow_gc_timer;
 233         unsigned long ageing_time;
 234         unsigned int numdisabled;
 235         struct list_head disabled;
 236         void *security;
 237         u32 flow_count;
 238         u32 rx_batched;
 239         struct tun_pcpu_stats __percpu *pcpu_stats;
 240         struct bpf_prog __rcu *xdp_prog;
 241         struct tun_prog __rcu *steering_prog;
 242         struct tun_prog __rcu *filter_prog;
 243 };
 244
 245 struct veth {
 246         __be16 h_vlan_proto;
 247         __be16 h_vlan_TCI;
 248 };
 249
 250 bool tun_is_xdp_buff(void *ptr)
 251 {
 252         return (unsigned long)ptr & TUN_XDP_FLAG;
 253 }
 254 EXPORT_SYMBOL(tun_is_xdp_buff);
 255
 256 void *tun_xdp_to_ptr(void *ptr)
 257 {
 258         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 259 }
 260 EXPORT_SYMBOL(tun_xdp_to_ptr);
 261
 262 void *tun_ptr_to_xdp(void *ptr)
 263 {
 264         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 265 }
 266 EXPORT_SYMBOL(tun_ptr_to_xdp);
 267
 268 static int tun_napi_receive(struct napi_struct *napi, int budget)
 269 {
 270         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 271         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 272         struct sk_buff_head process_queue;
 273         struct sk_buff *skb;
 274         int received = 0;
 275
 276         __skb_queue_head_init(&process_queue);
 277
 278         spin_lock(&queue->lock);
 279         skb_queue_splice_tail_init(queue, &process_queue);
 280         spin_unlock(&queue->lock);
 281
 282         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 283                 napi_gro_receive(napi, skb);
 284                 ++received;
 285         }
 286
 287         if (!skb_queue_empty(&process_queue)) {
 288                 spin_lock(&queue->lock);
 289                 skb_queue_splice(&process_queue, queue);
 290                 spin_unlock(&queue->lock);
 291         }
 292
 293         return received;
 294 }
 295
 296 static int tun_napi_poll(struct napi_struct *napi, int budget)
 297 {
 298         unsigned int received;
 299
 300         received = tun_napi_receive(napi, budget);
 301
 302         if (received < budget)
 303                 napi_complete_done(napi, received);
 304
 305         return received;
 306 }
 307
 308 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 309                           bool napi_en)
 310 {
 311         tfile->napi_enabled = napi_en;
 312         if (napi_en) {
 313                 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 314                                NAPI_POLL_WEIGHT);
 315                 napi_enable(&tfile->napi);
 316                 mutex_init(&tfile->napi_mutex);
 317         }
 318 }
 319
 320 static void tun_napi_disable(struct tun_struct *tun, struct tun_file *tfile)
 321 {
 322         if (tfile->napi_enabled)
 323                 napi_disable(&tfile->napi);
 324 }
 325
 326 static void tun_napi_del(struct tun_struct *tun, struct tun_file *tfile)
 327 {
 328         if (tfile->napi_enabled)
 329                 netif_napi_del(&tfile->napi);
 330 }
 331
 332 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
 333 {
 334         return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
 335 }
 336
 337 #ifdef CONFIG_TUN_VNET_CROSS_LE
 338 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 339 {
 340         return tun->flags & TUN_VNET_BE ? false :
 341                 virtio_legacy_is_little_endian();
 342 }
 343
 344 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 345 {
 346         int be = !!(tun->flags & TUN_VNET_BE);
 347
 348         if (put_user(be, argp))
 349                 return -EFAULT;
 350
 351         return 0;
 352 }
 353
 354 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 355 {
 356         int be;
 357
 358         if (get_user(be, argp))
 359                 return -EFAULT;
 360
 361         if (be)
 362                 tun->flags |= TUN_VNET_BE;
 363         else
 364                 tun->flags &= ~TUN_VNET_BE;
 365
 366         return 0;
 367 }
 368 #else
 369 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 370 {
 371         return virtio_legacy_is_little_endian();
 372 }
 373
 374 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 375 {
 376         return -EINVAL;
 377 }
 378
 379 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 380 {
 381         return -EINVAL;
 382 }
 383 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 384
 385 static inline bool tun_is_little_endian(struct tun_struct *tun)
 386 {
 387         return tun->flags & TUN_VNET_LE ||
 388                 tun_legacy_is_little_endian(tun);
 389 }
 390
 391 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 392 {
 393         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 394 }
 395
 396 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 397 {
 398         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 399 }
 400
 401 static inline u32 tun_hashfn(u32 rxhash)
 402 {
 403         return rxhash & 0x3ff;
 404 }
 405
 406 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 407 {
 408         struct tun_flow_entry *e;
 409
 410         hlist_for_each_entry_rcu(e, head, hash_link) {
 411                 if (e->rxhash == rxhash)
 412                         return e;
 413         }
 414         return NULL;
 415 }
 416
 417 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 418                                               struct hlist_head *head,
 419                                               u32 rxhash, u16 queue_index)
 420 {
 421         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 422
 423         if (e) {
 424                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 425                           rxhash, queue_index);
 426                 e->updated = jiffies;
 427                 e->rxhash = rxhash;
 428                 e->rps_rxhash = 0;
 429                 e->queue_index = queue_index;
 430                 e->tun = tun;
 431                 hlist_add_head_rcu(&e->hash_link, head);
 432                 ++tun->flow_count;
 433         }
 434         return e;
 435 }
 436
 437 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 438 {
 439         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 440                   e->rxhash, e->queue_index);
 441         hlist_del_rcu(&e->hash_link);
 442         kfree_rcu(e, rcu);
 443         --tun->flow_count;
 444 }
 445
 446 static void tun_flow_flush(struct tun_struct *tun)
 447 {
 448         int i;
 449
 450         spin_lock_bh(&tun->lock);
 451         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 452                 struct tun_flow_entry *e;
 453                 struct hlist_node *n;
 454
 455                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 456                         tun_flow_delete(tun, e);
 457         }
 458         spin_unlock_bh(&tun->lock);
 459 }
 460
 461 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 462 {
 463         int i;
 464
 465         spin_lock_bh(&tun->lock);
 466         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 467                 struct tun_flow_entry *e;
 468                 struct hlist_node *n;
 469
 470                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 471                         if (e->queue_index == queue_index)
 472                                 tun_flow_delete(tun, e);
 473                 }
 474         }
 475         spin_unlock_bh(&tun->lock);
 476 }
 477
 478 static void tun_flow_cleanup(struct timer_list *t)
 479 {
 480         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 481         unsigned long delay = tun->ageing_time;
 482         unsigned long next_timer = jiffies + delay;
 483         unsigned long count = 0;
 484         int i;
 485
 486         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 487
 488         spin_lock(&tun->lock);
 489         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 490                 struct tun_flow_entry *e;
 491                 struct hlist_node *n;
 492
 493                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 494                         unsigned long this_timer;
 495
 496                         this_timer = e->updated + delay;
 497                         if (time_before_eq(this_timer, jiffies)) {
 498                                 tun_flow_delete(tun, e);
 499                                 continue;
 500                         }
 501                         count++;
 502                         if (time_before(this_timer, next_timer))
 503                                 next_timer = this_timer;
 504                 }
 505         }
 506
 507         if (count)
 508                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 509         spin_unlock(&tun->lock);
 510 }
 511
 512 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 513                             struct tun_file *tfile)
 514 {
 515         struct hlist_head *head;
 516         struct tun_flow_entry *e;
 517         unsigned long delay = tun->ageing_time;
 518         u16 queue_index = tfile->queue_index;
 519
 520         if (!rxhash)
 521                 return;
 522         else
 523                 head = &tun->flows[tun_hashfn(rxhash)];
 524
 525         rcu_read_lock();
 526
 527         /* We may get a very small possibility of OOO during switching, not
 528          * worth to optimize.*/
 529         if (tun->numqueues == 1 || tfile->detached)
 530                 goto unlock;
 531
 532         e = tun_flow_find(head, rxhash);
 533         if (likely(e)) {
 534                 /* TODO: keep queueing to old queue until it's empty? */
 535                 e->queue_index = queue_index;
 536                 e->updated = jiffies;
 537                 sock_rps_record_flow_hash(e->rps_rxhash);
 538         } else {
 539                 spin_lock_bh(&tun->lock);
 540                 if (!tun_flow_find(head, rxhash) &&
 541                     tun->flow_count < MAX_TAP_FLOWS)
 542                         tun_flow_create(tun, head, rxhash, queue_index);
 543
 544                 if (!timer_pending(&tun->flow_gc_timer))
 545                         mod_timer(&tun->flow_gc_timer,
 546                                   round_jiffies_up(jiffies + delay));
 547                 spin_unlock_bh(&tun->lock);
 548         }
 549
 550 unlock:
 551         rcu_read_unlock();
 552 }
 553
 554 /**
 555  * Save the hash received in the stack receive path and update the
 556  * flow_hash table accordingly.
 557  */
 558 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 559 {
 560         if (unlikely(e->rps_rxhash != hash))
 561                 e->rps_rxhash = hash;
 562 }
 563
 564 /* We try to identify a flow through its rxhash first. The reason that
 565  * we do not check rxq no. is because some cards(e.g 82599), chooses
 566  * the rxq based on the txq where the last packet of the flow comes. As
 567  * the userspace application move between processors, we may get a
 568  * different rxq no. here. If we could not get rxhash, then we would
 569  * hope the rxq no. may help here.
