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 <net/xdp.h>
  74 #include <linux/seq_file.h>
  75 #include <linux/uio.h>
  76 #include <linux/skb_array.h>
  77 #include <linux/bpf.h>
  78 #include <linux/bpf_trace.h>
  79 #include <linux/mutex.h>
  80
  81 #include <linux/uaccess.h>
  82 #include <linux/proc_fs.h>
  83
  84 static void tun_default_link_ksettings(struct net_device *dev,
  85                                        struct ethtool_link_ksettings *cmd);
  86
  87 /* Uncomment to enable debugging */
  88 /* #define TUN_DEBUG 1 */
  89
  90 #ifdef TUN_DEBUG
  91 static int debug;
  92
  93 #define tun_debug(level, tun, fmt, args...)                     \
  94 do {                                                            \
  95         if (tun->debug)                                         \
  96                 netdev_printk(level, tun->dev, fmt, ##args);    \
  97 } while (0)
  98 #define DBG1(level, fmt, args...)                               \
  99 do {                                                            \
 100         if (debug == 2)                                         \
 101                 printk(level fmt, ##args);                      \
 102 } while (0)
 103 #else
 104 #define tun_debug(level, tun, fmt, args...)                     \
 105 do {                                                            \
 106         if (0)                                                  \
 107                 netdev_printk(level, tun->dev, fmt, ##args);    \
 108 } while (0)
 109 #define DBG1(level, fmt, args...)                               \
 110 do {                                                            \
 111         if (0)                                                  \
 112                 printk(level fmt, ##args);                      \
 113 } while (0)
 114 #endif
 115
 116 #define TUN_HEADROOM 256
 117 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 118
 119 /* TUN device flags */
 120
 121 /* IFF_ATTACH_QUEUE is never stored in device flags,
 122  * overload it to mean fasync when stored there.
 123  */
 124 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 125 /* High bits in flags field are unused. */
 126 #define TUN_VNET_LE     0x80000000
 127 #define TUN_VNET_BE     0x40000000
 128
 129 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 130                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 131
 132 #define GOODCOPY_LEN 128
 133
 134 #define FLT_EXACT_COUNT 8
 135 struct tap_filter {
 136         unsigned int    count;    /* Number of addrs. Zero means disabled */
 137         u32             mask[2];  /* Mask of the hashed addrs */
 138         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 139 };
 140
 141 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 142  * to max number of VCPUs in guest. */
 143 #define MAX_TAP_QUEUES 256
 144 #define MAX_TAP_FLOWS  4096
 145
 146 #define TUN_FLOW_EXPIRE (3 * HZ)
 147
 148 struct tun_pcpu_stats {
 149         u64 rx_packets;
 150         u64 rx_bytes;
 151         u64 tx_packets;
 152         u64 tx_bytes;
 153         struct u64_stats_sync syncp;
 154         u32 rx_dropped;
 155         u32 tx_dropped;
 156         u32 rx_frame_errors;
 157 };
 158
 159 /* A tun_file connects an open character device to a tuntap netdevice. It
 160  * also contains all socket related structures (except sock_fprog and tap_filter)
 161  * to serve as one transmit queue for tuntap device. The sock_fprog and
 162  * tap_filter were kept in tun_struct since they were used for filtering for the
 163  * netdevice not for a specific queue (at least I didn't see the requirement for
 164  * this).
 165  *
 166  * RCU usage:
 167  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 168  * other can only be read while rcu_read_lock or rtnl_lock is held.
 169  */
 170 struct tun_file {
 171         struct sock sk;
 172         struct socket socket;
 173         struct socket_wq wq;
 174         struct tun_struct __rcu *tun;
 175         struct fasync_struct *fasync;
 176         /* only used for fasnyc */
 177         unsigned int flags;
 178         union {
 179                 u16 queue_index;
 180                 unsigned int ifindex;
 181         };
 182         struct napi_struct napi;
 183         bool napi_enabled;
 184         struct mutex napi_mutex;        /* Protects access to the above napi */
 185         struct list_head next;
 186         struct tun_struct *detached;
 187         struct ptr_ring tx_ring;
 188         struct xdp_rxq_info xdp_rxq;
 189 };
 190
 191 struct tun_flow_entry {
 192         struct hlist_node hash_link;
 193         struct rcu_head rcu;
 194         struct tun_struct *tun;
 195
 196         u32 rxhash;
 197         u32 rps_rxhash;
 198         int queue_index;
 199         unsigned long updated;
 200 };
 201
 202 #define TUN_NUM_FLOW_ENTRIES 1024
 203
 204 struct tun_prog {
 205         struct rcu_head rcu;
 206         struct bpf_prog *prog;
 207 };
 208
 209 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 210  * device, socket filter, sndbuf and vnet header size were restore when the
 211  * file were attached to a persist device.
 212  */
 213 struct tun_struct {
 214         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 215         unsigned int            numqueues;
 216         unsigned int            flags;
 217         kuid_t                  owner;
 218         kgid_t                  group;
 219
 220         struct net_device       *dev;
 221         netdev_features_t       set_features;
 222 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 223                           NETIF_F_TSO6)
 224
 225         int                     align;
 226         int                     vnet_hdr_sz;
 227         int                     sndbuf;
 228         struct tap_filter       txflt;
 229         struct sock_fprog       fprog;
 230         /* protected by rtnl lock */
 231         bool                    filter_attached;
 232 #ifdef TUN_DEBUG
 233         int debug;
 234 #endif
 235         spinlock_t lock;
 236         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 237         struct timer_list flow_gc_timer;
 238         unsigned long ageing_time;
 239         unsigned int numdisabled;
 240         struct list_head disabled;
 241         void *security;
 242         u32 flow_count;
 243         u32 rx_batched;
 244         struct tun_pcpu_stats __percpu *pcpu_stats;
 245         struct bpf_prog __rcu *xdp_prog;
 246         struct tun_prog __rcu *steering_prog;
 247         struct tun_prog __rcu *filter_prog;
 248         struct ethtool_link_ksettings link_ksettings;
 249 };
 250
 251 struct veth {
 252         __be16 h_vlan_proto;
 253         __be16 h_vlan_TCI;
 254 };
 255
 256 bool tun_is_xdp_frame(void *ptr)
 257 {
 258         return (unsigned long)ptr & TUN_XDP_FLAG;
 259 }
 260 EXPORT_SYMBOL(tun_is_xdp_frame);
 261
 262 void *tun_xdp_to_ptr(void *ptr)
 263 {
 264         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 265 }
 266 EXPORT_SYMBOL(tun_xdp_to_ptr);
 267
 268 void *tun_ptr_to_xdp(void *ptr)
 269 {
 270         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 271 }
 272 EXPORT_SYMBOL(tun_ptr_to_xdp);
 273
 274 static int tun_napi_receive(struct napi_struct *napi, int budget)
 275 {
 276         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 277         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 278         struct sk_buff_head process_queue;
 279         struct sk_buff *skb;
 280         int received = 0;
 281
 282         __skb_queue_head_init(&process_queue);
 283
 284         spin_lock(&queue->lock);
 285         skb_queue_splice_tail_init(queue, &process_queue);
 286         spin_unlock(&queue->lock);
 287
 288         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 289                 napi_gro_receive(napi, skb);
 290                 ++received;
 291         }
 292
 293         if (!skb_queue_empty(&process_queue)) {
 294                 spin_lock(&queue->lock);
 295                 skb_queue_splice(&process_queue, queue);
 296                 spin_unlock(&queue->lock);
 297         }
 298
 299         return received;
 300 }
 301
 302 static int tun_napi_poll(struct napi_struct *napi, int budget)
 303 {
 304         unsigned int received;
 305
 306         received = tun_napi_receive(napi, budget);
 307
 308         if (received < budget)
 309                 napi_complete_done(napi, received);
 310
 311         return received;
 312 }
 313
 314 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 315                           bool napi_en)
 316 {
 317         tfile->napi_enabled = napi_en;
 318         if (napi_en) {
 319                 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 320                                NAPI_POLL_WEIGHT);
 321                 napi_enable(&tfile->napi);
 322                 mutex_init(&tfile->napi_mutex);
 323         }
 324 }
 325
 326 static void tun_napi_disable(struct tun_struct *tun, struct tun_file *tfile)
 327 {
 328         if (tfile->napi_enabled)
 329                 napi_disable(&tfile->napi);
 330 }
 331
 332 static void tun_napi_del(struct tun_struct *tun, struct tun_file *tfile)
 333 {
 334         if (tfile->napi_enabled)
 335                 netif_napi_del(&tfile->napi);
 336 }
 337
 338 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
 339 {
 340         return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
 341 }
 342
 343 #ifdef CONFIG_TUN_VNET_CROSS_LE
 344 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 345 {
 346         return tun->flags & TUN_VNET_BE ? false :
 347                 virtio_legacy_is_little_endian();
 348 }
 349
 350 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 351 {
 352         int be = !!(tun->flags & TUN_VNET_BE);
 353
 354         if (put_user(be, argp))
 355                 return -EFAULT;
 356
 357         return 0;
 358 }
 359
 360 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 361 {
 362         int be;
 363
 364         if (get_user(be, argp))
 365                 return -EFAULT;
 366
 367         if (be)
 368                 tun->flags |= TUN_VNET_BE;
 369         else
 370                 tun->flags &= ~TUN_VNET_BE;
 371
 372         return 0;
 373 }
 374 #else
 375 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 376 {
 377         return virtio_legacy_is_little_endian();
 378 }
 379
 380 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 381 {
 382         return -EINVAL;
 383 }
 384
 385 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 386 {
 387         return -EINVAL;
 388 }
 389 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 390
 391 static inline bool tun_is_little_endian(struct tun_struct *tun)
 392 {
 393         return tun->flags & TUN_VNET_LE ||
 394                 tun_legacy_is_little_endian(tun);
 395 }
 396
 397 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 398 {
 399         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 400 }
 401
 402 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 403 {
 404         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 405 }
 406
 407 static inline u32 tun_hashfn(u32 rxhash)
 408 {
 409         return rxhash & 0x3ff;
 410 }
 411
 412 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 413 {
 414         struct tun_flow_entry *e;
 415
 416         hlist_for_each_entry_rcu(e, head, hash_link) {
 417                 if (e->rxhash == rxhash)
 418                         return e;
 419         }
 420         return NULL;
 421 }
 422
 423 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 424                                               struct hlist_head *head,
 425                                               u32 rxhash, u16 queue_index)
 426 {
 427         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 428
 429         if (e) {
 430                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 431                           rxhash, queue_index);
 432                 e->updated = jiffies;
 433                 e->rxhash = rxhash;
 434                 e->rps_rxhash = 0;
 435                 e->queue_index = queue_index;
 436                 e->tun = tun;
 437                 hlist_add_head_rcu(&e->hash_link, head);
 438                 ++tun->flow_count;
 439         }
 440         return e;
 441 }
 442
 443 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 444 {
 445         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 446                   e->rxhash, e->queue_index);
 447         hlist_del_rcu(&e->hash_link);
 448         kfree_rcu(e, rcu);
 449         --tun->flow_count;
 450 }
 451
 452 static void tun_flow_flush(struct tun_struct *tun)
 453 {
 454         int i;
 455
 456         spin_lock_bh(&tun->lock);
 457         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 458                 struct tun_flow_entry *e;
 459                 struct hlist_node *n;
 460
 461                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 462                         tun_flow_delete(tun, e);
 463         }
 464         spin_unlock_bh(&tun->lock);
 465 }
 466
 467 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 468 {
 469         int i;
 470
 471         spin_lock_bh(&tun->lock);
 472         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 473                 struct tun_flow_entry *e;
 474                 struct hlist_node *n;
 475
 476                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 477                         if (e->queue_index == queue_index)
 478                                 tun_flow_delete(tun, e);
 479                 }
 480         }
 481         spin_unlock_bh(&tun->lock);
 482 }
 483
 484 static void tun_flow_cleanup(struct timer_list *t)
 485 {
 486         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 487         unsigned long delay = tun->ageing_time;
 488         unsigned long next_timer = jiffies + delay;
 489         unsigned long count = 0;
 490         int i;
 491
 492         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 493
 494         spin_lock(&tun->lock);
 495         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 496                 struct tun_flow_entry *e;
 497                 struct hlist_node *n;
 498
 499                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 500                         unsigned long this_timer;
 501
 502                         this_timer = e->updated + delay;
 503                         if (time_before_eq(this_timer, jiffies)) {
 504                                 tun_flow_delete(tun, e);
 505                                 continue;
 506                         }
 507                         count++;
 508                         if (time_before(this_timer, next_timer))
 509                                 next_timer = this_timer;
 510                 }
 511         }
 512
 513         if (count)