 570  */
 571 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 572 {
 573         struct tun_flow_entry *e;
 574         u32 txq = 0;
 575         u32 numqueues = 0;
 576
 577         numqueues = READ_ONCE(tun->numqueues);
 578
 579         txq = __skb_get_hash_symmetric(skb);
 580         if (txq) {
 581                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 582                 if (e) {
 583                         tun_flow_save_rps_rxhash(e, txq);
 584                         txq = e->queue_index;
 585                 } else
 586                         /* use multiply and shift instead of expensive divide */
 587                         txq = ((u64)txq * numqueues) >> 32;
 588         } else if (likely(skb_rx_queue_recorded(skb))) {
 589                 txq = skb_get_rx_queue(skb);
 590                 while (unlikely(txq >= numqueues))
 591                         txq -= numqueues;
 592         }
 593
 594         return txq;
 595 }
 596
 597 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 598 {
 599         struct tun_prog *prog;
 600         u16 ret = 0;
 601
 602         prog = rcu_dereference(tun->steering_prog);
 603         if (prog)
 604                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 605
 606         return ret % tun->numqueues;
 607 }
 608
 609 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 610                             void *accel_priv, select_queue_fallback_t fallback)
 611 {
 612         struct tun_struct *tun = netdev_priv(dev);
 613         u16 ret;
 614
 615         rcu_read_lock();
 616         if (rcu_dereference(tun->steering_prog))
 617                 ret = tun_ebpf_select_queue(tun, skb);
 618         else
 619                 ret = tun_automq_select_queue(tun, skb);
 620         rcu_read_unlock();
 621
 622         return ret;
 623 }
 624
 625 static inline bool tun_not_capable(struct tun_struct *tun)
 626 {
 627         const struct cred *cred = current_cred();
 628         struct net *net = dev_net(tun->dev);
 629
 630         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 631                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 632                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 633 }
 634
 635 static void tun_set_real_num_queues(struct tun_struct *tun)
 636 {
 637         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 638         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 639 }
 640
 641 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 642 {
 643         tfile->detached = tun;
 644         list_add_tail(&tfile->next, &tun->disabled);
 645         ++tun->numdisabled;
 646 }
 647
 648 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 649 {
 650         struct tun_struct *tun = tfile->detached;
 651
 652         tfile->detached = NULL;
 653         list_del_init(&tfile->next);
 654         --tun->numdisabled;
 655         return tun;
 656 }
 657
 658 static void tun_ptr_free(void *ptr)
 659 {
 660         if (!ptr)
 661                 return;
 662         if (tun_is_xdp_buff(ptr)) {
 663                 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
 664
 665                 put_page(virt_to_head_page(xdp->data));
 666         } else {
 667                 __skb_array_destroy_skb(ptr);
 668         }
 669 }
 670
 671 static void tun_queue_purge(struct tun_file *tfile)
 672 {
 673         void *ptr;
 674
 675         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 676                 tun_ptr_free(ptr);
 677
 678         skb_queue_purge(&tfile->sk.sk_write_queue);
 679         skb_queue_purge(&tfile->sk.sk_error_queue);
 680 }
 681
 682 static void tun_cleanup_tx_ring(struct tun_file *tfile)
 683 {
 684         if (tfile->tx_ring.queue) {
 685                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 686                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 687                 memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
 688         }
 689 }
 690
 691 static void __tun_detach(struct tun_file *tfile, bool clean)
 692 {
 693         struct tun_file *ntfile;
 694         struct tun_struct *tun;
 695
 696         tun = rtnl_dereference(tfile->tun);
 697
 698         if (tun && clean) {
 699                 tun_napi_disable(tun, tfile);
 700                 tun_napi_del(tun, tfile);
 701         }
 702
 703         if (tun && !tfile->detached) {
 704                 u16 index = tfile->queue_index;
 705                 BUG_ON(index >= tun->numqueues);
 706
 707                 rcu_assign_pointer(tun->tfiles[index],
 708                                    tun->tfiles[tun->numqueues - 1]);
 709                 ntfile = rtnl_dereference(tun->tfiles[index]);
 710                 ntfile->queue_index = index;
 711
 712                 --tun->numqueues;
 713                 if (clean) {
 714                         RCU_INIT_POINTER(tfile->tun, NULL);
 715                         sock_put(&tfile->sk);
 716                 } else
 717                         tun_disable_queue(tun, tfile);
 718
 719                 synchronize_net();
 720                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 721                 /* Drop read queue */
 722                 tun_queue_purge(tfile);
 723                 tun_set_real_num_queues(tun);
 724         } else if (tfile->detached && clean) {
 725                 tun = tun_enable_queue(tfile);
 726                 sock_put(&tfile->sk);
 727         }
 728
 729         if (clean) {
 730                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 731                         netif_carrier_off(tun->dev);
 732
 733                         if (!(tun->flags & IFF_PERSIST) &&
 734                             tun->dev->reg_state == NETREG_REGISTERED)
 735                                 unregister_netdevice(tun->dev);
 736                 }
 737                 tun_cleanup_tx_ring(tfile);
 738                 sock_put(&tfile->sk);
 739         }
 740 }
 741
 742 static void tun_detach(struct tun_file *tfile, bool clean)
 743 {
 744         rtnl_lock();
 745         __tun_detach(tfile, clean);
 746         rtnl_unlock();
 747 }
 748
 749 static void tun_detach_all(struct net_device *dev)
 750 {
 751         struct tun_struct *tun = netdev_priv(dev);
 752         struct tun_file *tfile, *tmp;
 753         int i, n = tun->numqueues;
 754
 755         for (i = 0; i < n; i++) {
 756                 tfile = rtnl_dereference(tun->tfiles[i]);
 757                 BUG_ON(!tfile);
 758                 tun_napi_disable(tun, tfile);
 759                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 760                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 761                 RCU_INIT_POINTER(tfile->tun, NULL);
 762                 --tun->numqueues;
 763         }
 764         list_for_each_entry(tfile, &tun->disabled, next) {
 765                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 766                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 767                 RCU_INIT_POINTER(tfile->tun, NULL);
 768         }
 769         BUG_ON(tun->numqueues != 0);
 770
 771         synchronize_net();
 772         for (i = 0; i < n; i++) {
 773                 tfile = rtnl_dereference(tun->tfiles[i]);
 774                 tun_napi_del(tun, tfile);
 775                 /* Drop read queue */
 776                 tun_queue_purge(tfile);
 777                 sock_put(&tfile->sk);
 778                 tun_cleanup_tx_ring(tfile);
 779         }
 780         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 781                 tun_enable_queue(tfile);
 782                 tun_queue_purge(tfile);
 783                 sock_put(&tfile->sk);
 784                 tun_cleanup_tx_ring(tfile);
 785         }
 786         BUG_ON(tun->numdisabled != 0);
 787
 788         if (tun->flags & IFF_PERSIST)
 789                 module_put(THIS_MODULE);
 790 }
 791
 792 static int tun_attach(struct tun_struct *tun, struct file *file,
 793                       bool skip_filter, bool napi)
 794 {
 795         struct tun_file *tfile = file->private_data;
 796         struct net_device *dev = tun->dev;
 797         int err;
 798
 799         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 800         if (err < 0)
 801                 goto out;
 802
 803         err = -EINVAL;
 804         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 805                 goto out;
 806
 807         err = -EBUSY;
 808         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 809                 goto out;
 810
 811         err = -E2BIG;
 812         if (!tfile->detached &&
 813             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 814                 goto out;
 815
 816         err = 0;
 817
 818         /* Re-attach the filter to persist device */
 819         if (!skip_filter && (tun->filter_attached == true)) {
 820                 lock_sock(tfile->socket.sk);
 821                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 822                 release_sock(tfile->socket.sk);
 823                 if (!err)
 824                         goto out;
 825         }
 826
 827         if (!tfile->detached &&
 828             ptr_ring_init(&tfile->tx_ring, dev->tx_queue_len, GFP_KERNEL)) {
 829                 err = -ENOMEM;
 830                 goto out;
 831         }
 832
 833         tfile->queue_index = tun->numqueues;
 834         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 835
 836         if (tfile->detached) {
 837                 /* Re-attach detached tfile, updating XDP queue_index */
 838                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 839
 840                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 841                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 842         } else {
 843                 /* Setup XDP RX-queue info, for new tfile getting attached */
 844                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 845                                        tun->dev, tfile->queue_index);
 846                 if (err < 0)
 847                         goto out;
 848                 err = 0;
 849         }
 850
 851         rcu_assign_pointer(tfile->tun, tun);
 852         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 853         tun->numqueues++;
 854
 855         if (tfile->detached) {
 856                 tun_enable_queue(tfile);
 857         } else {
 858                 sock_hold(&tfile->sk);
 859                 tun_napi_init(tun, tfile, napi);
 860         }
 861
 862         tun_set_real_num_queues(tun);
 863
 864         /* device is allowed to go away first, so no need to hold extra
 865          * refcnt.
 866          */
 867
 868 out:
 869         return err;
 870 }
 871
 872 static struct tun_struct *tun_get(struct tun_file *tfile)
 873 {
 874         struct tun_struct *tun;
 875
 876         rcu_read_lock();
 877         tun = rcu_dereference(tfile->tun);
 878         if (tun)
 879                 dev_hold(tun->dev);
 880         rcu_read_unlock();
 881
 882         return tun;
 883 }
 884
 885 static void tun_put(struct tun_struct *tun)
 886 {
 887         dev_put(tun->dev);
 888 }
 889
 890 /* TAP filtering */
 891 static void addr_hash_set(u32 *mask, const u8 *addr)
 892 {
 893         int n = ether_crc(ETH_ALEN, addr) >> 26;
 894         mask[n >> 5] |= (1 << (n & 31));
 895 }
 896
 897 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 898 {
 899         int n = ether_crc(ETH_ALEN, addr) >> 26;
 900         return mask[n >> 5] & (1 << (n & 31));
 901 }
 902
 903 static int update_filter(struct tap_filter *filter, void __user *arg)
 904 {
 905         struct { u8 u[ETH_ALEN]; } *addr;
 906         struct tun_filter uf;
 907         int err, alen, n, nexact;
 908
 909         if (copy_from_user(&uf, arg, sizeof(uf)))
 910                 return -EFAULT;
 911
 912         if (!uf.count) {
 913                 /* Disabled */
 914                 filter->count = 0;
 915                 return 0;
 916         }
 917
 918         alen = ETH_ALEN * uf.count;
 919         addr = memdup_user(arg + sizeof(uf), alen);
 920         if (IS_ERR(addr))
 921                 return PTR_ERR(addr);
 922
 923         /* The filter is updated without holding any locks. Which is
 924          * perfectly safe. We disable it first and in the worst
 925          * case we'll accept a few undesired packets. */
 926         filter->count = 0;
 927         wmb();
 928
 929         /* Use first set of addresses as an exact filter */
 930         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 931                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 932
 933         nexact = n;
 934
 935         /* Remaining multicast addresses are hashed,
 936          * unicast will leave the filter disabled. */
 937         memset(filter->mask, 0, sizeof(filter->mask));
 938         for (; n < uf.count; n++) {
 939                 if (!is_multicast_ether_addr(addr[n].u)) {
 940                         err = 0; /* no filter */
 941                         goto free_addr;
 942                 }
 943                 addr_hash_set(filter->mask, addr[n].u);
 944         }
 945
 946         /* For ALLMULTI just set the mask to all ones.
 947          * This overrides the mask populated above. */
 948         if ((uf.flags & TUN_FLT_ALLMULTI))
 949                 memset(filter->mask, ~0, sizeof(filter->mask));
 950
 951         /* Now enable the filter */
 952         wmb();
 953         filter->count = nexact;
 954
 955         /* Return the number of exact filters */
 956         err = nexact;
 957 free_addr:
 958         kfree(addr);
 959         return err;
 960 }
 961
 962 /* Returns: 0 - drop, !=0 - accept */
 963 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 964 {
 965         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 966          * at this point. */
 967         struct ethhdr *eh = (struct ethhdr *) skb->data;
 968         int i;
 969
 970         /* Exact match */
 971         for (i = 0; i < filter->count; i++)
 972                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 973                         return 1;
 974
 975         /* Inexact match (multicast only) */
 976         if (is_multicast_ether_addr(eh->h_dest))
 977                 return addr_hash_test(filter->mask, eh->h_dest);
 978
 979         return 0;
 980 }
 981
 982 /*
 983  * Checks whether the packet is accepted or not.
 984  * Returns: 0 - drop, !=0 - accept
 985  */
 986 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 987 {
 988         if (!filter->count)
 989                 return 1;
 990
 991         return run_filter(filter, skb);
 992 }
 993
 994 /* Network device part of the driver */
 995
 996 static const struct ethtool_ops tun_ethtool_ops;
 997
 998 /* Net device detach from fd. */
 999 static void tun_net_uninit(struct net_device *dev)
1000 {
1001         tun_detach_all(dev);
1002 }
1003
1004 /* Net device open. */
1005 static int tun_net_open(struct net_device *dev)
1006 {
1007         struct tun_struct *tun = netdev_priv(dev);
1008         int i;
1009
1010         netif_tx_start_all_queues(dev);
1011
1012         for (i = 0; i < tun->numqueues; i++) {
1013                 struct tun_file *tfile;
1014
1015                 tfile = rtnl_dereference(tun->tfiles[i]);
1016                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1017         }
1018
1019         return 0;
1020 }
1021
1022 /* Net device close. */
1023 static int tun_net_close(struct net_device *dev)
1024 {
1025         netif_tx_stop_all_queues(dev);
1026         return 0;
1027 }
1028
1029 /* Net device start xmit */
1030 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1031 {
1032 #ifdef CONFIG_RPS
1033         if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1034                 /* Select queue was not called for the skbuff, so we extract the
1035                  * RPS hash and save it into the flow_table here.