 514                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 515         spin_unlock(&tun->lock);
 516 }
 517
 518 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 519                             struct tun_file *tfile)
 520 {
 521         struct hlist_head *head;
 522         struct tun_flow_entry *e;
 523         unsigned long delay = tun->ageing_time;
 524         u16 queue_index = tfile->queue_index;
 525
 526         if (!rxhash)
 527                 return;
 528         else
 529                 head = &tun->flows[tun_hashfn(rxhash)];
 530
 531         rcu_read_lock();
 532
 533         e = tun_flow_find(head, rxhash);
 534         if (likely(e)) {
 535                 /* TODO: keep queueing to old queue until it's empty? */
 536                 e->queue_index = queue_index;
 537                 e->updated = jiffies;
 538                 sock_rps_record_flow_hash(e->rps_rxhash);
 539         } else {
 540                 spin_lock_bh(&tun->lock);
 541                 if (!tun_flow_find(head, rxhash) &&
 542                     tun->flow_count < MAX_TAP_FLOWS)
 543                         tun_flow_create(tun, head, rxhash, queue_index);
 544
 545                 if (!timer_pending(&tun->flow_gc_timer))
 546                         mod_timer(&tun->flow_gc_timer,
 547                                   round_jiffies_up(jiffies + delay));
 548                 spin_unlock_bh(&tun->lock);
 549         }
 550
 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 void tun_ptr_free(void *ptr)
 659 {
 660         if (!ptr)
 661                 return;
 662         if (tun_is_xdp_frame(ptr)) {
 663                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 664
 665                 xdp_return_frame(xdpf);
 666         } else {
 667                 __skb_array_destroy_skb(ptr);
 668         }
 669 }
 670 EXPORT_SYMBOL_GPL(tun_ptr_free);
 671
 672 static void tun_queue_purge(struct tun_file *tfile)
 673 {
 674         void *ptr;
 675
 676         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 677                 tun_ptr_free(ptr);
 678
 679         skb_queue_purge(&tfile->sk.sk_write_queue);
 680         skb_queue_purge(&tfile->sk.sk_error_queue);
 681 }
 682
 683 static void __tun_detach(struct tun_file *tfile, bool clean)
 684 {
 685         struct tun_file *ntfile;
 686         struct tun_struct *tun;
 687
 688         tun = rtnl_dereference(tfile->tun);
 689
 690         if (tun && clean) {
 691                 tun_napi_disable(tun, tfile);
 692                 tun_napi_del(tun, tfile);
 693         }
 694
 695         if (tun && !tfile->detached) {
 696                 u16 index = tfile->queue_index;
 697                 BUG_ON(index >= tun->numqueues);
 698
 699                 rcu_assign_pointer(tun->tfiles[index],
 700                                    tun->tfiles[tun->numqueues - 1]);
 701                 ntfile = rtnl_dereference(tun->tfiles[index]);
 702                 ntfile->queue_index = index;
 703
 704                 --tun->numqueues;
 705                 if (clean) {
 706                         RCU_INIT_POINTER(tfile->tun, NULL);
 707                         sock_put(&tfile->sk);
 708                 } else
 709                         tun_disable_queue(tun, tfile);
 710
 711                 synchronize_net();
 712                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 713                 /* Drop read queue */
 714                 tun_queue_purge(tfile);
 715                 tun_set_real_num_queues(tun);
 716         } else if (tfile->detached && clean) {
 717                 tun = tun_enable_queue(tfile);
 718                 sock_put(&tfile->sk);
 719         }
 720
 721         if (clean) {
 722                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 723                         netif_carrier_off(tun->dev);
 724
 725                         if (!(tun->flags & IFF_PERSIST) &&
 726                             tun->dev->reg_state == NETREG_REGISTERED)
 727                                 unregister_netdevice(tun->dev);
 728                 }
 729                 if (tun)
 730                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 731                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 732                 sock_put(&tfile->sk);
 733         }
 734 }
 735
 736 static void tun_detach(struct tun_file *tfile, bool clean)
 737 {
 738         struct tun_struct *tun;
 739         struct net_device *dev;
 740
 741         rtnl_lock();
 742         tun = rtnl_dereference(tfile->tun);
 743         dev = tun ? tun->dev : NULL;
 744         __tun_detach(tfile, clean);
 745         if (dev)
 746                 netdev_state_change(dev);
 747         rtnl_unlock();
 748 }
 749
 750 static void tun_detach_all(struct net_device *dev)
 751 {
 752         struct tun_struct *tun = netdev_priv(dev);
 753         struct tun_file *tfile, *tmp;
 754         int i, n = tun->numqueues;
 755
 756         for (i = 0; i < n; i++) {
 757                 tfile = rtnl_dereference(tun->tfiles[i]);
 758                 BUG_ON(!tfile);
 759                 tun_napi_disable(tun, tfile);
 760                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 761                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 762                 RCU_INIT_POINTER(tfile->tun, NULL);
 763                 --tun->numqueues;
 764         }
 765         list_for_each_entry(tfile, &tun->disabled, next) {
 766                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 767                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 768                 RCU_INIT_POINTER(tfile->tun, NULL);
 769         }
 770         BUG_ON(tun->numqueues != 0);
 771
 772         synchronize_net();
 773         for (i = 0; i < n; i++) {
 774                 tfile = rtnl_dereference(tun->tfiles[i]);
 775                 tun_napi_del(tun, tfile);
 776                 /* Drop read queue */
 777                 tun_queue_purge(tfile);
 778                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 779                 sock_put(&tfile->sk);
 780         }
 781         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 782                 tun_enable_queue(tfile);
 783                 tun_queue_purge(tfile);
 784                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 785                 sock_put(&tfile->sk);
 786         }
 787         BUG_ON(tun->numdisabled != 0);
 788
 789         if (tun->flags & IFF_PERSIST)
 790                 module_put(THIS_MODULE);
 791 }
 792
 793 static int tun_attach(struct tun_struct *tun, struct file *file,
 794                       bool skip_filter, bool napi)
 795 {
 796         struct tun_file *tfile = file->private_data;
 797         struct net_device *dev = tun->dev;
 798         int err;
 799
 800         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 801         if (err < 0)
 802                 goto out;
 803
 804         err = -EINVAL;
 805         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 806                 goto out;
 807
 808         err = -EBUSY;
 809         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 810                 goto out;
 811
 812         err = -E2BIG;
 813         if (!tfile->detached &&
 814             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 815                 goto out;
 816
 817         err = 0;
 818
 819         /* Re-attach the filter to persist device */
 820         if (!skip_filter && (tun->filter_attached == true)) {
 821                 lock_sock(tfile->socket.sk);
 822                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 823                 release_sock(tfile->socket.sk);
 824                 if (!err)
 825                         goto out;
 826         }
 827
 828         if (!tfile->detached &&
 829             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 830                             GFP_KERNEL, tun_ptr_free)) {
 831                 err = -ENOMEM;
 832                 goto out;
 833         }
 834
 835         tfile->queue_index = tun->numqueues;
 836         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 837
 838         if (tfile->detached) {
 839                 /* Re-attach detached tfile, updating XDP queue_index */
 840                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 841
 842                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 843                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 844         } else {
 845                 /* Setup XDP RX-queue info, for new tfile getting attached */
 846                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 847                                        tun->dev, tfile->queue_index);
 848                 if (err < 0)
 849                         goto out;
 850                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 851                                                  MEM_TYPE_PAGE_SHARED, NULL);
 852                 if (err < 0) {
 853                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 854                         goto out;
 855                 }
 856                 err = 0;
 857         }
 858
 859         rcu_assign_pointer(tfile->tun, tun);
 860         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 861         tun->numqueues++;
 862
 863         if (tfile->detached) {
 864                 tun_enable_queue(tfile);
 865         } else {
 866                 sock_hold(&tfile->sk);
 867                 tun_napi_init(tun, tfile, napi);
 868         }
 869
 870         tun_set_real_num_queues(tun);
 871
 872         /* device is allowed to go away first, so no need to hold extra
 873          * refcnt.
 874          */
 875
 876 out:
 877         return err;
 878 }
 879
 880 static struct tun_struct *tun_get(struct tun_file *tfile)
 881 {
 882         struct tun_struct *tun;
 883
 884         rcu_read_lock();
 885         tun = rcu_dereference(tfile->tun);
 886         if (tun)
 887                 dev_hold(tun->dev);
 888         rcu_read_unlock();
 889
 890         return tun;
 891 }
 892
 893 static void tun_put(struct tun_struct *tun)
 894 {
 895         dev_put(tun->dev);
 896 }
 897
 898 /* TAP filtering */
 899 static void addr_hash_set(u32 *mask, const u8 *addr)
 900 {
 901         int n = ether_crc(ETH_ALEN, addr) >> 26;
 902         mask[n >> 5] |= (1 << (n & 31));
 903 }
 904
 905 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 906 {
 907         int n = ether_crc(ETH_ALEN, addr) >> 26;
 908         return mask[n >> 5] & (1 << (n & 31));
 909 }
 910
 911 static int update_filter(struct tap_filter *filter, void __user *arg)
 912 {
 913         struct { u8 u[ETH_ALEN]; } *addr;
 914         struct tun_filter uf;
 915         int err, alen, n, nexact;
 916
 917         if (copy_from_user(&uf, arg, sizeof(uf)))
 918                 return -EFAULT;
 919
 920         if (!uf.count) {
 921                 /* Disabled */
 922                 filter->count = 0;
 923                 return 0;
 924         }
 925
 926         alen = ETH_ALEN * uf.count;
 927         addr = memdup_user(arg + sizeof(uf), alen);
 928         if (IS_ERR(addr))
 929                 return PTR_ERR(addr);
 930
 931         /* The filter is updated without holding any locks. Which is
 932          * perfectly safe. We disable it first and in the worst
 933          * case we'll accept a few undesired packets. */
 934         filter->count = 0;
 935         wmb();
 936
 937         /* Use first set of addresses as an exact filter */
 938         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 939                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 940
 941         nexact = n;
 942
 943         /* Remaining multicast addresses are hashed,
 944          * unicast will leave the filter disabled. */
 945         memset(filter->mask, 0, sizeof(filter->mask));
 946         for (; n < uf.count; n++) {
 947                 if (!is_multicast_ether_addr(addr[n].u)) {
 948                         err = 0; /* no filter */
 949                         goto free_addr;
 950                 }
 951                 addr_hash_set(filter->mask, addr[n].u);
 952         }
 953
 954         /* For ALLMULTI just set the mask to all ones.
 955          * This overrides the mask populated above. */
 956         if ((uf.flags & TUN_FLT_ALLMULTI))
 957                 memset(filter->mask, ~0, sizeof(filter->mask));
 958
 959         /* Now enable the filter */
 960         wmb();
 961         filter->count = nexact;
 962
 963         /* Return the number of exact filters */
 964         err = nexact;
 965 free_addr:
 966         kfree(addr);
 967         return err;
 968 }
 969
 970 /* Returns: 0 - drop, !=0 - accept */
 971 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 972 {
 973         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 974          * at this point. */
 975         struct ethhdr *eh = (struct ethhdr *) skb->data;
 976         int i;
 977
 978         /* Exact match */
 979         for (i = 0; i < filter->count; i++)
 980                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 981                         return 1;
 982
 983         /* Inexact match (multicast only) */
 984         if (is_multicast_ether_addr(eh->h_dest))
 985                 return addr_hash_test(filter->mask, eh->h_dest);
 986
 987         return 0;
 988 }
 989
 990 /*
 991  * Checks whether the packet is accepted or not.
 992  * Returns: 0 - drop, !=0 - accept
 993  */
 994 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 995 {
 996         if (!filter->count)
 997                 return 1;
 998
 999         return run_filter(filter, skb);
1000 }
1001
1002 /* Network device part of the driver */
1003
1004 static const struct ethtool_ops tun_ethtool_ops;
1005
1006 /* Net device detach from fd. */
1007 static void tun_net_uninit(struct net_device *dev)
1008 {
1009         tun_detach_all(dev);
1010 }
1011
1012 /* Net device open. */
1013 static int tun_net_open(struct net_device *dev)
1014 {
1015         struct tun_struct *tun = netdev_priv(dev);
1016         int i;
1017
1018         netif_tx_start_all_queues(dev);
1019
1020         for (i = 0; i < tun->numqueues; i++) {
1021                 struct tun_file *tfile;
1022
1023                 tfile = rtnl_dereference(tun->tfiles[i]);
1024                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1025         }
1026
1027         return 0;
1028 }
1029
1030 /* Net device close. */
1031 static int tun_net_close(struct net_device *dev)
1032 {
1033         netif_tx_stop_all_queues(dev);
1034         return 0;
1035 }
1036
1037 /* Net device start xmit */
1038 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1039 {
1040 #ifdef CONFIG_RPS
1041         if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1042                 /* Select queue was not called for the skbuff, so we extract the
1043                  * RPS hash and save it into the flow_table here.