1036                  */
1037                 __u32 rxhash;
1038
1039                 rxhash = __skb_get_hash_symmetric(skb);
1040                 if (rxhash) {
1041                         struct tun_flow_entry *e;
1042                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1043                                         rxhash);
1044                         if (e)
1045                                 tun_flow_save_rps_rxhash(e, rxhash);
1046                 }
1047         }
1048 #endif
1049 }
1050
1051 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1052                                     struct sk_buff *skb,
1053                                     int len)
1054 {
1055         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1056
1057         if (prog)
1058                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1059
1060         return len;
1061 }
1062
1063 /* Net device start xmit */
1064 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1065 {
1066         struct tun_struct *tun = netdev_priv(dev);
1067         int txq = skb->queue_mapping;
1068         struct tun_file *tfile;
1069         int len = skb->len;
1070
1071         rcu_read_lock();
1072         tfile = rcu_dereference(tun->tfiles[txq]);
1073
1074         /* Drop packet if interface is not attached */
1075         if (txq >= tun->numqueues)
1076                 goto drop;
1077
1078         if (!rcu_dereference(tun->steering_prog))
1079                 tun_automq_xmit(tun, skb);
1080
1081         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1082
1083         BUG_ON(!tfile);
1084
1085         /* Drop if the filter does not like it.
1086          * This is a noop if the filter is disabled.
1087          * Filter can be enabled only for the TAP devices. */
1088         if (!check_filter(&tun->txflt, skb))
1089                 goto drop;
1090
1091         if (tfile->socket.sk->sk_filter &&
1092             sk_filter(tfile->socket.sk, skb))
1093                 goto drop;
1094
1095         len = run_ebpf_filter(tun, skb, len);
1096
1097         /* Trim extra bytes since we may insert vlan proto & TCI
1098          * in tun_put_user().
1099          */
1100         len -= skb_vlan_tag_present(skb) ? sizeof(struct veth) : 0;
1101         if (len <= 0 || pskb_trim(skb, len))
1102                 goto drop;
1103
1104         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1105                 goto drop;
1106
1107         skb_tx_timestamp(skb);
1108
1109         /* Orphan the skb - required as we might hang on to it
1110          * for indefinite time.
1111          */
1112         skb_orphan(skb);
1113
1114         nf_reset(skb);
1115
1116         if (ptr_ring_produce(&tfile->tx_ring, skb))
1117                 goto drop;
1118
1119         /* Notify and wake up reader process */
1120         if (tfile->flags & TUN_FASYNC)
1121                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1122         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1123
1124         rcu_read_unlock();
1125         return NETDEV_TX_OK;
1126
1127 drop:
1128         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1129         skb_tx_error(skb);
1130         kfree_skb(skb);
1131         rcu_read_unlock();
1132         return NET_XMIT_DROP;
1133 }
1134
1135 static void tun_net_mclist(struct net_device *dev)
1136 {
1137         /*
1138          * This callback is supposed to deal with mc filter in
1139          * _rx_ path and has nothing to do with the _tx_ path.
1140          * In rx path we always accept everything userspace gives us.
1141          */
1142 }
1143
1144 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1145         netdev_features_t features)
1146 {
1147         struct tun_struct *tun = netdev_priv(dev);
1148
1149         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1150 }
1151 #ifdef CONFIG_NET_POLL_CONTROLLER
1152 static void tun_poll_controller(struct net_device *dev)
1153 {
1154         /*
1155          * Tun only receives frames when:
1156          * 1) the char device endpoint gets data from user space
1157          * 2) the tun socket gets a sendmsg call from user space
1158          * If NAPI is not enabled, since both of those are synchronous
1159          * operations, we are guaranteed never to have pending data when we poll
1160          * for it so there is nothing to do here but return.
1161          * We need this though so netpoll recognizes us as an interface that
1162          * supports polling, which enables bridge devices in virt setups to
1163          * still use netconsole
1164          * If NAPI is enabled, however, we need to schedule polling for all
1165          * queues unless we are using napi_gro_frags(), which we call in
1166          * process context and not in NAPI context.
1167          */
1168         struct tun_struct *tun = netdev_priv(dev);
1169
1170         if (tun->flags & IFF_NAPI) {
1171                 struct tun_file *tfile;
1172                 int i;
1173
1174                 if (tun_napi_frags_enabled(tun))
1175                         return;
1176
1177                 rcu_read_lock();
1178                 for (i = 0; i < tun->numqueues; i++) {
1179                         tfile = rcu_dereference(tun->tfiles[i]);
1180                         if (tfile->napi_enabled)
1181                                 napi_schedule(&tfile->napi);
1182                 }
1183                 rcu_read_unlock();
1184         }
1185         return;
1186 }
1187 #endif
1188
1189 static void tun_set_headroom(struct net_device *dev, int new_hr)
1190 {
1191         struct tun_struct *tun = netdev_priv(dev);
1192
1193         if (new_hr < NET_SKB_PAD)
1194                 new_hr = NET_SKB_PAD;
1195
1196         tun->align = new_hr;
1197 }
1198
1199 static void
1200 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1201 {
1202         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1203         struct tun_struct *tun = netdev_priv(dev);
1204         struct tun_pcpu_stats *p;
1205         int i;
1206
1207         for_each_possible_cpu(i) {
1208                 u64 rxpackets, rxbytes, txpackets, txbytes;
1209                 unsigned int start;
1210
1211                 p = per_cpu_ptr(tun->pcpu_stats, i);
1212                 do {
1213                         start = u64_stats_fetch_begin(&p->syncp);
1214                         rxpackets       = p->rx_packets;
1215                         rxbytes         = p->rx_bytes;
1216                         txpackets       = p->tx_packets;
1217                         txbytes         = p->tx_bytes;
1218                 } while (u64_stats_fetch_retry(&p->syncp, start));
1219
1220                 stats->rx_packets       += rxpackets;
1221                 stats->rx_bytes         += rxbytes;
1222                 stats->tx_packets       += txpackets;
1223                 stats->tx_bytes         += txbytes;
1224
1225                 /* u32 counters */
1226                 rx_dropped      += p->rx_dropped;
1227                 rx_frame_errors += p->rx_frame_errors;
1228                 tx_dropped      += p->tx_dropped;
1229         }
1230         stats->rx_dropped  = rx_dropped;
1231         stats->rx_frame_errors = rx_frame_errors;
1232         stats->tx_dropped = tx_dropped;
1233 }
1234
1235 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1236                        struct netlink_ext_ack *extack)
1237 {
1238         struct tun_struct *tun = netdev_priv(dev);
1239         struct bpf_prog *old_prog;
1240
1241         old_prog = rtnl_dereference(tun->xdp_prog);
1242         rcu_assign_pointer(tun->xdp_prog, prog);
1243         if (old_prog)
1244                 bpf_prog_put(old_prog);
1245
1246         return 0;
1247 }
1248
1249 static u32 tun_xdp_query(struct net_device *dev)
1250 {
1251         struct tun_struct *tun = netdev_priv(dev);
1252         const struct bpf_prog *xdp_prog;
1253
1254         xdp_prog = rtnl_dereference(tun->xdp_prog);
1255         if (xdp_prog)
1256                 return xdp_prog->aux->id;
1257
1258         return 0;
1259 }
1260
1261 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1262 {
1263         switch (xdp->command) {
1264         case XDP_SETUP_PROG:
1265                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1266         case XDP_QUERY_PROG:
1267                 xdp->prog_id = tun_xdp_query(dev);
1268                 xdp->prog_attached = !!xdp->prog_id;
1269                 return 0;
1270         default:
1271                 return -EINVAL;
1272         }
1273 }
1274
1275 static const struct net_device_ops tun_netdev_ops = {
1276         .ndo_uninit             = tun_net_uninit,
1277         .ndo_open               = tun_net_open,
1278         .ndo_stop               = tun_net_close,
1279         .ndo_start_xmit         = tun_net_xmit,
1280         .ndo_fix_features       = tun_net_fix_features,
1281         .ndo_select_queue       = tun_select_queue,
1282 #ifdef CONFIG_NET_POLL_CONTROLLER
1283         .ndo_poll_controller    = tun_poll_controller,
1284 #endif
1285         .ndo_set_rx_headroom    = tun_set_headroom,
1286         .ndo_get_stats64        = tun_net_get_stats64,
1287 };
1288
1289 static int tun_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp)
1290 {
1291         struct tun_struct *tun = netdev_priv(dev);
1292         struct xdp_buff *buff = xdp->data_hard_start;
1293         int headroom = xdp->data - xdp->data_hard_start;
1294         struct tun_file *tfile;
1295         u32 numqueues;
1296         int ret = 0;
1297
1298         /* Assure headroom is available and buff is properly aligned */
1299         if (unlikely(headroom < sizeof(*xdp) || tun_is_xdp_buff(xdp)))
1300                 return -ENOSPC;
1301
1302         *buff = *xdp;
1303
1304         rcu_read_lock();
1305
1306         numqueues = READ_ONCE(tun->numqueues);
1307         if (!numqueues) {
1308                 ret = -ENOSPC;
1309                 goto out;
1310         }
1311
1312         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1313                                             numqueues]);
1314         /* Encode the XDP flag into lowest bit for consumer to differ
1315          * XDP buffer from sk_buff.