1044                  */
1045                 __u32 rxhash;
1046
1047                 rxhash = __skb_get_hash_symmetric(skb);
1048                 if (rxhash) {
1049                         struct tun_flow_entry *e;
1050                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1051                                         rxhash);
1052                         if (e)
1053                                 tun_flow_save_rps_rxhash(e, rxhash);
1054                 }
1055         }
1056 #endif
1057 }
1058
1059 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1060                                     struct sk_buff *skb,
1061                                     int len)
1062 {
1063         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1064
1065         if (prog)
1066                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1067
1068         return len;
1069 }
1070
1071 /* Net device start xmit */
1072 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1073 {
1074         struct tun_struct *tun = netdev_priv(dev);
1075         int txq = skb->queue_mapping;
1076         struct tun_file *tfile;
1077         int len = skb->len;
1078
1079         rcu_read_lock();
1080         tfile = rcu_dereference(tun->tfiles[txq]);
1081
1082         /* Drop packet if interface is not attached */
1083         if (txq >= tun->numqueues)
1084                 goto drop;
1085
1086         if (!rcu_dereference(tun->steering_prog))
1087                 tun_automq_xmit(tun, skb);
1088
1089         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1090
1091         BUG_ON(!tfile);
1092
1093         /* Drop if the filter does not like it.
1094          * This is a noop if the filter is disabled.
1095          * Filter can be enabled only for the TAP devices. */
1096         if (!check_filter(&tun->txflt, skb))
1097                 goto drop;
1098
1099         if (tfile->socket.sk->sk_filter &&
1100             sk_filter(tfile->socket.sk, skb))
1101                 goto drop;
1102
1103         len = run_ebpf_filter(tun, skb, len);
1104         if (len == 0 || pskb_trim(skb, len))
1105                 goto drop;
1106
1107         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1108                 goto drop;
1109
1110         skb_tx_timestamp(skb);
1111
1112         /* Orphan the skb - required as we might hang on to it
1113          * for indefinite time.
1114          */
1115         skb_orphan(skb);
1116
1117         nf_reset(skb);
1118
1119         if (ptr_ring_produce(&tfile->tx_ring, skb))
1120                 goto drop;
1121
1122         /* Notify and wake up reader process */
1123         if (tfile->flags & TUN_FASYNC)
1124                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1125         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1126
1127         rcu_read_unlock();
1128         return NETDEV_TX_OK;
1129
1130 drop:
1131         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1132         skb_tx_error(skb);
1133         kfree_skb(skb);
1134         rcu_read_unlock();
1135         return NET_XMIT_DROP;
1136 }
1137
1138 static void tun_net_mclist(struct net_device *dev)
1139 {
1140         /*
1141          * This callback is supposed to deal with mc filter in
1142          * _rx_ path and has nothing to do with the _tx_ path.
1143          * In rx path we always accept everything userspace gives us.
1144          */
1145 }
1146
1147 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1148         netdev_features_t features)
1149 {
1150         struct tun_struct *tun = netdev_priv(dev);
1151
1152         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1153 }
1154 #ifdef CONFIG_NET_POLL_CONTROLLER
1155 static void tun_poll_controller(struct net_device *dev)
1156 {
1157         /*
1158          * Tun only receives frames when:
1159          * 1) the char device endpoint gets data from user space
1160          * 2) the tun socket gets a sendmsg call from user space
1161          * If NAPI is not enabled, since both of those are synchronous
1162          * operations, we are guaranteed never to have pending data when we poll
1163          * for it so there is nothing to do here but return.
1164          * We need this though so netpoll recognizes us as an interface that
1165          * supports polling, which enables bridge devices in virt setups to
1166          * still use netconsole
1167          * If NAPI is enabled, however, we need to schedule polling for all
1168          * queues unless we are using napi_gro_frags(), which we call in
1169          * process context and not in NAPI context.
1170          */
1171         struct tun_struct *tun = netdev_priv(dev);
1172
1173         if (tun->flags & IFF_NAPI) {
1174                 struct tun_file *tfile;
1175                 int i;
1176
1177                 if (tun_napi_frags_enabled(tun))
1178                         return;
1179
1180                 rcu_read_lock();
1181                 for (i = 0; i < tun->numqueues; i++) {
1182                         tfile = rcu_dereference(tun->tfiles[i]);
1183                         if (tfile->napi_enabled)
1184                                 napi_schedule(&tfile->napi);
1185                 }
1186                 rcu_read_unlock();
1187         }
1188         return;
1189 }
1190 #endif
1191
1192 static void tun_set_headroom(struct net_device *dev, int new_hr)
1193 {
1194         struct tun_struct *tun = netdev_priv(dev);
1195
1196         if (new_hr < NET_SKB_PAD)
1197                 new_hr = NET_SKB_PAD;
1198
1199         tun->align = new_hr;
1200 }
1201
1202 static void
1203 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1204 {
1205         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1206         struct tun_struct *tun = netdev_priv(dev);
1207         struct tun_pcpu_stats *p;
1208         int i;
1209
1210         for_each_possible_cpu(i) {
1211                 u64 rxpackets, rxbytes, txpackets, txbytes;
1212                 unsigned int start;
1213
1214                 p = per_cpu_ptr(tun->pcpu_stats, i);
1215                 do {
1216                         start = u64_stats_fetch_begin(&p->syncp);
1217                         rxpackets       = p->rx_packets;
1218                         rxbytes         = p->rx_bytes;
1219                         txpackets       = p->tx_packets;
1220                         txbytes         = p->tx_bytes;
1221                 } while (u64_stats_fetch_retry(&p->syncp, start));
1222
1223                 stats->rx_packets       += rxpackets;
1224                 stats->rx_bytes         += rxbytes;
1225                 stats->tx_packets       += txpackets;
1226                 stats->tx_bytes         += txbytes;
1227
1228                 /* u32 counters */
1229                 rx_dropped      += p->rx_dropped;
1230                 rx_frame_errors += p->rx_frame_errors;
1231                 tx_dropped      += p->tx_dropped;
1232         }
1233         stats->rx_dropped  = rx_dropped;
1234         stats->rx_frame_errors = rx_frame_errors;
1235         stats->tx_dropped = tx_dropped;
1236 }
1237
1238 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1239                        struct netlink_ext_ack *extack)
1240 {
1241         struct tun_struct *tun = netdev_priv(dev);
1242         struct bpf_prog *old_prog;
1243
1244         old_prog = rtnl_dereference(tun->xdp_prog);
1245         rcu_assign_pointer(tun->xdp_prog, prog);
1246         if (old_prog)
1247                 bpf_prog_put(old_prog);
1248
1249         return 0;
1250 }
1251
1252 static u32 tun_xdp_query(struct net_device *dev)
1253 {
1254         struct tun_struct *tun = netdev_priv(dev);
1255         const struct bpf_prog *xdp_prog;
1256
1257         xdp_prog = rtnl_dereference(tun->xdp_prog);
1258         if (xdp_prog)
1259                 return xdp_prog->aux->id;
1260
1261         return 0;
1262 }
1263
1264 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1265 {
1266         switch (xdp->command) {
1267         case XDP_SETUP_PROG:
1268                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1269         case XDP_QUERY_PROG:
1270                 xdp->prog_id = tun_xdp_query(dev);
1271                 xdp->prog_attached = !!xdp->prog_id;
1272                 return 0;
1273         default:
1274                 return -EINVAL;
1275         }
1276 }
1277
1278 static const struct net_device_ops tun_netdev_ops = {
1279         .ndo_uninit             = tun_net_uninit,
1280         .ndo_open               = tun_net_open,
1281         .ndo_stop               = tun_net_close,
1282         .ndo_start_xmit         = tun_net_xmit,
1283         .ndo_fix_features       = tun_net_fix_features,
1284         .ndo_select_queue       = tun_select_queue,
1285 #ifdef CONFIG_NET_POLL_CONTROLLER
1286         .ndo_poll_controller    = tun_poll_controller,
1287 #endif
1288         .ndo_set_rx_headroom    = tun_set_headroom,
1289         .ndo_get_stats64        = tun_net_get_stats64,
1290 };
1291
1292 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1293 {
1294         /* Notify and wake up reader process */
1295         if (tfile->flags & TUN_FASYNC)
1296                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1297         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1298 }
1299
1300 static int tun_xdp_xmit(struct net_device *dev, int n,
1301                         struct xdp_frame **frames, u32 flags)
1302 {
1303         struct tun_struct *tun = netdev_priv(dev);
1304         struct tun_file *tfile;
1305         u32 numqueues;
1306         int drops = 0;
1307         int cnt = n;
1308         int i;
1309
1310         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1311                 return -EINVAL;
1312
1313         rcu_read_lock();
1314
1315         numqueues = READ_ONCE(tun->numqueues);
1316         if (!numqueues) {
1317                 rcu_read_unlock();
1318                 return -ENXIO; /* Caller will free/return all frames */
1319         }
1320
1321         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1322                                             numqueues]);
1323
1324         spin_lock(&tfile->tx_ring.producer_lock);
1325         for (i = 0; i < n; i++) {
1326                 struct xdp_frame *xdp = frames[i];
1327                 /* Encode the XDP flag into lowest bit for consumer to differ
1328                  * XDP buffer from sk_buff.
1329                  */
1330                 void *frame = tun_xdp_to_ptr(xdp);
1331
1332                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1333                         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1334                         xdp_return_frame_rx_napi(xdp);
1335                         drops++;
1336                 }
1337         }
1338         spin_unlock(&tfile->tx_ring.producer_lock);
1339
1340         if (flags & XDP_XMIT_FLUSH)
1341                 __tun_xdp_flush_tfile(tfile);
1342
1343         rcu_read_unlock();
1344         return cnt - drops;
1345 }
1346
1347 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1348 {
1349         struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1350
1351         if (unlikely(!frame))
1352                 return -EOVERFLOW;
1353
1354         return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1355 }
1356
1357 static const struct net_device_ops tap_netdev_ops = {
1358         .ndo_uninit             = tun_net_uninit,
1359         .ndo_open               = tun_net_open,
1360         .ndo_stop               = tun_net_close,
1361         .ndo_start_xmit         = tun_net_xmit,
1362         .ndo_fix_features       = tun_net_fix_features,
1363         .ndo_set_rx_mode        = tun_net_mclist,
1364         .ndo_set_mac_address    = eth_mac_addr,
1365         .ndo_validate_addr      = eth_validate_addr,
1366         .ndo_select_queue       = tun_select_queue,
1367 #ifdef CONFIG_NET_POLL_CONTROLLER
1368         .ndo_poll_controller    = tun_poll_controller,
1369 #endif
1370         .ndo_features_check     = passthru_features_check,
1371         .ndo_set_rx_headroom    = tun_set_headroom,
1372         .ndo_get_stats64        = tun_net_get_stats64,
1373         .ndo_bpf                = tun_xdp,
1374         .ndo_xdp_xmit           = tun_xdp_xmit,
1375 };
1376
1377 static void tun_flow_init(struct tun_struct *tun)
1378 {
1379         int i;
1380
1381         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1382                 INIT_HLIST_HEAD(&tun->flows[i]);
1383
1384         tun->ageing_time = TUN_FLOW_EXPIRE;
1385         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1386         mod_timer(&tun->flow_gc_timer,
1387                   round_jiffies_up(jiffies + tun->ageing_time));
1388 }
1389
1390 static void tun_flow_uninit(struct tun_struct *tun)
1391 {
1392         del_timer_sync(&tun->flow_gc_timer);
1393         tun_flow_flush(tun);
1394 }
1395
1396 #define MIN_MTU 68
1397 #define MAX_MTU 65535
1398
1399 /* Initialize net device. */
1400 static void tun_net_init(struct net_device *dev)
1401 {
1402         struct tun_struct *tun = netdev_priv(dev);
1403
1404         switch (tun->flags & TUN_TYPE_MASK) {
1405         case IFF_TUN:
1406                 dev->netdev_ops = &tun_netdev_ops;
1407
1408                 /* Point-to-Point TUN Device */
1409                 dev->hard_header_len = 0;
1410                 dev->addr_len = 0;
1411                 dev->mtu = 1500;
1412
1413                 /* Zero header length */
1414                 dev->type = ARPHRD_NONE;
1415                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1416                 break;
1417
1418         case IFF_TAP:
1419                 dev->netdev_ops = &tap_netdev_ops;
1420                 /* Ethernet TAP Device */
1421                 ether_setup(dev);
1422                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1423                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1424
1425                 eth_hw_addr_random(dev);
1426
1427                 break;
1428         }
1429
1430         dev->min_mtu = MIN_MTU;
1431         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1432 }
1433
1434 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1435 {
1436         struct sock *sk = tfile->socket.sk;
1437
1438         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1439 }
1440
1441 /* Character device part */
1442
1443 /* Poll */
1444 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1445 {
1446         struct tun_file *tfile = file->private_data;
1447         struct tun_struct *tun = tun_get(tfile);
1448         struct sock *sk;
1449         __poll_t mask = 0;
1450
1451         if (!tun)
1452                 return EPOLLERR;
1453
1454         sk = tfile->socket.sk;
1455
1456         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1457
1458         poll_wait(file, sk_sleep(sk), wait);
1459
1460         if (!ptr_ring_empty(&tfile->tx_ring))
1461                 mask |= EPOLLIN | EPOLLRDNORM;
1462
1463         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1464          * guarantee EPOLLOUT to be raised by either here or
1465          * tun_sock_write_space(). Then process could get notification
1466          * after it writes to a down device and meets -EIO.
1467          */
1468         if (tun_sock_writeable(tun, tfile) ||
1469             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1470              tun_sock_writeable(tun, tfile)))
1471                 mask |= EPOLLOUT | EPOLLWRNORM;
1472
1473         if (tun->dev->reg_state != NETREG_REGISTERED)
1474                 mask = EPOLLERR;
1475
1476         tun_put(tun);
1477         return mask;
1478 }
1479
1480 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1481                                             size_t len,
1482                                             const struct iov_iter *it)
1483 {
1484         struct sk_buff *skb;
1485         size_t linear;
1486         int err;
1487         int i;
1488
1489         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1490                 return ERR_PTR(-ENOMEM);
1491
1492         local_bh_disable();
1493         skb = napi_get_frags(&tfile->napi);
1494         local_bh_enable();
1495         if (!skb)
1496                 return ERR_PTR(-ENOMEM);
1497
1498         linear = iov_iter_single_seg_count(it);
1499         err = __skb_grow(skb, linear);
1500         if (err)
1501                 goto free;
1502
1503         skb->len = len;
1504         skb->data_len = len - linear;
1505         skb->truesize += skb->data_len;
1506
1507         for (i = 1; i < it->nr_segs; i++) {
1508                 struct page_frag *pfrag = &current->task_frag;
1509                 size_t fragsz = it->iov[i].iov_len;
1510
1511                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1512                         err = -EINVAL;
1513                         goto free;
1514                 }
1515
1516                 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1517                         err = -ENOMEM;
1518                         goto free;
1519                 }
1520
1521                 skb_fill_page_desc(skb, i - 1, pfrag->page,
1522                                    pfrag->offset, fragsz);
1523                 page_ref_inc(pfrag->page);
1524                 pfrag->offset += fragsz;
1525         }
1526
1527         return skb;
1528 free:
1529         /* frees skb and all frags allocated with napi_alloc_frag() */
1530         napi_free_frags(&tfile->napi);
1531         return ERR_PTR(err);
1532 }
1533
1534 /* prepad is the amount to reserve at front.  len is length after that.
1535  * linear is a hint as to how much to copy (usually headers). */
1536 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1537                                      size_t prepad, size_t len,
1538                                      size_t linear, int noblock)
1539 {
1540         struct sock *sk = tfile->socket.sk;
1541         struct sk_buff *skb;
1542         int err;
1543
1544         /* Under a page?  Don't bother with paged skb. */
1545         if (prepad + len < PAGE_SIZE || !linear)
1546                 linear = len;
1547
1548         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1549                                    &err, 0);
1550         if (!skb)
1551                 return ERR_PTR(err);
1552
1553         skb_reserve(skb, prepad);
1554         skb_put(skb, linear);
1555         skb->data_len = len - linear;
1556         skb->len += len - linear;
1557
1558         return skb;
1559 }
1560
1561 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1562                            struct sk_buff *skb, int more)
1563 {
1564         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1565         struct sk_buff_head process_queue;
1566         u32 rx_batched = tun->rx_batched;
1567         bool rcv = false;
1568
1569         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1570                 local_bh_disable();
1571                 netif_receive_skb(skb);
1572                 local_bh_enable();
1573                 return;
1574         }
1575
1576         spin_lock(&queue->lock);
1577         if (!more || skb_queue_len(queue) == rx_batched) {
1578                 __skb_queue_head_init(&process_queue);
1579                 skb_queue_splice_tail_init(queue, &process_queue);
1580                 rcv = true;
1581         } else {
1582                 __skb_queue_tail(queue, skb);
1583         }
1584         spin_unlock(&queue->lock);
1585
1586         if (rcv) {
1587                 struct sk_buff *nskb;
1588
1589                 local_bh_disable();
1590                 while ((nskb = __skb_dequeue(&process_queue)))
1591                         netif_receive_skb(nskb);
1592                 netif_receive_skb(skb);
1593                 local_bh_enable();
1594         }
1595 }
1596
1597 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1598                               int len, int noblock, bool zerocopy)
1599 {
1600         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1601                 return false;
1602
1603         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1604                 return false;
1605
1606         if (!noblock)
1607                 return false;
1608
1609         if (zerocopy)
1610                 return false;
1611
1612         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1613             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1614                 return false;
1615
1616         return true;
1617 }
1618
1619 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1620                                      struct tun_file *tfile,
1621                                      struct iov_iter *from,
1622                                      struct virtio_net_hdr *hdr,
1623                                      int len, int *skb_xdp)
1624 {
1625         struct page_frag *alloc_frag = &current->task_frag;
1626         struct sk_buff *skb;
1627         struct bpf_prog *xdp_prog;
1628         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1629         unsigned int delta = 0;
1630         char *buf;
1631         size_t copied;
1632         int err, pad = TUN_RX_PAD;
1633
1634         rcu_read_lock();
1635         xdp_prog = rcu_dereference(tun->xdp_prog);
1636         if (xdp_prog)
1637                 pad += TUN_HEADROOM;
1638         buflen += SKB_DATA_ALIGN(len + pad);
1639         rcu_read_unlock();
1640
1641         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1642         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1643                 return ERR_PTR(-ENOMEM);
1644
1645         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1646         copied = copy_page_from_iter(alloc_frag->page,
1647                                      alloc_frag->offset + pad,
1648                                      len, from);
1649         if (copied != len)
1650                 return ERR_PTR(-EFAULT);
1651
1652         /* There's a small window that XDP may be set after the check
1653          * of xdp_prog above, this should be rare and for simplicity
1654          * we do XDP on skb in case the headroom is not enough.
1655          */
1656         if (hdr->gso_type || !xdp_prog)
1657                 *skb_xdp = 1;
1658         else
1659                 *skb_xdp = 0;
1660
1661         local_bh_disable();
1662         rcu_read_lock();
1663         xdp_prog = rcu_dereference(tun->xdp_prog);
1664         if (xdp_prog && !*skb_xdp) {
1665                 struct xdp_buff xdp;
1666                 void *orig_data;
1667                 u32 act;
1668
1669                 xdp.data_hard_start = buf;
1670                 xdp.data = buf + pad;
1671                 xdp_set_data_meta_invalid(&xdp);
1672                 xdp.data_end = xdp.data + len;
1673                 xdp.rxq = &tfile->xdp_rxq;
1674                 orig_data = xdp.data;
1675                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1676
1677                 switch (act) {
1678                 case XDP_REDIRECT:
1679                         get_page(alloc_frag->page);
1680                         alloc_frag->offset += buflen;
1681                         err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1682                         xdp_do_flush_map();
1683                         if (err)
1684                                 goto err_redirect;
1685                         rcu_read_unlock();
1686                         local_bh_enable();
1687                         return NULL;
1688                 case XDP_TX:
1689                         get_page(alloc_frag->page);
1690                         alloc_frag->offset += buflen;
1691                         if (tun_xdp_tx(tun->dev, &xdp))
1692                                 goto err_redirect;
1693                         rcu_read_unlock();
1694                         local_bh_enable();
1695                         return NULL;
1696                 case XDP_PASS:
1697                         delta = orig_data - xdp.data;
1698                         len = xdp.data_end - xdp.data;
1699                         break;
1700                 default:
1701                         bpf_warn_invalid_xdp_action(act);
1702                         /* fall through */
1703                 case XDP_ABORTED:
1704                         trace_xdp_exception(tun->dev, xdp_prog, act);
1705                         /* fall through */
1706                 case XDP_DROP:
1707                         goto err_xdp;
1708                 }
1709         }
1710
1711         skb = build_skb(buf, buflen);
1712         if (!skb) {
1713                 rcu_read_unlock();
1714                 local_bh_enable();
1715                 return ERR_PTR(-ENOMEM);
1716         }
1717
1718         skb_reserve(skb, pad - delta);
1719         skb_put(skb, len);
1720         get_page(alloc_frag->page);
1721         alloc_frag->offset += buflen;
1722
1723         rcu_read_unlock();
1724         local_bh_enable();
1725
1726         return skb;
1727
1728 err_redirect:
1729         put_page(alloc_frag->page);
1730 err_xdp:
1731         rcu_read_unlock();
1732         local_bh_enable();
1733         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1734         return NULL;
1735 }
1736
1737 /* Get packet from user space buffer */
1738 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1739                             void *msg_control, struct iov_iter *from,
1740                             int noblock, bool more)
1741 {
1742         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1743         struct sk_buff *skb;
1744         size_t total_len = iov_iter_count(from);
1745         size_t len = total_len, align = tun->align, linear;
1746         struct virtio_net_hdr gso = { 0 };
1747         struct tun_pcpu_stats *stats;
1748         int good_linear;
1749         int copylen;
1750         bool zerocopy = false;
1751         int err;
1752         u32 rxhash = 0;
1753         int skb_xdp = 1;
1754         bool frags = tun_napi_frags_enabled(tun);
1755
1756         if (!(tun->dev->flags & IFF_UP))
1757                 return -EIO;
1758
1759         if (!(tun->flags & IFF_NO_PI)) {
1760                 if (len < sizeof(pi))
1761                         return -EINVAL;
1762                 len -= sizeof(pi);
1763
1764                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1765                         return -EFAULT;
1766         }
1767
1768         if (tun->flags & IFF_VNET_HDR) {
1769                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1770
1771                 if (len < vnet_hdr_sz)
1772                         return -EINVAL;
1773                 len -= vnet_hdr_sz;
1774
1775                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1776                         return -EFAULT;
1777
1778                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1779                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1780                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1781
1782                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1783                         return -EINVAL;
1784                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1785         }
1786
1787         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1788                 align += NET_IP_ALIGN;
1789                 if (unlikely(len < ETH_HLEN ||
1790                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1791                         return -EINVAL;
1792         }
1793
1794         good_linear = SKB_MAX_HEAD(align);
1795
1796         if (msg_control) {
1797                 struct iov_iter i = *from;
1798
1799                 /* There are 256 bytes to be copied in skb, so there is
1800                  * enough room for skb expand head in case it is used.
1801                  * The rest of the buffer is mapped from userspace.
1802                  */
1803                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1804                 if (copylen > good_linear)
1805                         copylen = good_linear;
1806                 linear = copylen;
1807                 iov_iter_advance(&i, copylen);
1808                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1809                         zerocopy = true;
1810         }
1811
1812         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1813                 /* For the packet that is not easy to be processed
1814                  * (e.g gso or jumbo packet), we will do it at after
1815                  * skb was created with generic XDP routine.
1816                  */
1817                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1818                 if (IS_ERR(skb)) {
1819                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1820                         return PTR_ERR(skb);
1821                 }
1822                 if (!skb)
1823                         return total_len;
1824         } else {
1825                 if (!zerocopy) {
1826                         copylen = len;
1827                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1828                                 linear = good_linear;
1829                         else
1830                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1831                 }
1832
1833                 if (frags) {
1834                         mutex_lock(&tfile->napi_mutex);
1835                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1836                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1837                          * If zerocopy is enabled, then this layout will be
1838                          * overwritten by zerocopy_sg_from_iter().