1316          */
1317         if (ptr_ring_produce(&tfile->tx_ring, tun_xdp_to_ptr(buff))) {
1318                 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1319                 ret = -ENOSPC;
1320         }
1321
1322 out:
1323         rcu_read_unlock();
1324         return ret;
1325 }
1326
1327 static void tun_xdp_flush(struct net_device *dev)
1328 {
1329         struct tun_struct *tun = netdev_priv(dev);
1330         struct tun_file *tfile;
1331         u32 numqueues;
1332
1333         rcu_read_lock();
1334
1335         numqueues = READ_ONCE(tun->numqueues);
1336         if (!numqueues)
1337                 goto out;
1338
1339         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1340                                             numqueues]);
1341         /* Notify and wake up reader process */
1342         if (tfile->flags & TUN_FASYNC)
1343                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1344         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1345
1346 out:
1347         rcu_read_unlock();
1348 }
1349
1350 static const struct net_device_ops tap_netdev_ops = {
1351         .ndo_uninit             = tun_net_uninit,
1352         .ndo_open               = tun_net_open,
1353         .ndo_stop               = tun_net_close,
1354         .ndo_start_xmit         = tun_net_xmit,
1355         .ndo_fix_features       = tun_net_fix_features,
1356         .ndo_set_rx_mode        = tun_net_mclist,
1357         .ndo_set_mac_address    = eth_mac_addr,
1358         .ndo_validate_addr      = eth_validate_addr,
1359         .ndo_select_queue       = tun_select_queue,
1360 #ifdef CONFIG_NET_POLL_CONTROLLER
1361         .ndo_poll_controller    = tun_poll_controller,
1362 #endif
1363         .ndo_features_check     = passthru_features_check,
1364         .ndo_set_rx_headroom    = tun_set_headroom,
1365         .ndo_get_stats64        = tun_net_get_stats64,
1366         .ndo_bpf                = tun_xdp,
1367         .ndo_xdp_xmit           = tun_xdp_xmit,
1368         .ndo_xdp_flush          = tun_xdp_flush,
1369 };
1370
1371 static void tun_flow_init(struct tun_struct *tun)
1372 {
1373         int i;
1374
1375         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1376                 INIT_HLIST_HEAD(&tun->flows[i]);
1377
1378         tun->ageing_time = TUN_FLOW_EXPIRE;
1379         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1380         mod_timer(&tun->flow_gc_timer,
1381                   round_jiffies_up(jiffies + tun->ageing_time));
1382 }
1383
1384 static void tun_flow_uninit(struct tun_struct *tun)
1385 {
1386         del_timer_sync(&tun->flow_gc_timer);
1387         tun_flow_flush(tun);
1388 }
1389
1390 #define MIN_MTU 68
1391 #define MAX_MTU 65535
1392
1393 /* Initialize net device. */
1394 static void tun_net_init(struct net_device *dev)
1395 {
1396         struct tun_struct *tun = netdev_priv(dev);
1397
1398         switch (tun->flags & TUN_TYPE_MASK) {
1399         case IFF_TUN:
1400                 dev->netdev_ops = &tun_netdev_ops;
1401
1402                 /* Point-to-Point TUN Device */
1403                 dev->hard_header_len = 0;
1404                 dev->addr_len = 0;
1405                 dev->mtu = 1500;
1406
1407                 /* Zero header length */
1408                 dev->type = ARPHRD_NONE;
1409                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1410                 break;
1411
1412         case IFF_TAP:
1413                 dev->netdev_ops = &tap_netdev_ops;
1414                 /* Ethernet TAP Device */
1415                 ether_setup(dev);
1416                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1417                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1418
1419                 eth_hw_addr_random(dev);
1420
1421                 break;
1422         }
1423
1424         dev->min_mtu = MIN_MTU;
1425         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1426 }
1427
1428 /* Character device part */
1429
1430 /* Poll */
1431 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1432 {
1433         struct tun_file *tfile = file->private_data;
1434         struct tun_struct *tun = tun_get(tfile);
1435         struct sock *sk;
1436         __poll_t mask = 0;
1437
1438         if (!tun)
1439                 return POLLERR;
1440
1441         sk = tfile->socket.sk;
1442
1443         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1444
1445         poll_wait(file, sk_sleep(sk), wait);
1446
1447         if (!ptr_ring_empty(&tfile->tx_ring))
1448                 mask |= POLLIN | POLLRDNORM;
1449
1450         if (tun->dev->flags & IFF_UP &&
1451             (sock_writeable(sk) ||
1452              (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1453               sock_writeable(sk))))
1454                 mask |= POLLOUT | POLLWRNORM;
1455
1456         if (tun->dev->reg_state != NETREG_REGISTERED)
1457                 mask = POLLERR;
1458
1459         tun_put(tun);
1460         return mask;
1461 }
1462
1463 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1464                                             size_t len,
1465                                             const struct iov_iter *it)
1466 {
1467         struct sk_buff *skb;
1468         size_t linear;
1469         int err;
1470         int i;
1471
1472         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1473                 return ERR_PTR(-ENOMEM);
1474
1475         local_bh_disable();
1476         skb = napi_get_frags(&tfile->napi);
1477         local_bh_enable();
1478         if (!skb)
1479                 return ERR_PTR(-ENOMEM);
1480
1481         linear = iov_iter_single_seg_count(it);
1482         err = __skb_grow(skb, linear);
1483         if (err)
1484                 goto free;
1485
1486         skb->len = len;
1487         skb->data_len = len - linear;
1488         skb->truesize += skb->data_len;
1489
1490         for (i = 1; i < it->nr_segs; i++) {
1491                 size_t fragsz = it->iov[i].iov_len;
1492                 unsigned long offset;
1493                 struct page *page;
1494                 void *data;
1495
1496                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1497                         err = -EINVAL;
1498                         goto free;
1499                 }
1500
1501                 local_bh_disable();
1502                 data = napi_alloc_frag(fragsz);
1503                 local_bh_enable();
1504                 if (!data) {
1505                         err = -ENOMEM;
1506                         goto free;
1507                 }
1508
1509                 page = virt_to_head_page(data);
1510                 offset = data - page_address(page);
1511                 skb_fill_page_desc(skb, i - 1, page, offset, fragsz);
1512         }
1513
1514         return skb;
1515 free:
1516         /* frees skb and all frags allocated with napi_alloc_frag() */
1517         napi_free_frags(&tfile->napi);
1518         return ERR_PTR(err);
1519 }
1520
1521 /* prepad is the amount to reserve at front.  len is length after that.
1522  * linear is a hint as to how much to copy (usually headers). */
1523 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1524                                      size_t prepad, size_t len,
1525                                      size_t linear, int noblock)
1526 {
1527         struct sock *sk = tfile->socket.sk;
1528         struct sk_buff *skb;
1529         int err;
1530
1531         /* Under a page?  Don't bother with paged skb. */
1532         if (prepad + len < PAGE_SIZE || !linear)
1533                 linear = len;
1534
1535         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1536                                    &err, 0);
1537         if (!skb)
1538                 return ERR_PTR(err);
1539
1540         skb_reserve(skb, prepad);
1541         skb_put(skb, linear);
1542         skb->data_len = len - linear;
1543         skb->len += len - linear;
1544
1545         return skb;
1546 }
1547
1548 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1549                            struct sk_buff *skb, int more)
1550 {
1551         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1552         struct sk_buff_head process_queue;
1553         u32 rx_batched = tun->rx_batched;
1554         bool rcv = false;
1555
1556         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1557                 local_bh_disable();
1558                 netif_receive_skb(skb);
1559                 local_bh_enable();
1560                 return;
1561         }
1562
1563         spin_lock(&queue->lock);
1564         if (!more || skb_queue_len(queue) == rx_batched) {
1565                 __skb_queue_head_init(&process_queue);
1566                 skb_queue_splice_tail_init(queue, &process_queue);
1567                 rcv = true;
1568         } else {
1569                 __skb_queue_tail(queue, skb);
1570         }
1571         spin_unlock(&queue->lock);
1572
1573         if (rcv) {
1574                 struct sk_buff *nskb;
1575
1576                 local_bh_disable();
1577                 while ((nskb = __skb_dequeue(&process_queue)))
1578                         netif_receive_skb(nskb);
1579                 netif_receive_skb(skb);
1580                 local_bh_enable();
1581         }
1582 }
1583
1584 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1585                               int len, int noblock, bool zerocopy)
1586 {
1587         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1588                 return false;
1589
1590         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1591                 return false;
1592
1593         if (!noblock)
1594                 return false;
1595
1596         if (zerocopy)
1597                 return false;
1598
1599         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1600             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1601                 return false;
1602
1603         return true;
1604 }
1605
1606 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1607                                      struct tun_file *tfile,
1608                                      struct iov_iter *from,
1609                                      struct virtio_net_hdr *hdr,
1610                                      int len, int *skb_xdp)
1611 {
1612         struct page_frag *alloc_frag = &current->task_frag;
1613         struct sk_buff *skb;
1614         struct bpf_prog *xdp_prog;
1615         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1616         unsigned int delta = 0;
1617         char *buf;
1618         size_t copied;
1619         bool xdp_xmit = false;
1620         int err, pad = TUN_RX_PAD;
1621
1622         rcu_read_lock();
1623         xdp_prog = rcu_dereference(tun->xdp_prog);
1624         if (xdp_prog)
1625                 pad += TUN_HEADROOM;
1626         buflen += SKB_DATA_ALIGN(len + pad);
1627         rcu_read_unlock();
1628
1629         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1630         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1631                 return ERR_PTR(-ENOMEM);
1632
1633         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1634         copied = copy_page_from_iter(alloc_frag->page,
1635                                      alloc_frag->offset + pad,
1636                                      len, from);
1637         if (copied != len)
1638                 return ERR_PTR(-EFAULT);
1639
1640         /* There's a small window that XDP may be set after the check
1641          * of xdp_prog above, this should be rare and for simplicity
1642          * we do XDP on skb in case the headroom is not enough.
1643          */
1644         if (hdr->gso_type || !xdp_prog)
1645                 *skb_xdp = 1;
1646         else
1647                 *skb_xdp = 0;
1648
1649         rcu_read_lock();
1650         xdp_prog = rcu_dereference(tun->xdp_prog);
1651         if (xdp_prog && !*skb_xdp) {
1652                 struct xdp_buff xdp;
1653                 void *orig_data;
1654                 u32 act;
1655
1656                 xdp.data_hard_start = buf;
1657                 xdp.data = buf + pad;
1658                 xdp_set_data_meta_invalid(&xdp);
1659                 xdp.data_end = xdp.data + len;
1660                 xdp.rxq = &tfile->xdp_rxq;
1661                 orig_data = xdp.data;
1662                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1663
1664                 switch (act) {
1665                 case XDP_REDIRECT:
1666                         get_page(alloc_frag->page);
1667                         alloc_frag->offset += buflen;
1668                         err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1669                         if (err)
1670                                 goto err_redirect;
1671                         rcu_read_unlock();
1672                         return NULL;
1673                 case XDP_TX:
1674                         xdp_xmit = true;
1675                         /* fall through */
1676                 case XDP_PASS:
1677                         delta = orig_data - xdp.data;
1678                         break;
1679                 default:
1680                         bpf_warn_invalid_xdp_action(act);
1681                         /* fall through */
1682                 case XDP_ABORTED:
1683                         trace_xdp_exception(tun->dev, xdp_prog, act);
1684                         /* fall through */
1685                 case XDP_DROP:
1686                         goto err_xdp;
1687                 }
1688         }
1689
1690         skb = build_skb(buf, buflen);
1691         if (!skb) {
1692                 rcu_read_unlock();
1693                 return ERR_PTR(-ENOMEM);
1694         }
1695
1696         skb_reserve(skb, pad - delta);
1697         skb_put(skb, len + delta);
1698         get_page(alloc_frag->page);
1699         alloc_frag->offset += buflen;
1700
1701         if (xdp_xmit) {
1702                 skb->dev = tun->dev;
1703                 generic_xdp_tx(skb, xdp_prog);
1704                 rcu_read_unlock();
1705                 return NULL;
1706         }
1707
1708         rcu_read_unlock();
1709
1710         return skb;
1711
1712 err_redirect:
1713         put_page(alloc_frag->page);
1714 err_xdp:
1715         rcu_read_unlock();
1716         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1717         return NULL;
1718 }
1719
1720 /* Get packet from user space buffer */
1721 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1722                             void *msg_control, struct iov_iter *from,
1723                             int noblock, bool more)
1724 {
1725         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1726         struct sk_buff *skb;
1727         size_t total_len = iov_iter_count(from);
1728         size_t len = total_len, align = tun->align, linear;
1729         struct virtio_net_hdr gso = { 0 };
1730         struct tun_pcpu_stats *stats;
1731         int good_linear;
1732         int copylen;
1733         bool zerocopy = false;
1734         int err;
1735         u32 rxhash = 0;
1736         int skb_xdp = 1;
1737         bool frags = tun_napi_frags_enabled(tun);
1738
1739         if (!(tun->dev->flags & IFF_UP))
1740                 return -EIO;
1741
1742         if (!(tun->flags & IFF_NO_PI)) {
1743                 if (len < sizeof(pi))
1744                         return -EINVAL;
1745                 len -= sizeof(pi);
1746
1747                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1748                         return -EFAULT;
1749         }
1750
1751         if (tun->flags & IFF_VNET_HDR) {
1752                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1753
1754                 if (len < vnet_hdr_sz)
1755                         return -EINVAL;
1756                 len -= vnet_hdr_sz;
1757
1758                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1759                         return -EFAULT;
1760
1761                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1762                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1763                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1764
1765                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1766                         return -EINVAL;
1767                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1768         }
1769
1770         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1771                 align += NET_IP_ALIGN;
1772                 if (unlikely(len < ETH_HLEN ||
1773                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1774                         return -EINVAL;
1775         }
1776
1777         good_linear = SKB_MAX_HEAD(align);
1778
1779         if (msg_control) {
1780                 struct iov_iter i = *from;
1781
1782                 /* There are 256 bytes to be copied in skb, so there is
1783                  * enough room for skb expand head in case it is used.
1784                  * The rest of the buffer is mapped from userspace.
1785                  */
1786                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1787                 if (copylen > good_linear)
1788                         copylen = good_linear;
1789                 linear = copylen;
1790                 iov_iter_advance(&i, copylen);
1791                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1792                         zerocopy = true;
1793         }
1794
1795         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1796                 /* For the packet that is not easy to be processed
1797                  * (e.g gso or jumbo packet), we will do it at after
1798                  * skb was created with generic XDP routine.
1799                  */
1800                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1801                 if (IS_ERR(skb)) {
1802                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1803                         return PTR_ERR(skb);
1804                 }
1805                 if (!skb)
1806                         return total_len;
1807         } else {
1808                 if (!zerocopy) {
1809                         copylen = len;
1810                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1811                                 linear = good_linear;
1812                         else
1813                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1814                 }
1815
1816                 if (frags) {
1817                         mutex_lock(&tfile->napi_mutex);
1818                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1819                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1820                          * If zerocopy is enabled, then this layout will be
1821                          * overwritten by zerocopy_sg_from_iter().
1822                          */
1823                         zerocopy = false;
1824                 } else {
1825                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1826                                             noblock);
1827                 }
1828
1829                 if (IS_ERR(skb)) {
1830                         if (PTR_ERR(skb) != -EAGAIN)
1831                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1832                         if (frags)
1833                                 mutex_unlock(&tfile->napi_mutex);
1834                         return PTR_ERR(skb);
1835                 }
1836
1837                 if (zerocopy)
1838                         err = zerocopy_sg_from_iter(skb, from);
1839                 else
1840                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1841
1842                 if (err) {
1843                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1844                         kfree_skb(skb);
1845                         if (frags) {
1846                                 tfile->napi.skb = NULL;
1847                                 mutex_unlock(&tfile->napi_mutex);
1848                         }
1849
1850                         return -EFAULT;
1851                 }
1852         }
1853
1854         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1855                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1856                 kfree_skb(skb);
1857                 if (frags) {
1858                         tfile->napi.skb = NULL;
1859                         mutex_unlock(&tfile->napi_mutex);
1860                 }
1861
1862                 return -EINVAL;
1863         }
1864
1865         switch (tun->flags & TUN_TYPE_MASK) {
1866         case IFF_TUN:
1867                 if (tun->flags & IFF_NO_PI) {
1868                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1869
1870                         switch (ip_version) {
1871                         case 4:
1872                                 pi.proto = htons(ETH_P_IP);
1873                                 break;
1874                         case 6:
1875                                 pi.proto = htons(ETH_P_IPV6);
1876                                 break;
1877                         default:
1878                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1879                                 kfree_skb(skb);
1880                                 return -EINVAL;
1881                         }
1882                 }
1883
1884                 skb_reset_mac_header(skb);
1885                 skb->protocol = pi.proto;
1886                 skb->dev = tun->dev;
1887                 break;
1888         case IFF_TAP:
1889                 if (!frags)
1890                         skb->protocol = eth_type_trans(skb, tun->dev);
1891                 break;
1892         }
1893
1894         /* copy skb_ubuf_info for callback when skb has no error */
1895         if (zerocopy) {
1896                 skb_shinfo(skb)->destructor_arg = msg_control;
1897                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1898                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1899         } else if (msg_control) {
1900                 struct ubuf_info *uarg = msg_control;
1901                 uarg->callback(uarg, false);
1902         }
1903
1904         skb_reset_network_header(skb);
1905         skb_probe_transport_header(skb, 0);
1906
1907         if (skb_xdp) {
1908                 struct bpf_prog *xdp_prog;
1909                 int ret;
1910
1911                 rcu_read_lock();
1912                 xdp_prog = rcu_dereference(tun->xdp_prog);
1913                 if (xdp_prog) {
1914                         ret = do_xdp_generic(xdp_prog, skb);
1915                         if (ret != XDP_PASS) {
1916                                 rcu_read_unlock();
1917                                 return total_len;
1918                         }
1919                 }
1920                 rcu_read_unlock();
1921         }
1922
1923         rcu_read_lock();
1924         if (!rcu_dereference(tun->steering_prog))
1925                 rxhash = __skb_get_hash_symmetric(skb);
1926         rcu_read_unlock();
1927
1928         if (frags) {
1929                 /* Exercise flow dissector code path. */
1930                 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1931
1932                 if (unlikely(headlen > skb_headlen(skb))) {
1933                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1934                         napi_free_frags(&tfile->napi);
1935                         mutex_unlock(&tfile->napi_mutex);
1936                         WARN_ON(1);
1937                         return -ENOMEM;
1938                 }
1939
1940                 local_bh_disable();
1941                 napi_gro_frags(&tfile->napi);
1942                 local_bh_enable();
1943                 mutex_unlock(&tfile->napi_mutex);
1944         } else if (tfile->napi_enabled) {
1945                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1946                 int queue_len;
1947
1948                 spin_lock_bh(&queue->lock);
1949                 __skb_queue_tail(queue, skb);
1950                 queue_len = skb_queue_len(queue);
1951                 spin_unlock(&queue->lock);
1952
1953                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1954                         napi_schedule(&tfile->napi);
1955
1956                 local_bh_enable();
1957         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1958                 tun_rx_batched(tun, tfile, skb, more);
1959         } else {
1960                 netif_rx_ni(skb);
1961         }
1962
1963         stats = get_cpu_ptr(tun->pcpu_stats);
1964         u64_stats_update_begin(&stats->syncp);
1965         stats->rx_packets++;
1966         stats->rx_bytes += len;
1967         u64_stats_update_end(&stats->syncp);
1968         put_cpu_ptr(stats);
1969
1970         if (rxhash)
1971                 tun_flow_update(tun, rxhash, tfile);
1972
1973         return total_len;
1974 }
1975
1976 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1977 {
1978         struct file *file = iocb->ki_filp;
1979         struct tun_file *tfile = file->private_data;
1980         struct tun_struct *tun = tun_get(tfile);
1981         ssize_t result;
1982
1983         if (!tun)
1984                 return -EBADFD;
1985
1986         result = tun_get_user(tun, tfile, NULL, from,
1987                               file->f_flags & O_NONBLOCK, false);
1988
1989         tun_put(tun);
1990         return result;
1991 }
1992
1993 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1994                                 struct tun_file *tfile,
1995                                 struct xdp_buff *xdp,
1996                                 struct iov_iter *iter)
1997 {
1998         int vnet_hdr_sz = 0;
1999         size_t size = xdp->data_end - xdp->data;
2000         struct tun_pcpu_stats *stats;
2001         size_t ret;
2002
2003         if (tun->flags & IFF_VNET_HDR) {
2004                 struct virtio_net_hdr gso = { 0 };
2005
2006                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2007                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2008                         return -EINVAL;
2009                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2010                              sizeof(gso)))
2011                         return -EFAULT;
2012                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2013         }
2014
2015         ret = copy_to_iter(xdp->data, size, iter) + vnet_hdr_sz;
2016
2017         stats = get_cpu_ptr(tun->pcpu_stats);
2018         u64_stats_update_begin(&stats->syncp);
2019         stats->tx_packets++;
2020         stats->tx_bytes += ret;
2021         u64_stats_update_end(&stats->syncp);
2022         put_cpu_ptr(tun->pcpu_stats);
2023
2024         return ret;
2025 }
2026
2027 /* Put packet to the user space buffer */
2028 static ssize_t tun_put_user(struct tun_struct *tun,
2029                             struct tun_file *tfile,
2030                             struct sk_buff *skb,
2031                             struct iov_iter *iter)
2032 {
2033         struct tun_pi pi = { 0, skb->protocol };
2034         struct tun_pcpu_stats *stats;
2035         ssize_t total;
2036         int vlan_offset = 0;
2037         int vlan_hlen = 0;
2038         int vnet_hdr_sz = 0;
2039
2040         if (skb_vlan_tag_present(skb))
2041                 vlan_hlen = VLAN_HLEN;
2042
2043         if (tun->flags & IFF_VNET_HDR)
2044                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2045
2046         total = skb->len + vlan_hlen + vnet_hdr_sz;
2047
2048         if (!(tun->flags & IFF_NO_PI)) {
2049                 if (iov_iter_count(iter) < sizeof(pi))
2050                         return -EINVAL;
2051
2052                 total += sizeof(pi);
2053                 if (iov_iter_count(iter) < total) {
2054                         /* Packet will be striped */
2055                         pi.flags |= TUN_PKT_STRIP;
2056                 }
2057
2058                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2059                         return -EFAULT;
2060         }
2061
2062         if (vnet_hdr_sz) {
2063                 struct virtio_net_hdr gso;
2064
2065                 if (iov_iter_count(iter) < vnet_hdr_sz)
2066                         return -EINVAL;
2067
2068                 if (virtio_net_hdr_from_skb(skb, &gso,
2069                                             tun_is_little_endian(tun), true)) {
2070                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2071                         pr_err("unexpected GSO type: "
2072                                "0x%x, gso_size %d, hdr_len %d\n",
2073                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2074                                tun16_to_cpu(tun, gso.hdr_len));
2075                         print_hex_dump(KERN_ERR, "tun: ",
2076                                        DUMP_PREFIX_NONE,
2077                                        16, 1, skb->head,
2078                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2079                         WARN_ON_ONCE(1);
2080                         return -EINVAL;
2081                 }
2082
2083                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2084                         return -EFAULT;
2085
2086                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2087         }
2088
2089         if (vlan_hlen) {
2090                 int ret;
2091                 struct veth veth;
2092
2093                 veth.h_vlan_proto = skb->vlan_proto;
2094                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2095
2096                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2097
2098                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2099                 if (ret || !iov_iter_count(iter))
2100                         goto done;
2101
2102                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2103                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2104                         goto done;
2105         }
2106
2107         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2108
2109 done:
2110         /* caller is in process context, */
2111         stats = get_cpu_ptr(tun->pcpu_stats);
2112         u64_stats_update_begin(&stats->syncp);
2113         stats->tx_packets++;
2114         stats->tx_bytes += skb->len + vlan_hlen;
2115         u64_stats_update_end(&stats->syncp);
2116         put_cpu_ptr(tun->pcpu_stats);
2117
2118         return total;
2119 }
2120
2121 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2122 {
2123         DECLARE_WAITQUEUE(wait, current);
2124         void *ptr = NULL;
2125         int error = 0;
2126
2127         ptr = ptr_ring_consume(&tfile->tx_ring);
2128         if (ptr)
2129                 goto out;
2130         if (noblock) {
2131                 error = -EAGAIN;
2132                 goto out;
2133         }
2134
2135         add_wait_queue(&tfile->wq.wait, &wait);
2136         current->state = TASK_INTERRUPTIBLE;
2137
2138         while (1) {
2139                 ptr = ptr_ring_consume(&tfile->tx_ring);
2140                 if (ptr)
2141                         break;
2142                 if (signal_pending(current)) {
2143                         error = -ERESTARTSYS;
2144                         break;
2145                 }
2146                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2147                         error = -EFAULT;
2148                         break;
2149                 }
2150
2151                 schedule();
2152         }
2153
2154         current->state = TASK_RUNNING;
2155         remove_wait_queue(&tfile->wq.wait, &wait);
2156
2157 out:
2158         *err = error;
2159         return ptr;
2160 }
2161
2162 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2163                            struct iov_iter *to,
2164                            int noblock, void *ptr)
2165 {
2166         ssize_t ret;
2167         int err;
2168
2169         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2170
2171         if (!iov_iter_count(to)) {
2172                 tun_ptr_free(ptr);
2173                 return 0;
2174         }
2175
2176         if (!ptr) {
2177                 /* Read frames from ring */
2178                 ptr = tun_ring_recv(tfile, noblock, &err);
2179                 if (!ptr)
2180                         return err;
2181         }
2182
2183         if (tun_is_xdp_buff(ptr)) {
2184                 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2185
2186                 ret = tun_put_user_xdp(tun, tfile, xdp, to);
2187                 put_page(virt_to_head_page(xdp->data));
2188         } else {
2189                 struct sk_buff *skb = ptr;
2190
2191                 ret = tun_put_user(tun, tfile, skb, to);
2192                 if (unlikely(ret < 0))
2193                         kfree_skb(skb);
2194                 else
2195                         consume_skb(skb);
2196         }
2197
2198         return ret;
2199 }
2200
2201 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2202 {
2203         struct file *file = iocb->ki_filp;
2204         struct tun_file *tfile = file->private_data;
2205         struct tun_struct *tun = tun_get(tfile);
2206         ssize_t len = iov_iter_count(to), ret;
2207
2208         if (!tun)
2209                 return -EBADFD;
2210         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2211         ret = min_t(ssize_t, ret, len);
2212         if (ret > 0)
2213                 iocb->ki_pos = ret;
2214         tun_put(tun);
2215         return ret;
2216 }
2217
2218 static void tun_prog_free(struct rcu_head *rcu)
2219 {
2220         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2221
2222         bpf_prog_destroy(prog->prog);
2223         kfree(prog);
2224 }
2225
2226 static int __tun_set_ebpf(struct tun_struct *tun,
2227                           struct tun_prog __rcu **prog_p,
2228                           struct bpf_prog *prog)
2229 {
2230         struct tun_prog *old, *new = NULL;
2231
2232         if (prog) {
2233                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2234                 if (!new)
2235                         return -ENOMEM;
2236                 new->prog = prog;
2237         }
2238
2239         spin_lock_bh(&tun->lock);
2240         old = rcu_dereference_protected(*prog_p,
2241                                         lockdep_is_held(&tun->lock));
2242         rcu_assign_pointer(*prog_p, new);
2243         spin_unlock_bh(&tun->lock);
2244
2245         if (old)
2246                 call_rcu(&old->rcu, tun_prog_free);
2247
2248         return 0;
2249 }
2250
2251 static void tun_free_netdev(struct net_device *dev)
2252 {
2253         struct tun_struct *tun = netdev_priv(dev);
2254
2255         BUG_ON(!(list_empty(&tun->disabled)));
2256         free_percpu(tun->pcpu_stats);
2257         tun_flow_uninit(tun);
2258         security_tun_dev_free_security(tun->security);
2259         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2260         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2261 }
2262
2263 static void tun_setup(struct net_device *dev)
2264 {
2265         struct tun_struct *tun = netdev_priv(dev);
2266
2267         tun->owner = INVALID_UID;
2268         tun->group = INVALID_GID;
2269
2270         dev->ethtool_ops = &tun_ethtool_ops;
2271         dev->needs_free_netdev = true;
2272         dev->priv_destructor = tun_free_netdev;
2273         /* We prefer our own queue length */
2274         dev->tx_queue_len = TUN_READQ_SIZE;
2275 }
2276
2277 /* Trivial set of netlink ops to allow deleting tun or tap
2278  * device with netlink.