1839                          */
1840                         zerocopy = false;
1841                 } else {
1842                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1843                                             noblock);
1844                 }
1845
1846                 if (IS_ERR(skb)) {
1847                         if (PTR_ERR(skb) != -EAGAIN)
1848                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1849                         if (frags)
1850                                 mutex_unlock(&tfile->napi_mutex);
1851                         return PTR_ERR(skb);
1852                 }
1853
1854                 if (zerocopy)
1855                         err = zerocopy_sg_from_iter(skb, from);
1856                 else
1857                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1858
1859                 if (err) {
1860                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1861                         kfree_skb(skb);
1862                         if (frags) {
1863                                 tfile->napi.skb = NULL;
1864                                 mutex_unlock(&tfile->napi_mutex);
1865                         }
1866
1867                         return -EFAULT;
1868                 }
1869         }
1870
1871         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1872                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1873                 kfree_skb(skb);
1874                 if (frags) {
1875                         tfile->napi.skb = NULL;
1876                         mutex_unlock(&tfile->napi_mutex);
1877                 }
1878
1879                 return -EINVAL;
1880         }
1881
1882         switch (tun->flags & TUN_TYPE_MASK) {
1883         case IFF_TUN:
1884                 if (tun->flags & IFF_NO_PI) {
1885                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1886
1887                         switch (ip_version) {
1888                         case 4:
1889                                 pi.proto = htons(ETH_P_IP);
1890                                 break;
1891                         case 6:
1892                                 pi.proto = htons(ETH_P_IPV6);
1893                                 break;
1894                         default:
1895                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1896                                 kfree_skb(skb);
1897                                 return -EINVAL;
1898                         }
1899                 }
1900
1901                 skb_reset_mac_header(skb);
1902                 skb->protocol = pi.proto;
1903                 skb->dev = tun->dev;
1904                 break;
1905         case IFF_TAP:
1906                 if (!frags)
1907                         skb->protocol = eth_type_trans(skb, tun->dev);
1908                 break;
1909         }
1910
1911         /* copy skb_ubuf_info for callback when skb has no error */
1912         if (zerocopy) {
1913                 skb_shinfo(skb)->destructor_arg = msg_control;
1914                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1915                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1916         } else if (msg_control) {
1917                 struct ubuf_info *uarg = msg_control;
1918                 uarg->callback(uarg, false);
1919         }
1920
1921         skb_reset_network_header(skb);
1922         skb_probe_transport_header(skb, 0);
1923
1924         if (skb_xdp) {
1925                 struct bpf_prog *xdp_prog;
1926                 int ret;
1927
1928                 local_bh_disable();
1929                 rcu_read_lock();
1930                 xdp_prog = rcu_dereference(tun->xdp_prog);
1931                 if (xdp_prog) {
1932                         ret = do_xdp_generic(xdp_prog, skb);
1933                         if (ret != XDP_PASS) {
1934                                 rcu_read_unlock();
1935                                 local_bh_enable();
1936                                 return total_len;
1937                         }
1938                 }
1939                 rcu_read_unlock();
1940                 local_bh_enable();
1941         }
1942
1943         /* Compute the costly rx hash only if needed for flow updates.
1944          * We may get a very small possibility of OOO during switching, not
1945          * worth to optimize.
1946          */
1947         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1948             !tfile->detached)
1949                 rxhash = __skb_get_hash_symmetric(skb);
1950
1951         if (frags) {
1952                 /* Exercise flow dissector code path. */
1953                 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1954
1955                 if (unlikely(headlen > skb_headlen(skb))) {
1956                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1957                         napi_free_frags(&tfile->napi);
1958                         mutex_unlock(&tfile->napi_mutex);
1959                         WARN_ON(1);
1960                         return -ENOMEM;
1961                 }
1962
1963                 local_bh_disable();
1964                 napi_gro_frags(&tfile->napi);
1965                 local_bh_enable();
1966                 mutex_unlock(&tfile->napi_mutex);
1967         } else if (tfile->napi_enabled) {
1968                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1969                 int queue_len;
1970
1971                 spin_lock_bh(&queue->lock);
1972                 __skb_queue_tail(queue, skb);
1973                 queue_len = skb_queue_len(queue);
1974                 spin_unlock(&queue->lock);
1975
1976                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1977                         napi_schedule(&tfile->napi);
1978
1979                 local_bh_enable();
1980         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1981                 tun_rx_batched(tun, tfile, skb, more);
1982         } else {
1983                 netif_rx_ni(skb);
1984         }
1985
1986         stats = get_cpu_ptr(tun->pcpu_stats);
1987         u64_stats_update_begin(&stats->syncp);
1988         stats->rx_packets++;
1989         stats->rx_bytes += len;
1990         u64_stats_update_end(&stats->syncp);
1991         put_cpu_ptr(stats);
1992
1993         if (rxhash)
1994                 tun_flow_update(tun, rxhash, tfile);
1995
1996         return total_len;
1997 }
1998
1999 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2000 {
2001         struct file *file = iocb->ki_filp;
2002         struct tun_file *tfile = file->private_data;
2003         struct tun_struct *tun = tun_get(tfile);
2004         ssize_t result;
2005
2006         if (!tun)
2007                 return -EBADFD;
2008
2009         result = tun_get_user(tun, tfile, NULL, from,
2010                               file->f_flags & O_NONBLOCK, false);
2011
2012         tun_put(tun);
2013         return result;
2014 }
2015
2016 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2017                                 struct tun_file *tfile,
2018                                 struct xdp_frame *xdp_frame,
2019                                 struct iov_iter *iter)
2020 {
2021         int vnet_hdr_sz = 0;
2022         size_t size = xdp_frame->len;
2023         struct tun_pcpu_stats *stats;
2024         size_t ret;
2025
2026         if (tun->flags & IFF_VNET_HDR) {
2027                 struct virtio_net_hdr gso = { 0 };
2028
2029                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2030                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2031                         return -EINVAL;
2032                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2033                              sizeof(gso)))
2034                         return -EFAULT;
2035                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2036         }
2037
2038         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2039
2040         stats = get_cpu_ptr(tun->pcpu_stats);
2041         u64_stats_update_begin(&stats->syncp);
2042         stats->tx_packets++;
2043         stats->tx_bytes += ret;
2044         u64_stats_update_end(&stats->syncp);
2045         put_cpu_ptr(tun->pcpu_stats);
2046
2047         return ret;
2048 }
2049
2050 /* Put packet to the user space buffer */
2051 static ssize_t tun_put_user(struct tun_struct *tun,
2052                             struct tun_file *tfile,
2053                             struct sk_buff *skb,
2054                             struct iov_iter *iter)
2055 {
2056         struct tun_pi pi = { 0, skb->protocol };
2057         struct tun_pcpu_stats *stats;
2058         ssize_t total;
2059         int vlan_offset = 0;
2060         int vlan_hlen = 0;
2061         int vnet_hdr_sz = 0;
2062
2063         if (skb_vlan_tag_present(skb))
2064                 vlan_hlen = VLAN_HLEN;
2065
2066         if (tun->flags & IFF_VNET_HDR)
2067                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2068
2069         total = skb->len + vlan_hlen + vnet_hdr_sz;
2070
2071         if (!(tun->flags & IFF_NO_PI)) {
2072                 if (iov_iter_count(iter) < sizeof(pi))
2073                         return -EINVAL;
2074
2075                 total += sizeof(pi);
2076                 if (iov_iter_count(iter) < total) {
2077                         /* Packet will be striped */
2078                         pi.flags |= TUN_PKT_STRIP;
2079                 }
2080
2081                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2082                         return -EFAULT;
2083         }
2084
2085         if (vnet_hdr_sz) {
2086                 struct virtio_net_hdr gso;
2087
2088                 if (iov_iter_count(iter) < vnet_hdr_sz)
2089                         return -EINVAL;
2090
2091                 if (virtio_net_hdr_from_skb(skb, &gso,
2092                                             tun_is_little_endian(tun), true,
2093                                             vlan_hlen)) {
2094                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2095                         pr_err("unexpected GSO type: "
2096                                "0x%x, gso_size %d, hdr_len %d\n",
2097                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2098                                tun16_to_cpu(tun, gso.hdr_len));
2099                         print_hex_dump(KERN_ERR, "tun: ",
2100                                        DUMP_PREFIX_NONE,
2101                                        16, 1, skb->head,
2102                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2103                         WARN_ON_ONCE(1);
2104                         return -EINVAL;
2105                 }
2106
2107                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2108                         return -EFAULT;
2109
2110                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2111         }
2112
2113         if (vlan_hlen) {
2114                 int ret;
2115                 struct veth veth;
2116
2117                 veth.h_vlan_proto = skb->vlan_proto;
2118                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2119
2120                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2121
2122                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2123                 if (ret || !iov_iter_count(iter))
2124                         goto done;
2125
2126                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2127                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2128                         goto done;
2129         }
2130
2131         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2132
2133 done:
2134         /* caller is in process context, */
2135         stats = get_cpu_ptr(tun->pcpu_stats);
2136         u64_stats_update_begin(&stats->syncp);
2137         stats->tx_packets++;
2138         stats->tx_bytes += skb->len + vlan_hlen;
2139         u64_stats_update_end(&stats->syncp);
2140         put_cpu_ptr(tun->pcpu_stats);
2141
2142         return total;
2143 }
2144
2145 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2146 {
2147         DECLARE_WAITQUEUE(wait, current);
2148         void *ptr = NULL;
2149         int error = 0;
2150
2151         ptr = ptr_ring_consume(&tfile->tx_ring);
2152         if (ptr)
2153                 goto out;
2154         if (noblock) {
2155                 error = -EAGAIN;
2156                 goto out;
2157         }
2158
2159         add_wait_queue(&tfile->wq.wait, &wait);
2160         current->state = TASK_INTERRUPTIBLE;
2161
2162         while (1) {
2163                 ptr = ptr_ring_consume(&tfile->tx_ring);
2164                 if (ptr)
2165                         break;
2166                 if (signal_pending(current)) {
2167                         error = -ERESTARTSYS;
2168                         break;
2169                 }
2170                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2171                         error = -EFAULT;
2172                         break;
2173                 }
2174
2175                 schedule();
2176         }
2177
2178         current->state = TASK_RUNNING;
2179         remove_wait_queue(&tfile->wq.wait, &wait);
2180
2181 out:
2182         *err = error;
2183         return ptr;
2184 }
2185
2186 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2187                            struct iov_iter *to,
2188                            int noblock, void *ptr)
2189 {
2190         ssize_t ret;
2191         int err;
2192
2193         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2194
2195         if (!iov_iter_count(to)) {
2196                 tun_ptr_free(ptr);
2197                 return 0;
2198         }
2199
2200         if (!ptr) {
2201                 /* Read frames from ring */
2202                 ptr = tun_ring_recv(tfile, noblock, &err);
2203                 if (!ptr)
2204                         return err;
2205         }
2206
2207         if (tun_is_xdp_frame(ptr)) {
2208                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2209
2210                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2211                 xdp_return_frame(xdpf);
2212         } else {
2213                 struct sk_buff *skb = ptr;
2214
2215                 ret = tun_put_user(tun, tfile, skb, to);
2216                 if (unlikely(ret < 0))
2217                         kfree_skb(skb);
2218                 else
2219                         consume_skb(skb);
2220         }
2221
2222         return ret;
2223 }
2224
2225 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2226 {
2227         struct file *file = iocb->ki_filp;
2228         struct tun_file *tfile = file->private_data;
2229         struct tun_struct *tun = tun_get(tfile);
2230         ssize_t len = iov_iter_count(to), ret;
2231
2232         if (!tun)
2233                 return -EBADFD;
2234         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2235         ret = min_t(ssize_t, ret, len);
2236         if (ret > 0)
2237                 iocb->ki_pos = ret;
2238         tun_put(tun);
2239         return ret;
2240 }
2241
2242 static void tun_prog_free(struct rcu_head *rcu)
2243 {
2244         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2245
2246         bpf_prog_destroy(prog->prog);
2247         kfree(prog);
2248 }
2249
2250 static int __tun_set_ebpf(struct tun_struct *tun,
2251                           struct tun_prog __rcu **prog_p,
2252                           struct bpf_prog *prog)
2253 {
2254         struct tun_prog *old, *new = NULL;
2255
2256         if (prog) {
2257                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2258                 if (!new)
2259                         return -ENOMEM;
2260                 new->prog = prog;
2261         }
2262
2263         spin_lock_bh(&tun->lock);
2264         old = rcu_dereference_protected(*prog_p,
2265                                         lockdep_is_held(&tun->lock));
2266         rcu_assign_pointer(*prog_p, new);
2267         spin_unlock_bh(&tun->lock);
2268
2269         if (old)
2270                 call_rcu(&old->rcu, tun_prog_free);
2271
2272         return 0;
2273 }
2274
2275 static void tun_free_netdev(struct net_device *dev)
2276 {
2277         struct tun_struct *tun = netdev_priv(dev);
2278
2279         BUG_ON(!(list_empty(&tun->disabled)));
2280         free_percpu(tun->pcpu_stats);
2281         tun_flow_uninit(tun);
2282         security_tun_dev_free_security(tun->security);
2283         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2284         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2285 }
2286
2287 static void tun_setup(struct net_device *dev)
2288 {
2289         struct tun_struct *tun = netdev_priv(dev);
2290
2291         tun->owner = INVALID_UID;
2292         tun->group = INVALID_GID;
2293         tun_default_link_ksettings(dev, &tun->link_ksettings);
2294
2295         dev->ethtool_ops = &tun_ethtool_ops;
2296         dev->needs_free_netdev = true;
2297         dev->priv_destructor = tun_free_netdev;
2298         /* We prefer our own queue length */
2299         dev->tx_queue_len = TUN_READQ_SIZE;
2300 }
2301
2302 /* Trivial set of netlink ops to allow deleting tun or tap
2303  * device with netlink.