2279  */
2280 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2281                         struct netlink_ext_ack *extack)
2282 {
2283         return -EINVAL;
2284 }
2285
2286 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2287         .kind           = DRV_NAME,
2288         .priv_size      = sizeof(struct tun_struct),
2289         .setup          = tun_setup,
2290         .validate       = tun_validate,
2291 };
2292
2293 static void tun_sock_write_space(struct sock *sk)
2294 {
2295         struct tun_file *tfile;
2296         wait_queue_head_t *wqueue;
2297
2298         if (!sock_writeable(sk))
2299                 return;
2300
2301         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2302                 return;
2303
2304         wqueue = sk_sleep(sk);
2305         if (wqueue && waitqueue_active(wqueue))
2306                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
2307                                                 POLLWRNORM | POLLWRBAND);
2308
2309         tfile = container_of(sk, struct tun_file, sk);
2310         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2311 }
2312
2313 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2314 {
2315         int ret;
2316         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2317         struct tun_struct *tun = tun_get(tfile);
2318
2319         if (!tun)
2320                 return -EBADFD;
2321
2322         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2323                            m->msg_flags & MSG_DONTWAIT,
2324                            m->msg_flags & MSG_MORE);
2325         tun_put(tun);
2326         return ret;
2327 }
2328
2329 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2330                        int flags)
2331 {
2332         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2333         struct tun_struct *tun = tun_get(tfile);
2334         void *ptr = m->msg_control;
2335         int ret;
2336
2337         if (!tun) {
2338                 ret = -EBADFD;
2339                 goto out_free;
2340         }
2341
2342         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2343                 ret = -EINVAL;
2344                 goto out_put_tun;
2345         }
2346         if (flags & MSG_ERRQUEUE) {
2347                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2348                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2349                 goto out;
2350         }
2351         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2352         if (ret > (ssize_t)total_len) {
2353                 m->msg_flags |= MSG_TRUNC;
2354                 ret = flags & MSG_TRUNC ? ret : total_len;
2355         }
2356 out:
2357         tun_put(tun);
2358         return ret;
2359
2360 out_put_tun:
2361         tun_put(tun);
2362 out_free:
2363         tun_ptr_free(ptr);
2364         return ret;
2365 }
2366
2367 static int tun_ptr_peek_len(void *ptr)
2368 {
2369         if (likely(ptr)) {
2370                 if (tun_is_xdp_buff(ptr)) {
2371                         struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2372
2373                         return xdp->data_end - xdp->data;
2374                 }
2375                 return __skb_array_len_with_tag(ptr);
2376         } else {
2377                 return 0;
2378         }
2379 }
2380
2381 static int tun_peek_len(struct socket *sock)
2382 {
2383         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2384         struct tun_struct *tun;
2385         int ret = 0;
2386
2387         tun = tun_get(tfile);
2388         if (!tun)
2389                 return 0;
2390
2391         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2392         tun_put(tun);
2393
2394         return ret;
2395 }
2396
2397 /* Ops structure to mimic raw sockets with tun */
2398 static const struct proto_ops tun_socket_ops = {
2399         .peek_len = tun_peek_len,
2400         .sendmsg = tun_sendmsg,
2401         .recvmsg = tun_recvmsg,
2402 };
2403
2404 static struct proto tun_proto = {
2405         .name           = "tun",
2406         .owner          = THIS_MODULE,
2407         .obj_size       = sizeof(struct tun_file),
2408 };
2409
2410 static int tun_flags(struct tun_struct *tun)
2411 {
2412         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2413 }
2414
2415 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2416                               char *buf)
2417 {
2418         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2419         return sprintf(buf, "0x%x\n", tun_flags(tun));
2420 }
2421
2422 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2423                               char *buf)
2424 {
2425         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2426         return uid_valid(tun->owner)?
2427                 sprintf(buf, "%u\n",
2428                         from_kuid_munged(current_user_ns(), tun->owner)):
2429                 sprintf(buf, "-1\n");
2430 }
2431
2432 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2433                               char *buf)
2434 {
2435         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2436         return gid_valid(tun->group) ?
2437                 sprintf(buf, "%u\n",
2438                         from_kgid_munged(current_user_ns(), tun->group)):
2439                 sprintf(buf, "-1\n");
2440 }
2441
2442 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2443 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2444 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2445
2446 static struct attribute *tun_dev_attrs[] = {
2447         &dev_attr_tun_flags.attr,
2448         &dev_attr_owner.attr,
2449         &dev_attr_group.attr,
2450         NULL
2451 };
2452
2453 static const struct attribute_group tun_attr_group = {
2454         .attrs = tun_dev_attrs
2455 };
2456
2457 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2458 {
2459         struct tun_struct *tun;
2460         struct tun_file *tfile = file->private_data;
2461         struct net_device *dev;
2462         int err;
2463
2464         if (tfile->detached)
2465                 return -EINVAL;
2466
2467         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2468                 if (!capable(CAP_NET_ADMIN))
2469                         return -EPERM;
2470
2471                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2472                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2473                         return -EINVAL;
2474         }
2475
2476         dev = __dev_get_by_name(net, ifr->ifr_name);
2477         if (dev) {
2478                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2479                         return -EBUSY;
2480                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2481                         tun = netdev_priv(dev);
2482                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2483                         tun = netdev_priv(dev);
2484                 else
2485                         return -EINVAL;
2486
2487                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2488                     !!(tun->flags & IFF_MULTI_QUEUE))
2489                         return -EINVAL;
2490
2491                 if (tun_not_capable(tun))
2492                         return -EPERM;
2493                 err = security_tun_dev_open(tun->security);
2494                 if (err < 0)
2495                         return err;
2496
2497                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2498                                  ifr->ifr_flags & IFF_NAPI);
2499                 if (err < 0)
2500                         return err;
2501
2502                 if (tun->flags & IFF_MULTI_QUEUE &&
2503                     (tun->numqueues + tun->numdisabled > 1)) {
2504                         /* One or more queue has already been attached, no need
2505                          * to initialize the device again.
2506                          */
2507                         return 0;
2508                 }
2509         }
2510         else {
2511                 char *name;
2512                 unsigned long flags = 0;
2513                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2514                              MAX_TAP_QUEUES : 1;
2515
2516                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2517                         return -EPERM;
2518                 err = security_tun_dev_create();
2519                 if (err < 0)
2520                         return err;
2521
2522                 /* Set dev type */
2523                 if (ifr->ifr_flags & IFF_TUN) {
2524                         /* TUN device */
2525                         flags |= IFF_TUN;
2526                         name = "tun%d";
2527                 } else if (ifr->ifr_flags & IFF_TAP) {
2528                         /* TAP device */
2529                         flags |= IFF_TAP;
2530                         name = "tap%d";
2531                 } else
2532                         return -EINVAL;
2533
2534                 if (*ifr->ifr_name)
2535                         name = ifr->ifr_name;
2536
2537                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2538                                        NET_NAME_UNKNOWN, tun_setup, queues,
2539                                        queues);
2540
2541                 if (!dev)
2542                         return -ENOMEM;
2543                 err = dev_get_valid_name(net, dev, name);
2544                 if (err < 0)
2545                         goto err_free_dev;
2546
2547                 dev_net_set(dev, net);
2548                 dev->rtnl_link_ops = &tun_link_ops;
2549                 dev->ifindex = tfile->ifindex;
2550                 dev->sysfs_groups[0] = &tun_attr_group;
2551
2552                 tun = netdev_priv(dev);
2553                 tun->dev = dev;
2554                 tun->flags = flags;
2555                 tun->txflt.count = 0;
2556                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2557
2558                 tun->align = NET_SKB_PAD;
2559                 tun->filter_attached = false;
2560                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2561                 tun->rx_batched = 0;
2562                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2563
2564                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2565                 if (!tun->pcpu_stats) {
2566                         err = -ENOMEM;
2567                         goto err_free_dev;
2568                 }
2569
2570                 spin_lock_init(&tun->lock);
2571
2572                 err = security_tun_dev_alloc_security(&tun->security);
2573                 if (err < 0)
2574                         goto err_free_stat;
2575
2576                 tun_net_init(dev);
2577                 tun_flow_init(tun);
2578
2579                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2580                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2581                                    NETIF_F_HW_VLAN_STAG_TX;
2582                 dev->features = dev->hw_features | NETIF_F_LLTX;
2583                 dev->vlan_features = dev->features &
2584                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2585                                        NETIF_F_HW_VLAN_STAG_TX);
2586
2587                 INIT_LIST_HEAD(&tun->disabled);
2588                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2589                 if (err < 0)
2590                         goto err_free_flow;
2591
2592                 err = register_netdevice(tun->dev);
2593                 if (err < 0)
2594                         goto err_detach;
2595         }
2596
2597         netif_carrier_on(tun->dev);
2598
2599         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2600
2601         tun->flags = (tun->flags & ~TUN_FEATURES) |
2602                 (ifr->ifr_flags & TUN_FEATURES);
2603
2604         /* Make sure persistent devices do not get stuck in
2605          * xoff state.