2304  */
2305 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2306                         struct netlink_ext_ack *extack)
2307 {
2308         return -EINVAL;
2309 }
2310
2311 static size_t tun_get_size(const struct net_device *dev)
2312 {
2313         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2314         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2315
2316         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2317                nla_total_size(sizeof(gid_t)) + /* GROUP */
2318                nla_total_size(sizeof(u8)) + /* TYPE */
2319                nla_total_size(sizeof(u8)) + /* PI */
2320                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2321                nla_total_size(sizeof(u8)) + /* PERSIST */
2322                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2323                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2324                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2325                0;
2326 }
2327
2328 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2329 {
2330         struct tun_struct *tun = netdev_priv(dev);
2331
2332         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2333                 goto nla_put_failure;
2334         if (uid_valid(tun->owner) &&
2335             nla_put_u32(skb, IFLA_TUN_OWNER,
2336                         from_kuid_munged(current_user_ns(), tun->owner)))
2337                 goto nla_put_failure;
2338         if (gid_valid(tun->group) &&
2339             nla_put_u32(skb, IFLA_TUN_GROUP,
2340                         from_kgid_munged(current_user_ns(), tun->group)))
2341                 goto nla_put_failure;
2342         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2343                 goto nla_put_failure;
2344         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2345                 goto nla_put_failure;
2346         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2347                 goto nla_put_failure;
2348         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2349                        !!(tun->flags & IFF_MULTI_QUEUE)))
2350                 goto nla_put_failure;
2351         if (tun->flags & IFF_MULTI_QUEUE) {
2352                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2353                         goto nla_put_failure;
2354                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2355                                 tun->numdisabled))
2356                         goto nla_put_failure;
2357         }
2358
2359         return 0;
2360
2361 nla_put_failure:
2362         return -EMSGSIZE;
2363 }
2364
2365 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2366         .kind           = DRV_NAME,
2367         .priv_size      = sizeof(struct tun_struct),
2368         .setup          = tun_setup,
2369         .validate       = tun_validate,
2370         .get_size       = tun_get_size,
2371         .fill_info      = tun_fill_info,
2372 };
2373
2374 static void tun_sock_write_space(struct sock *sk)
2375 {
2376         struct tun_file *tfile;
2377         wait_queue_head_t *wqueue;
2378
2379         if (!sock_writeable(sk))
2380                 return;
2381
2382         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2383                 return;
2384
2385         wqueue = sk_sleep(sk);
2386         if (wqueue && waitqueue_active(wqueue))
2387                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2388                                                 EPOLLWRNORM | EPOLLWRBAND);
2389
2390         tfile = container_of(sk, struct tun_file, sk);
2391         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2392 }
2393
2394 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2395 {
2396         int ret;
2397         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2398         struct tun_struct *tun = tun_get(tfile);
2399
2400         if (!tun)
2401                 return -EBADFD;
2402
2403         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2404                            m->msg_flags & MSG_DONTWAIT,
2405                            m->msg_flags & MSG_MORE);
2406         tun_put(tun);
2407         return ret;
2408 }
2409
2410 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2411                        int flags)
2412 {
2413         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2414         struct tun_struct *tun = tun_get(tfile);
2415         void *ptr = m->msg_control;
2416         int ret;
2417
2418         if (!tun) {
2419                 ret = -EBADFD;
2420                 goto out_free;
2421         }
2422
2423         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2424                 ret = -EINVAL;
2425                 goto out_put_tun;
2426         }
2427         if (flags & MSG_ERRQUEUE) {
2428                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2429                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2430                 goto out;
2431         }
2432         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2433         if (ret > (ssize_t)total_len) {
2434                 m->msg_flags |= MSG_TRUNC;
2435                 ret = flags & MSG_TRUNC ? ret : total_len;
2436         }
2437 out:
2438         tun_put(tun);
2439         return ret;
2440
2441 out_put_tun:
2442         tun_put(tun);
2443 out_free:
2444         tun_ptr_free(ptr);
2445         return ret;
2446 }
2447
2448 static int tun_ptr_peek_len(void *ptr)
2449 {
2450         if (likely(ptr)) {
2451                 if (tun_is_xdp_frame(ptr)) {
2452                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2453
2454                         return xdpf->len;
2455                 }
2456                 return __skb_array_len_with_tag(ptr);
2457         } else {
2458                 return 0;
2459         }
2460 }
2461
2462 static int tun_peek_len(struct socket *sock)
2463 {
2464         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2465         struct tun_struct *tun;
2466         int ret = 0;
2467
2468         tun = tun_get(tfile);
2469         if (!tun)
2470                 return 0;
2471
2472         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2473         tun_put(tun);
2474
2475         return ret;
2476 }
2477
2478 /* Ops structure to mimic raw sockets with tun */
2479 static const struct proto_ops tun_socket_ops = {
2480         .peek_len = tun_peek_len,
2481         .sendmsg = tun_sendmsg,
2482         .recvmsg = tun_recvmsg,
2483 };
2484
2485 static struct proto tun_proto = {
2486         .name           = "tun",
2487         .owner          = THIS_MODULE,
2488         .obj_size       = sizeof(struct tun_file),
2489 };
2490
2491 static int tun_flags(struct tun_struct *tun)
2492 {
2493         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2494 }
2495
2496 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2497                               char *buf)
2498 {
2499         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2500         return sprintf(buf, "0x%x\n", tun_flags(tun));
2501 }
2502
2503 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2504                               char *buf)
2505 {
2506         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2507         return uid_valid(tun->owner)?
2508                 sprintf(buf, "%u\n",
2509                         from_kuid_munged(current_user_ns(), tun->owner)):
2510                 sprintf(buf, "-1\n");
2511 }
2512
2513 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2514                               char *buf)
2515 {
2516         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2517         return gid_valid(tun->group) ?
2518                 sprintf(buf, "%u\n",
2519                         from_kgid_munged(current_user_ns(), tun->group)):
2520                 sprintf(buf, "-1\n");
2521 }
2522
2523 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2524 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2525 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2526
2527 static struct attribute *tun_dev_attrs[] = {
2528         &dev_attr_tun_flags.attr,
2529         &dev_attr_owner.attr,
2530         &dev_attr_group.attr,
2531         NULL
2532 };
2533
2534 static const struct attribute_group tun_attr_group = {
2535         .attrs = tun_dev_attrs
2536 };
2537
2538 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2539 {
2540         struct tun_struct *tun;
2541         struct tun_file *tfile = file->private_data;
2542         struct net_device *dev;
2543         int err;
2544
2545         if (tfile->detached)
2546                 return -EINVAL;
2547
2548         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2549                 if (!capable(CAP_NET_ADMIN))
2550                         return -EPERM;
2551
2552                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2553                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2554                         return -EINVAL;
2555         }
2556
2557         dev = __dev_get_by_name(net, ifr->ifr_name);
2558         if (dev) {
2559                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2560                         return -EBUSY;
2561                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2562                         tun = netdev_priv(dev);
2563                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2564                         tun = netdev_priv(dev);
2565                 else
2566                         return -EINVAL;
2567
2568                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2569                     !!(tun->flags & IFF_MULTI_QUEUE))
2570                         return -EINVAL;
2571
2572                 if (tun_not_capable(tun))
2573                         return -EPERM;
2574                 err = security_tun_dev_open(tun->security);
2575                 if (err < 0)
2576                         return err;
2577
2578                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2579                                  ifr->ifr_flags & IFF_NAPI);
2580                 if (err < 0)
2581                         return err;
2582
2583                 if (tun->flags & IFF_MULTI_QUEUE &&
2584                     (tun->numqueues + tun->numdisabled > 1)) {
2585                         /* One or more queue has already been attached, no need
2586                          * to initialize the device again.
2587                          */
2588                         netdev_state_change(dev);
2589                         return 0;
2590                 }
2591
2592                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2593                               (ifr->ifr_flags & TUN_FEATURES);
2594
2595                 netdev_state_change(dev);
2596         } else {
2597                 char *name;
2598                 unsigned long flags = 0;
2599                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2600                              MAX_TAP_QUEUES : 1;
2601
2602                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2603                         return -EPERM;
2604                 err = security_tun_dev_create();
2605                 if (err < 0)
2606                         return err;
2607
2608                 /* Set dev type */
2609                 if (ifr->ifr_flags & IFF_TUN) {
2610                         /* TUN device */
2611                         flags |= IFF_TUN;
2612                         name = "tun%d";
2613                 } else if (ifr->ifr_flags & IFF_TAP) {
2614                         /* TAP device */
2615                         flags |= IFF_TAP;
2616                         name = "tap%d";
2617                 } else
2618                         return -EINVAL;
2619
2620                 if (*ifr->ifr_name)
2621                         name = ifr->ifr_name;
2622
2623                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2624                                        NET_NAME_UNKNOWN, tun_setup, queues,
2625                                        queues);
2626
2627                 if (!dev)
2628                         return -ENOMEM;
2629                 err = dev_get_valid_name(net, dev, name);
2630                 if (err < 0)
2631                         goto err_free_dev;
2632
2633                 dev_net_set(dev, net);
2634                 dev->rtnl_link_ops = &tun_link_ops;
2635                 dev->ifindex = tfile->ifindex;
2636                 dev->sysfs_groups[0] = &tun_attr_group;
2637
2638                 tun = netdev_priv(dev);
2639                 tun->dev = dev;
2640                 tun->flags = flags;
2641                 tun->txflt.count = 0;
2642                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2643
2644                 tun->align = NET_SKB_PAD;
2645                 tun->filter_attached = false;
2646                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2647                 tun->rx_batched = 0;
2648                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2649
2650                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2651                 if (!tun->pcpu_stats) {
2652                         err = -ENOMEM;
2653                         goto err_free_dev;
2654                 }
2655
2656                 spin_lock_init(&tun->lock);
2657
2658                 err = security_tun_dev_alloc_security(&tun->security);
2659                 if (err < 0)
2660                         goto err_free_stat;
2661
2662                 tun_net_init(dev);
2663                 tun_flow_init(tun);
2664
2665                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2666                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2667                                    NETIF_F_HW_VLAN_STAG_TX;
2668                 dev->features = dev->hw_features | NETIF_F_LLTX;
2669                 dev->vlan_features = dev->features &
2670                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2671                                        NETIF_F_HW_VLAN_STAG_TX);
2672
2673                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2674                               (ifr->ifr_flags & TUN_FEATURES);
2675
2676                 INIT_LIST_HEAD(&tun->disabled);
2677                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2678                 if (err < 0)
2679                         goto err_free_flow;
2680
2681                 err = register_netdevice(tun->dev);
2682                 if (err < 0)
2683                         goto err_detach;
2684         }
2685
2686         netif_carrier_on(tun->dev);
2687
2688         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2689
2690         /* Make sure persistent devices do not get stuck in
2691          * xoff state.
2692          */
2693         if (netif_running(tun->dev))
2694                 netif_tx_wake_all_queues(tun->dev);
2695
2696         strcpy(ifr->ifr_name, tun->dev->name);
2697         return 0;
2698
2699 err_detach:
2700         tun_detach_all(dev);
2701         /* register_netdevice() already called tun_free_netdev() */
2702         goto err_free_dev;
2703
2704 err_free_flow:
2705         tun_flow_uninit(tun);
2706         security_tun_dev_free_security(tun->security);
2707 err_free_stat:
2708         free_percpu(tun->pcpu_stats);
2709 err_free_dev:
2710         free_netdev(dev);
2711         return err;
2712 }
2713
2714 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2715                        struct ifreq *ifr)
2716 {
2717         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2718
2719         strcpy(ifr->ifr_name, tun->dev->name);
2720
2721         ifr->ifr_flags = tun_flags(tun);
2722
2723 }
2724
2725 /* This is like a cut-down ethtool ops, except done via tun fd so no
2726  * privs required. */
2727 static int set_offload(struct tun_struct *tun, unsigned long arg)
2728 {
2729         netdev_features_t features = 0;
2730
2731         if (arg & TUN_F_CSUM) {
2732                 features |= NETIF_F_HW_CSUM;
2733                 arg &= ~TUN_F_CSUM;
2734
2735                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2736                         if (arg & TUN_F_TSO_ECN) {
2737                                 features |= NETIF_F_TSO_ECN;
2738                                 arg &= ~TUN_F_TSO_ECN;
2739                         }
2740                         if (arg & TUN_F_TSO4)
2741                                 features |= NETIF_F_TSO;
2742                         if (arg & TUN_F_TSO6)
2743                                 features |= NETIF_F_TSO6;
2744                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2745                 }
2746
2747                 arg &= ~TUN_F_UFO;
2748         }
2749
2750         /* This gives the user a way to test for new features in future by
2751          * trying to set them. */
2752         if (arg)
2753                 return -EINVAL;
2754
2755         tun->set_features = features;
2756         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2757         tun->dev->wanted_features |= features;
2758         netdev_update_features(tun->dev);
2759
2760         return 0;
2761 }
2762
2763 static void tun_detach_filter(struct tun_struct *tun, int n)
2764 {
2765         int i;
2766         struct tun_file *tfile;
2767
2768         for (i = 0; i < n; i++) {
2769                 tfile = rtnl_dereference(tun->tfiles[i]);
2770                 lock_sock(tfile->socket.sk);
2771                 sk_detach_filter(tfile->socket.sk);
2772                 release_sock(tfile->socket.sk);
2773         }
2774
2775         tun->filter_attached = false;
2776 }
2777
2778 static int tun_attach_filter(struct tun_struct *tun)
2779 {
2780         int i, ret = 0;
2781         struct tun_file *tfile;
2782
2783         for (i = 0; i < tun->numqueues; i++) {
2784                 tfile = rtnl_dereference(tun->tfiles[i]);
2785                 lock_sock(tfile->socket.sk);
2786                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2787                 release_sock(tfile->socket.sk);
2788                 if (ret) {
2789                         tun_detach_filter(tun, i);
2790                         return ret;
2791                 }
2792         }
2793
2794         tun->filter_attached = true;
2795         return ret;
2796 }
2797
2798 static void tun_set_sndbuf(struct tun_struct *tun)
2799 {
2800         struct tun_file *tfile;
2801         int i;
2802
2803         for (i = 0; i < tun->numqueues; i++) {
2804                 tfile = rtnl_dereference(tun->tfiles[i]);
2805                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2806         }
2807 }
2808
2809 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2810 {
2811         struct tun_file *tfile = file->private_data;
2812         struct tun_struct *tun;
2813         int ret = 0;
2814
2815         rtnl_lock();
2816
2817         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2818                 tun = tfile->detached;
2819                 if (!tun) {
2820                         ret = -EINVAL;
2821                         goto unlock;
2822                 }
2823                 ret = security_tun_dev_attach_queue(tun->security);
2824                 if (ret < 0)
2825                         goto unlock;
2826                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2827         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2828                 tun = rtnl_dereference(tfile->tun);
2829                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2830                         ret = -EINVAL;
2831                 else
2832                         __tun_detach(tfile, false);
2833         } else
2834                 ret = -EINVAL;
2835
2836         if (ret >= 0)
2837                 netdev_state_change(tun->dev);
2838
2839 unlock:
2840         rtnl_unlock();
2841         return ret;
2842 }
2843
2844 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2845                         void __user *data)
2846 {
2847         struct bpf_prog *prog;
2848         int fd;
2849
2850         if (copy_from_user(&fd, data, sizeof(fd)))
2851                 return -EFAULT;
2852
2853         if (fd == -1) {
2854                 prog = NULL;
2855         } else {
2856                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2857                 if (IS_ERR(prog))
2858                         return PTR_ERR(prog);
2859         }
2860
2861         return __tun_set_ebpf(tun, prog_p, prog);
2862 }
2863
2864 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2865                             unsigned long arg, int ifreq_len)
2866 {
2867         struct tun_file *tfile = file->private_data;
2868         struct net *net = sock_net(&tfile->sk);
2869         struct tun_struct *tun;
2870         void __user* argp = (void __user*)arg;
2871         struct ifreq ifr;
2872         kuid_t owner;
2873         kgid_t group;
2874         int sndbuf;
2875         int vnet_hdr_sz;
2876         unsigned int ifindex;
2877         int le;
2878         int ret;
2879         bool do_notify = false;
2880
2881         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2882             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2883                 if (copy_from_user(&ifr, argp, ifreq_len))
2884                         return -EFAULT;
2885         } else {
2886                 memset(&ifr, 0, sizeof(ifr));
2887         }
2888         if (cmd == TUNGETFEATURES) {
2889                 /* Currently this just means: "what IFF flags are valid?".
2890                  * This is needed because we never checked for invalid flags on
2891                  * TUNSETIFF.
2892                  */
2893                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2894                                 (unsigned int __user*)argp);
2895         } else if (cmd == TUNSETQUEUE) {
2896                 return tun_set_queue(file, &ifr);
2897         } else if (cmd == SIOCGSKNS) {
2898                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2899                         return -EPERM;
2900                 return open_related_ns(&net->ns, get_net_ns);
2901         }
2902
2903         ret = 0;
2904         rtnl_lock();
2905
2906         tun = tun_get(tfile);
2907         if (cmd == TUNSETIFF) {
2908                 ret = -EEXIST;
2909                 if (tun)
2910                         goto unlock;
2911
2912                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2913
2914                 ret = tun_set_iff(net, file, &ifr);
2915
2916                 if (ret)
2917                         goto unlock;
2918
2919                 if (copy_to_user(argp, &ifr, ifreq_len))
2920                         ret = -EFAULT;
2921                 goto unlock;
2922         }
2923         if (cmd == TUNSETIFINDEX) {
2924                 ret = -EPERM;
2925                 if (tun)
2926                         goto unlock;
2927
2928                 ret = -EFAULT;
2929                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2930                         goto unlock;
2931
2932                 ret = 0;
2933                 tfile->ifindex = ifindex;
2934                 goto unlock;
2935         }
2936
2937         ret = -EBADFD;
2938         if (!tun)
2939                 goto unlock;
2940
2941         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2942
2943         ret = 0;
2944         switch (cmd) {
2945         case TUNGETIFF:
2946                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2947
2948                 if (tfile->detached)
2949                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2950                 if (!tfile->socket.sk->sk_filter)
2951                         ifr.ifr_flags |= IFF_NOFILTER;
2952
2953                 if (copy_to_user(argp, &ifr, ifreq_len))
2954                         ret = -EFAULT;
2955                 break;
2956
2957         case TUNSETNOCSUM:
2958                 /* Disable/Enable checksum */
2959
2960                 /* [unimplemented] */
2961                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2962                           arg ? "disabled" : "enabled");
2963                 break;
2964
2965         case TUNSETPERSIST:
2966                 /* Disable/Enable persist mode. Keep an extra reference to the
2967                  * module to prevent the module being unprobed.
2968                  */
2969                 if (arg && !(tun->flags & IFF_PERSIST)) {
2970                         tun->flags |= IFF_PERSIST;
2971                         __module_get(THIS_MODULE);
2972                         do_notify = true;
2973                 }
2974                 if (!arg && (tun->flags & IFF_PERSIST)) {
2975                         tun->flags &= ~IFF_PERSIST;
2976                         module_put(THIS_MODULE);
2977                         do_notify = true;
2978                 }
2979
2980                 tun_debug(KERN_INFO, tun, "persist %s\n",
2981                           arg ? "enabled" : "disabled");
2982                 break;
2983
2984         case TUNSETOWNER:
2985                 /* Set owner of the device */
2986                 owner = make_kuid(current_user_ns(), arg);
2987                 if (!uid_valid(owner)) {
2988                         ret = -EINVAL;
2989                         break;
2990                 }
2991                 tun->owner = owner;
2992                 do_notify = true;
2993                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2994                           from_kuid(&init_user_ns, tun->owner));
2995                 break;
2996
2997         case TUNSETGROUP:
2998                 /* Set group of the device */
2999                 group = make_kgid(current_user_ns(), arg);
3000                 if (!gid_valid(group)) {
3001                         ret = -EINVAL;
3002                         break;
3003                 }
3004                 tun->group = group;
3005                 do_notify = true;
3006                 tun_debug(KERN_INFO, tun, "group set to %u\n",
3007                           from_kgid(&init_user_ns, tun->group));
3008                 break;
3009
3010         case TUNSETLINK:
3011                 /* Only allow setting the type when the interface is down */
3012                 if (tun->dev->flags & IFF_UP) {
3013                         tun_debug(KERN_INFO, tun,
3014                                   "Linktype set failed because interface is up\n");
3015                         ret = -EBUSY;
3016                 } else {
3017                         tun->dev->type = (int) arg;
3018                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3019                                   tun->dev->type);
3020                         ret = 0;
3021                 }
3022                 break;
3023
3024 #ifdef TUN_DEBUG
3025         case TUNSETDEBUG:
3026                 tun->debug = arg;
3027                 break;
3028 #endif
3029         case TUNSETOFFLOAD:
3030                 ret = set_offload(tun, arg);
3031                 break;
3032
3033         case TUNSETTXFILTER:
3034                 /* Can be set only for TAPs */
3035                 ret = -EINVAL;
3036                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3037                         break;
3038                 ret = update_filter(&tun->txflt, (void __user *)arg);
3039                 break;
3040
3041         case SIOCGIFHWADDR:
3042                 /* Get hw address */
3043                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3044                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3045                 if (copy_to_user(argp, &ifr, ifreq_len))
3046                         ret = -EFAULT;
3047                 break;
3048
3049         case SIOCSIFHWADDR:
3050                 /* Set hw address */
3051                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3052                           ifr.ifr_hwaddr.sa_data);
3053
3054                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3055                 break;
3056
3057         case TUNGETSNDBUF:
3058                 sndbuf = tfile->socket.sk->sk_sndbuf;
3059                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3060                         ret = -EFAULT;
3061                 break;
3062
3063         case TUNSETSNDBUF:
3064                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3065                         ret = -EFAULT;
3066                         break;
3067                 }
3068                 if (sndbuf <= 0) {
3069                         ret = -EINVAL;
3070                         break;
3071                 }
3072
3073                 tun->sndbuf = sndbuf;
3074                 tun_set_sndbuf(tun);
3075                 break;
3076
3077         case TUNGETVNETHDRSZ:
3078                 vnet_hdr_sz = tun->vnet_hdr_sz;
3079                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3080                         ret = -EFAULT;
3081                 break;
3082
3083         case TUNSETVNETHDRSZ:
3084                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3085                         ret = -EFAULT;
3086                         break;
3087                 }
3088                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3089                         ret = -EINVAL;
3090                         break;
3091                 }
3092
3093                 tun->vnet_hdr_sz = vnet_hdr_sz;
3094                 break;
3095
3096         case TUNGETVNETLE:
3097                 le = !!(tun->flags & TUN_VNET_LE);
3098                 if (put_user(le, (int __user *)argp))
3099                         ret = -EFAULT;
3100                 break;
3101
3102         case TUNSETVNETLE:
3103                 if (get_user(le, (int __user *)argp)) {
3104                         ret = -EFAULT;
3105                         break;
3106                 }
3107                 if (le)
3108                         tun->flags |= TUN_VNET_LE;
3109                 else
3110                         tun->flags &= ~TUN_VNET_LE;
3111                 break;
3112
3113         case TUNGETVNETBE:
3114                 ret = tun_get_vnet_be(tun, argp);
3115                 break;
3116
3117         case TUNSETVNETBE:
3118                 ret = tun_set_vnet_be(tun, argp);
3119                 break;
3120
3121         case TUNATTACHFILTER:
3122                 /* Can be set only for TAPs */
3123                 ret = -EINVAL;
3124                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3125                         break;
3126                 ret = -EFAULT;
3127                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3128                         break;
3129
3130                 ret = tun_attach_filter(tun);
3131                 break;
3132
3133         case TUNDETACHFILTER:
3134                 /* Can be set only for TAPs */
3135                 ret = -EINVAL;
3136                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3137                         break;
3138                 ret = 0;
3139                 tun_detach_filter(tun, tun->numqueues);
3140                 break;
3141
3142         case TUNGETFILTER:
3143                 ret = -EINVAL;
3144                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3145                         break;
3146                 ret = -EFAULT;
3147                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3148                         break;
3149                 ret = 0;
3150                 break;
3151
3152         case TUNSETSTEERINGEBPF:
3153                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3154                 break;
3155
3156         case TUNSETFILTEREBPF:
3157                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3158                 break;
3159
3160         default:
3161                 ret = -EINVAL;
3162                 break;
3163         }
3164
3165         if (do_notify)
3166                 netdev_state_change(tun->dev);
3167
3168 unlock:
3169         rtnl_unlock();
3170         if (tun)
3171                 tun_put(tun);
3172         return ret;
3173 }
3174
3175 static long tun_chr_ioctl(struct file *file,
3176                           unsigned int cmd, unsigned long arg)
3177 {
3178         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3179 }
3180
3181 #ifdef CONFIG_COMPAT
3182 static long tun_chr_compat_ioctl(struct file *file,
3183                          unsigned int cmd, unsigned long arg)
3184 {
3185         switch (cmd) {
3186         case TUNSETIFF:
3187         case TUNGETIFF:
3188         case TUNSETTXFILTER:
3189         case TUNGETSNDBUF:
3190         case TUNSETSNDBUF:
3191         case SIOCGIFHWADDR:
3192         case SIOCSIFHWADDR:
3193                 arg = (unsigned long)compat_ptr(arg);
3194                 break;
3195         default:
3196                 arg = (compat_ulong_t)arg;
3197                 break;
3198         }
3199
3200         /*
3201          * compat_ifreq is shorter than ifreq, so we must not access beyond
3202          * the end of that structure. All fields that are used in this
3203          * driver are compatible though, we don't need to convert the
3204          * contents.