2606          */
2607         if (netif_running(tun->dev))
2608                 netif_tx_wake_all_queues(tun->dev);
2609
2610         strcpy(ifr->ifr_name, tun->dev->name);
2611         return 0;
2612
2613 err_detach:
2614         tun_detach_all(dev);
2615         /* register_netdevice() already called tun_free_netdev() */
2616         goto err_free_dev;
2617
2618 err_free_flow:
2619         tun_flow_uninit(tun);
2620         security_tun_dev_free_security(tun->security);
2621 err_free_stat:
2622         free_percpu(tun->pcpu_stats);
2623 err_free_dev:
2624         free_netdev(dev);
2625         return err;
2626 }
2627
2628 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2629                        struct ifreq *ifr)
2630 {
2631         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2632
2633         strcpy(ifr->ifr_name, tun->dev->name);
2634
2635         ifr->ifr_flags = tun_flags(tun);
2636
2637 }
2638
2639 /* This is like a cut-down ethtool ops, except done via tun fd so no
2640  * privs required. */
2641 static int set_offload(struct tun_struct *tun, unsigned long arg)
2642 {
2643         netdev_features_t features = 0;
2644
2645         if (arg & TUN_F_CSUM) {
2646                 features |= NETIF_F_HW_CSUM;
2647                 arg &= ~TUN_F_CSUM;
2648
2649                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2650                         if (arg & TUN_F_TSO_ECN) {
2651                                 features |= NETIF_F_TSO_ECN;
2652                                 arg &= ~TUN_F_TSO_ECN;
2653                         }
2654                         if (arg & TUN_F_TSO4)
2655                                 features |= NETIF_F_TSO;
2656                         if (arg & TUN_F_TSO6)
2657                                 features |= NETIF_F_TSO6;
2658                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2659                 }
2660
2661                 arg &= ~TUN_F_UFO;
2662         }
2663
2664         /* This gives the user a way to test for new features in future by
2665          * trying to set them. */
2666         if (arg)
2667                 return -EINVAL;
2668
2669         tun->set_features = features;
2670         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2671         tun->dev->wanted_features |= features;
2672         netdev_update_features(tun->dev);
2673
2674         return 0;
2675 }
2676
2677 static void tun_detach_filter(struct tun_struct *tun, int n)
2678 {
2679         int i;
2680         struct tun_file *tfile;
2681
2682         for (i = 0; i < n; i++) {
2683                 tfile = rtnl_dereference(tun->tfiles[i]);
2684                 lock_sock(tfile->socket.sk);
2685                 sk_detach_filter(tfile->socket.sk);
2686                 release_sock(tfile->socket.sk);
2687         }
2688
2689         tun->filter_attached = false;
2690 }
2691
2692 static int tun_attach_filter(struct tun_struct *tun)
2693 {
2694         int i, ret = 0;
2695         struct tun_file *tfile;
2696
2697         for (i = 0; i < tun->numqueues; i++) {
2698                 tfile = rtnl_dereference(tun->tfiles[i]);
2699                 lock_sock(tfile->socket.sk);
2700                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2701                 release_sock(tfile->socket.sk);
2702                 if (ret) {
2703                         tun_detach_filter(tun, i);
2704                         return ret;
2705                 }
2706         }
2707
2708         tun->filter_attached = true;
2709         return ret;
2710 }
2711
2712 static void tun_set_sndbuf(struct tun_struct *tun)
2713 {
2714         struct tun_file *tfile;
2715         int i;
2716
2717         for (i = 0; i < tun->numqueues; i++) {
2718                 tfile = rtnl_dereference(tun->tfiles[i]);
2719                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2720         }
2721 }
2722
2723 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2724 {
2725         struct tun_file *tfile = file->private_data;
2726         struct tun_struct *tun;
2727         int ret = 0;
2728
2729         rtnl_lock();
2730
2731         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2732                 tun = tfile->detached;
2733                 if (!tun) {
2734                         ret = -EINVAL;
2735                         goto unlock;
2736                 }
2737                 ret = security_tun_dev_attach_queue(tun->security);
2738                 if (ret < 0)
2739                         goto unlock;
2740                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2741         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2742                 tun = rtnl_dereference(tfile->tun);
2743                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2744                         ret = -EINVAL;
2745                 else
2746                         __tun_detach(tfile, false);
2747         } else
2748                 ret = -EINVAL;
2749
2750 unlock:
2751         rtnl_unlock();
2752         return ret;
2753 }
2754
2755 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2756                         void __user *data)
2757 {
2758         struct bpf_prog *prog;
2759         int fd;
2760
2761         if (copy_from_user(&fd, data, sizeof(fd)))
2762                 return -EFAULT;
2763
2764         if (fd == -1) {
2765                 prog = NULL;
2766         } else {
2767                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2768                 if (IS_ERR(prog))
2769                         return PTR_ERR(prog);
2770         }
2771
2772         return __tun_set_ebpf(tun, prog_p, prog);
2773 }
2774
2775 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2776                             unsigned long arg, int ifreq_len)
2777 {
2778         struct tun_file *tfile = file->private_data;
2779         struct tun_struct *tun;
2780         void __user* argp = (void __user*)arg;
2781         struct ifreq ifr;
2782         kuid_t owner;
2783         kgid_t group;
2784         int sndbuf;
2785         int vnet_hdr_sz;
2786         unsigned int ifindex;
2787         int le;
2788         int ret;
2789
2790         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2791                 if (copy_from_user(&ifr, argp, ifreq_len))
2792                         return -EFAULT;
2793         } else {
2794                 memset(&ifr, 0, sizeof(ifr));
2795         }
2796         if (cmd == TUNGETFEATURES) {
2797                 /* Currently this just means: "what IFF flags are valid?".
2798                  * This is needed because we never checked for invalid flags on
2799                  * TUNSETIFF.
2800                  */
2801                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2802                                 (unsigned int __user*)argp);
2803         } else if (cmd == TUNSETQUEUE)
2804                 return tun_set_queue(file, &ifr);
2805
2806         ret = 0;
2807         rtnl_lock();
2808
2809         tun = tun_get(tfile);
2810         if (cmd == TUNSETIFF) {
2811                 ret = -EEXIST;
2812                 if (tun)
2813                         goto unlock;
2814
2815                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2816
2817                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2818
2819                 if (ret)
2820                         goto unlock;
2821
2822                 if (copy_to_user(argp, &ifr, ifreq_len))
2823                         ret = -EFAULT;
2824                 goto unlock;
2825         }
2826         if (cmd == TUNSETIFINDEX) {
2827                 ret = -EPERM;
2828                 if (tun)
2829                         goto unlock;
2830
2831                 ret = -EFAULT;
2832                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2833                         goto unlock;
2834
2835                 ret = 0;
2836                 tfile->ifindex = ifindex;
2837                 goto unlock;
2838         }
2839
2840         ret = -EBADFD;
2841         if (!tun)
2842                 goto unlock;
2843
2844         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2845
2846         ret = 0;
2847         switch (cmd) {
2848         case TUNGETIFF:
2849                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2850
2851                 if (tfile->detached)
2852                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2853                 if (!tfile->socket.sk->sk_filter)
2854                         ifr.ifr_flags |= IFF_NOFILTER;
2855
2856                 if (copy_to_user(argp, &ifr, ifreq_len))
2857                         ret = -EFAULT;
2858                 break;
2859
2860         case TUNSETNOCSUM:
2861                 /* Disable/Enable checksum */
2862
2863                 /* [unimplemented] */
2864                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2865                           arg ? "disabled" : "enabled");
2866                 break;
2867
2868         case TUNSETPERSIST:
2869                 /* Disable/Enable persist mode. Keep an extra reference to the
2870                  * module to prevent the module being unprobed.
2871                  */
2872                 if (arg && !(tun->flags & IFF_PERSIST)) {
2873                         tun->flags |= IFF_PERSIST;
2874                         __module_get(THIS_MODULE);
2875                 }
2876                 if (!arg && (tun->flags & IFF_PERSIST)) {
2877                         tun->flags &= ~IFF_PERSIST;
2878                         module_put(THIS_MODULE);
2879                 }
2880
2881                 tun_debug(KERN_INFO, tun, "persist %s\n",
2882                           arg ? "enabled" : "disabled");
2883                 break;
2884
2885         case TUNSETOWNER:
2886                 /* Set owner of the device */
2887                 owner = make_kuid(current_user_ns(), arg);
2888                 if (!uid_valid(owner)) {
2889                         ret = -EINVAL;
2890                         break;
2891                 }
2892                 tun->owner = owner;
2893                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2894                           from_kuid(&init_user_ns, tun->owner));
2895                 break;
2896
2897         case TUNSETGROUP:
2898                 /* Set group of the device */
2899                 group = make_kgid(current_user_ns(), arg);
2900                 if (!gid_valid(group)) {
2901                         ret = -EINVAL;
2902                         break;
2903                 }
2904                 tun->group = group;
2905                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2906                           from_kgid(&init_user_ns, tun->group));
2907                 break;
2908
2909         case TUNSETLINK:
2910                 /* Only allow setting the type when the interface is down */
2911                 if (tun->dev->flags & IFF_UP) {
2912                         tun_debug(KERN_INFO, tun,
2913                                   "Linktype set failed because interface is up\n");
2914                         ret = -EBUSY;
2915                 } else {
2916                         tun->dev->type = (int) arg;
2917                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2918                                   tun->dev->type);
2919                         ret = 0;
2920                 }
2921                 break;
2922
2923 #ifdef TUN_DEBUG
2924         case TUNSETDEBUG:
2925                 tun->debug = arg;
2926                 break;
2927 #endif
2928         case TUNSETOFFLOAD:
2929                 ret = set_offload(tun, arg);
2930                 break;
2931
2932         case TUNSETTXFILTER:
2933                 /* Can be set only for TAPs */
2934                 ret = -EINVAL;
2935                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2936                         break;
2937                 ret = update_filter(&tun->txflt, (void __user *)arg);
2938                 break;
2939
2940         case SIOCGIFHWADDR:
2941                 /* Get hw address */
2942                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2943                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2944                 if (copy_to_user(argp, &ifr, ifreq_len))
2945                         ret = -EFAULT;
2946                 break;
2947
2948         case SIOCSIFHWADDR:
2949                 /* Set hw address */
2950                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2951                           ifr.ifr_hwaddr.sa_data);
2952
2953                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2954                 break;
2955
2956         case TUNGETSNDBUF:
2957                 sndbuf = tfile->socket.sk->sk_sndbuf;
2958                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2959                         ret = -EFAULT;
2960                 break;
2961
2962         case TUNSETSNDBUF:
2963                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2964                         ret = -EFAULT;
2965                         break;
2966                 }
2967                 if (sndbuf <= 0) {
2968                         ret = -EINVAL;
2969                         break;
2970                 }
2971
2972                 tun->sndbuf = sndbuf;
2973                 tun_set_sndbuf(tun);
2974                 break;
2975
2976         case TUNGETVNETHDRSZ:
2977                 vnet_hdr_sz = tun->vnet_hdr_sz;
2978                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2979                         ret = -EFAULT;
2980                 break;
2981
2982         case TUNSETVNETHDRSZ:
2983                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2984                         ret = -EFAULT;
2985                         break;
2986                 }
2987                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2988                         ret = -EINVAL;
2989                         break;
2990                 }
2991
2992                 tun->vnet_hdr_sz = vnet_hdr_sz;
2993                 break;
2994
2995         case TUNGETVNETLE:
2996                 le = !!(tun->flags & TUN_VNET_LE);
2997                 if (put_user(le, (int __user *)argp))
2998                         ret = -EFAULT;
2999                 break;
3000
3001         case TUNSETVNETLE:
3002                 if (get_user(le, (int __user *)argp)) {
3003                         ret = -EFAULT;
3004                         break;
3005                 }
3006                 if (le)
3007                         tun->flags |= TUN_VNET_LE;
3008                 else
3009                         tun->flags &= ~TUN_VNET_LE;
3010                 break;
3011
3012         case TUNGETVNETBE:
3013                 ret = tun_get_vnet_be(tun, argp);
3014                 break;
3015
3016         case TUNSETVNETBE:
3017                 ret = tun_set_vnet_be(tun, argp);
3018                 break;
3019
3020         case TUNATTACHFILTER:
3021                 /* Can be set only for TAPs */
3022                 ret = -EINVAL;
3023                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3024                         break;
3025                 ret = -EFAULT;
3026                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3027                         break;
3028
3029                 ret = tun_attach_filter(tun);
3030                 break;
3031
3032         case TUNDETACHFILTER:
3033                 /* Can be set only for TAPs */
3034                 ret = -EINVAL;
3035                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3036                         break;
3037                 ret = 0;
3038                 tun_detach_filter(tun, tun->numqueues);
3039                 break;
3040
3041         case TUNGETFILTER:
3042                 ret = -EINVAL;
3043                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3044                         break;
3045                 ret = -EFAULT;
3046                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3047                         break;
3048                 ret = 0;
3049                 break;
3050
3051         case TUNSETSTEERINGEBPF:
3052                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3053                 break;
3054
3055         case TUNSETFILTEREBPF:
3056                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3057                 break;
3058
3059         default:
3060                 ret = -EINVAL;
3061                 break;
3062         }
3063
3064 unlock:
3065         rtnl_unlock();
3066         if (tun)
3067                 tun_put(tun);
3068         return ret;
3069 }
3070
3071 static long tun_chr_ioctl(struct file *file,
3072                           unsigned int cmd, unsigned long arg)
3073 {
3074         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3075 }
3076
3077 #ifdef CONFIG_COMPAT
3078 static long tun_chr_compat_ioctl(struct file *file,
3079                          unsigned int cmd, unsigned long arg)
3080 {
3081         switch (cmd) {
3082         case TUNSETIFF:
3083         case TUNGETIFF:
3084         case TUNSETTXFILTER:
3085         case TUNGETSNDBUF:
3086         case TUNSETSNDBUF:
3087         case SIOCGIFHWADDR:
3088         case SIOCSIFHWADDR:
3089                 arg = (unsigned long)compat_ptr(arg);
3090                 break;
3091         default:
3092                 arg = (compat_ulong_t)arg;
3093                 break;
3094         }
3095
3096         /*
3097          * compat_ifreq is shorter than ifreq, so we must not access beyond
3098          * the end of that structure. All fields that are used in this
3099          * driver are compatible though, we don't need to convert the
3100          * contents.