3205          */
3206         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3207 }
3208 #endif /* CONFIG_COMPAT */
3209
3210 static int tun_chr_fasync(int fd, struct file *file, int on)
3211 {
3212         struct tun_file *tfile = file->private_data;
3213         int ret;
3214
3215         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3216                 goto out;
3217
3218         if (on) {
3219                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3220                 tfile->flags |= TUN_FASYNC;
3221         } else
3222                 tfile->flags &= ~TUN_FASYNC;
3223         ret = 0;
3224 out:
3225         return ret;
3226 }
3227
3228 static int tun_chr_open(struct inode *inode, struct file * file)
3229 {
3230         struct net *net = current->nsproxy->net_ns;
3231         struct tun_file *tfile;
3232
3233         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3234
3235         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3236                                             &tun_proto, 0);
3237         if (!tfile)
3238                 return -ENOMEM;
3239         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3240                 sk_free(&tfile->sk);
3241                 return -ENOMEM;
3242         }
3243
3244         RCU_INIT_POINTER(tfile->tun, NULL);
3245         tfile->flags = 0;
3246         tfile->ifindex = 0;
3247
3248         init_waitqueue_head(&tfile->wq.wait);
3249         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3250
3251         tfile->socket.file = file;
3252         tfile->socket.ops = &tun_socket_ops;
3253
3254         sock_init_data(&tfile->socket, &tfile->sk);
3255
3256         tfile->sk.sk_write_space = tun_sock_write_space;
3257         tfile->sk.sk_sndbuf = INT_MAX;
3258
3259         file->private_data = tfile;
3260         INIT_LIST_HEAD(&tfile->next);
3261
3262         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3263
3264         return 0;
3265 }
3266
3267 static int tun_chr_close(struct inode *inode, struct file *file)
3268 {
3269         struct tun_file *tfile = file->private_data;
3270
3271         tun_detach(tfile, true);
3272
3273         return 0;
3274 }
3275
3276 #ifdef CONFIG_PROC_FS
3277 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3278 {
3279         struct tun_file *tfile = file->private_data;
3280         struct tun_struct *tun;
3281         struct ifreq ifr;
3282
3283         memset(&ifr, 0, sizeof(ifr));
3284
3285         rtnl_lock();
3286         tun = tun_get(tfile);
3287         if (tun)
3288                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3289         rtnl_unlock();
3290
3291         if (tun)
3292                 tun_put(tun);
3293
3294         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3295 }
3296 #endif
3297
3298 static const struct file_operations tun_fops = {
3299         .owner  = THIS_MODULE,
3300         .llseek = no_llseek,
3301         .read_iter  = tun_chr_read_iter,
3302         .write_iter = tun_chr_write_iter,
3303         .poll   = tun_chr_poll,
3304         .unlocked_ioctl = tun_chr_ioctl,
3305 #ifdef CONFIG_COMPAT
3306         .compat_ioctl = tun_chr_compat_ioctl,
3307 #endif
3308         .open   = tun_chr_open,
3309         .release = tun_chr_close,
3310         .fasync = tun_chr_fasync,
3311 #ifdef CONFIG_PROC_FS
3312         .show_fdinfo = tun_chr_show_fdinfo,
3313 #endif
3314 };
3315
3316 static struct miscdevice tun_miscdev = {
3317         .minor = TUN_MINOR,
3318         .name = "tun",
3319         .nodename = "net/tun",
3320         .fops = &tun_fops,
3321 };
3322
3323 /* ethtool interface */
3324
3325 static void tun_default_link_ksettings(struct net_device *dev,
3326                                        struct ethtool_link_ksettings *cmd)
3327 {
3328         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3329         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3330         cmd->base.speed         = SPEED_10;
3331         cmd->base.duplex        = DUPLEX_FULL;
3332         cmd->base.port          = PORT_TP;
3333         cmd->base.phy_address   = 0;
3334         cmd->base.autoneg       = AUTONEG_DISABLE;
3335 }
3336
3337 static int tun_get_link_ksettings(struct net_device *dev,
3338                                   struct ethtool_link_ksettings *cmd)
3339 {
3340         struct tun_struct *tun = netdev_priv(dev);
3341
3342         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3343         return 0;
3344 }
3345
3346 static int tun_set_link_ksettings(struct net_device *dev,
3347                                   const struct ethtool_link_ksettings *cmd)
3348 {
3349         struct tun_struct *tun = netdev_priv(dev);
3350
3351         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3352         return 0;
3353 }
3354
3355 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3356 {
3357         struct tun_struct *tun = netdev_priv(dev);
3358
3359         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3360         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3361
3362         switch (tun->flags & TUN_TYPE_MASK) {
3363         case IFF_TUN:
3364                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3365                 break;
3366         case IFF_TAP:
3367                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3368                 break;
3369         }
3370 }
3371
3372 static u32 tun_get_msglevel(struct net_device *dev)
3373 {
3374 #ifdef TUN_DEBUG
3375         struct tun_struct *tun = netdev_priv(dev);
3376         return tun->debug;
3377 #else
3378         return -EOPNOTSUPP;
3379 #endif
3380 }
3381
3382 static void tun_set_msglevel(struct net_device *dev, u32 value)
3383 {
3384 #ifdef TUN_DEBUG
3385         struct tun_struct *tun = netdev_priv(dev);
3386         tun->debug = value;
3387 #endif
3388 }
3389
3390 static int tun_get_coalesce(struct net_device *dev,
3391                             struct ethtool_coalesce *ec)
3392 {
3393         struct tun_struct *tun = netdev_priv(dev);
3394
3395         ec->rx_max_coalesced_frames = tun->rx_batched;
3396
3397         return 0;
3398 }
3399
3400 static int tun_set_coalesce(struct net_device *dev,
3401                             struct ethtool_coalesce *ec)
3402 {
3403         struct tun_struct *tun = netdev_priv(dev);
3404
3405         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3406                 tun->rx_batched = NAPI_POLL_WEIGHT;
3407         else
3408                 tun->rx_batched = ec->rx_max_coalesced_frames;
3409
3410         return 0;
3411 }
3412
3413 static const struct ethtool_ops tun_ethtool_ops = {
3414         .get_drvinfo    = tun_get_drvinfo,
3415         .get_msglevel   = tun_get_msglevel,
3416         .set_msglevel   = tun_set_msglevel,
3417         .get_link       = ethtool_op_get_link,
3418         .get_ts_info    = ethtool_op_get_ts_info,
3419         .get_coalesce   = tun_get_coalesce,
3420         .set_coalesce   = tun_set_coalesce,
3421         .get_link_ksettings = tun_get_link_ksettings,
3422         .set_link_ksettings = tun_set_link_ksettings,
3423 };
3424
3425 static int tun_queue_resize(struct tun_struct *tun)
3426 {
3427         struct net_device *dev = tun->dev;
3428         struct tun_file *tfile;
3429         struct ptr_ring **rings;
3430         int n = tun->numqueues + tun->numdisabled;
3431         int ret, i;
3432
3433         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3434         if (!rings)
3435                 return -ENOMEM;
3436
3437         for (i = 0; i < tun->numqueues; i++) {
3438                 tfile = rtnl_dereference(tun->tfiles[i]);
3439                 rings[i] = &tfile->tx_ring;
3440         }
3441         list_for_each_entry(tfile, &tun->disabled, next)
3442                 rings[i++] = &tfile->tx_ring;
3443
3444         ret = ptr_ring_resize_multiple(rings, n,
3445                                        dev->tx_queue_len, GFP_KERNEL,
3446                                        tun_ptr_free);
3447
3448         kfree(rings);
3449         return ret;
3450 }
3451
3452 static int tun_device_event(struct notifier_block *unused,
3453                             unsigned long event, void *ptr)
3454 {
3455         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3456         struct tun_struct *tun = netdev_priv(dev);
3457
3458         if (dev->rtnl_link_ops != &tun_link_ops)
3459                 return NOTIFY_DONE;
3460
3461         switch (event) {
3462         case NETDEV_CHANGE_TX_QUEUE_LEN:
3463                 if (tun_queue_resize(tun))
3464                         return NOTIFY_BAD;
3465                 break;
3466         default:
3467                 break;
3468         }
3469
3470         return NOTIFY_DONE;
3471 }
3472
3473 static struct notifier_block tun_notifier_block __read_mostly = {
3474         .notifier_call  = tun_device_event,
3475 };
3476
3477 static int __init tun_init(void)
3478 {
3479         int ret = 0;
3480
3481         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3482
3483         ret = rtnl_link_register(&tun_link_ops);
3484         if (ret) {
3485                 pr_err("Can't register link_ops\n");
3486                 goto err_linkops;
3487         }
3488
3489         ret = misc_register(&tun_miscdev);
3490         if (ret) {
3491                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3492                 goto err_misc;
3493         }
3494
3495         ret = register_netdevice_notifier(&tun_notifier_block);
3496         if (ret) {
3497                 pr_err("Can't register netdevice notifier\n");
3498                 goto err_notifier;
3499         }
3500
3501         return  0;
3502
3503 err_notifier:
3504         misc_deregister(&tun_miscdev);
3505 err_misc:
3506         rtnl_link_unregister(&tun_link_ops);
3507 err_linkops:
3508         return ret;
3509 }
3510
3511 static void tun_cleanup(void)
3512 {
3513         misc_deregister(&tun_miscdev);
3514         rtnl_link_unregister(&tun_link_ops);
3515         unregister_netdevice_notifier(&tun_notifier_block);
3516 }
3517
3518 /* Get an underlying socket object from tun file.  Returns error unless file is
3519  * attached to a device.  The returned object works like a packet socket, it
3520  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3521  * holding a reference to the file for as long as the socket is in use. */
3522 struct socket *tun_get_socket(struct file *file)
3523 {
3524         struct tun_file *tfile;
3525         if (file->f_op != &tun_fops)
3526                 return ERR_PTR(-EINVAL);
3527         tfile = file->private_data;
3528         if (!tfile)
3529                 return ERR_PTR(-EBADFD);
3530         return &tfile->socket;
3531 }
3532 EXPORT_SYMBOL_GPL(tun_get_socket);
3533
3534 struct ptr_ring *tun_get_tx_ring(struct file *file)
3535 {
3536         struct tun_file *tfile;
3537
3538         if (file->f_op != &tun_fops)
3539                 return ERR_PTR(-EINVAL);
3540         tfile = file->private_data;
3541         if (!tfile)
3542                 return ERR_PTR(-EBADFD);
3543         return &tfile->tx_ring;
3544 }
3545 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3546
3547 module_init(tun_init);
3548 module_exit(tun_cleanup);
3549 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3550 MODULE_AUTHOR(DRV_COPYRIGHT);
3551 MODULE_LICENSE("GPL");
3552 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3553 MODULE_ALIAS("devname:net/tun");