3101          */
3102         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3103 }
3104 #endif /* CONFIG_COMPAT */
3105
3106 static int tun_chr_fasync(int fd, struct file *file, int on)
3107 {
3108         struct tun_file *tfile = file->private_data;
3109         int ret;
3110
3111         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3112                 goto out;
3113
3114         if (on) {
3115                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3116                 tfile->flags |= TUN_FASYNC;
3117         } else
3118                 tfile->flags &= ~TUN_FASYNC;
3119         ret = 0;
3120 out:
3121         return ret;
3122 }
3123
3124 static int tun_chr_open(struct inode *inode, struct file * file)
3125 {
3126         struct net *net = current->nsproxy->net_ns;
3127         struct tun_file *tfile;
3128
3129         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3130
3131         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3132                                             &tun_proto, 0);
3133         if (!tfile)
3134                 return -ENOMEM;
3135         RCU_INIT_POINTER(tfile->tun, NULL);
3136         tfile->flags = 0;
3137         tfile->ifindex = 0;
3138
3139         init_waitqueue_head(&tfile->wq.wait);
3140         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3141
3142         tfile->socket.file = file;
3143         tfile->socket.ops = &tun_socket_ops;
3144
3145         sock_init_data(&tfile->socket, &tfile->sk);
3146
3147         tfile->sk.sk_write_space = tun_sock_write_space;
3148         tfile->sk.sk_sndbuf = INT_MAX;
3149
3150         file->private_data = tfile;
3151         INIT_LIST_HEAD(&tfile->next);
3152
3153         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3154
3155         memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
3156
3157         return 0;
3158 }
3159
3160 static int tun_chr_close(struct inode *inode, struct file *file)
3161 {
3162         struct tun_file *tfile = file->private_data;
3163
3164         tun_detach(tfile, true);
3165
3166         return 0;
3167 }
3168
3169 #ifdef CONFIG_PROC_FS
3170 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3171 {
3172         struct tun_file *tfile = file->private_data;
3173         struct tun_struct *tun;
3174         struct ifreq ifr;
3175
3176         memset(&ifr, 0, sizeof(ifr));
3177
3178         rtnl_lock();
3179         tun = tun_get(tfile);
3180         if (tun)
3181                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3182         rtnl_unlock();
3183
3184         if (tun)
3185                 tun_put(tun);
3186
3187         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3188 }
3189 #endif
3190
3191 static const struct file_operations tun_fops = {
3192         .owner  = THIS_MODULE,
3193         .llseek = no_llseek,
3194         .read_iter  = tun_chr_read_iter,
3195         .write_iter = tun_chr_write_iter,
3196         .poll   = tun_chr_poll,
3197         .unlocked_ioctl = tun_chr_ioctl,
3198 #ifdef CONFIG_COMPAT
3199         .compat_ioctl = tun_chr_compat_ioctl,
3200 #endif
3201         .open   = tun_chr_open,
3202         .release = tun_chr_close,
3203         .fasync = tun_chr_fasync,
3204 #ifdef CONFIG_PROC_FS
3205         .show_fdinfo = tun_chr_show_fdinfo,
3206 #endif
3207 };
3208
3209 static struct miscdevice tun_miscdev = {
3210         .minor = TUN_MINOR,
3211         .name = "tun",
3212         .nodename = "net/tun",
3213         .fops = &tun_fops,
3214 };
3215
3216 /* ethtool interface */
3217
3218 static int tun_get_link_ksettings(struct net_device *dev,
3219                                   struct ethtool_link_ksettings *cmd)
3220 {
3221         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3222         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3223         cmd->base.speed         = SPEED_10;
3224         cmd->base.duplex        = DUPLEX_FULL;
3225         cmd->base.port          = PORT_TP;
3226         cmd->base.phy_address   = 0;
3227         cmd->base.autoneg       = AUTONEG_DISABLE;
3228         return 0;
3229 }
3230
3231 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3232 {
3233         struct tun_struct *tun = netdev_priv(dev);
3234
3235         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3236         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3237
3238         switch (tun->flags & TUN_TYPE_MASK) {
3239         case IFF_TUN:
3240                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3241                 break;
3242         case IFF_TAP:
3243                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3244                 break;
3245         }
3246 }
3247
3248 static u32 tun_get_msglevel(struct net_device *dev)
3249 {
3250 #ifdef TUN_DEBUG
3251         struct tun_struct *tun = netdev_priv(dev);
3252         return tun->debug;
3253 #else
3254         return -EOPNOTSUPP;
3255 #endif
3256 }
3257
3258 static void tun_set_msglevel(struct net_device *dev, u32 value)
3259 {
3260 #ifdef TUN_DEBUG
3261         struct tun_struct *tun = netdev_priv(dev);
3262         tun->debug = value;
3263 #endif
3264 }
3265
3266 static int tun_get_coalesce(struct net_device *dev,
3267                             struct ethtool_coalesce *ec)
3268 {
3269         struct tun_struct *tun = netdev_priv(dev);
3270
3271         ec->rx_max_coalesced_frames = tun->rx_batched;
3272
3273         return 0;
3274 }
3275
3276 static int tun_set_coalesce(struct net_device *dev,
3277                             struct ethtool_coalesce *ec)
3278 {
3279         struct tun_struct *tun = netdev_priv(dev);
3280
3281         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3282                 tun->rx_batched = NAPI_POLL_WEIGHT;
3283         else
3284                 tun->rx_batched = ec->rx_max_coalesced_frames;
3285
3286         return 0;
3287 }
3288
3289 static const struct ethtool_ops tun_ethtool_ops = {
3290         .get_drvinfo    = tun_get_drvinfo,
3291         .get_msglevel   = tun_get_msglevel,
3292         .set_msglevel   = tun_set_msglevel,
3293         .get_link       = ethtool_op_get_link,
3294         .get_ts_info    = ethtool_op_get_ts_info,
3295         .get_coalesce   = tun_get_coalesce,
3296         .set_coalesce   = tun_set_coalesce,
3297         .get_link_ksettings = tun_get_link_ksettings,
3298 };
3299
3300 static int tun_queue_resize(struct tun_struct *tun)
3301 {
3302         struct net_device *dev = tun->dev;
3303         struct tun_file *tfile;
3304         struct ptr_ring **rings;
3305         int n = tun->numqueues + tun->numdisabled;
3306         int ret, i;
3307
3308         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3309         if (!rings)
3310                 return -ENOMEM;
3311
3312         for (i = 0; i < tun->numqueues; i++) {
3313                 tfile = rtnl_dereference(tun->tfiles[i]);
3314                 rings[i] = &tfile->tx_ring;
3315         }
3316         list_for_each_entry(tfile, &tun->disabled, next)
3317                 rings[i++] = &tfile->tx_ring;
3318
3319         ret = ptr_ring_resize_multiple(rings, n,
3320                                        dev->tx_queue_len, GFP_KERNEL,
3321                                        tun_ptr_free);
3322
3323         kfree(rings);
3324         return ret;
3325 }
3326
3327 static int tun_device_event(struct notifier_block *unused,
3328                             unsigned long event, void *ptr)
3329 {
3330         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3331         struct tun_struct *tun = netdev_priv(dev);
3332
3333         if (dev->rtnl_link_ops != &tun_link_ops)
3334                 return NOTIFY_DONE;
3335
3336         switch (event) {
3337         case NETDEV_CHANGE_TX_QUEUE_LEN:
3338                 if (tun_queue_resize(tun))
3339                         return NOTIFY_BAD;
3340                 break;
3341         default:
3342                 break;
3343         }
3344
3345         return NOTIFY_DONE;
3346 }
3347
3348 static struct notifier_block tun_notifier_block __read_mostly = {
3349         .notifier_call  = tun_device_event,
3350 };
3351
3352 static int __init tun_init(void)
3353 {
3354         int ret = 0;
3355
3356         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3357
3358         ret = rtnl_link_register(&tun_link_ops);
3359         if (ret) {
3360                 pr_err("Can't register link_ops\n");
3361                 goto err_linkops;
3362         }
3363
3364         ret = misc_register(&tun_miscdev);
3365         if (ret) {
3366                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3367                 goto err_misc;
3368         }
3369
3370         ret = register_netdevice_notifier(&tun_notifier_block);
3371         if (ret) {
3372                 pr_err("Can't register netdevice notifier\n");
3373                 goto err_notifier;
3374         }
3375
3376         return  0;
3377
3378 err_notifier:
3379         misc_deregister(&tun_miscdev);
3380 err_misc:
3381         rtnl_link_unregister(&tun_link_ops);
3382 err_linkops:
3383         return ret;
3384 }
3385
3386 static void tun_cleanup(void)
3387 {
3388         misc_deregister(&tun_miscdev);
3389         rtnl_link_unregister(&tun_link_ops);
3390         unregister_netdevice_notifier(&tun_notifier_block);
3391 }
3392
3393 /* Get an underlying socket object from tun file.  Returns error unless file is
3394  * attached to a device.  The returned object works like a packet socket, it
3395  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3396  * holding a reference to the file for as long as the socket is in use. */
3397 struct socket *tun_get_socket(struct file *file)
3398 {
3399         struct tun_file *tfile;
3400         if (file->f_op != &tun_fops)
3401                 return ERR_PTR(-EINVAL);
3402         tfile = file->private_data;
3403         if (!tfile)
3404                 return ERR_PTR(-EBADFD);
3405         return &tfile->socket;
3406 }
3407 EXPORT_SYMBOL_GPL(tun_get_socket);
3408
3409 struct ptr_ring *tun_get_tx_ring(struct file *file)
3410 {
3411         struct tun_file *tfile;
3412
3413         if (file->f_op != &tun_fops)
3414                 return ERR_PTR(-EINVAL);
3415         tfile = file->private_data;
3416         if (!tfile)
3417                 return ERR_PTR(-EBADFD);
3418         return &tfile->tx_ring;
3419 }
3420 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3421
3422 module_init(tun_init);
3423 module_exit(tun_cleanup);
3424 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3425 MODULE_AUTHOR(DRV_COPYRIGHT);
3426 MODULE_LICENSE("GPL");
3427 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3428 MODULE_ALIAS("devname:net/tun");